w1: allow master IO commands

This small patchset extendes existing commands with reset, master IO and
status messages.  Reset is used to reset the bus for given master device,
master IO command allows to initiate IO against bus itself not selecting
slave device first, which can be used to probe the device for example.
And status messages carry command completion status back to the userspace
(namely very useful to get -ENODEV from when requested device was not
found).

Great thanks to Paul Alfille of OWFS for testing and commands suggestions.

This patch:

Allow starting of IO not against already found slave devices, but against
the bus itself, which can be used for example to probe devices.

[akpm@linux-foundation.org: reindent switch statements]
Signed-off-by: Evgeniy Polyakov <zbr@ioremap.net>
Cc: Paul Alfille <paul.alfille@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

w1: documentation update

Signed-off-by: Evgeniy Polyakov <zbr@ioremap.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

w1: list slaves commands

Initiates search (or alarm search) and returns all found devices to
userspace.  Found devices are not added into the system (i.e.  they are
not attached to family devices or bus masters), it will be done via (if
was not done yet) usual timed searching.

Signed-off-by: Evgeniy Polyakov <zbr@ioremap.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

w1: add touch block command

Writes and returns sampled data back to userspace.

Signed-off-by: Evgeniy Polyakov <zbr@ioremap.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

w1: add list masters w1 command

This patch series introduces and extends several userspace commands
used with netlink protocol.

Touch block command allows to write data and return sampled data to
the userspace.

Extended search and alarm seach commands to return list of slave
devices found during given search.

List masters command allows to send all registered master IDs to the
userspace.

Great thanks to Paul Alfille (owfs) who
tested this implementation and wrote w1-to-network daemon
http://sourceforge.net/projects/w1repeater/ and

Frederik Deweerdt and Randy Dunlap for review.

This patch:

Returns list of registered bus master devices.

Signed-off-by: Evgeniy Polyakov <zbr@ioremap.net>
Cc: Paul Alfille <paul.alfille@gmail.com>
Cc: Frederik Deweerdt <frederik.deweerdt@xprog.eu>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

w1: add 1-wire master driver for i.MX27 / i.MX31

This patch adds support for the 1-wire master interface for i.MX27 and
i.MX31.

Signed-off-by: Luotao Fu <l.fu@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: Evgeniy Polyakov <zbr@ioremap.net>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tpm: clean up tpm_nsc driver for platform_device suspend/resume compliance

Signed-off-by: Marcel Selhorst <tpm@selhorst.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

misc: add dell-laptop driver

Add a driver for controlling Dell-specific backlight and rfkill interfaces.
This driver makes use of the dcdbas interface to the Dell firmware to
allow the backlight and rfkill interfaces on Dell systems to be driven
through the standardised sysfs interfaces.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
Cc: Matt Domsch <Matt_Domsch@dell.com>
Cc: Ivo van Doorn <ivdoorn@gmail.com>
Cc: Len Brown <lenb@kernel.org>
Cc: Richard Purdie <rpurdie@rpsys.net>
Cc: Henrique de Moraes Holschuh <hmh@hmh.eng.br>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

dcdbas: export functionality for use in other drivers

The dcdbas code allows calls to be made into the firmware on Dell systems.
 Exporting this to other drivers allows them to implement Dell-specific
functionality in a safe way.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
Cc: Matt Domsch <Matt_Domsch@dell.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ELF: implement AT_RANDOM for glibc PRNG seeding

While discussing[1] the need for glibc to have access to random bytes
during program load, it seems that an earlier attempt to implement
AT_RANDOM got stalled.  This implements a random 16 byte string, available
to every ELF program via a new auxv AT_RANDOM vector.

[1] http://sourceware.org/ml/libc-alpha/2008-10/msg00006.html

Ulrich said:

glibc needs right after startup a bit of random data for internal
protections (stack canary etc).  What is now in upstream glibc is that we
always unconditionally open /dev/urandom, read some data, and use it.  For
every process startup.  That's slow.

...

The solution is to provide a limited amount of random data to the
starting process in the aux vector.  I suggested 16 bytes and this is
what the patch implements.  If we need only 16 bytes or less we use the
data directly.  If we need more we'll use the 16 bytes to see a PRNG.
This avoids the costly /dev/urandom use and it allows the kernel to use
the most adequate source of random data for this purpose.  It might not
be the same pool as that for /dev/urandom.

Concerns were expressed about the depletion of the randomness pool.  But
this patch doesn't make the situation worse, it doesn't deplete entropy
more than happens now.

Signed-off-by: Kees Cook <kees.cook@canonical.com>
Cc: Jakub Jelinek <jakub@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Ulrich Drepper <drepper@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mqueue: fix si_pid value in mqueue do_notify()

If a process registers for asynchronous notification on a POSIX message
queue, it gets a signal and a siginfo_t structure when a message arrives
on the message queue.  The si_pid in the siginfo_t structure is set to the
PID of the process that sent the message to the message queue.

The principle is the following:
. when mq_notify(SIGEV_SIGNAL) is called, the caller registers for
  notification when a msg arrives. The associated pid structure is stroed into
  inode_info->notify_owner. Let's call this process P1.
. when mq_send() is called by say P2, P2 sends a signal to P1 to notify
  him about msg arrival.

The way .si_pid is set today is not correct, since it doesn't take into account
the fact that the process that is sending the message might not be in the
same namespace as the notified one.

This patch proposes to set si_pid to the sender's pid into the notify_owner
namespace.

Signed-off-by: Nadia Derbey <Nadia.Derbey@bull.net>
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Bastian Blank <bastian@waldi.eu.org>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pid: generalize task_active_pid_ns

Currently task_active_pid_ns is not safe to call after a task becomes a
zombie and exit_task_namespaces is called, as nsproxy becomes NULL.  By
reading the pid namespace from the pid of the task we can trivially solve
this problem at the cost of one extra memory read in what should be the
same cacheline as we read the namespace from.

When moving things around I have made task_active_pid_ns out of line
because keeping it in pid_namespace.h would require adding includes of
pid.h and sched.h that I don't think we want.

This change does make task_active_pid_ns unsafe to call during
copy_process until we attach a pid on the task_struct which seems to be a
reasonable trade off.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Bastian Blank <bastian@waldi.eu.org>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Nadia Derbey <Nadia.Derbey@bull.net>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pid: implement ns_of_pid

A current problem with the pid namespace is that it is easy to do pid
related work after exit_task_namespaces which drops the nsproxy pointer.

However if we are doing pid namespace related work we are always operating
on some struct pid which retains the pid_namespace pointer of the pid
namespace it was allocated in.

So provide ns_of_pid which allows us to find the pid namespace a pid was
allocated in.

