upstream/illumos/illumos-gate: comparison usr/src/uts/common/fs/zfs/zil.c

equal deleted inserted replaced

-:8aac17999e4d
+:e2081f502306
 * Use is subject to license terms.
 */
 #include <sys/zfs_context.h>
 #include <sys/spa.h>
-#include <sys/spa_impl.h>
 #include <sys/dmu.h>
 #include <sys/zap.h>
 #include <sys/arc.h>
 #include <sys/stat.h>
 #include <sys/resource.h>
 static kmem_cache_t *zil_lwb_cache;
 static boolean_t zil_empty(zilog_t *zilog);
 static int
-zil_dva_compare(const void *x1, const void *x2)
+zil_bp_compare(const void *x1, const void *x2)
 {
-	const dva_t *dva1 = x1;
+	const dva_t *dva1 = &((zil_bp_node_t *)x1)->zn_dva;
-	const dva_t *dva2 = x2;
+	const dva_t *dva2 = &((zil_bp_node_t *)x2)->zn_dva;
 	if (DVA_GET_VDEV(dva1) < DVA_GET_VDEV(dva2))
 		return (-1);
 	if (DVA_GET_VDEV(dva1) > DVA_GET_VDEV(dva2))
 		return (1);
 	return (0);
 }
 static void
-zil_dva_tree_init(avl_tree_t *t)
+zil_bp_tree_init(zilog_t *zilog)
 {
-	avl_create(t, zil_dva_compare, sizeof (zil_dva_node_t),
+	avl_create(&zilog->zl_bp_tree, zil_bp_compare,
-	    offsetof(zil_dva_node_t, zn_node));
+	    sizeof (zil_bp_node_t), offsetof(zil_bp_node_t, zn_node));
 }
 static void
-zil_dva_tree_fini(avl_tree_t *t)
+zil_bp_tree_fini(zilog_t *zilog)
 {
-	zil_dva_node_t *zn;
+	avl_tree_t *t = &zilog->zl_bp_tree;
+	zil_bp_node_t *zn;
 	void *cookie = NULL;
 	while ((zn = avl_destroy_nodes(t, &cookie)) != NULL)
-		kmem_free(zn, sizeof (zil_dva_node_t));
+		kmem_free(zn, sizeof (zil_bp_node_t));
 	avl_destroy(t);
 }
-static int
+int
-zil_dva_tree_add(avl_tree_t *t, dva_t *dva)
+zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp)
 {
-	zil_dva_node_t *zn;
+	avl_tree_t *t = &zilog->zl_bp_tree;
+	const dva_t *dva = BP_IDENTITY(bp);
+	zil_bp_node_t *zn;
 	avl_index_t where;
 	if (avl_find(t, dva, &where) != NULL)
 		return (EEXIST);
-	zn = kmem_alloc(sizeof (zil_dva_node_t), KM_SLEEP);
+	zn = kmem_alloc(sizeof (zil_bp_node_t), KM_SLEEP);
 	zn->zn_dva = *dva;
 	avl_insert(t, zn, where);
 	return (0);
 }
 	zc->zc_word[ZIL_ZC_OBJSET] = dmu_objset_id(zilog->zl_os);
 	zc->zc_word[ZIL_ZC_SEQ] = 1ULL;
 }
 /*
-* Read a log block, make sure it's valid, and byteswap it if necessary.
+* Read a log block and make sure it's valid.
 */
 static int
-zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, arc_buf_t **abufpp)
+zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, blkptr_t *nbp, void *dst)
 {
-	blkptr_t blk = *bp;
+	enum zio_flag zio_flags = ZIO_FLAG_CANFAIL;
+	uint32_t aflags = ARC_WAIT;
+	arc_buf_t *abuf = NULL;
 	zbookmark_t zb;
-	uint32_t aflags = ARC_WAIT;
 	int error;
-	zb.zb_objset = bp->blk_cksum.zc_word[ZIL_ZC_OBJSET];
+	if (zilog->zl_header->zh_claim_txg == 0)
-	zb.zb_object = 0;
+		zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB;
-	zb.zb_level = -1;
-	zb.zb_blkid = bp->blk_cksum.zc_word[ZIL_ZC_SEQ];
+	if (!(zilog->zl_header->zh_flags & ZIL_CLAIM_LR_SEQ_VALID))
+		zio_flags |= ZIO_FLAG_SPECULATIVE;
-	*abufpp = NULL;
+	SET_BOOKMARK(&zb, bp->blk_cksum.zc_word[ZIL_ZC_OBJSET],
-	/*
+	    ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
-	 * We shouldn't be doing any scrubbing while we're doing log
-	 * replay, it's OK to not lock.
+	error = arc_read_nolock(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf,
-	 */
+	    ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb);
-	error = arc_read_nolock(NULL, zilog->zl_spa, &blk,
-	    arc_getbuf_func, abufpp, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL |
-	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB, &aflags, &zb);
 	if (error == 0) {
-		char *data = (*abufpp)->b_data;
+		char *data = abuf->b_data;
-		uint64_t blksz = BP_GET_LSIZE(bp);
+		uint64_t size = BP_GET_LSIZE(bp);
-		zil_trailer_t *ztp = (zil_trailer_t *)(data + blksz) - 1;
+		zil_trailer_t *ztp = (zil_trailer_t *)(data + size) - 1;
 		zio_cksum_t cksum = bp->blk_cksum;
+		bcopy(data, dst, size);
+		*nbp = ztp->zit_next_blk;
 		/*
 		 * Validate the checksummed log block.
 		 *
 		 * Sequence numbers should be... sequential.  The checksum
 		 */
 		cksum.zc_word[ZIL_ZC_SEQ]++;
 		if (bcmp(&cksum, &ztp->zit_next_blk.blk_cksum,
 		    sizeof (cksum)) || BP_IS_HOLE(&ztp->zit_next_blk) ||
-		    (ztp->zit_nused > (blksz - sizeof (zil_trailer_t)))) {
+		    (ztp->zit_nused > (size - sizeof (zil_trailer_t))))
 			error = ECKSUM;
-		}
+		VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1);
-		if (error) {
+	}
-			VERIFY(arc_buf_remove_ref(*abufpp, abufpp) == 1);
-			*abufpp = NULL;
-		}
-	}
-	dprintf("error %d on %llu:%llu\n", error, zb.zb_objset, zb.zb_blkid);
 	return (error);
 }
 /*
+* Read a TX_WRITE log data block.
+*/
+static int
+zil_read_log_data(zilog_t *zilog, const lr_write_t *lr, void *wbuf)
+{
+	enum zio_flag zio_flags = ZIO_FLAG_CANFAIL;
+	const blkptr_t *bp = &lr->lr_blkptr;
+	uint32_t aflags = ARC_WAIT;
+	arc_buf_t *abuf = NULL;
+	zbookmark_t zb;
+	int error;
+	if (BP_IS_HOLE(bp)) {
+		if (wbuf != NULL)
+			bzero(wbuf, MAX(BP_GET_LSIZE(bp), lr->lr_length));
+		return (0);
+	}
+	if (zilog->zl_header->zh_claim_txg == 0)
+		zio_flags |= ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB;
+	SET_BOOKMARK(&zb, dmu_objset_id(zilog->zl_os), lr->lr_foid,
+	    ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
+	error = arc_read_nolock(NULL, zilog->zl_spa, bp, arc_getbuf_func, &abuf,
+	    ZIO_PRIORITY_SYNC_READ, zio_flags, &aflags, &zb);
+	if (error == 0) {
+		if (wbuf != NULL)
+			bcopy(abuf->b_data, wbuf, arc_buf_size(abuf));
+		(void) arc_buf_remove_ref(abuf, &abuf);
+	}
+	return (error);
+}
+/*
 * Parse the intent log, and call parse_func for each valid record within.
-* Return the highest sequence number.
+*/
-*/
+int
-uint64_t
 zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
 zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg)
 {
 	const zil_header_t *zh = zilog->zl_header;
-	uint64_t claim_seq = zh->zh_claim_seq;
+	boolean_t claimed = !!zh->zh_claim_txg;
-	uint64_t seq = 0;
+	uint64_t claim_blk_seq = claimed ? zh->zh_claim_blk_seq : UINT64_MAX;
-	uint64_t max_seq = 0;
+	uint64_t claim_lr_seq = claimed ? zh->zh_claim_lr_seq : UINT64_MAX;
-	blkptr_t blk = zh->zh_log;
+	uint64_t max_blk_seq = 0;
-	arc_buf_t *abuf;
+	uint64_t max_lr_seq = 0;
+	uint64_t blk_count = 0;
+	uint64_t lr_count = 0;
+	blkptr_t blk, next_blk;
 	char *lrbuf, *lrp;
-	zil_trailer_t *ztp;
+	int error = 0;
-	int reclen, error;
+	/*
-	if (BP_IS_HOLE(&blk))
+	 * Old logs didn't record the maximum zh_claim_lr_seq.
-		return (max_seq);
+	 */
+	if (!(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID))
+		claim_lr_seq = UINT64_MAX;
 	/*
 	 * Starting at the block pointed to by zh_log we read the log chain.
 	 * For each block in the chain we strongly check that block to
 	 * ensure its validity.  We stop when an invalid block is found.
 	 * For each block pointer in the chain we call parse_blk_func().
 	 * For each record in each valid block we call parse_lr_func().
 	 * If the log has been claimed, stop if we encounter a sequence
 	 * number greater than the highest claimed sequence number.
 	 */
-	zil_dva_tree_init(&zilog->zl_dva_tree);
+	lrbuf = zio_buf_alloc(SPA_MAXBLOCKSIZE);
-	for (;;) {
+	zil_bp_tree_init(zilog);
-		seq = blk.blk_cksum.zc_word[ZIL_ZC_SEQ];
+	for (blk = zh->zh_log; !BP_IS_HOLE(&blk); blk = next_blk) {
-		if (claim_seq != 0 && seq > claim_seq)
+		zil_trailer_t *ztp =
+		    (zil_trailer_t *)(lrbuf + BP_GET_LSIZE(&blk)) - 1;
+		uint64_t blk_seq = blk.blk_cksum.zc_word[ZIL_ZC_SEQ];
+		int reclen;
+		if (blk_seq > claim_blk_seq)
 			break;
+		if ((error = parse_blk_func(zilog, &blk, arg, txg)) != 0)
-		ASSERT(max_seq < seq);
+			break;
-		max_seq = seq;
+		ASSERT(max_blk_seq < blk_seq);
+		max_blk_seq = blk_seq;
-		error = zil_read_log_block(zilog, &blk, &abuf);
+		blk_count++;
-		if (parse_blk_func != NULL)
+		if (max_lr_seq == claim_lr_seq && max_blk_seq == claim_blk_seq)
-			parse_blk_func(zilog, &blk, arg, txg);
+			break;
+		error = zil_read_log_block(zilog, &blk, &next_blk, lrbuf);
 		if (error)
 			break;
-		lrbuf = abuf->b_data;
-		ztp = (zil_trailer_t *)(lrbuf + BP_GET_LSIZE(&blk)) - 1;
-		blk = ztp->zit_next_blk;
-		if (parse_lr_func == NULL) {
-			VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1);
-			continue;
-		}
 		for (lrp = lrbuf; lrp < lrbuf + ztp->zit_nused; lrp += reclen) {
 			lr_t *lr = (lr_t *)lrp;
 			reclen = lr->lrc_reclen;
 			ASSERT3U(reclen, >=, sizeof (lr_t));
-			parse_lr_func(zilog, lr, arg, txg);
+			if (lr->lrc_seq > claim_lr_seq)
+				goto done;
+			if ((error = parse_lr_func(zilog, lr, arg, txg)) != 0)
+				goto done;
+			ASSERT(max_lr_seq < lr->lrc_seq);
+			max_lr_seq = lr->lrc_seq;
+			lr_count++;
 		}
-		VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1);
+	}
-	}
+done:
-	zil_dva_tree_fini(&zilog->zl_dva_tree);
+	zilog->zl_parse_error = error;
+	zilog->zl_parse_blk_seq = max_blk_seq;
-	return (max_seq);
+	zilog->zl_parse_lr_seq = max_lr_seq;
+	zilog->zl_parse_blk_count = blk_count;
+	zilog->zl_parse_lr_count = lr_count;
+	ASSERT(!claimed || !(zh->zh_flags & ZIL_CLAIM_LR_SEQ_VALID) ||
+	    (max_blk_seq == claim_blk_seq && max_lr_seq == claim_lr_seq));
+	zil_bp_tree_fini(zilog);
+	zio_buf_free(lrbuf, SPA_MAXBLOCKSIZE);
+	return (error);
+}
+static int
+zil_claim_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg)
+{
+	/*
+	 * Claim log block if not already committed and not already claimed.
+	 * If tx == NULL, just verify that the block is claimable.
+	 */
+	if (bp->blk_birth < first_txg || zil_bp_tree_add(zilog, bp) != 0)
+		return (0);
+	return (zio_wait(zio_claim(NULL, zilog->zl_spa,
+	    tx == NULL ? 0 : first_txg, bp, spa_claim_notify, NULL,
+	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB)));
+}
+static int
+zil_claim_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg)
+{
+	lr_write_t *lr = (lr_write_t *)lrc;
+	int error;
+	if (lrc->lrc_txtype != TX_WRITE)
+		return (0);
+	/*
+	 * If the block is not readable, don't claim it.  This can happen
+	 * in normal operation when a log block is written to disk before
+	 * some of the dmu_sync() blocks it points to.  In this case, the
+	 * transaction cannot have been committed to anyone (we would have
+	 * waited for all writes to be stable first), so it is semantically
+	 * correct to declare this the end of the log.
+	 */
+	if (lr->lr_blkptr.blk_birth >= first_txg &&
+	    (error = zil_read_log_data(zilog, lr, NULL)) != 0)
+		return (error);
+	return (zil_claim_log_block(zilog, &lr->lr_blkptr, tx, first_txg));
 }
 /* ARGSUSED */
-static void
+static int
-zil_claim_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t first_txg)
-{
-	spa_t *spa = zilog->zl_spa;
-	int err;
-	/*
-	 * Claim log block if not already committed and not already claimed.
