zfs list: Allow more fields in ZFS_ITER_SIMPLE mode

If the fields to be listed and sorted by are constrained to those
populated by dsl_dataset_fast_stat(), then zfs list is much faster,
as it does not need to open each objset and reads its properties.

A previous optimization by Pawel Dawidek
(0cee24064a79f9c01fc4521543c37acea538405f) took advantage
of this to make listing snapshot names sorted only by name much faster.

However, it was limited to `-o name -s name`, this work extends this
optimization to work with:
  - name
  - guid
  - createtxg
  - numclones
  - inconsistent
  - redacted
  - origin
and could be further extended to any other properties supported by
dsl_dataset_fast_stat() or similar, that do not require extra locking
or reading from disk.

This was committed before (9a9e2e343dfa2af28bf7910de77ae73aa006de62),
but was reverted due to a regression when used with an older kernel.

If the kernel does not populate zc->zc_objset_stats, we now fallback
to getting the properties via the slower interface, to avoid problems
with newer userland and older kernels.

Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Allan Jude <[email protected]>
Closes #14110

3 files changed, 78 insertions, 45 deletions

diff --git a/cmd/zfs/zfs_iter.c b/cmd/zfs/zfs_iter.c
index a0a80d481..0f8ddd93a 100644
--- a/cmd/zfs/zfs_iter.c
+++ b/cmd/zfs/zfs_iter.c
@@ -143,19 +143,20 @@ zfs_callback(zfs_handle_t *zhp, void *data)
 		    (cb->cb_types &
 		    (ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME))) &&
 		    zfs_get_type(zhp) == ZFS_TYPE_FILESYSTEM) {
-			(void) zfs_iter_filesystems(zhp, zfs_callback, data);
+			(void) zfs_iter_filesystems(zhp, cb->cb_flags,
+			    zfs_callback, data);
 		}
 
 		if (((zfs_get_type(zhp) & (ZFS_TYPE_SNAPSHOT |
 		    ZFS_TYPE_BOOKMARK)) == 0) && include_snaps) {
-			(void) zfs_iter_snapshots(zhp,
-			    (cb->cb_flags & ZFS_ITER_SIMPLE) != 0,
+			(void) zfs_iter_snapshots(zhp, cb->cb_flags,
 			    zfs_callback, data, 0, 0);
 		}
 
 		if (((zfs_get_type(zhp) & (ZFS_TYPE_SNAPSHOT |
 		    ZFS_TYPE_BOOKMARK)) == 0) && include_bmarks) {
-			(void) zfs_iter_bookmarks(zhp, zfs_callback, data);
+			(void) zfs_iter_bookmarks(zhp, cb->cb_flags,
+			    zfs_callback, data);
 		}
 
 		cb->cb_depth--;
@@ -211,18 +212,58 @@ zfs_free_sort_columns(zfs_sort_column_t *sc)
 	}
 }
 
-int
-zfs_sort_only_by_name(const zfs_sort_column_t *sc)
+/*
+ * Return true if all of the properties to be sorted are populated by
+ * dsl_dataset_fast_stat(). Note that sc == NULL (no sort) means we
+ * don't need any extra properties, so returns true.
+ */
+boolean_t
+zfs_sort_only_by_fast(const zfs_sort_column_t *sc)
 {
-	return (sc != NULL && sc->sc_next == NULL &&
-	    sc->sc_prop == ZFS_PROP_NAME);
+	while (sc != NULL) {
+		switch (sc->sc_prop) {
+		case ZFS_PROP_NAME:
+		case ZFS_PROP_GUID:
+		case ZFS_PROP_CREATETXG:
+		case ZFS_PROP_NUMCLONES:
+		case ZFS_PROP_INCONSISTENT:
+		case ZFS_PROP_REDACTED:
+		case ZFS_PROP_ORIGIN:
+			break;
+		default:
+			return (B_FALSE);
+		}
+		sc = sc->sc_next;
+	}
+
+	return (B_TRUE);
 }
 
-int
-zfs_sort_only_by_createtxg(const zfs_sort_column_t *sc)
+boolean_t
+zfs_list_only_by_fast(const zprop_list_t *p)
 {
-	return (sc != NULL && sc->sc_next == NULL &&
-	    sc->sc_prop == ZFS_PROP_CREATETXG);
+	if (p == NULL) {
+		/* NULL means 'all' so we can't use simple mode */
+		return (B_FALSE);
+	}
+
+	while (p != NULL) {
+		switch (p->pl_prop) {
+		case ZFS_PROP_NAME:
+		case ZFS_PROP_GUID:
+		case ZFS_PROP_CREATETXG:
+		case ZFS_PROP_NUMCLONES:
+		case ZFS_PROP_INCONSISTENT:
+		case ZFS_PROP_REDACTED:
+		case ZFS_PROP_ORIGIN:
+			break;
+		default:
+			return (B_FALSE);
+		}
+		p = p->pl_next;
+	}
+
+	return (B_TRUE);
 }
 
