qemu/qemu-img.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

5607 lines
165 KiB
C
Raw Normal View History

/*
* QEMU disk image utility
*
* Copyright (c) 2003-2008 Fabrice Bellard
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
#include "qemu/osdep.h"
#include <getopt.h>
#include "qemu/help-texts.h"
#include "qemu/qemu-progress.h"
#include "qemu-version.h"
2016-03-14 11:01:28 +03:00
#include "qapi/error.h"
2020-05-13 04:16:45 +03:00
#include "qapi/qapi-commands-block-core.h"
#include "qapi/qapi-visit-block-core.h"
#include "qapi/qobject-output-visitor.h"
#include "qapi/qmp/qjson.h"
#include "qapi/qmp/qdict.h"
#include "qemu/cutils.h"
#include "qemu/config-file.h"
#include "qemu/option.h"
#include "qemu/error-report.h"
#include "qemu/log.h"
#include "qemu/main-loop.h"
#include "qemu/module.h"
#include "qemu/sockets.h"
#include "qemu/units.h"
#include "qemu/memalign.h"
#include "qom/object_interfaces.h"
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
#include "sysemu/block-backend.h"
#include "block/block_int.h"
#include "block/blockjob.h"
#include "block/dirty-bitmap.h"
#include "block/qapi.h"
#include "crypto/init.h"
#include "trace/control.h"
#include "qemu/throttle.h"
#include "block/throttle-groups.h"
#define QEMU_IMG_VERSION "qemu-img version " QEMU_FULL_VERSION \
"\n" QEMU_COPYRIGHT "\n"
typedef struct img_cmd_t {
const char *name;
int (*handler)(int argc, char **argv);
} img_cmd_t;
enum {
OPTION_OUTPUT = 256,
OPTION_BACKING_CHAIN = 257,
OPTION_OBJECT = 258,
OPTION_IMAGE_OPTS = 259,
OPTION_PATTERN = 260,
OPTION_FLUSH_INTERVAL = 261,
OPTION_NO_DRAIN = 262,
OPTION_TARGET_IMAGE_OPTS = 263,
OPTION_SIZE = 264,
OPTION_PREALLOCATION = 265,
OPTION_SHRINK = 266,
OPTION_SALVAGE = 267,
OPTION_TARGET_IS_ZERO = 268,
2020-05-13 04:16:45 +03:00
OPTION_ADD = 269,
OPTION_REMOVE = 270,
OPTION_CLEAR = 271,
OPTION_ENABLE = 272,
OPTION_DISABLE = 273,
OPTION_MERGE = 274,
OPTION_BITMAPS = 275,
OPTION_FORCE = 276,
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps' The point of 'qemu-img convert --bitmaps' is to be a convenience for actions that are already possible through a string of smaller 'qemu-img bitmap' sub-commands. One situation not accounted for already is that if a source image contains an inconsistent bitmap (for example, because a qemu process died abruptly before flushing bitmap state), the user MUST delete those inconsistent bitmaps before anything else useful can be done with the image. We don't want to delete inconsistent bitmaps by default: although a corrupt bitmap is only a loss of optimization rather than a corruption of user-visible data, it is still nice to require the user to opt in to the fact that they are aware of the loss of the bitmap. Still, requiring the user to check 'qemu-img info' to see whether bitmaps are consistent, then use 'qemu-img bitmap --remove' to remove offenders, all before using 'qemu-img convert', is a lot more work than just adding a knob 'qemu-img convert --bitmaps --skip-broken-bitmaps' which opts in to skipping the broken bitmaps. After testing the new option, also demonstrate the way to manually fix things (either deleting bad bitmaps, or re-creating them as empty) so that it is possible to convert without the option. Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1946084 Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210709153951.2801666-4-eblake@redhat.com> [eblake: warning message tweak, test enhancements] Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2021-07-21 18:53:48 +03:00
OPTION_SKIP_BROKEN = 277,
};
typedef enum OutputFormat {
OFORMAT_JSON,
OFORMAT_HUMAN,
} OutputFormat;
/* Default to cache=writeback as data integrity is not important for qemu-img */
#define BDRV_DEFAULT_CACHE "writeback"
static void format_print(void *opaque, const char *name)
{
printf(" %s", name);
}
static G_NORETURN G_GNUC_PRINTF(1, 2)
void error_exit(const char *fmt, ...)
{
va_list ap;
va_start(ap, fmt);
error_vreport(fmt, ap);
va_end(ap);
error_printf("Try 'qemu-img --help' for more information\n");
exit(EXIT_FAILURE);
}
static G_NORETURN
void missing_argument(const char *option)
{
error_exit("missing argument for option '%s'", option);
}
static G_NORETURN
void unrecognized_option(const char *option)
{
error_exit("unrecognized option '%s'", option);
}
/* Please keep in synch with docs/tools/qemu-img.rst */
static G_NORETURN
void help(void)
{
const char *help_msg =
QEMU_IMG_VERSION
"usage: qemu-img [standard options] command [command options]\n"
"QEMU disk image utility\n"
"\n"
" '-h', '--help' display this help and exit\n"
" '-V', '--version' output version information and exit\n"
" '-T', '--trace' [[enable=]<pattern>][,events=<file>][,file=<file>]\n"
" specify tracing options\n"
"\n"
"Command syntax:\n"
#define DEF(option, callback, arg_string) \
" " arg_string "\n"
#include "qemu-img-cmds.h"
#undef DEF
"\n"
"Command parameters:\n"
" 'filename' is a disk image filename\n"
" 'objectdef' is a QEMU user creatable object definition. See the qemu(1)\n"
" manual page for a description of the object properties. The most common\n"
" object type is a 'secret', which is used to supply passwords and/or\n"
" encryption keys.\n"
" 'fmt' is the disk image format. It is guessed automatically in most cases\n"
" 'cache' is the cache mode used to write the output disk image, the valid\n"
" options are: 'none', 'writeback' (default, except for convert), 'writethrough',\n"
" 'directsync' and 'unsafe' (default for convert)\n"
" 'src_cache' is the cache mode used to read input disk images, the valid\n"
" options are the same as for the 'cache' option\n"
" 'size' is the disk image size in bytes. Optional suffixes\n"
" 'k' or 'K' (kilobyte, 1024), 'M' (megabyte, 1024k), 'G' (gigabyte, 1024M),\n"
" 'T' (terabyte, 1024G), 'P' (petabyte, 1024T) and 'E' (exabyte, 1024P) are\n"
" supported. 'b' is ignored.\n"
" 'output_filename' is the destination disk image filename\n"
" 'output_fmt' is the destination format\n"
" 'options' is a comma separated list of format specific options in a\n"
" name=value format. Use -o help for an overview of the options supported by\n"
" the used format\n"
" 'snapshot_param' is param used for internal snapshot, format\n"
" is 'snapshot.id=[ID],snapshot.name=[NAME]', or\n"
" '[ID_OR_NAME]'\n"
" '-c' indicates that target image must be compressed (qcow format only)\n"
" '-u' allows unsafe backing chains. For rebasing, it is assumed that old and\n"
" new backing file match exactly. The image doesn't need a working\n"
" backing file before rebasing in this case (useful for renaming the\n"
" backing file). For image creation, allow creating without attempting\n"
" to open the backing file.\n"
" '-h' with or without a command shows this help and lists the supported formats\n"
" '-p' show progress of command (only certain commands)\n"
" '-q' use Quiet mode - do not print any output (except errors)\n"
" '-S' indicates the consecutive number of bytes (defaults to 4k) that must\n"
" contain only zeros for qemu-img to create a sparse image during\n"
" conversion. If the number of bytes is 0, the source will not be scanned for\n"
" unallocated or zero sectors, and the destination image will always be\n"
" fully allocated\n"
" '--output' takes the format in which the output must be done (human or json)\n"
" '-n' skips the target volume creation (useful if the volume is created\n"
" prior to running qemu-img)\n"
"\n"
2020-05-13 04:16:45 +03:00
"Parameters to bitmap subcommand:\n"
" 'bitmap' is the name of the bitmap to manipulate, through one or more\n"
" actions from '--add', '--remove', '--clear', '--enable', '--disable',\n"
" or '--merge source'\n"
" '-g granularity' sets the granularity for '--add' actions\n"
" '-b source' and '-F src_fmt' tell '--merge' actions to find the source\n"
" bitmaps from an alternative file\n"
"\n"
"Parameters to check subcommand:\n"
" '-r' tries to repair any inconsistencies that are found during the check.\n"
" '-r leaks' repairs only cluster leaks, whereas '-r all' fixes all\n"
" kinds of errors, with a higher risk of choosing the wrong fix or\n"
" hiding corruption that has already occurred.\n"
"\n"
"Parameters to convert subcommand:\n"
" '--bitmaps' copies all top-level persistent bitmaps to destination\n"
" '-m' specifies how many coroutines work in parallel during the convert\n"
" process (defaults to 8)\n"
" '-W' allow to write to the target out of order rather than sequential\n"
"\n"
"Parameters to snapshot subcommand:\n"
" 'snapshot' is the name of the snapshot to create, apply or delete\n"
" '-a' applies a snapshot (revert disk to saved state)\n"
" '-c' creates a snapshot\n"
" '-d' deletes a snapshot\n"
" '-l' lists all snapshots in the given image\n"
"\n"
"Parameters to compare subcommand:\n"
" '-f' first image format\n"
" '-F' second image format\n"
" '-s' run in Strict mode - fail on different image size or sector allocation\n"
"\n"
"Parameters to dd subcommand:\n"
" 'bs=BYTES' read and write up to BYTES bytes at a time "
"(default: 512)\n"
" 'count=N' copy only N input blocks\n"
" 'if=FILE' read from FILE\n"
" 'of=FILE' write to FILE\n"
" 'skip=N' skip N bs-sized blocks at the start of input\n";
printf("%s\nSupported formats:", help_msg);
bdrv_iterate_format(format_print, NULL, false);
printf("\n\n" QEMU_HELP_BOTTOM "\n");
exit(EXIT_SUCCESS);
}
/*
* Is @list safe for accumulate_options()?
* It is when multiple of them can be joined together separated by ','.
* To make that work, @list must not start with ',' (or else a
* separating ',' preceding it gets escaped), and it must not end with
* an odd number of ',' (or else a separating ',' following it gets
* escaped), or be empty (or else a separating ',' preceding it can
* escape a separating ',' following it).
*
*/
static bool is_valid_option_list(const char *list)
{
size_t len = strlen(list);
size_t i;
if (!list[0] || list[0] == ',') {
return false;
}
for (i = len; i > 0 && list[i - 1] == ','; i--) {
}
if ((len - i) % 2) {
return false;
}
return true;
}
static int accumulate_options(char **options, char *list)
{
char *new_options;
if (!is_valid_option_list(list)) {
error_report("Invalid option list: %s", list);
return -1;
}
if (!*options) {
*options = g_strdup(list);
} else {
new_options = g_strdup_printf("%s,%s", *options, list);
g_free(*options);
*options = new_options;
}
return 0;
}
static QemuOptsList qemu_source_opts = {
.name = "source",
.implied_opt_name = "file",
.head = QTAILQ_HEAD_INITIALIZER(qemu_source_opts.head),
.desc = {
{ }
},
};
static int G_GNUC_PRINTF(2, 3) qprintf(bool quiet, const char *fmt, ...)
{
int ret = 0;
if (!quiet) {
va_list args;
va_start(args, fmt);
ret = vprintf(fmt, args);
va_end(args);
}
return ret;
}
static int print_block_option_help(const char *filename, const char *fmt)
{
BlockDriver *drv, *proto_drv;
QemuOptsList *create_opts = NULL;
Error *local_err = NULL;
/* Find driver and parse its options */
drv = bdrv_find_format(fmt);
if (!drv) {
error_report("Unknown file format '%s'", fmt);
return 1;
}
if (!drv->create_opts) {
error_report("Format driver '%s' does not support image creation", fmt);
return 1;
}
create_opts = qemu_opts_append(create_opts, drv->create_opts);
if (filename) {
proto_drv = bdrv_find_protocol(filename, true, &local_err);
if (!proto_drv) {
error_report_err(local_err);
qemu_opts_free(create_opts);
return 1;
}
if (!proto_drv->create_opts) {
error_report("Protocol driver '%s' does not support image creation",
proto_drv->format_name);
qemu_opts_free(create_opts);
return 1;
}
create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
}
if (filename) {
printf("Supported options:\n");
} else {
printf("Supported %s options:\n", fmt);
}
qemu_opts_print_help(create_opts, false);
qemu_opts_free(create_opts);
if (!filename) {
printf("\n"
"The protocol level may support further options.\n"
"Specify the target filename to include those options.\n");
}
return 0;
}
static BlockBackend *img_open_opts(const char *optstr,
QemuOpts *opts, int flags, bool writethrough,
bool quiet, bool force_share)
{
QDict *options;
Error *local_err = NULL;
BlockBackend *blk;
options = qemu_opts_to_qdict(opts, NULL);
if (force_share) {
if (qdict_haskey(options, BDRV_OPT_FORCE_SHARE)
&& strcmp(qdict_get_str(options, BDRV_OPT_FORCE_SHARE), "on")) {
error_report("--force-share/-U conflicts with image options");
qobject_unref(options);
return NULL;
}
qdict_put_str(options, BDRV_OPT_FORCE_SHARE, "on");
}
blk = blk_new_open(NULL, NULL, options, flags, &local_err);
if (!blk) {
error_reportf_err(local_err, "Could not open '%s': ", optstr);
return NULL;
}
blk_set_enable_write_cache(blk, !writethrough);
return blk;
}
static BlockBackend *img_open_file(const char *filename,
QDict *options,
const char *fmt, int flags,
bool writethrough, bool quiet,
bool force_share)
{
BlockBackend *blk;
Error *local_err = NULL;
if (!options) {
options = qdict_new();
}
if (fmt) {
qdict_put_str(options, "driver", fmt);
}
if (force_share) {
qdict_put_bool(options, BDRV_OPT_FORCE_SHARE, true);
}
blk = blk_new_open(filename, NULL, options, flags, &local_err);
if (!blk) {
error_reportf_err(local_err, "Could not open '%s': ", filename);
return NULL;
}
blk_set_enable_write_cache(blk, !writethrough);
return blk;
}
static int img_add_key_secrets(void *opaque,
const char *name, const char *value,
Error **errp)
{
QDict *options = opaque;
if (g_str_has_suffix(name, "key-secret")) {
qdict_put_str(options, name, value);
}
return 0;
}
static BlockBackend *img_open(bool image_opts,
const char *filename,
const char *fmt, int flags, bool writethrough,
bool quiet, bool force_share)
{
BlockBackend *blk;
if (image_opts) {
QemuOpts *opts;
if (fmt) {
error_report("--image-opts and --format are mutually exclusive");
return NULL;
}
opts = qemu_opts_parse_noisily(qemu_find_opts("source"),
filename, true);
if (!opts) {
return NULL;
}
blk = img_open_opts(filename, opts, flags, writethrough, quiet,
force_share);
} else {
blk = img_open_file(filename, NULL, fmt, flags, writethrough, quiet,
force_share);
}
if (blk) {
blk_set_force_allow_inactivate(blk);
}
return blk;
}
static int add_old_style_options(const char *fmt, QemuOpts *opts,
const char *base_filename,
const char *base_fmt)
{
if (base_filename) {
if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename,
NULL)) {
error_report("Backing file not supported for file format '%s'",
fmt);
return -1;
}
}
if (base_fmt) {
if (!qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt, NULL)) {
error_report("Backing file format not supported for file "
"format '%s'", fmt);
return -1;
}
}
return 0;
}
static int64_t cvtnum_full(const char *name, const char *value, int64_t min,
int64_t max)
{
int err;
uint64_t res;
err = qemu_strtosz(value, NULL, &res);
if (err < 0 && err != -ERANGE) {
error_report("Invalid %s specified. You may use "
"k, M, G, T, P or E suffixes for", name);
error_report("kilobytes, megabytes, gigabytes, terabytes, "
"petabytes and exabytes.");
return err;
}
if (err == -ERANGE || res > max || res < min) {
error_report("Invalid %s specified. Must be between %" PRId64
" and %" PRId64 ".", name, min, max);
return -ERANGE;
}
return res;
}
static int64_t cvtnum(const char *name, const char *value)
{
return cvtnum_full(name, value, 0, INT64_MAX);
}
static int img_create(int argc, char **argv)
{
int c;
uint64_t img_size = -1;
const char *fmt = "raw";
const char *base_fmt = NULL;
const char *filename;
const char *base_filename = NULL;
char *options = NULL;
Error *local_err = NULL;
bool quiet = false;
int flags = 0;
for(;;) {
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"object", required_argument, 0, OPTION_OBJECT},
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, ":F:b:f:ho:qu",
long_options, NULL);
if (c == -1) {
break;
}
switch(c) {
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
break;
case 'F':
base_fmt = optarg;
break;
case 'b':
base_filename = optarg;
break;
case 'f':
fmt = optarg;
break;
case 'o':
if (accumulate_options(&options, optarg) < 0) {
goto fail;
}
break;
case 'q':
quiet = true;
break;
case 'u':
flags |= BDRV_O_NO_BACKING;
break;
case OPTION_OBJECT:
user_creatable_process_cmdline(optarg);
break;
}
}
/* Get the filename */
filename = (optind < argc) ? argv[optind] : NULL;
if (options && has_help_option(options)) {
g_free(options);
return print_block_option_help(filename, fmt);
}
if (optind >= argc) {
error_exit("Expecting image file name");
}
optind++;
/* Get image size, if specified */
if (optind < argc) {
int64_t sval;
sval = cvtnum("image size", argv[optind++]);
if (sval < 0) {
goto fail;
}
img_size = (uint64_t)sval;
}
if (optind != argc) {
error_exit("Unexpected argument: %s", argv[optind]);
}
bdrv_img_create(filename, fmt, base_filename, base_fmt,
options, img_size, flags, quiet, &local_err);
if (local_err) {
error_reportf_err(local_err, "%s: ", filename);
goto fail;
}
g_free(options);
return 0;
fail:
g_free(options);
return 1;
}
static void dump_json_image_check(ImageCheck *check, bool quiet)
{
GString *str;
QObject *obj;
Visitor *v = qobject_output_visitor_new(&obj);
qapi: Add new visit_complete() function Making each output visitor provide its own output collection function was the only remaining reason for exposing visitor sub-types to the rest of the code base. Add a polymorphic visit_complete() function which is a no-op for input visitors, and which populates an opaque pointer for output visitors. For maximum type-safety, also add a parameter to the output visitor constructors with a type-correct version of the output pointer, and assert that the two uses match. This approach was considered superior to either passing the output parameter only during construction (action at a distance during visit_free() feels awkward) or only during visit_complete() (defeating type safety makes it easier to use incorrectly). Most callers were function-local, and therefore a mechanical conversion; the testsuite was a bit trickier, but the previous cleanup patch minimized the churn here. The visit_complete() function may be called at most once; doing so lets us use transfer semantics rather than duplication or ref-count semantics to get the just-built output back to the caller, even though it means our behavior is not idempotent. Generated code is simplified as follows for events: |@@ -26,7 +26,7 @@ void qapi_event_send_acpi_device_ost(ACP | QDict *qmp; | Error *err = NULL; | QMPEventFuncEmit emit; |- QmpOutputVisitor *qov; |+ QObject *obj; | Visitor *v; | q_obj_ACPI_DEVICE_OST_arg param = { | info |@@ -39,8 +39,7 @@ void qapi_event_send_acpi_device_ost(ACP | | qmp = qmp_event_build_dict("ACPI_DEVICE_OST"); | |- qov = qmp_output_visitor_new(); |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(&obj); | | visit_start_struct(v, "ACPI_DEVICE_OST", NULL, 0, &err); | if (err) { |@@ -55,7 +54,8 @@ void qapi_event_send_acpi_device_ost(ACP | goto out; | } | |- qdict_put_obj(qmp, "data", qmp_output_get_qobject(qov)); |+ visit_complete(v, &obj); |+ qdict_put_obj(qmp, "data", obj); | emit(QAPI_EVENT_ACPI_DEVICE_OST, qmp, &err); and for commands: | { | Error *err = NULL; |- QmpOutputVisitor *qov = qmp_output_visitor_new(); | Visitor *v; | |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(ret_out); | visit_type_AddfdInfo(v, "unused", &ret_in, &err); |- if (err) { |- goto out; |+ if (!err) { |+ visit_complete(v, ret_out); | } |- *ret_out = qmp_output_get_qobject(qov); |- |-out: | error_propagate(errp, err); Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <1465490926-28625-13-git-send-email-eblake@redhat.com> Reviewed-by: Markus Armbruster <armbru@redhat.com> Signed-off-by: Markus Armbruster <armbru@redhat.com>
2016-06-09 19:48:43 +03:00
visit_type_ImageCheck(v, NULL, &check, &error_abort);
visit_complete(v, &obj);
str = qobject_to_json_pretty(obj, true);
assert(str != NULL);
qprintf(quiet, "%s\n", str->str);
qobject_unref(obj);
qapi: Add new visit_complete() function Making each output visitor provide its own output collection function was the only remaining reason for exposing visitor sub-types to the rest of the code base. Add a polymorphic visit_complete() function which is a no-op for input visitors, and which populates an opaque pointer for output visitors. For maximum type-safety, also add a parameter to the output visitor constructors with a type-correct version of the output pointer, and assert that the two uses match. This approach was considered superior to either passing the output parameter only during construction (action at a distance during visit_free() feels awkward) or only during visit_complete() (defeating type safety makes it easier to use incorrectly). Most callers were function-local, and therefore a mechanical conversion; the testsuite was a bit trickier, but the previous cleanup patch minimized the churn here. The visit_complete() function may be called at most once; doing so lets us use transfer semantics rather than duplication or ref-count semantics to get the just-built output back to the caller, even though it means our behavior is not idempotent. Generated code is simplified as follows for events: |@@ -26,7 +26,7 @@ void qapi_event_send_acpi_device_ost(ACP | QDict *qmp; | Error *err = NULL; | QMPEventFuncEmit emit; |- QmpOutputVisitor *qov; |+ QObject *obj; | Visitor *v; | q_obj_ACPI_DEVICE_OST_arg param = { | info |@@ -39,8 +39,7 @@ void qapi_event_send_acpi_device_ost(ACP | | qmp = qmp_event_build_dict("ACPI_DEVICE_OST"); | |- qov = qmp_output_visitor_new(); |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(&obj); | | visit_start_struct(v, "ACPI_DEVICE_OST", NULL, 0, &err); | if (err) { |@@ -55,7 +54,8 @@ void qapi_event_send_acpi_device_ost(ACP | goto out; | } | |- qdict_put_obj(qmp, "data", qmp_output_get_qobject(qov)); |+ visit_complete(v, &obj); |+ qdict_put_obj(qmp, "data", obj); | emit(QAPI_EVENT_ACPI_DEVICE_OST, qmp, &err); and for commands: | { | Error *err = NULL; |- QmpOutputVisitor *qov = qmp_output_visitor_new(); | Visitor *v; | |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(ret_out); | visit_type_AddfdInfo(v, "unused", &ret_in, &err); |- if (err) { |- goto out; |+ if (!err) { |+ visit_complete(v, ret_out); | } |- *ret_out = qmp_output_get_qobject(qov); |- |-out: | error_propagate(errp, err); Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <1465490926-28625-13-git-send-email-eblake@redhat.com> Reviewed-by: Markus Armbruster <armbru@redhat.com> Signed-off-by: Markus Armbruster <armbru@redhat.com>
2016-06-09 19:48:43 +03:00
visit_free(v);
g_string_free(str, true);
}
static void dump_human_image_check(ImageCheck *check, bool quiet)
{
if (!(check->corruptions || check->leaks || check->check_errors)) {
qprintf(quiet, "No errors were found on the image.\n");
} else {
if (check->corruptions) {
qprintf(quiet, "\n%" PRId64 " errors were found on the image.\n"
"Data may be corrupted, or further writes to the image "
"may corrupt it.\n",
check->corruptions);
}
if (check->leaks) {
qprintf(quiet,
"\n%" PRId64 " leaked clusters were found on the image.\n"
"This means waste of disk space, but no harm to data.\n",
check->leaks);
}
if (check->check_errors) {
qprintf(quiet,
"\n%" PRId64
" internal errors have occurred during the check.\n",
check->check_errors);
}
}
if (check->total_clusters != 0 && check->allocated_clusters != 0) {
qprintf(quiet, "%" PRId64 "/%" PRId64 " = %0.2f%% allocated, "
"%0.2f%% fragmented, %0.2f%% compressed clusters\n",
check->allocated_clusters, check->total_clusters,
check->allocated_clusters * 100.0 / check->total_clusters,
check->fragmented_clusters * 100.0 / check->allocated_clusters,
check->compressed_clusters * 100.0 /
check->allocated_clusters);
}
if (check->image_end_offset) {
qprintf(quiet,
"Image end offset: %" PRId64 "\n", check->image_end_offset);
}
}
static int collect_image_check(BlockDriverState *bs,
ImageCheck *check,
const char *filename,
const char *fmt,
int fix)
{
int ret;
BdrvCheckResult result;
ret = bdrv_check(bs, &result, fix);
if (ret < 0) {
return ret;
}
check->filename = g_strdup(filename);
check->format = g_strdup(bdrv_get_format_name(bs));
check->check_errors = result.check_errors;
check->corruptions = result.corruptions;
check->has_corruptions = result.corruptions != 0;
check->leaks = result.leaks;
check->has_leaks = result.leaks != 0;
check->corruptions_fixed = result.corruptions_fixed;
check->has_corruptions_fixed = result.corruptions_fixed != 0;
check->leaks_fixed = result.leaks_fixed;
check->has_leaks_fixed = result.leaks_fixed != 0;
check->image_end_offset = result.image_end_offset;
check->has_image_end_offset = result.image_end_offset != 0;
check->total_clusters = result.bfi.total_clusters;
check->has_total_clusters = result.bfi.total_clusters != 0;
check->allocated_clusters = result.bfi.allocated_clusters;
check->has_allocated_clusters = result.bfi.allocated_clusters != 0;
check->fragmented_clusters = result.bfi.fragmented_clusters;
check->has_fragmented_clusters = result.bfi.fragmented_clusters != 0;
check->compressed_clusters = result.bfi.compressed_clusters;
check->has_compressed_clusters = result.bfi.compressed_clusters != 0;
return 0;
}
/*
* Checks an image for consistency. Exit codes:
*
* 0 - Check completed, image is good
* 1 - Check not completed because of internal errors
* 2 - Check completed, image is corrupted
* 3 - Check completed, image has leaked clusters, but is good otherwise
* 63 - Checks are not supported by the image format
*/
static int img_check(int argc, char **argv)
{
int c, ret;
OutputFormat output_format = OFORMAT_HUMAN;
const char *filename, *fmt, *output, *cache;
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
BlockBackend *blk;
BlockDriverState *bs;
int fix = 0;
int flags = BDRV_O_CHECK;
bool writethrough;
ImageCheck *check;
bool quiet = false;
bool image_opts = false;
bool force_share = false;
fmt = NULL;
output = NULL;
cache = BDRV_DEFAULT_CACHE;
for(;;) {
int option_index = 0;
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"format", required_argument, 0, 'f'},
{"repair", required_argument, 0, 'r'},
{"output", required_argument, 0, OPTION_OUTPUT},
{"object", required_argument, 0, OPTION_OBJECT},
{"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{"force-share", no_argument, 0, 'U'},
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, ":hf:r:T:qU",
long_options, &option_index);
if (c == -1) {
break;
}
switch(c) {
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
break;
case 'f':
fmt = optarg;
break;
case 'r':
flags |= BDRV_O_RDWR;
if (!strcmp(optarg, "leaks")) {
fix = BDRV_FIX_LEAKS;
} else if (!strcmp(optarg, "all")) {
fix = BDRV_FIX_LEAKS | BDRV_FIX_ERRORS;
} else {
error_exit("Unknown option value for -r "
"(expecting 'leaks' or 'all'): %s", optarg);
}
break;
case OPTION_OUTPUT:
output = optarg;
break;
case 'T':
cache = optarg;
break;
case 'q':
quiet = true;
break;
case 'U':
force_share = true;
break;
case OPTION_OBJECT:
user_creatable_process_cmdline(optarg);
break;
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
}
}
if (optind != argc - 1) {
error_exit("Expecting one image file name");
}
filename = argv[optind++];
if (output && !strcmp(output, "json")) {
output_format = OFORMAT_JSON;
} else if (output && !strcmp(output, "human")) {
output_format = OFORMAT_HUMAN;
} else if (output) {
error_report("--output must be used with human or json as argument.");
return 1;
}
ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
if (ret < 0) {
error_report("Invalid source cache option: %s", cache);
return 1;
}
blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
force_share);
if (!blk) {
return 1;
}
bs = blk_bs(blk);
check = g_new0(ImageCheck, 1);
ret = collect_image_check(bs, check, filename, fmt, fix);
if (ret == -ENOTSUP) {
error_report("This image format does not support checks");
ret = 63;
goto fail;
}
if (check->corruptions_fixed || check->leaks_fixed) {
int corruptions_fixed, leaks_fixed;
bool has_leaks_fixed, has_corruptions_fixed;
leaks_fixed = check->leaks_fixed;
has_leaks_fixed = check->has_leaks_fixed;
corruptions_fixed = check->corruptions_fixed;
has_corruptions_fixed = check->has_corruptions_fixed;
if (output_format == OFORMAT_HUMAN) {
qprintf(quiet,
"The following inconsistencies were found and repaired:\n\n"
" %" PRId64 " leaked clusters\n"
" %" PRId64 " corruptions\n\n"
"Double checking the fixed image now...\n",
check->leaks_fixed,
check->corruptions_fixed);
}
qapi_free_ImageCheck(check);
check = g_new0(ImageCheck, 1);
ret = collect_image_check(bs, check, filename, fmt, 0);
check->leaks_fixed = leaks_fixed;
check->has_leaks_fixed = has_leaks_fixed;
check->corruptions_fixed = corruptions_fixed;
check->has_corruptions_fixed = has_corruptions_fixed;
}
if (!ret) {
switch (output_format) {
case OFORMAT_HUMAN:
dump_human_image_check(check, quiet);
break;
case OFORMAT_JSON:
dump_json_image_check(check, quiet);
break;
}
}
if (ret || check->check_errors) {
if (ret) {
error_report("Check failed: %s", strerror(-ret));
} else {
error_report("Check failed");
}
ret = 1;
goto fail;
}
if (check->corruptions) {
ret = 2;
} else if (check->leaks) {
ret = 3;
} else {
ret = 0;
}
fail:
qapi_free_ImageCheck(check);
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
blk_unref(blk);
return ret;
}
typedef struct CommonBlockJobCBInfo {
BlockDriverState *bs;
