linux/fs/autofs4/root.c

1065 lines
27 KiB
C
Raw Normal View History

/* -*- c -*- --------------------------------------------------------------- *
*
* linux/fs/autofs/root.c
*
* Copyright 1997-1998 Transmeta Corporation -- All Rights Reserved
* Copyright 1999-2000 Jeremy Fitzhardinge <jeremy@goop.org>
* Copyright 2001-2006 Ian Kent <raven@themaw.net>
*
* This file is part of the Linux kernel and is made available under
* the terms of the GNU General Public License, version 2, or at your
* option, any later version, incorporated herein by reference.
*
* ------------------------------------------------------------------------- */
#include <linux/capability.h>
#include <linux/errno.h>
#include <linux/stat.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 16:04:11 +08:00
#include <linux/slab.h>
#include <linux/param.h>
#include <linux/time.h>
#include <linux/compat.h>
#include <linux/mutex.h>
#include "autofs_i.h"
DEFINE_SPINLOCK(autofs4_lock);
static int autofs4_dir_symlink(struct inode *,struct dentry *,const char *);
static int autofs4_dir_unlink(struct inode *,struct dentry *);
static int autofs4_dir_rmdir(struct inode *,struct dentry *);
static int autofs4_dir_mkdir(struct inode *,struct dentry *,int);
static long autofs4_root_ioctl(struct file *,unsigned int,unsigned long);
#ifdef CONFIG_COMPAT
static long autofs4_root_compat_ioctl(struct file *,unsigned int,unsigned long);
#endif
static int autofs4_dir_open(struct inode *inode, struct file *file);
static struct dentry *autofs4_lookup(struct inode *,struct dentry *, struct nameidata *);
autofs4: use lookup intent flags to trigger mounts When an open(2) call is made on an autofs mount point directory that already exists and the O_DIRECTORY flag is not used the needed mount callback to the daemon is not done. This leads to the path walk continuing resulting in a callback to the daemon with an incorrect key. open(2) is called without O_DIRECTORY by the "find" utility but this should be handled properly anyway. This happens because autofs needs to use the lookup flags to decide when to callback to the daemon to perform a mount to prevent mount storms. For example, an autofs indirect mount map that has the "browse" option will have the mount point directories are pre-created and the stat(2) call made by a color ls against each directory will cause all these directories to be mounted. It is unfortunate we need to resort to this but mount maps can be quite large. Additionally, if a user manually umounts an autofs indirect mount the directory isn't removed which also leads to this situation. To resolve this autofs needs to use the lookup intent flags to enable it to make this decision. This patch adds this check and triggers a call back if any of the lookup intent flags are set as all these calls warrant a mount attempt be requested. I know that external VFS code which uses the lookup flags is something that the VFS would like to eliminate but I have no choice as I can't see any other way to do this. A VFS dentry or inode operation callback which returns the lookup "type" (requires a definition) would be sufficient. But this change is needed now and I'm not aware of the form that coming VFS changes will take so I'm not willing to propose anything along these lines. If anyone can provide an alternate method I would be happy to use it. [akpm@linux-foundation.org: fix build for concurrent VFS changes] Signed-off-by: Ian Kent <raven@themaw.net> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:15 +08:00
#define TRIGGER_FLAGS (LOOKUP_CONTINUE | LOOKUP_DIRECTORY)
#define TRIGGER_INTENTS (LOOKUP_OPEN | LOOKUP_CREATE)
const struct file_operations autofs4_root_operations = {
.open = dcache_dir_open,
.release = dcache_dir_close,
.read = generic_read_dir,
.readdir = dcache_readdir,
.llseek = dcache_dir_lseek,
.unlocked_ioctl = autofs4_root_ioctl,
#ifdef CONFIG_COMPAT
.compat_ioctl = autofs4_root_compat_ioctl,
#endif
};
const struct file_operations autofs4_dir_operations = {
.open = autofs4_dir_open,
.release = dcache_dir_close,
.read = generic_read_dir,
.readdir = dcache_readdir,
.llseek = dcache_dir_lseek,
};
const struct inode_operations autofs4_indirect_root_inode_operations = {
.lookup = autofs4_lookup,
.unlink = autofs4_dir_unlink,
.symlink = autofs4_dir_symlink,
.mkdir = autofs4_dir_mkdir,
.rmdir = autofs4_dir_rmdir,
};
const struct inode_operations autofs4_direct_root_inode_operations = {
.lookup = autofs4_lookup,
.unlink = autofs4_dir_unlink,
.mkdir = autofs4_dir_mkdir,
.rmdir = autofs4_dir_rmdir,
};
const struct inode_operations autofs4_dir_inode_operations = {
.lookup = autofs4_lookup,
.unlink = autofs4_dir_unlink,
.symlink = autofs4_dir_symlink,
.mkdir = autofs4_dir_mkdir,
.rmdir = autofs4_dir_rmdir,
};
static void autofs4_add_active(struct dentry *dentry)
{
struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
struct autofs_info *ino = autofs4_dentry_ino(dentry);
if (ino) {
spin_lock(&sbi->lookup_lock);
if (!ino->active_count) {
if (list_empty(&ino->active))
list_add(&ino->active, &sbi->active_list);
}
ino->active_count++;
spin_unlock(&sbi->lookup_lock);
}
return;
}
static void autofs4_del_active(struct dentry *dentry)
{
struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
struct autofs_info *ino = autofs4_dentry_ino(dentry);
if (ino) {
spin_lock(&sbi->lookup_lock);
ino->active_count--;
if (!ino->active_count) {
if (!list_empty(&ino->active))
list_del_init(&ino->active);
}
spin_unlock(&sbi->lookup_lock);
}
return;
}
static unsigned int autofs4_need_mount(unsigned int flags)
{
unsigned int res = 0;
if (flags & (TRIGGER_FLAGS | TRIGGER_INTENTS))
res = 1;
return res;
}
static int autofs4_dir_open(struct inode *inode, struct file *file)
{
struct dentry *dentry = file->f_path.dentry;
struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
DPRINTK("file=%p dentry=%p %.*s",
file, dentry, dentry->d_name.len, dentry->d_name.name);
if (autofs4_oz_mode(sbi))
goto out;
/*
* An empty directory in an autofs file system is always a
* mount point. The daemon must have failed to mount this
* during lookup so it doesn't exist. This can happen, for
* example, if user space returns an incorrect status for a
* mount request. Otherwise we're doing a readdir on the
* autofs file system so just let the libfs routines handle
* it.