 static int
diff --git a/cmd/zfs/zfs_iter.h b/cmd/zfs/zfs_iter.h
index effb22ded..d742ec7b2 100644
--- a/cmd/zfs/zfs_iter.h
+++ b/cmd/zfs/zfs_iter.h
@@ -40,20 +40,12 @@ typedef struct zfs_sort_column {
 	boolean_t		sc_reverse;
 } zfs_sort_column_t;
 
-#define	ZFS_ITER_RECURSE	   (1 << 0)
-#define	ZFS_ITER_ARGS_CAN_BE_PATHS (1 << 1)
-#define	ZFS_ITER_PROP_LISTSNAPS    (1 << 2)
-#define	ZFS_ITER_DEPTH_LIMIT	   (1 << 3)
-#define	ZFS_ITER_RECVD_PROPS	   (1 << 4)
-#define	ZFS_ITER_LITERAL_PROPS	   (1 << 5)
-#define	ZFS_ITER_SIMPLE		   (1 << 6)
-
 int zfs_for_each(int, char **, int options, zfs_type_t,
     zfs_sort_column_t *, zprop_list_t **, int, zfs_iter_f, void *);
 int zfs_add_sort_column(zfs_sort_column_t **, const char *, boolean_t);
 void zfs_free_sort_columns(zfs_sort_column_t *);
-int zfs_sort_only_by_name(const zfs_sort_column_t *);
-int zfs_sort_only_by_createtxg(const zfs_sort_column_t *);
+boolean_t zfs_sort_only_by_fast(const zfs_sort_column_t *);
+boolean_t zfs_list_only_by_fast(const zprop_list_t *);
 
 #ifdef	__cplusplus
 }
diff --git a/cmd/zfs/zfs_main.c b/cmd/zfs/zfs_main.c
index 03640a4cd..44440dc3d 100644
--- a/cmd/zfs/zfs_main.c
+++ b/cmd/zfs/zfs_main.c
@@ -1531,7 +1531,7 @@ destroy_print_snapshots(zfs_handle_t *fs_zhp, destroy_cbdata_t *cb)
 	int err;
 	assert(cb->cb_firstsnap == NULL);
 	assert(cb->cb_prevsnap == NULL);
-	err = zfs_iter_snapshots_sorted(fs_zhp, destroy_print_cb, cb, 0, 0);
+	err = zfs_iter_snapshots_sorted(fs_zhp, 0, destroy_print_cb, cb, 0, 0);
 	if (cb->cb_firstsnap != NULL) {
 		uint64_t used = 0;
 		if (err == 0) {
@@ -1557,7 +1557,7 @@ snapshot_to_nvl_cb(zfs_handle_t *zhp, void *arg)
 	if (!cb->cb_doclones && !cb->cb_defer_destroy) {
 		cb->cb_target = zhp;
 		cb->cb_first = B_TRUE;
-		err = zfs_iter_dependents(zhp, B_TRUE,
+		err = zfs_iter_dependents(zhp, 0, B_TRUE,
 		    destroy_check_dependent, cb);
 	}
 
@@ -1575,7 +1575,8 @@ gather_snapshots(zfs_handle_t *zhp, void *arg)
 	destroy_cbdata_t *cb = arg;
 	int err = 0;
 
-	err = zfs_iter_snapspec(zhp, cb->cb_snapspec, snapshot_to_nvl_cb, cb);
+	err = zfs_iter_snapspec(zhp, 0, cb->cb_snapspec,
+	    snapshot_to_nvl_cb, cb);
 	if (err == ENOENT)
 		err = 0;
 	if (err != 0)
@@ -1588,7 +1589,7 @@ gather_snapshots(zfs_handle_t *zhp, void *arg)
 	}
 
 	if (cb->cb_recurse)
-		err = zfs_iter_filesystems(zhp, gather_snapshots, cb);
+		err = zfs_iter_filesystems(zhp, 0, gather_snapshots, cb);
 
