mc/lib/vfs/direntry.c

1739 lines
47 KiB
C

/*
Directory cache support
Copyright (C) 1998-2022
Free Software Foundation, Inc.
Written by:
Pavel Machek <pavel@ucw.cz>, 1998
Slava Zanko <slavazanko@gmail.com>, 2010-2013
Andrew Borodin <aborodin@vmail.ru> 2010-2022
This file is part of the Midnight Commander.
The Midnight Commander 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 3 of the License,
or (at your option) any later version.
The Midnight Commander 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 program. If not, see <http://www.gnu.org/licenses/>.
\warning Paths here do _not_ begin with '/', so root directory of
archive/site is simply "".
*/
/** \file
* \brief Source: directory cache support
*
* So that you do not have copy of this in each and every filesystem.
*
* Very loosely based on tar.c from midnight and archives.[ch] from
* avfs by Miklos Szeredi (mszeredi@inf.bme.hu)
*
* Unfortunately, I was unable to keep all filesystems
* uniform. tar-like filesystems use tree structure where each
* directory has pointers to its subdirectories. We can do this
* because we have full information about our archive.
*
* At ftp-like filesystems, situation is a little bit different. When
* you cd /usr/src/linux/drivers/char, you do _not_ want /usr,
* /usr/src, /usr/src/linux and /usr/src/linux/drivers to be
* listed. That means that we do not have complete information, and if
* /usr is symlink to /4, we will not know. Also we have to time out
* entries and things would get messy with tree-like approach. So we
* do different trick: root directory is completely special and
* completely fake, it contains entries such as 'usr', 'usr/src', ...,
* and we'll try to use custom find_entry function.
*
* \author Pavel Machek <pavel@ucw.cz>
* \date 1998
*
*/
#include <config.h>
#include <errno.h>
#include <inttypes.h> /* uintmax_t */
#include <stdarg.h>
#ifdef HAVE_SYS_SELECT_H
#include <sys/select.h>
#endif
#include <sys/types.h>
#include <unistd.h>
#include "lib/global.h"
#include "lib/tty/tty.h" /* enable/disable interrupt key */
#include "lib/util.h" /* canonicalize_pathname_custom() */
#if 0
#include "lib/widget.h" /* message() */
#endif
#include "vfs.h"
#include "utilvfs.h"
#include "xdirentry.h"
#include "gc.h" /* vfs_rmstamp */
/*** global variables ****************************************************************************/
/*** file scope macro definitions ****************************************************************/
#define CALL(x) \
if (VFS_SUBCLASS (me)->x != NULL) \
VFS_SUBCLASS (me)->x
/*** file scope type declarations ****************************************************************/
struct dirhandle
{
GList *cur;
struct vfs_s_inode *dir;
};
/*** file scope variables ************************************************************************/
/* --------------------------------------------------------------------------------------------- */
/*** file scope functions ************************************************************************/
/* --------------------------------------------------------------------------------------------- */
/* We were asked to create entries automagically */
static struct vfs_s_entry *
vfs_s_automake (struct vfs_class *me, struct vfs_s_inode *dir, char *path, int flags)
{
struct vfs_s_entry *res;
char *sep;
sep = strchr (path, PATH_SEP);
if (sep != NULL)
*sep = '\0';
res = vfs_s_generate_entry (me, path, dir, (flags & FL_MKDIR) != 0 ? (0777 | S_IFDIR) : 0777);
vfs_s_insert_entry (me, dir, res);
if (sep != NULL)
*sep = PATH_SEP;
return res;
}
/* --------------------------------------------------------------------------------------------- */
/* If the entry is a symlink, find the entry for its target */
static struct vfs_s_entry *
vfs_s_resolve_symlink (struct vfs_class *me, struct vfs_s_entry *entry, int follow)
{
char *linkname;
char *fullname = NULL;
struct vfs_s_entry *target;
if (follow == LINK_NO_FOLLOW)
return entry;
if (follow == 0)
ERRNOR (ELOOP, NULL);
if (entry == NULL)
ERRNOR (ENOENT, NULL);
if (!S_ISLNK (entry->ino->st.st_mode))
return entry;
linkname = entry->ino->linkname;
if (linkname == NULL)
ERRNOR (EFAULT, NULL);
/* make full path from relative */
if (!IS_PATH_SEP (*linkname))
{
char *fullpath;
fullpath = vfs_s_fullpath (me, entry->dir);
if (fullpath != NULL)
{
fullname = g_strconcat (fullpath, PATH_SEP_STR, linkname, (char *) NULL);
linkname = fullname;
g_free (fullpath);
}
}
target =
VFS_SUBCLASS (me)->find_entry (me, entry->dir->super->root, linkname, follow - 1, FL_NONE);
g_free (fullname);
return target;
}
/* --------------------------------------------------------------------------------------------- */
/*
* Follow > 0: follow links, serves as loop protect,
* == -1: do not follow links
*/
static struct vfs_s_entry *
vfs_s_find_entry_tree (struct vfs_class *me, struct vfs_s_inode *root,
const char *a_path, int follow, int flags)
{
size_t pseg;
struct vfs_s_entry *ent = NULL;
char *const pathref = g_strdup (a_path);
char *path = pathref;
