Debug traces show that in per-bdi writeback, the inode under writeback
almost always get redirtied by a busy dirtier. We used to call
redirty_tail() in this case, which could delay inode for up to 30s.
This is unacceptable because it now happens so frequently for plain cp/dd,
that the accumulated delays could make writeback of big files very slow.
So let's distinguish between data redirty and metadata only redirty.
The first one is caused by a busy dirtier, while the latter one could
happen in XFS, NFS, etc. when they are doing delalloc or updating isize.
The inode being busy dirtied will now be requeued for next io, while
the inode being redirtied by fs will continue to be delayed to avoid
repeated IO.
CC: Jan Kara <jack@suse.cz>
CC: Theodore Ts'o <tytso@mit.edu>
CC: Dave Chinner <david@fromorbit.com>
CC: Chris Mason <chris.mason@oracle.com>
CC: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Currently we pin the inode->i_sb for every single inode. This
increases cache traffic on sb->s_umount sem. Lets instead
cache the inode sb pin state and keep the super_block pinned
for as long as keep writing out inodes from the same
super_block.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
If we only moved inodes from a single super_block to the temporary
list, there's no point in doing a resort for multiple super_blocks.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
__mark_inode_dirty adds inode to wb dirty list in random order. If a disk has
several partitions, writeback might keep spindle moving between partitions.
To reduce the move, better write big chunk of one partition and then move to
another. Inodes from one fs usually are in one partion, so idealy move indoes
from one fs together should reduce spindle move. This patch tries to address
this. Before per-bdi writeback is added, the behavior is write indoes
from one fs first and then another, so the patch restores previous behavior.
The loop in the patch is a bit ugly, should we add a dirty list for each
superblock in bdi_writeback?
Test in a two partition disk with attached fio script shows about 3% ~ 6%
improvement.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Reviewed-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Fix the kupdate case, which disregards wbc.more_io and stop writeback
prematurely even when there are more inodes to be synced.
wbc.more_io should always be respected.
Also remove the pages_skipped check. It will set when some page(s) of some
inode(s) cannot be written for now. Such inodes will be delayed for a while.
This variable has nothing to do with whether there are other writeable inodes.
CC: Jan Kara <jack@suse.cz>
CC: Dave Chinner <david@fromorbit.com>
CC: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Treat bdi_start_writeback(0) as a special request to do background write,
and stop such work when we are below the background dirty threshold.
Also simplify the (nr_pages <= 0) checks. Since we already pass in
nr_pages=LONG_MAX for WB_SYNC_ALL and background writes, we don't
need to worry about it being decreased to zero.
Reported-by: Richard Kennedy <richard@rsk.demon.co.uk>
CC: Jan Kara <jack@suse.cz>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Some filesystem may choose to write much more than ratelimit_pages
before calling balance_dirty_pages_ratelimited_nr(). So it is safer to
determine number to write based on real number of dirtied pages.
Otherwise it is possible that
loop {
btrfs_file_write(): dirty 1024 pages
balance_dirty_pages(): write up to 48 pages (= ratelimit_pages * 1.5)
}
in which the writeback rate cannot keep up with dirty rate, and the
dirty pages go all the way beyond dirty_thresh.
The increased write_chunk may make the dirtier more bumpy.
So filesystems shall be take care not to dirty too much at
a time (eg. > 4MB) without checking the ratelimit.
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
If all inodes are under writeback (e.g. in case when there's only one inode
with dirty pages), wb_writeback() with WB_SYNC_NONE work basically degrades
to busylooping until I_SYNC flags of the inode is cleared. Fix the problem by
waiting on I_SYNC flags of an inode on b_more_io list in case we failed to
write anything.
Tested-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Back in January 2008 Nick Piggin implemented "ticket" spinlocks
for X86 (See commit 314cdbefd1).
IA64 implementation has a couple of differences because of the
available atomic operations ... e.g. we have no fetchadd2 instruction
that operates on a 16-bit quantity so we make ticket locks use
a 32-bit word for each of the current ticket and now-serving values.
Performance on uncontended locks is about 8% worse than the previous
implementation, but this seems a good trade for determinism in the
contended case. Performance impact on macro-level benchmarks is in
the noise.
Signed-off-by: Tony Luck <tony.luck@intel.com>
* git://git.kernel.org/pub/scm/linux/kernel/git/lethal/sh-2.6:
sh_mobile_ceu_camera: fix compile breakage, caused by a bad merge
sh: Add support DMA Engine to SH7780
sh: Add support DMA Engine to SH7722
sh: enable onenand support in kfr2r09 defconfig.
sh: update defconfigs.
sh: add FSI driver support for ms7724se
sh: Fix up uninitialized variable use caught by gcc 4.4.
sh: Handle unaligned 16-bit instructions on SH-2A.
sh: mach-ecovec24: Add active low setting for sh_eth
sh: includecheck fix: dwarf.c
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6: (94 commits)
genetlink: fix netns vs. netlink table locking (2)
3c59x: Get rid of "Trying to free already-free IRQ"
tunnel: eliminate recursion field
ems_pci: fix size of CAN controllers BAR mapping for CPC-PCI v2
net: fix htmldocs sunrpc, clnt.c
Phonet: error on broadcast sending (unimplemented)
Phonet: fix race for port number in concurrent bind()
pktgen: better scheduler friendliness
pktgen: T_TERMINATE flag is unused
ipv4: check optlen for IP_MULTICAST_IF option
ath9k: Initialize txgain and rxgain for newer AR9287 chipsets.
iwlagn: fix panic in iwl{5000,4965}_rx_reply_tx
ath9k: Fix RFKILL bugs
drivers/net/wireless: Use usb_endpoint_dir_out
cfg80211: don't overwrite privacy setting
wl12xx: fix kconfig/link errors
rt2x00: fix the definition of rt2x00crypto_rx_insert_iv
iwlwifi: reduce noise when skb allocation fails
iwlwifi: do not send sync command while holding spinlock
mac80211: fix DTIM setting
...
Wu Fengguang
Jens Axboe
Shaohua Li
Jan Kara
Tony Luck
Linus Torvalds
Ingo Molnar
Thierry Reding
Guennadi Liakhovetski
Nobuhiro Iwamatsu
Paul Mundt
Steve French
Jeff Layton