2 * SPDX-License-Identifier: BSD-2-Clause
4 * Copyright (c) 2017 Kyle J. Kneitinger <kyle@kneit.in>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/param.h>
29 #include <sys/module.h>
30 #include <sys/mount.h>
32 #include <sys/ucred.h>
33 #include <sys/queue.h>
34 #include <sys/zfs_context.h>
35 #include <sys/mntent.h>
36 #include <sys/zfs_ioctl.h>
41 #include <libzfs_core.h>
42 #include <libzfs_impl.h>
47 #include <libzfsbootenv.h>
52 struct promote_entry {
53 char name[BE_MAXPATHLEN];
54 SLIST_ENTRY(promote_entry) link;
57 struct be_destroy_data {
59 char target_name[BE_MAXPATHLEN];
61 SLIST_HEAD(, promote_entry) promotelist;
65 static int be_create_child_noent(libbe_handle_t *lbh, const char *active,
66 const char *child_path);
67 static int be_create_child_cloned(libbe_handle_t *lbh, const char *active);
70 /* Arbitrary... should tune */
71 #define BE_SNAP_SERIAL_MAX 1024
74 * Iterator function for locating the rootfs amongst the children of the
75 * zfs_be_root set by loader(8). data is expected to be a libbe_handle_t *.
78 be_locate_rootfs(libbe_handle_t *lbh)
85 * Check first if root is ZFS; if not, we'll bail on rootfs capture.
86 * Unfortunately needed because zfs_path_to_zhandle will emit to
87 * stderr if / isn't actually a ZFS filesystem, which we'd like
90 if (statfs("/", &sfs) == 0) {
91 statfs2mnttab(&sfs, &entry);
92 if (strcmp(entry.mnt_fstype, MNTTYPE_ZFS) != 0)
96 zfs = zfs_path_to_zhandle(lbh->lzh, "/", ZFS_TYPE_FILESYSTEM);
100 strlcpy(lbh->rootfs, zfs_get_name(zfs), sizeof(lbh->rootfs));
106 * Initializes the libbe context to operate in the root boot environment
107 * dataset, for example, zroot/ROOT.
110 libbe_init(const char *root)
112 char altroot[MAXPATHLEN];
114 char *poolname, *pos;
118 poolname = pos = NULL;
121 * If the zfs kmod's not loaded then the later libzfs_init() will load
122 * the module for us, but that's not desirable for a couple reasons. If
123 * the module's not loaded, there's no pool imported and we're going to
124 * fail anyways. We also don't really want libbe consumers to have that
125 * kind of side-effect (module loading) in the general case.
127 if (modfind("zfs") < 0)
130 if ((lbh = calloc(1, sizeof(libbe_handle_t))) == NULL)
133 if ((lbh->lzh = libzfs_init()) == NULL)
137 * Grab rootfs, we'll work backwards from there if an optional BE root
138 * has not been passed in.
140 if (be_locate_rootfs(lbh) != 0) {
146 /* Strip off the final slash from rootfs to get the be root */
147 strlcpy(lbh->root, lbh->rootfs, sizeof(lbh->root));
148 pos = strrchr(lbh->root, '/');
153 strlcpy(lbh->root, root, sizeof(lbh->root));
155 if ((pos = strchr(lbh->root, '/')) == NULL)
158 pnamelen = pos - lbh->root;
159 poolname = malloc(pnamelen + 1);
160 if (poolname == NULL)
163 strlcpy(poolname, lbh->root, pnamelen + 1);
164 if ((lbh->active_phandle = zpool_open(lbh->lzh, poolname)) == NULL)
169 if (zpool_get_prop(lbh->active_phandle, ZPOOL_PROP_BOOTFS, lbh->bootfs,
170 sizeof(lbh->bootfs), NULL, true) != 0)
173 if (zpool_get_prop(lbh->active_phandle, ZPOOL_PROP_ALTROOT,
174 altroot, sizeof(altroot), NULL, true) == 0 &&
175 strcmp(altroot, "-") != 0)
176 lbh->altroot_len = strlen(altroot);
178 (void) lzbe_get_boot_device(zpool_get_name(lbh->active_phandle),
184 if (lbh->active_phandle != NULL)
185 zpool_close(lbh->active_phandle);
186 if (lbh->lzh != NULL)
187 libzfs_fini(lbh->lzh);
196 * Free memory allocated by libbe_init()
199 libbe_close(libbe_handle_t *lbh)
202 if (lbh->active_phandle != NULL)
203 zpool_close(lbh->active_phandle);
204 libzfs_fini(lbh->lzh);
211 * Proxy through to libzfs for the moment.
