Reduce blkptr logical and physical birth to 48 bits

This PR reduces the blkptr's logical and physical birth times from 64-bit to 48-bits. The upper 16-bits are reserved for future work. Consumers of the blkptr's logical or physical birth times must use the newly provided macros when accessing those members of the blkptr structure. Signed-off-by: George Wilson <[email protected]>
openzfs · Mar 11, 2024 · 2cecdbb · 2cecdbb
1 parent 8f2f6cd
commit 2cecdbb
Show file tree

Hide file tree

Showing 32 changed files with 231 additions and 199 deletions.
diff --git a/cmd/zdb/zdb.c b/cmd/zdb/zdb.c
@@ -199,7 +199,8 @@ sublivelist_verify_blkptr(void *arg, const blkptr_t *bp, boolean_t free,
 					break;
 				sublivelist_verify_block_t svb = {
 				    .svb_dva = bp->blk_dva[i],
-				    .svb_allocated_txg = bp->blk_birth
+				    .svb_allocated_txg =
+				    BP_GET_LOGICAL_BIRTH(bp)
 				};
 
 				if (zfs_btree_find(&sv->sv_leftover, &svb,
@@ -2340,7 +2341,7 @@ snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp,
 		    (int)BPE_GET_ETYPE(bp),
 		    (u_longlong_t)BPE_GET_LSIZE(bp),
 		    (u_longlong_t)BPE_GET_PSIZE(bp),
-		    (u_longlong_t)bp->blk_birth);
+		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
 		return;
 	}
 
@@ -2358,16 +2359,16 @@ snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp,
 		    buflen - strlen(blkbuf),
 		    "%llxL B=%llu",
 		    (u_longlong_t)BP_GET_LSIZE(bp),
-		    (u_longlong_t)bp->blk_birth);
+		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));
 	} else {
 		(void) snprintf(blkbuf + strlen(blkbuf),
 		    buflen - strlen(blkbuf),
 		    "%llxL/%llxP F=%llu B=%llu/%llu",
 		    (u_longlong_t)BP_GET_LSIZE(bp),
 		    (u_longlong_t)BP_GET_PSIZE(bp),
 		    (u_longlong_t)BP_GET_FILL(bp),
-		    (u_longlong_t)bp->blk_birth,
-		    (u_longlong_t)BP_PHYSICAL_BIRTH(bp));
+		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp),
+		    (u_longlong_t)BP_GET_BIRTH(bp));
 		if (bp_freed)
 			(void) snprintf(blkbuf + strlen(blkbuf),
 			    buflen - strlen(blkbuf), " %s", "FREE");
@@ -2417,7 +2418,7 @@ visit_indirect(spa_t *spa, const dnode_phys_t *dnp,
 {
 	int err = 0;
 
-	if (bp->blk_birth == 0)
+	if (BP_GET_LOGICAL_BIRTH(bp) == 0)
 		return (0);
 
 	print_indirect(spa, bp, zb, dnp);
@@ -2605,7 +2606,7 @@ dump_bptree_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
 	(void) arg, (void) tx;
 	char blkbuf[BP_SPRINTF_LEN];
 
-	if (bp->blk_birth != 0) {
+	if (BP_GET_LOGICAL_BIRTH(bp) != 0) {
 		snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
 		(void) printf("\t%s\n", blkbuf);
 	}
@@ -2646,7 +2647,7 @@ dump_bpobj_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx)
 	(void) arg, (void) tx;
 	char blkbuf[BP_SPRINTF_LEN];
 
-	ASSERT(bp->blk_birth != 0);
+	ASSERT(BP_GET_LOGICAL_BIRTH(bp) != 0);
 	snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp, bp_freed);
 	(void) printf("\t%s\n", blkbuf);
 	return (0);
@@ -5788,7 +5789,7 @@ zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 	if (zb->zb_level == ZB_DNODE_LEVEL)
 		return (0);
 
-	if (dump_opt['b'] >= 5 && bp->blk_birth > 0) {
+	if (dump_opt['b'] >= 5 && BP_GET_LOGICAL_BIRTH(bp) > 0) {
 		char blkbuf[BP_SPRINTF_LEN];
 		snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
 		(void) printf("objset %llu object %llu "

diff --git a/cmd/zdb/zdb_il.c b/cmd/zdb/zdb_il.c
@@ -173,8 +173,8 @@ zil_prt_rec_write(zilog_t *zilog, int txtype, const void *arg)
 
