linux.git/kernel/exit.c, branch v2.6.21.7

pi-futex: Fix exit races and locking problems

2007-08-04T16:10:23+00:00

1. New entries can be added to tsk->pi_state_list after task completed
   exit_pi_state_list(). The result is memory leakage and deadlocks.

2. handle_mm_fault() is called under spinlock. The result is obvious.

3. results in self-inflicted deadlock inside glibc.
   Sometimes futex_lock_pi returns -ESRCH, when it is not expected
   and glibc enters to for(;;) sleep() to simulate deadlock. This problem
   is quite obvious and I think the patch is right. Though it looks like
   each "if" in futex_lock_pi() got some stupid special case "else if". :-)

4. sometimes futex_lock_pi() returns -EDEADLK,
   when nobody has the lock. The reason is also obvious (see comment
   in the patch), but correct fix is far beyond my comprehension.
   I guess someone already saw this, the chunk:

                        if (rt_mutex_trylock(&q.pi_state->pi_mutex))
                                ret = 0;

   is obviously from the same opera. But it does not work, because the
   rtmutex is really taken at this point: wake_futex_pi() of previous
   owner reassigned it to us. My fix works. But it looks very stupid.
   I would think about removal of shift of ownership in wake_futex_pi()
   and making all the work in context of process taking lock.

From: Thomas Gleixner 

Fix 1) Avoid the tasklist lock variant of the exit race fix by adding
    an additional state transition to the exit code.

    This fixes also the issue, when a task with recursive segfaults
    is not able to release the futexes.

Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH
    problem finally.

Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup
    in the lock protected section by using the in_atomic userspace access
    functions.
	
    This removes also the ugly lock drop / unqueue inside of fixup_pi_state()

Fix 4) Fix a stale lock in the error path of futex_wake_pi()

Added some error checks for verification.

The -EDEADLK problem is solved by the rtmutex fixups.

Cc: Alexey Kuznetsov 
Signed-off-by: Thomas Gleixner 
Acked-by: Ingo Molnar 
Signed-off-by: Chris Wright 
Signed-off-by: Greg Kroah-Hartman

[PATCH] pid: Properly detect orphaned process groups in exit_notify

2007-03-29T15:16:23+00:00

In commit 0475ac0845f9295bc5f69af45f58dff2c104c8d1 when converting the
orphaned process group handling to use struct pid I made a small
mistake.  I accidentally replaced an == with a !=.

Besides just being a dumb thing to do apparently this has a bad side
effect.  The improper orphaned process group detection causes kwin to
die after a suspend/resume cycle.

I'm amazed this patch has been around as long as it has without anyone
else noticing something funny going on.

And the following people deserve credit for spotting and helping
to reproduce this.

Thanks to: Sid Boyce 
Thanks to: "Michael Wu"

Signed-off-by: "Eric W. Biederman" 
Signed-off-by: Linus Torvalds

[PATCH] pid: replace is_orphaned_pgrp with is_current_pgrp_orphaned

2007-02-12T17:48:32+00:00

Every call to is_orphaned_pgrp passed in process_group(current) which is racy
with respect to another thread changing our process group.  It didn't bite us
because we were dealing with integers and the worse we would get would be a
stale answer.

In switching the checks to use struct pid to be a little more efficient and
prepare the way for pid namespaces this race became apparent.

So I simplified the calls to the more specialized is_current_pgrp_orphaned so
I didn't have to worry about making logic changes to avoid the race.

Signed-off-by: Eric W. Biederman 
Cc: Alan Cox 
Cc: Oleg Nesterov 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

[PATCH] pid: use struct pid for talking about process groups in exitc

2007-02-12T17:48:32+00:00

Modify has_stopped_jobs and will_become_orphan_pgrp to use struct pid based
process groups.  This reduces the number of hash tables looks ups and paves
the way for multiple pid spaces.

Signed-off-by: Eric W. Biederman 
Cc: Alan Cox 
Cc: Oleg Nesterov 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

[PATCH] pid: make session_of_pgrp use struct pid instead of pid_t

2007-02-12T17:48:31+00:00

To properly implement a pid namespace I need to deal exclusively in terms of
struct pid, because pid_t values become ambiguous.

To this end session_of_pgrp is transformed to take and return a struct pid
pointer.  To avoid the need to worry about reference counting I now require my
caller to hold the appropriate locks.  Leaving callers repsonsible for
increasing the reference count if they need access to the result outside of
the locks.

Since session_of_pgrp currently only has one caller and that caller simply
uses only test the result for equality with another process group, the locking
change means I don't actually have to acquire the tasklist_lock at all.

tiocspgrp is also modified to take and release the lock.  The logic there is a
little more complicated but nothing I won't need when I convert pgrp of a tty
to a struct pid pointer.

Signed-off-by: Eric W. Biederman 
Cc: Alan Cox 
Cc: Oleg Nesterov 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

[PATCH] close_files(): add scheduling point

2007-02-12T17:48:30+00:00

close_files() can sometimes take long enough to trigger the soft lockup
detector.

Cc: Eric Dumazet 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

[PATCH] Numerous fixes to kernel-doc info in source files.

2007-02-11T18:51:32+00:00

A variety of (mostly) innocuous fixes to the embedded kernel-doc content in
source files, including:

  * make multi-line initial descriptions single line
  * denote some function names, constants and structs as such
  * change erroneous opening '/*' to '/**' in a few places
  * reword some text for clarity

Signed-off-by: Robert P. J. Day 
Cc: "Randy.Dunlap" 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

[PATCH] namespaces: fix task exit disaster

2007-01-30T21:40:36+00:00

This is based on a patch by Eric W.  Biederman, who pointed out that pid
namespaces are still fake, and we only have one ever active.

So for the time being, we can modify any code which could access
tsk->nsproxy->pid_ns during task exit to just use &init_pid_ns instead,
and move the exit_task_namespaces call in do_exit() back above
exit_notify(), so that an exiting nfs server has a valid tsk->sighand to
work with.

Long term, pulling pid_ns out of nsproxy might be the cleanest solution.

Signed-off-by: Eric W. Biederman 

[ Eric's patch fixed to take care of free_pid() too ]

Signed-off-by: Serge E. Hallyn 
Signed-off-by: Linus Torvalds

Revert "[PATCH] namespaces: fix exit race by splitting exit"

2007-01-30T21:35:18+00:00

This reverts commit 7a238fcba0629b6f2edbcd37458bae56fcf36be5 in
preparation for a better and simpler fix proposed by Eric Biederman
(and fixed up by Serge Hallyn)

Acked-by: Serge E. Hallyn 
Signed-off-by: Linus Torvalds

[PATCH] namespaces: fix exit race by splitting exit

2007-01-30T16:26:44+00:00

Fix exit race by splitting the nsproxy putting into two pieces.  First
piece reduces the nsproxy refcount.  If we dropped the last reference, then
it puts the mnt_ns, and returns the nsproxy as a hint to the caller.  Else
it returns NULL.  The second piece of exiting task namespaces sets
tsk->nsproxy to NULL, and drops the references to other namespaces and
frees the nsproxy only if an nsproxy was passed in.

A little awkward and should probably be reworked, but hopefully it fixes
the NFS oops.

Signed-off-by: Serge E. Hallyn 
Cc: Herbert Poetzl 
Cc: Oleg Nesterov 
Cc: "Eric W. Biederman" 
Cc: Cedric Le Goater 
Cc: Daniel Hokka Zakrisson 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds