Teach zpool scrub to scrub only blocks in error log

Added a flag '-e' in zpool scrub to scrub only blocks in error log. A
user can pause, resume and cancel the error scrub by passing additional
command line arguments -p -s just like a regular scrub. This involves
adding a new flag, creating new libzfs interfaces, a new ioctl, and the
actual iteration and read-issuing logic. Error scrubbing is executed in
multiple txg to make sure pool performance is not affected.

Co-authored-by: TulsiJain tulsi.jain@delphix.com
Signed-off-by: George Amanakis <gamanakis@gmail.com>

cmd/zpool/zpool_main.c

@@ -401,7 +401,7 @@ get_usage(zpool_help_t idx)
 		return (gettext("\tinitialize [-c | -s] [-w] <pool> "
 		    "[<device> ...]\n"));
 	case HELP_SCRUB:
-		return (gettext("\tscrub [-s | -p] [-w] <pool> ...\n"));
+		return (gettext("\tscrub [-s | -p] [-w] [-e] <pool> ...\n"));
 	case HELP_RESILVER:
 		return (gettext("\tresilver <pool> ...\n"));
 	case HELP_TRIM:
@@ -7297,8 +7297,9 @@ wait_callback(zpool_handle_t *zhp, void *data)
 }
 
 /*
- * zpool scrub [-s | -p] [-w] <pool> ...
+ * zpool scrub [-s | -p] [-w] [-e] <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.
@@ -7314,14 +7315,21 @@ zpool_do_scrub(int argc, char **argv)
 	cb.cb_type = POOL_SCAN_SCRUB;
 	cb.cb_scrub_cmd = POOL_SCRUB_NORMAL;
 
+	boolean_t is_error_scrub = B_FALSE;
+	boolean_t is_pause = B_FALSE;
+	boolean_t is_stop = B_FALSE;
+
 	/* check options */
-	while ((c = getopt(argc, argv, "spw")) != -1) {
+	while ((c = getopt(argc, argv, "spwe")) != -1) {
 		switch (c) {
+		case 'e':
+			is_error_scrub = B_TRUE;
+			break;
 		case 's':
-			cb.cb_type = POOL_SCAN_NONE;
+			is_stop = B_TRUE;
 			break;
 		case 'p':
-			cb.cb_scrub_cmd = POOL_SCRUB_PAUSE;
+			is_pause = B_TRUE;
 			break;
 		case 'w':
 			wait = B_TRUE;
@@ -7333,11 +7341,21 @@ zpool_do_scrub(int argc, char **argv)
 		}
 	}
 
-	if (cb.cb_type == POOL_SCAN_NONE &&
-	    cb.cb_scrub_cmd == POOL_SCRUB_PAUSE) {
-		(void) fprintf(stderr, gettext("invalid option combination: "
-		    "-s and -p are mutually exclusive\n"));
+	if (is_pause && is_stop) {
+		(void) fprintf(stderr, gettext("invalid option "
+		    "combination :-s and -p are mutually exclusive\n"));
 		usage(B_FALSE);
+	} else {
+		if (is_error_scrub)
+			cb.cb_type = POOL_SCAN_ERRORSCRUB;
+
+		if (is_pause) {
+			cb.cb_scrub_cmd = POOL_SCRUB_PAUSE;
+		} else if (is_stop) {
+			cb.cb_type = POOL_SCAN_NONE;
+		} else {
+			cb.cb_scrub_cmd = POOL_SCRUB_NORMAL;
+		}
 	}
 
 	if (wait && (cb.cb_type == POOL_SCAN_NONE ||
@@ -7561,6 +7579,70 @@ secs_to_dhms(uint64_t total, char *buf)
 	}
 }
 
