Illumos 4958 zdb trips assert on pools with ashift >= 0xe

4958 zdb trips assert on pools with ashift >= 0xe
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Max Grossman <max.grossman@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Approved by: Garrett D'Amore <garrett@damore.org>

References:
  https://www.illumos.org/issues/4958
  https://github.com/illumos/illumos-gate/commit/2a104a5

Porting notes:

Keep the ZIO_FLAG_FASTWRITE define.  This is for a feature present
in Linux but not yet in *BSD.

Ported by: Turbo Fredriksson <turbo@bayour.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #2697

cmd/ztest/ztest.c

@@ -859,7 +859,7 @@ static uint64_t
 ztest_get_ashift(void)
 {
 	if (ztest_opts.zo_ashift == 0)
-		return (SPA_MINBLOCKSHIFT + ztest_random(3));
+		return (SPA_MINBLOCKSHIFT + ztest_random(5));
 	return (ztest_opts.zo_ashift);
 }
 
@@ -1021,11 +1021,28 @@ ztest_random_spa_version(uint64_t initial_version)
 	return (version);
 }
 
+/*
+ * Find the largest ashift used
+ */
+static uint64_t
+ztest_spa_get_ashift() {
+	uint64_t i;
+	uint64_t ashift = SPA_MINBLOCKSHIFT;
+	vdev_t *rvd = ztest_spa->spa_root_vdev;
+
+	for (i = 0; i < rvd->vdev_children; i++) {
+		ashift = MAX(ashift, rvd->vdev_child[i]->vdev_ashift);
+	}
+	return (ashift);
+}
+
 static int
 ztest_random_blocksize(void)
 {
-	return (1 << (SPA_MINBLOCKSHIFT +
+	// Choose a block size >= the ashift.
-	    ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
+	uint64_t block_shift =
+	    ztest_random(SPA_MAXBLOCKSHIFT - ztest_spa_get_ashift() + 1);
+	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
 }
 
 static int
@@ -5963,17 +5980,31 @@ ztest_freeze(void)
 	 */
 	spa_freeze(spa);
 
+	/*
+	 * Because it is hard to predict how much space a write will actually
+	 * require beforehand, we leave ourselves some fudge space to write over
+	 * capacity.
+	 */
+	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
+
 	/*
 	 * Run tests that generate log records but don't alter the pool config
 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
 	 * We do a txg_wait_synced() after each iteration to force the txg
 	 * to increase well beyond the last synced value in the uberblock.
 	 * The ZIL should be OK with that.
+	 *
+	 * Run a random number of times less than zo_maxloops and ensure we do
+	 * not run out of space on the pool.
 	 */
 	while (ztest_random(10) != 0 &&
-	    numloops++ < ztest_opts.zo_maxloops) {
+	    numloops++ < ztest_opts.zo_maxloops &&
-		ztest_dmu_write_parallel(zd, 0);
+	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
-		ztest_dmu_object_alloc_free(zd, 0);
+		ztest_od_t od;
+		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
+		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
+		ztest_io(zd, od.od_object,
+		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
 		txg_wait_synced(spa_get_dsl(spa), 0);
 	}
 

include/sys/vdev_impl.h

@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
  */
 
 #ifndef _SYS_VDEV_IMPL_H
@@ -239,8 +239,11 @@ struct vdev {
 #define	VDEV_PHYS_SIZE		(112 << 10)
 #define	VDEV_UBERBLOCK_RING	(128 << 10)
 
+/* The largest uberblock we support is 8k. */
+#define	MAX_UBERBLOCK_SHIFT (13)
 #define	VDEV_UBERBLOCK_SHIFT(vd)	\
-	MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT)
+	MIN(MAX((vd)->vdev_top->vdev_ashift, UBERBLOCK_SHIFT), \
+	    MAX_UBERBLOCK_SHIFT)
 #define	VDEV_UBERBLOCK_COUNT(vd)	\
 	(VDEV_UBERBLOCK_RING >> VDEV_UBERBLOCK_SHIFT(vd))
 #define	VDEV_UBERBLOCK_OFFSET(vd, n)	\

include/sys/zio.h

@@ -167,19 +167,20 @@ enum zio_flag {
 	ZIO_FLAG_RESILVER	= 1 << 3,
 	ZIO_FLAG_SCRUB		= 1 << 4,
 	ZIO_FLAG_SCAN_THREAD	= 1 << 5,
+	ZIO_FLAG_PHYSICAL	= 1 << 6,
 
 #define	ZIO_FLAG_AGG_INHERIT	(ZIO_FLAG_CANFAIL - 1)
 
 	/*
 	 * Flags inherited by ddt, gang, and vdev children.
 	 */
-	ZIO_FLAG_CANFAIL	= 1 << 6,	/* must be first for INHERIT */
+	ZIO_FLAG_CANFAIL	= 1 << 7,	/* must be first for INHERIT */
-	ZIO_FLAG_SPECULATIVE	= 1 << 7,
+	ZIO_FLAG_SPECULATIVE	= 1 << 8,
-	ZIO_FLAG_CONFIG_WRITER	= 1 << 8,
+	ZIO_FLAG_CONFIG_WRITER	= 1 << 9,
-	ZIO_FLAG_DONT_RETRY	= 1 << 9,
+	ZIO_FLAG_DONT_RETRY	= 1 << 10,
-	ZIO_FLAG_DONT_CACHE	= 1 << 10,
+	ZIO_FLAG_DONT_CACHE	= 1 << 11,
-	ZIO_FLAG_NODATA		= 1 << 11,
+	ZIO_FLAG_NODATA		= 1 << 12,
-	ZIO_FLAG_INDUCE_DAMAGE	= 1 << 12,
+	ZIO_FLAG_INDUCE_DAMAGE	= 1 << 13,
 
 #define	ZIO_FLAG_DDT_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
 #define	ZIO_FLAG_GANG_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
@@ -187,28 +188,28 @@ enum zio_flag {
 	/*
 	 * Flags inherited by vdev children.
 	 */
-	ZIO_FLAG_IO_RETRY	= 1 << 13,	/* must be first for INHERIT */
+	ZIO_FLAG_IO_RETRY	= 1 << 14,	/* must be first for INHERIT */
-	ZIO_FLAG_PROBE		= 1 << 14,
+	ZIO_FLAG_PROBE		= 1 << 15,
-	ZIO_FLAG_TRYHARD	= 1 << 15,
+	ZIO_FLAG_TRYHARD	= 1 << 16,
-	ZIO_FLAG_OPTIONAL	= 1 << 16,
+	ZIO_FLAG_OPTIONAL	= 1 << 17,
 
 #define	ZIO_FLAG_VDEV_INHERIT	(ZIO_FLAG_DONT_QUEUE - 1)
 
 	/*
 	 * Flags not inherited by any children.
 	 */
-	ZIO_FLAG_DONT_QUEUE	= 1 << 17,	/* must be first for INHERIT */
+	ZIO_FLAG_DONT_QUEUE	= 1 << 18,	/* must be first for INHERIT */
-	ZIO_FLAG_DONT_PROPAGATE	= 1 << 18,
+	ZIO_FLAG_DONT_PROPAGATE	= 1 << 19,
-	ZIO_FLAG_IO_BYPASS	= 1 << 19,
+	ZIO_FLAG_IO_BYPASS	= 1 << 20,
-	ZIO_FLAG_IO_REWRITE	= 1 << 20,
+	ZIO_FLAG_IO_REWRITE	= 1 << 21,
-	ZIO_FLAG_RAW		= 1 << 21,
+	ZIO_FLAG_RAW		= 1 << 22,
-	ZIO_FLAG_GANG_CHILD	= 1 << 22,
+	ZIO_FLAG_GANG_CHILD	= 1 << 23,
-	ZIO_FLAG_DDT_CHILD	= 1 << 23,
+	ZIO_FLAG_DDT_CHILD	= 1 << 24,
-	ZIO_FLAG_GODFATHER	= 1 << 24,
+	ZIO_FLAG_GODFATHER	= 1 << 25,
-	ZIO_FLAG_NOPWRITE	= 1 << 25,
+	ZIO_FLAG_NOPWRITE	= 1 << 26,
-	ZIO_FLAG_REEXECUTED	= 1 << 26,
+	ZIO_FLAG_REEXECUTED	= 1 << 27,
-	ZIO_FLAG_DELEGATED	= 1 << 27,
+	ZIO_FLAG_DELEGATED	= 1 << 28,
-	ZIO_FLAG_FASTWRITE	= 1 << 28
+	ZIO_FLAG_FASTWRITE	= 1 << 29,
 };
 
 #define	ZIO_FLAG_MUSTSUCCEED		0

module/zfs/dsl_pool.c

@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  * Copyright (c) 2013 Steven Hartland. All rights reserved.
  */
 

module/zfs/metaslab.c

@@ -64,6 +64,21 @@ uint64_t metaslab_gang_bang = SPA_MAXBLOCKSIZE + 1;	/* force gang blocks */
  */
 int zfs_condense_pct = 200;
 
+/*
+ * Condensing a metaslab is not guaranteed to actually reduce the amount of
+ * space used on disk. In particular, a space map uses data in increments of
+ * MAX(1 << ashift, SPACE_MAP_INITIAL_BLOCKSIZE), so a metaslab might use the
+ * same number of blocks after condensing. Since the goal of condensing is to
+ * reduce the number of IOPs required to read the space map, we only want to
+ * condense when we can be sure we will reduce the number of blocks used by the
+ * space map. Unfortunately, we cannot precisely compute whether or not this is
+ * the case in metaslab_should_condense since we are holding ms_lock. Instead,
+ * we apply the following heuristic: do not condense a spacemap unless the
+ * uncondensed size consumes greater than zfs_metaslab_condense_block_threshold
+ * blocks.
+ */
+int zfs_metaslab_condense_block_threshold = 4;
+
 /*
  * The zfs_mg_noalloc_threshold defines which metaslab groups should
  * be eligible for allocation. The value is defined as a percentage of
@@ -1633,6 +1648,8 @@ metaslab_group_preload(metaslab_group_t *mg)
  * times the size than the free space range tree representation
  * (i.e. zfs_condense_pct = 110 and in-core = 1MB, minimal = 1.1.MB).
  *
+ * 3. The on-disk size of the space map should actually decrease.
+ *
  * Checking the first condition is tricky since we don't want to walk
  * the entire AVL tree calculating the estimated on-disk size. Instead we
  * use the size-ordered range tree in the metaslab and calculate the
@@ -1643,13 +1660,21 @@ metaslab_group_preload(metaslab_group_t *mg)
  * To determine the second criterion we use a best-case estimate and assume
  * each segment can be represented on-disk as a single 64-bit entry. We refer
  * to this best-case estimate as the space map's minimal form.
+ *
+ * Unfortunately, we cannot compute the on-disk size of the space map in this
+ * context because we cannot accurately compute the effects of compression, etc.
+ * Instead, we apply the heuristic described in the block comment for
+ * zfs_metaslab_condense_block_threshold - we only condense if the space used
+ * is greater than a threshold number of blocks.
  */
 static boolean_t
 metaslab_should_condense(metaslab_t *msp)
 {
 	space_map_t *sm = msp->ms_sm;
 	range_seg_t *rs;
-	uint64_t size, entries, segsz;
+	uint64_t size, entries, segsz, object_size, optimal_size, record_size;
+	dmu_object_info_t doi;
+	uint64_t vdev_blocksize = 1 << msp->ms_group->mg_vd->vdev_ashift;
 
 	ASSERT(MUTEX_HELD(&msp->ms_lock));
 	ASSERT(msp->ms_loaded);
@@ -1674,9 +1699,15 @@ metaslab_should_condense(metaslab_t *msp)
 	entries = size / (MIN(size, SM_RUN_MAX));
 	segsz = entries * sizeof (uint64_t);
 
