Redesign scan/scrub interfact to allow scrubing range of TXGs

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

include/sys/dsl_scan.h

@@ -189,7 +189,8 @@ void dsl_scan_setup_sync(void *, dmu_tx_t *);
 void dsl_scan_fini(struct dsl_pool *dp);
 void dsl_scan_sync(struct dsl_pool *, dmu_tx_t *);
 int dsl_scan_cancel(struct dsl_pool *);
-int dsl_scan(struct dsl_pool *, pool_scan_func_t);
+int dsl_scan(struct dsl_pool *, pool_scan_func_t, uint64_t starttxg,
+    uint64_t txgend);
 void dsl_scan_assess_vdev(struct dsl_pool *dp, vdev_t *vd);
 boolean_t dsl_scan_scrubbing(const struct dsl_pool *dp);
 boolean_t dsl_errorscrubbing(const struct dsl_pool *dp);

include/sys/spa.h

@@ -821,6 +821,8 @@ extern void spa_l2cache_drop(spa_t *spa);
 
 /* scanning */
 extern int spa_scan(spa_t *spa, pool_scan_func_t func);
+extern int spa_scan_range(spa_t *spa, pool_scan_func_t func, uint64_t txgstart,
+    uint64_t txgend);
 extern int spa_scan_stop(spa_t *spa);
 extern int spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t flag);
 

module/zfs/dsl_scan.c

@@ -848,18 +848,24 @@ dsl_scan_setup_check(void *arg, dmu_tx_t *tx)
 	return (0);
 }
 
+typedef struct {
+	pool_scan_func_t func;
+	uint64_t	 txgstart;
+	uint64_t	 txgend;
+} setup_sync_arg_t;
+
 void
 dsl_scan_setup_sync(void *arg, dmu_tx_t *tx)
 {
-	(void) arg;
+	setup_sync_arg_t *setup_sync_arg = (setup_sync_arg_t *)arg;
 	dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
-	pool_scan_func_t *funcp = arg;
 	dmu_object_type_t ot = 0;
 	dsl_pool_t *dp = scn->scn_dp;
 	spa_t *spa = dp->dp_spa;
 
 	ASSERT(!dsl_scan_is_running(scn));
-	ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS);
+	ASSERT(setup_sync_arg->func > POOL_SCAN_NONE &&
+	    setup_sync_arg->func < POOL_SCAN_FUNCS);
 	memset(&scn->scn_phys, 0, sizeof (scn->scn_phys));
 
 	/*
@@ -869,10 +875,14 @@ dsl_scan_setup_sync(void *arg, dmu_tx_t *tx)
 	memset(&scn->errorscrub_phys, 0, sizeof (scn->errorscrub_phys));
 	dsl_errorscrub_sync_state(scn, tx);
 
-	scn->scn_phys.scn_func = *funcp;
+	scn->scn_phys.scn_func = setup_sync_arg->func;
 	scn->scn_phys.scn_state = DSS_SCANNING;
-	scn->scn_phys.scn_min_txg = 0;
+	scn->scn_phys.scn_min_txg = setup_sync_arg->txgstart;
-	scn->scn_phys.scn_max_txg = tx->tx_txg;
+	if (setup_sync_arg->txgend == 0) {
+		scn->scn_phys.scn_max_txg = tx->tx_txg;
+	} else {
+		scn->scn_phys.scn_max_txg = setup_sync_arg->txgend;
+	}
 	scn->scn_phys.scn_ddt_class_max = DDT_CLASSES - 1; /* the entire DDT */
 	scn->scn_phys.scn_start_time = gethrestime_sec();
 	scn->scn_phys.scn_errors = 0;
@@ -959,7 +969,7 @@ dsl_scan_setup_sync(void *arg, dmu_tx_t *tx)
 
 	spa_history_log_internal(spa, "scan setup", tx,
 	    "func=%u mintxg=%llu maxtxg=%llu",
-	    *funcp, (u_longlong_t)scn->scn_phys.scn_min_txg,
+	    setup_sync_arg->func, (u_longlong_t)scn->scn_phys.scn_min_txg,
 	    (u_longlong_t)scn->scn_phys.scn_max_txg);
 }
 
