[illumos-gate merge]

commit 4a20c933b148de8a1c1d3538391c64284e636653
    6111 zfs send should ignore datasets created after the ending snapshot
commit 1fdcbd00c9cbac286b5f92e08877e8cb3c448420
    6874 rollback and receive need to reset ZPL state to what's on disk

2 files changed, 200 insertions, 103 deletions

diff --git a/usr/src/lib/libzfs/common/libzfs_sendrecv.c b/usr/src/lib/libzfs/common/libzfs_sendrecv.c
index 988bbe3dcc..81d79d8976 100644
--- a/usr/src/lib/libzfs/common/libzfs_sendrecv.c
+++ b/usr/src/lib/libzfs/common/libzfs_sendrecv.c
@@ -571,13 +571,30 @@ fsavl_create(nvlist_t *fss)
  * Routines for dealing with the giant nvlist of fs-nvlists, etc.
  */
 typedef struct send_data {
+	/*
+	 * assigned inside every recursive call,
+	 * restored from *_save on return:
+	 *
+	 * guid of fromsnap snapshot in parent dataset
+	 * txg of fromsnap snapshot in current dataset
+	 * txg of tosnap snapshot in current dataset
+	 */
+
 	uint64_t parent_fromsnap_guid;
+	uint64_t fromsnap_txg;
+	uint64_t tosnap_txg;
+
+	/* the nvlists get accumulated during depth-first traversal */
 	nvlist_t *parent_snaps;
 	nvlist_t *fss;
 	nvlist_t *snapprops;
+
+	/* send-receive configuration, does not change during traversal */
+	const char *fsname;
 	const char *fromsnap;
 	const char *tosnap;
 	boolean_t recursive;
+	boolean_t verbose;
 
 	/*
 	 * The header nvlist is of the following format:
@@ -610,11 +627,23 @@ send_iterate_snap(zfs_handle_t *zhp, void *arg)
 {
 	send_data_t *sd = arg;
 	uint64_t guid = zhp->zfs_dmustats.dds_guid;
+	uint64_t txg = zhp->zfs_dmustats.dds_creation_txg;
 	char *snapname;
 	nvlist_t *nv;
 
 	snapname = strrchr(zhp->zfs_name, '@')+1;
 
+	if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) {
+		if (sd->verbose) {
+			(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
+			    "skipping snapshot %s because it was created "
+			    "after the destination snapshot (%s)\n"),
+			    zhp->zfs_name, sd->tosnap);
+		}
+		zfs_close(zhp);
+		return (0);
+	}
+
 	VERIFY(0 == nvlist_add_uint64(sd->parent_snaps, snapname, guid));
 	/*
 	 * NB: if there is no fromsnap here (it's a newly created fs in
@@ -708,6 +737,31 @@ send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 }
 
 /*
+ * returns snapshot creation txg
+ * and returns 0 if the snapshot does not exist
+ */
+static uint64_t
+get_snap_txg(libzfs_handle_t *hdl, const char *fs, const char *snap)
+{
+	char name[ZFS_MAXNAMELEN];
+	uint64_t txg = 0;
+
+	if (fs == NULL || fs[0] == '\0' || snap == NULL || snap[0] == '\0')
+		return (txg);
+
+	(void) snprintf(name, sizeof (name), "%s@%s", fs, snap);
+	if (zfs_dataset_exists(hdl, name, ZFS_TYPE_SNAPSHOT)) {
+		zfs_handle_t *zhp = zfs_open(hdl, name, ZFS_TYPE_SNAPSHOT);
+		if (zhp != NULL) {
+			txg = zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG);
+			zfs_close(zhp);
+		}
+	}
+
+	return (txg);
+}
+
+/*
  * recursively generate nvlists describing datasets.  See comment
  * for the data structure send_data_t above for description of contents
  * of the nvlist.
@@ -719,9 +773,48 @@ send_iterate_fs(zfs_handle_t *zhp, void *arg)
 	nvlist_t *nvfs, *nv;
 	int rv = 0;
 	uint64_t parent_fromsnap_guid_save = sd->parent_fromsnap_guid;
+	uint64_t fromsnap_txg_save = sd->fromsnap_txg;
+	uint64_t tosnap_txg_save = sd->tosnap_txg;
+	uint64_t txg = zhp->zfs_dmustats.dds_creation_txg;
 	uint64_t guid = zhp->zfs_dmustats.dds_guid;
+	uint64_t fromsnap_txg, tosnap_txg;
 	char guidstring[64];
 
