The EFI is initialized with a reference count of 2. One for the EFI to
ensure the item makes it to the AIL and one for the subsequently created
EFD to release the EFI once the EFD is committed. Log recovery uses the
EFI in a similar manner, but implements a hack to remove both references
in one call once the EFD is handled.
Update log recovery to use EFI reference counting in a manner consistent
with the log. When an EFI is encountered during recovery, an EFI item is
allocated and inserted to the AIL directly. Since the EFI reference is
typically dropped when the EFI is unpinned and this is analogous with
AIL insertion, drop the EFI reference at this point.
When a corresponding EFD is encountered in the log, this indicates that
the extents were freed, no processing is required and the EFI can be
dropped. Update xlog_recover_efd_pass2() to simply drop the EFD
reference at this point rather than open code the AIL removal and EFI
free.
Remaining EFIs (i.e., with no corresponding EFD) are processed in
xlog_recover_finish(). An EFD transaction is allocated and the extents
are freed, which transfers ownership of the EFI reference to the EFD
item in the log.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Log recovery attempts to free extents with leftover EFIs in the AIL
after initial processing. If the extent free fails (e.g., due to
unrelated fs corruption), the transaction is cancelled, though it
might not be dirtied at the time. If this is the case, the EFD does
not abort and thus does not release the EFI. This can lead to hangs
as the EFI pins the AIL.
Update xlog_recover_process_efi() to log the EFD in the transaction
before xfs_free_extent() errors are handled to ensure the
transaction is dirty, aborts the EFD and releases the EFI on error.
Since this is a requirement for EFD processing (and consistent with
xfs_bmap_finish()), update the EFD logging helper to do the extent
free and unconditionally log the EFD. This encodes the required EFD
logging behavior into the helper and reduces the likelihood of
errors down the road.
[dchinner: re-add xfs_alloc.h to xfs_log_recover.c to fix build
failure.]
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Freeing an extent in XFS involves logging an EFI (extent free
intention), freeing the actual extent, and logging an EFD (extent
free done). The EFI object is created with a reference count of 2:
one for the current transaction and one for the subsequently created
EFD. Under normal circumstances, the first reference is dropped when
the EFI is unpinned and the second reference is dropped when the EFD
is committed to the on-disk log.
In event of errors or filesystem shutdown, there are various
potential cleanup scenarios depending on the state of the EFI/EFD.
The cleanup scenarios are confusing and racy, as demonstrated by the
following test sequence:
# mount $dev $mnt
# fsstress -d $mnt -n 99999 -p 16 -z -f fallocate=1 \
-f punch=1 -f creat=1 -f unlink=1 &
# sleep 5
# killall -9 fsstress; wait
# godown -f $mnt
# umount
... in which the final umount can hang due to the AIL being pinned
indefinitely by one or more EFI items. This can occur due to several
conditions. For example, if the shutdown occurs after the EFI is
committed to the on-disk log and the EFD committed to the CIL, but
before the EFD committed to the log, the EFD iop_committed() abort
handler does not drop its reference to the EFI. Alternatively,
manual error injection in the xfs_bmap_finish() codepath shows that
if an error occurs after the EFI transaction is committed but before
the EFD is constructed and logged, the EFI is never released from
the AIL.
Update the EFI/EFD item handling code to use a more straightforward
and reliable approach to error handling. If an error occurs after
the EFI transaction is committed and before the EFD is constructed,
release the EFI explicitly from xfs_bmap_finish(). If the EFI
transaction is cancelled, release the EFI in the unlock handler.
Once the EFD is constructed, it is responsible for releasing the EFI
under any circumstances (including whether the EFI item aborts due
to log I/O error). Update the EFD item handlers to release the EFI
if the transaction is cancelled or aborts due to log I/O error.
Finally, update xfs_bmap_finish() to log at least one EFD extent to
the transaction before xfs_free_extent() errors are handled to
ensure the transaction is dirty and EFD item error handling is
triggered.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Some callers need to make error handling decisions based on whether
the current transaction successfully committed or not. Rename
xfs_trans_roll(), add a new parameter and provide a wrapper to
preserve existing callers.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Release of the EFI either occurs based on the reference count or the
extent count. The extent count used is either the count tracked in
the EFI or EFD, depending on the particular situation. In either
case, the count is initialized to the final value and thus always
matches the current efi_next_extent value once the EFI is completely
constructed. For example, the EFI extent count is increased as the
extents are logged in xfs_bmap_finish() and the full free list is
always completely processed. Therefore, the count is guaranteed to
be complete once the EFI transaction is committed. The EFD uses the
efd_nextents counter to release the EFI. This counter is initialized
to the count of the EFI when the EFD is created. Thus the EFD, as
currently used, has no concept of partial EFI release based on
extent count.
