diff options
author | Christoph Hellwig <hch@infradead.org> | 2012-04-23 15:58:39 +1000 |
---|---|---|
committer | Ben Myers <bpm@sgi.com> | 2012-05-14 16:20:31 -0500 |
commit | 43ff2122e6492bcc88b065c433453dce88223b30 (patch) | |
tree | 0f762cfb753edd73402b8830e0927d9efba30c61 /fs/xfs/xfs_trans_buf.c | |
parent | 960c60af8b9481595e68875e79b2602e73169c29 (diff) | |
download | linux-stable-43ff2122e6492bcc88b065c433453dce88223b30.tar.gz linux-stable-43ff2122e6492bcc88b065c433453dce88223b30.tar.bz2 linux-stable-43ff2122e6492bcc88b065c433453dce88223b30.zip |
xfs: on-stack delayed write buffer lists
Queue delwri buffers on a local on-stack list instead of a per-buftarg one,
and write back the buffers per-process instead of by waking up xfsbufd.
This is now easily doable given that we have very few places left that write
delwri buffers:
- log recovery:
Only done at mount time, and already forcing out the buffers
synchronously using xfs_flush_buftarg
- quotacheck:
Same story.
- dquot reclaim:
Writes out dirty dquots on the LRU under memory pressure. We might
want to look into doing more of this via xfsaild, but it's already
more optimal than the synchronous inode reclaim that writes each
buffer synchronously.
- xfsaild:
This is the main beneficiary of the change. By keeping a local list
of buffers to write we reduce latency of writing out buffers, and
more importably we can remove all the delwri list promotions which
were hitting the buffer cache hard under sustained metadata loads.
The implementation is very straight forward - xfs_buf_delwri_queue now gets
a new list_head pointer that it adds the delwri buffers to, and all callers
need to eventually submit the list using xfs_buf_delwi_submit or
xfs_buf_delwi_submit_nowait. Buffers that already are on a delwri list are
skipped in xfs_buf_delwri_queue, assuming they already are on another delwri
list. The biggest change to pass down the buffer list was done to the AIL
pushing. Now that we operate on buffers the trylock, push and pushbuf log
item methods are merged into a single push routine, which tries to lock the
item, and if possible add the buffer that needs writeback to the buffer list.
This leads to much simpler code than the previous split but requires the
individual IOP_PUSH instances to unlock and reacquire the AIL around calls
to blocking routines.
Given that xfsailds now also handle writing out buffers, the conditions for
log forcing and the sleep times needed some small changes. The most
important one is that we consider an AIL busy as long we still have buffers
to push, and the other one is that we do increment the pushed LSN for
buffers that are under flushing at this moment, but still count them towards
the stuck items for restart purposes. Without this we could hammer on stuck
items without ever forcing the log and not make progress under heavy random
delete workloads on fast flash storage devices.
[ Dave Chinner:
- rebase on previous patches.
