mirrors/qemu
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
dd684ead29
qemu/tests/test-util-filemonitor.c

718 lines
22 KiB
C
Raw Normal View History

util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
/*
* Tests for util/filemonitor-*.c
*
* Copyright 2018 Red Hat, Inc.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this library; if not, see <http://www.gnu.org/licenses/>.
*
*/
#include "qemu/osdep.h"
#include "qemu/main-loop.h"
#include "qapi/error.h"
#include "qemu/filemonitor.h"
#include <utime.h>
enum {
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
QFILE_MONITOR_TEST_OP_ADD_WATCH,
QFILE_MONITOR_TEST_OP_DEL_WATCH,
QFILE_MONITOR_TEST_OP_EVENT,
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
QFILE_MONITOR_TEST_OP_CREATE,
QFILE_MONITOR_TEST_OP_APPEND,
QFILE_MONITOR_TEST_OP_TRUNC,
QFILE_MONITOR_TEST_OP_RENAME,
QFILE_MONITOR_TEST_OP_TOUCH,
QFILE_MONITOR_TEST_OP_UNLINK,
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
QFILE_MONITOR_TEST_OP_MKDIR,
QFILE_MONITOR_TEST_OP_RMDIR,
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
};
typedef struct {
int type;
const char *filesrc;
const char *filedst;
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
int64_t *watchid;
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
int eventid;
tests: make filemonitor test more robust to event ordering The ordering of events that are emitted during the rmdir test have changed with kernel >= 5.3. Semantically both new & old orderings are correct, so we must be able to cope with either. To cope with this, when we see an unexpected event, we push it back onto the queue and look and the subsequent event to see if that matches instead. Tested-by: Peter Xu <peterx@redhat.com> Tested-by: Wei Yang <richardw.yang@linux.intel.com> Tested-by: Cornelia Huck <cohuck@redhat.com> Reviewed-by: Thomas Huth <thuth@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-08-21 18:14:27 +03:00
/*
* Only valid with OP_EVENT - this event might be
* swapped with the next OP_EVENT
*/
bool swapnext;
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
} QFileMonitorTestOp;
typedef struct {
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
int64_t id;
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
QFileMonitorEvent event;
char *filename;
} QFileMonitorTestRecord;
typedef struct {
QemuMutex lock;
GList *records;
} QFileMonitorTestData;
static QemuMutex evlock;
static bool evstopping;
static bool evrunning;
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
static bool debug;
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
/*
* Main function for a background thread that is
* running the event loop during the test
*/
static void *
qemu_file_monitor_test_event_loop(void *opaque G_GNUC_UNUSED)
{
qemu_mutex_lock(&evlock);
while (!evstopping) {
qemu_mutex_unlock(&evlock);
main_loop_wait(true);
qemu_mutex_lock(&evlock);
}
evrunning = false;
qemu_mutex_unlock(&evlock);
return NULL;
}
/*
* File monitor event handler which simply maintains
* an ordered list of all events that it receives
*/
static void
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
qemu_file_monitor_test_handler(int64_t id,
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
QFileMonitorEvent event,
const char *filename,
void *opaque)
{
QFileMonitorTestData *data = opaque;
QFileMonitorTestRecord *rec = g_new0(QFileMonitorTestRecord, 1);
tests: make filemonitor test more robust to event ordering The ordering of events that are emitted during the rmdir test have changed with kernel >= 5.3. Semantically both new & old orderings are correct, so we must be able to cope with either. To cope with this, when we see an unexpected event, we push it back onto the queue and look and the subsequent event to see if that matches instead. Tested-by: Peter Xu <peterx@redhat.com> Tested-by: Wei Yang <richardw.yang@linux.intel.com> Tested-by: Cornelia Huck <cohuck@redhat.com> Reviewed-by: Thomas Huth <thuth@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-08-21 18:14:27 +03:00
if (debug) {
g_printerr("Queue event id %" PRIx64 " event %d file %s\n",
id, event, filename);
}
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
rec->id = id;
rec->event = event;
rec->filename = g_strdup(filename);
qemu_mutex_lock(&data->lock);
data->records = g_list_append(data->records, rec);
qemu_mutex_unlock(&data->lock);
}
static void
qemu_file_monitor_test_record_free(QFileMonitorTestRecord *rec)
{
g_free(rec->filename);
g_free(rec);
}
/*
* Get the next event record that has been received by
* the file monitor event handler. Since events are
* emitted in the background thread running the event
* loop, we can't assume there is a record available
* immediately. Thus we will sleep for upto 5 seconds
* to wait for the event to be queued for us.
