xfs_trans_ail_cursor_set() doesn't set the cursor to the current log
item, it sets it to the next item. There is already a function for
doing this - xfs_trans_ail_cursor_next() - and the _set function is
simply a two line wrapper. Remove it and open code the setting of
the cursor in the two locations that call it to remove the
confusion.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
Delayed logging can insert tens of thousands of log items into the
AIL at the same LSN. When the committing of log commit records
occur, we can get insertions occurring at an LSN that is not at the
end of the AIL. If there are thousands of items in the AIL on the
tail LSN, each insertion has to walk the AIL to find the correct
place to insert the new item into the AIL. This can consume large
amounts of CPU time and block other operations from occurring while
the traversals are in progress.
To avoid this repeated walk, use a AIL cursor to record
where we should be inserting the new items into the AIL without
having to repeat the walk. The cursor infrastructure already
provides this functionality for push walks, so is a simple extension
of existing code. While this will not avoid the initial walk, it
will avoid repeating it tens of thousands of times during a single
checkpoint commit.
This version includes logic improvements from Christoph Hellwig.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
The recent conversion of the xfsaild functionality to a work queue
introduced a hard-to-hit log space grant hang. One is caused by a
race condition in determining whether there is a psh in progress or
not.
The XFS_AIL_PUSHING_BIT is used to determine whether a push is
currently in progress. When the AIL push work completes, it checked
whether the target changed and cleared the PUSHING bit to allow a
new push to be requeued. The race condition is as follows:
Thread 1 push work
smp_wmb()
smp_rmb()
check ailp->xa_target unchanged
update ailp->xa_target
test/set PUSHING bit
does not queue
clear PUSHING bit
does not requeue
Now that the push target is updated, new attempts to push the AIL
will not trigger as the push target will be the same, and hence
despite trying to push the AIL we won't ever wake it again.
The fix is to ensure that the AIL push work clears the PUSHING bit
before it checks if the target is unchanged.
As a result, both push triggers operate on the same test/set bit
criteria, so even if we race in the push work and miss the target
update, the thread requesting the push will still set the PUSHING
bit and queue the push work to occur. For safety sake, the same
queue check is done if the push work detects the target change,
though only one of the two will will queue new work due to the use
of test_and_set_bit() checks.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
(cherry picked from commit e4d3c4a43b)
The recent conversion of the xfsaild functionality to a work queue
introduced a hard-to-hit log space grant hang. One of the problems
noticed was that updates of the push target are not 32 bit safe as
the target is a 64 bit value.
We cannot copy a 64 bit LSN without the possibility of corrupting
the result when racing with another updating thread. We have
function to do this update safely without needing to care about
32/64 bit issues - xfs_trans_ail_copy_lsn() - so use that when
updating the AIL push target.
Also move the reading of the target in the push work inside the AIL
lock, and use XFS_LSN_CMP() for the unlocked comparison during work
termination to close read holes as well.
Signed-off-by: Dave Chinner <david@fromorbit.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
(cherry picked from commit fd5670f22f)
The recent conversion of the xfsaild functionality to a work queue
introduced a hard-to-hit log space grant hang. One of the problems
discovered is a target mismatch between the item pushing loop and
the target itself.
The push trigger checks for the target increasing (i.e. new target >
current) while the push loop only pushes items that have a LSN <
current. As a result, we can get the situation where the push target
is X, the items at the tail of the AIL have LSN X and they don't get
pushed. The push work then completes thinking it is done, and cannot
be restarted until the push target increases to >= X + 1. If the
push target then never increases (because the tail is not moving),
then we never run the push work again and we stall.
Fix it by making sure log items with a LSN that matches the target
exactly are pushed during the loop.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
(cherry picked from commit cb64026b6e)
The recent conversion of the xfsaild functionality to a work queue
introduced a hard-to-hit log space grant hang. The main cause is a
regression where a work exit path fails to clear the PUSHING state
and recheck the target correctly.
