workqueue: fix is_chained_work() regression

c9e7cf273f ("workqueue: move busy_hash from global_cwq to
worker_pool") incorrectly converted is_chained_work() to use
get_gcwq() inside for_each_gcwq_cpu() while removing get_gcwq().

As cwq might not exist for all possible workqueue CPUs, @cwq can be
NULL and the following cwq deferences can lead to oops.

Fix it by using for_each_cwq_cpu() instead, which is the better one to
use anyway as we only need to check pools that the wq is associated
with.

Signed-off-by: Tejun Heo <tj@kernel.org>

workqueue: pick cwq instead of pool in __queue_work()

Currently, __queue_work() chooses the pool to queue a work item to and
then determines cwq from the target wq and the chosen pool.  This is a
bit backwards in that we can determine cwq first and simply use
cwq->pool.  This way, we can skip get_std_worker_pool() in queueing
path which will be a hurdle when implementing custom worker pools.

Update __queue_work() such that it chooses the target cwq and then use
cwq->pool instead of the other way around.  While at it, add missing
{} in an if statement.

This patch doesn't introduce any functional changes.

tj: The original patch had two get_cwq() calls - the first to
    determine the pool by doing get_cwq(cpu, wq)->pool and the second
    to determine the matching cwq from get_cwq(pool->cpu, wq).
    Updated the function such that it chooses cwq instead of pool and
    removed the second call.  Rewrote the description.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

workqueue: make get_work_pool_id() cheaper

get_work_pool_id() currently first obtains pool using get_work_pool()
and then return pool->id.  For an off-queue work item, this involves
obtaining pool ID from worker->data, performing idr_find() to find the
matching pool and then returning its pool->id which of course is the
same as the one which went into idr_find().

Just open code WORK_STRUCT_CWQ case and directly return pool ID from
work->data.

tj: The original patch dropped on-queue work item handling and renamed
    the function to offq_work_pool_id().  There isn't much benefit in
    doing so.  Handling it only requires a single if() and we need at
    least BUG_ON(), which is also a branch, even if we drop on-queue
    handling.  Open code WORK_STRUCT_CWQ case and keep the function in
    line with get_work_pool().  Rewrote the description.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

workqueue: move nr_running into worker_pool

As nr_running is likely to be accessed from other CPUs during
try_to_wake_up(), it was kept outside worker_pool; however, while less
frequent, other fields in worker_pool are accessed from other CPUs
for, e.g., non-reentrancy check.  Also, with recent pool related
changes, accessing nr_running matching the worker_pool isn't as simple
as it used to be.

Move nr_running inside worker_pool.  Keep it aligned to cacheline and
define CPU pools using DEFINE_PER_CPU_SHARED_ALIGNED().  This should
give at least the same cacheline behavior.

get_pool_nr_running() is replaced with direct pool->nr_running
accesses.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Joonsoo Kim <js1304@gmail.com>

workqueue: cosmetic update in try_to_grab_pending()

With the recent is-work-queued-here test simplification, the nested
if() in try_to_grab_pending() can be collapsed.  Collapse it.

This patch is purely cosmetic.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: simplify is-work-item-queued-here test

Currently, determining whether a work item is queued on a locked pool
involves somewhat convoluted memory barrier dancing.  It goes like the
following.

* When a work item is queued on a pool, work->data is updated before
  work->entry is linked to the pending list with a wmb() inbetween.

* When trying to determine whether a work item is currently queued on
  a pool pointed to by work->data, it locks the pool and looks at
  work->entry.  If work->entry is linked, we then do rmb() and then
  check whether work->data points to the current pool.

This works because, work->data can only point to a pool if it
currently is or were on the pool and,

* If it currently is on the pool, the tests would obviously succeed.

* It it left the pool, its work->entry was cleared under pool->lock,
  so if we're seeing non-empty work->entry, it has to be from the work
  item being linked on another pool.  Because work->data is updated
  before work->entry is linked with wmb() inbetween, work->data update
  from another pool is guaranteed to be visible if we do rmb() after
  seeing non-empty work->entry.  So, we either see empty work->entry
  or we see updated work->data pointin to another pool.

While this works, it's convoluted, to put it mildly.  With recent
updates, it's now guaranteed that work->data points to cwq only while
the work item is queued and that updating work->data to point to cwq
or back to pool is done under pool->lock, so we can simply test
whether work->data points to cwq which is associated with the
currently locked pool instead of the convoluted memory barrier
dancing.

This patch replaces the memory barrier based "are you still here,
really?" test with much simpler "does work->data points to me?" test -
if work->data points to a cwq which is associated with the currently
locked pool, the work item is guaranteed to be queued on the pool as
work->data can start and stop pointing to such cwq only under
pool->lock and the start and stop coincide with queue and dequeue.

tj: Rewrote the comments and description.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

workqueue: make work->data point to pool after try_to_grab_pending()

We plan to use work->data pointing to cwq as the synchronization
invariant when determining whether a given work item is on a locked
pool or not, which requires work->data pointing to cwq only while the
work item is queued on the associated pool.

With delayed_work updated not to overload work->data for target
workqueue recording, the only case where we still have off-queue
work->data pointing to cwq is try_to_grab_pending() which doesn't
update work->data after stealing a queued work item.  There's no
reason for try_to_grab_pending() to not update work->data to point to
the pool instead of cwq, like the normal execution does.

This patch adds set_work_pool_and_keep_pending() which makes
work->data point to pool instead of cwq but keeps the pending bit
unlike set_work_pool_and_clear_pending() (surprise!).

After this patch, it's guaranteed that only queued work items point to
cwqs.

This patch doesn't introduce any visible behavior change.

tj: Renamed the new helper function to match
    set_work_pool_and_clear_pending() and rewrote the description.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

workqueue: add delayed_work->wq to simplify reentrancy handling

To avoid executing the same work item from multiple CPUs concurrently,
a work_struct records the last pool it was on in its ->data so that,
on the next queueing, the pool can be queried to determine whether the
work item is still executing or not.

A delayed_work goes through timer before actually being queued on the
target workqueue and the timer needs to know the target workqueue and
CPU.  This is currently achieved by modifying delayed_work->work.data
such that it points to the cwq which points to the target workqueue
and the last CPU the work item was on.  __queue_delayed_work()
extracts the last CPU from delayed_work->work.data and then combines
it with the target workqueue to create new work.data.

The only thing this rather ugly hack achieves is encoding the target
workqueue into delayed_work->work.data without using a separate field,
which could be a trade off one can make; unfortunately, this entangles
work->data management between regular workqueue and delayed_work code
by setting cwq pointer before the work item is actually queued and
becomes a hindrance for further improvements of work->data handling.

