1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
24 */
25
26 /*
27 * ZFS control directory (a.k.a. ".zfs")
28 *
29 * This directory provides a common location for all ZFS meta-objects.
30 * Currently, this is only the 'snapshot' directory, but this may expand in the
31 * future. The elements are built using the GFS primitives, as the hierarchy
32 * does not actually exist on disk.
33 *
34 * For 'snapshot', we don't want to have all snapshots always mounted, because
35 * this would take up a huge amount of space in /etc/mnttab. We have three
36 * types of objects:
37 *
38 * ctldir ------> snapshotdir -------> snapshot
39 * |
40 * |
41 * V
42 * mounted fs
43 *
44 * The 'snapshot' node contains just enough information to lookup '..' and act
45 * as a mountpoint for the snapshot. Whenever we lookup a specific snapshot, we
46 * perform an automount of the underlying filesystem and return the
47 * corresponding vnode.
48 *
49 * All mounts are handled automatically by the kernel, but unmounts are
50 * (currently) handled from user land. The main reason is that there is no
51 * reliable way to auto-unmount the filesystem when it's "no longer in use".
52 * When the user unmounts a filesystem, we call zfsctl_unmount(), which
53 * unmounts any snapshots within the snapshot directory.
54 *
55 * The '.zfs', '.zfs/snapshot', and all directories created under
56 * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') are all GFS nodes and
57 * share the same vfs_t as the head filesystem (what '.zfs' lives under).
58 *
59 * File systems mounted ontop of the GFS nodes '.zfs/snapshot/<snapname>'
60 * (ie: snapshots) are ZFS nodes and have their own unique vfs_t.
61 * However, vnodes within these mounted on file systems have their v_vfsp
62 * fields set to the head filesystem to make NFS happy (see
63 * zfsctl_snapdir_lookup()). We VFS_HOLD the head filesystem's vfs_t
64 * so that it cannot be freed until all snapshots have been unmounted.
65 */
66
67 #include <fs/fs_subr.h>
68 #include <sys/zfs_ctldir.h>
69 #include <sys/zfs_ioctl.h>
70 #include <sys/zfs_vfsops.h>
71 #include <sys/vfs_opreg.h>
72 #include <sys/gfs.h>
73 #include <sys/stat.h>
74 #include <sys/dmu.h>
75 #include <sys/dsl_deleg.h>
76 #include <sys/mount.h>
77 #include <sys/sunddi.h>
78
79 #include "zfs_namecheck.h"
80
81 typedef struct zfsctl_node {
82 gfs_dir_t zc_gfs_private;
83 uint64_t zc_id;
84 timestruc_t zc_cmtime; /* ctime and mtime, always the same */
85 } zfsctl_node_t;
86
87 typedef struct zfsctl_snapdir {
88 zfsctl_node_t sd_node;
89 kmutex_t sd_lock;
90 avl_tree_t sd_snaps;
91 } zfsctl_snapdir_t;
92
93 typedef struct {
94 char *se_name;
95 vnode_t *se_root;
96 avl_node_t se_node;
97 } zfs_snapentry_t;
98
99 static int
100 snapentry_compare(const void *a, const void *b)
101 {
102 const zfs_snapentry_t *sa = a;
103 const zfs_snapentry_t *sb = b;
104 int ret = strcmp(sa->se_name, sb->se_name);
105
106 if (ret < 0)
107 return (-1);
108 else if (ret > 0)
109 return (1);
110 else
111 return (0);
112 }
113
114 vnodeops_t *zfsctl_ops_root;
115 vnodeops_t *zfsctl_ops_snapdir;
116 vnodeops_t *zfsctl_ops_snapshot;
117 vnodeops_t *zfsctl_ops_shares;
118 vnodeops_t *zfsctl_ops_shares_dir;
119
120 static const fs_operation_def_t zfsctl_tops_root[];
121 static const fs_operation_def_t zfsctl_tops_snapdir[];
122 static const fs_operation_def_t zfsctl_tops_snapshot[];
123 static const fs_operation_def_t zfsctl_tops_shares[];
124
125 static vnode_t *zfsctl_mknode_snapdir(vnode_t *);
126 static vnode_t *zfsctl_mknode_shares(vnode_t *);
127 static vnode_t *zfsctl_snapshot_mknode(vnode_t *, uint64_t objset);
128 static int zfsctl_unmount_snap(zfs_snapentry_t *, int, cred_t *);
129
130 static gfs_opsvec_t zfsctl_opsvec[] = {
131 { ".zfs", zfsctl_tops_root, &zfsctl_ops_root },
132 { ".zfs/snapshot", zfsctl_tops_snapdir, &zfsctl_ops_snapdir },
133 { ".zfs/snapshot/vnode", zfsctl_tops_snapshot, &zfsctl_ops_snapshot },
134 { ".zfs/shares", zfsctl_tops_shares, &zfsctl_ops_shares_dir },
135 { ".zfs/shares/vnode", zfsctl_tops_shares, &zfsctl_ops_shares },
136 { NULL }
137 };
138
139 /*
140 * Root directory elements. We only have two entries
141 * snapshot and shares.