-	 */
-	if (bp->blk_birth >= first_txg &&
-	    zil_dva_tree_add(&zilog->zl_dva_tree, BP_IDENTITY(bp)) == 0) {
-		err = zio_wait(zio_claim(NULL, spa, first_txg, bp, NULL, NULL,
-		    ZIO_FLAG_MUSTSUCCEED));
-		ASSERT(err == 0);
-	}
-}
-static void
-zil_claim_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t first_txg)
-{
-	if (lrc->lrc_txtype == TX_WRITE) {
-		lr_write_t *lr = (lr_write_t *)lrc;
-		zil_claim_log_block(zilog, &lr->lr_blkptr, tx, first_txg);
-	}
-}
-/* ARGSUSED */
-static void
 zil_free_log_block(zilog_t *zilog, blkptr_t *bp, void *tx, uint64_t claim_txg)
 {
-	zio_free_blk(zilog->zl_spa, bp, dmu_tx_get_txg(tx));
+	zio_free_zil(zilog->zl_spa, dmu_tx_get_txg(tx), bp);
-}
+	return (0);
-static void
+}
+static int
 zil_free_log_record(zilog_t *zilog, lr_t *lrc, void *tx, uint64_t claim_txg)
 {
+	lr_write_t *lr = (lr_write_t *)lrc;
+	blkptr_t *bp = &lr->lr_blkptr;
 	/*
 	 * If we previously claimed it, we need to free it.
 	 */
-	if (claim_txg != 0 && lrc->lrc_txtype == TX_WRITE) {
+	if (claim_txg != 0 && lrc->lrc_txtype == TX_WRITE &&
-		lr_write_t *lr = (lr_write_t *)lrc;
+	    bp->blk_birth >= claim_txg && zil_bp_tree_add(zilog, bp) == 0)
-		blkptr_t *bp = &lr->lr_blkptr;
+		zio_free(zilog->zl_spa, dmu_tx_get_txg(tx), bp);
-		if (bp->blk_birth >= claim_txg &&
-		    !zil_dva_tree_add(&zilog->zl_dva_tree, BP_IDENTITY(bp))) {
+	return (0);
-			(void) arc_free(NULL, zilog->zl_spa,
-			    dmu_tx_get_txg(tx), bp, NULL, NULL, ARC_WAIT);
-		}
-	}
 }
 /*
 * Create an on-disk intent log.
 */
 	 * If we don't already have an initial log block or we have one
 	 * but it's the wrong endianness then allocate one.
 	 */
 	if (BP_IS_HOLE(&blk) || BP_SHOULD_BYTESWAP(&blk)) {
 		tx = dmu_tx_create(zilog->zl_os);
-		(void) dmu_tx_assign(tx, TXG_WAIT);
+		VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0);
 		dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
 		txg = dmu_tx_get_txg(tx);
 		if (!BP_IS_HOLE(&blk)) {
-			zio_free_blk(zilog->zl_spa, &blk, txg);
+			zio_free_zil(zilog->zl_spa, txg, &blk);
 			BP_ZERO(&blk);
 		}
-		error = zio_alloc_blk(zilog->zl_spa, ZIL_MIN_BLKSZ, &blk,
+		error = zio_alloc_zil(zilog->zl_spa, txg, &blk, NULL,
-		    NULL, txg, zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
+		    ZIL_MIN_BLKSZ, zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
 		if (error == 0)
 			zil_init_log_chain(zilog, &blk);
 	}
 		lwb->lwb_nused = 0;
 		lwb->lwb_sz = BP_GET_LSIZE(&lwb->lwb_blk);
 		lwb->lwb_buf = zio_buf_alloc(lwb->lwb_sz);
 		lwb->lwb_max_txg = txg;
 		lwb->lwb_zio = NULL;
+		lwb->lwb_tx = NULL;
 		mutex_enter(&zilog->zl_lock);
 		list_insert_tail(&zilog->zl_lwb_list, lwb);
 		mutex_exit(&zilog->zl_lock);
 	}
 	/*
 	 * Wait for any previous destroy to complete.
 	 */
 	txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
+	zilog->zl_old_header = *zh;		/* debugging aid */
 	if (BP_IS_HOLE(&zh->zh_log))
 		return;
 	tx = dmu_tx_create(zilog->zl_os);
-	(void) dmu_tx_assign(tx, TXG_WAIT);
+	VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0);
 	dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
 	txg = dmu_tx_get_txg(tx);
 	mutex_enter(&zilog->zl_lock);
 	ASSERT3U(zilog->zl_destroy_txg, <, txg);
 	zilog->zl_destroy_txg = txg;
+	zilog->zl_keep_first = keep_first;
 	if (!list_is_empty(&zilog->zl_lwb_list)) {
 		ASSERT(zh->zh_claim_txg == 0);
-		zilog->zl_keep_first = B_FALSE;
+		ASSERT(!keep_first);
 		while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
 			list_remove(&zilog->zl_lwb_list, lwb);
 			if (lwb->lwb_buf != NULL)
 				zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
-			zio_free_blk(zilog->zl_spa, &lwb->lwb_blk, txg);
+			zio_free_zil(zilog->zl_spa, txg, &lwb->lwb_blk);
 			kmem_cache_free(zil_lwb_cache, lwb);
 		}
-	} else {
+	} else if (!keep_first) {
-		zilog->zl_keep_first = keep_first;
+		(void) zil_parse(zilog, zil_free_log_block,
-		if (zh->zh_flags & ZIL_REPLAY_NEEDED) {
+		    zil_free_log_record, tx, zh->zh_claim_txg);
-			ASSERT(!keep_first);
-			(void) zil_parse(zilog, zil_free_log_block,
-			    zil_free_log_record, tx, zh->zh_claim_txg);
-		} else {
-			/*
-			 * Would like to assert zil_empty() but that
-			 * would force us to read the log chain which
-			 * requires us to do I/O to the log. This is
-			 * overkill since we really just want to destroy
-			 * the chain anyway.
-			 */
-			if (!keep_first) {
-				blkptr_t bp = zh->zh_log;
-				zio_free_blk(zilog->zl_spa, &bp, txg);
-			}
-		}
 	}
 	mutex_exit(&zilog->zl_lock);
 	dmu_tx_commit(tx);
-}
-/*
-* return true if the initial log block is not valid
-*/
-static boolean_t
-zil_empty(zilog_t *zilog)
-{
-	const zil_header_t *zh = zilog->zl_header;
-	arc_buf_t *abuf = NULL;
-	if (BP_IS_HOLE(&zh->zh_log))
-		return (B_TRUE);
-	if (zil_read_log_block(zilog, &zh->zh_log, &abuf) != 0)
-		return (B_TRUE);
-	VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1);
-	return (B_FALSE);
 }
 int
 zil_claim(char *osname, void *txarg)
 {
 	}
 	zilog = dmu_objset_zil(os);
 	zh = zil_header_in_syncing_context(zilog);
-	if (zilog->zl_spa->spa_log_state == SPA_LOG_CLEAR) {
+	if (spa_get_log_state(zilog->zl_spa) == SPA_LOG_CLEAR) {
 		if (!BP_IS_HOLE(&zh->zh_log))
-			zio_free_blk(zilog->zl_spa, &zh->zh_log, first_txg);
+			zio_free_zil(zilog->zl_spa, first_txg, &zh->zh_log);
 		BP_ZERO(&zh->zh_log);
 		dsl_dataset_dirty(dmu_objset_ds(os), tx);
 		dmu_objset_rele(os, FTAG);
 		return (0);
-	}
-	/*
-	 * Record here whether the zil has any records to replay.
