linux.git/mm/page-writeback.c, branch v4.9.209

mm/page-writeback.c: fix range_cyclic writeback vs writepages deadlock

2019-11-28T17:28:42+00:00

[ Upstream commit 64081362e8ff4587b4554087f3cfc73d3e0a4cd7 ]

We've recently seen a workload on XFS filesystems with a repeatable
deadlock between background writeback and a multi-process application
doing concurrent writes and fsyncs to a small range of a file.

range_cyclic
writeback		Process 1		Process 2

xfs_vm_writepages
  write_cache_pages
    writeback_index = 2
    cycled = 0
    ....
    find page 2 dirty
    lock Page 2
    ->writepage
      page 2 writeback
      page 2 clean
      page 2 added to bio
    no more pages
			write()
			locks page 1
			dirties page 1
			locks page 2
			dirties page 1
			fsync()
			....
			xfs_vm_writepages
			write_cache_pages
			  start index 0
			  find page 1 towrite
			  lock Page 1
			  ->writepage
			    page 1 writeback
			    page 1 clean
			    page 1 added to bio
			  find page 2 towrite
			  lock Page 2
			  page 2 is writeback
			  
						write()
						locks page 1
						dirties page 1
						fsync()
						....
						xfs_vm_writepages
						write_cache_pages
						  start index 0

    !done && !cycled
      sets index to 0, restarts lookup
    find page 1 dirty
						  find page 1 towrite
						  lock Page 1
						  page 1 is writeback
						  

    lock Page 1
    

DEADLOCK because:

	- process 1 needs page 2 writeback to complete to make
	  enough progress to issue IO pending for page 1
	- writeback needs page 1 writeback to complete so process 2
	  can progress and unlock the page it is blocked on, then it
	  can issue the IO pending for page 2
	- process 2 can't make progress until process 1 issues IO
	  for page 1

The underlying cause of the problem here is that range_cyclic writeback is
processing pages in descending index order as we hold higher index pages
in a structure controlled from above write_cache_pages().  The
write_cache_pages() caller needs to be able to submit these pages for IO
before write_cache_pages restarts writeback at mapping index 0 to avoid
wcp inverting the page lock/writeback wait order.

generic_writepages() is not susceptible to this bug as it has no private
context held across write_cache_pages() - filesystems using this
infrastructure always submit pages in ->writepage immediately and so there
is no problem with range_cyclic going back to mapping index 0.

However:
	mpage_writepages() has a private bio context,
	exofs_writepages() has page_collect
	fuse_writepages() has fuse_fill_wb_data
	nfs_writepages() has nfs_pageio_descriptor
	xfs_vm_writepages() has xfs_writepage_ctx

All of these ->writepages implementations can hold pages under writeback
in their private structures until write_cache_pages() returns, and hence
they are all susceptible to this deadlock.

Also worth noting is that ext4 has it's own bastardised version of
write_cache_pages() and so it /may/ have an equivalent deadlock.  I looked
at the code long enough to understand that it has a similar retry loop for
range_cyclic writeback reaching the end of the file and then promptly ran
away before my eyes bled too much.  I'll leave it for the ext4 developers
to determine if their code is actually has this deadlock and how to fix it
if it has.

There's a few ways I can see avoid this deadlock.  There's probably more,
but these are the first I've though of:

1. get rid of range_cyclic altogether

2. range_cyclic always stops at EOF, and we start again from
writeback index 0 on the next call into write_cache_pages()

2a. wcp also returns EAGAIN to ->writepages implementations to
indicate range cyclic has hit EOF. writepages implementations can
then flush the current context and call wpc again to continue. i.e.
lift the retry into the ->writepages implementation

3. range_cyclic uses trylock_page() rather than lock_page(), and it
skips pages it can't lock without blocking. It will already do this
for pages under writeback, so this seems like a no-brainer

3a. all non-WB_SYNC_ALL writeback uses trylock_page() to avoid
blocking as per pages under writeback.