*/
static QFileMonitorTestRecord *
tests: make filemonitor test more robust to event ordering The ordering of events that are emitted during the rmdir test have changed with kernel >= 5.3. Semantically both new & old orderings are correct, so we must be able to cope with either. To cope with this, when we see an unexpected event, we push it back onto the queue and look and the subsequent event to see if that matches instead. Tested-by: Peter Xu <peterx@redhat.com> Tested-by: Wei Yang <richardw.yang@linux.intel.com> Tested-by: Cornelia Huck <cohuck@redhat.com> Reviewed-by: Thomas Huth <thuth@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-08-21 18:14:27 +03:00
qemu_file_monitor_test_next_record(QFileMonitorTestData *data,
QFileMonitorTestRecord *pushback)
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
{
GTimer *timer = g_timer_new();
QFileMonitorTestRecord *record = NULL;
GList *tmp;
qemu_mutex_lock(&data->lock);
while (!data->records && g_timer_elapsed(timer, NULL) < 5) {
qemu_mutex_unlock(&data->lock);
usleep(10 * 1000);
qemu_mutex_lock(&data->lock);
}
if (data->records) {
record = data->records->data;
tests: make filemonitor test more robust to event ordering The ordering of events that are emitted during the rmdir test have changed with kernel >= 5.3. Semantically both new & old orderings are correct, so we must be able to cope with either. To cope with this, when we see an unexpected event, we push it back onto the queue and look and the subsequent event to see if that matches instead. Tested-by: Peter Xu <peterx@redhat.com> Tested-by: Wei Yang <richardw.yang@linux.intel.com> Tested-by: Cornelia Huck <cohuck@redhat.com> Reviewed-by: Thomas Huth <thuth@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-08-21 18:14:27 +03:00
if (pushback) {
data->records->data = pushback;
} else {
tmp = data->records;
data->records = g_list_remove_link(data->records, tmp);
g_list_free(tmp);
}
} else if (pushback) {
qemu_file_monitor_test_record_free(pushback);
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
}
qemu_mutex_unlock(&data->lock);
g_timer_destroy(timer);
return record;
}
/*
* Check whether the event record we retrieved matches
* data we were expecting to see for the event
*/
static bool
qemu_file_monitor_test_expect(QFileMonitorTestData *data,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
int64_t id,
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
QFileMonitorEvent event,
tests: make filemonitor test more robust to event ordering The ordering of events that are emitted during the rmdir test have changed with kernel >= 5.3. Semantically both new & old orderings are correct, so we must be able to cope with either. To cope with this, when we see an unexpected event, we push it back onto the queue and look and the subsequent event to see if that matches instead. Tested-by: Peter Xu <peterx@redhat.com> Tested-by: Wei Yang <richardw.yang@linux.intel.com> Tested-by: Cornelia Huck <cohuck@redhat.com> Reviewed-by: Thomas Huth <thuth@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-08-21 18:14:27 +03:00
const char *filename,
bool swapnext)
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
{
QFileMonitorTestRecord *rec;
bool ret = false;
tests: make filemonitor test more robust to event ordering The ordering of events that are emitted during the rmdir test have changed with kernel >= 5.3. Semantically both new & old orderings are correct, so we must be able to cope with either. To cope with this, when we see an unexpected event, we push it back onto the queue and look and the subsequent event to see if that matches instead. Tested-by: Peter Xu <peterx@redhat.com> Tested-by: Wei Yang <richardw.yang@linux.intel.com> Tested-by: Cornelia Huck <cohuck@redhat.com> Reviewed-by: Thomas Huth <thuth@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-08-21 18:14:27 +03:00
rec = qemu_file_monitor_test_next_record(data, NULL);
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
tests: make filemonitor test more robust to event ordering The ordering of events that are emitted during the rmdir test have changed with kernel >= 5.3. Semantically both new & old orderings are correct, so we must be able to cope with either. To cope with this, when we see an unexpected event, we push it back onto the queue and look and the subsequent event to see if that matches instead. Tested-by: Peter Xu <peterx@redhat.com> Tested-by: Wei Yang <richardw.yang@linux.intel.com> Tested-by: Cornelia Huck <cohuck@redhat.com> Reviewed-by: Thomas Huth <thuth@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-08-21 18:14:27 +03:00
retry:
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
if (!rec) {
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
g_printerr("Missing event watch id %" PRIx64 " event %d file %s\n",
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
id, event, filename);
return false;
}
if (id != rec->id) {