*/
spin_lock(&autofs4_lock);
spin_lock(&dentry->d_lock);
if (!d_mountpoint(dentry) && list_empty(&dentry->d_subdirs)) {
spin_unlock(&dentry->d_lock);
spin_unlock(&autofs4_lock);
return -ENOENT;
}
spin_unlock(&dentry->d_lock);
spin_unlock(&autofs4_lock);
out:
return dcache_dir_open(inode, file);
}
static int try_to_fill_dentry(struct dentry *dentry, int flags)
{
struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
struct autofs_info *ino = autofs4_dentry_ino(dentry);
int status;
DPRINTK("dentry=%p %.*s ino=%p",
dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
/*
* Wait for a pending mount, triggering one if there
* isn't one already
*/
if (dentry->d_inode == NULL) {
DPRINTK("waiting for mount name=%.*s",
dentry->d_name.len, dentry->d_name.name);
status = autofs4_wait(sbi, dentry, NFY_MOUNT);
DPRINTK("mount done status=%d", status);
/* Turn this into a real negative dentry? */
if (status == -ENOENT) {
spin_lock(&sbi->fs_lock);
ino->flags &= ~AUTOFS_INF_PENDING;
spin_unlock(&sbi->fs_lock);
return status;
} else if (status) {
/* Return a negative dentry, but leave it "pending" */
return status;
}
/* Trigger mount for path component or follow link */
} else if (ino->flags & AUTOFS_INF_PENDING ||
autofs4_need_mount(flags)) {
DPRINTK("waiting for mount name=%.*s",
dentry->d_name.len, dentry->d_name.name);
spin_lock(&sbi->fs_lock);
ino->flags |= AUTOFS_INF_PENDING;
spin_unlock(&sbi->fs_lock);
status = autofs4_wait(sbi, dentry, NFY_MOUNT);
DPRINTK("mount done status=%d", status);
if (status) {
spin_lock(&sbi->fs_lock);
ino->flags &= ~AUTOFS_INF_PENDING;
spin_unlock(&sbi->fs_lock);
return status;
}
}
/* Initialize expiry counter after successful mount */
ino->last_used = jiffies;
spin_lock(&sbi->fs_lock);
ino->flags &= ~AUTOFS_INF_PENDING;
spin_unlock(&sbi->fs_lock);
return 0;
}
/* For autofs direct mounts the follow link triggers the mount */
static void *autofs4_follow_link(struct dentry *dentry, struct nameidata *nd)
{
struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
struct autofs_info *ino = autofs4_dentry_ino(dentry);
int oz_mode = autofs4_oz_mode(sbi);
unsigned int lookup_type;
int status;
DPRINTK("dentry=%p %.*s oz_mode=%d nd->flags=%d",
dentry, dentry->d_name.len, dentry->d_name.name, oz_mode,
nd->flags);
/*
* For an expire of a covered direct or offset mount we need
Add a dentry op to allow processes to be held during pathwalk transit Add a dentry op (d_manage) to permit a filesystem to hold a process and make it sleep when it tries to transit away from one of that filesystem's directories during a pathwalk. The operation is keyed off a new dentry flag (DCACHE_MANAGE_TRANSIT). The filesystem is allowed to be selective about which processes it holds and which it permits to continue on or prohibits from transiting from each flagged directory. This will allow autofs to hold up client processes whilst letting its userspace daemon through to maintain the directory or the stuff behind it or mounted upon it. The ->d_manage() dentry operation: int (*d_manage)(struct path *path, bool mounting_here); takes a pointer to the directory about to be transited away from and a flag indicating whether the transit is undertaken by do_add_mount() or do_move_mount() skipping through a pile of filesystems mounted on a mountpoint. It should return 0 if successful and to let the process continue on its way; -EISDIR to prohibit the caller from skipping to overmounted filesystems or automounting, and to use this directory; or some other error code to return to the user. ->d_manage() is called with namespace_sem writelocked if mounting_here is true and no other locks held, so it may sleep. However, if mounting_here is true, it may not initiate or wait for a mount or unmount upon the parameter directory, even if the act is actually performed by userspace. Within fs/namei.c, follow_managed() is extended to check with d_manage() first on each managed directory, before transiting away from it or attempting to automount upon it. follow_down() is renamed follow_down_one() and should only be used where the filesystem deliberately intends to avoid management steps (e.g. autofs). A new follow_down() is added that incorporates the loop done by all other callers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS and CIFS do use it, their use is removed by converting them to use d_automount()). The new follow_down() calls d_manage() as appropriate. It also takes an extra parameter to indicate if it is being called from mount code (with namespace_sem writelocked) which it passes to d_manage(). follow_down() ignores automount points so that it can be used to mount on them. __follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with DCACHE_MANAGE_TRANSIT set on the basis that we're probably going to have to sleep. It would be possible to enter d_manage() in rcu-walk mode too, and have that determine whether to abort or not itself. That would allow the autofs daemon to continue on in rcu-walk mode. Note that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn't required as every tranist from that directory will cause d_manage() to be invoked. It can always be set again when necessary. ========================== WHAT THIS MEANS FOR AUTOFS ========================== Autofs currently uses the lookup() inode op and the d_revalidate() dentry op to trigger the automounting of indirect mounts, and both of these can be called with i_mutex held. autofs knows that the i_mutex will be held by the caller in lookup(), and so can drop it before invoking the daemon - but this isn't so for d_revalidate(), since the lock is only held on _some_ of the code paths that call it. This means that autofs can't risk dropping i_mutex from its d_revalidate() function before it calls the daemon. The bug could manifest itself as, for example, a process that's trying to validate an automount dentry that gets made to wait because that dentry is expired and needs cleaning up: mkdir S ffffffff8014e05a 0 32580 24956 Call Trace: [<ffffffff885371fd>] :autofs4:autofs4_wait+0x674/0x897 [<ffffffff80127f7d>] avc_has_perm+0x46/0x58 [<ffffffff8009fdcf>] autoremove_wake_function+0x0/0x2e [<ffffffff88537be6>] :autofs4:autofs4_expire_wait+0x41/0x6b [<ffffffff88535cfc>] :autofs4:autofs4_revalidate+0x91/0x149 [<ffffffff80036d96>] __lookup_hash+0xa0/0x12f [<ffffffff80057a2f>] lookup_create+0x46/0x80 [<ffffffff800e6e31>] sys_mkdirat+0x56/0xe4 versus the automount daemon which wants to remove that dentry, but can't because the normal process is holding the i_mutex lock: automount D ffffffff8014e05a 0 32581 1 32561 Call Trace: [<ffffffff80063c3f>] __mutex_lock_slowpath+0x60/0x9b [<ffffffff8000ccf1>] do_path_lookup+0x2ca/0x2f1 [<ffffffff80063c89>] .text.lock.mutex+0xf/0x14 [<ffffffff800e6d55>] do_rmdir+0x77/0xde [<ffffffff8005d229>] tracesys+0x71/0xe0 [<ffffffff8005d28d>] tracesys+0xd5/0xe0 which means that the system is deadlocked. This patch allows autofs to hold up normal processes whilst the daemon goes ahead and does things to the dentry tree behind the automouter point without risking a deadlock as almost no locks are held in d_manage() and none in d_automount(). Signed-off-by: David Howells <dhowells@redhat.com> Was-Acked-by: Ian Kent <raven@themaw.net> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-15 02:45:26 +08:00
* to break out of follow_down_one() at the autofs mount trigger
* (d_mounted--), so we can see the expiring flag, and manage
* the blocking and following here until the expire is completed.
