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) 2013 by Delphix. All rights reserved.
24 * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved.
25 */
26
27 /*
28 * ZFS control directory (a.k.a. ".zfs")
29 *
30 * This directory provides a common location for all ZFS meta-objects.
31 * Currently, this is only the 'snapshot' directory, but this may expand in the
32 * future. The elements are built using the GFS primitives, as the hierarchy
33 * does not actually exist on disk.
34 *
35 * For 'snapshot', we don't want to have all snapshots always mounted, because
36 * this would take up a huge amount of space in /etc/mnttab. We have three
37 * types of objects:
38 *
39 * ctldir ------> snapshotdir -------> snapshot
40 * |
41 * |
42 * V
43 * mounted fs
44 *
45 * The 'snapshot' node contains just enough information to lookup '..' and act
46 * as a mountpoint for the snapshot. Whenever we lookup a specific snapshot, we
47 * perform an automount of the underlying filesystem and return the
48 * corresponding vnode.
49 *
50 * All mounts are handled automatically by the kernel, but unmounts are
51 * (currently) handled from user land. The main reason is that there is no
52 * reliable way to auto-unmount the filesystem when it's "no longer in use".
53 * When the user unmounts a filesystem, we call zfsctl_unmount(), which
54 * unmounts any snapshots within the snapshot directory.
55 *
56 * The '.zfs', '.zfs/snapshot', and all directories created under
57 * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') are all GFS nodes and
58 * share the same vfs_t as the head filesystem (what '.zfs' lives under).
59 *
60 * File systems mounted ontop of the GFS nodes '.zfs/snapshot/<snapname>'
61 * (ie: snapshots) are ZFS nodes and have their own unique vfs_t.
62 * However, vnodes within these mounted on file systems have their v_vfsp
63 * fields set to the head filesystem to make NFS happy (see
64 * zfsctl_snapdir_lookup()). We VFS_HOLD the head filesystem's vfs_t
65 * so that it cannot be freed until all snapshots have been unmounted.
66 */
67
68 #include <sys/zfs_context.h>
69 #include <sys/zfs_ctldir.h>
70 #include <sys/zfs_ioctl.h>
71 #include <sys/zfs_vfsops.h>
72 #include <sys/namei.h>
73 #include <sys/gfs.h>
74 #include <sys/stat.h>
75 #include <sys/dmu.h>
76 #include <sys/dsl_destroy.h>
77 #include <sys/dsl_deleg.h>
78 #include <sys/mount.h>
79 #include <sys/sunddi.h>
80
81 #include "zfs_namecheck.h"
82
83 typedef struct zfsctl_node {
84 gfs_dir_t zc_gfs_private;
85 uint64_t zc_id;
86 timestruc_t zc_cmtime; /* ctime and mtime, always the same */
87 } zfsctl_node_t;
88
89 typedef struct zfsctl_snapdir {
90 zfsctl_node_t sd_node;
91 kmutex_t sd_lock;
92 avl_tree_t sd_snaps;
93 } zfsctl_snapdir_t;
94
95 typedef struct {
96 char *se_name;
97 vnode_t *se_root;
98 avl_node_t se_node;
99 } zfs_snapentry_t;
100
101 static int
snapentry_compare(const void * a,const void * b)102 snapentry_compare(const void *a, const void *b)
103 {
104 const zfs_snapentry_t *sa = a;
105 const zfs_snapentry_t *sb = b;
106 int ret = strcmp(sa->se_name, sb->se_name);
107
108 if (ret < 0)
109 return (-1);
110 else if (ret > 0)
111 return (1);
112 else
113 return (0);
114 }
115
116 #ifdef illumos
117 vnodeops_t *zfsctl_ops_root;
118 vnodeops_t *zfsctl_ops_snapdir;
119 vnodeops_t *zfsctl_ops_snapshot;
120 vnodeops_t *zfsctl_ops_shares;
121 vnodeops_t *zfsctl_ops_shares_dir;
122
123 static const fs_operation_def_t zfsctl_tops_root[];
124 static const fs_operation_def_t zfsctl_tops_snapdir[];
125 static const fs_operation_def_t zfsctl_tops_snapshot[];
126 static const fs_operation_def_t zfsctl_tops_shares[];
127 #else
128 static struct vop_vector zfsctl_ops_root;
129 static struct vop_vector zfsctl_ops_snapdir;
130 static struct vop_vector zfsctl_ops_snapshot;
131 static struct vop_vector zfsctl_ops_shares;
132 static struct vop_vector zfsctl_ops_shares_dir;
133 #endif
134
135 static vnode_t *zfsctl_mknode_snapdir(vnode_t *);
136 static vnode_t *zfsctl_mknode_shares(vnode_t *);
137 static vnode_t *zfsctl_snapshot_mknode(vnode_t *, uint64_t objset);
138 static int zfsctl_unmount_snap(zfs_snapentry_t *, int, cred_t *);
139
140 #ifdef illumos
141 static gfs_opsvec_t zfsctl_opsvec[] = {
142 { ".zfs", zfsctl_tops_root, &zfsctl_ops_root },
143 { ".zfs/snapshot", zfsctl_tops_snapdir, &zfsctl_ops_snapdir },
144 { ".zfs/snapshot/vnode", zfsctl_tops_snapshot, &zfsctl_ops_snapshot },
145 { ".zfs/shares", zfsctl_tops_shares, &zfsctl_ops_shares_dir },
146 { ".zfs/shares/vnode", zfsctl_tops_shares, &zfsctl_ops_shares },
147 { NULL }
148 };
149 #endif
150
151 /*
152 * Root directory elements. We only have two entries
153 * snapshot and shares.
154 */
155 static gfs_dirent_t zfsctl_root_entries[] = {
156 { "snapshot", zfsctl_mknode_snapdir, GFS_CACHE_VNODE },
157 { "shares", zfsctl_mknode_shares, GFS_CACHE_VNODE },
158 { NULL }
159 };
160
161 /* include . and .. in the calculation */
162 #define NROOT_ENTRIES ((sizeof (zfsctl_root_entries) / \
163 sizeof (gfs_dirent_t)) + 1)
164
165
166 /*
167 * Initialize the various GFS pieces we'll need to create and manipulate .zfs
168 * directories. This is called from the ZFS init routine, and initializes the
169 * vnode ops vectors that we'll be using.