tests: make filemonitor test more robust to event ordering The ordering of events that are emitted during the rmdir test have changed with kernel >= 5.3. Semantically both new & old orderings are correct, so we must be able to cope with either. To cope with this, when we see an unexpected event, we push it back onto the queue and look and the subsequent event to see if that matches instead. Tested-by: Peter Xu <peterx@redhat.com> Tested-by: Wei Yang <richardw.yang@linux.intel.com> Tested-by: Cornelia Huck <cohuck@redhat.com> Reviewed-by: Thomas Huth <thuth@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-08-21 18:14:27 +03:00
if (swapnext) {
rec = qemu_file_monitor_test_next_record(data, rec);
swapnext = false;
goto retry;
}
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
g_printerr("Expected watch id %" PRIx64 " but got %" PRIx64 "\n",
id, rec->id);
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
goto cleanup;
}
if (event != rec->event) {
g_printerr("Expected event %d but got %d\n", event, rec->event);
goto cleanup;
}
if (!g_str_equal(filename, rec->filename)) {
g_printerr("Expected filename %s but got %s\n",
filename, rec->filename);
goto cleanup;
}
ret = true;
cleanup:
qemu_file_monitor_test_record_free(rec);
return ret;
}
static void
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
test_file_monitor_events(void)
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
{
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
int64_t watch0 = 0;
int64_t watch1 = 0;
int64_t watch2 = 0;
int64_t watch3 = 0;
int64_t watch4 = 0;
int64_t watch5 = 0;
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
QFileMonitorTestOp ops[] = {
{ .type = QFILE_MONITOR_TEST_OP_ADD_WATCH,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = NULL, .watchid = &watch0 },
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
{ .type = QFILE_MONITOR_TEST_OP_ADD_WATCH,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch1 },
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
{ .type = QFILE_MONITOR_TEST_OP_ADD_WATCH,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch2 },
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
{ .type = QFILE_MONITOR_TEST_OP_CREATE,
.filesrc = "one.txt", },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch0,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch1,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_CREATE,
.filesrc = "two.txt", },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch0,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch2,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_CREATE,
.filesrc = "three.txt", },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "three.txt", .watchid = &watch0,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_UNLINK,
.filesrc = "three.txt", },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "three.txt", .watchid = &watch0,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_DELETED },
{ .type = QFILE_MONITOR_TEST_OP_RENAME,
.filesrc = "one.txt", .filedst = "two.txt" },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch0,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_DELETED },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch1,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_DELETED },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch0,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch2,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_APPEND,
.filesrc = "two.txt", },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch0,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_MODIFIED },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch2,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_MODIFIED },
{ .type = QFILE_MONITOR_TEST_OP_TOUCH,
.filesrc = "two.txt", },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch0,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_ATTRIBUTES },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch2,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_ATTRIBUTES },
{ .type = QFILE_MONITOR_TEST_OP_DEL_WATCH,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch1 },
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
{ .type = QFILE_MONITOR_TEST_OP_ADD_WATCH,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch3 },
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
{ .type = QFILE_MONITOR_TEST_OP_CREATE,
.filesrc = "one.txt", },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch0,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch3,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_DEL_WATCH,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch3 },
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
{ .type = QFILE_MONITOR_TEST_OP_UNLINK,
.filesrc = "one.txt", },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch0,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_DELETED },
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
{ .type = QFILE_MONITOR_TEST_OP_MKDIR,
.filesrc = "fish", },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "fish", .watchid = &watch0,