Given that the EFI extent count is always released in whole, use of
the extent count for reference counting is unnecessary. Remove this
level of the API and release the EFI based on the core reference
count. The efi_next_extent counter remains because it is still used
to track the slot to log the next extent to free.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.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>
Add a function to determine the path length of a kernfs node. This
for now will be used by writeback tracepoint updates.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
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>
The key_put() function tests whether its argument is NULL and then
returns immediately. Thus the test around this call might not be needed.
This issue was detected by using the Coccinelle software.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Consider eCryptfs dcache entries to be stale when the corresponding
lower inode's i_nlink count is zero. This solves a problem caused by the
lower inode being directly modified, without going through the eCryptfs
mount, leaving stale eCryptfs dentries cached and the eCryptfs inode's
i_nlink count not being cleared.
Signed-off-by: Tyler Hicks <tyhicks@canonical.com>
Reported-by: Richard Weinberger <richard@nod.at>
Cc: stable@vger.kernel.org
On a box with a lot of ram (148gb) I can make the box softlockup after running
an fs_mark job that creates hundreds of millions of empty files. This is
because we never generate enough memory pressure to keep the number of inodes on
our unused list low, so when we go to unmount we have to evict ~100 million
inodes. This makes one processor a very unhappy person, so add a cond_resched()
in dispose_list() and if we need a resched when processing the s_inodes list do
that and run dispose_list() on what we've currently culled. Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Jan Kara <jack@suse.cz>
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>
generic_file_write_iter() will already do an fsync on our behalf
if the file descriptor is O_SYNC or the file is marked as IS_SYNC.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
There are cases on which lowcomms_connect_sock() is called directly,
which caused the CF_WRITE_PENDING flag to not bet set upon reconnect,
specially on send_to_sock() error handling. On this last, the flag was
already cleared and no further attempt on transmitting would be done.
As dlm tends to connect when it needs to transmit something, it makes
sense to always mark this flag right after the connect.
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David Teigland <teigland@redhat.com>
BUG_ON() is a severe action for this case, specially now that DLM with
SCTP will use 1 socket per association. Instead, we can just close the
socket on this error condition and return from the function.
Also move the check to an earlier stage as it won't change and thus we
can abort as soon as possible.
Although this issue was reported when still using SCTP with 1-to-many
API, this cleanup wouldn't be that simple back then because we couldn't
close the socket and making sure such event would cease would be hard.
And actually, previous code was closing the association, yet SCTP layer
is still raising the new data event. Probably a bug to be fixed in SCTP.
Reported-by: <tan.hu@zte.com.cn>
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David Teigland <teigland@redhat.com>
DLM is using 1-to-many API but in a 1-to-1 fashion. That is, it's not
needed but this causes it to use sctp_do_peeloff() to mimic an
kernel_accept() and this causes a symbol dependency on sctp module.
By switching it to 1-to-1 API we can avoid this dependency and also
reduce quite a lot of SCTP-specific code in lowcomms.c.
The caveat is that now DLM won't always use the same src port. It will
choose a random one, just like TCP code. This allows the peers to
attempt simultaneous connections, which now are handled just like for
TCP.
Even more sharing between TCP and SCTP code on DLM is possible, but it
is intentionally left for a later commit.
Note that for using nodes with this commit, you have to have at least
the early fixes on this patchset otherwise it will trigger some issues
on old nodes.
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David Teigland <teigland@redhat.com>
If we don't clear that bit, lowcomms_connect_sock() will not schedule
another attempt, and no further attempt will be done.
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David Teigland <teigland@redhat.com>
When a connection have issues DLM may need to close it. Therefore we
should also cancel pending workqueues for such connection at that time,
and not just when dlm is not willing to use this connection anymore.
Also, if we don't clear CF_CONNECT_PENDING flag, the error handling
routines won't be able to re-connect as lowcomms_connect_sock() will
check for it.
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David Teigland <teigland@redhat.com>
When using SCTP and accepting a new connection, DLM currently validates
if the peer trying to connect to it is one of the cluster nodes, but it
doesn't check if it already has a connection to it or not.
If it already had a connection, it will be overwritten, and the new one
will be used for writes, possibly causing the node to leave the cluster
due to communication breakage.
Still, one could DoS the node by attempting N connections and keeping
them open.