142 */
143 static gfs_dirent_t zfsctl_root_entries[] = {
144 { "snapshot", zfsctl_mknode_snapdir, GFS_CACHE_VNODE },
145 { "shares", zfsctl_mknode_shares, GFS_CACHE_VNODE },
146 { NULL }
147 };
148
149 /* include . and .. in the calculation */
150 #define NROOT_ENTRIES ((sizeof (zfsctl_root_entries) / \
151 sizeof (gfs_dirent_t)) + 1)
152
153
154 /*
155 * Initialize the various GFS pieces we'll need to create and manipulate .zfs
156 * directories. This is called from the ZFS init routine, and initializes the
157 * vnode ops vectors that we'll be using.
158 */
159 void
160 zfsctl_init(void)
161 {
162 VERIFY(gfs_make_opsvec(zfsctl_opsvec) == 0);
163 }
164
165 void
166 zfsctl_fini(void)
167 {
168 /*
169 * Remove vfsctl vnode ops
170 */
171 if (zfsctl_ops_root)
172 vn_freevnodeops(zfsctl_ops_root);
173 if (zfsctl_ops_snapdir)
174 vn_freevnodeops(zfsctl_ops_snapdir);
175 if (zfsctl_ops_snapshot)
176 vn_freevnodeops(zfsctl_ops_snapshot);
177 if (zfsctl_ops_shares)
178 vn_freevnodeops(zfsctl_ops_shares);
179 if (zfsctl_ops_shares_dir)
180 vn_freevnodeops(zfsctl_ops_shares_dir);
181
182 zfsctl_ops_root = NULL;
183 zfsctl_ops_snapdir = NULL;
184 zfsctl_ops_snapshot = NULL;
185 zfsctl_ops_shares = NULL;
186 zfsctl_ops_shares_dir = NULL;
187 }
188
189 boolean_t
190 zfsctl_is_node(vnode_t *vp)
191 {
192 return (vn_matchops(vp, zfsctl_ops_root) ||
193 vn_matchops(vp, zfsctl_ops_snapdir) ||
194 vn_matchops(vp, zfsctl_ops_snapshot) ||
195 vn_matchops(vp, zfsctl_ops_shares) ||
196 vn_matchops(vp, zfsctl_ops_shares_dir));
197
198 }
199
200 /*
201 * Return the inode number associated with the 'snapshot' or
202 * 'shares' directory.
203 */
204 /* ARGSUSED */
205 static ino64_t
206 zfsctl_root_inode_cb(vnode_t *vp, int index)
207 {
208 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
209
210 ASSERT(index <= 2);
211
212 if (index == 0)
213 return (ZFSCTL_INO_SNAPDIR);
214
215 return (zfsvfs->z_shares_dir);
216 }
217
218 /*
219 * Create the '.zfs' directory. This directory is cached as part of the VFS
220 * structure. This results in a hold on the vfs_t. The code in zfs_umount()
221 * therefore checks against a vfs_count of 2 instead of 1. This reference
222 * is removed when the ctldir is destroyed in the unmount.
223 */
224 void
225 zfsctl_create(zfsvfs_t *zfsvfs)
226 {
227 vnode_t *vp, *rvp;
228 zfsctl_node_t *zcp;
229 uint64_t crtime[2];
230
231 ASSERT(zfsvfs->z_ctldir == NULL);
232
233 vp = gfs_root_create(sizeof (zfsctl_node_t), zfsvfs->z_vfs,
234 zfsctl_ops_root, ZFSCTL_INO_ROOT, zfsctl_root_entries,
235 zfsctl_root_inode_cb, MAXNAMELEN, NULL, NULL);
236 zcp = vp->v_data;
237 zcp->zc_id = ZFSCTL_INO_ROOT;
238
239 VERIFY(VFS_ROOT(zfsvfs->z_vfs, &rvp) == 0);
240 VERIFY(0 == sa_lookup(VTOZ(rvp)->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
241 &crtime, sizeof (crtime)));
242 ZFS_TIME_DECODE(&zcp->zc_cmtime, crtime);
243 VN_RELE(rvp);
244
245 /*
246 * We're only faking the fact that we have a root of a filesystem for
247 * the sake of the GFS interfaces. Undo the flag manipulation it did
248 * for us.
249 */
250 vp->v_flag &= ~(VROOT | VNOCACHE | VNOMAP | VNOSWAP | VNOMOUNT);
251
252 zfsvfs->z_ctldir = vp;
253 }
254
255 /*
256 * Destroy the '.zfs' directory. Only called when the filesystem is unmounted.
257 * There might still be more references if we were force unmounted, but only
258 * new zfs_inactive() calls can occur and they don't reference .zfs
259 */
260 void
261 zfsctl_destroy(zfsvfs_t *zfsvfs)
262 {
263 VN_RELE(zfsvfs->z_ctldir);
264 zfsvfs->z_ctldir = NULL;
265 }
266
267 /*
268 * Given a root znode, retrieve the associated .zfs directory.
269 * Add a hold to the vnode and return it.
270 */
271 vnode_t *
272 zfsctl_root(znode_t *zp)
273 {
274 ASSERT(zfs_has_ctldir(zp));
275 VN_HOLD(zp->z_zfsvfs->z_ctldir);
276 return (zp->z_zfsvfs->z_ctldir);
277 }
278
279 /*
280 * Common open routine. Disallow any write access.
