Merge remote-tracking branch 'upstream/master' into merge

cmd/zdb/zdb.c

@@ -3590,8 +3590,15 @@ dump_object(objset_t *os, uint64_t object, int verbosity,
 		*print_header = B_TRUE;
 	}
 
-	if (verbosity >= 5)
+	if (verbosity >= 5) {
+		if (dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
+			char blkbuf[BP_SPRINTF_LEN];
+			snprintf_blkptr_compact(blkbuf, sizeof (blkbuf),
+			    DN_SPILL_BLKPTR(dn->dn_phys), B_FALSE);
+			(void) printf("\nSpill block: %s\n", blkbuf);
+		}
 		dump_indirect(dn);
+	}
 
 	if (verbosity >= 5) {
 		/*

cmd/ztest.c

@@ -3027,7 +3027,7 @@ ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
 	zil_close(zd->zd_zilog);
 
 	/* zfsvfs_setup() */
-	VERIFY3P(zil_open(os, ztest_get_data), ==, zd->zd_zilog);
+	VERIFY3P(zil_open(os, ztest_get_data, NULL), ==, zd->zd_zilog);
 	zil_replay(os, zd, ztest_replay_vector);
 
 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
@@ -4534,7 +4534,7 @@ ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
 	/*
 	 * Open the intent log for it.
 	 */
-	zilog = zil_open(os, ztest_get_data);
+	zilog = zil_open(os, ztest_get_data, NULL);
 
 	/*
 	 * Put some objects in there, do a little I/O to them,
@@ -7508,7 +7508,7 @@ ztest_dataset_open(int d)
 		    zilog->zl_parse_lr_count,
 		    zilog->zl_replaying_seq);
 
-	zilog = zil_open(os, ztest_get_data);
+	zilog = zil_open(os, ztest_get_data, NULL);
 
 	if (zilog->zl_replaying_seq != 0 &&
 	    zilog->zl_replaying_seq < committed_seq)

include/sys/dataset_kstats.h

@@ -30,6 +30,7 @@
 #include <sys/wmsum.h>
 #include <sys/dmu.h>
 #include <sys/kstat.h>
+#include <sys/zil.h>
 
 typedef struct dataset_sum_stats_t {
 	wmsum_t dss_writes;
@@ -56,14 +57,19 @@ typedef struct dataset_kstat_values {
 	 * entry is removed from the unlinked set
 	 */
 	kstat_named_t dkv_nunlinked;
+	/*
+	 * Per dataset zil kstats
+	 */
+	zil_kstat_values_t dkv_zil_stats;
 } dataset_kstat_values_t;
 
 typedef struct dataset_kstats {
 	dataset_sum_stats_t dk_sums;
+	zil_sums_t dk_zil_sums;
 	kstat_t *dk_kstats;
 } dataset_kstats_t;
 
-void dataset_kstats_create(dataset_kstats_t *, objset_t *);
+int dataset_kstats_create(dataset_kstats_t *, objset_t *);
 void dataset_kstats_destroy(dataset_kstats_t *);
 
 void dataset_kstats_update_write_kstats(dataset_kstats_t *, int64_t);

include/sys/zil.h

@@ -33,6 +33,7 @@
 #include <sys/zio.h>
 #include <sys/dmu.h>
 #include <sys/zio_crypt.h>
+#include <sys/wmsum.h>
 
 #ifdef	__cplusplus
 extern "C" {
@@ -485,12 +486,36 @@ typedef struct zil_stats {
 	 * zil_commit call. Keep a count of how often that occurs.
 	 */
 	kstat_named_t zil_skip_zil_commit;
-} zil_stats_t;
-
-#define	ZIL_STAT_INCR(stat, val) \
-    atomic_add_64(&zil_stats.stat.value.ui64, (val));
-#define	ZIL_STAT_BUMP(stat) \
-    ZIL_STAT_INCR(stat, 1);
+} zil_kstat_values_t;
+
+typedef struct zil_sums {
+	wmsum_t zil_commit_count;
+	wmsum_t zil_commit_writer_count;
+	wmsum_t zil_itx_count;
+	wmsum_t zil_itx_indirect_count;
+	wmsum_t zil_itx_indirect_bytes;
+	wmsum_t zil_itx_copied_count;
+	wmsum_t zil_itx_copied_bytes;
+	wmsum_t zil_itx_needcopy_count;
+	wmsum_t zil_itx_needcopy_bytes;
+	wmsum_t zil_itx_metaslab_normal_count;
+	wmsum_t zil_itx_metaslab_normal_bytes;
+	wmsum_t zil_itx_metaslab_slog_count;
+	wmsum_t zil_itx_metaslab_slog_bytes;
+	wmsum_t zil_slog_alloc_failures;
+	wmsum_t zil_skip_zil_commit;
+} zil_sums_t;
+
+#define	ZIL_STAT_INCR(zil, stat, val) \
+	do { \
+		int64_t tmpval = (val); \
+		wmsum_add(&(zil_sums_global.stat), tmpval); \
+		if ((zil)->zl_sums) \
+			wmsum_add(&((zil)->zl_sums->stat), tmpval); \
+	} while (0)
+
+#define	ZIL_STAT_BUMP(zil, stat) \
+    ZIL_STAT_INCR(zil, stat, 1);
 