@@ -969,10 +979,16 @@ dsl_scan_setup_sync(void *arg, dmu_tx_t *tx)
  * error scrub.
  */
 int
-dsl_scan(dsl_pool_t *dp, pool_scan_func_t func)
+dsl_scan(dsl_pool_t *dp, pool_scan_func_t func, uint64_t txgstart,
+    uint64_t txgend)
 {
 	spa_t *spa = dp->dp_spa;
 	dsl_scan_t *scn = dp->dp_scan;
+	setup_sync_arg_t setup_sync_arg;
+
+	if (func != POOL_SCAN_SCRUB && (txgstart != 0 || txgend != 0)) {
+		return (EINVAL);
+	}
 
 	/*
 	 * Purge all vdev caches and probe all devices.  We do this here
@@ -1023,8 +1039,13 @@ dsl_scan(dsl_pool_t *dp, pool_scan_func_t func)
 		return (SET_ERROR(err));
 	}
 
+	setup_sync_arg.func = func;
+	setup_sync_arg.txgstart = txgstart;
+	setup_sync_arg.txgend = txgend;
+
 	return (dsl_sync_task(spa_name(spa), dsl_scan_setup_check,
-	    dsl_scan_setup_sync, &func, 0, ZFS_SPACE_CHECK_EXTRA_RESERVED));
+	    dsl_scan_setup_sync, &setup_sync_arg, 0,
+	    ZFS_SPACE_CHECK_EXTRA_RESERVED));
 }
 
 static void
@@ -4301,13 +4322,16 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx)
 	 */
 	if (dsl_scan_restarting(scn, tx) ||
 	    (spa->spa_resilver_deferred && zfs_resilver_disable_defer)) {
-		pool_scan_func_t func = POOL_SCAN_SCRUB;
+		setup_sync_arg_t setup_sync_arg = {
+			.func = POOL_SCAN_SCRUB,
+		};
 		dsl_scan_done(scn, B_FALSE, tx);
 		if (vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
-			func = POOL_SCAN_RESILVER;
+			setup_sync_arg.func = POOL_SCAN_RESILVER;
 		zfs_dbgmsg("restarting scan func=%u on %s txg=%llu",
-		    func, dp->dp_spa->spa_name, (longlong_t)tx->tx_txg);
+		    setup_sync_arg.func, dp->dp_spa->spa_name,
-		dsl_scan_setup_sync(&func, tx);
+		    (longlong_t)tx->tx_txg);
+		dsl_scan_setup_sync(&setup_sync_arg, tx);
 	}
 
 	/*

module/zfs/spa.c

@@ -8877,6 +8877,13 @@ spa_scan_stop(spa_t *spa)
 
 int
 spa_scan(spa_t *spa, pool_scan_func_t func)
+{
+	return (spa_scan_range(spa, func, 0, 0));
+}
+
+int
+spa_scan_range(spa_t *spa, pool_scan_func_t func, uint64_t txgstart,
+    uint64_t txgend)
 {
 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
 
@@ -8887,6 +8894,9 @@ spa_scan(spa_t *spa, pool_scan_func_t func)
 	    !spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER))
 		return (SET_ERROR(ENOTSUP));
 
+	if (func != POOL_SCAN_SCRUB && (txgstart != 0 || txgend != 0))
+		return (SET_ERROR(ENOTSUP));
+
 	/*
 	 * If a resilver was requested, but there is no DTL on a
 	 * writeable leaf device, we have nothing to do.
@@ -8901,7 +8911,7 @@ spa_scan(spa_t *spa, pool_scan_func_t func)
 	    !spa_feature_is_enabled(spa, SPA_FEATURE_HEAD_ERRLOG))
 		return (SET_ERROR(ENOTSUP));
 
-	return (dsl_scan(spa->spa_dsl_pool, func));
+	return (dsl_scan(spa->spa_dsl_pool, func, txgstart, txgend));
 }
 
 /*
@@ -10982,6 +10992,7 @@ EXPORT_SYMBOL(spa_l2cache_drop);
 
 /* scanning */
 EXPORT_SYMBOL(spa_scan);
+EXPORT_SYMBOL(spa_scan_range);
 EXPORT_SYMBOL(spa_scan_stop);
 
 /* spa syncing */