- improved comments for XBF_DELWRI_Q handling
- fix XBF_ASYNC handling in queue submission (test 106 failure)
- rename delwri submit function buffer list parameters for clarity
- xfs_efd_item_push() should return XFS_ITEM_PINNED ]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
Diffstat (limited to 'fs/xfs/xfs_trans_buf.c')
-rw-r--r-- | fs/xfs/xfs_trans_buf.c | 84 |
1 files changed, 27 insertions, 57 deletions
diff --git a/fs/xfs/xfs_trans_buf.c b/fs/xfs/xfs_trans_buf.c index 296a7995a007..9132d162c4b8 100644 --- a/fs/xfs/xfs_trans_buf.c +++ b/fs/xfs/xfs_trans_buf.c @@ -165,14 +165,6 @@ xfs_trans_get_buf(xfs_trans_t *tp, XFS_BUF_DONE(bp); } - /* - * If the buffer is stale then it was binval'ed - * since last read. This doesn't matter since the - * caller isn't allowed to use the data anyway. - */ - else if (XFS_BUF_ISSTALE(bp)) - ASSERT(!XFS_BUF_ISDELAYWRITE(bp)); - ASSERT(bp->b_transp == tp); bip = bp->b_fspriv; ASSERT(bip != NULL); @@ -418,19 +410,6 @@ xfs_trans_read_buf( return 0; shutdown_abort: - /* - * the theory here is that buffer is good but we're - * bailing out because the filesystem is being forcibly - * shut down. So we should leave the b_flags alone since - * the buffer's not staled and just get out. - */ -#if defined(DEBUG) - if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp)) - xfs_notice(mp, "about to pop assert, bp == 0x%p", bp); -#endif - ASSERT((bp->b_flags & (XBF_STALE|XBF_DELWRI)) != - (XBF_STALE|XBF_DELWRI)); - trace_xfs_trans_read_buf_shut(bp, _RET_IP_); xfs_buf_relse(bp); *bpp = NULL; @@ -649,22 +628,33 @@ xfs_trans_log_buf(xfs_trans_t *tp, /* - * This called to invalidate a buffer that is being used within - * a transaction. Typically this is because the blocks in the - * buffer are being freed, so we need to prevent it from being - * written out when we're done. Allowing it to be written again - * might overwrite data in the free blocks if they are reallocated - * to a file. + * Invalidate a buffer that is being used within a transaction. + * + * Typically this is because the blocks in the buffer are being freed, so we + * need to prevent it from being written out when we're done. Allowing it + * to be written again might overwrite data in the free blocks if they are + * reallocated to a file. * - * We prevent the buffer from being written out by clearing the - * B_DELWRI flag. We can't always - * get rid of the buf log item at this point, though, because - * the buffer may still be pinned by another transaction. If that - * is the case, then we'll wait until the buffer is committed to - * disk for the last time (we can tell by the ref count) and - * free it in xfs_buf_item_unpin(). Until it is cleaned up we - * will keep the buffer locked so that the buffer and buf log item - * are not reused. + * We prevent the buffer from being written out by marking it stale. We can't + * get rid of the buf log item at this point because the buffer may still be + * pinned by another transaction. If that is the case, then we'll wait until + * the buffer is committed to disk for the last time (we can tell by the ref + * count) and free it in xfs_buf_item_unpin(). Until that happens we will + * keep the buffer locked so that the buffer and buf log item are not reused. + * + * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log + * the buf item. This will be used at recovery time to determine that copies + * of the buffer in the log before this should not be replayed. + * + * We mark the item descriptor and the transaction dirty so that we'll hold + * the buffer until after the commit. + * + * Since we're invalidating the buffer, we also clear the state about which + * parts of the buffer have been logged. We also clear the flag indicating + * that this is an inode buffer since the data in the buffer will no longer + * be valid. + * + * We set the stale bit in the buffer as well since we're getting rid of it. */ void xfs_trans_binval( @@ -684,7 +674,6 @@ xfs_trans_binval( * If the buffer is already invalidated, then * just return. */ - ASSERT(!(XFS_BUF_ISDELAYWRITE(bp))); ASSERT(XFS_BUF_ISSTALE(bp)); ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY))); ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF)); @@ -694,27 +683,8 @@ xfs_trans_binval( return; } - /* - * Clear the dirty bit in the buffer and set the STALE flag - * in the buf log item. The STALE flag will be used in - * xfs_buf_item_unpin() to determine if it should clean up - * when the last reference to the buf item is given up. - * We set the XFS_BLF_CANCEL flag in the buf log format structure - * and log the buf item. This will be used at recovery time - * to determine that copies of the buffer in the log before - * this should not be replayed. - * We mark the item descriptor and the transaction dirty so - * that we'll hold the buffer until after the commit. - * - * Since we're invalidating the buffer, we also clear the state - * about which parts of the buffer have been logged. We also - * clear the flag indicating that this is an inode buffer since - * the data in the buffer will no longer be valid. - * - * We set the stale bit in the buffer as well since we're getting - * rid of it. - */ xfs_buf_stale(bp); + bip->bli_flags |= XFS_BLI_STALE; bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY); bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF; |