From 7db37c5e6575b229a5051be1d3ef15257ae0ba5d Mon Sep 17 00:00:00 2001 From: Dave Chinner Date: Thu, 27 Jan 2011 12:02:00 +1100 Subject: xfs: fix log ticket leak on forced shutdown. The kmemleak detector shows this after test 139: unreferenced object 0xffff880079b88bb0 (size 264): comm "xfs_io", pid 4904, jiffies 4294909382 (age 276.824s) hex dump (first 32 bytes): 00 00 00 00 ad 4e ad de ff ff ff ff 00 00 00 00 .....N.......... ff ff ff ff ff ff ff ff 48 7b c9 82 ff ff ff ff ........H{...... backtrace: [] kmemleak_alloc+0x2d/0x60 [] kmem_cache_alloc+0x13f/0x2b0 [] kmem_zone_alloc+0x77/0xf0 [] kmem_zone_zalloc+0x1e/0x50 [] xlog_ticket_alloc+0x34/0x170 [] xlog_cil_push+0xa4/0x3f0 [] xlog_cil_force_lsn+0x15a/0x160 [] _xfs_log_force_lsn+0x75/0x2d0 [] _xfs_trans_commit+0x2bd/0x2f0 [] xfs_iomap_write_allocate+0x1ad/0x350 [] xfs_map_blocks+0x21f/0x370 [] xfs_vm_writepage+0x1c7/0x550 [] __writepage+0x1a/0x50 [] write_cache_pages+0x1c2/0x4c0 [] generic_writepages+0x27/0x30 [] xfs_vm_writepages+0x5d/0x80 By inspection, the leak occurs when xlog_write() returns and error and we jump to the abort path without dropping the reference on the active ticket. Signed-off-by: Dave Chinner Reviewed-by: Christoph Hellwig Reviewed-by: Alex Elder --- fs/xfs/xfs_log_cil.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) (limited to 'fs') diff --git a/fs/xfs/xfs_log_cil.c b/fs/xfs/xfs_log_cil.c index 9dc8125d04e5..c7eac5acbfea 100644 --- a/fs/xfs/xfs_log_cil.c +++ b/fs/xfs/xfs_log_cil.c @@ -543,7 +543,7 @@ xlog_cil_push( error = xlog_write(log, &lvhdr, tic, &ctx->start_lsn, NULL, 0); if (error) - goto out_abort; + goto out_abort_free_ticket; /* * now that we've written the checkpoint into the log, strictly @@ -569,8 +569,9 @@ restart: } spin_unlock(&cil->xc_cil_lock); + /* xfs_log_done always frees the ticket on error. */ commit_lsn = xfs_log_done(log->l_mp, tic, &commit_iclog, 0); - if (error || commit_lsn == -1) + if (commit_lsn == -1) goto out_abort; /* attach all the transactions w/ busy extents to iclog */ @@ -600,6 +601,8 @@ out_free_ticket: kmem_free(new_ctx); return 0; +out_abort_free_ticket: + xfs_log_ticket_put(tic); out_abort: xlog_cil_committed(ctx, XFS_LI_ABORTED); return XFS_ERROR(EIO); -- cgit v1.2.3 From e34a314c5e49fe6b763568f6576b19f1299c33c2 Mon Sep 17 00:00:00 2001 From: Dave Chinner Date: Thu, 27 Jan 2011 12:13:35 +1100 Subject: xfs: fix efi item leak on forced shutdown After test 139, kmemleak shows: unreferenced object 0xffff880078b405d8 (size 400): comm "xfs_io", pid 4904, jiffies 4294909383 (age 1186.728s) hex dump (first 32 bytes): 60 c1 17 79 00 88 ff ff 60 c1 17 79 00 88 ff ff `..y....`..y.... 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ backtrace: [] kmemleak_alloc+0x2d/0x60 [] kmem_cache_alloc+0x13f/0x2b0 [] kmem_zone_alloc+0x77/0xf0 [] kmem_zone_zalloc+0x1e/0x50 [] xfs_efi_init+0x4b/0xb0 [] xfs_trans_get_efi+0x58/0x90 [] xfs_bmap_finish+0x8b/0x1d0 [] xfs_itruncate_finish+0x2c4/0x5d0 [] xfs_setattr+0x8df/0xa70 [] xfs_vn_setattr+0x1b/0x20 [] notify_change+0x170/0x2e0 [] do_truncate+0x66/0xa0 [] sys_ftruncate+0xdb/0xe0 [] system_call_fastpath+0x16/0x1b [] 0xffffffffffffffff The cause of the leak is that the "remove" parameter of IOP_UNPIN() is never set when a CIL push is aborted. This means that the EFI item is never freed if it was in the push being cancelled. The problem is specific to delayed logging, but has uncovered a couple of problems with the handling