I don't think #1 is an option - range_cyclic prevents frequently
dirtied lower file offset from starving background writeback of
rarely touched higher file offsets.

#2 is simple, and I don't think it will have any impact on
performance as going back to the start of the file implies an
immediate seek. We'll have exactly the same number of seeks if we
switch writeback to another inode, and then come back to this one
later and restart from index 0.

#2a is pretty much "status quo without the deadlock". Moving the
retry loop up into the wcp caller means we can issue IO on the
pending pages before calling wcp again, and so avoid locking or
waiting on pages in the wrong order. I'm not convinced we need to do
this given that we get the same thing from #2 on the next writeback
call from the writeback infrastructure.

#3 is really just a band-aid - it doesn't fix the access/wait
inversion problem, just prevents it from becoming a deadlock
situation. I'd prefer we fix the inversion, not sweep it under the
carpet like this.

#3a is really an optimisation that just so happens to include the
band-aid fix of #3.

So it seems that the simplest way to fix this issue is to implement
solution #2

Link: http://lkml.kernel.org/r/20181005054526.21507-1-david@fromorbit.com
Signed-off-by: Dave Chinner 
Reviewed-by: Jan Kara 
Cc: Nicholas Piggin 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Sasha Levin

mm/page-writeback.c: don't break integrity writeback on ->writepage() error

2019-01-26T08:38:35+00:00

[ Upstream commit 3fa750dcf29e8606e3969d13d8e188cc1c0f511d ]

write_cache_pages() is used in both background and integrity writeback
scenarios by various filesystems.  Background writeback is mostly
concerned with cleaning a certain number of dirty pages based on various
mm heuristics.  It may not write the full set of dirty pages or wait for
I/O to complete.  Integrity writeback is responsible for persisting a set
of dirty pages before the writeback job completes.  For example, an
fsync() call must perform integrity writeback to ensure data is on disk
before the call returns.

write_cache_pages() unconditionally breaks out of its processing loop in
the event of a ->writepage() error.  This is fine for background
writeback, which had no strict requirements and will eventually come
around again.  This can cause problems for integrity writeback on
filesystems that might need to clean up state associated with failed page
writeouts.  For example, XFS performs internal delayed allocation
accounting before returning a ->writepage() error, where applicable.  If
the current writeback happens to be associated with an unmount and
write_cache_pages() completes the writeback prematurely due to error, the
filesystem is unmounted in an inconsistent state if dirty+delalloc pages
still exist.

To handle this problem, update write_cache_pages() to always process the
full set of pages for integrity writeback regardless of ->writepage()
errors.  Save the first encountered error and return it to the caller once
complete.  This facilitates XFS (or any other fs that expects integrity
writeback to process the entire set of dirty pages) to clean up its
internal state completely in the event of persistent mapping errors.
Background writeback continues to exit on the first error encountered.

[akpm@linux-foundation.org: fix typo in comment]
Link: http://lkml.kernel.org/r/20181116134304.32440-1-bfoster@redhat.com
Signed-off-by: Brian Foster 
Reviewed-by: Jan Kara 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Sasha Levin

writeback: safer lock nesting

2018-04-24T07:34:18+00:00

commit 2e898e4c0a3897ccd434adac5abb8330194f527b upstream.

lock_page_memcg()/unlock_page_memcg() use spin_lock_irqsave/restore() if
the page's memcg is undergoing move accounting, which occurs when a
process leaves its memcg for a new one that has
memory.move_charge_at_immigrate set.

unlocked_inode_to_wb_begin,end() use spin_lock_irq/spin_unlock_irq() if
the given inode is switching writeback domains.  Switches occur when
enough writes are issued from a new domain.

