The recent ->lookup() deadlock correction required the directory inode
mutex to be dropped while waiting for expire completion. We were
concerned about side effects from this change and one has been identified.
I saw several error messages.
They cause autofs to become quite confused and don't really point to the
actual problem.
Things like:
handle_packet_missing_direct:1376: can't find map entry for (43,1827932)
which is usually totally fatal (although in this case it wouldn't be
except that I treat is as such because it normally is).
do_mount_direct: direct trigger not valid or already mounted
/test/nested/g3c/s1/ss1
which is recoverable, however if this problem is at play it can cause
autofs to become quite confused as to the dependencies in the mount tree
because mount triggers end up mounted multiple times. It's hard to
accurately check for this over mounting case and automount shouldn't need
to if the kernel module is doing its job.
There was one other message, similar in consequence of this last one but I
can't locate a log example just now.
When checking if a mount has already completed prior to adding a new mount
request to the wait queue we check if the dentry is hashed and, if so, if
it is a mount point. But, if a mount successfully completed while we
slept on the wait queue mutex the dentry must exist for the mount to have
completed so the test is not really needed.
Mounts can also be done on top of a global root dentry, so for the above
case, where a mount request completes and the wait queue entry has already
been removed, the hashed test returning false can cause an incorrect
callback to the daemon. Also, d_mountpoint() is not sufficient to check
if a mount has completed for the multi-mount case when we don't have a
real mount at the base of the tree.
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>
Fix an obvious incorrect return status in autofs4_mount_busy().
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>
Missing conversion from kernel to userland dev_t; this sucker
breaks as soon as we get sufficiently many autofs mounts for
new_encode_dev(s_dev) != s_dev.
Note: this is the minimal fix.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
A deadlock can occur when user space uses a signal (autofs version 4 uses
SIGCHLD for this) to effect expire completion.
The order of events is:
Expire process completes, but before being able to send SIGCHLD to it's parent
...
Another process walks onto a different mount point and drops the directory
inode semaphore prior to sending the request to the daemon as it must ...
A third process does an lstat on on the expired mount point causing it to wait
on expire completion (unfortunately) holding the directory semaphore.
The mount request then arrives at the daemon which does an lstat and,
deadlock.
For some time I was concerned about releasing the directory semaphore around
the expire wait in autofs4_lookup as well as for the mount call back. I
finally realized that the last round of changes in this function made the
expiring dentry and the lookup dentry separate and distinct so the check and
possible wait can be done anywhere prior to the mount call back. This patch
moves the check to just before the mount call back and inside the directory
inode mutex release.
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>
A significant portion of the autofs_dev_ioctl_expire() and
autofs4_expire_multi() functions is duplicated code. This patch cleans that
up.
Signed-off-by: Ian Kent <raven@themaw.net>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In function validate_dev_ioctl() we check that the string we've been sent
is a valid path. The function that does this check assumes the string is
NULL terminated but our NULL termination check isn't done until after this
call. This patch changes the order of the check.
Signed-off-by: Ian Kent <raven@themaw.net>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
- the type assigned at mount when no type is given is changed
from 0 to AUTOFS_TYPE_INDIRECT. This was done because 0 and
AUTOFS_TYPE_INDIRECT were being treated implicitly as the same
type.
- previously, an offset mount had it's type set to
AUTOFS_TYPE_DIRECT|AUTOFS_TYPE_OFFSET but the mount control
re-implementation needs to be able distinguish all three types.
So this was changed to make the type setting explicit.
- a type AUTOFS_TYPE_ANY was added for use by the re-implementation
when checking if a given path is a mountpoint. It's not really a
type as we use this to ask if a given path is a mountpoint in the
autofs_dev_ioctl_ismountpoint() function.
- functions to set and test the autofs mount types have been added to
improve readability and make the type usage explicit.
- the mount type is used from user space for the mount control
re-implementtion so, for consistency, all the definitions have
been moved to the user space include file include/linux/auto_fs4.h.
Signed-off-by: Ian Kent <raven@themaw.net>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A local definition of devid in autofs_dev_ioctl_ismountpoint() shadows
the fuction wide definition.
