The AIO interface is fairly complex because it tries to allow
filesystems to always work async and then wakeup a synchronous
caller through aio_complete. It turns out that basically no one
was doing this to avoid the complexity and context switches,
and we've already fixed up the remaining users and can now
get rid of this case.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
This follows up "nfs: fix dio deadlock when O_DIRECT flag is flipped"
and removes the unnecessary CONFIG_NFS_SWAP switch.
Signed-off-by: Peng Tao <tao.peng@primarydata.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
There is no need to pass the total request length in the kiocb, as
we already get passed in through the iov_iter argument.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
With FAN_ONDIR set, the user can end up getting events, which it hasn't
marked. This was revealed with fanotify04 testcase failure on
Linux-4.0-rc1, and is a regression from 3.19, revealed with 66ba93c0d7
("fanotify: don't set FAN_ONDIR implicitly on a marks ignored mask").
# /opt/ltp/testcases/bin/fanotify04
[ ... ]
fanotify04 7 TPASS : event generated properly for type 100000
fanotify04 8 TFAIL : fanotify04.c:147: got unexpected event 30
fanotify04 9 TPASS : No event as expected
The testcase sets the adds the following marks : FAN_OPEN | FAN_ONDIR for
a fanotify on a dir. Then does an open(), followed by close() of the
directory and expects to see an event FAN_OPEN(0x20). However, the
fanotify returns (FAN_OPEN|FAN_CLOSE_NOWRITE(0x10)). This happens due to
the flaw in the check for event_mask in fanotify_should_send_event() which
does:
if (event_mask & marks_mask & ~marks_ignored_mask)
return true;
where, event_mask == (FAN_ONDIR | FAN_CLOSE_NOWRITE),
marks_mask == (FAN_ONDIR | FAN_OPEN),
marks_ignored_mask == 0
Fix this by masking the outgoing events to the user, as we already take
care of FAN_ONDIR and FAN_EVENT_ON_CHILD.
Signed-off-by: Suzuki K. Poulose <suzuki.poulose@arm.com>
Tested-by: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Eric Paris <eparis@redhat.com>
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>
According to a report from Yuxuan Shui, nilfs2 in kernel 3.19 got stuck
during recovery at mount time. The code path that caused the deadlock was
as follows:
nilfs_fill_super()
load_nilfs()
nilfs_salvage_orphan_logs()
* Do roll-forwarding, attach segment constructor for recovery,
and kick it.
nilfs_segctor_thread()
nilfs_segctor_thread_construct()
* A lock is held with nilfs_transaction_lock()
nilfs_segctor_do_construct()
nilfs_segctor_drop_written_files()
iput()
iput_final()
write_inode_now()
writeback_single_inode()
__writeback_single_inode()
do_writepages()
nilfs_writepage()
nilfs_construct_dsync_segment()
nilfs_transaction_lock() --> deadlock
This can happen if commit 7ef3ff2fea ("nilfs2: fix deadlock of segment
constructor over I_SYNC flag") is applied and roll-forward recovery was
performed at mount time. The roll-forward recovery can happen if datasync
write is done and the file system crashes immediately after that. For
instance, we can reproduce the issue with the following steps:
< nilfs2 is mounted on /nilfs (device: /dev/sdb1) >
# dd if=/dev/zero of=/nilfs/test bs=4k count=1 && sync
# dd if=/dev/zero of=/nilfs/test conv=notrunc oflag=dsync bs=4k
count=1 && reboot -nfh
< the system will immediately reboot >
# mount -t nilfs2 /dev/sdb1 /nilfs
The deadlock occurs because iput() can run segment constructor through
writeback_single_inode() if MS_ACTIVE flag is not set on sb->s_flags. The
above commit changed segment constructor so that it calls iput()
asynchronously for inodes with i_nlink == 0, but that change was
imperfect.
This fixes the another deadlock by deferring iput() in segment constructor
even for the case that mount is not finished, that is, for the case that
MS_ACTIVE flag is not set.
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Reported-by: Yuxuan Shui <yshuiv7@gmail.com>
Tested-by: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It turns out that making this feature ro_compat isn't quite enough to
prevent accidental corruption on mount from older kernels. Ocfs2 (like
other file systems) will process orphaned inodes even when the user mounts
in 'ro' mode. So for the case of a filesystem not knowing the append_dio
feature, mounting the filesystem could result in orphaned-for-dio files
being deleted, which we clearly don't want.
