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
<interrupts mistakenly enabled>
<interrupt>
end_page_writeback
test_clear_page_writeback
lock_page_memcg
<deadlock>
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: 682aa8e1a6 ("writeback: implement unlocked_inode_to_wb transaction and use it for stat updates")
Signed-off-by: Greg Thelen <gthelen@google.com>
Reported-by: Wang Long <wanglong19@meituan.com>
Acked-by: Wang Long <wanglong19@meituan.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: <stable@vger.kernel.org> [v4.2+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Remove the address_space ->tree_lock and use the xa_lock newly added to
the radix_tree_root. Rename the address_space ->page_tree to ->i_pages,
since we don't really care that it's a tree.
[willy@infradead.org: fix nds32, fs/dax.c]
Link: http://lkml.kernel.org/r/20180406145415.GB20605@bombadil.infradead.orgLink: http://lkml.kernel.org/r/20180313132639.17387-9-willy@infradead.org
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Acked-by: Jeff Layton <jlayton@redhat.com>
Cc: Darrick J. Wong <darrick.wong@oracle.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
And use it in a few more places rather than opencoding the values.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
The @head can be wb->b_dirty_time, so update the comment.
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Wang Long <wanglong19@meituan.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This is a pure automated search-and-replace of the internal kernel
superblock flags.
The s_flags are now called SB_*, with the names and the values for the
moment mirroring the MS_* flags that they're equivalent to.
Note how the MS_xyz flags are the ones passed to the mount system call,
while the SB_xyz flags are what we then use in sb->s_flags.
The script to do this was:
# places to look in; re security/*: it generally should *not* be
# touched (that stuff parses mount(2) arguments directly), but
# there are two places where we really deal with superblock flags.
FILES="drivers/mtd drivers/staging/lustre fs ipc mm \
include/linux/fs.h include/uapi/linux/bfs_fs.h \
security/apparmor/apparmorfs.c security/apparmor/include/lib.h"
# the list of MS_... constants
SYMS="RDONLY NOSUID NODEV NOEXEC SYNCHRONOUS REMOUNT MANDLOCK \
DIRSYNC NOATIME NODIRATIME BIND MOVE REC VERBOSE SILENT \
POSIXACL UNBINDABLE PRIVATE SLAVE SHARED RELATIME KERNMOUNT \
I_VERSION STRICTATIME LAZYTIME SUBMOUNT NOREMOTELOCK NOSEC BORN \
ACTIVE NOUSER"
SED_PROG=
for i in $SYMS; do SED_PROG="$SED_PROG -e s/MS_$i/SB_$i/g"; done
# we want files that contain at least one of MS_...,
# with fs/namespace.c and fs/pnode.c excluded.
L=$(for i in $SYMS; do git grep -w -l MS_$i $FILES; done| sort|uniq|grep -v '^fs/namespace.c'|grep -v '^fs/pnode.c')
for f in $L; do sed -i $f $SED_PROG; done
Requested-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Since commit 925a6efb8f ("Btrfs: stop using
try_to_writeback_inodes_sb_nr to flush delalloc") this function hasn't
been used outside so stop exporting it.
In addition we merge it into try_to_writeback_inodes_sb() which is the
only caller. Also change return type of try_to_writeback_inodes_sb to
void as the only user ext4 doesn't care.
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Rakesh Pandit <rakesh@tuxera.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Handle start-all writeback like we do periodic or kupdate
style writeback - by marking the bdi_writeback as needing a full
flush, and simply waking the thread. This eliminates the need to
allocate and queue a specific work item just for this purpose.
After this change, we truly only ever have one of them running at
any point in time. We mark the need to start all flushes, and the
writeback thread will clear it once it has processed the request.
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When someone calls wakeup_flusher_threads() or
wakeup_flusher_threads_bdi(), they schedule writeback of all dirty
pages in the system (or on that bdi). If we are tight on memory, we
can get tons of these queued from kswapd/vmscan. This causes (at
least) two problems:
1) We consume a ton of memory just allocating writeback work items.
We've seen as much as 600 million of these writeback work items
pending. That's a lot of memory to pointlessly hold hostage,
while the box is under memory pressure.
2) We spend so much time processing these work items, that we
introduce a softlockup in writeback processing. This is because
each of the writeback work items don't end up doing any work (it's
hard when you have millions of identical ones coming in to the
flush machinery), so we just sit in a tight loop pulling work
items and deleting/freeing them.
Fix this by adding a 'start_all' bit to the writeback structure, and
set that when someone attempts to flush all dirty pages. The bit is
cleared when we start writeback on that work item. If the bit is
already set when we attempt to queue !nr_pages writeback, then we
simply ignore it.
This provides us one full flush in flight, with one pending as well,
and makes for more efficient handling of this type of writeback.