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>
The parameter usage in the device node ioctl code uses arg1 and arg2 as
parameter names. This patch redefines the parameter names to reflect what
they actually are in an effort to make the code more readable.
Signed-off-by: Ian Kent <raven@themaw.net>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
... and don't bother in callers. Don't bother with zeroing i_blocks,
while we are at it - it's already been zeroed.
i_mode is not worth the effort; it has no common default value.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Conflicts:
security/keys/internal.h
security/keys/process_keys.c
security/keys/request_key.c
Fixed conflicts above by using the non 'tsk' versions.
Signed-off-by: James Morris <jmorris@namei.org>
Pass credentials through dentry_open() so that the COW creds patch can have
SELinux's flush_unauthorized_files() pass the appropriate creds back to itself
when it opens its null chardev.
The security_dentry_open() call also now takes a creds pointer, as does the
dentry_open hook in struct security_operations.
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: James Morris <jmorris@namei.org>
Signed-off-by: James Morris <jmorris@namei.org>
Wrap access to task credentials so that they can be separated more easily from
the task_struct during the introduction of COW creds.
Change most current->(|e|s|fs)[ug]id to current_(|e|s|fs)[ug]id().
Change some task->e?[ug]id to task_e?[ug]id(). In some places it makes more
sense to use RCU directly rather than a convenient wrapper; these will be
addressed by later patches.
Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: James Morris <jmorris@namei.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: Ian Kent <raven@themaw.net>
Cc: autofs@linux.kernel.org
Signed-off-by: James Morris <jmorris@namei.org>
The function check_dev_ioctl_version() returns an error code upon fail but
it isn't captured and returned in validate_dev_ioctl() as it should be.
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Ian Kent <raven@themaw.net>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When checking a directory tree in autofs_tree_busy() we can incorrectly
decide that the tree isn't busy. This happens for the case of an active
offset mount as autofs4_follow_mount() follows past the active offset
mount, which has an open file handle used for expires, causing the file
handle not to count toward the busyness check.
Signed-off-by: Ian Kent <raven@themaw.net>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add a miscellaneous device to the autofs4 module for routing ioctls. This
provides the ability to obtain an ioctl file handle for an autofs mount
point that is possibly covered by another mount.
The actual problem with autofs is that it can't reconnect to existing
mounts. Immediately one things of just adding the ability to remount
autofs file systems would solve it, but alas, that can't work. This is
because autofs direct mounts and the implementation of "on demand mount
and expire" of nested mount trees have the file system mounted on top of
the mount trigger dentry.
To resolve this a miscellaneous device node for routing ioctl commands to
these mount points has been implemented in the autofs4 kernel module and a
library added to autofs. This provides the ability to open a file
descriptor for these over mounted autofs mount points.
Please refer to Documentation/filesystems/autofs4-mount-control.txt for a
discussion of the problem, implementation alternatives considered and a
description of the interface.
[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: build fix]
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>
Track the uid and gid of the last process to request a mount for on an
autofs dentry.
[akpm@linux-foundation.org: fix tpyo in comment]
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>
Usage of the AUTOFS_TYPE_* defines is a little confusing and appears
inconsistent.
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>
This is a much better version of a previous patch to make the parser
tables constant. Rather than changing the typedef, we put the "const" in
all the various places where its required, allowing the __initconst
exception for nfsroot which was the cause of the previous trouble.
This was posted for review some time ago and I believe its been in -mm
since then.
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Cc: Alexander Viro <aviro@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The ioctls AUTOFS_IOC_TOGGLEREGHOST and AUTOFS_IOC_ASKREGHOST were added
several years ago but what they were intended for has never been
implemented (as far as I'm aware noone uses them) so remove them.
Signed-off-by: Ian Kent <raven@themaw.net>
Reviewed-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch re-orgnirzes the checking for and waiting on active expires and
elininates redundant checks.
Signed-off-by: Ian Kent <raven@themaw.net>
Cc: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Appologies, somehow I seem to have sent an out dated version of this
patch. Here is an additional patch that brings the patch up to date.