Using this we have the needed infrastructure to do pid namespace related
work at anytime we have a struct pid, removing the chance of accidentally
having a NULL pointer dereference when accessing current->nsproxy.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Bastian Blank <bastian@waldi.eu.org>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Nadia Derbey <Nadia.Derbey@bull.net>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpuset: remove remaining pointers to cpumask_t

Impact: cleanups, use new cpumask API

Final trivial cleanups: mainly s/cpumask_t/struct cpumask

Note there is a FIXME in generate_sched_domains(). A future patch will
change struct cpumask *doms to struct cpumask *doms[].
(I suppose Rusty will do this.)

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Mike Travis <travis@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>

cpuset: convert cpuset->cpus_allowed to cpumask_var_t

Impact: use new cpumask API

This patch mainly does the following things:
- change cs->cpus_allowed from cpumask_t to cpumask_var_t
- call alloc_bootmem_cpumask_var() for top_cpuset in cpuset_init_early()
- call alloc_cpumask_var() for other cpusets
- replace cpus_xxx() to cpumask_xxx()

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Mike Travis <travis@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>

cpuset: don't allocate trial cpuset on stack

Impact: cleanups, reduce stack usage

This patch prepares for the next patch.  When we convert
cpuset.cpus_allowed to cpumask_var_t, (trialcs = *cs) no longer works.

Another result of this patch is reducing stack usage of trialcs.
sizeof(*cs) can be as large as 148 bytes on x86_64, so it's really not
good to have it on stack.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Mike Travis <travis@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>

cpuset: convert cpuset_attach() to use cpumask_var_t

Impact: reduce stack usage

Allocate a global cpumask_var_t at boot, and use it in cpuset_attach(), so
we won't fail cpuset_attach().

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Mike Travis <travis@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>

cpuset: remove on stack cpumask_t in cpuset_can_attach()

Impact: reduce stack usage

Just use cs->cpus_allowed, and no need to allocate a cpumask_var_t.

Signed-off-by: Li Zefan <lizf@cn.fujistu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Mike Travis <travis@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>

cpuset: remove on stack cpumask_t in cpuset_sprintf_cpulist()

This patchset converts cpuset to use new cpumask API, and thus
remove on stack cpumask_t to reduce stack usage.

Before:
 # cat kernel/cpuset.c include/linux/cpuset.h | grep -c cpumask_t
 21
After:
 # cat kernel/cpuset.c include/linux/cpuset.h | grep -c cpumask_t
 0

This patch:

Impact: reduce stack usage

It's safe to call cpulist_scnprintf inside callback_mutex, and thus we can
just remove the cpumask_t and no need to allocate a cpumask_var_t.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Mike Travis <travis@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: set task's cpu_allowed to cpu_possible_map when attaching it into top cpuset

I found a bug on my dual-cpu box.  I created a sub cpuset in top 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 top cpuset automatically because there is no cpu in sub
cpuset.  Then we online CPU1, we find all the tasks which doesn't belong
to top cpuset originally just run on CPU0.

We fix this bug by setting task's cpu_allowed to cpu_possible_map when
attaching it into top cpuset.  This method needn't modify the current
behavior of cpusets on CPU hotplug, and all of tasks in top cpuset use
cpu_possible_map to initialize their cpu_allowed.

Signed-off-by: Miao Xie <miaox@cn.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>

cpuset: rcu_read_lock() to protect task_cs()

task_cs() calls task_subsys_state().

We must use rcu_read_lock() to protect cgroup_subsys_state().

It's correct that top_cpuset is never freed, but cgroup_subsys_state()
accesses css_set, this css_set maybe freed when task_cs() called.

We use use rcu_read_lock() to protect it.

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

cgroups: add css_tryget()

Add css_tryget(), that obtains a counted reference on a CSS.  It is used
in situations where the caller has a "weak" reference to the CSS, i.e.
one that does not protect the cgroup from removal via a reference count,
but would instead be cleaned up by a destroy() callback.

css_tryget() will return true on success, or false if the cgroup is being
removed.

This is similar to Kamezawa Hiroyuki's patch from a week or two ago, but
with the difference that in the event of css_tryget() racing with a
cgroup_rmdir(), css_tryget() will only return false if the cgroup really
does get removed.

This implementation is done by biasing css->refcnt, so that a refcnt of 1
means "releasable" and 0 means "released or releasing".  In the event of a
race, css_tryget() distinguishes between "released" and "releasing" by
checking for the CSS_REMOVED flag in css->flags.

Signed-off-by: Paul Menage <menage@google.com>
Tested-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroups: use hierarchy_mutex in memory controller

Update the memory controller to use its hierarchy_mutex rather than
calling cgroup_lock() to protected against cgroup_mkdir()/cgroup_rmdir()
from occurring in its hierarchy.

Signed-off-by: Paul Menage <menage@google.com>
Tested-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroups: add a per-subsystem hierarchy_mutex

These patches introduce new locking/refcount support for cgroups to
reduce the need for subsystems to call cgroup_lock(). This will
ultimately allow the atomicity of cgroup_rmdir() (which was removed
recently) to be restored.

These three patches give:

1/3 - introduce a per-subsystem hierarchy_mutex which a subsystem can
     use to prevent changes to its own cgroup tree

2/3 - use hierarchy_mutex in place of calling cgroup_lock() in the
     memory controller

3/3 - introduce a css_tryget() function similar to the one recently
      proposed by Kamezawa, but avoiding spurious refcount failures in
      the event of a race between a css_tryget() and an unsuccessful
      cgroup_rmdir()

Future patches will likely involve:

- using hierarchy mutex in place of cgroup_lock() in more subsystems
 where appropriate

- restoring the atomicity of cgroup_rmdir() with respect to cgroup_create()

This patch:

Add a hierarchy_mutex to the cgroup_subsys object that protects changes to
the hierarchy observed by that subsystem.  It is taken by the cgroup
subsystem (in addition to cgroup_mutex) for the following operations:

- linking a cgroup into that subsystem's cgroup tree
- unlinking a cgroup from that subsystem's cgroup tree
- moving the subsystem to/from a hierarchy (including across the
  bind() callback)

Thus if the subsystem holds its own hierarchy_mutex, it can safely
traverse its own hierarchy.

Signed-off-by: Paul Menage <menage@google.com>
Tested-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: fix shmem's swap accounting

Now, you can see following even when swap accounting is enabled.

 1. Create Group 01, and 02.
 2. allocate a "file" on tmpfs by a task under 01.
 3. swap out the "file" (by memory pressure)
 4. Read "file" from a task in group 02.
 5. the charge of "file" is moved to group 02.

This is not ideal behavior. This is because SwapCache which was loaded
by read-ahead is not taken into account..