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Tested-by: Chris Mason <clm@fb.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Now that we have no external callers of wb_start_writeback(), we
can shuffle the passing in of 'nr_pages'. Everybody passes in 0
at this point, so just kill the argument and move the dirty
count retrieval to that function.
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Tested-by: Chris Mason <clm@fb.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We don't have any callers outside of fs-writeback.c anymore,
make it private.
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Tested-by: Chris Mason <clm@fb.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Similar to wakeup_flusher_threads(), except that we only wake
up the flusher threads on the specified backing device.
No functional changes in this patch.
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Tested-by: Chris Mason <clm@fb.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We're writing back the full range of dirty pages on the devices,
there's no point in making this special and not do normal range
cyclic writeback.
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Everybody is passing in 0 now, let's get rid of the argument.
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently the writeback statistics code uses a percpu counters to hold
various statistics. Furthermore we have 2 families of functions - those
which disable local irq and those which doesn't and whose names begin
with double underscore. However, they both end up calling
__add_wb_stats which in turn calls percpu_counter_add_batch which is
already irq-safe.
Exploiting this fact allows to eliminated the __wb_* functions since
they don't add any further protection than we already have.
Furthermore, refactor the wb_* function to call __add_wb_stat directly
without the irq-disabling dance. This will likely result in better
runtime of code which deals with modifying the stat counters.
While at it also document why percpu_counter_add_batch is in fact
preempt and irq-safe since at least 3 people got confused.
Link: http://lkml.kernel.org/r/1498029937-27293-1-git-send-email-nborisov@suse.com
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Acked-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Sphinx gets confused when it finds identation without a
good reason for it and without a preceding blank line:
./fs/mpage.c:347: ERROR: Unexpected indentation.
./fs/namei.c:4303: ERROR: Unexpected indentation.
./fs/fs-writeback.c:2060: ERROR: Unexpected indentation.
No functional changes.
Signed-off-by: Mauro Carvalho Chehab <mchehab@s-opensource.com>
When WB_registered flag is not set, wb_queue_work() skips queuing the
work, but does not perform the necessary clean up. In particular, if
work->auto_free is true, it should free the memory.
The leak condition can be reprouced by following these steps:
mount /dev/sdb /mnt/sdb
/* In qemu console: device_del sdb */
umount /dev/sdb
Above will result in a wb_queue_work() call on an unregistered wb and
thus leak memory.
Reported-by: John Sperbeck <jsperbeck@google.com>
Signed-off-by: Tahsin Erdogan <tahsin@google.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
I've found funny live-lock between raid10 barriers during resync and
memory controller hard limits. Inside mpage_readpages() task holds on to
its plug bio which blocks the barrier in raid10. Its memory cgroup have
no free memory thus the task goes into reclaimer but all reclaimable
pages are dirty and cannot be written because raid10 is rebuilding and
stuck on the barrier.
Common flush of such IO in schedule() never happens, because the caller
doesn't go to sleep.
Lock is 'live' because changing memory limit or killing tasks which
holds that stuck bio unblock whole progress.
That was what happened in 3.18.x but I see no difference in upstream
logic. Theoretically this might happen even without memory cgroup.
Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Signed-off-by: Jens Axboe <axboe@fb.com>
There are now a number of accounting oddities such as mapped file pages
being accounted for on the node while the total number of file pages are
accounted on the zone. This can be coped with to some extent but it's
confusing so this patch moves the relevant file-based accounted. Due to
throttling logic in the page allocator for reliable OOM detection, it is
still necessary to track dirty and writeback pages on a per-zone basis.
[mgorman@techsingularity.net: fix NR_ZONE_WRITE_PENDING accounting]
Link: http://lkml.kernel.org/r/1468404004-5085-5-git-send-email-mgorman@techsingularity.net
Link: http://lkml.kernel.org/r/1467970510-21195-20-git-send-email-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The per-sb inode writeback list tracks inodes currently under writeback
to facilitate efficient sync processing. In particular, it ensures that
sync only needs to walk through a list of inodes that were cleaned by
the sync.
Add a couple tracepoints to help identify when inodes are added/removed
to and from the writeback lists. Piggyback off of the writeback
lazytime tracepoint template as it already tracks the relevant inode
information.
Link: http://lkml.kernel.org/r/1466594593-6757-3-git-send-email-bfoster@redhat.com
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Dave Chinner <dchinner@redhat.com>
cc: Josef Bacik <jbacik@fb.com>
Cc: Holger Hoffstätte <holger.hoffstaette@applied-asynchrony.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
wait_sb_inodes() currently does a walk of all inodes in the filesystem
to find dirty one to wait on during sync. This is highly inefficient
and wastes a lot of CPU when there are lots of clean cached inodes that
we don't need to wait on.
To avoid this "all inode" walk, we need to track inodes that are
currently under writeback that we need to wait for. We do this by
adding inodes to a writeback list on the sb when the mapping is first
tagged as having pages under writeback. wait_sb_inodes() can then walk
this list of "inodes under IO" and wait specifically just for the inodes
that the current sync(2) needs to wait for.
Define a couple helpers to add/remove an inode from the writeback list
and call them when the overall mapping is tagged for or cleared from
writeback. Update wait_sb_inodes() to walk only the inodes under
writeback due to the sync.
With this change, filesystem sync times are significantly reduced for
fs' with largely populated inode caches and otherwise no other work to
do. For example, on a 16xcpu 2GHz x86-64 server, 10TB XFS filesystem
with a ~10m entry inode cache, sync times are reduced from ~7.3s to less
than 0.1s when the filesystem is fully clean.
Link: http://lkml.kernel.org/r/1466594593-6757-2-git-send-email-bfoster@redhat.com
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Tested-by: Holger Hoffstätte <holger.hoffstaette@applied-asynchrony.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time
ago with promise that one day it will be possible to implement page
cache with bigger chunks than PAGE_SIZE.
This promise never materialized. And unlikely will.
We have many places where PAGE_CACHE_SIZE assumed to be equal to
PAGE_SIZE. And it's constant source of confusion on whether
PAGE_CACHE_* or PAGE_* constant should be used in a particular case,
especially on the border between fs and mm.
Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much
breakage to be doable.
Let's stop pretending that pages in page cache are special. They are
not.
The changes are pretty straight-forward:
- <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>;
- <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>;
- PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN};
- page_cache_get() -> get_page();
- page_cache_release() -> put_page();
This patch contains automated changes generated with coccinelle using
script below. For some reason, coccinelle doesn't patch header files.
I've called spatch for them manually.
The only adjustment after coccinelle is revert of changes to
PAGE_CAHCE_ALIGN definition: we are going to drop it later.
There are few places in the code where coccinelle didn't reach. I'll
fix them manually in a separate patch. Comments and documentation also
will be addressed with the separate patch.
virtual patch
@@
expression E;
@@
- E << (PAGE_CACHE_SHIFT - PAGE_SHIFT)
+ E
@@
expression E;
@@
- E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT)
+ E
@@
@@
- PAGE_CACHE_SHIFT
+ PAGE_SHIFT
@@
@@
- PAGE_CACHE_SIZE
+ PAGE_SIZE
@@
@@
- PAGE_CACHE_MASK
+ PAGE_MASK
@@
expression E;
@@
- PAGE_CACHE_ALIGN(E)
+ PAGE_ALIGN(E)
@@
expression E;
@@
- page_cache_get(E)
+ get_page(E)
@@
expression E;
@@
- page_cache_release(E)
+ put_page(E)
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When cgroup writeback is in use, there can be multiple wb's
(bdi_writeback's) per bdi and an inode may switch among them
dynamically. In a couple places, the wrong wb was used leading to
performing operations on the wrong list under the wrong lock
corrupting the io lists.
* writeback_single_inode() was taking @wb parameter and used it to
remove the inode from io lists if it becomes clean after writeback.
The callers of this function were always passing in the root wb
regardless of the actual wb that the inode was associated with,
which could also change while writeback is in progress.
Fix it by dropping the @wb parameter and using
inode_to_wb_and_lock_list() to determine and lock the associated wb.
* After writeback_sb_inodes() writes out an inode, it re-locks @wb and
inode to remove it from or move it to the right io list. It assumes
that the inode is still associated with @wb; however, the inode may
have switched to another wb while writeback was in progress.
Fix it by using inode_to_wb_and_lock_list() to determine and lock
the associated wb after writeback is complete. As the function
requires the original @wb->list_lock locked for the next iteration,
in the unlikely case where the inode has changed association, switch
the locks.
Kudos to Tahsin for pinpointing these subtle breakages.
Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: d10c809552 ("writeback: implement foreign cgroup inode bdi_writeback switching")
Link: http://lkml.kernel.org/g/CAAeU0aMYeM_39Y2+PaRvyB1nqAPYZSNngJ1eBRmrxn7gKAt2Mg@mail.gmail.com
Reported-and-diagnosed-by: Tahsin Erdogan <tahsin@google.com>
Tested-by: Tahsin Erdogan <tahsin@google.com>
Cc: stable@vger.kernel.org # v4.2+
Signed-off-by: Jens Axboe <axboe@fb.com>
locked_inode_to_wb_and_lock_list() wb_get()'s the wb associated with
the target inode, unlocks inode, locks the wb's list_lock and verifies
that the inode is still associated with the wb. To prevent the wb
going away between dropping inode lock and acquiring list_lock, the wb
is pinned while inode lock is held. The wb reference is put right
after acquiring list_lock citing that the wb won't be dereferenced
anymore.