 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
 		(void) printf("%shas blkptr, %s\n", tab_prefix,
-		    !BP_IS_HOLE(bp) &&
-		    bp->blk_birth >= spa_min_claim_txg(zilog->zl_spa) ?
+		    !BP_IS_HOLE(bp) && BP_GET_LOGICAL_BIRTH(bp) >=
+		    spa_min_claim_txg(zilog->zl_spa) ?
 		    "will claim" : "won't claim");
 		print_log_bp(bp, tab_prefix);
 
@@ -186,7 +186,7 @@ zil_prt_rec_write(zilog_t *zilog, int txtype, const void *arg)
 			(void) printf("%s<hole>\n", tab_prefix);
 			return;
 		}
-		if (bp->blk_birth < zilog->zl_header->zh_claim_txg) {
+		if (BP_GET_LOGICAL_BIRTH(bp) < zilog->zl_header->zh_claim_txg) {
 			(void) printf("%s<block already committed>\n",
 			    tab_prefix);
 			return;
@@ -237,8 +237,8 @@ zil_prt_rec_write_enc(zilog_t *zilog, int txtype, const void *arg)
 
 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
 		(void) printf("%shas blkptr, %s\n", tab_prefix,
-		    !BP_IS_HOLE(bp) &&
-		    bp->blk_birth >= spa_min_claim_txg(zilog->zl_spa) ?
+		    !BP_IS_HOLE(bp) && BP_GET_LOGICAL_BIRTH(bp) >=
+		    spa_min_claim_txg(zilog->zl_spa) ?
 		    "will claim" : "won't claim");
 		print_log_bp(bp, tab_prefix);
 	}
@@ -473,7 +473,7 @@ print_log_block(zilog_t *zilog, const blkptr_t *bp, void *arg,
 
 	if (claim_txg != 0)
 		claim = "already claimed";
-	else if (bp->blk_birth >= spa_min_claim_txg(zilog->zl_spa))
+	else if (BP_GET_LOGICAL_BIRTH(bp) >= spa_min_claim_txg(zilog->zl_spa))
 		claim = "will claim";
 	else
 		claim = "won't claim";

diff --git a/cmd/zhack.c b/cmd/zhack.c
@@ -612,8 +612,8 @@ zhack_repair_undetach(uberblock_t *ub, nvlist_t *cfg, const int l)
 	 * Uberblock root block pointer has valid birth TXG.
 	 * Copying it to the label NVlist
 	 */
-	if (ub->ub_rootbp.blk_birth != 0) {
-		const uint64_t txg = ub->ub_rootbp.blk_birth;
+	if (BP_GET_LOGICAL_BIRTH(&ub->ub_rootbp) != 0) {
+		const uint64_t txg = BP_GET_LOGICAL_BIRTH(&ub->ub_rootbp);
 		ub->ub_txg = txg;
 
