scrub: add option to scrub only recent data

Sponsored-By: Wasabi Technology, Inc.
Sponsored-By: Klara Inc.
Signed-off-by: Mariusz Zaborski <[email protected]>

Author: oshogbo

cmd/zpool/zpool_main.c

@@ -512,7 +512,7 @@ get_usage(zpool_help_t idx)
 		return (gettext("\tinitialize [-c | -s | -u] [-w] <pool> "
 		    "[<device> ...]\n"));
 	case HELP_SCRUB:
-		return (gettext("\tscrub [-e | -s | -p | -C] [-w] "
+		return (gettext("\tscrub [-e | -s | -p | -C | -R] [-w] "
 		    "<pool> ...\n"));
 	case HELP_RESILVER:
 		return (gettext("\tresilver <pool> ...\n"));
@@ -8413,13 +8413,14 @@ wait_callback(zpool_handle_t *zhp, void *data)
 }
 
 /*
- * zpool scrub [-e | -s | -p | -C] [-w] <pool> ...
+ * zpool scrub [-e | -s | -p | -C | -R] [-w] <pool> ...
  *
  *	-e	Only scrub blocks in the error log.
  *	-s	Stop.  Stops any in-progress scrub.
  *	-p	Pause. Pause in-progress scrub.
  *	-w	Wait.  Blocks until scrub has completed.
  *	-C	Scrub from last saved txg.
+ *	-R	Scrub only recent data.
  */
 int
 zpool_do_scrub(int argc, char **argv)
@@ -8436,9 +8437,10 @@ 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 'e':
 			is_error_scrub = B_TRUE;
@@ -8455,6 +8457,9 @@ zpool_do_scrub(int argc, char **argv)
 		case 'C':
 			is_txg_continue = B_TRUE;
 			break;
+		case 'R':
+			is_recent_scrub = B_TRUE;
+			break;
 		case '?':
 			(void) fprintf(stderr, gettext("invalid option '%c'\n"),
 			    optopt);
@@ -8478,11 +8483,30 @@ 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_recent_scrub && is_txg_continue) {
+		(void) fprintf(stderr, gettext("invalid option "
+		    "combination :-R 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

@@ -1090,6 +1090,7 @@ 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;
 

lib/libzfs/libzfs.abi

@@ -3132,7 +3132,8 @@
       <enumerator name='ZPOOL_PROP_DEDUP_TABLE_SIZE' value='36'/>
       <enumerator name='ZPOOL_PROP_DEDUP_TABLE_QUOTA' value='37'/>
       <enumerator name='ZPOOL_PROP_DEDUPCACHED' value='38'/>
-      <enumerator name='ZPOOL_NUM_PROPS' value='39'/>
+      <enumerator name='ZPOOL_PROP_LAST_SCRUBBED_TXG' value='39'/>
+      <enumerator name='ZPOOL_NUM_PROPS' value='40'/>
     </enum-decl>
     <typedef-decl name='zpool_prop_t' type-id='af1ba157' id='5d0c23fb'/>
     <typedef-decl name='regoff_t' type-id='95e97e5e' id='54a2a2a8'/>
@@ -5985,7 +5986,8 @@
       <enumerator name='POOL_SCRUB_NORMAL' value='0'/>
       <enumerator name='POOL_SCRUB_PAUSE' value='1'/>
       <enumerator name='POOL_SCRUB_FROM_LAST_TXG' value='2'/>
-      <enumerator name='POOL_SCRUB_FLAGS_END' value='3'/>
+      <enumerator name='POOL_SCRUB_RECENT_TXGS' value='3'/>
+      <enumerator name='POOL_SCRUB_FLAGS_END' value='4'/>
     </enum-decl>
     <typedef-decl name='pool_scrub_cmd_t' type-id='a1474cbd' id='b51cf3c2'/>
     <enum-decl name='zpool_errata' id='d9abbf54'>

lib/libzfsbootenv/libzfsbootenv.abi

@@ -1,6 +1,6 @@
 <abi-corpus version='2.0' architecture='elf-amd-x86_64' soname='libzfsbootenv.so.1'>
   <elf-needed>
-    <dependency name='libzfs.so.4'/>
+    <dependency name='libzfs.so.6'/>
     <dependency name='libnvpair.so.3'/>
     <dependency name='libc.so.6'/>
   </elf-needed>

man/man4/zfs.4

@@ -2044,6 +2044,9 @@ 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

@@ -36,7 +36,7 @@
 .Sh SYNOPSIS
 .Nm zpool
 .Cm scrub
-.Op Ns Fl e | Ns Fl p | Fl s Ns | Fl C Ns
+.Op Ns Fl e | Ns Fl p | Fl s Ns | Fl C Ns | Fl R Ns
 .Op Fl w
 .Ar pool Ns …
 .
@@ -117,6 +117,10 @@ 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

@@ -242,6 +242,9 @@ 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,
@@ -5349,4 +5352,7 @@ ZFS_MODULE_PARAM(zfs, zfs_, resilver_defer_percent, UINT, 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

@@ -1721,6 +1721,16 @@ 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);
 	}