So instead, turn this into an incompat flag.
Btw, this is kind of my fault - initially I asked that we add a flag to
cover the feature and even suggested that we use an ro flag. It wasn't
until I was looking through our commits for v4.0-rc1 that I realized we
actually want this to be incompat.
Signed-off-by: Mark Fasheh <mfasheh@suse.de>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Do so on the assumption that for most use cases, that list will turn into
a more or less LRU-ordered list, and so the list traversals in
nfs_client_return_marked_delegations() are likely to be shorter before
hitting a candidate to return.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
A remount that alters security flavors can appear to succeed when it should
instead return -EINVAL. Check to see if the current security flavor exists
within the flavors specified in the remount options, and if not fail the
remount.
Signed-off-by: Benjamin Coddington <bcodding@redhat.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
The function nfs4_pnfs_v3_ds_connect_unload is exported so it can be
used by other modules, but is also marked '__exit' and will be
discarded when built into the kernel, as pointed out by this
linker error:
`nfs4_pnfs_v3_ds_connect_unload' referenced in section `___ksymtab_gpl+nfs4_pnfs_v3_ds_connect_unload' of fs/built-in.o: defined in discarded section `.exit.text' of fs/built-in.o
This removes the __exit annotation to make it safe to call this function.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Fixes: 5f01d95394 ("nfs41: create NFSv3 DS connection if specified")
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
The semantic patch that fixes this problem is as follows:
(http://coccinelle.lip6.fr/)
// <smpl>
@r@
type T;
identifier f;
@@
static T f (...) { ... }
@@
identifier r.f;
declarer name EXPORT_SYMBOL_GPL;
@@
-EXPORT_SYMBOL_GPL(f);
// </smpl>
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Pull btrfs fixes from Chris Mason:
"Outside of misc fixes, Filipe has a few fsync corners and we're
pulling in one more of Josef's fixes from production use here"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs:
Btrfs:__add_inode_ref: out of bounds memory read when looking for extended ref.
Btrfs: fix data loss in the fast fsync path
Btrfs: remove extra run_delayed_refs in update_cowonly_root
Btrfs: incremental send, don't rename a directory too soon
btrfs: fix lost return value due to variable shadowing
Btrfs: do not ignore errors from btrfs_lookup_xattr in do_setxattr
Btrfs: fix off-by-one logic error in btrfs_realloc_node
Btrfs: add missing inode update when punching hole
Btrfs: abort the transaction if we fail to update the free space cache inode
Btrfs: fix fsync race leading to ordered extent memory leaks
Pull file locking fix from Jeff Layton:
"Just a single patch to fix a memory leak that Daniel Wagner discovered
while doing some testing with leases"
* tag 'locks-v4.0-3' of git://git.samba.org/jlayton/linux:
locks: fix fasync_struct memory leak in lease upgrade/downgrade handling
Pull NFS client bugfixes from Trond Myklebust:
"Highlights include:
- Fix a regression in the NFSv4 open state recovery code
- Fix a regression in the NFSv4 close code
- Fix regressions and side-effects of the loop-back mounted NFS fixes
in 3.18, that cause the NFS read() syscall to return EBUSY.
- Fix regressions around the readdirplus code and how it interacts
with the VFS lazy unmount changes that went into v3.18.
- Fix issues with out-of-order RPC call replies replacing updated
attributes with stale ones (particularly after a truncate()).
- Fix an underflow checking issue with RPC/RDMA credits
- Fix a number of issues with the NFSv4 delegation return/free code.