*/
if (oz_mode) {
spin_lock(&sbi->fs_lock);
if (ino->flags & AUTOFS_INF_EXPIRING) {
spin_unlock(&sbi->fs_lock);
/* Follow down to our covering mount. */
Add a dentry op to allow processes to be held during pathwalk transit Add a dentry op (d_manage) to permit a filesystem to hold a process and make it sleep when it tries to transit away from one of that filesystem's directories during a pathwalk. The operation is keyed off a new dentry flag (DCACHE_MANAGE_TRANSIT). The filesystem is allowed to be selective about which processes it holds and which it permits to continue on or prohibits from transiting from each flagged directory. This will allow autofs to hold up client processes whilst letting its userspace daemon through to maintain the directory or the stuff behind it or mounted upon it. The ->d_manage() dentry operation: int (*d_manage)(struct path *path, bool mounting_here); takes a pointer to the directory about to be transited away from and a flag indicating whether the transit is undertaken by do_add_mount() or do_move_mount() skipping through a pile of filesystems mounted on a mountpoint. It should return 0 if successful and to let the process continue on its way; -EISDIR to prohibit the caller from skipping to overmounted filesystems or automounting, and to use this directory; or some other error code to return to the user. ->d_manage() is called with namespace_sem writelocked if mounting_here is true and no other locks held, so it may sleep. However, if mounting_here is true, it may not initiate or wait for a mount or unmount upon the parameter directory, even if the act is actually performed by userspace. Within fs/namei.c, follow_managed() is extended to check with d_manage() first on each managed directory, before transiting away from it or attempting to automount upon it. follow_down() is renamed follow_down_one() and should only be used where the filesystem deliberately intends to avoid management steps (e.g. autofs). A new follow_down() is added that incorporates the loop done by all other callers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS and CIFS do use it, their use is removed by converting them to use d_automount()). The new follow_down() calls d_manage() as appropriate. It also takes an extra parameter to indicate if it is being called from mount code (with namespace_sem writelocked) which it passes to d_manage(). follow_down() ignores automount points so that it can be used to mount on them. __follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with DCACHE_MANAGE_TRANSIT set on the basis that we're probably going to have to sleep. It would be possible to enter d_manage() in rcu-walk mode too, and have that determine whether to abort or not itself. That would allow the autofs daemon to continue on in rcu-walk mode. Note that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn't required as every tranist from that directory will cause d_manage() to be invoked. It can always be set again when necessary. ========================== WHAT THIS MEANS FOR AUTOFS ========================== Autofs currently uses the lookup() inode op and the d_revalidate() dentry op to trigger the automounting of indirect mounts, and both of these can be called with i_mutex held. autofs knows that the i_mutex will be held by the caller in lookup(), and so can drop it before invoking the daemon - but this isn't so for d_revalidate(), since the lock is only held on _some_ of the code paths that call it. This means that autofs can't risk dropping i_mutex from its d_revalidate() function before it calls the daemon. The bug could manifest itself as, for example, a process that's trying to validate an automount dentry that gets made to wait because that dentry is expired and needs cleaning up: mkdir S ffffffff8014e05a 0 32580 24956 Call Trace: [<ffffffff885371fd>] :autofs4:autofs4_wait+0x674/0x897 [<ffffffff80127f7d>] avc_has_perm+0x46/0x58 [<ffffffff8009fdcf>] autoremove_wake_function+0x0/0x2e [<ffffffff88537be6>] :autofs4:autofs4_expire_wait+0x41/0x6b [<ffffffff88535cfc>] :autofs4:autofs4_revalidate+0x91/0x149 [<ffffffff80036d96>] __lookup_hash+0xa0/0x12f [<ffffffff80057a2f>] lookup_create+0x46/0x80 [<ffffffff800e6e31>] sys_mkdirat+0x56/0xe4 versus the automount daemon which wants to remove that dentry, but can't because the normal process is holding the i_mutex lock: automount D ffffffff8014e05a 0 32581 1 32561 Call Trace: [<ffffffff80063c3f>] __mutex_lock_slowpath+0x60/0x9b [<ffffffff8000ccf1>] do_path_lookup+0x2ca/0x2f1 [<ffffffff80063c89>] .text.lock.mutex+0xf/0x14 [<ffffffff800e6d55>] do_rmdir+0x77/0xde [<ffffffff8005d229>] tracesys+0x71/0xe0 [<ffffffff8005d28d>] tracesys+0xd5/0xe0 which means that the system is deadlocked. This patch allows autofs to hold up normal processes whilst the daemon goes ahead and does things to the dentry tree behind the automouter point without risking a deadlock as almost no locks are held in d_manage() and none in d_automount(). Signed-off-by: David Howells <dhowells@redhat.com> Was-Acked-by: Ian Kent <raven@themaw.net> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-15 02:45:26 +08:00
if (!follow_down_one(&nd->path))
goto done;
goto follow;
}
spin_unlock(&sbi->fs_lock);
goto done;
}
/* If an expire request is pending everyone must wait. */
autofs4_expire_wait(dentry);
/* We trigger a mount for almost all flags */
lookup_type = autofs4_need_mount(nd->flags);
spin_lock(&sbi->fs_lock);
spin_lock(&autofs4_lock);
spin_lock(&dentry->d_lock);
if (!(lookup_type || ino->flags & AUTOFS_INF_PENDING)) {
spin_unlock(&dentry->d_lock);
spin_unlock(&autofs4_lock);
spin_unlock(&sbi->fs_lock);
goto follow;
}
/*
* If the dentry contains directories then it is an autofs
* multi-mount with no root mount offset. So don't try to
* mount it again.