214 be_nicenum(uint64_t num, char *buf, size_t buflen)
217 zfs_nicenum(num, buf, buflen);
221 be_should_promote_clones(zfs_handle_t *zfs_hdl, struct be_destroy_data *bdd)
225 if (zfs_get_type(zfs_hdl) != ZFS_TYPE_SNAPSHOT)
229 * If we're deleting a snapshot, we need to make sure we only promote
230 * clones that are derived from one of the snapshots we're deleting,
231 * rather than that of a snapshot we're not touching. This keeps stuff
232 * in a consistent state, making sure that we don't error out unless
235 if (bdd->snapname == NULL)
238 atpos = strchr(zfs_get_name(zfs_hdl), '@');
239 return (strcmp(atpos + 1, bdd->snapname) == 0);
243 * This is executed from be_promote_dependent_clones via zfs_iter_dependents,
244 * It checks if the dependent type is a snapshot then attempts to find any
245 * clones associated with it. Any clones not related to the destroy target are
246 * added to the promote list.
249 be_dependent_clone_cb(zfs_handle_t *zfs_hdl, void *data)
256 struct be_destroy_data *bdd;
257 struct promote_entry *entry, *newentry;
261 bdd = (struct be_destroy_data *)data;
263 if (be_should_promote_clones(zfs_hdl, bdd) &&
264 (nvl = zfs_get_clones_nvl(zfs_hdl)) != NULL) {
265 while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
266 name = nvpair_name(nvp);
269 * Skip if the clone is equal to, or a child of, the
272 if (strncmp(name, bdd->target_name,
273 strlen(bdd->target_name)) == 0 ||
274 strstr(name, bdd->target_name) == name) {
279 SLIST_FOREACH(entry, &bdd->promotelist, link) {
280 if (strcmp(entry->name, name) == 0) {
289 newentry = malloc(sizeof(struct promote_entry));
290 if (newentry == NULL) {
295 #define BE_COPY_NAME(entry, src) \
296 strlcpy((entry)->name, (src), sizeof((entry)->name))
297 if (BE_COPY_NAME(newentry, name) >=
298 sizeof(newentry->name)) {
299 /* Shouldn't happen. */
307 * We're building up a SLIST here to make sure both that
308 * we get the order right and so that we don't
309 * inadvertently observe the wrong state by promoting
310 * datasets while we're still walking the tree. The
311 * latter can lead to situations where we promote a BE
312 * then effectively demote it again.
314 SLIST_INSERT_HEAD(&bdd->promotelist, newentry, link);
323 * This is called before a destroy, so that any datasets(environments) that are
324 * dependent on this one get promoted before destroying the target.
327 be_promote_dependent_clones(zfs_handle_t *zfs_hdl, struct be_destroy_data *bdd)
331 struct promote_entry *entry;
333 snprintf(bdd->target_name, BE_MAXPATHLEN, "%s/", zfs_get_name(zfs_hdl));
334 err = zfs_iter_dependents(zfs_hdl, true, be_dependent_clone_cb, bdd);
337 * Drain the list and walk away from it if we're only deleting a
340 if (bdd->snapname != NULL && !SLIST_EMPTY(&bdd->promotelist))
341 err = BE_ERR_HASCLONES;
342 while (!SLIST_EMPTY(&bdd->promotelist)) {
343 entry = SLIST_FIRST(&bdd->promotelist);
344 SLIST_REMOVE_HEAD(&bdd->promotelist, link);
346 #define ZFS_GRAB_CLONE() \
347 zfs_open(bdd->lbh->lzh, entry->name, ZFS_TYPE_FILESYSTEM)
349 * Just skip this part on error, we still want to clean up the
350 * promotion list after the first error. We'll then preserve it
353 if (err == 0 && (clone = ZFS_GRAB_CLONE()) != NULL) {
354 err = zfs_promote(clone);
356 err = BE_ERR_DESTROYMNT;
359 #undef ZFS_GRAB_CLONE
367 be_destroy_cb(zfs_handle_t *zfs_hdl, void *data)
369 char path[BE_MAXPATHLEN];
370 struct be_destroy_data *bdd;
374 bdd = (struct be_destroy_data *)data;
375 if (bdd->snapname == NULL) {
376 err = zfs_iter_children(zfs_hdl, be_destroy_cb, data);
379 return (zfs_destroy(zfs_hdl, false));
381 /* If we're dealing with snapshots instead, delete that one alone */
382 err = zfs_iter_filesystems(zfs_hdl, be_destroy_cb, data);
386 * This part is intentionally glossing over any potential errors,
387 * because there's a lot less potential for errors when we're cleaning
388 * up snapshots rather than a full deep BE. The primary error case
389 * here being if the snapshot doesn't exist in the first place, which
390 * the caller will likely deem insignificant as long as it doesn't
391 * exist after the call. Thus, such a missing snapshot shouldn't jam
392 * up the destruction.