- Fix issues around stale NFSv4.1 leases when doing a mount"
* tag 'nfs-for-4.0-3' of git://git.linux-nfs.org/projects/trondmy/linux-nfs: (24 commits)
NFSv4.1: Clear the old state by our client id before establishing a new lease
NFSv4: Fix a race in NFSv4.1 server trunking discovery
NFS: Don't write enable new pages while an invalidation is proceeding
NFS: Fix a regression in the read() syscall
NFSv4: Ensure we skip delegations that are already being returned
NFSv4: Pin the superblock while we're returning the delegation
NFSv4: Ensure we honour NFS_DELEGATION_RETURNING in nfs_inode_set_delegation()
NFSv4: Ensure that we don't reap a delegation that is being returned
NFS: Fix stateid used for NFS v4 closes
NFSv4: Don't call put_rpccred() under the rcu_read_lock()
NFS: Don't require a filehandle to refresh the inode in nfs_prime_dcache()
NFSv3: Use the readdir fileid as the mounted-on-fileid
NFS: Don't invalidate a submounted dentry in nfs_prime_dcache()
NFSv4: Set a barrier in the update_changeattr() helper
NFS: Fix nfs_post_op_update_inode() to set an attribute barrier
NFS: Remove size hack in nfs_inode_attrs_need_update()
NFSv4: Add attribute update barriers to delegreturn and pNFS layoutcommit
NFS: Add attribute update barriers to NFS writebacks
NFS: Set an attribute barrier on all updates
NFS: Add attribute update barriers to nfs_setattr_update_inode()
...
Improper arithmetics when calculting the address of the extended ref could
lead to an out of bounds memory read and kernel panic.
Signed-off-by: Quentin Casasnovas <quentin.casasnovas@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.cz>
cc: stable@vger.kernel.org # v3.7+
Signed-off-by: Chris Mason <clm@fb.com>
When using the fast file fsync code path we can miss the fact that new
writes happened since the last file fsync and therefore return without
waiting for the IO to finish and write the new extents to the fsync log.
Here's an example scenario where the fsync will miss the fact that new
file data exists that wasn't yet durably persisted:
1. fs_info->last_trans_committed == N - 1 and current transaction is
transaction N (fs_info->generation == N);
2. do a buffered write;
3. fsync our inode, this clears our inode's full sync flag, starts
an ordered extent and waits for it to complete - when it completes
at btrfs_finish_ordered_io(), the inode's last_trans is set to the
value N (via btrfs_update_inode_fallback -> btrfs_update_inode ->
btrfs_set_inode_last_trans);
4. transaction N is committed, so fs_info->last_trans_committed is now
set to the value N and fs_info->generation remains with the value N;
5. do another buffered write, when this happens btrfs_file_write_iter
sets our inode's last_trans to the value N + 1 (that is
fs_info->generation + 1 == N + 1);
6. transaction N + 1 is started and fs_info->generation now has the
value N + 1;
7. transaction N + 1 is committed, so fs_info->last_trans_committed
is set to the value N + 1;
8. fsync our inode - because it doesn't have the full sync flag set,
we only start the ordered extent, we don't wait for it to complete
(only in a later phase) therefore its last_trans field has the
value N + 1 set previously by btrfs_file_write_iter(), and so we
have:
inode->last_trans <= fs_info->last_trans_committed
(N + 1) (N + 1)
Which made us not log the last buffered write and exit the fsync
handler immediately, returning success (0) to user space and resulting
in data loss after a crash.
This can actually be triggered deterministically and the following excerpt
from a testcase I made for xfstests triggers the issue. It moves a dummy
file across directories and then fsyncs the old parent directory - this
is just to trigger a transaction commit, so moving files around isn't
directly related to the issue but it was chosen because running 'sync' for
example does more than just committing the current transaction, as it
flushes/waits for all file data to be persisted. The issue can also happen
at random periods, since the transaction kthread periodicaly commits the
current transaction (about every 30 seconds by default).
The body of the test is:
_scratch_mkfs >> $seqres.full 2>&1
_init_flakey
_mount_flakey
# Create our main test file 'foo', the one we check for data loss.
# By doing an fsync against our file, it makes btrfs clear the 'needs_full_sync'
# bit from its flags (btrfs inode specific flags).
$XFS_IO_PROG -f -c "pwrite -S 0xaa 0 8K" \
-c "fsync" $SCRATCH_MNT/foo | _filter_xfs_io
# Now create one other file and 2 directories. We will move this second file
# from one directory to the other later because it forces btrfs to commit its
# currently open transaction if we fsync the old parent directory. This is
# necessary to trigger the data loss bug that affected btrfs.
mkdir $SCRATCH_MNT/testdir_1
touch $SCRATCH_MNT/testdir_1/bar
mkdir $SCRATCH_MNT/testdir_2
# Make sure everything is durably persisted.
sync
# Write more 8Kb of data to our file.
$XFS_IO_PROG -c "pwrite -S 0xbb 8K 8K" $SCRATCH_MNT/foo | _filter_xfs_io
# Move our 'bar' file into a new directory.
mv $SCRATCH_MNT/testdir_1/bar $SCRATCH_MNT/testdir_2/bar
# Fsync our first directory. Because it had a file moved into some other
# directory, this made btrfs commit the currently open transaction. This is
# a condition necessary to trigger the data loss bug.
$XFS_IO_PROG -c "fsync" $SCRATCH_MNT/testdir_1
# Now fsync our main test file. If the fsync succeeds, we expect the 8Kb of
# data we wrote previously to be persisted and available if a crash happens.
# This did not happen with btrfs, because of the transaction commit that
# happened when we fsynced the parent directory.
$XFS_IO_PROG -c "fsync" $SCRATCH_MNT/foo
# Simulate a crash/power loss.
_load_flakey_table $FLAKEY_DROP_WRITES
_unmount_flakey
_load_flakey_table $FLAKEY_ALLOW_WRITES
_mount_flakey
# Now check that all data we wrote before are available.
echo "File content after log replay:"
od -t x1 $SCRATCH_MNT/foo
status=0
exit
The expected golden output for the test, which is what we get with this
fix applied (or when running against ext3/4 and xfs), is:
wrote 8192/8192 bytes at offset 0
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
wrote 8192/8192 bytes at offset 8192
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
File content after log replay:
0000000 aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa
*
0020000 bb bb bb bb bb bb bb bb bb bb bb bb bb bb bb bb
*
0040000
Without this fix applied, the output shows the test file does not have
the second 8Kb extent that we successfully fsynced:
wrote 8192/8192 bytes at offset 0
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
wrote 8192/8192 bytes at offset 8192
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
File content after log replay:
0000000 aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa
*
0020000
So fix this by skipping the fsync only if we're doing a full sync and
if the inode's last_trans is <= fs_info->last_trans_committed, or if
the inode is already in the log. Also remove setting the inode's
last_trans in btrfs_file_write_iter since it's useless/unreliable.
Also because btrfs_file_write_iter no longer sets inode->last_trans to
fs_info->generation + 1, don't set last_trans to 0 if we bail out and don't
bail out if last_trans is 0, otherwise something as simple as the following
example wouldn't log the second write on the last fsync:
1. write to file
2. fsync file
3. fsync file
|--> btrfs_inode_in_log() returns true and it set last_trans to 0
4. write to file
|--> btrfs_file_write_iter() no longers sets last_trans, so it
remained with a value of 0
5. fsync
|--> inode->last_trans == 0, so it bails out without logging the
second write
A test case for xfstests will be sent soon.
CC: <stable@vger.kernel.org>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
Commit 8634b51f6c (locks: convert lease handling to file_lock_context)
introduced a regression in the handling of lease upgrade/downgrades.
In the event that we already have a lease on a file and are going to
either upgrade or downgrade it, we skip doing any list insertion or
deletion and simply re-call lm_setup on the existing lease.
As of commit 8634b51f6c however, we end up calling lm_setup on the
lease that was passed in, instead of on the existing lease. This causes
us to leak the fasync_struct that was allocated in the event that there
was not already an existing one (as it always appeared that there
wasn't one).
Fixes: 8634b51f6c (locks: convert lease handling to file_lock_context)
Reported-and-Tested-by: Daniel Wagner <daniel.wagner@bmw-carit.de>
Signed-off-by: Jeff Layton <jeff.layton@primarydata.com>
Pull eCryptfs fixes from Tyler Hicks:
"Fixes for proper ioctl handling and an untriggerable buffer overflow
- The eCryptfs ioctl handling functions should only pass known-good
ioctl commands to the lower filesystem
- A static checker found a potential buffer overflow. Upon
inspection, it is not triggerable due to input validation performed
on the mount parameters"
* tag 'ecryptfs-4.0-rc3-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tyhicks/ecryptfs:
eCryptfs: don't pass fs-specific ioctl commands through
eCryptfs: ensure copy to crypt_stat->cipher does not overrun
Currently quota format that supports 64-bit usage sets maximum quota
limit as 2^64-1. However quota core code uses signed numbers to track
usage and even limits themselves are stored in long long. Checking of
maximum allowable limits worked by luck until commit 14bf61ffe6
(quota: Switch ->get_dqblk() and ->set_dqblk() to use bytes as space
units) because variable we compared with was unsigned. After that commit
the type we compared against changed to signed and thus checks for
maximum limits with the newest VFS quota format started to refuse any
non-negative value. Later the problem was inadvertedly fixed by commit
b10a08194c (quota: Store maximum space limit in bytes) because we
started to compare against unsigned type as well.