/* canonicalize as well, but don't remove '../' from path */
canonicalize_pathname_custom (path, CANON_PATH_ALL & (~CANON_PATH_REMDOUBLEDOTS));
while (root != NULL)
{
GList *iter;
while (IS_PATH_SEP (*path)) /* Strip leading '/' */
path++;
if (path[0] == '\0')
{
g_free (pathref);
return ent;
}
for (pseg = 0; path[pseg] != '\0' && !IS_PATH_SEP (path[pseg]); pseg++)
;
for (iter = g_queue_peek_head_link (root->subdir); iter != NULL; iter = g_list_next (iter))
{
ent = VFS_ENTRY (iter->data);
if (strlen (ent->name) == pseg && strncmp (ent->name, path, pseg) == 0)
/* FOUND! */
break;
}
ent = iter != NULL ? VFS_ENTRY (iter->data) : NULL;
if (ent == NULL && (flags & (FL_MKFILE | FL_MKDIR)) != 0)
ent = vfs_s_automake (me, root, path, flags);
if (ent == NULL)
{
me->verrno = ENOENT;
goto cleanup;
}
path += pseg;
/* here we must follow leading directories always;
only the actual file is optional */
ent = vfs_s_resolve_symlink (me, ent,
strchr (path, PATH_SEP) != NULL ? LINK_FOLLOW : follow);
if (ent == NULL)
goto cleanup;
root = ent->ino;
}
cleanup:
g_free (pathref);
return NULL;
}
/* --------------------------------------------------------------------------------------------- */
static struct vfs_s_entry *
vfs_s_find_entry_linear (struct vfs_class *me, struct vfs_s_inode *root,
const char *a_path, int follow, int flags)
{
struct vfs_s_entry *ent = NULL;
char *const path = g_strdup (a_path);
GList *iter;
if (root->super->root != root)
vfs_die ("We have to use _real_ root. Always. Sorry.");
/* canonicalize as well, but don't remove '../' from path */
canonicalize_pathname_custom (path, CANON_PATH_ALL & (~CANON_PATH_REMDOUBLEDOTS));
if ((flags & FL_DIR) == 0)
{
char *dirname, *name;
struct vfs_s_inode *ino;
dirname = g_path_get_dirname (path);
name = g_path_get_basename (path);
ino = vfs_s_find_inode (me, root->super, dirname, follow, flags | FL_DIR);
ent = vfs_s_find_entry_tree (me, ino, name, follow, flags);
g_free (dirname);
g_free (name);
g_free (path);
return ent;
}
iter = g_queue_find_custom (root->subdir, path, (GCompareFunc) vfs_s_entry_compare);
ent = iter != NULL ? VFS_ENTRY (iter->data) : NULL;
if (ent != NULL && !VFS_SUBCLASS (me)->dir_uptodate (me, ent->ino))
{
#if 1
vfs_print_message (_("Directory cache expired for %s"), path);
#endif
vfs_s_free_entry (me, ent);
ent = NULL;
}
if (ent == NULL)
{
struct vfs_s_inode *ino;
ino = vfs_s_new_inode (me, root->super, vfs_s_default_stat (me, S_IFDIR | 0755));
ent = vfs_s_new_entry (me, path, ino);
if (VFS_SUBCLASS (me)->dir_load (me, ino, path) == -1)
{
vfs_s_free_entry (me, ent);
g_free (path);
return NULL;
}
vfs_s_insert_entry (me, root, ent);
iter = g_queue_find_custom (root->subdir, path, (GCompareFunc) vfs_s_entry_compare);
ent = iter != NULL ? VFS_ENTRY (iter->data) : NULL;
}
if (ent == NULL)
vfs_die ("find_linear: success but directory is not there\n");
#if 0
if (vfs_s_resolve_symlink (me, ent, follow) == NULL)
{
g_free (path);
return NULL;
}
#endif
g_free (path);
return ent;
}
/* --------------------------------------------------------------------------------------------- */
/* Ook, these were functions around directory entries / inodes */
/* -------------------------------- superblock games -------------------------- */
static struct vfs_s_super *
vfs_s_new_super (struct vfs_class *me)
{
struct vfs_s_super *super;
super = g_new0 (struct vfs_s_super, 1);
super->me = me;
return super;
}
/* --------------------------------------------------------------------------------------------- */
static inline void
vfs_s_insert_super (struct vfs_class *me, struct vfs_s_super *super)
{
VFS_SUBCLASS (me)->supers = g_list_prepend (VFS_SUBCLASS (me)->supers, super);
}
/* --------------------------------------------------------------------------------------------- */
static void
vfs_s_free_super (struct vfs_class *me, struct vfs_s_super *super)
{
if (super->root != NULL)
{
vfs_s_free_inode (me, super->root);
super->root = NULL;
}
#if 0
/* FIXME: We currently leak small ammount of memory, sometimes. Fix it if you can. */
if (super->ino_usage != 0)
message (D_ERROR, "Direntry warning",
"Super ino_usage is %d, memory leak", super->ino_usage);
if (super->want_stale)
message (D_ERROR, "Direntry warning", "%s", "Super has want_stale set");
#endif
VFS_SUBCLASS (me)->supers = g_list_remove (VFS_SUBCLASS (me)->supers, super);
CALL (free_archive) (me, super);
#ifdef ENABLE_VFS_NET
vfs_path_element_free (super->path_element);
#endif
g_free (super->name);
g_free (super);
}
/* --------------------------------------------------------------------------------------------- */
static vfs_file_handler_t *
vfs_s_new_fh (struct vfs_s_inode *ino, gboolean changed)
{
vfs_file_handler_t *fh;
fh = g_new0 (vfs_file_handler_t, 1);
vfs_s_init_fh (fh, ino, changed);
return fh;
}
/* --------------------------------------------------------------------------------------------- */
static void
vfs_s_free_fh (struct vfs_s_subclass *s, vfs_file_handler_t * fh)
{
if (s->fh_free != NULL)
s->fh_free (fh);