170 */
171 void
zfsctl_init(void)172 zfsctl_init(void)
173 {
174 #ifdef illumos
175 VERIFY(gfs_make_opsvec(zfsctl_opsvec) == 0);
176 #endif
177 }
178
179 void
zfsctl_fini(void)180 zfsctl_fini(void)
181 {
182 #ifdef illumos
183 /*
184 * Remove vfsctl vnode ops
185 */
186 if (zfsctl_ops_root)
187 vn_freevnodeops(zfsctl_ops_root);
188 if (zfsctl_ops_snapdir)
189 vn_freevnodeops(zfsctl_ops_snapdir);
190 if (zfsctl_ops_snapshot)
191 vn_freevnodeops(zfsctl_ops_snapshot);
192 if (zfsctl_ops_shares)
193 vn_freevnodeops(zfsctl_ops_shares);
194 if (zfsctl_ops_shares_dir)
195 vn_freevnodeops(zfsctl_ops_shares_dir);
196
197 zfsctl_ops_root = NULL;
198 zfsctl_ops_snapdir = NULL;
199 zfsctl_ops_snapshot = NULL;
200 zfsctl_ops_shares = NULL;
201 zfsctl_ops_shares_dir = NULL;
202 #endif /* illumos */
203 }
204
205 boolean_t
zfsctl_is_node(vnode_t * vp)206 zfsctl_is_node(vnode_t *vp)
207 {
208 return (vn_matchops(vp, zfsctl_ops_root) ||
209 vn_matchops(vp, zfsctl_ops_snapdir) ||
210 vn_matchops(vp, zfsctl_ops_snapshot) ||
211 vn_matchops(vp, zfsctl_ops_shares) ||
212 vn_matchops(vp, zfsctl_ops_shares_dir));
213
214 }
215
216 /*
217 * Return the inode number associated with the 'snapshot' or
218 * 'shares' directory.
219 */
220 /* ARGSUSED */
221 static ino64_t
zfsctl_root_inode_cb(vnode_t * vp,int index)222 zfsctl_root_inode_cb(vnode_t *vp, int index)
223 {
224 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
225
226 ASSERT(index <= 2);
227
228 if (index == 0)
229 return (ZFSCTL_INO_SNAPDIR);
230
231 return (zfsvfs->z_shares_dir);
232 }
233
234 /*
235 * Create the '.zfs' directory. This directory is cached as part of the VFS
236 * structure. This results in a hold on the vfs_t. The code in zfs_umount()
237 * therefore checks against a vfs_count of 2 instead of 1. This reference
238 * is removed when the ctldir is destroyed in the unmount.
239 */
240 void
zfsctl_create(zfsvfs_t * zfsvfs)241 zfsctl_create(zfsvfs_t *zfsvfs)
242 {
243 vnode_t *vp, *rvp;
244 zfsctl_node_t *zcp;
245 uint64_t crtime[2];
246
247 ASSERT(zfsvfs->z_ctldir == NULL);
248
249 vp = gfs_root_create(sizeof (zfsctl_node_t), zfsvfs->z_vfs,
250 &zfsctl_ops_root, ZFSCTL_INO_ROOT, zfsctl_root_entries,
251 zfsctl_root_inode_cb, MAXNAMELEN, NULL, NULL);
252 zcp = vp->v_data;
253 zcp->zc_id = ZFSCTL_INO_ROOT;
254
255 VERIFY(VFS_ROOT(zfsvfs->z_vfs, LK_EXCLUSIVE, &rvp) == 0);
256 VERIFY(0 == sa_lookup(VTOZ(rvp)->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
257 &crtime, sizeof (crtime)));
258 ZFS_TIME_DECODE(&zcp->zc_cmtime, crtime);
259 VN_URELE(rvp);
260
261 /*
262 * We're only faking the fact that we have a root of a filesystem for
263 * the sake of the GFS interfaces. Undo the flag manipulation it did
264 * for us.
265 */
266 vp->v_vflag &= ~VV_ROOT;
267
268 zfsvfs->z_ctldir = vp;
269
270 VOP_UNLOCK(vp, 0);
271 }
272
273 /*
274 * Destroy the '.zfs' directory. Only called when the filesystem is unmounted.
275 * There might still be more references if we were force unmounted, but only
276 * new zfs_inactive() calls can occur and they don't reference .zfs
277 */
278 void
zfsctl_destroy(zfsvfs_t * zfsvfs)279 zfsctl_destroy(zfsvfs_t *zfsvfs)
280 {
281 VN_RELE(zfsvfs->z_ctldir);
282 zfsvfs->z_ctldir = NULL;
283 }
284
285 /*
286 * Given a root znode, retrieve the associated .zfs directory.
287 * Add a hold to the vnode and return it.
288 */
289 vnode_t *
zfsctl_root(znode_t * zp)290 zfsctl_root(znode_t *zp)
291 {
292 ASSERT(zfs_has_ctldir(zp));
293 VN_HOLD(zp->z_zfsvfs->z_ctldir);
294 return (zp->z_zfsvfs->z_ctldir);
295 }
296
297 /*
298 * Common open routine. Disallow any write access.
299 */
300 /* ARGSUSED */
301 static int
zfsctl_common_open(struct vop_open_args * ap)302 zfsctl_common_open(struct vop_open_args *ap)
303 {
304 int flags = ap->a_mode;
305
306 if (flags & FWRITE)
307 return (SET_ERROR(EACCES));
308
309 return (0);
310 }
311
312 /*
313 * Common close routine. Nothing to do here.
314 */
315 /* ARGSUSED */
316 static int
zfsctl_common_close(struct vop_close_args * ap)317 zfsctl_common_close(struct vop_close_args *ap)
318 {
319 return (0);
320 }
321
322 /*
323 * Common access routine. Disallow writes.
324 */
325 /* ARGSUSED */
326 static int
zfsctl_common_access(ap)327 zfsctl_common_access(ap)
328 struct vop_access_args /* {
329 struct vnode *a_vp;
330 accmode_t a_accmode;
331 struct ucred *a_cred;
332 struct thread *a_td;
333 } */ *ap;
334 {
335 accmode_t accmode = ap->a_accmode;
336
337 #ifdef TODO
338 if (flags & V_ACE_MASK) {
339 if (accmode & ACE_ALL_WRITE_PERMS)
340 return (SET_ERROR(EACCES));
341 } else {
342 #endif
343 if (accmode & VWRITE)
344 return (SET_ERROR(EACCES));
345 #ifdef TODO
346 }
347 #endif
348
349 return (0);
350 }
351
352 /*
353 * Common getattr function. Fill in basic information.
354 */
355 static void
zfsctl_common_getattr(vnode_t * vp,vattr_t * vap)356 zfsctl_common_getattr(vnode_t *vp, vattr_t *vap)
357 {
358 timestruc_t now;
359
360 vap->va_uid = 0;
361 vap->va_gid = 0;
362 vap->va_rdev = 0;
363 /*
364 * We are a purely virtual object, so we have no
365 * blocksize or allocated blocks.
366 */
367 vap->va_blksize = 0;
368 vap->va_nblocks = 0;
369 vap->va_seq = 0;
370 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
371 vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP |
372 S_IROTH | S_IXOTH;
373 vap->va_type = VDIR;
374 /*
375 * We live in the now (for atime).
376 */
377 gethrestime(&now);
378 vap->va_atime = now;
379 /* FreeBSD: Reset chflags(2) flags. */
380 vap->va_flags = 0;
381 }
382
383 /*ARGSUSED*/
384 static int
zfsctl_common_fid(ap)385 zfsctl_common_fid(ap)
386 struct vop_fid_args /* {
387 struct vnode *a_vp;
388 struct fid *a_fid;
389 } */ *ap;
390 {
391 vnode_t *vp = ap->a_vp;
392 fid_t *fidp = (void *)ap->a_fid;
393 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
394 zfsctl_node_t *zcp = vp->v_data;
395 uint64_t object = zcp->zc_id;
396 zfid_short_t *zfid;
397 int i;
398
399 ZFS_ENTER(zfsvfs);
400
401 #ifdef illumos
402 if (fidp->fid_len < SHORT_FID_LEN) {
403 fidp->fid_len = SHORT_FID_LEN;
404 ZFS_EXIT(zfsvfs);
405 return (SET_ERROR(ENOSPC));
406 }
407 #else
408 fidp->fid_len = SHORT_FID_LEN;
409 #endif
410
411 zfid = (zfid_short_t *)fidp;
412
413 zfid->zf_len = SHORT_FID_LEN;
414
415 for (i = 0; i < sizeof (zfid->zf_object); i++)
416 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
417
418 /* .zfs znodes always have a generation number of 0 */
419 for (i = 0; i < sizeof (zfid->zf_gen); i++)
420 zfid->zf_gen[i] = 0;
421
422 ZFS_EXIT(zfsvfs);
423 return (0);
424 }
425
426
427 /*ARGSUSED*/
428 static int
zfsctl_shares_fid(ap)429 zfsctl_shares_fid(ap)
430 struct vop_fid_args /* {
431 struct vnode *a_vp;
432 struct fid *a_fid;
433 } */ *ap;
434 {
435 vnode_t *vp = ap->a_vp;
436 fid_t *fidp = (void *)ap->a_fid;
437 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
438 znode_t *dzp;
439 int error;
440
441 ZFS_ENTER(zfsvfs);
442
443 if (zfsvfs->z_shares_dir == 0) {
444 ZFS_EXIT(zfsvfs);
445 return (SET_ERROR(ENOTSUP));
446 }
447
448 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
449 error = VOP_FID(ZTOV(dzp), fidp);
450 VN_RELE(ZTOV(dzp));
451 }
452
453 ZFS_EXIT(zfsvfs);
454 return (error);
455 }
456
457 static int
zfsctl_common_reclaim(ap)458 zfsctl_common_reclaim(ap)
459 struct vop_reclaim_args /* {
460 struct vnode *a_vp;
461 struct thread *a_td;
462 } */ *ap;
463 {
464 vnode_t *vp = ap->a_vp;
465
466 /*
467 * Destroy the vm object and flush associated pages.
468 */
469 vnode_destroy_vobject(vp);
470 VI_LOCK(vp);
471 vp->v_data = NULL;
472 VI_UNLOCK(vp);
473 return (0);
474 }
475
476 /*
477 * .zfs inode namespace
478 *
479 * We need to generate unique inode numbers for all files and directories
480 * within the .zfs pseudo-filesystem. We use the following scheme:
481 *
482 * ENTRY ZFSCTL_INODE
483 * .zfs 1
484 * .zfs/snapshot 2
485 * .zfs/snapshot/<snap> objectid(snap)
486 */
487
488 #define ZFSCTL_INO_SNAP(id) (id)
489
490 /*
491 * Get root directory attributes.
492 */
493 /* ARGSUSED */
494 static int
zfsctl_root_getattr(ap)495 zfsctl_root_getattr(ap)
496 struct vop_getattr_args /* {
497 struct vnode *a_vp;
498 struct vattr *a_vap;
499 struct ucred *a_cred;
500 } */ *ap;
501 {
502 struct vnode *vp = ap->a_vp;
503 struct vattr *vap = ap->a_vap;
504 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
505 zfsctl_node_t *zcp = vp->v_data;
506
507 ZFS_ENTER(zfsvfs);
508 vap->va_nodeid = ZFSCTL_INO_ROOT;
509 vap->va_nlink = vap->va_size = NROOT_ENTRIES;
510 vap->va_mtime = vap->va_ctime = zcp->zc_cmtime;
511 vap->va_birthtime = vap->va_ctime;
512
513 zfsctl_common_getattr(vp, vap);
514 ZFS_EXIT(zfsvfs);
515
516 return (0);
517 }
518
519 /*
520 * Special case the handling of "..".
521 */
522 /* ARGSUSED */
523 int
zfsctl_root_lookup(vnode_t * dvp,char * nm,vnode_t ** vpp,pathname_t * pnp,int flags,vnode_t * rdir,cred_t * cr,caller_context_t * ct,int * direntflags,pathname_t * realpnp)524 zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
525 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
526 int *direntflags, pathname_t *realpnp)
527 {
528 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
529 int err;
530
531 /*
532 * No extended attributes allowed under .zfs
533 */
534 if (flags & LOOKUP_XATTR)
535 return (SET_ERROR(EINVAL));
536
537 ZFS_ENTER(zfsvfs);
538
539 if (strcmp(nm, "..") == 0) {
540 err = VFS_ROOT(dvp->v_vfsp, LK_EXCLUSIVE, vpp);
541 if (err == 0)
542 VOP_UNLOCK(*vpp, 0);
543 } else {
544 err = gfs_vop_lookup(dvp, nm, vpp, pnp, flags, rdir,
545 cr, ct, direntflags, realpnp);
546 }
547
548 ZFS_EXIT(zfsvfs);
549
550 return (err);
551 }
552
553 #ifdef illumos
554 static int
zfsctl_pathconf(vnode_t * vp,int cmd,ulong_t * valp,cred_t * cr,caller_context_t * ct)555 zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
556 caller_context_t *ct)
557 {
558 /*
559 * We only care about ACL_ENABLED so that libsec can
560 * display ACL correctly and not default to POSIX draft.