Make both exit paths do the same PUSHING bit clearing and target
checking when the "no more work to be done" condition is hit.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
(cherry picked from commit ea35a20021)
When we are short on memory, we want to expedite the cleaning of
dirty objects. Hence when we run short on memory, we need to kick
the AIL flushing into action to clean as many dirty objects as
quickly as possible. To implement this, sample the lsn of the log
item at the head of the AIL and use that as the push target for the
AIL flush.
Further, we keep items in the AIL that are dirty that are not
tracked any other way, so we can get objects sitting in the AIL that
don't get written back until the AIL is pushed. Hence to get the
filesystem to the idle state, we might need to push the AIL to flush
out any remaining dirty objects sitting in the AIL. This requires
the same push mechanism as the reclaim push.
This patch also renames xfs_trans_ail_tail() to xfs_ail_min_lsn() to
match the new xfs_ail_max_lsn() function introduced in this patch.
Similarly for xfs_trans_ail_push -> xfs_ail_push.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Alex Elder <aelder@sgi.com>
This patch rearranges the location of functions in xfs_trans_ail.c
to remove the need for forward declarations of those functions in
preparation for adding new functions without the need for forward
declarations.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Alex Elder <aelder@sgi.com>
Similar to the xfssyncd, the per-filesystem xfsaild threads can be
converted to a global workqueue and run periodically by delayed
works. This makes sense for the AIL pushing because it uses
variable timeouts depending on the work that needs to be done.
By removing the xfsaild, we simplify the AIL pushing code and
remove the need to spread the code to implement the threading
and pushing across multiple files.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Alex Elder <aelder@sgi.com>
Continue the conversion of the old cmn_err interface be converting
all the conditional panic tag errors to xfs_alert_tag() and then
removing xfs_cmn_err().
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Alex Elder <aelder@sgi.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
We now have two copies of AIL delete operations that are mostly
duplicate functionality. The single log item deletes can be
implemented via the bulk updates by turning xfs_trans_ail_delete()
into a simple wrapper. This removes all the duplicate delete
functionality and associated helpers.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
We now have two copies of AIL insert operations that are mostly
duplicate functionality. The single log item updates can be
implemented via the bulk updates by turning xfs_trans_ail_update()
into a simple wrapper. This removes all the duplicate insert
functionality and associated helpers.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
When inode buffer IO completes, usually all of the inodes are removed from the
AIL. This involves processing them one at a time and taking the AIL lock once
for every inode. When all CPUs are processing inode IO completions, this causes
excessive amount sof contention on the AIL lock.
Instead, change the way we process inode IO completion in the buffer
IO done callback. Allow the inode IO done callback to walk the list
of IO done callbacks and pull all the inodes off the buffer in one
go and then process them as a batch.
Once all the inodes for removal are collected, take the AIL lock
once and do a bulk removal operation to minimise traffic on the AIL
lock.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
When inserting items into the AIL from the transaction committed
callbacks, we take the AIL lock for every single item that is to be
inserted. For a CIL checkpoint commit, this can be tens of thousands
of individual inserts, yet almost all of the items will be inserted
at the same point in the AIL because they have the same index.
To reduce the overhead and contention on the AIL lock for such
operations, introduce a "bulk insert" operation which allows a list
of log items with the same LSN to be inserted in a single operation
via a list splice. To do this, we need to pre-sort the log items
being committed into a temporary list for insertion.
The complexity is that not every log item will end up with the same
LSN, and not every item is actually inserted into the AIL. Items
that don't match the commit LSN will be inserted and unpinned as per
the current one-at-a-time method (relatively rare), while items that
are not to be inserted will be unpinned and freed immediately. Items
that are to be inserted at the given commit lsn are placed in a
temporary array and inserted into the AIL in bulk each time the
array fills up.
As a result of this, we trade off AIL hold time for a significant
reduction in traffic. lock_stat output shows that the worst case
hold time is unchanged, but contention from AIL inserts drops by an
order of magnitude and the number of lock traversal decreases
significantly.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
xfs_ail_delete() has a needlessly complex interface. It returns the log item
that was passed in for deletion (which the callers then assert is identical to
the one passed in), and callers of xfs_ail_delete() still need to invalidate
current traversal cursors.