394 snprintf(path, sizeof(path), "%s@%s", zfs_get_name(zfs_hdl),
396 if (!zfs_dataset_exists(bdd->lbh->lzh, path, ZFS_TYPE_SNAPSHOT))
398 snap = zfs_open(bdd->lbh->lzh, path, ZFS_TYPE_SNAPSHOT);
400 zfs_destroy(snap, false);
404 #define BE_DESTROY_WANTORIGIN (BE_DESTROY_ORIGIN | BE_DESTROY_AUTOORIGIN)
406 * Destroy the boot environment or snapshot specified by the name
407 * parameter. Options are or'd together with the possible values:
408 * BE_DESTROY_FORCE : forces operation on mounted datasets
409 * BE_DESTROY_ORIGIN: destroy the origin snapshot as well
412 be_destroy_internal(libbe_handle_t *lbh, const char *name, int options,
415 struct be_destroy_data bdd;
416 char origin[BE_MAXPATHLEN], path[BE_MAXPATHLEN];
419 int err, force, mounted;
424 SLIST_INIT(&bdd.promotelist);
425 force = options & BE_DESTROY_FORCE;
428 be_root_concat(lbh, name, path);
430 if ((snapdelim = strchr(path, '@')) == NULL) {
431 if (!zfs_dataset_exists(lbh->lzh, path, ZFS_TYPE_FILESYSTEM))
432 return (set_error(lbh, BE_ERR_NOENT));
434 if (strcmp(path, lbh->rootfs) == 0 ||
435 strcmp(path, lbh->bootfs) == 0)
436 return (set_error(lbh, BE_ERR_DESTROYACT));
438 fs = zfs_open(lbh->lzh, path, ZFS_TYPE_FILESYSTEM);
440 return (set_error(lbh, BE_ERR_ZFSOPEN));
442 /* Don't destroy a mounted dataset unless force is specified */
443 if ((mounted = zfs_is_mounted(fs, NULL)) != 0) {
445 zfs_unmount(fs, NULL, 0);
448 return (set_error(lbh, BE_ERR_DESTROYMNT));
452 /* Handle destroying bootonce */
453 if (lbh->bootonce != NULL &&
454 strcmp(path, lbh->bootonce) == 0)
455 (void) lzbe_set_boot_device(
456 zpool_get_name(lbh->active_phandle), lzbe_add, NULL);
459 * If we're initially destroying a snapshot, origin options do
460 * not make sense. If we're destroying the origin snapshot of
461 * a BE, we want to maintain the options in case we need to
462 * fake success after failing to promote.
465 options &= ~BE_DESTROY_WANTORIGIN;
466 if (!zfs_dataset_exists(lbh->lzh, path, ZFS_TYPE_SNAPSHOT))
467 return (set_error(lbh, BE_ERR_NOENT));
469 bdd.snapname = strdup(snapdelim + 1);
470 if (bdd.snapname == NULL)
471 return (set_error(lbh, BE_ERR_NOMEM));
473 fs = zfs_open(lbh->lzh, path, ZFS_TYPE_DATASET);
476 return (set_error(lbh, BE_ERR_ZFSOPEN));
481 * Whether we're destroying a BE or a single snapshot, we need to walk
482 * the tree of what we're going to destroy and promote everything in our
483 * path so that we can make it happen.
485 if ((err = be_promote_dependent_clones(fs, &bdd)) != 0) {
489 * If we're just destroying the origin of some other dataset
490 * we were invoked to destroy, then we just ignore
491 * BE_ERR_HASCLONES and return success unless the caller wanted
492 * to force the issue.
494 if (odestroyer && err == BE_ERR_HASCLONES &&
495 (options & BE_DESTROY_AUTOORIGIN) != 0)
497 return (set_error(lbh, err));
501 * This was deferred until after we promote all of the derivatives so
502 * that we grab the new origin after everything's settled down.
504 if ((options & BE_DESTROY_WANTORIGIN) != 0 &&
505 zfs_prop_get(fs, ZFS_PROP_ORIGIN, origin, sizeof(origin),
506 NULL, NULL, 0, 1) != 0 &&
507 (options & BE_DESTROY_ORIGIN) != 0)
508 return (set_error(lbh, BE_ERR_NOORIGIN));
511 * If the caller wants auto-origin destruction and the origin
512 * name matches one of our automatically created snapshot names
513 * (i.e. strftime("%F-%T") with a serial at the end), then
514 * we'll set the DESTROY_ORIGIN flag and nuke it
515 * be_is_auto_snapshot_name is exported from libbe(3) so that
516 * the caller can determine if it needs to warn about the origin
517 * not being destroyed or not.
519 if ((options & BE_DESTROY_AUTOORIGIN) != 0 && *origin != '\0' &&
520 be_is_auto_snapshot_name(lbh, origin))
521 options |= BE_DESTROY_ORIGIN;
523 err = be_destroy_cb(fs, &bdd);
527 /* Children are still present or the mount is referenced */
529 return (set_error(lbh, BE_ERR_DESTROYMNT));
530 return (set_error(lbh, BE_ERR_UNKNOWN));
533 if ((options & BE_DESTROY_ORIGIN) == 0)
536 /* The origin can't possibly be shorter than the BE root */
537 rootlen = strlen(lbh->root);
538 if (*origin == '\0' || strlen(origin) <= rootlen + 1)
539 return (set_error(lbh, BE_ERR_INVORIGIN));
542 * We'll be chopping off the BE root and running this back through
543 * be_destroy, so that we properly handle the origin snapshot whether
544 * it be that of a deep BE or not.