of IOP_UNPIN(remove). Firstly, we cannot safely call xfs_trans_del_item() from IOP_UNPIN() in the CIL commit failure path or the iclog write failure path because for delayed loging we have no transaction context. Hence we must only call xfs_trans_del_item() if the log item being unpinned has an active log item descriptor. Secondly, xfs_trans_uncommit() does not handle log item descriptor freeing during the traversal of log items on a transaction. It can reference a freed log item descriptor when unpinning an EFI item. Hence it needs to use a safe list traversal method to allow items to be removed from the transaction during IOP_UNPIN(). Signed-off-by: Dave Chinner Reviewed-by: Alex Elder --- fs/xfs/xfs_buf_item.c | 12 +++++++----- fs/xfs/xfs_extfree_item.c | 3 ++- fs/xfs/xfs_trans.c | 36 +++++++++++++++++++++++++++++------- 3 files changed, 38 insertions(+), 13 deletions(-) (limited to 'fs') diff --git a/fs/xfs/xfs_buf_item.c b/fs/xfs/xfs_buf_item.c index 98c6f73b6752..6f8c21ce0d6d 100644 --- a/fs/xfs/xfs_buf_item.c +++ b/fs/xfs/xfs_buf_item.c @@ -427,13 +427,15 @@ xfs_buf_item_unpin( if (remove) { /* - * We have to remove the log item from the transaction - * as we are about to release our reference to the - * buffer. If we don't, the unlock that occurs later - * in xfs_trans_uncommit() will ry to reference the + * If we are in a transaction context, we have to + * remove the log item from the transaction as we are + * about to release our reference to the buffer. If we + * don't, the unlock that occurs later in + * xfs_trans_uncommit() will try to reference the * buffer which we no longer have a hold on. */ - xfs_trans_del_item(lip); + if (lip->li_desc) + xfs_trans_del_item(lip); /* * Since the transaction no longer refers to the buffer, diff --git a/fs/xfs/xfs_extfree_item.c b/fs/xfs/xfs_extfree_item.c index 75f2ef60e579..d22e62623437 100644 --- a/fs/xfs/xfs_extfree_item.c +++ b/fs/xfs/xfs_extfree_item.c @@ -138,7 +138,8 @@ xfs_efi_item_unpin( if (remove) { ASSERT(!(lip->li_flags & XFS_LI_IN_AIL)); - xfs_trans_del_item(lip); + if (lip->li_desc) + xfs_trans_del_item(lip); xfs_efi_item_free(efip); return; } diff --git a/fs/xfs/xfs_trans.c b/fs/xfs/xfs_trans.c index 33dbc4e0ad62..29f5e5424897 100644 --- a/fs/xfs/xfs_trans.c +++ b/fs/xfs/xfs_trans.c @@ -1446,6 +1446,14 @@ xfs_log_item_batch_insert( * Bulk operation version of xfs_trans_committed that takes a log vector of * items to insert into the AIL. This uses bulk AIL insertion techniques to * minimise lock traffic. + * + * If we are called with the aborted flag set, it is because a log write during + * a CIL checkpoint commit has failed. In this case, all the items in the + * checkpoint have already gone through IOP_COMMITED and IOP_UNLOCK, which + * means that checkpoint commit abort handling is treated exactly the same + * as an iclog write error even though we haven't started any IO yet. Hence in + * this case all we need to do is IOP_COMMITTED processing, followed by an + * IOP_UNPIN(aborted) call. */ void xfs_trans_committed_bulk( @@ -1472,6 +1480,16 @@ xfs_trans_committed_bulk( if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) continue; + /* + * if we are aborting the operation, no point in inserting the + * object into the AIL as we are in a shutdown situation. + */ + if (aborted) { + ASSERT(XFS_FORCED_SHUTDOWN(ailp->xa_mount)); + IOP_UNPIN(lip, 1); + continue; + } + if (item_lsn != commit_lsn) { /* @@ -1503,20 +1521,24 @@ xfs_trans_committed_bulk( } /* - * Called from the trans_commit code when we notice that - * the filesystem is in the middle of a forced shutdown. + * Called from the trans_commit code when we notice that the filesystem is in + * the middle of a forced shutdown. + * + * When we are called here, we have already pinned all the items in the + * transaction. However, neither IOP_COMMITTING or IOP_UNLOCK has been called + * so we can simply walk the items in the transaction, unpin them with an abort + * flag and then free the items. Note that unpinning the items can result in + * them being freed immediately, so we need to use a safe list traversal method + * here. */ STATIC void xfs_trans_uncommit( struct xfs_trans *tp, uint flags) { - struct xfs_log_item_desc *lidp; + struct xfs_log_item_desc *lidp, *n; - list_for_each_entry(lidp, &tp->t_items, lid_trans) { - /* - * Unpin all but those that aren't dirty. - */ + list_for_each_entry_safe(lidp, n, &tp->t_items, lid_trans) { if (lidp->lid_flags & XFS_LID_DIRTY) IOP_UNPIN(lidp->lid_item, 1); } -- cgit v1.2.3 From b8fc82630ae289bb4e661567808afc59e3298dce Mon Sep 17 00:00:00 2001 From: Dave Chinner Date: Thu, 27 Jan 2011 12:14:12 +1100 Subject: xfs: speculative delayed allocation uses rounddown_power_of_2 badly rounddown_power_of_2() returns an undefined result when passed a value of zero. The specualtive delayed allocation code is doing this when the inode is zero length. Hence occasionally the preallocation is much, much larger than is necessary (e.g. 8GB for a 270 _byte_ file). Ensure we don't even pass a zero value to this function so the result of preallocation is always the desired size. Signed-off-by: Dave Chinner Reviewed-by: Christoph Hellwig Reviewed-by: Alex Elder --- fs/xfs/xfs_iomap.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) (limited to 'fs') diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c index 55582bd66659..8a0f044750c3 100644 --- a/fs/xfs/xfs_iomap.c +++ b/fs/xfs/xfs_iomap.c @@ -337,7 +337,12 @@ xfs_iomap_prealloc_size( int shift = 0; int64_t freesp; - alloc_blocks = XFS_B_TO_FSB(mp, ip->i_size); + /* + * rounddown_pow_of_two() returns an undefined result + * if we pass in alloc_blocks = 0. Hence the "+ 1" to + * ensure we always pass in a non-zero value. + */ + alloc_blocks = XFS_B_TO_FSB(mp, ip->i_size) + 1; alloc_blocks = XFS_FILEOFF_MIN(MAXEXTLEN, rounddown_pow_of_two(alloc_blocks)); -- cgit v1.2.3 From 14b064ceaa6f51a7426cc45b4b43685b94380658 Mon Sep 17 00:00:00 2001 From: Dave Chinner Date: Thu, 27 Jan 2011 12:16:28 +1100 Subject: xfs: limit extent length for allocation to AG size Delayed allocation extents can be larger than AGs, so when trying to convert a large range we may scan every AG inside xfs_bmap_alloc_nullfb() trying to find an AG with a size larger than an AG. We should stop when we find the first AG with a maximum possible allocation size. This causes excessive CPU usage when there are lots of AGs. The same problem occurs when doing preallocation of a range larger than an AG. Fix the problem by limiting real allocation lengths to the maximum that an AG can support. This means if we have empty AGs, we'll stop the search at the first of them. If there are no empty AGs, we'll still scan them all, but that is a different problem.... Signed-off-by: Dave Chinner Reviewed-by: Christoph Hellwig Reviewed-by: Alex Elder --- fs/xfs/xfs_alloc.h | 16 ++++++++++++++++ fs/xfs/xfs_bmap.c | 18 ++++++++++-------- 2 files changed, 26 insertions(+), 8 deletions(-) (limited to 'fs') diff --git a/fs/xfs/xfs_alloc.h b/fs/xfs/xfs_alloc.h index 0ab56b32c7eb..d0b3bc72005b 100644 --- a/fs/xfs/xfs_alloc.h +++ b/fs/xfs/xfs_alloc.h @@ -74,6 +74,22 @@ typedef unsigned int xfs_alloctype_t; */ #define XFS_ALLOC_SET_ASIDE(mp) (4 + ((mp)->m_sb.sb_agcount * 4)) +/* + * When deciding how much space to allocate out of an AG, we limit the + * allocation maximum size to the size the AG. However, we cannot use all the + * blocks in the AG - some are permanently used by metadata. These + * blocks are generally: + * - the AG superblock, AGF, AGI and AGFL + * - the AGF (bno and cnt) and AGI btree root blocks + * - 4 blocks on the AGFL according to XFS_ALLOC_SET_ASIDE() limits + * + * The AG headers are sector sized, so the amount of space they take up is + * dependent on filesystem geometry. The others are all single blocks. + */ +#define XFS_ALLOC_AG_MAX_USABLE(mp) \ + ((mp)->m_sb.sb_agblocks - XFS_BB_TO_FSB(mp, XFS_FSS_TO_BB(mp, 4)) - 7) + + /* * Argument structure for xfs_alloc routines. * This is turned into a structure to avoid having 20 arguments passed diff --git a/fs/xfs/xfs_bmap.c b/fs/xfs/xfs_bmap.c index 4111cd3966c7..f3a3768189bb 100644 --- a/fs/xfs/xfs_bmap.c +++ b/fs/xfs/xfs_bmap.c @@ -2430,7 +2430,7 @@ xfs_bmap_btalloc_nullfb( startag = ag = 0; pag = xfs_perag_get(mp, ag); - while (*blen < ap->alen) { + while (*blen < args->maxlen) { if (!pag->pagf_init) { error = xfs_alloc_pagf_init(mp, args->tp, ag, XFS_ALLOC_FLAG_TRYLOCK); @@ -2452,7 +2452,7 @@ xfs_bmap_btalloc_nullfb( notinit = 1; if (xfs_inode_is_filestream(ap->ip)) { - if (*blen >= ap->alen) + if (*blen >= args->maxlen) break; if (ap->userdata) { @@ -2498,14 +2498,14 @@ xfs_bmap_btalloc_nullfb( * If the best seen length is less than the request * length, use the best as the minimum. */ - else if (*blen < ap->alen) + else if (*blen < args->maxlen) args->minlen = *blen; /* - * Otherwise we've seen an extent as big as alen, + * Otherwise we've seen an extent as big as maxlen, * use that as the minimum. */ else - args->minlen = ap->alen; + args->minlen = args->maxlen; /* * set the failure fallback case to look in the selected @@ -2573,7 +2573,9 @@ xfs_bmap_btalloc( args.tp = ap->tp; args.mp = mp; args.fsbno = ap->rval; - args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks); + + /* Trim the allocation back to the maximum an AG can fit. */ + args.maxlen = MIN(ap->alen, XFS_ALLOC_AG_MAX_USABLE(mp)); args.firstblock = ap->firstblock; blen = 0; if (nullfb) { @@ -2621,7 +2623,7 @@ xfs_bmap_btalloc( /* * Adjust for alignment */ - if (blen > args.alignment && blen <= ap->alen) + if (blen > args.alignment && blen <= args.maxlen) args.minlen = blen - args.alignment; args.minalignslop = 0; } else { @@ -2640,7 +2642,7 @@ xfs_bmap_btalloc( * of minlen+alignment+slop doesn't go up * between the calls. */ - if (blen > mp->m_dalign && blen <= ap->alen) + if (blen > mp->m_dalign && blen <= args.maxlen) nextminlen = blen - mp->m_dalign; else nextminlen = args.minlen; -- cgit v1.2.3 From 4ce159890c00e2cc705e955a939bf1dca7b07ab8 Mon Sep 17 00:00:00 2001 From: Dave Chinner Date: Thu, 27 Jan 2011 12:17:58 +1100 Subject: xfs: prevent extsize alignment from exceeding maximum extent size When doing delayed allocation, if the allocation size is for a maximally sized extent, extent size alignment can push it over this limit. This