 		if (nvlist_remove_all(cfg, ZPOOL_CONFIG_CREATE_TXG) != 0) {

diff --git a/include/sys/spa.h b/include/sys/spa.h
@@ -125,15 +125,15 @@ typedef struct zio_cksum_salt {
  *
  *	64	56	48	40	32	24	16	8	0
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
- * 0	|  pad  |	  vdev1         | GRID  |	  ASIZE		|
+ * 0	|  pad  |	  vdev1         | pad   |	  ASIZE		|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
  * 1	|G|			 offset1				|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
- * 2	|  pad  |	  vdev2         | GRID  |	  ASIZE		|
+ * 2	|  pad  |	  vdev2         | pad   |	  ASIZE		|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
  * 3	|G|			 offset2				|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
- * 4	|  pad  |	  vdev3         | GRID  |	  ASIZE		|
+ * 4	|  pad  |	  vdev3         | pad   |	  ASIZE		|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
  * 5	|G|			 offset3				|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
@@ -143,9 +143,9 @@ typedef struct zio_cksum_salt {
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
  * 8	|			padding					|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
- * 9	|			physical birth txg			|
+ * 9	|	RSVD	|	physical birth txg			|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
- * a	|			logical birth txg			|
+ * a	|	RSVD	|	logical birth txg			|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
  * b	|			fill count				|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
@@ -165,7 +165,6 @@ typedef struct zio_cksum_salt {
  * LSIZE	logical size
  * PSIZE	physical size (after compression)
  * ASIZE	allocated size (including RAID-Z parity and gang block headers)
- * GRID		RAID-Z layout information (reserved for future use)
  * cksum	checksum function
  * comp		compression function
  * G		gang block indicator
@@ -175,6 +174,7 @@ typedef struct zio_cksum_salt {
  * E		blkptr_t contains embedded data (see below)
  * lvl		level of indirection
  * type		DMU object type
+ * RSVD		reserved for future use
  * phys birth	txg when dva[0] was written; zero if same as logical birth txg
  *              note that typically all the dva's would be written in this
  *              txg, but they could be different if they were moved by
@@ -190,11 +190,11 @@ typedef struct zio_cksum_salt {
  *
  *	64	56	48	40	32	24	16	8	0
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
- * 0	|		vdev1		| GRID  |	  ASIZE		|
+ * 0	|		vdev1		| pad   |	  ASIZE		|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
  * 1	|G|			 offset1				|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
- * 2	|		vdev2		| GRID  |	  ASIZE		|
+ * 2	|		vdev2		| pad   |	  ASIZE		|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
  * 3	|G|			 offset2				|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
@@ -208,9 +208,9 @@ typedef struct zio_cksum_salt {
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
  * 8	|			padding					|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
- * 9	|			physical birth txg			|
+ * 9	|	RSVD	|	physical birth txg			|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
- * a	|			logical birth txg			|
+ * a	|	RSVD	|	logical birth txg			|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
  * b	|		IV2		|	    fill count		|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
@@ -283,7 +283,7 @@ typedef struct zio_cksum_salt {
  * 8	|      payload                                                  |
  * 9	|      payload                                                  |
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
- * a	|			logical birth txg			|
+ * a	|	RSVD	|	logical birth txg			|
  *	+-------+-------+-------+-------+-------+-------+-------+-------+
  * b	|      payload                                                  |
  * c	|      payload                                                  |
@@ -307,6 +307,7 @@ typedef struct zio_cksum_salt {
  * LSIZE		logical size of payload, in bytes
  *			note that 25 bits is enough to store the largest
  *			"normal" BP's LSIZE (2^16 * 2^9) in bytes
+ * RSVD			reserved for future use
  * log. birth		transaction group in which the block was logically born
  *
  * Note that LSIZE and PSIZE are stored in bytes, whereas for non-embedded
@@ -355,7 +356,7 @@ typedef enum bp_embedded_type {
 #define	BPE_NUM_WORDS 14
 #define	BPE_PAYLOAD_SIZE (BPE_NUM_WORDS * sizeof (uint64_t))
 #define	BPE_IS_PAYLOADWORD(bp, wp) \
-	((wp) != &(bp)->blk_prop && (wp) != &(bp)->blk_birth)
+	((wp) != &(bp)->blk_prop && (wp) != (&(bp)->blk_birth_word[1]))
 
 #define	SPA_BLKPTRSHIFT	7		/* blkptr_t is 128 bytes	*/
 #define	SPA_DVAS_PER_BP	3		/* Number of DVAs in a bp	*/
@@ -374,8 +375,19 @@ typedef struct blkptr {
 	dva_t		blk_dva[SPA_DVAS_PER_BP]; /* Data Virtual Addresses */
 	uint64_t	blk_prop;	/* size, compression, type, etc	    */
 	uint64_t	blk_pad[2];	/* Extra space for the future	    */
-	uint64_t	blk_phys_birth;	/* txg when block was allocated	    */
-	uint64_t	blk_birth;	/* transaction group at birth	    */
+
+	/*
+	 * The following two words MUST NOT be accessed directly since
+	 * only 48 bits of the word contain the corresponding birth txg.
+	 * The BP_[GET|SET]_BIRTH macros should be used to manipulate the
+	 * physical and logical birth times. Those macros will set/get
+	 * the value in both of the words.
+	 *
+	 * To retrieve or manipulate one of the words, BP_GET_LOGICAL_BIRTH
+	 * and BP_GET_PHYSICAL_BIRTH (and a corresponding SET macro) must be
+	 * used.
+	 */
+	uint64_t	blk_birth_word[2];
 	uint64_t	blk_fill;	/* fill count			    */
 	zio_cksum_t	blk_cksum;	/* 256-bit checksum		    */
 } blkptr_t;
@@ -395,9 +407,6 @@ typedef struct blkptr {
 	BF64_SET_SB((dva)->dva_word[0], 0, SPA_ASIZEBITS, \
 	SPA_MINBLOCKSHIFT, 0, x)
 