-	 * If the header block pointer is null or the block points
-	 * to the stubby then we know there are no valid log records.
-	 * We use the header to store this state as the the zilog gets
-	 * freed later in dmu_objset_close().
-	 * The flags (and the rest of the header fields) are cleared in
-	 * zil_sync() as a result of a zil_destroy(), after replaying the log.
-	 *
-	 * Note, the intent log can be empty but still need the
-	 * stubby to be claimed.
-	 */
-	if (!zil_empty(zilog)) {
-		zh->zh_flags |= ZIL_REPLAY_NEEDED;
-		dsl_dataset_dirty(dmu_objset_ds(os), tx);
 	}
 	/*
 	 * Claim all log blocks if we haven't already done so, and remember
 	 * the highest claimed sequence number.  This ensures that if we can
 	 * but we can read the entire log later, we will not try to replay
 	 * or destroy beyond the last block we successfully claimed.
 	 */
 	ASSERT3U(zh->zh_claim_txg, <=, first_txg);
 	if (zh->zh_claim_txg == 0 && !BP_IS_HOLE(&zh->zh_log)) {
+		(void) zil_parse(zilog, zil_claim_log_block,
+		    zil_claim_log_record, tx, first_txg);
 		zh->zh_claim_txg = first_txg;
-		zh->zh_claim_seq = zil_parse(zilog, zil_claim_log_block,
+		zh->zh_claim_blk_seq = zilog->zl_parse_blk_seq;
-		    zil_claim_log_record, tx, first_txg);
+		zh->zh_claim_lr_seq = zilog->zl_parse_lr_seq;
+		if (zilog->zl_parse_lr_count || zilog->zl_parse_blk_count > 1)
+			zh->zh_flags |= ZIL_REPLAY_NEEDED;
+		zh->zh_flags |= ZIL_CLAIM_LR_SEQ_VALID;
 		dsl_dataset_dirty(dmu_objset_ds(os), tx);
 	}
 	ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1));
 	dmu_objset_rele(os, FTAG);
 /*
 * Check the log by walking the log chain.
 * Checksum errors are ok as they indicate the end of the chain.
 * Any other error (no device or read failure) returns an error.
 */
-/* ARGSUSED */
 int
-zil_check_log_chain(char *osname, void *txarg)
+zil_check_log_chain(char *osname, void *tx)
 {
 	zilog_t *zilog;
-	zil_header_t *zh;
-	blkptr_t blk;
-	arc_buf_t *abuf;
 	objset_t *os;
-	char *lrbuf;
-	zil_trailer_t *ztp;
 	int error;
+	ASSERT(tx == NULL);
 	error = dmu_objset_hold(osname, FTAG, &os);
 	if (error) {
 		cmn_err(CE_WARN, "can't open objset for %s", osname);
 		return (0);
 	}
 	zilog = dmu_objset_zil(os);
-	zh = zil_header_in_syncing_context(zilog);
-	blk = zh->zh_log;
+	/*
-	if (BP_IS_HOLE(&blk)) {
+	 * Because tx == NULL, zil_claim_log_block() will not actually claim
-		dmu_objset_rele(os, FTAG);
+	 * any blocks, but just determine whether it is possible to do so.
-		return (0); /* no chain */
+	 * In addition to checking the log chain, zil_claim_log_block()
-	}
+	 * will invoke zio_claim() with a done func of spa_claim_notify(),
+	 * which will update spa_max_claim_txg.  See spa_load() for details.
-	for (;;) {
+	 */
-		error = zil_read_log_block(zilog, &blk, &abuf);
+	error = zil_parse(zilog, zil_claim_log_block, zil_claim_log_record, tx,
-		if (error)
+	    zilog->zl_header->zh_claim_txg ? -1ULL : spa_first_txg(os->os_spa));
-			break;
-		lrbuf = abuf->b_data;
-		ztp = (zil_trailer_t *)(lrbuf + BP_GET_LSIZE(&blk)) - 1;
-		blk = ztp->zit_next_blk;
-		VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1);
-	}
 	dmu_objset_rele(os, FTAG);
-	if (error == ECKSUM)
-		return (0); /* normal end of chain */
+	return ((error == ECKSUM || error == ENOENT) ? 0 : error);
-	return (error);
 }
 static int
 zil_vdev_compare(const void *x1, const void *x2)
 {
 	return (0);
 }
 void
-zil_add_block(zilog_t *zilog, blkptr_t *bp)
+zil_add_block(zilog_t *zilog, const blkptr_t *bp)
 {
 	avl_tree_t *t = &zilog->zl_vdev_tree;
 	avl_index_t where;
 	zil_vdev_node_t *zv, zvsearch;
 	int ndvas = BP_GET_NDVAS(bp);
 static void
 zil_lwb_write_done(zio_t *zio)
 {
 	lwb_t *lwb = zio->io_private;
 	zilog_t *zilog = lwb->lwb_zilog;
+	dmu_tx_t *tx = lwb->lwb_tx;
 	ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF);
 	ASSERT(BP_GET_CHECKSUM(zio->io_bp) == ZIO_CHECKSUM_ZILOG);
 	ASSERT(BP_GET_TYPE(zio->io_bp) == DMU_OT_INTENT_LOG);
 	ASSERT(BP_GET_LEVEL(zio->io_bp) == 0);
 	 * the lwb if lwb_buf is null.
 	 */
 	zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
 	mutex_enter(&zilog->zl_lock);
 	lwb->lwb_buf = NULL;
-	if (zio->io_error)
+	lwb->lwb_tx = NULL;
-		zilog->zl_log_error = B_TRUE;
+	mutex_exit(&zilog->zl_lock);
 	/*
 	 * Now that we've written this log block, we have a stable pointer
 	 * to the next block in the chain, so it's OK to let the txg in
-	 * which we allocated the next block sync. We still have the
+	 * which we allocated the next block sync.
-	 * zl_lock to ensure zil_sync doesn't kmem free the lwb.
+	 */
-	 */
+	dmu_tx_commit(tx);
-	txg_rele_to_sync(&lwb->lwb_txgh);
-	mutex_exit(&zilog->zl_lock);
 }
 /*
 * Initialize the io for a log block.
 */
 static void
 zil_lwb_write_init(zilog_t *zilog, lwb_t *lwb)
 {
 	zbookmark_t zb;
-	zb.zb_objset = lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_OBJSET];
+	SET_BOOKMARK(&zb, lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_OBJSET],
-	zb.zb_object = 0;
+	    ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
-	zb.zb_level = -1;
+	    lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_SEQ]);
-	zb.zb_blkid = lwb->lwb_blk.blk_cksum.zc_word[ZIL_ZC_SEQ];
 	if (zilog->zl_root_zio == NULL) {
 		zilog->zl_root_zio = zio_root(zilog->zl_spa, NULL, NULL,
 		    ZIO_FLAG_CANFAIL);
 	}
 {
 	lwb_t *nlwb;
 	zil_trailer_t *ztp = (zil_trailer_t *)(lwb->lwb_buf + lwb->lwb_sz) - 1;
 	spa_t *spa = zilog->zl_spa;
 	blkptr_t *bp = &ztp->zit_next_blk;
+	dmu_tx_t *tx;
 	uint64_t txg;
 	uint64_t zil_blksz;
 	int error;
 	ASSERT(lwb->lwb_nused <= ZIL_BLK_DATA_SZ(lwb));
 	/*
 	 * Allocate the next block and save its address in this block
 	 * before writing it in order to establish the log chain.
