workqueues: implement flush_work()

Most of users of flush_workqueue() can be changed to use cancel_work_sync(),
but sometimes we really need to wait for the completion and cancelling is not
an option. schedule_on_each_cpu() is good example.

Add the new helper, flush_work(work), which waits for the completion of the
specific work_struct. More precisely, it "flushes" the result of of the last
queue_work() which is visible to the caller.

For example, this code

queue_work(wq, work);
/* WINDOW */
queue_work(wq, work);

flush_work(work);

doesn't necessary work "as expected". What can happen in the WINDOW above is

- wq starts the execution of work->func()

- the caller migrates to another CPU

now, after the 2nd queue_work() this work is active on the previous CPU, and
at the same time it is queued on another. In this case flush_work(work) may
return before the first work->func() completes.

It is trivial to add another helper

int flush_work_sync(struct work_struct *work)
{
return flush_work(work) || wait_on_work(work);
}

which works "more correctly", but it has to iterate over all CPUs and thus
it much slower than flush_work().

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Max Krasnyansky <maxk@qualcomm.com>
Acked-by: Jarek Poplawski <jarkao2@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

workqueues: insert_work: use "list_head *" instead of "int tail"

insert_work() inserts the new work_struct before or after cwq->worklist,
depending on the "int tail" parameter. Change it to accept "list_head *"
instead, this shrinks .text a bit and allows us to insert the barrier
after specific work_struct.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Jarek Poplawski <jarkao2@gmail.com>
Cc: Max Krasnyansky <maxk@qualcomm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: format_corename: fix the "core_uses_pid" logic

I don't understand why the multi-thread coredump implies the core_uses_pid
behaviour, but we shouldn't use mm->mm_users for that.  This counter can
be incremented by get_task_mm().  Use the valued returned by
coredump_wait() instead.

Also, remove the "const char *pattern" argument, format_corename() can use
core_pattern directly.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: kill mm->core_done

Now that we have core_state->dumper list we can use it to wake up the
sub-threads waiting for the coredump completion.

This uglifies the code and .text grows by 47 bytes, but otoh mm_struct
lessens by sizeof(struct completion).  Also, with this change we can
decouple exit_mm() from the coredumping code.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: elf_fdpic_core_dump: use core_state->dumper list

Kill the nasty rcu_read_lock() + do_each_thread() loop, use the list
encoded in mm->core_state instead, s/GFP_ATOMIC/GFP_KERNEL/.

This patch allows futher cleanups in binfmt_elf_fdpic.c.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: elf_core_dump: use core_state->dumper list

Kill the nasty rcu_read_lock() + do_each_thread() loop, use the list
encoded in mm->core_state instead, s/GFP_ATOMIC/GFP_KERNEL/.

This patch allows futher cleanups in binfmt_elf.c, in particular we can
kill the parallel info->threads list.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: construct the list of coredumping threads at startup time

binfmt->core_dump() has to iterate over the all threads in system in order
to find the coredumping threads and construct the list using the
GFP_ATOMIC allocations.

With this patch each thread allocates the list node on exit_mm()'s stack and
adds itself to the list.

This allows us to do further changes:

- simplify ->core_dump()

- change exit_mm() to clear ->mm first, then wait for ->core_done.
  this makes the coredumping process visible to oom_kill

- kill mm->core_done

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: make mm->core_state visible to ->core_dump()

Move the "struct core_state core_state" from coredump_wait() to
do_coredump(), this makes mm->core_state visible to binfmt->core_dump().

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: turn core_state->nr_threads into atomic_t

Turn core_state->nr_threads into atomic_t and kill now unneeded
down_write(&mm->mmap_sem) in exit_mm().

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: simplify core_state->nr_threads calculation

Change zap_process() to return int instead of incrementing
mm->core_state->nr_threads directly.  Change zap_threads() to set
mm->core_state only on success.

This patch restores the original size of .text, and more importantly now
->nr_threads is used in two places only.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: move mm->core_waiters into struct core_state

Move mm->core_waiters into "struct core_state" allocated on stack.  This
shrinks mm_struct a little bit and allows further changes.

This patch mostly does s/core_waiters/core_state.  The only essential
change is that coredump_wait() must clear mm->core_state before return.

The coredump_wait()'s path is uglified and .text grows by 30 bytes, this
is fixed by the next patch.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: turn mm->core_startup_done into the pointer to struct core_state

mm->core_startup_done points to "struct completion startup_done" allocated
on the coredump_wait()'s stack.  Introduce the new structure, core_state,
which holds this "struct completion".  This way we can add more info
visible to the threads participating in coredump without enlarging
mm_struct.

No changes in affected .o files.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: elf_core_dump: skip kernel threads

linux_binfmt->core_dump() runs before the process does exit_aio(), this
means that we can hit the kernel thread which shares the same ->mm.
Afaics, nothing really bad can happen, but perhaps it makes sense to fix
this minor bug.

It is sad we have to iterate over all threads in system and use
GFP_ATOMIC.  Hopefully we can kill theses ugly do_each_thread()s, but this
needs some nontrivial changes in mm_struct and do_coredump.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: zap_threads() must skip kernel threads

The main loop in zap_threads() must skip kthreads which may use the same
mm.  Otherwise we "kill" this thread erroneously (for example, it can not
fork or exec after that), and the coredumping task stucks in the
TASK_UNINTERRUPTIBLE state forever because of the wrong ->core_waiters
count.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kill PF_BORROWED_MM in favour of PF_KTHREAD

Kill PF_BORROWED_MM.  Change use_mm/unuse_mm to not play with ->flags, and
do s/PF_BORROWED_MM/PF_KTHREAD/ for a couple of other users.

No functional changes yet.  But this allows us to do further
fixes/cleanups.

oom_kill/ptrace/etc often check "p->mm != NULL" to filter out the
kthreads, this is wrong because of use_mm().  The problem with
PF_BORROWED_MM is that we need task_lock() to avoid races.  With this
patch we can check PF_KTHREAD directly, or use a simple lockless helper:

/* The result must not be dereferenced !!! */
struct mm_struct *__get_task_mm(struct task_struct *tsk)
{
if (tsk->flags & PF_KTHREAD)
return NULL;
return tsk->mm;
}

Note also ecard_task().  It runs with ->mm != NULL, but it's the kernel
thread without PF_BORROWED_MM.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

introduce PF_KTHREAD flag

Introduce the new PF_KTHREAD flag to mark the kernel threads.  It is set
by INIT_TASK() and copied to the forked childs (we could set it in
kthreadd() along with PF_NOFREEZE instead).

daemonize() was changed as well.  In that case testing of PF_KTHREAD is
racy, but daemonize() is hopeless anyway.

This flag is cleared in do_execve(), before search_binary_handler().
Probably not the best place, we can do this in exec_mmap() or in
start_thread(), or clear it along with PF_FORKNOEXEC.  But I think this
doesn't matter in practice, and if do_execve() fails kthread should die
soon.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ptrace: simplify ptrace_stop()->sigkill_pending() path

1. SIGKILL can't be blocked, remove this check from sigkill_pending().

2. When ptrace_stop() sees sigkill_pending() == T, it can just return.
   Kill "int killed" and simplify the code. This also is more correct,
   the tracer shouldn't see us in TASK_TRACED if we are not going to
   stop.

I strongly believe this code needs further changes.  We should do the "was
this task killed" check unconditionally, currently it depends on
arch_ptrace_stop_needed().  On the other hand, sigkill_pending() isn't
very clever.  If the task was killed tkill(SIGKILL), the signal can be
already dequeued if the caller is do_exit().

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ptrace: give more respect to SIGKILL

ptrace_stop() has some complicated checks to prevent the scheduling in the
TASK_TRACED state with the pending SIGKILL, but these checks are racy, and
they depend on arch_ptrace_stop_needed().

This patch assumes that the traced task should die asap if it was killed by
SIGKILL, in that case schedule()->signal_pending_state() has no reason to
ignore the TASK_WAKEKILL part of TASK_TRACED, and we can kill this nasty
special case.

Note: do_exit()->ptrace_notify() is special, the killed task can already
dequeue SIGKILL at this point. Another indication that fatal_signal_pending()
is not exactly right.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

include/asm/ptrace.h userspace headers cleanup

This patch contains the following cleanups for the asm/ptrace.h
userspace headers:

- include/asm-generic/Kbuild.asm already lists ptrace.h, remove
  the superfluous listings in the Kbuild files of the following
  architectures:
  - cris
  - frv
  - powerpc
  - x86
- don't expose function prototypes and macros to userspace:
  - arm
  - blackfin
  - cris
  - mn10300
  - parisc
- remove #ifdef CONFIG_'s around #define's:
  - blackfin
  - m68knommu
- sh: AFAIK __SH5__ should work in both kernel and userspace,
      no need to leak CONFIG_SUPERH64 to userspace
- xtensa: cosmetical change to remove empty
            #ifndef __ASSEMBLY__ #else #endif
          from the userspace headers

Not changed by this patch is the fact that the following architectures
have a different struct pt_regs depending on CONFIG_ variables:
- h8300
- m68knommu
- mips

This does not work in userspace.

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Cc: <linux-arch@vger.kernel.org>
Cc: Roland McGrath <roland@redhat.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Greg Ungerer <gerg@uclinux.org>
Acked-by: Paul Mundt <lethal@linux-sh.org>
Acked-by: Grant Grundler <grundler@parisc-linux.org>
Acked-by: Jesper Nilsson <jesper.nilsson@axis.com>
Acked-by: Chris Zankel <chris@zankel.net>
Acked-by: David Howells <dhowells@redhat.com>
Acked-by: Paul Mackerras <paulus@samba.org>
Acked-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel/signal.c: change vars pid and tgid types to pid_t

Change the type of pid and tgid variables from int to the POSIX type
pid_t.

Signed-off-by: Gustavo F. Padovan <gustavo@las.ic.unicamp.br>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: make siginfo_t si_utime + si_sstime report times in USER_HZ, not HZ

In the switch to configurable HZ in 2.6, the treatment of the si_utime and
si_stime fields that are exposed to userland via the siginfo structure
looks to have been botched.  As things stand, these fields report times in
units of HZ, so that userland gets information that varies depending on
the HZ that the kernel was configured with.  This patch changes the
reported values to use USER_HZ units.

Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com>
Acked-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coredump: zap_threads: comments && use while_each_thread()

No changes in fs/exec.o

The for_each_process() loop in zap_threads() is very subtle, it is not
clear why we don't race with fork/exit/exec.  Add the fat comment.

Also, change the code to use while_each_thread().

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: do_signal_stop: kill the SIGNAL_UNKILLABLE check

fae5fa44f1fd079ffbed8e0add929dd7bbd1347f changed do_signal_stop() to check
SIGNAL_UNKILLABLE, this wasn't needed.  If signal_group_exit() == F, the
signal sent to SIGNAL_UNKILLABLE task must be already filtered out by the
caller, get_signal_to_deliver().  And if signal_group_exit() == T we are
not going to stop.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: dequeue_signal: don't check SIGNAL_GROUP_EXIT when setting SIGNAL_STOP_DEQUEUED

dequeue_signal() checks SIGNAL_GROUP_EXIT before setting
SIGNAL_STOP_DEQUEUED.  This was added by
788e05a67c343fa22f2ae1d3ca264e7f15c25eaf a long ago to avoid the
coredump/SIGSTOP race.

Since then the related code was changed, and now this subtle check is both
incomplete and unneeded at the same time.  It is incomplete because
nowadays exec() doesn't set SIGNAL_GROUP_EXIT, so in fact we should check
signal_group_exit() to avoid a similar race.  Fortunately, we doesn't need
the check at all.  The only function which relies on SIGNAL_STOP_DEQUEUED
is do_signal_stop(), and it ignores this flag if signal_group_exit() == T,
this covers the SIGNAL_GROUP_EXIT case.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

__exit_signal: don't take rcu lock

There is no reason for rcu_read_lock() in __exit_signal().  tsk->sighand
can only be changed if tsk does exec, obviously this is not possible.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: change collect_signal() to return void

With the recent changes collect_signal() always returns true.  Change it
to return void and update the single caller.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: collect_signal: simplify the "still_pending" logic

Factor out sigdelset() calls and remove the "still_pending" variable.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

signals: collect_signal: remove the unneeded sigismember() check

collect_signal() checks sigismember(&list->signal, sig), this is not
needed.  This "sig" was just found by next_signal(), so it must be valid.