This isn't true. If the inode is still associated with the wb, the
inode has reference and it's safe to return the wb; however, if inode
has been switched, the wb still needs to be unlocked which is a
dereference and can lead to use-after-free if it it races with wb
destruction.
Fix it by putting the reference after releasing list_lock.
Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: 87e1d789bf ("writeback: implement [locked_]inode_to_wb_and_lock_list()")
Cc: stable@vger.kernel.org # v4.2+
Tested-by: Tahsin Erdogan <tahsin@google.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
If cgroup writeback is in use, inodes can be scheduled for
asynchronous wb switching. Before 5ff8eaac16 ("writeback: keep
superblock pinned during cgroup writeback association switches"), this
could race with umount leading to super_block being destroyed while
inodes are pinned for wb switching. 5ff8eaac16 fixed it by bumping
s_active while wb switches are in flight; however, this allowed
in-flight wb switches to make umounts asynchronous when the userland
expected synchronosity - e.g. fsck immediately following umount may
fail because the device is still busy.
This patch removes the problematic super_block pinning and instead
makes generic_shutdown_super() flush in-flight wb switches. wb
switches are now executed on a dedicated isw_wq so that they can be
flushed and isw_nr_in_flight keeps track of the number of in-flight wb
switches so that flushing can be avoided in most cases.
v2: Move cgroup_writeback_umount() further below and add MS_ACTIVE
check in inode_switch_wbs() as Jan an Al suggested.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Tahsin Erdogan <tahsin@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Link: http://lkml.kernel.org/g/CAAeU0aNCq7LGODvVGRU-oU_o-6enii5ey0p1c26D1ZzYwkDc5A@mail.gmail.com
Fixes: 5ff8eaac16 ("writeback: keep superblock pinned during cgroup writeback association switches")
Cc: stable@vger.kernel.org #v4.5
Reviewed-by: Jan Kara <jack@suse.cz>
Tested-by: Tahsin Erdogan <tahsin@google.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
If cgroup writeback is in use, an inode is associated with a cgroup
for writeback. If the inode's main dirtier changes to another cgroup,
the association gets updated asynchronously. Nothing was pinning the
superblock while such switches are in progress and superblock could go
away while async switching is pending or in progress leading to
crashes like the following.
kernel BUG at fs/jbd2/transaction.c:319!
invalid opcode: 0000 [#1] SMP DEBUG_PAGEALLOC
CPU: 1 PID: 29158 Comm: kworker/1:10 Not tainted 4.5.0-rc3 #51
Hardware name: Google Google, BIOS Google 01/01/2011
Workqueue: events inode_switch_wbs_work_fn
task: ffff880213dbbd40 ti: ffff880209264000 task.ti: ffff880209264000
RIP: 0010:[<ffffffff803e6922>] [<ffffffff803e6922>] start_this_handle+0x382/0x3e0
RSP: 0018:ffff880209267c30 EFLAGS: 00010202
...
Call Trace:
[<ffffffff803e6be4>] jbd2__journal_start+0xf4/0x190
[<ffffffff803cfc7e>] __ext4_journal_start_sb+0x4e/0x70
[<ffffffff803b31ec>] ext4_evict_inode+0x12c/0x3d0
[<ffffffff8035338b>] evict+0xbb/0x190
[<ffffffff80354190>] iput+0x130/0x190
[<ffffffff80360223>] inode_switch_wbs_work_fn+0x343/0x4c0
[<ffffffff80279819>] process_one_work+0x129/0x300
[<ffffffff80279b16>] worker_thread+0x126/0x480
[<ffffffff8027ed14>] kthread+0xc4/0xe0
[<ffffffff809771df>] ret_from_fork+0x3f/0x70
Fix it by bumping s_active while cgroup association switching is in
flight.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-and-tested-by: Tahsin Erdogan <tahsin@google.com>
Link: http://lkml.kernel.org/g/CAAeU0aNCq7LGODvVGRU-oU_o-6enii5ey0p1c26D1ZzYwkDc5A@mail.gmail.com
Fixes: d10c809552 ("writeback: implement foreign cgroup inode bdi_writeback switching")
Cc: stable@vger.kernel.org #v4.5+
Signed-off-by: Jens Axboe <axboe@fb.com>
In earlier versions, mem_cgroup_css_from_page() could return non-root
css on a legacy hierarchy which can go away and required rcu locking;
however, the eventual version simply returns the root cgroup if memcg is
on a legacy hierarchy and thus doesn't need rcu locking around or in it.
Remove spurious rcu lockings.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov@virtuozzo.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>
Fix kernel-doc warnings in fs/fs-writeback.c by moving a #define macro to
after the function's opening brace. Also #undef this macro at the end of
the function.