 typedef int zil_parse_blk_func_t(zilog_t *zilog, const blkptr_t *bp, void *arg,
     uint64_t txg);
@@ -510,7 +535,8 @@ extern void	zil_fini(void);
 extern zilog_t	*zil_alloc(objset_t *os, zil_header_t *zh_phys);
 extern void	zil_free(zilog_t *zilog);
 
-extern zilog_t	*zil_open(objset_t *os, zil_get_data_t *get_data);
+extern zilog_t	*zil_open(objset_t *os, zil_get_data_t *get_data,
+    zil_sums_t *zil_sums);
 extern void	zil_close(zilog_t *zilog);
 
 extern void	zil_replay(objset_t *os, void *arg,
@@ -550,6 +576,11 @@ extern void	zil_set_logbias(zilog_t *zilog, uint64_t slogval);
 extern uint64_t	zil_max_copied_data(zilog_t *zilog);
 extern uint64_t	zil_max_log_data(zilog_t *zilog);
 
+extern void zil_sums_init(zil_sums_t *zs);
+extern void zil_sums_fini(zil_sums_t *zs);
+extern void zil_kstat_values_update(zil_kstat_values_t *zs,
+    zil_sums_t *zil_sums);
+
 extern int zil_replay_disable;
 
 #ifdef	__cplusplus

include/sys/zil_impl.h

@@ -222,6 +222,9 @@ struct zilog {
 	 * (see zil_max_copied_data()).
 	 */
 	uint64_t	zl_max_block_size;
+
+	/* Pointer for per dataset zil sums */
+	zil_sums_t *zl_sums;
 };
 
 typedef struct zil_bp_node {

include/sys/zio.h

@@ -709,6 +709,8 @@ extern void spa_handle_ignored_writes(spa_t *spa);
 /* zbookmark_phys functions */
 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
+boolean_t zbookmark_subtree_tbd(const struct dnode_phys *dnp,
+    const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
     uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
 

module/os/freebsd/zfs/zfs_vfsops.c

@@ -1027,8 +1027,6 @@ zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
 	if (error)
 		return (error);
 
-	zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
-
 	/*
 	 * If we are not mounting (ie: online recv), then we don't
 	 * have to worry about replaying the log as we blocked all
@@ -1038,7 +1036,11 @@ zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
 		boolean_t readonly;
 
 		ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL);
-		dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
+		error = dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
+		if (error)
+			return (error);
+		zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
+		    &zfsvfs->z_kstat.dk_zil_sums);
 
 		/*
 		 * During replay we remove the read only flag to
@@ -1109,6 +1111,10 @@ zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
 		/* restore readonly bit */
 		if (readonly != 0)
 			zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
+	} else {
+		ASSERT3P(zfsvfs->z_kstat.dk_kstats, !=, NULL);
+		zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
+		    &zfsvfs->z_kstat.dk_zil_sums);
 	}
 
 	/*

module/os/freebsd/zfs/zvol_os.c

@@ -1189,7 +1189,7 @@ zvol_ensure_zilog(zvol_state_t *zv)
 		}
 		if (zv->zv_zilog == NULL) {
 			zv->zv_zilog = zil_open(zv->zv_objset,
-			    zvol_get_data);
+			    zvol_get_data, &zv->zv_kstat.dk_zil_sums);
 			zv->zv_flags |= ZVOL_WRITTEN_TO;
 			/* replay / destroy done in zvol_os_create_minor() */
 			VERIFY0(zv->zv_zilog->zl_header->zh_flags &
@@ -1422,8 +1422,12 @@ zvol_os_create_minor(const char *name)
 	zv->zv_volsize = volsize;
 	zv->zv_objset = os;
 