We have a (completely broken) call to ->notifier in between, but it must
not play with sigpending->signal bits or unlock ->siglock.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

posix timers: release_posix_timer: kill the bogus put_task_struct(->it_process);

release_posix_timer() can't be called with ->it_process != NULL.  Once
sys_timer_create() sets ->it_process it must not call
release_posix_timer(), otherwise we can race with another thread doing
sys_timer_delete(), this timer is visible to idr_find() and unlocked.

The same is true for two other callers (actually, for any possible
caller), sys_timer_delete() and itimer_delete().  They must clear
->it_process before unlock_timer() + release_posix_timer().

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

posix timers: timer_delete: remove the bogus "->it_process != NULL" check

sys_timer_delete() and itimer_delete() check "timer->it_process != NULL",
this looks completely bogus.  ->it_process == NULL means that this timer
is already under destruction or it is not fully initialized, this must not
happen.

sys_timer_delete: the timer is locked, and lock_timer() can't succeed
if ->it_process == NULL.

itimer_delete: it is called by exit_itimers() when there are no other
threads which can play with signal_struct->posix_timers.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: john stultz <johnstul@us.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpuset: two minor code-cleanups

In cpuset_update_task_memory_state() local variable struct task_struct
*tsk = current;

And local variable tsk is used 14 times and statement task_cs(tsk) is used
twice in this function.  So using task_cs(tsk) instead of task_cs(current)
is better for readability.

And "(struct cgroup_scanner *)&scan" is not good for readability also.
(and "container_of" is used in cpuset_do_move_task(), not
"(cpuset_hotplug_scanner *)scan")

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Acked-by: Paul Menage <menage@google.com>
Cc: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpuset: code-cleanup for started_after

cgroup(cgroup_scan_tasks) will initialize heap->gt for us.  This patch
removes started_after() and its helper-function.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Acked-by: Paul Menage <menage@google.com>
Cc: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpuset: don't pass empty cpumasks to partition_sched_domains()

I create lots of empty cpusets(empty cpumasks) and turn off the
"sched_load_balance" in top cpuset.

I found that all these empty cpumasks are passed to
partition_sched_domains() in rebuild_sched_domains(), it's very
time-consuming for partition_sched_domains() and it's not need.

It also reduce memory consumed and some works in rebuild_sched_domains()
too.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Acked-by: Paul Menage <menage@google.com>
Cc: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpuset: avoid unnecessary sched domains rebuilding

When changing 'sched_relax_domain_level', don't rebuild sched domains if
'cpus' is empty or 'sched_load_balance' is not set.

Also make the comments of rebuild_sched_domains() more readable.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpusets: update task's cpus_allowed and mems_allowed after CPU/NODE offline/online

The bug is that a task may run on the cpu/node which is not in its
cpuset.cpus/ cpuset.mems.

It can be reproduced by the following commands:
-----------------------------------
# mkdir /dev/cpuset
# mount -t cpuset xxx /dev/cpuset
# mkdir /dev/cpuset/0
# echo 0-1 > /dev/cpuset/0/cpus
# echo 0 > /dev/cpuset/0/mems
# echo $$ > /dev/cpuset/0/tasks
# echo 0 > /sys/devices/system/cpu/cpu1/online
# echo 1 > /sys/devices/system/cpu/cpu1/online
-----------------------------------

There is only CPU0 in cpuset.cpus, but the task in this cpuset runs on
both CPU0 and CPU1.

It is because the task's cpu_allowed didn't get updated after we did CPU
offline/online manipulation.  Similar for mem_allowed.

This patch fixes this bug expect for root cpuset.  Because there is a
problem about root cpuset, in that whether it is necessary to update all
the tasks in root cpuset or not after cpu/node offline/online.

If updating, some kernel threads which is bound into a specified cpu will
be unbound.

If not updating, there is a bug in root cpuset.  This bug is also caused
by offline/online manipulation.  For example, there is a dual-cpu machine.
 we create a sub cpuset in root cpuset and assign 1 to its cpus.  And then
we attach some tasks into this sub cpuset.  After this, we offline CPU1.
Now, the tasks in this new cpuset are moved into root cpuset automatically
because there is no cpu in sub cpuset.  Then we online CPU1, we find all
the tasks which doesn't belong to root cpuset originally just run on CPU0.

Maybe we need to add a flag in the task_struct to mark which task can't be
unbound?

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Acked-by: Paul Jackson <pj@sgi.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpusets: restructure the function update_cpumask() and update_nodemask()

Extract two functions from update_cpumask() and update_nodemask().They
will be used later for updating tasks' cpus_allowed and mems_allowed after
CPU/NODE offline/online.

[lizf@cn.fujitsu.com: build fix]
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Acked-by: Paul Jackson <pj@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: limit change shrink usage

Shrinking memory usage at limit change.

[akpm@linux-foundation.org: coding-style fixes]
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

res_counter: limit change support ebusy

Add an interface to set limit.  This is necessary to memory resource
controller because it shrinks usage at set limit.

Other controllers may not need this interface to shrink usage because
shrinking is not necessary or impossible.

Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: clean up checking of the disabled flag

Those checks are unnecessary, because when the subsystem is disabled
it can't be mounted, so those functions won't get called.

The check is needed in functions which will be called in other places
except cgroup.

[hugh@veritas.com: further checking of disabled flag]
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: remove a redundant check

Because of remove refcnt patch, it's very rare case to that
mem_cgroup_charge_common() is called against a page which is accounted.

mem_cgroup_charge_common() is called when.
 1. a page is added into file cache.
 2. an anon page is _newly_ mapped.

A racy case is that a newly-swapped-in anonymous page is referred from
prural threads in do_swap_page() at the same time.
(a page is not Locked when mem_cgroup_charge() is called from do_swap_page.)

Another case is shmem. It charges its page before calling add_to_page_cache().
Then, mem_cgroup_charge_cache() is called twice. This case is handled in
mem_cgroup_cache_charge(). But this check may be too hacky...

Signed-off-by : KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: add hints for branch

Showing brach direction for obvious conditions.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: helper function for relcaim from shmem.

A new call, mem_cgroup_shrink_usage() is added for shmem handling and
relacing non-standard usage of mem_cgroup_charge/uncharge.