g_free (fh);
}
/* --------------------------------------------------------------------------------------------- */
/* Support of archives */
/* ------------------------ readdir & friends ----------------------------- */
static struct vfs_s_inode *
vfs_s_inode_from_path (const vfs_path_t * vpath, int flags)
{
struct vfs_s_super *super;
struct vfs_s_inode *ino;
const char *q;
const vfs_path_element_t *path_element;
q = vfs_s_get_path (vpath, &super, 0);
if (q == NULL)
return NULL;
path_element = vfs_path_get_by_index (vpath, -1);
ino =
vfs_s_find_inode (path_element->class, super, q,
(flags & FL_FOLLOW) != 0 ? LINK_FOLLOW : LINK_NO_FOLLOW,
flags & ~FL_FOLLOW);
if (ino == NULL && *q == '\0')
/* We are asking about / directory of ftp server: assume it exists */
ino =
vfs_s_find_inode (path_element->class, super, q,
(flags & FL_FOLLOW) != 0 ? LINK_FOLLOW : LINK_NO_FOLLOW,
FL_DIR | (flags & ~FL_FOLLOW));
return ino;
}
/* --------------------------------------------------------------------------------------------- */
static void *
vfs_s_opendir (const vfs_path_t * vpath)
{
struct vfs_s_inode *dir;
struct dirhandle *info;
const vfs_path_element_t *path_element;
dir = vfs_s_inode_from_path (vpath, FL_DIR | FL_FOLLOW);
if (dir == NULL)
return NULL;
path_element = vfs_path_get_by_index (vpath, -1);
if (!S_ISDIR (dir->st.st_mode))
{
path_element->class->verrno = ENOTDIR;
return NULL;
}
dir->st.st_nlink++;
#if 0
if (dir->subdir == NULL) /* This can actually happen if we allow empty directories */
{
path_element->class->verrno = EAGAIN;
return NULL;
}
#endif
info = g_new (struct dirhandle, 1);
info->cur = g_queue_peek_head_link (dir->subdir);
info->dir = dir;
return info;
}
/* --------------------------------------------------------------------------------------------- */
static struct vfs_dirent *
vfs_s_readdir (void *data)
{
struct vfs_dirent *dir = NULL;
struct dirhandle *info = (struct dirhandle *) data;
const char *name;
if (info->cur == NULL || info->cur->data == NULL)
return NULL;
name = VFS_ENTRY (info->cur->data)->name;
if (name != NULL)
dir = vfs_dirent_init (NULL, name, 0);
else
vfs_die ("Null in structure-cannot happen");
info->cur = g_list_next (info->cur);
return dir;
}
/* --------------------------------------------------------------------------------------------- */
static int
vfs_s_closedir (void *data)
{
struct dirhandle *info = (struct dirhandle *) data;
struct vfs_s_inode *dir = info->dir;
vfs_s_free_inode (dir->super->me, dir);
g_free (data);
return 0;
}
/* --------------------------------------------------------------------------------------------- */
static int
vfs_s_chdir (const vfs_path_t * vpath)
{
void *data;
data = vfs_s_opendir (vpath);
if (data == NULL)
return (-1);
vfs_s_closedir (data);
return 0;
}
/* --------------------------------------------------------------------------------------------- */
/* --------------------------- stat and friends ---------------------------- */
static int
vfs_s_internal_stat (const vfs_path_t * vpath, struct stat *buf, int flag)
{
struct vfs_s_inode *ino;
ino = vfs_s_inode_from_path (vpath, flag);
if (ino == NULL)
return (-1);
*buf = ino->st;
return 0;
}
/* --------------------------------------------------------------------------------------------- */
static int
vfs_s_readlink (const vfs_path_t * vpath, char *buf, size_t size)
{
struct vfs_s_inode *ino;
size_t len;
const vfs_path_element_t *path_element;
ino = vfs_s_inode_from_path (vpath, 0);
if (ino == NULL)
return (-1);
path_element = vfs_path_get_by_index (vpath, -1);
if (!S_ISLNK (ino->st.st_mode))
{
path_element->class->verrno = EINVAL;
return (-1);
}
if (ino->linkname == NULL)
{
path_element->class->verrno = EFAULT;
return (-1);
}
len = strlen (ino->linkname);
if (size < len)
len = size;
/* readlink() does not append a NUL character to buf */
memcpy (buf, ino->linkname, len);
return len;
}
/* --------------------------------------------------------------------------------------------- */
static ssize_t
vfs_s_read (void *fh, char *buffer, size_t count)
{
vfs_file_handler_t *file = VFS_FILE_HANDLER (fh);
struct vfs_class *me = VFS_FILE_HANDLER_SUPER (fh)->me;
if (file->linear == LS_LINEAR_PREOPEN)
{
if (VFS_SUBCLASS (me)->linear_start (me, file, file->pos) == 0)
return (-1);
}
if (file->linear == LS_LINEAR_CLOSED)
vfs_die ("linear_start() did not set linear_state!");
if (file->linear == LS_LINEAR_OPEN)
return VFS_SUBCLASS (me)->linear_read (me, file, buffer, count);
if (file->handle != -1)
{
ssize_t n;
n = read (file->handle, buffer, count);
if (n < 0)
me->verrno = errno;
return n;
}
vfs_die ("vfs_s_read: This should not happen\n");
return (-1);
}
/* --------------------------------------------------------------------------------------------- */
static ssize_t
vfs_s_write (void *fh, const char *buffer, size_t count)
{
vfs_file_handler_t *file = VFS_FILE_HANDLER (fh);
struct vfs_class *me = VFS_FILE_HANDLER_SUPER (fh)->me;
if (file->linear != LS_NOT_LINEAR)
vfs_die ("no writing to linear files, please");