+	ASSERT3P(zv->zv_kstat.dk_kstats, ==, NULL);
+	error = dataset_kstats_create(&zv->zv_kstat, zv->zv_objset);
+	if (error)
+		goto out_dmu_objset_disown;
 	ASSERT3P(zv->zv_zilog, ==, NULL);
-	zv->zv_zilog = zil_open(os, zvol_get_data);
+	zv->zv_zilog = zil_open(os, zvol_get_data, &zv->zv_kstat.dk_zil_sums);
 	if (spa_writeable(dmu_objset_spa(os))) {
 		if (zil_replay_disable)
 			zil_destroy(zv->zv_zilog, B_FALSE);
@@ -1432,8 +1436,6 @@ zvol_os_create_minor(const char *name)
 	}
 	zil_close(zv->zv_zilog);
 	zv->zv_zilog = NULL;
-	ASSERT3P(zv->zv_kstat.dk_kstats, ==, NULL);
-	dataset_kstats_create(&zv->zv_kstat, zv->zv_objset);
 
 	/* TODO: prefetch for geom tasting */
 

module/os/linux/zfs/zfs_vfsops.c

@@ -848,16 +848,18 @@ zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
 	if (error)
 		return (error);
 
-	zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
-
 	/*
 	 * If we are not mounting (ie: online recv), then we don't
 	 * have to worry about replaying the log as we blocked all
 	 * operations out since we closed the ZIL.
 	 */
 	if (mounting) {
 		ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL);
-		dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
+		error = dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
+		if (error)
+			return (error);
+		zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
+		    &zfsvfs->z_kstat.dk_zil_sums);
 
 		/*
 		 * During replay we remove the read only flag to
@@ -921,6 +923,10 @@ zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
 		/* restore readonly bit */
 		if (readonly != 0)
 			readonly_changed_cb(zfsvfs, B_TRUE);
+	} else {
+		ASSERT3P(zfsvfs->z_kstat.dk_kstats, !=, NULL);
+		zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
+		    &zfsvfs->z_kstat.dk_zil_sums);
 	}
 
 	/*

module/os/linux/zfs/zvol_os.c

@@ -576,7 +576,7 @@ zvol_request_impl(zvol_state_t *zv, struct bio *bio, struct request *rq,
 			rw_enter(&zv->zv_suspend_lock, RW_WRITER);
 			if (zv->zv_zilog == NULL) {
 				zv->zv_zilog = zil_open(zv->zv_objset,
-				    zvol_get_data);
+				    zvol_get_data, &zv->zv_kstat.dk_zil_sums);
 				zv->zv_flags |= ZVOL_WRITTEN_TO;
 				/* replay / destroy done in zvol_create_minor */
 				VERIFY0((zv->zv_zilog->zl_header->zh_flags &
@@ -1426,8 +1426,12 @@ zvol_os_create_minor(const char *name)
 	blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, zv->zv_zso->zvo_queue);
 #endif
 
+	ASSERT3P(zv->zv_kstat.dk_kstats, ==, NULL);
+	error = dataset_kstats_create(&zv->zv_kstat, zv->zv_objset);
+	if (error)
+		goto out_dmu_objset_disown;
 	ASSERT3P(zv->zv_zilog, ==, NULL);
-	zv->zv_zilog = zil_open(os, zvol_get_data);
+	zv->zv_zilog = zil_open(os, zvol_get_data, &zv->zv_kstat.dk_zil_sums);
 	if (spa_writeable(dmu_objset_spa(os))) {
 		if (zil_replay_disable)
 			zil_destroy(zv->zv_zilog, B_FALSE);
@@ -1436,8 +1440,6 @@ zvol_os_create_minor(const char *name)
 	}
 	zil_close(zv->zv_zilog);
 	zv->zv_zilog = NULL;
-	ASSERT3P(zv->zv_kstat.dk_kstats, ==, NULL);
-	dataset_kstats_create(&zv->zv_kstat, zv->zv_objset);
 
 	/*
 	 * When udev detects the addition of the device it will immediately