results in an assert failure in xfs_bmbt_set_allf() as the extent length is too large to find in the extent record. Fix this by ensuring that we allow for space that extent size alignment requires (up to 2 * (extsize -1) blocks as we have to handle both head and tail alignment) when limiting the maximum size of the extent. Signed-off-by: Dave Chinner Reviewed-by: Christoph Hellwig Reviewed-by: Alex Elder --- fs/xfs/xfs_bmap.c | 10 ++++++++++ 1 file changed, 10 insertions(+) (limited to 'fs') diff --git a/fs/xfs/xfs_bmap.c b/fs/xfs/xfs_bmap.c index f3a3768189bb..3e9c278a8f78 100644 --- a/fs/xfs/xfs_bmap.c +++ b/fs/xfs/xfs_bmap.c @@ -4487,6 +4487,16 @@ xfs_bmapi( /* Figure out the extent size, adjust alen */ extsz = xfs_get_extsz_hint(ip); if (extsz) { + /* + * make sure we don't exceed a single + * extent length when we align the + * extent by reducing length we are + * going to allocate by the maximum + * amount extent size aligment may + * require. + */ + alen = XFS_FILBLKS_MIN(len, + MAXEXTLEN - (2 * extsz - 1)); error = xfs_bmap_extsize_align(mp, &got, &prev, extsz, rt, eof, -- cgit v1.2.3 From 5315837daee7ed76c31ef643915f7d76ef8c1aa3 Mon Sep 17 00:00:00 2001 From: Dave Chinner Date: Thu, 27 Jan 2011 12:18:18 +1100 Subject: xfs: limit extsize to size of AGs and/or MAXEXTLEN MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The extent size hint can be set to larger than an AG. This means that the alignment process can push the range to be allocated outside the bounds of the AG, resulting in assert failures or corrupted bmbt records. Similarly, if the extsize is larger than the maximum extent size supported, the alignment process will produce extents that are too large to fit into the bmbt records, resulting in a different type of assert/corruption failure. Fix this by limiting extsize at the time іt is set firstly to be less than MAXEXTLEN, then to be a maximum of half the size of the AGs in the filesystem for non-realtime inodes. Realtime inodes do not allocate out of AGs, so don't have to be restricted by the size of AGs. Signed-off-by: Dave Chinner Reviewed-by: Christoph Hellwig Reviewed-by: Alex Elder --- fs/xfs/linux-2.6/xfs_ioctl.c | 20 ++++++++++++++++++-- 1 file changed, 18 insertions(+), 2 deletions(-) (limited to 'fs') diff --git a/fs/xfs/linux-2.6/xfs_ioctl.c b/fs/xfs/linux-2.6/xfs_ioctl.c index b06ede1d0bed..f5e2a19e0f8e 100644 --- a/fs/xfs/linux-2.6/xfs_ioctl.c +++ b/fs/xfs/linux-2.6/xfs_ioctl.c @@ -985,10 +985,22 @@ xfs_ioctl_setattr( /* * Extent size must be a multiple of the appropriate block - * size, if set at all. + * size, if set at all. It must also be smaller than the + * maximum extent size supported by the filesystem. + * + * Also, for non-realtime files, limit the extent size hint to + * half the size of the AGs in the filesystem so alignment + * doesn't result in extents larger than an AG. */ if (fa->fsx_extsize != 0) { - xfs_extlen_t size; + xfs_extlen_t size; + xfs_fsblock_t extsize_fsb; + + extsize_fsb = XFS_B_TO_FSB(mp, fa->fsx_extsize); + if (extsize_fsb > MAXEXTLEN) { + code = XFS_ERROR(EINVAL); + goto error_return; + } if (XFS_IS_REALTIME_INODE(ip) || ((mask & FSX_XFLAGS) && @@ -997,6 +1009,10 @@ xfs_ioctl_setattr( mp->m_sb.sb_blocklog; } else { size = mp->m_sb.sb_blocksize; + if (extsize_fsb > mp->m_sb.sb_agblocks / 2) { + code = XFS_ERROR(EINVAL); + goto error_return; + } } if (fa->fsx_extsize % size) { -- cgit v1.2.3 From