Fix possible future problems of this kind by setting maximum limits to
2^63-1 to avoid overflow issues.
Reported-by: Carlos Carvalho <carlos@fisica.ufpr.br>
Signed-off-by: Jan Kara <jack@suse.cz>
Root level in quota tree must be already allocated otherwise this block
could be used for something else.
Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Signed-off-by: Jan Kara <jack@suse.cz>
Setting timers or warning counts for id 0 via Q_XSETQLIM is used to
actually set time limits and warning limits for all users. Hook up
->set_info to this so that VFS quota time limits get set the same
way as XFS ones.
When doing this Q_XSETQLIM for XFS is effectively split into two
independent transactions - one for setting timers and warning limits and
one for setting space and inode limits. Although this is inefficient, it
is rare enough that it does not matter.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Add support to XFS so that time limits can be set through Q_SETINFO
quotactl.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Change ->set_info to take new qc_info structure which contains all the
necessary information both for XFS and VFS. Convert Q_SETINFO handler
to use this structure.
Signed-off-by: Jan Kara <jack@suse.cz>
Convert xfs to use ->get_state callback instead of ->get_xstate and
->get_xstatev.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Add appropriate conversion functions so that filesystems supporting
->get_state() method can be queried using Q_GETXSTATE and Q_GETXSTATV
calls.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Create new internal interface for getting information about quota which
contains everything needed for both VFS quotas and XFS quotas. Make VFS
use this and hook it up to Q_GETINFO.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
If the call to exchange-id returns with the EXCHGID4_FLAG_CONFIRMED_R flag
set, then that means our lease was established by a previous mount instance.
Ensure that we detect this situation, and that we clear the state held by
that mount.
Reported-by: Jorge Mora <Jorge.Mora@netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
We do not want to allow a race with another NFS mount to cause
nfs41_walk_client_list() to establish a lease on our nfs_client before
we're done checking for trunking.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Pull nfsd fixes from Bruce Fields:
"Three miscellaneous bugfixes, most importantly the clp->cl_revoked
bug, which we've seen several reports of people hitting"
* 'for-4.0' of git://linux-nfs.org/~bfields/linux:
sunrpc: integer underflow in rsc_parse()
nfsd: fix clp->cl_revoked list deletion causing softlock in nfsd
svcrpc: fix memory leak in gssp_accept_sec_context_upcall
nfs_vm_page_mkwrite() should wait until the page cache invalidation
is finished. This is the second patch in a 2 patch series to deprecate
the NFS client's reliance on nfs_release_page() in the context of
nfs_invalidate_mapping().
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
When invalidating the page cache for a regular file, we want to first
sync all dirty data to disk and then call invalidate_inode_pages2().
The latter relies on nfs_launder_page() and nfs_release_page() to deal
respectively with dirty pages, and unstable written pages.
When commit 9590544694 ("NFS: avoid deadlocks with loop-back mounted
NFS filesystems.") changed the behaviour of nfs_release_page(), then it
made it possible for invalidate_inode_pages2() to fail with an EBUSY.
Unfortunately, that error is then propagated back to read().
Let's therefore work around the problem for now by protecting the call
to sync the data and invalidate_inode_pages2() so that they are atomic
w.r.t. the addition of new writes.
Later on, we can revisit whether or not we still need nfs_launder_page()
and nfs_release_page().
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
During the roll-forward recovery, -ENOENT for f2fs_iget can be skipped.
So, this error value should not be propagated.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
We update extent cache for all user inode of f2fs including dir inode, so this
patch gives another chance to try to get physical address of page from extent
cache for dir inode.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch switch to check FI_NO_EXTENT in f2fs_{lookup,update}_extent_cache
instead of f2fs_{lookup,update}_extent_tree or {lookup,update}_extent_info.