561 */
562 if (cmd == _PC_ACL_ENABLED) {
563 *valp = _ACL_ACE_ENABLED;
564 return (0);
565 }
566
567 return (fs_pathconf(vp, cmd, valp, cr, ct));
568 }
569 #endif /* illumos */
570
571 #ifdef illumos
572 static const fs_operation_def_t zfsctl_tops_root[] = {
573 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
574 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
575 { VOPNAME_IOCTL, { .error = fs_inval } },
576 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_root_getattr } },
577 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
578 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } },
579 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_root_lookup } },
580 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
581 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } },
582 { VOPNAME_PATHCONF, { .vop_pathconf = zfsctl_pathconf } },
583 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } },
584 { NULL }
585 };
586 #endif /* illumos */
587
588 /*
589 * Special case the handling of "..".
590 */
591 /* ARGSUSED */
592 int
zfsctl_freebsd_root_lookup(ap)593 zfsctl_freebsd_root_lookup(ap)
594 struct vop_lookup_args /* {
595 struct vnode *a_dvp;
596 struct vnode **a_vpp;
597 struct componentname *a_cnp;
598 } */ *ap;
599 {
600 vnode_t *dvp = ap->a_dvp;
601 vnode_t **vpp = ap->a_vpp;
602 cred_t *cr = ap->a_cnp->cn_cred;
603 int flags = ap->a_cnp->cn_flags;
604 int nameiop = ap->a_cnp->cn_nameiop;
605 char nm[NAME_MAX + 1];
606 int err;
607 int ltype;
608
609 if ((flags & ISLASTCN) && (nameiop == RENAME || nameiop == CREATE))
610 return (EOPNOTSUPP);
611
612 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm));
613 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1);
614 err = zfsctl_root_lookup(dvp, nm, vpp, NULL, 0, NULL, cr, NULL, NULL, NULL);
615 if (err == 0 && (nm[0] != '.' || nm[1] != '\0')) {
616 ltype = VOP_ISLOCKED(dvp);
617 if (flags & ISDOTDOT) {
618 VN_HOLD(*vpp);
619 VOP_UNLOCK(dvp, 0);
620 }
621 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
622 if (flags & ISDOTDOT) {
623 VN_RELE(*vpp);
624 vn_lock(dvp, ltype| LK_RETRY);
625 }
626 }
627
628 return (err);
629 }
630
631 static struct vop_vector zfsctl_ops_root = {
632 .vop_default = &default_vnodeops,
633 .vop_open = zfsctl_common_open,
634 .vop_close = zfsctl_common_close,
635 .vop_ioctl = VOP_EINVAL,
636 .vop_getattr = zfsctl_root_getattr,
637 .vop_access = zfsctl_common_access,
638 .vop_readdir = gfs_vop_readdir,
639 .vop_lookup = zfsctl_freebsd_root_lookup,
640 .vop_inactive = VOP_NULL,
641 .vop_reclaim = gfs_vop_reclaim,
642 #ifdef TODO
643 .vop_pathconf = zfsctl_pathconf,
644 #endif
645 .vop_fid = zfsctl_common_fid,
646 };
647
648 /*
649 * Gets the full dataset name that corresponds to the given snapshot name
650 * Example:
651 * zfsctl_snapshot_zname("snap1") -> "mypool/myfs@snap1"
652 */
653 static int
zfsctl_snapshot_zname(vnode_t * vp,const char * name,int len,char * zname)654 zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname)
655 {
656 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os;
657
658 if (zfs_component_namecheck(name, NULL, NULL) != 0)
659 return (SET_ERROR(EILSEQ));
660 dmu_objset_name(os, zname);
661 if (strlen(zname) + 1 + strlen(name) >= len)
662 return (SET_ERROR(ENAMETOOLONG));
663 (void) strcat(zname, "@");
664 (void) strcat(zname, name);
665 return (0);
666 }
667
668 static int
zfsctl_unmount_snap(zfs_snapentry_t * sep,int fflags,cred_t * cr)669 zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr)
670 {
671 vnode_t *svp = sep->se_root;
672 int error;
673
674 ASSERT(vn_ismntpt(svp));
675
676 /* this will be dropped by dounmount() */
677 if ((error = vn_vfswlock(svp)) != 0)
678 return (error);
679
680 #ifdef illumos
681 VN_HOLD(svp);
682 error = dounmount(vn_mountedvfs(svp), fflags, cr);
683 if (error) {
684 VN_RELE(svp);
685 return (error);
686 }
687
688 /*
689 * We can't use VN_RELE(), as that will try to invoke
690 * zfsctl_snapdir_inactive(), which would cause us to destroy
691 * the sd_lock mutex held by our caller.
692 */
693 ASSERT(svp->v_count == 1);
694 gfs_vop_reclaim(svp, cr, NULL);
695
696 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
697 kmem_free(sep, sizeof (zfs_snapentry_t));
698
699 return (0);
700 #else
701 vfs_ref(vn_mountedvfs(svp));
702 return (dounmount(vn_mountedvfs(svp), fflags, curthread));
703 #endif
704 }
705
706 #ifdef illumos
707 static void
zfsctl_rename_snap(zfsctl_snapdir_t * sdp,zfs_snapentry_t * sep,const char * nm)708 zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm)
709 {
710 avl_index_t where;
711 vfs_t *vfsp;
712 refstr_t *pathref;
713 char newpath[MAXNAMELEN];
714 char *tail;
715
716 ASSERT(MUTEX_HELD(&sdp->sd_lock));
717 ASSERT(sep != NULL);
718
719 vfsp = vn_mountedvfs(sep->se_root);
720 ASSERT(vfsp != NULL);
721
722 vfs_lock_wait(vfsp);
723
724 /*
725 * Change the name in the AVL tree.
