| Age | Commit message (Collapse) | Author |
|---|
|
There were two places where return value from bdi_init was not tested.
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
The feature prevents mistrusted filesystems (ie: FUSE mounts created by
unprivileged users) to grow a large number of dirty pages before
throttling. For such filesystems balance_dirty_pages always check bdi
counters against bdi limits. I.e. even if global "nr_dirty" is under
"freerun", it's not allowed to skip bdi checks. The only use case for now
is fuse: it sets bdi max_ratio to 1% by default and system administrators
are supposed to expect that this limit won't be exceeded.
The feature is on if a BDI is marked by BDI_CAP_STRICTLIMIT flag. A
filesystem may set the flag when it initializes its BDI.
The problematic scenario comes from the fact that nobody pays attention to
the NR_WRITEBACK_TEMP counter (i.e. number of pages under fuse
writeback). The implementation of fuse writeback releases original page
(by calling end_page_writeback) almost immediately. A fuse request queued
for real processing bears a copy of original page. Hence, if userspace
fuse daemon doesn't finalize write requests in timely manner, an
aggressive mmap writer can pollute virtually all memory by those temporary
fuse page copies. They are carefully accounted in NR_WRITEBACK_TEMP, but
nobody cares.
To make further explanations shorter, let me use "NR_WRITEBACK_TEMP
problem" as a shortcut for "a possibility of uncontrolled grow of amount
of RAM consumed by temporary pages allocated by kernel fuse to process
writeback".
The problem was very easy to reproduce. There is a trivial example
filesystem implementation in fuse userspace distribution: fusexmp_fh.c. I
added "sleep(1);" to the write methods, then recompiled and mounted it.
Then created a huge file on the mount point and run a simple program which
mmap-ed the file to a memory region, then wrote a data to the region. An
hour later I observed almost all RAM consumed by fuse writeback. Since
then some unrelated changes in kernel fuse made it more difficult to
reproduce, but it is still possible now.
Putting this theoretical happens-in-the-lab thing aside, there is another
thing that really hurts real world (FUSE) users. This is write-through
page cache policy FUSE currently uses. I.e. handling write(2), kernel
fuse populates page cache and flushes user data to the server
synchronously. This is excessively suboptimal. Pavel Emelyanov's patches
("writeback cache policy") solve the problem, but they also make resolving
NR_WRITEBACK_TEMP problem absolutely necessary. Otherwise, simply copying
a huge file to a fuse mount would result in memory starvation. Miklos,
the maintainer of FUSE, believes strictlimit feature the way to go.
And eventually putting FUSE topics aside, there is one more use-case for
strictlimit feature. Using a slow USB stick (mass storage) in a machine
with huge amount of RAM installed is a well-known pain. Let's make simple
computations. Assuming 64GB of RAM installed, existing implementation of
balance_dirty_pages will start throttling only after 9.6GB of RAM becomes
dirty (freerun == 15% of total RAM). So, the command "cp 9GB_file
/media/my-usb-storage/" may return in a few seconds, but subsequent
"umount /media/my-usb-storage/" will take more than two hours if effective
throughput of the storage is, to say, 1MB/sec.
After inclusion of strictlimit feature, it will be trivial to add a knob
(e.g. /sys/devices/virtual/bdi/x:y/strictlimit) to enable it on demand.
Manually or via udev rule. May be I'm wrong, but it seems to be quite a
natural desire to limit the amount of dirty memory for some devices we are
not fully trust (in the sense of sustainable throughput).
[akpm@linux-foundation.org: fix warning in page-writeback.c]
Signed-off-by: Maxim Patlasov <MPatlasov@parallels.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Writeback implements its own worker pool - each bdi can be associated
with a worker thread which is created and destroyed dynamically. The
worker thread for the default bdi is always present and serves as the
"forker" thread which forks off worker threads for other bdis.
there's no reason for writeback to implement its own worker pool when
using unbound workqueue instead is much simpler and more efficient.
This patch replaces custom worker pool implementation in writeback
with an unbound workqueue.
The conversion isn't too complicated but the followings are worth
mentioning.
* bdi_writeback->last_active, task and wakeup_timer are removed.
delayed_work ->dwork is added instead. Explicit timer handling is
no longer necessary. Everything works by either queueing / modding
/ flushing / canceling the delayed_work item.
* bdi_writeback_thread() becomes bdi_writeback_workfn() which runs off
bdi_writeback->dwork. On each execution, it processes
bdi->work_list and reschedules itself if there are more things to
do.
The function also handles low-mem condition, which used to be
handled by the forker thread. If the function is running off a
rescuer thread, it only writes out limited number of pages so that
the rescuer can serve other bdis too. This preserves the flusher
creation failure behavior of the forker thread.
* INIT_LIST_HEAD(&bdi->bdi_list) is used to tell
bdi_writeback_workfn() about on-going bdi unregistration so that it
always drains work_list even if it's running off the rescuer. Note
that the original code was broken in this regard. Under memory
pressure, a bdi could finish unregistration with non-empty
work_list.
* The default bdi is no longer special. It now is treated the same as
any other bdi and bdi_cap_flush_forker() is removed.
* BDI_pending is no longer used. Removed.