This is a patch to fix shmem's swapcache behavior.
  - remove mem_cgroup_cache_charge_swapin().
  - Add SwapCache handler routine to mem_cgroup_cache_charge().
    By this, shmem's file cache is charged at add_to_page_cache()
    with GFP_NOWAIT.
  - pass the page of swapcache to shrink_mem_cgroup.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: fix LRU accounting for SwapCache

Now, a page can be deleted from SwapCache while do_swap_page().
memcg-fix-swap-accounting-leak-v3.patch handles that, but, LRU handling is
still broken.  (above behavior broke assumption of memcg-synchronized-lru
patch.)

This patch is a fix for LRU handling (especially for per-zone counters).
At charging SwapCache,
 - Remove page_cgroup from LRU if it's not used.
 - Add page cgroup to LRU if it's not linked to.

Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.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>

memcg: use css_tryget in memcg

From:KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

css_tryget() newly is added and we can know css is alive or not and get
refcnt of css in very safe way.  ("alive" here means "rmdir/destroy" is
not called.)

This patch replaces css_get() to css_tryget(), where I cannot explain
why css_get() is safe. And removes memcg->obsolete flag.

Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
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: fix double free and make refcnt sane

1. Fix double-free BUG in error route of mem_cgroup_create().
    mem_cgroup_free() itself frees per-zone-info.
 2. Making refcnt of memcg simple.
    Add 1 refcnt at creation and call free when refcnt goes down to 0.

Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Paul Menage <menage@google.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
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: fix swap accounting leak

Fix swapin charge operation of memcg.

Now, memcg has hooks to swap-out operation and checks SwapCache is really
unused or not.  That check depends on contents of struct page.  I.e.  If
PageAnon(page) && page_mapped(page), the page is recoginized as
still-in-use.

Now, reuse_swap_page() calles delete_from_swap_cache() before establishment
of any rmap. Then, in followinig sequence

(Page fault with WRITE)
try_charge() (charge += PAGESIZE)
commit_charge() (Check page_cgroup is used or not..)
reuse_swap_page()
-> delete_from_swapcache()
-> mem_cgroup_uncharge_swapcache() (charge -= PAGESIZE)
......
New charge is uncharged soon....
To avoid this,  move commit_charge() after page_mapcount() goes up to 1.
By this,

try_charge() (usage += PAGESIZE)
reuse_swap_page() (may usage -= PAGESIZE if PCG_USED is set)
commit_charge() (If page_cgroup is not marked as PCG_USED,
 add new charge.)
Accounting will be correct.

Changelog (v2) -> (v3)
  - fixed invalid charge to swp_entry==0.
  - updated documentation.
Changelog (v1) -> (v2)
  - fixed comment.

[nishimura@mxp.nes.nec.co.jp: swap accounting leak doc fix]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Tested-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: change try_to_free_pages to hierarchical_reclaim

mem_cgroup_hierarchicl_reclaim() works properly even when !use_hierarchy
now (by memcg-hierarchy-avoid-unnecessary-reclaim.patch), so, instead of
try_to_free_mem_cgroup_pages(), it should be used in many cases.

The only exception is force_empty.  The group has no children in this
case.

Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: 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>

memcg: avoid deadlock caused by race between oom and cpuset_attach

mpol_rebind_mm(), which can be called from cpuset_attach(), does
down_write(mm->mmap_sem).  This means down_write(mm->mmap_sem) can be
called under cgroup_mutex.

OTOH, page fault path does down_read(mm->mmap_sem) and calls
mem_cgroup_try_charge_xxx(), which may eventually calls
mem_cgroup_out_of_memory().  And mem_cgroup_out_of_memory() calls
cgroup_lock().  This means cgroup_lock() can be called under
down_read(mm->mmap_sem).

If those two paths race, deadlock can happen.

This patch avoid this deadlock by:
  - remove cgroup_lock() from mem_cgroup_out_of_memory().
  - define new mutex (memcg_tasklist) and serialize mem_cgroup_move_task()
    (->attach handler of memory cgroup) and mem_cgroup_out_of_memory.

Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: 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>

memcg: remove mem_cgroup_try_charge

After previous patch, mem_cgroup_try_charge is not used by anyone, so we
can remove it.

Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: don't trigger oom at page migration

I think triggering OOM at mem_cgroup_prepare_migration would be just a bit
overkill.  Returning -ENOMEM would be enough for
mem_cgroup_prepare_migration.  The caller would handle the case anyway.

Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: 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>

memcg: explain details and test document

Documentation for implementation details and how to test.

Just an example. feel free to modify, add, remove lines.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: show real limit under hierarchy mode

Show "real" limit of memcg.  This helps my debugging and maybe useful for
users.

While testing hierarchy like this

mount -t cgroup none /cgroup -t memory
mkdir /cgroup/A
set use_hierarchy==1 to "A"
mkdir /cgroup/A/01
mkdir /cgroup/A/01/02
mkdir /cgroup/A/01/03
mkdir /cgroup/A/01/03/04
mkdir /cgroup/A/08
mkdir /cgroup/A/08/01
....
and set each own limit to them, "real" limit of each memcg is unclear.
This patch shows real limit by checking all ancestors.

Changelog: (v1) -> (v2)
- remove "if" and use "min(a,b)"

Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: 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>

memcg: fix calculation of active_ratio

Currently, inactive_ratio of memcg is calculated at setting limit.
because page_alloc.c does so and current implementation is straightforward
porting.

However, memcg introduced hierarchy feature recently.  In hierarchy
restriction, memory limit is not only decided memory.limit_in_bytes of
current cgroup, but also parent limit and sibling memory usage.

Then, The optimal inactive_ratio is changed frequently.  So, everytime
calculation is better.

Tested-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.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>

memcg: swappiness

Currently, /proc/sys/vm/swappiness can change swappiness ratio for global
reclaim.  However, memcg reclaim doesn't have tuning parameter for itself.

In general, the optimal swappiness depend on workload.  (e.g.  hpc
workload need to low swappiness than the others.)

Then, per cgroup swappiness improve administrator tunability.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: protect prev_priority

Currently, mem_cgroup doesn't have own lock and almost its member doesn't
need.  (e.g.  mem_cgroup->info is protected by zone lock, mem_cgroup->stat
is per cpu variable)

However, there is one explict exception.  mem_cgroup->prev_priorit need
lock, but doesn't protect.  Luckly, this is NOT bug because prev_priority
isn't used for current reclaim code.

However, we plan to use prev_priority future again.  Therefore, fixing is
better.

In addition, we plan to reuse this lock for another member.  Then
"reclaim_param_lock" name is better than "prev_priority_lock".

Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: rename scan global lru

Rename scan_global_lru() to scanning_global_lru().

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: show reclaim stat

Add the following four fields to memory.stat file:

  - inactive_ratio
  - recent_rotated_anon
  - recent_rotated_file
  - recent_scanned_anon
  - recent_scanned_file

Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: remove mem_cgroup_cal_reclaim()

Now, get_scan_ratio() return correct value although memcg reclaim.  Then,
mem_cgroup_calc_reclaim() can be removed.