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_ADD_WATCH,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "fish/", .watchid = &watch4 },
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
{ .type = QFILE_MONITOR_TEST_OP_ADD_WATCH,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "fish/one.txt", .watchid = &watch5 },
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
{ .type = QFILE_MONITOR_TEST_OP_CREATE,
.filesrc = "fish/one.txt", },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch4,
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch5,
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_DEL_WATCH,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "fish/one.txt", .watchid = &watch5 },
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
{ .type = QFILE_MONITOR_TEST_OP_RENAME,
.filesrc = "fish/one.txt", .filedst = "two.txt", },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "one.txt", .watchid = &watch4,
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
.eventid = QFILE_MONITOR_EVENT_DELETED },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch0,
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch2,
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
.eventid = QFILE_MONITOR_EVENT_CREATED },
{ .type = QFILE_MONITOR_TEST_OP_RMDIR,
.filesrc = "fish", },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "", .watchid = &watch4,
tests: make filemonitor test more robust to event ordering The ordering of events that are emitted during the rmdir test have changed with kernel >= 5.3. Semantically both new & old orderings are correct, so we must be able to cope with either. To cope with this, when we see an unexpected event, we push it back onto the queue and look and the subsequent event to see if that matches instead. Tested-by: Peter Xu <peterx@redhat.com> Tested-by: Wei Yang <richardw.yang@linux.intel.com> Tested-by: Cornelia Huck <cohuck@redhat.com> Reviewed-by: Thomas Huth <thuth@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-08-21 18:14:27 +03:00
.eventid = QFILE_MONITOR_EVENT_IGNORED,
.swapnext = true },
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "fish", .watchid = &watch0,
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
.eventid = QFILE_MONITOR_EVENT_DELETED },
{ .type = QFILE_MONITOR_TEST_OP_DEL_WATCH,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "fish", .watchid = &watch4 },
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
{ .type = QFILE_MONITOR_TEST_OP_UNLINK,
.filesrc = "two.txt", },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch0,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_DELETED },
{ .type = QFILE_MONITOR_TEST_OP_EVENT,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch2,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
.eventid = QFILE_MONITOR_EVENT_DELETED },
{ .type = QFILE_MONITOR_TEST_OP_DEL_WATCH,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = "two.txt", .watchid = &watch2 },
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
{ .type = QFILE_MONITOR_TEST_OP_DEL_WATCH,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
.filesrc = NULL, .watchid = &watch0 },
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
};
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
Error *local_err = NULL;
GError *gerr = NULL;
QFileMonitor *mon = qemu_file_monitor_new(&local_err);
QemuThread th;
GTimer *timer;
gchar *dir = NULL;
int err = -1;
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
gsize i;
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
char *pathsrc = NULL;
char *pathdst = NULL;
QFileMonitorTestData data;
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
GHashTable *ids = g_hash_table_new(g_int64_hash, g_int64_equal);
tests/test-util-filemonitor: Skip test on non-x86 Travis containers test-util-filemonitor fails in restricted non-x86 Travis containers since they apparently blacklisted some required system calls there. Let's simply skip the test if we detect such an environment. Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Thomas Huth <thuth@redhat.com> Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Cleber Rosa <crosa@redhat.com> Tested-by: Cleber Rosa <crosa@redhat.com> Message-Id: <20191204154618.23560-6-thuth@redhat.com>
2019-12-04 18:46:16 +03:00
char *travis_arch;
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
qemu_mutex_init(&data.lock);
data.records = NULL;
tests/test-util-filemonitor: Skip test on non-x86 Travis containers test-util-filemonitor fails in restricted non-x86 Travis containers since they apparently blacklisted some required system calls there. Let's simply skip the test if we detect such an environment. Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Thomas Huth <thuth@redhat.com> Signed-off-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Cleber Rosa <crosa@redhat.com> Tested-by: Cleber Rosa <crosa@redhat.com> Message-Id: <20191204154618.23560-6-thuth@redhat.com>
2019-12-04 18:46:16 +03:00
/*
* This test does not work on Travis LXD containers since some
* syscalls are blocked in that environment.