Make xfs_ail_delete() return void, move the cursor invalidation inside it, and
clean up the callers just to use the log item pointer they passed in.
While cleaning up, remove the messy and unnecessary "/* ARGUSED */" comments
around all these functions.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Dmapi support was never merged upstream, but we still have a lot of hooks
bloating XFS for it, all over the fast pathes of the filesystem.
This patch drops over 700 lines of dmapi overhead. If we'll ever get HSM
support in mainline at least the namespace events can be done much saner
in the VFS instead of the individual filesystem, so it's not like this
is much help for future work.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
All buffers logged into the AIL are marked as delayed write.
When the AIL needs to push the buffer out, it issues an async write of the
buffer. This means that IO patterns are dependent on the order of
buffers in the AIL.
Instead of flushing the buffer, promote the buffer in the delayed
write list so that the next time the xfsbufd is run the buffer will
be flushed by the xfsbufd. Return the state to the xfsaild that the
buffer was promoted so that the xfsaild knows that it needs to cause
the xfsbufd to run to flush the buffers that were promoted.
Using the xfsbufd for issuing the IO allows us to dispatch all
buffer IO from the one queue. This means that we can make much more
enlightened decisions on what order to flush buffers to disk as
we don't have multiple places issuing IO. Optimisations to xfsbufd
will be in a future patch.
Version 2
- kill XFS_ITEM_FLUSHING as it is now unused.
Signed-off-by: Dave Chinner <david@fromorbit.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Remove the XFS_LOG_FORCE argument which was always set, and the
XFS_LOG_URGE define, which was never used.
Split xfs_log_force into a two helpers - xfs_log_force which forces
the whole log, and xfs_log_force_lsn which forces up to the
specified LSN. The underlying implementations already were entirely
separate, as were the users.
Also re-indent the new _xfs_log_force/_xfs_log_force which
previously had a weird coding style.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
Now that the AIL push algorithm is traversal safe, we don't need a
watchdog function in the xfsaild to catch pushes that fail to make
progress. Remove the watchdog timeout and make pushes purely driven
by demand. This will remove the once-per-second wakeup that is seen
when the filesystem is idle and make laptop power misers happy.
Signed-off-by: Dave Chinner <david@fromorbit.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
Access to log items on the AIL is generally protected by m_ail_lock;
this is particularly needed when we're getting or setting the 64-bit
li_lsn on a 32-bit platform. This patch fixes a couple places where we
were accessing the log item after dropping the AIL lock on 32-bit
machines.
This can result in a partially-zeroed log->l_tail_lsn if
xfs_trans_ail_delete is racing with xfs_trans_ail_update, and in at
least some cases, this can leave the l_tail_lsn with a zero cycle
number, which means xlog_space_left will think the log is full (unless
CONFIG_XFS_DEBUG is set, in which case we'll trip an ASSERT), leading to
processes stuck forever in xlog_grant_log_space.
Thanks to Adrian VanderSpek for first spotting the race potential and to
Dave Chinner for debug assistance.
Signed-off-by: Nathaniel W. Turner <nate@houseofnate.net>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
Under heavy metadata load we are seeing log hangs. The AIL has items in it
ready to be pushed, and they are within the push target window. However,
we are not pushing them when the last pushed LSN is less than the LSN of
the first log item on the AIL. This is a regression introduced by the AIL
push cursor modifications.
SGI-PV: 987246
SGI-Modid: xfs-linux-melb:xfs-kern:32409a
Signed-off-by: David Chinner <david@fromorbit.com>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Signed-off-by: Tim Shimmin <tes@sgi.com>
Change all the remaining AIL API functions that are passed struct
xfs_mount pointers to pass pointers directly to the struct xfs_ail being
used. With this conversion, all external access to the AIL is via the
struct xfs_ail. Hence the operation and referencing of the AIL is almost
entirely independent of the xfs_mount that is using it - it is now much
more tightly tied to the log and the items it is tracking in the log than
it is tied to the xfs_mount.