This existing pattern is thus suspicious:
    lock_page_memcg(page);
    unlocked_inode_to_wb_begin(inode, &locked);
    ...
    unlocked_inode_to_wb_end(inode, locked);
    unlock_page_memcg(page);

If both inode switch and process memcg migration are both in-flight then
unlocked_inode_to_wb_end() will unconditionally enable interrupts while
still holding the lock_page_memcg() irq spinlock.  This suggests the
possibility of deadlock if an interrupt occurs before unlock_page_memcg().

    truncate
    __cancel_dirty_page
    lock_page_memcg
    unlocked_inode_to_wb_begin
    unlocked_inode_to_wb_end
    
                                    
                                    end_page_writeback
                                    test_clear_page_writeback
                                    lock_page_memcg
                                    
    unlock_page_memcg

Due to configuration limitations this deadlock is not currently possible
because we don't mix cgroup writeback (a cgroupv2 feature) and
memory.move_charge_at_immigrate (a cgroupv1 feature).

If the kernel is hacked to always claim inode switching and memcg
moving_account, then this script triggers lockup in less than a minute:

  cd /mnt/cgroup/memory
  mkdir a b
  echo 1 > a/memory.move_charge_at_immigrate
  echo 1 > b/memory.move_charge_at_immigrate
  (
    echo $BASHPID > a/cgroup.procs
    while true; do
      dd if=/dev/zero of=/mnt/big bs=1M count=256
    done
  ) &
  while true; do
    sync
  done &
  sleep 1h &
  SLEEP=$!
  while true; do
    echo $SLEEP > a/cgroup.procs
    echo $SLEEP > b/cgroup.procs
  done

The deadlock does not seem possible, so it's debatable if there's any
reason to modify the kernel.  I suggest we should to prevent future
surprises.  And Wang Long said "this deadlock occurs three times in our
environment", so there's more reason to apply this, even to stable.
Stable 4.4 has minor conflicts applying this patch.  For a clean 4.4 patch
see "[PATCH for-4.4] writeback: safer lock nesting"
https://lkml.org/lkml/2018/4/11/146

Wang Long said "this deadlock occurs three times in our environment"

[gthelen@google.com: v4]
  Link: http://lkml.kernel.org/r/20180411084653.254724-1-gthelen@google.com
[akpm@linux-foundation.org: comment tweaks, struct initialization simplification]
Change-Id: Ibb773e8045852978f6207074491d262f1b3fb613
Link: http://lkml.kernel.org/r/20180410005908.167976-1-gthelen@google.com
Fixes: 682aa8e1a6a1 ("writeback: implement unlocked_inode_to_wb transaction and use it for stat updates")
Signed-off-by: Greg Thelen 
Reported-by: Wang Long 
Acked-by: Wang Long 
Acked-by: Michal Hocko 
Reviewed-by: Andrew Morton 
Cc: Johannes Weiner 
Cc: Tejun Heo 
Cc: Nicholas Piggin 
Cc: 	[v4.2+]
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
[natechancellor: Adjust context due to lack of b93b016313b3b]
Signed-off-by: Nathan Chancellor 
Signed-off-by: Greg Kroah-Hartman

mm: don't use radix tree writeback tags for pages in swap cache

2016-10-08T01:46:28+00:00

File pages use a set of radix tree tags (DIRTY, TOWRITE, WRITEBACK,
etc.) to accelerate finding the pages with a specific tag in the radix
tree during inode writeback.  But for anonymous pages in the swap cache,
there is no inode writeback.  So there is no need to find the pages with
some writeback tags in the radix tree.  It is not necessary to touch
radix tree writeback tags for pages in the swap cache.

Per Rik van Riel's suggestion, a new flag AS_NO_WRITEBACK_TAGS is
introduced for address spaces which don't need to update the writeback
tags.  The flag is set for swap caches.  It may be used for DAX file
systems, etc.