c6f990d1ff8e4e53b12f4175eb7d7ea710c3ca73 Mon Sep 17 00:00:00 2001 From: Dave Chinner Date: Thu, 27 Jan 2011 13:23:28 +1100 Subject: xfs: handle CIl transaction commit failures correctly Failure to commit a transaction into the CIL is not handled correctly. This currently can only happen when racing with a shutdown and requires an explicit shutdown check, so it rare and can be avoided. Remove the shutdown check and make the CIL commit a void function to indicate it will always succeed, thereby removing the incorrectly handled failure case. Signed-off-by: Dave Chinner Reviewed-by: Christoph Hellwig Reviewed-by: Alex Elder --- fs/xfs/xfs_log.h | 2 +- fs/xfs/xfs_log_cil.c | 8 +------- fs/xfs/xfs_trans.c | 5 +---- 3 files changed, 3 insertions(+), 12 deletions(-) (limited to 'fs') diff --git a/fs/xfs/xfs_log.h b/fs/xfs/xfs_log.h index 916eb7db14d9..3bd3291ef8d2 100644 --- a/fs/xfs/xfs_log.h +++ b/fs/xfs/xfs_log.h @@ -191,7 +191,7 @@ void xfs_log_ticket_put(struct xlog_ticket *ticket); xlog_tid_t xfs_log_get_trans_ident(struct xfs_trans *tp); -int xfs_log_commit_cil(struct xfs_mount *mp, struct xfs_trans *tp, +void xfs_log_commit_cil(struct xfs_mount *mp, struct xfs_trans *tp, struct xfs_log_vec *log_vector, xfs_lsn_t *commit_lsn, int flags); bool xfs_log_item_in_current_chkpt(struct xfs_log_item *lip); diff --git a/fs/xfs/xfs_log_cil.c b/fs/xfs/xfs_log_cil.c index c7eac5acbfea..9ca59be08977 100644 --- a/fs/xfs/xfs_log_cil.c +++ b/fs/xfs/xfs_log_cil.c @@ -625,7 +625,7 @@ out_abort: * background commit, returns without it held once background commits are * allowed again. */ -int +void xfs_log_commit_cil( struct xfs_mount *mp, struct xfs_trans *tp, @@ -640,11 +640,6 @@ xfs_log_commit_cil( if (flags & XFS_TRANS_RELEASE_LOG_RES) log_flags = XFS_LOG_REL_PERM_RESERV; - if (XLOG_FORCED_SHUTDOWN(log)) { - xlog_cil_free_logvec(log_vector); - return XFS_ERROR(EIO); - } - /* * do all the hard work of formatting items (including memory * allocation) outside the CIL context lock. This prevents stalling CIL @@ -704,7 +699,6 @@ xfs_log_commit_cil( */ if (push) xlog_cil_push(log, 0); - return 0; } /* diff --git a/fs/xfs/xfs_trans.c b/fs/xfs/xfs_trans.c index 29f5e5424897..76922793f64f 100644 --- a/fs/xfs/xfs_trans.c +++ b/fs/xfs/xfs_trans.c @@ -1755,7 +1755,6 @@ xfs_trans_commit_cil( int flags) { struct xfs_log_vec *log_vector; - int error; /* * Get each log item to allocate a vector structure for @@ -1766,9 +1765,7 @@ xfs_trans_commit_cil( if (!log_vector) return ENOMEM; - error = xfs_log_commit_cil(mp, tp, log_vector, commit_lsn, flags); - if (error) - return error; + xfs_log_commit_cil(mp, tp, log_vector, commit_lsn, flags); current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); xfs_trans_free(tp); -- cgit v1.2.3 From 0fbca4d1c3932c27c4794bf5c2b5fc961cf5a54f Mon Sep 17 00:00:00 2001 From: Dave Chinner Date: Fri, 28 Jan 2011 11:20:46 +1100 Subject: xfs: fix dquot shaker deadlock Commit 368e136 ("xfs: remove duplicate code from dquot reclaim") fails to unlock the dquot freelist when the number of loop restarts is exceeded in xfs_qm_dqreclaim_one(). This causes hangs in memory reclaim. Rework the loop control logic into an unwind stack that all the different cases jump into. This means there is only one set of code that processes the loop exit criteria, and simplifies the unlocking of all the items from different points in the loop. It also fixes a double increment of the restart counter from the