linux.git/mm/filemap.c, branch v4.13

page waitqueue: always add new entries at the end

2017-08-28T23:45:40+00:00

Commit 3510ca20ece0 ("Minor page waitqueue cleanups") made the page
queue code always add new waiters to the back of the queue, which helps
upcoming patches to batch the wakeups for some horrid loads where the
wait queues grow to thousands of entries.

However, I forgot about the nasrt add_page_wait_queue() special case
code that is only used by the cachefiles code.  That one still continued
to add the new wait queue entries at the beginning of the list.

Fix it, because any sane batched wakeup will require that we don't
suddenly start getting new entries at the beginning of the list that we
already handled in a previous batch.

[ The current code always does the whole list while holding the lock, so
  wait queue ordering doesn't matter for correctness, but even then it's
  better to add later entries at the end from a fairness standpoint ]

Signed-off-by: Linus Torvalds

Avoid page waitqueue race leaving possible page locker waiting

2017-08-27T23:25:09+00:00

The "lock_page_killable()" function waits for exclusive access to the
page lock bit using the WQ_FLAG_EXCLUSIVE bit in the waitqueue entry
set.

That means that if it gets woken up, other waiters may have been
skipped.

That, in turn, means that if it sees the page being unlocked, it *must*
take that lock and return success, even if a lethal signal is also
pending.

So instead of checking for lethal signals first, we need to check for
them after we've checked the actual bit that we were waiting for.  Even
if that might then delay the killing of the process.

This matches the order of the old "wait_on_bit_lock()" infrastructure
that the page locking used to use (and is still used in a few other
areas).

Note that if we still return an error after having unsuccessfully tried
to acquire the page lock, that is ok: that means that some other thread
was able to get ahead of us and lock the page, and when that other
thread then unlocks the page, the wakeup event will be repeated.  So any
other pending waiters will now get properly woken up.

Fixes: 62906027091f ("mm: add PageWaiters indicating tasks are waiting for a page bit")
Cc: Nick Piggin 
Cc: Peter Zijlstra 
Cc: Mel Gorman 
Cc: Jan Kara 
Cc: Davidlohr Bueso 
Cc: Andi Kleen 
Signed-off-by: Linus Torvalds

Minor page waitqueue cleanups

2017-08-27T20:55:12+00:00

Tim Chen and Kan Liang have been battling a customer load that shows
extremely long page wakeup lists.  The cause seems to be constant NUMA
migration of a hot page that is shared across a lot of threads, but the
actual root cause for the exact behavior has not been found.

Tim has a patch that batches the wait list traversal at wakeup time, so
that we at least don't get long uninterruptible cases where we traverse
and wake up thousands of processes and get nasty latency spikes.  That
is likely 4.14 material, but we're still discussing the page waitqueue
specific parts of it.

In the meantime, I've tried to look at making the page wait queues less
expensive, and failing miserably.  If you have thousands of threads
waiting for the same page, it will be painful.  We'll need to try to
figure out the NUMA balancing issue some day, in addition to avoiding
the excessive spinlock hold times.

That said, having tried to rewrite the page wait queues, I can at least
fix up some of the braindamage in the current situation. In particular:

 (a) we don't want to continue walking the page wait list if the bit
     we're waiting for already got set again (which seems to be one of
     the patterns of the bad load).  That makes no progress and just
     causes pointless cache pollution chasing the pointers.

 (b) we don't want to put the non-locking waiters always on the front of
     the queue, and the locking waiters always on the back.  Not only is
     that unfair, it means that we wake up thousands of reading threads
     that will just end up being blocked by the writer later anyway.

Also add a comment about the layout of 'struct wait_page_key' - there is
an external user of it in the cachefiles code that means that it has to
match the layout of 'struct wait_bit_key' in the two first members.  It
so happens to match, because 'struct page *' and 'unsigned long *' end
up having the same values simply because the page flags are the first
member in struct page.

Cc: Tim Chen 
Cc: Kan Liang 
Cc: Mel Gorman 
Cc: Christopher Lameter 
Cc: Andi Kleen 
Cc: Davidlohr Bueso 
Cc: Peter Zijlstra 
Signed-off-by: Linus Torvalds

mm: hugetlb: return immediately for hugetlb page in __delete_from_page_cache()

2017-07-10T23:32:30+00:00

We avoid calling __mod_node_page_state(NR_FILE_PAGES) for hugetlb page
now, but it's not enough because later code doesn't handle hugetlb
properly.  Actually in our testing, WARN_ON_ONCE(PageDirty(page)) at the
end of this function fires for hugetlb, which makes no sense.  So we
should return immediately for hugetlb pages.

