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.

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

cmd/zpool/zpool_main.c

@@ -395,7 +395,7 @@ 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:
@@ -7145,8 +7145,9 @@ 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.
@@ -7162,14 +7163,21 @@ 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;
@@ -7181,11 +7189,27 @@ 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_ERRORSCRUB;
+			if (is_pause) {
+				cb.cb_scrub_cmd = POOL_SCRUB_PAUSE;
+			} else if (is_stop) {
+				cb.cb_scrub_cmd = POOL_ERRORSCRUB_STOP;
+			} else {
+				cb.cb_scrub_cmd = POOL_SCRUB_NORMAL;
+			}
+		} else {
+			if (is_pause) {
+				cb.cb_scrub_cmd = POOL_SCRUB_PAUSE;
+			} else if (is_stop) {
+				cb.cb_type = POOL_SCAN_NONE;
+			}
+		}
 	}
 
 	if (wait && (cb.cb_type == POOL_SCAN_NONE ||
@@ -7408,6 +7432,70 @@ 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_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_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.
  */
@@ -7733,6 +7821,7 @@ 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;
@@ -7747,6 +7836,7 @@ print_scan_status(zpool_handle_t *zhp, nvlist_t *nvroot)
 
 		have_resilver = (ps->pss_func == POOL_SCAN_RESILVER);
 		have_scrub = (ps->pss_func == POOL_SCAN_SCRUB);
+		have_errorscrub = (ps->pss_error_scrub_func == POOL_ERRORSCRUB);
 	}
 
 	boolean_t active_rebuild = check_rebuilding(nvroot, &rebuild_end_time);
@@ -7755,6 +7845,8 @@ print_scan_status(zpool_handle_t *zhp, nvlist_t *nvroot)
 	/* Always print the scrub status when available. */
 	if (have_scrub)
 		print_scan_scrub_resilver_status(ps);
+	if (have_errorscrub)
+		print_err_scrub_status(ps);
 
 	/*
 	 * When there is an active resilver or rebuild print its status.
@@ -8463,6 +8555,7 @@ status_callback(zpool_handle_t *zhp, void *data)
 
 		(void) nvlist_lookup_uint64_array(nvroot,
 		    ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t **)&prs, &c);
+
 		print_removal_status(zhp, prs);
 
 		(void) nvlist_lookup_uint64_array(nvroot,

include/libzfs.h

@@ -126,11 +126,15 @@ 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_NO_ERRORSCRUB,	/* no active error 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

@@ -143,6 +143,9 @@ _LIBZFS_CORE_H int lzc_wait_fs(const char *, zfs_wait_activity_t, boolean_t *);
 
 _LIBZFS_CORE_H int lzc_set_bootenv(const char *, const nvlist_t *);
 _LIBZFS_CORE_H int lzc_get_bootenv(const char *, nvlist_t **);
+
+_LIBZFS_CORE_H int lzc_scrub(zfs_ioc_t, const char *, nvlist_t *, nvlist_t **);
+
 #ifdef	__cplusplus
 }
 #endif

include/sys/dmu.h

@@ -370,6 +370,7 @@ 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 +167,16 @@ 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_bytes_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;
@@ -167,12 +188,16 @@ int dsl_scan_setup_check(void *, dmu_tx_t *);
 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_cancel(struct dsl_pool *, pool_scan_func_t func);
 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, pool_scan_func_t func);
+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 +209,7 @@ 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

@@ -942,6 +942,7 @@ typedef enum pool_scan_func {
 	POOL_SCAN_NONE,
 	POOL_SCAN_SCRUB,
 	POOL_SCAN_RESILVER,
+	POOL_ERRORSCRUB,
 	POOL_SCAN_FUNCS
 } pool_scan_func_t;
 
@@ -951,6 +952,7 @@ typedef enum pool_scan_func {
 typedef enum pool_scrub_cmd {
 	POOL_SCRUB_NORMAL = 0,
 	POOL_SCRUB_PAUSE,
+	POOL_ERRORSCRUB_STOP,
 	POOL_SCRUB_FLAGS_END
 } pool_scrub_cmd_t;
 
@@ -1005,6 +1007,20 @@ 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 {
@@ -1026,6 +1042,7 @@ typedef enum dsl_scan_state {
 	DSS_SCANNING,
 	DSS_FINISHED,
 	DSS_CANCELED,
+	DSS_ERRORSCRUBBING,
 	DSS_NUM_STATES
 } dsl_scan_state_t;
 
@@ -1357,6 +1374,7 @@ typedef enum zfs_ioc {
 	ZFS_IOC_GET_BOOKMARK_PROPS,		/* 0x5a52 */
 	ZFS_IOC_WAIT,				/* 0x5a53 */
 	ZFS_IOC_WAIT_FS,			/* 0x5a54 */
+	ZFS_IOC_POOL_SCRUB,			/* 0x5a55 */
 
 	/*
 	 * Per-platform (Optional) - 8/128 numbers reserved.

include/sys/spa.h

@@ -818,8 +818,9 @@ 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_stop(spa_t *spa);
-extern int spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t flag);
+extern int spa_scan_stop(spa_t *spa, pool_scan_func_t func);
+extern int spa_scrub_pause_resume(spa_t *spa, pool_scan_func_t func,
+    pool_scrub_cmd_t flag);
 
 /* spa syncing */
 extern void spa_sync(spa_t *spa, uint64_t txg); /* only for DMU use */
@@ -1140,6 +1141,7 @@ extern void zfs_post_remove(spa_t *spa, vdev_t *vd);
 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_get_errlog_size(spa_t *spa);
+extern uint64_t spa_get_last_errlog_size(spa_t *spa);
 extern int spa_get_errlog(spa_t *spa, void *uaddr, size_t *count);
 extern void spa_errlog_rotate(spa_t *spa);
 extern void spa_errlog_drain(spa_t *spa);

include/sys/spa_impl.h

@@ -291,6 +291,10 @@ 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	"error_scrub_start"
+#define	ESC_ZFS_ERRORSCRUB_FINISH	"error_scrub_finish"
+#define	ESC_ZFS_ERRORSCRUB_ABORT	"error_scrub_abort"
+#define	ESC_ZFS_ERRORSCRUB_RESUME	"error_scrub_resume"
+#define	ESC_ZFS_ERRORSCRUB_PAUSED	"error_scrub_paused"
 
 /*
  * datalink subclass definitions.