281 */
282 /* ARGSUSED */
283 static int
284 zfsctl_common_open(vnode_t **vpp, int flags, cred_t *cr, caller_context_t *ct)
285 {
286 if (flags & FWRITE)
287 return (EACCES);
288
289 return (0);
290 }
291
292 /*
293 * Common close routine. Nothing to do here.
294 */
295 /* ARGSUSED */
296 static int
297 zfsctl_common_close(vnode_t *vpp, int flags, int count, offset_t off,
298 cred_t *cr, caller_context_t *ct)
299 {
300 return (0);
301 }
302
303 /*
304 * Common access routine. Disallow writes.
305 */
306 /* ARGSUSED */
307 static int
308 zfsctl_common_access(vnode_t *vp, int mode, int flags, cred_t *cr,
309 caller_context_t *ct)
310 {
311 if (flags & V_ACE_MASK) {
312 if (mode & ACE_ALL_WRITE_PERMS)
313 return (EACCES);
314 } else {
315 if (mode & VWRITE)
316 return (EACCES);
317 }
318
319 return (0);
320 }
321
322 /*
323 * Common getattr function. Fill in basic information.
324 */
325 static void
326 zfsctl_common_getattr(vnode_t *vp, vattr_t *vap)
327 {
328 timestruc_t now;
329
330 vap->va_uid = 0;
331 vap->va_gid = 0;
332 vap->va_rdev = 0;
333 /*
334 * We are a purely virtual object, so we have no
335 * blocksize or allocated blocks.
336 */
337 vap->va_blksize = 0;
338 vap->va_nblocks = 0;
339 vap->va_seq = 0;
340 vap->va_fsid = vp->v_vfsp->vfs_dev;
341 vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP |
342 S_IROTH | S_IXOTH;
343 vap->va_type = VDIR;
344 /*
345 * We live in the now (for atime).
346 */
347 gethrestime(&now);
348 vap->va_atime = now;
349 }
350
351 /*ARGSUSED*/
352 static int
353 zfsctl_common_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
354 {
355 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
356 zfsctl_node_t *zcp = vp->v_data;
357 uint64_t object = zcp->zc_id;
358 zfid_short_t *zfid;
359 int i;
360
361 ZFS_ENTER(zfsvfs);
362
363 if (fidp->fid_len < SHORT_FID_LEN) {
364 fidp->fid_len = SHORT_FID_LEN;
365 ZFS_EXIT(zfsvfs);
366 return (ENOSPC);
367 }
368
369 zfid = (zfid_short_t *)fidp;
370
371 zfid->zf_len = SHORT_FID_LEN;
372
373 for (i = 0; i < sizeof (zfid->zf_object); i++)
374 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
375
376 /* .zfs znodes always have a generation number of 0 */
377 for (i = 0; i < sizeof (zfid->zf_gen); i++)
378 zfid->zf_gen[i] = 0;
379
380 ZFS_EXIT(zfsvfs);
381 return (0);
382 }
383
384
385 /*ARGSUSED*/
386 static int
387 zfsctl_shares_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
388 {
389 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
390 znode_t *dzp;
391 int error;
392
393 ZFS_ENTER(zfsvfs);
394
395 if (zfsvfs->z_shares_dir == 0) {
396 ZFS_EXIT(zfsvfs);
397 return (ENOTSUP);
398 }
399
400 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
401 error = VOP_FID(ZTOV(dzp), fidp, ct);
402 VN_RELE(ZTOV(dzp));
403 }
404
405 ZFS_EXIT(zfsvfs);
406 return (error);
407 }
408 /*
409 * .zfs inode namespace
410 *
411 * We need to generate unique inode numbers for all files and directories
412 * within the .zfs pseudo-filesystem. We use the following scheme:
413 *
414 * ENTRY ZFSCTL_INODE
415 * .zfs 1
416 * .zfs/snapshot 2
417 * .zfs/snapshot/<snap> objectid(snap)
418 */
419
420 #define ZFSCTL_INO_SNAP(id) (id)
421
422 /*
423 * Get root directory attributes.
424 */
425 /* ARGSUSED */
426 static int
427 zfsctl_root_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
428 caller_context_t *ct)
429 {
430 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
431 zfsctl_node_t *zcp = vp->v_data;
432
433 ZFS_ENTER(zfsvfs);
434 vap->va_nodeid = ZFSCTL_INO_ROOT;
435 vap->va_nlink = vap->va_size = NROOT_ENTRIES;
436 vap->va_mtime = vap->va_ctime = zcp->zc_cmtime;
437
438 zfsctl_common_getattr(vp, vap);
439 ZFS_EXIT(zfsvfs);
440
441 return (0);
442 }
443
444 /*
445 * Special case the handling of "..".
446 */
447 /* ARGSUSED */
448 int
449 zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
450 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
451 int *direntflags, pathname_t *realpnp)
452 {
453 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
454 int err;
455
456 /*
457 * No extended attributes allowed under .zfs
458 */
459 if (flags & LOOKUP_XATTR)
460 return (EINVAL);
461
462 ZFS_ENTER(zfsvfs);
463
464 if (strcmp(nm, "..") == 0) {
465 err = VFS_ROOT(dvp->v_vfsp, vpp);
466 } else {
467 err = gfs_vop_lookup(dvp, nm, vpp, pnp, flags, rdir,
468 cr, ct, direntflags, realpnp);
469 }
470
471 ZFS_EXIT(zfsvfs);
472
473 return (err);
474 }
475
476 static int
477 zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
478 caller_context_t *ct)
479 {
480 /*
481 * We only care about ACL_ENABLED so that libsec can
482 * display ACL correctly and not default to POSIX draft.