Signed-off-by: Ian Kent <raven@themaw.net>
Cc: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
For direct and offset type mounts that are covered by another mount we
cannot check the AUTOFS_INF_EXPIRING flag during a path walk which leads
to lookups walking into an expiring mount while it is being expired.
For example, for the direct multi-mount map entry with a couple of
offsets:
/race/mm1 / <server1>:/<path1>
/om1 <server2>:/<path2>
/om2 <server1>:/<path3>
an autofs trigger mount is mounted on /race/mm1 and when accessed it is
over mounted and trigger mounts made for /race/mm1/om1 and /race/mm1/om2.
So it isn't possible for path walks to see the expiring flag at all and
they happily walk into the file system while it is expiring.
When expiring these mounts follow_down() must stop at the autofs mount and
all processes must block in the ->follow_link() method (except the daemon)
until the expire is complete. This is done by decrementing the d_mounted
field of the autofs trigger mount root dentry until the expire is
completed. In ->follow_link() all processes wait on the expire and the
mount following is completed for the daemon until the expire is complete.
Signed-off-by: Ian Kent <raven@themaw.net>
Cc: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The selection of a dentry for expiration and the setting of the
AUTOFS_INF_EXPIRING flag isn't done atomically which can lead to lookups
walking into an expiring mount.
What happens is that an expire is initiated by the daemon and a dentry is
selected for expire but, since there is no lock held between the selection
and setting of the expiring flag, a process may find the flag clear and
continue walking into the mount tree at the same time the daemon attempts
the expire it.
Signed-off-by: Ian Kent <raven@themaw.net>
Reviewed-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There are two cases for which a dentry that has a pending mount request
does not wait for completion. One is via autofs4_revalidate() and the
other via autofs4_follow_link().
In revalidate, after the mount point directory is created, but before the
mount is done, the check in try_to_fill_dentry() can can fail to send the
dentry to the wait queue since the dentry is positive and the lookup flags
may contain only LOOKUP_FOLLOW. Although we don't trigger a mount for the
LOOKUP_FOLLOW flag, if ther's one pending we might as well wait and use
the mounted dentry for the lookup.
In autofs4_follow_link() the dentry is not checked to see if it is pending
so it may fail to call try_to_fill_dentry() and not wait for mount
completion.
A dentry that is pending must always be sent to the wait queue.
Signed-off-by: Ian Kent <raven@themaw.net>
Reviewed-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The mount triggering functionality of readdir and related functions is no
longer used (and is quite broken as well). The unused portions have been
removed.
Signed-off-by: Ian Kent <raven@themaw.net>
Reviewed-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We have been seeing mount requests comming to the automount daemon for
keys of the form "<map key>/<non key directory>" which are lookups for
invalid map keys. But we can check for this in the kernel module and
return a fail immediately, without having to send a request to the daemon.
It is possible to recognise these requests are invalid based on whether
the request dentry is negative and its relation to the autofs file system
root.
For example, given the indirect multi-mount map entry:
idm1 \
/mm1 <server>:/<path1>
/mm2 <server>:/<path2>
For a request to mount idm1, IS_ROOT((idm1)->d_parent) will be always be
true and the dentry may be negative. But directories idm1/mm1 and
idm1/mm2 will always be created as part of the mount request for idm1. So
any mount request within idm1 itself must have a positive dentry otherwise
the map key is invalid.
In version 4 these multi-mount entries are all mounted and umounted as a
single request and in version 5 the directories idm1/mm1 and idm1/mm2 are
created and an autofs fs mounted on them to act as a mount trigger so the
above is also true.
This also holds true for the autofs version 4 pseudo direct mount feature.
When this feature is used without the "--ghost" option automount(8) will
create internal submounts as we go down the map key paths which are
essentially normal indirect mounts for which the above holds. If the
"--ghost" option is given the directories for map keys are created at
daemon startup so valid map entries correspond to postive dentries in the
autofs fs.
autofs version 5 direct mount maps are similar except that the IS_ROOT
check is not needed. This has been addressed in a previous patch tittled
"autofs4 - detect invalid direct mount requests".