*/
travis_arch = getenv("TRAVIS_ARCH");
if (travis_arch && !g_str_equal(travis_arch, "x86_64")) {
g_test_skip("Test does not work on non-x86 Travis containers.");
return;
}
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
/*
* The file monitor needs the main loop running in
* order to receive events from inotify. We must
* thus spawn a background thread to run an event
* loop impl, while this thread triggers the
* actual file operations we're testing
*/
evrunning = 1;
evstopping = 0;
qemu_thread_create(&th, "event-loop",
qemu_file_monitor_test_event_loop, NULL,
QEMU_THREAD_JOINABLE);
if (local_err) {
g_printerr("File monitoring not available: %s",
error_get_pretty(local_err));
error_free(local_err);
return;
}
dir = g_dir_make_tmp("test-util-filemonitor-XXXXXX",
&gerr);
if (!dir) {
g_printerr("Unable to create tmp dir %s",
gerr->message);
g_error_free(gerr);
abort();
}
/*
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
* Run through the operation sequence validating events
* as we go
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
*/
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
for (i = 0; i < G_N_ELEMENTS(ops); i++) {
const QFileMonitorTestOp *op = &(ops[i]);
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
int fd;
struct utimbuf ubuf;
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
char *watchdir;
const char *watchfile;
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
pathsrc = g_strdup_printf("%s/%s", dir, op->filesrc);
if (op->filedst) {
pathdst = g_strdup_printf("%s/%s", dir, op->filedst);
}
switch (op->type) {
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
case QFILE_MONITOR_TEST_OP_ADD_WATCH:
if (debug) {
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
g_printerr("Add watch %s %s\n",
dir, op->filesrc);
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
}
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
if (op->filesrc && strchr(op->filesrc, '/')) {
watchdir = g_strdup_printf("%s/%s", dir, op->filesrc);
watchfile = strrchr(watchdir, '/');
*(char *)watchfile = '\0';
watchfile++;
if (*watchfile == '\0') {
watchfile = NULL;
}
} else {
watchdir = g_strdup(dir);
watchfile = op->filesrc;
}
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
*op->watchid =
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
qemu_file_monitor_add_watch(mon,
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
watchdir,
watchfile,
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
qemu_file_monitor_test_handler,
&data,
&local_err);
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
g_free(watchdir);
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
if (*op->watchid < 0) {
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
g_printerr("Unable to add watch %s",
error_get_pretty(local_err));
goto cleanup;
}
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
if (debug) {
g_printerr("Watch ID %" PRIx64 "\n", *op->watchid);
}
if (g_hash_table_contains(ids, op->watchid)) {
g_printerr("Watch ID %" PRIx64 "already exists", *op->watchid);
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
goto cleanup;
}
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
g_hash_table_add(ids, op->watchid);
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
break;
case QFILE_MONITOR_TEST_OP_DEL_WATCH:
if (debug) {
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
g_printerr("Del watch %s %" PRIx64 "\n", dir, *op->watchid);
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
}
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
if (op->filesrc && strchr(op->filesrc, '/')) {
watchdir = g_strdup_printf("%s/%s", dir, op->filesrc);
watchfile = strrchr(watchdir, '/');
*(char *)watchfile = '\0';
} else {
watchdir = g_strdup(dir);
}
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
g_hash_table_remove(ids, op->watchid);
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
qemu_file_monitor_remove_watch(mon,
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
watchdir,
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
*op->watchid);
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
g_free(watchdir);