483 */
484 if (cmd == _PC_ACL_ENABLED) {
485 *valp = _ACL_ACE_ENABLED;
486 return (0);
487 }
488
489 return (fs_pathconf(vp, cmd, valp, cr, ct));
490 }
491
492 static const fs_operation_def_t zfsctl_tops_root[] = {
493 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
494 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
495 { VOPNAME_IOCTL, { .error = fs_inval } },
496 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_root_getattr } },
497 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
498 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } },
499 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_root_lookup } },
500 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
501 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } },
502 { VOPNAME_PATHCONF, { .vop_pathconf = zfsctl_pathconf } },
503 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } },
504 { NULL }
505 };
506
507 static int
508 zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname)
509 {
510 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os;
511
512 if (snapshot_namecheck(name, NULL, NULL) != 0)
513 return (EILSEQ);
514 dmu_objset_name(os, zname);
515 if (strlen(zname) + 1 + strlen(name) >= len)
516 return (ENAMETOOLONG);
517 (void) strcat(zname, "@");
518 (void) strcat(zname, name);
519 return (0);
520 }
521
522 static int
523 zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr)
524 {
525 vnode_t *svp = sep->se_root;
526 int error;
527
528 ASSERT(vn_ismntpt(svp));
529
530 /* this will be dropped by dounmount() */
531 if ((error = vn_vfswlock(svp)) != 0)
532 return (error);
533
534 VN_HOLD(svp);
535 error = dounmount(vn_mountedvfs(svp), fflags, cr);
536 if (error) {
537 VN_RELE(svp);
538 return (error);
539 }
540
541 /*
542 * We can't use VN_RELE(), as that will try to invoke
543 * zfsctl_snapdir_inactive(), which would cause us to destroy
544 * the sd_lock mutex held by our caller.
545 */
546 ASSERT(svp->v_count == 1);
547 gfs_vop_inactive(svp, cr, NULL);
548
549 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
550 kmem_free(sep, sizeof (zfs_snapentry_t));
551
552 return (0);
553 }
554
555 static void
556 zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm)
557 {
558 avl_index_t where;
559 vfs_t *vfsp;
560 refstr_t *pathref;
561 char newpath[MAXNAMELEN];
562 char *tail;
563
564 ASSERT(MUTEX_HELD(&sdp->sd_lock));
565 ASSERT(sep != NULL);
566
567 vfsp = vn_mountedvfs(sep->se_root);
568 ASSERT(vfsp != NULL);
569
570 vfs_lock_wait(vfsp);
571
572 /*
573 * Change the name in the AVL tree.
574 */
575 avl_remove(&sdp->sd_snaps, sep);
576 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
577 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
578 (void) strcpy(sep->se_name, nm);
579 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL);
580 avl_insert(&sdp->sd_snaps, sep, where);
581
582 /*
583 * Change the current mountpoint info:
584 * - update the tail of the mntpoint path
585 * - update the tail of the resource path
586 */
587 pathref = vfs_getmntpoint(vfsp);
588 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
589 VERIFY((tail = strrchr(newpath, '/')) != NULL);
590 *(tail+1) = '\0';
591 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
592 (void) strcat(newpath, nm);
593 refstr_rele(pathref);
594 vfs_setmntpoint(vfsp, newpath, 0);
595
596 pathref = vfs_getresource(vfsp);
597 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
598 VERIFY((tail = strrchr(newpath, '@')) != NULL);
599 *(tail+1) = '\0';
600 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
601 (void) strcat(newpath, nm);
602 refstr_rele(pathref);
603 vfs_setresource(vfsp, newpath, 0);
604
605 vfs_unlock(vfsp);
606 }
607
608 /*ARGSUSED*/
609 static int
610 zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm,
611 cred_t *cr, caller_context_t *ct, int flags)
612 {
613 zfsctl_snapdir_t *sdp = sdvp->v_data;
614 zfs_snapentry_t search, *sep;
615 zfsvfs_t *zfsvfs;
616 avl_index_t where;
617 char from[MAXNAMELEN], to[MAXNAMELEN];
618 char real[MAXNAMELEN];
619 int err;
620
621 zfsvfs = sdvp->v_vfsp->vfs_data;
622 ZFS_ENTER(zfsvfs);
623
624 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
625 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real,
626 MAXNAMELEN, NULL);
627 if (err == 0) {
628 snm = real;
629 } else if (err != ENOTSUP) {
630 ZFS_EXIT(zfsvfs);
631 return (err);
632 }
633 }
634
635 ZFS_EXIT(zfsvfs);
636
637 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from);
638 if (!err)
639 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to);
640 if (!err)
641 err = zfs_secpolicy_rename_perms(from, to, cr);
642 if (err)
643 return (err);
644
645 /*
646 * Cannot move snapshots out of the snapdir.