So, memcg reclaim get the same capability of anon/file reclaim balancing
as global reclaim now.

Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@redhat.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: add zone_reclaim_stat

Introduce mem_cgroup_per_zone::reclaim_stat member and its statics
collecting function.

Now, get_scan_ratio() can calculate correct value on memcg reclaim.

[hugh@veritas.com: avoid reclaim_stat oops when disabled]
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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: add mem_cgroup_zone_nr_pages()

Introduce mem_cgroup_zone_nr_pages().  It is called by zone_nr_pages()
helper function.

This patch doesn't have any behavior change.

Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: add inactive_anon_is_low()

The inactive_anon_is_low() is key component of active/inactive anon
balancing on reclaim.  However current inactive_anon_is_low() function
only consider global reclaim.

Therefore, we need following ugly scan_global_lru() condition.

if (lru == LRU_ACTIVE_ANON &&
    (!scan_global_lru(sc) || inactive_anon_is_low(zone))) {
shrink_active_list(nr_to_scan, zone, sc, priority, file);
return 0;

it cause that memcg reclaim always deactivate pages when shrink_list() is
called.  To make mem_cgroup_inactive_anon_is_low() improve active/inactive
anon balancing of memcgroup.

Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Cc: "Pekka Enberg" <penberg@cs.helsinki.fi>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: add null check to page_cgroup_zoneinfo()

If CONFIG_CGROUP_MEM_RES_CTLR_SWAP=y, page_cgroup::mem_cgroup can be NULL.
Therefore null checking is better.

A later patch uses this function.

Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

mm: make get_scan_ratio() safe for memcg

Currently, get_scan_ratio() always calculate the balancing value for
global reclaim and memcg reclaim doesn't use it.  Therefore it doesn't
have scan_global_lru() condition.

However, we plan to expand get_scan_ratio() to be usable for memcg too,
latter.  Then, The dependency code of global reclaim in the
get_scan_ratio() insert into scan_global_lru() condision explictly.

This patch doesn't have any functional change.

Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

mm: add zone nr_pages helper function

Add zone_nr_pages() helper function.

It is used by a later patch.  This patch doesn't have any functional
change.

Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

mm: introduce zone_reclaim struct

Add zone_reclam_stat struct for later enhancement.

A later patch uses this.  This patch doesn't any behavior change (yet).

Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

inactive_anon_is_low: move to vmscan

The inactive_anon_is_low() is called only vmscan.  Then it can move to
vmscan.c

This patch doesn't have any functional change.

Reviewd-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: hierarchy avoid unnecessary reclaim

If hierarchy is not used, no tree-walk is necessary.

Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: swapout refcnt fix

css's refcnt is dropped before end of following access.
Hold it until end of access.

Reported-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: memory swap controller: fix limit check

There are scatterd calls of res_counter_check_under_limit(), and most of
them don't take mem+swap accounting into account.

define mem_cgroup_check_under_limit() and avoid direct use of
res_counter_check_limit().

Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: check group leader fix

Remove unnecessary codes (...fragments of not-implemented
functionalilty...)

Reported-by: Nikanth Karthikesan <knikanth@suse.de>
Signed-off-by: Nikanth Karthikesan <knikanth@suse.de>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: revert gfp mask fix

My patch, memcg-fix-gfp_mask-of-callers-of-charge.patch changed gfp_mask
of callers of charge to be GFP_HIGHUSER_MOVABLE for showing what will
happen at memory reclaim.

But in recent discussion, it's NACKed because it sounds ugly.

This patch is for reverting it and add some clean up to gfp_mask of
callers of charge.  No behavior change but need review before generating
HUNK in deep queue.

This patch also adds explanation to meaning of gfp_mask passed to charge
functions in memcontrol.h.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: 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>

memcg: fix reclaim result checks

check_under_limit logic was wrong and this check should be against
mem_over_limit rather than mem.

Reported-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Jan Blunck <jblunck@suse.de>
Cc: Hirokazu Takahashi <taka@valinux.co.jp>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: avoid unnecessary system-wide-oom-killer

Current mmtom has new oom function as pagefault_out_of_memory().  It's
added for select bad process rathar than killing current.

When memcg hit limit and calls OOM at page_fault, this handler called and
system-wide-oom handling happens.  (means kernel panics if panic_on_oom is
true....)

To avoid overkill, check memcg's recent behavior before starting
system-wide-oom.

And this patch also fixes to guarantee "don't accnout against process with
TIF_MEMDIE".  This is necessary for smooth OOM.

[akpm@linux-foundation.org: build fix]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Jan Blunck <jblunck@suse.de>
Cc: Hirokazu Takahashi <taka@valinux.co.jp>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcontrol: rcu_read_lock() to protect mm_match_cgroup()

mm_match_cgroup() calls cgroup_subsys_state().

We must use rcu_read_lock() to protect cgroup_subsys_state().

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

memcg: memory cgroup hierarchy feature selector

Don't enable multiple hierarchy support by default.  This patch introduces
a features element that can be set to enable the nested depth hierarchy
feature.  This feature can only be enabled when the cgroup for which the
feature this is enabled, has no children.

Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelianov <xemul@openvz.org>
Cc: Dhaval Giani <dhaval@linux.vnet.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>

memcg: memory cgroup hierarchical reclaim

This patch introduces hierarchical reclaim.  When an ancestor goes over
its limit, the charging routine points to the parent that is above its
limit.  The reclaim process then starts from the last scanned child of the
ancestor and reclaims until the ancestor goes below its limit.

[akpm@linux-foundation.org: coding-style fixes]
[d-nishimura@mtf.biglobe.ne.jp: mem_cgroup_from_res_counter should handle both mem->res and mem->memsw]
Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelianov <xemul@openvz.org>
Cc: Dhaval Giani <dhaval@linux.vnet.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: memory cgroup resource counters for hierarchy

Add support for building hierarchies in resource counters.  Cgroups allows
us to build a deep hierarchy, but we currently don't link the resource
counters belonging to the memory controller control groups, in the same
fashion as the corresponding cgroup entries in the cgroup hierarchy.  This
patch provides the infrastructure for resource counters that have the same
hiearchy as their cgroup counter parts.

These set of patches are based on the resource counter hiearchy patches
posted by Pavel Emelianov.

NOTE: Building hiearchies is expensive, deeper hierarchies imply charging
the all the way up to the root.  It is known that hiearchies are
expensive, so the user needs to be careful and aware of the trade-offs
before creating very deep ones.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelianov <xemul@openvz.org>
Cc: Dhaval Giani <dhaval@linux.vnet.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>

memcg: memory cgroup hierarchy documentation

Documentation updates for hierarchy support

Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: YAMAMOTO Takashi <yamamoto@valinux.co.jp>
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelianov <xemul@openvz.org>
Cc: Dhaval Giani <dhaval@linux.vnet.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>

memcg: add mem_cgroup_disabled()

We check mem_cgroup is disabled or not by checking
mem_cgroup_subsys.disabled.  I think it has more references than expected,
now.

replacing
   if (mem_cgroup_subsys.disabled)
with
   if (mem_cgroup_disabled())

give us good look, I think.