This can be easily made sane by adding a target workqueue field to
delayed_work.  While delayed_work is used widely in the kernel and
this does make it a bit larger (<5%), I think this is the right
trade-off especially given the prospect of much saner handling of
work->data which currently involves quite tricky memory barrier
dancing, and don't expect to see any measureable effect.

Add delayed_work->wq and drop the delayed_work->work.data overloading.

tj: Rewrote the description.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

workqueue: make work_busy() test WORK_STRUCT_PENDING first

Currently, work_busy() first tests whether the work has a pool
associated with it and if not, considers it idle.  This works fine
even for delayed_work.work queued on timer, as __queue_delayed_work()
sets cwq on delayed_work.work - a queued delayed_work always has its
cwq and thus pool associated with it.

However, we're about to update delayed_work queueing and this won't
hold.  Update work_busy() such that it tests WORK_STRUCT_PENDING
before the associated pool.  This doesn't make any noticeable behavior
difference now.

With work_pending() test moved, the function read a lot better with
"if (!pool)" test flipped to positive.  Flip it.

While at it, lose the comment about now non-existent reentrant
workqueues.

tj: Reorganized the function and rewrote the description.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

workqueue: replace WORK_CPU_NONE/LAST with WORK_CPU_END

Now that workqueue has moved away from gcwqs, workqueue no longer has
the need to have a CPU identifier indicating "no cpu associated" - we
now use WORK_OFFQ_POOL_NONE instead - and most uses of WORK_CPU_NONE
are gone.

The only left usage is as the end marker for for_each_*wq*()
iterators, where the name WORK_CPU_NONE is confusing w/o actual
WORK_CPU_NONE usages.  Similarly, WORK_CPU_LAST which equals
WORK_CPU_NONE no longer makes sense.

Replace both WORK_CPU_NONE and LAST with WORK_CPU_END.  This patch
doesn't introduce any functional difference.

tj: s/WORK_CPU_LAST/WORK_CPU_END/ and rewrote the description.

Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

workqueue: post global_cwq removal cleanups

Remove remaining references to gcwq.

* __next_gcwq_cpu() steals __next_wq_cpu() name.  The original
  __next_wq_cpu() became __next_cwq_cpu().

* s/for_each_gcwq_cpu/for_each_wq_cpu/
  s/for_each_online_gcwq_cpu/for_each_online_wq_cpu/

* s/gcwq_mayday_timeout/pool_mayday_timeout/

* s/gcwq_unbind_fn/wq_unbind_fn/

* Drop references to gcwq in comments.

This patch doesn't introduce any functional changes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: rename nr_running variables

Rename per-cpu and unbound nr_running variables such that they match
the pool variables.

This patch doesn't introduce any functional changes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: remove global_cwq

global_cwq is now nothing but a container for per-cpu standard
worker_pools.  Declare the worker pools directly as
cpu/unbound_std_worker_pools[] and remove global_cwq.

* ____cacheline_aligned_in_smp moved from global_cwq to worker_pool.
  This probably would have made sense even before this change as we
  want each pool to be aligned.

* get_gcwq() is replaced with std_worker_pools() which returns the
  pointer to the standard pool array for a given CPU.

* __alloc_workqueue_key() updated to use get_std_worker_pool() instead
  of open-coding pool determination.

This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.

v2: Joonsoo pointed out that it'd better to align struct worker_pool
    rather than the array so that every pool is aligned.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Joonsoo Kim <js1304@gmail.com>

workqueue: remove worker_pool->gcwq

The only remaining user of pool->gcwq is std_worker_pool_pri().
Reimplement it using get_gcwq() and remove worker_pool->gcwq.

This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: replace for_each_worker_pool() with for_each_std_worker_pool()

for_each_std_worker_pool() takes @cpu instead of @gcwq.

This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: make freezing/thawing per-pool

Instead of holding locks from both pools and then processing the pools
together, make freezing/thwaing per-pool - grab locks of one pool,
process it, release it and then proceed to the next pool.

While this patch changes processing order across pools, order within
each pool remains the same.  As each pool is independent, this
shouldn't break anything.

This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: make hotplug processing per-pool

Instead of holding locks from both pools and then processing the pools
together, make hotplug processing per-pool - grab locks of one pool,
process it, release it and then proceed to the next pool.

rebind_workers() is updated to take and process @pool instead of @gcwq
which results in a lot of de-indentation.  gcwq_claim_assoc_and_lock()
and its counterpart are replaced with in-line per-pool locking.

While this patch changes processing order across pools, order within
each pool remains the same.  As each pool is independent, this
shouldn't break anything.

This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: move global_cwq->lock to worker_pool

Move gcwq->lock to pool->lock.  The conversion is mostly
straight-forward.  Things worth noting are

* In many places, this removes the need to use gcwq completely.  pool
  is used directly instead.  get_std_worker_pool() is added to help
  some of these conversions.  This also leaves get_work_gcwq() without
  any user.  Removed.

* In hotplug and freezer paths, the pools belonging to a CPU are often
  processed together.  This patch makes those paths hold locks of all
  pools, with highpri lock nested inside, to keep the conversion
  straight-forward.  These nested lockings will be removed by
  following patches.

This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: move global_cwq->cpu to worker_pool

Move gcwq->cpu to pool->cpu.  This introduces a couple places where
gcwq->pools[0].cpu is used.  These will soon go away as gcwq is
further reduced.

This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: move busy_hash from global_cwq to worker_pool

There's no functional necessity for the two pools on the same CPU to
share the busy hash table.  It's also likely to be a bottleneck when
implementing pools with user-specified attributes.

This patch makes busy_hash per-pool.  The conversion is mostly
straight-forward.  Changes worth noting are,

* Large block of changes in rebind_workers() is moving the block
  inside for_each_worker_pool() as now there are separate hash tables
  for each pool.  This changes the order of operations but doesn't
  break anything.

* Thre for_each_worker_pool() loops in gcwq_unbind_fn() are combined
  into one.  This again changes the order of operaitons but doesn't
  break anything.

This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: record pool ID instead of CPU in work->data when off-queue

Currently, when a work item is off-queue, work->data records the CPU
it was last on, which is used to locate the last executing instance
for non-reentrance, flushing, etc.

We're in the process of removing global_cwq and making worker_pool the
top level abstraction.  This patch makes work->data point to the pool
it was last associated with instead of CPU.

After the previous WORK_OFFQ_POOL_CPU and worker_poo->id additions,
the conversion is fairly straight-forward.  WORK_OFFQ constants and
functions are modified to record and read back pool ID instead.
worker_pool_by_id() is added to allow looking up pool from ID.
get_work_pool() replaces get_work_gcwq(), which is reimplemented using
get_work_pool().  get_work_pool_id() replaces work_cpu().