Error **errp;
} CommonBlockJobCBInfo;
static void common_block_job_cb(void *opaque, int ret)
{
CommonBlockJobCBInfo *cbi = opaque;
if (ret < 0) {
error_setg_errno(cbi->errp, -ret, "Block job failed");
}
}
static void run_block_job(BlockJob *job, Error **errp)
{
uint64_t progress_current, progress_total;
AioContext *aio_context = block_job_get_aio_context(job);
int ret = 0;
job_lock();
job_ref_locked(&job->job);
do {
float progress = 0.0f;
job_unlock();
aio_poll(aio_context, true);
progress_get_snapshot(&job->job.progress, &progress_current,
&progress_total);
if (progress_total) {
progress = (float)progress_current / progress_total * 100.f;
}
qemu_progress_print(progress, 0);
job_lock();
} while (!job_is_ready_locked(&job->job) &&
!job_is_completed_locked(&job->job));
if (!job_is_completed_locked(&job->job)) {
ret = job_complete_sync_locked(&job->job, errp);
} else {
ret = job->job.ret;
}
job_unref_locked(&job->job);
job_unlock();
/* publish completion progress only when success */
if (!ret) {
qemu_progress_print(100.f, 0);
}
}
static int img_commit(int argc, char **argv)
{
int c, ret, flags;
const char *filename, *fmt, *cache, *base;
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
BlockBackend *blk;
BlockDriverState *bs, *base_bs;
BlockJob *job;
bool progress = false, quiet = false, drop = false;
bool writethrough;
Error *local_err = NULL;
CommonBlockJobCBInfo cbi;
bool image_opts = false;
AioContext *aio_context;
int64_t rate_limit = 0;
fmt = NULL;
cache = BDRV_DEFAULT_CACHE;
base = NULL;
for(;;) {
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"object", required_argument, 0, OPTION_OBJECT},
{"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, ":f:ht:b:dpqr:",
long_options, NULL);
if (c == -1) {
break;
}
switch(c) {
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
break;
case 'f':
fmt = optarg;
break;
case 't':
cache = optarg;
break;
case 'b':
base = optarg;
/* -b implies -d */
drop = true;
break;
case 'd':
drop = true;
break;
case 'p':
progress = true;
break;
case 'q':
quiet = true;
break;
case 'r':
rate_limit = cvtnum("rate limit", optarg);
if (rate_limit < 0) {
return 1;
}
break;
case OPTION_OBJECT:
user_creatable_process_cmdline(optarg);
break;
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
}
}
/* Progress is not shown in Quiet mode */
if (quiet) {
progress = false;
}
if (optind != argc - 1) {
error_exit("Expecting one image file name");
}
filename = argv[optind++];
flags = BDRV_O_RDWR | BDRV_O_UNMAP;
ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
if (ret < 0) {
error_report("Invalid cache option: %s", cache);
return 1;
}
blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
false);
if (!blk) {
return 1;
}
bs = blk_bs(blk);
qemu_progress_init(progress, 1.f);
qemu_progress_print(0.f, 100);
if (base) {
base_bs = bdrv_find_backing_image(bs, base);
if (!base_bs) {
error_setg(&local_err,
"Did not find '%s' in the backing chain of '%s'",
base, filename);
goto done;
}
} else {
/* This is different from QMP, which by default uses the deepest file in
* the backing chain (i.e., the very base); however, the traditional
* behavior of qemu-img commit is using the immediate backing file. */
base_bs = bdrv_backing_chain_next(bs);
if (!base_bs) {
error_setg(&local_err, "Image does not have a backing file");
goto done;
}
}
cbi = (CommonBlockJobCBInfo){
.errp = &local_err,
.bs = bs,
};
aio_context = bdrv_get_aio_context(bs);
aio_context_acquire(aio_context);
commit_active_start("commit", bs, base_bs, JOB_DEFAULT, rate_limit,
BLOCKDEV_ON_ERROR_REPORT, NULL, common_block_job_cb,
&cbi, false, &local_err);
aio_context_release(aio_context);
if (local_err) {
goto done;
}
/* When the block job completes, the BlockBackend reference will point to
* the old backing file. In order to avoid that the top image is already
* deleted, so we can still empty it afterwards, increment the reference
* counter here preemptively. */
if (!drop) {
bdrv_ref(bs);
}
job = block_job_get("commit");
assert(job);
run_block_job(job, &local_err);
if (local_err) {
goto unref_backing;
}
if (!drop) {
BlockBackend *old_backing_blk;
old_backing_blk = blk_new_with_bs(bs, BLK_PERM_WRITE, BLK_PERM_ALL,
&local_err);
if (!old_backing_blk) {
goto unref_backing;
}
ret = blk_make_empty(old_backing_blk, &local_err);
blk_unref(old_backing_blk);
if (ret == -ENOTSUP) {
error_free(local_err);
local_err = NULL;
} else if (ret < 0) {
goto unref_backing;
}
}
unref_backing:
if (!drop) {
bdrv_unref(bs);
}
done:
qemu_progress_end();
/*
* Manually inactivate the image first because this way we can know whether
* an error occurred. blk_unref() doesn't tell us about failures.
*/
ret = bdrv_inactivate_all();
if (ret < 0 && !local_err) {
error_setg_errno(&local_err, -ret, "Error while closing the image");
}
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
blk_unref(blk);
if (local_err) {
error_report_err(local_err);
return 1;
}
qprintf(quiet, "Image committed.\n");
return 0;
}
/*
* Returns -1 if 'buf' contains only zeroes, otherwise the byte index
* of the first sector boundary within buf where the sector contains a
* non-zero byte. This function is robust to a buffer that is not
* sector-aligned.
*/
static int64_t find_nonzero(const uint8_t *buf, int64_t n)
{
int64_t i;
int64_t end = QEMU_ALIGN_DOWN(n, BDRV_SECTOR_SIZE);
for (i = 0; i < end; i += BDRV_SECTOR_SIZE) {
if (!buffer_is_zero(buf + i, BDRV_SECTOR_SIZE)) {
return i;
}
}
if (i < n && !buffer_is_zero(buf + i, n - end)) {
return i;
}
return -1;
}
/*
* Returns true iff the first sector pointed to by 'buf' contains at least
* a non-NUL byte.
*
* 'pnum' is set to the number of sectors (including and immediately following
* the first one) that are known to be in the same allocated/unallocated state.
* The function will try to align the end offset to alignment boundaries so
* that the request will at least end aligned and consecutive requests will
* also start at an aligned offset.
*/
static int is_allocated_sectors(const uint8_t *buf, int n, int *pnum,
int64_t sector_num, int alignment)
{
bool is_zero;
int i, tail;
if (n <= 0) {
*pnum = 0;
return 0;
}
is_zero = buffer_is_zero(buf, BDRV_SECTOR_SIZE);
for(i = 1; i < n; i++) {
buf += BDRV_SECTOR_SIZE;
if (is_zero != buffer_is_zero(buf, BDRV_SECTOR_SIZE)) {
break;
}
}
if (i == n) {
/*
* The whole buf is the same.
* No reason to split it into chunks, so return now.
*/
*pnum = i;
return !is_zero;
}
tail = (sector_num + i) & (alignment - 1);
if (tail) {
if (is_zero && i <= tail) {
/*
* For sure next sector after i is data, and it will rewrite this
* tail anyway due to RMW. So, let's just write data now.
*/
is_zero = false;
}
if (!is_zero) {
/* If possible, align up end offset of allocated areas. */
i += alignment - tail;
i = MIN(i, n);
} else {
/*
* For sure next sector after i is data, and it will rewrite this
* tail anyway due to RMW. Better is avoid RMW and write zeroes up
* to aligned bound.
*/
i -= tail;
}
}
*pnum = i;
return !is_zero;
}
/*
* Like is_allocated_sectors, but if the buffer starts with a used sector,
* up to 'min' consecutive sectors containing zeros are ignored. This avoids
* breaking up write requests for only small sparse areas.
*/
static int is_allocated_sectors_min(const uint8_t *buf, int n, int *pnum,
int min, int64_t sector_num, int alignment)
{
int ret;
int num_checked, num_used;
if (n < min) {
min = n;
}
ret = is_allocated_sectors(buf, n, pnum, sector_num, alignment);
if (!ret) {
return ret;
}
num_used = *pnum;
buf += BDRV_SECTOR_SIZE * *pnum;
n -= *pnum;
sector_num += *pnum;
num_checked = num_used;
while (n > 0) {
ret = is_allocated_sectors(buf, n, pnum, sector_num, alignment);
buf += BDRV_SECTOR_SIZE * *pnum;
n -= *pnum;
sector_num += *pnum;
num_checked += *pnum;
if (ret) {
num_used = num_checked;
} else if (*pnum >= min) {
break;
}
}
*pnum = num_used;
return 1;
}
/*
* Compares two buffers chunk by chunk, where @chsize is the chunk size.
* If @chsize is 0, default chunk size of BDRV_SECTOR_SIZE is used.
* Returns 0 if the first chunk of each buffer matches, non-zero otherwise.
*
* @pnum is set to the size of the buffer prefix aligned to @chsize that
* has the same matching status as the first chunk.
*/
static int compare_buffers(const uint8_t *buf1, const uint8_t *buf2,
int64_t bytes, uint64_t chsize, int64_t *pnum)
{
bool res;
int64_t i;
assert(bytes > 0);
if (!chsize) {
chsize = BDRV_SECTOR_SIZE;
}
i = MIN(bytes, chsize);
res = !!memcmp(buf1, buf2, i);
while (i < bytes) {
int64_t len = MIN(bytes - i, chsize);
if (!!memcmp(buf1 + i, buf2 + i, len) != res) {
break;
}
i += len;
}
*pnum = i;
return res;
}
#define IO_BUF_SIZE (2 * MiB)
/*
* Check if passed sectors are empty (not allocated or contain only 0 bytes)
*
* Intended for use by 'qemu-img compare': Returns 0 in case sectors are
* filled with 0, 1 if sectors contain non-zero data (this is a comparison
* failure), and 4 on error (the exit status for read errors), after emitting
* an error message.
*
* @param blk: BlockBackend for the image
* @param offset: Starting offset to check
* @param bytes: Number of bytes to check
* @param filename: Name of disk file we are checking (logging purpose)
* @param buffer: Allocated buffer for storing read data
* @param quiet: Flag for quiet mode
*/
static int check_empty_sectors(BlockBackend *blk, int64_t offset,
int64_t bytes, const char *filename,
uint8_t *buffer, bool quiet)
{
int ret = 0;
int64_t idx;
ret = blk_pread(blk, offset, bytes, buffer, 0);
if (ret < 0) {
error_report("Error while reading offset %" PRId64 " of %s: %s",
offset, filename, strerror(-ret));
return 4;
}
idx = find_nonzero(buffer, bytes);
if (idx >= 0) {
qprintf(quiet, "Content mismatch at offset %" PRId64 "!\n",
offset + idx);
return 1;
}
return 0;
}
/*
* Compares two images. Exit codes:
*
* 0 - Images are identical or the requested help was printed
* 1 - Images differ
* >1 - Error occurred
*/
static int img_compare(int argc, char **argv)
{
const char *fmt1 = NULL, *fmt2 = NULL, *cache, *filename1, *filename2;
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
BlockBackend *blk1, *blk2;
BlockDriverState *bs1, *bs2;
int64_t total_size1, total_size2;
uint8_t *buf1 = NULL, *buf2 = NULL;
block: Convert bdrv_get_block_status_above() to bytes We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access. Changing the name of the function from bdrv_get_block_status_above() to bdrv_block_status_above() ensures that the compiler enforces that all callers are updated. Likewise, since it a byte interface allows an offset mapping that might not be sector aligned, split the mapping out of the return value and into a pass-by-reference parameter. For now, the io.c layer still assert()s that all uses are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers. For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), plus updates for the new split return interface. But some code, particularly bdrv_block_status(), gets a lot simpler because it no longer has to mess with sectors. Likewise, mirror code no longer computes s->granularity >> BDRV_SECTOR_BITS, and can therefore drop an assertion about alignment because the loop no longer depends on alignment (never mind that we don't really have a driver that reports sub-sector alignments, so it's not really possible to test the effect of sub-sector mirroring). Fix a neighboring assertion to use is_power_of_2 while there. For ease of review, bdrv_get_block_status() was tackled separately. Signed-off-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 06:47:08 +03:00
int64_t pnum1, pnum2;
int allocated1, allocated2;
int ret = 0; /* return value - 0 Ident, 1 Different, >1 Error */
bool progress = false, quiet = false, strict = false;
int flags;
bool writethrough;
int64_t total_size;
int64_t offset = 0;
int64_t chunk;
int c;
uint64_t progress_base;
bool image_opts = false;
bool force_share = false;
cache = BDRV_DEFAULT_CACHE;
for (;;) {
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"object", required_argument, 0, OPTION_OBJECT},
{"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{"force-share", no_argument, 0, 'U'},
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, ":hf:F:T:pqsU",
long_options, NULL);
if (c == -1) {
break;
}
switch (c) {
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
break;
case 'f':
fmt1 = optarg;
break;
case 'F':
fmt2 = optarg;
break;
case 'T':
cache = optarg;
break;
case 'p':
progress = true;
break;
case 'q':
quiet = true;
break;
case 's':
strict = true;
break;
case 'U':
force_share = true;
break;
case OPTION_OBJECT:
{
Error *local_err = NULL;
if (!user_creatable_add_from_str(optarg, &local_err)) {
if (local_err) {
error_report_err(local_err);
exit(2);
} else {
/* Help was printed */
exit(EXIT_SUCCESS);
}
}
break;
}
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
}
}
/* Progress is not shown in Quiet mode */
if (quiet) {
progress = false;
}
if (optind != argc - 2) {
error_exit("Expecting two image file names");
}
filename1 = argv[optind++];
filename2 = argv[optind++];
/* Initialize before goto out */
qemu_progress_init(progress, 2.0);
flags = 0;
ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
if (ret < 0) {
error_report("Invalid source cache option: %s", cache);
ret = 2;
goto out3;
}
blk1 = img_open(image_opts, filename1, fmt1, flags, writethrough, quiet,
force_share);
if (!blk1) {
ret = 2;
goto out3;
}
blk2 = img_open(image_opts, filename2, fmt2, flags, writethrough, quiet,
force_share);
if (!blk2) {
ret = 2;
goto out2;
}
bs1 = blk_bs(blk1);
bs2 = blk_bs(blk2);
buf1 = blk_blockalign(blk1, IO_BUF_SIZE);
buf2 = blk_blockalign(blk2, IO_BUF_SIZE);
total_size1 = blk_getlength(blk1);
if (total_size1 < 0) {
error_report("Can't get size of %s: %s",
filename1, strerror(-total_size1));
ret = 4;
goto out;
}
total_size2 = blk_getlength(blk2);
if (total_size2 < 0) {
error_report("Can't get size of %s: %s",
filename2, strerror(-total_size2));
ret = 4;
goto out;
}
total_size = MIN(total_size1, total_size2);
progress_base = MAX(total_size1, total_size2);
qemu_progress_print(0, 100);
if (strict && total_size1 != total_size2) {
ret = 1;
qprintf(quiet, "Strict mode: Image size mismatch!\n");
goto out;
}
while (offset < total_size) {
block: Convert bdrv_get_block_status_above() to bytes We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access. Changing the name of the function from bdrv_get_block_status_above() to bdrv_block_status_above() ensures that the compiler enforces that all callers are updated. Likewise, since it a byte interface allows an offset mapping that might not be sector aligned, split the mapping out of the return value and into a pass-by-reference parameter. For now, the io.c layer still assert()s that all uses are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers. For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), plus updates for the new split return interface. But some code, particularly bdrv_block_status(), gets a lot simpler because it no longer has to mess with sectors. Likewise, mirror code no longer computes s->granularity >> BDRV_SECTOR_BITS, and can therefore drop an assertion about alignment because the loop no longer depends on alignment (never mind that we don't really have a driver that reports sub-sector alignments, so it's not really possible to test the effect of sub-sector mirroring). Fix a neighboring assertion to use is_power_of_2 while there. For ease of review, bdrv_get_block_status() was tackled separately. Signed-off-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 06:47:08 +03:00
int status1, status2;
status1 = bdrv_block_status_above(bs1, NULL, offset,
total_size1 - offset, &pnum1, NULL,
NULL);
if (status1 < 0) {
ret = 3;
error_report("Sector allocation test failed for %s", filename1);
goto out;
}
allocated1 = status1 & BDRV_BLOCK_ALLOCATED;
status2 = bdrv_block_status_above(bs2, NULL, offset,
total_size2 - offset, &pnum2, NULL,
NULL);
if (status2 < 0) {
ret = 3;
error_report("Sector allocation test failed for %s", filename2);
goto out;
}
allocated2 = status2 & BDRV_BLOCK_ALLOCATED;
assert(pnum1 && pnum2);
chunk = MIN(pnum1, pnum2);
if (strict) {
block: Convert bdrv_get_block_status_above() to bytes We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access. Changing the name of the function from bdrv_get_block_status_above() to bdrv_block_status_above() ensures that the compiler enforces that all callers are updated. Likewise, since it a byte interface allows an offset mapping that might not be sector aligned, split the mapping out of the return value and into a pass-by-reference parameter. For now, the io.c layer still assert()s that all uses are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers. For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), plus updates for the new split return interface. But some code, particularly bdrv_block_status(), gets a lot simpler because it no longer has to mess with sectors. Likewise, mirror code no longer computes s->granularity >> BDRV_SECTOR_BITS, and can therefore drop an assertion about alignment because the loop no longer depends on alignment (never mind that we don't really have a driver that reports sub-sector alignments, so it's not really possible to test the effect of sub-sector mirroring). Fix a neighboring assertion to use is_power_of_2 while there. For ease of review, bdrv_get_block_status() was tackled separately. Signed-off-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 06:47:08 +03:00
if (status1 != status2) {
ret = 1;
qprintf(quiet, "Strict mode: Offset %" PRId64
" block status mismatch!\n", offset);
goto out;
}
}
if ((status1 & BDRV_BLOCK_ZERO) && (status2 & BDRV_BLOCK_ZERO)) {
/* nothing to do */
} else if (allocated1 == allocated2) {
if (allocated1) {
int64_t pnum;
chunk = MIN(chunk, IO_BUF_SIZE);
ret = blk_pread(blk1, offset, chunk, buf1, 0);
if (ret < 0) {
error_report("Error while reading offset %" PRId64
" of %s: %s",
offset, filename1, strerror(-ret));
ret = 4;
goto out;
}
ret = blk_pread(blk2, offset, chunk, buf2, 0);
if (ret < 0) {
error_report("Error while reading offset %" PRId64
" of %s: %s",
offset, filename2, strerror(-ret));
ret = 4;
goto out;
}
ret = compare_buffers(buf1, buf2, chunk, 0, &pnum);
if (ret || pnum != chunk) {
qprintf(quiet, "Content mismatch at offset %" PRId64 "!\n",
offset + (ret ? 0 : pnum));
ret = 1;
goto out;
}
}
} else {
chunk = MIN(chunk, IO_BUF_SIZE);
if (allocated1) {
ret = check_empty_sectors(blk1, offset, chunk,
filename1, buf1, quiet);
} else {
ret = check_empty_sectors(blk2, offset, chunk,
filename2, buf1, quiet);
}
if (ret) {
goto out;
}
}
offset += chunk;
qemu_progress_print(((float) chunk / progress_base) * 100, 100);
}
if (total_size1 != total_size2) {
BlockBackend *blk_over;
const char *filename_over;
qprintf(quiet, "Warning: Image size mismatch!\n");
if (total_size1 > total_size2) {
blk_over = blk1;
filename_over = filename1;
} else {
blk_over = blk2;
filename_over = filename2;
}
while (offset < progress_base) {
ret = bdrv_block_status_above(blk_bs(blk_over), NULL, offset,
progress_base - offset, &chunk,
NULL, NULL);
if (ret < 0) {
ret = 3;
error_report("Sector allocation test failed for %s",
filename_over);
goto out;
}
2017-10-12 06:47:10 +03:00
if (ret & BDRV_BLOCK_ALLOCATED && !(ret & BDRV_BLOCK_ZERO)) {
chunk = MIN(chunk, IO_BUF_SIZE);
ret = check_empty_sectors(blk_over, offset, chunk,
filename_over, buf1, quiet);
if (ret) {
goto out;
}
}
offset += chunk;
qemu_progress_print(((float) chunk / progress_base) * 100, 100);
}
}
qprintf(quiet, "Images are identical.\n");
ret = 0;
out:
qemu_vfree(buf1);
qemu_vfree(buf2);
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
blk_unref(blk2);
out2:
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
blk_unref(blk1);
out3:
qemu_progress_end();
return ret;
}
/* Convenience wrapper around qmp_block_dirty_bitmap_merge */
static void do_dirty_bitmap_merge(const char *dst_node, const char *dst_name,
const char *src_node, const char *src_name,
Error **errp)
{
BlockDirtyBitmapOrStr *merge_src;
BlockDirtyBitmapOrStrList *list = NULL;
merge_src = g_new0(BlockDirtyBitmapOrStr, 1);
merge_src->type = QTYPE_QDICT;
merge_src->u.external.node = g_strdup(src_node);
merge_src->u.external.name = g_strdup(src_name);
QAPI_LIST_PREPEND(list, merge_src);
qmp_block_dirty_bitmap_merge(dst_node, dst_name, list, errp);
qapi_free_BlockDirtyBitmapOrStrList(list);
}
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
enum ImgConvertBlockStatus {
BLK_DATA,
BLK_ZERO,
BLK_BACKING_FILE,
};
#define MAX_COROUTINES 16
#define CONVERT_THROTTLE_GROUP "img_convert"
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
typedef struct ImgConvertState {
BlockBackend **src;
int64_t *src_sectors;
int *src_alignment;
int src_num;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
int64_t total_sectors;
int64_t allocated_sectors;
int64_t allocated_done;
int64_t sector_num;
int64_t wr_offs;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
enum ImgConvertBlockStatus status;
int64_t sector_next_status;
BlockBackend *target;
bool has_zero_init;
bool compressed;
bool target_is_new;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
bool target_has_backing;
int64_t target_backing_sectors; /* negative if unknown */
bool wr_in_order;
bool copy_range;
bool salvage;
bool quiet;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
int min_sparse;
int alignment;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
size_t cluster_sectors;
size_t buf_sectors;
long num_coroutines;
int running_coroutines;
Coroutine *co[MAX_COROUTINES];
int64_t wait_sector_num[MAX_COROUTINES];
CoMutex lock;
int ret;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
} ImgConvertState;
static void convert_select_part(ImgConvertState *s, int64_t sector_num,
int *src_cur, int64_t *src_cur_offset)
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
{
*src_cur = 0;
*src_cur_offset = 0;
while (sector_num - *src_cur_offset >= s->src_sectors[*src_cur]) {
*src_cur_offset += s->src_sectors[*src_cur];
(*src_cur)++;
assert(*src_cur < s->src_num);
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
}
}
static int convert_iteration_sectors(ImgConvertState *s, int64_t sector_num)
{
block: Convert bdrv_get_block_status_above() to bytes We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access. Changing the name of the function from bdrv_get_block_status_above() to bdrv_block_status_above() ensures that the compiler enforces that all callers are updated. Likewise, since it a byte interface allows an offset mapping that might not be sector aligned, split the mapping out of the return value and into a pass-by-reference parameter. For now, the io.c layer still assert()s that all uses are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers. For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), plus updates for the new split return interface. But some code, particularly bdrv_block_status(), gets a lot simpler because it no longer has to mess with sectors. Likewise, mirror code no longer computes s->granularity >> BDRV_SECTOR_BITS, and can therefore drop an assertion about alignment because the loop no longer depends on alignment (never mind that we don't really have a driver that reports sub-sector alignments, so it's not really possible to test the effect of sub-sector mirroring). Fix a neighboring assertion to use is_power_of_2 while there. For ease of review, bdrv_get_block_status() was tackled separately. Signed-off-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 06:47:08 +03:00
int64_t src_cur_offset;
int ret, n, src_cur;
bool post_backing_zero = false;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
assert(s->total_sectors > sector_num);
n = MIN(s->total_sectors - sector_num, BDRV_REQUEST_MAX_SECTORS);
if (s->target_backing_sectors >= 0) {
if (sector_num >= s->target_backing_sectors) {
post_backing_zero = true;
} else if (sector_num + n > s->target_backing_sectors) {
/* Split requests around target_backing_sectors (because
* starting from there, zeros are handled differently) */
n = s->target_backing_sectors - sector_num;
}
}
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
if (s->sector_next_status <= sector_num) {
uint64_t offset = (sector_num - src_cur_offset) * BDRV_SECTOR_SIZE;
int64_t count;
int tail;
BlockDriverState *src_bs = blk_bs(s->src[src_cur]);
BlockDriverState *base;
if (s->target_has_backing) {
base = bdrv_cow_bs(bdrv_skip_filters(src_bs));
} else {
base = NULL;
}
block: Convert bdrv_get_block_status_above() to bytes We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access. Changing the name of the function from bdrv_get_block_status_above() to bdrv_block_status_above() ensures that the compiler enforces that all callers are updated. Likewise, since it a byte interface allows an offset mapping that might not be sector aligned, split the mapping out of the return value and into a pass-by-reference parameter. For now, the io.c layer still assert()s that all uses are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers. For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), plus updates for the new split return interface. But some code, particularly bdrv_block_status(), gets a lot simpler because it no longer has to mess with sectors. Likewise, mirror code no longer computes s->granularity >> BDRV_SECTOR_BITS, and can therefore drop an assertion about alignment because the loop no longer depends on alignment (never mind that we don't really have a driver that reports sub-sector alignments, so it's not really possible to test the effect of sub-sector mirroring). Fix a neighboring assertion to use is_power_of_2 while there. For ease of review, bdrv_get_block_status() was tackled separately. Signed-off-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 06:47:08 +03:00
do {
count = n * BDRV_SECTOR_SIZE;
ret = bdrv_block_status_above(src_bs, base, offset, count, &count,
NULL, NULL);
if (ret < 0) {
if (s->salvage) {
if (n == 1) {
if (!s->quiet) {
warn_report("error while reading block status at "
"offset %" PRIu64 ": %s", offset,
strerror(-ret));
}
/* Just try to read the data, then */
ret = BDRV_BLOCK_DATA;
count = BDRV_SECTOR_SIZE;
} else {
/* Retry on a shorter range */
n = DIV_ROUND_UP(n, 4);
}
} else {
error_report("error while reading block status at offset "
"%" PRIu64 ": %s", offset, strerror(-ret));
return ret;
}
}
} while (ret < 0);
block: Convert bdrv_get_block_status() to bytes We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access. Changing the name of the function from bdrv_get_block_status() to bdrv_block_status() ensures that the compiler enforces that all callers are updated. For now, the io.c layer still assert()s that all callers are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers. There was an inherent limitation in returning the offset via the return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which means an offset can only be mapped for sector-aligned queries (or, if we declare that non-aligned input is at the same relative position modulo 512 of the answer), so the new interface also changes things to return the offset via output through a parameter by reference rather than mashed into the return value. We'll have some glue code that munges between the two styles until we finish converting all uses. For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), coupled with the tweak in calling convention. But some code, particularly bdrv_is_allocated(), gets a lot simpler because it no longer has to mess with sectors. For ease of review, bdrv_get_block_status_above() will be tackled separately. Signed-off-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 06:47:03 +03:00
block: Convert bdrv_get_block_status_above() to bytes We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access. Changing the name of the function from bdrv_get_block_status_above() to bdrv_block_status_above() ensures that the compiler enforces that all callers are updated. Likewise, since it a byte interface allows an offset mapping that might not be sector aligned, split the mapping out of the return value and into a pass-by-reference parameter. For now, the io.c layer still assert()s that all uses are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers. For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), plus updates for the new split return interface. But some code, particularly bdrv_block_status(), gets a lot simpler because it no longer has to mess with sectors. Likewise, mirror code no longer computes s->granularity >> BDRV_SECTOR_BITS, and can therefore drop an assertion about alignment because the loop no longer depends on alignment (never mind that we don't really have a driver that reports sub-sector alignments, so it's not really possible to test the effect of sub-sector mirroring). Fix a neighboring assertion to use is_power_of_2 while there. For ease of review, bdrv_get_block_status() was tackled separately. Signed-off-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 06:47:08 +03:00
n = DIV_ROUND_UP(count, BDRV_SECTOR_SIZE);
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
/*
* Avoid that s->sector_next_status becomes unaligned to the source
* request alignment and/or cluster size to avoid unnecessary read
* cycles.