[kamezawa.hiroyu@jp.fujitsu.com: fix typo]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: synchronized LRU

A big patch for changing memcg's LRU semantics.

Now,
  - page_cgroup is linked to mem_cgroup's its own LRU (per zone).

  - LRU of page_cgroup is not synchronous with global LRU.

  - page and page_cgroup is one-to-one and statically allocated.

  - To find page_cgroup is on what LRU, you have to check pc->mem_cgroup as
    - lru = page_cgroup_zoneinfo(pc, nid_of_pc, zid_of_pc);

  - SwapCache is handled.

And, when we handle LRU list of page_cgroup, we do following.

pc = lookup_page_cgroup(page);
lock_page_cgroup(pc); .....................(1)
mz = page_cgroup_zoneinfo(pc);
spin_lock(&mz->lru_lock);
.....add to LRU
spin_unlock(&mz->lru_lock);
unlock_page_cgroup(pc);

But (1) is spin_lock and we have to be afraid of dead-lock with zone->lru_lock.
So, trylock() is used at (1), now. Without (1), we can't trust "mz" is correct.

This is a trial to remove this dirty nesting of locks.
This patch changes mz->lru_lock to be zone->lru_lock.
Then, above sequence will be written as

        spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU
mem_cgroup_add/remove/etc_lru() {
pc = lookup_page_cgroup(page);
mz = page_cgroup_zoneinfo(pc);
if (PageCgroupUsed(pc)) {
....add to LRU
}
        spin_lock(&zone->lru_lock); # in vmscan.c or swap.c via global LRU

This is much simpler.
(*) We're safe even if we don't take lock_page_cgroup(pc). Because..
    1. When pc->mem_cgroup can be modified.
       - at charge.
       - at account_move().
    2. at charge
       the PCG_USED bit is not set before pc->mem_cgroup is fixed.
    3. at account_move()
       the page is isolated and not on LRU.

Pros.
  - easy for maintenance.
  - memcg can make use of laziness of pagevec.
  - we don't have to duplicated LRU/Active/Unevictable bit in page_cgroup.
  - LRU status of memcg will be synchronized with global LRU's one.
  - # of locks are reduced.
  - account_move() is simplified very much.
Cons.
  - may increase cost of LRU rotation.
    (no impact if memcg is not configured.)

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

memcg: mem+swap controller core

This patch implements per cgroup limit for usage of memory+swap.  However
there are SwapCache, double counting of swap-cache and swap-entry is
avoided.

Mem+Swap controller works as following.
  - memory usage is limited by memory.limit_in_bytes.
  - memory + swap usage is limited by memory.memsw_limit_in_bytes.

This has following benefits.
  - A user can limit total resource usage of mem+swap.

    Without this, because memory resource controller doesn't take care of
    usage of swap, a process can exhaust all the swap (by memory leak.)
    We can avoid this case.

    And Swap is shared resource but it cannot be reclaimed (goes back to memory)
    until it's used. This characteristic can be trouble when the memory
    is divided into some parts by cpuset or memcg.
    Assume group A and group B.
    After some application executes, the system can be..

    Group A -- very large free memory space but occupy 99% of swap.
    Group B -- under memory shortage but cannot use swap...it's nearly full.

    Ability to set appropriate swap limit for each group is required.

Maybe someone wonder "why not swap but mem+swap ?"

  - The global LRU(kswapd) can swap out arbitrary pages. Swap-out means
    to move account from memory to swap...there is no change in usage of
    mem+swap.

    In other words, when we want to limit the usage of swap without affecting
    global LRU, mem+swap limit is better than just limiting swap.

Accounting target information is stored in swap_cgroup which is
per swap entry record.

Charge is done as following.
  map
    - charge  page and memsw.

  unmap
    - uncharge page/memsw if not SwapCache.

  swap-out (__delete_from_swap_cache)
    - uncharge page
    - record mem_cgroup information to swap_cgroup.

  swap-in (do_swap_page)
    - charged as page and memsw.
      record in swap_cgroup is cleared.
      memsw accounting is decremented.

  swap-free (swap_free())
    - if swap entry is freed, memsw is uncharged by PAGE_SIZE.

There are people work under never-swap environments and consider swap as
something bad. For such people, this mem+swap controller extension is just an
overhead.  This overhead is avoided by config or boot option.
(see Kconfig. detail is not in this patch.)

TODO:
 - maybe more optimization can be don in swap-in path. (but not very safe.)
   But we just do simple accounting at this stage.

[nishimura@mxp.nes.nec.co.jp: make resize limit hold mutex]
[hugh@veritas.com: memswap controller core swapcache fixes]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
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: swap cgroup for remembering usage

For accounting swap, we need a record per swap entry, at least.

This patch adds following function.
  - swap_cgroup_swapon() .... called from swapon
  - swap_cgroup_swapoff() ... called at the end of swapoff.

  - swap_cgroup_record() .... record information of swap entry.
  - swap_cgroup_lookup() .... lookup information of swap entry.

This patch just implements "how to record information".  No actual method
for limit the usage of swap.  These routine uses flat table to record and
lookup.  "wise" lookup system like radix-tree requires requires memory
allocation at new records but swap-out is usually called under memory
shortage (or memcg hits limit.) So, I used static allocation.  (maybe
dynamic allocation is not very hard but it adds additional memory
allocation in memory shortage path.)

Note1: In this, we use pointer to record information and this means
      8bytes per swap entry. I think we can reduce this when we
      create "id of cgroup" in the range of 0-65535 or 0-255.

Reported-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reported-by: Hugh Dickins <hugh@veritas.com>
Reported-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Reported-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Pavel Emelianov <xemul@openvz.org>
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: mem+swap controller Kconfig

Config and control variable for mem+swap controller.

This patch adds CONFIG_CGROUP_MEM_RES_CTLR_SWAP
(memory resource controller swap extension.)

For accounting swap, it's obvious that we have to use additional memory to
remember "who uses swap".  This adds more overhead.  So, it's better to
offer "choice" to users.  This patch adds 2 choices.

This patch adds 2 parameters to enable swap extension or not.
  - CONFIG
  - boot option

Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: handle swap caches

SwapCache support for memory resource controller (memcg)

Before mem+swap controller, memcg itself should handle SwapCache in proper
way.  This is cut-out from it.

In current memcg, SwapCache is just leaked and the user can create tons of
SwapCache.  This is a leak of account and should be handled.