*/
if (ino->flags & AUTOFS_INF_PENDING ||
(!d_mountpoint(dentry) && list_empty(&dentry->d_subdirs))) {
spin_unlock(&dentry->d_lock);
spin_unlock(&autofs4_lock);
spin_unlock(&sbi->fs_lock);
status = try_to_fill_dentry(dentry, nd->flags);
if (status)
goto out_error;
goto follow;
}
spin_unlock(&dentry->d_lock);
spin_unlock(&autofs4_lock);
spin_unlock(&sbi->fs_lock);
follow:
/*
* If there is no root mount it must be an autofs
* multi-mount with no root offset so we don't need
* to follow it.
*/
Add a dentry op to allow processes to be held during pathwalk transit Add a dentry op (d_manage) to permit a filesystem to hold a process and make it sleep when it tries to transit away from one of that filesystem's directories during a pathwalk. The operation is keyed off a new dentry flag (DCACHE_MANAGE_TRANSIT). The filesystem is allowed to be selective about which processes it holds and which it permits to continue on or prohibits from transiting from each flagged directory. This will allow autofs to hold up client processes whilst letting its userspace daemon through to maintain the directory or the stuff behind it or mounted upon it. The ->d_manage() dentry operation: int (*d_manage)(struct path *path, bool mounting_here); takes a pointer to the directory about to be transited away from and a flag indicating whether the transit is undertaken by do_add_mount() or do_move_mount() skipping through a pile of filesystems mounted on a mountpoint. It should return 0 if successful and to let the process continue on its way; -EISDIR to prohibit the caller from skipping to overmounted filesystems or automounting, and to use this directory; or some other error code to return to the user. ->d_manage() is called with namespace_sem writelocked if mounting_here is true and no other locks held, so it may sleep. However, if mounting_here is true, it may not initiate or wait for a mount or unmount upon the parameter directory, even if the act is actually performed by userspace. Within fs/namei.c, follow_managed() is extended to check with d_manage() first on each managed directory, before transiting away from it or attempting to automount upon it. follow_down() is renamed follow_down_one() and should only be used where the filesystem deliberately intends to avoid management steps (e.g. autofs). A new follow_down() is added that incorporates the loop done by all other callers of follow_down() (do_add/move_mount(), autofs and NFSD; whilst AFS, NFS and CIFS do use it, their use is removed by converting them to use d_automount()). The new follow_down() calls d_manage() as appropriate. It also takes an extra parameter to indicate if it is being called from mount code (with namespace_sem writelocked) which it passes to d_manage(). follow_down() ignores automount points so that it can be used to mount on them. __follow_mount_rcu() is made to abort rcu-walk mode if it hits a directory with DCACHE_MANAGE_TRANSIT set on the basis that we're probably going to have to sleep. It would be possible to enter d_manage() in rcu-walk mode too, and have that determine whether to abort or not itself. That would allow the autofs daemon to continue on in rcu-walk mode. Note that DCACHE_MANAGE_TRANSIT on a directory should be cleared when it isn't required as every tranist from that directory will cause d_manage() to be invoked. It can always be set again when necessary. ========================== WHAT THIS MEANS FOR AUTOFS ========================== Autofs currently uses the lookup() inode op and the d_revalidate() dentry op to trigger the automounting of indirect mounts, and both of these can be called with i_mutex held. autofs knows that the i_mutex will be held by the caller in lookup(), and so can drop it before invoking the daemon - but this isn't so for d_revalidate(), since the lock is only held on _some_ of the code paths that call it. This means that autofs can't risk dropping i_mutex from its d_revalidate() function before it calls the daemon. The bug could manifest itself as, for example, a process that's trying to validate an automount dentry that gets made to wait because that dentry is expired and needs cleaning up: mkdir S ffffffff8014e05a 0 32580 24956 Call Trace: [<ffffffff885371fd>] :autofs4:autofs4_wait+0x674/0x897 [<ffffffff80127f7d>] avc_has_perm+0x46/0x58 [<ffffffff8009fdcf>] autoremove_wake_function+0x0/0x2e [<ffffffff88537be6>] :autofs4:autofs4_expire_wait+0x41/0x6b [<ffffffff88535cfc>] :autofs4:autofs4_revalidate+0x91/0x149 [<ffffffff80036d96>] __lookup_hash+0xa0/0x12f [<ffffffff80057a2f>] lookup_create+0x46/0x80 [<ffffffff800e6e31>] sys_mkdirat+0x56/0xe4 versus the automount daemon which wants to remove that dentry, but can't because the normal process is holding the i_mutex lock: automount D ffffffff8014e05a 0 32581 1 32561 Call Trace: [<ffffffff80063c3f>] __mutex_lock_slowpath+0x60/0x9b [<ffffffff8000ccf1>] do_path_lookup+0x2ca/0x2f1 [<ffffffff80063c89>] .text.lock.mutex+0xf/0x14 [<ffffffff800e6d55>] do_rmdir+0x77/0xde [<ffffffff8005d229>] tracesys+0x71/0xe0 [<ffffffff8005d28d>] tracesys+0xd5/0xe0 which means that the system is deadlocked. This patch allows autofs to hold up normal processes whilst the daemon goes ahead and does things to the dentry tree behind the automouter point without risking a deadlock as almost no locks are held in d_manage() and none in d_automount(). Signed-off-by: David Howells <dhowells@redhat.com> Was-Acked-by: Ian Kent <raven@themaw.net> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-15 02:45:26 +08:00
if (d_managed(dentry)) {
status = follow_down(&nd->path, false);
if (status < 0)
goto out_error;
}
done:
return NULL;
out_error:
path_put(&nd->path);
return ERR_PTR(status);
}
/*
* Revalidate is called on every cache lookup. Some of those
* cache lookups may actually happen while the dentry is not
* yet completely filled in, and revalidate has to delay such
* lookups..