546 if (strncmp(origin, lbh->root, rootlen) != 0 || origin[rootlen] != '/')
549 return (be_destroy_internal(lbh, origin + rootlen + 1,
550 options & ~BE_DESTROY_ORIGIN, true));
554 be_destroy(libbe_handle_t *lbh, const char *name, int options)
558 * The consumer must not set both BE_DESTROY_AUTOORIGIN and
559 * BE_DESTROY_ORIGIN. Internally, we'll set the latter from the former.
560 * The latter should imply that we must succeed at destroying the
561 * origin, or complain otherwise.
563 if ((options & BE_DESTROY_WANTORIGIN) == BE_DESTROY_WANTORIGIN)
564 return (set_error(lbh, BE_ERR_UNKNOWN));
565 return (be_destroy_internal(lbh, name, options, false));
569 be_setup_snapshot_name(libbe_handle_t *lbh, char *buf, size_t buflen)
576 len += strftime(buf + len, buflen - len, "@%F-%T", localtime(&rawtime));
577 /* No room for serial... caller will do its best */
578 if (buflen - len < 2)
581 for (serial = 0; serial < BE_SNAP_SERIAL_MAX; ++serial) {
582 snprintf(buf + len, buflen - len, "-%d", serial);
583 if (!zfs_dataset_exists(lbh->lzh, buf, ZFS_TYPE_SNAPSHOT))
589 be_is_auto_snapshot_name(libbe_handle_t *lbh __unused, const char *name)
592 int day, hour, minute, month, second, serial, year;
594 if ((snap = strchr(name, '@')) == NULL)
597 /* We'll grab the individual components and do some light validation. */
598 if (sscanf(snap, "%d-%d-%d-%d:%d:%d-%d", &year, &month, &day, &hour,
599 &minute, &second, &serial) != 7)
601 return (year >= 1970) && (month >= 1 && month <= 12) &&
602 (day >= 1 && day <= 31) && (hour >= 0 && hour <= 23) &&
603 (minute >= 0 && minute <= 59) && (second >= 0 && second <= 60) &&
608 be_snapshot(libbe_handle_t *lbh, const char *source, const char *snap_name,
609 bool recursive, char *result)
611 char buf[BE_MAXPATHLEN];
614 be_root_concat(lbh, source, buf);
616 if ((err = be_exists(lbh, buf)) != 0)
617 return (set_error(lbh, err));
619 if (snap_name != NULL) {
620 if (strlcat(buf, "@", sizeof(buf)) >= sizeof(buf))
621 return (set_error(lbh, BE_ERR_INVALIDNAME));
623 if (strlcat(buf, snap_name, sizeof(buf)) >= sizeof(buf))
624 return (set_error(lbh, BE_ERR_INVALIDNAME));
627 snprintf(result, BE_MAXPATHLEN, "%s@%s", source,
630 be_setup_snapshot_name(lbh, buf, sizeof(buf));
632 if (result != NULL && strlcpy(result, strrchr(buf, '/') + 1,
633 sizeof(buf)) >= sizeof(buf))
634 return (set_error(lbh, BE_ERR_INVALIDNAME));
636 if ((err = zfs_snapshot(lbh->lzh, buf, recursive, NULL)) != 0) {
638 case EZFS_INVALIDNAME:
639 return (set_error(lbh, BE_ERR_INVALIDNAME));
643 * The other errors that zfs_ioc_snapshot might return
644 * shouldn't happen if we've set things up properly, so
645 * we'll gloss over them and call it UNKNOWN as it will
646 * require further triage.
648 if (errno == ENOTSUP)
649 return (set_error(lbh, BE_ERR_NOPOOL));
650 return (set_error(lbh, BE_ERR_UNKNOWN));
654 return (BE_ERR_SUCCESS);
659 * Create the boot environment specified by the name parameter
662 be_create(libbe_handle_t *lbh, const char *name)
666 err = be_create_from_existing(lbh, name, be_active_path(lbh));
668 return (set_error(lbh, err));
672 be_deep_clone_prop(int prop, void *cb)
675 struct libbe_dccb *dccb;
677 char pval[BE_MAXPATHLEN];
678 char source[BE_MAXPATHLEN];
682 /* Skip some properties we don't want to touch */
683 if (prop == ZFS_PROP_CANMOUNT)
686 /* Don't copy readonly properties */
687 if (zfs_prop_readonly(prop))
690 if ((err = zfs_prop_get(dccb->zhp, prop, (char *)&pval,
691 sizeof(pval), &src, (char *)&source, sizeof(source), false)))
692 /* Just continue if we fail to read a property */
696 * Only copy locally defined or received properties. This continues
697 * to avoid temporary/default/local properties intentionally without
698 * breaking received datasets.
