linux_dsm_epyc7002/fs
Dave Chinner 7271d243f9 xfs: don't serialise adjacent concurrent direct IO appending writes
For append write workloads, extending the file requires a certain
amount of exclusive locking to be done up front to ensure sanity in
things like ensuring that we've zeroed any allocated regions
between the old EOF and the start of the new IO.

For single threads, this typically isn't a problem, and for large
IOs we don't serialise enough for it to be a problem for two
threads on really fast block devices. However for smaller IO and
larger thread counts we have a problem.

Take 4 concurrent sequential, single block sized and aligned IOs.
After the first IO is submitted but before it completes, we end up
with this state:

        IO 1    IO 2    IO 3    IO 4
      +-------+-------+-------+-------+
      ^       ^
      |       |
      |       |
      |       |
      |       \- ip->i_new_size
      \- ip->i_size

And the IO is done without exclusive locking because offset <=
ip->i_size. When we submit IO 2, we see offset > ip->i_size, and
grab the IO lock exclusive, because there is a chance we need to do
EOF zeroing. However, there is already an IO in progress that avoids
the need for IO zeroing because offset <= ip->i_new_size. hence we
could avoid holding the IO lock exlcusive for this. Hence after
submission of the second IO, we'd end up this state:

        IO 1    IO 2    IO 3    IO 4
      +-------+-------+-------+-------+
      ^               ^
      |               |
      |               |
      |               |
      |               \- ip->i_new_size
      \- ip->i_size

There is no need to grab the i_mutex of the IO lock in exclusive
mode if we don't need to invalidate the page cache. Taking these
locks on every direct IO effective serialises them as taking the IO
lock in exclusive mode has to wait for all shared holders to drop
the lock. That only happens when IO is complete, so effective it
prevents dispatch of concurrent direct IO writes to the same inode.

And so you can see that for the third concurrent IO, we'd avoid
exclusive locking for the same reason we avoided the exclusive lock
for the second IO.

Fixing this is a bit more complex than that, because we need to hold
a write-submission local value of ip->i_new_size to that clearing
the value is only done if no other thread has updated it before our
IO completes.....