*/
tail = (sector_num - src_cur_offset + n) % s->src_alignment[src_cur];
if (n > tail) {
n -= tail;
}
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
if (ret & BDRV_BLOCK_ZERO) {
s->status = post_backing_zero ? BLK_BACKING_FILE : BLK_ZERO;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
} else if (ret & BDRV_BLOCK_DATA) {
s->status = BLK_DATA;
} else {
s->status = s->target_has_backing ? BLK_BACKING_FILE : BLK_DATA;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
}
s->sector_next_status = sector_num + n;
}
n = MIN(n, s->sector_next_status - sector_num);
if (s->status == BLK_DATA) {
n = MIN(n, s->buf_sectors);
}
/* We need to write complete clusters for compressed images, so if an
* unallocated area is shorter than that, we must consider the whole
* cluster allocated. */
if (s->compressed) {
if (n < s->cluster_sectors) {
n = MIN(s->cluster_sectors, s->total_sectors - sector_num);
s->status = BLK_DATA;
} else {
n = QEMU_ALIGN_DOWN(n, s->cluster_sectors);
}
}
return n;
}
static int coroutine_fn convert_co_read(ImgConvertState *s, int64_t sector_num,
int nb_sectors, uint8_t *buf)
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
{
uint64_t single_read_until = 0;
int n, ret;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
assert(nb_sectors <= s->buf_sectors);
while (nb_sectors > 0) {
BlockBackend *blk;
int src_cur;
int64_t bs_sectors, src_cur_offset;
uint64_t offset;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
/* In the case of compression with multiple source files, we can get a
* nb_sectors that spreads into the next part. So we must be able to
* read across multiple BDSes for one convert_read() call. */
convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
blk = s->src[src_cur];
bs_sectors = s->src_sectors[src_cur];
offset = (sector_num - src_cur_offset) << BDRV_SECTOR_BITS;
n = MIN(nb_sectors, bs_sectors - (sector_num - src_cur_offset));
if (single_read_until > offset) {
n = 1;
}
ret = blk_co_pread(blk, offset, n << BDRV_SECTOR_BITS, buf, 0);
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
if (ret < 0) {
if (s->salvage) {
if (n > 1) {
single_read_until = offset + (n << BDRV_SECTOR_BITS);
continue;
} else {
if (!s->quiet) {
warn_report("error while reading offset %" PRIu64
": %s", offset, strerror(-ret));
}
memset(buf, 0, BDRV_SECTOR_SIZE);
}
} else {
return ret;
}
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
}
sector_num += n;
nb_sectors -= n;
buf += n * BDRV_SECTOR_SIZE;
}
return 0;
}
static int coroutine_fn convert_co_write(ImgConvertState *s, int64_t sector_num,
int nb_sectors, uint8_t *buf,
enum ImgConvertBlockStatus status)
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
{
int ret;
while (nb_sectors > 0) {
int n = nb_sectors;
BdrvRequestFlags flags = s->compressed ? BDRV_REQ_WRITE_COMPRESSED : 0;
switch (status) {
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
case BLK_BACKING_FILE:
/* If we have a backing file, leave clusters unallocated that are
* unallocated in the source image, so that the backing file is
* visible at the respective offset. */
assert(s->target_has_backing);
break;
case BLK_DATA:
/* If we're told to keep the target fully allocated (-S 0) or there
* is real non-zero data, we must write it. Otherwise we can treat
* it as zero sectors.
* Compressed clusters need to be written as a whole, so in that
* case we can only save the write if the buffer is completely
* zeroed. */
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
if (!s->min_sparse ||
(!s->compressed &&
is_allocated_sectors_min(buf, n, &n, s->min_sparse,
sector_num, s->alignment)) ||
(s->compressed &&
!buffer_is_zero(buf, n * BDRV_SECTOR_SIZE)))
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
{
ret = blk_co_pwrite(s->target, sector_num << BDRV_SECTOR_BITS,
n << BDRV_SECTOR_BITS, buf, flags);
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
if (ret < 0) {
return ret;
}
break;
}
/* fall-through */
case BLK_ZERO:
if (s->has_zero_init) {
assert(!s->target_has_backing);
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
break;
}
ret = blk_co_pwrite_zeroes(s->target,
sector_num << BDRV_SECTOR_BITS,
qemu-img: Enable BDRV_REQ_MAY_UNMAP in convert With Kevin's "block: Fix slow pre-zeroing in qemu-img convert"[1] (commit c9fdcf202f, 'qemu-img: Use BDRV_REQ_NO_FALLBACK for pre-zeroing') we skip the pre zero step called like this: blk_make_zero(s->target, BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK) And we write zeroes later using: blk_co_pwrite_zeroes(s->target, sector_num << BDRV_SECTOR_BITS, n << BDRV_SECTOR_BITS, 0); Since we use flags=0, this is translated to NBD_CMD_WRITE_ZEROES with NBD_CMD_FLAG_NO_HOLE flag, which cause the NBD server to allocated space instead of punching a hole. Here is an example failure: $ dd if=/dev/urandom of=src.img bs=1M count=5 $ truncate -s 50m src.img $ truncate -s 50m dst.img $ nbdkit -f -v -e '' -U nbd.sock file file=dst.img $ ./qemu-img convert -n src.img nbd:unix:nbd.sock We can see in nbdkit log that it received the NBD_CMD_FLAG_NO_HOLE (may_trim=0): nbdkit: file[1]: debug: newstyle negotiation: flags: export 0x4d nbdkit: file[1]: debug: pwrite count=2097152 offset=0 nbdkit: file[1]: debug: pwrite count=2097152 offset=2097152 nbdkit: file[1]: debug: pwrite count=1048576 offset=4194304 nbdkit: file[1]: debug: zero count=33554432 offset=5242880 may_trim=0 nbdkit: file[1]: debug: zero count=13631488 offset=38797312 may_trim=0 nbdkit: file[1]: debug: flush And the image became fully allocated: $ qemu-img info dst.img virtual size: 50M (52428800 bytes) disk size: 50M With this change we see that nbdkit did not receive the NBD_CMD_FLAG_NO_HOLE (may_trim=1): nbdkit: file[1]: debug: newstyle negotiation: flags: export 0x4d nbdkit: file[1]: debug: pwrite count=2097152 offset=0 nbdkit: file[1]: debug: pwrite count=2097152 offset=2097152 nbdkit: file[1]: debug: pwrite count=1048576 offset=4194304 nbdkit: file[1]: debug: zero count=33554432 offset=5242880 may_trim=1 nbdkit: file[1]: debug: zero count=13631488 offset=38797312 may_trim=1 nbdkit: file[1]: debug: flush And the file is sparse as expected: $ qemu-img info dst.img virtual size: 50M (52428800 bytes) disk size: 5.0M [1] http://lists.nongnu.org/archive/html/qemu-block/2019-03/msg00761.html Signed-off-by: Nir Soffer <nsoffer@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2019-03-24 03:20:12 +03:00
n << BDRV_SECTOR_BITS,
BDRV_REQ_MAY_UNMAP);
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
if (ret < 0) {
return ret;
}
break;
}
sector_num += n;
nb_sectors -= n;
buf += n * BDRV_SECTOR_SIZE;
}
return 0;
}
static int coroutine_fn convert_co_copy_range(ImgConvertState *s, int64_t sector_num,
int nb_sectors)
{
int n, ret;
while (nb_sectors > 0) {
BlockBackend *blk;
int src_cur;
int64_t bs_sectors, src_cur_offset;
int64_t offset;
convert_select_part(s, sector_num, &src_cur, &src_cur_offset);
offset = (sector_num - src_cur_offset) << BDRV_SECTOR_BITS;
blk = s->src[src_cur];
bs_sectors = s->src_sectors[src_cur];
n = MIN(nb_sectors, bs_sectors - (sector_num - src_cur_offset));
ret = blk_co_copy_range(blk, offset, s->target,
sector_num << BDRV_SECTOR_BITS,
n << BDRV_SECTOR_BITS, 0, 0);
if (ret < 0) {
return ret;
}
sector_num += n;
nb_sectors -= n;
}
return 0;
}
static void coroutine_fn convert_co_do_copy(void *opaque)
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
{
ImgConvertState *s = opaque;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
uint8_t *buf = NULL;
int ret, i;
int index = -1;
for (i = 0; i < s->num_coroutines; i++) {
if (s->co[i] == qemu_coroutine_self()) {
index = i;
break;
}
}
assert(index >= 0);
s->running_coroutines++;
buf = blk_blockalign(s->target, s->buf_sectors * BDRV_SECTOR_SIZE);
while (1) {
int n;
int64_t sector_num;
enum ImgConvertBlockStatus status;
bool copy_range;
qemu_co_mutex_lock(&s->lock);
if (s->ret != -EINPROGRESS || s->sector_num >= s->total_sectors) {
qemu_co_mutex_unlock(&s->lock);
break;
}
WITH_GRAPH_RDLOCK_GUARD() {
n = convert_iteration_sectors(s, s->sector_num);
}
if (n < 0) {
qemu_co_mutex_unlock(&s->lock);
s->ret = n;
break;
}
/* save current sector and allocation status to local variables */
sector_num = s->sector_num;
status = s->status;
if (!s->min_sparse && s->status == BLK_ZERO) {
n = MIN(n, s->buf_sectors);
}
/* increment global sector counter so that other coroutines can
* already continue reading beyond this request */
s->sector_num += n;
qemu_co_mutex_unlock(&s->lock);
if (status == BLK_DATA || (!s->min_sparse && status == BLK_ZERO)) {
s->allocated_done += n;
qemu_progress_print(100.0 * s->allocated_done /
s->allocated_sectors, 0);
}
retry:
copy_range = s->copy_range && s->status == BLK_DATA;
if (status == BLK_DATA && !copy_range) {
ret = convert_co_read(s, sector_num, n, buf);
if (ret < 0) {
error_report("error while reading at byte %lld: %s",
sector_num * BDRV_SECTOR_SIZE, strerror(-ret));
s->ret = ret;
}
} else if (!s->min_sparse && status == BLK_ZERO) {
status = BLK_DATA;
memset(buf, 0x00, n * BDRV_SECTOR_SIZE);
}
if (s->wr_in_order) {
/* keep writes in order */
while (s->wr_offs != sector_num && s->ret == -EINPROGRESS) {
s->wait_sector_num[index] = sector_num;
qemu_coroutine_yield();
}
s->wait_sector_num[index] = -1;
}
if (s->ret == -EINPROGRESS) {
if (copy_range) {
WITH_GRAPH_RDLOCK_GUARD() {
ret = convert_co_copy_range(s, sector_num, n);
}
if (ret) {
s->copy_range = false;
goto retry;
}
} else {
ret = convert_co_write(s, sector_num, n, buf, status);
}
if (ret < 0) {
error_report("error while writing at byte %lld: %s",
sector_num * BDRV_SECTOR_SIZE, strerror(-ret));
s->ret = ret;
}
}
if (s->wr_in_order) {
/* reenter the coroutine that might have waited
* for this write to complete */
s->wr_offs = sector_num + n;
for (i = 0; i < s->num_coroutines; i++) {
if (s->co[i] && s->wait_sector_num[i] == s->wr_offs) {
/*
* A -> B -> A cannot occur because A has
* s->wait_sector_num[i] == -1 during A -> B. Therefore
* B will never enter A during this time window.
*/
qemu_coroutine_enter(s->co[i]);
break;
}
}
}
}
qemu_vfree(buf);
s->co[index] = NULL;
s->running_coroutines--;
if (!s->running_coroutines && s->ret == -EINPROGRESS) {
/* the convert job finished successfully */
s->ret = 0;
}
}
static int convert_do_copy(ImgConvertState *s)
{
int ret, i, n;
int64_t sector_num = 0;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
/* Check whether we have zero initialisation or can get it efficiently */
if (!s->has_zero_init && s->target_is_new && s->min_sparse &&
!s->target_has_backing) {
s->has_zero_init = bdrv_has_zero_init(blk_bs(s->target));
}
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
/* Allocate buffer for copied data. For compressed images, only one cluster
* can be copied at a time. */
if (s->compressed) {
if (s->cluster_sectors <= 0 || s->cluster_sectors > s->buf_sectors) {
error_report("invalid cluster size");
return -EINVAL;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
}
s->buf_sectors = s->cluster_sectors;
}
while (sector_num < s->total_sectors) {
n = convert_iteration_sectors(s, sector_num);
if (n < 0) {
return n;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
}
if (s->status == BLK_DATA || (!s->min_sparse && s->status == BLK_ZERO))
{
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
s->allocated_sectors += n;
}
sector_num += n;
}
/* Do the copy */
s->sector_next_status = 0;
s->ret = -EINPROGRESS;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
qemu_co_mutex_init(&s->lock);
for (i = 0; i < s->num_coroutines; i++) {
s->co[i] = qemu_coroutine_create(convert_co_do_copy, s);
s->wait_sector_num[i] = -1;
qemu_coroutine_enter(s->co[i]);
}
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
while (s->running_coroutines) {
main_loop_wait(false);
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
}
if (s->compressed && !s->ret) {
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
/* signal EOF to align */
ret = blk_pwrite_compressed(s->target, 0, 0, NULL);
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
if (ret < 0) {
return ret;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
}
}
return s->ret;
qemu-img convert: Rewrite copying logic The implementation of qemu-img convert is (a) messy, (b) buggy, and (c) less efficient than possible. The changes required to beat some sense into it are massive enough that incremental changes would only make my and the reviewers' life harder. So throw it away and reimplement it from scratch. Let me give some examples what I mean by messy, buggy and inefficient: (a) The copying logic of qemu-img convert has two separate branches for compressed and normal target images, which roughly do the same - except for a little code that handles actual differences between compressed and uncompressed images, and much more code that implements just a different set of optimisations and bugs. This is unnecessary code duplication, and makes the code for compressed output (unsurprisingly) suffer from bitrot. The code for uncompressed ouput is run twice to count the the total length for the progress bar. In the first run it just takes a shortcut and runs only half the loop, and when it's done, it toggles a boolean, jumps out of the loop with a backwards goto and starts over. Works, but pretty is something different. (b) Converting while keeping a backing file (-B option) is broken in several ways. This includes not writing to the image file if the input has zero clusters or data filled with zeros (ignoring that the backing file will be visible instead). It also doesn't correctly limit every iteration of the copy loop to sectors of the same status so that too many sectors may be copied to in the target image. For -B this gives an unexpected result, for other images it just does more work than necessary. Conversion with a compressed target completely ignores any target backing file. (c) qemu-img convert skips reading and writing an area if it knows from metadata that copying isn't needed (except for the bug mentioned above that ignores a status change in some cases). It does, however, read from the source even if it knows that it will read zeros, and then search for non-zero bytes in the read buffer, if it's possible that a write might be needed. This reimplementation of the copying core reorganises the code to remove the duplication and have a much more obvious code flow, by essentially splitting the copy iteration loop into three parts: 1. Find the number of contiguous sectors of the same status at the current offset (This can also be called in a separate loop before the copying loop in order to determine the total sectors for the progress bar.) 2. Read sectors. If the status implies that there is no data there to read (zero or unallocated cluster), don't do anything. 3. Write sectors depending on the status. If it's data, write it. If we want the backing file to be visible (with -B), don't write it. If it's zeroed, skip it if you can, otherwise use bdrv_write_zeroes() to optimise the write at least where possible. Signed-off-by: Kevin Wolf <kwolf@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com>
2015-03-19 15:33:32 +03:00
}
/* Check that bitmaps can be copied, or output an error */
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps' The point of 'qemu-img convert --bitmaps' is to be a convenience for actions that are already possible through a string of smaller 'qemu-img bitmap' sub-commands. One situation not accounted for already is that if a source image contains an inconsistent bitmap (for example, because a qemu process died abruptly before flushing bitmap state), the user MUST delete those inconsistent bitmaps before anything else useful can be done with the image. We don't want to delete inconsistent bitmaps by default: although a corrupt bitmap is only a loss of optimization rather than a corruption of user-visible data, it is still nice to require the user to opt in to the fact that they are aware of the loss of the bitmap. Still, requiring the user to check 'qemu-img info' to see whether bitmaps are consistent, then use 'qemu-img bitmap --remove' to remove offenders, all before using 'qemu-img convert', is a lot more work than just adding a knob 'qemu-img convert --bitmaps --skip-broken-bitmaps' which opts in to skipping the broken bitmaps. After testing the new option, also demonstrate the way to manually fix things (either deleting bad bitmaps, or re-creating them as empty) so that it is possible to convert without the option. Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1946084 Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210709153951.2801666-4-eblake@redhat.com> [eblake: warning message tweak, test enhancements] Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2021-07-21 18:53:48 +03:00
static int convert_check_bitmaps(BlockDriverState *src, bool skip_broken)
{
BdrvDirtyBitmap *bm;
if (!bdrv_supports_persistent_dirty_bitmap(src)) {
error_report("Source lacks bitmap support");
return -1;
}
FOR_EACH_DIRTY_BITMAP(src, bm) {
if (!bdrv_dirty_bitmap_get_persistence(bm)) {
continue;
}
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps' The point of 'qemu-img convert --bitmaps' is to be a convenience for actions that are already possible through a string of smaller 'qemu-img bitmap' sub-commands. One situation not accounted for already is that if a source image contains an inconsistent bitmap (for example, because a qemu process died abruptly before flushing bitmap state), the user MUST delete those inconsistent bitmaps before anything else useful can be done with the image. We don't want to delete inconsistent bitmaps by default: although a corrupt bitmap is only a loss of optimization rather than a corruption of user-visible data, it is still nice to require the user to opt in to the fact that they are aware of the loss of the bitmap. Still, requiring the user to check 'qemu-img info' to see whether bitmaps are consistent, then use 'qemu-img bitmap --remove' to remove offenders, all before using 'qemu-img convert', is a lot more work than just adding a knob 'qemu-img convert --bitmaps --skip-broken-bitmaps' which opts in to skipping the broken bitmaps. After testing the new option, also demonstrate the way to manually fix things (either deleting bad bitmaps, or re-creating them as empty) so that it is possible to convert without the option. Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1946084 Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210709153951.2801666-4-eblake@redhat.com> [eblake: warning message tweak, test enhancements] Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2021-07-21 18:53:48 +03:00
if (!skip_broken && bdrv_dirty_bitmap_inconsistent(bm)) {
error_report("Cannot copy inconsistent bitmap '%s'",
bdrv_dirty_bitmap_name(bm));
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps' The point of 'qemu-img convert --bitmaps' is to be a convenience for actions that are already possible through a string of smaller 'qemu-img bitmap' sub-commands. One situation not accounted for already is that if a source image contains an inconsistent bitmap (for example, because a qemu process died abruptly before flushing bitmap state), the user MUST delete those inconsistent bitmaps before anything else useful can be done with the image. We don't want to delete inconsistent bitmaps by default: although a corrupt bitmap is only a loss of optimization rather than a corruption of user-visible data, it is still nice to require the user to opt in to the fact that they are aware of the loss of the bitmap. Still, requiring the user to check 'qemu-img info' to see whether bitmaps are consistent, then use 'qemu-img bitmap --remove' to remove offenders, all before using 'qemu-img convert', is a lot more work than just adding a knob 'qemu-img convert --bitmaps --skip-broken-bitmaps' which opts in to skipping the broken bitmaps. After testing the new option, also demonstrate the way to manually fix things (either deleting bad bitmaps, or re-creating them as empty) so that it is possible to convert without the option. Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1946084 Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210709153951.2801666-4-eblake@redhat.com> [eblake: warning message tweak, test enhancements] Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2021-07-21 18:53:48 +03:00
error_printf("Try --skip-broken-bitmaps, or "
"use 'qemu-img bitmap --remove' to delete it\n");
return -1;
}
}
return 0;
}
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps' The point of 'qemu-img convert --bitmaps' is to be a convenience for actions that are already possible through a string of smaller 'qemu-img bitmap' sub-commands. One situation not accounted for already is that if a source image contains an inconsistent bitmap (for example, because a qemu process died abruptly before flushing bitmap state), the user MUST delete those inconsistent bitmaps before anything else useful can be done with the image. We don't want to delete inconsistent bitmaps by default: although a corrupt bitmap is only a loss of optimization rather than a corruption of user-visible data, it is still nice to require the user to opt in to the fact that they are aware of the loss of the bitmap. Still, requiring the user to check 'qemu-img info' to see whether bitmaps are consistent, then use 'qemu-img bitmap --remove' to remove offenders, all before using 'qemu-img convert', is a lot more work than just adding a knob 'qemu-img convert --bitmaps --skip-broken-bitmaps' which opts in to skipping the broken bitmaps. After testing the new option, also demonstrate the way to manually fix things (either deleting bad bitmaps, or re-creating them as empty) so that it is possible to convert without the option. Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1946084 Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210709153951.2801666-4-eblake@redhat.com> [eblake: warning message tweak, test enhancements] Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2021-07-21 18:53:48 +03:00
static int convert_copy_bitmaps(BlockDriverState *src, BlockDriverState *dst,
bool skip_broken)
{
BdrvDirtyBitmap *bm;
Error *err = NULL;
FOR_EACH_DIRTY_BITMAP(src, bm) {
const char *name;
if (!bdrv_dirty_bitmap_get_persistence(bm)) {
continue;
}
name = bdrv_dirty_bitmap_name(bm);
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps' The point of 'qemu-img convert --bitmaps' is to be a convenience for actions that are already possible through a string of smaller 'qemu-img bitmap' sub-commands. One situation not accounted for already is that if a source image contains an inconsistent bitmap (for example, because a qemu process died abruptly before flushing bitmap state), the user MUST delete those inconsistent bitmaps before anything else useful can be done with the image. We don't want to delete inconsistent bitmaps by default: although a corrupt bitmap is only a loss of optimization rather than a corruption of user-visible data, it is still nice to require the user to opt in to the fact that they are aware of the loss of the bitmap. Still, requiring the user to check 'qemu-img info' to see whether bitmaps are consistent, then use 'qemu-img bitmap --remove' to remove offenders, all before using 'qemu-img convert', is a lot more work than just adding a knob 'qemu-img convert --bitmaps --skip-broken-bitmaps' which opts in to skipping the broken bitmaps. After testing the new option, also demonstrate the way to manually fix things (either deleting bad bitmaps, or re-creating them as empty) so that it is possible to convert without the option. Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1946084 Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210709153951.2801666-4-eblake@redhat.com> [eblake: warning message tweak, test enhancements] Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2021-07-21 18:53:48 +03:00
if (skip_broken && bdrv_dirty_bitmap_inconsistent(bm)) {
warn_report("Skipping inconsistent bitmap '%s'", name);
continue;
}
qmp_block_dirty_bitmap_add(dst->node_name, name,
true, bdrv_dirty_bitmap_granularity(bm),
true, true,
true, !bdrv_dirty_bitmap_enabled(bm),
&err);
if (err) {
error_reportf_err(err, "Failed to create bitmap %s: ", name);
return -1;
}
do_dirty_bitmap_merge(dst->node_name, name, src->node_name, name,
&err);
if (err) {
error_reportf_err(err, "Failed to populate bitmap %s: ", name);
qmp_block_dirty_bitmap_remove(dst->node_name, name, NULL);
return -1;
}
}
return 0;
}
#define MAX_BUF_SECTORS 32768
static void set_rate_limit(BlockBackend *blk, int64_t rate_limit)
{
ThrottleConfig cfg;
throttle_config_init(&cfg);
cfg.buckets[THROTTLE_BPS_WRITE].avg = rate_limit;
blk_io_limits_enable(blk, CONVERT_THROTTLE_GROUP);
blk_set_io_limits(blk, &cfg);
}
static int img_convert(int argc, char **argv)
{
int c, bs_i, flags, src_flags = BDRV_O_NO_SHARE;
const char *fmt = NULL, *out_fmt = NULL, *cache = "unsafe",
*src_cache = BDRV_DEFAULT_CACHE, *out_baseimg = NULL,
*out_filename, *out_baseimg_param, *snapshot_name = NULL,
*backing_fmt = NULL;
BlockDriver *drv = NULL, *proto_drv = NULL;
BlockDriverInfo bdi;
BlockDriverState *out_bs;
QemuOpts *opts = NULL, *sn_opts = NULL;
QemuOptsList *create_opts = NULL;
QDict *open_opts = NULL;
char *options = NULL;
Error *local_err = NULL;
bool writethrough, src_writethrough, image_opts = false,
skip_create = false, progress = false, tgt_image_opts = false;
int64_t ret = -EINVAL;
bool force_share = false;
bool explict_min_sparse = false;
bool bitmaps = false;
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps' The point of 'qemu-img convert --bitmaps' is to be a convenience for actions that are already possible through a string of smaller 'qemu-img bitmap' sub-commands. One situation not accounted for already is that if a source image contains an inconsistent bitmap (for example, because a qemu process died abruptly before flushing bitmap state), the user MUST delete those inconsistent bitmaps before anything else useful can be done with the image. We don't want to delete inconsistent bitmaps by default: although a corrupt bitmap is only a loss of optimization rather than a corruption of user-visible data, it is still nice to require the user to opt in to the fact that they are aware of the loss of the bitmap. Still, requiring the user to check 'qemu-img info' to see whether bitmaps are consistent, then use 'qemu-img bitmap --remove' to remove offenders, all before using 'qemu-img convert', is a lot more work than just adding a knob 'qemu-img convert --bitmaps --skip-broken-bitmaps' which opts in to skipping the broken bitmaps. After testing the new option, also demonstrate the way to manually fix things (either deleting bad bitmaps, or re-creating them as empty) so that it is possible to convert without the option. Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1946084 Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210709153951.2801666-4-eblake@redhat.com> [eblake: warning message tweak, test enhancements] Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2021-07-21 18:53:48 +03:00
bool skip_broken = false;
int64_t rate_limit = 0;
ImgConvertState s = (ImgConvertState) {
/* Need at least 4k of zeros for sparse detection */
.min_sparse = 8,
.copy_range = false,
.buf_sectors = IO_BUF_SIZE / BDRV_SECTOR_SIZE,
.wr_in_order = true,
.num_coroutines = 8,
};
for(;;) {
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"object", required_argument, 0, OPTION_OBJECT},
{"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{"force-share", no_argument, 0, 'U'},
{"target-image-opts", no_argument, 0, OPTION_TARGET_IMAGE_OPTS},
{"salvage", no_argument, 0, OPTION_SALVAGE},
{"target-is-zero", no_argument, 0, OPTION_TARGET_IS_ZERO},
{"bitmaps", no_argument, 0, OPTION_BITMAPS},
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps' The point of 'qemu-img convert --bitmaps' is to be a convenience for actions that are already possible through a string of smaller 'qemu-img bitmap' sub-commands. One situation not accounted for already is that if a source image contains an inconsistent bitmap (for example, because a qemu process died abruptly before flushing bitmap state), the user MUST delete those inconsistent bitmaps before anything else useful can be done with the image. We don't want to delete inconsistent bitmaps by default: although a corrupt bitmap is only a loss of optimization rather than a corruption of user-visible data, it is still nice to require the user to opt in to the fact that they are aware of the loss of the bitmap. Still, requiring the user to check 'qemu-img info' to see whether bitmaps are consistent, then use 'qemu-img bitmap --remove' to remove offenders, all before using 'qemu-img convert', is a lot more work than just adding a knob 'qemu-img convert --bitmaps --skip-broken-bitmaps' which opts in to skipping the broken bitmaps. After testing the new option, also demonstrate the way to manually fix things (either deleting bad bitmaps, or re-creating them as empty) so that it is possible to convert without the option. Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1946084 Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210709153951.2801666-4-eblake@redhat.com> [eblake: warning message tweak, test enhancements] Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2021-07-21 18:53:48 +03:00
{"skip-broken-bitmaps", no_argument, 0, OPTION_SKIP_BROKEN},
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, ":hf:O:B:CcF:o:l:S:pt:T:qnm:WUr:",
long_options, NULL);
if (c == -1) {
break;
}
switch(c) {
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
break;
case 'f':
fmt = optarg;
break;
case 'O':
out_fmt = optarg;
break;
case 'B':
out_baseimg = optarg;
break;
case 'C':
s.copy_range = true;
break;
case 'c':
s.compressed = true;
break;
case 'F':
backing_fmt = optarg;
break;
case 'o':
if (accumulate_options(&options, optarg) < 0) {
goto fail_getopt;
}
break;
case 'l':
if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
QemuOpts: Wean off qerror_report_err() qerror_report_err() is a transitional interface to help with converting existing monitor commands to QMP. It should not be used elsewhere. The only remaining user in qemu-option.c is qemu_opts_parse(). Is it used in QMP context? If not, we can simply replace qerror_report_err() by error_report_err(). The uses in qemu-img.c, qemu-io.c, qemu-nbd.c and under tests/ are clearly not in QMP context. The uses in vl.c aren't either, because the only QMP command handlers there are qmp_query_status() and qmp_query_machines(), and they don't call it. Remaining uses: * drive_def(): Command line -drive and such, HMP drive_add and pci_add * hmp_chardev_add(): HMP chardev-add * monitor_parse_command(): HMP core * tmp_config_parse(): Command line -tpmdev * net_host_device_add(): HMP host_net_add * net_client_parse(): Command line -net and -netdev * qemu_global_option(): Command line -global * vnc_parse_func(): Command line -display, -vnc, default display, HMP change, QMP change. Bummer. * qemu_pci_hot_add_nic(): HMP pci_add * usb_net_init(): Command line -usbdevice, HMP usb_add Propagate errors through qemu_opts_parse(). Create a convenience function qemu_opts_parse_noisily() that passes errors to error_report_err(). Switch all non-QMP users outside tests to it. That leaves vnc_parse_func(). Propagate errors through it. Since I'm touching it anyway, rename it to vnc_parse(). Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Eric Blake <eblake@redhat.com> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> Reviewed-by: Luiz Capitulino <lcapitulino@redhat.com>
2015-02-13 14:50:26 +03:00
sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
optarg, false);
if (!sn_opts) {
error_report("Failed in parsing snapshot param '%s'",
optarg);
goto fail_getopt;
}
} else {
snapshot_name = optarg;
}
break;
case 'S':
{
int64_t sval;
sval = cvtnum("buffer size for sparse output", optarg);
if (sval < 0) {
goto fail_getopt;
} else if (!QEMU_IS_ALIGNED(sval, BDRV_SECTOR_SIZE) ||
sval / BDRV_SECTOR_SIZE > MAX_BUF_SECTORS) {
error_report("Invalid buffer size for sparse output specified. "
"Valid sizes are multiples of %llu up to %llu. Select "
"0 to disable sparse detection (fully allocates output).",
BDRV_SECTOR_SIZE, MAX_BUF_SECTORS * BDRV_SECTOR_SIZE);
goto fail_getopt;
}
s.min_sparse = sval / BDRV_SECTOR_SIZE;
explict_min_sparse = true;
break;
}
case 'p':
progress = true;
break;
case 't':
cache = optarg;
break;
case 'T':
src_cache = optarg;
break;
case 'q':
s.quiet = true;
break;
case 'n':
skip_create = true;
break;
case 'm':
if (qemu_strtol(optarg, NULL, 0, &s.num_coroutines) ||
s.num_coroutines < 1 || s.num_coroutines > MAX_COROUTINES) {
error_report("Invalid number of coroutines. Allowed number of"
" coroutines is between 1 and %d", MAX_COROUTINES);
goto fail_getopt;
}
break;
case 'W':
s.wr_in_order = false;
break;
case 'U':
force_share = true;
break;
case 'r':
rate_limit = cvtnum("rate limit", optarg);
if (rate_limit < 0) {
goto fail_getopt;
}
break;
case OPTION_OBJECT:
user_creatable_process_cmdline(optarg);
break;
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
case OPTION_SALVAGE:
s.salvage = true;
break;
case OPTION_TARGET_IMAGE_OPTS:
tgt_image_opts = true;
break;
case OPTION_TARGET_IS_ZERO:
/*
* The user asserting that the target is blank has the
* same effect as the target driver supporting zero
* initialisation.
*/
s.has_zero_init = true;
break;
case OPTION_BITMAPS:
bitmaps = true;
break;
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps' The point of 'qemu-img convert --bitmaps' is to be a convenience for actions that are already possible through a string of smaller 'qemu-img bitmap' sub-commands. One situation not accounted for already is that if a source image contains an inconsistent bitmap (for example, because a qemu process died abruptly before flushing bitmap state), the user MUST delete those inconsistent bitmaps before anything else useful can be done with the image. We don't want to delete inconsistent bitmaps by default: although a corrupt bitmap is only a loss of optimization rather than a corruption of user-visible data, it is still nice to require the user to opt in to the fact that they are aware of the loss of the bitmap. Still, requiring the user to check 'qemu-img info' to see whether bitmaps are consistent, then use 'qemu-img bitmap --remove' to remove offenders, all before using 'qemu-img convert', is a lot more work than just adding a knob 'qemu-img convert --bitmaps --skip-broken-bitmaps' which opts in to skipping the broken bitmaps. After testing the new option, also demonstrate the way to manually fix things (either deleting bad bitmaps, or re-creating them as empty) so that it is possible to convert without the option. Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1946084 Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210709153951.2801666-4-eblake@redhat.com> [eblake: warning message tweak, test enhancements] Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2021-07-21 18:53:48 +03:00
case OPTION_SKIP_BROKEN:
skip_broken = true;
break;
}
}
if (!out_fmt && !tgt_image_opts) {
out_fmt = "raw";
}
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps' The point of 'qemu-img convert --bitmaps' is to be a convenience for actions that are already possible through a string of smaller 'qemu-img bitmap' sub-commands. One situation not accounted for already is that if a source image contains an inconsistent bitmap (for example, because a qemu process died abruptly before flushing bitmap state), the user MUST delete those inconsistent bitmaps before anything else useful can be done with the image. We don't want to delete inconsistent bitmaps by default: although a corrupt bitmap is only a loss of optimization rather than a corruption of user-visible data, it is still nice to require the user to opt in to the fact that they are aware of the loss of the bitmap. Still, requiring the user to check 'qemu-img info' to see whether bitmaps are consistent, then use 'qemu-img bitmap --remove' to remove offenders, all before using 'qemu-img convert', is a lot more work than just adding a knob 'qemu-img convert --bitmaps --skip-broken-bitmaps' which opts in to skipping the broken bitmaps. After testing the new option, also demonstrate the way to manually fix things (either deleting bad bitmaps, or re-creating them as empty) so that it is possible to convert without the option. Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1946084 Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210709153951.2801666-4-eblake@redhat.com> [eblake: warning message tweak, test enhancements] Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2021-07-21 18:53:48 +03:00
if (skip_broken && !bitmaps) {
error_report("Use of --skip-broken-bitmaps requires --bitmaps");
goto fail_getopt;
}
if (s.compressed && s.copy_range) {
error_report("Cannot enable copy offloading when -c is used");
goto fail_getopt;
}
if (explict_min_sparse && s.copy_range) {
error_report("Cannot enable copy offloading when -S is used");
goto fail_getopt;
}
if (s.copy_range && s.salvage) {
error_report("Cannot use copy offloading in salvaging mode");
goto fail_getopt;
}
if (tgt_image_opts && !skip_create) {
error_report("--target-image-opts requires use of -n flag");
goto fail_getopt;
}
if (skip_create && options) {
error_report("-o has no effect when skipping image creation");
goto fail_getopt;
}
if (s.has_zero_init && !skip_create) {
error_report("--target-is-zero requires use of -n flag");
goto fail_getopt;
}
s.src_num = argc - optind - 1;
out_filename = s.src_num >= 1 ? argv[argc - 1] : NULL;
if (options && has_help_option(options)) {
if (out_fmt) {
ret = print_block_option_help(out_filename, out_fmt);
goto fail_getopt;
} else {
error_report("Option help requires a format be specified");
goto fail_getopt;
}
}
if (s.src_num < 1) {
error_report("Must specify image file name");
goto fail_getopt;
}
/* ret is still -EINVAL until here */
ret = bdrv_parse_cache_mode(src_cache, &src_flags, &src_writethrough);
if (ret < 0) {
error_report("Invalid source cache option: %s", src_cache);
goto fail_getopt;
}
/* Initialize before goto out */
if (s.quiet) {
progress = false;
}
qemu_progress_init(progress, 1.0);
qemu_progress_print(0, 100);
s.src = g_new0(BlockBackend *, s.src_num);
s.src_sectors = g_new(int64_t, s.src_num);
s.src_alignment = g_new(int, s.src_num);
for (bs_i = 0; bs_i < s.src_num; bs_i++) {
BlockDriverState *src_bs;
s.src[bs_i] = img_open(image_opts, argv[optind + bs_i],
fmt, src_flags, src_writethrough, s.quiet,
force_share);
if (!s.src[bs_i]) {
ret = -1;
goto out;
}
s.src_sectors[bs_i] = blk_nb_sectors(s.src[bs_i]);
if (s.src_sectors[bs_i] < 0) {
error_report("Could not get size of %s: %s",
argv[optind + bs_i], strerror(-s.src_sectors[bs_i]));
ret = -1;
goto out;
}
src_bs = blk_bs(s.src[bs_i]);
s.src_alignment[bs_i] = DIV_ROUND_UP(src_bs->bl.request_alignment,
BDRV_SECTOR_SIZE);
if (!bdrv_get_info(src_bs, &bdi)) {
s.src_alignment[bs_i] = MAX(s.src_alignment[bs_i],
bdi.cluster_size / BDRV_SECTOR_SIZE);
}
s.total_sectors += s.src_sectors[bs_i];
}
if (sn_opts) {
bdrv_snapshot_load_tmp(blk_bs(s.src[0]),
qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
&local_err);
} else if (snapshot_name != NULL) {
if (s.src_num > 1) {
error_report("No support for concatenating multiple snapshot");
ret = -1;
goto out;
}
bdrv_snapshot_load_tmp_by_id_or_name(blk_bs(s.src[0]), snapshot_name,
&local_err);
}
if (local_err) {
error_reportf_err(local_err, "Failed to load snapshot: ");
ret = -1;
goto out;
}
if (!skip_create) {
/* Find driver and parse its options */
drv = bdrv_find_format(out_fmt);
if (!drv) {
error_report("Unknown file format '%s'", out_fmt);
ret = -1;
goto out;
}
proto_drv = bdrv_find_protocol(out_filename, true, &local_err);
if (!proto_drv) {
error_report_err(local_err);
ret = -1;
goto out;
}
if (!drv->create_opts) {
error_report("Format driver '%s' does not support image creation",
drv->format_name);
ret = -1;
goto out;
}
if (!proto_drv->create_opts) {
error_report("Protocol driver '%s' does not support image creation",
proto_drv->format_name);
ret = -1;
goto out;
}
create_opts = qemu_opts_append(create_opts, drv->create_opts);
create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
if (options) {
if (!qemu_opts_do_parse(opts, options, NULL, &local_err)) {
error_report_err(local_err);
ret = -1;
goto out;
}
}
qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
s.total_sectors * BDRV_SECTOR_SIZE, &error_abort);
ret = add_old_style_options(out_fmt, opts, out_baseimg, backing_fmt);
if (ret < 0) {
goto out;
}
}
/* Get backing file name if -o backing_file was used */
out_baseimg_param = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
if (out_baseimg_param) {
out_baseimg = out_baseimg_param;
}
s.target_has_backing = (bool) out_baseimg;
if (s.has_zero_init && s.target_has_backing) {
error_report("Cannot use --target-is-zero when the destination "
"image has a backing file");
goto out;
}
if (s.src_num > 1 && out_baseimg) {
error_report("Having a backing file for the target makes no sense when "
"concatenating multiple input images");
ret = -1;
goto out;
}
qemu-img: Deprecate use of -b without -F Creating an image that requires format probing of the backing image is potentially unsafe (we've had several CVEs over the years based on probes leaking information to the guest on a subsequent boot, although these days tools like libvirt are aware of the issue enough to prevent the worst effects). For example, if our probing algorithm ever changes, or if other tools like libvirt determine a different probe result than we do, then subsequent use of that backing file under a different format will present corrupted data to the guest. Fortunately, the worst effects occur only when the backing image is originally raw, and we at least prevent commit into a probed raw backing file that would change its probed type. Still, it is worth starting a deprecation clock so that future qemu-img can refuse to create backing chains that would rely on probing, to encourage clients to avoid unsafe practices. Most warnings are intentionally emitted from bdrv_img_create() in the block layer, but qemu-img convert uses bdrv_create() which cannot emit its own warning without causing spurious warnings on other code paths. In the end, all command-line image creation or backing file rewriting now performs a check. Furthermore, if we probe a backing file as non-raw, then it is safe to explicitly record that result (rather than relying on future probes); only where we probe a raw image do we care about further warnings to the user when using such an image (for example, commits into a probed-raw backing file are prevented), to help them improve their tooling. But whether or not we make the probe results explicit, we still warn the user to remind them to upgrade their workflow to supply -F always. iotest 114 specifically wants to create an unsafe image for later amendment rather than defaulting to our new default of recording a probed format, so it needs an update. While touching it, expand it to cover all of the various warnings enabled by this patch. iotest 301 also shows a change to qcow messages. Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20200706203954.341758-11-eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2020-07-06 23:39:54 +03:00
if (out_baseimg_param) {
if (!qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT)) {
error_report("Use of backing file requires explicit "
"backing format");
ret = -1;
goto out;
qemu-img: Deprecate use of -b without -F Creating an image that requires format probing of the backing image is potentially unsafe (we've had several CVEs over the years based on probes leaking information to the guest on a subsequent boot, although these days tools like libvirt are aware of the issue enough to prevent the worst effects). For example, if our probing algorithm ever changes, or if other tools like libvirt determine a different probe result than we do, then subsequent use of that backing file under a different format will present corrupted data to the guest. Fortunately, the worst effects occur only when the backing image is originally raw, and we at least prevent commit into a probed raw backing file that would change its probed type. Still, it is worth starting a deprecation clock so that future qemu-img can refuse to create backing chains that would rely on probing, to encourage clients to avoid unsafe practices. Most warnings are intentionally emitted from bdrv_img_create() in the block layer, but qemu-img convert uses bdrv_create() which cannot emit its own warning without causing spurious warnings on other code paths. In the end, all command-line image creation or backing file rewriting now performs a check. Furthermore, if we probe a backing file as non-raw, then it is safe to explicitly record that result (rather than relying on future probes); only where we probe a raw image do we care about further warnings to the user when using such an image (for example, commits into a probed-raw backing file are prevented), to help them improve their tooling. But whether or not we make the probe results explicit, we still warn the user to remind them to upgrade their workflow to supply -F always. iotest 114 specifically wants to create an unsafe image for later amendment rather than defaulting to our new default of recording a probed format, so it needs an update. While touching it, expand it to cover all of the various warnings enabled by this patch. iotest 301 also shows a change to qcow messages. Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20200706203954.341758-11-eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2020-07-06 23:39:54 +03:00
}
}
/* Check if compression is supported */
if (s.compressed) {
bool encryption =
qemu_opt_get_bool(opts, BLOCK_OPT_ENCRYPT, false);
const char *encryptfmt =
qemu_opt_get(opts, BLOCK_OPT_ENCRYPT_FORMAT);
const char *preallocation =
qemu_opt_get(opts, BLOCK_OPT_PREALLOC);
if (drv && !block_driver_can_compress(drv)) {
error_report("Compression not supported for this file format");
ret = -1;
goto out;
}
if (encryption || encryptfmt) {
error_report("Compression and encryption not supported at "
"the same time");
ret = -1;
goto out;
}
if (preallocation
&& strcmp(preallocation, "off"))
{
error_report("Compression and preallocation not supported at "
"the same time");
ret = -1;
goto out;
}
}
/* Determine if bitmaps need copying */
if (bitmaps) {
if (s.src_num > 1) {
error_report("Copying bitmaps only possible with single source");
ret = -1;
goto out;
}
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps' The point of 'qemu-img convert --bitmaps' is to be a convenience for actions that are already possible through a string of smaller 'qemu-img bitmap' sub-commands. One situation not accounted for already is that if a source image contains an inconsistent bitmap (for example, because a qemu process died abruptly before flushing bitmap state), the user MUST delete those inconsistent bitmaps before anything else useful can be done with the image. We don't want to delete inconsistent bitmaps by default: although a corrupt bitmap is only a loss of optimization rather than a corruption of user-visible data, it is still nice to require the user to opt in to the fact that they are aware of the loss of the bitmap. Still, requiring the user to check 'qemu-img info' to see whether bitmaps are consistent, then use 'qemu-img bitmap --remove' to remove offenders, all before using 'qemu-img convert', is a lot more work than just adding a knob 'qemu-img convert --bitmaps --skip-broken-bitmaps' which opts in to skipping the broken bitmaps. After testing the new option, also demonstrate the way to manually fix things (either deleting bad bitmaps, or re-creating them as empty) so that it is possible to convert without the option. Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1946084 Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210709153951.2801666-4-eblake@redhat.com> [eblake: warning message tweak, test enhancements] Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2021-07-21 18:53:48 +03:00
ret = convert_check_bitmaps(blk_bs(s.src[0]), skip_broken);
if (ret < 0) {
goto out;
}
}
/*
* The later open call will need any decryption secrets, and
* bdrv_create() will purge "opts", so extract them now before
* they are lost.
*/
if (!skip_create) {
open_opts = qdict_new();
qemu_opt_foreach(opts, img_add_key_secrets, open_opts, &error_abort);
/* Create the new image */
ret = bdrv_create(drv, out_filename, opts, &local_err);
if (ret < 0) {
error_reportf_err(local_err, "%s: error while converting %s: ",
out_filename, out_fmt);
goto out;
}
}
s.target_is_new = !skip_create;
flags = s.min_sparse ? (BDRV_O_RDWR | BDRV_O_UNMAP) : BDRV_O_RDWR;
ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
if (ret < 0) {
error_report("Invalid cache option: %s", cache);
goto out;
}
if (flags & BDRV_O_NOCACHE) {
/*
* If we open the target with O_DIRECT, it may be necessary to
* extend its size to align to the physical sector size.
*/
flags |= BDRV_O_RESIZE;
}
if (skip_create) {
s.target = img_open(tgt_image_opts, out_filename, out_fmt,
flags, writethrough, s.quiet, false);
} else {
/* TODO ultimately we should allow --target-image-opts
* to be used even when -n is not given.