-#define	DVA_GET_GRID(dva)	BF64_GET((dva)->dva_word[0], 24, 8)
-#define	DVA_SET_GRID(dva, x)	BF64_SET((dva)->dva_word[0], 24, 8, x)
-
 #define	DVA_GET_VDEV(dva)	BF64_GET((dva)->dva_word[0], 32, SPA_VDEVBITS)
 #define	DVA_SET_VDEV(dva, x)	\
 	BF64_SET((dva)->dva_word[0], 32, SPA_VDEVBITS, x)
@@ -480,15 +489,27 @@ typedef struct blkptr {
 #define	BP_GET_FREE(bp)			BF64_GET((bp)->blk_fill, 0, 1)
 #define	BP_SET_FREE(bp, x)		BF64_SET((bp)->blk_fill, 0, 1, x)
 
-#define	BP_PHYSICAL_BIRTH(bp)		\
-	(BP_IS_EMBEDDED(bp) ? 0 : \
-	(bp)->blk_phys_birth ? (bp)->blk_phys_birth : (bp)->blk_birth)
+#define	BP_GET_LOGICAL_BIRTH(bp)	\
+	BF64_GET((bp)->blk_birth_word[1], 0, 48)
+#define	BP_SET_LOGICAL_BIRTH(bp, x)	\
+	BF64_SET((bp)->blk_birth_word[1], 0, 48, x)
+
+#define	BP_GET_PHYSICAL_BIRTH(bp)	\
+	BF64_GET((bp)->blk_birth_word[0], 0, 48)
+#define	BP_SET_PHYSICAL_BIRTH(bp, x)	\
+	BF64_SET((bp)->blk_birth_word[0], 0, 48, x)
+
+#define	BP_GET_BIRTH(bp)					\
+	(BP_IS_EMBEDDED(bp) ? 0 : 				\
+	BP_GET_PHYSICAL_BIRTH(bp) ? BP_GET_PHYSICAL_BIRTH(bp) :	\
+	BP_GET_LOGICAL_BIRTH(bp))
 
 #define	BP_SET_BIRTH(bp, logical, physical)	\
 {						\
 	ASSERT(!BP_IS_EMBEDDED(bp));		\
-	(bp)->blk_birth = (logical);		\
-	(bp)->blk_phys_birth = ((logical) == (physical) ? 0 : (physical)); \
+	BP_SET_LOGICAL_BIRTH(bp, logical);	\
+	BP_SET_PHYSICAL_BIRTH(bp, 		\
+	    ((logical) == (physical) ? 0 : (physical))); \
 }
 
 #define	BP_GET_FILL(bp)				\
@@ -541,8 +562,8 @@ typedef struct blkptr {
 	(dva1)->dva_word[0] == (dva2)->dva_word[0])
 
