scrub: add option to scrub only recent data

Recent data is defined as the last known point in the TXG database,
minus a user-defined time interval (default: 4h).

This feature can be triggered using either of the following commands:
`zpool clean -s` or `zpool scrub -R`.

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

Author: oshogbo

cmd/zpool/zpool_main.c

@@ -458,7 +458,8 @@ get_usage(zpool_help_t idx)
 		return (gettext("\tattach [-fsw] [-o property=value] "
 		    "<pool> <device> <new-device>\n"));
 	case HELP_CLEAR:
-		return (gettext("\tclear [[--power]|[-nF]] <pool> [device]\n"));
+		return (gettext("\tclear [[--power]|[-nsF]] <pool> "
+		    "[device]\n"));
 	case HELP_CREATE:
 		return (gettext("\tcreate [-fnd] [-o property=value] ... \n"
 		    "\t    [-O file-system-property=value] ... \n"
@@ -513,8 +514,8 @@ get_usage(zpool_help_t idx)
 		return (gettext("\tinitialize [-c | -s | -u] [-w] <-a | <pool> "
 		    "[<device> ...]>\n"));
 	case HELP_SCRUB:
-		return (gettext("\tscrub [-e | -s | -p | -C | -E | -S] [-w] "
-		    "<-a | <pool> [<pool> ...]>\n"));
+		return (gettext("\tscrub [-e | -s | -p | -C | -E | -S | -R] "
+		    "[-w] <-a | <pool> [<pool> ...]>\n"));
 	case HELP_RESILVER:
 		return (gettext("\tresilver <pool> ...\n"));
 	case HELP_TRIM:
@@ -8155,8 +8156,74 @@ zpool_do_offline(int argc, char **argv)
 	return (ret);
 }
 
+typedef struct scrub_cbdata {
+	int	cb_type;
+	pool_scrub_cmd_t cb_scrub_cmd;
+	time_t	cb_date_start;
+	time_t	cb_date_end;
+} scrub_cbdata_t;
+
+static boolean_t
+zpool_has_checkpoint(zpool_handle_t *zhp)
+{
+	nvlist_t *config, *nvroot;
+
+	config = zpool_get_config(zhp, NULL);
+
+	if (config != NULL) {
+		pool_checkpoint_stat_t *pcs = NULL;
+		uint_t c;
+
+		nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE);
+		(void) nvlist_lookup_uint64_array(nvroot,
+		    ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t **)&pcs, &c);
+
+		if (pcs == NULL || pcs->pcs_state == CS_NONE)
+			return (B_FALSE);
+
+		assert(pcs->pcs_state == CS_CHECKPOINT_EXISTS ||
+		    pcs->pcs_state == CS_CHECKPOINT_DISCARDING);
+		return (B_TRUE);
+	}
+
+	return (B_FALSE);
+}
+
+static int
+zpool_scrub(zpool_handle_t *zhp, scrub_cbdata_t *cb)
+{
+	int err;
+
+	/*
+	 * Ignore faulted pools.
+	 */
+	if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
+		(void) fprintf(stderr, gettext("cannot scan '%s': pool is "
+		    "currently unavailable\n"), zpool_get_name(zhp));
+		return (1);
+	}
+
+	err = zpool_scan_range(zhp, cb->cb_type, cb->cb_scrub_cmd,
+	    cb->cb_date_start, cb->cb_date_end);
+	if (err == 0 && zpool_has_checkpoint(zhp) &&
+	    cb->cb_type == POOL_SCAN_SCRUB) {
+		(void) printf(gettext("warning: will not scrub state that "
+		    "belongs to the checkpoint of pool '%s'\n"),
+		    zpool_get_name(zhp));
+	}
+
+	return (err != 0);
+}
+
+static int
+scrub_callback(zpool_handle_t *zhp, void *data)
+{
+	scrub_cbdata_t *cb = data;
+	return (zpool_scrub(zhp, cb));
+}
+
 /*
- * zpool clear [-nF]|[--power] <pool> [device]
+ * zpool clear [-nsF]|[--power] <pool> [device]
  *
  * Clear all errors associated with a pool or a particular device.
  */
@@ -8169,6 +8236,7 @@ zpool_do_clear(int argc, char **argv)
 	boolean_t do_rewind = B_FALSE;
 	boolean_t xtreme_rewind = B_FALSE;
 	boolean_t is_power_on = B_FALSE;
+	boolean_t scrub = B_FALSE;
 	uint32_t rewind_policy = ZPOOL_NO_REWIND;
 	nvlist_t *policy = NULL;
 	zpool_handle_t *zhp;
@@ -8180,7 +8248,7 @@ zpool_do_clear(int argc, char **argv)
 	};
 