700 if (src != ZPROP_SRC_LOCAL && src != ZPROP_SRC_RECEIVED)
703 /* Augment mountpoint with altroot, if needed */
705 if (prop == ZFS_PROP_MOUNTPOINT)
706 val = be_mountpoint_augmented(dccb->lbh, val);
708 nvlist_add_string(dccb->props, zfs_prop_to_name(prop), val);
714 * Return the corresponding boot environment path for a given
715 * dataset path, the constructed path is placed in 'result'.
717 * example: say our new boot environment name is 'bootenv' and
718 * the dataset path is 'zroot/ROOT/default/data/set'.
720 * result should produce: 'zroot/ROOT/bootenv/data/set'
723 be_get_path(struct libbe_deep_clone *ldc, const char *dspath, char *result, int result_size)
728 /* match the root path for the boot environments */
729 pos = strstr(dspath, ldc->lbh->root);
731 /* no match, different pools? */
733 return (BE_ERR_BADPATH);
735 /* root path of the new boot environment */
736 snprintf(result, result_size, "%s/%s", ldc->lbh->root, ldc->bename);
738 /* gets us to the parent dataset, the +1 consumes a trailing slash */
739 pos += strlen(ldc->lbh->root) + 1;
741 /* skip the parent dataset */
742 if ((child_dataset = strchr(pos, '/')) != NULL)
743 strlcat(result, child_dataset, result_size);
745 return (BE_ERR_SUCCESS);
749 be_clone_cb(zfs_handle_t *ds, void *data)
752 char be_path[BE_MAXPATHLEN];
753 char snap_path[BE_MAXPATHLEN];
755 zfs_handle_t *snap_hdl;
757 struct libbe_deep_clone *ldc;
758 struct libbe_dccb dccb;
760 ldc = (struct libbe_deep_clone *)data;
761 dspath = zfs_get_name(ds);
763 snprintf(snap_path, sizeof(snap_path), "%s@%s", dspath, ldc->snapname);
765 /* construct the boot environment path from the dataset we're cloning */
766 if (be_get_path(ldc, dspath, be_path, sizeof(be_path)) != BE_ERR_SUCCESS)
767 return (BE_ERR_UNKNOWN);
769 /* the dataset to be created (i.e. the boot environment) already exists */
770 if (zfs_dataset_exists(ldc->lbh->lzh, be_path, ZFS_TYPE_DATASET))
771 return (BE_ERR_EXISTS);
773 /* no snapshot found for this dataset, silently skip it */
774 if (!zfs_dataset_exists(ldc->lbh->lzh, snap_path, ZFS_TYPE_SNAPSHOT))
778 zfs_open(ldc->lbh->lzh, snap_path, ZFS_TYPE_SNAPSHOT)) == NULL)
779 return (BE_ERR_ZFSOPEN);
781 nvlist_alloc(&props, NV_UNIQUE_NAME, KM_SLEEP);
782 nvlist_add_string(props, "canmount", "noauto");
787 if (zprop_iter(be_deep_clone_prop, &dccb, B_FALSE, B_FALSE,
788 ZFS_TYPE_FILESYSTEM) == ZPROP_INVAL)
791 if ((err = zfs_clone(snap_hdl, be_path, props)) != 0)
792 return (BE_ERR_ZFSCLONE);
797 if (ldc->depth_limit == -1 || ldc->depth < ldc->depth_limit) {
799 err = zfs_iter_filesystems(ds, be_clone_cb, ldc);
807 * Create a boot environment with a given name from a given snapshot.
808 * Snapshots can be in the format 'zroot/ROOT/default@snapshot' or
809 * 'default@snapshot'. In the latter case, 'default@snapshot' will be prepended
810 * with the root path that libbe was initailized with.
813 be_clone(libbe_handle_t *lbh, const char *bename, const char *snapshot, int depth)
816 char snap_path[BE_MAXPATHLEN];
817 char *parentname, *snapname;
818 zfs_handle_t *parent_hdl;
819 struct libbe_deep_clone ldc;
821 /* ensure the boot environment name is valid */
822 if ((err = be_validate_name(lbh, bename)) != 0)
823 return (set_error(lbh, err));
826 * prepend the boot environment root path if we're
827 * given a partial snapshot name.
829 if ((err = be_root_concat(lbh, snapshot, snap_path)) != 0)
830 return (set_error(lbh, err));
832 /* ensure the snapshot exists */
833 if ((err = be_validate_snap(lbh, snap_path)) != 0)
834 return (set_error(lbh, err));
836 /* get a copy of the snapshot path so we can disect it */
837 if ((parentname = strdup(snap_path)) == NULL)
838 return (set_error(lbh, BE_ERR_UNKNOWN));
840 /* split dataset name from snapshot name */
841 snapname = strchr(parentname, '@');
842 if (snapname == NULL) {
844 return (set_error(lbh, BE_ERR_UNKNOWN));
849 /* set-up the boot environment */
852 ldc.snapname = snapname;
854 ldc.depth_limit = depth;
856 /* the boot environment will be cloned from this dataset */
857 parent_hdl = zfs_open(lbh->lzh, parentname, ZFS_TYPE_DATASET);
859 /* create the boot environment */
860 err = be_clone_cb(parent_hdl, &ldc);
863 return (set_error(lbh, err));
867 * Create a boot environment from pre-existing snapshot, specifying a depth.