+	fromsnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->fromsnap);
+	if (fromsnap_txg != 0)
+		sd->fromsnap_txg = fromsnap_txg;
+
+	tosnap_txg = get_snap_txg(zhp->zfs_hdl, zhp->zfs_name, sd->tosnap);
+	if (tosnap_txg != 0)
+		sd->tosnap_txg = tosnap_txg;
+
+	/*
+	 * on the send side, if the current dataset does not have tosnap,
+	 * perform two additional checks:
+	 *
+	 * - skip sending the current dataset if it was created later than
+	 *   the parent tosnap
+	 * - return error if the current dataset was created earlier than
+	 *   the parent tosnap
+	 */
+	if (sd->tosnap != NULL && tosnap_txg == 0) {
+		if (sd->tosnap_txg != 0 && txg > sd->tosnap_txg) {
+			if (sd->verbose) {
+				(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
+				    "skipping dataset %s: snapshot %s does "
+				    "not exist\n"), zhp->zfs_name, sd->tosnap);
+			}
+		} else {
+			(void) fprintf(stderr, dgettext(TEXT_DOMAIN,
+			    "cannot send %s@%s%s: snapshot %s@%s does not "
+			    "exist\n"), sd->fsname, sd->tosnap, sd->recursive ?
+			    dgettext(TEXT_DOMAIN, " recursively") : "",
+			    zhp->zfs_name, sd->tosnap);
+			rv = -1;
+		}
+		goto out;
+	}
+
 	VERIFY(0 == nvlist_alloc(&nvfs, NV_UNIQUE_NAME, 0));
 	VERIFY(0 == nvlist_add_string(nvfs, "name", zhp->zfs_name));
 	VERIFY(0 == nvlist_add_uint64(nvfs, "parentfromsnap",
@@ -730,8 +823,10 @@ send_iterate_fs(zfs_handle_t *zhp, void *arg)
 	if (zhp->zfs_dmustats.dds_origin[0]) {
 		zfs_handle_t *origin = zfs_open(zhp->zfs_hdl,
 		    zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
-		if (origin == NULL)
-			return (-1);
+		if (origin == NULL) {
+			rv = -1;
+			goto out;
+		}
 		VERIFY(0 == nvlist_add_uint64(nvfs, "origin",
 		    origin->zfs_dmustats.dds_guid));
 	}
@@ -762,7 +857,10 @@ send_iterate_fs(zfs_handle_t *zhp, void *arg)
 	if (sd->recursive)
 		rv = zfs_iter_filesystems(zhp, send_iterate_fs, sd);
 
+out:
 	sd->parent_fromsnap_guid = parent_fromsnap_guid_save;
+	sd->fromsnap_txg = fromsnap_txg_save;
+	sd->tosnap_txg = tosnap_txg_save;
 
 	zfs_close(zhp);
 	return (rv);
@@ -770,7 +868,8 @@ send_iterate_fs(zfs_handle_t *zhp, void *arg)
 
 static int
 gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
-    const char *tosnap, boolean_t recursive, nvlist_t **nvlp, avl_tree_t **avlp)
+    const char *tosnap, boolean_t recursive, boolean_t verbose,
+    nvlist_t **nvlp, avl_tree_t **avlp)
 {
 	zfs_handle_t *zhp;
 	send_data_t sd = { 0 };
@@ -781,9 +880,11 @@ gather_nvlist(libzfs_handle_t *hdl, const char *fsname, const char *fromsnap,
 		return (EZFS_BADTYPE);
 
 	VERIFY(0 == nvlist_alloc(&sd.fss, NV_UNIQUE_NAME, 0));
+	sd.fsname = fsname;
 	sd.fromsnap = fromsnap;
 	sd.tosnap = tosnap;
 	sd.recursive = recursive;
+	sd.verbose = verbose;
 
 	if ((error = send_iterate_fs(zhp, &sd)) != 0) {
 		nvlist_free(sd.fss);
@@ -1688,7 +1789,8 @@ zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
 			}
 
 			err = gather_nvlist(zhp->zfs_hdl, zhp->zfs_name,
-			    fromsnap, tosnap, flags->replicate, &fss, &fsavl);
+			    fromsnap, tosnap, flags->replicate, flags->verbose,
+			    &fss, &fsavl);
 			if (err)
 				goto err_out;
 			VERIFY(0 == nvlist_add_nvlist(hdrnv, "fss", fss));
@@ -2299,7 +2401,7 @@ again:
 	needagain = progress = B_FALSE;
 
 	if ((error = gather_nvlist(hdl, tofs, fromsnap, NULL,
-	    recursive, &local_nv, &local_avl)) != 0)
+	    recursive, B_FALSE, &local_nv, &local_avl)) != 0)
 		return (error);
 