../fs/fs-writeback.c:1984: warning: Excess function parameter 'inode' description in 'I_DIRTY_INODE'
../fs/fs-writeback.c:1984: warning: Excess function parameter 'flags' description in 'I_DIRTY_INODE'
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
filemap_fdatawait() is a function to wait for on-going writeback to
complete but also consume and clear error status of the mapping set during
writeback.
The latter functionality is critical for applications to detect writeback
error with system calls like fsync(2)/fdatasync(2).
However filemap_fdatawait() is also used by sync(2) or FIFREEZE ioctl,
which don't check error status of individual mappings.
As a result, fsync() may not be able to detect writeback error if events
happen in the following order:
Application System admin
----------------------------------------------------------
write data on page cache
Run sync command
writeback completes with error
filemap_fdatawait() clears error
fsync returns success
(but the data is not on disk)
This patch adds filemap_fdatawait_keep_errors() for call sites where
writeback error is not handled so that they don't clear error status.
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Acked-by: Andi Kleen <ak@linux.intel.com>
Reviewed-by: Tejun Heo <tj@kernel.org>
Cc: Fengguang Wu <fengguang.wu@gmail.com>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
bdi_split_work_to_wbs() uses list_for_each_entry_rcu_continue()
to walk @bdi->wb_list. To set up the initial iteration
condition, it uses list_entry_rcu() to calculate the entry
pointer corresponding to the list head; however, this isn't an
actual RCU dereference and using list_entry_rcu() for it ended
up breaking a proposed list_entry_rcu() change because it was
feeding an non-lvalue pointer into the macro.
Don't use the RCU variant for simple pointer offsetting. Use
list_entry() instead.
Reported-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Darren Hart <dvhart@linux.intel.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Lai Jiangshan <jiangshanlai@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Patrick Marlier <patrick.marlier@gmail.com>
Cc: Paul McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: pranith kumar <bobby.prani@gmail.com>
Link: http://lkml.kernel.org/r/20151027051939.GA19355@mtj.duckdns.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
bdi_for_each_wb() is used in several places to wake up or issue
writeback work items to all wb's (bdi_writeback's) on a given bdi.
The iteration is performed by walking bdi->cgwb_tree; however, the
tree only indexes wb's which are currently active.
For example, when a memcg gets associated with a different blkcg, the
old wb is removed from the tree so that the new one can be indexed.
The old wb starts dying from then on but will linger till all its
inodes are drained. As these dying wb's may still host dirty inodes,
writeback operations which affect all wb's must include them.
bdi_for_each_wb() skipping dying wb's led to sync(2) missing and
failing to sync the inodes belonging to those wb's.
This patch adds a RCU protected @bdi->wb_list which lists all wb's
beloinging to that bdi. wb's are added on creation and removed on
release rather than on the start of destruction. bdi_for_each_wb()
usages are replaced with list_for_each[_continue]_rcu() iterations
over @bdi->wb_list and bdi_for_each_wb() and its helpers are removed.
v2: Updated as per Jan. last_wb ref leak in bdi_split_work_to_wbs()
fixed and unnecessary list head severing in cgwb_bdi_destroy()
removed.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-and-tested-by: Artem Bityutskiy <dedekind1@gmail.com>
Fixes: ebe41ab0c7 ("writeback: implement bdi_for_each_wb()")
Link: http://lkml.kernel.org/g/1443012552.19983.209.camel@gmail.com
Cc: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
wakeup_dirtytime_writeback() walks and wakes up all wb's of all bdi's;
unfortunately, it was always waking up bdi->wb instead of the wb being
walked. Fix it.
Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: 001fe6f617 ("writeback: make wakeup_dirtytime_writeback() handle multiple bdi_writeback's")
Reviewed-by: Jan Kara <jack@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Commit 505a666ee3 ("writeback: plug writeback in wb_writeback() and
writeback_inodes_wb()") has us holding a plug during writeback_sb_inodes,
which increases the merge rate when relatively contiguous small files
are written by the filesystem. It helps both on flash and spindles.
For an fs_mark workload creating 4K files in parallel across 8 drives,
this commit improves performance ~9% more by unplugging before calling
cond_resched(). cond_resched() doesn't trigger an implicit unplug, so
explicitly getting the IO down to the device before scheduling reduces
latencies for anyone waiting on clean pages.
It also cuts down on how often we use kblockd to unplug, which means
less work bouncing from one workqueue to another.
Many more details about how we got here:
https://lkml.org/lkml/2015/9/11/570
Signed-off-by: Chris Mason <clm@fb.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We had to revert the pluggin in writeback_sb_inodes() because the
wb->list_lock is held, but we could easily plug at a higher level before
taking that lock, and unplug after releasing it. This does that.
Chris will run performance numbers, just to verify that this approach is
comparable to the alternative (we could just drop and re-take the lock
around the blk_finish_plug() rather than these two commits.