Now, shmem calls mem_cgroup_charge() just for reclaim some pages from
mem_cgroup.  In general, shmem is used by some process group and not for
global resource (like file caches).  So, it's reasonable to reclaim pages
from mem_cgroup where shmem is mainly used.

[hugh@veritas.com: shmem_getpage release page sooner]
[hugh@veritas.com: mem_cgroup_shrink_usage css_put]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: remove refcnt from page_cgroup

memcg: performance improvements

Patch Description
 1/5 ... remove refcnt fron page_cgroup patch (shmem handling is fixed)
 2/5 ... swapcache handling patch
 3/5 ... add helper function for shmem's memory reclaim patch
 4/5 ... optimize by likely/unlikely ppatch
 5/5 ... remove redundunt check patch (shmem handling is fixed.)

Unix bench result.

== 2.6.26-rc2-mm1 + memory resource controller
Execl Throughput                           2915.4 lps   (29.6 secs, 3 samples)
C Compiler Throughput                      1019.3 lpm   (60.0 secs, 3 samples)
Shell Scripts (1 concurrent)               5796.0 lpm   (60.0 secs, 3 samples)
Shell Scripts (8 concurrent)               1097.7 lpm   (60.0 secs, 3 samples)
Shell Scripts (16 concurrent)               565.3 lpm   (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks    1022128.0 KBps  (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks   544057.0 KBps  (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks    346481.0 KBps  (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks      319325.0 KBps  (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks     148788.0 KBps  (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks       99051.0 KBps  (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks    2058917.0 KBps  (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks   1606109.0 KBps  (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks    854789.0 KBps  (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places         126145.2 lpm   (30.0 secs, 3 samples)

                     INDEX VALUES
TEST                                        BASELINE     RESULT      INDEX

Execl Throughput                                43.0     2915.4      678.0
File Copy 1024 bufsize 2000 maxblocks         3960.0   346481.0      875.0
File Copy 256 bufsize 500 maxblocks           1655.0    99051.0      598.5
File Copy 4096 bufsize 8000 maxblocks         5800.0   854789.0     1473.8
Shell Scripts (8 concurrent)                     6.0     1097.7     1829.5
                                                                 =========
     FINAL SCORE                                                     991.3

== 2.6.26-rc2-mm1 + this set ==
Execl Throughput                           3012.9 lps   (29.9 secs, 3 samples)
C Compiler Throughput                       981.0 lpm   (60.0 secs, 3 samples)
Shell Scripts (1 concurrent)               5872.0 lpm   (60.0 secs, 3 samples)
Shell Scripts (8 concurrent)               1120.3 lpm   (60.0 secs, 3 samples)
Shell Scripts (16 concurrent)               578.0 lpm   (60.0 secs, 3 samples)
File Read 1024 bufsize 2000 maxblocks    1003993.0 KBps  (30.0 secs, 3 samples)
File Write 1024 bufsize 2000 maxblocks   550452.0 KBps  (30.0 secs, 3 samples)
File Copy 1024 bufsize 2000 maxblocks    347159.0 KBps  (30.0 secs, 3 samples)
File Read 256 bufsize 500 maxblocks      314644.0 KBps  (30.0 secs, 3 samples)
File Write 256 bufsize 500 maxblocks     151852.0 KBps  (30.0 secs, 3 samples)
File Copy 256 bufsize 500 maxblocks      101000.0 KBps  (30.0 secs, 3 samples)
File Read 4096 bufsize 8000 maxblocks    2033256.0 KBps  (30.0 secs, 3 samples)
File Write 4096 bufsize 8000 maxblocks   1611814.0 KBps  (30.0 secs, 3 samples)
File Copy 4096 bufsize 8000 maxblocks    847979.0 KBps  (30.0 secs, 3 samples)
Dc: sqrt(2) to 99 decimal places         128148.7 lpm   (30.0 secs, 3 samples)

                     INDEX VALUES
TEST                                        BASELINE     RESULT      INDEX

Execl Throughput                                43.0     3012.9      700.7
File Copy 1024 bufsize 2000 maxblocks         3960.0   347159.0      876.7
File Copy 256 bufsize 500 maxblocks           1655.0   101000.0      610.3
File Copy 4096 bufsize 8000 maxblocks         5800.0   847979.0     1462.0
Shell Scripts (8 concurrent)                     6.0     1120.3     1867.2
                                                                 =========
     FINAL SCORE                                                    1004.6

This patch:

Remove refcnt from page_cgroup().

After this,

 * A page is charged only when !page_mapped() && no page_cgroup is assigned.
* Anon page is newly mapped.
* File page is added to mapping->tree.

 * A page is uncharged only when
* Anon page is fully unmapped.
* File page is removed from LRU.

There is no change in behavior from user's view.

This patch also removes unnecessary calls in rmap.c which was used only for
refcnt mangement.

[akpm@linux-foundation.org: fix warning]
[hugh@veritas.com: fix shmem_unuse_inode charging]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: better migration handling

This patch changes page migration under memory controller to use a
different algorithm.  (thanks to Christoph for new idea.)

Before:
 - page_cgroup is migrated from an old page to a new page.
After:
 - a new page is accounted , no reuse of page_cgroup.

Pros:

 - We can avoid compliated lock depndencies and races in migration.

Cons:

 - new param to mem_cgroup_charge_common().

 - mem_cgroup_getref() is added for handling ref_cnt ping-pong.

This version simplifies complicated lock dependency in page migraiton
under memory resource controller.

  new refcnt sequence is following.

a mapped page:
  prepage_migration() ..... +1 to NEW page
  try_to_unmap()      ..... all refs to OLD page is gone.
  move_pages()        ..... +1 to NEW page if page cache.
  remap...            ..... all refs from *map* is added to NEW one.
  end_migration()     ..... -1 to New page.

  page's mapcount + (page_is_cache) refs are added to NEW one.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: avoid unnecessary initialization

* remove over-killing initialization (in fast path)
* makeing the condition for PAGE_CGROUP_FLAG_ACTIVE be more obvious.

Signed-off-by: KAMEAZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: make global var read_mostly

mem_cgroup_subsys and page_cgroup_cache should be read_mostly and
MEM_CGROUP_RECLAIM_RETRIES can be just a fixed number.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

devcgroup: code cleanup

- clean up set_majmin()
- use simple_strtoul() to parse major/minor

[akpm@linux-foundation.org: fix simple_strtoul() usage]
[kosaki.motohiro@jp.fujitsu.com: fix warnings]
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: Serge Hallyn <serue@us.ibm.com>
Cc: Paul Menage <menage@google.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

devcgroup: relax white-list protection down to RCU

Currently this list is protected with a simple spinlock, even for reading
from one.  This is OK, but can be better.