* Some tracepoints become non-applicable. The following TPs are
removed - writeback_nothread, writeback_wake_thread,
writeback_wake_forker_thread, writeback_thread_start,
writeback_thread_stop.
Everything, including devices coming and going away and rescuer
operation under simulated memory pressure, seems to work fine in my
test setup.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
|
|
There's no user left. Remove it.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Fengguang Wu <fengguang.wu@intel.com>
|
|
This patchset ("stable page writes, part 2") makes some key
modifications to the original 'stable page writes' patchset. First, it
provides creators (devices and filesystems) of a backing_dev_info a flag
that declares whether or not it is necessary to ensure that page
contents cannot change during writeout. It is no longer assumed that
this is true of all devices (which was never true anyway). Second, the
flag is used to relaxed the wait_on_page_writeback calls so that wait
only occurs if the device needs it. Third, it fixes up the remaining
disk-backed filesystems to use this improved conditional-wait logic to
provide stable page writes on those filesystems.
It is hoped that (for people not using checksumming devices, anyway)
this patchset will give back unnecessary performance decreases since the
original stable page write patchset went into 3.0. Sorry about not
fixing it sooner.
Complaints were registered by several people about the long write
latencies introduced by the original stable page write patchset.
Generally speaking, the kernel ought to allocate as little extra memory
as possible to facilitate writeout, but for people who simply cannot
wait, a second page stability strategy is (re)introduced: snapshotting
page contents. The waiting behavior is still the default strategy; to
enable page snapshotting, a superblock flag (MS_SNAP_STABLE) must be
set. This flag is used to bandaid^Henable stable page writeback on
ext3[1], and is not used anywhere else.
Given that there are already a few storage devices and network FSes that
have rolled their own page stability wait/page snapshot code, it would
be nice to move towards consolidating all of these. It seems possible
that iscsi and raid5 may wish to use the new stable page write support
to enable zero-copy writeout.
Thank you to Jan Kara for helping fix a couple more filesystems.
Per Andrew Morton's request, here are the result of using dbench to measure
latencies on ext2:
3.8.0-rc3:
Operation Count AvgLat MaxLat
----------------------------------------
WriteX 109347 0.028 59.817
ReadX 347180 0.004 3.391
Flush 15514 29.828 287.283
Throughput 57.429 MB/sec 4 clients 4 procs max_latency=287.290 ms
3.8.0-rc3 + patches:
WriteX 105556 0.029 4.273
ReadX 335004 0.005 4.112
Flush 14982 30.540 298.634
Throughput 55.4496 MB/sec 4 clients 4 procs max_latency=298.650 ms
As you can see, for ext2 the maximum write latency decreases from ~60ms
on a laptop hard disk to ~4ms. I'm not sure why the flush latencies
increase, though I suspect that being able to dirty pages faster gives
the flusher more work to do.
On ext4, the average write latency decreases as well as all the maximum
latencies:
3.8.0-rc3:
WriteX 85624 0.152 33.078
ReadX 272090 0.010 61.210
Flush 12129 36.219 168.260
Throughput 44.8618 MB/sec 4 clients 4 procs max_latency=168.276 ms
3.8.0-rc3 + patches:
WriteX 86082 0.141 30.928
ReadX 273358 0.010 36.124
Flush 12214 34.800 165.689
Throughput 44.9941 MB/sec 4 clients 4 procs max_latency=165.722 ms
XFS seems to exhibit similar latency improvements as ext2:
3.8.0-rc3:
WriteX 125739 0.028 104.343
ReadX 399070 0.005 4.115
Flush 17851 25.004 131.390
Throughput 66.0024 MB/sec 4 clients 4 procs max_latency=131.406 ms
3.8.0-rc3 + patches:
WriteX 123529 0.028 6.299
ReadX 392434 0.005 4.287
Flush 17549 25.120 188.687
Throughput 64.9113 MB/sec 4 clients 4 procs max_latency=188.704 ms
...and btrfs, just to round things out, also shows some latency
decreases:
3.8.0-rc3:
WriteX 67122 0.083 82.355
ReadX 212719 0.005 2.828
Flush 9547 47.561 147.418
Throughput 35.3391 MB/sec 4 clients 4 procs max_latency=147.433 ms
3.8.0-rc3 + patches:
WriteX 64898 0.101 71.631
ReadX 206673 0.005 7.123
Flush 9190 47.963 219.034
Throughput 34.0795 MB/sec 4 clients 4 procs max_latency=219.044 ms
Before this patchset, all filesystems would block, regardless of whether
or not it was necessary. ext3 would wait, but still generate occasional
checksum errors. The network filesystems were left to do their own
thing, so they'd wait too.
After this patchset, all the disk filesystems except ext3 and btrfs will
wait only if the hardware requires it. ext3 (if necessary) snapshots
pages instead of blocking, and btrfs provides its own bdi so the mm will
never wait. Network filesystems haven't been touched, so either they
provide their own wait code, or they don't block at all. The blocking
behavior is back to what it was before 3.0 if you don't have a disk
requiring stable page writes.
This patchset has been tested on 3.8.0-rc3 on x64 with ext3, ext4, and
xfs. I've spot-checked 3.8.0-rc4 and seem to be getting the same
results as -rc3.
