From 53c7411919a64d6f0889aa0d6974610f6cd35744 Mon Sep 17 00:00:00 2001 From: Brian Behlendorf Date: Wed, 5 Dec 2012 13:00:49 -0800 Subject: [PATCH] Revert "Fix unlink/xattr deadlock" This reverts commit b00131d43ca344d4b205a03ab3eb771a060e5087 which is no longer needed due to e89260a1c8851ce05ea04b23606ba438b271d890. This change forces all xattr znodes to hold a reference on their parent which ensures prune_icache() will never attempt to evict both the parent and child concurrently. This effectively prevents the deadlock condition from ever occuring. Therefore we can safely revert back to the upstream synchronous cleanup code. This is nice because it keeps our code base closer to upstream and resolves the performance issues introduced by the original deadlock fix. Signed-off-by: Brian Behlendorf Closes #457 --- module/zfs/zfs_dir.c | 139 ++++++++++++++++------------------------ module/zfs/zfs_vfsops.c | 6 -- 2 files changed, 55 insertions(+), 90 deletions(-) diff --git a/module/zfs/zfs_dir.c b/module/zfs/zfs_dir.c index 670e31398..a373537dd 100644 --- a/module/zfs/zfs_dir.c +++ b/module/zfs/zfs_dir.c @@ -482,6 +482,57 @@ zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx) zap_add_int(zsb->z_os, zsb->z_unlinkedobj, zp->z_id, tx)); } +/* + * Clean up any znodes that had no links when we either crashed or + * (force) umounted the file system. + */ +void +zfs_unlinked_drain(zfs_sb_t *zsb) +{ + zap_cursor_t zc; + zap_attribute_t zap; + dmu_object_info_t doi; + znode_t *zp; + int error; + + /* + * Interate over the contents of the unlinked set. + */ + for (zap_cursor_init(&zc, zsb->z_os, zsb->z_unlinkedobj); + zap_cursor_retrieve(&zc, &zap) == 0; + zap_cursor_advance(&zc)) { + + /* + * See what kind of object we have in list + */ + + error = dmu_object_info(zsb->z_os, zap.za_first_integer, &doi); + if (error != 0) + continue; + + ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) || + (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS)); + /* + * We need to re-mark these list entries for deletion, + * so we pull them back into core and set zp->z_unlinked. + */ + error = zfs_zget(zsb, zap.za_first_integer, &zp); + + /* + * We may pick up znodes that are already marked for deletion. + * This could happen during the purge of an extended attribute + * directory. All we need to do is skip over them, since they + * are already in the system marked z_unlinked. + */ + if (error != 0) + continue; + + zp->z_unlinked = B_TRUE; + iput(ZTOI(zp)); + } + zap_cursor_fini(&zc); +} + /* * Delete the entire contents of a directory. Return a count * of the number of entries that could not be deleted. If we encounter @@ -548,71 +599,6 @@ zfs_purgedir(znode_t *dzp) return (skipped); } -/* - * Clean up any znodes that had no links when we either crashed or - * (force) umounted the file system. - */ -void -zfs_unlinked_drain(zfs_sb_t *zsb) -{ - zap_cursor_t zc; - zap_attribute_t zap; - dmu_object_info_t doi; - znode_t *zp; - int error; - - /* - * Interate over the contents of the unlinked set. - */ - for (zap_cursor_init(&zc, zsb->z_os, zsb->z_unlinkedobj); - zap_cursor_retrieve(&zc, &zap) == 0; - zap_cursor_advance(&zc)) { - - /* - * See what kind of object we have in list - */ - - error = dmu_object_info(zsb->z_os, zap.za_first_integer, &doi); - if (error != 0) - continue; - - ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) || - (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS)); - /* - * We need to re-mark these list entries for deletion, - * so we pull them back into core and set zp->z_unlinked. - */ - error = zfs_zget(zsb, zap.za_first_integer, &zp); - - /* - * We may pick up znodes that are already marked for deletion. - * This could happen during the purge of an extended attribute - * directory. All we need to do is skip over them, since they - * are already in the system marked z_unlinked. - */ - if (error != 0) - continue; - - zp->z_unlinked = B_TRUE; - - /* - * If this is an attribute directory, purge its contents. - */ - if (S_ISDIR(ZTOI(zp)->i_mode) && (zp->z_pflags & ZFS_XATTR)) { - /* - * We don't need to check the return value of - * zfs_purgedir here, because zfs_rmnode will just - * return this xattr directory to the unlinked set - * until all of its xattrs are gone. - */ - (void) zfs_purgedir(zp); - } - - iput(ZTOI(zp)); - } - zap_cursor_fini(&zc); -} - void zfs_rmnode(znode_t *zp) { @@ -622,7 +608,6 @@ zfs_rmnode(znode_t *zp) dmu_tx_t *tx; uint64_t acl_obj; uint64_t xattr_obj; - uint64_t count; int error; ASSERT(zp->z_links == 0); @@ -632,27 +617,13 @@ zfs_rmnode(znode_t *zp) * If this is an attribute directory, purge its contents. */ if (S_ISDIR(ZTOI(zp)->i_mode) && (zp->z_pflags & ZFS_XATTR)) { - error = zap_count(os, zp->z_id, &count); - if (error) { - zfs_znode_dmu_fini(zp); - return; - } - - if (count > 0) { - taskq_t *taskq; - + if (zfs_purgedir(zp) != 0) { /* - * There are still directory entries in this xattr - * directory. Let zfs_unlinked_drain() deal with - * them to avoid deadlocking this process in the - * zfs_purgedir()->zfs_zget()->ilookup() callpath - * on the xattr inode's I_FREEING bit. + * Not enough space to delete some xattrs. + * Leave it in the unlinked set. */ - taskq = dsl_pool_iput_taskq(dmu_objset_pool(os)); - taskq_dispatch(taskq, (task_func_t *) - zfs_unlinked_drain, zsb, TQ_SLEEP); - zfs_znode_dmu_fini(zp); + return; } } diff --git a/module/zfs/zfs_vfsops.c b/module/zfs/zfs_vfsops.c index 90f9055af..175dca821 100644 --- a/module/zfs/zfs_vfsops.c +++ b/module/zfs/zfs_vfsops.c @@ -1040,12 +1040,6 @@ zfs_sb_teardown(zfs_sb_t *zsb, boolean_t unmounting) (void) spl_invalidate_inodes(zsb->z_parent->z_sb, 0); } - /* - * Drain the iput_taskq to ensure all active references to the - * zfs_sb_t have been handled only then can it be safely destroyed. - */ - taskq_wait(dsl_pool_iput_taskq(dmu_objset_pool(zsb->z_os))); - /* * Close the zil. NB: Can't close the zil while zfs_inactive * threads are blocked as zil_close can call zfs_inactive.