This patch shouldn't introduce any observable behavior changes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: add worker_pool->id

Add worker_pool->id which is allocated from worker_pool_idr.  This
will be used to record the last associated worker_pool in work->data.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: introduce WORK_OFFQ_CPU_NONE

Currently, when a work item is off queue, high bits of its data
encodes the last CPU it was on.  This is scheduled to be changed to
pool ID, which will make it impossible to use WORK_CPU_NONE to
indicate no association.

This patch limits the number of bits which are used for off-queue cpu
number to 31 (so that the max fits in an int) and uses the highest
possible value - WORK_OFFQ_CPU_NONE - to indicate no association.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: make GCWQ_FREEZING a pool flag

Make GCWQ_FREEZING a pool flag POOL_FREEZING.  This patch doesn't
change locking - FREEZING on both pools of a CPU are set or clear
together while holding gcwq->lock.  It shouldn't cause any functional
difference.

This leaves gcwq->flags w/o any flags.  Removed.

While at it, convert BUG_ON()s in freeze_workqueue_begin() and
thaw_workqueues() to WARN_ON_ONCE().

This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: make GCWQ_DISASSOCIATED a pool flag

Make GCWQ_DISASSOCIATED a pool flag POOL_DISASSOCIATED.  This patch
doesn't change locking - DISASSOCIATED on both pools of a CPU are set
or clear together while holding gcwq->lock.  It shouldn't cause any
functional difference.

This is part of an effort to remove global_cwq and make worker_pool
the top level abstraction, which in turn will help implementing worker
pools with user-specified attributes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: use std_ prefix for the standard per-cpu pools

There are currently two worker pools per cpu (including the unbound
cpu) and they are the only pools in use.  New class of pools are
scheduled to be added and some pool related APIs will be added
inbetween.  Call the existing pools the standard pools and prefix them
with std_.  Do this early so that new APIs can use std_ prefix from
the beginning.

This patch doesn't introduce any functional difference.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: unexport work_cpu()

This function no longer has any external users.  Unexport it.  It will
be removed later on.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>

workqueue: implement current_is_async()

This function queries whether %current is an async worker executing an
async item.  This will be used to implement warning on synchronous
request_module() from async workers.

Signed-off-by: Tejun Heo <tj@kernel.org>

workqueue: move struct worker definition to workqueue_internal.h

This will be used to implement an inline function to query whether
%current is a workqueue worker and, if so, allow determining which
work item it's executing.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>

workqueue: rename kernel/workqueue_sched.h to kernel/workqueue_internal.h

Workqueue wants to expose more interface internal to kernel/.  Instead
of adding a new header file, repurpose kernel/workqueue_sched.h.
Rename it to workqueue_internal.h and add include protector.

This patch doesn't introduce any functional changes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>

workqueue: set PF_WQ_WORKER on rescuers

PF_WQ_WORKER is used to tell scheduler that the task is a workqueue
worker and needs wq_worker_sleeping/waking_up() invoked on it for
concurrency management.  As rescuers never participate in concurrency
management, PF_WQ_WORKER wasn't set on them.

There's a need for an interface which can query whether %current is
executing a work item and if so which.  Such interface requires a way
to identify all tasks which may execute work items and PF_WQ_WORKER
will be used for that.  As all normal workers always have PF_WQ_WORKER
set, we only need to add it to rescuers.

As rescuers start with WORKER_PREP but never clear it, it's always
NOT_RUNNING and there's no need to worry about it interfering with
concurrency management even if PF_WQ_WORKER is set; however, unlike
normal workers, rescuers currently don't have its worker struct as
kthread_data().  It uses the associated workqueue_struct instead.
This is problematic as wq_worker_sleeping/waking_up() expect struct
worker at kthread_data().

This patch adds worker->rescue_wq and start rescuer kthreads with
worker struct as kthread_data and sets PF_WQ_WORKER on rescuers.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>

workqueue: fix find_worker_executing_work() brekage from hashtable conversion

42f8570f43 ("workqueue: use new hashtable implementation") incorrectly
made busy workers hashed by the pointer value of worker instead of
work.  This broke find_worker_executing_work() which in turn broke a
lot of fundamental operations of workqueue - non-reentrancy and
flushing among others.  The flush malfunction triggered warning in
disk event code in Fengguang's automated test.

 write_dev_root_ (3265) used greatest stack depth: 2704 bytes left
 ------------[ cut here ]------------
 WARNING: at /c/kernel-tests/src/stable/block/genhd.c:1574 disk_clear_events+0x\
cf/0x108()
 Hardware name: Bochs
 Modules linked in:
 Pid: 3328, comm: ata_id Not tainted 3.7.0-01930-gbff6343 #1167
 Call Trace:
  [<ffffffff810997c4>] warn_slowpath_common+0x83/0x9c
  [<ffffffff810997f7>] warn_slowpath_null+0x1a/0x1c
  [<ffffffff816aea77>] disk_clear_events+0xcf/0x108
  [<ffffffff811bd8be>] check_disk_change+0x27/0x59
  [<ffffffff822e48e2>] cdrom_open+0x49/0x68b
  [<ffffffff81ab0291>] idecd_open+0x88/0xb7
  [<ffffffff811be58f>] __blkdev_get+0x102/0x3ec
  [<ffffffff811bea08>] blkdev_get+0x18f/0x30f
  [<ffffffff811bebfd>] blkdev_open+0x75/0x80
  [<ffffffff8118f510>] do_dentry_open+0x1ea/0x295
  [<ffffffff8118f5f0>] finish_open+0x35/0x41
  [<ffffffff8119c720>] do_last+0x878/0xa25
  [<ffffffff8119c993>] path_openat+0xc6/0x333
  [<ffffffff8119cf37>] do_filp_open+0x38/0x86
  [<ffffffff81190170>] do_sys_open+0x6c/0xf9
  [<ffffffff8119021e>] sys_open+0x21/0x23
  [<ffffffff82c1c3d9>] system_call_fastpath+0x16/0x1b

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Cc: Sasha Levin <sasha.levin@oracle.com>

workqueue: consider work function when searching for busy work items

To avoid executing the same work item concurrenlty, workqueue hashes
currently busy workers according to their current work items and looks
up the the table when it wants to execute a new work item.  If there
already is a worker which is executing the new work item, the new item
is queued to the found worker so that it gets executed only after the
current execution finishes.

Unfortunately, a work item may be freed while being executed and thus
recycled for different purposes.  If it gets recycled for a different
work item and queued while the previous execution is still in
progress, workqueue may make the new work item wait for the old one
although the two aren't really related in any way.