[1] The alternative fixes to ext3 include fixing the locking order and
page bit handling like we did for ext4 (but then why not just use
ext4?), or setting PG_writeback so early that ext3 becomes extremely
slow. I tried that, but the number of write()s I could initiate dropped
by nearly an order of magnitude. That was a bit much even for the
author of the stable page series! :)
This patch:
Creates a per-backing-device flag that tracks whether or not pages must
be held immutable during writeout. Eventually it will be used to waive
wait_for_page_writeback() if nothing requires stable pages.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Artem Bityutskiy <dedekind1@gmail.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Eric Van Hensbergen <ericvh@gmail.com>
Cc: Ron Minnich <rminnich@sandia.gov>
Cc: Latchesar Ionkov <lucho@ionkov.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
__printf is useful to verify format and arguments.
Signed-off-by: Joe Perches <joe@perches.com>
Reviewed-by: Alex Elder <elder@inktank.com>
|
|
Finally we can kill the 'sync_supers' kernel thread along with the
'->write_super()' superblock operation because all the users are gone.
Now every file-system is supposed to self-manage own superblock and
its dirty state.
The nice thing about killing this thread is that it improves power management.
Indeed, 'sync_supers' is a source of monotonic system wake-ups - it woke up
every 5 seconds no matter what - even if there were no dirty superblocks and
even if there were no file-systems using this service (e.g., btrfs and
journalled ext4 do not need it). So it was wasting power most of the time. And
because the thread was in the core of the kernel, all systems had to have it.
So I am quite happy to make it go away.
Interestingly, this thread is a left-over from the pdflush kernel thread which
was a self-forking kernel thread responsible for all the write-back in old
Linux kernels. It was turned into per-block device BDI threads, and
'sync_supers' was a left-over. Thus, R.I.P, pdflush as well.
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
|
|
Since per-BDI flusher threads were introduced in 2.6, the pdflush
mechanism is not used any more. But the old interface exported through
/proc/sys/vm/nr_pdflush_threads still exists and is obviously useless.
For back-compatibility, printk warning information and return 2 to notify
the users that the interface is removed.
Signed-off-by: Wanpeng Li <liwp@linux.vnet.ibm.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Convert calculations of proportion of writeback each bdi does to new flexible
proportion code. That allows us to use aging period of fixed wallclock time
which gives better proportion estimates given the hugely varying throughput of
different devices.
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
|
|
This creates a new 'reason' field in a wb_writeback_work
structure, which unambiguously identifies who initiates
writeback activity. A 'wb_reason' enumeration has been
added to writeback.h, to enumerate the possible reasons.
The 'writeback_work_class' and tracepoint event class and
'writeback_queue_io' tracepoints are updated to include the
symbolic 'reason' in all trace events.
And the 'writeback_inodes_sbXXX' family of routines has had
a wb_stats parameter added to them, so callers can specify
why writeback is being started.
Acked-by: Jan Kara <jack@suse.cz>
Signed-off-by: Curt Wohlgemuth <curtw@google.com>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
There are some imperfections in balanced_dirty_ratelimit.
1) large fluctuations
The dirty_rate used for computing balanced_dirty_ratelimit is merely
averaged in the past 200ms (very small comparing to the 3s estimation
period for write_bw), which makes rather dispersed distribution of
balanced_dirty_ratelimit.
It's pretty hard to average out the singular points by increasing the
estimation period. Considering that the averaging technique will
introduce very undesirable time lags, I give it up totally. (btw, the 3s
write_bw averaging time lag is much more acceptable because its impact
is one-way and therefore won't lead to oscillations.)
The more practical way is filtering -- most singular
balanced_dirty_ratelimit points can be filtered out by remembering some
prev_balanced_rate and prev_prev_balanced_rate. However the more
reliable way is to guard balanced_dirty_ratelimit with task_ratelimit.
2) due to truncates and fs redirties, the (write_bw <=> dirty_rate)
match could become unbalanced, which may lead to large systematical
errors in balanced_dirty_ratelimit. The truncates, due to its possibly
bumpy nature, can hardly be compensated smoothly. So let's face it. When
some over-estimated balanced_dirty_ratelimit brings dirty_ratelimit
high, dirty pages will go higher than the setpoint. task_ratelimit will
in turn become lower than dirty_ratelimit. So if we consider both
balanced_dirty_ratelimit and task_ratelimit and update dirty_ratelimit
only when they are on the same side of dirty_ratelimit, the systematical
errors in balanced_dirty_ratelimit won't be able to bring
dirty_ratelimit far away.
The balanced_dirty_ratelimit estimation may also be inaccurate near
@limit or @freerun, however is less an issue.
3) since we ultimately want to
- keep the fluctuations of task ratelimit as small as possible
- keep the dirty pages around the setpoint as long time as possible
the update policy used for (2) also serves the above goals nicely:
if for some reason the dirty pages are high (task_ratelimit < dirty_ratelimit),
and dirty_ratelimit is low (dirty_ratelimit < balanced_dirty_ratelimit),
there is no point to bring up dirty_ratelimit in a hurry only to hurt
both the above two goals.
So, we make use of task_ratelimit to limit the update of dirty_ratelimit
in two ways:
1) avoid changing dirty rate when it's against the position control target
(the adjusted rate will slow down the progress of dirty pages going
back to setpoint).
2) limit the step size. task_ratelimit is changing values step by step,
leaving a consistent trace comparing to the randomly jumping
balanced_dirty_ratelimit. task_ratelimit also has the nice smaller
errors in stable state and typically larger errors when there are big
errors in rate. So it's a pretty good limiting factor for the step
size of dirty_ratelimit.
Note that bdi->dirty_ratelimit is always tracking balanced_dirty_ratelimit.
task_ratelimit is merely used as a limiting factor.
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
It's all about bdi->dirty_ratelimit, which aims to be (write_bw / N)
when there are N dd tasks.
On write() syscall, use bdi->dirty_ratelimit
============================================
balance_dirty_pages(pages_dirtied)
{
task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio();
pause = pages_dirtied / task_ratelimit;
sleep(pause);
}
On every 200ms, update bdi->dirty_ratelimit
===========================================
bdi_update_dirty_ratelimit()
{
task_ratelimit = bdi->dirty_ratelimit * bdi_position_ratio();
balanced_dirty_ratelimit = task_ratelimit * write_bw / dirty_rate;
bdi->dirty_ratelimit = balanced_dirty_ratelimit
}
Estimation of balanced bdi->dirty_ratelimit
===========================================
balanced task_ratelimit
-----------------------
balance_dirty_pages() needs to throttle tasks dirtying pages such that
the total amount of dirty pages stays below the specified dirty limit in
order to avoid memory deadlocks. Furthermore we desire fairness in that
tasks get throttled proportionally to the amount of pages they dirty.
IOW we want to throttle tasks such that we match the dirty rate to the
writeout bandwidth, this yields a stable amount of dirty pages:
dirty_rate == write_bw (1)
The fairness requirement gives us:
task_ratelimit = balanced_dirty_ratelimit
== write_bw / N (2)
where N is the number of dd tasks. We don't know N beforehand, but
still can estimate balanced_dirty_ratelimit within 200ms.
Start by throttling each dd task at rate
task_ratelimit = task_ratelimit_0 (3)
(any non-zero initial value is OK)
After 200ms, we measured
dirty_rate = # of pages dirtied by all dd's / 200ms
write_bw = # of pages written to the disk / 200ms
For the aggressive dd dirtiers, the equality holds
dirty_rate == N * task_rate
== N * task_ratelimit_0 (4)
Or
task_ratelimit_0 == dirty_rate / N (5)
Now we conclude that the balanced task ratelimit can be estimated by
write_bw
balanced_dirty_ratelimit = task_ratelimit_0 * ---------- (6)
dirty_rate
Because with (4) and (5) we can get the desired equality (1):
write_bw
balanced_dirty_ratelimit == (dirty_rate / N) * ----------
dirty_rate
== write_bw / N
Then using the balanced task ratelimit we can compute task pause times like:
task_pause = task->nr_dirtied / task_ratelimit
task_ratelimit with position control
------------------------------------
However, while the above gives us means of matching the dirty rate to
the writeout bandwidth, it at best provides us with a stable dirty page
count (assuming a static system). In order to control the dirty page
count such that it is high enough to provide performance, but does not
exceed the specified limit we need another control.
The dirty position control works by extending (2) to
task_ratelimit = balanced_dirty_ratelimit * pos_ratio (7)
where pos_ratio is a negative feedback function that subjects to
1) f(setpoint) = 1.0
2) df/dx < 0
That is, if the dirty pages are ABOVE the setpoint, we throttle each
task a bit more HEAVY than balanced_dirty_ratelimit, so that the dirty
pages are created less fast than they are cleaned, thus DROP to the
setpoints (and the reverse).
Based on (7) and the assumption that both dirty_ratelimit and pos_ratio
remains CONSTANT for the past 200ms, we get
task_ratelimit_0 = balanced_dirty_ratelimit * pos_ratio (8)
Putting (8) into (6), we get the formula used in
bdi_update_dirty_ratelimit():
write_bw
balanced_dirty_ratelimit *= pos_ratio * ---------- (9)
dirty_rate
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
Introduce the BDI_DIRTIED counter. It will be used for estimating the
bdi's dirty bandwidth.
CC: Jan Kara <jack@suse.cz>
CC: Michael Rubin <mrubin@google.com>
CC: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
This allows us to move duplicated code in <asm/atomic.h>
(atomic_inc_not_zero() for now) to <linux/atomic.h>
Signed-off-by: Arun Sharma <asharma@fb.com>
Reviewed-by: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: David Miller <davem@davemloft.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Mike Frysinger <vapier@gentoo.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
The estimation value will start from 100MB/s and adapt to the real
bandwidth in seconds.
It tries to update the bandwidth only when disk is fully utilized.
Any inactive period of more than one second will be skipped.
The estimated bandwidth will be reflecting how fast the device can
writeout when _fully utilized_, and won't drop to 0 when it goes idle.