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
break;
case QFILE_MONITOR_TEST_OP_EVENT:
if (debug) {
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
g_printerr("Event id=%" PRIx64 " event=%d file=%s\n",
*op->watchid, op->eventid, op->filesrc);
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
}
tests: make filemonitor test more robust to event ordering The ordering of events that are emitted during the rmdir test have changed with kernel >= 5.3. Semantically both new & old orderings are correct, so we must be able to cope with either. To cope with this, when we see an unexpected event, we push it back onto the queue and look and the subsequent event to see if that matches instead. Tested-by: Peter Xu <peterx@redhat.com> Tested-by: Wei Yang <richardw.yang@linux.intel.com> Tested-by: Cornelia Huck <cohuck@redhat.com> Reviewed-by: Thomas Huth <thuth@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-08-21 18:14:27 +03:00
if (!qemu_file_monitor_test_expect(&data, *op->watchid,
op->eventid, op->filesrc,
op->swapnext))
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
goto cleanup;
break;
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
case QFILE_MONITOR_TEST_OP_CREATE:
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
if (debug) {
g_printerr("Create %s\n", pathsrc);
}
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
fd = open(pathsrc, O_WRONLY | O_CREAT, 0700);
if (fd < 0) {
g_printerr("Unable to create %s: %s",
pathsrc, strerror(errno));
goto cleanup;
}
close(fd);
break;
case QFILE_MONITOR_TEST_OP_APPEND:
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
if (debug) {
g_printerr("Append %s\n", pathsrc);
}
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
fd = open(pathsrc, O_WRONLY | O_APPEND, 0700);
if (fd < 0) {
g_printerr("Unable to open %s: %s",
pathsrc, strerror(errno));
goto cleanup;
}
if (write(fd, "Hello World", 10) != 10) {
g_printerr("Unable to write %s: %s",
pathsrc, strerror(errno));
close(fd);
goto cleanup;
}
close(fd);
break;
case QFILE_MONITOR_TEST_OP_TRUNC:
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
if (debug) {
g_printerr("Truncate %s\n", pathsrc);
}
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
if (truncate(pathsrc, 4) < 0) {
g_printerr("Unable to truncate %s: %s",
pathsrc, strerror(errno));
goto cleanup;
}
break;
case QFILE_MONITOR_TEST_OP_RENAME:
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
if (debug) {
g_printerr("Rename %s -> %s\n", pathsrc, pathdst);
}
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
if (rename(pathsrc, pathdst) < 0) {
g_printerr("Unable to rename %s to %s: %s",
pathsrc, pathdst, strerror(errno));
goto cleanup;
}
break;
case QFILE_MONITOR_TEST_OP_UNLINK:
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
if (debug) {
g_printerr("Unlink %s\n", pathsrc);
}
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
if (unlink(pathsrc) < 0) {
g_printerr("Unable to unlink %s: %s",
pathsrc, strerror(errno));
goto cleanup;
}
break;
case QFILE_MONITOR_TEST_OP_TOUCH:
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
if (debug) {
g_printerr("Touch %s\n", pathsrc);
}
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
ubuf.actime = 1024;
ubuf.modtime = 1025;
if (utime(pathsrc, &ubuf) < 0) {
g_printerr("Unable to touch %s: %s",
pathsrc, strerror(errno));
goto cleanup;
}
break;
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
case QFILE_MONITOR_TEST_OP_MKDIR:
if (debug) {
g_printerr("Mkdir %s\n", pathsrc);
}
if (mkdir(pathsrc, 0700) < 0) {
g_printerr("Unable to mkdir %s: %s",
pathsrc, strerror(errno));
goto cleanup;
}
break;
case QFILE_MONITOR_TEST_OP_RMDIR:
if (debug) {
g_printerr("Rmdir %s\n", pathsrc);
}
if (rmdir(pathsrc) < 0) {
g_printerr("Unable to rmdir %s: %s",
pathsrc, strerror(errno));
goto cleanup;
}
break;
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
default:
g_assert_not_reached();
}
g_free(pathsrc);
g_free(pathdst);
pathsrc = pathdst = NULL;
}
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
g_assert_cmpint(g_hash_table_size(ids), ==, 0);
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
err = 0;
cleanup:
g_free(pathsrc);
g_free(pathdst);