SwapCache accounting is done as following.

  charge (anon)
- charged when it's mapped.
  (because of readahead, charge at add_to_swap_cache() is not sane)
  uncharge (anon)
- uncharged when it's dropped from swapcache and fully unmapped.
  means it's not uncharged at unmap.
  Note: delete from swap cache at swap-in is done after rmap information
        is established.
  charge (shmem)
- charged at swap-in. this prevents charge at add_to_page_cache().

  uncharge (shmem)
- uncharged when it's dropped from swapcache and not on shmem's
  radix-tree.

  at migration, check against 'old page' is modified to handle shmem.

Comparing to the old version discussed (and caused troubles), we have
advantages of
  - PCG_USED bit.
  - simple migrating handling.

So, situation is much easier than several months ago, maybe.

[hugh@veritas.com: memcg: handle swap caches build fix]
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
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: new force_empty to free pages under group

By memcg-move-all-accounts-to-parent-at-rmdir.patch, there is no leak of
memory usage and force_empty is removed.

This patch adds "force_empty" again, in reasonable manner.

memory.force_empty file works when

  #echo 0 (or some) > memory.force_empty
  and have following function.

  1. only works when there are no task in this cgroup.
  2. free all page under this cgroup as much as possible.
  3. page which cannot be freed will be moved up to parent.
  4. Then, memcg will be empty after above echo returns.

This is much better behavior than old "force_empty" which just forget
all accounts. This patch also check signal_pending() and above "echo"
can be stopped by "Ctrl-C".

[akpm@linux-foundation.org: cleanup]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: reduce size of mem_cgroup by using nr_cpu_ids

As Jan Blunck <jblunck@suse.de> pointed out, allocating per-cpu stat for
memcg to the size of NR_CPUS is not good.

This patch changes mem_cgroup's cpustat allocation not based on NR_CPUS
but based on nr_cpu_ids.

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

memcg: move all acccounting to parent at rmdir()

This patch provides a function to move account information of a page
between mem_cgroups and rewrite force_empty to make use of this.

This moving of page_cgroup is done under
 - lru_lock of source/destination mem_cgroup is held.
 - lock_page_cgroup() is held.

Then, a routine which touches pc->mem_cgroup without lock_page_cgroup()
should confirm pc->mem_cgroup is still valid or not.  Typical code can be
following.

(while page is not under lock_page())
mem = pc->mem_cgroup;
mz = page_cgroup_zoneinfo(pc)
spin_lock_irqsave(&mz->lru_lock);
if (pc->mem_cgroup == mem)
...../* some list handling */
spin_unlock_irqrestore(&mz->lru_lock);

Of course, better way is
lock_page_cgroup(pc);
....
unlock_page_cgroup(pc);

But you should confirm the nest of lock and avoid deadlock.

If you treats page_cgroup from mem_cgroup's LRU under mz->lru_lock,
you don't have to worry about what pc->mem_cgroup points to.
moved pages are added to head of lru, not to tail.

Expected users of this routine is:
  - force_empty (rmdir)
  - moving tasks between cgroup (for moving account information.)
  - hierarchy (maybe useful.)

force_empty(rmdir) uses this move_account and move pages to its parent.
This "move" will not cause OOM (I added "oom" parameter to try_charge().)

If the parent is busy (not enough memory), force_empty calls try_to_free_page()
and reduce usage.

Purpose of this behavior is
  - Fix "forget all" behavior of force_empty and avoid leak of accounting.
  - By "moving first, free if necessary", keep pages on memory as much as
    possible.

Adding a switch to change behavior of force_empty to
  - free first, move if necessary
  - free all, if there is mlocked/busy pages, return -EBUSY.
is under consideration. (I'll add if someone requtests.)

This patch also removes memory.force_empty file, a brutal debug-only interface.

Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Tested-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.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: do not recalculate section unnecessarily in init_section_page_cgroup

In init_section_page_cgroup() the section a given pfn belongs to is
calculated at the top of the function and, despite the fact that the
pfn/section correspondence does not change, it is recalculated further
down the same function.  By computing this just once and reusing that
value we save some bytes in the object file and do not waste CPU cycles.

Signed-off-by: Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: simple migration handling

Now, management of "charge" under page migration is done under following
manner. (Assume migrate page contents from oldpage to newpage)

 before
  - "newpage" is charged before migration.
 at success.
  - "oldpage" is uncharged at somewhere(unmap, radix-tree-replace)
 at failure
  - "newpage" is uncharged.
  - "oldpage" is charged if necessary (*1)

But (*1) is not reliable....because of GFP_ATOMIC.

This patch tries to change behavior as following by charge/commit/cancel ops.

 before
  - charge PAGE_SIZE (no target page)
 success
  - commit charge against "newpage".
 failure
  - commit charge against "oldpage".
    (PCG_USED bit works effectively to avoid double-counting)
  - if "oldpage" is obsolete, cancel charge of PAGE_SIZE.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: fix gfp_mask of callers of charge

Fix misuse of gfp_kernel.

Now, most of callers of mem_cgroup_charge_xxx functions uses GFP_KERNEL.

I think that this is from the fact that page_cgroup *was* dynamically
allocated.

But now, we allocate all page_cgroup at boot.  And
mem_cgroup_try_to_free_pages() reclaim memory from GFP_HIGHUSER_MOVABLE +
specified GFP_RECLAIM_MASK.

  * This is because we just want to reduce memory usage.
    "Where we should reclaim from ?" is not a problem in memcg.

This patch modifies gfp masks to be GFP_HIGUSER_MOVABLE if possible.

Note: This patch is not for fixing behavior but for showing sane information
      in source code.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: introduce charge-commit-cancel style of functions

There is a small race in do_swap_page().  When the page swapped-in is
charged, the mapcount can be greater than 0.  But, at the same time some
process (shares it ) call unmap and make mapcount 1->0 and the page is
uncharged.

      CPUA  CPUB
       mapcount == 1.
   (1) charge if mapcount==0     zap_pte_range()
                                (2) mapcount 1 => 0.
        (3) uncharge(). (success)
   (4) set page's rmap()
       mapcount 0=>1

Then, this swap page's account is leaked.

For fixing this, I added a new interface.
  - charge
   account to res_counter by PAGE_SIZE and try to free pages if necessary.
  - commit
   register page_cgroup and add to LRU if necessary.
  - cancel
   uncharge PAGE_SIZE because of do_swap_page failure.

     CPUA
  (1) charge (always)
  (2) set page's rmap (mapcount > 0)
  (3) commit charge was necessary or not after set_pte().

This protocol uses PCG_USED bit on page_cgroup for avoiding over accounting.
Usual mem_cgroup_charge_common() does charge -> commit at a time.

And this patch also adds following function to clarify all charges.

  - mem_cgroup_newpage_charge() ....replacement for mem_cgroup_charge()
called against newly allocated anon pages.

  - mem_cgroup_charge_migrate_fixup()
        called only from remove_migration_ptes().
we'll have to rewrite this later.(this patch just keeps old behavior)
This function will be removed by additional patch to make migration
clearer.