 	 * Note that if the allocation of nlwb synced before we wrote
 	 * the block that points at it (lwb), we'd leak it if we crashed.
-	 * Therefore, we don't do txg_rele_to_sync() until zil_lwb_write_done().
+	 * Therefore, we don't do dmu_tx_commit() until zil_lwb_write_done().
-	 */
+	 * We dirty the dataset to ensure that zil_sync() will be called
-	txg = txg_hold_open(zilog->zl_dmu_pool, &lwb->lwb_txgh);
+	 * to clean up in the event of allocation failure or I/O failure.
-	txg_rele_to_quiesce(&lwb->lwb_txgh);
+	 */
+	tx = dmu_tx_create(zilog->zl_os);
+	VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0);
+	dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
+	txg = dmu_tx_get_txg(tx);
+	lwb->lwb_tx = tx;
 	/*
 	 * Pick a ZIL blocksize. We request a size that is the
 	 * maximum of the previous used size, the current used size and
 	 * the amount waiting in the queue.
 	if (zil_blksz > ZIL_MAX_BLKSZ)
 		zil_blksz = ZIL_MAX_BLKSZ;
 	BP_ZERO(bp);
 	/* pass the old blkptr in order to spread log blocks across devs */
-	error = zio_alloc_blk(spa, zil_blksz, bp, &lwb->lwb_blk, txg,
+	error = zio_alloc_zil(spa, txg, bp, &lwb->lwb_blk, zil_blksz,
 	    USE_SLOG(zilog));
 	if (error) {
-		dmu_tx_t *tx = dmu_tx_create_assigned(zilog->zl_dmu_pool, txg);
 		/*
-		 * We dirty the dataset to ensure that zil_sync() will
+		 * Since we've just experienced an allocation failure,
-		 * be called to remove this lwb from our zl_lwb_list.
-		 * Failing to do so, may leave an lwb with a NULL lwb_buf
-		 * hanging around on the zl_lwb_list.
-		 */
-		dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
-		dmu_tx_commit(tx);
-		/*
-		 * Since we've just experienced an allocation failure so we
 		 * terminate the current lwb and send it on its way.
 		 */
 		ztp->zit_pad = 0;
 		ztp->zit_nused = lwb->lwb_nused;
 		ztp->zit_bt.zbt_cksum = lwb->lwb_blk.blk_cksum;
 	/*
 	 * Allocate a new log write buffer (lwb).
 	 */
 	nlwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP);
 	nlwb->lwb_zilog = zilog;
 	nlwb->lwb_blk = *bp;
 	nlwb->lwb_nused = 0;
 	nlwb->lwb_sz = BP_GET_LSIZE(&nlwb->lwb_blk);
 	nlwb->lwb_buf = zio_buf_alloc(nlwb->lwb_sz);
 	nlwb->lwb_max_txg = txg;
 	nlwb->lwb_zio = NULL;
+	nlwb->lwb_tx = NULL;
 	/*
 	 * Put new lwb at the end of the log chain
 	 */
 	mutex_enter(&zilog->zl_lock);
 	zil_add_block(zilog, &lwb->lwb_blk);
 	/*
 	 * kick off the write for the old log block
 	 */
-	dprintf_bp(&lwb->lwb_blk, "lwb %p txg %llu: ", lwb, txg);
 	ASSERT(lwb->lwb_zio);
 	zio_nowait(lwb->lwb_zio);
 	return (nlwb);
 }
 static lwb_t *
 zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
 {
 	lr_t *lrc = &itx->itx_lr; /* common log record */
-	lr_write_t *lr = (lr_write_t *)lrc;
+	lr_write_t *lrw = (lr_write_t *)lrc;
+	char *lr_buf;
 	uint64_t txg = lrc->lrc_txg;
 	uint64_t reclen = lrc->lrc_reclen;
-	uint64_t dlen;
+	uint64_t dlen = 0;
 	if (lwb == NULL)
 		return (NULL);
 	ASSERT(lwb->lwb_buf != NULL);
 	if (lrc->lrc_txtype == TX_WRITE && itx->itx_wr_state == WR_NEED_COPY)
 		dlen = P2ROUNDUP_TYPED(
-		    lr->lr_length, sizeof (uint64_t), uint64_t);
+		    lrw->lr_length, sizeof (uint64_t), uint64_t);
-	else
-		dlen = 0;
 	zilog->zl_cur_used += (reclen + dlen);
 	zil_lwb_write_init(zilog, lwb);
 			txg_wait_synced(zilog->zl_dmu_pool, txg);
 			return (lwb);
 		}
 	}
-	/*
+	lr_buf = lwb->lwb_buf + lwb->lwb_nused;
-	 * Update the lrc_seq, to be log record sequence number. See zil.h
+	bcopy(lrc, lr_buf, reclen);
-	 * Then copy the record to the log buffer.
+	lrc = (lr_t *)lr_buf;
-	 */
+	lrw = (lr_write_t *)lrc;
-	lrc->lrc_seq = ++zilog->zl_lr_seq; /* we are single threaded */
-	bcopy(lrc, lwb->lwb_buf + lwb->lwb_nused, reclen);
 	/*
 	 * If it's a write, fetch the data or get its blkptr as appropriate.
 	 */
 	if (lrc->lrc_txtype == TX_WRITE) {
 			txg_wait_synced(zilog->zl_dmu_pool, txg);
 		if (itx->itx_wr_state != WR_COPIED) {
 			char *dbuf;
 			int error;
-			/* alignment is guaranteed */
-			lr = (lr_write_t *)(lwb->lwb_buf + lwb->lwb_nused);
 			if (dlen) {
 				ASSERT(itx->itx_wr_state == WR_NEED_COPY);
-				dbuf = lwb->lwb_buf + lwb->lwb_nused + reclen;
+				dbuf = lr_buf + reclen;
-				lr->lr_common.lrc_reclen += dlen;
+				lrw->lr_common.lrc_reclen += dlen;
 			} else {
 				ASSERT(itx->itx_wr_state == WR_INDIRECT);
 				dbuf = NULL;
 			}
 			error = zilog->zl_get_data(
-			    itx->itx_private, lr, dbuf, lwb->lwb_zio);
+			    itx->itx_private, lrw, dbuf, lwb->lwb_zio);
 			if (error == EIO) {
 				txg_wait_synced(zilog->zl_dmu_pool, txg);
 				return (lwb);
 			}
 			if (error) {
 				return (lwb);
 			}
 		}
 	}
+	/*
+	 * We're actually making an entry, so update lrc_seq to be the
+	 * log record sequence number.  Note that this is generally not
+	 * equal to the itx sequence number because not all transactions
+	 * are synchronous, and sometimes spa_sync() gets there first.
+	 */
+	lrc->lrc_seq = ++zilog->zl_lr_seq; /* we are single threaded */
 	lwb->lwb_nused += reclen + dlen;
 	lwb->lwb_max_txg = MAX(lwb->lwb_max_txg, txg);
 	ASSERT3U(lwb->lwb_nused, <=, ZIL_BLK_DATA_SZ(lwb));
 	ASSERT3U(P2PHASE(lwb->lwb_nused, sizeof (uint64_t)), ==, 0);
 	itx->itx_lr.lrc_seq = 0;	/* defensive */
 	return (itx);
 }
+void
+zil_itx_destroy(itx_t *itx)
+{
+	kmem_free(itx, offsetof(itx_t, itx_lr) + itx->itx_lr.lrc_reclen);
+}
 uint64_t
 zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx)
 {
 	uint64_t seq;
 	ASSERT(itx->itx_lr.lrc_seq == 0);
+	ASSERT(!zilog->zl_replay);
 	mutex_enter(&zilog->zl_lock);
 	list_insert_tail(&zilog->zl_itx_list, itx);
 	zilog->zl_itx_list_sz += itx->itx_sod;
 	itx->itx_lr.lrc_txg = dmu_tx_get_txg(tx);
 	mutex_exit(&zilog->zl_lock);
 	/* destroy sync'd log transactions */
 	while ((itx = list_head(&clean_list)) != NULL) {
 		list_remove(&clean_list, itx);
-		kmem_free(itx, offsetof(itx_t, itx_lr)
+		zil_itx_destroy(itx);
-		    + itx->itx_lr.lrc_reclen);
 	}
 	list_destroy(&clean_list);
 }
 /*
 static void
 zil_commit_writer(zilog_t *zilog, uint64_t seq, uint64_t foid)
 {
 	uint64_t txg;
 	uint64_t commit_seq = 0;
-	itx_t *itx, *itx_next = (itx_t *)-1;
+	itx_t *itx, *itx_next;
 	lwb_t *lwb;
 	spa_t *spa;
+	int error = 0;
 	zilog->zl_writer = B_TRUE;
 	ASSERT(zilog->zl_root_zio == NULL);
 	spa = zilog->zl_spa;
 		}
 	}
 	/* Loop through in-memory log transactions filling log blocks. */
 	DTRACE_PROBE1(zil__cw1, zilog_t *, zilog);
-	for (;;) {
+	for (itx = list_head(&zilog->zl_itx_list); itx; itx = itx_next) {
 		/*
-		 * Find the next itx to push:
+		 * Save the next pointer.  Even though we drop zl_lock below,
-		 * Push all transactions related to specified foid and all
+		 * all threads that can remove itx list entries (other writers
-		 * other transactions except TX_WRITE, TX_TRUNCATE,
+		 * and zil_itx_clean()) can't do so until they have zl_writer.
-		 * TX_SETATTR and TX_ACL for all other files.
 		 */
-		if (itx_next != (itx_t *)-1)
+		itx_next = list_next(&zilog->zl_itx_list, itx);
-			itx = itx_next;
-		else
+		/*
-			itx = list_head(&zilog->zl_itx_list);
+		 * Determine whether to push this itx.
-		for (; itx != NULL; itx = list_next(&zilog->zl_itx_list, itx)) {
+		 * Push all transactions related to specified foid and
-			if (foid == 0) /* push all foids? */
+		 * all other transactions except those that can be logged
-				break;
+		 * out of order (TX_WRITE, TX_TRUNCATE, TX_SETATTR, TX_ACL)
-			if (itx->itx_sync) /* push all O_[D]SYNC */
+		 * for all other files.
-				break;
+		 *
-			switch (itx->itx_lr.lrc_txtype) {
+		 * If foid == 0 (meaning "push all foids") or
-			case TX_SETATTR:
+		 * itx->itx_sync is set (meaning O_[D]SYNC), push regardless.
-			case TX_WRITE:
+		 */
-			case TX_TRUNCATE:
+		if (foid != 0 && !itx->itx_sync &&
-			case TX_ACL:
+		    TX_OOO(itx->itx_lr.lrc_txtype) &&
-				/* lr_foid is same offset for these records */
+		    ((lr_ooo_t *)&itx->itx_lr)->lr_foid != foid)
-				if (((lr_write_t *)&itx->itx_lr)->lr_foid
+			continue; /* skip this record */
-				    != foid) {
-					continue; /* skip this record */
-				}
-			}
-			break;
-		}
-		if (itx == NULL)
-			break;
 		if ((itx->itx_lr.lrc_seq > seq) &&
 		    ((lwb == NULL) || (lwb->lwb_nused == 0) ||
-		    (lwb->lwb_nused + itx->itx_sod > ZIL_BLK_DATA_SZ(lwb)))) {
+		    (lwb->lwb_nused + itx->itx_sod > ZIL_BLK_DATA_SZ(lwb))))
 			break;
-		}
-		/*
-		 * Save the next pointer.  Even though we soon drop
-		 * zl_lock all threads that may change the list
-		 * (another writer or zil_itx_clean) can't do so until
-		 * they have zl_writer.
-		 */
-		itx_next = list_next(&zilog->zl_itx_list, itx);
 		list_remove(&zilog->zl_itx_list, itx);
 		zilog->zl_itx_list_sz -= itx->itx_sod;
 		mutex_exit(&zilog->zl_lock);
 		txg = itx->itx_lr.lrc_txg;
 		ASSERT(txg);
 		if (txg > spa_last_synced_txg(spa) ||
 		    txg > spa_freeze_txg(spa))
 			lwb = zil_lwb_commit(zilog, itx, lwb);
-		kmem_free(itx, offsetof(itx_t, itx_lr)
-		    + itx->itx_lr.lrc_reclen);
+		zil_itx_destroy(itx);
 		mutex_enter(&zilog->zl_lock);
 	}
 	DTRACE_PROBE1(zil__cw2, zilog_t *, zilog);
 	/* determine commit sequence number */
 	itx = list_head(&zilog->zl_itx_list);
 	if (itx)
-		commit_seq = itx->itx_lr.lrc_seq;
+		commit_seq = itx->itx_lr.lrc_seq - 1;
 	else
 		commit_seq = zilog->zl_itx_seq;
 	mutex_exit(&zilog->zl_lock);
 	/* write the last block out */
 	/*
 	 * Wait if necessary for the log blocks to be on stable storage.
 	 */
 	if (zilog->zl_root_zio) {
 		DTRACE_PROBE1(zil__cw3, zilog_t *, zilog);
-		(void) zio_wait(zilog->zl_root_zio);
+		error = zio_wait(zilog->zl_root_zio);
 		zilog->zl_root_zio = NULL;
 		DTRACE_PROBE1(zil__cw4, zilog_t *, zilog);
 		zil_flush_vdevs(zilog);
 	}
-	if (zilog->zl_log_error || lwb == NULL) {
+	if (error || lwb == NULL)
-		zilog->zl_log_error = 0;
 		txg_wait_synced(zilog->zl_dmu_pool, 0);
-	}
 	mutex_enter(&zilog->zl_lock);
 	zilog->zl_writer = B_FALSE;
 	ASSERT3U(commit_seq, >=, zilog->zl_commit_seq);
 	zilog->zl_commit_seq = commit_seq;
+	/*
+	 * Remember the highest committed log sequence number for ztest.
+	 * We only update this value when all the log writes succeeded,
+	 * because ztest wants to ASSERT that it got the whole log chain.
+	 */
+	if (error == 0 && lwb != NULL)
+		zilog->zl_commit_lr_seq = zilog->zl_lr_seq;
 }
 /*
 * Push zfs transactions to stable storage up to the supplied sequence number.
 * If foid is 0 push out all transactions, otherwise push only those
 	seq = MIN(seq, zilog->zl_itx_seq);	/* cap seq at largest itx seq */
 	while (zilog->zl_writer) {
 		cv_wait(&zilog->zl_cv_writer, &zilog->zl_lock);
-		if (seq < zilog->zl_commit_seq) {
+		if (seq <= zilog->zl_commit_seq) {
 			mutex_exit(&zilog->zl_lock);
 			return;
 		}
 	}
 	zil_commit_writer(zilog, seq, foid); /* drops zl_lock */
 	cv_broadcast(&zilog->zl_cv_writer);
 	mutex_exit(&zilog->zl_lock);
 }
 /*
+* Report whether all transactions are committed.
+*/
+static boolean_t
+zil_is_committed(zilog_t *zilog)
+{
+	lwb_t *lwb;
+	boolean_t committed;
+	mutex_enter(&zilog->zl_lock);
+	while (zilog->zl_writer)
+		cv_wait(&zilog->zl_cv_writer, &zilog->zl_lock);
+	if (!list_is_empty(&zilog->zl_itx_list))
+		committed = B_FALSE;		/* unpushed transactions */
+	else if ((lwb = list_head(&zilog->zl_lwb_list)) == NULL)
+		committed = B_TRUE;		/* intent log never used */
+	else if (list_next(&zilog->zl_lwb_list, lwb) != NULL)
+		committed = B_FALSE;		/* zil_sync() not done yet */
+	else
+		committed = B_TRUE;		/* everything synced */
+	mutex_exit(&zilog->zl_lock);
+	return (committed);
+}
+/*
 * Called in syncing context to free committed log blocks and update log header.
 */
 void
 zil_sync(zilog_t *zilog, dmu_tx_t *tx)
 {
 	zil_header_t *zh = zil_header_in_syncing_context(zilog);
 	uint64_t txg = dmu_tx_get_txg(tx);
 	spa_t *spa = zilog->zl_spa;
+	uint64_t *replayed_seq = &zilog->zl_replayed_seq[txg & TXG_MASK];
 	lwb_t *lwb;
 	/*
 	 * We don't zero out zl_destroy_txg, so make sure we don't try
 	 * to destroy it twice.
 	mutex_enter(&zilog->zl_lock);
 	ASSERT(zilog->zl_stop_sync == 0);
-	zh->zh_replay_seq = zilog->zl_replayed_seq[txg & TXG_MASK];
+	if (*replayed_seq != 0) {
+		ASSERT(zh->zh_replay_seq < *replayed_seq);
+		zh->zh_replay_seq = *replayed_seq;
+		*replayed_seq = 0;
+	}
 	if (zilog->zl_destroy_txg == txg) {
 		blkptr_t blk = zh->zh_log;
 		ASSERT(list_head(&zilog->zl_lwb_list) == NULL);
 	while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
 		zh->zh_log = lwb->lwb_blk;
 		if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg)
 			break;
 		list_remove(&zilog->zl_lwb_list, lwb);
-		zio_free_blk(spa, &lwb->lwb_blk, txg);
+		zio_free_zil(spa, txg, &lwb->lwb_blk);
 		kmem_cache_free(zil_lwb_cache, lwb);
 		/*
 		 * If we don't have anything left in the lwb list then
 		 * we've had an allocation failure and we need to zero
 	 * (so zil_sync() will be called) and wait for that txg to sync.
 	 */
 	if (!zil_is_committed(zilog)) {
 		uint64_t txg;
 		dmu_tx_t *tx = dmu_tx_create(zilog->zl_os);
-		(void) dmu_tx_assign(tx, TXG_WAIT);
+		VERIFY(dmu_tx_assign(tx, TXG_WAIT) == 0);
 		dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
 		txg = dmu_tx_get_txg(tx);
 		dmu_tx_commit(tx);
 		txg_wait_synced(zilog->zl_dmu_pool, txg);
 	}
 	ASSERT(zilog->zl_suspend != 0);
 	zilog->zl_suspend--;
 	mutex_exit(&zilog->zl_lock);
 }
-/*
-* Read in the data for the dmu_sync()ed block, and change the log
-* record to write this whole block.
-*/
-void
-zil_get_replay_data(zilog_t *zilog, lr_write_t *lr)
-{
-	blkptr_t *wbp = &lr->lr_blkptr;
-	char *wbuf = (char *)(lr + 1); /* data follows lr_write_t */
-	uint64_t blksz;
-	if (BP_IS_HOLE(wbp)) {	/* compressed to a hole */
-		blksz = BP_GET_LSIZE(&lr->lr_blkptr);
-		/*
-		 * If the blksz is zero then we must be replaying a log
-		 * from an version prior to setting the blksize of null blocks.
-		 * So we just zero the actual write size reqeusted.
-		 */
-		if (blksz == 0) {
-			bzero(wbuf, lr->lr_length);
-			return;
-		}
-		bzero(wbuf, blksz);
-	} else {
-		/*
-		 * A subsequent write may have overwritten this block, in which
-		 * case wbp may have been been freed and reallocated, and our
-		 * read of wbp may fail with a checksum error.  We can safely
-		 * ignore this because the later write will provide the
-		 * correct data.
-		 */
-		zbookmark_t zb;
-		zb.zb_objset = dmu_objset_id(zilog->zl_os);
-		zb.zb_object = lr->lr_foid;
-		zb.zb_level = 0;
-		zb.zb_blkid = -1; /* unknown */
-		blksz = BP_GET_LSIZE(&lr->lr_blkptr);
-		(void) zio_wait(zio_read(NULL, zilog->zl_spa, wbp, wbuf, blksz,
-		    NULL, NULL, ZIO_PRIORITY_SYNC_READ,
-		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &zb));
-	}
-	lr->lr_offset -= lr->lr_offset % blksz;
-	lr->lr_length = blksz;
-}
 typedef struct zil_replay_arg {
-	objset_t	*zr_os;
 	zil_replay_func_t **zr_replay;
 	void		*zr_arg;
 	boolean_t	zr_byteswap;
-	char		*zr_lrbuf;
+	char		*zr_lr;
 } zil_replay_arg_t;
-static void
+static int
+zil_replay_error(zilog_t *zilog, lr_t *lr, int error)
+{
+	char name[MAXNAMELEN];
+	zilog->zl_replaying_seq--;	/* didn't actually replay this one */
+	dmu_objset_name(zilog->zl_os, name);
+	cmn_err(CE_WARN, "ZFS replay transaction error %d, "
+	    "dataset %s, seq 0x%llx, txtype %llu %s\n", error, name,
+	    (u_longlong_t)lr->lrc_seq,
+	    (u_longlong_t)(lr->lrc_txtype & ~TX_CI),
+	    (lr->lrc_txtype & TX_CI) ? "CI" : "");
+	return (error);
+}
+static int
 zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg)
 {
 	zil_replay_arg_t *zr = zra;
 	const zil_header_t *zh = zilog->zl_header;
 	uint64_t reclen = lr->lrc_reclen;
 	uint64_t txtype = lr->lrc_txtype;
-	char *name;
+	int error = 0;
-	int pass, error;
+	zilog->zl_replaying_seq = lr->lrc_seq;
-	if (!zilog->zl_replay)			/* giving up */
-		return;
+	if (lr->lrc_seq <= zh->zh_replay_seq)	/* already replayed */
+		return (0);
 	if (lr->lrc_txg < claim_txg)		/* already committed */
-		return;
+		return (0);
-	if (lr->lrc_seq <= zh->zh_replay_seq)	/* already replayed */
-		return;
 	/* Strip case-insensitive bit, still present in log record */
 	txtype &= ~TX_CI;
-	if (txtype == 0 || txtype >= TX_MAX_TYPE) {
+	if (txtype == 0 || txtype >= TX_MAX_TYPE)
-		error = EINVAL;
+		return (zil_replay_error(zilog, lr, EINVAL));
-		goto bad;
+	/*
+	 * If this record type can be logged out of order, the object
+	 * (lr_foid) may no longer exist.  That's legitimate, not an error.