869 int be_create_depth(libbe_handle_t *lbh, const char *bename,
870 const char *snap, int depth)
872 return (be_clone(lbh, bename, snap, depth));
876 * Create the boot environment from pre-existing snapshot
879 be_create_from_existing_snap(libbe_handle_t *lbh, const char *bename,
882 return (be_clone(lbh, bename, snap, -1));
887 * Create a boot environment from an existing boot environment
890 be_create_from_existing(libbe_handle_t *lbh, const char *bename, const char *old)
893 char snap[BE_MAXPATHLEN];
895 if ((err = be_snapshot(lbh, old, NULL, true, snap)) != 0)
896 return (set_error(lbh, err));
898 err = be_clone(lbh, bename, snap, -1);
900 return (set_error(lbh, err));
905 * Verifies that a snapshot has a valid name, exists, and has a mountpoint of
906 * '/'. Returns BE_ERR_SUCCESS (0), upon success, or the relevant BE_ERR_* upon
907 * failure. Does not set the internal library error state.
910 be_validate_snap(libbe_handle_t *lbh, const char *snap_name)
913 if (strlen(snap_name) >= BE_MAXPATHLEN)
914 return (BE_ERR_PATHLEN);
916 if (!zfs_name_valid(snap_name, ZFS_TYPE_SNAPSHOT))
917 return (BE_ERR_INVALIDNAME);
919 if (!zfs_dataset_exists(lbh->lzh, snap_name,
921 return (BE_ERR_NOENT);
923 return (BE_ERR_SUCCESS);
928 * Idempotently appends the name argument to the root boot environment path
929 * and copies the resulting string into the result buffer (which is assumed
930 * to be at least BE_MAXPATHLEN characters long. Returns BE_ERR_SUCCESS upon
931 * success, BE_ERR_PATHLEN if the resulting path is longer than BE_MAXPATHLEN,
932 * or BE_ERR_INVALIDNAME if the name is a path that does not begin with
933 * zfs_be_root. Does not set internal library error state.
936 be_root_concat(libbe_handle_t *lbh, const char *name, char *result)
938 size_t name_len, root_len;
940 name_len = strlen(name);
941 root_len = strlen(lbh->root);
943 /* Act idempotently; return be name if it is already a full path */
944 if (strrchr(name, '/') != NULL) {
945 if (strstr(name, lbh->root) != name)
946 return (BE_ERR_INVALIDNAME);
948 if (name_len >= BE_MAXPATHLEN)
949 return (BE_ERR_PATHLEN);
951 strlcpy(result, name, BE_MAXPATHLEN);
952 return (BE_ERR_SUCCESS);
953 } else if (name_len + root_len + 1 < BE_MAXPATHLEN) {
954 snprintf(result, BE_MAXPATHLEN, "%s/%s", lbh->root,
956 return (BE_ERR_SUCCESS);
959 return (BE_ERR_PATHLEN);
964 * Verifies the validity of a boot environment name (A-Za-z0-9-_.). Returns
965 * BE_ERR_SUCCESS (0) if name is valid, otherwise returns BE_ERR_INVALIDNAME
967 * Does not set internal library error state.
970 be_validate_name(libbe_handle_t *lbh, const char *name)
974 * Impose the additional restriction that the entire dataset name must
975 * not exceed the maximum length of a dataset, i.e. MAXNAMELEN.
977 if (strlen(lbh->root) + 1 + strlen(name) > MAXNAMELEN)
978 return (BE_ERR_PATHLEN);
980 if (!zfs_name_valid(name, ZFS_TYPE_DATASET))
981 return (BE_ERR_INVALIDNAME);
984 * ZFS allows spaces in boot environment names, but the kernel can't
985 * handle booting from such a dataset right now. vfs.root.mountfrom
986 * is defined to be a space-separated list, and there's no protocol for
987 * escaping whitespace in the path component of a dev:path spec. So
988 * while loader can handle this situation alright, it can't safely pass
989 * it on to mountroot.