-	return (segsz <= space_map_length(msp->ms_sm) &&
+	optimal_size = sizeof (uint64_t) * avl_numnodes(&msp->ms_tree->rt_root);
-	    space_map_length(msp->ms_sm) >= (zfs_condense_pct *
+	object_size = space_map_length(msp->ms_sm);
-	    sizeof (uint64_t) * avl_numnodes(&msp->ms_tree->rt_root)) / 100);
+
+	dmu_object_info_from_db(sm->sm_dbuf, &doi);
+	record_size = MAX(doi.doi_data_block_size, vdev_blocksize);
+
+	return (segsz <= object_size &&
+	    object_size >= (optimal_size * zfs_condense_pct / 100) &&
+	    object_size > zfs_metaslab_condense_block_threshold * record_size);
 }
 
 /*

module/zfs/spa_misc.c

@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  */
 

module/zfs/zfs_debug.c

@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
  */
 
 #include <sys/zfs_context.h>
@@ -136,7 +136,10 @@ zfs_dbgmsg_fini(void)
  * echo ::zfs_dbgmsg | mdb -k
  *
  * Monitor these messages by running:
- * 	dtrace -q -n 'zfs-dbgmsg{printf("%s\n", stringof(arg0))}'
+ * dtrace -qn 'zfs-dbgmsg{printf("%s\n", stringof(arg0))}'
+ *
+ * When used with libzpool, monitor with:
+ * dtrace -qn 'zfs$pid::zfs_dbgmsg:probe1{printf("%s\n", copyinstr(arg1))}'
  */
 void
 zfs_dbgmsg(const char *fmt, ...)

module/zfs/zio.c

@@ -889,8 +889,8 @@ zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
 	ASSERT3U(offset + size, <=, vd->vdev_psize);
 
 	zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private,
-	    ZIO_TYPE_READ, priority, flags, vd, offset, NULL,
+	    ZIO_TYPE_READ, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset,
-	    ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE);
+	    NULL, ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE);
 
 	zio->io_prop.zp_checksum = checksum;
 
@@ -910,8 +910,8 @@ zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
 	ASSERT3U(offset + size, <=, vd->vdev_psize);
 
 	zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private,
-	    ZIO_TYPE_WRITE, priority, flags, vd, offset, NULL,
+	    ZIO_TYPE_WRITE, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset,
-	    ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE);
+	    NULL, ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE);
 
 	zio->io_prop.zp_checksum = checksum;
 
@@ -2642,7 +2642,9 @@ zio_vdev_io_start(zio_t *zio)
 
 	align = 1ULL << vd->vdev_top->vdev_ashift;
 
-	if (P2PHASE(zio->io_size, align) != 0) {
+	if (!(zio->io_flags & ZIO_FLAG_PHYSICAL) &&
+	    P2PHASE(zio->io_size, align) != 0) {
+		/* Transform logical writes to be a full physical block size. */
 		uint64_t asize = P2ROUNDUP(zio->io_size, align);
 		char *abuf = zio_buf_alloc(asize);
 		ASSERT(vd == vd->vdev_top);
@@ -2653,8 +2655,22 @@ zio_vdev_io_start(zio_t *zio)
 		zio_push_transform(zio, abuf, asize, asize, zio_subblock);
 	}
 
-	ASSERT(P2PHASE(zio->io_offset, align) == 0);
+	/*
-	ASSERT(P2PHASE(zio->io_size, align) == 0);
+	 * If this is not a physical io, make sure that it is properly aligned
+	 * before proceeding.
+	 */
+	if (!(zio->io_flags & ZIO_FLAG_PHYSICAL)) {
+		ASSERT0(P2PHASE(zio->io_offset, align));
+		ASSERT0(P2PHASE(zio->io_size, align));
+	} else {
+		/*
+		 * For physical writes, we allow 512b aligned writes and assume
+		 * the device will perform a read-modify-write as necessary.
+		 */
+		ASSERT0(P2PHASE(zio->io_offset, SPA_MINBLOCKSIZE));
+		ASSERT0(P2PHASE(zio->io_size, SPA_MINBLOCKSIZE));
+	}
+
 	VERIFY(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa));
 
 	/*