scrub: add option to scrub only recent data

Sponsored-By: Wasabi Technology, Inc.
Sponsored-By: Klara Inc.
Signed-off-by: Mariusz Zaborski <mariusz.zaborski@klarasystems.com>

cmd/zpool/zpool_main.c

@@ -506,7 +506,7 @@ get_usage(zpool_help_t idx)
 		return (gettext("\tinitialize [-c | -s | -u] [-w] <pool> "
 		    "[<device> ...]\n"));
 	case HELP_SCRUB:
-		return (gettext("\tscrub [-s | -p] [-w] [-e] [-C] "
+		return (gettext("\tscrub [-s | -p] [-w] [-e] [-C] [-R] "
 		    "<pool> ...\n"));
 	case HELP_RESILVER:
 		return (gettext("\tresilver <pool> ...\n"));
@@ -8398,6 +8398,7 @@ wait_callback(zpool_handle_t *zhp, void *data)
  *	-e	Only scrub blocks in the error log.
  *	-s	Stop.  Stops any in-progress scrub.
  *	-p	Pause. Pause in-progress scrub.
+ *	-R	Scrub only recent data.
  *	-w	Wait.  Blocks until scrub has completed.
  */
 int
@@ -8415,9 +8416,10 @@ zpool_do_scrub(int argc, char **argv)
 	boolean_t is_pause = B_FALSE;
 	boolean_t is_stop = B_FALSE;
 	boolean_t is_txg_continue = B_FALSE;
+	boolean_t is_recent_scrub = B_FALSE;
 
 	/* check options */
-	while ((c = getopt(argc, argv, "spweC")) != -1) {
+	while ((c = getopt(argc, argv, "spweCR")) != -1) {
 		switch (c) {
 		case 'C':
 			is_txg_continue = B_TRUE;
@@ -8431,6 +8433,9 @@ zpool_do_scrub(int argc, char **argv)
 		case 'p':
 			is_pause = B_TRUE;
 			break;
+		case 'R':
+			is_recent_scrub = B_TRUE;
+			break;
 		case 'w':
 			wait = B_TRUE;
 			break;
@@ -8457,11 +8462,26 @@ zpool_do_scrub(int argc, char **argv)
 		(void) fprintf(stderr, gettext("invalid option "
 		    "combination :-e and -C are mutually exclusive\n"));
 		usage(B_FALSE);
+	} else if (is_pause && is_recent_scrub) {
+		(void) fprintf(stderr, gettext("invalid option "
+		    "combination :-p and -R are mutually exclusive\n"));
+		usage(B_FALSE);
+	} else if (is_stop && is_recent_scrub) {
+		(void) fprintf(stderr, gettext("invalid option "
+		    "combination :-s and -R are mutually exclusive\n"));
+		usage(B_FALSE);
+	} else if (is_error_scrub && is_recent_scrub) {
+		(void) fprintf(stderr, gettext("invalid option "
+		    "combination :-e and -R are mutually exclusive\n"));
+		usage(B_FALSE);
 	} else {
-		if (is_error_scrub)
+		if (is_error_scrub) {
 			cb.cb_type = POOL_SCAN_ERRORSCRUB;
+		}
 
-		if (is_pause) {
+		if (is_recent_scrub) {
+			cb.cb_scrub_cmd = POOL_SCRUB_RECENT_TXGS;
+		} else if (is_pause) {
 			cb.cb_scrub_cmd = POOL_SCRUB_PAUSE;
 		} else if (is_stop) {
 			cb.cb_type = POOL_SCAN_NONE;

include/sys/dsl_scan.h

@@ -44,6 +44,7 @@ struct dsl_pool;
 struct dmu_tx;
 
 extern int zfs_scan_suspend_progress;
+extern uint_t zfs_scrub_recent_txgs;
 
 /*
  * All members of this structure must be uint64_t, for byteswap

include/sys/fs/zfs.h

@@ -1077,6 +1077,7 @@ typedef enum pool_scrub_cmd {
 	POOL_SCRUB_NORMAL = 0,
 	POOL_SCRUB_PAUSE,
 	POOL_SCRUB_FROM_LAST_TXG,
+	POOL_SCRUB_RECENT_TXGS,
 	POOL_SCRUB_FLAGS_END
 } pool_scrub_cmd_t;
 

man/man4/zfs.4

@@ -1986,6 +1986,9 @@ working on a scrub between TXG flushes.
 .It Sy zfs_scrub_error_blocks_per_txg Ns = Ns Sy 4096 Pq uint
 Error blocks to be scrubbed in one txg.
 .
+.It Sy zfs_scrub_recent_txgs Ns = Ns Sy 256 Pq uint
+Number of txgs to be considered as recent when performing a recent data scrub.
+.
 .It Sy zfs_scan_checkpoint_intval Ns = Ns Sy 7200 Ns s Po 2 hour Pc Pq uint
 To preserve progress across reboots, the sequential scan algorithm periodically
 needs to stop metadata scanning and issue all the verification I/O to disk.

man/man8/zpool-scrub.8

@@ -40,6 +40,7 @@
 .Op Fl w
 .Op Fl e
 .Op Fl C
+.Op Fl R
 .Ar pool Ns …
 .
 .Sh DESCRIPTION
@@ -119,6 +120,10 @@ resilvering, nor can it be run when a regular scrub is paused.
 Continue scrub from last saved txg (see zpool
 .Sy last_scrubbed_txg
 property).
+.It Fl R
+Scrub only recent data (this can be controlled by the
+.Sy zfs_scrub_recent_txgs
+parameter).
 .El
 .Sh EXAMPLES
 .Ss Example 1

module/zfs/dsl_scan.c

@@ -239,6 +239,9 @@ static int zfs_free_bpobj_enabled = 1;
 /* Error blocks to be scrubbed in one txg. */
 static uint_t zfs_scrub_error_blocks_per_txg = 1 << 12;
 
+/* The number of TXGs should be scrubbed while scrubbing recent data. */
+uint_t zfs_scrub_recent_txgs = 256;
+
 /* the order has to match pool_scan_type */
 static scan_cb_t *scan_funcs[POOL_SCAN_FUNCS] = {
 	NULL,
@@ -5323,4 +5326,7 @@ ZFS_MODULE_PARAM(zfs, zfs_, resilver_disable_defer, INT, ZMOD_RW,
 
 ZFS_MODULE_PARAM(zfs, zfs_, scrub_error_blocks_per_txg, UINT, ZMOD_RW,
 	"Error blocks to be scrubbed in one txg");
+
+ZFS_MODULE_PARAM(zfs, zfs_, scrub_recent_txgs, UINT, ZMOD_RW,
+	"The number of TXGs should be scrubbed while scrubbing recent data");
 /* END CSTYLED */

module/zfs/zfs_ioctl.c

@@ -1722,6 +1722,16 @@ zfs_ioc_pool_scrub(const char *poolname, nvlist_t *innvl, nvlist_t *outnvl)
 	} else if (scan_cmd == POOL_SCRUB_FROM_LAST_TXG) {
 		error = spa_scan_range(spa, scan_type,
 		    spa_get_last_scrubbed_txg(spa), 0);
+	} else if (scan_cmd == POOL_SCRUB_RECENT_TXGS) {
+		uint64_t start;
+
+		start = 0;
+		if (spa_last_synced_txg(spa) > zfs_scrub_recent_txgs) {
+			start = spa_last_synced_txg(spa) -
+			    zfs_scrub_recent_txgs;
+		}
+
+		error = spa_scan_range(spa, scan_type, start, 0);
 	} else {
 		error = spa_scan(spa, scan_type);
 	}