I'd have preferred waiting for actual performance numbers before picking
one approach over the other, but I don't want to release rc1 with the
known "sleeping function called from invalid context" issue, so I'll
pick this cleanup version for now. But if the numbers show that we
really want to plug just at the writeback_sb_inodes() level, and we
should just play ugly games with the spinlock, we'll switch to that.
Cc: Chris Mason <clm@fb.com>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Jan Kara <jack@suse.cz>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This reverts commit d353d7587d.
Doing the block layer plug/unplug inside writeback_sb_inodes() is
broken, because that function is actually called with a spinlock held:
wb->list_lock, as pointed out by Chris Mason.
Chris suggested just dropping and re-taking the spinlock around the
blk_finish_plug() call (the plgging itself can happen under the
spinlock), and that would technically work, but is just disgusting.
We do something fairly similar - but not quite as disgusting because we
at least have a better reason for it - in writeback_single_inode(), so
it's not like the caller can depend on the lock being held over the
call, but in this case there just isn't any good reason for that
"release and re-take the lock" pattern.
[ In general, we should really strive to avoid the "release and retake"
pattern for locks, because in the general case it can easily cause
subtle bugs when the caller caches any state around the call that
might be invalidated by dropping the lock even just temporarily. ]
But in this case, the plugging should be easy to just move up to the
callers before the spinlock is taken, which should even improve the
effectiveness of the plug. So there is really no good reason to play
games with locking here.
I'll send off a test-patch so that Dave Chinner can verify that that
plug movement works. In the meantime this just reverts the problematic
commit and adds a comment to the function so that we hopefully don't
make this mistake again.
Reported-by: Chris Mason <clm@fb.com>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Jan Kara <jack@suse.cz>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull blk-cg updates from Jens Axboe:
"A bit later in the cycle, but this has been in the block tree for a a
while. This is basically four patchsets from Tejun, that improve our
buffered cgroup writeback. It was dependent on the other cgroup
changes, but they went in earlier in this cycle.
Series 1 is set of 5 patches that has cgroup writeback updates:
- bdi_writeback iteration fix which could lead to some wb's being
skipped or repeated during e.g. sync under memory pressure.
- Simplification of wb work wait mechanism.
- Writeback tracepoints updated to report cgroup.
Series 2 is is a set of updates for the CFQ cgroup writeback handling:
cfq has always charged all async IOs to the root cgroup. It didn't
have much choice as writeback didn't know about cgroups and there
was no way to tell who to blame for a given writeback IO.
writeback finally grew support for cgroups and now tags each
writeback IO with the appropriate cgroup to charge it against.
This patchset updates cfq so that it follows the blkcg each bio is
tagged with. Async cfq_queues are now shared across cfq_group,
which is per-cgroup, instead of per-request_queue cfq_data. This
makes all IOs follow the weight based IO resource distribution
implemented by cfq.
- Switched from GFP_ATOMIC to GFP_NOWAIT as suggested by Jeff.
- Other misc review points addressed, acks added and rebased.
Series 3 is the blkcg policy cleanup patches:
This patchset contains assorted cleanups for blkcg_policy methods
and blk[c]g_policy_data handling.
- alloc/free added for blkg_policy_data. exit dropped.
- alloc/free added for blkcg_policy_data.
- blk-throttle's async percpu allocation is replaced with direct
allocation.
- all methods now take blk[c]g_policy_data instead of blkcg_gq or
blkcg.
And finally, series 4 is a set of patches cleaning up the blkcg stats
handling:
blkcg's stats have always been somwhat of a mess. This patchset
tries to improve the situation a bit.
- The following patches added to consolidate blkcg entry point and
blkg creation. This is in itself is an improvement and helps
colllecting common stats on bio issue.
- per-blkg stats now accounted on bio issue rather than request
completion so that bio based and request based drivers can behave
the same way. The issue was spotted by Vivek.
- cfq-iosched implements custom recursive stats and blk-throttle
implements custom per-cpu stats. This patchset make blkcg core
support both by default.