Actually I want it to be better very much, since after replacing the
OpenVZ device permissions engine with the cgroup-based one I noticed, that
we set 12 default device permissions for each newly created container (for
/dev/null, full, terminals, ect devices), and people sometimes have up to
20 perms more, so traversing the ~30-40 elements list under a spinlock
doesn't seem very good.

Here's the RCU protection for white-list - dev_whitelist_item-s are added
and removed under the devcg->lock, but are looked up in permissions
checking under the rcu_read_lock.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroup_clone: use pid of newly created task for new cgroup

cgroup_clone creates a new cgroup with the pid of the task.  This works
correctly for unshare, but for clone cgroup_clone is called from
copy_namespaces inside copy_process, which happens before the new pid is
created.  As a result, the new cgroup was created with current's pid.
This patch:

1. Moves the call inside copy_process to after the new pid
   is created
2. Passes the struct pid into ns_cgroup_clone (as it is not
   yet attached to the task)
3. Passes a name from ns_cgroup_clone() into cgroup_clone()
   so as to keep cgroup_clone() itself simpler
4. Uses pid_vnr() to get the process id value, so that the
   pid used to name the new cgroup is always the pid as it
   would be known to the task which did the cloning or
   unsharing.  I think that is the most intuitive thing to
   do.  This way, task t1 does clone(CLONE_NEWPID) to get
   t2, which does clone(CLONE_NEWPID) to get t3, then the
   cgroup for t3 will be named for the pid by which t2 knows
   t3.

(Thanks to Dan Smith for finding the main bug)

Changelog:
June 11: Incorporate Paul Menage's feedback:  don't pass
         NULL to ns_cgroup_clone from unshare, and reduce
 patch size by using 'nodename' in cgroup_clone.
June 10: Original version

[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Serge Hallyn <serge@us.ibm.com>
Acked-by: Paul Menage <menage@google.com>
Tested-by: Dan Smith <danms@us.ibm.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroup files: convert res_counter_write() to be a cgroups write_string() handler

Currently res_counter_write() is a raw file handler even though it's
ultimately taking a number, since in some cases it wants to
pre-process the string when converting it to a number.

This patch converts res_counter_write() from a raw file handler to a
write_string() handler; this allows some of the boilerplate
copying/locking/checking to be removed, and simplies the cleanup path,
since these functions are now performed by the cgroups framework.

[lizf@cn.fujitsu.com: build fix]
Signed-off-by: Paul Menage <menage@google.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Serge Hallyn <serue@us.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroup files: convert devcgroup_access_write() into a cgroup write_string() handler

This patch converts devcgroup_access_write() from a raw file handler
into a handler for the cgroup write_string() method. This allows some
boilerplate copying/locking/checking to be removed and simplifies the
cleanup path, since these functions are performed by the cgroups
framework before calling the handler.

Signed-off-by: Paul Menage <menage@google.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroup files: remove cpuset_common_file_write()

This patch tweaks the signatures of the update_cpumask() and
update_nodemask() functions so that they can be called directly as
handlers for the new cgroups write_string() method.

This allows cpuset_common_file_write() to be removed.

Signed-off-by: Paul Menage <menage@google.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Serge Hallyn <serue@us.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroup files: turn attach_task_by_pid directly into a cgroup write handler

This patch changes attach_task_by_pid() to take a u64 rather than a
string; as a result it can be called directly as a control groups
write_u64 handler, and cgroup_common_file_write() can be removed.

Signed-off-by: Paul Menage <menage@google.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Serge Hallyn <serue@us.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroup files: move notify_on_release file to separate write handler

This patch moves the write handler for the cgroups notify_on_release
file into a separate handler. This handler requires no cgroups locking
since it relies on atomic bitops for synchronization.

Signed-off-by: Paul Menage <menage@google.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Serge Hallyn <serue@us.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroups: misc cleanups to write_string patchset

This patch contains cleanups suggested by reviewers for the recent
write_string() patchset:

- pair cgroup_lock_live_group() with cgroup_unlock() in cgroup.c for
  clarity, rather than directly unlocking cgroup_mutex.

- make the return type of cgroup_lock_live_group() a bool

- use a #define'd constant for the local buffer size in read/write functions

Signed-off-by: Paul Menage <menage@google.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroup files: move the release_agent file to use typed handlers

Adds cgroup_release_agent_write() and cgroup_release_agent_show()
methods to handle writing/reading the path to a cgroup hierarchy's
release agent. As a result, cgroup_common_file_read() is now unnecessary.

As part of the change, a previously-tolerated race in
cgroup_release_agent() is avoided by copying the current
release_agent_path prior to calling call_usermode_helper().

Signed-off-by: Paul Menage <menage@google.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroup files: add write_string cgroup control file method

This patch adds a write_string() method for cgroups control files. The
semantics are that a buffer is copied from userspace to kernelspace
and the handler function invoked on that buffer.  The buffer is
guaranteed to be nul-terminated, and no longer than max_write_len
(defaulting to 64 bytes if unspecified). Later patches will convert
existing raw file write handlers in control group subsystems to use
this method.

Signed-off-by: Paul Menage <menage@google.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Acked-by: Balbir Singh <balbir@in.ibm.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroup files: clean up whitespace in struct cftype

This patch removes some extraneous spaces from method declarations in
struct cftype, to fit in with conventional kernel style.

Signed-off-by: Paul Menage <menage@google.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Serge Hallyn <serue@us.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroups: annotate two variables with __read_mostly

- need_forkexit_callback will be read only after system boot.
- use_task_css_set_links will be read only after it's set.

And these 2 variables are checked when a new process is forked.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Paul Menage <menage@google.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroup: list_for_each cleanup

--------------------------
while() {
list_entry();
...
}
--------------------------

is equivalent to following code.

--------------------------
list_for_each_entry(){
...
}
--------------------------

later can review easily more.

this patch is just clean up.
it doesn't have any behavor change.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Mark res_counter_charge(_locked) with __must_check

Ignoring their return values may result in counter underflow in the future -
when the value charged will be uncharged (or in "leaks" - when the value is
not uncharged).