qemu_mutex_lock(&evlock);
evstopping = 1;
timer = g_timer_new();
while (evrunning && g_timer_elapsed(timer, NULL) < 5) {
qemu_mutex_unlock(&evlock);
usleep(10 * 1000);
qemu_mutex_lock(&evlock);
}
qemu_mutex_unlock(&evlock);
if (g_timer_elapsed(timer, NULL) >= 5) {
g_printerr("Event loop failed to quit after 5 seconds\n");
}
g_timer_destroy(timer);
qemu_file_monitor_free(mon);
g_list_foreach(data.records,
(GFunc)qemu_file_monitor_test_record_free, NULL);
g_list_free(data.records);
qemu_mutex_destroy(&data.lock);
if (dir) {
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
for (i = 0; i < G_N_ELEMENTS(ops); i++) {
const QFileMonitorTestOp *op = &(ops[i]);
char *path = g_strdup_printf("%s/%s",
dir, op->filesrc);
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
if (op->type == QFILE_MONITOR_TEST_OP_MKDIR) {
rmdir(path);
g_free(path);
} else {
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
unlink(path);
g_free(path);
filemon: ensure watch IDs are unique to QFileMonitor scope The watch IDs are mistakenly only unique within the scope of the directory being monitored. This is not useful for clients which are monitoring multiple directories. They require watch IDs to be unique globally within the QFileMonitor scope. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Tested-by: Bandan Das <bsd@redhat.com> Reviewed-by: Bandan Das <bsd@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-13 20:36:18 +03:00
if (op->filedst) {
path = g_strdup_printf("%s/%s",
dir, op->filedst);
unlink(path);
g_free(path);
}
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
}
}
if (rmdir(dir) < 0) {
g_printerr("Failed to remove %s: %s\n",
dir, strerror(errno));
abort();
}
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
}
filemon: fix watch IDs to avoid potential wraparound issues Watch IDs are allocated from incrementing a int counter against the QFileMonitor object. In very long life QEMU processes with a huge amount of USB MTP activity creating & deleting directories it is just about conceivable that the int counter can wrap around. This would result in incorrect behaviour of the file monitor watch APIs due to clashing watch IDs. Instead of trying to detect this situation, this patch changes the way watch IDs are allocated. It is turned into an int64_t variable where the high 32 bits are set from the underlying inotify "int" ID. This gives an ID that is guaranteed unique for the directory as a whole, and we can rely on the kernel to enforce this. QFileMonitor then sets the low 32 bits from a per-directory counter. The USB MTP device only sets watches on the directory as a whole, not files within, so there is no risk of guest triggered wrap around on the low 32 bits. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-19 18:47:47 +03:00
g_hash_table_unref(ids);
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
g_free(dir);
g_assert(err == 0);
}
int main(int argc, char **argv)
{
g_test_init(&argc, &argv, NULL);
qemu_init_main_loop(&error_abort);
qemu_mutex_init(&evlock);
tests: refactor file monitor test to make it more understandable The current file monitor unit tests are too clever for their own good making it hard to understand the desired output. Instead of trying to infer the expected events, explicitly list the events we expect in the operation sequence. Instead of dynamically building a matrix of tests, just have one giant operation sequence that validates all scenarios in a single test. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2019-03-14 17:45:08 +03:00
debug = getenv("FILEMONITOR_DEBUG") != NULL;
g_test_add_func("/util/filemonitor", test_file_monitor_events);
util: add helper APIs for dealing with inotify in portable manner The inotify userspace API for reading events is quite horrible, so it is useful to wrap it in a more friendly API to avoid duplicating code across many users in QEMU. Wrapping it also allows introduction of a platform portability layer, so that we can add impls for non-Linux based equivalents in future. Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
2018-06-08 19:24:57 +03:00
return g_test_run();
}
Reference in New Issue Copy Permalink