No functionality modification in this patch.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds a fast lookup path for rb-tree extent cache.
In this patch we add a recently accessed extent node pointer 'cached_en' in
extent tree. In lookup path of extent cache, we will firstly lookup the last
accessed extent node which cached_en points, if we do not hit in this node,
we will try to lookup extent node in rb-tree.
By this way we can avoid unnecessary slow lookup in rb-tree sometimes.
Note that, side-effect of this patch is that we will increase memory cost,
because we will store a pointer variable in each struct extent tree
additionally.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds trace for lookup/update/shrink/destroy ops in rb-tree extent cache.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch add and show stat info of total memory footprint for extent tree,node
in debugfs.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch enables rb-tree based extent cache in f2fs.
When we mount with "-o extent_cache", f2fs will try to add recently accessed
page-block mappings into rb-tree based extent cache as much as possible, instead
of original one extent info cache.
By this way, f2fs can support more effective cache between dnode page cache and
disk. It will supply high hit ratio in the cache with fewer memory when dnode
page cache are reclaimed in environment of low memory.
Storage: Sandisk sd card 64g
1.append write file (offset: 0, size: 128M);
2.override write file (offset: 2M, size: 1M);
3.override write file (offset: 4M, size: 1M);
...
4.override write file (offset: 48M, size: 1M);
...
5.override write file (offset: 112M, size: 1M);
6.sync
7.echo 3 > /proc/sys/vm/drop_caches
8.read file (size:128M, unit: 4k, count: 32768)
(time dd if=/mnt/f2fs/128m bs=4k count=32768)
Extent Hit Ratio:
before patched
Hit Ratio 121 / 1071 1071 / 1071
Performance:
before patched
real 0m37.051s 0m35.556s
user 0m0.040s 0m0.026s
sys 0m2.990s 0m2.251s
Memory Cost:
before patched
Tree Count: 0 1 (size: 24 bytes)
Node Count: 0 45 (size: 1440 bytes)
v3:
o retest and given more details of test result.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds a mount option 'extent_cache' in f2fs.
It is try to use a rb-tree based extent cache to cache more mapping information
with less memory if this option is set, otherwise we will use the original one
extent info cache.
Suggested-by: Changman Lee <cm224.lee@samsung.com>
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds core functions including slab cache init function and
init/lookup/update/shrink/destroy function for rb-tree based extent cache.
Thank Jaegeuk Kim and Changman Lee as they gave much suggestion about detail
design and implementation of extent cache.
Todo:
* register rb-based extent cache shrink with mm shrink interface.
v2:
o move set_extent_info and __is_{extent,back,front}_mergeable into f2fs.h.
o introduce __{attach,detach}_extent_node for code readability.
o add cond_resched() when fail to invoke kmem_cache_alloc/radix_tree_insert.
o fix some coding style and typo issues.
v3:
o fix oops due to using an unassigned pointer.
o use list_del to remove extent node in shrink list.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Changman Lee <cm224.lee@samsung.com>
[Jaegeuk Kim: add static for some funcitons and declare in f2fs.h]
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Introduce infra macro and data structure for rb-tree based extent cache:
Macros:
* EXT_TREE_VEC_SIZE: indicate vector size for gang lookup in extent tree.
* F2FS_MIN_EXTENT_LEN: indicate minimum length of extent managed in cache.
* EXTENT_CACHE_SHRINK_NUMBER: indicate number of extent in cache will be shrunk.
Basic data structures for extent cache:
* struct extent_tree: extent tree entry per inode.
* struct extent_node: extent info node linked in extent tree.
Besides, adding new extent cache related fields in f2fs_sb_info.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Changman Lee <cm224.lee@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In this patch, we do these jobs:
1. rename {check,update}_extent_cache to {lookup,update}_extent_info;
2. introduce universal lookup/update interface of extent cache:
f2fs_{lookup,update}_extent_cache including above two real functions, then
export them to function callers.
So after above cleanup, we can add new rb-tree based extent cache into exported
interfaces.
v2:
o remove "f2fs_" for inner function {lookup,update}_extent_info suggested by
Jaegeuk Kim.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