647 */
648 if (sdvp != tdvp)
649 return (EINVAL);
650
651 if (strcmp(snm, tnm) == 0)
652 return (0);
653
654 mutex_enter(&sdp->sd_lock);
655
656 search.se_name = (char *)snm;
657 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) {
658 mutex_exit(&sdp->sd_lock);
659 return (ENOENT);
660 }
661
662 err = dmu_objset_rename(from, to, B_FALSE);
663 if (err == 0)
664 zfsctl_rename_snap(sdp, sep, tnm);
665
666 mutex_exit(&sdp->sd_lock);
667
668 return (err);
669 }
670
671 /* ARGSUSED */
672 static int
673 zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
674 caller_context_t *ct, int flags)
675 {
676 zfsctl_snapdir_t *sdp = dvp->v_data;
677 zfs_snapentry_t *sep;
678 zfs_snapentry_t search;
679 zfsvfs_t *zfsvfs;
680 char snapname[MAXNAMELEN];
681 char real[MAXNAMELEN];
682 int err;
683
684 zfsvfs = dvp->v_vfsp->vfs_data;
685 ZFS_ENTER(zfsvfs);
686
687 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
688
689 err = dmu_snapshot_realname(zfsvfs->z_os, name, real,
690 MAXNAMELEN, NULL);
691 if (err == 0) {
692 name = real;
693 } else if (err != ENOTSUP) {
694 ZFS_EXIT(zfsvfs);
695 return (err);
696 }
697 }
698
699 ZFS_EXIT(zfsvfs);
700
701 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname);
702 if (!err)
703 err = zfs_secpolicy_destroy_perms(snapname, cr);
704 if (err)
705 return (err);
706
707 mutex_enter(&sdp->sd_lock);
708
709 search.se_name = name;
710 sep = avl_find(&sdp->sd_snaps, &search, NULL);
711 if (sep) {
712 avl_remove(&sdp->sd_snaps, sep);
713 err = zfsctl_unmount_snap(sep, MS_FORCE, cr);
714 if (err)
715 avl_add(&sdp->sd_snaps, sep);
716 else
717 err = dmu_objset_destroy(snapname, B_FALSE);
718 } else {
719 err = ENOENT;
720 }
721
722 mutex_exit(&sdp->sd_lock);
723
724 return (err);
725 }
726
727 /*
728 * This creates a snapshot under '.zfs/snapshot'.
729 */
730 /* ARGSUSED */
731 static int
732 zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp,
733 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp)
734 {
735 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
736 char name[MAXNAMELEN];
737 int err;
738 static enum symfollow follow = NO_FOLLOW;
739 static enum uio_seg seg = UIO_SYSSPACE;
740
741 if (snapshot_namecheck(dirname, NULL, NULL) != 0)
742 return (EILSEQ);
743
744 dmu_objset_name(zfsvfs->z_os, name);
745
746 *vpp = NULL;
747
748 err = zfs_secpolicy_snapshot_perms(name, cr);
749 if (err)
750 return (err);
751
752 if (err == 0) {
753 err = dmu_objset_snapshot_one(name, dirname);
754 if (err)
755 return (err);
756 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp);
757 }
758
759 return (err);
760 }
761
762 /*
763 * Lookup entry point for the 'snapshot' directory. Try to open the
764 * snapshot if it exist, creating the pseudo filesystem vnode as necessary.
765 * Perform a mount of the associated dataset on top of the vnode.
766 */
767 /* ARGSUSED */
768 static int
769 zfsctl_snapdir_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
770 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
771 int *direntflags, pathname_t *realpnp)
772 {
773 zfsctl_snapdir_t *sdp = dvp->v_data;
774 objset_t *snap;
775 char snapname[MAXNAMELEN];
776 char real[MAXNAMELEN];
777 char *mountpoint;
778 zfs_snapentry_t *sep, search;
779 struct mounta margs;
780 vfs_t *vfsp;
781 size_t mountpoint_len;
782 avl_index_t where;
783 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
784 int err;
785
786 /*
787 * No extended attributes allowed under .zfs
788 */
789 if (flags & LOOKUP_XATTR)
790 return (EINVAL);
791
792 ASSERT(dvp->v_type == VDIR);
793
794 /*
795 * If we get a recursive call, that means we got called
796 * from the domount() code while it was trying to look up the
797 * spec (which looks like a local path for zfs). We need to
798 * add some flag to domount() to tell it not to do this lookup.
799 */
800 if (MUTEX_HELD(&sdp->sd_lock))
801 return (ENOENT);
802
803 ZFS_ENTER(zfsvfs);
804
805 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
806 ZFS_EXIT(zfsvfs);
807 return (0);
808 }
809
810 if (flags & FIGNORECASE) {
811 boolean_t conflict = B_FALSE;
812
813 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real,
814 MAXNAMELEN, &conflict);
815 if (err == 0) {
816 nm = real;
817 } else if (err != ENOTSUP) {
818 ZFS_EXIT(zfsvfs);
819 return (err);
820 }
821 if (realpnp)
822 (void) strlcpy(realpnp->pn_buf, nm,
823 realpnp->pn_bufsize);
824 if (conflict && direntflags)
825 *direntflags = ED_CASE_CONFLICT;
826 }
827
828 mutex_enter(&sdp->sd_lock);
829 search.se_name = (char *)nm;
830 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) {
831 *vpp = sep->se_root;
832 VN_HOLD(*vpp);
833 err = traverse(vpp);
834 if (err) {
835 VN_RELE(*vpp);
836 *vpp = NULL;
837 } else if (*vpp == sep->se_root) {
838 /*
839 * The snapshot was unmounted behind our backs,
840 * try to remount it.