*/
static int autofs4_revalidate(struct dentry *dentry, struct nameidata *nd)
{
struct inode *dir;
struct autofs_sb_info *sbi;
int oz_mode;
int flags = nd ? nd->flags : 0;
int status = 1;
if (flags & LOOKUP_RCU)
return -ECHILD;
dir = dentry->d_parent->d_inode;
sbi = autofs4_sbi(dir->i_sb);
oz_mode = autofs4_oz_mode(sbi);
/* Pending dentry */
spin_lock(&sbi->fs_lock);
if (autofs4_ispending(dentry)) {
/* The daemon never causes a mount to trigger */
spin_unlock(&sbi->fs_lock);
if (oz_mode)
return 1;
/*
* If the directory has gone away due to an expire
* we have been called as ->d_revalidate() and so
* we need to return false and proceed to ->lookup().
*/
if (autofs4_expire_wait(dentry) == -EAGAIN)
return 0;
/*
* A zero status is success otherwise we have a
* negative error code.
*/
status = try_to_fill_dentry(dentry, flags);
if (status == 0)
return 1;
return status;
}
spin_unlock(&sbi->fs_lock);
/* Negative dentry.. invalidate if "old" */
if (dentry->d_inode == NULL)
return 0;
/* Check for a non-mountpoint directory with no contents */
spin_lock(&autofs4_lock);
spin_lock(&dentry->d_lock);
if (S_ISDIR(dentry->d_inode->i_mode) &&
!d_mountpoint(dentry) && list_empty(&dentry->d_subdirs)) {
DPRINTK("dentry=%p %.*s, emptydir",
dentry, dentry->d_name.len, dentry->d_name.name);
spin_unlock(&dentry->d_lock);
spin_unlock(&autofs4_lock);
/* The daemon never causes a mount to trigger */
if (oz_mode)
return 1;
/*
* A zero status is success otherwise we have a
* negative error code.
*/
status = try_to_fill_dentry(dentry, flags);
if (status == 0)
return 1;
return status;
}
spin_unlock(&dentry->d_lock);
spin_unlock(&autofs4_lock);
return 1;
}
void autofs4_dentry_release(struct dentry *de)
{
struct autofs_info *inf;
DPRINTK("releasing %p", de);
inf = autofs4_dentry_ino(de);
de->d_fsdata = NULL;
if (inf) {
struct autofs_sb_info *sbi = autofs4_sbi(de->d_sb);
if (sbi) {
spin_lock(&sbi->lookup_lock);
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
if (!list_empty(&inf->active))
list_del(&inf->active);
if (!list_empty(&inf->expiring))
list_del(&inf->expiring);
spin_unlock(&sbi->lookup_lock);
}
inf->dentry = NULL;
inf->inode = NULL;
autofs4_free_ino(inf);
}
}
/* For dentries of directories in the root dir */
static const struct dentry_operations autofs4_root_dentry_operations = {
.d_release = autofs4_dentry_release,
};
/* For other dentries */
static const struct dentry_operations autofs4_dentry_operations = {
.d_automount = autofs4_d_automount,
.d_release = autofs4_dentry_release,
};
static struct dentry *autofs4_lookup_active(struct dentry *dentry)
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
{
struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
struct dentry *parent = dentry->d_parent;
struct qstr *name = &dentry->d_name;
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
unsigned int len = name->len;
unsigned int hash = name->hash;
const unsigned char *str = name->name;
struct list_head *p, *head;
spin_lock(&autofs4_lock);
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
spin_lock(&sbi->lookup_lock);
head = &sbi->active_list;
list_for_each(p, head) {
struct autofs_info *ino;
struct dentry *active;
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
struct qstr *qstr;
ino = list_entry(p, struct autofs_info, active);
active = ino->dentry;
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
spin_lock(&active->d_lock);
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
/* Already gone? */
if (active->d_count == 0)
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
goto next;
qstr = &active->d_name;
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
if (active->d_name.hash != hash)
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
goto next;
if (active->d_parent != parent)
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
goto next;
if (qstr->len != len)
goto next;
if (memcmp(qstr->name, str, len))
goto next;
if (d_unhashed(active)) {
dget_dlock(active);
spin_unlock(&active->d_lock);
spin_unlock(&sbi->lookup_lock);
spin_unlock(&autofs4_lock);
return active;
}
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
next:
spin_unlock(&active->d_lock);
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
}
spin_unlock(&sbi->lookup_lock);
spin_unlock(&autofs4_lock);
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
return NULL;
}
static struct dentry *autofs4_lookup_expiring(struct dentry *dentry)
{
struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
struct dentry *parent = dentry->d_parent;
struct qstr *name = &dentry->d_name;
unsigned int len = name->len;
unsigned int hash = name->hash;
const unsigned char *str = name->name;
struct list_head *p, *head;
spin_lock(&autofs4_lock);
spin_lock(&sbi->lookup_lock);
head = &sbi->expiring_list;
list_for_each(p, head) {
struct autofs_info *ino;
struct dentry *expiring;
struct qstr *qstr;
ino = list_entry(p, struct autofs_info, expiring);
expiring = ino->dentry;
spin_lock(&expiring->d_lock);
/* Bad luck, we've already been dentry_iput */
if (!expiring->d_inode)
goto next;
qstr = &expiring->d_name;
if (expiring->d_name.hash != hash)
goto next;
if (expiring->d_parent != parent)
goto next;
if (qstr->len != len)
goto next;
if (memcmp(qstr->name, str, len))
goto next;
if (d_unhashed(expiring)) {
dget_dlock(expiring);
spin_unlock(&expiring->d_lock);
spin_unlock(&sbi->lookup_lock);
spin_unlock(&autofs4_lock);
return expiring;
}
next:
spin_unlock(&expiring->d_lock);
}
spin_unlock(&sbi->lookup_lock);
spin_unlock(&autofs4_lock);
return NULL;
}
static int autofs4_mount_wait(struct dentry *dentry)
{
struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
struct autofs_info *ino = autofs4_dentry_ino(dentry);
int status;
if (ino->flags & AUTOFS_INF_PENDING) {
DPRINTK("waiting for mount name=%.*s",
dentry->d_name.len, dentry->d_name.name);
status = autofs4_wait(sbi, dentry, NFY_MOUNT);
DPRINTK("mount wait done status=%d", status);
ino->last_used = jiffies;
return status;
}
return 0;
}
static int do_expire_wait(struct dentry *dentry)
{
struct dentry *expiring;
expiring = autofs4_lookup_expiring(dentry);
if (!expiring)
return autofs4_expire_wait(dentry);
else {
/*
* If we are racing with expire the request might not
* be quite complete, but the directory has been removed
* so it must have been successful, just wait for it.