726 */
727 avl_remove(&sdp->sd_snaps, sep);
728 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
729 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
730 (void) strcpy(sep->se_name, nm);
731 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL);
732 avl_insert(&sdp->sd_snaps, sep, where);
733
734 /*
735 * Change the current mountpoint info:
736 * - update the tail of the mntpoint path
737 * - update the tail of the resource path
738 */
739 pathref = vfs_getmntpoint(vfsp);
740 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
741 VERIFY((tail = strrchr(newpath, '/')) != NULL);
742 *(tail+1) = '\0';
743 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
744 (void) strcat(newpath, nm);
745 refstr_rele(pathref);
746 vfs_setmntpoint(vfsp, newpath, 0);
747
748 pathref = vfs_getresource(vfsp);
749 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
750 VERIFY((tail = strrchr(newpath, '@')) != NULL);
751 *(tail+1) = '\0';
752 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
753 (void) strcat(newpath, nm);
754 refstr_rele(pathref);
755 vfs_setresource(vfsp, newpath, 0);
756
757 vfs_unlock(vfsp);
758 }
759 #endif /* illumos */
760
761 #ifdef illumos
762 /*ARGSUSED*/
763 static int
zfsctl_snapdir_rename(vnode_t * sdvp,char * snm,vnode_t * tdvp,char * tnm,cred_t * cr,caller_context_t * ct,int flags)764 zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm,
765 cred_t *cr, caller_context_t *ct, int flags)
766 {
767 zfsctl_snapdir_t *sdp = sdvp->v_data;
768 zfs_snapentry_t search, *sep;
769 zfsvfs_t *zfsvfs;
770 avl_index_t where;
771 char from[MAXNAMELEN], to[MAXNAMELEN];
772 char real[MAXNAMELEN], fsname[MAXNAMELEN];
773 int err;
774
775 zfsvfs = sdvp->v_vfsp->vfs_data;
776 ZFS_ENTER(zfsvfs);
777
778 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
779 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real,
780 MAXNAMELEN, NULL);
781 if (err == 0) {
782 snm = real;
783 } else if (err != ENOTSUP) {
784 ZFS_EXIT(zfsvfs);
785 return (err);
786 }
787 }
788
789 ZFS_EXIT(zfsvfs);
790
791 dmu_objset_name(zfsvfs->z_os, fsname);
792
793 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from);
794 if (err == 0)
795 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to);
796 if (err == 0)
797 err = zfs_secpolicy_rename_perms(from, to, cr);
798 if (err != 0)
799 return (err);
800
801 /*
802 * Cannot move snapshots out of the snapdir.
803 */
804 if (sdvp != tdvp)
805 return (SET_ERROR(EINVAL));
806
807 if (strcmp(snm, tnm) == 0)
808 return (0);
809
810 mutex_enter(&sdp->sd_lock);
811
812 search.se_name = (char *)snm;
813 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) {
814 mutex_exit(&sdp->sd_lock);
815 return (SET_ERROR(ENOENT));
816 }
817
818 err = dsl_dataset_rename_snapshot(fsname, snm, tnm, 0);
819 if (err == 0)
820 zfsctl_rename_snap(sdp, sep, tnm);
821
822 mutex_exit(&sdp->sd_lock);
823
824 return (err);
825 }
826 #endif /* illumos */
827
828 #ifdef illumos
829 /* ARGSUSED */
830 static int
zfsctl_snapdir_remove(vnode_t * dvp,char * name,vnode_t * cwd,cred_t * cr,caller_context_t * ct,int flags)831 zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
832 caller_context_t *ct, int flags)
833 {
834 zfsctl_snapdir_t *sdp = dvp->v_data;
835 zfs_snapentry_t *sep;
836 zfs_snapentry_t search;
837 zfsvfs_t *zfsvfs;
838 char snapname[MAXNAMELEN];
839 char real[MAXNAMELEN];
840 int err;
841
842 zfsvfs = dvp->v_vfsp->vfs_data;
843 ZFS_ENTER(zfsvfs);
844
845 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
846
847 err = dmu_snapshot_realname(zfsvfs->z_os, name, real,
848 MAXNAMELEN, NULL);
849 if (err == 0) {
850 name = real;
851 } else if (err != ENOTSUP) {
852 ZFS_EXIT(zfsvfs);
853 return (err);
854 }
855 }
856
857 ZFS_EXIT(zfsvfs);
858
859 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname);
860 if (err == 0)
861 err = zfs_secpolicy_destroy_perms(snapname, cr);
862 if (err != 0)
863 return (err);
864
865 mutex_enter(&sdp->sd_lock);
866
867 search.se_name = name;
868 sep = avl_find(&sdp->sd_snaps, &search, NULL);
869 if (sep) {
870 avl_remove(&sdp->sd_snaps, sep);
871 err = zfsctl_unmount_snap(sep, MS_FORCE, cr);
872 if (err != 0)
873 avl_add(&sdp->sd_snaps, sep);
874 else
875 err = dsl_destroy_snapshot(snapname, B_FALSE);
876 } else {
877 err = SET_ERROR(ENOENT);
878 }
879
880 mutex_exit(&sdp->sd_lock);
881
882 return (err);
883 }
884 #endif /* illumos */
885
886 /*
887 * This creates a snapshot under '.zfs/snapshot'.
888 */
889 /* ARGSUSED */
890 static int
zfsctl_snapdir_mkdir(vnode_t * dvp,char * dirname,vattr_t * vap,vnode_t ** vpp,cred_t * cr,caller_context_t * cc,int flags,vsecattr_t * vsecp)891 zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp,
892 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp)
893 {
894 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
895 char name[MAXNAMELEN];
896 int err;
897 static enum symfollow follow = NO_FOLLOW;
898 static enum uio_seg seg = UIO_SYSSPACE;
899
900 if (zfs_component_namecheck(dirname, NULL, NULL) != 0)
901 return (SET_ERROR(EILSEQ));
902
903 dmu_objset_name(zfsvfs->z_os, name);
904
905 *vpp = NULL;
906
907 err = zfs_secpolicy_snapshot_perms(name, cr);
908 if (err != 0)
909 return (err);
910
911 if (err == 0) {
912 err = dmu_objset_snapshot_one(name, dirname);
913 if (err != 0)
914 return (err);
915 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp);
916 }
917
918 return (err);
919 }
920
921 static int
zfsctl_freebsd_snapdir_mkdir(ap)922 zfsctl_freebsd_snapdir_mkdir(ap)
923 struct vop_mkdir_args /* {
924 struct vnode *a_dvp;
925 struct vnode **a_vpp;
926 struct componentname *a_cnp;
927 struct vattr *a_vap;
928 } */ *ap;
929 {
930
931 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
932
933 return (zfsctl_snapdir_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, NULL,
934 ap->a_vpp, ap->a_cnp->cn_cred, NULL, 0, NULL));
935 }
936
937 /*
938 * Lookup entry point for the 'snapshot' directory. Try to open the
939 * snapshot if it exist, creating the pseudo filesystem vnode as necessary.
