From: Dave Chinner Date: Fri, 24 Sep 2010 09:51:50 +0000 (+1000) Subject: xfs: batch inode reclaim lookup X-Git-Tag: v2.6.37-rc1~162^2~17 X-Git-Url: https://git.karo-electronics.de/?a=commitdiff_plain;h=e3a20c0b02e1704ab115dfa9d012caf0fbc45ed0;p=karo-tx-linux.git xfs: batch inode reclaim lookup Batch and optimise the per-ag inode lookup for reclaim to minimise scanning overhead. This involves gang lookups on the radix trees to get multiple inodes during each tree walk, and tighter validation of what inodes can be reclaimed without blocking befor we take any locks. This is based on ideas suggested in a proof-of-concept patch posted by Nick Piggin. Signed-off-by: Dave Chinner Reviewed-by: Christoph Hellwig Reviewed-by: Alex Elder --- diff --git a/fs/xfs/linux-2.6/xfs_sync.c b/fs/xfs/linux-2.6/xfs_sync.c index 0ed3d0ae3c28..754bc591a247 100644 --- a/fs/xfs/linux-2.6/xfs_sync.c +++ b/fs/xfs/linux-2.6/xfs_sync.c @@ -630,6 +630,43 @@ __xfs_inode_clear_reclaim_tag( __xfs_inode_clear_reclaim(pag, ip); } +/* + * Grab the inode for reclaim exclusively. + * Return 0 if we grabbed it, non-zero otherwise. + */ +STATIC int +xfs_reclaim_inode_grab( + struct xfs_inode *ip, + int flags) +{ + + /* + * do some unlocked checks first to avoid unnecceary lock traffic. + * The first is a flush lock check, the second is a already in reclaim + * check. Only do these checks if we are not going to block on locks. + */ + if ((flags & SYNC_TRYLOCK) && + (!ip->i_flush.done || __xfs_iflags_test(ip, XFS_IRECLAIM))) { + return 1; + } + + /* + * The radix tree lock here protects a thread in xfs_iget from racing + * with us starting reclaim on the inode. Once we have the + * XFS_IRECLAIM flag set it will not touch us. + */ + spin_lock(&ip->i_flags_lock); + ASSERT_ALWAYS(__xfs_iflags_test(ip, XFS_IRECLAIMABLE)); + if (__xfs_iflags_test(ip, XFS_IRECLAIM)) { + /* ignore as it is already under reclaim */ + spin_unlock(&ip->i_flags_lock); + return 1; + } + __xfs_iflags_set(ip, XFS_IRECLAIM); + spin_unlock(&ip->i_flags_lock); + return 0; +} + /* * Inodes in different states need to be treated differently, and the return * value of xfs_iflush is not sufficient to get this right. The following table @@ -688,23 +725,6 @@ xfs_reclaim_inode( { int error = 0; - /* - * The radix tree lock here protects a thread in xfs_iget from racing - * with us starting reclaim on the inode. Once we have the - * XFS_IRECLAIM flag set it will not touch us. - */ - spin_lock(&ip->i_flags_lock); - ASSERT_ALWAYS(__xfs_iflags_test(ip, XFS_IRECLAIMABLE)); - if (__xfs_iflags_test(ip, XFS_IRECLAIM)) { - /* ignore as it is already under reclaim */ - spin_unlock(&ip->i_flags_lock); - write_unlock(&pag->pag_ici_lock); - return 0; - } - __xfs_iflags_set(ip, XFS_IRECLAIM); - spin_unlock(&ip->i_flags_lock); - write_unlock(&pag->pag_ici_lock); - xfs_ilock(ip, XFS_ILOCK_EXCL); if (!xfs_iflock_nowait(ip)) { if (!(sync_mode & SYNC_WAIT)) @@ -822,16 +842,19 @@ xfs_reclaim_inodes_ag( while ((pag = xfs_perag_get_tag(mp, ag, XFS_ICI_RECLAIM_TAG))) { unsigned long first_index = 0; int done = 0; + int nr_found = 0; ag = pag->pag_agno + 1; do { - struct xfs_inode *ip; - int nr_found; + struct xfs_inode *batch[XFS_LOOKUP_BATCH]; + int i; write_lock(&pag->pag_ici_lock); - nr_found = radix_tree_gang_lookup_tag(&pag->pag_ici_root, - (void **)&ip, first_index, 1, + nr_found = radix_tree_gang_lookup_tag( + &pag->pag_ici_root, + (void **)batch, first_index, + XFS_LOOKUP_BATCH, XFS_ICI_RECLAIM_TAG); if (!nr_found) { write_unlock(&pag->pag_ici_lock); @@ -839,20 +862,41 @@ xfs_reclaim_inodes_ag( } /* - * Update the index for the next lookup. Catch overflows - * into the next AG range which can occur if we have inodes - * in the last block of the AG and we are currently - * pointing to the last inode. + * Grab the inodes before we drop the lock. if we found + * nothing, nr == 0 and the loop will be skipped. */ - first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); - if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) - done = 1; + for (i = 0; i < nr_found; i++) { + struct xfs_inode *ip = batch[i]; + + if (done || xfs_reclaim_inode_grab(ip, flags)) + batch[i] = NULL; + + /* + * Update the index for the next lookup. Catch + * overflows into the next AG range which can + * occur if we have inodes in the last block of + * the AG and we are currently pointing to the + * last inode. + */ + first_index = XFS_INO_TO_AGINO(mp, ip->i_ino + 1); + if (first_index < XFS_INO_TO_AGINO(mp, ip->i_ino)) + done = 1; + } - error = xfs_reclaim_inode(ip, pag, flags); - if (error && last_error != EFSCORRUPTED) - last_error = error; + /* unlock now we've grabbed the inodes. */ + write_unlock(&pag->pag_ici_lock); + + for (i = 0; i < nr_found; i++) { + if (!batch[i]) + continue; + error = xfs_reclaim_inode(batch[i], pag, flags); + if (error && last_error != EFSCORRUPTED) + last_error = error; + } + + *nr_to_scan -= XFS_LOOKUP_BATCH; - } while (!done && (*nr_to_scan)--); + } while (nr_found && !done && *nr_to_scan > 0); xfs_perag_put(pag); } @@ -888,7 +932,7 @@ xfs_reclaim_inode_shrink( if (!(gfp_mask & __GFP_FS)) return -1; - xfs_reclaim_inodes_ag(mp, 0, &nr_to_scan); + xfs_reclaim_inodes_ag(mp, SYNC_TRYLOCK, &nr_to_scan); /* terminate if we don't exhaust the scan */ if (nr_to_scan > 0) return -1;