On an error path in inotify_init1 a normal user can trigger a double
free of struct user. This is a regression introduced by a2ae4cc9a1
("inotify: stop kernel memory leak on file creation failure").
We fix this by making sure that if a group exists the user reference is
dropped when the group is cleaned up. We should not explictly drop the
reference on error and also drop the reference when the group is cleaned
up.
The new lifetime rules are that an inotify group lives from
inotify_new_group to the last fsnotify_put_group. Since the struct user
and inotify_devs are directly tied to this lifetime they are only
changed/updated in those two locations. We get rid of all special
casing of struct user or user->inotify_devs.
Signed-off-by: Eric Paris <eparis@redhat.com>
Cc: stable@kernel.org (2.6.37 and up)
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If inotify_init is unable to allocate a new file for the new inotify
group we leak the new group. This patch drops the reference on the
group on file allocation failure.
Reported-by: Vegard Nossum <vegard.nossum@gmail.com>
cc: stable@kernel.org
Signed-off-by: Eric Paris <eparis@redhat.com>
The _IN_ in the naming is reserved for flags only used by inotify. Since I
am about to use this flag for fanotify rename it to be generic like the
rest.
Signed-off-by: Eric Paris <eparis@redhat.com>
All file_operations should get a .llseek operation so we can make
nonseekable_open the default for future file operations without a
.llseek pointer.
The three cases that we can automatically detect are no_llseek, seq_lseek
and default_llseek. For cases where we can we can automatically prove that
the file offset is always ignored, we use noop_llseek, which maintains
the current behavior of not returning an error from a seek.
New drivers should normally not use noop_llseek but instead use no_llseek
and call nonseekable_open at open time. Existing drivers can be converted
to do the same when the maintainer knows for certain that no user code
relies on calling seek on the device file.
The generated code is often incorrectly indented and right now contains
comments that clarify for each added line why a specific variant was
chosen. In the version that gets submitted upstream, the comments will
be gone and I will manually fix the indentation, because there does not
seem to be a way to do that using coccinelle.
Some amount of new code is currently sitting in linux-next that should get
the same modifications, which I will do at the end of the merge window.
Many thanks to Julia Lawall for helping me learn to write a semantic
patch that does all this.
===== begin semantic patch =====
// This adds an llseek= method to all file operations,
// as a preparation for making no_llseek the default.
//
// The rules are
// - use no_llseek explicitly if we do nonseekable_open
// - use seq_lseek for sequential files
// - use default_llseek if we know we access f_pos
// - use noop_llseek if we know we don't access f_pos,
// but we still want to allow users to call lseek
//
@ open1 exists @
identifier nested_open;
@@
nested_open(...)
{
<+...
nonseekable_open(...)
...+>
}
@ open exists@
identifier open_f;
identifier i, f;
identifier open1.nested_open;
@@
int open_f(struct inode *i, struct file *f)
{
<+...
(
nonseekable_open(...)
|
nested_open(...)
)
...+>
}
@ read disable optional_qualifier exists @
identifier read_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
expression E;
identifier func;
@@
ssize_t read_f(struct file *f, char *p, size_t s, loff_t *off)
{
<+...
(
*off = E
|
*off += E
|
func(..., off, ...)
|
E = *off
)
...+>
}
@ read_no_fpos disable optional_qualifier exists @
identifier read_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
@@
ssize_t read_f(struct file *f, char *p, size_t s, loff_t *off)
{
... when != off
}
@ write @
identifier write_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
expression E;
identifier func;
@@
ssize_t write_f(struct file *f, const char *p, size_t s, loff_t *off)
{
<+...
(
*off = E
|
*off += E
|
func(..., off, ...)
|
E = *off
)
...+>
}
@ write_no_fpos @
identifier write_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
@@
ssize_t write_f(struct file *f, const char *p, size_t s, loff_t *off)
{
... when != off
}
@ fops0 @
identifier fops;
@@
struct file_operations fops = {
...
};
@ has_llseek depends on fops0 @
identifier fops0.fops;
identifier llseek_f;
@@
struct file_operations fops = {
...
.llseek = llseek_f,
...
};
@ has_read depends on fops0 @
identifier fops0.fops;
identifier read_f;
@@
struct file_operations fops = {
...