In extreme cases, this false dependency may lead to deadlock although
it's extremely unlikely given that there aren't too many self-freeing
work item users and they usually don't wait for other work items.

To alleviate the problem, record the current work function in each
busy worker and match it together with the work item address in
find_worker_executing_work().  While this isn't complete, it ensures
that unrelated work items don't interact with each other and in the
very unlikely case where a twisted wq user triggers it, it's always
onto itself making the culprit easy to spot.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Andrey Isakov <andy51@gmx.ru>
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=51701
Cc: stable@vger.kernel.org

workqueue: use new hashtable implementation

Switch workqueues to use the new hashtable implementation. This reduces the
amount of generic unrelated code in the workqueues.

This patch depends on d9b482c ("hashtable: introduce a small and naive
hashtable") which was merged in v3.6.

Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

Merge branch 'akpm' (Andrew's patch-bomb)

Merge misc patches from Andrew Morton:
 "Incoming:

   - lots of misc stuff

   - backlight tree updates

   - lib/ updates

   - Oleg's percpu-rwsem changes

   - checkpatch

   - rtc

   - aoe

   - more checkpoint/restart support

  I still have a pile of MM stuff pending - Pekka should be merging
  later today after which that is good to go.  A number of other things
  are twiddling thumbs awaiting maintainer merges."

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (180 commits)
  scatterlist: don't BUG when we can trivially return a proper error.
  docs: update documentation about /proc/<pid>/fdinfo/<fd> fanotify output
  fs, fanotify: add @mflags field to fanotify output
  docs: add documentation about /proc/<pid>/fdinfo/<fd> output
  fs, notify: add procfs fdinfo helper
  fs, exportfs: add exportfs_encode_inode_fh() helper
  fs, exportfs: escape nil dereference if no s_export_op present
  fs, epoll: add procfs fdinfo helper
  fs, eventfd: add procfs fdinfo helper
  procfs: add ability to plug in auxiliary fdinfo providers
  tools/testing/selftests/kcmp/kcmp_test.c: print reason for failure in kcmp_test
  breakpoint selftests: print failure status instead of cause make error
  kcmp selftests: print fail status instead of cause make error
  kcmp selftests: make run_tests fix
  mem-hotplug selftests: print failure status instead of cause make error
  cpu-hotplug selftests: print failure status instead of cause make error
  mqueue selftests: print failure status instead of cause make error
  vm selftests: print failure status instead of cause make error
  ubifs: use prandom_bytes
  mtd: nandsim: use prandom_bytes
  ...

efi: Fix the build with user namespaces enabled.

When compiling efivars.c the build fails with:

   CC      drivers/firmware/efivars.o
  drivers/firmware/efivars.c: In function ‘efivarfs_get_inode’:
  drivers/firmware/efivars.c:886:31: error: incompatible types when assigning to type ‘kgid_t’ from type ‘int’
  make[2]: *** [drivers/firmware/efivars.o] Error 1
  make[1]: *** [drivers/firmware/efivars.o] Error 2

Fix the build error by removing the duplicate initialization of i_uid and
i_gid inode_init_always has already initialized them to 0.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm,numa: fix update_mmu_cache_pmd call

This build error is currently hidden by the fact that the x86
implementation of 'update_mmu_cache_pmd()' is a macro that doesn't use
its last argument, but commit b32967ff101a ("mm: numa: Add THP migration
for the NUMA working set scanning fault case") introduced a call with
the wrong third argument.

In the akpm tree, it causes this build error:

  mm/migrate.c: In function 'migrate_misplaced_transhuge_page_put':
  mm/migrate.c:1666:2: error: incompatible type for argument 3 of 'update_mmu_cache_pmd'
  arch/x86/include/asm/pgtable.h:792:20: note: expected 'struct pmd_t *' but argument is of type 'pmd_t'

Fix it.

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

scatterlist: don't BUG when we can trivially return a proper error.

There is absolutely no reason to crash the kernel when we have a
perfectly good return value already available to use for conveying
failure status.

Let's return an error code instead of crashing the kernel: that sounds
like a much better plan.

[akpm@linux-foundation.org: s/E2BIG/EINVAL/]
Signed-off-by: Nick Bowler <nbowler@elliptictech.com>
Cc: Maxim Levitsky <maximlevitsky@gmail.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

docs: update documentation about /proc/<pid>/fdinfo/<fd> fanotify output

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: James Bottomley <jbottomley@parallels.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Matthew Helsley <matt.helsley@gmail.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@onelan.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs, fanotify: add @mflags field to fanotify output

The kernel keeps FAN_MARK_IGNORED_SURV_MODIFY bit separately from
fsnotify_mark::mask|ignored_mask thus put it in @mflags (mark flags)
field so the user-space reader will be able to detect if such bit were
used on mark creation procedure.

 | pos: 0
 | flags: 04002
 | fanotify flags:10 event-flags:0
 | fanotify mnt_id:12 mflags:40 mask:38 ignored_mask:40000003
 | fanotify ino:4f969 sdev:800013 mflags:0 mask:3b ignored_mask:40000000 fhandle-bytes:8 fhandle-type:1 f_handle:69f90400c275b5b4

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: James Bottomley <jbottomley@parallels.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Matthew Helsley <matt.helsley@gmail.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Tvrtko Ursulin <tvrtko.ursulin@onelan.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

docs: add documentation about /proc/<pid>/fdinfo/<fd> output

[akpm@linux-foundation.org: tweak documentation]
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: James Bottomley <jbottomley@parallels.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Matthew Helsley <matt.helsley@gmail.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@onelan.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs, notify: add procfs fdinfo helper

This allow us to print out fsnotify details such as watchee inode, device,
mask and optionally a file handle.

For inotify objects if kernel compiled with exportfs support the output
will be

 | pos: 0
 | flags: 02000000
 | inotify wd:3 ino:9e7e sdev:800013 mask:800afce ignored_mask:0 fhandle-bytes:8 fhandle-type:1 f_handle:7e9e0000640d1b6d
 | inotify wd:2 ino:a111 sdev:800013 mask:800afce ignored_mask:0 fhandle-bytes:8 fhandle-type:1 f_handle:11a1000020542153
 | inotify wd:1 ino:6b149 sdev:800013 mask:800afce ignored_mask:0 fhandle-bytes:8 fhandle-type:1 f_handle:49b1060023552153

If kernel compiled without exportfs support, the file handle
won't be provided but inode and device only.

 | pos: 0
 | flags: 02000000
 | inotify wd:3 ino:9e7e sdev:800013 mask:800afce ignored_mask:0
 | inotify wd:2 ino:a111 sdev:800013 mask:800afce ignored_mask:0
 | inotify wd:1 ino:6b149 sdev:800013 mask:800afce ignored_mask:0

For fanotify the output is like

 | pos: 0
 | flags: 04002
 | fanotify flags:10 event-flags:0
 | fanotify mnt_id:12 mask:3b ignored_mask:0
 | fanotify ino:50205 sdev:800013 mask:3b ignored_mask:40000000 fhandle-bytes:8 fhandle-type:1 f_handle:05020500fb1d47e7

To minimize impact on general fsnotify code the new functionality
is gathered in fs/notify/fdinfo.c file.