+/*
+ * Print out detailed error scrub status.
+ */
+static void
+print_err_scrub_status(pool_scan_stat_t *ps)
+{
+	time_t start, end, pause;
+	uint64_t total_secs_left;
+	uint64_t secs_left, mins_left, hours_left, days_left;
+	uint64_t examined, to_be_examined;
+
+	if (ps == NULL || ps->pss_error_scrub_func != POOL_SCAN_ERRORSCRUB) {
+		return;
+	}
+
+	(void) printf(gettext(" scrub: "));
+
+	start = ps->pss_error_scrub_start;
+	end = ps->pss_error_scrub_end;
+	pause = ps->pss_pass_error_scrub_pause;
+	examined = ps->pss_error_scrub_examined;
+	to_be_examined = ps->pss_error_scrub_to_be_examined;
+
+	assert(ps->pss_error_scrub_func == POOL_SCAN_ERRORSCRUB);
+
+	if (ps->pss_error_scrub_state == DSS_FINISHED) {
+		total_secs_left = end - start;
+		days_left = total_secs_left / 60 / 60 / 24;
+		hours_left = (total_secs_left / 60 / 60) % 24;
+		mins_left = (total_secs_left / 60) % 60;
+		secs_left = (total_secs_left % 60);
+
+		(void) printf(gettext("scrubbed %llu error blocks in %llu days "
+		    "%02llu:%02llu:%02llu on %s"), (u_longlong_t)examined,
+		    (u_longlong_t)days_left, (u_longlong_t)hours_left,
+		    (u_longlong_t)mins_left, (u_longlong_t)secs_left,
+		    ctime(&end));
+
+		return;
+	} else if (ps->pss_error_scrub_state == DSS_CANCELED) {
+		(void) printf(gettext("error scrub canceled on %s"),
+		    ctime(&end));
+		return;
+	}
+	assert(ps->pss_error_scrub_state == DSS_ERRORSCRUBBING);
+
+	/* Error scrub is in progress. */
+	if (pause == 0) {
+		(void) printf(gettext("error scrub in progress since %s"),
+		    ctime(&start));
+	} else {
+		(void) printf(gettext("error scrub paused since %s"),
+		    ctime(&pause));
+		(void) printf(gettext("\terror scrub started on %s"),
+		    ctime(&start));
+	}
+
+	double fraction_done = (double)examined / (to_be_examined + examined);
+	(void) printf(gettext("\t%.2f%% done, issued I/O for %llu error"
+	    " blocks"), 100 * fraction_done, (u_longlong_t)examined);
+
+	(void) printf("\n");
+}
+
 /*
  * Print out detailed scrub status.
  */
@@ -7897,10 +7979,12 @@ print_scan_status(zpool_handle_t *zhp, nvlist_t *nvroot)
 {
 	uint64_t rebuild_end_time = 0, resilver_end_time = 0;
 	boolean_t have_resilver = B_FALSE, have_scrub = B_FALSE;
+	boolean_t have_errorscrub = B_FALSE;
 	boolean_t active_resilver = B_FALSE;
 	pool_checkpoint_stat_t *pcs = NULL;
 	pool_scan_stat_t *ps = NULL;
 	uint_t c;
+	time_t scrub_start = 0, errorscrub_start = 0;
 
 	if (nvlist_lookup_uint64_array(nvroot, ZPOOL_CONFIG_SCAN_STATS,
 	    (uint64_t **)&ps, &c) == 0) {
@@ -7909,16 +7993,23 @@ print_scan_status(zpool_handle_t *zhp, nvlist_t *nvroot)
 			active_resilver = (ps->pss_state == DSS_SCANNING);
 		}
 
+
 		have_resilver = (ps->pss_func == POOL_SCAN_RESILVER);
 		have_scrub = (ps->pss_func == POOL_SCAN_SCRUB);
+		scrub_start = ps->pss_start_time;
+		have_errorscrub = (ps->pss_error_scrub_func ==
+		    POOL_SCAN_ERRORSCRUB);
+		errorscrub_start = ps->pss_error_scrub_start;
 	}
 
 	boolean_t active_rebuild = check_rebuilding(nvroot, &rebuild_end_time);
 	boolean_t have_rebuild = (active_rebuild || (rebuild_end_time > 0));
 
 	/* Always print the scrub status when available. */
-	if (have_scrub)
+	if (have_scrub && scrub_start > errorscrub_start)
 		print_scan_scrub_resilver_status(ps);
+	else if (have_errorscrub && errorscrub_start >= scrub_start)
+		print_err_scrub_status(ps);
 
 	/*
 	 * When there is an active resilver or rebuild print its status.