841 */
842 goto domount;
843 } else {
844 /*
845 * VROOT was set during the traverse call. We need
846 * to clear it since we're pretending to be part
847 * of our parent's vfs.
848 */
849 (*vpp)->v_flag &= ~VROOT;
850 }
851 mutex_exit(&sdp->sd_lock);
852 ZFS_EXIT(zfsvfs);
853 return (err);
854 }
855
856 /*
857 * The requested snapshot is not currently mounted, look it up.
858 */
859 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname);
860 if (err) {
861 mutex_exit(&sdp->sd_lock);
862 ZFS_EXIT(zfsvfs);
863 /*
864 * handle "ls *" or "?" in a graceful manner,
865 * forcing EILSEQ to ENOENT.
866 * Since shell ultimately passes "*" or "?" as name to lookup
867 */
868 return (err == EILSEQ ? ENOENT : err);
869 }
870 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) {
871 mutex_exit(&sdp->sd_lock);
872 ZFS_EXIT(zfsvfs);
873 return (ENOENT);
874 }
875
876 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP);
877 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
878 (void) strcpy(sep->se_name, nm);
879 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap));
880 avl_insert(&sdp->sd_snaps, sep, where);
881
882 dmu_objset_rele(snap, FTAG);
883 domount:
884 mountpoint_len = strlen(refstr_value(dvp->v_vfsp->vfs_mntpt)) +
885 strlen("/.zfs/snapshot/") + strlen(nm) + 1;
886 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP);
887 (void) snprintf(mountpoint, mountpoint_len, "%s/.zfs/snapshot/%s",
888 refstr_value(dvp->v_vfsp->vfs_mntpt), nm);
889
890 margs.spec = snapname;
891 margs.dir = mountpoint;
892 margs.flags = MS_SYSSPACE | MS_NOMNTTAB;
893 margs.fstype = "zfs";
894 margs.dataptr = NULL;
895 margs.datalen = 0;
896 margs.optptr = NULL;
897 margs.optlen = 0;
898
899 err = domount("zfs", &margs, *vpp, kcred, &vfsp);
900 kmem_free(mountpoint, mountpoint_len);
901
902 if (err == 0) {
903 /*
904 * Return the mounted root rather than the covered mount point.
905 * Takes the GFS vnode at .zfs/snapshot/<snapname> and returns
906 * the ZFS vnode mounted on top of the GFS node. This ZFS
907 * vnode is the root of the newly created vfsp.
908 */
909 VFS_RELE(vfsp);
910 err = traverse(vpp);
911 }
912
913 if (err == 0) {
914 /*
915 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>.
916 *
917 * This is where we lie about our v_vfsp in order to
918 * make .zfs/snapshot/<snapname> accessible over NFS
919 * without requiring manual mounts of <snapname>.
920 */
921 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs);
922 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs;
923 (*vpp)->v_vfsp = zfsvfs->z_vfs;
924 (*vpp)->v_flag &= ~VROOT;
925 }
926 mutex_exit(&sdp->sd_lock);
927 ZFS_EXIT(zfsvfs);
928
929 /*
930 * If we had an error, drop our hold on the vnode and
931 * zfsctl_snapshot_inactive() will clean up.
932 */
933 if (err) {
934 VN_RELE(*vpp);
935 *vpp = NULL;
936 }
937 return (err);
938 }
939
940 /* ARGSUSED */
941 static int
942 zfsctl_shares_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
943 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
944 int *direntflags, pathname_t *realpnp)
945 {
946 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
947 znode_t *dzp;
948 int error;
949
950 ZFS_ENTER(zfsvfs);
951
952 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
953 ZFS_EXIT(zfsvfs);
954 return (0);
955 }
956
957 if (zfsvfs->z_shares_dir == 0) {
958 ZFS_EXIT(zfsvfs);
959 return (ENOTSUP);
960 }
961 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0)
962 error = VOP_LOOKUP(ZTOV(dzp), nm, vpp, pnp,
963 flags, rdir, cr, ct, direntflags, realpnp);
964
965 VN_RELE(ZTOV(dzp));
966 ZFS_EXIT(zfsvfs);
967
968 return (error);
969 }
970
971 /* ARGSUSED */
972 static int
973 zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp,
974 offset_t *offp, offset_t *nextp, void *data, int flags)
975 {
976 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
977 char snapname[MAXNAMELEN];
978 uint64_t id, cookie;
979 boolean_t case_conflict;
980 int error;
981
982 ZFS_ENTER(zfsvfs);
983
984 cookie = *offp;
985 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id,
986 &cookie, &case_conflict);
987 if (error) {
988 ZFS_EXIT(zfsvfs);
989 if (error == ENOENT) {
990 *eofp = 1;
991 return (0);
992 }
993 return (error);
994 }
995
996 if (flags & V_RDDIR_ENTFLAGS) {
997 edirent_t *eodp = dp;
998
999 (void) strcpy(eodp->ed_name, snapname);
1000 eodp->ed_ino = ZFSCTL_INO_SNAP(id);
1001 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0;
1002 } else {
1003 struct dirent64 *odp = dp;
1004
1005 (void) strcpy(odp->d_name, snapname);
1006 odp->d_ino = ZFSCTL_INO_SNAP(id);
1007 }
1008 *nextp = cookie;
1009
1010 ZFS_EXIT(zfsvfs);
1011
1012 return (0);
1013 }
1014
1015 /* ARGSUSED */
1016 static int
1017 zfsctl_shares_readdir(vnode_t *vp, uio_t *uiop, cred_t *cr, int *eofp,
1018 caller_context_t *ct, int flags)
1019 {
1020 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1021 znode_t *dzp;
1022 int error;
1023
1024 ZFS_ENTER(zfsvfs);
1025
1026 if (zfsvfs->z_shares_dir == 0) {
1027 ZFS_EXIT(zfsvfs);
1028 return (ENOTSUP);
1029 }
1030 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1031 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ct, flags);
1032 VN_RELE(ZTOV(dzp));
1033 } else {
1034 *eofp = 1;
1035 error = ENOENT;
1036 }
1037
1038 ZFS_EXIT(zfsvfs);
1039 return (error);
1040 }
1041
1042 /*
1043 * pvp is the '.zfs' directory (zfsctl_node_t).