file->changed = TRUE;
if (file->handle != -1)
{
ssize_t n;
n = write (file->handle, buffer, count);
if (n < 0)
me->verrno = errno;
return n;
}
vfs_die ("vfs_s_write: This should not happen\n");
return 0;
}
/* --------------------------------------------------------------------------------------------- */
static off_t
vfs_s_lseek (void *fh, off_t offset, int whence)
{
vfs_file_handler_t *file = VFS_FILE_HANDLER (fh);
off_t size = file->ino->st.st_size;
if (file->linear == LS_LINEAR_OPEN)
vfs_die ("cannot lseek() after linear_read!");
if (file->handle != -1)
{ /* If we have local file opened, we want to work with it */
off_t retval;
retval = lseek (file->handle, offset, whence);
if (retval == -1)
VFS_FILE_HANDLER_SUPER (fh)->me->verrno = errno;
return retval;
}
switch (whence)
{
case SEEK_CUR:
offset += file->pos;
break;
case SEEK_END:
offset += size;
break;
default:
break;
}
if (offset < 0)
file->pos = 0;
else if (offset < size)
file->pos = offset;
else
file->pos = size;
return file->pos;
}
/* --------------------------------------------------------------------------------------------- */
static int
vfs_s_close (void *fh)
{
vfs_file_handler_t *file = VFS_FILE_HANDLER (fh);
struct vfs_s_super *super = VFS_FILE_HANDLER_SUPER (fh);
struct vfs_class *me = super->me;
struct vfs_s_subclass *sub = VFS_SUBCLASS (me);
int res = 0;
if (me == NULL)
return (-1);
super->fd_usage--;
if (super->fd_usage == 0)
vfs_stamp_create (me, VFS_FILE_HANDLER_SUPER (fh));
if (file->linear == LS_LINEAR_OPEN)
sub->linear_close (me, fh);
if (sub->fh_close != NULL)
res = sub->fh_close (me, fh);
if ((me->flags & VFSF_USETMP) != 0 && file->changed && sub->file_store != NULL)
{
char *s;
s = vfs_s_fullpath (me, file->ino);
if (s == NULL)
res = -1;
else
{
res = sub->file_store (me, fh, s, file->ino->localname);
g_free (s);
}
vfs_s_invalidate (me, super);
}
if (file->handle != -1)
{
close (file->handle);
file->handle = -1;
}
vfs_s_free_inode (me, file->ino);
vfs_s_free_fh (sub, fh);
return res;
}
/* --------------------------------------------------------------------------------------------- */
static void
vfs_s_print_stats (const char *fs_name, const char *action,
const char *file_name, off_t have, off_t need)
{
if (need != 0)
vfs_print_message (_("%s: %s: %s %3d%% (%lld) bytes transferred"), fs_name, action,
file_name, (int) ((double) have * 100 / need), (long long) have);
else
vfs_print_message (_("%s: %s: %s %lld bytes transferred"), fs_name, action, file_name,
(long long) have);
}
/* --------------------------------------------------------------------------------------------- */
/* ------------------------------- mc support ---------------------------- */
static void
vfs_s_fill_names (struct vfs_class *me, fill_names_f func)
{
GList *iter;
for (iter = VFS_SUBCLASS (me)->supers; iter != NULL; iter = g_list_next (iter))
{
const struct vfs_s_super *super = (const struct vfs_s_super *) iter->data;
char *name;
name = g_strconcat (super->name, PATH_SEP_STR, me->prefix, VFS_PATH_URL_DELIMITER,
/* super->current_dir->name, */ (char *) NULL);
func (name);
g_free (name);
}
}
/* --------------------------------------------------------------------------------------------- */
static int
vfs_s_ferrno (struct vfs_class *me)
{
return me->verrno;
}
/* --------------------------------------------------------------------------------------------- */
/**
* Get local copy of the given file. We reuse the existing file cache
* for remote filesystems. Archives use standard VFS facilities.
*/
static vfs_path_t *
vfs_s_getlocalcopy (const vfs_path_t * vpath)
{
vfs_file_handler_t *fh;
vfs_path_t *local = NULL;
if (vpath == NULL)
return NULL;
fh = vfs_s_open (vpath, O_RDONLY, 0);
if (fh != NULL)
{
const struct vfs_class *me;
me = vfs_path_get_by_index (vpath, -1)->class;
if ((me->flags & VFSF_USETMP) != 0 && fh->ino != NULL)
local = vfs_path_from_str_flags (fh->ino->localname, VPF_NO_CANON);
vfs_s_close (fh);
}
return local;
}
/* --------------------------------------------------------------------------------------------- */
/**
* Return the local copy. Since we are using our cache, we do nothing -
* the cache will be removed when the archive is closed.
*/
static int
vfs_s_ungetlocalcopy (const vfs_path_t * vpath, const vfs_path_t * local, gboolean has_changed)
{
(void) vpath;
(void) local;
(void) has_changed;
return 0;
}
/* --------------------------------------------------------------------------------------------- */
static int
vfs_s_setctl (const vfs_path_t * vpath, int ctlop, void *arg)
{
const vfs_path_element_t *path_element;
path_element = vfs_path_get_by_index (vpath, -1);
switch (ctlop)
{
case VFS_SETCTL_STALE_DATA:
{
struct vfs_s_inode *ino;
ino = vfs_s_inode_from_path (vpath, 0);
if (ino == NULL)
return 0;
if (arg != NULL)
ino->super->want_stale = TRUE;
else
{
ino->super->want_stale = FALSE;
vfs_s_invalidate (path_element->class, ino->super);
}
return 1;
}
case VFS_SETCTL_LOGFILE:
path_element->class->logfile = fopen ((char *) arg, "w");