The value will remain constant at disk idle time. At busy write time, if
not considering fluctuations, it will also remain high unless be knocked
down by possible concurrent reads that compete for the disk time and
bandwidth with async writes.
The estimation is not done purely in the flusher because there is no
guarantee for write_cache_pages() to return timely to update bandwidth.
The bdi->avg_write_bandwidth smoothing is very effective for filtering
out sudden spikes, however may be a little biased in long term.
The overheads are low because the bdi bandwidth update only occurs at
200ms intervals.
The 200ms update interval is suitable, because it's not possible to get
the real bandwidth for the instance at all, due to large fluctuations.
The NFS commits can be as large as seconds worth of data. One XFS
completion may be as large as half second worth of data if we are going
to increase the write chunk to half second worth of data. In ext4,
fluctuations with time period of around 5 seconds is observed. And there
is another pattern of irregular periods of up to 20 seconds on SSD tests.
That's why we are not only doing the estimation at 200ms intervals, but
also averaging them over a period of 3 seconds and then go further to do
another level of smoothing in avg_write_bandwidth.
CC: Li Shaohua <shaohua.li@intel.com>
CC: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
Introduce the BDI_WRITTEN counter. It will be used for estimating the
bdi's write bandwidth.
Peter Zijlstra <a.p.zijlstra@chello.nl>:
Move BDI_WRITTEN accounting into __bdi_writeout_inc().
This will cover and fix fuse, which only calls bdi_writeout_inc().
CC: Michael Rubin <mrubin@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
Split the global inode_wb_list_lock into a per-bdi_writeback list_lock,
as it's currently the most contended lock in the system for metadata
heavy workloads. It won't help for single-filesystem workloads for
which we'll need the I/O-less balance_dirty_pages, but at least we
can dedicate a cpu to spinning on each bdi now for larger systems.
Based on earlier patches from Nick Piggin and Dave Chinner.
It reduces lock contentions to 1/4 in this test case:
10 HDD JBOD, 100 dd on each disk, XFS, 6GB ram
lock_stat version 0.3
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
class name con-bounces contentions waittime-min waittime-max waittime-total acq-bounces acquisitions holdtime-min holdtime-max holdtime-total
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
vanilla 2.6.39-rc3:
inode_wb_list_lock: 42590 44433 0.12 147.74 144127.35 252274 886792 0.08 121.34 917211.23
------------------
inode_wb_list_lock 2 [<ffffffff81165da5>] bdev_inode_switch_bdi+0x29/0x85
inode_wb_list_lock 34 [<ffffffff8115bd0b>] inode_wb_list_del+0x22/0x49
inode_wb_list_lock 12893 [<ffffffff8115bb53>] __mark_inode_dirty+0x170/0x1d0
inode_wb_list_lock 10702 [<ffffffff8115afef>] writeback_single_inode+0x16d/0x20a
------------------
inode_wb_list_lock 2 [<ffffffff81165da5>] bdev_inode_switch_bdi+0x29/0x85
inode_wb_list_lock 19 [<ffffffff8115bd0b>] inode_wb_list_del+0x22/0x49
inode_wb_list_lock 5550 [<ffffffff8115bb53>] __mark_inode_dirty+0x170/0x1d0
inode_wb_list_lock 8511 [<ffffffff8115b4ad>] writeback_sb_inodes+0x10f/0x157
2.6.39-rc3 + patch:
&(&wb->list_lock)->rlock: 11383 11657 0.14 151.69 40429.51 90825 527918 0.11 145.90 556843.37
------------------------
&(&wb->list_lock)->rlock 10 [<ffffffff8115b189>] inode_wb_list_del+0x5f/0x86
&(&wb->list_lock)->rlock 1493 [<ffffffff8115b1ed>] writeback_inodes_wb+0x3d/0x150
&(&wb->list_lock)->rlock 3652 [<ffffffff8115a8e9>] writeback_sb_inodes+0x123/0x16f
&(&wb->list_lock)->rlock 1412 [<ffffffff8115a38e>] writeback_single_inode+0x17f/0x223
------------------------
&(&wb->list_lock)->rlock 3 [<ffffffff8110b5af>] bdi_lock_two+0x46/0x4b
&(&wb->list_lock)->rlock 6 [<ffffffff8115b189>] inode_wb_list_del+0x5f/0x86
&(&wb->list_lock)->rlock 2061 [<ffffffff8115af97>] __mark_inode_dirty+0x173/0x1cf
&(&wb->list_lock)->rlock 2629 [<ffffffff8115a8e9>] writeback_sb_inodes+0x123/0x16f
hughd@google.com: fix recursive lock when bdi_lock_two() is called with new the same as old
akpm@linux-foundation.org: cleanup bdev_inode_switch_bdi() comment
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
|
|
Code has been converted over to the new explicit on-stack plugging,
and delay users have been converted to use the new API for that.
So lets kill off the old plugging along with aops->sync_page().
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
|
|
Declare 'bdi_pending_list' and 'tag_pages_for_writeback()' to remove
following sparse warnings:
mm/backing-dev.c:46:1: warning: symbol 'bdi_pending_list' was not declared. Should it be static?
mm/page-writeback.c:825:6: warning: symbol 'tag_pages_for_writeback' was not declared. Should it be static?
