From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from cuda.sgi.com (cuda3.sgi.com [192.48.176.15]) by oss.sgi.com (8.14.3/8.14.3/SuSE Linux 0.8) with ESMTP id nBK4NtC4148831 for ; Sat, 19 Dec 2009 22:23:56 -0600 Received: from mail.internode.on.net (localhost [127.0.0.1]) by cuda.sgi.com (Spam Firewall) with ESMTP id 76F9219C24E7 for ; Sat, 19 Dec 2009 20:24:34 -0800 (PST) Received: from mail.internode.on.net (bld-mail16.adl2.internode.on.net [150.101.137.101]) by cuda.sgi.com with ESMTP id GHoNEu2wDtoH5uf6 for ; Sat, 19 Dec 2009 20:24:34 -0800 (PST) Received: from discord (unverified [121.44.162.26]) by mail.internode.on.net (SurgeMail 3.8f2) with ESMTP id 10370739-1927428 for ; Sun, 20 Dec 2009 14:54:33 +1030 (CDT) Received: from dave by discord with local (Exim 4.69) (envelope-from ) id 1NMDLL-0000Ie-C0 for xfs@oss.sgi.com; Sun, 20 Dec 2009 15:24:31 +1100 Date: Sun, 20 Dec 2009 15:24:31 +1100 From: Dave Chinner Subject: Re: [PATCH 2/6] XFS: Don't directly reference m_perag in allocation code Message-ID: <20091220042431.GI4850@discord.disaster> References: <1260857477-2368-1-git-send-email-david@fromorbit.com> <1260857477-2368-3-git-send-email-david@fromorbit.com> MIME-Version: 1.0 Content-Disposition: inline In-Reply-To: <1260857477-2368-3-git-send-email-david@fromorbit.com> List-Id: XFS Filesystem from SGI List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Sender: xfs-bounces@oss.sgi.com Errors-To: xfs-bounces@oss.sgi.com To: xfs@oss.sgi.com On Tue, Dec 15, 2009 at 05:11:13PM +1100, Dave Chinner wrote: > Start abstracting the perag references so that the indexing of the > structures is not directly coded into all the places that uses the > perag structures. This will allow us to separate the use of the > perag structure and the way it is indexed and hence avoid the known > deadlocks related to growing a busy filesystem. > > Signed-off-by: Dave Chinner I screwed up the rebase so this patch is missing a xfs_perag_put() call in xfs_alloc_get_freelist(), so will assert fail on unmount in test 042. Corrected patch below. Cheers, Dave. -- Dave Chinner david@fromorbit.com XFS: Don't directly reference m_perag in allocation code Start abstracting the perag references so that the indexing of the structures is not directly coded into all the places that uses the perag structures. This will allow us to separate the use of the perag structure and the way it is indexed and hence avoid the known deadlocks related to growing a busy filesystem. Signed-off-by: Dave Chinner --- fs/xfs/xfs_alloc.c | 82 ++++++++++++++++++++++++++-------------------- fs/xfs/xfs_alloc_btree.c | 9 ++++- 2 files changed, 53 insertions(+), 38 deletions(-) diff --git a/fs/xfs/xfs_alloc.c b/fs/xfs/xfs_alloc.c index a1c65fc..d39ee2d 100644 --- a/fs/xfs/xfs_alloc.c +++ b/fs/xfs/xfs_alloc.c @@ -1662,11 +1662,13 @@ xfs_free_ag_extent( xfs_agf_t *agf; xfs_perag_t *pag; /* per allocation group data */ + pag = xfs_perag_get(mp, agno); + pag->pagf_freeblks += len; + xfs_perag_put(pag); + agf = XFS_BUF_TO_AGF(agbp); - pag = &mp->m_perag[agno]; be32_add_cpu(&agf->agf_freeblks, len); xfs_trans_agblocks_delta(tp, len); - pag->pagf_freeblks += len; XFS_WANT_CORRUPTED_GOTO( be32_to_cpu(agf->agf_freeblks) <= be32_to_cpu(agf->agf_length), @@ -1969,10 +1971,12 @@ xfs_alloc_get_freelist( xfs_trans_brelse(tp, agflbp); if (be32_to_cpu(agf->agf_flfirst) == XFS_AGFL_SIZE(mp)) agf->agf_flfirst = 0; - pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)]; + + pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno)); be32_add_cpu(&agf->agf_flcount, -1); xfs_trans_agflist_delta(tp, -1); pag->pagf_flcount--; + xfs_perag_put(pag); logflags = XFS_AGF_FLFIRST | XFS_AGF_FLCOUNT; if (btreeblk) { @@ -2078,7 +2082,8 @@ xfs_alloc_put_freelist( be32_add_cpu(&agf->agf_fllast, 1); if (be32_to_cpu(agf->agf_fllast) == XFS_AGFL_SIZE(mp)) agf->agf_fllast = 0; - pag = &mp->m_perag[be32_to_cpu(agf->agf_seqno)]; + + pag = xfs_perag_get(mp, be32_to_cpu(agf->agf_seqno)); be32_add_cpu(&agf->agf_flcount, 1); xfs_trans_agflist_delta(tp, 1); pag->pagf_flcount++; @@ -2089,6 +2094,7 @@ xfs_alloc_put_freelist( pag->pagf_btreeblks--; logflags |= XFS_AGF_BTREEBLKS; } + xfs_perag_put(pag); xfs_alloc_log_agf(tp, agbp, logflags); @@ -2152,7 +2158,6 @@ xfs_read_agf( xfs_trans_brelse(tp, *bpp); return XFS_ERROR(EFSCORRUPTED); } - XFS_BUF_SET_VTYPE_REF(*bpp, B_FS_AGF, XFS_AGF_REF); return 0; } @@ -2184,7 +2189,7 @@ xfs_alloc_read_agf( ASSERT(!XFS_BUF_GETERROR(*bpp)); agf = XFS_BUF_TO_AGF(*bpp); - pag = &mp->m_perag[agno]; + pag = xfs_perag_get(mp, agno); if (!pag->pagf_init) { pag->pagf_freeblks = be32_to_cpu(agf->agf_freeblks); pag->pagf_btreeblks = be32_to_cpu(agf->agf_btreeblks); @@ -2211,6 +2216,7 @@ xfs_alloc_read_agf( be32_to_cpu(agf->agf_levels[XFS_BTNUM_CNTi])); } #endif + xfs_perag_put(pag); return 0; } @@ -2271,7 +2277,7 @@ xfs_alloc_vextent( */ args->agno = XFS_FSB_TO_AGNO(mp, args->fsbno); down_read(&mp->m_peraglock); - args->pag = &mp->m_perag[args->agno]; + args->pag = xfs_perag_get(mp, args->agno); args->minleft = 0; error = xfs_alloc_fix_freelist(args, 0); args->minleft = minleft; @@ -2341,7 +2347,7 @@ xfs_alloc_vextent( */ down_read(&mp->m_peraglock); for (;;) { - args->pag = &mp->m_perag[args->agno]; + args->pag = xfs_perag_get(mp, args->agno); if (no_min) args->minleft = 0; error = xfs_alloc_fix_freelist(args, flags); args->minleft = minleft; @@ -2400,6 +2406,7 @@ xfs_alloc_vextent( } } } + xfs_perag_put(args->pag); } up_read(&mp->m_peraglock); if (bump_rotor || (type == XFS_ALLOCTYPE_ANY_AG)) { @@ -2427,8 +2434,10 @@ xfs_alloc_vextent( args->len); #endif } + xfs_perag_put(args->pag); return 0; error0: + xfs_perag_put(args->pag); up_read(&mp->m_peraglock); return error; } @@ -2455,7 +2464,7 @@ xfs_free_extent( ASSERT(args.agno < args.mp->m_sb.sb_agcount); args.agbno = XFS_FSB_TO_AGBNO(args.mp, bno); down_read(&args.mp->m_peraglock); - args.pag = &args.mp->m_perag[args.agno]; + args.pag = xfs_perag_get(args.mp, args.agno); if ((error = xfs_alloc_fix_freelist(&args, XFS_ALLOC_FLAG_FREEING))) goto error0; #ifdef DEBUG @@ -2465,6 +2474,7 @@ xfs_free_extent( #endif error = xfs_free_ag_extent(tp, args.agbp, args.agno, args.agbno, len, 0); error0: + xfs_perag_put(args.pag); up_read(&args.mp->m_peraglock); return error; } @@ -2486,15 +2496,15 @@ xfs_alloc_mark_busy(xfs_trans_t *tp, xfs_agblock_t bno, xfs_extlen_t len) { - xfs_mount_t *mp; xfs_perag_busy_t *bsy; + struct xfs_perag *pag; int n; - mp = tp->t_mountp; - spin_lock(&mp->m_perag[agno].pagb_lock); + pag = xfs_perag_get(tp->t_mountp, agno); + spin_lock(&pag->pagb_lock); /* search pagb_list for an open slot */ - for (bsy = mp->m_perag[agno].pagb_list, n = 0; + for (bsy = pag->pagb_list, n = 0; n < XFS_PAGB_NUM_SLOTS; bsy++, n++) { if (bsy->busy_tp == NULL) { @@ -2502,11 +2512,11 @@ xfs_alloc_mark_busy(xfs_trans_t *tp, } } - trace_xfs_alloc_busy(mp, agno, bno, len, n); + trace_xfs_alloc_busy(tp->t_mountp, agno, bno, len, n); if (n < XFS_PAGB_NUM_SLOTS) { - bsy = &mp->m_perag[agno].pagb_list[n]; - mp->m_perag[agno].pagb_count++; + bsy = &pag->pagb_list[n]; + pag->pagb_count++; bsy->busy_start = bno; bsy->busy_length = len; bsy->busy_tp = tp; @@ -2521,7 +2531,8 @@ xfs_alloc_mark_busy(xfs_trans_t *tp, xfs_trans_set_sync(tp); } - spin_unlock(&mp->m_perag[agno].pagb_lock); + spin_unlock(&pag->pagb_lock); + xfs_perag_put(pag); } void @@ -2529,24 +2540,23 @@ xfs_alloc_clear_busy(xfs_trans_t *tp, xfs_agnumber_t agno, int idx) { - xfs_mount_t *mp; + struct xfs_perag *pag; xfs_perag_busy_t *list; - mp = tp->t_mountp; - - spin_lock(&mp->m_perag[agno].pagb_lock); - list = mp->m_perag[agno].pagb_list; - ASSERT(idx < XFS_PAGB_NUM_SLOTS); + pag = xfs_perag_get(tp->t_mountp, agno); + spin_lock(&pag->pagb_lock); + list = pag->pagb_list; - trace_xfs_alloc_unbusy(mp, agno, idx, list[idx].busy_tp == tp); + trace_xfs_alloc_unbusy(tp->t_mountp, agno, idx, list[idx].busy_tp == tp); if (list[idx].busy_tp == tp) { list[idx].busy_tp = NULL; - mp->m_perag[agno].pagb_count--; + pag->pagb_count--; } - spin_unlock(&mp->m_perag[agno].pagb_lock); + spin_unlock(&pag->pagb_lock); + xfs_perag_put(pag); } @@ -2560,21 +2570,20 @@ xfs_alloc_search_busy(xfs_trans_t *tp, xfs_agblock_t bno, xfs_extlen_t len) { - xfs_mount_t *mp; + struct xfs_perag *pag; xfs_perag_busy_t *bsy; xfs_agblock_t uend, bend; xfs_lsn_t lsn; int cnt; - mp = tp->t_mountp; - - spin_lock(&mp->m_perag[agno].pagb_lock); - cnt = mp->m_perag[agno].pagb_count; + pag = xfs_perag_get(tp->t_mountp, agno); + spin_lock(&pag->pagb_lock); + cnt = pag->pagb_count; uend = bno + len - 1; /* search pagb_list for this slot, skipping open slots */ - for (bsy = mp->m_perag[agno].pagb_list; cnt; bsy++) { + for (bsy = pag->pagb_list; cnt; bsy++) { /* * (start1,length1) within (start2, length2) @@ -2589,7 +2598,7 @@ xfs_alloc_search_busy(xfs_trans_t *tp, } } - trace_xfs_alloc_busysearch(mp, agno, bno, len, !!cnt); + trace_xfs_alloc_busysearch(tp->t_mountp, agno, bno, len, !!cnt); /* * If a block was found, force the log through the LSN of the @@ -2597,9 +2606,10 @@ xfs_alloc_search_busy(xfs_trans_t *tp, */ if (cnt) { lsn = bsy->busy_tp->t_commit_lsn; - spin_unlock(&mp->m_perag[agno].pagb_lock); - xfs_log_force(mp, lsn, XFS_LOG_FORCE|XFS_LOG_SYNC); + spin_unlock(&pag->pagb_lock); + xfs_log_force(tp->t_mountp, lsn, XFS_LOG_FORCE|XFS_LOG_SYNC); } else { - spin_unlock(&mp->m_perag[agno].pagb_lock); + spin_unlock(&pag->pagb_lock); } + xfs_perag_put(pag); } diff --git a/fs/xfs/xfs_alloc_btree.c b/fs/xfs/xfs_alloc_btree.c index adbd914..b726e10 100644 --- a/fs/xfs/xfs_alloc_btree.c +++ b/fs/xfs/xfs_alloc_btree.c @@ -61,12 +61,14 @@ xfs_allocbt_set_root( struct xfs_agf *agf = XFS_BUF_TO_AGF(agbp); xfs_agnumber_t seqno = be32_to_cpu(agf->agf_seqno); int btnum = cur->bc_btnum; + struct xfs_perag *pag = xfs_perag_get(cur->bc_mp, seqno); ASSERT(ptr->s != 0); agf->agf_roots[btnum] = ptr->s; be32_add_cpu(&agf->agf_levels[btnum], inc); - cur->bc_mp->m_perag[seqno].pagf_levels[btnum] += inc; + pag->pagf_levels[btnum] += inc; + xfs_perag_put(pag); xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS); } @@ -150,6 +152,7 @@ xfs_allocbt_update_lastrec( { struct xfs_agf *agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp); xfs_agnumber_t seqno = be32_to_cpu(agf->agf_seqno); + struct xfs_perag *pag; __be32 len; int numrecs; @@ -193,7 +196,9 @@ xfs_allocbt_update_lastrec( } agf->agf_longest = len; - cur->bc_mp->m_perag[seqno].pagf_longest = be32_to_cpu(len); + pag = xfs_perag_get(cur->bc_mp, seqno); + pag->pagf_longest = be32_to_cpu(len); + xfs_perag_put(pag); xfs_alloc_log_agf(cur->bc_tp, cur->bc_private.a.agbp, XFS_AGF_LONGEST); } _______________________________________________ xfs mailing list xfs@oss.sgi.com http://oss.sgi.com/mailman/listinfo/xfs