Date: Wed, 12 May 2010 09:51:57 +0000 (UTC) From: Martin Matuska <mm@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r207956 - head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs Message-ID: <201005120951.o4C9pvWs065377@svn.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: mm Date: Wed May 12 09:51:57 2010 New Revision: 207956 URL: http://svn.freebsd.org/changeset/base/207956 Log: Fix possible hang when replaying large truncations. OpenSolaris onnv revision: 7904:6a124a4ca9c5 Approved by: pjd, delphij (mentor) Obtained from: OpenSolaris (Bug ID 6761624) MFC after: 3 days Modified: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zil.c Modified: head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zil.c ============================================================================== --- head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zil.c Wed May 12 09:34:10 2010 (r207955) +++ head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zil.c Wed May 12 09:51:57 2010 (r207956) @@ -1567,6 +1567,29 @@ zil_replay_log_record(zilog_t *zilog, lr } /* + * Replay of large truncates can end up needing additional txs + * and a different txg. If they are nested within the replay tx + * as below then a hang is possible. So we do the truncate here + * and redo the truncate later (a no-op) and update the sequence + * number whilst in the replay tx. Fortunately, it's safe to repeat + * a truncate if we crash and the truncate commits. A create over + * an existing file will also come in as a TX_TRUNCATE record. + * + * Note, remove of large files and renames over large files is + * handled by putting the deleted object on a stable list + * and if necessary force deleting the object outside of the replay + * transaction using the zr_replay_cleaner. + */ + if (txtype == TX_TRUNCATE) { + *zr->zr_txgp = TXG_NOWAIT; + error = zr->zr_replay[TX_TRUNCATE](zr->zr_arg, zr->zr_lrbuf, + zr->zr_byteswap); + if (error) + goto bad; + zr->zr_byteswap = 0; /* only byteswap once */ + } + + /* * We must now do two things atomically: replay this log record, * and update the log header to reflect the fact that we did so. * We use the DMU's ability to assign into a specific txg to do this. @@ -1636,6 +1659,7 @@ zil_replay_log_record(zilog_t *zilog, lr dprintf("pass %d, retrying\n", pass); } +bad: ASSERT(error && error != ERESTART); name = kmem_alloc(MAXNAMELEN, KM_SLEEP); dmu_objset_name(zr->zr_os, name);
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201005120951.o4C9pvWs065377>