940 * Perform a mount of the associated dataset on top of the vnode.
941 */
942 /* ARGSUSED */
943 int
zfsctl_snapdir_lookup(ap)944 zfsctl_snapdir_lookup(ap)
945 struct vop_lookup_args /* {
946 struct vnode *a_dvp;
947 struct vnode **a_vpp;
948 struct componentname *a_cnp;
949 } */ *ap;
950 {
951 vnode_t *dvp = ap->a_dvp;
952 vnode_t **vpp = ap->a_vpp;
953 struct componentname *cnp = ap->a_cnp;
954 char nm[NAME_MAX + 1];
955 zfsctl_snapdir_t *sdp = dvp->v_data;
956 objset_t *snap;
957 char snapname[MAXNAMELEN];
958 char real[MAXNAMELEN];
959 char *mountpoint;
960 zfs_snapentry_t *sep, search;
961 size_t mountpoint_len;
962 avl_index_t where;
963 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
964 int err;
965 int ltype, flags = 0;
966
967 /*
968 * No extended attributes allowed under .zfs
969 */
970 if (flags & LOOKUP_XATTR)
971 return (SET_ERROR(EINVAL));
972 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm));
973 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1);
974
975 ASSERT(dvp->v_type == VDIR);
976
977 *vpp = NULL;
978
979 /*
980 * If we get a recursive call, that means we got called
981 * from the domount() code while it was trying to look up the
982 * spec (which looks like a local path for zfs). We need to
983 * add some flag to domount() to tell it not to do this lookup.
984 */
985 if (MUTEX_HELD(&sdp->sd_lock))
986 return (SET_ERROR(ENOENT));
987
988 ZFS_ENTER(zfsvfs);
989 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
990 ZFS_EXIT(zfsvfs);
991 return (0);
992 }
993
994 if (flags & FIGNORECASE) {
995 boolean_t conflict = B_FALSE;
996
997 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real,
998 MAXNAMELEN, &conflict);
999 if (err == 0) {
1000 strlcpy(nm, real, sizeof(nm));
1001 } else if (err != ENOTSUP) {
1002 ZFS_EXIT(zfsvfs);
1003 return (err);
1004 }
1005 #if 0
1006 if (realpnp)
1007 (void) strlcpy(realpnp->pn_buf, nm,
1008 realpnp->pn_bufsize);
1009 if (conflict && direntflags)
1010 *direntflags = ED_CASE_CONFLICT;
1011 #endif
1012 }
1013
1014 mutex_enter(&sdp->sd_lock);
1015 search.se_name = (char *)nm;
1016 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) {
1017 *vpp = sep->se_root;
1018 VN_HOLD(*vpp);
1019 err = traverse(vpp, LK_EXCLUSIVE | LK_RETRY);
1020 if (err != 0) {
1021 VN_RELE(*vpp);
1022 *vpp = NULL;
1023 } else if (*vpp == sep->se_root) {
1024 /*
1025 * The snapshot was unmounted behind our backs,
1026 * try to remount it.
1027 */
1028 VERIFY(zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname) == 0);
1029 goto domount;
1030 } else {
1031 /*
1032 * VROOT was set during the traverse call. We need
1033 * to clear it since we're pretending to be part
1034 * of our parent's vfs.
1035 */
1036 (*vpp)->v_flag &= ~VROOT;
1037 }
1038 mutex_exit(&sdp->sd_lock);
1039 ZFS_EXIT(zfsvfs);
1040 return (err);
1041 }
1042
1043 /*
1044 * The requested snapshot is not currently mounted, look it up.
1045 */
1046 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname);
1047 if (err != 0) {
1048 mutex_exit(&sdp->sd_lock);
1049 ZFS_EXIT(zfsvfs);
1050 /*
1051 * handle "ls *" or "?" in a graceful manner,
1052 * forcing EILSEQ to ENOENT.
1053 * Since shell ultimately passes "*" or "?" as name to lookup
1054 */
1055 return (err == EILSEQ ? ENOENT : err);
1056 }
1057 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) {
1058 mutex_exit(&sdp->sd_lock);
1059 #ifdef illumos
1060 ZFS_EXIT(zfsvfs);
1061 return (SET_ERROR(ENOENT));
1062 #else /* !illumos */
1063 /* Translate errors and add SAVENAME when needed. */
1064 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) {
1065 err = EJUSTRETURN;
1066 cnp->cn_flags |= SAVENAME;
1067 } else {
1068 err = SET_ERROR(ENOENT);
1069 }
1070 ZFS_EXIT(zfsvfs);
1071 return (err);
1072 #endif /* illumos */
1073 }
1074
1075 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP);
1076 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
1077 (void) strcpy(sep->se_name, nm);
1078 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap));
1079 VN_HOLD(*vpp);
1080 avl_insert(&sdp->sd_snaps, sep, where);
1081
1082 dmu_objset_rele(snap, FTAG);
1083 domount:
1084 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) +
1085 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(nm) + 1;
1086 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP);
1087 (void) snprintf(mountpoint, mountpoint_len,
1088 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s",
1089 dvp->v_vfsp->mnt_stat.f_mntonname, nm);
1090 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, snapname, 0);
1091 kmem_free(mountpoint, mountpoint_len);
1092 if (err == 0) {
1093 /*
1094 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>.
1095 *
1096 * This is where we lie about our v_vfsp in order to
1097 * make .zfs/snapshot/<snapname> accessible over NFS
1098 * without requiring manual mounts of <snapname>.
1099 */
1100 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs);
1101 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs;
1102 }
1103 mutex_exit(&sdp->sd_lock);
1104 ZFS_EXIT(zfsvfs);
1105
1106 #ifdef illumos
1107 /*
1108 * If we had an error, drop our hold on the vnode and
1109 * zfsctl_snapshot_inactive() will clean up.