For example, given the direct multi-mount map entry:
/test/dm1 \
/mm1 <server>:/<path1>
/mm2 <server>:/<path2>
An autofs fs is mounted on /test/dm1 as a trigger mount and when a mount
is triggered for /test/dm1, the multi-mount offset directories
/test/dm1/mm1 and /test/dm1/mm2 are created and an autofs fs is mounted on
them to act as mount triggers. So valid direct mount requests must always
have a positive dentry if they correspond to a valid map entry.
Signed-off-by: Ian Kent <raven@themaw.net>
Acked-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
autofs v5 direct and offset mounts within an autofs filesystem are
triggered by existing autofs triger mounts so the mount point dentry must
be positive. If the mount point dentry is negative then the trigger
doesn't exist so we can return fail immediately.
Signed-off-by: Ian Kent <raven@themaw.net>
Cc: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If an autofs mount becomes catatonic before autofs4_wait_release() is
called the wait queue counter will not be decremented down to zero and the
entry will never be freed. There are also races decrementing the wait
counter in the wait release function. To deal with this the counter needs
to be updated while holding the wait queue mutex and waiters need to be
woken up unconditionally when the wait is removed from the queue to ensure
we eventually free the wait.
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>
It is possible for an autofs mount to become catatonic (and for the daemon
communication pipe to become NULL) after a wait has been initiallized but
before the request has been sent to the daemon. We need to check for this
before sending the request packet.
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>
It see that the patch tittled "autofs4 - fix pending mount race" is
missing a change that I had recently made.
It's missing a kfree for the case mutex_lock_interruptible() fails
to aquire the wait queue mutex.
Signed-off-by: Ian Kent <raven@themaw.net>
Cc: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Close a race between a pending mount that is about to finish and a new
lookup for the same directory.
Process P1 triggers a mount of directory foo. It sets
DCACHE_AUTOFS_PENDING in the ->lookup routine, creates a waitq entry for
'foo', and calls out to the daemon to perform the mount. The autofs
daemon will then create the directory 'foo', using a new dentry that will
be hashed in the dcache.
Before the mount completes, another process, P2, tries to walk into the
'foo' directory. The vfs path walking code finds an entry for 'foo' and
calls the revalidate method. Revalidate finds that the entry is not
PENDING (because PENDING was never set on the dentry created by the
mkdir), but it does find the directory is empty. Revalidate calls
try_to_fill_dentry, which sets the PENDING flag and then calls into the
autofs4 wait code to trigger or wait for a mount of 'foo'. The wait code
finds the entry for 'foo' and goes to sleep waiting for the completion of
the mount.
Yet another process, P3, tries to walk into the 'foo' directory. This
process again finds a dentry in the dcache for 'foo', and calls into the
autofs revalidate code.
The revalidate code finds that the PENDING flag is set, and so calls
try_to_fill_dentry.
a) try_to_fill_dentry sets the PENDING flag redundantly for this
dentry, then calls into the autofs4 wait code.
b) the autofs4 wait code takes the waitq mutex and searches for an
entry for 'foo'
Between a and b, P1 is woken up because the mount completed. P1 takes the
wait queue mutex, clears the PENDING flag from the dentry, and removes the
waitqueue entry for 'foo' from the list.
When it releases the waitq mutex, P3 (eventually) acquires it. At this
time, it looks for an existing waitq for 'foo', finds none, and so creates
a new one and calls out to the daemon to mount the 'foo' directory.
Now, the reason that three processes are required to trigger this race is
that, because the PENDING flag is not set on the dentry created by mkdir,
the window for the race would be way to slim for it to ever occur.
Basically, between the testing of d_mountpoint(dentry) and the taking of
the waitq mutex, the mount would have to complete and the daemon would
have to be woken up, and that in turn would have to wake up P1. This is
simply impossible. Add the third process, though, and it becomes slightly
more likely.
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
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>
The autofs4_catatonic_mode() function accesses the wait queue without any
locking but can be called at any time. This could lead to a possible
double free of the name field of the wait and a double fput of the daemon
communication pipe or an fput of a NULL file pointer.
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>
The autofs_wait_queue already contains all of the fields of the
struct qstr, so change it into a qstr.
This patch, from Jeff Moyer, has been modified a liitle by myself.