1044 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t).
1045 *
1046 * This function is the callback to create a GFS vnode for '.zfs/snapshot'
1047 * when a lookup is performed on .zfs for "snapshot".
1048 */
1049 vnode_t *
1050 zfsctl_mknode_snapdir(vnode_t *pvp)
1051 {
1052 vnode_t *vp;
1053 zfsctl_snapdir_t *sdp;
1054
1055 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp,
1056 zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN,
1057 zfsctl_snapdir_readdir_cb, NULL);
1058 sdp = vp->v_data;
1059 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR;
1060 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1061 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL);
1062 avl_create(&sdp->sd_snaps, snapentry_compare,
1063 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node));
1064 return (vp);
1065 }
1066
1067 vnode_t *
1068 zfsctl_mknode_shares(vnode_t *pvp)
1069 {
1070 vnode_t *vp;
1071 zfsctl_node_t *sdp;
1072
1073 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp,
1074 zfsctl_ops_shares, NULL, NULL, MAXNAMELEN,
1075 NULL, NULL);
1076 sdp = vp->v_data;
1077 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1078 return (vp);
1079
1080 }
1081
1082 /* ARGSUSED */
1083 static int
1084 zfsctl_shares_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
1085 caller_context_t *ct)
1086 {
1087 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1088 znode_t *dzp;
1089 int error;
1090
1091 ZFS_ENTER(zfsvfs);
1092 if (zfsvfs->z_shares_dir == 0) {
1093 ZFS_EXIT(zfsvfs);
1094 return (ENOTSUP);
1095 }
1096 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1097 error = VOP_GETATTR(ZTOV(dzp), vap, flags, cr, ct);
1098 VN_RELE(ZTOV(dzp));
1099 }
1100 ZFS_EXIT(zfsvfs);
1101 return (error);
1102
1103
1104 }
1105
1106 /* ARGSUSED */
1107 static int
1108 zfsctl_snapdir_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
1109 caller_context_t *ct)
1110 {
1111 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1112 zfsctl_snapdir_t *sdp = vp->v_data;
1113
1114 ZFS_ENTER(zfsvfs);
1115 zfsctl_common_getattr(vp, vap);
1116 vap->va_nodeid = gfs_file_inode(vp);
1117 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2;
1118 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os);
1119 ZFS_EXIT(zfsvfs);
1120
1121 return (0);
1122 }
1123
1124 /* ARGSUSED */
1125 static void
1126 zfsctl_snapdir_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
1127 {
1128 zfsctl_snapdir_t *sdp = vp->v_data;
1129 void *private;
1130
1131 private = gfs_dir_inactive(vp);
1132 if (private != NULL) {
1133 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0);
1134 mutex_destroy(&sdp->sd_lock);
1135 avl_destroy(&sdp->sd_snaps);
1136 kmem_free(private, sizeof (zfsctl_snapdir_t));
1137 }
1138 }
1139
1140 static const fs_operation_def_t zfsctl_tops_snapdir[] = {
1141 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
1142 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
1143 { VOPNAME_IOCTL, { .error = fs_inval } },
1144 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } },
1145 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
1146 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } },
1147 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } },
1148 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } },
1149 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } },
1150 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } },
1151 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
1152 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } },
1153 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } },
1154 { NULL }
1155 };
1156
1157 static const fs_operation_def_t zfsctl_tops_shares[] = {
1158 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
1159 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
1160 { VOPNAME_IOCTL, { .error = fs_inval } },
1161 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } },
1162 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
1163 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } },
1164 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } },
1165 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
1166 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } },
1167 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } },
1168 { NULL }
1169 };
1170
1171 /*
1172 * pvp is the GFS vnode '.zfs/snapshot'.
1173 *
1174 * This creates a GFS node under '.zfs/snapshot' representing each
1175 * snapshot. This newly created GFS node is what we mount snapshot
1176 * vfs_t's ontop of.