This also prevents from using charging routines to decrement the
counter value (i.e. uncharge it) ;)

(Current code works OK with res_counter, however :) )

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Paul Menage <menage@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroup: use read lock to guard find_existing_css_set()

The function does not modify anything (except the temporary css template), so
it's sufficient to hold read lock.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Paul Menage <menage@google.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

procfs-guide: drop pointless &nbsp; entities

Having trailing &nbsp; entities in a revision numer seems pretty pointless
to me.  More so, it's causing me pains, so just drop them since no other
guide is doing this.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

quota: implement sending information via netlink about user below quota

Sometimes it may be useful for userspace to know (e.g.  for some hosting
guys) that some user stopped exceeding his hardlimit or softlimit in
quotas.  Implement sending of such events to userspace via quota netlink
protocol so that they don't have to poll for such events.  Based on idea
and initial implementation by Vladislav Bogdanov.

Cc: Vladislav Bogdanov <slava@nsys.by>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

quota: convert macros to inline functions

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

quota: move function-macros from quota.h to quotaops.h

Move declarations of some macros, which should be in fact functions to
quotaops.h.  This way they can be later converted to inline functions
because we can now use declarations from quota.h.  Also add necessary
includes of quotaops.h to a few files.

[akpm@linux-foundation.org: fix JFS build]
[akpm@linux-foundation.org: fix UFS build]
[vegard.nossum@gmail.com: fix QUOTA=n build]
Signed-off-by: Jan Kara <jack@suse.cz>
Cc: Vegard Nossum <vegard.nossum@gmail.com>
Cc: Arjen Pool <arjenpool@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

quota: cleanup loop in sync_dquots()

Make loop in sync_dquots() checking whether there's something to write
more readable, remove useless variable and macro info_any_dirty() which
is used only in this place.

Signed-off-by: Jan Kara <jack@suse.cz>
Cc: "Vegard Nossum" <vegard.nossum@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

quota: rename quota functions from upper case, make bigger ones non-inline

Cleanup quotaops.h: Rename functions from uppercase to lowercase (and
define backward compatibility macros), move larger functions to dquot.c
and make them non-inline.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

quota: fix possible infinite loop in quota code

When quota structure is going to be dropped and it is dirty, quota code tries
to write it.  If the write fails for some reason (e.  g.  transaction cannot
be started because the journal is aborted), we try writing again and again and
again...  Fix the problem by clearing the dirty bit even if the write failed.

(akpm: for 2.6.27, 2.6.26.x and 2.6.25.x)

Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: dingdinghua <dingdinghua85@gmail.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

UTC timestamp option for FAT filesystems fix

Signed-off-by: Joe Peterson <joe@skyrush.com>
Acked-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fatfs: add UTC timestamp option

Provide a new mount option ("tz=UTC") for DOS (vfat/msdos) filesystems,
allowing timestamps to be in coordinated universal time (UTC) rather than
local time in applications where doing this is advantageous.

In particular, portable devices that use fat/vfat (such as digital
cameras) can benefit from using UTC in their internal clocks, thus
avoiding daylight saving time errors and general time ambiguity issues.
The user of the device does not have to worry about changing the time when
moving from place or when daylight saving changes.

The new mount option, when set, disables the counter-adjustment that Linux
currently makes to FAT timestamp info in anticipation of the normal
userspace time zone correction.  When used in this new mode, all daylight
saving time and time zone handling is done in userspace as is normal for
many other filesystems (like ext3).  The default mode, which remains
unchanged, is still appropriate when mounting volumes written in Windows
(because of its use of local time).

I originally based this patch on one submitted last year by Paul Collins,
but I updated it to work with current source and changed variable/option
naming.  Ogawa Hirofumi (who maintains these filesystems) and I discussed
this patch at length on lkml, and he suggested using the option name in
the attached version of the patch.  Barry Bouwsma pointed out a good
addition to the patch as well.

Signed-off-by: Joe Peterson <joe@skyrush.com>
Signed-off-by: Paul Collins <paul@ondioline.org>
Acked-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Cc: Barry Bouwsma <free_beer_for_all@yahoo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

remove unused #include <linux/dirent.h>'s

Remove some unused #include <linux/dirent.h>'s.

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

remove the in-kernel struct dirent{,64}

The kernel struct dirent{,64} were different from the ones in
userspace.

Even worse, we exported the kernel ones to userspace.

But after the fat usages are fixed we can remove the conflicting
kernel versions.

Reviewed-by: H. Peter Anvin <hpa@kernel.org>
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Cc: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

msdos fs: remove unsettable atari option

It has been impossible to set the option 'atari' of the MSDOS filesystem
for several years.  Since nobody seems to have missed it, let's remove its
remains.

Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Acked-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fat: small optimization to __fat_readdir()

This removes unnecessary parsing for directory entries.

If short_only, we don't need to parse longname.  And if !both and it found
the longname, we don't need shortname.

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fat: use same logic in fat_search_long() and __fat_readdir()

This uses uses stack for shortname, and uses __getname() for longname in
fat_search_long() and __fat_readdir().  By this, it removes unneeded
__getname() for shortname.

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fat: cleanup fs/fat/dir.c

This is no logic changes, just cleans fs/fat/dir.c up.

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fat/dir.c: switch to struct __fat_dirent

struct __fat_dirent is what was formerly the kernel struct dirent (that
was different from the userspace struct dirent).

Converting all fat users to struct __fat_dirent will allow us to get rid
of the conflicting struct dirent definition.

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fat: fix VFAT_IOCTL_READDIR_xxx and cleanup for userland

"struct dirent" is a kernel type here, but is a **different type** in
userspace!  This means both the structure and the IOCTL number is wrong!

So, this adds new "struct __fat_dirent" to generate correct IOCTL number.
And kernel stuff moves to under __KERNEL__.

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fat: fix parse_options()

Current parse_options() exits too early.  We need to run the code of
bottom in this function even if users doesn't specify options.