Link: http://lkml.kernel.org/r/1496305019-5493-3-git-send-email-n-horiguchi@ah.jp.nec.com
Signed-off-by: Naoya Horiguchi 
Cc: Michal Hocko 
Cc: "Aneesh Kumar K.V" 
Cc: Anshuman Khandual 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

Merge tag 'for-linus-v4.13-2' of git://git.kernel.org/pub/scm/linux/kernel/git/jlayton/linux

2017-07-08T02:38:17+00:00

Pull Writeback error handling updates from Jeff Layton:
 "This pile represents the bulk of the writeback error handling fixes
  that I have for this cycle. Some of the earlier patches in this pile
  may look trivial but they are prerequisites for later patches in the
  series.

  The aim of this set is to improve how we track and report writeback
  errors to userland. Most applications that care about data integrity
  will periodically call fsync/fdatasync/msync to ensure that their
  writes have made it to the backing store.

  For a very long time, we have tracked writeback errors using two flags
  in the address_space: AS_EIO and AS_ENOSPC. Those flags are set when a
  writeback error occurs (via mapping_set_error) and are cleared as a
  side-effect of filemap_check_errors (as you noted yesterday). This
  model really sucks for userland.

  Only the first task to call fsync (or msync or fdatasync) will see the
  error. Any subsequent task calling fsync on a file will get back 0
  (unless another writeback error occurs in the interim). If I have
  several tasks writing to a file and calling fsync to ensure that their
  writes got stored, then I need to have them coordinate with one
  another. That's difficult enough, but in a world of containerized
  setups that coordination may even not be possible.

  But wait...it gets worse!

  The calls to filemap_check_errors can be buried pretty far down in the
  call stack, and there are internal callers of filemap_write_and_wait
  and the like that also end up clearing those errors. Many of those
  callers ignore the error return from that function or return it to
  userland at nonsensical times (e.g. truncate() or stat()). If I get
  back -EIO on a truncate, there is no reason to think that it was
  because some previous writeback failed, and a subsequent fsync() will
  (incorrectly) return 0.

  This pile aims to do three things:

   1) ensure that when a writeback error occurs that that error will be
      reported to userland on a subsequent fsync/fdatasync/msync call,
      regardless of what internal callers are doing

   2) report writeback errors on all file descriptions that were open at
      the time that the error occurred. This is a user-visible change,
      but I think most applications are written to assume this behavior
      anyway. Those that aren't are unlikely to be hurt by it.

   3) document what filesystems should do when there is a writeback
      error. Today, there is very little consistency between them, and a
      lot of cargo-cult copying. We need to make it very clear what
      filesystems should do in this situation.

  To achieve this, the set adds a new data type (errseq_t) and then
  builds new writeback error tracking infrastructure around that. Once
  all of that is in place, we change the filesystems to use the new
  infrastructure for reporting wb errors to userland.

  Note that this is just the initial foray into cleaning up this mess.
  There is a lot of work remaining here:

   1) convert the rest of the filesystems in a similar fashion. Once the
      initial set is in, then I think most other fs' will be fairly
      simple to convert. Hopefully most of those can in via individual
      filesystem trees.

   2) convert internal waiters on writeback to use errseq_t for
      detecting errors instead of relying on the AS_* flags. I have some
      draft patches for this for ext4, but they are not quite ready for
      prime time yet.

  This was a discussion topic this year at LSF/MM too. If you're
  interested in the gory details, LWN has some good articles about this:

      https://lwn.net/Articles/718734/
      https://lwn.net/Articles/724307/"

* tag 'for-linus-v4.13-2' of git://git.kernel.org/pub/scm/linux/kernel/git/jlayton/linux:
  btrfs: minimal conversion to errseq_t writeback error reporting on fsync
  xfs: minimal conversion to errseq_t writeback error reporting
  ext4: use errseq_t based error handling for reporting data writeback errors
  fs: convert __generic_file_fsync to use errseq_t based reporting
  block: convert to errseq_t based writeback error tracking
  dax: set errors in mapping when writeback fails
  Documentation: flesh out the section in vfs.txt on storing and reporting writeback errors
  mm: set both AS_EIO/AS_ENOSPC and errseq_t in mapping_set_error
  fs: new infrastructure for writeback error handling and reporting
  lib: add errseq_t type and infrastructure for handling it
  mm: don't TestClearPageError in __filemap_fdatawait_range
  mm: clear AS_EIO/AS_ENOSPC when writeback initiation fails
  jbd2: don't clear and reset errors after waiting on writeback
  buffer: set errors in mapping at the time that the error occurs
  fs: check for writeback errors after syncing out buffers in generic_file_fsync
  buffer: use mapping_set_error instead of setting the flag
  mm: fix mapping_set_error call in me_pagecache_dirty

mm: per-cgroup memory reclaim stats

2017-07-06T23:24:35+00:00

Track the following reclaim counters for every memory cgroup: PGREFILL,
PGSCAN, PGSTEAL, PGACTIVATE, PGDEACTIVATE, PGLAZYFREE and PGLAZYFREED.

These values are exposed using the memory.stats interface of cgroup v2.

The meaning of each value is the same as for global counters, available
using /proc/vmstat.

Also, for consistency, rename mem_cgroup_count_vm_event() to
count_memcg_event_mm().