With this patch, the swap out bandwidth improved 22.3% (from ~1.2GB/s to
~1.48GBps) in the vm-scalability swap-w-seq test case with 8 processes.
The test is done on a Xeon E5 v3 system.  The swap device used is a RAM
simulated PMEM (persistent memory) device.  The improvement comes from
the reduced contention on the swap cache radix tree lock.  To test
sequential swapping out, the test case uses 8 processes, which
sequentially allocate and write to the anonymous pages until RAM and
part of the swap device is used up.

Details of comparison is as follow,

base             base+patch
---------------- --------------------------
         %stddev     %change         %stddev
             \          |                \
   2506952 Â±  2%     +28.1%    3212076 Â±  7%  vm-scalability.throughput
   1207402 Â±  7%     +22.3%    1476578 Â±  6%  vmstat.swap.so
     10.86 Â± 12%     -23.4%       8.31 Â± 16%  perf-profile.cycles-pp._raw_spin_lock_irq.__add_to_swap_cache.add_to_swap_cache.add_to_swap.shrink_page_list
     10.82 Â± 13%     -33.1%       7.24 Â± 14%  perf-profile.cycles-pp._raw_spin_lock_irqsave.__remove_mapping.shrink_page_list.shrink_inactive_list.shrink_zone_memcg
     10.36 Â± 11%    -100.0%       0.00 Â± -1%  perf-profile.cycles-pp._raw_spin_lock_irqsave.__test_set_page_writeback.bdev_write_page.__swap_writepage.swap_writepage
     10.52 Â± 12%    -100.0%       0.00 Â± -1%  perf-profile.cycles-pp._raw_spin_lock_irqsave.test_clear_page_writeback.end_page_writeback.page_endio.pmem_rw_page

Link: http://lkml.kernel.org/r/1472578089-5560-1-git-send-email-ying.huang@intel.com
Signed-off-by: "Huang, Ying" 
Acked-by: Rik van Riel 
Cc: Hugh Dickins 
Cc: Shaohua Li 
Cc: Minchan Kim 
Cc: Mel Gorman 
Cc: Tejun Heo 
Cc: Wu Fengguang 
Cc: Dave Hansen 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mm, vmscan: get rid of throttle_vm_writeout

2016-10-08T01:46:27+00:00

throttle_vm_writeout() was introduced back in 2005 to fix OOMs caused by
excessive pageout activity during the reclaim.  Too many pages could be
put under writeback therefore LRUs would be full of unreclaimable pages
until the IO completes and in turn the OOM killer could be invoked.

There have been some important changes introduced since then in the
reclaim path though.  Writers are throttled by balance_dirty_pages when
initiating the buffered IO and later during the memory pressure, the
direct reclaim is throttled by wait_iff_congested if the node is
considered congested by dirty pages on LRUs and the underlying bdi is
congested by the queued IO.  The kswapd is throttled as well if it
encounters pages marked for immediate reclaim or under writeback which
signals that that there are too many pages under writeback already.
Finally should_reclaim_retry does congestion_wait if the reclaim cannot
make any progress and there are too many dirty/writeback pages.

Another important aspect is that we do not issue any IO from the direct
reclaim context anymore.  In a heavy parallel load this could queue a
lot of IO which would be very scattered and thus unefficient which would
just make the problem worse.

This three mechanisms should throttle and keep the amount of IO in a
steady state even under heavy IO and memory pressure so yet another
throttling point doesn't really seem helpful.  Quite contrary, Mikulas
Patocka has reported that swap backed by dm-crypt doesn't work properly
because the swapout IO cannot make sufficient progress as the writeout
path depends on dm_crypt worker which has to allocate memory to perform
the encryption.  In order to guarantee a forward progress it relies on
the mempool allocator.  mempool_alloc(), however, prefers to use the
underlying (usually page) allocator before it grabs objects from the
pool.  Such an allocation can dive into the memory reclaim and
consequently to throttle_vm_writeout.  If there are too many dirty or
pages under writeback it will get throttled even though it is in fact a
flusher to clear pending pages.