Good for clarifying "what we do"

Then, we have 4 following charge points.
  - newpage
  - swap-in
  - add-to-cache.
  - migration.

[akpm@linux-foundation.org: add missing inline directives to stubs]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

devices cgroup: allow mkfifo

The devcgroup_inode_permission() hook in the devices whitelist cgroup has
always bypassed access checks on fifos.  But the mknod hook did not.  The
devices whitelist is only about block and char devices, and fifos can't
even be added to the whitelist, so fifos can't be created at all except by
tasks which have 'a' in their whitelist (meaning they have access to all
devices).

Fix the behavior by bypassing access checks to mkfifo.

Signed-off-by: Serge E. Hallyn <serue@us.ibm.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Cc: Pavel Emelyanov <xemul@openvz.org>
Cc: Paul Menage <menage@google.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: James Morris <jmorris@namei.org>
Reported-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Cc: <stable@kernel.org> [2.6.27.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

devcgroup: use list_for_each_entry_rcu()

We should use list_for_each_entry_rcu in RCU read site.

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

cgroups: make cgroup_path() RCU-safe

Fix races between /proc/sched_debug by freeing cgroup objects via an RCU
callback.  Thus any cgroup reference obtained from an RCU-safe source will
remain valid during the RCU section.  Since dentries are also RCU-safe,
this allows us to traverse up the tree safely.

Additionally, make cgroup_path() check for a NULL cgrp->dentry to avoid
trying to report a path for a partially-created cgroup.

[lizf@cn.fujitsu.com: call deactive_super() in cgroup_diput()]
Signed-off-by: Paul Menage <menage@google.com>
Reviewed-by: Li Zefan <lizf@cn.fujitsu.com>
Tested-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
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>

cgroups: skip processes from other namespaces when listing a cgroup

Once tasks are populated from system namespace inside cgroup, container
replaces other namespace task with 0 while listing tasks, inside
container.

Though this is expected behaviour from container end, there is no use of
showing unwanted 0s.

In this patch, we check if a process is in same namespace before loading
into pid array.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Gowrishankar M <gowrishankar.m@in.ibm.com>
Acked-by: Paul Menage <menage@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cgroups: introduce link_css_set() to remove duplicate code

Add a common function link_css_set() to link a css_set to a cgroup.

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

cgroups: add inactive subsystems to rootnode.subsys_list

Though for an inactive hierarchy, we have subsys->root == &rootnode, but
rootnode's subsys_list is always empty.

This conflicts with the code in find_css_set():

for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
...
if (ss->root->subsys_list.next == &ss->sibling) {
...
}
}
if (list_empty(&rootnode.subsys_list)) {
...
}

The above code assumes rootnode.subsys_list links all inactive
hierarchies.

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

cgroups: make root_list contains active hierarchies only

Don't link rootnode to the root list, so root_list contains active
hierarchies only as the comment indicates.  And rename for_each_root() to
for_each_active_root().

Also remove redundant check in cgroup_kill_sb().

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

cgroups: remove rcu_read_lock() in cgroupstats_build()

cgroup_iter_* do not need rcu_read_lock().

In cgroup_enable_task_cg_lists(), do_each_thread() and while_each_thread()
are protected by RCU, it's OK, for write_lock(&css_set_lock) implies
rcu_read_lock() in non-RT kernel.

If we need explicit rcu_read_lock(), we should add rcu_read_lock() in
cgroup_enable_task_cg_lists(), not cgroup_iter_*.

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

cgroups: call find_css_set() safely in cgroup_attach_task()

In cgroup_attach_task(), tsk maybe exit when we call find_css_set().  and
find_css_set() will access to invalid css_set.

This patch increases the count before get_css_set(), and decreases it
after find_css_set().

NOTE:

css_set's refcount is also taskcount, after this patch applied, taskcount
may be off-by-one WHEN cgroup_lock() is not held.  but I reviewed other
code which use taskcount, they are still correct.  No regression found by
reviewing and simply testing.

So I do not use two counters in css_set.  (one counter for taskcount, the
other for refcount.  like struct mm_struct) If this fix cause regression,
we will use two counters in css_set.

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

cgroups: use task_lock() for access tsk->cgroups safe in cgroup_clone()

Use task_lock() protect tsk->cgroups and get_css_set(tsk->cgroups).

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

cgroups: don't put struct cgroupfs_root protected by RCU

We don't access struct cgroupfs_root in fast path, so we should not put
struct cgroupfs_root protected by RCU

But the comment in struct cgroup_subsys.root confuse us.

struct cgroup_subsys.root is used in these places:

1 find_css_set(): if (ss->root->subsys_list.next == &ss->sibling)
2 rebind_subsystems(): if (ss->root != &rootnode)
                       rcu_assign_pointer(ss->root, root);
                       rcu_assign_pointer(subsys[i]->root, &rootnode);
3 cgroup_has_css_refs(): if (ss->root != cgrp->root)
4 cgroup_init_subsys(): ss->root = &rootnode;
5 proc_cgroupstats_show(): ss->name, ss->root->subsys_bits,
                           ss->root->number_of_cgroups, !ss->disabled);
6 cgroup_clone(): root = subsys->root;
                  if ((root != subsys->root) ||

All these place we have held cgroup_lock() or we don't dereference to
struct cgroupfs_root.  It's means wo don't need RCU when use struct
cgroup_subsys.root, and we should not put struct cgroupfs_root protected
by RCU.

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

cgroups: fix cgroup_iter_next() bug

We access res->cgroups without the task_lock(), so res->cgroups may be
changed.  it's unreliable, and "if (l == &res->cgroups->tasks)" may be
false forever.

We don't need add any lock for fixing this bug.  we just access to struct
css_set by struct cg_cgroup_link, not by struct task_struct.

Since we hold css_set_lock, struct cg_cgroup_link is reliable.

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

cgroups: add lock for child->cgroups in cgroup_post_fork()

When cgroup_post_fork() is called, child is seen by find_task_by_vpid(),
so child->cgroups maybe be changed, It'll incorrect.

child->cgroups<old>'s refcnt is decreased
child->cgroups<new>'s refcnt is increased
but child->cg_list is added to child->cgroups<old>'s list.

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

memcg: fix a typo in Kconfig

s/contoller/controller/

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

ns_cgroup: remove unused spinlock

I happened to find the spinlock in struct ns_cgroup is never used.

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

cgroups: remove some redundant NULL checks

- In cgroup_clone(), if vfs_mkdir() returns successfully,
  dentry->d_fsdata will be the pointer to the newly created
  cgroup and won't be NULL.

- a cgroup file's dentry->d_fsdata won't be NULL, guaranteed
  by cgroup_add_file().