include/libzfs.h

@@ -125,11 +125,14 @@ typedef enum zfs_error {
 	EZFS_THREADCREATEFAILED, /* thread create failed */
 	EZFS_POSTSPLIT_ONLINE,	/* onlining a disk after splitting it */
 	EZFS_SCRUBBING,		/* currently scrubbing */
+	EZFS_ERRORSCRUBBING,	/* currently error scrubbing */
+	EZFS_ERRORSCRUB_PAUSED,	/* error scrub currently paused */
 	EZFS_NO_SCRUB,		/* no active scrub */
 	EZFS_DIFF,		/* general failure of zfs diff */
 	EZFS_DIFFDATA,		/* bad zfs diff data */
 	EZFS_POOLREADONLY,	/* pool is in read-only mode */
 	EZFS_SCRUB_PAUSED,	/* scrub currently paused */
+	EZFS_SCRUB_PAUSED_TO_CANCEL,	/* scrub currently paused */
 	EZFS_ACTIVE_POOL,	/* pool is imported on a different system */
 	EZFS_CRYPTOFAILED,	/* failed to setup encryption */
 	EZFS_NO_PENDING,	/* cannot cancel, no operation is pending */

include/libzfs_core.h

@@ -155,6 +155,8 @@ _LIBZFS_CORE_H int lzc_get_bootenv(const char *, nvlist_t **);
 _LIBZFS_CORE_H int lzc_get_vdev_prop(const char *, nvlist_t *, nvlist_t **);
 _LIBZFS_CORE_H int lzc_set_vdev_prop(const char *, nvlist_t *, nvlist_t **);
 
+_LIBZFS_CORE_H int lzc_scrub(zfs_ioc_t, const char *, nvlist_t *, nvlist_t **);
+
 #ifdef	__cplusplus
 }
 #endif

include/sys/dmu.h

@@ -378,6 +378,7 @@ typedef struct dmu_buf {
 #define	DMU_POOL_DDT_STATS		"DDT-statistics"
 #define	DMU_POOL_CREATION_VERSION	"creation_version"
 #define	DMU_POOL_SCAN			"scan"
+#define	DMU_POOL_ERRORSCRUB		"error_scrub"
 #define	DMU_POOL_FREE_BPOBJ		"free_bpobj"
 #define	DMU_POOL_BPTREE_OBJ		"bptree_obj"
 #define	DMU_POOL_EMPTY_BPOBJ		"empty_bpobj"

include/sys/dsl_scan.h

@@ -29,6 +29,7 @@
 
 #include <sys/zfs_context.h>
 #include <sys/zio.h>
+#include <sys/zap.h>
 #include <sys/ddt.h>
 #include <sys/bplist.h>
 
@@ -78,6 +79,21 @@ typedef enum dsl_scan_flags {
 
 #define	DSL_SCAN_FLAGS_MASK (DSF_VISIT_DS_AGAIN)
 
+typedef struct dsl_errorscrub_phys {
+	uint64_t dep_func; /* pool_scan_func_t */
+	uint64_t dep_state; /* dsl_scan_state_t */
+	uint64_t dep_cursor; /* serialized zap cursor for tracing progress */
+	uint64_t dep_start_time; /* error scrub start time, unix timestamp */
+	uint64_t dep_end_time; /* error scrub end time, unix timestamp */
+	uint64_t dep_to_examine; /* total error blocks to be scrubbed */
+	uint64_t dep_examined; /* blocks scrubbed so far */
+	uint64_t dep_errors;	/* error scrub I/O error count */
+	uint64_t dep_paused_flags; /* flag for paused */
+} dsl_errorscrub_phys_t;
+
+#define	ERRORSCRUB_PHYS_NUMINTS (sizeof (dsl_errorscrub_phys_t) \
+	/ sizeof (uint64_t))
+
 /*
  * Every pool will have one dsl_scan_t and this structure will contain
  * in-memory information about the scan and a pointer to the on-disk
@@ -151,11 +167,15 @@ typedef struct dsl_scan {
 	uint64_t scn_avg_zio_size_this_txg;
 	uint64_t scn_zios_this_txg;
 