Signed-off-by: Namhyung Kim <namhyung@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
BDIs or if significant congestion is not being encountered in the current zone
If congestion_wait() is called with no BDI congested, the caller will
sleep for the full timeout and this may be an unnecessary sleep. This
patch adds a wait_iff_congested() that checks congestion and only sleeps
if a BDI is congested else, it calls cond_resched() to ensure the caller
is not hogging the CPU longer than its quota but otherwise will not sleep.
This is aimed at reducing some of the major desktop stalls reported during
IO. For example, while kswapd is operating, it calls congestion_wait()
but it could just have been reclaiming clean page cache pages with no
congestion. Without this patch, it would sleep for a full timeout but
after this patch, it'll just call schedule() if it has been on the CPU too
long. Similar logic applies to direct reclaimers that are not making
enough progress.
Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.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>
|
|
Commit 83ba7b071f3 ("writeback: simplify the write back thread queue")
broke writeback_in_progress() as in that commit we started to remove work
items from the list at the moment we start working on them and not at the
moment they are finished. Thus if the flusher thread was doing some work
but there was no other work queued, writeback_in_progress() returned
false. This could in particular cause unnecessary queueing of background
writeback from balance_dirty_pages() or writeout work from
writeback_sb_if_idle().
This patch fixes the problem by introducing a bit in the bdi state which
indicates that the flusher thread is processing some work and uses this
bit for writeback_in_progress() test.
NOTE: Both callsites of writeback_in_progress() (namely,
writeback_inodes_sb_if_idle() and balance_dirty_pages()) would actually
need a different information than what writeback_in_progress() provides.
They would need to know whether *the kind of writeback they are going to
submit* is already queued. But this information isn't that simple to
provide so let's fix writeback_in_progress() for the time being.
Signed-off-by: Jan Kara <jack@suse.cz>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Acked-by: Jens Axboe <jaxboe@fusionio.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Whe the first inode for a bdi is marked dirty, we wake up the bdi thread which
should take care of the periodic background write-out. However, the write-out
will actually start only 'dirty_writeback_interval' centisecs later, so we can
delay the wake-up.
This change was requested by Nick Piggin who pointed out that if we delay the
wake-up, we weed out 2 unnecessary contex switches, which matters because
'__mark_inode_dirty()' is a hot-path function.
This patch introduces a new function - 'bdi_wakeup_thread_delayed()', which
sets up a timer to wake-up the bdi thread and returns. So the wake-up is
delayed.
We also delete the timer in bdi threads just before writing-back. And
synchronously delete it when unregistering bdi. At the unregister point the bdi
does not have any users, so no one can arm it again.
Since now we take 'bdi->wb_lock' in the timer, which can execute in softirq
context, we have to use 'spin_lock_bh()' for 'bdi->wb_lock'. This patch makes
this change as well.
This patch also moves the 'bdi_wb_init()' function down in the file to avoid
forward-declaration of 'bdi_wakeup_thread_delayed()'.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
|
|
Currently bdi threads use local variable 'last_active' which stores last time
when the bdi thread did some useful work. Move this local variable to 'struct
bdi_writeback'. This is just a preparation for the further patches which will
make the forker thread decide when bdi threads should be killed.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
|
|
This patch simplifies bdi code a little by removing the 'pending_list' which is
redundant. Indeed, currently the forker thread ('bdi_forker_thread()') is
working like this:
1. In a loop, fetch all bdi's which have works but have no writeback thread and
move them to the 'pending_list'.
2. If the list is empty, sleep for 5 sec.
3. Otherwise, take one bdi from the list, fork the writeback thread for this
bdi, and repeat the loop.
IOW, it first moves everything to the 'pending_list', then process only one
element, and so on. This patch simplifies the algorithm, which is now as
follows.
1. Find the first bdi which has a work and remove it from the global list of
bdi's (bdi_list).
2. If there was not such bdi, sleep 5 sec.
3. Fork the writeback thread for this bdi and repeat the loop.
IOW, now we find the first bdi to process, process it, and so on. This is
simpler and involves less lists.
The bonus now is that we can get rid of a couple of functions, as well as
remove complications which involve 'rcu_call()' and 'bdi->rcu_head'.
This patch also makes sure we use 'list_add_tail_rcu()', instead of plain
'list_add_tail()', but this piece of code is going to be removed in the next
patch anyway.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
|
|
The write-back code mixes words "thread" and "task" for the same things. This
is not a big deal, but still an inconsistency.
hch: a convention I tend to use and I've seen in various places
is to always use _task for the storage of the task_struct pointer,
and thread everywhere else. This especially helps with having
foo_thread for the actual thread and foo_task for a global
variable keeping the task_struct pointer
This patch renames:
* 'bdi_add_default_flusher_task()' -> 'bdi_add_default_flusher_thread()'
* 'bdi_forker_task()' -> 'bdi_forker_thread()'
because bdi threads are 'bdi_writeback_thread()', so these names are more
consistent.
This patch also amends commentaries and makes them refer the forker and bdi
threads as "thread", not "task".
Also, while on it, make 'bdi_add_default_flusher_thread()' declaration use
'static void' instead of 'void static' and make checkpatch.pl happy.
Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
|
|
Move all code for the writeback thread into fs/fs-writeback.c instead of
splitting it over two functions in two files.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
|
|
The wb_list member of struct backing_device_info always has exactly one
element. Just use the direct bdi->wb pointer instead and simplify some
code.
Also remove bdi_task_init which is now trivial to prepare for the next
patch.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
|
|
First remove items from work_list as soon as we start working on them. This
means we don't have to track any pending or visited state and can get
rid of all the RCU magic freeing the work items - we can simply free
them once the operation has finished. Second use a real completion for
tracking synchronous requests - if the caller sets the completion pointer
we complete it, otherwise use it as a boolean indicator that we can free
the work item directly. Third unify struct wb_writeback_args and struct
bdi_work into a single data structure, wb_writeback_work. Previous we
set all parameters into a struct wb_writeback_args, copied it into
struct bdi_work, copied it again on the stack to use it there. Instead
of just allocate one structure dynamically or on the stack and use it
all the way through the stack.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
|
|
bdi_start_writeback now never gets a superblock passed, so we can just remove
that case. And to further untangle the code and flatten the call stack
split it into two trivial helpers for it's two callers.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
|
|
This reverts commit e913fc825dc685a444cb4c1d0f9d32f372f59861.
We are investigating a hang associated with the WB_SYNC_NONE changes,
so revert them for now.
Conflicts:
fs/fs-writeback.c
mm/page-writeback.c
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
|
|
Commit 69b62d01 fixed up most of the places where we would enter
busy schedule() spins when disabling the periodic background
writeback. This fixes up the sb timer so that it doesn't get
hammered on with the delay disabled, and ensures that it gets
rearmed if needed when /proc/sys/vm/dirty_writeback_centisecs
gets modified.
bdi_forker_task() also needs to check for !dirty_writeback_centisecs
and use schedule() appropriately, fix that up too.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
When umount calls sync_filesystem(), we first do a WB_SYNC_NONE
writeback to kick off writeback of pending dirty inodes, then follow
that up with a WB_SYNC_ALL to wait for it. Since umount already holds
the sb s_umount mutex, WB_SYNC_NONE ends up doing nothing and all
writeback happens as WB_SYNC_ALL. This can greatly slow down umount,
since WB_SYNC_ALL writeback is a data integrity operation and thus
a bigger hammer than simple WB_SYNC_NONE. For barrier aware file systems
it's a lot slower.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
Conflicts:
fs/block_dev.c
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
noop_backing_dev_info is used only as a flag to mark filesystems that
don't have any backing store, like tmpfs, procfs, spufs, etc.
Signed-off-by: Joern Engel <joern@logfs.org>
Changed the BUG_ON() to a WARN_ON(). Note that adding dirty inodes
to the noop_backing_dev_info is not legal and will not result in
them being flushed, but we already catch this condition in
__mark_inode_dirty() when checking for a registered bdi.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
Pretty trivial helper, just sets up the bdi and registers it. An atomic
sequence count is used to ensure that the registered sysfs names are
unique.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
One of the features of laptop-mode is that it forces a writeout of dirty
pages if something else triggers a physical read or write from a device.
The current implementation flushes pages on all devices, rather than only
the one that triggered the flush. This patch alters the behaviour so that
only the recently accessed block device is flushed, preventing other
disks being spun up for no terribly good reason.
Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
There's nothing block related about them, the backing device
is used by things like NFS etc as well. This gets rid of the
need to protect such calls by CONFIG_BLOCK.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
Sometimes we only want to write pages from a specific super_block,
so allow that to be passed in.
This fixes a problem with commit 56a131dcf7ed36c3c6e36bea448b674ea85ed5bb
causing writeback on all super_blocks on a bdi, where we only really
want to sync a specific sb from writeback_inodes_sb().
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
bdi_start_writeback() is currently split into two paths, one for
WB_SYNC_NONE and one for WB_SYNC_ALL. Add bdi_sync_writeback()
for WB_SYNC_ALL writeback and let bdi_start_writeback() handle
only WB_SYNC_NONE.
Push down the writeback_control allocation and only accept the
parameters that make sense for each function. This cleans up
the API considerably.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
Now that bdi_writeback_all() no longer handles integrity writeback,
it doesn't have to block anymore. This means that we can switch
bdi_list reader side protection to RCU.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
Also a debugging aid. We want to catch dirty inodes being added to
backing devices that don't do writeback.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
This enables us to track who does what and print info. Its main use
is catching dirty inodes on the default_backing_dev_info, so we can
fix that up.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
This gets rid of pdflush for bdi writeout and kupdated style cleaning.