qi_dqlist_lock case. Reported-by: Malcolm Scott Signed-off-by: Dave Chinner Reviewed-by: Alex Elder --- fs/xfs/quota/xfs_qm.c | 46 +++++++++++++++++++++------------------------- 1 file changed, 21 insertions(+), 25 deletions(-) (limited to 'fs') diff --git a/fs/xfs/quota/xfs_qm.c b/fs/xfs/quota/xfs_qm.c index f8e854b4fde8..206a2815ced6 100644 --- a/fs/xfs/quota/xfs_qm.c +++ b/fs/xfs/quota/xfs_qm.c @@ -1863,12 +1863,14 @@ xfs_qm_dqreclaim_one(void) xfs_dquot_t *dqpout; xfs_dquot_t *dqp; int restarts; + int startagain; restarts = 0; dqpout = NULL; /* lockorder: hashchainlock, freelistlock, mplistlock, dqlock, dqflock */ -startagain: +again: + startagain = 0; mutex_lock(&xfs_Gqm->qm_dqfrlist_lock); list_for_each_entry(dqp, &xfs_Gqm->qm_dqfrlist, q_freelist) { @@ -1885,13 +1887,10 @@ startagain: ASSERT(! (dqp->dq_flags & XFS_DQ_INACTIVE)); trace_xfs_dqreclaim_want(dqp); - - xfs_dqunlock(dqp); - mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock); - if (++restarts >= XFS_QM_RECLAIM_MAX_RESTARTS) - return NULL; XQM_STATS_INC(xqmstats.xs_qm_dqwants); - goto startagain; + restarts++; + startagain = 1; + goto dqunlock; } /* @@ -1906,23 +1905,20 @@ startagain: ASSERT(list_empty(&dqp->q_mplist)); list_del_init(&dqp->q_freelist); xfs_Gqm->qm_dqfrlist_cnt--; - xfs_dqunlock(dqp); dqpout = dqp; XQM_STATS_INC(xqmstats.xs_qm_dqinact_reclaims); - break; + goto dqunlock; } ASSERT(dqp->q_hash); ASSERT(!list_empty(&dqp->q_mplist)); /* - * Try to grab the flush lock. If this dquot is in the process of - * getting flushed to disk, we don't want to reclaim it. + * Try to grab the flush lock. If this dquot is in the process + * of getting flushed to disk, we don't want to reclaim it. */ - if (!xfs_dqflock_nowait(dqp)) { - xfs_dqunlock(dqp); - continue; - } + if (!xfs_dqflock_nowait(dqp)) + goto dqunlock; /* * We have the flush lock so we know that this is not in the @@ -1944,8 +1940,7 @@ startagain: xfs_fs_cmn_err(CE_WARN, mp, "xfs_qm_dqreclaim: dquot %p flush failed", dqp); } - xfs_dqunlock(dqp); /* dqflush unlocks dqflock */ - continue; + goto dqunlock; } /* @@ -1967,13 +1962,8 @@ startagain: */ if (!mutex_trylock(&mp->m_quotainfo->qi_dqlist_lock)) { restarts++; - mutex_unlock(&dqp->q_hash->qh_lock); - xfs_dqfunlock(dqp); - xfs_dqunlock(dqp); - mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock); - if (restarts++ >= XFS_QM_RECLAIM_MAX_RESTARTS) - return NULL; - goto startagain; + startagain = 1; + goto qhunlock; } ASSERT(dqp->q_nrefs == 0); @@ -1986,14 +1976,20 @@ startagain: xfs_Gqm->qm_dqfrlist_cnt--; dqpout = dqp; mutex_unlock(&mp->m_quotainfo->qi_dqlist_lock); +qhunlock: mutex_unlock(&dqp->q_hash->qh_lock); dqfunlock: xfs_dqfunlock(dqp); +dqunlock: xfs_dqunlock(dqp); if (dqpout) break; if (restarts >= XFS_QM_RECLAIM_MAX_RESTARTS) - return NULL; + break; + if (startagain) { + mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock); + goto again; + } } mutex_unlock(&xfs_Gqm->qm_dqfrlist_lock); return dqpout; -- cgit v1.2.3 From 24446fc66fdebbdd8baca0f44fd2a47ad77ba580 Mon Sep 17 00:00:00 2001 From: "bpm@sgi.com" Date: Wed, 19 Jan 2011 17:41:58 +0000 Subject: xfs: xfs_bmap_add_extent_delay_real should init br_startblock When filling in the middle of a previous delayed allocation in xfs_bmap_add_extent_delay_real, set br_startblock of the new delay extent to the right to nullstartblock instead of 0 before inserting the extent into the ifork (xfs_iext_insert), rather than setting