991 if (strchr(name, ' ') != NULL)
992 return (BE_ERR_INVALIDNAME);
994 return (BE_ERR_SUCCESS);
1002 be_rename(libbe_handle_t *lbh, const char *old, const char *new)
1004 char full_old[BE_MAXPATHLEN];
1005 char full_new[BE_MAXPATHLEN];
1006 zfs_handle_t *zfs_hdl;
1010 * be_validate_name is documented not to set error state, so we should
1013 if ((err = be_validate_name(lbh, new)) != 0)
1014 return (set_error(lbh, err));
1015 if ((err = be_root_concat(lbh, old, full_old)) != 0)
1016 return (set_error(lbh, err));
1017 if ((err = be_root_concat(lbh, new, full_new)) != 0)
1018 return (set_error(lbh, err));
1020 if (!zfs_dataset_exists(lbh->lzh, full_old, ZFS_TYPE_DATASET))
1021 return (set_error(lbh, BE_ERR_NOENT));
1023 if (zfs_dataset_exists(lbh->lzh, full_new, ZFS_TYPE_DATASET))
1024 return (set_error(lbh, BE_ERR_EXISTS));
1026 if ((zfs_hdl = zfs_open(lbh->lzh, full_old,
1027 ZFS_TYPE_FILESYSTEM)) == NULL)
1028 return (set_error(lbh, BE_ERR_ZFSOPEN));
1030 /* recurse, nounmount, forceunmount */
1031 struct renameflags flags = {
1034 err = zfs_rename(zfs_hdl, full_new, flags);
1038 /* handle renaming bootonce */
1039 if (lbh->bootonce != NULL &&
1040 strcmp(full_old, lbh->bootonce) == 0)
1041 err = be_activate(lbh, new, true);
1045 return (set_error(lbh, err));
1050 be_export(libbe_handle_t *lbh, const char *bootenv, int fd)
1052 char snap_name[BE_MAXPATHLEN];
1053 char buf[BE_MAXPATHLEN];
1055 sendflags_t flags = { 0 };
1058 if ((err = be_snapshot(lbh, bootenv, NULL, true, snap_name)) != 0)
1059 /* Use the error set by be_snapshot */
1062 be_root_concat(lbh, snap_name, buf);
1064 if ((zfs = zfs_open(lbh->lzh, buf, ZFS_TYPE_DATASET)) == NULL)
1065 return (set_error(lbh, BE_ERR_ZFSOPEN));
1067 err = zfs_send_one(zfs, NULL, fd, &flags, /* redactbook */ NULL);
1075 be_import(libbe_handle_t *lbh, const char *bootenv, int fd)
1077 char buf[BE_MAXPATHLEN];
1080 recvflags_t flags = { .nomount = 1 };
1083 be_root_concat(lbh, bootenv, buf);
1085 if ((err = zfs_receive(lbh->lzh, buf, NULL, &flags, fd, NULL)) != 0) {
1088 return (set_error(lbh, BE_ERR_NOORIGIN));
1090 return (set_error(lbh, BE_ERR_NOENT));
1092 return (set_error(lbh, BE_ERR_IO));
1094 return (set_error(lbh, BE_ERR_UNKNOWN));
1098 if ((zfs = zfs_open(lbh->lzh, buf, ZFS_TYPE_FILESYSTEM)) == NULL)
1099 return (set_error(lbh, BE_ERR_ZFSOPEN));
1101 nvlist_alloc(&props, NV_UNIQUE_NAME, KM_SLEEP);
1102 nvlist_add_string(props, "canmount", "noauto");
1103 nvlist_add_string(props, "mountpoint", "none");
1105 err = zfs_prop_set_list(zfs, props);
1111 return (set_error(lbh, BE_ERR_UNKNOWN));
1118 be_create_child_noent(libbe_handle_t *lbh, const char *active,
1119 const char *child_path)
1125 nvlist_alloc(&props, NV_UNIQUE_NAME, KM_SLEEP);
1126 nvlist_add_string(props, "canmount", "noauto");
1127 nvlist_add_string(props, "mountpoint", child_path);
1130 if ((err = zfs_create(lbh->lzh, active, ZFS_TYPE_DATASET,
1134 return (set_error(lbh, BE_ERR_EXISTS));
1136 return (set_error(lbh, BE_ERR_NOENT));
1138 case EZFS_BADVERSION:
1139 return (set_error(lbh, BE_ERR_NOPOOL));
1142 /* We set something up wrong, probably... */
1143 return (set_error(lbh, BE_ERR_UNKNOWN));
1148 if ((zfs = zfs_open(lbh->lzh, active, ZFS_TYPE_DATASET)) == NULL)
1149 return (set_error(lbh, BE_ERR_ZFSOPEN));
1152 if ((err = zfs_prop_set(zfs, "canmount", "noauto")) != 0) {
1155 * Similar to other cases, this shouldn't fail unless we've
1156 * done something wrong. This is a new dataset that shouldn't
1157 * have been mounted anywhere between creation and now.
1159 if (err == EZFS_NOMEM)
1160 return (set_error(lbh, BE_ERR_NOMEM));
1161 return (set_error(lbh, BE_ERR_UNKNOWN));
1164 return (BE_ERR_SUCCESS);
1168 be_create_child_cloned(libbe_handle_t *lbh, const char *active)
1170 char buf[BE_MAXPATHLEN], tmp[BE_MAXPATHLEN];;
1177 * Establish if the existing path is a zfs dataset or just
1178 * the subdirectory of one
1180 strlcpy(tmp, "tmp/be_snap.XXXXX", sizeof(tmp));
1181 if (mktemp(tmp) == NULL)
1182 return (set_error(lbh, BE_ERR_UNKNOWN));
1184 be_root_concat(lbh, tmp, buf);
1185 printf("Here %s?\n", buf);
1186 if ((err = zfs_snapshot(lbh->lzh, buf, false, NULL)) != 0) {
1188 case EZFS_INVALIDNAME:
1189 return (set_error(lbh, BE_ERR_INVALIDNAME));
1193 * The other errors that zfs_ioc_snapshot might return
1194 * shouldn't happen if we've set things up properly, so
1195 * we'll gloss over them and call it UNKNOWN as it will
1196 * require further triage.