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: James Bottomley <jbottomley@parallels.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Matthew Helsley <matt.helsley@gmail.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@onelan.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs, exportfs: add exportfs_encode_inode_fh() helper

We will need this helper in the next patch to provide a file handle for
inotify marks in /proc/pid/fdinfo output.

The patch is rather providing the way to use inodes directly when dentry
is not available (like in case of inotify system).

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: James Bottomley <jbottomley@parallels.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Matthew Helsley <matt.helsley@gmail.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@onelan.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs, exportfs: escape nil dereference if no s_export_op present

This routine will be used to generate a file handle in fdinfo output for
inotify subsystem, where if no s_export_op present the general
export_encode_fh should be used.  Thus add a test if s_export_op present
inside exportfs_encode_fh itself.

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: James Bottomley <jbottomley@parallels.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Matthew Helsley <matt.helsley@gmail.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@onelan.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs, epoll: add procfs fdinfo helper

This allows us to print out eventpoll target file descriptor, events and
data, the /proc/pid/fdinfo/fd consists of

 | pos: 0
 | flags: 02
 | tfd:        5 events:       1d data: ffffffffffffffff enabled: 1

[avagin@: fix for unitialized ret variable]

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: James Bottomley <jbottomley@parallels.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Matthew Helsley <matt.helsley@gmail.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@onelan.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs, eventfd: add procfs fdinfo helper

This allows us to print out raw counter value.  The /proc/pid/fdinfo/fd
output is

 | pos: 0
 | flags: 04002
 | eventfd-count:               5a

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: James Bottomley <jbottomley@parallels.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Matthew Helsley <matt.helsley@gmail.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@onelan.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

procfs: add ability to plug in auxiliary fdinfo providers

This patch brings ability to print out auxiliary data associated with
file in procfs interface /proc/pid/fdinfo/fd.

In particular further patches make eventfd, evenpoll, signalfd and
fsnotify to print additional information complete enough to restore
these objects after checkpoint.

To simplify the code we add show_fdinfo callback inside struct
file_operations (as Al and Pavel are proposing).

Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Andrey Vagin <avagin@openvz.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: James Bottomley <jbottomley@parallels.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Matthew Helsley <matt.helsley@gmail.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@onelan.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

tools/testing/selftests/kcmp/kcmp_test.c: print reason for failure in kcmp_test

I was curious why sys_kcmp wasn't working, which led me to the testcase.
It turned out I hadn't enabled CHECKPOINT_RESTORE in the kernel I was
testing.  Add a decoding of errno to the testcase to make that obvious.

Signed-off-by: Dave Jones <davej@redhat.com>
Acked-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

breakpoint selftests: print failure status instead of cause make error

In case breakpoint test exit non zero value it will cause make error.
Better way is just print the test failure status.

Signed-off-by: Dave Young <dyoung@redhat.com>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kcmp selftests: print fail status instead of cause make error

In case kcmp_test exit non zero value it will cause make error.
Better way is just print the test failure status.

Signed-off-by: Dave Young <dyoung@redhat.com>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kcmp selftests: make run_tests fix

make run_tests need the target is run_tests instead of run-tests
Also gcc output should be kcmp_test. Fix these two issues.

Signed-off-by: Dave Young <dyoung@redhat.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mem-hotplug selftests: print failure status instead of cause make error

Original behavior:
  bash-4.1$ make -C memory-hotplug run_tests
  make: Entering directory `/home/dave/git/linux-2.6/tools/testing/selftests/memory-hotplug'
  ./on-off-test.sh
  make: execvp: ./on-off-test.sh: Permission denied
  make: *** [run_tests] Error 127
  make: Leaving directory `/home/dave/git/linux-2.6/tools/testing/selftests/memory-hotplug'

After applying the patch:
  bash-4.1$ make -C memory-hotplug run_tests
  make: Entering directory `/home/dave/git/linux-2.6/tools/testing/selftests/memory-hotplug'
  /bin/sh: ./on-off-test.sh: Permission denied
  memory-hotplug selftests: [FAIL]
  make: Leaving directory `/home/dave/git/linux-2.6/tools/testing/selftests/memory-hotplug'

Signed-off-by: Dave Young <dyoung@redhat.com>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

cpu-hotplug selftests: print failure status instead of cause make error

Original behavior:
  bash-4.1$ make -C cpu-hotplug run_tests
  make: Entering directory `/home/dave/git/linux-2.6/tools/testing/selftests/cpu-hotplug'
  ./on-off-test.sh
  make: execvp: ./on-off-test.sh: Permission denied
  make: *** [run_tests] Error 127
  make: Leaving directory `/home/dave/git/linux-2.6/tools/testing/selftests/cpu-hotplug'

After applying the patch:
  bash-4.1$ make -C cpu-hotplug run_tests
  make: Entering directory `/home/dave/git/linux-2.6/tools/testing/selftests/cpu-hotplug'
  /bin/sh: ./on-off-test.sh: Permission denied
  cpu-hotplug selftests: [FAIL]
  make: Leaving directory `/home/dave/git/linux-2.6/tools/testing/selftests/cpu-hotplug'

Signed-off-by: Dave Young <dyoung@redhat.com>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mqueue selftests: print failure status instead of cause make error

Original behavior:
  bash-4.1$ make -C mqueue run_tests
  make: Entering directory `/home/dave/git/linux-2.6/tools/testing/selftests/mqueue'
  ./mq_open_tests /test1
  Not running as root, but almost all tests require root in order to modify
  system settings.  Exiting.
  make: *** [run_tests] Error 1
  make: Leaving directory `/home/dave/git/linux-2.6/tools/testing/selftests/mqueue'

After applying the patch:
  bash-4.1$ make -C mqueue run_tests
  make: Entering directory `/home/dave/git/linux-2.6/tools/testing/selftests/mqueue'
  Not running as root, but almost all tests require root in order to modify
  system settings.  Exiting.
  mq_open_tests: [FAIL]
  Not running as root, but almost all tests require root in order to modify
  system settings.  Exiting.
  mq_perf_tests: [FAIL]
  make: Leaving directory `/home/dave/git/linux-2.6/tools/testing/selftests/mqueue'