br_startblock afterward. Adding the extent into the ifork with br_startblock=0 can lead to the extent being copied into the btree by xfs_bmap_extent_to_btree if we happen to convert from extents format to btree format before updating br_startblock with the correct value. The unexpected addition of this delay extent to the btree can cause subsequent XFS_WANT_CORRUPTED_GOTO filesystem shutdown in several xfs_bmap_add_extent_delay_real cases where we are converting a delay extent to real and unexpectedly find an extent already inserted. For example: 911 case BMAP_LEFT_FILLING: 912 /* 913 * Filling in the first part of a previous delayed allocation. 914 * The left neighbor is not contiguous. 915 */ 916 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_); 917 xfs_bmbt_set_startoff(ep, new_endoff); 918 temp = PREV.br_blockcount - new->br_blockcount; 919 xfs_bmbt_set_blockcount(ep, temp); 920 xfs_iext_insert(ip, idx, 1, new, state); 921 ip->i_df.if_lastex = idx; 922 ip->i_d.di_nextents++; 923 if (cur == NULL) 924 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT; 925 else { 926 rval = XFS_ILOG_CORE; 927 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff, 928 new->br_startblock, new->br_blockcount, 929 &i))) 930 goto done; 931 XFS_WANT_CORRUPTED_GOTO(i == 0, done); With the bogus extent in the btree we shutdown the filesystem at 931. The conversion from extents to btree format happens when the number of extents in the inode increases above ip->i_df.if_ext_max. xfs_bmap_extent_to_btree copies extents from the ifork into the btree, ignoring all delalloc extents which are denoted by br_startblock having some value of nullstartblock. SGI-PV: 1013221 Signed-off-by: Ben Myers Reviewed-by: Dave Chinner Signed-off-by: Alex Elder --- fs/xfs/xfs_bmap.c | 33 +++++++++++++++++++++++++-------- 1 file changed, 25 insertions(+), 8 deletions(-) (limited to 'fs') diff --git a/fs/xfs/xfs_bmap.c b/fs/xfs/xfs_bmap.c index 3e9c278a8f78..dc3afd7739ff 100644 --- a/fs/xfs/xfs_bmap.c +++ b/fs/xfs/xfs_bmap.c @@ -1038,17 +1038,34 @@ xfs_bmap_add_extent_delay_real( * Filling in the middle part of a previous delayed allocation. * Contiguity is impossible here. * This case is avoided almost all the time. + * + * We start with a delayed allocation: + * + * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+ + * PREV @ idx + * + * and we are allocating: + * +rrrrrrrrrrrrrrrrr+ + * new + * + * and we set it up for insertion as: + * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+ + * new + * PREV @ idx LEFT RIGHT + * inserted at idx + 1 */ temp = new->br_startoff - PREV.br_startoff; - trace_xfs_bmap_pre_update(ip, idx, 0, _THIS_IP_); - xfs_bmbt_set_blockcount(ep, temp); - r[0] = *new; - r[1].br_state = PREV.br_state; - r[1].br_startblock = 0; - r[1].br_startoff = new_endoff; temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff; - r[1].br_blockcount = temp2; - xfs_iext_insert(ip, idx + 1, 2, &r[0], state); + trace_xfs_bmap_pre_update(ip, idx, 0, _THIS_IP_); + xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */ + LEFT = *new; + RIGHT.br_state = PREV.br_state; + RIGHT.br_startblock = nullstartblock( + (int)xfs_bmap_worst_indlen(ip, temp2)); + RIGHT.br_startoff = new_endoff; + RIGHT.br_blockcount = temp2; + /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */ + xfs_iext_insert(ip, idx + 1, 2, &LEFT, state); ip->i_df.if_lastex = idx + 1; ip->i_d.di_nextents++; if (cur == NULL) -- cgit v1.2.3