* That has to wait for bdrv_create to be improved
* to allow filenames in option syntax
*/
s.target = img_open_file(out_filename, open_opts, out_fmt,
flags, writethrough, s.quiet, false);
open_opts = NULL; /* blk_new_open will have freed it */
}
if (!s.target) {
ret = -1;
goto out;
}
out_bs = blk_bs(s.target);
if (bitmaps && !bdrv_supports_persistent_dirty_bitmap(out_bs)) {
error_report("Format driver '%s' does not support bitmaps",
out_bs->drv->format_name);
ret = -1;
goto out;
}
if (s.compressed && !block_driver_can_compress(out_bs->drv)) {
error_report("Compression not supported for this file format");
ret = -1;
goto out;
}
/* increase bufsectors from the default 4096 (2M) if opt_transfer
* or discard_alignment of the out_bs is greater. Limit to
* MAX_BUF_SECTORS as maximum which is currently 32768 (16MB). */
s.buf_sectors = MIN(MAX_BUF_SECTORS,
MAX(s.buf_sectors,
MAX(out_bs->bl.opt_transfer >> BDRV_SECTOR_BITS,
out_bs->bl.pdiscard_alignment >>
BDRV_SECTOR_BITS)));
/* try to align the write requests to the destination to avoid unnecessary
* RMW cycles. */
s.alignment = MAX(pow2floor(s.min_sparse),
DIV_ROUND_UP(out_bs->bl.request_alignment,
BDRV_SECTOR_SIZE));
assert(is_power_of_2(s.alignment));
if (skip_create) {
int64_t output_sectors = blk_nb_sectors(s.target);
if (output_sectors < 0) {
error_report("unable to get output image length: %s",
strerror(-output_sectors));
ret = -1;
goto out;
} else if (output_sectors < s.total_sectors) {
error_report("output file is smaller than input file");
ret = -1;
goto out;
}
}
if (s.target_has_backing && s.target_is_new) {
/* Errors are treated as "backing length unknown" (which means
* s.target_backing_sectors has to be negative, which it will
* be automatically). The backing file length is used only
* for optimizations, so such a case is not fatal. */
s.target_backing_sectors =
bdrv_nb_sectors(bdrv_backing_chain_next(out_bs));
} else {
s.target_backing_sectors = -1;
}
ret = bdrv_get_info(out_bs, &bdi);
if (ret < 0) {
if (s.compressed) {
error_report("could not get block driver info");
goto out;
}
} else {
s.compressed = s.compressed || bdi.needs_compressed_writes;
s.cluster_sectors = bdi.cluster_size / BDRV_SECTOR_SIZE;
}
if (rate_limit) {
set_rate_limit(s.target, rate_limit);
}
ret = convert_do_copy(&s);
/* Now copy the bitmaps */
if (bitmaps && ret == 0) {
qemu-img: Add --skip-broken-bitmaps for 'convert --bitmaps' The point of 'qemu-img convert --bitmaps' is to be a convenience for actions that are already possible through a string of smaller 'qemu-img bitmap' sub-commands. One situation not accounted for already is that if a source image contains an inconsistent bitmap (for example, because a qemu process died abruptly before flushing bitmap state), the user MUST delete those inconsistent bitmaps before anything else useful can be done with the image. We don't want to delete inconsistent bitmaps by default: although a corrupt bitmap is only a loss of optimization rather than a corruption of user-visible data, it is still nice to require the user to opt in to the fact that they are aware of the loss of the bitmap. Still, requiring the user to check 'qemu-img info' to see whether bitmaps are consistent, then use 'qemu-img bitmap --remove' to remove offenders, all before using 'qemu-img convert', is a lot more work than just adding a knob 'qemu-img convert --bitmaps --skip-broken-bitmaps' which opts in to skipping the broken bitmaps. After testing the new option, also demonstrate the way to manually fix things (either deleting bad bitmaps, or re-creating them as empty) so that it is possible to convert without the option. Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1946084 Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210709153951.2801666-4-eblake@redhat.com> [eblake: warning message tweak, test enhancements] Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2021-07-21 18:53:48 +03:00
ret = convert_copy_bitmaps(blk_bs(s.src[0]), out_bs, skip_broken);
}
out:
if (!ret) {
qemu_progress_print(100, 0);
}
qemu_progress_end();
qemu_opts_del(opts);
qemu_opts_free(create_opts);
qobject_unref(open_opts);
blk_unref(s.target);
if (s.src) {
for (bs_i = 0; bs_i < s.src_num; bs_i++) {
blk_unref(s.src[bs_i]);
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
}
g_free(s.src);
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
}
g_free(s.src_sectors);
g_free(s.src_alignment);
fail_getopt:
qemu_opts_del(sn_opts);
g_free(options);
return !!ret;
}
static void dump_snapshots(BlockDriverState *bs)
{
QEMUSnapshotInfo *sn_tab, *sn;
int nb_sns, i;
nb_sns = bdrv_snapshot_list(bs, &sn_tab);
if (nb_sns <= 0)
return;
printf("Snapshot list:\n");
bdrv_snapshot_dump(NULL);
printf("\n");
for(i = 0; i < nb_sns; i++) {
sn = &sn_tab[i];
bdrv_snapshot_dump(sn);
printf("\n");
}
g_free(sn_tab);
}
static void dump_json_block_graph_info_list(BlockGraphInfoList *list)
{
GString *str;
QObject *obj;
Visitor *v = qobject_output_visitor_new(&obj);
qapi: Add new visit_complete() function Making each output visitor provide its own output collection function was the only remaining reason for exposing visitor sub-types to the rest of the code base. Add a polymorphic visit_complete() function which is a no-op for input visitors, and which populates an opaque pointer for output visitors. For maximum type-safety, also add a parameter to the output visitor constructors with a type-correct version of the output pointer, and assert that the two uses match. This approach was considered superior to either passing the output parameter only during construction (action at a distance during visit_free() feels awkward) or only during visit_complete() (defeating type safety makes it easier to use incorrectly). Most callers were function-local, and therefore a mechanical conversion; the testsuite was a bit trickier, but the previous cleanup patch minimized the churn here. The visit_complete() function may be called at most once; doing so lets us use transfer semantics rather than duplication or ref-count semantics to get the just-built output back to the caller, even though it means our behavior is not idempotent. Generated code is simplified as follows for events: |@@ -26,7 +26,7 @@ void qapi_event_send_acpi_device_ost(ACP | QDict *qmp; | Error *err = NULL; | QMPEventFuncEmit emit; |- QmpOutputVisitor *qov; |+ QObject *obj; | Visitor *v; | q_obj_ACPI_DEVICE_OST_arg param = { | info |@@ -39,8 +39,7 @@ void qapi_event_send_acpi_device_ost(ACP | | qmp = qmp_event_build_dict("ACPI_DEVICE_OST"); | |- qov = qmp_output_visitor_new(); |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(&obj); | | visit_start_struct(v, "ACPI_DEVICE_OST", NULL, 0, &err); | if (err) { |@@ -55,7 +54,8 @@ void qapi_event_send_acpi_device_ost(ACP | goto out; | } | |- qdict_put_obj(qmp, "data", qmp_output_get_qobject(qov)); |+ visit_complete(v, &obj); |+ qdict_put_obj(qmp, "data", obj); | emit(QAPI_EVENT_ACPI_DEVICE_OST, qmp, &err); and for commands: | { | Error *err = NULL; |- QmpOutputVisitor *qov = qmp_output_visitor_new(); | Visitor *v; | |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(ret_out); | visit_type_AddfdInfo(v, "unused", &ret_in, &err); |- if (err) { |- goto out; |+ if (!err) { |+ visit_complete(v, ret_out); | } |- *ret_out = qmp_output_get_qobject(qov); |- |-out: | error_propagate(errp, err); Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <1465490926-28625-13-git-send-email-eblake@redhat.com> Reviewed-by: Markus Armbruster <armbru@redhat.com> Signed-off-by: Markus Armbruster <armbru@redhat.com>
2016-06-09 19:48:43 +03:00
visit_type_BlockGraphInfoList(v, NULL, &list, &error_abort);
qapi: Add new visit_complete() function Making each output visitor provide its own output collection function was the only remaining reason for exposing visitor sub-types to the rest of the code base. Add a polymorphic visit_complete() function which is a no-op for input visitors, and which populates an opaque pointer for output visitors. For maximum type-safety, also add a parameter to the output visitor constructors with a type-correct version of the output pointer, and assert that the two uses match. This approach was considered superior to either passing the output parameter only during construction (action at a distance during visit_free() feels awkward) or only during visit_complete() (defeating type safety makes it easier to use incorrectly). Most callers were function-local, and therefore a mechanical conversion; the testsuite was a bit trickier, but the previous cleanup patch minimized the churn here. The visit_complete() function may be called at most once; doing so lets us use transfer semantics rather than duplication or ref-count semantics to get the just-built output back to the caller, even though it means our behavior is not idempotent. Generated code is simplified as follows for events: |@@ -26,7 +26,7 @@ void qapi_event_send_acpi_device_ost(ACP | QDict *qmp; | Error *err = NULL; | QMPEventFuncEmit emit; |- QmpOutputVisitor *qov; |+ QObject *obj; | Visitor *v; | q_obj_ACPI_DEVICE_OST_arg param = { | info |@@ -39,8 +39,7 @@ void qapi_event_send_acpi_device_ost(ACP | | qmp = qmp_event_build_dict("ACPI_DEVICE_OST"); | |- qov = qmp_output_visitor_new(); |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(&obj); | | visit_start_struct(v, "ACPI_DEVICE_OST", NULL, 0, &err); | if (err) { |@@ -55,7 +54,8 @@ void qapi_event_send_acpi_device_ost(ACP | goto out; | } | |- qdict_put_obj(qmp, "data", qmp_output_get_qobject(qov)); |+ visit_complete(v, &obj); |+ qdict_put_obj(qmp, "data", obj); | emit(QAPI_EVENT_ACPI_DEVICE_OST, qmp, &err); and for commands: | { | Error *err = NULL; |- QmpOutputVisitor *qov = qmp_output_visitor_new(); | Visitor *v; | |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(ret_out); | visit_type_AddfdInfo(v, "unused", &ret_in, &err); |- if (err) { |- goto out; |+ if (!err) { |+ visit_complete(v, ret_out); | } |- *ret_out = qmp_output_get_qobject(qov); |- |-out: | error_propagate(errp, err); Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <1465490926-28625-13-git-send-email-eblake@redhat.com> Reviewed-by: Markus Armbruster <armbru@redhat.com> Signed-off-by: Markus Armbruster <armbru@redhat.com>
2016-06-09 19:48:43 +03:00
visit_complete(v, &obj);
str = qobject_to_json_pretty(obj, true);
assert(str != NULL);
printf("%s\n", str->str);
qobject_unref(obj);
qapi: Add new visit_complete() function Making each output visitor provide its own output collection function was the only remaining reason for exposing visitor sub-types to the rest of the code base. Add a polymorphic visit_complete() function which is a no-op for input visitors, and which populates an opaque pointer for output visitors. For maximum type-safety, also add a parameter to the output visitor constructors with a type-correct version of the output pointer, and assert that the two uses match. This approach was considered superior to either passing the output parameter only during construction (action at a distance during visit_free() feels awkward) or only during visit_complete() (defeating type safety makes it easier to use incorrectly). Most callers were function-local, and therefore a mechanical conversion; the testsuite was a bit trickier, but the previous cleanup patch minimized the churn here. The visit_complete() function may be called at most once; doing so lets us use transfer semantics rather than duplication or ref-count semantics to get the just-built output back to the caller, even though it means our behavior is not idempotent. Generated code is simplified as follows for events: |@@ -26,7 +26,7 @@ void qapi_event_send_acpi_device_ost(ACP | QDict *qmp; | Error *err = NULL; | QMPEventFuncEmit emit; |- QmpOutputVisitor *qov; |+ QObject *obj; | Visitor *v; | q_obj_ACPI_DEVICE_OST_arg param = { | info |@@ -39,8 +39,7 @@ void qapi_event_send_acpi_device_ost(ACP | | qmp = qmp_event_build_dict("ACPI_DEVICE_OST"); | |- qov = qmp_output_visitor_new(); |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(&obj); | | visit_start_struct(v, "ACPI_DEVICE_OST", NULL, 0, &err); | if (err) { |@@ -55,7 +54,8 @@ void qapi_event_send_acpi_device_ost(ACP | goto out; | } | |- qdict_put_obj(qmp, "data", qmp_output_get_qobject(qov)); |+ visit_complete(v, &obj); |+ qdict_put_obj(qmp, "data", obj); | emit(QAPI_EVENT_ACPI_DEVICE_OST, qmp, &err); and for commands: | { | Error *err = NULL; |- QmpOutputVisitor *qov = qmp_output_visitor_new(); | Visitor *v; | |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(ret_out); | visit_type_AddfdInfo(v, "unused", &ret_in, &err); |- if (err) { |- goto out; |+ if (!err) { |+ visit_complete(v, ret_out); | } |- *ret_out = qmp_output_get_qobject(qov); |- |-out: | error_propagate(errp, err); Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <1465490926-28625-13-git-send-email-eblake@redhat.com> Reviewed-by: Markus Armbruster <armbru@redhat.com> Signed-off-by: Markus Armbruster <armbru@redhat.com>
2016-06-09 19:48:43 +03:00
visit_free(v);
g_string_free(str, true);
}
static void dump_json_block_graph_info(BlockGraphInfo *info)
{
GString *str;
QObject *obj;
Visitor *v = qobject_output_visitor_new(&obj);
qapi: Add new visit_complete() function Making each output visitor provide its own output collection function was the only remaining reason for exposing visitor sub-types to the rest of the code base. Add a polymorphic visit_complete() function which is a no-op for input visitors, and which populates an opaque pointer for output visitors. For maximum type-safety, also add a parameter to the output visitor constructors with a type-correct version of the output pointer, and assert that the two uses match. This approach was considered superior to either passing the output parameter only during construction (action at a distance during visit_free() feels awkward) or only during visit_complete() (defeating type safety makes it easier to use incorrectly). Most callers were function-local, and therefore a mechanical conversion; the testsuite was a bit trickier, but the previous cleanup patch minimized the churn here. The visit_complete() function may be called at most once; doing so lets us use transfer semantics rather than duplication or ref-count semantics to get the just-built output back to the caller, even though it means our behavior is not idempotent. Generated code is simplified as follows for events: |@@ -26,7 +26,7 @@ void qapi_event_send_acpi_device_ost(ACP | QDict *qmp; | Error *err = NULL; | QMPEventFuncEmit emit; |- QmpOutputVisitor *qov; |+ QObject *obj; | Visitor *v; | q_obj_ACPI_DEVICE_OST_arg param = { | info |@@ -39,8 +39,7 @@ void qapi_event_send_acpi_device_ost(ACP | | qmp = qmp_event_build_dict("ACPI_DEVICE_OST"); | |- qov = qmp_output_visitor_new(); |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(&obj); | | visit_start_struct(v, "ACPI_DEVICE_OST", NULL, 0, &err); | if (err) { |@@ -55,7 +54,8 @@ void qapi_event_send_acpi_device_ost(ACP | goto out; | } | |- qdict_put_obj(qmp, "data", qmp_output_get_qobject(qov)); |+ visit_complete(v, &obj); |+ qdict_put_obj(qmp, "data", obj); | emit(QAPI_EVENT_ACPI_DEVICE_OST, qmp, &err); and for commands: | { | Error *err = NULL; |- QmpOutputVisitor *qov = qmp_output_visitor_new(); | Visitor *v; | |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(ret_out); | visit_type_AddfdInfo(v, "unused", &ret_in, &err); |- if (err) { |- goto out; |+ if (!err) { |+ visit_complete(v, ret_out); | } |- *ret_out = qmp_output_get_qobject(qov); |- |-out: | error_propagate(errp, err); Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <1465490926-28625-13-git-send-email-eblake@redhat.com> Reviewed-by: Markus Armbruster <armbru@redhat.com> Signed-off-by: Markus Armbruster <armbru@redhat.com>
2016-06-09 19:48:43 +03:00
visit_type_BlockGraphInfo(v, NULL, &info, &error_abort);
qapi: Add new visit_complete() function Making each output visitor provide its own output collection function was the only remaining reason for exposing visitor sub-types to the rest of the code base. Add a polymorphic visit_complete() function which is a no-op for input visitors, and which populates an opaque pointer for output visitors. For maximum type-safety, also add a parameter to the output visitor constructors with a type-correct version of the output pointer, and assert that the two uses match. This approach was considered superior to either passing the output parameter only during construction (action at a distance during visit_free() feels awkward) or only during visit_complete() (defeating type safety makes it easier to use incorrectly). Most callers were function-local, and therefore a mechanical conversion; the testsuite was a bit trickier, but the previous cleanup patch minimized the churn here. The visit_complete() function may be called at most once; doing so lets us use transfer semantics rather than duplication or ref-count semantics to get the just-built output back to the caller, even though it means our behavior is not idempotent. Generated code is simplified as follows for events: |@@ -26,7 +26,7 @@ void qapi_event_send_acpi_device_ost(ACP | QDict *qmp; | Error *err = NULL; | QMPEventFuncEmit emit; |- QmpOutputVisitor *qov; |+ QObject *obj; | Visitor *v; | q_obj_ACPI_DEVICE_OST_arg param = { | info |@@ -39,8 +39,7 @@ void qapi_event_send_acpi_device_ost(ACP | | qmp = qmp_event_build_dict("ACPI_DEVICE_OST"); | |- qov = qmp_output_visitor_new(); |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(&obj); | | visit_start_struct(v, "ACPI_DEVICE_OST", NULL, 0, &err); | if (err) { |@@ -55,7 +54,8 @@ void qapi_event_send_acpi_device_ost(ACP | goto out; | } | |- qdict_put_obj(qmp, "data", qmp_output_get_qobject(qov)); |+ visit_complete(v, &obj); |+ qdict_put_obj(qmp, "data", obj); | emit(QAPI_EVENT_ACPI_DEVICE_OST, qmp, &err); and for commands: | { | Error *err = NULL; |- QmpOutputVisitor *qov = qmp_output_visitor_new(); | Visitor *v; | |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(ret_out); | visit_type_AddfdInfo(v, "unused", &ret_in, &err); |- if (err) { |- goto out; |+ if (!err) { |+ visit_complete(v, ret_out); | } |- *ret_out = qmp_output_get_qobject(qov); |- |-out: | error_propagate(errp, err); Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <1465490926-28625-13-git-send-email-eblake@redhat.com> Reviewed-by: Markus Armbruster <armbru@redhat.com> Signed-off-by: Markus Armbruster <armbru@redhat.com>
2016-06-09 19:48:43 +03:00
visit_complete(v, &obj);
str = qobject_to_json_pretty(obj, true);
assert(str != NULL);
printf("%s\n", str->str);
qobject_unref(obj);
qapi: Add new visit_complete() function Making each output visitor provide its own output collection function was the only remaining reason for exposing visitor sub-types to the rest of the code base. Add a polymorphic visit_complete() function which is a no-op for input visitors, and which populates an opaque pointer for output visitors. For maximum type-safety, also add a parameter to the output visitor constructors with a type-correct version of the output pointer, and assert that the two uses match. This approach was considered superior to either passing the output parameter only during construction (action at a distance during visit_free() feels awkward) or only during visit_complete() (defeating type safety makes it easier to use incorrectly). Most callers were function-local, and therefore a mechanical conversion; the testsuite was a bit trickier, but the previous cleanup patch minimized the churn here. The visit_complete() function may be called at most once; doing so lets us use transfer semantics rather than duplication or ref-count semantics to get the just-built output back to the caller, even though it means our behavior is not idempotent. Generated code is simplified as follows for events: |@@ -26,7 +26,7 @@ void qapi_event_send_acpi_device_ost(ACP | QDict *qmp; | Error *err = NULL; | QMPEventFuncEmit emit; |- QmpOutputVisitor *qov; |+ QObject *obj; | Visitor *v; | q_obj_ACPI_DEVICE_OST_arg param = { | info |@@ -39,8 +39,7 @@ void qapi_event_send_acpi_device_ost(ACP | | qmp = qmp_event_build_dict("ACPI_DEVICE_OST"); | |- qov = qmp_output_visitor_new(); |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(&obj); | | visit_start_struct(v, "ACPI_DEVICE_OST", NULL, 0, &err); | if (err) { |@@ -55,7 +54,8 @@ void qapi_event_send_acpi_device_ost(ACP | goto out; | } | |- qdict_put_obj(qmp, "data", qmp_output_get_qobject(qov)); |+ visit_complete(v, &obj); |+ qdict_put_obj(qmp, "data", obj); | emit(QAPI_EVENT_ACPI_DEVICE_OST, qmp, &err); and for commands: | { | Error *err = NULL; |- QmpOutputVisitor *qov = qmp_output_visitor_new(); | Visitor *v; | |- v = qmp_output_get_visitor(qov); |+ v = qmp_output_visitor_new(ret_out); | visit_type_AddfdInfo(v, "unused", &ret_in, &err); |- if (err) { |- goto out; |+ if (!err) { |+ visit_complete(v, ret_out); | } |- *ret_out = qmp_output_get_qobject(qov); |- |-out: | error_propagate(errp, err); Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <1465490926-28625-13-git-send-email-eblake@redhat.com> Reviewed-by: Markus Armbruster <armbru@redhat.com> Signed-off-by: Markus Armbruster <armbru@redhat.com>
2016-06-09 19:48:43 +03:00
visit_free(v);
g_string_free(str, true);
}
static void dump_human_image_info(BlockGraphInfo *info, int indentation,
const char *path)
{
BlockChildInfoList *children_list;
qemu-img: Change info key names for protocol nodes Currently, when querying a qcow2 image, qemu-img info reports something like this: image: test.qcow2 file format: qcow2 virtual size: 64 MiB (67108864 bytes) disk size: 196 KiB cluster_size: 65536 Format specific information: compat: 1.1 compression type: zlib lazy refcounts: false refcount bits: 16 corrupt: false extended l2: false Child node '/file': image: test.qcow2 file format: file virtual size: 192 KiB (197120 bytes) disk size: 196 KiB Format specific information: extent size hint: 1048576 Notably, the way the keys are named is specific for image files: The filename is shown under "image", the BDS driver under "file format", and the BDS length under "virtual size". This does not make much sense for nodes that are not actually supposed to be guest images, like the /file child node shown above. Give bdrv_node_info_dump() a @protocol parameter that gives a hint that the respective node is probably just used for data storage and does not necessarily present the data for a VM guest disk. This renames the keys so that with this patch, the output becomes: image: test.qcow2 [...] Child node '/file': filename: test.qcow2 protocol type: file file length: 192 KiB (197120 bytes) disk size: 196 KiB Format specific information: extent size hint: 1048576 (Perhaps we should also rename "Format specific information", but I could not come up with anything better that will not become problematic if we guess wrong with the protocol "heuristic".) This change affects iotest 302, which has protocol node information in its reference output. Signed-off-by: Hanna Reitz <hreitz@redhat.com> Message-Id: <20220620162704.80987-13-hreitz@redhat.com> Reviewed-by: Kevin Wolf <kwolf@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2022-06-20 19:27:04 +03:00
bdrv_node_info_dump(qapi_BlockGraphInfo_base(info), indentation,
info->children == NULL);
for (children_list = info->children; children_list;
children_list = children_list->next)
{
BlockChildInfo *child = children_list->value;
g_autofree char *child_path = NULL;
printf("%*sChild node '%s%s':\n",
indentation * 4, "", path, child->name);
child_path = g_strdup_printf("%s%s/", path, child->name);
dump_human_image_info(child->info, indentation + 1, child_path);
}
}
static void dump_human_image_info_list(BlockGraphInfoList *list)
{
BlockGraphInfoList *elem;
bool delim = false;
for (elem = list; elem; elem = elem->next) {
if (delim) {
printf("\n");
}
delim = true;
dump_human_image_info(elem->value, 0, "/");
}
}
static gboolean str_equal_func(gconstpointer a, gconstpointer b)
{
return strcmp(a, b) == 0;
}
/**
* Open an image file chain and return an BlockGraphInfoList
*
* @filename: topmost image filename
* @fmt: topmost image format (may be NULL to autodetect)
* @chain: true - enumerate entire backing file chain
* false - only topmost image file
*
* Returns a list of BlockNodeInfo objects or NULL if there was an error
* opening an image file. If there was an error a message will have been
* printed to stderr.
*/
static BlockGraphInfoList *collect_image_info_list(bool image_opts,
const char *filename,
const char *fmt,
bool chain, bool force_share)
{
BlockGraphInfoList *head = NULL;
BlockGraphInfoList **tail = &head;
GHashTable *filenames;
Error *err = NULL;
filenames = g_hash_table_new_full(g_str_hash, str_equal_func, NULL, NULL);
while (filename) {
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
BlockBackend *blk;
BlockDriverState *bs;
BlockGraphInfo *info;
if (g_hash_table_lookup_extended(filenames, filename, NULL, NULL)) {
error_report("Backing file '%s' creates an infinite loop.",
filename);
goto err;
}
g_hash_table_insert(filenames, (gpointer)filename, NULL);
blk = img_open(image_opts, filename, fmt,
BDRV_O_NO_BACKING | BDRV_O_NO_IO, false, false,
force_share);
if (!blk) {
goto err;
}
bs = blk_bs(blk);
/*
* Note that the returned BlockGraphInfo object will not have
* information about this image's backing node, because we have opened
* it with BDRV_O_NO_BACKING. Printing this object will therefore not
* duplicate the backing chain information that we obtain by walking
* the chain manually here.
*/
bdrv_graph_rdlock_main_loop();
bdrv_query_block_graph_info(bs, &info, &err);
bdrv_graph_rdunlock_main_loop();
if (err) {
error_report_err(err);
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
blk_unref(blk);
goto err;
}
QAPI_LIST_APPEND(tail, info);
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
blk_unref(blk);
/* Clear parameters that only apply to the topmost image */
filename = fmt = NULL;
image_opts = false;
if (chain) {
if (info->full_backing_filename) {
filename = info->full_backing_filename;
} else if (info->backing_filename) {
error_report("Could not determine absolute backing filename,"
" but backing filename '%s' present",
info->backing_filename);
goto err;
}
if (info->backing_filename_format) {
fmt = info->backing_filename_format;
}
}
}
g_hash_table_destroy(filenames);
return head;
err:
qapi_free_BlockGraphInfoList(head);
g_hash_table_destroy(filenames);
return NULL;
}
static int img_info(int argc, char **argv)
{
int c;
OutputFormat output_format = OFORMAT_HUMAN;
bool chain = false;
const char *filename, *fmt, *output;
BlockGraphInfoList *list;
bool image_opts = false;
bool force_share = false;
fmt = NULL;
output = NULL;
for(;;) {
int option_index = 0;
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"format", required_argument, 0, 'f'},
{"output", required_argument, 0, OPTION_OUTPUT},
{"backing-chain", no_argument, 0, OPTION_BACKING_CHAIN},
{"object", required_argument, 0, OPTION_OBJECT},
{"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{"force-share", no_argument, 0, 'U'},
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, ":f:hU",
long_options, &option_index);
if (c == -1) {
break;
}
switch(c) {
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
break;
case 'f':
fmt = optarg;
break;
case 'U':
force_share = true;
break;
case OPTION_OUTPUT:
output = optarg;
break;
case OPTION_BACKING_CHAIN:
chain = true;
break;
case OPTION_OBJECT:
user_creatable_process_cmdline(optarg);
break;
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
}
}
if (optind != argc - 1) {
error_exit("Expecting one image file name");
}
filename = argv[optind++];
if (output && !strcmp(output, "json")) {
output_format = OFORMAT_JSON;
} else if (output && !strcmp(output, "human")) {
output_format = OFORMAT_HUMAN;
} else if (output) {
error_report("--output must be used with human or json as argument.");
return 1;
}
list = collect_image_info_list(image_opts, filename, fmt, chain,
force_share);
if (!list) {
return 1;
}
switch (output_format) {
case OFORMAT_HUMAN:
dump_human_image_info_list(list);
break;
case OFORMAT_JSON:
if (chain) {
dump_json_block_graph_info_list(list);
} else {
dump_json_block_graph_info(list->value);
}
break;
}
qapi_free_BlockGraphInfoList(list);
return 0;
}
static int dump_map_entry(OutputFormat output_format, MapEntry *e,
MapEntry *next)
{
switch (output_format) {
case OFORMAT_HUMAN:
if (e->data && !e->has_offset) {
error_report("File contains external, encrypted or compressed clusters.");
return -1;
}
if (e->data && !e->zero) {
printf("%#-16"PRIx64"%#-16"PRIx64"%#-16"PRIx64"%s\n",
e->start, e->length,
e->has_offset ? e->offset : 0,
e->filename ?: "");
}
/* This format ignores the distinction between 0, ZERO and ZERO|DATA.
* Modify the flags here to allow more coalescing.