 	/* check options */
-	while ((c = getopt_long(argc, argv, "FnX", long_options,
+	while ((c = getopt_long(argc, argv, "FnsX", long_options,
 	    NULL)) != -1) {
 		switch (c) {
 		case 'F':
@@ -8189,6 +8257,9 @@ zpool_do_clear(int argc, char **argv)
 		case 'n':
 			dryrun = B_TRUE;
 			break;
+		case 's':
+			scrub = B_TRUE;
+			break;
 		case 'X':
 			xtreme_rewind = B_TRUE;
 			break;
@@ -8256,6 +8327,14 @@ zpool_do_clear(int argc, char **argv)
 	if (zpool_clear(zhp, device, policy) != 0)
 		ret = 1;
 
+	if (ret == 0 && !dryrun && scrub) {
+		scrub_cbdata_t cbdata = {
+			.cb_type = POOL_SCAN_SCRUB,
+			.cb_scrub_cmd = POOL_SCRUB_RECENT,
+		};
+		ret = scrub_callback(zhp, &cbdata);
+	}
+
 	zpool_close(zhp);
 
 	nvlist_free(policy);
@@ -8357,66 +8436,6 @@ zpool_do_reopen(int argc, char **argv)
 	return (ret);
 }
 
-typedef struct scrub_cbdata {
-	int	cb_type;
-	pool_scrub_cmd_t cb_scrub_cmd;
-	time_t	cb_date_start;
-	time_t	cb_date_end;
-} scrub_cbdata_t;
-
-static boolean_t
-zpool_has_checkpoint(zpool_handle_t *zhp)
-{
-	nvlist_t *config, *nvroot;
-
-	config = zpool_get_config(zhp, NULL);
-
-	if (config != NULL) {
-		pool_checkpoint_stat_t *pcs = NULL;
-		uint_t c;
-
-		nvroot = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE);
-		(void) nvlist_lookup_uint64_array(nvroot,
-		    ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t **)&pcs, &c);
-
-		if (pcs == NULL || pcs->pcs_state == CS_NONE)
-			return (B_FALSE);
-
-		assert(pcs->pcs_state == CS_CHECKPOINT_EXISTS ||
-		    pcs->pcs_state == CS_CHECKPOINT_DISCARDING);
-		return (B_TRUE);
-	}
-
-	return (B_FALSE);
-}
-
-static int
-scrub_callback(zpool_handle_t *zhp, void *data)
-{
-	scrub_cbdata_t *cb = data;
-	int err;
-
-	/*
-	 * Ignore faulted pools.
-	 */
-	if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
-		(void) fprintf(stderr, gettext("cannot scan '%s': pool is "
-		    "currently unavailable\n"), zpool_get_name(zhp));
-		return (1);
-	}
-
-	err = zpool_scan_range(zhp, cb->cb_type, cb->cb_scrub_cmd,
-	    cb->cb_date_start, cb->cb_date_end);
-	if (err == 0 && zpool_has_checkpoint(zhp) &&
-	    cb->cb_type == POOL_SCAN_SCRUB) {
-		(void) printf(gettext("warning: will not scrub state that "
-		    "belongs to the checkpoint of pool '%s'\n"),
-		    zpool_get_name(zhp));
-	}
-
-	return (err != 0);
-}
-
 static int
 wait_callback(zpool_handle_t *zhp, void *data)
 {
@@ -8452,14 +8471,15 @@ struct zpool_scrub_option {
 };
 
 /*
- * zpool scrub [-e | -s | -p | -C | -E | -S] [-w] [-a | <pool> ...]
+ * zpool scrub [-e | -s | -p | -C | -E | -S | -R] [-w] [-a | <pool> ...]
  *
  *	-a	Scrub all pools.
  *	-e	Only scrub blocks in the error log.
  *	-E	End date of scrub.
  *	-S	Start date of scrub.
  *	-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.
  *	-C	Scrub from last saved txg.
  */
@@ -8478,11 +8498,12 @@ zpool_do_scrub(int argc, char **argv)
 	struct zpool_scrub_option is_error_scrub = {'e', B_FALSE};
 	struct zpool_scrub_option is_pause = {'p', B_FALSE};
 	struct zpool_scrub_option is_stop = {'s', B_FALSE};
+	struct zpool_scrub_option is_recent = {'R', B_FALSE};
 	struct zpool_scrub_option is_txg_continue = {'C', B_FALSE};
 	struct zpool_scrub_option scrub_all = {'a', B_FALSE};
 
 	/* check options */
-	while ((c = getopt(argc, argv, "aspweCE:S:")) != -1) {
+	while ((c = getopt(argc, argv, "aspweCE:S:R")) != -1) {
 		switch (c) {
 		case 'a':
 			scrub_all.enabled = B_TRUE;
@@ -8506,6 +8527,9 @@ zpool_do_scrub(int argc, char **argv)
 		case 'p':
 			is_pause.enabled = B_TRUE;
 			break;
+		case 'R':
+			is_recent.enabled = B_TRUE;
+			break;
 		case 'w':
 			wait.enabled = B_TRUE;
 			break;
@@ -8526,9 +8550,13 @@ zpool_do_scrub(int argc, char **argv)
 		{&is_stop, &is_pause},
 		{&is_stop, &is_txg_continue},
 		{&is_stop, &is_error_scrub},
+		{&is_stop, &is_recent},
 		{&is_pause, &is_txg_continue},
 		{&is_pause, &is_error_scrub},
+		{&is_pause, &is_recent},
 		{&is_error_scrub, &is_txg_continue},
+		{&is_error_scrub, &is_recent},
+		{&is_recent, &is_txg_continue},
 	};
 
 	for (int i = 0; i < sizeof (scrub_exclusive_options) /
@@ -8553,6 +8581,8 @@ zpool_do_scrub(int argc, char **argv)
 		cb.cb_type = POOL_SCAN_NONE;
 	} else if (is_txg_continue.enabled) {
 		cb.cb_scrub_cmd = POOL_SCRUB_FROM_LAST_TXG;
+	} else if (is_recent.enabled) {
+		cb.cb_scrub_cmd = POOL_SCRUB_RECENT;
 	} else {
 		cb.cb_scrub_cmd = POOL_SCRUB_NORMAL;
 	}

include/os/freebsd/spl/sys/mod.h

@@ -83,6 +83,15 @@
 #define	param_set_deadman_ziotime_args(var) \
     CTLTYPE_U64, NULL, 0, param_set_deadman_ziotime, "QU"
 
+#define	scrub_param_set_recent_time_args(var) \
+    CTLTYPE_U64, NULL, 0, scrub_param_set_recent_time, "QU"
+
+#define	scrub_param_set_recent_time_hours_args(var) \
+    CTLTYPE_U64, NULL, 0, scrub_param_set_recent_time_hours, "QU"
+
+#define	scrub_param_set_recent_time_days_args(var) \
+    CTLTYPE_U64, NULL, 0, scrub_param_set_recent_time_days, "QU"
+
 #define	param_set_multihost_interval_args(var) \
     CTLTYPE_U64, NULL, 0, param_set_multihost_interval, "QU"
 

include/sys/dsl_scan.h

@@ -45,6 +45,9 @@ struct dsl_pool;
 struct dmu_tx;
 
 extern int zfs_scan_suspend_progress;
+extern uint64_t zfs_scrub_recent_time;
+extern uint64_t zfs_scrub_recent_time_hours;
+extern uint64_t zfs_scrub_recent_time_days;
 
 /*
  * All members of this structure must be uint64_t, for byteswap
@@ -222,6 +225,10 @@ void dsl_scan_freed(spa_t *spa, const blkptr_t *bp);
 void dsl_scan_io_queue_destroy(dsl_scan_io_queue_t *queue);
 void dsl_scan_io_queue_vdev_xfer(vdev_t *svd, vdev_t *tvd);
 
+int scrub_param_set_recent_time(ZFS_MODULE_PARAM_ARGS);
+int scrub_param_set_recent_time_hours(ZFS_MODULE_PARAM_ARGS);
+int scrub_param_set_recent_time_days(ZFS_MODULE_PARAM_ARGS);
+
 #ifdef	__cplusplus
 }
 #endif

include/sys/fs/zfs.h

@@ -1098,6 +1098,7 @@ typedef enum pool_scrub_cmd {
 	POOL_SCRUB_NORMAL = 0,
 	POOL_SCRUB_PAUSE,
 	POOL_SCRUB_FROM_LAST_TXG,
+	POOL_SCRUB_RECENT,
 	POOL_SCRUB_FLAGS_END
 } pool_scrub_cmd_t;
 

include/zfs_crrd.h

@@ -71,5 +71,6 @@ void rrd_add(rrd_t *rrd, hrtime_t time, uint64_t txg);
 
 void dbrrd_add(dbrrd_t *db, hrtime_t time, uint64_t txg);
 uint64_t dbrrd_query(dbrrd_t *r, hrtime_t tv, dbrrd_rounding_t rouding);
+hrtime_t dbrrd_tail_time(dbrrd_t *r);
 
 #endif

man/man4/zfs.4

@@ -1967,6 +1967,24 @@ simply doing a metadata crawl of the pool instead.
 .It Sy zfs_no_scrub_prefetch Ns = Ns Sy 0 Ns | Ns 1 Pq int
 Set to disable block prefetching for scrubs.
 .
+.It Sy zfs_scrub_recent_time Ns = Ns Sy 14400 Pq uint
+.It Sy zfs_scrub_recent_time_hours Ns = Ns Sy 4 Pq uint
+.It Sy zfs_scrub_recent_time_days Ns = Ns Sy 0 Pq uint
+Set the time interval that defines which data is considered most recent.
+The interval is measured relative to the last known TXG in the TXG database.
+.Pp
+This setting can be specified using one of three properties: seconds, hours,
+or days.
+These properties are provided for user convenience.
+Each represents the same value, expressed in different units of time.
+For example, setting the
+.Sy zfs_scrub_recent_time_hours
+property to 1 will automatically set the
+.Sy zfs_scrub_recent_time
+to 3600 and
+.Sy zfs_scrub_recent_time_days
+to 0.
+.
 .It Sy zfs_nocacheflush Ns = Ns Sy 0 Ns | Ns 1 Pq int
 Disable cache flush operations on disks when writing.
 Setting this will cause pool corruption on power loss