As said, but being explicit, both situations are only triggerable from
other cluster nodes, but are doable with only user-level perms.
Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David Teigland <teigland@redhat.com>
Chuck reports seeing cases where a GETATTR that happens to race
with an asynchronous WRITE is overriding the file size, despite
the attribute barrier being set by the writeback code.
The culprit turns out to be the check in nfs_ctime_need_update(),
which sees that the ctime is newer than the cached ctime, and
assumes that it is safe to override the attribute barrier.
This patch removes that override, and ensures that attribute
barriers are always respected.
Reported-by: Chuck Lever <chuck.lever@oracle.com>
Fixes: a08a8cd375 ("NFS: Add attribute update barriers to NFS writebacks")
Cc: stable@vger.kernel.org # v4.0+
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
* bugfixes:
SUNRPC: Fix a thinko in xs_connect()
NFSv4.1/pNFS: Fix borken function _same_data_server_addrs_locked()
NFS: nfs_set_pgio_error sometimes misses errors
NFS: NFS over RDMA Client Side Changes
These patches improve both client performance and scalability, most notably
by increasing the maixmum allowed rsize and wsize and by increasing the number
of RDMA "credits". There are also several bugfixes, such as correcting how
WRITE compounds are encoded and fixing large NFS symlink operations.
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
And call nfs_file_clear_open_context() directly. This makes it obvious
that nfs_file_release() will always return 0.
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
All nfs_write_inode() does is pass its arguments to
nfs_commit_unstable_pages(). Let's cut out the middle man and have
nfs_write_pages() do the work directly.
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
All these functions do is call nfs41_ping_server() without adding
anything. Let's remove them and give nfs41_ping_server() a better name
instead.
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
The idmap_init() and idmap_quit() functions only exist to call the
_keyring() version. Let's just call the keyring() functions directly.
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
nfs_readdir_xdr_to_array() uses both a cache array and an array of
pages, so I rename these functions to make it clearer how the code
works. nfs_readdir_large_page() becomes nfs_readdir_alloc_pages()
because this function has absolutely nothing to do with setting up a
large page. nfs_readdir_free_pagearray() becomes
nfs_readdir_free_pages() to stay consistent with the new alloc_pages()
function.
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
This variable is initialized to NULL and is never modified before being
passed to nfs_readdir_free_large_page(). But that's okay, because
nfs_readdir_free_large_page() only seems to exist as a way of calling
nfs_readdir_free_pagearray() without this parameter. Let's simplify by
removing pages_ptr and nfs_readdir_free_pagearray().
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
We generally want to read and write to a block device that's used by
the pNFS block layout client (and even if it's read only the server
has no way of telling us). Add FMODE_WRITE to the mode argument
so that we don't incorrectly tell the block driver that we want a
read-only open.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Instead of overwriting kernel memory reject too long signatures.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
We need to replace the __be32 with a void pointer to do proper arithmentics
on the virtual addresses so that we can get the right page pointers.
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
We need to include the first u32 for the number of entries. Add a helper
for the calculation instead of opencoding it so that it's in one place.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
- Switch back to using list_for_each_entry(). Fixes an incorrect test
for list NULL termination.
- Do not assume that lists are sorted.
- Finally, consider an existing entry to match if it consists of a subset
of the addresses in the new entry.
Cc: stable@vger.kernel.org # 4.0+
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
We should ensure that we always set the pgio_header's error field
if a READ or WRITE RPC call returns an error. The current code depends
on 'hdr->good_bytes' always being initialised to a large value, which
is not always done correctly by callers.
When this happens, applications may end up missing important errors.
Cc: stable@vger.kernel.org
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
fuse_dev_ioctl() performed fuse_get_dev() on a user-supplied fd,
leading to a type confusion issue. Fix it by checking file->f_op.
Signed-off-by: Jann Horn <jann@thejh.net>
Acked-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The xfstests ext4/305 will mount and unmount the same file system over
4,000 times, and each one of these will cause a system log message.
Ratelimit this message since if we are getting more than a few dozen
of these messages, they probably aren't going to be helpful.
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Static checkers complain that the format string should be "%s". It does
not make a difference for the current code.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
My static check complains because we have:
if (!*bh)
return -ENOMEM;
if (*bh) {
The second check is unnecessary.
I've simplified this code by moving the "if (!*bh)" checks around. Also
Andreas Dilger says we should probably print a warning if sb_getblk()
fails.
[ Restructured the code so that we print a warning message as well if
the mmp block doesn't check out, and to print the error code to
disambiguate between the error cases. - TYT ]
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