- When walking through the subsystems of a cgroup_fs (using
  for_each_subsys), cgrp->subsys[ss->subsys_id] won't be NULL,
  guaranteed by cgroup_create().

(Also remove 2 unused variables in cgroup_rmdir().

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

cgroups: documentation updates

- remove 'releasable' since it has been moved to the debug subsys.
- update lock requirements of subsys callbacks.

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

cgroups: make cgroup config a submenu

Making CGROUP related configs be a sub-menu.

This patch make CGROUP related configs be a sub-menu and makes 1st level
configs of "General Setup" shorter.

 including following additional changes
  - add help comment about CGROUPS and GROUP_SCHED.
  - moved MM_OWNER config to the bottom.
    (for good indent in menuconfig)

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@in.ibm.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>

quota: don't set grace time when user isn't above softlimit

do_set_dqblk() allowed SETDQBLK quotactl to set user's grace time even if
user was not above his softlimit.  This does not make much sence and by
coincidence causes quota code to omit softlimit warning when user really
exceeds softlimit.  This patch makes do_set_dqblk() reset user's grace
time if he has not exceeded softlimit.

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>

coda: fix fs/coda/sysctl.c build warnings when !CONFIG_SYSCTL

Fix
fs/coda/sysctl.c:14: warning: 'fs_table_header' defined but not used
fs/coda/sysctl.c:44: warning: 'fs_table' defined but not used

these are only used when CONFIG_SYSCTL is defined.

Signed-off-by: Richard A. Holden III <aciddeath@gmail.com>
Cc: Jan Harkes <jaharkes@cs.cmu.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

jbd: remove excess kernel-doc notation

Remove excess kernel-doc from fs/jbd/transaction.c:

Warning(linux-2.6.28-git5//fs/jbd/transaction.c:764): Excess function parameter 'credits' description in 'journal_get_write_access'

Signed-off-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>

ext3: tighten restrictions on inode flags

At the moment there are few restrictions on which flags may be set on
which inodes.  Specifically DIRSYNC may only be set on directories and
IMMUTABLE and APPEND may not be set on links.  Tighten that to disallow
TOPDIR being set on non-directories and only NODUMP and NOATIME to be set
on non-regular file, non-directories.

Introduces a flags masking function which masks flags based on mode and
use it during inode creation and when flags are set via the ioctl to
facilitate future consistency.

Signed-off-by: Duane Griffin <duaneg@dghda.com>
Acked-by: Andreas Dilger <adilger@sun.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 inherit inappropriate inode flags from parent

At present INDEX is the only flag that new ext3 inodes do NOT inherit from
their parent.  In addition prevent the flags DIRTY, ECOMPR, IMAGIC and
TOPDIR from being inherited.  List inheritable flags explicitly to prevent
future flags from accidentally being inherited.

This fixes the TOPDIR flag inheritance bug reported at
http://bugzilla.kernel.org/show_bug.cgi?id=9866.

Signed-off-by: Duane Griffin <duaneg@dghda.com>
Acked-by: Andreas Dilger <adilger@sun.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: allocate ->s_blockgroup_lock separately

As spotted by kmemtrace, struct ext3_sb_info is 17152 bytes on 64-bit
which makes it a very bad fit for SLAB allocators.  The culprit of the
wasted memory is ->s_blockgroup_lock which can be as big as 16 KB when
NR_CPUS >= 32.

To fix that, allocate ->s_blockgroup_lock, which fits nicely in a order 2
page in the worst case, separately.  This shinks down struct ext3_sb_info
enough to fit a 1 KB slab cache so now we allocate 16 KB + 1 KB instead of
32 KB saving 15 KB of memory.

Acked-by: Andreas Dilger <adilger@sun.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
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: improve fsync batching

There is a flaw with the way jbd handles fsync batching.  If we fsync() a
file and we were not the last person to run fsync() on this fs then we
automatically sleep for 1 jiffie in order to wait for new writers to join
into the transaction before forcing the commit.  The problem with this is
that with really fast storage (ie a Clariion) the time it takes to commit
a transaction to disk is way faster than 1 jiffie in most cases, so
sleeping means waiting longer with nothing to do than if we just committed
the transaction and kept going.  Ric Wheeler noticed this when using
fs_mark with more than 1 thread, the throughput would plummet as he added
more threads.

This patch attempts to fix this problem by recording the average time in
nanoseconds that it takes to commit a transaction to disk, and what time
we started the transaction.  If we run an fsync() and we have been running
for less time than it takes to commit the transaction to disk, we sleep
for the delta amount of time and then commit to disk.  We acheive
sub-jiffie sleeping using schedule_hrtimeout.  This means that the wait
time is auto-tuned to the speed of the underlying disk, instead of having
this static timeout.  I weighted the average according to somebody's
comments (Andreas Dilger I think) in order to help normalize random
outliers where we take way longer or way less time to commit than the
average.  I also have a min() check in there to make sure we don't sleep
longer than a jiffie in case our storage is super slow, this was requested
by Andrew.

I unfortunately do not have access to a Clariion, so I had to use a
ramdisk to represent a super fast array.  I tested with a SATA drive with
barrier=1 to make sure there was no regression with local disks, I tested
with a 4 way multipathed Apple Xserve RAID array and of course the
ramdisk.  I ran the following command

fs_mark -d /mnt/ext3-test -s 4096 -n 2000 -D 64 -t $i

where $i was 2, 4, 8, 16 and 32.  I mkfs'ed the fs each time.  Here are my
results

type threads with patch without patch
sata 2 24.6 26.3
sata 4 49.2 48.1
sata 8 70.1 67.0
sata 16 104.0 94.1
sata 32 153.6 142.7

xserve 2 246.4 222.0
xserve 4 480.0 440.8
xserve 8 829.5 730.8
xserve 16 1172.7 1026.9
xserve 32 1816.3 1650.5

ramdisk 2 2538.3 1745.6
ramdisk 4 2942.3 661.9
ramdisk 8 2882.5 999.8
ramdisk 16 2738.7 1801.9
ramdisk 32 2541.9 2394.0

Signed-off-by: Josef Bacik <jbacik@redhat.com>
Cc: Andreas Dilger <adilger@sun.com>
Cc: Arjan van de Ven <arjan@infradead.org>
Cc: Ric Wheeler <rwheeler@redhat.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>

ext2: tighten restrictions on inode flags

At the moment there are few restrictions on which flags may be set on
which inodes.  Specifically DIRSYNC may only be set on directories and
IMMUTABLE and APPEND may not be set on links.  Tighten that to disallow
TOPDIR being set on non-directories and only NODUMP and NOATIME to be set
on non-regular file, non-directories.

Introduces a flags masking function which masks flags based on mode and
use it during inode creation and when flags are set via the ioctl to
facilitate future consistency.

Signed-off-by: Duane Griffin <duaneg@dghda.com>
Acked-by: Andreas Dilger <adilger@sun.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>