 #define	BP_EQUAL(bp1, bp2)	\
-	(BP_PHYSICAL_BIRTH(bp1) == BP_PHYSICAL_BIRTH(bp2) &&	\
-	(bp1)->blk_birth == (bp2)->blk_birth &&			\
+	(BP_GET_BIRTH(bp1) == BP_GET_BIRTH(bp2) &&	\
+	BP_GET_LOGICAL_BIRTH(bp1) == BP_GET_LOGICAL_BIRTH(bp2) &&	\
 	DVA_EQUAL(&(bp1)->blk_dva[0], &(bp2)->blk_dva[0]) &&	\
 	DVA_EQUAL(&(bp1)->blk_dva[1], &(bp2)->blk_dva[1]) &&	\
 	DVA_EQUAL(&(bp1)->blk_dva[2], &(bp2)->blk_dva[2]))
@@ -581,8 +602,8 @@ typedef struct blkptr {
 	(bp)->blk_prop = 0;			\
 	(bp)->blk_pad[0] = 0;			\
 	(bp)->blk_pad[1] = 0;			\
-	(bp)->blk_phys_birth = 0;		\
-	(bp)->blk_birth = 0;			\
+	(bp)->blk_birth_word[0] = 0;		\
+	(bp)->blk_birth_word[1] = 0;		\
 	(bp)->blk_fill = 0;			\
 	ZIO_SET_CHECKSUM(&(bp)->blk_cksum, 0, 0, 0, 0);	\
 }
@@ -631,7 +652,7 @@ typedef struct blkptr {
 		    (u_longlong_t)BP_GET_LEVEL(bp),			\
 		    type,						\
 		    (u_longlong_t)BP_GET_LSIZE(bp),			\
-		    (u_longlong_t)bp->blk_birth);			\
+		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));		\
 	} else if (BP_IS_EMBEDDED(bp)) {				\
 		len = func(buf + len, size - len,			\
 		    "EMBEDDED [L%llu %s] et=%u %s "			\
@@ -642,14 +663,14 @@ typedef struct blkptr {
 		    compress,						\
 		    (u_longlong_t)BPE_GET_LSIZE(bp),			\
 		    (u_longlong_t)BPE_GET_PSIZE(bp),			\
-		    (u_longlong_t)bp->blk_birth);			\
+		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));		\
 	} else if (BP_IS_REDACTED(bp)) {				\
 		len += func(buf + len, size - len,			\
 		    "REDACTED [L%llu %s] size=%llxL birth=%lluL",	\
 		    (u_longlong_t)BP_GET_LEVEL(bp),			\
 		    type,						\
 		    (u_longlong_t)BP_GET_LSIZE(bp),			\
-		    (u_longlong_t)bp->blk_birth);			\
+		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp));		\
 	} else {							\
 		for (int d = 0; d < BP_GET_NDVAS(bp); d++) {		\
 			const dva_t *dva = &bp->blk_dva[d];		\
@@ -691,8 +712,8 @@ typedef struct blkptr {
 		    ws,							\
 		    (u_longlong_t)BP_GET_LSIZE(bp),			\
 		    (u_longlong_t)BP_GET_PSIZE(bp),			\
-		    (u_longlong_t)bp->blk_birth,			\
-		    (u_longlong_t)BP_PHYSICAL_BIRTH(bp),		\
+		    (u_longlong_t)BP_GET_LOGICAL_BIRTH(bp),		\
+		    (u_longlong_t)BP_GET_BIRTH(bp),			\
 		    (u_longlong_t)BP_GET_FILL(bp),			\
 		    ws,							\
 		    (u_longlong_t)bp->blk_cksum.zc_word[0],		\
@@ -1142,9 +1163,9 @@ extern const char *spa_state_to_name(spa_t *spa);
 /* error handling */
 struct zbookmark_phys;
 extern void spa_log_error(spa_t *spa, const zbookmark_phys_t *zb,
-    const uint64_t *birth);
+    const uint64_t birth);
 extern void spa_remove_error(spa_t *spa, zbookmark_phys_t *zb,
-    const uint64_t *birth);
+    const uint64_t birth);
 extern int zfs_ereport_post(const char *clazz, spa_t *spa, vdev_t *vd,
     const zbookmark_phys_t *zb, zio_t *zio, uint64_t state);
 extern boolean_t zfs_ereport_is_valid(const char *clazz, spa_t *spa, vdev_t *vd,

diff --git a/include/sys/uberblock_impl.h b/include/sys/uberblock_impl.h
@@ -165,7 +165,7 @@ struct uberblock {
 	 * pool from a checkpointed uberblock [see spa_ld_select_uberblock()],
 	 * the value of the field is used to determine which ZIL blocks have
 	 * been allocated according to the ms_sm when we are rewinding to a
-	 * checkpoint. Specifically, if blk_birth > ub_checkpoint_txg, then
+	 * checkpoint. Specifically, if logical birth > ub_checkpoint_txg,then
 	 * the ZIL block is not allocated [see uses of spa_min_claim_txg()].
 	 */
 	uint64_t	ub_checkpoint_txg;

diff --git a/lib/libzdb/libzdb.c b/lib/libzdb/libzdb.c
@@ -93,9 +93,9 @@ livelist_compare(const void *larg, const void *rarg)
 	 * Since we're storing blkptrs without cancelling FREE/ALLOC pairs,
 	 * it's possible the offsets are equal. In that case, sort by txg
 	 */
-	if (l->blk_birth < r->blk_birth) {
+	if (BP_GET_LOGICAL_BIRTH(l) < BP_GET_LOGICAL_BIRTH(r)) {
 		return (-1);
-	} else if (l->blk_birth > r->blk_birth) {
+	} else if (BP_GET_LOGICAL_BIRTH(l) > BP_GET_LOGICAL_BIRTH(r)) {
 		return (+1);
 	}
 	return (0);