Link: http://lkml.kernel.org/r/1494530183-30808-1-git-send-email-guro@fb.com
Signed-off-by: Roman Gushchin 
Suggested-by: Johannes Weiner 
Acked-by: Michal Hocko 
Acked-by: Vladimir Davydov 
Acked-by: Johannes Weiner 
Cc: Tejun Heo 
Cc: Li Zefan 
Cc: Balbir Singh 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds

fs: new infrastructure for writeback error handling and reporting

2017-07-06T11:02:25+00:00

Most filesystems currently use mapping_set_error and
filemap_check_errors for setting and reporting/clearing writeback errors
at the mapping level. filemap_check_errors is indirectly called from
most of the filemap_fdatawait_* functions and from
filemap_write_and_wait*. These functions are called from all sorts of
contexts to wait on writeback to finish -- e.g. mostly in fsync, but
also in truncate calls, getattr, etc.

The non-fsync callers are problematic. We should be reporting writeback
errors during fsync, but many places spread over the tree clear out
errors before they can be properly reported, or report errors at
nonsensical times.

If I get -EIO on a stat() call, there is no reason for me to assume that
it is because some previous writeback failed. The fact that it also
clears out the error such that a subsequent fsync returns 0 is a bug,
and a nasty one since that's potentially silent data corruption.

This patch adds a small bit of new infrastructure for setting and
reporting errors during address_space writeback. While the above was my
original impetus for adding this, I think it's also the case that
current fsync semantics are just problematic for userland. Most
applications that call fsync do so to ensure that the data they wrote
has hit the backing store.

In the case where there are multiple writers to the file at the same
time, this is really hard to determine. The first one to call fsync will
see any stored error, and the rest get back 0. The processes with open
fds may not be associated with one another in any way. They could even
be in different containers, so ensuring coordination between all fsync
callers is not really an option.

One way to remedy this would be to track what file descriptor was used
to dirty the file, but that's rather cumbersome and would likely be
slow. However, there is a simpler way to improve the semantics here
without incurring too much overhead.

This set adds an errseq_t to struct address_space, and a corresponding
one is added to struct file. Writeback errors are recorded in the
mapping's errseq_t, and the one in struct file is used as the "since"
value.

This changes the semantics of the Linux fsync implementation such that
applications can now use it to determine whether there were any
writeback errors since fsync(fd) was last called (or since the file was
opened in the case of fsync having never been called).

Note that those writeback errors may have occurred when writing data
that was dirtied via an entirely different fd, but that's the case now
with the current mapping_set_error/filemap_check_error infrastructure.
This will at least prevent you from getting a false report of success.

The new behavior is still consistent with the POSIX spec, and is more
reliable for application developers. This patch just adds some basic
infrastructure for doing this, and ensures that the f_wb_err "cursor"
is properly set when a file is opened. Later patches will change the
existing code to use this new infrastructure for reporting errors at
fsync time.

Signed-off-by: Jeff Layton 
Reviewed-by: Jan Kara

mm: don't TestClearPageError in __filemap_fdatawait_range

2017-07-06T11:02:24+00:00

The -EIO returned here can end up overriding whatever error is marked in
the address space, and be returned at fsync time, even when there is a
more appropriate error stored in the mapping.

Read errors are also sometimes tracked on a per-page level using
PG_error. Suppose we have a read error on a page, and then that page is
subsequently dirtied by overwriting the whole page. Writeback doesn't
clear PG_error, so we can then end up successfully writing back that
page and still return -EIO on fsync.

Worse yet, PG_error is cleared during a sync() syscall, but the -EIO
return from that is silently discarded. Any subsystem that is relying on
PG_error to report errors during fsync can easily lose writeback errors
due to this. All you need is a stray sync() call to wait for writeback
to complete and you've lost the error.

Since the handling of the PG_error flag is somewhat inconsistent across
subsystems, let's just rely on marking the address space when there are
writeback errors. Change the TestClearPageError call to ClearPageError,
and make __filemap_fdatawait_range a void return function.

Signed-off-by: Jeff Layton

mm: clear AS_EIO/AS_ENOSPC when writeback initiation fails

2017-07-06T11:02:23+00:00

filemap_write_and_wait{_range} will return an error if writeback
initiation fails, but won't clear errors in the address_space. This is
particularly problematic on DAX, as filemap_fdatawrite* is
effectively synchronous there. Ensure that we clear the AS_EIO/AS_ENOSPC
flags when filemap_fdatawrite* returns an error.

Signed-off-by: Jeff Layton

jbd2: don't clear and reset errors after waiting on writeback

2017-07-06T11:02:22+00:00

Resetting this flag is almost certainly racy, and will be problematic
with some coming changes.

Make filemap_fdatawait_keep_errors return int, but not clear the flag(s).
Have jbd2 call it instead of filemap_fdatawait and don't attempt to
re-set the error flag if it fails.

Reviewed-by: Jan Kara 
Reviewed-by: Carlos Maiolino 
Signed-off-by: Jeff Layton