+	/* zap cursor for tracing error scrub progress */
+	zap_cursor_t errorscrub_cursor;
 	/* members needed for syncing scan status to disk */
 	dsl_scan_phys_t scn_phys;	/* on disk representation of scan */
 	dsl_scan_phys_t scn_phys_cached;
 	avl_tree_t scn_queue;		/* queue of datasets to scan */
 	uint64_t scn_queues_pending;	/* outstanding data to issue */
+	/* members needed for syncing error scrub status to disk */
+	dsl_errorscrub_phys_t errorscrub_phys;
 } dsl_scan_t;
 
 typedef struct dsl_scan_io_queue dsl_scan_io_queue_t;
@@ -171,8 +191,12 @@ int dsl_scan_cancel(struct dsl_pool *);
 int dsl_scan(struct dsl_pool *, pool_scan_func_t);
 void dsl_scan_assess_vdev(struct dsl_pool *dp, vdev_t *vd);
 boolean_t dsl_scan_scrubbing(const struct dsl_pool *dp);
-int dsl_scrub_set_pause_resume(const struct dsl_pool *dp, pool_scrub_cmd_t cmd);
+boolean_t dsl_errorscrubbing(const struct dsl_pool *dp);
+boolean_t dsl_errorscrub_active(dsl_scan_t *scn);
 void dsl_scan_restart_resilver(struct dsl_pool *, uint64_t txg);
+int dsl_scrub_set_pause_resume(const struct dsl_pool *dp,
+    pool_scrub_cmd_t cmd);
+void dsl_errorscrub_sync(struct dsl_pool *, dmu_tx_t *);
 boolean_t dsl_scan_resilvering(struct dsl_pool *dp);
 boolean_t dsl_scan_resilver_scheduled(struct dsl_pool *dp);
 boolean_t dsl_dataset_unstable(struct dsl_dataset *ds);
@@ -184,6 +208,7 @@ void dsl_scan_ds_clone_swapped(struct dsl_dataset *ds1, struct dsl_dataset *ds2,
     struct dmu_tx *tx);
 boolean_t dsl_scan_active(dsl_scan_t *scn);
 boolean_t dsl_scan_is_paused_scrub(const dsl_scan_t *scn);
+boolean_t dsl_errorscrub_is_paused(const dsl_scan_t *scn);
 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);

include/sys/fs/zfs.h

@@ -1036,6 +1036,7 @@ typedef enum pool_scan_func {
 	POOL_SCAN_NONE,
 	POOL_SCAN_SCRUB,
 	POOL_SCAN_RESILVER,
+	POOL_SCAN_ERRORSCRUB,
 	POOL_SCAN_FUNCS
 } pool_scan_func_t;
 
@@ -1099,6 +1100,20 @@ typedef struct pool_scan_stat {
 	uint64_t	pss_pass_scrub_spent_paused;
 	uint64_t	pss_pass_issued; /* issued bytes per scan pass */
 	uint64_t	pss_issued;	/* total bytes checked by scanner */
+
+	/* error scrub values stored on disk */
+	uint64_t	pss_error_scrub_func;	/* pool_scan_func_t */
+	uint64_t	pss_error_scrub_state;	/* dsl_scan_state_t */
+	uint64_t	pss_error_scrub_start;	/* error scrub start time */
+	uint64_t	pss_error_scrub_end;	/* error scrub end time */
+	uint64_t	pss_error_scrub_examined; /* error blocks issued I/O */
+	/* error blocks to be issued I/O */
+	uint64_t	pss_error_scrub_to_be_examined;
+
+	/* error scrub values not stored on disk */
+	/* error scrub pause time in milliseconds */
+	uint64_t	pss_pass_error_scrub_pause;
+
 } pool_scan_stat_t;
 
 typedef struct pool_removal_stat {
@@ -1120,6 +1135,7 @@ typedef enum dsl_scan_state {
 	DSS_SCANNING,
 	DSS_FINISHED,
 	DSS_CANCELED,
+	DSS_ERRORSCRUBBING,
 	DSS_NUM_STATES
 } dsl_scan_state_t;
 
@@ -1359,7 +1375,7 @@ typedef enum {
  */
 typedef enum zfs_ioc {
 	/*
-	 * Core features - 81/128 numbers reserved.
+	 * Core features - 88/128 numbers reserved.
 	 */
 #ifdef __FreeBSD__
 	ZFS_IOC_FIRST =	0,
@@ -1454,6 +1470,7 @@ typedef enum zfs_ioc {
 	ZFS_IOC_WAIT_FS,			/* 0x5a54 */
 	ZFS_IOC_VDEV_GET_PROPS,			/* 0x5a55 */
 	ZFS_IOC_VDEV_SET_PROPS,			/* 0x5a56 */
+	ZFS_IOC_POOL_SCRUB,			/* 0x5a57 */
 