 out:
 	zfs_close(zhp);
@@ -1613,7 +1614,7 @@ destroy_clones(destroy_cbdata_t *cb)
 			 * false while destroying the clones.
 			 */
 			cb->cb_defer_destroy = B_FALSE;
-			err = zfs_iter_dependents(zhp, B_FALSE,
+			err = zfs_iter_dependents(zhp, 0, B_FALSE,
 			    destroy_callback, cb);
 			cb->cb_defer_destroy = defer;
 			zfs_close(zhp);
@@ -1824,7 +1825,7 @@ zfs_do_destroy(int argc, char **argv)
 		 */
 		cb.cb_first = B_TRUE;
 		if (!cb.cb_doclones &&
-		    zfs_iter_dependents(zhp, B_TRUE, destroy_check_dependent,
+		    zfs_iter_dependents(zhp, 0, B_TRUE, destroy_check_dependent,
 		    &cb) != 0) {
 			rv = 1;
 			goto out;
@@ -1835,7 +1836,7 @@ zfs_do_destroy(int argc, char **argv)
 			goto out;
 		}
 		cb.cb_batchedsnaps = fnvlist_alloc();
-		if (zfs_iter_dependents(zhp, B_FALSE, destroy_callback,
+		if (zfs_iter_dependents(zhp, 0, B_FALSE, destroy_callback,
 		    &cb) != 0) {
 			rv = 1;
 			goto out;
@@ -3660,16 +3661,6 @@ found3:;
 	argv += optind;
 
 	/*
-	 * If we are only going to list snapshot names and sort by name or
-	 * by createtxg, then we can use faster version.
-	 */
-	if (strcmp(fields, "name") == 0 &&
-	    (zfs_sort_only_by_name(sortcol) ||
-	    zfs_sort_only_by_createtxg(sortcol))) {
-		flags |= ZFS_ITER_SIMPLE;
-	}
-
-	/*
 	 * If "-o space" and no types were specified, don't display snapshots.
 	 */
 	if (strcmp(fields, "space") == 0 && types_specified == B_FALSE)
@@ -3696,6 +3687,15 @@ found3:;
 
 	cb.cb_first = B_TRUE;
 
+	/*
+	 * If we are only going to list and sort by properties that are "fast"
+	 * then we can use "simple" mode and avoid populating the properties
+	 * nvlist.
+	 */
+	if (zfs_list_only_by_fast(cb.cb_proplist) &&
+	    zfs_sort_only_by_fast(sortcol))
+		flags |= ZFS_ITER_SIMPLE;
+
 	ret = zfs_for_each(argc, argv, flags, types, sortcol, &cb.cb_proplist,
 	    limit, list_callback, &cb);
 
@@ -4006,7 +4006,7 @@ rollback_check(zfs_handle_t *zhp, void *data)
 		}
 
 		if (cbp->cb_recurse) {
-			if (zfs_iter_dependents(zhp, B_TRUE,
+			if (zfs_iter_dependents(zhp, 0, B_TRUE,
 			    rollback_check_dependent, cbp) != 0) {
 				zfs_close(zhp);
 				return (-1);
@@ -4105,10 +4105,10 @@ zfs_do_rollback(int argc, char **argv)
 	if (cb.cb_create > 0)
 		min_txg = cb.cb_create;
 
-	if ((ret = zfs_iter_snapshots(zhp, B_FALSE, rollback_check, &cb,
+	if ((ret = zfs_iter_snapshots(zhp, 0, rollback_check, &cb,
 	    min_txg, 0)) != 0)
 		goto out;
-	if ((ret = zfs_iter_bookmarks(zhp, rollback_check, &cb)) != 0)
+	if ((ret = zfs_iter_bookmarks(zhp, 0, rollback_check, &cb)) != 0)
 		goto out;
 
 	if ((ret = cb.cb_error) != 0)
@@ -4250,7 +4250,7 @@ zfs_snapshot_cb(zfs_handle_t *zhp, void *arg)
 	free(name);
 
 	if (sd->sd_recursive)
-		rv = zfs_iter_filesystems(zhp, zfs_snapshot_cb, sd);
+		rv = zfs_iter_filesystems(zhp, 0, zfs_snapshot_cb, sd);
 	zfs_close(zhp);
 	return (rv);
 }
@@ -6310,7 +6310,7 @@ zfs_do_allow_unallow_impl(int argc, char **argv, boolean_t un)
 
 		if (un && opts.recursive) {
 			struct deleg_perms data = { un, update_perm_nvl };
-			if (zfs_iter_filesystems(zhp, set_deleg_perms,
+			if (zfs_iter_filesystems(zhp, 0, set_deleg_perms,
 			    &data) != 0)
 				goto cleanup0;
 		}
@@ -6688,7 +6688,7 @@ get_one_dataset(zfs_handle_t *zhp, void *data)
 	/*
 	 * Iterate over any nested datasets.
 	 */
-	if (zfs_iter_filesystems(zhp, get_one_dataset, data) != 0) {
+	if (zfs_iter_filesystems(zhp, 0, get_one_dataset, data) != 0) {
 		zfs_close(zhp);
 		return (1);
 	}