  kworker/u4:0    D ffff88003df7f438 10488     6      2	0x00000000
  Workqueue: kcryptd kcryptd_crypt [dm_crypt]
  Call Trace:
    schedule+0x3c/0x90
    schedule_timeout+0x1d8/0x360
    io_schedule_timeout+0xa4/0x110
    congestion_wait+0x86/0x1f0
    throttle_vm_writeout+0x44/0xd0
    shrink_zone_memcg+0x613/0x720
    shrink_zone+0xe0/0x300
    do_try_to_free_pages+0x1ad/0x450
    try_to_free_pages+0xef/0x300
    __alloc_pages_nodemask+0x879/0x1210
    alloc_pages_current+0xa1/0x1f0
    new_slab+0x2d7/0x6a0
    ___slab_alloc+0x3fb/0x5c0
    __slab_alloc+0x51/0x90
    kmem_cache_alloc+0x27b/0x310
    mempool_alloc_slab+0x1d/0x30
    mempool_alloc+0x91/0x230
    bio_alloc_bioset+0xbd/0x260
    kcryptd_crypt+0x114/0x3b0 [dm_crypt]

Let's just drop throttle_vm_writeout altogether.  It is not very much
helpful anymore.

I have tried to test a potential writeback IO runaway similar to the one
described in the original patch which has introduced that [1].  Small
virtual machine (512MB RAM, 4 CPUs, 2G of swap space and disk image on a
rather slow NFS in a sync mode on the host) with 8 parallel writers each
writing 1G worth of data.  As soon as the pagecache fills up and the
direct reclaim hits then I start anon memory consumer in a loop
(allocating 300M and exiting after populating it) in the background to
make the memory pressure even stronger as well as to disrupt the steady
state for the IO.  The direct reclaim is throttled because of the
congestion as well as kswapd hitting congestion_wait due to nr_immediate
but throttle_vm_writeout doesn't ever trigger the sleep throughout the
test.  Dirty+writeback are close to nr_dirty_threshold with some
fluctuations caused by the anon consumer.

[1] https://www2.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.9-rc1/2.6.9-rc1-mm3/broken-out/vm-pageout-throttling.patch
Link: http://lkml.kernel.org/r/1471171473-21418-1-git-send-email-mhocko@kernel.org
Signed-off-by: Michal Hocko 
Reported-by: Mikulas Patocka 
Cc: Marcelo Tosatti 
Cc: NeilBrown 
Cc: Ondrej Kozina 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mm/writeback: Convert to hotplug state machine

2016-09-06T16:30:20+00:00

Install the callbacks via the state machine and let the core invoke
the callbacks on the already online CPUs.

Signed-off-by: Sebastian Andrzej Siewior 
Cc: Peter Zijlstra 
Cc: Jens Axboe 
Cc: linux-mm@kvack.org
Cc: rt@linutronix.de
Cc: Tejun Heo 
Link: http://lkml.kernel.org/r/20160818125731.27256-6-bigeasy@linutronix.de
Signed-off-by: Thomas Gleixner

mm: remove reclaim and compaction retry approximations

2016-07-28T23:07:41+00:00

If per-zone LRU accounting is available then there is no point
approximating whether reclaim and compaction should retry based on pgdat
statistics.  This is effectively a revert of "mm, vmstat: remove zone
and node double accounting by approximating retries" with the difference
that inactive/active stats are still available.  This preserves the
history of why the approximation was retried and why it had to be
reverted to handle OOM kills on 32-bit systems.

Link: http://lkml.kernel.org/r/1469110261-7365-4-git-send-email-mgorman@techsingularity.net
Signed-off-by: Mel Gorman 
Acked-by: Johannes Weiner 
Acked-by: Minchan Kim 
Cc: Michal Hocko 
Cc: Vlastimil Babka 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mm, vmscan: remove highmem_file_pages

2016-07-28T23:07:41+00:00

With the reintroduction of per-zone LRU stats, highmem_file_pages is
redundant so remove it.