*/
if (next && (!next->data || next->zero)) {
next->data = false;
next->zero = true;
}
break;
case OFORMAT_JSON:
printf("{ \"start\": %"PRId64", \"length\": %"PRId64","
qemu-img: Make unallocated part of backing chain obvious in map The recently-added NBD context qemu:allocation-depth is able to distinguish between locally-present data (even when that data is sparse) [shown as depth 1 over NBD], and data that could not be found anywhere in the backing chain [shown as depth 0]; and the libnbd project was recently patched to give the human-readable name "absent" to an allocation-depth of 0. But qemu-img map --output=json predates that addition, and has the unfortunate behavior that all portions of the backing chain that resolve without finding a hit in any backing layer report the same depth as the final backing layer. This makes it harder to reconstruct a qcow2 backing chain using just 'qemu-img map' output, especially when using "backing":null to artificially limit a backing chain, because it is impossible to distinguish between a QCOW2_CLUSTER_UNALLOCATED (which defers to a [missing] backing file) and a QCOW2_CLUSTER_ZERO_PLAIN cluster (which would override any backing file), since both types of clusters otherwise show as "data":false,"zero":true" (but note that we can distinguish a QCOW2_CLUSTER_ZERO_ALLOCATED, which would also have an "offset": listing). The task of reconstructing a qcow2 chain was made harder in commit 0da9856851 (nbd: server: Report holes for raw images), because prior to that point, it was possible to abuse NBD's block status command to see which portions of a qcow2 file resulted in BDRV_BLOCK_ALLOCATED (showing up as NBD_STATE_ZERO in isolation) vs. missing from the chain (showing up as NBD_STATE_ZERO|NBD_STATE_HOLE); but now qemu reports more accurate sparseness information over NBD. An obvious solution is to make 'qemu-img map --output=json' add an additional "present":false designation to any cluster lacking an allocation anywhere in the chain, without any change to the "depth" parameter to avoid breaking existing clients. The iotests have several examples where this distinction demonstrates the additional accuracy. Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210701190655.2131223-3-eblake@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> [eblake: fix more iotest fallout] Signed-off-by: Eric Blake <eblake@redhat.com>
2021-07-01 22:06:55 +03:00
" \"depth\": %"PRId64", \"present\": %s, \"zero\": %s,"
" \"data\": %s, \"compressed\": %s",
e->start, e->length, e->depth,
qemu-img: Make unallocated part of backing chain obvious in map The recently-added NBD context qemu:allocation-depth is able to distinguish between locally-present data (even when that data is sparse) [shown as depth 1 over NBD], and data that could not be found anywhere in the backing chain [shown as depth 0]; and the libnbd project was recently patched to give the human-readable name "absent" to an allocation-depth of 0. But qemu-img map --output=json predates that addition, and has the unfortunate behavior that all portions of the backing chain that resolve without finding a hit in any backing layer report the same depth as the final backing layer. This makes it harder to reconstruct a qcow2 backing chain using just 'qemu-img map' output, especially when using "backing":null to artificially limit a backing chain, because it is impossible to distinguish between a QCOW2_CLUSTER_UNALLOCATED (which defers to a [missing] backing file) and a QCOW2_CLUSTER_ZERO_PLAIN cluster (which would override any backing file), since both types of clusters otherwise show as "data":false,"zero":true" (but note that we can distinguish a QCOW2_CLUSTER_ZERO_ALLOCATED, which would also have an "offset": listing). The task of reconstructing a qcow2 chain was made harder in commit 0da9856851 (nbd: server: Report holes for raw images), because prior to that point, it was possible to abuse NBD's block status command to see which portions of a qcow2 file resulted in BDRV_BLOCK_ALLOCATED (showing up as NBD_STATE_ZERO in isolation) vs. missing from the chain (showing up as NBD_STATE_ZERO|NBD_STATE_HOLE); but now qemu reports more accurate sparseness information over NBD. An obvious solution is to make 'qemu-img map --output=json' add an additional "present":false designation to any cluster lacking an allocation anywhere in the chain, without any change to the "depth" parameter to avoid breaking existing clients. The iotests have several examples where this distinction demonstrates the additional accuracy. Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210701190655.2131223-3-eblake@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> [eblake: fix more iotest fallout] Signed-off-by: Eric Blake <eblake@redhat.com>
2021-07-01 22:06:55 +03:00
e->present ? "true" : "false",
e->zero ? "true" : "false",
e->data ? "true" : "false",
e->compressed ? "true" : "false");
if (e->has_offset) {
printf(", \"offset\": %"PRId64"", e->offset);
}
putchar('}');
if (next) {
puts(",");
}
break;
}
return 0;
}
static int get_block_status(BlockDriverState *bs, int64_t offset,
int64_t bytes, MapEntry *e)
{
block: Convert bdrv_get_block_status() to bytes We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access. Changing the name of the function from bdrv_get_block_status() to bdrv_block_status() ensures that the compiler enforces that all callers are updated. For now, the io.c layer still assert()s that all callers are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers. There was an inherent limitation in returning the offset via the return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which means an offset can only be mapped for sector-aligned queries (or, if we declare that non-aligned input is at the same relative position modulo 512 of the answer), so the new interface also changes things to return the offset via output through a parameter by reference rather than mashed into the return value. We'll have some glue code that munges between the two styles until we finish converting all uses. For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), coupled with the tweak in calling convention. But some code, particularly bdrv_is_allocated(), gets a lot simpler because it no longer has to mess with sectors. For ease of review, bdrv_get_block_status_above() will be tackled separately. Signed-off-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 06:47:03 +03:00
int ret;
int depth;
BlockDriverState *file;
bool has_offset;
block: Convert bdrv_get_block_status() to bytes We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access. Changing the name of the function from bdrv_get_block_status() to bdrv_block_status() ensures that the compiler enforces that all callers are updated. For now, the io.c layer still assert()s that all callers are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers. There was an inherent limitation in returning the offset via the return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which means an offset can only be mapped for sector-aligned queries (or, if we declare that non-aligned input is at the same relative position modulo 512 of the answer), so the new interface also changes things to return the offset via output through a parameter by reference rather than mashed into the return value. We'll have some glue code that munges between the two styles until we finish converting all uses. For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), coupled with the tweak in calling convention. But some code, particularly bdrv_is_allocated(), gets a lot simpler because it no longer has to mess with sectors. For ease of review, bdrv_get_block_status_above() will be tackled separately. Signed-off-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 06:47:03 +03:00
int64_t map;
char *filename = NULL;
/* As an optimization, we could cache the current range of unallocated
* clusters in each file of the chain, and avoid querying the same
* range repeatedly.
*/
depth = 0;
for (;;) {
bs = bdrv_skip_filters(bs);
block: Convert bdrv_get_block_status() to bytes We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access. Changing the name of the function from bdrv_get_block_status() to bdrv_block_status() ensures that the compiler enforces that all callers are updated. For now, the io.c layer still assert()s that all callers are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers. There was an inherent limitation in returning the offset via the return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which means an offset can only be mapped for sector-aligned queries (or, if we declare that non-aligned input is at the same relative position modulo 512 of the answer), so the new interface also changes things to return the offset via output through a parameter by reference rather than mashed into the return value. We'll have some glue code that munges between the two styles until we finish converting all uses. For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), coupled with the tweak in calling convention. But some code, particularly bdrv_is_allocated(), gets a lot simpler because it no longer has to mess with sectors. For ease of review, bdrv_get_block_status_above() will be tackled separately. Signed-off-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 06:47:03 +03:00
ret = bdrv_block_status(bs, offset, bytes, &bytes, &map, &file);
if (ret < 0) {
return ret;
}
block: Convert bdrv_get_block_status() to bytes We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access. Changing the name of the function from bdrv_get_block_status() to bdrv_block_status() ensures that the compiler enforces that all callers are updated. For now, the io.c layer still assert()s that all callers are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers. There was an inherent limitation in returning the offset via the return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which means an offset can only be mapped for sector-aligned queries (or, if we declare that non-aligned input is at the same relative position modulo 512 of the answer), so the new interface also changes things to return the offset via output through a parameter by reference rather than mashed into the return value. We'll have some glue code that munges between the two styles until we finish converting all uses. For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), coupled with the tweak in calling convention. But some code, particularly bdrv_is_allocated(), gets a lot simpler because it no longer has to mess with sectors. For ease of review, bdrv_get_block_status_above() will be tackled separately. Signed-off-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 06:47:03 +03:00
assert(bytes);
if (ret & (BDRV_BLOCK_ZERO|BDRV_BLOCK_DATA)) {
break;
}
bs = bdrv_cow_bs(bs);
if (bs == NULL) {
ret = 0;
break;
}
depth++;
}
has_offset = !!(ret & BDRV_BLOCK_OFFSET_VALID);
if (file && has_offset) {
bdrv_graph_rdlock_main_loop();
bdrv_refresh_filename(file);
bdrv_graph_rdunlock_main_loop();
filename = file->filename;
}
*e = (MapEntry) {
.start = offset,
block: Convert bdrv_get_block_status() to bytes We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access. Changing the name of the function from bdrv_get_block_status() to bdrv_block_status() ensures that the compiler enforces that all callers are updated. For now, the io.c layer still assert()s that all callers are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers. There was an inherent limitation in returning the offset via the return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which means an offset can only be mapped for sector-aligned queries (or, if we declare that non-aligned input is at the same relative position modulo 512 of the answer), so the new interface also changes things to return the offset via output through a parameter by reference rather than mashed into the return value. We'll have some glue code that munges between the two styles until we finish converting all uses. For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), coupled with the tweak in calling convention. But some code, particularly bdrv_is_allocated(), gets a lot simpler because it no longer has to mess with sectors. For ease of review, bdrv_get_block_status_above() will be tackled separately. Signed-off-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 06:47:03 +03:00
.length = bytes,
.data = !!(ret & BDRV_BLOCK_DATA),
.zero = !!(ret & BDRV_BLOCK_ZERO),
.compressed = !!(ret & BDRV_BLOCK_COMPRESSED),
block: Convert bdrv_get_block_status() to bytes We are gradually moving away from sector-based interfaces, towards byte-based. In the common case, allocation is unlikely to ever use values that are not naturally sector-aligned, but it is possible that byte-based values will let us be more precise about allocation at the end of an unaligned file that can do byte-based access. Changing the name of the function from bdrv_get_block_status() to bdrv_block_status() ensures that the compiler enforces that all callers are updated. For now, the io.c layer still assert()s that all callers are sector-aligned, but that can be relaxed when a later patch implements byte-based block status in the drivers. There was an inherent limitation in returning the offset via the return value: we only have room for BDRV_BLOCK_OFFSET_MASK bits, which means an offset can only be mapped for sector-aligned queries (or, if we declare that non-aligned input is at the same relative position modulo 512 of the answer), so the new interface also changes things to return the offset via output through a parameter by reference rather than mashed into the return value. We'll have some glue code that munges between the two styles until we finish converting all uses. For the most part this patch is just the addition of scaling at the callers followed by inverse scaling at bdrv_block_status(), coupled with the tweak in calling convention. But some code, particularly bdrv_is_allocated(), gets a lot simpler because it no longer has to mess with sectors. For ease of review, bdrv_get_block_status_above() will be tackled separately. Signed-off-by: Eric Blake <eblake@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2017-10-12 06:47:03 +03:00
.offset = map,
.has_offset = has_offset,
.depth = depth,
qemu-img: Make unallocated part of backing chain obvious in map The recently-added NBD context qemu:allocation-depth is able to distinguish between locally-present data (even when that data is sparse) [shown as depth 1 over NBD], and data that could not be found anywhere in the backing chain [shown as depth 0]; and the libnbd project was recently patched to give the human-readable name "absent" to an allocation-depth of 0. But qemu-img map --output=json predates that addition, and has the unfortunate behavior that all portions of the backing chain that resolve without finding a hit in any backing layer report the same depth as the final backing layer. This makes it harder to reconstruct a qcow2 backing chain using just 'qemu-img map' output, especially when using "backing":null to artificially limit a backing chain, because it is impossible to distinguish between a QCOW2_CLUSTER_UNALLOCATED (which defers to a [missing] backing file) and a QCOW2_CLUSTER_ZERO_PLAIN cluster (which would override any backing file), since both types of clusters otherwise show as "data":false,"zero":true" (but note that we can distinguish a QCOW2_CLUSTER_ZERO_ALLOCATED, which would also have an "offset": listing). The task of reconstructing a qcow2 chain was made harder in commit 0da9856851 (nbd: server: Report holes for raw images), because prior to that point, it was possible to abuse NBD's block status command to see which portions of a qcow2 file resulted in BDRV_BLOCK_ALLOCATED (showing up as NBD_STATE_ZERO in isolation) vs. missing from the chain (showing up as NBD_STATE_ZERO|NBD_STATE_HOLE); but now qemu reports more accurate sparseness information over NBD. An obvious solution is to make 'qemu-img map --output=json' add an additional "present":false designation to any cluster lacking an allocation anywhere in the chain, without any change to the "depth" parameter to avoid breaking existing clients. The iotests have several examples where this distinction demonstrates the additional accuracy. Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210701190655.2131223-3-eblake@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> [eblake: fix more iotest fallout] Signed-off-by: Eric Blake <eblake@redhat.com>
2021-07-01 22:06:55 +03:00
.present = !!(ret & BDRV_BLOCK_ALLOCATED),
.filename = filename,
};
return 0;
}
static inline bool entry_mergeable(const MapEntry *curr, const MapEntry *next)
{
if (curr->length == 0) {
return false;
}
if (curr->zero != next->zero ||
curr->data != next->data ||
curr->compressed != next->compressed ||
curr->depth != next->depth ||
qemu-img: Make unallocated part of backing chain obvious in map The recently-added NBD context qemu:allocation-depth is able to distinguish between locally-present data (even when that data is sparse) [shown as depth 1 over NBD], and data that could not be found anywhere in the backing chain [shown as depth 0]; and the libnbd project was recently patched to give the human-readable name "absent" to an allocation-depth of 0. But qemu-img map --output=json predates that addition, and has the unfortunate behavior that all portions of the backing chain that resolve without finding a hit in any backing layer report the same depth as the final backing layer. This makes it harder to reconstruct a qcow2 backing chain using just 'qemu-img map' output, especially when using "backing":null to artificially limit a backing chain, because it is impossible to distinguish between a QCOW2_CLUSTER_UNALLOCATED (which defers to a [missing] backing file) and a QCOW2_CLUSTER_ZERO_PLAIN cluster (which would override any backing file), since both types of clusters otherwise show as "data":false,"zero":true" (but note that we can distinguish a QCOW2_CLUSTER_ZERO_ALLOCATED, which would also have an "offset": listing). The task of reconstructing a qcow2 chain was made harder in commit 0da9856851 (nbd: server: Report holes for raw images), because prior to that point, it was possible to abuse NBD's block status command to see which portions of a qcow2 file resulted in BDRV_BLOCK_ALLOCATED (showing up as NBD_STATE_ZERO in isolation) vs. missing from the chain (showing up as NBD_STATE_ZERO|NBD_STATE_HOLE); but now qemu reports more accurate sparseness information over NBD. An obvious solution is to make 'qemu-img map --output=json' add an additional "present":false designation to any cluster lacking an allocation anywhere in the chain, without any change to the "depth" parameter to avoid breaking existing clients. The iotests have several examples where this distinction demonstrates the additional accuracy. Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20210701190655.2131223-3-eblake@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com> [eblake: fix more iotest fallout] Signed-off-by: Eric Blake <eblake@redhat.com>
2021-07-01 22:06:55 +03:00
curr->present != next->present ||
!curr->filename != !next->filename ||
curr->has_offset != next->has_offset) {
return false;
}
if (curr->filename && strcmp(curr->filename, next->filename)) {
return false;
}
if (curr->has_offset && curr->offset + curr->length != next->offset) {
return false;
}
return true;
}
static int img_map(int argc, char **argv)
{
int c;
OutputFormat output_format = OFORMAT_HUMAN;
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
BlockBackend *blk;
BlockDriverState *bs;
const char *filename, *fmt, *output;
int64_t length;
MapEntry curr = { .length = 0 }, next;
int ret = 0;
bool image_opts = false;
bool force_share = false;
int64_t start_offset = 0;
int64_t max_length = -1;
fmt = NULL;
output = NULL;
for (;;) {
int option_index = 0;
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"format", required_argument, 0, 'f'},
{"output", required_argument, 0, OPTION_OUTPUT},
{"object", required_argument, 0, OPTION_OBJECT},
{"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{"force-share", no_argument, 0, 'U'},
{"start-offset", required_argument, 0, 's'},
{"max-length", required_argument, 0, 'l'},
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, ":f:s:l:hU",
long_options, &option_index);
if (c == -1) {
break;
}
switch (c) {
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
break;
case 'f':
fmt = optarg;
break;
case 'U':
force_share = true;
break;
case OPTION_OUTPUT:
output = optarg;
break;
case 's':
start_offset = cvtnum("start offset", optarg);
if (start_offset < 0) {
return 1;
}
break;
case 'l':
max_length = cvtnum("max length", optarg);
if (max_length < 0) {
return 1;
}
break;
case OPTION_OBJECT:
user_creatable_process_cmdline(optarg);
break;
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
}
}
if (optind != argc - 1) {
error_exit("Expecting one image file name");
}
filename = argv[optind];
if (output && !strcmp(output, "json")) {
output_format = OFORMAT_JSON;
} else if (output && !strcmp(output, "human")) {
output_format = OFORMAT_HUMAN;
} else if (output) {
error_report("--output must be used with human or json as argument.");
return 1;
}
blk = img_open(image_opts, filename, fmt, 0, false, false, force_share);
if (!blk) {
return 1;
}
bs = blk_bs(blk);
if (output_format == OFORMAT_HUMAN) {
printf("%-16s%-16s%-16s%s\n", "Offset", "Length", "Mapped to", "File");
} else if (output_format == OFORMAT_JSON) {
putchar('[');
}
length = blk_getlength(blk);
if (length < 0) {
error_report("Failed to get size for '%s'", filename);
return 1;
}
if (max_length != -1) {
length = MIN(start_offset + max_length, length);
}
curr.start = start_offset;
while (curr.start + curr.length < length) {
int64_t offset = curr.start + curr.length;
int64_t n = length - offset;
ret = get_block_status(bs, offset, n, &next);
if (ret < 0) {
error_report("Could not read file metadata: %s", strerror(-ret));
goto out;
}
if (entry_mergeable(&curr, &next)) {
curr.length += next.length;
continue;
}
if (curr.length > 0) {
ret = dump_map_entry(output_format, &curr, &next);
if (ret < 0) {
goto out;
}
}
curr = next;
}
ret = dump_map_entry(output_format, &curr, NULL);
if (output_format == OFORMAT_JSON) {
puts("]");
}
out:
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
blk_unref(blk);
return ret < 0;
}
#define SNAPSHOT_LIST 1
#define SNAPSHOT_CREATE 2
#define SNAPSHOT_APPLY 3
#define SNAPSHOT_DELETE 4
static int img_snapshot(int argc, char **argv)
{
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
BlockBackend *blk;
BlockDriverState *bs;
QEMUSnapshotInfo sn;
char *filename, *snapshot_name = NULL;
int c, ret = 0, bdrv_oflags;
int action = 0;
bool quiet = false;
Error *err = NULL;
bool image_opts = false;
bool force_share = false;
int64_t rt;
bdrv_oflags = BDRV_O_RDWR;
/* Parse commandline parameters */
for(;;) {
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"object", required_argument, 0, OPTION_OBJECT},
{"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{"force-share", no_argument, 0, 'U'},
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, ":la:c:d:hqU",
long_options, NULL);
if (c == -1) {
break;
}
switch(c) {
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
return 0;
case 'l':
if (action) {
error_exit("Cannot mix '-l', '-a', '-c', '-d'");
return 0;
}
action = SNAPSHOT_LIST;
bdrv_oflags &= ~BDRV_O_RDWR; /* no need for RW */
break;
case 'a':
if (action) {
error_exit("Cannot mix '-l', '-a', '-c', '-d'");
return 0;
}
action = SNAPSHOT_APPLY;
snapshot_name = optarg;
break;
case 'c':
if (action) {
error_exit("Cannot mix '-l', '-a', '-c', '-d'");
return 0;
}
action = SNAPSHOT_CREATE;
snapshot_name = optarg;
break;
case 'd':
if (action) {
error_exit("Cannot mix '-l', '-a', '-c', '-d'");
return 0;
}
action = SNAPSHOT_DELETE;
snapshot_name = optarg;
break;
case 'q':
quiet = true;
break;
case 'U':
force_share = true;
break;
case OPTION_OBJECT:
user_creatable_process_cmdline(optarg);
break;
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
}
}
if (optind != argc - 1) {
error_exit("Expecting one image file name");
}
filename = argv[optind++];
/* Open the image */
blk = img_open(image_opts, filename, NULL, bdrv_oflags, false, quiet,
force_share);
if (!blk) {
return 1;
}
bs = blk_bs(blk);
/* Perform the requested action */
switch(action) {
case SNAPSHOT_LIST:
dump_snapshots(bs);
break;
case SNAPSHOT_CREATE:
memset(&sn, 0, sizeof(sn));
pstrcpy(sn.name, sizeof(sn.name), snapshot_name);
rt = g_get_real_time();
sn.date_sec = rt / G_USEC_PER_SEC;
sn.date_nsec = (rt % G_USEC_PER_SEC) * 1000;
bdrv_graph_rdlock_main_loop();
ret = bdrv_snapshot_create(bs, &sn);
bdrv_graph_rdunlock_main_loop();
if (ret) {
error_report("Could not create snapshot '%s': %s",
snapshot_name, strerror(-ret));
}
break;
case SNAPSHOT_APPLY:
ret = bdrv_snapshot_goto(bs, snapshot_name, &err);
if (ret) {
error_reportf_err(err, "Could not apply snapshot '%s': ",
snapshot_name);
}
break;
case SNAPSHOT_DELETE:
bdrv_graph_rdlock_main_loop();
ret = bdrv_snapshot_find(bs, &sn, snapshot_name);
if (ret < 0) {
error_report("Could not delete snapshot '%s': snapshot not "
"found", snapshot_name);
ret = 1;
} else {
ret = bdrv_snapshot_delete(bs, sn.id_str, sn.name, &err);
if (ret < 0) {
error_reportf_err(err, "Could not delete snapshot '%s': ",
snapshot_name);
ret = 1;
}
}
bdrv_graph_rdunlock_main_loop();
break;
}
/* Cleanup */
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
blk_unref(blk);
if (ret) {
return 1;
}
return 0;
}
static int img_rebase(int argc, char **argv)
{
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
BlockBackend *blk = NULL, *blk_old_backing = NULL, *blk_new_backing = NULL;
uint8_t *buf_old = NULL;
uint8_t *buf_new = NULL;
BlockDriverState *bs = NULL, *prefix_chain_bs = NULL;
BlockDriverState *unfiltered_bs;
2023-09-19 19:58:01 +03:00
BlockDriverInfo bdi = {0};
char *filename;
const char *fmt, *cache, *src_cache, *out_basefmt, *out_baseimg;
int c, flags, src_flags, ret;
BdrvRequestFlags write_flags = 0;
bool writethrough, src_writethrough;
int unsafe = 0;
bool force_share = false;
int progress = 0;
bool quiet = false;
bool compress = false;
Error *local_err = NULL;
bool image_opts = false;
2023-09-19 19:58:01 +03:00
int64_t write_align;
/* Parse commandline parameters */
fmt = NULL;
cache = BDRV_DEFAULT_CACHE;
src_cache = BDRV_DEFAULT_CACHE;
out_baseimg = NULL;
out_basefmt = NULL;
for(;;) {
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"object", required_argument, 0, OPTION_OBJECT},
{"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{"force-share", no_argument, 0, 'U'},
{"compress", no_argument, 0, 'c'},
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, ":hf:F:b:upt:T:qUc",
long_options, NULL);
if (c == -1) {
break;
}
switch(c) {
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
return 0;
case 'f':
fmt = optarg;
break;
case 'F':
out_basefmt = optarg;
break;
case 'b':
out_baseimg = optarg;
break;
case 'u':
unsafe = 1;
break;
case 'p':
progress = 1;
break;
case 't':
cache = optarg;
break;
case 'T':
src_cache = optarg;
break;
case 'q':
quiet = true;
break;
case OPTION_OBJECT:
user_creatable_process_cmdline(optarg);
break;
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
case 'U':
force_share = true;
break;
case 'c':
compress = true;
break;
}
}
if (quiet) {
progress = 0;
}
if (optind != argc - 1) {
error_exit("Expecting one image file name");
}
if (!unsafe && !out_baseimg) {
error_exit("Must specify backing file (-b) or use unsafe mode (-u)");
}
filename = argv[optind++];
qemu_progress_init(progress, 2.0);
qemu_progress_print(0, 100);
flags = BDRV_O_RDWR | (unsafe ? BDRV_O_NO_BACKING : 0);
ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
if (ret < 0) {
error_report("Invalid cache option: %s", cache);
goto out;
}
src_flags = 0;
ret = bdrv_parse_cache_mode(src_cache, &src_flags, &src_writethrough);
if (ret < 0) {
error_report("Invalid source cache option: %s", src_cache);
goto out;
}
/* The source files are opened read-only, don't care about WCE */
assert((src_flags & BDRV_O_RDWR) == 0);
(void) src_writethrough;
/*
* Open the images.
*
* Ignore the old backing file for unsafe rebase in case we want to correct
* the reference to a renamed or moved backing file.
*/
blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
false);
if (!blk) {
ret = -1;
goto out;
}
bs = blk_bs(blk);
unfiltered_bs = bdrv_skip_filters(bs);
if (compress && !block_driver_can_compress(unfiltered_bs->drv)) {
error_report("Compression not supported for this file format");
ret = -1;
goto out;
} else if (compress) {
write_flags |= BDRV_REQ_WRITE_COMPRESSED;
}
if (out_basefmt != NULL) {
if (bdrv_find_format(out_basefmt) == NULL) {
error_report("Invalid format name: '%s'", out_basefmt);
ret = -1;
goto out;
}
}
2023-09-19 19:58:01 +03:00
/*
* We need overlay subcluster size (or cluster size in case writes are
* compressed) to make sure write requests are aligned.
2023-09-19 19:58:01 +03:00
*/
ret = bdrv_get_info(unfiltered_bs, &bdi);
if (ret < 0) {
error_report("could not get block driver info");
goto out;
} else if (bdi.subcluster_size == 0) {
bdi.cluster_size = bdi.subcluster_size = 1;
2023-09-19 19:58:01 +03:00
}
write_align = compress ? bdi.cluster_size : bdi.subcluster_size;
2023-09-19 19:58:01 +03:00
/* For safe rebasing we need to compare old and new backing file */
if (!unsafe) {
QDict *options = NULL;
BlockDriverState *base_bs = bdrv_cow_bs(unfiltered_bs);
if (base_bs) {
blk_old_backing = blk_new(qemu_get_aio_context(),
BLK_PERM_CONSISTENT_READ,
BLK_PERM_ALL);
ret = blk_insert_bs(blk_old_backing, base_bs,
&local_err);
if (ret < 0) {
error_reportf_err(local_err,
"Could not reuse old backing file '%s': ",
base_bs->filename);
goto out;
}
} else {
blk_old_backing = NULL;
}
if (out_baseimg[0]) {
const char *overlay_filename;
char *out_real_path;
options = qdict_new();
if (out_basefmt) {
qdict_put_str(options, "driver", out_basefmt);
}
if (force_share) {
qdict_put_bool(options, BDRV_OPT_FORCE_SHARE, true);
}
bdrv_graph_rdlock_main_loop();
bdrv_refresh_filename(bs);
bdrv_graph_rdunlock_main_loop();
overlay_filename = bs->exact_filename[0] ? bs->exact_filename
: bs->filename;
out_real_path =
bdrv_get_full_backing_filename_from_filename(overlay_filename,
out_baseimg,
&local_err);
if (local_err) {
qobject_unref(options);
error_reportf_err(local_err,
"Could not resolve backing filename: ");
ret = -1;
goto out;
}
/*
* Find out whether we rebase an image on top of a previous image
* in its chain.
*/
prefix_chain_bs = bdrv_find_backing_image(bs, out_real_path);
if (prefix_chain_bs) {
qobject_unref(options);
g_free(out_real_path);
blk_new_backing = blk_new(qemu_get_aio_context(),
BLK_PERM_CONSISTENT_READ,
BLK_PERM_ALL);
ret = blk_insert_bs(blk_new_backing, prefix_chain_bs,
&local_err);
if (ret < 0) {
error_reportf_err(local_err,
"Could not reuse backing file '%s': ",
out_baseimg);
goto out;
}
} else {
blk_new_backing = blk_new_open(out_real_path, NULL,
options, src_flags, &local_err);
g_free(out_real_path);
if (!blk_new_backing) {
error_reportf_err(local_err,
"Could not open new backing file '%s': ",
out_baseimg);
ret = -1;
goto out;
}
}
}
}
/*
* Check each unallocated cluster in the COW file. If it is unallocated,
* accesses go to the backing file. We must therefore compare this cluster
* in the old and new backing file, and if they differ we need to copy it
* from the old backing file into the COW file.