return 1;
case VFS_SETCTL_FLUSH:
path_element->class->flush = TRUE;
return 1;
default:
return 0;
}
}
/* --------------------------------------------------------------------------------------------- */
/* ----------------------------- Stamping support -------------------------- */
static vfsid
vfs_s_getid (const vfs_path_t * vpath)
{
struct vfs_s_super *archive = NULL;
const char *p;
p = vfs_s_get_path (vpath, &archive, FL_NO_OPEN);
if (p == NULL)
return NULL;
return (vfsid) archive;
}
/* --------------------------------------------------------------------------------------------- */
static gboolean
vfs_s_nothingisopen (vfsid id)
{
return (VFS_SUPER (id)->fd_usage <= 0);
}
/* --------------------------------------------------------------------------------------------- */
static void
vfs_s_free (vfsid id)
{
vfs_s_free_super (VFS_SUPER (id)->me, VFS_SUPER (id));
}
/* --------------------------------------------------------------------------------------------- */
static gboolean
vfs_s_dir_uptodate (struct vfs_class *me, struct vfs_s_inode *ino)
{
gint64 tim;
if (me->flush)
{
me->flush = FALSE;
return 0;
}
tim = g_get_monotonic_time ();
return (tim < ino->timestamp);
}
/* --------------------------------------------------------------------------------------------- */
/*** public functions ****************************************************************************/
/* --------------------------------------------------------------------------------------------- */
struct vfs_s_inode *
vfs_s_new_inode (struct vfs_class *me, struct vfs_s_super *super, struct stat *initstat)
{
struct vfs_s_inode *ino;
ino = g_try_new0 (struct vfs_s_inode, 1);
if (ino == NULL)
return NULL;
if (initstat != NULL)
ino->st = *initstat;
ino->super = super;
ino->subdir = g_queue_new ();
ino->st.st_nlink = 0;
ino->st.st_ino = VFS_SUBCLASS (me)->inode_counter++;
ino->st.st_dev = VFS_SUBCLASS (me)->rdev;
super->ino_usage++;
CALL (init_inode) (me, ino);
return ino;
}
/* --------------------------------------------------------------------------------------------- */
void
vfs_s_free_inode (struct vfs_class *me, struct vfs_s_inode *ino)
{
if (ino == NULL)
vfs_die ("Don't pass NULL to me");
/* ==0 can happen if freshly created entry is deleted */
if (ino->st.st_nlink > 1)
{
ino->st.st_nlink--;
return;
}
while (g_queue_get_length (ino->subdir) != 0)
{
struct vfs_s_entry *entry;
entry = VFS_ENTRY (g_queue_peek_head (ino->subdir));
vfs_s_free_entry (me, entry);
}
g_queue_free (ino->subdir);
ino->subdir = NULL;
CALL (free_inode) (me, ino);
g_free (ino->linkname);
if ((me->flags & VFSF_USETMP) != 0 && ino->localname != NULL)
{
unlink (ino->localname);
g_free (ino->localname);
}
ino->super->ino_usage--;
g_free (ino);
}
/* --------------------------------------------------------------------------------------------- */
struct vfs_s_entry *
vfs_s_new_entry (struct vfs_class *me, const char *name, struct vfs_s_inode *inode)
{
struct vfs_s_entry *entry;
entry = g_new0 (struct vfs_s_entry, 1);
entry->name = g_strdup (name);
entry->ino = inode;
entry->ino->ent = entry;
CALL (init_entry) (me, entry);
return entry;
}
/* --------------------------------------------------------------------------------------------- */
void
vfs_s_free_entry (struct vfs_class *me, struct vfs_s_entry *ent)
{
if (ent->dir != NULL)
g_queue_remove (ent->dir->subdir, ent);
MC_PTR_FREE (ent->name);
if (ent->ino != NULL)
{
ent->ino->ent = NULL;
vfs_s_free_inode (me, ent->ino);
}
g_free (ent);
}
/* --------------------------------------------------------------------------------------------- */
void
vfs_s_insert_entry (struct vfs_class *me, struct vfs_s_inode *dir, struct vfs_s_entry *ent)
{
(void) me;
ent->dir = dir;
ent->ino->st.st_nlink++;
g_queue_push_tail (dir->subdir, ent);
}
/* --------------------------------------------------------------------------------------------- */
int
vfs_s_entry_compare (const void *a, const void *b)
{
const struct vfs_s_entry *e = (const struct vfs_s_entry *) a;
const char *name = (const char *) b;
return strcmp (e->name, name);
}
/* --------------------------------------------------------------------------------------------- */
struct stat *
vfs_s_default_stat (struct vfs_class *me, mode_t mode)
{
static struct stat st;
mode_t myumask;
(void) me;
myumask = umask (022);
umask (myumask);
mode &= ~myumask;
st.st_mode = mode;
st.st_ino = 0;
st.st_dev = 0;
#ifdef HAVE_STRUCT_STAT_ST_RDEV
st.st_rdev = 0;
#endif
st.st_uid = getuid ();
st.st_gid = getgid ();
#ifdef HAVE_STRUCT_STAT_ST_BLKSIZE
st.st_blksize = 512;
#endif
st.st_size = 0;
st.st_mtime = st.st_atime = st.st_ctime = time (NULL);
#ifdef HAVE_STRUCT_STAT_ST_MTIM
st.st_atim.tv_nsec = st.st_mtim.tv_nsec = st.st_ctim.tv_nsec = 0;
#endif
vfs_adjust_stat (&st);
return &st;
}
/* --------------------------------------------------------------------------------------------- */
/**
* Calculate number of st_blocks using st_size and st_blksize.
* In according to stat(2), st_blocks is the size in 512-byte units.
*
* @param s stat info
*/
void
vfs_adjust_stat (struct stat *s)
{
#ifdef HAVE_STRUCT_STAT_ST_BLOCKS
if (s->st_size == 0)
s->st_blocks = 0;
else
{
#ifdef HAVE_STRUCT_STAT_ST_BLKSIZE
blkcnt_t ioblocks;
blksize_t ioblock_size;
/* 1. Calculate how many IO blocks are occupied */
ioblocks = 1 + (s->st_size - 1) / s->st_blksize;
/* 2. Calculate size of st_blksize in 512-byte units */
ioblock_size = 1 + (s->st_blksize - 1) / 512;
/* 3. Calculate number of blocks */
s->st_blocks = ioblocks * ioblock_size;
#else
/* Let IO block size is 512 bytes */
s->st_blocks = 1 + (s->st_size - 1) / 512;
#endif /* HAVE_STRUCT_STAT_ST_BLKSIZE */
}
#endif /* HAVE_STRUCT_STAT_ST_BLOCKS */
}
/* --------------------------------------------------------------------------------------------- */
struct vfs_s_entry *
vfs_s_generate_entry (struct vfs_class *me, const char *name, struct vfs_s_inode *parent,
mode_t mode)
{
struct vfs_s_inode *inode;
struct stat *st;
st = vfs_s_default_stat (me, mode);
inode = vfs_s_new_inode (me, parent->super, st);
return vfs_s_new_entry (me, name, inode);
}
/* --------------------------------------------------------------------------------------------- */
struct vfs_s_inode *
vfs_s_find_inode (struct vfs_class *me, const struct vfs_s_super *super,
const char *path, int follow, int flags)
{
struct vfs_s_entry *ent;
if (((me->flags & VFSF_REMOTE) == 0) && (*path == '\0'))
return super->root;
ent = VFS_SUBCLASS (me)->find_entry (me, super->root, path, follow, flags);
return (ent != NULL ? ent->ino : NULL);
}
/* --------------------------------------------------------------------------------------------- */
/* Ook, these were functions around directory entries / inodes */
/* -------------------------------- superblock games -------------------------- */
/**
* get superlock object by vpath
*
* @param vpath path
* @return superlock object or NULL if not found
*/
struct vfs_s_super *
vfs_get_super_by_vpath (const vfs_path_t * vpath)
{
GList *iter;
void *cookie = NULL;
const vfs_path_element_t *path_element;
struct vfs_s_subclass *subclass;
struct vfs_s_super *super = NULL;
vfs_path_t *vpath_archive;
path_element = vfs_path_get_by_index (vpath, -1);
subclass = VFS_SUBCLASS (path_element->class);
vpath_archive = vfs_path_clone (vpath);
vfs_path_remove_element_by_index (vpath_archive, -1);
if (subclass->archive_check != NULL)
{
cookie = subclass->archive_check (vpath_archive);
if (cookie == NULL)
goto ret;
}
if (subclass->archive_same == NULL)
goto ret;
for (iter = subclass->supers; iter != NULL; iter = g_list_next (iter))
{
int i;
super = VFS_SUPER (iter->data);
/* 0 == other, 1 == same, return it, 2 == other but stop scanning */
i = subclass->archive_same (path_element, super, vpath_archive, cookie);
if (i == 1)
goto ret;
if (i != 0)
break;
super = NULL;
}
ret:
vfs_path_free (vpath_archive, TRUE);
return super;
}
/* --------------------------------------------------------------------------------------------- */
/**
* get path from last VFS-element and create corresponding superblock
*
* @param vpath source path object
* @param archive pointer to object for store newly created superblock
* @param flags flags
*
* @return path from last VFS-element
*/
const char *
vfs_s_get_path (const vfs_path_t * vpath, struct vfs_s_super **archive, int flags)
{
const char *retval = "";
int result = -1;
struct vfs_s_super *super;
const vfs_path_element_t *path_element;
struct vfs_s_subclass *subclass;
path_element = vfs_path_get_by_index (vpath, -1);
if (path_element->path != NULL)
retval = path_element->path;
super = vfs_get_super_by_vpath (vpath);
if (super != NULL)
goto return_success;
if ((flags & FL_NO_OPEN) != 0)
{
path_element->class->verrno = EIO;
return NULL;
}
subclass = VFS_SUBCLASS (path_element->class);
super = subclass->new_archive != NULL ?
subclass->new_archive (path_element->class) : vfs_s_new_super (path_element->class);
if (subclass->open_archive != NULL)
{
vfs_path_t *vpath_archive;
vpath_archive = vfs_path_clone (vpath);
vfs_path_remove_element_by_index (vpath_archive, -1);
result = subclass->open_archive (super, vpath_archive, path_element);
vfs_path_free (vpath_archive, TRUE);
}
if (result == -1)
{
vfs_s_free_super (path_element->class, super);
path_element->class->verrno = EIO;
return NULL;
}
if (super->name == NULL)
vfs_die ("You have to fill name\n");
if (super->root == NULL)
vfs_die ("You have to fill root inode\n");
vfs_s_insert_super (path_element->class, super);
vfs_stamp_create (path_element->class, super);
return_success:
*archive = super;
return retval;
}
/* --------------------------------------------------------------------------------------------- */
void
vfs_s_invalidate (struct vfs_class *me, struct vfs_s_super *super)
{
if (!super->want_stale)
{
vfs_s_free_inode (me, super->root);
super->root = vfs_s_new_inode (me, super, vfs_s_default_stat (me, S_IFDIR | 0755));
}
}
/* --------------------------------------------------------------------------------------------- */
char *
vfs_s_fullpath (struct vfs_class *me, struct vfs_s_inode *ino)
{
if (ino->ent == NULL)
ERRNOR (EAGAIN, NULL);
if ((me->flags & VFSF_USETMP) == 0)
{
/* archives */
char *path;
path = g_strdup (ino->ent->name);
while (TRUE)
{
char *newpath;
ino = ino->ent->dir;
if (ino == ino->super->root)
break;
newpath = g_strconcat (ino->ent->name, PATH_SEP_STR, path, (char *) NULL);
g_free (path);
path = newpath;
}
return path;
}
/* remote systems */
if (ino->ent->dir == NULL || ino->ent->dir->ent == NULL)
return g_strdup (ino->ent->name);
return g_strconcat (ino->ent->dir->ent->name, PATH_SEP_STR, ino->ent->name, (char *) NULL);
}
/* --------------------------------------------------------------------------------------------- */
void
vfs_s_init_fh (vfs_file_handler_t * fh, struct vfs_s_inode *ino, gboolean changed)
{
fh->ino = ino;
fh->handle = -1;
fh->changed = changed;
fh->linear = LS_NOT_LINEAR;
}
/* --------------------------------------------------------------------------------------------- */
/* --------------------------- stat and friends ---------------------------- */
void *
vfs_s_open (const vfs_path_t * vpath, int flags, mode_t mode)
{
gboolean was_changed = FALSE;
vfs_file_handler_t *fh;
struct vfs_s_super *super;
const char *q;
struct vfs_s_inode *ino;
const vfs_path_element_t *path_element;
struct vfs_s_subclass *s;
q = vfs_s_get_path (vpath, &super, 0);
if (q == NULL)
return NULL;
path_element = vfs_path_get_by_index (vpath, -1);
ino = vfs_s_find_inode (path_element->class, super, q, LINK_FOLLOW, FL_NONE);
if (ino != NULL && (flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
{
path_element->class->verrno = EEXIST;
return NULL;
}
s = VFS_SUBCLASS (path_element->class);
if (ino == NULL)
{
char *dirname, *name;
struct vfs_s_entry *ent;
struct vfs_s_inode *dir;
/* If the filesystem is read-only, disable file creation */
if ((flags & O_CREAT) == 0 || path_element->class->write == NULL)
return NULL;
dirname = g_path_get_dirname (q);
dir = vfs_s_find_inode (path_element->class, super, dirname, LINK_FOLLOW, FL_DIR);
g_free (dirname);
if (dir == NULL)
return NULL;
name = g_path_get_basename (q);
ent = vfs_s_generate_entry (path_element->class, name, dir, 0755);
ino = ent->ino;
vfs_s_insert_entry (path_element->class, dir, ent);
if ((VFS_CLASS (s)->flags & VFSF_USETMP) != 0)
{
int tmp_handle;
vfs_path_t *tmp_vpath;
tmp_handle = vfs_mkstemps (&tmp_vpath, path_element->class->name, name);
ino->localname = vfs_path_free (tmp_vpath, FALSE);
if (tmp_handle == -1)
{
g_free (name);
return NULL;
}
close (tmp_handle);
}
g_free (name);
was_changed = TRUE;
}
if (S_ISDIR (ino->st.st_mode))
{
path_element->class->verrno = EISDIR;
return NULL;
}
fh = s->fh_new != NULL ? s->fh_new (ino, was_changed) : vfs_s_new_fh (ino, was_changed);
if (IS_LINEAR (flags))
{
if (s->linear_start != NULL)
{
vfs_print_message ("%s", _("Starting linear transfer..."));
fh->linear = LS_LINEAR_PREOPEN;
}
}
else
{
if (s->fh_open != NULL && s->fh_open (path_element->class, fh, flags, mode) != 0)
{
vfs_s_free_fh (s, fh);
return NULL;
}
}
if ((VFS_CLASS (s)->flags & VFSF_USETMP) != 0 && fh->ino->localname != NULL)
{
fh->handle = open (fh->ino->localname, NO_LINEAR (flags), mode);
if (fh->handle == -1)
{
vfs_s_free_fh (s, fh);
path_element->class->verrno = errno;
return NULL;
}
}
/* i.e. we had no open files and now we have one */
vfs_rmstamp (path_element->class, (vfsid) super);
super->fd_usage++;
fh->ino->st.st_nlink++;
return fh;
}
/* --------------------------------------------------------------------------------------------- */
int
vfs_s_stat (const vfs_path_t * vpath, struct stat *buf)
{
return vfs_s_internal_stat (vpath, buf, FL_FOLLOW);
}
/* --------------------------------------------------------------------------------------------- */
int
vfs_s_lstat (const vfs_path_t * vpath, struct stat *buf)
{
return vfs_s_internal_stat (vpath, buf, FL_NONE);
}
/* --------------------------------------------------------------------------------------------- */
int
vfs_s_fstat (void *fh, struct stat *buf)
{
*buf = VFS_FILE_HANDLER (fh)->ino->st;
return 0;
}
/* --------------------------------------------------------------------------------------------- */
int
vfs_s_retrieve_file (struct vfs_class *me, struct vfs_s_inode *ino)
{
/* If you want reget, you'll have to open file with O_LINEAR */
off_t total = 0;
char buffer[8192];
int handle;
ssize_t n;
off_t stat_size = ino->st.st_size;
vfs_file_handler_t *fh = NULL;
vfs_path_t *tmp_vpath;
struct vfs_s_subclass *s = VFS_SUBCLASS (me);
if ((me->flags & VFSF_USETMP) == 0)
return (-1);
handle = vfs_mkstemps (&tmp_vpath, me->name, ino->ent->name);
ino->localname = vfs_path_free (tmp_vpath, FALSE);
if (handle == -1)
{
me->verrno = errno;
goto error_4;
}
fh = s->fh_new != NULL ? s->fh_new (ino, FALSE) : vfs_s_new_fh (ino, FALSE);
if (s->linear_start (me, fh, 0) == 0)
goto error_3;
/* Clear the interrupt status */
tty_got_interrupt ();
tty_enable_interrupt_key ();
while ((n = s->linear_read (me, fh, buffer, sizeof (buffer))) != 0)
{
int t;
if (n < 0)
goto error_1;
total += n;
vfs_s_print_stats (me->name, _("Getting file"), ino->ent->name, total, stat_size);
if (tty_got_interrupt ())
goto error_1;
t = write (handle, buffer, n);
if (t != n)
{
if (t == -1)
me->verrno = errno;
goto error_1;
}
}
s->linear_close (me, fh);
close (handle);
tty_disable_interrupt_key ();
vfs_s_free_fh (s, fh);
return 0;
error_1:
s->linear_close (me, fh);
error_3:
tty_disable_interrupt_key ();
close (handle);
unlink (ino->localname);
error_4:
MC_PTR_FREE (ino->localname);
if (fh != NULL)
vfs_s_free_fh (s, fh);
return (-1);
}
/* --------------------------------------------------------------------------------------------- */
/* ----------------------------- Stamping support -------------------------- */
/* Initialize one of our subclasses - fill common functions */
void
vfs_init_class (struct vfs_class *vclass, const char *name, vfs_flags_t flags, const char *prefix)
{
memset (vclass, 0, sizeof (struct vfs_class));
vclass->name = name;
vclass->flags = flags;
vclass->prefix = prefix;
vclass->fill_names = vfs_s_fill_names;
vclass->open = vfs_s_open;
vclass->close = vfs_s_close;
vclass->read = vfs_s_read;
if ((vclass->flags & VFSF_READONLY) == 0)
vclass->write = vfs_s_write;
vclass->opendir = vfs_s_opendir;
vclass->readdir = vfs_s_readdir;
vclass->closedir = vfs_s_closedir;
vclass->stat = vfs_s_stat;
vclass->lstat = vfs_s_lstat;
vclass->fstat = vfs_s_fstat;
vclass->readlink = vfs_s_readlink;
vclass->chdir = vfs_s_chdir;
vclass->ferrno = vfs_s_ferrno;
vclass->lseek = vfs_s_lseek;
vclass->getid = vfs_s_getid;
vclass->nothingisopen = vfs_s_nothingisopen;
vclass->free = vfs_s_free;
vclass->setctl = vfs_s_setctl;
if ((vclass->flags & VFSF_USETMP) != 0)
{
vclass->getlocalcopy = vfs_s_getlocalcopy;
vclass->ungetlocalcopy = vfs_s_ungetlocalcopy;
}
}
/* --------------------------------------------------------------------------------------------- */
void
vfs_init_subclass (struct vfs_s_subclass *sub, const char *name, vfs_flags_t flags,
const char *prefix)
{
struct vfs_class *vclass = VFS_CLASS (sub);
size_t len;
char *start;
vfs_init_class (vclass, name, flags, prefix);
len = sizeof (struct vfs_s_subclass) - sizeof (struct vfs_class);
start = (char *) sub + sizeof (struct vfs_class);
memset (start, 0, len);
if ((vclass->flags & VFSF_USETMP) != 0)
sub->find_entry = vfs_s_find_entry_linear;
else if ((vclass->flags & VFSF_REMOTE) != 0)
sub->find_entry = vfs_s_find_entry_linear;
else
sub->find_entry = vfs_s_find_entry_tree;
sub->dir_uptodate = vfs_s_dir_uptodate;
}
/* --------------------------------------------------------------------------------------------- */
/** Find VFS id for given directory name */
vfsid
vfs_getid (const vfs_path_t * vpath)
{
const vfs_path_element_t *path_element;
path_element = vfs_path_get_by_index (vpath, -1);
if (!vfs_path_element_valid (path_element) || path_element->class->getid == NULL)
return NULL;
return (*path_element->class->getid) (vpath);
}
/* --------------------------------------------------------------------------------------------- */
/* ----------- Utility functions for networked filesystems -------------- */
#ifdef ENABLE_VFS_NET
int
vfs_s_select_on_two (int fd1, int fd2)
{
fd_set set;
struct timeval time_out;
int v;
int maxfd = MAX (fd1, fd2) + 1;
time_out.tv_sec = 1;
time_out.tv_usec = 0;
FD_ZERO (&set);
FD_SET (fd1, &set);
FD_SET (fd2, &set);
v = select (maxfd, &set, 0, 0, &time_out);
if (v <= 0)
return v;
if (FD_ISSET (fd1, &set))
return 1;
if (FD_ISSET (fd2, &set))
return 2;
return (-1);
}
/* --------------------------------------------------------------------------------------------- */
int
vfs_s_get_line (struct vfs_class *me, int sock, char *buf, int buf_len, char term)
{
FILE *logfile = me->logfile;
int i;
char c;
for (i = 0; i < buf_len - 1; i++, buf++)
{
if (read (sock, buf, sizeof (char)) <= 0)
return 0;
if (logfile != NULL)
{
size_t ret1;
int ret2;
ret1 = fwrite (buf, 1, 1, logfile);
ret2 = fflush (logfile);
(void) ret1;
(void) ret2;
}
if (*buf == term)
{
*buf = '\0';
return 1;
}
}
/* Line is too long - terminate buffer and discard the rest of line */
*buf = '\0';
while (read (sock, &c, sizeof (c)) > 0)
{
if (logfile != NULL)
{
size_t ret1;
int ret2;
ret1 = fwrite (&c, 1, 1, logfile);
ret2 = fflush (logfile);
(void) ret1;
(void) ret2;
}
if (c == '\n')
return 1;
}
return 0;
}
/* --------------------------------------------------------------------------------------------- */
int
vfs_s_get_line_interruptible (struct vfs_class *me, char *buffer, int size, int fd)
{
int i;
int res = 0;
(void) me;
tty_enable_interrupt_key ();
for (i = 0; i < size - 1; i++)
{
ssize_t n;
n = read (fd, &buffer[i], 1);
if (n == -1 && errno == EINTR)
{
buffer[i] = '\0';
res = EINTR;
goto ret;
}
if (n == 0)
{
buffer[i] = '\0';
goto ret;
}
if (buffer[i] == '\n')
{
buffer[i] = '\0';
res = 1;
goto ret;
}
}
buffer[size - 1] = '\0';
ret:
tty_disable_interrupt_key ();
return res;
}
#endif /* ENABLE_VFS_NET */
/* --------------------------------------------------------------------------------------------- */
/**
* Normalize filenames start position
*/
void
vfs_s_normalize_filename_leading_spaces (struct vfs_s_inode *root_inode, size_t final_num_spaces)
{
GList *iter;
for (iter = g_queue_peek_head_link (root_inode->subdir); iter != NULL;
iter = g_list_next (iter))
{
struct vfs_s_entry *entry = VFS_ENTRY (iter->data);
if ((size_t) entry->leading_spaces > final_num_spaces)
{
char *source_name, *spacer;
source_name = entry->name;
spacer = g_strnfill ((size_t) entry->leading_spaces - final_num_spaces, ' ');
entry->name = g_strconcat (spacer, source_name, (char *) NULL);
g_free (spacer);
g_free (source_name);
}
entry->leading_spaces = -1;
}
}
/* --------------------------------------------------------------------------------------------- */