[mgorman@techsingularity.net: wrong stat is being accumulated in highmem_dirtyable_memory]
  Link: http://lkml.kernel.org/r/20160725092324.GM10438@techsingularity.netLink: http://lkml.kernel.org/r/1469110261-7365-3-git-send-email-mgorman@techsingularity.net
Signed-off-by: Mel Gorman 
Acked-by: Johannes Weiner 
Cc: Minchan Kim 
Cc: Michal Hocko 
Cc: Vlastimil Babka 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mm, page_alloc: fix dirtyable highmem calculation

2016-07-28T23:07:41+00:00

When I tested vmscale in mmtest in 32bit, I found the benchmark was slow
down 0.5 times.

                base        node
                   1    global-1
User           12.98       16.04
System        147.61      166.42
Elapsed        26.48       38.08

With vmstat, I found IO wait avg is much increased compared to base.

The reason was highmem_dirtyable_memory accumulates free pages and
highmem_file_pages from HIGHMEM to MOVABLE zones which was wrong.  With
that, dirth_thresh in throtlle_vm_write is always 0 so that it calls
congestion_wait frequently if writeback starts.

With this patch, it is much recovered.

                base        node          fi
                   1    global-1         fix
User           12.98       16.04       13.78
System        147.61      166.42      143.92
Elapsed        26.48       38.08       29.64

Link: http://lkml.kernel.org/r/1468404004-5085-4-git-send-email-mgorman@techsingularity.net
Signed-off-by: Minchan Kim 
Signed-off-by: Mel Gorman 
Acked-by: Johannes Weiner 
Acked-by: Vlastimil Babka 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

mm, vmstat: remove zone and node double accounting by approximating retries

2016-07-28T23:07:41+00:00

The number of LRU pages, dirty pages and writeback pages must be
accounted for on both zones and nodes because of the reclaim retry
logic, compaction retry logic and highmem calculations all depending on
per-zone stats.

Many lowmem allocations are immune from OOM kill due to a check in
__alloc_pages_may_oom for (ac->high_zoneidx < ZONE_NORMAL) since commit
03668b3ceb0c ("oom: avoid oom killer for lowmem allocations").  The
exception is costly high-order allocations or allocations that cannot
fail.  If the __alloc_pages_may_oom avoids OOM-kill for low-order lowmem
allocations then it would fall through to __alloc_pages_direct_compact.

This patch will blindly retry reclaim for zone-constrained allocations
in should_reclaim_retry up to MAX_RECLAIM_RETRIES.  This is not ideal
but without per-zone stats there are not many alternatives.  The impact
it that zone-constrained allocations may delay before considering the
OOM killer.

As there is no guarantee enough memory can ever be freed to satisfy
compaction, this patch avoids retrying compaction for zone-contrained
allocations.

In combination, that means that the per-node stats can be used when
deciding whether to continue reclaim using a rough approximation.  While
it is possible this will make the wrong decision on occasion, it will
not infinite loop as the number of reclaim attempts is capped by
MAX_RECLAIM_RETRIES.

The final step is calculating the number of dirtyable highmem pages.  As
those calculations only care about the global count of file pages in
highmem.  This patch uses a global counter used instead of per-zone
stats as it is sufficient.

In combination, this allows the per-zone LRU and dirty state counters to
be removed.

[mgorman@techsingularity.net: fix acct_highmem_file_pages()]
  Link: http://lkml.kernel.org/r/1468853426-12858-4-git-send-email-mgorman@techsingularity.netLink: http://lkml.kernel.org/r/1467970510-21195-35-git-send-email-mgorman@techsingularity.net
Signed-off-by: Mel Gorman 
Suggested by: Michal Hocko 
Acked-by: Hillf Danton 
Cc: Johannes Weiner 
Cc: Joonsoo Kim 
Cc: Michal Hocko 
Cc: Minchan Kim 
Cc: Rik van Riel 
Cc: Vlastimil Babka 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds