Implement dynamic gang header sizes

ZFS gang block headers are currently fixed at 512 bytes. This is
increasingly wasteful in the era of larger disk sector sizes. This PR
allows any size allocation to work as a gang header. It also contains
supporting changes to ZDB to make gang headers easier to work with.

Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Signed-off-by: Paul Dagnelie <[email protected]>

Author: Paul Dagnelie

cmd/zdb/zdb.c

@@ -8588,9 +8588,9 @@ zdb_dump_indirect(blkptr_t *bp, int nbps, int flags)
 }
 
 static void
-zdb_dump_gbh(void *buf, int flags)
+zdb_dump_gbh(void *buf, uint64_t size, int flags)
 {
-	zdb_dump_indirect((blkptr_t *)buf, SPA_GBH_NBLKPTRS, flags);
+	zdb_dump_indirect((blkptr_t *)buf, gbh_nblkptrs(size), flags);
 }
 
 static void
@@ -9073,7 +9073,7 @@ zdb_read_block(char *thing, spa_t *spa)
 		zdb_dump_indirect((blkptr_t *)buf,
 		    orig_lsize / sizeof (blkptr_t), flags);
 	else if (flags & ZDB_FLAG_GBH)
-		zdb_dump_gbh(buf, flags);
+		zdb_dump_gbh(buf, lsize, flags);
 	else
 		zdb_dump_block(thing, buf, lsize, flags);
 

include/sys/vdev.h

@@ -148,7 +148,7 @@ extern uint64_t vdev_psize_to_asize(vdev_t *vd, uint64_t psize);
 static inline uint64_t
 vdev_gang_header_asize(vdev_t *vd)
 {
-	return (vdev_psize_to_asize_txg(vd, SPA_GANGBLOCKSIZE, 0));
+	return (vdev_psize_to_asize_txg(vd, SPA_OLD_GANGBLOCKSIZE, 0));
 }
 
 extern int vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux);

include/sys/zio.h

@@ -61,19 +61,38 @@ typedef struct zio_eck {
  * Gang block headers are self-checksumming and contain an array
  * of block pointers.
  */
-#define	SPA_GANGBLOCKSIZE	SPA_MINBLOCKSIZE
-#define	SPA_GBH_NBLKPTRS	((SPA_GANGBLOCKSIZE - \
-	sizeof (zio_eck_t)) / sizeof (blkptr_t))
-#define	SPA_GBH_FILLER		((SPA_GANGBLOCKSIZE - \
-	sizeof (zio_eck_t) - \
-	(SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\
-	sizeof (uint64_t))
-
-typedef struct zio_gbh {
-	blkptr_t		zg_blkptr[SPA_GBH_NBLKPTRS];
-	uint64_t		zg_filler[SPA_GBH_FILLER];
-	zio_eck_t		zg_tail;
-} zio_gbh_phys_t;
+
+typedef enum zio_gb_version {
+	ZIO_GB_OLD = 0,
+	ZIO_GB_SIZED,
+	ZIO_GB_VERSIONS,
+} zio_gb_version_t;
+
+typedef struct zio_gb_tail {
+	uint64_t	zgt_version;	/* gang block type */
+	zio_eck_t	zgt_eck;	/* embedded checksum */
+} zio_gb_tail_t;
+
+#define	SPA_OLD_GANGBLOCKSIZE	SPA_MINBLOCKSIZE
+
+static inline uint64_t
+gbh_nblkptrs(uint64_t size) {
+	return ((size - sizeof (zio_gb_tail_t)) /
+	    sizeof (blkptr_t));
+}
+
+static inline uint64_t
+gbh_filler(uint64_t size) {
+	return ((size - sizeof (zio_gb_tail_t) -
+	    (gbh_nblkptrs(size) * sizeof (blkptr_t))) /
+	    sizeof (uint64_t));
+}
+
+static inline zio_gb_tail_t *
+gbh_tail(void *gbh, uint64_t size) {
+	return ((zio_gb_tail_t *)((uintptr_t)gbh + size -
+	    sizeof (zio_gb_tail_t)));
+}
 
 enum zio_checksum {
 	ZIO_CHECKSUM_INHERIT = 0,
@@ -398,8 +417,10 @@ typedef struct zio_vsd_ops {
 } zio_vsd_ops_t;
 
 typedef struct zio_gang_node {
-	zio_gbh_phys_t		*gn_gbh;
-	struct zio_gang_node	*gn_child[SPA_GBH_NBLKPTRS];
+	void			*gn_gbh;
+	uint64_t		gn_gangblocksize;
+	uint64_t		gn_orig_gangblocksize;
+	struct zio_gang_node	*gn_child[];
 } zio_gang_node_t;
 
 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,

include/zfeature_common.h

@@ -87,6 +87,7 @@ typedef enum spa_feature {
 	SPA_FEATURE_FAST_DEDUP,
 	SPA_FEATURE_LONGNAME,
 	SPA_FEATURE_LARGE_MICROZAP,
+	SPA_FEATURE_DYNAMIC_GANG_HEADER,
 	SPA_FEATURES
 } spa_feature_t;
 
@@ -103,7 +104,14 @@ typedef enum zfeature_flags {
 	/* Activate this feature at the same time it is enabled. */
 	ZFEATURE_FLAG_ACTIVATE_ON_ENABLE =	(1 << 2),
 	/* Each dataset has a field set if it has ever used this feature. */
-	ZFEATURE_FLAG_PER_DATASET =		(1 << 3)
+	ZFEATURE_FLAG_PER_DATASET =		(1 << 3),
+	/*
+	 * This feature isn't enabled by zpool upgrade; it must be explicitly
+	 * listed to be enabled. This also applies to compatibility lists. This
+	 * flag can be removed from a given feature once support is sufficiently
+	 * widespread.
+	 */
+	ZFEATURE_FLAG_NO_UPGRADE = 		(1 << 4)
 } zfeature_flags_t;
 
 typedef enum zfeature_type {

lib/libnvpair/libnvpair.abi

@@ -5122,9 +5122,13 @@ zpool_load_compat(const char *compat, boolean_t *features, char *report,
 	/* special cases (unset), "" and "off" => enable all features */
 	if (compat == NULL || compat[0] == '\0' ||
 	    strcmp(compat, ZPOOL_COMPAT_OFF) == 0) {
-		if (features != NULL)
-			for (uint_t i = 0; i < SPA_FEATURES; i++)
-				features[i] = B_TRUE;
+		if (features != NULL) {
+			for (uint_t i = 0; i < SPA_FEATURES; i++) {
+				if (!(spa_feature_table[i].fi_flags &
+				    ZFEATURE_FLAG_NO_UPGRADE))
+					features[i] = B_TRUE;
+			}
+		}
 		if (report != NULL)
 			strlcpy(report, gettext("all features enabled"), rlen);
 		return (ZPOOL_COMPATIBILITY_OK);