Signed-off-by: Dave Young <dyoung@redhat.com>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vm selftests: print failure status instead of cause make error

Original behavior:
  bash-4.1$ make -C vm run_tests
  make: Entering directory `/home/dave/git/linux-2.6/tools/testing/selftests/vm'
  /bin/sh ./run_vmtests
  ./run_vmtests: line 24: /proc/sys/vm/nr_hugepages: Permission denied
  Please run this test as root
  make: *** [run_tests] Error 1
  make: Leaving directory `/home/dave/git/linux-2.6/tools/testing/selftests/vm'

After applying the patch:
  bash-4.1$ make -C vm run_tests
  make: Entering directory `/home/dave/git/linux-2.6/tools/testing/selftests/vm'
  ./run_vmtests: line 24: /proc/sys/vm/nr_hugepages: Permission denied
  Please run this test as root
  vmtests: [FAIL]
  make: Leaving directory `/home/dave/git/linux-2.6/tools/testing/selftests/vm'

Signed-off-by: Dave Young <dyoung@redhat.com>
Cc: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

ubifs: use prandom_bytes

This also converts filling memory loop to use memset.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Artem Bityutskiy <dedekind1@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: David Laight <david.laight@aculab.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Eilon Greenstein <eilong@broadcom.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Robert Love <robert.w.love@intel.com>
Cc: Valdis Kletnieks <valdis.kletnieks@vt.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mtd: nandsim: use prandom_bytes

This also removes unnecessary memset call which is immediately overwritten
with random bytes.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Artem Bityutskiy <dedekind1@gmail.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Laight <david.laight@aculab.com>
Cc: Eilon Greenstein <eilong@broadcom.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Robert Love <robert.w.love@intel.com>
Cc: Valdis Kletnieks <valdis.kletnieks@vt.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

bnx2x: use prandom_bytes()

Use prandom_bytes() to fill rss key with pseudo-random bytes.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Eilon Greenstein <eilong@broadcom.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Artem Bityutskiy <dedekind1@gmail.com>
Cc: David Laight <david.laight@aculab.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Michel Lespinasse <walken@google.com>
Cc: Robert Love <robert.w.love@intel.com>
Cc: Valdis Kletnieks <valdis.kletnieks@vt.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

prandom: introduce prandom_bytes() and prandom_bytes_state()

Add functions to get the requested number of pseudo-random bytes.

The difference from get_random_bytes() is that it generates pseudo-random
numbers by prandom_u32().  It doesn't consume the entropy pool, and the
sequence is reproducible if the same rnd_state is used.  So it is suitable
for generating random bytes for testing.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Artem Bityutskiy <dedekind1@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Eilon Greenstein <eilong@broadcom.com>
Cc: David Laight <david.laight@aculab.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Robert Love <robert.w.love@intel.com>
Cc: Valdis Kletnieks <valdis.kletnieks@vt.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

random32: rename random32 to prandom

This renames all random32 functions to have 'prandom_' prefix as follows:

  void prandom_seed(u32 seed); /* rename from srandom32() */
  u32 prandom_u32(void); /* rename from random32() */
  void prandom_seed_state(struct rnd_state *state, u64 seed);
   /* rename from prandom32_seed() */
  u32 prandom_u32_state(struct rnd_state *state);
   /* rename from prandom32() */

The purpose of this renaming is to prevent some kernel developers from
assuming that prandom32() and random32() might imply that only
prandom32() was the one using a pseudo-random number generator by
prandom32's "p", and the result may be a very embarassing security
exposure.  This concern was expressed by Theodore Ts'o.

And furthermore, I'm going to introduce new functions for getting the
requested number of pseudo-random bytes.  If I continue to use both
prandom32 and random32 prefixes for these functions, the confusion
is getting worse.

As a result of this renaming, "prandom_" is the common prefix for
pseudo-random number library.

Currently, srandom32() and random32() are preserved because it is
difficult to rename too many users at once.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Robert Love <robert.w.love@intel.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: Valdis Kletnieks <valdis.kletnieks@vt.edu>
Cc: David Laight <david.laight@aculab.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Artem Bityutskiy <dedekind1@gmail.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: fix use after free in aoedev_by_aoeaddr()

We should return NULL on failure instead of returning a freed pointer.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: update internal version number to 81

This version number is printed to the console on module initialization
and is available in sysfs, which is where the userland aoe-version tool
looks for it.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: identify source of runt AoE packets

This change only affects experimental AoE storage networks.

It modifies the console message about runt packets detected so that the
AoE major and minor addresses of the AoE target that generated the runt
are mentioned.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: allow comma separator in aoe_iflist value

By default, the aoe driver uses any ethernet interface for AoE, but the
aoe_iflist module parameter provides a convenient way to limit AoE
traffic to a specific list of local network interfaces.

This change allows a list to be specified using the comma character as a
separator.  For example,

  modprobe aoe aoe_iflist=eth2,eth3

Before, it was inconvenient to get the quoting right in shell scripts
when setting aoe_iflist to have more than one network interface.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: allow user to disable target failure timeout

With this change, the aoe driver treats the value zero as special for
the aoe_deadsecs module parameter.  Normally, this value specifies the
number of seconds during which the driver will continue to attempt
retransmits to an unresponsive AoE target.  After aoe_deadsecs has
elapsed, the aoe driver marks the aoe device as "down" and fails all
I/O.

The new meaning of an aoe_deadsecs of zero is for the driver to
retransmit commands indefinitely.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: use dynamic number of remote ports for AoE storage target

Many AoE targets have four or fewer network ports, but some existing
storage devices have many, and the AoE protocol sets no limit.

This patch allows the use of more than eight remote MAC addresses per AoE
target, while reducing the amount of memory used by the aoe driver in
cases where there are many AoE targets with fewer than eight MAC addresses
each.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: avoid races between device destruction and discovery

This change avoids a race that could result in a NULL pointer derference
following a WARNing from kobject_add_internal, "don't try to register
things with the same name in the same directory."

The problem was found with a test that forgets and discovers an
aoe device in a loop:

  while test ! -r /tmp/stop; do
aoe-flush -a
aoe-discover
  done

The race was between aoedev_flush taking aoedevs out of the devlist,
allowing a new discovery of the same AoE target to take place before the
driver gets around to calling sysfs_remove_group.  Fixing that one
revealed another race between do_open and add_disk, and this patch avoids
that, too.

The fix required some care, because for flushing (forgetting) an aoedev,
some of the steps must be performed under lock and some must be able to
sleep.  Also, for discovering a new aoedev, some steps might sleep.