.read = read_f,
...
};
@ has_write depends on fops0 @
identifier fops0.fops;
identifier write_f;
@@
struct file_operations fops = {
...
.write = write_f,
...
};
@ has_open depends on fops0 @
identifier fops0.fops;
identifier open_f;
@@
struct file_operations fops = {
...
.open = open_f,
...
};
// use no_llseek if we call nonseekable_open
////////////////////////////////////////////
@ nonseekable1 depends on !has_llseek && has_open @
identifier fops0.fops;
identifier nso ~= "nonseekable_open";
@@
struct file_operations fops = {
... .open = nso, ...
+.llseek = no_llseek, /* nonseekable */
};
@ nonseekable2 depends on !has_llseek @
identifier fops0.fops;
identifier open.open_f;
@@
struct file_operations fops = {
... .open = open_f, ...
+.llseek = no_llseek, /* open uses nonseekable */
};
// use seq_lseek for sequential files
/////////////////////////////////////
@ seq depends on !has_llseek @
identifier fops0.fops;
identifier sr ~= "seq_read";
@@
struct file_operations fops = {
... .read = sr, ...
+.llseek = seq_lseek, /* we have seq_read */
};
// use default_llseek if there is a readdir
///////////////////////////////////////////
@ fops1 depends on !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier readdir_e;
@@
// any other fop is used that changes pos
struct file_operations fops = {
... .readdir = readdir_e, ...
+.llseek = default_llseek, /* readdir is present */
};
// use default_llseek if at least one of read/write touches f_pos
/////////////////////////////////////////////////////////////////
@ fops2 depends on !fops1 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier read.read_f;
@@
// read fops use offset
struct file_operations fops = {
... .read = read_f, ...
+.llseek = default_llseek, /* read accesses f_pos */
};
@ fops3 depends on !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier write.write_f;
@@
// write fops use offset
struct file_operations fops = {
... .write = write_f, ...
+ .llseek = default_llseek, /* write accesses f_pos */
};
// Use noop_llseek if neither read nor write accesses f_pos
///////////////////////////////////////////////////////////
@ fops4 depends on !fops1 && !fops2 && !fops3 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier read_no_fpos.read_f;
identifier write_no_fpos.write_f;
@@
// write fops use offset
struct file_operations fops = {
...
.write = write_f,
.read = read_f,
...
+.llseek = noop_llseek, /* read and write both use no f_pos */
};
@ depends on has_write && !has_read && !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier write_no_fpos.write_f;
@@
struct file_operations fops = {
... .write = write_f, ...
+.llseek = noop_llseek, /* write uses no f_pos */
};
@ depends on has_read && !has_write && !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier read_no_fpos.read_f;
@@
struct file_operations fops = {
... .read = read_f, ...
+.llseek = noop_llseek, /* read uses no f_pos */
};
@ depends on !has_read && !has_write && !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
@@
struct file_operations fops = {
...
+.llseek = noop_llseek, /* no read or write fn */
};
===== End semantic patch =====
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Julia Lawall <julia@diku.dk>
Cc: Christoph Hellwig <hch@infradead.org>
group->mask is now useless. It was originally a shortcut for fsnotify to
save on performance. These checks are now redundant, so we remove them.
Signed-off-by: Eric Paris <eparis@redhat.com>
Rather than the horrific void ** argument and such just to pass the
fanotify_merge event back to the caller of fsnotify_add_notify_event() have
those things return an event if it was different than the event suggusted to
be added.
Signed-off-by: Eric Paris <eparis@redhat.com>
It can be hard to debug fsnotify since there are so few printks. Use
pr_debug to allow for dynamic debugging.
Signed-off-by: Eric Paris <eparis@redhat.com>
The mask checks in inotify_update_existing_watch() and
inotify_new_watch() are useless because inotify_arg_to_mask() sets
FS_IN_IGNORED and FS_EVENT_ON_CHILD bits anyway.
Signed-off-by: Eric Paris <eparis@redhat.com>
inotify uses bits called IN_* and fsnotify uses bits called FS_*. These
need to line up. This patch adds build time checks to make sure noone can
change these bits so they are not the same.
