Tony's feedback

Signed-off-by: Paul Dagnelie <[email protected]>

Author: Paul Dagnelie

cmd/zdb/zdb.c

@@ -9394,7 +9394,7 @@ zdb_print_anyraid_tile_layout(vdev_t *vd)
 	// Create and populate table with all the values we need to print.
 	char ***table = malloc(sizeof (*table) * cols);
 	for (int i = 0; i < cols; i++) {
-		table[i] = calloc(var->vd_children[i]->van_capacity,
+		table[i] = calloc(var->vd_children[i]->van_capacity + 1,
 		    sizeof (**table));
 	}
 
@@ -9440,7 +9440,7 @@ zdb_print_anyraid_tile_layout(vdev_t *vd)
 		for (int v = 0; v < cols; v++) {
 			if (final[v]) {
 				ASSERT3U(i, >=,
-				    var->vd_children[v]->van_capacity);
+				    var->vd_children[v]->van_capacity + 1);
 				int extra_width = 0;
 				if (v == 0 || !printed[v - 1])
 					extra_width++;
@@ -9449,7 +9449,7 @@ zdb_print_anyraid_tile_layout(vdev_t *vd)
 				printed[v] = B_FALSE;
 				continue;
 			}
-			if (i + 1 == var->vd_children[v]->van_capacity)
+			if (i + 1 == var->vd_children[v]->van_capacity + 1)
 				final[v] = B_TRUE;
 			if (v - 1 != last_printed)
 				(void) printf("│");
@@ -9466,7 +9466,7 @@ zdb_print_anyraid_tile_layout(vdev_t *vd)
 	}
 	(void) printf("\n");
 	for (int i = 0; i < cols; i++) {
-		for (int j = 0; j < var->vd_children[i]->van_capacity; j++)
+		for (int j = 0; j < var->vd_children[i]->van_capacity + 1; j++)
 			if (table[i][j])
 				free(table[i][j]);
 		free(table[i]);
@@ -9539,7 +9539,7 @@ print_anyraid_mapping(vdev_t *vd, int child, int mapping,
 	    &disk_id) != 0)
 		(void) printf("No valid disk ID\n");
 
-	(void) printf("version:    %6d\ttile size: %8lx\ttxg: %lu\n",
+	(void) printf("version:    %6d\ttile size: %#8lx\ttxg: %lu\n",
 	    version, tile_size, written_txg);
 	(void) printf("map length: %6u\tdisk id: %3u\n", map_length, disk_id);
 
@@ -9722,12 +9722,18 @@ zdb_dump_anyraid_map_vdev(vdev_t *vd, int verbosity)
 	ASSERT3P(vd->vdev_ops, ==, &vdev_anyraid_ops);
 	vdev_anyraid_t *var = vd->vdev_tsd;
 
-	(void) printf("\t%-5s%11llu   %s %16llx\n",
+	(void) printf("\t%-5s%11llu   %s %#16llx\n",
 	    "vdev", (u_longlong_t)vd->vdev_id,
 	    "tile_size", (u_longlong_t)var->vd_tile_size);
-	(void) printf("\t%-8s%8llu   %-12s %10u\n", "tiles",
-	    (u_longlong_t)avl_numnodes(&var->vd_tile_map),
-	    "checkpoint tile", var->vd_checkpoint_tile);
+	(void) printf("\t%-8s%8llu", "tiles",
+	    (u_longlong_t)avl_numnodes(&var->vd_tile_map));
+	if (var->vd_checkpoint_tile != UINT32_MAX) {
+		(void) printf(".  %-12s %10u\n", "checkpoint tile",
+		    var->vd_checkpoint_tile);
+	} else {
+		(void) printf("\n");
+	}
+
 	(void) printf("\t%16s   %12s   %13s\n", "----------------",
 	    "------------", "-------------");
 
@@ -9759,23 +9765,26 @@ zdb_dump_anyraid_map(char *vdev_str, spa_t *spa, int verbosity)
 {
 	vdev_t *rvd, *vd;
 
-	(void) printf("\nAnyRAID tiles:\n");
-
 	/* A specific vdev. */
 	if (vdev_str != NULL) {
 		vd = zdb_vdev_lookup(spa->spa_root_vdev, vdev_str);
 		if (vd == NULL) {
 			(void) printf("Invalid vdev: %s\n", vdev_str);
 			return (EINVAL);
 		}
-		if (vd->vdev_ops != &vdev_anyraid_ops) {
+		if (vd->vdev_ops != &vdev_anyraid_ops &&
+		    (vd->vdev_parent == NULL ||
+		    (vd = vd->vdev_parent)->vdev_ops != &vdev_anyraid_ops)) {
 			(void) printf("Not an anyraid vdev: %s\n", vdev_str);
 			return (EINVAL);
 		}
+
+		(void) printf("\nAnyRAID tiles:\n");
 		zdb_dump_anyraid_map_vdev(vd, verbosity);
 		return (0);
 	}
 
+	(void) printf("\nAnyRAID tiles:\n");
 	/* All anyraid vdevs. */
 	rvd = spa->spa_root_vdev;
 	for (uint64_t c = 0; c < rvd->vdev_children; c++) {

cmd/zpool/zpool_vdev.c

@@ -459,7 +459,7 @@ is_raidz_mirror(replication_level_t *a, replication_level_t *b,
 	if ((strcmp(a->zprl_type, "raidz") == 0 ||
 	    strcmp(a->zprl_type, "draid") == 0) &&
 	    (strcmp(b->zprl_type, "mirror") == 0 ||
-	    strcmp(b->zprl_type, "anyraid") == 0)) {
+	    strcmp(b->zprl_type, "anymirror") == 0)) {
 		*raidz = a;
 		*mirror = b;
 		return (B_TRUE);
@@ -555,11 +555,11 @@ get_replication(nvlist_t *nvroot, boolean_t fatal)
 			rep.zprl_children = 0;
 
 			if (strcmp(type, VDEV_TYPE_RAIDZ) == 0 ||
-			    strcmp(type, VDEV_TYPE_DRAID) == 0) {
+			    strcmp(type, VDEV_TYPE_DRAID) == 0 ||
+			    strcmp(type, VDEV_TYPE_ANYRAID) == 0) {
 				verify(nvlist_lookup_uint64(nv,
 				    ZPOOL_CONFIG_NPARITY,
 				    &rep.zprl_parity) == 0);
-				assert(rep.zprl_parity != 0);
 			} else {
 				rep.zprl_parity = 0;
 			}
@@ -771,7 +771,9 @@ get_replication(nvlist_t *nvroot, boolean_t fatal)
 					    rep.zprl_type);
 				else
 					return (NULL);
-			} else if (lastrep.zprl_children != rep.zprl_children) {
+			} else if (lastrep.zprl_children !=
+			    rep.zprl_children && strcmp(rep.zprl_type,
+			    VDEV_TYPE_ANYRAID) != 0) {
 				if (ret)
 					free(ret);
 				ret = NULL;
@@ -1258,7 +1260,7 @@ get_parity(const char *type)
 			parity = strtol(p, &end, 10);
 			if (errno != 0 || *end != '\0' ||
 			    parity < 0 || parity > VDEV_ANYRAID_MAXPARITY) {
-				return (0);
+				return (-1);
 			}
 		}
 	} else if (strncmp(type, VDEV_TYPE_DRAID,
@@ -1320,6 +1322,8 @@ is_grouping(const char *type, int *mindev, int *maxdev)
 
 	if (strncmp(type, VDEV_TYPE_ANYRAID, strlen(VDEV_TYPE_ANYRAID)) == 0) {
 		nparity = get_parity(type);
+		if (nparity < 0)
+			return (NULL);
 		if (mindev != NULL)
 			*mindev = nparity + 1;
 		if (maxdev != NULL)

cmd/ztest.c

@@ -189,7 +189,7 @@ typedef struct ztest_shared_opts {
 	int zo_raid_do_expand;
 	int zo_raid_children;
 	int zo_raid_parity;
-	char zo_raid_type[8];
+	char zo_raid_type[16];
 	int zo_draid_data;
 	int zo_draid_spares;
 	int zo_datasets;
@@ -773,7 +773,7 @@ static ztest_option_t option_table[] = {
 	    DEFAULT_RAID_CHILDREN, NULL},
 	{ 'R',	"raid-parity", "INTEGER", "Raid parity",
 	    DEFAULT_RAID_PARITY, NULL},
-	{ 'K',  "raid-kind", "raidz|eraidz|draid|anyraid|random", "Raid kind",
+	{ 'K',  "raid-kind", "raidz|eraidz|draid|anymirror|random", "Raid kind",
 	    NO_DEFAULT, "random"},
 	{ 'D',	"draid-data", "INTEGER", "Number of draid data drives",
 	    DEFAULT_DRAID_DATA, NULL},
@@ -1134,7 +1134,7 @@ process_options(int argc, char **argv)
 			raid_kind = "draid";
 			break;
 		case 3:
-			raid_kind = "anyraid";
+			raid_kind = "anymirror";
 			break;
 		}
 
@@ -1190,7 +1190,7 @@ process_options(int argc, char **argv)
 	} else if (strcmp(raid_kind, "raidz") == 0) {
 		zo->zo_raid_parity = MIN(zo->zo_raid_parity,
 		    zo->zo_raid_children - 1);
-	} else if (strcmp(raid_kind, "anyraid") == 0) {
+	} else if (strcmp(raid_kind, "anymirror") == 0) {
 		uint64_t min_devsize;
 
 		/* With fewer disks use 1G, otherwise 512M is OK */
@@ -3814,7 +3814,8 @@ ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
 	if (ztest_opts.zo_raid_children > 1) {
 		if (strcmp(oldvd->vdev_ops->vdev_op_type, "raidz") == 0)
 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_raidz_ops);
-		else if (strcmp(oldvd->vdev_ops->vdev_op_type, "anyraid") == 0)
+		else if (strcmp(oldvd->vdev_ops->vdev_op_type, "anymirror") ==
+		    0)
 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_anyraid_ops);
 		else
 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_draid_ops);
@@ -3838,11 +3839,7 @@ ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
 	}
 
 	oldguid = oldvd->vdev_guid;
-	if (oldvd->vdev_ops != &vdev_anyraid_ops)
-		oldsize = vdev_get_min_asize(oldvd);
-	else
-		oldsize = oldvd->vdev_child[
-		    ztest_random(oldvd->vdev_children)]->vdev_asize;
+	oldsize = vdev_get_min_attach_size(oldvd);
 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
 	oldvd_is_special =
 	    oldvd->vdev_top->vdev_alloc_bias == VDEV_BIAS_SPECIAL ||

include/sys/fs/zfs.h

@@ -916,7 +916,7 @@ typedef struct zpool_load_policy {
 #define	VDEV_TYPE_MIRROR		"mirror"
 #define	VDEV_TYPE_REPLACING		"replacing"
 #define	VDEV_TYPE_RAIDZ			"raidz"
-#define	VDEV_TYPE_ANYRAID		"anyraid"
+#define	VDEV_TYPE_ANYRAID		"anymirror"
 #define	VDEV_TYPE_DRAID			"draid"
 #define	VDEV_TYPE_DRAID_SPARE		"dspare"
 #define	VDEV_TYPE_DISK			"disk"

include/sys/vdev_anyraid.h

@@ -44,6 +44,7 @@ typedef struct vdev_anyraid_node {
 	avl_node_t	van_node;
 	uint8_t		van_id;
 	uint16_t	van_next_offset;
+	// Note: store capacity - 1 for rollover reasons
 	uint16_t	van_capacity;
 } vdev_anyraid_node_t;
 
@@ -275,7 +276,6 @@ _Static_assert(VDEV_ANYRAID_MAP_SIZE % SPA_MAXBLOCKSIZE == 0, "");
 void vdev_anyraid_write_map_sync(vdev_t *vd, zio_t *pio, uint64_t txg,
     uint64_t *good_writes, int flags, vdev_config_sync_status_t status);
 
-uint64_t vdev_anyraid_min_newsize(vdev_t *vd, uint64_t ashift);
 void vdev_anyraid_expand(vdev_t *tvd, vdev_t *newvd);
 boolean_t vdev_anyraid_mapped(vdev_t *vd, uint64_t offset);
 

include/sys/vdev_impl.h

@@ -71,6 +71,7 @@ typedef int	vdev_open_func_t(vdev_t *vd, uint64_t *size, uint64_t *max_size,
 typedef void	vdev_close_func_t(vdev_t *vd);
 typedef uint64_t vdev_asize_func_t(vdev_t *vd, uint64_t psize, uint64_t txg);
 typedef uint64_t vdev_min_asize_func_t(vdev_t *pvd, vdev_t *cvd);
+typedef uint64_t vdev_min_attach_size_func_t(vdev_t *vd);
 typedef uint64_t vdev_min_alloc_func_t(vdev_t *vd);
 typedef void	vdev_io_start_func_t(zio_t *zio);
 typedef void	vdev_io_done_func_t(zio_t *zio);
@@ -107,6 +108,7 @@ typedef const struct vdev_ops {
 	vdev_asize_func_t		*vdev_op_psize_to_asize;
 	vdev_asize_func_t		*vdev_op_asize_to_psize;
 	vdev_min_asize_func_t		*vdev_op_min_asize;
+	vdev_min_attach_size_func_t		*vdev_op_min_attach_size;
 	vdev_min_alloc_func_t		*vdev_op_min_alloc;
 	vdev_io_start_func_t		*vdev_op_io_start;
 	vdev_io_done_func_t		*vdev_op_io_done;
@@ -630,7 +632,9 @@ extern void vdev_default_xlate(vdev_t *vd, const zfs_range_seg64_t *logical_rs,
 extern uint64_t vdev_default_psize(vdev_t *vd, uint64_t asize, uint64_t txg);
 extern uint64_t vdev_default_asize(vdev_t *vd, uint64_t psize, uint64_t txg);
 extern uint64_t vdev_default_min_asize(vdev_t *pvd, vdev_t *cvd);
+extern uint64_t vdev_default_min_attach_size(vdev_t *vd);
 extern uint64_t vdev_get_min_asize(vdev_t *vd);
+extern uint64_t vdev_get_min_attach_size(vdev_t *vd);
 extern void vdev_set_min_asize(vdev_t *vd);
 extern uint64_t vdev_get_nparity(vdev_t *vd);
 extern uint64_t vdev_get_ndisks(vdev_t *vd);

lib/libzfs/libzfs_pool.c

@@ -1223,7 +1223,8 @@ zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 	    strncmp(pool, "raidz", 5) == 0 ||
 	    strncmp(pool, "draid", 5) == 0 ||
 	    strncmp(pool, "spare", 5) == 0 ||
-	    strcmp(pool, "log") == 0)) {
+	    strcmp(pool, "log") == 0 ||
+	    strncmp(pool, "anymirror", 9) == 0)) {
 		if (hdl != NULL)
 			zfs_error_aux(hdl,
 			    dgettext(TEXT_DOMAIN, "name is reserved"));

man/man4/zfs.4

@@ -655,6 +655,15 @@ Logical ashift for file-based devices.
 .It Sy vdev_file_physical_ashift Ns = Ns Sy 9 Po 512 B Pc Pq u64
 Physical ashift for file-based devices.
 .
+.It Sy zfs_anyraid_min_tile_size Ns = Ns Sy 16 GiB Pq u64
+Minimum size of the tiles that anyraid will use to do its mapping.
+Smaller tile sizes let data be spread more evenly across devices, and makes
+smaller devices use more of their capacity.
+Larger tile sizes allow for larger disks to be used in the future, since a given
+device can only store 16384 tiles.
+The minimum valid tile size is 16MiB, since a metaslab always needs to be able
+to fit in a single tile.
+.
 .It Sy zap_iterate_prefetch Ns = Ns Sy 1 Ns | Ns 0 Pq int
 If set, when we start iterating over a ZAP object,
 prefetch the entire object (all leaf blocks).

man/man7/vdevprops.7

@@ -188,6 +188,21 @@ If this device should perform new allocations, used to disable a device
 when it is scheduled for later removal.
 See
 .Xr zpool-remove 8 .
+.It anyraid_tile_capacity
+Only valid for
+.Sy AnyRAID
+vdevs and their leaf vdevs.
+The number of physical tiles that the vdev can hold.
+.It anyraid_tile_count
+Only valid for
+.Sy AnyRAID
+vdevs and their leaf vdevs.
+The number of physical tiles that are currently allocated on the vdev.
+.It anyraid_tile_size
+Only valid for
+.Sy AnyRAID
+vdevs and their leaf vdevs.
+The size of the tiles in use on this vdev.
 .El
 .Ss User Properties
 In addition to the standard native properties, ZFS supports arbitrary user

man/man7/zpoolconcepts.7

@@ -165,7 +165,7 @@ An error is returned when the provided number of children differs.
 The number of distributed hot spares.
 Defaults to zero.
 .El
-.It Sy anyraid , anyraid0 , anyraid1 , anyraid2
+.It Sy anymirror , anymirror0 , anymirror1 , anymirror2
 A new device type that allows for mirror-parity redundancy while using devices
 of different sizes.
 An AnyRAID vdev works by dividing each of the underlying disks that make it up
@@ -177,13 +177,17 @@ disks, while enabling maximum space usage by allocating more tiles from the
 disks with the most free space.
 In addition, the device can be expanded by attaching new disks, and new tiles
 will be allocated from those disks.
-.Sy anyraid
+The vdev class as a whole is referred to as AnyRAID; anymirror vdevs
+specifically use mirror-style parity.
+Future work will also add anyraidz, which will use the same basic tile
+architecture, but use raidz-style parity.
+.Sy anymirror
 is a synonym for
-.Sy anyraid1
+.Sy anymirror1
 , which is the 2-way mirror parity version (1 parity tile).
-.Sy anyraid2
+.Sy anymirror2
 is a 3-way mirror (2 parity tiles), while
-.Sy anyraid0
+.Sy anymirror0
 is striped (no parity tiles), and is primarily intended for testing.
 .It Sy spare
 A pseudo-vdev which keeps track of available hot spares for a pool.