The check for a bad aoedev pointer remains from a time when about half of
this patch was done, and it was possible for the
bdev->bd_disk->private_data to become corrupted.  The check should be
removed eventually, but it is not expected to add significant overhead,
occurring in the aoeblk_open routine.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: improve handling of misbehaving network paths

An AoE target can have multiple network ports used for AoE, and in the
aoe driver, those are tracked by the aoetgt struct.  These changes allow
the aoe driver to handle network paths, or aoetgts, that are not working
well, compared to the others.

Paths that do not get responses despite the retransmission of AoE
commands are marked as "tainted", and non-tainted paths are preferred.

Meanwhile, the aoe driver attempts to "probe" the tainted path in the
background by issuing reads of LBA 0 that are padded out to full
(possibly jumbo-frame) size.  If the probes get responses, then the path
is "redeemed", and its taint is removed.

This mechanism has been shown to be helpful in transparently handling
and recovering from real-world network "brown outs" in ways that the
earlier "shoot the help-needing target in the head" mechanism could not.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: return real minor number for static minors

The value returned by the static minor device number number allocator is
the real minor number, so it must be multiplied by the supported number
of partitions per aoedev.

Without this fix the support for systems without udev is incomplete, and
the few users of aoe on such systems will have surprising results when
device nodes names do not match the AoE target.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: initialize sysminor to avoid compiler warning

Because the minor_get and related functions use the return values for
errors, the compiler doesn't know that sysminor will always either 1) be
initialized in aoedev_by_aoeaddr by the call to minor_get, or 2) be
unused as the "goto out" is executed.

This patch avoids the compiler warning.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: make error messages more specific in static minor allocation

For some special-purpose systems where udev isn't present, static
allocation of minor numbers is desirable.  This update distinguishes
different failure scenarios, to help the user understand what went
wrong.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: remove call to request handler from I/O completion

There is no need to call the request handler function in the I/O
completion routine.  The user impact of not doing it is a more "nice" aoe
driver that is less susceptible to causing soft lockups.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: cleanup: correct comment for aoetgt nout

A misplaced comment was attached to the nout member of the aoetgt.  This
change corrects the comment.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: increase default cap on outstanding AoE commands in the network

The aoe driver will never be waiting for more than aoe_maxout AoE
commands from a given remote network port on an AoE target.  Increasing
the cap increases performance.  Users can tighten the setting to reduce
the amount of memory used for handling AoE traffic or the network
bandwidth used for AoE.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: remove vestigial request queue allocation

Before the aoe driver was an I/O request handler, it was a
make_request-style block driver.  Even so, there was a problem where
sysfs expected a request queue to exist, so one was provided in commit
7135a71b19be ("aoe: allocate unused request_queue for sysfs").

During the transition to the request-handler style, a patch was merged
that was based on a driver without the noop queue, and the noop queue
remained in place after the patch was merged, even though a new
functional queue was introduced by the patch, allocated through
blk_init_queue.

The user impact is a memory leak proportional to the number of AoE
targets discovered.  This patch removes the memory leak and cleans up
vestiges of the old do-nothing queue from the aoeblk_gdalloc function.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: copy fallback timing information on destination failover

Commit f3b8e07af774 ("aoe: commands in retransmit queue use new
destination on failure") omits the copying of the coarse-grained time
when an AoE command was sent during the failover from one destination
MAC address on the AoE target to another.

The coarse-grained timing is only used when the system time changes or
an unlikely length of time has passed since the sending of the AoE
command.  Users will not be impacted unless their system clock is very
inaccurate or something unusual (e.g., 10 GbE link reset) happens during
the period when the aoe driver is handling the failure of a port on the
AoE target.  Being effected will mean that an AoE target could be
considered "down" too eagerly.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: update driver-internal version to 64+

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: commands in retransmit queue use new destination on failure

When one remote MAC address isn't working as a destination for AoE
commands, the frames used to track information associated with the AoE
commands are moved to a new aoetgt (defined by the tuple of {AoE major,
AoE minor, target MAC address}).

This patch makes sure that the frames on the queue for retransmits that
need to be done are updated to use the new destination, so that
retransmits will be sent through a working network path.

Without this change, packets on the retransmit queue will be needlessly
retransmitted to the unresponsive destination MAC, possibly causing
premature target failure before there's time for the retransmit timer to
run again, decide to retransmit again, and finally update the destination
to a working MAC address on the AoE target.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: use high-resolution RTTs with fallback to low-res

These changes improve the accuracy of the decision about whether it's time
to retransmit an AoE command by using the microsecond-resolution
gettimeofday instead of jiffies.

Because the system time can jump suddenly, the decision reverts to using
jiffies if the high-resolution time difference is relatively large.
Otherwise the AoE targets could be considered failed inappropriately.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: manipulate aoedev network stats under lock

With this bugfix in place the calculation of the criterion for "lateness"
is performed under lock.  Without the lock, there is a chance that one of
the non-atomic operations performed on the round trip time statistics
could be incomplete, such that an incorrect lateness criterion would be
calculated.

Without this change, the effect of the bug would be rare unecessary but
benign retransmissions.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: err device: include MAC addresses for unexpected responses

The /dev/etherd/err character device provides low-level information about
normal but sometimes interesting AoE command retransmits and "unexpected
responses", i.e., responses for packets that have already been
retransmitted.

This change adds MAC addresses to the messages about unexpected responses,
so that when they occur, it's more easy to determine the network paths to
which they belong.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: improve network congestion handling

The aoe driver already had some congestion handling, but it was limited in
its ability to cope with the kind of congestion that can arise on more
complex networks such as those involving paths through multiple ethernet
switches.

Some of the lessons from TCP's history of development can be applied to
improving the congestion control and avoidance on AoE storage networks.
These changes use familar concepts from Van Jacobson's "Congestion
Avoidance and Control" paper from '88, without adding significant
overhead.

This patch depends on an upcoming patch that covers the failover case when
AoE commands being retransmitted are transferred from one retransmit queue
to another.  Another upcoming patch increases the timing accuracy.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: provide ATA identify device content to user on request

Make the aoe driver follow expected behavior when the user uses ioctl to
get the ATA device identify information, allowing access to model, serial
number, etc.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: update driver-internal version number to 60

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: whitespace cleanup

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: cleanup: remove unused ata_scnt function

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: "payload" sysfs file exports per-AoE-command data transfer size

The userland aoetools package includes an "aoe-stat" command that can
display a "payload size" column when the aoe driver exports this
information.  Users can quickly see what amount of user data is
transferred inside each AoE command on the network, network headers
excluded.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: support larger I/O requests via aoe_maxsectors module param

The GPFS filesystem is an example of an aoe user that requires the aoe
driver to support I/O request sizes larger than the default.  Most users
will not need large I/O request sizes, because they would need to be split
up into multiple AoE commands anyway.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: support the forgetting (flushing) of a user-specified AoE target

Users sometimes want to cause the aoe driver to forget a particular
previously discovered device when it is no longer online.  The aoetools
provide an "aoe-flush" command that users run to perform this
administrative task.  The changes below provide the support needed in the
driver.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: update cap on outstanding commands based on config query response

The ATA over Ethernet config query response contains a "buffer count"
field reflecting the AoE target's capacity to buffer incoming AoE
commands.