1110 */
1111 if (err != 0) {
1112 VN_RELE(*vpp);
1113 *vpp = NULL;
1114 }
1115 #else
1116 if (err != 0)
1117 *vpp = NULL;
1118 #endif
1119 return (err);
1120 }
1121
1122 /* ARGSUSED */
1123 int
zfsctl_shares_lookup(ap)1124 zfsctl_shares_lookup(ap)
1125 struct vop_lookup_args /* {
1126 struct vnode *a_dvp;
1127 struct vnode **a_vpp;
1128 struct componentname *a_cnp;
1129 } */ *ap;
1130 {
1131 vnode_t *dvp = ap->a_dvp;
1132 vnode_t **vpp = ap->a_vpp;
1133 struct componentname *cnp = ap->a_cnp;
1134 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
1135 char nm[NAME_MAX + 1];
1136 znode_t *dzp;
1137 int error;
1138
1139 ZFS_ENTER(zfsvfs);
1140
1141 ASSERT(cnp->cn_namelen < sizeof(nm));
1142 strlcpy(nm, cnp->cn_nameptr, cnp->cn_namelen + 1);
1143
1144 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
1145 ZFS_EXIT(zfsvfs);
1146 return (0);
1147 }
1148
1149 if (zfsvfs->z_shares_dir == 0) {
1150 ZFS_EXIT(zfsvfs);
1151 return (SET_ERROR(ENOTSUP));
1152 }
1153 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1154 error = VOP_LOOKUP(ZTOV(dzp), vpp, cnp);
1155 VN_RELE(ZTOV(dzp));
1156 }
1157
1158 ZFS_EXIT(zfsvfs);
1159
1160 return (error);
1161 }
1162
1163 /* ARGSUSED */
1164 static int
zfsctl_snapdir_readdir_cb(vnode_t * vp,void * dp,int * eofp,offset_t * offp,offset_t * nextp,void * data,int flags)1165 zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp,
1166 offset_t *offp, offset_t *nextp, void *data, int flags)
1167 {
1168 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1169 char snapname[MAXNAMELEN];
1170 uint64_t id, cookie;
1171 boolean_t case_conflict;
1172 int error;
1173
1174 ZFS_ENTER(zfsvfs);
1175
1176 cookie = *offp;
1177 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG);
1178 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id,
1179 &cookie, &case_conflict);
1180 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG);
1181 if (error) {
1182 ZFS_EXIT(zfsvfs);
1183 if (error == ENOENT) {
1184 *eofp = 1;
1185 return (0);
1186 }
1187 return (error);
1188 }
1189
1190 if (flags & V_RDDIR_ENTFLAGS) {
1191 edirent_t *eodp = dp;
1192
1193 (void) strcpy(eodp->ed_name, snapname);
1194 eodp->ed_ino = ZFSCTL_INO_SNAP(id);
1195 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0;
1196 } else {
1197 struct dirent64 *odp = dp;
1198
1199 (void) strcpy(odp->d_name, snapname);
1200 odp->d_ino = ZFSCTL_INO_SNAP(id);
1201 }
1202 *nextp = cookie;
1203
1204 ZFS_EXIT(zfsvfs);
1205
1206 return (0);
1207 }
1208
1209 /* ARGSUSED */
1210 static int
zfsctl_shares_readdir(ap)1211 zfsctl_shares_readdir(ap)
1212 struct vop_readdir_args /* {
1213 struct vnode *a_vp;
1214 struct uio *a_uio;
1215 struct ucred *a_cred;
1216 int *a_eofflag;
1217 int *a_ncookies;
1218 u_long **a_cookies;
1219 } */ *ap;
1220 {
1221 vnode_t *vp = ap->a_vp;
1222 uio_t *uiop = ap->a_uio;
1223 cred_t *cr = ap->a_cred;
1224 int *eofp = ap->a_eofflag;
1225 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1226 znode_t *dzp;
1227 int error;
1228
1229 ZFS_ENTER(zfsvfs);
1230
1231 if (zfsvfs->z_shares_dir == 0) {
1232 ZFS_EXIT(zfsvfs);
1233 return (SET_ERROR(ENOTSUP));
1234 }
1235 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1236 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY);
1237 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ap->a_ncookies, ap->a_cookies);
1238 VN_URELE(ZTOV(dzp));
1239 } else {
1240 *eofp = 1;
1241 error = SET_ERROR(ENOENT);
1242 }
1243
1244 ZFS_EXIT(zfsvfs);
1245 return (error);
1246 }
1247
1248 /*
1249 * pvp is the '.zfs' directory (zfsctl_node_t).
1250 *
1251 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t).
1252 *
1253 * This function is the callback to create a GFS vnode for '.zfs/snapshot'
1254 * when a lookup is performed on .zfs for "snapshot".
1255 */
1256 vnode_t *
zfsctl_mknode_snapdir(vnode_t * pvp)1257 zfsctl_mknode_snapdir(vnode_t *pvp)
1258 {
1259 vnode_t *vp;
1260 zfsctl_snapdir_t *sdp;
1261
1262 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp,
1263 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN,
1264 zfsctl_snapdir_readdir_cb, NULL);
1265 sdp = vp->v_data;
1266 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR;
1267 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1268 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL);
1269 avl_create(&sdp->sd_snaps, snapentry_compare,
1270 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node));
1271 VOP_UNLOCK(vp, 0);
1272 return (vp);
1273 }
1274
1275 vnode_t *
zfsctl_mknode_shares(vnode_t * pvp)1276 zfsctl_mknode_shares(vnode_t *pvp)
1277 {
1278 vnode_t *vp;
1279 zfsctl_node_t *sdp;
1280
1281 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp,
1282 &zfsctl_ops_shares, NULL, NULL, MAXNAMELEN,
1283 NULL, NULL);
1284 sdp = vp->v_data;
1285 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1286 VOP_UNLOCK(vp, 0);
1287 return (vp);
1288
1289 }
1290
1291 /* ARGSUSED */
1292 static int
zfsctl_shares_getattr(ap)1293 zfsctl_shares_getattr(ap)
1294 struct vop_getattr_args /* {
1295 struct vnode *a_vp;
1296 struct vattr *a_vap;
1297 struct ucred *a_cred;
1298 struct thread *a_td;
1299 } */ *ap;
1300 {
1301 vnode_t *vp = ap->a_vp;
1302 vattr_t *vap = ap->a_vap;
1303 cred_t *cr = ap->a_cred;
1304 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1305 znode_t *dzp;
1306 int error;
1307
1308 ZFS_ENTER(zfsvfs);
1309 if (zfsvfs->z_shares_dir == 0) {
1310 ZFS_EXIT(zfsvfs);
1311 return (SET_ERROR(ENOTSUP));
1312 }
1313 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1314 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY);
1315 error = VOP_GETATTR(ZTOV(dzp), vap, cr);
1316 VN_URELE(ZTOV(dzp));
1317 }
1318 ZFS_EXIT(zfsvfs);
1319 return (error);
1320
1321
1322 }
1323
1324 /* ARGSUSED */
1325 static int
zfsctl_snapdir_getattr(ap)1326 zfsctl_snapdir_getattr(ap)
1327 struct vop_getattr_args /* {
1328 struct vnode *a_vp;
1329 struct vattr *a_vap;
1330 struct ucred *a_cred;
1331 } */ *ap;
1332 {
1333 vnode_t *vp = ap->a_vp;
1334 vattr_t *vap = ap->a_vap;
1335 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1336 zfsctl_snapdir_t *sdp = vp->v_data;
1337
1338 ZFS_ENTER(zfsvfs);
1339 zfsctl_common_getattr(vp, vap);
1340 vap->va_nodeid = gfs_file_inode(vp);
1341 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2;
1342 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os);
1343 vap->va_birthtime = vap->va_ctime;
1344 ZFS_EXIT(zfsvfs);
1345
1346 return (0);
1347 }
1348
1349 /* ARGSUSED */
1350 static int
zfsctl_snapdir_inactive(ap)1351 zfsctl_snapdir_inactive(ap)
1352 struct vop_inactive_args /* {
1353 struct vnode *a_vp;
1354 struct thread *a_td;
1355 } */ *ap;
1356 {
1357 vnode_t *vp = ap->a_vp;
1358 zfsctl_snapdir_t *sdp = vp->v_data;
1359 zfs_snapentry_t *sep;
1360
1361 /*
1362 * On forced unmount we have to free snapshots from here.