SGI-PV: 988143
SGI-Modid: xfs-linux-melb:xfs-kern:32353a
Signed-off-by: David Chinner <david@fromorbit.com>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Signed-off-by: Christoph Hellwig <hch@infradead.org>
Bring the ail lock inside the struct xfs_ail. This means the AIL can be
entirely manipulated via the struct xfs_ail rather than needing both the
struct xfs_mount and the struct xfs_ail.
SGI-PV: 988143
SGI-Modid: xfs-linux-melb:xfs-kern:32350a
Signed-off-by: David Chinner <david@fromorbit.com>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Signed-off-by: Christoph Hellwig <hch@infradead.org>
With the new cursor interface, it makes sense to make all the traversing
code use the cursor interface and make the old one go away. This means
more of the AIL interfacing is done by passing struct xfs_ail pointers
around the place instead of struct xfs_mount pointers.
We can replace the use of xfs_trans_first_ail() in xfs_log_need_covered()
as it is only checking if the AIL is empty. We can do that with a call to
xfs_trans_ail_tail() instead, where a zero LSN returned indicates and
empty AIL...
SGI-PV: 988143
SGI-Modid: xfs-linux-melb:xfs-kern:32348a
Signed-off-by: David Chinner <david@fromorbit.com>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Signed-off-by: Christoph Hellwig <hch@infradead.org>
To replace the current generation number ensuring sanity of the AIL
traversal, replace it with an external cursor that is linked to the AIL.
Basically, we store the next item in the cursor whenever we want to drop
the AIL lock to do something to the current item. When we regain the lock.
the current item may already be free, so we can't reference it, but the
next item in the traversal is already held in the cursor.
When we move or delete an object, we search all the active cursors and if
there is an item match we clear the cursor(s) that point to the object.
This forces the traversal to restart transparently.
We don't invalidate the cursor on insert because the cursor still points
to a valid item. If the intem is inserted between the current item and the
cursor it does not matter; the traversal is considered to be past the
insertion point so it will be picked up in the next traversal.
Hence traversal restarts pretty much disappear altogether with this method
of traversal, which should substantially reduce the overhead of pushing on
a busy AIL.
Version 2 o add restart logic o comment cursor interface o minor cleanups
SGI-PV: 988143
SGI-Modid: xfs-linux-melb:xfs-kern:32347a
Signed-off-by: David Chinner <david@fromorbit.com>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Signed-off-by: Christoph Hellwig <hch@infradead.org>
Rather than embedding the struct xfs_ail in the struct xfs_mount, allocate
it during AIL initialisation. Add a back pointer to the struct xfs_ail so
that we can pass around the xfs_ail and still be able to access the
xfs_mount if need be. This is th first step involved in isolating the AIL
implementation from the surrounding filesystem code.
SGI-PV: 988143
SGI-Modid: xfs-linux-melb:xfs-kern:32346a
Signed-off-by: David Chinner <david@fromorbit.com>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Signed-off-by: Christoph Hellwig <hch@infradead.org>
Replace the xfs_ail_entry_t with a struct list_head and clean the
surrounding code up. Also fixes a livelock in xfs_trans_first_push_ail()
by terminating the loop at the head of the list correctly.
SGI-PV: 978682
SGI-Modid: xfs-linux-melb:xfs-kern:30636a
Signed-off-by: Josef 'Jeff' Sipek <jeffpc@josefsipek.net>
Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
wakeups
Idle state is not being detected properly by the xfsaild push code. The
current idle state is detected by an empty list which may never happen
with mostly idle filesystem or one using lazy superblock counters. A
single dirty item in the list that exists beyond the push target can
result repeated looping attempting to push up to the target because it
fails to check if the push target has been acheived or not.
Fix by considering a dirty list with everything past the target as an idle
state and set the timeout appropriately.