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
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>
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>
Since we now delay hashing of dentrys until the ->mkdir() call, droping
and re-taking the directory mutex within the ->lookup() function when we
are being called by user space is not needed. This can lead to a race
when other processes are attempting to access the same directory during
mount point directory creation.
In this case we need to hang onto the mutex to ensure we don't get user
processes trying to create a mount request for a newly created dentry
after the mount point entry has already been created. This ensures that
when we need to check a dentry passed to autofs4_wait(), if it is hashed,
it is always the mount point dentry and not a new dentry created by
another lookup during directory creation.
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>
The length of the symlink name has been moved but it needs to be set
before allocating space for it in the dentry info struct. This corrects a
mistake in a recent patch.
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>
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>
This patch series enables the use of a single dentry for lookups prior to
the dentry being hashed and so we no longer need to redo the lookup. This
patch reverts the patch of commit
033790449b.
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>
Correct the error of making a positive dentry negative after it has been
instantiated.
The code that makes this error attempts to re-use the dentry from a
concurrent expire and mount to resolve a race and the dentry used for the
lookup must be negative for mounts to trigger in the required cases. The
fact is that the dentry doesn't need to be re-used because all that is
needed is to preserve the flag that indicates an expire is still
incomplete at the time of the mount request.
This change uses the the dentry to check the flag and wait for the expire
to complete then discards it instead of attempting to re-use it.
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>
Here are some more places where path_{get,put}() can be used instead of
dput()/mntput() pair. Besides that it fixes a bug in autofs4_mount_busy()
where mntput() was called before dput().
Signed-off-by: Jan Blunck <jblunck@suse.de>
Cc: Ian Kent <raven@themaw.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Jeff Moyer has identified a case where the autofs4 function
root.c:try_to_fill_dentry() can return -EBUSY when it should return 0.
Jeff's description of the way this happens is:
"automount starts an expire for directory d. after the callout to the daemon,
but before the rmdir, another process tries to walk into the same directory.
It puts itself onto the waitq, pending the expiration.
When the expire finishes, the second process is woken up. In
try_to_fill_dentry, it does this check:
status = d_invalidate(dentry);
if (status != -EBUSY)
return -EAGAIN;
And status is EBUSY. The dentry still has a non-zero d_inode, and the
flags do not contain LOOKUP_CONTINUE or LOOKUP_DIRECTORY
So, we fall through and return -EBUSY to the caller."
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
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>
Jeff Moyer has identified a race in due to an execution order dependency
in the autofs4 function root.c:try_to_fill_dentry().
Jeff's description of this race is:
"P1 does a lookup of /mount/submount/foo. Since the VFS can't find an entry
for "foo" under /mount/submount, it calls into the autofs4 kernel module to
allocate a new dentry, D1. The kernel creates a new waitq for this lookup and
calls the daemon to perform the mount.
The daemon performs a mkdir of the "foo" directory under /mount/submount,
which ends up creating a *new* dentry, D2.
Then, P2 does a lookup of /mount/submount/foo. The VFS path walking logic
finds a dentry in the dcache, D2, and calls the revalidate function with this.
In the autofs4 revalidate code, we then trigger a mount, since the dentry is
an empty directory that isn't a mountpoint, and so set DCACHE_AUTOFS_PENDING
and call into the wait code to trigger the mount.
The wait code finds our existing waitq entry (since it is keyed off of the
directory name) and adds itself to the list of waiters.
After the daemon finishes the mount, it calls back into the kernel to release
the waiters. When this happens, P1 is woken up and goes about clearing the
DCACHE_AUTOFS_PENDING flag, but it does this in D1! So, given that P1 in our
case is a program that will immediately try to access a file under
/mount/submount/foo, we end up finding the dentry D2 which still has the
pending flag set, and we set out to wait for a mount *again*!
So, one way to address this is to re-do the lookup at the end of
try_to_fill_dentry, and to clear the pending flag on the hashed dentry. This
seems a sane approach to me."
And Jeff's patch does this.
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by-by: Ian Kent <raven@themaw.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Al Viro
Ian Kent
Alexey Dobriyan
James Morris
David Howells
Steven Whitehouse
Jeff Moyer
Jan Blunck