 	/*
@@ -3330,7 +3432,7 @@ zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
 		 */
 		*cp = '\0';
 		if (gather_nvlist(hdl, zc.zc_value, NULL, NULL, B_FALSE,
-		    &local_nv, &local_avl) == 0) {
+		    B_FALSE, &local_nv, &local_avl) == 0) {
 			*cp = '@';
 			fs = fsavl_find(local_avl, drrb->drr_toguid, NULL);
 			fsavl_destroy(local_avl);
diff --git a/usr/src/uts/common/fs/zfs/zfs_vfsops.c b/usr/src/uts/common/fs/zfs/zfs_vfsops.c
index 6b9d9f90f5..ace4bf8173 100644
--- a/usr/src/uts/common/fs/zfs/zfs_vfsops.c
+++ b/usr/src/uts/common/fs/zfs/zfs_vfsops.c
@@ -21,7 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
- * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
  * Copyright (c) 2014 Integros [integros.com]
  */
 
@@ -853,61 +853,46 @@ zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
 	return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
 }
 
-int
-zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
+/*
+ * Associate this zfsvfs with the given objset, which must be owned.
+ * This will cache a bunch of on-disk state from the objset in the
+ * zfsvfs.
+ */
+static int
+zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
 {
-	objset_t *os;
-	zfsvfs_t *zfsvfs;
-	uint64_t zval;
-	int i, error;
-	uint64_t sa_obj;
-
-	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
-
-	/*
-	 * We claim to always be readonly so we can open snapshots;
-	 * other ZPL code will prevent us from writing to snapshots.
-	 */
-	error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
-	if (error) {
-		kmem_free(zfsvfs, sizeof (zfsvfs_t));
-		return (error);
-	}
+	int error;
+	uint64_t val;
 
-	/*
-	 * Initialize the zfs-specific filesystem structure.
-	 * Should probably make this a kmem cache, shuffle fields,
-	 * and just bzero up to z_hold_mtx[].
-	 */
-	zfsvfs->z_vfs = NULL;
-	zfsvfs->z_parent = zfsvfs;
 	zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
 	zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
 	zfsvfs->z_os = os;
 
 	error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
-	if (error) {
-		goto out;
-	} else if (zfsvfs->z_version >
+	if (error != 0)
+		return (error);
+	if (zfsvfs->z_version >
 	    zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
 		(void) printf("Can't mount a version %lld file system "
 		    "on a version %lld pool\n. Pool must be upgraded to mount "
 		    "this file system.", (u_longlong_t)zfsvfs->z_version,
 		    (u_longlong_t)spa_version(dmu_objset_spa(os)));
-		error = SET_ERROR(ENOTSUP);
-		goto out;
+		return (SET_ERROR(ENOTSUP));
 	}
-	if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
-		goto out;
-	zfsvfs->z_norm = (int)zval;
+	error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
+	if (error != 0)
+		return (error);
+	zfsvfs->z_norm = (int)val;
 
-	if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
-		goto out;
-	zfsvfs->z_utf8 = (zval != 0);
+	error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
+	if (error != 0)
+		return (error);
+	zfsvfs->z_utf8 = (val != 0);
 
-	if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
-		goto out;
-	zfsvfs->z_case = (uint_t)zval;
+	error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
+	if (error != 0)
+		return (error);
+	zfsvfs->z_case = (uint_t)val;
 
 	/*
 	 * Fold case on file systems that are always or sometimes case
@@ -920,60 +905,88 @@ zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
 
+	uint64_t sa_obj = 0;
 	if (zfsvfs->z_use_sa) {
 		/* should either have both of these objects or none */
 		error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
 		    &sa_obj);
-		if (error)
-			goto out;
-	} else {
-		/*
-		 * Pre SA versions file systems should never touch
-		 * either the attribute registration or layout objects.
-		 */
-		sa_obj = 0;
+		if (error != 0)
+			return (error);
 	}
 
 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
 	    &zfsvfs->z_attr_table);
-	if (error)
-		goto out;
+	if (error != 0)
+		return (error);
 
 	if (zfsvfs->z_version >= ZPL_VERSION_SA)
 		sa_register_update_callback(os, zfs_sa_upgrade);
 