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
2011-10-11 21:14:59 -05:00
..
9p fs/9p: Use protocol-defined value for lock/getlock 'type' field. 2011-09-06 08:17:16 -05:00
adfs
affs
afs
autofs4 autofs4: fix debug printk warning uncovered by cleanup 2011-08-08 12:02:43 -07:00
befs befs: Validate length of long symbolic links. 2011-08-17 13:31:24 -07:00
bfs
btrfs Merge branch 'btrfs-3.0' of git://github.com/chrismason/linux 2011-10-03 12:17:44 -07:00
cachefiles
ceph Merge branch 'for-linus' of git://ceph.newdream.net/git/ceph-client 2011-09-09 15:48:34 -07:00
cifs cifs: Fix broken sec=ntlmv2/i sec option (try #2) 2011-09-19 21:16:58 -05:00
coda
configfs
cramfs
debugfs
devpts
dlm
ecryptfs Ecryptfs: Add mount option to check uid of device being mounted = expect uid 2011-08-09 23:29:01 -05:00
efs
exofs Merge branch 'for-linus' of git://git.open-osd.org/linux-open-osd 2011-08-06 22:56:03 -07:00
exportfs
ext2 Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6 2011-08-01 13:48:31 -10:00
ext3 block: separate priority boosting from REQ_META 2011-08-23 14:50:29 +02:00
ext4 Merge branch 'for-linus' of git://git.kernel.dk/linux-block 2011-09-21 13:20:21 -07:00
fat Merge git://git.kernel.org/pub/scm/linux/kernel/git/hirofumi/fatfs-2.6 2011-08-18 14:16:13 -07:00
freevxfs
fscache
fuse fuse: fix memory leak 2011-09-12 11:47:10 -07:00
gfs2 block: separate priority boosting from REQ_META 2011-08-23 14:50:29 +02:00
hfs
hfsplus hfsplus: fix filesystem size checks 2011-09-15 09:03:17 -07:00
hostfs
hpfs
hppfs hppfs: missing include 2011-07-27 22:21:58 -04:00
hugetlbfs lockdep: Add helper function for dir vs file i_mutex annotation 2011-08-25 10:50:18 -07:00
isofs
jbd
jbd2
jffs2 switch posix_acl_equiv_mode() to umode_t * 2011-08-01 02:10:06 -04:00
jfs Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/shaggy/jfs-2.6 2011-08-15 08:40:24 -07:00
lockd Merge branch 'nfs-for-3.1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs 2011-07-27 13:23:02 -07:00
logfs
minix
ncpfs
nfs VFS: Fix the remaining automounter semantics regressions 2011-09-26 19:16:46 -07:00
nfs_common
nfsd
nilfs2
nls
notify atomic: use <linux/atomic.h> 2011-07-26 16:49:47 -07:00
ntfs atomic: use <linux/atomic.h> 2011-07-26 16:49:47 -07:00
ocfs2 switch posix_acl_equiv_mode() to umode_t * 2011-08-01 02:10:06 -04:00
omfs
openpromfs
partitions
proc teach /proc/$pid/numa_maps about transparent hugepages 2011-09-21 13:15:44 -07:00
pstore
qnx4
quota VFS: Fix the remaining automounter semantics regressions 2011-09-26 19:16:46 -07:00
ramfs
reiserfs switch posix_acl_equiv_mode() to umode_t * 2011-08-01 02:10:06 -04:00
romfs
squashfs
sysfs
sysv
ubifs UBIFS: not build debug messages with CONFIG_UBIFS_FS_DEBUG disabled 2011-08-19 18:58:58 +03:00
udf
ufs
xfs xfs: don't serialise adjacent concurrent direct IO appending writes 2011-10-11 21:14:59 -05:00
aio.c
anon_inodes.c
attr.c
bad_inode.c
binfmt_aout.c
binfmt_elf_fdpic.c
binfmt_elf.c
binfmt_em86.c
binfmt_flat.c
binfmt_misc.c
binfmt_script.c
binfmt_som.c
bio-integrity.c
bio.c
block_dev.c Avoid dereferencing a 'request_queue' after last close. 2011-09-10 17:20:21 +10:00
buffer.c
char_dev.c
compat_binfmt_elf.c
compat_ioctl.c compat_ioctl: add compat handler for PPPIOCGL2TPSTATS 2011-08-07 22:24:41 -07:00
compat.c All Arch: remove linkage for sys_nfsservctl system call 2011-08-26 15:09:58 -07:00
dcache.c vfs: renumber DCACHE_xyz flags, remove some stale ones 2011-08-06 22:52:40 -07:00
dcookies.c
direct-io.c atomic: use <linux/atomic.h> 2011-07-26 16:49:47 -07:00
drop_caches.c
eventfd.c
eventpoll.c atomic: use <linux/atomic.h> 2011-07-26 16:49:47 -07:00
exec.c move RLIMIT_NPROC check from set_user() to do_execve_common() 2011-08-11 11:24:42 -07:00
fcntl.c
fhandle.c
fifo.c
file_table.c atomic: use <linux/atomic.h> 2011-07-26 16:49:47 -07:00
file.c
filesystems.c
fs_struct.c
fs-writeback.c don't busy retry the inode on failed grab_super_passive() 2011-07-31 22:52:08 +08:00
generic_acl.c switch posix_acl_equiv_mode() to umode_t * 2011-08-01 02:10:06 -04:00
inode.c lockdep: Add helper function for dir vs file i_mutex annotation 2011-08-25 10:50:18 -07:00
internal.h
ioctl.c
ioprio.c
Kconfig tmpfs: expand "help" to explain value of TMPFS_POSIX_ACL 2011-08-03 14:25:25 -10:00
Kconfig.binfmt
libfs.c
locks.c
Makefile
mbcache.c
mpage.c
namei.c vfs: remove LOOKUP_NO_AUTOMOUNT flag 2011-09-27 08:12:33 -07:00
namespace.c VFS: Fix the remaining automounter semantics regressions 2011-09-26 19:16:46 -07:00
no-block.c
open.c
pipe.c
pnode.c
pnode.h
posix_acl.c switch posix_acl_chmod() to umode_t 2011-08-01 02:10:32 -04:00
read_write.c
read_write.h
readdir.c
select.c
seq_file.c
signalfd.c
splice.c
stack.c mm: a few small updates for radix-swap 2011-08-03 14:25:24 -10:00
stat.c vfs: remove LOOKUP_NO_AUTOMOUNT flag 2011-09-27 08:12:33 -07:00
statfs.c
super.c
sync.c
timerfd.c
utimes.c
xattr_acl.c
xattr.c