Signed-off-by: Eric Paris <eparis@redhat.com>
An inotify watch on a directory will send events for children even if those
children have been unlinked. This patch add a new inotify flag IN_EXCL_UNLINK
which allows a watch to specificy they don't care about unlinked children.
This should fix performance problems seen by tasks which add a watch to
/tmp and then are overrun with events when other processes are reading and
writing to unlinked files they created in /tmp.
https://bugzilla.kernel.org/show_bug.cgi?id=16296
Requested-by: Matthias Clasen <mclasen@redhat.com>
Signed-off-by: Eric Paris <eparis@redhat.com>
Since the .31 or so notify rewrite inotify has not sent events about
inodes which are unmounted. This patch restores those events.
Signed-off-by: Eric Paris <eparis@redhat.com>
The symbol inotify_max_user_watches is not used outside this
file and should be static.
Signed-off-by: H Hartley Sweeten <hsweeten@visionengravers.com>
Cc: John McCutchan <john@johnmccutchan.com>
Cc: Robert Love <rlove@rlove.org>
Cc: Eric Paris <eparis@parisplace.org>
Signed-off-by: Eric Paris <eparis@redhat.com>
Each group can define their own notification (and secondary_q) merge
function. Inotify does tail drop, fanotify does matching and drop which
can actually allocate a completely new event. But for fanotify to properly
deal with permissions events it needs to know the new event which was
ultimately added to the notification queue. This patch just implements a
void ** argument which is passed to the merge function. fanotify can use
this field to pass the new event back to higher layers.
Signed-off-by: Eric Paris <eparis@redhat.com>
for fanotify to properly deal with permissions events
inotify marks must pin inodes in core. dnotify doesn't technically need to
since they are closed when the directory is closed. fanotify also need to
pin inodes in core as it works today. But the next step is to introduce
the concept of 'ignored masks' which is actually a mask of events for an
inode of no interest. I claim that these should be liberally sent to the
kernel and should not pin the inode in core. If the inode is brought back
in the listener will get an event it may have thought excluded, but this is
not a serious situation and one any listener should deal with.
This patch lays the ground work for non-pinning inode marks by using lazy
inode pinning. We do not pin a mark until it has a non-zero mask entry. If a
listener new sets a mask we never pin the inode.
Signed-off-by: Eric Paris <eparis@redhat.com>
currently all marking is done by functions in inode-mark.c. Some of this
is pretty generic and should be instead done in a generic function and we
should only put the inode specific code in inode-mark.c
Signed-off-by: Eric Paris <eparis@redhat.com>
All callers to fsnotify_find_mark_entry() except one take and
release inode->i_lock around the call. Take the lock inside
fsnotify_find_mark_entry() instead.
Signed-off-by: Andreas Gruenbacher <agruen@suse.de>
Signed-off-by: Eric Paris <eparis@redhat.com>
The addition of marks on vfs mounts will be simplified if the inode
specific parts of a mark and the vfsmnt specific parts of a mark are
actually in a union so naming can be easy. This patch just implements the
inode struct and the union.
Signed-off-by: Eric Paris <eparis@redhat.com>
fsnotify_obtain_group was intended to be able to find an already existing
group. Nothing uses that functionality. This just renames it to
fsnotify_alloc_group so it is clear what it is doing.
Signed-off-by: Eric Paris <eparis@redhat.com>
The original fsnotify interface has a group-num which was intended to be
able to find a group after it was added. I no longer think this is a
necessary thing to do and so we remove the group_num.
Signed-off-by: Eric Paris <eparis@redhat.com>
inotify only wishes to merge a new event with the last event on the
notification fifo. fanotify is willing to merge any events including by
means of bitwise OR masks of multiple events together. This patch moves
the inotify event merging logic out of the generic fsnotify notification.c
and into the inotify code. This allows each use of fsnotify to provide
their own merge functionality.
Signed-off-by: Eric Paris <eparis@redhat.com>
inotify was supposed to have a dmesg printk ratelimitor which would cause
inotify to only emit one message per boot. The static bool was never set
so it kept firing messages. This patch correctly limits warnings in multiple
places.
