Replace i_pin_wait, which is only used during synchronous inode flushing
with a bit waitqueue.  This trades off a much smaller inode against
slightly slower wakeup performance, and saves 12 (32-bit) or 20 (64-bit)
bytes in the XFS inode.
Reviewed-by: NAlex Elder <aelder@sgi.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NBen Myers <bpm@sgi.com>

We almost never block on i_flock, the exception is synchronous inode
flushing.  Instead of bloating the inode with a 16/24-byte completion
that we abuse as a semaphore just implement it as a bitlock that uses
a bit waitqueue for the rare sleeping path.  This primarily is a
tradeoff between a much smaller inode and a faster non-blocking
path vs faster wakeups, and we are much better off with the former.

A small downside is that we will lose lockdep checking for i_flock, but
given that it's always taken inside the ilock that should be acceptable.

Note that for example the inode writeback locking is implemented in a
very similar way.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NAlex Elder <aelder@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

To be used for bit wakeup i_flags needs to be an unsigned long or we'll
run into trouble on big endian systems.  Because of the 1-byte i_update
field right after it this actually causes a fairly large size increase
on its own (4 or 8 bytes), but that increase will be more than offset
by the next two patches.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NAlex Elder <aelder@sgi.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

We spent a lot of effort to maintain this field, but it always equals to the
fork size divided by the constant size of an extent.  The prime use of it is
to assert that the two stay in sync.  Just divide the fork size by the extent
size in the few places that we actually use it and remove the overhead
of maintaining it.  Also introduce a few helpers to consolidate the places
where we actually care about the value.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

.. and the just as dead bhv_desc forward declaration while we're at it.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NAlex Elder <aelder@sgi.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Replace the nasty if, else if, elseif condition with more natural C flow
that expressed the logic we want here better.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

This wrapper isn't overly useful, not to say rather confusing.

Around the call to xfs_itruncate_extents it does:

 - add tracing
 - add a few asserts in debug builds
 - conditionally update the inode size in two places
 - log the inode

Both the tracing and the inode logging can be moved to xfs_itruncate_extents
as they are useful for the attribute fork as well - in fact the attr code
already does an equivalent xfs_trans_log_inode call just after calling
xfs_itruncate_extents.  The conditional size updates are a mess, and there
was no reason to do them in two places anyway, as the first one was
conditional on the inode having extents - but without extents we
xfs_itruncate_extents would be a no-op and the placement wouldn't matter
anyway.  Instead move the size assignments and the asserts that make sense
to the callers that want it.

As a side effect of this clean up xfs_setattr_size by introducing variables
for the old and new inode size, and moving the size updates into a common
place.
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NBen Myers <bpm@sgi.com>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

vfs_create() ignores everything outside of 16bit subset of its
mode argument; switching it to umode_t is obviously equivalent
and it's the only caller of the method
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

vfs_mkdir() gets int, but immediately drops everything that might not
fit into umode_t and that's the only caller of ->mkdir()...
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Seeing that just about every destructor got that INIT_LIST_HEAD() copied into
it, there is no point whatsoever keeping this INIT_LIST_HEAD in inode_init_once();
the cost of taking it into inode_init_always() will be negligible for pipes
and sockets and negative for everything else.  Not to mention the removal of
boilerplate code from ->destroy_inode() instances...
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

new helper (wrapper around mnt_drop_write()) to be used in pair with
mnt_want_write_file().
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

it's both faster (in case when file has been opened for write) and cleaner.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

When finding the longest extent in an AG, we read the value directly
out of the AGF buffer without endian conversion. This will give an
incorrect length, resulting in FITRIM operations potentially not
trimming everything that it should.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NBen Myers <bpm@sgi.com>

Since Linux 2.6.36 the writeback code has introduces various measures for
live lock prevention during sync().  Unfortunately some of these are
actively harmful for the XFS model, where the inode gets marked dirty for
metadata from the data I/O handler.

The older_than_this checks that are now more strictly enforced since

    writeback: avoid livelocking WB_SYNC_ALL writeback

by only calling into __writeback_inodes_sb and thus only sampling the
current cut off time once.  But on a slow enough devices the previous
asynchronous sync pass might not have fully completed yet, and thus XFS
might mark metadata dirty only after that sampling of the cut off time for
the blocking pass already happened.  I have not myself reproduced this
myself on a real system, but by introducing artificial delay into the
XFS I/O completion workqueues it can be reproduced easily.