 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
 	    &zfsvfs->z_root);
-	if (error)
-		goto out;
+	if (error != 0)
+		return (error);
 	ASSERT(zfsvfs->z_root != 0);
 
 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
 	    &zfsvfs->z_unlinkedobj);
-	if (error)
-		goto out;
+	if (error != 0)
+		return (error);
 
 	error = zap_lookup(os, MASTER_NODE_OBJ,
 	    zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
 	    8, 1, &zfsvfs->z_userquota_obj);
-	if (error && error != ENOENT)
-		goto out;
+	if (error == ENOENT)
+		zfsvfs->z_userquota_obj = 0;
+	else if (error != 0)
+		return (error);
 
 	error = zap_lookup(os, MASTER_NODE_OBJ,
 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
 	    8, 1, &zfsvfs->z_groupquota_obj);
-	if (error && error != ENOENT)
-		goto out;
+	if (error == ENOENT)
+		zfsvfs->z_groupquota_obj = 0;
+	else if (error != 0)
+		return (error);
 
 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
 	    &zfsvfs->z_fuid_obj);
-	if (error && error != ENOENT)
-		goto out;
+	if (error == ENOENT)
+		zfsvfs->z_fuid_obj = 0;
+	else if (error != 0)
+		return (error);
 
 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
 	    &zfsvfs->z_shares_dir);
-	if (error && error != ENOENT)
-		goto out;
+	if (error == ENOENT)
+		zfsvfs->z_shares_dir = 0;
+	else if (error != 0)
+		return (error);
+
+	return (0);
+}
+
+int
+zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
+{
+	objset_t *os;
+	zfsvfs_t *zfsvfs;
+	int error;
+
+	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
+
+	/*
+	 * We claim to always be readonly so we can open snapshots;
+	 * other ZPL code will prevent us from writing to snapshots.
+	 */
+	error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
+	if (error) {
+		kmem_free(zfsvfs, sizeof (zfsvfs_t));
+		return (error);
+	}
+
+	zfsvfs->z_vfs = NULL;
+	zfsvfs->z_parent = zfsvfs;
 
 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
@@ -982,17 +995,19 @@ zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
 	rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
 	rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
 	rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
-	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
+	for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++)
 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
 
+	error = zfsvfs_init(zfsvfs, os);
+	if (error != 0) {
+		dmu_objset_disown(os, zfsvfs);
+		*zfvp = NULL;
+		kmem_free(zfsvfs, sizeof (zfsvfs_t));
+		return (error);
+	}
+
 	*zfvp = zfsvfs;
 	return (0);
-
-out:
-	dmu_objset_disown(os, zfsvfs);
-	*zfvp = NULL;
-	kmem_free(zfsvfs, sizeof (zfsvfs_t));
-	return (error);
 }
 
 static int
@@ -2020,7 +2035,6 @@ zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
 {
 	int err;
 	znode_t *zp;
-	uint64_t sa_obj = 0;
 
 	ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
 	ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
@@ -2029,35 +2043,16 @@ zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
 	 * We already own this, so just hold and rele it to update the
 	 * objset_t, as the one we had before may have been evicted.
 	 */
-	VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os));
-	VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs);
-	VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset));
-	dmu_objset_rele(zfsvfs->z_os, zfsvfs);
-
-	/*
-	 * Make sure version hasn't changed
-	 */
-
-	err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION,
-	    &zfsvfs->z_version);
-
-	if (err)
-		goto bail;
-
-	err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
-	    ZFS_SA_ATTRS, 8, 1, &sa_obj);
-
-	if (err && zfsvfs->z_version >= ZPL_VERSION_SA)
-		goto bail;
+	objset_t *os;
+	VERIFY0(dmu_objset_hold(osname, zfsvfs, &os));
+	VERIFY3P(os->os_dsl_dataset->ds_owner, ==, zfsvfs);
+	VERIFY(dsl_dataset_long_held(os->os_dsl_dataset));
+	dmu_objset_rele(os, zfsvfs);
 
-	if ((err = sa_setup(zfsvfs->z_os, sa_obj,
-	    zfs_attr_table,  ZPL_END, &zfsvfs->z_attr_table)) != 0)
+	err = zfsvfs_init(zfsvfs, os);
+	if (err != 0)
 		goto bail;
 
-	if (zfsvfs->z_version >= ZPL_VERSION_SA)
-		sa_register_update_callback(zfsvfs->z_os,
-		    zfs_sa_upgrade);
-
 	VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
 
 	zfs_set_fuid_feature(zfsvfs);