+	 */
+	if (TX_OOO(txtype)) {
+		error = dmu_object_info(zilog->zl_os,
+		    ((lr_ooo_t *)lr)->lr_foid, NULL);
+		if (error == ENOENT || error == EEXIST)
+			return (0);
 	}
 	/*
 	 * Make a copy of the data so we can revise and extend it.
 	 */
-	bcopy(lr, zr->zr_lrbuf, reclen);
+	bcopy(lr, zr->zr_lr, reclen);
+	/*
+	 * If this is a TX_WRITE with a blkptr, suck in the data.
+	 */
+	if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) {
+		error = zil_read_log_data(zilog, (lr_write_t *)lr,
+		    zr->zr_lr + reclen);
+		if (error)
+			return (zil_replay_error(zilog, lr, error));
+	}
 	/*
 	 * The log block containing this lr may have been byteswapped
 	 * so that we can easily examine common fields like lrc_txtype.
-	 * However, the log is a mix of different data types, and only the
+	 * However, the log is a mix of different record types, and only the
 	 * replay vectors know how to byteswap their records.  Therefore, if
 	 * the lr was byteswapped, undo it before invoking the replay vector.
 	 */
 	if (zr->zr_byteswap)
-		byteswap_uint64_array(zr->zr_lrbuf, reclen);
+		byteswap_uint64_array(zr->zr_lr, reclen);
 	/*
 	 * We must now do two things atomically: replay this log record,
 	 * and update the log header sequence number to reflect the fact that
 	 * we did so. At the end of each replay function the sequence number
 	 * is updated if we are in replay mode.
 	 */
-	for (pass = 1; pass <= 2; pass++) {
+	error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, zr->zr_byteswap);
-		zilog->zl_replaying_seq = lr->lrc_seq;
+	if (error) {
-		/* Only byteswap (if needed) on the 1st pass.  */
-		error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lrbuf,
-		    zr->zr_byteswap && pass == 1);
-		if (!error)
-			return;
 		/*
 		 * The DMU's dnode layer doesn't see removes until the txg
 		 * commits, so a subsequent claim can spuriously fail with
 		 * EEXIST. So if we receive any error we try syncing out
-		 * any removes then retry the transaction.
+		 * any removes then retry the transaction.  Note that we
+		 * specify B_FALSE for byteswap now, so we don't do it twice.
 		 */
-		if (pass == 1)
+		txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0);
-			txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0);
+		error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, B_FALSE);
-	}
+		if (error)
+			return (zil_replay_error(zilog, lr, error));
-bad:
+	}
-	ASSERT(error);
+	return (0);
-	name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
-	dmu_objset_name(zr->zr_os, name);
-	cmn_err(CE_WARN, "ZFS replay transaction error %d, "
-	    "dataset %s, seq 0x%llx, txtype %llu %s\n",
-	    error, name, (u_longlong_t)lr->lrc_seq, (u_longlong_t)txtype,
-	    (lr->lrc_txtype & TX_CI) ? "CI" : "");
-	zilog->zl_replay = B_FALSE;
-	kmem_free(name, MAXNAMELEN);
 }
 /* ARGSUSED */
-static void
+static int
 zil_incr_blks(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
 {
 	zilog->zl_replay_blks++;
+	return (0);
 }
 /*
 * If this dataset has a non-empty intent log, replay it and destroy it.
 */
 	if ((zh->zh_flags & ZIL_REPLAY_NEEDED) == 0) {
 		zil_destroy(zilog, B_TRUE);
 		return;
 	}
-	zr.zr_os = os;
 	zr.zr_replay = replay_func;
 	zr.zr_arg = arg;
 	zr.zr_byteswap = BP_SHOULD_BYTESWAP(&zh->zh_log);
-	zr.zr_lrbuf = kmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP);
+	zr.zr_lr = kmem_alloc(2 * SPA_MAXBLOCKSIZE, KM_SLEEP);
 	/*
 	 * Wait for in-progress removes to sync before starting replay.
 	 */
 	txg_wait_synced(zilog->zl_dmu_pool, 0);
 	zilog->zl_replay = B_TRUE;
 	zilog->zl_replay_time = lbolt;
 	ASSERT(zilog->zl_replay_blks == 0);
 	(void) zil_parse(zilog, zil_incr_blks, zil_replay_log_record, &zr,
 	    zh->zh_claim_txg);
-	kmem_free(zr.zr_lrbuf, 2 * SPA_MAXBLOCKSIZE);
+	kmem_free(zr.zr_lr, 2 * SPA_MAXBLOCKSIZE);
 	zil_destroy(zilog, B_FALSE);
 	txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
 	zilog->zl_replay = B_FALSE;
 }
-/*
+boolean_t
-* Report whether all transactions are committed
+zil_replaying(zilog_t *zilog, dmu_tx_t *tx)
-*/
+{
-int
+	if (zilog == NULL)
-zil_is_committed(zilog_t *zilog)
+		return (B_TRUE);
-{
-	lwb_t *lwb;
+	if (zilog->zl_replay) {
-	int ret;
+		dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
+		zilog->zl_replayed_seq[dmu_tx_get_txg(tx) & TXG_MASK] =
-	mutex_enter(&zilog->zl_lock);
+		    zilog->zl_replaying_seq;
-	while (zilog->zl_writer)
+		return (B_TRUE);
-		cv_wait(&zilog->zl_cv_writer, &zilog->zl_lock);
+	}
-	/* recent unpushed intent log transactions? */
+	return (B_FALSE);
-	if (!list_is_empty(&zilog->zl_itx_list)) {
-		ret = B_FALSE;
-		goto out;
-	}
-	/* intent log never used? */
-	lwb = list_head(&zilog->zl_lwb_list);
-	if (lwb == NULL) {
-		ret = B_TRUE;
-		goto out;
-	}
-	/*
-	 * more than 1 log buffer means zil_sync() hasn't yet freed
-	 * entries after a txg has committed
-	 */
-	if (list_next(&zilog->zl_lwb_list, lwb)) {
-		ret = B_FALSE;
-		goto out;
-	}
-	ASSERT(zil_empty(zilog));
-	ret = B_TRUE;
-out:
-	cv_broadcast(&zilog->zl_cv_writer);
-	mutex_exit(&zilog->zl_lock);
-	return (ret);
 }
 /* ARGSUSED */
 int
 zil_vdev_offline(char *osname, void *arg)

changeset 10922	e2081f502306
parent 10921	8aac17999e4d
child 11066	cebb50cbe4f9