*/
autofs4_expire_wait(expiring);
autofs4_del_expiring(expiring);
dput(expiring);
}
return 0;
}
static struct dentry *autofs4_mountpoint_changed(struct path *path)
{
struct dentry *dentry = path->dentry;
struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
/*
* If this is an indirect mount the dentry could have gone away
* as a result of an expire and a new one created.
*/
if (autofs_type_indirect(sbi->type) && d_unhashed(dentry)) {
struct dentry *parent = dentry->d_parent;
struct dentry *new = d_lookup(parent, &dentry->d_name);
if (!new)
return NULL;
dput(path->dentry);
path->dentry = new;
}
return path->dentry;
}
struct vfsmount *autofs4_d_automount(struct path *path)
{
struct dentry *dentry = path->dentry;
struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
struct autofs_info *ino = autofs4_dentry_ino(dentry);
int status;
DPRINTK("dentry=%p %.*s",
dentry, dentry->d_name.len, dentry->d_name.name);
/* The daemon never triggers a mount. */
if (autofs4_oz_mode(sbi))
return NULL;
/*
* If an expire request is pending everyone must wait.
* If the expire fails we're still mounted so continue
* the follow and return. A return of -EAGAIN (which only
* happens with indirect mounts) means the expire completed
* and the directory was removed, so just go ahead and try
* the mount.
*/
status = do_expire_wait(dentry);
if (status && status != -EAGAIN)
return NULL;
/* Callback to the daemon to perform the mount or wait */
spin_lock(&sbi->fs_lock);
if (ino->flags & AUTOFS_INF_PENDING) {
spin_unlock(&sbi->fs_lock);
status = autofs4_mount_wait(dentry);
if (status)
return ERR_PTR(status);
spin_lock(&sbi->fs_lock);
goto done;
}
/*
* If the dentry is a symlink it's equivalent to a directory
* having d_mounted() true, so there's no need to call back
* to the daemon.
*/
if (dentry->d_inode && S_ISLNK(dentry->d_inode->i_mode))
goto done;
spin_lock(&dentry->d_lock);
if (!d_mountpoint(dentry) && list_empty(&dentry->d_subdirs)) {
ino->flags |= AUTOFS_INF_PENDING;
spin_unlock(&dentry->d_lock);
spin_unlock(&sbi->fs_lock);
status = autofs4_mount_wait(dentry);
if (status)
return ERR_PTR(status);
spin_lock(&sbi->fs_lock);
ino->flags &= ~AUTOFS_INF_PENDING;
goto done;
}
spin_unlock(&dentry->d_lock);
done:
/*
* Any needed mounting has been completed and the path updated
* so turn this into a normal dentry so we don't continually
* call ->d_automount().
*/
managed_dentry_clear_automount(dentry);
spin_unlock(&sbi->fs_lock);
/* Mount succeeded, check if we ended up with a new dentry */
dentry = autofs4_mountpoint_changed(path);
if (!dentry)
return ERR_PTR(-ENOENT);
return NULL;
}
/* Lookups in the root directory */
static struct dentry *autofs4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
{
struct autofs_sb_info *sbi;
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
struct autofs_info *ino;
struct dentry *active;
DPRINTK("name = %.*s", dentry->d_name.len, dentry->d_name.name);
/* File name too long to exist */
if (dentry->d_name.len > NAME_MAX)
return ERR_PTR(-ENAMETOOLONG);
sbi = autofs4_sbi(dir->i_sb);
DPRINTK("pid = %u, pgrp = %u, catatonic = %d, oz_mode = %d",
current->pid, task_pgrp_nr(current), sbi->catatonic, oz_mode);
active = autofs4_lookup_active(dentry);
if (active) {
return active;
} else {
d_set_d_op(dentry, &autofs4_root_dentry_operations);
/*
* A dentry that is not within the root can never trigger a
* mount operation, unless the directory already exists, so we
* can return fail immediately. The daemon however does need
* to create directories within the file system.
*/
if (!autofs4_oz_mode(sbi) && !IS_ROOT(dentry->d_parent))
return ERR_PTR(-ENOENT);
/* Mark entries in the root as mount triggers */
if (autofs_type_indirect(sbi->type) && IS_ROOT(dentry->d_parent)) {
d_set_d_op(dentry, &autofs4_dentry_operations);
managed_dentry_set_automount(dentry);
}
ino = autofs4_init_ino(NULL, sbi, 0555);
if (!ino)
return ERR_PTR(-ENOMEM);
dentry->d_fsdata = ino;
ino->dentry = dentry;
autofs4_add_active(dentry);
d_instantiate(dentry, NULL);
}
return NULL;
}
static int autofs4_dir_symlink(struct inode *dir,
struct dentry *dentry,
const char *symname)
{
struct autofs_sb_info *sbi = autofs4_sbi(dir->i_sb);
struct autofs_info *ino = autofs4_dentry_ino(dentry);
struct autofs_info *p_ino;
struct inode *inode;
char *cp;
DPRINTK("%s <- %.*s", symname,
dentry->d_name.len, dentry->d_name.name);
if (!autofs4_oz_mode(sbi))
return -EACCES;
ino = autofs4_init_ino(ino, sbi, S_IFLNK | 0555);
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
if (!ino)
return -ENOMEM;
autofs4_del_active(dentry);
ino->size = strlen(symname);
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
cp = kmalloc(ino->size + 1, GFP_KERNEL);
if (!cp) {
if (!dentry->d_fsdata)
kfree(ino);
return -ENOMEM;
}
strcpy(cp, symname);
inode = autofs4_get_inode(dir->i_sb, ino);
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
if (!inode) {
kfree(cp);
if (!dentry->d_fsdata)
kfree(ino);
return -ENOMEM;
}
d_add(dentry, inode);
dentry->d_fsdata = ino;
ino->dentry = dget(dentry);
atomic_inc(&ino->count);
p_ino = autofs4_dentry_ino(dentry->d_parent);
if (p_ino && dentry->d_parent != dentry)
atomic_inc(&p_ino->count);
ino->inode = inode;
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
ino->u.symlink = cp;
dir->i_mtime = CURRENT_TIME;
return 0;
}
/*
* NOTE!