 	/*
 	 * Per-platform (Optional) - 8/128 numbers reserved.

include/sys/spa.h

@@ -1154,6 +1154,7 @@ extern void zfs_post_state_change(spa_t *spa, vdev_t *vd, uint64_t laststate);
 extern void zfs_post_autoreplace(spa_t *spa, vdev_t *vd);
 extern uint64_t spa_approx_errlog_size(spa_t *spa);
 extern int spa_get_errlog(spa_t *spa, void *uaddr, uint64_t *count);
+extern uint64_t spa_get_last_errlog_size(spa_t *spa);
 extern void spa_errlog_rotate(spa_t *spa);
 extern void spa_errlog_drain(spa_t *spa);
 extern void spa_errlog_sync(spa_t *spa, uint64_t txg);
@@ -1164,6 +1165,13 @@ extern void spa_swap_errlog(spa_t *spa, uint64_t new_head_ds,
 extern void sync_error_list(spa_t *spa, avl_tree_t *t, uint64_t *obj,
     dmu_tx_t *tx);
 extern void spa_upgrade_errlog(spa_t *spa, dmu_tx_t *tx);
+extern int find_top_affected_fs(spa_t *spa, uint64_t head_ds,
+    zbookmark_err_phys_t *zep, uint64_t *top_affected_fs);
+extern int find_birth_txg(struct dsl_dataset *ds, zbookmark_err_phys_t *zep,
+    uint64_t *birth_txg);
+extern void zep_to_zb(uint64_t dataset, zbookmark_err_phys_t *zep,
+    zbookmark_phys_t *zb);
+extern void name_to_errphys(char *buf, zbookmark_err_phys_t *zep);
 
 /* vdev cache */
 extern void vdev_cache_stat_init(void);

include/sys/spa_impl.h

@@ -295,6 +295,10 @@ struct spa {
 	uint64_t	spa_scan_pass_exam;	/* examined bytes per pass */
 	uint64_t	spa_scan_pass_issued;	/* issued bytes per pass */
 
+	/* error scrub pause time in milliseconds */
+	uint64_t	spa_scan_pass_errorscrub_pause;
+	/* total error scrub paused time in milliseconds */
+	uint64_t	spa_scan_pass_errorscrub_spent_paused;
 	/*
 	 * We are in the middle of a resilver, and another resilver
 	 * is needed once this one completes. This is set iff any

include/sys/sysevent/eventdefs.h

@@ -123,6 +123,11 @@ extern "C" {
 #define	ESC_ZFS_TRIM_CANCEL		"trim_cancel"
 #define	ESC_ZFS_TRIM_RESUME		"trim_resume"
 #define	ESC_ZFS_TRIM_SUSPEND		"trim_suspend"
+#define	ESC_ZFS_ERRORSCRUB_START	"errorscrub_start"
+#define	ESC_ZFS_ERRORSCRUB_FINISH	"errorscrub_finish"
+#define	ESC_ZFS_ERRORSCRUB_ABORT	"errorscrub_abort"
+#define	ESC_ZFS_ERRORSCRUB_RESUME	"errorscrub_resume"
+#define	ESC_ZFS_ERRORSCRUB_PAUSED	"errorscrub_paused"
 
 /*
  * datalink subclass definitions.

lib/libzfs/libzfs.abi

@@ -5717,7 +5717,8 @@
       <enumerator name='POOL_SCAN_NONE' value='0'/>
       <enumerator name='POOL_SCAN_SCRUB' value='1'/>
       <enumerator name='POOL_SCAN_RESILVER' value='2'/>
-      <enumerator name='POOL_SCAN_FUNCS' value='3'/>
+      <enumerator name='POOL_SCAN_ERRORSCRUB' value='3'/>
+      <enumerator name='POOL_SCAN_FUNCS' value='4'/>
     </enum-decl>
     <typedef-decl name='pool_scan_func_t' type-id='1b092565' id='7313fbe2'/>
     <enum-decl name='pool_scrub_cmd' id='a1474cbd'>