pdflush writeout suffers from lack of locality and also requires more
threads to handle the same workload, since it has to work in a
non-blocking fashion against each queue. This also introduces lumpy
behaviour and potential request starvation, since pdflush can be starved
for queue access if others are accessing it. A sample ffsb workload that
does random writes to files is about 8% faster here on a simple SATA drive
during the benchmark phase. File layout also seems a LOT more smooth in
vmstat:
r b swpd free buff cache si so bi bo in cs us sy id wa
0 1 0 608848 2652 375372 0 0 0 71024 604 24 1 10 48 42
0 1 0 549644 2712 433736 0 0 0 60692 505 27 1 8 48 44
1 0 0 476928 2784 505192 0 0 4 29540 553 24 0 9 53 37
0 1 0 457972 2808 524008 0 0 0 54876 331 16 0 4 38 58
0 1 0 366128 2928 614284 0 0 4 92168 710 58 0 13 53 34
0 1 0 295092 3000 684140 0 0 0 62924 572 23 0 9 53 37
0 1 0 236592 3064 741704 0 0 4 58256 523 17 0 8 48 44
0 1 0 165608 3132 811464 0 0 0 57460 560 21 0 8 54 38
0 1 0 102952 3200 873164 0 0 4 74748 540 29 1 10 48 41
0 1 0 48604 3252 926472 0 0 0 53248 469 29 0 7 47 45
where vanilla tends to fluctuate a lot in the creation phase:
r b swpd free buff cache si so bi bo in cs us sy id wa
1 1 0 678716 5792 303380 0 0 0 74064 565 50 1 11 52 36
1 0 0 662488 5864 319396 0 0 4 352 302 329 0 2 47 51
0 1 0 599312 5924 381468 0 0 0 78164 516 55 0 9 51 40
0 1 0 519952 6008 459516 0 0 4 78156 622 56 1 11 52 37
1 1 0 436640 6092 541632 0 0 0 82244 622 54 0 11 48 41
0 1 0 436640 6092 541660 0 0 0 8 152 39 0 0 51 49
0 1 0 332224 6200 644252 0 0 4 102800 728 46 1 13 49 36
1 0 0 274492 6260 701056 0 0 4 12328 459 49 0 7 50 43
0 1 0 211220 6324 763356 0 0 0 106940 515 37 1 10 51 39
1 0 0 160412 6376 813468 0 0 0 8224 415 43 0 6 49 45
1 1 0 85980 6452 886556 0 0 4 113516 575 39 1 11 54 34
0 2 0 85968 6452 886620 0 0 0 1640 158 211 0 0 46 54
A 10 disk test with btrfs performs 26% faster with per-bdi flushing. A
SSD based writeback test on XFS performs over 20% better as well, with
the throughput being very stable around 1GB/sec, where pdflush only
manages 750MB/sec and fluctuates wildly while doing so. Random buffered
writes to many files behave a lot better as well, as does random mmap'ed
writes.
A separate thread is added to sync the super blocks. In the long term,
adding sync_supers_bdi() functionality could get rid of this thread again.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
This is a first step at introducing per-bdi flusher threads. We should
have no change in behaviour, although sb_has_dirty_inodes() is now
ridiculously expensive, as there's no easy way to answer that question.
Not a huge problem, since it'll be deleted in subsequent patches.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
Move the definition of BLK_RW_ASYNC/BLK_RW_SYNC into linux/backing-dev.h
so that it is available to all callers of set/clear_bdi_congested().
This replaces commit 097041e576ee3a50d92dd643ee8ca65bf6a62e21 ("fuse:
Fix build error"), which will be reverted.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Acked-by: Larry Finger <Larry.Finger@lwfinger.net>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Commit 1faa16d22877f4839bd433547d770c676d1d964c accidentally broke
the bdi congestion wait queue logic, causing us to wait on congestion
for WRITE (== 1) when we really wanted BLK_RW_ASYNC (== 0) instead.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
|
|
This makes sure that we never wait on async IO for sync requests, instead
of doing the split on writes vs reads.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Split the LRU lists in two, one set for pages that are backed by real file
systems ("file") and one for pages that are backed by memory and swap
("anon"). The latter includes tmpfs.
The advantage of doing this is that the VM will not have to scan over lots
of anonymous pages (which we generally do not want to swap out), just to
find the page cache pages that it should evict.
This patch has the infrastructure and a basic policy to balance how much
we scan the anon lists and how much we scan the file lists. The big
policy changes are in separate patches.
[lee.schermerhorn@hp.com: collect lru meminfo statistics from correct offset]
[kosaki.motohiro@jp.fujitsu.com: prevent incorrect oom under split_lru]
[kosaki.motohiro@jp.fujitsu.com: fix pagevec_move_tail() doesn't treat unevictable page]
[hugh@veritas.com: memcg swapbacked pages active]
[hugh@veritas.com: splitlru: BDI_CAP_SWAP_BACKED]
[akpm@linux-foundation.org: fix /proc/vmstat units]
[nishimura@mxp.nes.nec.co.jp: memcg: fix handling of shmem migration]
[kosaki.motohiro@jp.fujitsu.com: adjust Quicklists field of /proc/meminfo]
[kosaki.motohiro@jp.fujitsu.com: fix style issue of get_scan_ratio()]
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Fuse needs this for writable mmap support.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
Add a new BDI capability flag: BDI_CAP_NO_ACCT_WB. If this flag is
set, then don't update the per-bdi writeback stats from
test_set_page_writeback() and test_clear_page_writeback().
Misc cleanups:
- convert bdi_cap_writeback_dirty() and friends to static inline functions
- create a flag that includes all three dirty/writeback related flags,
since almst all users will want to have them toghether
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