*
* Normal filesystems would do a "d_delete()" to tell the VFS dcache
* that the file no longer exists. However, doing that means that the
* VFS layer can turn the dentry into a negative dentry. We don't want
* this, because the unlink is probably the result of an expire.
* We simply d_drop it and add it to a expiring list in the super block,
* which allows the dentry lookup to check for an incomplete expire.
*
* If a process is blocked on the dentry waiting for the expire to finish,
* it will invalidate the dentry and try to mount with a new one.
*
* Also see autofs4_dir_rmdir()..
*/
static int autofs4_dir_unlink(struct inode *dir, struct dentry *dentry)
{
struct autofs_sb_info *sbi = autofs4_sbi(dir->i_sb);
struct autofs_info *ino = autofs4_dentry_ino(dentry);
struct autofs_info *p_ino;
/* This allows root to remove symlinks */
if (!autofs4_oz_mode(sbi) && !capable(CAP_SYS_ADMIN))
return -EACCES;
if (atomic_dec_and_test(&ino->count)) {
p_ino = autofs4_dentry_ino(dentry->d_parent);
if (p_ino && dentry->d_parent != dentry)
atomic_dec(&p_ino->count);
}
dput(ino->dentry);
dentry->d_inode->i_size = 0;
clear_nlink(dentry->d_inode);
dir->i_mtime = CURRENT_TIME;
spin_lock(&autofs4_lock);
autofs4_add_expiring(dentry);
spin_lock(&dentry->d_lock);
__d_drop(dentry);
spin_unlock(&dentry->d_lock);
spin_unlock(&autofs4_lock);
return 0;
}
static int autofs4_dir_rmdir(struct inode *dir, struct dentry *dentry)
{
struct autofs_sb_info *sbi = autofs4_sbi(dir->i_sb);
struct autofs_info *ino = autofs4_dentry_ino(dentry);
struct autofs_info *p_ino;
DPRINTK("dentry %p, removing %.*s",
dentry, dentry->d_name.len, dentry->d_name.name);
if (!autofs4_oz_mode(sbi))
return -EACCES;
spin_lock(&autofs4_lock);
spin_lock(&sbi->lookup_lock);
spin_lock(&dentry->d_lock);
if (!list_empty(&dentry->d_subdirs)) {
spin_unlock(&dentry->d_lock);
spin_unlock(&sbi->lookup_lock);
spin_unlock(&autofs4_lock);
return -ENOTEMPTY;
}
__autofs4_add_expiring(dentry);
spin_unlock(&sbi->lookup_lock);
__d_drop(dentry);
spin_unlock(&dentry->d_lock);
spin_unlock(&autofs4_lock);
if (atomic_dec_and_test(&ino->count)) {
p_ino = autofs4_dentry_ino(dentry->d_parent);
if (p_ino && dentry->d_parent != dentry)
atomic_dec(&p_ino->count);
}
dput(ino->dentry);
dentry->d_inode->i_size = 0;
clear_nlink(dentry->d_inode);
if (dir->i_nlink)
drop_nlink(dir);
return 0;
}
static int autofs4_dir_mkdir(struct inode *dir, struct dentry *dentry, int mode)
{
struct autofs_sb_info *sbi = autofs4_sbi(dir->i_sb);
struct autofs_info *ino = autofs4_dentry_ino(dentry);
struct autofs_info *p_ino;
struct inode *inode;
if (!autofs4_oz_mode(sbi))
return -EACCES;
DPRINTK("dentry %p, creating %.*s",
dentry, dentry->d_name.len, dentry->d_name.name);
ino = autofs4_init_ino(ino, sbi, S_IFDIR | 0555);
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
if (!ino)
return -ENOMEM;
autofs4_del_active(dentry);
inode = autofs4_get_inode(dir->i_sb, ino);
autofs4: use look aside list for lookups A while ago a patch to resolve a deadlock during directory creation was merged. This delayed the hashing of lookup dentrys until the ->mkdir() (or ->symlink()) operation completed to ensure we always went through ->lookup() instead of also having processes go through ->revalidate() so our VFS locking remained consistent. Now we are seeing a couple of side affects of that change in situations with heavy mount activity. Two cases have been identified: 1) When a mount request is triggered, due to the delayed hashing, the directory created by user space for the mount point doesn't have the DCACHE_AUTOFS_PENDING flag set. In the case of an autofs multi-mount where a tree of mount point directories are created this can lead to the path walk continuing rather than the dentry being sent to the wait queue to wait for request completion. This is because, if the pending flag isn't set, the criteria for deciding this is a mount in progress fails to hold, namely that the dentry is not a mount point and has no subdirectories. 2) A mount request dentry is initially created negative and unhashed. It remains this way until the ->mkdir() callback completes. Since it is unhashed a fresh dentry is used when the user space mount request creates the mount point directory. This leaves the original dentry negative and unhashed. But revalidate has no way to tell the VFS that the dentry has changed, other than to force another ->lookup() by returning false, which is at best wastefull and at worst not possible. This results in an -ENOENT return from the original path walk when in fact the mount succeeded. To resolve this we need to ensure that the same dentry is used in all calls to ->lookup() during the course of a mount request. This patch achieves that by adding the initial dentry to a look aside list and removes it at ->mkdir() or ->symlink() completion (or when the dentry is released), since these are the only create operations autofs4 supports. Signed-off-by: Ian Kent <raven@themaw.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:30:12 +08:00