*
* If qemu-img crashes during this step, no harm is done. The content of
* the image is the same as the original one at any time.
*/
if (!unsafe) {
int64_t size;
int64_t old_backing_size = 0;
int64_t new_backing_size = 0;
uint64_t offset;
2023-09-19 19:58:01 +03:00
int64_t n, n_old = 0, n_new = 0;
float local_progress = 0;
if (blk_old_backing && bdrv_opt_mem_align(blk_bs(blk_old_backing)) >
bdrv_opt_mem_align(blk_bs(blk))) {
buf_old = blk_blockalign(blk_old_backing, IO_BUF_SIZE);
} else {
buf_old = blk_blockalign(blk, IO_BUF_SIZE);
}
buf_new = blk_blockalign(blk_new_backing, IO_BUF_SIZE);
size = blk_getlength(blk);
if (size < 0) {
error_report("Could not get size of '%s': %s",
filename, strerror(-size));
ret = -1;
goto out;
}
if (blk_old_backing) {
old_backing_size = blk_getlength(blk_old_backing);
if (old_backing_size < 0) {
char backing_name[PATH_MAX];
bdrv_get_backing_filename(bs, backing_name,
sizeof(backing_name));
error_report("Could not get size of '%s': %s",
backing_name, strerror(-old_backing_size));
ret = -1;
goto out;
}
}
if (blk_new_backing) {
new_backing_size = blk_getlength(blk_new_backing);
if (new_backing_size < 0) {
error_report("Could not get size of '%s': %s",
out_baseimg, strerror(-new_backing_size));
ret = -1;
goto out;
}
}
if (size != 0) {
local_progress = (float)100 / (size / MIN(size, IO_BUF_SIZE));
}
for (offset = 0; offset < size; offset += n) {
2023-09-19 19:58:01 +03:00
bool old_backing_eof = false;
int64_t n_alloc;
/* How many bytes can we handle with the next read? */
n = MIN(IO_BUF_SIZE, size - offset);
/* If the cluster is allocated, we don't need to take action */
ret = bdrv_is_allocated(unfiltered_bs, offset, n, &n);
if (ret < 0) {
error_report("error while reading image metadata: %s",
strerror(-ret));
goto out;
}
if (ret) {
continue;
}
if (prefix_chain_bs) {
uint64_t bytes = n;
/*
* If cluster wasn't changed since prefix_chain, we don't need
* to take action
*/
ret = bdrv_is_allocated_above(bdrv_cow_bs(unfiltered_bs),
prefix_chain_bs, false,
offset, n, &n);
if (ret < 0) {
error_report("error while reading image metadata: %s",
strerror(-ret));
goto out;
}
if (!ret && n) {
continue;
}
if (!n) {
/*
* If we've reached EOF of the old backing, it means that
* offsets beyond the old backing size were read as zeroes.
* Now we will need to explicitly zero the cluster in
* order to preserve that state after the rebase.
*/
n = bytes;
}
}
2023-09-19 19:58:01 +03:00
/*
* At this point we know that the region [offset; offset + n)
* is unallocated within the target image. This region might be
* unaligned to the target image's (sub)cluster boundaries, as
* old backing may have smaller clusters (or have subclusters).
* We extend it to the aligned boundaries to avoid CoW on
* partial writes in blk_pwrite(),
*/
n += offset - QEMU_ALIGN_DOWN(offset, write_align);
offset = QEMU_ALIGN_DOWN(offset, write_align);
n += QEMU_ALIGN_UP(offset + n, write_align) - (offset + n);
n = MIN(n, size - offset);
assert(!bdrv_is_allocated(unfiltered_bs, offset, n, &n_alloc) &&
n_alloc == n);
/*
* Much like with the target image, we'll try to read as much
* of the old and new backings as we can.
*/
n_old = MIN(n, MAX(0, old_backing_size - (int64_t) offset));
n_new = MIN(n, MAX(0, new_backing_size - (int64_t) offset));
/*
* Read old and new backing file and take into consideration that
* backing files may be smaller than the COW image.
*/
2023-09-19 19:58:01 +03:00
memset(buf_old + n_old, 0, n - n_old);
if (!n_old) {
old_backing_eof = true;
} else {
2023-09-19 19:58:01 +03:00
ret = blk_pread(blk_old_backing, offset, n_old, buf_old, 0);
if (ret < 0) {
error_report("error while reading from old backing file");
goto out;
}
}
2023-09-19 19:58:01 +03:00
memset(buf_new + n_new, 0, n - n_new);
if (n_new) {
ret = blk_pread(blk_new_backing, offset, n_new, buf_new, 0);
if (ret < 0) {
error_report("error while reading from new backing file");
goto out;
}
}
/* If they differ, we need to write to the COW file */
uint64_t written = 0;
while (written < n) {
int64_t pnum;
if (compare_buffers(buf_old + written, buf_new + written,
2023-09-19 19:58:01 +03:00
n - written, write_align, &pnum))
{
2023-09-19 19:58:01 +03:00
if (old_backing_eof) {
ret = blk_pwrite_zeroes(blk, offset + written, pnum, 0);
} else {
2023-09-19 19:58:01 +03:00
assert(written + pnum <= IO_BUF_SIZE);
ret = blk_pwrite(blk, offset + written, pnum,
buf_old + written, write_flags);
}
if (ret < 0) {
error_report("Error while writing to COW image: %s",
strerror(-ret));
goto out;
}
}
written += pnum;
2023-09-19 19:58:01 +03:00
if (offset + written >= old_backing_size) {
old_backing_eof = true;
}
}
qemu_progress_print(local_progress, 100);
}
}
/*
* Change the backing file. All clusters that are different from the old
* backing file are overwritten in the COW file now, so the visible content
* doesn't change when we switch the backing file.
*/
if (out_baseimg && *out_baseimg) {
ret = bdrv_change_backing_file(unfiltered_bs, out_baseimg, out_basefmt,
true);
} else {
ret = bdrv_change_backing_file(unfiltered_bs, NULL, NULL, false);
}
if (ret == -ENOSPC) {
error_report("Could not change the backing file to '%s': No "
"space left in the file header", out_baseimg);
} else if (ret == -EINVAL && out_baseimg && !out_basefmt) {
error_report("Could not change the backing file to '%s': backing "
"format must be specified", out_baseimg);
} else if (ret < 0) {
error_report("Could not change the backing file to '%s': %s",
out_baseimg, strerror(-ret));
}
qemu_progress_print(100, 0);
/*
* TODO At this point it is possible to check if any clusters that are
* allocated in the COW file are the same in the backing file. If so, they
* could be dropped from the COW file. Don't do this before switching the
* backing file, in case of a crash this would lead to corruption.
*/
out:
qemu_progress_end();
/* Cleanup */
if (!unsafe) {
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
blk_unref(blk_old_backing);
blk_unref(blk_new_backing);
}
qemu_vfree(buf_old);
qemu_vfree(buf_new);
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
blk_unref(blk);
if (ret) {
return 1;
}
return 0;
}
static int img_resize(int argc, char **argv)
{
Error *err = NULL;
int c, ret, relative;
const char *filename, *fmt, *size;
int64_t n, total_size, current_size;
bool quiet = false;
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
BlockBackend *blk = NULL;
PreallocMode prealloc = PREALLOC_MODE_OFF;
QemuOpts *param;
static QemuOptsList resize_options = {
.name = "resize_options",
.head = QTAILQ_HEAD_INITIALIZER(resize_options.head),
.desc = {
{
.name = BLOCK_OPT_SIZE,
.type = QEMU_OPT_SIZE,
.help = "Virtual disk size"
}, {
/* end of list */
}
},
};
bool image_opts = false;
bool shrink = false;
/* Remove size from argv manually so that negative numbers are not treated
* as options by getopt. */
if (argc < 3) {
error_exit("Not enough arguments");
return 1;
}
size = argv[--argc];
/* Parse getopt arguments */
fmt = NULL;
for(;;) {
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"object", required_argument, 0, OPTION_OBJECT},
{"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{"preallocation", required_argument, 0, OPTION_PREALLOCATION},
{"shrink", no_argument, 0, OPTION_SHRINK},
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, ":f:hq",
long_options, NULL);
if (c == -1) {
break;
}
switch(c) {
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
break;
case 'f':
fmt = optarg;
break;
case 'q':
quiet = true;
break;
case OPTION_OBJECT:
user_creatable_process_cmdline(optarg);
break;
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
case OPTION_PREALLOCATION:
prealloc = qapi_enum_parse(&PreallocMode_lookup, optarg,
PREALLOC_MODE__MAX, NULL);
if (prealloc == PREALLOC_MODE__MAX) {
error_report("Invalid preallocation mode '%s'", optarg);
return 1;
}
break;
case OPTION_SHRINK:
shrink = true;
break;
}
}
if (optind != argc - 1) {
error_exit("Expecting image file name and size");
}
filename = argv[optind++];
/* Choose grow, shrink, or absolute resize mode */
switch (size[0]) {
case '+':
relative = 1;
size++;
break;
case '-':
relative = -1;
size++;
break;
default:
relative = 0;
break;
}
/* Parse size */
param = qemu_opts_create(&resize_options, NULL, 0, &error_abort);
if (!qemu_opt_set(param, BLOCK_OPT_SIZE, size, &err)) {
error_report_err(err);
ret = -1;
qemu_opts_del(param);
goto out;
}
n = qemu_opt_get_size(param, BLOCK_OPT_SIZE, 0);
qemu_opts_del(param);
blk = img_open(image_opts, filename, fmt,
BDRV_O_RDWR | BDRV_O_RESIZE, false, quiet,
false);
if (!blk) {
ret = -1;
goto out;
}
current_size = blk_getlength(blk);
if (current_size < 0) {
error_report("Failed to inquire current image length: %s",
strerror(-current_size));
ret = -1;
goto out;
}
if (relative) {
total_size = current_size + n * relative;
} else {
total_size = n;
}
if (total_size <= 0) {
error_report("New image size must be positive");
ret = -1;
goto out;
}
if (total_size <= current_size && prealloc != PREALLOC_MODE_OFF) {
error_report("Preallocation can only be used for growing images");
ret = -1;
goto out;
}
if (total_size < current_size && !shrink) {
error_report("Use the --shrink option to perform a shrink operation.");
warn_report("Shrinking an image will delete all data beyond the "
"shrunken image's end. Before performing such an "
"operation, make sure there is no important data there.");
ret = -1;
goto out;
}
/*
* The user expects the image to have the desired size after
* resizing, so pass @exact=true. It is of no use to report
* success when the image has not actually been resized.
*/
ret = blk_truncate(blk, total_size, true, prealloc, 0, &err);
if (!ret) {
qprintf(quiet, "Image resized.\n");
} else {
error_report_err(err);
}
out:
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
blk_unref(blk);
if (ret) {
return 1;
}
return 0;
}
static void amend_status_cb(BlockDriverState *bs,
int64_t offset, int64_t total_work_size,
void *opaque)
{
qemu_progress_print(100.f * offset / total_work_size, 0);
}
static int print_amend_option_help(const char *format)
{
BlockDriver *drv;
GRAPH_RDLOCK_GUARD_MAINLOOP();
/* Find driver and parse its options */
drv = bdrv_find_format(format);
if (!drv) {
error_report("Unknown file format '%s'", format);
return 1;
}
if (!drv->bdrv_amend_options) {
error_report("Format driver '%s' does not support option amendment",
format);
return 1;
}
/* Every driver supporting amendment must have amend_opts */
assert(drv->amend_opts);
printf("Amend options for '%s':\n", format);
qemu_opts_print_help(drv->amend_opts, false);
return 0;
}
static int img_amend(int argc, char **argv)
{
Error *err = NULL;
int c, ret = 0;
char *options = NULL;
QemuOptsList *amend_opts = NULL;
QemuOpts *opts = NULL;
const char *fmt = NULL, *filename, *cache;
int flags;
bool writethrough;
bool quiet = false, progress = false;
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
BlockBackend *blk = NULL;
BlockDriverState *bs = NULL;
bool image_opts = false;
bool force = false;
cache = BDRV_DEFAULT_CACHE;
for (;;) {
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"object", required_argument, 0, OPTION_OBJECT},
{"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{"force", no_argument, 0, OPTION_FORCE},
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, ":ho:f:t:pq",
long_options, NULL);
if (c == -1) {
break;
}
switch (c) {
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
break;
case 'o':
if (accumulate_options(&options, optarg) < 0) {
ret = -1;
goto out_no_progress;
}
break;
case 'f':
fmt = optarg;
break;
case 't':
cache = optarg;
break;
case 'p':
progress = true;
break;
case 'q':
quiet = true;
break;
case OPTION_OBJECT:
user_creatable_process_cmdline(optarg);
break;
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
case OPTION_FORCE:
force = true;
break;
}
}
if (!options) {
error_exit("Must specify options (-o)");
}
if (quiet) {
progress = false;
}
qemu_progress_init(progress, 1.0);
filename = (optind == argc - 1) ? argv[argc - 1] : NULL;
if (fmt && has_help_option(options)) {
/* If a format is explicitly specified (and possibly no filename is
* given), print option help here */
ret = print_amend_option_help(fmt);
goto out;
}
if (optind != argc - 1) {
error_report("Expecting one image file name");
ret = -1;
goto out;
}
flags = BDRV_O_RDWR;
ret = bdrv_parse_cache_mode(cache, &flags, &writethrough);
if (ret < 0) {
error_report("Invalid cache option: %s", cache);
goto out;
}
blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
false);
if (!blk) {
ret = -1;
goto out;
}
bs = blk_bs(blk);
fmt = bs->drv->format_name;
if (has_help_option(options)) {
/* If the format was auto-detected, print option help here */
ret = print_amend_option_help(fmt);
goto out;
}
bdrv_graph_rdlock_main_loop();
if (!bs->drv->bdrv_amend_options) {
error_report("Format driver '%s' does not support option amendment",
fmt);
bdrv_graph_rdunlock_main_loop();
ret = -1;
goto out;
}
/* Every driver supporting amendment must have amend_opts */
assert(bs->drv->amend_opts);
amend_opts = qemu_opts_append(amend_opts, bs->drv->amend_opts);
opts = qemu_opts_create(amend_opts, NULL, 0, &error_abort);
if (!qemu_opts_do_parse(opts, options, NULL, &err)) {
/* Try to parse options using the create options */
amend_opts = qemu_opts_append(amend_opts, bs->drv->create_opts);
qemu_opts_del(opts);
opts = qemu_opts_create(amend_opts, NULL, 0, &error_abort);
if (qemu_opts_do_parse(opts, options, NULL, NULL)) {
error_append_hint(&err,
"This option is only supported for image creation\n");
}
bdrv_graph_rdunlock_main_loop();
error_report_err(err);
ret = -1;
goto out;
}
/* In case the driver does not call amend_status_cb() */
qemu_progress_print(0.f, 0);
ret = bdrv_amend_options(bs, opts, &amend_status_cb, NULL, force, &err);
qemu_progress_print(100.f, 0);
bdrv_graph_rdunlock_main_loop();
if (ret < 0) {
error_report_err(err);
goto out;
}
out:
qemu_progress_end();
out_no_progress:
block: New BlockBackend A block device consists of a frontend device model and a backend. A block backend has a tree of block drivers doing the actual work. The tree is managed by the block layer. We currently use a single abstraction BlockDriverState both for tree nodes and the backend as a whole. Drawbacks: * Its API includes both stuff that makes sense only at the block backend level (root of the tree) and stuff that's only for use within the block layer. This makes the API bigger and more complex than necessary. Moreover, it's not obvious which interfaces are meant for device models, and which really aren't. * Since device models keep a reference to their backend, the backend object can't just be destroyed. But for media change, we need to replace the tree. Our solution is to make the BlockDriverState generic, with actual driver state in a separate object, pointed to by member opaque. That lets us replace the tree by deinitializing and reinitializing its root. This special need of the root makes the data structure awkward everywhere in the tree. The general plan is to separate the APIs into "block backend", for use by device models, monitor and whatever other code dealing with block backends, and "block driver", for use by the block layer and whatever other code (if any) dealing with trees and tree nodes. Code dealing with block backends, device models in particular, should become completely oblivious of BlockDriverState. This should let us clean up both APIs, and the tree data structures. This commit is a first step. It creates a minimal "block backend" API: type BlockBackend and functions to create, destroy and find them. BlockBackend objects are created and destroyed exactly when root BlockDriverState objects are created and destroyed. "Root" in the sense of "in bdrv_states". They're not yet used for anything; that'll come shortly. A root BlockDriverState is created with bdrv_new_root(), so where to create a BlockBackend is obvious. Where these roots get destroyed isn't always as obvious. It is obvious in qemu-img.c, qemu-io.c and qemu-nbd.c, and in error paths of blockdev_init(), blk_connect(). That leaves destruction of objects successfully created by blockdev_init() and blk_connect(). blockdev_init() is used only by drive_new() and qmp_blockdev_add(). Objects created by the latter are currently indestructible (see commit 48f364d "blockdev: Refuse to drive_del something added with blockdev-add" and commit 2d246f0 "blockdev: Introduce DriveInfo.enable_auto_del"). Objects created by the former get destroyed by drive_del(). Objects created by blk_connect() get destroyed by blk_disconnect(). BlockBackend is reference-counted. Its reference count never exceeds one so far, but that's going to change. In drive_del(), the BB's reference count is surely one now. The BDS's reference count is greater than one when something else is holding a reference, such as a block job. In this case, the BB is destroyed right away, but the BDS lives on until all extra references get dropped. Signed-off-by: Markus Armbruster <armbru@redhat.com> Reviewed-by: Max Reitz <mreitz@redhat.com> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
2014-10-07 15:59:04 +04:00
blk_unref(blk);
qemu_opts_del(opts);
qemu_opts_free(amend_opts);
g_free(options);
if (ret) {
return 1;
}
return 0;
}
typedef struct BenchData {
BlockBackend *blk;
uint64_t image_size;
bool write;
int bufsize;
int step;
int nrreq;
int n;
int flush_interval;
bool drain_on_flush;
uint8_t *buf;
QEMUIOVector *qiov;
int in_flight;
bool in_flush;
uint64_t offset;
} BenchData;
static void bench_undrained_flush_cb(void *opaque, int ret)
{
if (ret < 0) {
error_report("Failed flush request: %s", strerror(-ret));
exit(EXIT_FAILURE);
}
}
static void bench_cb(void *opaque, int ret)
{
BenchData *b = opaque;
BlockAIOCB *acb;
if (ret < 0) {
error_report("Failed request: %s", strerror(-ret));
exit(EXIT_FAILURE);
}
if (b->in_flush) {
/* Just finished a flush with drained queue: Start next requests */
assert(b->in_flight == 0);
b->in_flush = false;
} else if (b->in_flight > 0) {
int remaining = b->n - b->in_flight;
b->n--;
b->in_flight--;
/* Time for flush? Drain queue if requested, then flush */
if (b->flush_interval && remaining % b->flush_interval == 0) {
if (!b->in_flight || !b->drain_on_flush) {
BlockCompletionFunc *cb;
if (b->drain_on_flush) {
b->in_flush = true;
cb = bench_cb;
} else {
cb = bench_undrained_flush_cb;
}
acb = blk_aio_flush(b->blk, cb, b);
if (!acb) {
error_report("Failed to issue flush request");
exit(EXIT_FAILURE);
}
}
if (b->drain_on_flush) {
return;
}
}
}
while (b->n > b->in_flight && b->in_flight < b->nrreq) {
int64_t offset = b->offset;
/* blk_aio_* might look for completed I/Os and kick bench_cb
* again, so make sure this operation is counted by in_flight
* and b->offset is ready for the next submission.
*/
b->in_flight++;
b->offset += b->step;
b->offset %= b->image_size;
if (b->write) {
acb = blk_aio_pwritev(b->blk, offset, b->qiov, 0, bench_cb, b);
} else {
acb = blk_aio_preadv(b->blk, offset, b->qiov, 0, bench_cb, b);
}
if (!acb) {
error_report("Failed to issue request");
exit(EXIT_FAILURE);
}
}
}
static int img_bench(int argc, char **argv)
{
int c, ret = 0;
const char *fmt = NULL, *filename;
bool quiet = false;
bool image_opts = false;
bool is_write = false;
int count = 75000;
int depth = 64;
int64_t offset = 0;
size_t bufsize = 4096;
int pattern = 0;
size_t step = 0;
int flush_interval = 0;
bool drain_on_flush = true;
int64_t image_size;
BlockBackend *blk = NULL;
BenchData data = {};
int flags = 0;
bool writethrough = false;
struct timeval t1, t2;
int i;
bool force_share = false;
size_t buf_size = 0;
for (;;) {
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"flush-interval", required_argument, 0, OPTION_FLUSH_INTERVAL},
{"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{"pattern", required_argument, 0, OPTION_PATTERN},
{"no-drain", no_argument, 0, OPTION_NO_DRAIN},
{"force-share", no_argument, 0, 'U'},
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, ":hc:d:f:ni:o:qs:S:t:wU", long_options,
NULL);
if (c == -1) {
break;
}
switch (c) {
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
break;
case 'c':
{
unsigned long res;
if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > INT_MAX) {
error_report("Invalid request count specified");
return 1;
}
count = res;
break;
}
case 'd':
{
unsigned long res;
if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > INT_MAX) {
error_report("Invalid queue depth specified");
return 1;
}
depth = res;
break;
}
case 'f':
fmt = optarg;
break;
case 'n':
flags |= BDRV_O_NATIVE_AIO;
break;
case 'i':
ret = bdrv_parse_aio(optarg, &flags);
if (ret < 0) {
error_report("Invalid aio option: %s", optarg);
ret = -1;
goto out;
}
break;
case 'o':
{
offset = cvtnum("offset", optarg);
if (offset < 0) {
return 1;
}
break;
}
break;
case 'q':
quiet = true;
break;
case 's':
{
int64_t sval;
sval = cvtnum_full("buffer size", optarg, 0, INT_MAX);
if (sval < 0) {
return 1;
}
bufsize = sval;
break;
}
case 'S':
{
int64_t sval;
sval = cvtnum_full("step_size", optarg, 0, INT_MAX);
if (sval < 0) {
return 1;
}
step = sval;
break;
}
case 't':
ret = bdrv_parse_cache_mode(optarg, &flags, &writethrough);
if (ret < 0) {
error_report("Invalid cache mode");
ret = -1;
goto out;
}
break;
case 'w':
flags |= BDRV_O_RDWR;
is_write = true;
break;
case 'U':
force_share = true;
break;
case OPTION_PATTERN:
{
unsigned long res;
if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > 0xff) {
error_report("Invalid pattern byte specified");
return 1;
}
pattern = res;
break;
}
case OPTION_FLUSH_INTERVAL:
{
unsigned long res;
if (qemu_strtoul(optarg, NULL, 0, &res) < 0 || res > INT_MAX) {
error_report("Invalid flush interval specified");
return 1;
}
flush_interval = res;
break;
}
case OPTION_NO_DRAIN:
drain_on_flush = false;
break;
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
}
}
if (optind != argc - 1) {
error_exit("Expecting one image file name");
}
filename = argv[argc - 1];
if (!is_write && flush_interval) {
error_report("--flush-interval is only available in write tests");
ret = -1;
goto out;
}
if (flush_interval && flush_interval < depth) {
error_report("Flush interval can't be smaller than depth");
ret = -1;
goto out;
}
blk = img_open(image_opts, filename, fmt, flags, writethrough, quiet,
force_share);
if (!blk) {
ret = -1;
goto out;
}
image_size = blk_getlength(blk);
if (image_size < 0) {
ret = image_size;
goto out;
}
data = (BenchData) {
.blk = blk,
.image_size = image_size,
.bufsize = bufsize,
.step = step ?: bufsize,
.nrreq = depth,
.n = count,
.offset = offset,
.write = is_write,
.flush_interval = flush_interval,
.drain_on_flush = drain_on_flush,
};
printf("Sending %d %s requests, %d bytes each, %d in parallel "
"(starting at offset %" PRId64 ", step size %d)\n",
data.n, data.write ? "write" : "read", data.bufsize, data.nrreq,
data.offset, data.step);
if (flush_interval) {
printf("Sending flush every %d requests\n", flush_interval);
}
buf_size = data.nrreq * data.bufsize;
data.buf = blk_blockalign(blk, buf_size);
memset(data.buf, pattern, data.nrreq * data.bufsize);
blk_register_buf(blk, data.buf, buf_size, &error_fatal);
data.qiov = g_new(QEMUIOVector, data.nrreq);
for (i = 0; i < data.nrreq; i++) {
qemu_iovec_init(&data.qiov[i], 1);
qemu_iovec_add(&data.qiov[i],
data.buf + i * data.bufsize, data.bufsize);
}
gettimeofday(&t1, NULL);
bench_cb(&data, 0);
while (data.n > 0) {
main_loop_wait(false);
}
gettimeofday(&t2, NULL);
printf("Run completed in %3.3f seconds.\n",
(t2.tv_sec - t1.tv_sec)
+ ((double)(t2.tv_usec - t1.tv_usec) / 1000000));
out:
if (data.buf) {
blk_unregister_buf(blk, data.buf, buf_size);
}
qemu_vfree(data.buf);
blk_unref(blk);
if (ret) {
return 1;
}
return 0;
}
2020-05-13 04:16:45 +03:00
enum ImgBitmapAct {
BITMAP_ADD,
BITMAP_REMOVE,
BITMAP_CLEAR,
BITMAP_ENABLE,
BITMAP_DISABLE,
BITMAP_MERGE,
};
typedef struct ImgBitmapAction {
enum ImgBitmapAct act;
const char *src; /* only used for merge */
QSIMPLEQ_ENTRY(ImgBitmapAction) next;
} ImgBitmapAction;
static int img_bitmap(int argc, char **argv)
{
Error *err = NULL;
int c, ret = 1;
QemuOpts *opts = NULL;
const char *fmt = NULL, *src_fmt = NULL, *src_filename = NULL;
const char *filename, *bitmap;
BlockBackend *blk = NULL, *src = NULL;
BlockDriverState *bs = NULL, *src_bs = NULL;
bool image_opts = false;
int64_t granularity = 0;
bool add = false, merge = false;
QSIMPLEQ_HEAD(, ImgBitmapAction) actions;
ImgBitmapAction *act, *act_next;
const char *op;
int inactivate_ret;
2020-05-13 04:16:45 +03:00
QSIMPLEQ_INIT(&actions);
for (;;) {
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"object", required_argument, 0, OPTION_OBJECT},
{"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{"add", no_argument, 0, OPTION_ADD},
{"remove", no_argument, 0, OPTION_REMOVE},
{"clear", no_argument, 0, OPTION_CLEAR},
{"enable", no_argument, 0, OPTION_ENABLE},
{"disable", no_argument, 0, OPTION_DISABLE},
{"merge", required_argument, 0, OPTION_MERGE},
{"granularity", required_argument, 0, 'g'},
{"source-file", required_argument, 0, 'b'},
{"source-format", required_argument, 0, 'F'},
{0, 0, 0, 0}
};
c = getopt_long(argc, argv, ":b:f:F:g:h", long_options, NULL);
if (c == -1) {
break;
}
switch (c) {
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
break;
case 'b':
src_filename = optarg;
break;
case 'f':
fmt = optarg;
break;
case 'F':
src_fmt = optarg;
break;
case 'g':
granularity = cvtnum("granularity", optarg);
if (granularity < 0) {
return 1;
}
break;
case OPTION_ADD:
act = g_new0(ImgBitmapAction, 1);
act->act = BITMAP_ADD;
QSIMPLEQ_INSERT_TAIL(&actions, act, next);
add = true;
break;
case OPTION_REMOVE:
act = g_new0(ImgBitmapAction, 1);
act->act = BITMAP_REMOVE;
QSIMPLEQ_INSERT_TAIL(&actions, act, next);
break;
case OPTION_CLEAR:
act = g_new0(ImgBitmapAction, 1);
act->act = BITMAP_CLEAR;
QSIMPLEQ_INSERT_TAIL(&actions, act, next);
break;
case OPTION_ENABLE:
act = g_new0(ImgBitmapAction, 1);
act->act = BITMAP_ENABLE;
QSIMPLEQ_INSERT_TAIL(&actions, act, next);
break;
case OPTION_DISABLE:
act = g_new0(ImgBitmapAction, 1);
act->act = BITMAP_DISABLE;
QSIMPLEQ_INSERT_TAIL(&actions, act, next);
break;
case OPTION_MERGE:
act = g_new0(ImgBitmapAction, 1);
act->act = BITMAP_MERGE;
act->src = optarg;
QSIMPLEQ_INSERT_TAIL(&actions, act, next);
merge = true;
break;
case OPTION_OBJECT:
user_creatable_process_cmdline(optarg);
2020-05-13 04:16:45 +03:00
break;
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
}
}
if (QSIMPLEQ_EMPTY(&actions)) {
error_report("Need at least one of --add, --remove, --clear, "
"--enable, --disable, or --merge");
goto out;
}
if (granularity && !add) {
error_report("granularity only supported with --add");
goto out;
}
if (src_fmt && !src_filename) {
error_report("-F only supported with -b");
goto out;
}
if (src_filename && !merge) {
error_report("Merge bitmap source file only supported with "
"--merge");
goto out;
}
if (optind != argc - 2) {
error_report("Expecting filename and bitmap name");
goto out;
}
filename = argv[optind];
bitmap = argv[optind + 1];
qemu-img: Support bitmap --merge into backing image If you have the chain 'base.qcow2 <- top.qcow2' and want to merge a bitmap from top into base, qemu-img was failing with: qemu-img: Could not open 'top.qcow2': Could not open backing file: Failed to get shared "write" lock Is another process using the image [base.qcow2]? The easiest fix is to not open the entire backing chain of either image (source or destination); after all, the point of 'qemu-img bitmap' is solely to manipulate bitmaps directly within a single qcow2 image, and this is made more precise if we don't pay attention to other images in the chain that may happen to have a bitmap by the same name. However, note that on a case-by-case analysis, there _are_ times where we treat it as a feature that we can access a bitmap from a backing layer in association with an overlay BDS. A demonstration of this is using NBD to expose both an overlay BDS (for constant contents) and a bitmap (for learning which blocks are interesting) during an incremental backup: Base <- Active <- Temporary \--block job ->/ where Temporary is being fed by a backup 'sync=none' job. When exposing Temporary over NBD, referring to a bitmap that lives only in Active is less effort than having to copy a bitmap into Temporary [1]. So the testsuite additions in this patch check both where bitmaps get allocated (the qemu-img info output), and that qemu-nbd is indeed able to access a bitmap inherited from the backing chain since it is a different use case than 'qemu-img bitmap'. [1] Full disclosure: prior to the recent commit 374eedd1c4 and friends, we were NOT able to see bitmaps through filters, which meant that we actually did not have nice clean semantics for uniformly being able to pick up bitmaps from anywhere in the backing chain (seen as a change in behavior between qemu 4.1 and 4.2 at commit 00e30f05de, when block-copy swapped from a one-off to a filter). Which means libvirt was already coded to copy bitmaps around for the sake of older qemu, even though modern qemu no longer needs it. Oh well. Fixes: http://bugzilla.redhat.com/1877209 Reported-by: Eyal Shenitzky <eshenitz@redhat.com> Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20200914191009.644842-1-eblake@redhat.com> [eblake: more commit message tweaks, per Max Reitz review] Reviewed-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2020-09-14 22:10:09 +03:00
/*
* No need to open backing chains; we will be manipulating bitmaps
* directly in this image without reference to image contents.