- cfq-iosched and blk-throttle keep track of the same stats
multiple times. Unify them"
* 'for-4.3/blkcg' of git://git.kernel.dk/linux-block: (45 commits)
blkcg: use CGROUP_WEIGHT_* scale for io.weight on the unified hierarchy
blkcg: s/CFQ_WEIGHT_*/CFQ_WEIGHT_LEGACY_*/
blkcg: implement interface for the unified hierarchy
blkcg: misc preparations for unified hierarchy interface
blkcg: separate out tg_conf_updated() from tg_set_conf()
blkcg: move body parsing from blkg_conf_prep() to its callers
blkcg: mark existing cftypes as legacy
blkcg: rename subsystem name from blkio to io
blkcg: refine error codes returned during blkcg configuration
blkcg: remove unnecessary NULL checks from __cfqg_set_weight_device()
blkcg: reduce stack usage of blkg_rwstat_recursive_sum()
blkcg: remove cfqg_stats->sectors
blkcg: move io_service_bytes and io_serviced stats into blkcg_gq
blkcg: make blkg_[rw]stat_recursive_sum() to be able to index into blkcg_gq
blkcg: make blkcg_[rw]stat per-cpu
blkcg: add blkg_[rw]stat->aux_cnt and replace cfq_group->dead_stats with it
blkcg: consolidate blkg creation in blkcg_bio_issue_check()
blk-throttle: improve queue bypass handling
blkcg: move root blkg lookup optimization from throtl_lookup_tg() to __blkg_lookup()
blkcg: inline [__]blkg_lookup()
...
Pull vfs updates from Al Viro:
"In this one:
- d_move fixes (Eric Biederman)
- UFS fixes (me; locking is mostly sane now, a bunch of bugs in error
handling ought to be fixed)
- switch of sb_writers to percpu rwsem (Oleg Nesterov)
- superblock scalability (Josef Bacik and Dave Chinner)
- swapon(2) race fix (Hugh Dickins)"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: (65 commits)
vfs: Test for and handle paths that are unreachable from their mnt_root
dcache: Reduce the scope of i_lock in d_splice_alias
dcache: Handle escaped paths in prepend_path
mm: fix potential data race in SyS_swapon
inode: don't softlockup when evicting inodes
inode: rename i_wb_list to i_io_list
sync: serialise per-superblock sync operations
inode: convert inode_sb_list_lock to per-sb
inode: add hlist_fake to avoid the inode hash lock in evict
writeback: plug writeback at a high level
change sb_writers to use percpu_rw_semaphore
shift percpu_counter_destroy() into destroy_super_work()
percpu-rwsem: kill CONFIG_PERCPU_RWSEM
percpu-rwsem: introduce percpu_rwsem_release() and percpu_rwsem_acquire()
percpu-rwsem: introduce percpu_down_read_trylock()
document rwsem_release() in sb_wait_write()
fix the broken lockdep logic in __sb_start_write()
introduce __sb_writers_{acquired,release}() helpers
ufs_inode_get{frag,block}(): get rid of 'phys' argument
ufs_getfrag_block(): tidy up a bit
...
e79729123f ("writeback: don't issue wb_writeback_work if clean")
updated writeback path to avoid kicking writeback work items if there
are no inodes to be written out; unfortunately, the avoidance logic
was too aggressive and broke sync_inodes_sb().
* sync_inodes_sb() must write out I_DIRTY_TIME inodes but I_DIRTY_TIME
inodes dont't contribute to bdi/wb_has_dirty_io() tests and were
being skipped over.
* inodes are taken off wb->b_dirty/io/more_io lists after writeback
starts on them. sync_inodes_sb() skipping wait_sb_inodes() when
bdi_has_dirty_io() breaks it by making it return while writebacks
are in-flight.
This patch fixes the breakages by
* Removing bdi_has_dirty_io() shortcut from bdi_split_work_to_wbs().
The callers are already testing the condition.
* Removing bdi_has_dirty_io() shortcut from sync_inodes_sb() so that
it always calls into bdi_split_work_to_wbs() and wait_sb_inodes().
* Making bdi_split_work_to_wbs() consider the b_dirty_time list for
WB_SYNC_ALL writebacks.
Kudos to Eryu, Dave and Jan for tracking down the issue.
Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: e79729123f ("writeback: don't issue wb_writeback_work if clean")
Link: http://lkml.kernel.org/g/20150812101204.GE17933@dhcp-13-216.nay.redhat.com
Reported-and-bisected-by: Eryu Guan <eguan@redhat.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jan Kara <jack@suse.com>
Cc: Ted Ts'o <tytso@google.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
The following tracepoints are updated to report the cgroup used during
cgroup writeback.
* writeback_write_inode[_start]
* writeback_queue
* writeback_exec
* writeback_start
* writeback_written
* writeback_wait
* writeback_nowork
* writeback_wake_background
* wbc_writepage
* writeback_queue_io
* bdi_dirty_ratelimit
* balance_dirty_pages
* writeback_sb_inodes_requeue
* writeback_single_inode[_start]
Note that writeback_bdi_register is separated out from writeback_class
as reporting cgroup doesn't make sense to it. Tracepoints which take
bdi are updated to take bdi_writeback instead.
Signed-off-by: Tejun Heo <tj@kernel.org>
Suggested-by: Jan Kara <jack@suse.cz>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
wb_writeback_work->single_wait/done are used for the wait mechanism
for synchronous wb_work (wb_writeback_work) items which are issued
when bdi_split_work_to_wbs() fails to allocate memory for asynchronous
wb_work items; however, there's no reason to use a separate wait
mechanism for this. bdi_split_work_to_wbs() can simply use on-stack
fallback wb_work item and separate wb_completion to wait for it.