By taking the current value of this field into accound, we increase
performance throughput or avoid network congestion, when the value
has increased or decreased, respectively.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: print warning regarding a common reason for dropped transmits

Dropped transmits are not common, but when they do occur, increasing
the transmit queue length often helps.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

aoe: describe the behavior of the "err" character device

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Documentation/sparse.txt: document context annotations for lock checking

The context feature of sparse is used with the Linux kernel sources to
check for imbalanced uses of locks.  Document the annotations defined in
include/linux/compiler.h that tell sparse what to expect when a lock is
held on function entry, exit, or both.

Signed-off-by: Ed Cashin <ecashin@coraid.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Acked-by: Christopher Li <sparse@chrisli.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

linux/compiler.h: add __must_hold macro for functions called with a lock held

linux/compiler.h has macros to denote functions that acquire or release
locks, but not to denote functions called with a lock held that return
with the lock still held.  Add a __must_hold macro to cover that case.

Signed-off-by: Josh Triplett <josh@joshtriplett.org>
Reported-by: Ed Cashin <ecashin@coraid.com>
Tested-by: Ed Cashin <ecashin@coraid.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

pidns: remove unused is_container_init()

Since commit 1cdcbec1a337 ("CRED: Neuter sys_capset()")
is_container_init() has no callers.

Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Cc: David Howells <dhowells@redhat.com>
Acked-by: Serge Hallyn <serge.hallyn@canonical.com>
Cc: James Morris <jmorris@namei.org>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

exec: use -ELOOP for max recursion depth

To avoid an explosion of request_module calls on a chain of abusive
scripts, fail maximum recursion with -ELOOP instead of -ENOEXEC. As soon
as maximum recursion depth is hit, the error will fail all the way back
up the chain, aborting immediately.

This also has the side-effect of stopping the user's shell from attempting
to reexecute the top-level file as a shell script. As seen in the
dash source:

        if (cmd != path_bshell && errno == ENOEXEC) {
                *argv-- = cmd;
                *argv = cmd = path_bshell;
                goto repeat;
        }

The above logic was designed for running scripts automatically that lacked
the "#!" header, not to re-try failed recursion. On a legitimate -ENOEXEC,
things continue to behave as the shell expects.

Additionally, when tracking recursion, the binfmt handlers should not be
involved. The recursion being tracked is the depth of calls through
search_binary_handler(), so that function should be exclusively responsible
for tracking the depth.

Signed-off-by: Kees Cook <keescook@chromium.org>
Cc: halfdog <me@halfdog.net>
Cc: P J P <ppandit@redhat.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

proc: pid/status: show all supplementary groups

We display a list of supplementary group for each process in
/proc/<pid>/status.  However, we show only the first 32 groups, not all of
them.

Although this is rare, but sometimes processes do have more than 32
supplementary groups, and this kernel limitation breaks user-space apps
that rely on the group list in /proc/<pid>/status.

Number 32 comes from the internal NGROUPS_SMALL macro which defines the
length for the internal kernel "small" groups buffer.  There is no
apparent reason to limit to this value.

This patch removes the 32 groups printing limit.

The Linux kernel limits the amount of supplementary groups by NGROUPS_MAX,
which is currently set to 65536.  And this is the maximum count of groups
we may possibly print.

Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

/proc/pid/status: add "Seccomp" field

It is currently impossible to examine the state of seccomp for a given
process.  While attaching with gdb and attempting "call
prctl(PR_GET_SECCOMP,...)" will work with some situations, it is not
reliable.  If the process is in seccomp mode 1, this query will kill the
process (prctl not allowed), if the process is in mode 2 with prctl not
allowed, it will similarly be killed, and in weird cases, if prctl is
filtered to return errno 0, it can look like seccomp is disabled.

When reviewing the state of running processes, there should be a way to
externally examine the seccomp mode.  ("Did this build of Chrome end up
using seccomp?" "Did my distro ship ssh with seccomp enabled?")

This adds the "Seccomp" line to /proc/$pid/status.

Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: James Morris <jmorris@namei.org>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

procfs: add VmFlags field in smaps output

During c/r sessions we've found that there is no way at the moment to
fetch some VMA associated flags, such as mlock() and madvise().

This leads us to a problem -- we don't know if we should call for mlock()
and/or madvise() after restore on the vma area we're bringing back to
life.

This patch intorduces a new field into "smaps" output called VmFlags,
where all set flags associated with the particular VMA is shown as two
letter mnemonics.

[ Strictly speaking for c/r we only need mlock/madvise bits but it has been
  said that providing just a few flags looks somehow inconsistent.  So all
  flags are here now. ]

This feature is made available on CONFIG_CHECKPOINT_RESTORE=n kernels, as
other applications may start to use these fields.

The data is encoded in a somewhat awkward two letters mnemonic form, to
encourage userspace to be prepared for fields being added or removed in
the future.

[a.p.zijlstra@chello.nl: props to use for_each_set_bit]
[sfr@canb.auug.org.au: props to use array instead of struct]
[akpm@linux-foundation.org: overall redesign and simplification]
[akpm@linux-foundation.org: remove unneeded braces per sfr, avoid using bloaty for_each_set_bit()]
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

proc: don't show nonexistent capabilities

Without this patch it is really hard to interpret a bounding set, if
CAP_LAST_CAP is unknown for a current kernel.

Non-existant capabilities can not be deleted from a bounding set with help
of prctl.

E.g.: Here are two examples without/with this patch.

  CapBnd: ffffffe0fdecffff
  CapBnd: 00000000fdecffff

I suggest to hide non-existent capabilities. Here is two reasons.
* It's logically and easier for using.
* It helps to checkpoint-restore capabilities of tasks, because tasks
can be restored on another kernel, where CAP_LAST_CAP is bigger.

Signed-off-by: Andrew Vagin <avagin@openvz.org>
Cc: Andrew G. Morgan <morgan@kernel.org>
Reviewed-by: Serge E. Hallyn <serge.hallyn@canonical.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: James Morris <jmorris@namei.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>