Signed-off-by: Eric Paris <eparis@redhat.com>
Prior to 2.6.31 inotify would not reuse watch descriptors until all of
them had been used at least once. After the rewrite inotify would reuse
watch descriptors. The selinux utility 'restorecond' was found to have
problems when watch descriptors were reused. This patch reverts to the
pre inotify rewrite behavior to not reuse watch descriptors.
Signed-off-by: Eric Paris <eparis@redhat.com>
inotify_free_mark casts directly from an fsnotify_mark_entry to an
inotify_inode_mark_entry. This works, but should use container_of instead
for future proofing.
Signed-off-by: Eric Paris <eparis@redhat.com>
This patch allows a task to add a second fsnotify mark to an inode for the
same group. This mark will be added to the end of the inode's list and
this will never be found by the stand fsnotify_find_mark() function. This
is useful if a user wants to add a new mark before removing the old one.
Signed-off-by: Eric Paris <eparis@redhat.com>
This patch moves all of the idr editing operations into their own idr
functions. It makes it easier to prove locking correctness and to to
understand the code flow.
Signed-off-by: Eric Paris <eparis@redhat.com>
There is a race in the inotify add/rm watch code. A task can find and
remove a mark which doesn't have all of it's references. This can
result in a use after free/double free situation.
Task A Task B
------------ -----------
inotify_new_watch()
allocate a mark (refcnt == 1)
add it to the idr
inotify_rm_watch()
inotify_remove_from_idr()
fsnotify_put_mark()
refcnt hits 0, free
take reference because we are on idr
[at this point it is a use after free]
[time goes on]
refcnt may hit 0 again, double free
The fix is to take the reference BEFORE the object can be found in the
idr.
Signed-off-by: Eric Paris <eparis@redhat.com>
Cc: <stable@kernel.org>
inotify_add_watch explictly frees the unused inode mark, but it can just
use the generic code. Just do that.
Signed-off-by: Eric Paris <eparis@redhat.com>
Since commit 7e790dd5fc ("inotify: fix
error paths in inotify_update_watch") inotify changed the manor in which
it gave watch descriptors back to userspace. Previous to this commit
inotify acted like the following:
inotify_add_watch(X, Y, Z) = 1
inotify_rm_watch(X, 1);
inotify_add_watch(X, Y, Z) = 2
but after this patch inotify would return watch descriptors like so:
inotify_add_watch(X, Y, Z) = 1
inotify_rm_watch(X, 1);
inotify_add_watch(X, Y, Z) = 1
which I saw as equivalent to opening an fd where
open(file) = 1;
close(1);
open(file) = 1;
seemed perfectly reasonable. The issue is that quite a bit of userspace
apparently relies on the behavior in which watch descriptors will not be
quickly reused. KDE relies on it, I know some selinux packages rely on
it, and I have heard complaints from other random sources such as debian
bug 558981.
Although the man page implies what we do is ok, we broke userspace so
this patch almost reverts us to the old behavior. It is still slightly
racey and I have patches that would fix that, but they are rather large
and this will fix it for all real world cases. The race is as follows:
- task1 creates a watch and blocks in idr_new_watch() before it updates
the hint.
- task2 creates a watch and updates the hint.
- task1 updates the hint with it's older wd
- task removes the watch created by task2
- task adds a new watch and will reuse the wd originally given to task2
it requires moving some locking around the hint (last_wd) but this should
solve it for the real world and be -stable safe.
As a side effect this patch papers over a bug in the lib/idr code which
is causing a large number WARN's to pop on people's system and many
reports in kerneloops.org. I'm working on the root cause of that idr
bug seperately but this should make inotify immune to that issue.
Signed-off-by: Eric Paris <eparis@redhat.com>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
For consistency drop & in front of every proc_handler. Explicity
taking the address is unnecessary and it prevents optimizations
like stubbing the proc_handlers to NULL.
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Joe Perches <joe@perches.com>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Now that sys_sysctl is a generic wrapper around /proc/sys .ctl_name
and .strategy members of sysctl tables are dead code. Remove them.
Cc: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Seperating the addition and update of marks in inotify resulted in a
regression in that inotify never gets events. The inotify group mask is
always 0. This mask should be updated any time a new mark is added.