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

reiserfs: remove double definitions of xattr macros

remove the definitions of macros:
XATTR_SECURITY_PREFIX
XATTR_TRUSTED_PREFIX
XATTR_USER_PREFIX
since they are defined in linux/xattr.h

Signed-off-by: Shen Feng <shen@cn.fujitsu.com>
Signed-off-by: Mingming Cao <cmm@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

reiserfs: convert j_commit_lock to mutex

j_commit_lock is a semaphore but uses it as if it were a mutex.  This patch
converts it to a mutex.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Edward Shishkin <edward.shishkin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

reiserfs: convert j_flush_sem to mutex

j_flush_sem is a semaphore but uses it as if it were a mutex.  This patch
converts it to a mutex.

[akpm@linux-foundation.org: fix mutex_trylock retval treatment]
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Edward Shishkin <edward.shishkin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

reiserfs: convert j_lock to mutex

j_lock is a semaphore but uses it as if it were a mutex.  This patch converts
it to a mutex.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Edward Shishkin <edward.shishkin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

reiserfs: correct mount option parsing to detect when quota options can be changed

We should not allow user to change quota mount options when quota is just
suspended.  It would make mount options and internal quota state inconsistent.

Also we should not allow user to change quota format when quota is turned on.
On the other hand we can just silently ignore when some option is set to the
value it already has (some mount versions do this on remount).  Finally, we
should not discard current quota options if parsing of mount options fails.

Cc: <reiserfs-devel@vger.kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

reiserfs: fix typos in messages and comments (journalled -> journaled)

Cc: <reiserfs-devel@vger.kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

reiserfs: fix synchronization of quota files in journal=data mode

In journal=data mode, it is not enough to do write_inode_now() as done in
vfs_quota_on() to write all data to their final location (which is needed for
quota_read to work correctly).  Calling journal_end_sync() before calling
vfs_quota_on() does it's job because transactions are committed to the journal
and data marked as dirty in memory so write_inode_now() writes them to their
final locations.

Cc: <reiserfs-devel@vger.kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hfsplus: convert the extents_lock in a mutex

Apple Extended HFS file system: The semaphore extents lock is used as a
mutex.  Convert it to the mutex API.

Signed-off-by: Matthias Kaehlcke <matthias@kaehlcke.net>
Cc: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hfs: convert extents_lock in a mutex

Apple Macintosh file system: The semaphore extens_lock is used as a mutex.
Convert it to the mutex API

Signed-off-by: Matthias Kaehlcke <matthias@kaehlcke.net>
Cc: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hfs: convert bitmap_lock in a mutex

Apple Macintosh file system: The semaphore bitmap_lock is used as a mutex.
Convert it to the mutex API

Signed-off-by: Matthias Kaehlcke <matthias@kaehlcke.net>
Cc: Roman Zippel <zippel@linux-m68k.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

coda: remove CODA_FS_OLD_API

While fixing CONFIG_ leakages to the userspace kernel headers I ran into
CODA_FS_OLD_API.

After five years, are there still people using the old API left?
Especially considering that you have to choose at compile time which API
to support in the kernel (and distributions tend to offer the new API for
some time).

Jan: "The old API can definitely go.  Around the time the new
      interface went in there were some non-Coda userspace file system
      implementations that took a while longer to convert to the new API,
      but by now they all switched to the new interface or in some cases
      to a FUSE-based solution."

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Acked-by: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

isofs: fix minor filesystem corruption

Some iso9660 images contain files with rockridge data that is either
incorrect or incompletely parsed.  Prior to commit
f2966632a134e865db3c819346a1dc7d96e05309 ("[PATCH] rock: handle directory
overflows") (included with kernel 2.6.13) the kernel ignored the rockridge
data for these files, while still allowing the files to be accessed under
their non-rockridge names.  That commit inadvertently changed things so
that files with invalid rockridge data could not be accessed at all.  (I
ran across the problem when comparing some old CDs with hard disk copies I
had made long ago under kernel 2.4: a few of the files on the hard disk
copies were no longer visible on the CDs.)

This change reverts to the pre-2.6.13 behavior.

Signed-off-by: Adam Greenblatt <adam.greenblatt@gmail.com>
Reviewed-by: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: <stable@kernel.org> [2.6.25.x, 2.6.26.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ext3: validate directory entry data before use

ext3_dx_find_entry uses ext3_next_entry without verifying that the entry
is valid.  If its rec_len == 0 this causes an infinite loop.  Refactor the
loop to check the validity of entries before checking whether they match
and moving onto the next one.

There are other uses of ext3_next_entry in this file which also look
problematic.  They should be reviewed and fixed if/when we have a
test-case that triggers them.

This patch fixes the first case (image hdb.25.softlockup.gz) reported in
http://bugzilla.kernel.org/show_bug.cgi?id=10882.

Signed-off-by: Duane Griffin <duaneg@dghda.com>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

jbd: don't abort if flushing file data failed

In ordered mode, the current jbd aborts the journal if a file data buffer
has an error.  But this behavior is unintended, and we found that it has
been adopted accidentally.

This patch undoes it and just calls printk() instead of aborting the
journal.  Additionally, set AS_EIO into the address_space object of the
failed buffer which is submitted by journal_do_submit_data() so that
fsync() can get -EIO.

Missing error checkings are also added to inform errors on file data
buffers to the user.  The following buffers are targeted.

  (a) the buffer which has already been written out by pdflush
  (b) the buffer which has been unlocked before scanned in the
      t_locked_list loop

[akpm@linux-foundation.org: improve grammar in a printk]
Signed-off-by: Hidehiro Kawai <hidehiro.kawai.ez@hitachi.com>
Acked-by: Jan Kara <jack@suse.cz>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ext3: kill 2 useless magic numbers

dx_root_limit() will never return 20, and I can't figure out what 20
stands for.  This function has never changed since htree directory
indexing was merged.

Similar for dx_node_limit() and the magic 22.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Acked-by: Andreas Dilger <adilger@sun.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

jbd: positively dispose the unmapped data buffers in journal_commit_transaction()

After ext3-ordered files are truncated, there is a possibility that the
pages which cannot be estimated still remain.  Remaining pages can be
released when the system has really few memory.  So, it is not memory
leakage.  But the resource management software etc.  may not work
correctly.

It is possible that journal_unmap_buffer() cannot release the buffers, and
the pages to which they belong because they are attached to a commiting
transaction and journal_unmap_buffer() cannot release them.  To release
such the buffers and the pages later, journal_unmap_buffer() leaves it to
journal_commit_transaction().  (journal_unmap_buffer() puts the mark
'BH_Freed' to the buffers so that journal_commit_transaction() can
identify whether they can be released or not.)