Fix this by iterating over all XFS inodes in ->sync_fs and log all that
are dirty.  This might log inode that only got redirtied after the
previous pass, but given how cheap delayed logging of inodes is it
isn't a major concern for performance.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Tested-by: NMark Tinguely <tinguely@sgi.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

If the writeback code writes back an inode because it has expired we currently
use the non-blockin ->write_inode path.  This means any inode that is pinned
is skipped.  With delayed logging and a workload that has very little log
traffic otherwise it is very likely that an inode that gets constantly
written to is always pinned, and thus we keep refusing to write it.  The VM
writeback code at that point redirties it and doesn't try to write it again
for another 30 seconds.  This means under certain scenarious time based
metadata writeback never happens.

Fix this by calling into xfs_log_inode for kupdate in addition to data
integrity syncs, and thus transfer the inode to the log ASAP.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Tested-by: NMark Tinguely <tinguely@sgi.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

On a system with lots of memory pressure that is stuck on synchronous inode
reclaim the workqueue code will run one instance of the inode reclaim work
item on every CPU. which is not what we want.  Make sure to mark the
xfssyncd workqueue as non-reentrant to make sure there only is one instace
of each running globally.  Also stop using special paramater for the
workqueue; now that we guarantee each fs has only running one of each works
at a time there is no need to artificially lower max_active and compensate
for that by setting the WQ_CPU_INTENSIVE flag.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

There is no reason to drop qi_dqlist_lock around calls to xfs_qm_dqrele
because the free list lock now nests inside qi_dqlist_lock and the
dquot lock.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Commit fa8b18ed didn't prevent the integer overflow and possible
memory corruption.  "count" can go negative and bypass the check.
Signed-off-by: NXi Wang <xi.wang@gmail.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NBen Myers <bpm@sgi.com>

XBT_FORCE_SLEEP is no longer ever tested; it is only set
and cleared.  Remove it.
Signed-off-by: NEric Sandeen <sandeen@sandeen.net>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Just read the id 0 dquot from disk directly in xfs_qm_init_quotainfo instead
of going through dqget and requiring a special flag to not add the dquot to
any lists.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

This function doesn't help the code flow, so merge the dquot allocation and
transaction handling into xfs_qm_dqread.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Factor the common pattern of:

	xfs_dqlock(dqp);
	XFS_DQHOLD(dqp);
	xfs_dqunlock(dqp);

into a new helper, and remove XFS_DQHOLD now that only one other caller
is left.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

No need to play games with the qlock now that the freelist lock nests inside
it.  Also clean up various outdated comments.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Allow xfs_qm_dqput to work without trylock loops by nesting the freelist lock
inside the dquot qlock.  In turn that requires trylocks in the reclaim path
instead, but given it's a classic tradeoff between fast and slow path, and
we follow the model of the inode and dentry caches.

Document our new lock order now that it has settled.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Introduce a new XFS_DQ_FREEING flag that tells lookup and mplist walks
to skip a dquot that is beeing freed, and use this avoid the trylock
on the hash and mplist locks in xfs_qm_dqreclaim_one.  Also simplify
xfs_dqpurge by moving the inodes to a dispose list after marking them
XFS_DQ_FREEING and avoid the locker ordering constraints.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Do not remove dquots from the freelist when we grab a reference to them in
xfs_qm_dqlookup, but leave them on the freelist util scanning notices that
they have a reference.  This speeds up the lookup fastpath, and greatly
simplifies the lock ordering constraints.  Note that the same scheme is
used by the VFS inode and dentry caches.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Free dquots when purging them during umount instead of keeping them around
on the freelist in a degraded state.  The out of order locking in
xfs_qm_dqpurge will be removed again later in this series.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Rearrange the code to avoid the conditional locking around the flist_locked
variable.  This means we lose a (rather pointless) assert, and hold the
freelist lock a bit longer for one corner case.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Mark the trivial lock wrappers as inline, and make the naming consistent
for all of them.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

It always is zero, and removing it will make future changes easier.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Now that we can't have any dirty dquots around that aren't in the AIL we
can get rid of the explicit dquot syncing from xfssyncd and xfs_fs_sync_fs
and instead rely on AIL pushing to write out any quota updates.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Make sure we do not skip any dquots when flushing them out after a
quotacheck to make sure that we will never have any dirty dquots on a
live filesystem.  At this point no dquot should be pinnable, but lets
be pedantic about it.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Only skip pinned dquots if SYNC_TRYLOCK is specified, and adjust the callers
to keep the behaviour unchanged.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Outside the now removed nodelaylog code this field is only used for
asserts and can be safely removed now.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Now that the nodelaylog mode is gone we can simplify the transaction commit
path a bit by removing the xfs_trans_commit_cil routine.  Restoring the
process flags is merged into xfs_trans_commit which already does it for
the error path, and allocating the log vectors is merged into
xlog_cil_format_items, which already fills them with data, thus avoiding
one loop over all log items.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

The delaylog mode has been the default for a long time, and the nodelaylog
option has been scheduled for removal in Linux 3.3.  Remove it and code
only used by it now that we have opened the 3.3 window.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>