*/
blk = img_open(image_opts, filename, fmt, BDRV_O_RDWR | BDRV_O_NO_BACKING,
false, false, false);
2020-05-13 04:16:45 +03:00
if (!blk) {
goto out;
}
bs = blk_bs(blk);
if (src_filename) {
qemu-img: Support bitmap --merge into backing image If you have the chain 'base.qcow2 <- top.qcow2' and want to merge a bitmap from top into base, qemu-img was failing with: qemu-img: Could not open 'top.qcow2': Could not open backing file: Failed to get shared "write" lock Is another process using the image [base.qcow2]? The easiest fix is to not open the entire backing chain of either image (source or destination); after all, the point of 'qemu-img bitmap' is solely to manipulate bitmaps directly within a single qcow2 image, and this is made more precise if we don't pay attention to other images in the chain that may happen to have a bitmap by the same name. However, note that on a case-by-case analysis, there _are_ times where we treat it as a feature that we can access a bitmap from a backing layer in association with an overlay BDS. A demonstration of this is using NBD to expose both an overlay BDS (for constant contents) and a bitmap (for learning which blocks are interesting) during an incremental backup: Base <- Active <- Temporary \--block job ->/ where Temporary is being fed by a backup 'sync=none' job. When exposing Temporary over NBD, referring to a bitmap that lives only in Active is less effort than having to copy a bitmap into Temporary [1]. So the testsuite additions in this patch check both where bitmaps get allocated (the qemu-img info output), and that qemu-nbd is indeed able to access a bitmap inherited from the backing chain since it is a different use case than 'qemu-img bitmap'. [1] Full disclosure: prior to the recent commit 374eedd1c4 and friends, we were NOT able to see bitmaps through filters, which meant that we actually did not have nice clean semantics for uniformly being able to pick up bitmaps from anywhere in the backing chain (seen as a change in behavior between qemu 4.1 and 4.2 at commit 00e30f05de, when block-copy swapped from a one-off to a filter). Which means libvirt was already coded to copy bitmaps around for the sake of older qemu, even though modern qemu no longer needs it. Oh well. Fixes: http://bugzilla.redhat.com/1877209 Reported-by: Eyal Shenitzky <eshenitz@redhat.com> Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20200914191009.644842-1-eblake@redhat.com> [eblake: more commit message tweaks, per Max Reitz review] Reviewed-by: Max Reitz <mreitz@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2020-09-14 22:10:09 +03:00
src = img_open(false, src_filename, src_fmt, BDRV_O_NO_BACKING,
false, false, false);
2020-05-13 04:16:45 +03:00
if (!src) {
goto out;
}
src_bs = blk_bs(src);
} else {
src_bs = bs;
}
QSIMPLEQ_FOREACH_SAFE(act, &actions, next, act_next) {
switch (act->act) {
case BITMAP_ADD:
qmp_block_dirty_bitmap_add(bs->node_name, bitmap,
!!granularity, granularity, true, true,
false, false, &err);
op = "add";
break;
case BITMAP_REMOVE:
qmp_block_dirty_bitmap_remove(bs->node_name, bitmap, &err);
op = "remove";
break;
case BITMAP_CLEAR:
qmp_block_dirty_bitmap_clear(bs->node_name, bitmap, &err);
op = "clear";
break;
case BITMAP_ENABLE:
qmp_block_dirty_bitmap_enable(bs->node_name, bitmap, &err);
op = "enable";
break;
case BITMAP_DISABLE:
qmp_block_dirty_bitmap_disable(bs->node_name, bitmap, &err);
op = "disable";
break;
case BITMAP_MERGE:
do_dirty_bitmap_merge(bs->node_name, bitmap, src_bs->node_name,
act->src, &err);
2020-05-13 04:16:45 +03:00
op = "merge";
break;
default:
g_assert_not_reached();
}
if (err) {
error_reportf_err(err, "Operation %s on bitmap %s failed: ",
op, bitmap);
goto out;
}
g_free(act);
}
ret = 0;
out:
/*
* Manually inactivate the images first because this way we can know whether
* an error occurred. blk_unref() doesn't tell us about failures.
*/
inactivate_ret = bdrv_inactivate_all();
if (inactivate_ret < 0) {
error_report("Error while closing the image: %s", strerror(-inactivate_ret));
ret = 1;
}
2020-05-13 04:16:45 +03:00
blk_unref(src);
blk_unref(blk);
qemu_opts_del(opts);
return ret;
}
#define C_BS 01
#define C_COUNT 02
#define C_IF 04
#define C_OF 010
#define C_SKIP 020
struct DdInfo {
unsigned int flags;
int64_t count;
};
struct DdIo {
int bsz; /* Block size */
char *filename;
uint8_t *buf;
int64_t offset;
};
struct DdOpts {
const char *name;
int (*f)(const char *, struct DdIo *, struct DdIo *, struct DdInfo *);
unsigned int flag;
};
static int img_dd_bs(const char *arg,
struct DdIo *in, struct DdIo *out,
struct DdInfo *dd)
{
int64_t res;
res = cvtnum_full("bs", arg, 1, INT_MAX);
if (res < 0) {
return 1;
}
in->bsz = out->bsz = res;
return 0;
}
static int img_dd_count(const char *arg,
struct DdIo *in, struct DdIo *out,
struct DdInfo *dd)
{
dd->count = cvtnum("count", arg);
if (dd->count < 0) {
return 1;
}
return 0;
}
static int img_dd_if(const char *arg,
struct DdIo *in, struct DdIo *out,
struct DdInfo *dd)
{
in->filename = g_strdup(arg);
return 0;
}
static int img_dd_of(const char *arg,
struct DdIo *in, struct DdIo *out,
struct DdInfo *dd)
{
out->filename = g_strdup(arg);
return 0;
}
static int img_dd_skip(const char *arg,
struct DdIo *in, struct DdIo *out,
struct DdInfo *dd)
{
in->offset = cvtnum("skip", arg);
if (in->offset < 0) {
return 1;
}
return 0;
}
static int img_dd(int argc, char **argv)
{
int ret = 0;
char *arg = NULL;
char *tmp;
BlockDriver *drv = NULL, *proto_drv = NULL;
BlockBackend *blk1 = NULL, *blk2 = NULL;
QemuOpts *opts = NULL;
QemuOptsList *create_opts = NULL;
Error *local_err = NULL;
bool image_opts = false;
int c, i;
const char *out_fmt = "raw";
const char *fmt = NULL;
int64_t size = 0;
int64_t out_pos, in_pos;
bool force_share = false;
struct DdInfo dd = {
.flags = 0,
.count = 0,
};
struct DdIo in = {
.bsz = 512, /* Block size is by default 512 bytes */
.filename = NULL,
.buf = NULL,
.offset = 0
};
struct DdIo out = {
.bsz = 512,
.filename = NULL,
.buf = NULL,
.offset = 0
};
const struct DdOpts options[] = {
{ "bs", img_dd_bs, C_BS },
{ "count", img_dd_count, C_COUNT },
{ "if", img_dd_if, C_IF },
{ "of", img_dd_of, C_OF },
{ "skip", img_dd_skip, C_SKIP },
{ NULL, NULL, 0 }
};
const struct option long_options[] = {
{ "help", no_argument, 0, 'h'},
{ "object", required_argument, 0, OPTION_OBJECT},
{ "image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{ "force-share", no_argument, 0, 'U'},
{ 0, 0, 0, 0 }
};
while ((c = getopt_long(argc, argv, ":hf:O:U", long_options, NULL))) {
if (c == EOF) {
break;
}
switch (c) {
case 'O':
out_fmt = optarg;
break;
case 'f':
fmt = optarg;
break;
case ':':
missing_argument(argv[optind - 1]);
break;
case '?':
unrecognized_option(argv[optind - 1]);
break;
case 'h':
help();
break;
case 'U':
force_share = true;
break;
case OPTION_OBJECT:
user_creatable_process_cmdline(optarg);
break;
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
}
}
for (i = optind; i < argc; i++) {
int j;
arg = g_strdup(argv[i]);
tmp = strchr(arg, '=');
if (tmp == NULL) {
error_report("unrecognized operand %s", arg);
ret = -1;
goto out;
}
*tmp++ = '\0';
for (j = 0; options[j].name != NULL; j++) {
if (!strcmp(arg, options[j].name)) {
break;
}
}
if (options[j].name == NULL) {
error_report("unrecognized operand %s", arg);
ret = -1;
goto out;
}
if (options[j].f(tmp, &in, &out, &dd) != 0) {
ret = -1;
goto out;
}
dd.flags |= options[j].flag;
g_free(arg);
arg = NULL;
}
if (!(dd.flags & C_IF && dd.flags & C_OF)) {
error_report("Must specify both input and output files");
ret = -1;
goto out;
}
blk1 = img_open(image_opts, in.filename, fmt, 0, false, false,
force_share);
if (!blk1) {
ret = -1;
goto out;
}
drv = bdrv_find_format(out_fmt);
if (!drv) {
error_report("Unknown file format");
ret = -1;
goto out;
}
proto_drv = bdrv_find_protocol(out.filename, true, &local_err);
if (!proto_drv) {
error_report_err(local_err);
ret = -1;
goto out;
}
if (!drv->create_opts) {
error_report("Format driver '%s' does not support image creation",
drv->format_name);
ret = -1;
goto out;
}
if (!proto_drv->create_opts) {
error_report("Protocol driver '%s' does not support image creation",
proto_drv->format_name);
ret = -1;
goto out;
}
create_opts = qemu_opts_append(create_opts, drv->create_opts);
create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
size = blk_getlength(blk1);
if (size < 0) {
error_report("Failed to get size for '%s'", in.filename);
ret = -1;
goto out;
}
if (dd.flags & C_COUNT && dd.count <= INT64_MAX / in.bsz &&
dd.count * in.bsz < size) {
size = dd.count * in.bsz;
}
/* Overflow means the specified offset is beyond input image's size */
if (dd.flags & C_SKIP && (in.offset > INT64_MAX / in.bsz ||
size < in.bsz * in.offset)) {
qemu_opt_set_number(opts, BLOCK_OPT_SIZE, 0, &error_abort);
} else {
qemu_opt_set_number(opts, BLOCK_OPT_SIZE,
size - in.bsz * in.offset, &error_abort);
}
ret = bdrv_create(drv, out.filename, opts, &local_err);
if (ret < 0) {
error_reportf_err(local_err,
"%s: error while creating output image: ",
out.filename);
ret = -1;
goto out;
}
/* TODO, we can't honour --image-opts for the target,
* since it needs to be given in a format compatible
* with the bdrv_create() call above which does not
* support image-opts style.
*/
blk2 = img_open_file(out.filename, NULL, out_fmt, BDRV_O_RDWR,
false, false, false);
if (!blk2) {
ret = -1;
goto out;
}
if (dd.flags & C_SKIP && (in.offset > INT64_MAX / in.bsz ||
size < in.offset * in.bsz)) {
/* We give a warning if the skip option is bigger than the input
* size and create an empty output disk image (i.e. like dd(1)).
*/
error_report("%s: cannot skip to specified offset", in.filename);
in_pos = size;
} else {
in_pos = in.offset * in.bsz;
}
in.buf = g_new(uint8_t, in.bsz);
for (out_pos = 0; in_pos < size; ) {
int bytes = (in_pos + in.bsz > size) ? size - in_pos : in.bsz;
ret = blk_pread(blk1, in_pos, bytes, in.buf, 0);
if (ret < 0) {
error_report("error while reading from input image file: %s",
strerror(-ret));
goto out;
}
in_pos += bytes;
ret = blk_pwrite(blk2, out_pos, bytes, in.buf, 0);
if (ret < 0) {
error_report("error while writing to output image file: %s",
strerror(-ret));
goto out;
}
out_pos += bytes;
}
out:
g_free(arg);
qemu_opts_del(opts);
qemu_opts_free(create_opts);
blk_unref(blk1);
blk_unref(blk2);
g_free(in.filename);
g_free(out.filename);
g_free(in.buf);
g_free(out.buf);
if (ret) {
return 1;
}
return 0;
}
static void dump_json_block_measure_info(BlockMeasureInfo *info)
{
GString *str;
QObject *obj;
Visitor *v = qobject_output_visitor_new(&obj);
visit_type_BlockMeasureInfo(v, NULL, &info, &error_abort);
visit_complete(v, &obj);
str = qobject_to_json_pretty(obj, true);
assert(str != NULL);
printf("%s\n", str->str);
qobject_unref(obj);
visit_free(v);
g_string_free(str, true);
}
static int img_measure(int argc, char **argv)
{
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"image-opts", no_argument, 0, OPTION_IMAGE_OPTS},
{"object", required_argument, 0, OPTION_OBJECT},
{"output", required_argument, 0, OPTION_OUTPUT},
{"size", required_argument, 0, OPTION_SIZE},
{"force-share", no_argument, 0, 'U'},
{0, 0, 0, 0}
};
OutputFormat output_format = OFORMAT_HUMAN;
BlockBackend *in_blk = NULL;
BlockDriver *drv;
const char *filename = NULL;
const char *fmt = NULL;
const char *out_fmt = "raw";
char *options = NULL;
char *snapshot_name = NULL;
bool force_share = false;
QemuOpts *opts = NULL;
QemuOpts *object_opts = NULL;
QemuOpts *sn_opts = NULL;
QemuOptsList *create_opts = NULL;
bool image_opts = false;
uint64_t img_size = UINT64_MAX;
BlockMeasureInfo *info = NULL;
Error *local_err = NULL;
int ret = 1;
int c;
while ((c = getopt_long(argc, argv, "hf:O:o:l:U",
long_options, NULL)) != -1) {
switch (c) {
case '?':
case 'h':
help();
break;
case 'f':
fmt = optarg;
break;
case 'O':
out_fmt = optarg;
break;
case 'o':
if (accumulate_options(&options, optarg) < 0) {
goto out;
}
break;
case 'l':
if (strstart(optarg, SNAPSHOT_OPT_BASE, NULL)) {
sn_opts = qemu_opts_parse_noisily(&internal_snapshot_opts,
optarg, false);
if (!sn_opts) {
error_report("Failed in parsing snapshot param '%s'",
optarg);
goto out;
}
} else {
snapshot_name = optarg;
}
break;
case 'U':
force_share = true;
break;
case OPTION_OBJECT:
user_creatable_process_cmdline(optarg);
break;
case OPTION_IMAGE_OPTS:
image_opts = true;
break;
case OPTION_OUTPUT:
if (!strcmp(optarg, "json")) {
output_format = OFORMAT_JSON;
} else if (!strcmp(optarg, "human")) {
output_format = OFORMAT_HUMAN;
} else {
error_report("--output must be used with human or json "
"as argument.");
goto out;
}
break;
case OPTION_SIZE:
{
int64_t sval;
sval = cvtnum("image size", optarg);
if (sval < 0) {
goto out;
}
img_size = (uint64_t)sval;
}
break;
}
}
if (argc - optind > 1) {
error_report("At most one filename argument is allowed.");
goto out;
} else if (argc - optind == 1) {
filename = argv[optind];
}
if (!filename && (image_opts || fmt || snapshot_name || sn_opts)) {
error_report("--image-opts, -f, and -l require a filename argument.");
goto out;
}
if (filename && img_size != UINT64_MAX) {
error_report("--size N cannot be used together with a filename.");
goto out;
}
if (!filename && img_size == UINT64_MAX) {
error_report("Either --size N or one filename must be specified.");
goto out;
}
if (filename) {
in_blk = img_open(image_opts, filename, fmt, 0,
false, false, force_share);
if (!in_blk) {
goto out;
}
if (sn_opts) {
bdrv_snapshot_load_tmp(blk_bs(in_blk),
qemu_opt_get(sn_opts, SNAPSHOT_OPT_ID),
qemu_opt_get(sn_opts, SNAPSHOT_OPT_NAME),
&local_err);
} else if (snapshot_name != NULL) {
bdrv_snapshot_load_tmp_by_id_or_name(blk_bs(in_blk),
snapshot_name, &local_err);
}
if (local_err) {
error_reportf_err(local_err, "Failed to load snapshot: ");
goto out;
}
}
drv = bdrv_find_format(out_fmt);
if (!drv) {
error_report("Unknown file format '%s'", out_fmt);
goto out;
}
if (!drv->create_opts) {
error_report("Format driver '%s' does not support image creation",
drv->format_name);
goto out;
}
create_opts = qemu_opts_append(create_opts, drv->create_opts);
create_opts = qemu_opts_append(create_opts, bdrv_file.create_opts);
opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
if (options) {
if (!qemu_opts_do_parse(opts, options, NULL, &local_err)) {
error_report_err(local_err);
error_report("Invalid options for file format '%s'", out_fmt);
goto out;
}
}
if (img_size != UINT64_MAX) {
qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size, &error_abort);
}
info = bdrv_measure(drv, opts, in_blk ? blk_bs(in_blk) : NULL, &local_err);
if (local_err) {
error_report_err(local_err);
goto out;
}
if (output_format == OFORMAT_HUMAN) {
printf("required size: %" PRIu64 "\n", info->required);
printf("fully allocated size: %" PRIu64 "\n", info->fully_allocated);
qcow2: Expose bitmaps' size during measure It's useful to know how much space can be occupied by qcow2 persistent bitmaps, even though such metadata is unrelated to the guest-visible data. Report this value as an additional QMP field, present when measuring an existing image and output format that both support bitmaps. Update iotest 178 and 190 to updated output, as well as new coverage in 190 demonstrating non-zero values made possible with the recently-added qemu-img bitmap command (see 3b51ab4b). The new 'bitmaps size:' field is displayed automatically as part of 'qemu-img measure' any time it is present in QMP (that is, any time both the source image being measured and destination format support bitmaps, even if the measurement is 0 because there are no bitmaps present). If the field is absent, it means that no bitmaps can be copied (source, destination, or both lack bitmaps, including when measuring based on size rather than on a source image). This behavior is compatible with an upcoming patch adding 'qemu-img convert --bitmaps': that command will fail in the same situations where this patch omits the field. The addition of a new field demonstrates why we should always zero-initialize qapi C structs; while the qcow2 driver still fully populates all fields, the raw and crypto drivers had to be tweaked to avoid uninitialized data. Consideration was also given towards having a 'qemu-img measure --bitmaps' which errors out when bitmaps are not possible, and otherwise sums the bitmaps into the existing allocation totals rather than displaying as a separate field, as a potential convenience factor. But this was ultimately decided to be more complexity than necessary when the QMP interface was sufficient enough with bitmaps remaining a separate field. See also: https://bugzilla.redhat.com/1779904 Reported-by: Nir Soffer <nsoffer@redhat.com> Signed-off-by: Eric Blake <eblake@redhat.com> Message-Id: <20200521192137.1120211-3-eblake@redhat.com> Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
2020-05-21 22:21:34 +03:00
if (info->has_bitmaps) {
printf("bitmaps size: %" PRIu64 "\n", info->bitmaps);
}
} else {
dump_json_block_measure_info(info);
}
ret = 0;
out:
qapi_free_BlockMeasureInfo(info);
qemu_opts_del(object_opts);
qemu_opts_del(opts);
qemu_opts_del(sn_opts);
qemu_opts_free(create_opts);
g_free(options);
blk_unref(in_blk);
return ret;
}
static const img_cmd_t img_cmds[] = {
#define DEF(option, callback, arg_string) \
{ option, callback },
#include "qemu-img-cmds.h"
#undef DEF
{ NULL, NULL, },
};
int main(int argc, char **argv)
{
const img_cmd_t *cmd;
const char *cmdname;
int c;
static const struct option long_options[] = {
{"help", no_argument, 0, 'h'},
{"version", no_argument, 0, 'V'},
{"trace", required_argument, NULL, 'T'},
{0, 0, 0, 0}
};
#ifdef CONFIG_POSIX
signal(SIGPIPE, SIG_IGN);
#endif
socket_init();
error_init(argv[0]);
module_call_init(MODULE_INIT_TRACE);
qemu_init_exec_dir(argv[0]);
qemu_init_main_loop(&error_fatal);
qcrypto_init(&error_fatal);
module_call_init(MODULE_INIT_QOM);
bdrv_init();
if (argc < 2) {
error_exit("Not enough arguments");
}
qemu_add_opts(&qemu_source_opts);
qemu_add_opts(&qemu_trace_opts);
while ((c = getopt_long(argc, argv, "+:hVT:", long_options, NULL)) != -1) {
switch (c) {
case ':':
missing_argument(argv[optind - 1]);
return 0;
case '?':
unrecognized_option(argv[optind - 1]);
return 0;
case 'h':
help();
return 0;
case 'V':
printf(QEMU_IMG_VERSION);
return 0;
case 'T':
trace_opt_parse(optarg);
break;
}
}
cmdname = argv[optind];
/* reset getopt_long scanning */
argc -= optind;
if (argc < 1) {
return 0;
}
argv += optind;
qemu-io: Add generic function for reinitializing optind. On FreeBSD 11.2: $ nbdkit memory size=1M --run './qemu-io -f raw -c "aio_write 0 512" $nbd' Parsing error: non-numeric argument, or extraneous/unrecognized suffix -- aio_write After main option parsing, we reinitialize optind so we can parse each command. However reinitializing optind to 0 does not work on FreeBSD. What happens when you do this is optind remains 0 after the option parsing loop, and the result is we try to parse argv[optind] == argv[0] == "aio_write" as if it was the first parameter. The FreeBSD manual page says: In order to use getopt() to evaluate multiple sets of arguments, or to evaluate a single set of arguments multiple times, the variable optreset must be set to 1 before the second and each additional set of calls to getopt(), and the variable optind must be reinitialized. (From the rest of the man page it is clear that optind must be reinitialized to 1). The glibc man page says: A program that scans multiple argument vectors, or rescans the same vector more than once, and wants to make use of GNU extensions such as '+' and '-' at the start of optstring, or changes the value of POSIXLY_CORRECT between scans, must reinitialize getopt() by resetting optind to 0, rather than the traditional value of 1. (Resetting to 0 forces the invocation of an internal initialization routine that rechecks POSIXLY_CORRECT and checks for GNU extensions in optstring.) This commit introduces an OS-portability function called qemu_reset_optind which provides a way of resetting optind that works on FreeBSD and platforms that use optreset, while keeping it the same as now on other platforms. Note that the qemu codebase sets optind in many other places, but in those other places it's setting a local variable and not using getopt. This change is only needed in places where we are using getopt and the associated global variable optind. Signed-off-by: Richard W.M. Jones <rjones@redhat.com> Message-id: 20190118101114.11759-2-rjones@redhat.com Reviewed-by: Daniel P. Berrangé <berrange@redhat.com> Reviewed-by: Eric Blake <eblake@redhat.com> Signed-off-by: Max Reitz <mreitz@redhat.com>
2019-01-18 13:11:14 +03:00
qemu_reset_optind();
if (!trace_init_backends()) {
exit(1);
}
trace_init_file();
qemu_set_log(LOG_TRACE, &error_fatal);
/* find the command */
for (cmd = img_cmds; cmd->name != NULL; cmd++) {
if (!strcmp(cmdname, cmd->name)) {
return cmd->handler(argc, argv);
}
}
/* not found */
error_exit("Command not found: %s", cmdname);
}