This patch removes wb_work->single_wait/done and the related code and
make bdi_split_work_to_wbs() use on-stack fallback wb_work and
wb_completion instead.
Signed-off-by: Tejun Heo <tj@kernel.org>
Suggested-by: Jan Kara <jack@suse.cz>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
wb's (bdi_writeback's) are currently keyed by memcg ID; however, in an
earlier implementation, wb's were keyed by blkcg ID.
bdi_for_each_wb() walks bdi->cgwb_tree in the ascending ID order and
allows iterations to start from an arbitrary ID which is used to
interrupt and resume iterations.
Unfortunately, while changing wb to be keyed by memcg ID instead of
blkcg, bdi_for_each_wb() was missed and is still assuming that wb's
are keyed by blkcg ID. This doesn't affect iterations which don't get
interrupted but bdi_split_work_to_wbs() makes use of iteration
resuming on allocation failures and thus may incorrectly skip or
repeat wb's.
Fix it by changing bdi_for_each_wb() to take memcg IDs instead of
blkcg IDs and updating bdi_split_work_to_wbs() accordingly.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@fb.com>
There's a small consistency problem between the inode and writeback
naming. Writeback calls the "for IO" inode queues b_io and
b_more_io, but the inode calls these the "writeback list" or
i_wb_list. This makes it hard to an new "under writeback" list to
the inode, or call it an "under IO" list on the bdi because either
way we'll have writeback on IO and IO on writeback and it'll just be
confusing. I'm getting confused just writing this!
So, rename the inode "for IO" list variable to i_io_list so we can
add a new "writeback list" in a subsequent patch.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: Dave Chinner <dchinner@redhat.com>
When competing sync(2) calls walk the same filesystem, they need to
walk the list of inodes on the superblock to find all the inodes
that we need to wait for IO completion on. However, when multiple
wait_sb_inodes() calls do this at the same time, they contend on the
the inode_sb_list_lock and the contention causes system wide
slowdowns. In effect, concurrent sync(2) calls can take longer and
burn more CPU than if they were serialised.
Stop the worst of the contention by adding a per-sb mutex to wrap
around wait_sb_inodes() so that we only execute one sync(2) IO
completion walk per superblock superblock at a time and hence avoid
contention being triggered by concurrent sync(2) calls.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: Dave Chinner <dchinner@redhat.com>
The process of reducing contention on per-superblock inode lists
starts with moving the locking to match the per-superblock inode
list. This takes the global lock out of the picture and reduces the
contention problems to within a single filesystem. This doesn't get
rid of contention as the locks still have global CPU scope, but it
does isolate operations on different superblocks form each other.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: Dave Chinner <dchinner@redhat.com>
Doing writeback on lots of little files causes terrible IOPS storms
because of the per-mapping writeback plugging we do. This
essentially causes imeediate dispatch of IO for each mapping,
regardless of the context in which writeback is occurring.
IOWs, running a concurrent write-lots-of-small 4k files using fsmark
on XFS results in a huge number of IOPS being issued for data
writes. Metadata writes are sorted and plugged at a high level by
XFS, so aggregate nicely into large IOs. However, data writeback IOs
are dispatched in individual 4k IOs, even when the blocks of two
consecutively written files are adjacent.
Test VM: 8p, 8GB RAM, 4xSSD in RAID0, 100TB sparse XFS filesystem,
metadata CRCs enabled.
Kernel: 3.10-rc5 + xfsdev + my 3.11 xfs queue (~70 patches)
Test:
$ ./fs_mark -D 10000 -S0 -n 10000 -s 4096 -L 120 -d
/mnt/scratch/0 -d /mnt/scratch/1 -d /mnt/scratch/2 -d
/mnt/scratch/3 -d /mnt/scratch/4 -d /mnt/scratch/5 -d
/mnt/scratch/6 -d /mnt/scratch/7
Result:
wall sys create rate Physical write IO
time CPU (avg files/s) IOPS Bandwidth
----- ----- ------------ ------ ---------
unpatched 6m56s 15m47s 24,000+/-500 26,000 130MB/s
patched 5m06s 13m28s 32,800+/-600 1,500 180MB/s
improvement -26.44% -14.68% +36.67% -94.23% +38.46%
If I use zero length files, this workload at about 500 IOPS, so
plugging drops the data IOs from roughly 25,500/s to 1000/s.
3 lines of code, 35% better throughput for 15% less CPU.
The benefits of plugging at this layer are likely to be higher for
spinning media as the IO patterns for this workload are going make a
much bigger difference on high IO latency devices.....
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Tested-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Greg Thelen