Signed-off-by: Eric Paris <eparis@redhat.com>
0db501bd06 introduced a regresion in that it now sends a nul
terminator but the length accounting when checking for space or
reporting to userspace did not take this into account. This corrects
all of the rounding logic.
Signed-off-by: Eric Paris <eparis@redhat.com>
When an event has no pathname, there's no need to pad it with a null byte and
therefore generate an inotify_event sized block of zeros. This fixes a
regression introduced by commit 0db501bd06 where
my system wouldn't finish booting because some process was being confused by
this.
Signed-off-by: Brian Rogers <brian@xyzw.org>
Signed-off-by: Eric Paris <eparis@redhat.com>
Before the rewrite copy_event_to_user always wrote a terqminating '\0'
byte to user space after the filename. Since the rewrite that
terminating byte was skipped if your filename is exactly a multiple of
event_size. Ouch!
So add one byte to name_size before we round up and use clear_user to
set userspace to zero like /dev/zero does instead of copying the
strange nul_inotify_event. I can't quite convince myself len_to_zero
will never exceed 16 and even if it doesn't clear_user should be more
efficient and a more accurate reflection of what the code is trying to
do.
Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
Signed-off-by: Eric Paris <eparis@redhat.com>
The are races around the idr storage of inotify watches. It's possible
that a watch could be found from sys_inotify_rm_watch() in the idr, but it
could be removed from the idr before that code does it's removal. Move the
locking and the refcnt'ing so that these have to happen atomically.
Signed-off-by: Eric Paris <eparis@redhat.com>
There is nothing known wrong with the inotify watch addition/modification
but this patch seperates the two code paths to make them each easy to
verify as correct.
Signed-off-by: Eric Paris <eparis@redhat.com>
The inotify_add_watch man page specifies that inotify_add_watch() will
return a non-negative integer. However, historically the inotify
watches started at 1, not at 0.
Turns out that the inotifywait program provided by the inotify-tools
package doesn't properly handle a 0 watch descriptor. In 7e790dd5 we
changed from starting at 1 to starting at 0. This patch starts at 1,
just like in previous kernels, but also just like in previous kernels
it's possible for it to wrap back to 0. This preserves the kernel
functionality exactly like it was before the patch (neither method broke
the spec)
Signed-off-by: Eric Paris <eparis@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
inotify decides if private data it passed to get added to an event was
used by checking list_empty(). But it's possible that the event may
have been dequeued and the private event removed so it would look empty.
The fix is to use the return code from fsnotify_add_notify_event rather
than looking at the list.
Signed-off-by: Eric Paris <eparis@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
inotify can have a watchs removed under filesystem reclaim.
=================================
[ INFO: inconsistent lock state ]
2.6.31-rc2 #16
---------------------------------
inconsistent {IN-RECLAIM_FS-W} -> {RECLAIM_FS-ON-W} usage.
khubd/217 [HC0[0]:SC0[0]:HE1:SE1] takes:
(iprune_mutex){+.+.?.}, at: [<c10ba899>] invalidate_inodes+0x20/0xe3
{IN-RECLAIM_FS-W} state was registered at:
[<c10536ab>] __lock_acquire+0x2c9/0xac4
[<c1053f45>] lock_acquire+0x9f/0xc2
[<c1308872>] __mutex_lock_common+0x2d/0x323
[<c1308c00>] mutex_lock_nested+0x2e/0x36
[<c10ba6ff>] shrink_icache_memory+0x38/0x1b2
[<c108bfb6>] shrink_slab+0xe2/0x13c
[<c108c3e1>] kswapd+0x3d1/0x55d
[<c10449b5>] kthread+0x66/0x6b
[<c1003fdf>] kernel_thread_helper+0x7/0x10
[<ffffffff>] 0xffffffff
Two things are needed to fix this. First we need a method to tell
fsnotify_create_event() to use GFP_NOFS and second we need to stop using
one global IN_IGNORED event and allocate them one at a time. This solves
current issues with multiple IN_IGNORED on a queue having tail drop
problems and simplifies the allocations since we don't have to worry about
two tasks opperating on the IGNORED event concurrently.
Signed-off-by: Eric Paris <eparis@redhat.com>