In the journalled mode and the writeback mode, jbd does with only metadata
buffers.  But in the ordered mode, jbd does with metadata buffers and also
data buffers.

Actually, journal_commit_transaction() releases only the metadata buffers
of which release is demanded by journal_unmap_buffer(), and also releases
the pages to which they belong if possible.

As a result, the data buffers of which release is demanded by
journal_unmap_buffer() remain after a transaction commits.  And also the
pages to which they belong remain.

Such the remained pages don't have mapping any longer.  Due to this fact,
there is a possibility that the pages which cannot be estimated remain.

The metadata buffers marked 'BH_Freed' and the pages to which
they belong can be released at 'JBD: commit phase 7'.

Therefore, by applying the same code into 'JBD: commit phase 2' (where the
data buffers are done with), journal_commit_transaction() can also release
the data buffers marked 'BH_Freed' and the pages to which they belong.

As a result, all the buffers marked 'BH_Freed' can be released, and also
all the pages to which these buffers belong can be released at
journal_commit_transaction().  So, the page which cannot be estimated is
lost.

<<Excerpt of code at 'JBD: commit phase 7'>>
 >         spin_lock(&journal->j_list_lock);
 >         while (commit_transaction->t_forget) {
 >                 transaction_t *cp_transaction;
 >                 struct buffer_head *bh;
 >
 >                 jh = commit_transaction->t_forget;
 >...
 >                 if (buffer_freed(bh)) {
 >                 ^^^^^^^^^^^^^^^^^^^^^^^^
 >                         clear_buffer_freed(bh);
 >                        ^^^^^^^^^^^^^^^^^^^^^^^^
 >                         clear_buffer_jbddirty(bh);
 >                 }
 >
 >                 if (buffer_jbddirty(bh)) {
 >                         JBUFFER_TRACE(jh, "add to new checkpointing trans");
 >                         __journal_insert_checkpoint(jh, commit_transaction);
 >                         JBUFFER_TRACE(jh, "refile for checkpoint writeback");
 >                         __journal_refile_buffer(jh);
 >                         jbd_unlock_bh_state(bh);
 >                 } else {
 >                         J_ASSERT_BH(bh, !buffer_dirty(bh));
 > ...
 >                         JBUFFER_TRACE(jh, "refile or unfile freed buffer");
 >                         __journal_refile_buffer(jh);
 >                         if (!jh->b_transaction) {
 >                                 jbd_unlock_bh_state(bh);
 >                                  /* needs a brelse */
 >                                 journal_remove_journal_head(bh);
 >                                 release_buffer_page(bh);
 >                                 ^^^^^^^^^^^^^^^^^^^^^^^^
 >                         } else
 >                 }
****************************************************************
* Apply the code of "^^^^^^" lines into 'JBD: commit phase 2' *
****************************************************************

At journal_commit_transaction() code, there is one extra message in the
series of jbd debug messages.  ("JBD: commit phase 2") This patch fixes
it, too.

Signed-off-by: Toshiyuki Okajima <toshi.okajima@jp.fujitsu.com>
Acked-by: Jan Kara <jack@suse.cz>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

jbd: unexport journal_update_superblock

Remove the unused EXPORT_SYMBOL(journal_update_superblock).

Signed-off-by: Adrian Bunk <bunk@kernel.org>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ext3: handle deleting corrupted indirect blocks

While freeing indirect blocks we attach a journal head to the parent
buffer head, free the blocks, then journal the parent.  If the indirect
block list is corrupted and points to the parent the journal head will be
detached when the block is cleared, causing an OOPS.

Check for that explicitly and handle it gracefully.

This patch fixes the third case (image hdb.20000057.nullderef.gz)
reported in http://bugzilla.kernel.org/show_bug.cgi?id=10882.

Immediately above the change, in the ext3_free_data function, we call
ext3_clear_blocks to clear the indirect blocks in this parent block.  If
one of those blocks happens to actually be the parent block it will clear
b_private / BH_JBD.

I did the check at the end rather than earlier as it seemed more elegant.
I don't think there should be much practical difference, although it is
possible the FS may not be quite so badly corrupted if we did it the other
way (and didn't clear the block at all).  To be honest, I'm not convinced
there aren't other similar failure modes lurking in this code, although I
couldn't find any with a quick review.

[akpm@linux-foundation.org: fix printk warning]
Signed-off-by: Duane Griffin <duaneg@dghda.com>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ext3: don't read inode block if the buffer has a write error

A transient I/O error can corrupt inode data.  Here is the scenario:

(1) update inode_A at the block_B
(2) pdflush writes out new inode_A to the filesystem, but it results
    in write I/O error, at this point, BH_Uptodate flag of the buffer
    for block_B is cleared and BH_Write_EIO is set
(3) create new inode_C which located at block_B, and
    __ext3_get_inode_loc() tries to read on-disk block_B because the
    buffer is not uptodate
(4) if it can read on-disk block_B successfully, inode_A is
    overwritten by old data

This patch makes __ext3_get_inode_loc() not read the inode block if the
buffer has BH_Write_EIO flag.  In this case, the buffer should have the
latest information, so setting the uptodate flag to the buffer (this
avoids WARN_ON_ONCE() in mark_buffer_dirty().)

According to this change, we would need to test BH_Write_EIO flag for the
error checking.  Currently nobody checks write I/O errors on metadata
buffers, but it will be done in other patches I'm working on.

Signed-off-by: Hidehiro Kawai <hidehiro.kawai.ez@hitachi.com>
Cc: sugita <yumiko.sugita.yf@hitachi.com>
Cc: Satoshi OSHIMA <satoshi.oshima.fk@hitachi.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Jan Kara <jack@ucw.cz>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ext3: handle corrupted orphan list at mount

If the orphan node list includes valid, untruncatable nodes with nlink > 0
the ext3_orphan_cleanup loop which attempts to delete them will not do so,
causing it to loop forever. Fix by checking for such nodes in the
ext3_orphan_get function.

This patch fixes the second case (image hdb.20000009.softlockup.gz)
reported in http://bugzilla.kernel.org/show_bug.cgi?id=10882.

[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: printk warning fix]
Signed-off-by: Duane Griffin <duaneg@dghda.com>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>