1198 if (errno == ENOTSUP)
1199 return (set_error(lbh, BE_ERR_NOPOOL));
1200 return (set_error(lbh, BE_ERR_UNKNOWN));
1205 if ((zfs = zfs_open(lbh->lzh, buf, ZFS_TYPE_SNAPSHOT)) == NULL)
1206 return (BE_ERR_ZFSOPEN);
1208 if ((err = zfs_clone(zfs, active, NULL)) != 0)
1209 /* XXX TODO correct error */
1210 return (set_error(lbh, BE_ERR_UNKNOWN));
1214 return (BE_ERR_SUCCESS);
1218 be_add_child(libbe_handle_t *lbh, const char *child_path, bool cp_if_exists)
1221 char active[BE_MAXPATHLEN], buf[BE_MAXPATHLEN];
1225 /* Require absolute paths */
1226 if (*child_path != '/')
1227 return (set_error(lbh, BE_ERR_BADPATH));
1229 strlcpy(active, be_active_path(lbh), BE_MAXPATHLEN);
1230 strcpy(buf, active);
1232 /* Create non-mountable parent dataset(s) */
1234 for (char *p; (p = strchr(s+1, '/')) != NULL; s = p) {
1236 strncat(buf, s, len);
1238 nvlist_alloc(&props, NV_UNIQUE_NAME, KM_SLEEP);
1239 nvlist_add_string(props, "canmount", "off");
1240 nvlist_add_string(props, "mountpoint", "none");
1241 zfs_create(lbh->lzh, buf, ZFS_TYPE_DATASET, props);
1245 /* Path does not exist as a descendent of / yet */
1246 if (strlcat(active, child_path, BE_MAXPATHLEN) >= BE_MAXPATHLEN)
1247 return (set_error(lbh, BE_ERR_PATHLEN));
1249 if (stat(child_path, &sb) != 0) {
1250 /* Verify that error is ENOENT */
1251 if (errno != ENOENT)
1252 return (set_error(lbh, BE_ERR_UNKNOWN));
1253 return (be_create_child_noent(lbh, active, child_path));
1254 } else if (cp_if_exists)
1255 /* Path is already a descendent of / and should be copied */
1256 return (be_create_child_cloned(lbh, active));
1257 return (set_error(lbh, BE_ERR_EXISTS));
1262 * Deactivate old BE dataset; currently just sets canmount=noauto or
1263 * resets boot once configuration.
1266 be_deactivate(libbe_handle_t *lbh, const char *ds, bool temporary)
1271 return (lzbe_set_boot_device(
1272 zpool_get_name(lbh->active_phandle), lzbe_add, NULL));
1275 if ((zfs = zfs_open(lbh->lzh, ds, ZFS_TYPE_DATASET)) == NULL)
1277 if (zfs_prop_set(zfs, "canmount", "noauto") != 0)
1284 be_zfs_promote_cb(zfs_handle_t *zhp, void *data)
1286 char origin[BE_MAXPATHLEN];
1287 bool *found_origin = (bool *)data;
1290 if (zfs_prop_get(zhp, ZFS_PROP_ORIGIN, origin, sizeof(origin),
1291 NULL, NULL, 0, true) == 0) {
1292 *found_origin = true;
1293 err = zfs_promote(zhp);
1298 return (zfs_iter_filesystems(zhp, be_zfs_promote_cb, data));
1302 be_zfs_promote(zfs_handle_t *zhp, bool *found_origin)
1304 *found_origin = false;
1305 return (be_zfs_promote_cb(zhp, (void *)found_origin));
1309 be_activate(libbe_handle_t *lbh, const char *bootenv, bool temporary)
1311 char be_path[BE_MAXPATHLEN];
1316 be_root_concat(lbh, bootenv, be_path);
1318 /* Note: be_exists fails if mountpoint is not / */
1319 if ((err = be_exists(lbh, be_path)) != 0)
1320 return (set_error(lbh, err));
1323 return (lzbe_set_boot_device(
1324 zpool_get_name(lbh->active_phandle), lzbe_add, be_path));
1326 if (strncmp(lbh->bootfs, "-", 1) != 0 &&
1327 be_deactivate(lbh, lbh->bootfs, false) != 0)
1330 /* Obtain bootenv zpool */
1331 err = zpool_set_prop(lbh->active_phandle, "bootfs", be_path);
1336 zhp = zfs_open(lbh->lzh, be_path, ZFS_TYPE_FILESYSTEM);
1340 err = be_zfs_promote(zhp, &found_origin);
1353 return (BE_ERR_SUCCESS);