1177 */
1178 static vnode_t *
1179 zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset)
1180 {
1181 vnode_t *vp;
1182 zfsctl_node_t *zcp;
1183
1184 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp,
1185 zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL);
1186 zcp = vp->v_data;
1187 zcp->zc_id = objset;
1188
1189 return (vp);
1190 }
1191
1192 static void
1193 zfsctl_snapshot_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
1194 {
1195 zfsctl_snapdir_t *sdp;
1196 zfs_snapentry_t *sep, *next;
1197 vnode_t *dvp;
1198
1199 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0);
1200 sdp = dvp->v_data;
1201
1202 mutex_enter(&sdp->sd_lock);
1203
1204 if (vp->v_count > 1) {
1205 mutex_exit(&sdp->sd_lock);
1206 return;
1207 }
1208 ASSERT(!vn_ismntpt(vp));
1209
1210 sep = avl_first(&sdp->sd_snaps);
1211 while (sep != NULL) {
1212 next = AVL_NEXT(&sdp->sd_snaps, sep);
1213
1214 if (sep->se_root == vp) {
1215 avl_remove(&sdp->sd_snaps, sep);
1216 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
1217 kmem_free(sep, sizeof (zfs_snapentry_t));
1218 break;
1219 }
1220 sep = next;
1221 }
1222 ASSERT(sep != NULL);
1223
1224 mutex_exit(&sdp->sd_lock);
1225 VN_RELE(dvp);
1226
1227 /*
1228 * Dispose of the vnode for the snapshot mount point.
1229 * This is safe to do because once this entry has been removed
1230 * from the AVL tree, it can't be found again, so cannot become
1231 * "active". If we lookup the same name again we will end up
1232 * creating a new vnode.
1233 */
1234 gfs_vop_inactive(vp, cr, ct);
1235 }
1236
1237
1238 /*
1239 * These VP's should never see the light of day. They should always
1240 * be covered.
1241 */
1242 static const fs_operation_def_t zfsctl_tops_snapshot[] = {
1243 VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapshot_inactive },
1244 NULL, NULL
1245 };
1246
1247 int
1248 zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp)
1249 {
1250 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1251 vnode_t *dvp, *vp;
1252 zfsctl_snapdir_t *sdp;
1253 zfsctl_node_t *zcp;
1254 zfs_snapentry_t *sep;
1255 int error;
1256
1257 ASSERT(zfsvfs->z_ctldir != NULL);
1258 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1259 NULL, 0, NULL, kcred, NULL, NULL, NULL);
1260 if (error != 0)
1261 return (error);
1262 sdp = dvp->v_data;
1263
1264 mutex_enter(&sdp->sd_lock);
1265 sep = avl_first(&sdp->sd_snaps);
1266 while (sep != NULL) {
1267 vp = sep->se_root;
1268 zcp = vp->v_data;
1269 if (zcp->zc_id == objsetid)
1270 break;
1271
1272 sep = AVL_NEXT(&sdp->sd_snaps, sep);
1273 }
1274
1275 if (sep != NULL) {
1276 VN_HOLD(vp);
1277 /*
1278 * Return the mounted root rather than the covered mount point.
1279 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid>
1280 * and returns the ZFS vnode mounted on top of the GFS node.
1281 * This ZFS vnode is the root of the vfs for objset 'objsetid'.
1282 */
1283 error = traverse(&vp);
1284 if (error == 0) {
1285 if (vp == sep->se_root)
1286 error = EINVAL;
1287 else
1288 *zfsvfsp = VTOZ(vp)->z_zfsvfs;
1289 }
1290 mutex_exit(&sdp->sd_lock);
1291 VN_RELE(vp);
1292 } else {
1293 error = EINVAL;
1294 mutex_exit(&sdp->sd_lock);
1295 }
1296
1297 VN_RELE(dvp);
1298
1299 return (error);
1300 }
1301
1302 /*
1303 * Unmount any snapshots for the given filesystem. This is called from
1304 * zfs_umount() - if we have a ctldir, then go through and unmount all the
1305 * snapshots.
1306 */
1307 int
1308 zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr)
1309 {
1310 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1311 vnode_t *dvp;
1312 zfsctl_snapdir_t *sdp;
1313 zfs_snapentry_t *sep, *next;
1314 int error;
1315
1316 ASSERT(zfsvfs->z_ctldir != NULL);
1317 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1318 NULL, 0, NULL, cr, NULL, NULL, NULL);
1319 if (error != 0)
1320 return (error);
1321 sdp = dvp->v_data;
1322
1323 mutex_enter(&sdp->sd_lock);
1324
1325 sep = avl_first(&sdp->sd_snaps);
1326 while (sep != NULL) {
1327 next = AVL_NEXT(&sdp->sd_snaps, sep);
1328
1329 /*
1330 * If this snapshot is not mounted, then it must
1331 * have just been unmounted by somebody else, and
1332 * will be cleaned up by zfsctl_snapdir_inactive().
1333 */
1334 if (vn_ismntpt(sep->se_root)) {
1335 avl_remove(&sdp->sd_snaps, sep);
1336 error = zfsctl_unmount_snap(sep, fflags, cr);
1337 if (error) {
1338 avl_add(&sdp->sd_snaps, sep);
1339 break;
1340 }
1341 }
1342 sep = next;
1343 }
1344
1345 mutex_exit(&sdp->sd_lock);
1346 VN_RELE(dvp);
1347
1348 return (error);
1349 }