SGI-PV: 977545
SGI-Modid: xfs-linux-melb:xfs-kern:30532a
Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Checking the entire AIL on every insert and remove is prohibitively
expensive - the sustained sequntial create rate on a single disk drops
from about 1800/s to 60/s because of this checking resulting in the
xfslogd becoming cpu bound.
By default on debug builds, only check the next and previous entries in
the list to ensure they are ordered correctly. If you really want, define
XFS_TRANS_DEBUG to use the old behaviour.
SGI-PV: 972759
SGI-Modid: xfs-linux-melb:xfs-kern:30372a
Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
When many hundreds to thousands of threads all try to do simultaneous
transactions and the log is in a tail-pushing situation (i.e. full), we
can get multiple threads walking the AIL list and contending on the AIL
lock.
The AIL push is, in effect, a simple I/O dispatch algorithm complicated by
the ordering constraints placed on it by the transaction subsystem. It
really does not need multiple threads to push on it - even when only a
single CPU is pushing the AIL, it can push the I/O out far faster that
pretty much any disk subsystem can handle.
So, to avoid contention problems stemming from multiple list walkers, move
the list walk off into another thread and simply provide a "target" to
push to. When a thread requires a push, it sets the target and wakes the
push thread, then goes to sleep waiting for the required amount of space
to become available in the log.
This mechanism should also be a lot fairer under heavy load as the waiters
will queue in arrival order, rather than queuing in "who completed a push
first" order.
Also, by moving the pushing to a separate thread we can do more
effectively overload detection and prevention as we can keep context from
loop iteration to loop iteration. That is, we can push only part of the
list each loop and not have to loop back to the start of the list every
time we run. This should also help by reducing the number of items we try
to lock and/or push items that we cannot move.
Note that this patch is not intended to solve the inefficiencies in the
AIL structure and the associated issues with extremely large list
contents. That needs to be addresses separately; parallel access would
cause problems to any new structure as well, so I'm only aiming to isolate
the structure from unbounded parallelism here.
SGI-PV: 972759
SGI-Modid: xfs-linux-melb:xfs-kern:30371a
Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
One of the perpetual scaling problems XFS has is indexing it's incore
inodes. We currently uses hashes and the default hash sizes chosen can
only ever be a tradeoff between memory consumption and the maximum
realistic size of the cache.
As a result, anyone who has millions of inodes cached on a filesystem
needs to tunes the size of the cache via the ihashsize mount option to
allow decent scalability with inode cache operations.
A further problem is the separate inode cluster hash, whose size is based
on the ihashsize but is smaller, and so under certain conditions (sparse
cluster cache population) this can become a limitation long before the
inode hash is causing issues.
The following patchset removes the inode hash and cluster hash and
replaces them with radix trees to avoid the scalability limitations of the
hashes. It also reduces the size of the inodes by 3 pointers....
SGI-PV: 969561
SGI-Modid: xfs-linux-melb:xfs-kern:29481a
Signed-off-by: David Chinner <dgc@sgi.com>
Signed-off-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Tim Shimmin <tes@sgi.com>
xfs_trans_delete_ail
xfs_trans_update_ail and xfs_trans_delete_ail get called with the AIL lock
held, and release it. Add lock annotations to these two functions so that
sparse can check callers for lock pairing, and so that sparse will not
complain about these functions since they intentionally use locks in this
manner.
SGI-PV: 954580
SGI-Modid: xfs-linux-melb:xfs-kern:26807a
Signed-off-by: Josh Triplett <josh@freedesktop.org>
Signed-off-by: Nathan Scott <nathans@sgi.com>
Signed-off-by: Tim Shimmin <tes@sgi.com>
when it goes to force out the log, and get the tail lsn, it will want to
get the AIL lock.
SGI-PV: 940076
SGI-Modid: xfs-linux:xfs-kern:23260a
Signed-off-by: Tim Shimmin <tes@sgi.com>
Signed-off-by: Nathan Scott <nathans@sgi.com>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!