if (!inode) {
if (!dentry->d_fsdata)
kfree(ino);
return -ENOMEM;
}
d_add(dentry, inode);
dentry->d_fsdata = ino;
ino->dentry = dget(dentry);
atomic_inc(&ino->count);
p_ino = autofs4_dentry_ino(dentry->d_parent);
if (p_ino && dentry->d_parent != dentry)
atomic_inc(&p_ino->count);
ino->inode = inode;
inc_nlink(dir);
dir->i_mtime = CURRENT_TIME;
return 0;
}
/* Get/set timeout ioctl() operation */
#ifdef CONFIG_COMPAT
static inline int autofs4_compat_get_set_timeout(struct autofs_sb_info *sbi,
compat_ulong_t __user *p)
{
int rv;
unsigned long ntimeout;
if ((rv = get_user(ntimeout, p)) ||
(rv = put_user(sbi->exp_timeout/HZ, p)))
return rv;
if (ntimeout > UINT_MAX/HZ)
sbi->exp_timeout = 0;
else
sbi->exp_timeout = ntimeout * HZ;
return 0;
}
#endif
static inline int autofs4_get_set_timeout(struct autofs_sb_info *sbi,
unsigned long __user *p)
{
int rv;
unsigned long ntimeout;
if ((rv = get_user(ntimeout, p)) ||
(rv = put_user(sbi->exp_timeout/HZ, p)))
return rv;
if (ntimeout > ULONG_MAX/HZ)
sbi->exp_timeout = 0;
else
sbi->exp_timeout = ntimeout * HZ;
return 0;
}
/* Return protocol version */
static inline int autofs4_get_protover(struct autofs_sb_info *sbi, int __user *p)
{
return put_user(sbi->version, p);
}
/* Return protocol sub version */
static inline int autofs4_get_protosubver(struct autofs_sb_info *sbi, int __user *p)
{
return put_user(sbi->sub_version, p);
}
/*
* Tells the daemon whether it can umount the autofs mount.
*/
static inline int autofs4_ask_umount(struct vfsmount *mnt, int __user *p)
{
int status = 0;
if (may_umount(mnt))
status = 1;
DPRINTK("returning %d", status);
status = put_user(status, p);
return status;
}
/* Identify autofs4_dentries - this is so we can tell if there's
an extra dentry refcount or not. We only hold a refcount on the
dentry if its non-negative (ie, d_inode != NULL)
*/
int is_autofs4_dentry(struct dentry *dentry)
{
return dentry && dentry->d_inode &&
(dentry->d_op == &autofs4_root_dentry_operations ||
dentry->d_op == &autofs4_dentry_operations) &&
dentry->d_fsdata != NULL;
}
/*
* ioctl()'s on the root directory is the chief method for the daemon to
* generate kernel reactions
*/
static int autofs4_root_ioctl_unlocked(struct inode *inode, struct file *filp,
unsigned int cmd, unsigned long arg)
{
struct autofs_sb_info *sbi = autofs4_sbi(inode->i_sb);
void __user *p = (void __user *)arg;
DPRINTK("cmd = 0x%08x, arg = 0x%08lx, sbi = %p, pgrp = %u",
cmd,arg,sbi,task_pgrp_nr(current));
if (_IOC_TYPE(cmd) != _IOC_TYPE(AUTOFS_IOC_FIRST) ||
_IOC_NR(cmd) - _IOC_NR(AUTOFS_IOC_FIRST) >= AUTOFS_IOC_COUNT)
return -ENOTTY;
if (!autofs4_oz_mode(sbi) && !capable(CAP_SYS_ADMIN))
return -EPERM;
switch(cmd) {
case AUTOFS_IOC_READY: /* Wait queue: go ahead and retry */
return autofs4_wait_release(sbi,(autofs_wqt_t)arg,0);
case AUTOFS_IOC_FAIL: /* Wait queue: fail with ENOENT */
return autofs4_wait_release(sbi,(autofs_wqt_t)arg,-ENOENT);
case AUTOFS_IOC_CATATONIC: /* Enter catatonic mode (daemon shutdown) */
autofs4_catatonic_mode(sbi);
return 0;
case AUTOFS_IOC_PROTOVER: /* Get protocol version */
return autofs4_get_protover(sbi, p);
case AUTOFS_IOC_PROTOSUBVER: /* Get protocol sub version */
return autofs4_get_protosubver(sbi, p);
case AUTOFS_IOC_SETTIMEOUT:
return autofs4_get_set_timeout(sbi, p);
#ifdef CONFIG_COMPAT
case AUTOFS_IOC_SETTIMEOUT32:
return autofs4_compat_get_set_timeout(sbi, p);
#endif
case AUTOFS_IOC_ASKUMOUNT:
return autofs4_ask_umount(filp->f_path.mnt, p);
/* return a single thing to expire */
case AUTOFS_IOC_EXPIRE:
return autofs4_expire_run(inode->i_sb,filp->f_path.mnt,sbi, p);
/* same as above, but can send multiple expires through pipe */
case AUTOFS_IOC_EXPIRE_MULTI:
return autofs4_expire_multi(inode->i_sb,filp->f_path.mnt,sbi, p);
default:
return -ENOSYS;
}
}
static long autofs4_root_ioctl(struct file *filp,
unsigned int cmd, unsigned long arg)
{
struct inode *inode = filp->f_dentry->d_inode;
return autofs4_root_ioctl_unlocked(inode, filp, cmd, arg);
}
#ifdef CONFIG_COMPAT
static long autofs4_root_compat_ioctl(struct file *filp,
unsigned int cmd, unsigned long arg)
{
struct inode *inode = filp->f_path.dentry->d_inode;
int ret;
if (cmd == AUTOFS_IOC_READY || cmd == AUTOFS_IOC_FAIL)
ret = autofs4_root_ioctl_unlocked(inode, filp, cmd, arg);
else
ret = autofs4_root_ioctl_unlocked(inode, filp, cmd,
(unsigned long)compat_ptr(arg));
return ret;
}
#endif