1363 */
1364 mutex_enter(&sdp->sd_lock);
1365 while ((sep = avl_first(&sdp->sd_snaps)) != NULL) {
1366 avl_remove(&sdp->sd_snaps, sep);
1367 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
1368 kmem_free(sep, sizeof (zfs_snapentry_t));
1369 }
1370 mutex_exit(&sdp->sd_lock);
1371 gfs_dir_inactive(vp);
1372 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0);
1373 mutex_destroy(&sdp->sd_lock);
1374 avl_destroy(&sdp->sd_snaps);
1375 kmem_free(sdp, sizeof (zfsctl_snapdir_t));
1376
1377 return (0);
1378 }
1379
1380 #ifdef illumos
1381 static const fs_operation_def_t zfsctl_tops_snapdir[] = {
1382 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
1383 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
1384 { VOPNAME_IOCTL, { .error = fs_inval } },
1385 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } },
1386 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
1387 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } },
1388 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } },
1389 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } },
1390 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } },
1391 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } },
1392 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
1393 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } },
1394 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } },
1395 { NULL }
1396 };
1397
1398 static const fs_operation_def_t zfsctl_tops_shares[] = {
1399 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
1400 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
1401 { VOPNAME_IOCTL, { .error = fs_inval } },
1402 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } },
1403 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
1404 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } },
1405 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } },
1406 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
1407 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } },
1408 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } },
1409 { NULL }
1410 };
1411 #else /* !illumos */
1412 static struct vop_vector zfsctl_ops_snapdir = {
1413 .vop_default = &default_vnodeops,
1414 .vop_open = zfsctl_common_open,
1415 .vop_close = zfsctl_common_close,
1416 .vop_ioctl = VOP_EINVAL,
1417 .vop_getattr = zfsctl_snapdir_getattr,
1418 .vop_access = zfsctl_common_access,
1419 .vop_mkdir = zfsctl_freebsd_snapdir_mkdir,
1420 .vop_readdir = gfs_vop_readdir,
1421 .vop_lookup = zfsctl_snapdir_lookup,
1422 .vop_inactive = zfsctl_snapdir_inactive,
1423 .vop_reclaim = zfsctl_common_reclaim,
1424 .vop_fid = zfsctl_common_fid,
1425 };
1426
1427 static struct vop_vector zfsctl_ops_shares = {
1428 .vop_default = &default_vnodeops,
1429 .vop_open = zfsctl_common_open,
1430 .vop_close = zfsctl_common_close,
1431 .vop_ioctl = VOP_EINVAL,
1432 .vop_getattr = zfsctl_shares_getattr,
1433 .vop_access = zfsctl_common_access,
1434 .vop_readdir = zfsctl_shares_readdir,
1435 .vop_lookup = zfsctl_shares_lookup,
1436 .vop_inactive = VOP_NULL,
1437 .vop_reclaim = gfs_vop_reclaim,
1438 .vop_fid = zfsctl_shares_fid,
1439 };
1440 #endif /* illumos */
1441
1442 /*
1443 * pvp is the GFS vnode '.zfs/snapshot'.
1444 *
1445 * This creates a GFS node under '.zfs/snapshot' representing each
1446 * snapshot. This newly created GFS node is what we mount snapshot
1447 * vfs_t's ontop of.
1448 */
1449 static vnode_t *
zfsctl_snapshot_mknode(vnode_t * pvp,uint64_t objset)1450 zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset)
1451 {
1452 vnode_t *vp;
1453 zfsctl_node_t *zcp;
1454
1455 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp,
1456 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL);
1457 VN_HOLD(vp);
1458 zcp = vp->v_data;
1459 zcp->zc_id = objset;
1460 VOP_UNLOCK(vp, 0);
1461
1462 return (vp);
1463 }
1464
1465
1466 static int
zfsctl_snapshot_reclaim(ap)1467 zfsctl_snapshot_reclaim(ap)
1468 struct vop_inactive_args /* {
1469 struct vnode *a_vp;
1470 struct thread *a_td;
1471 } */ *ap;
1472 {
1473 vnode_t *vp = ap->a_vp;
1474 cred_t *cr = ap->a_td->td_ucred;
1475 struct vop_reclaim_args iap;
1476 zfsctl_snapdir_t *sdp;
1477 zfs_snapentry_t *sep, *next;
1478 int locked;
1479 vnode_t *dvp;
1480
1481 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0);
1482 sdp = dvp->v_data;
1483 VOP_UNLOCK(dvp, 0);
1484 /* this may already have been unmounted */
1485 if (sdp == NULL) {
1486 VN_RELE(dvp);
1487 return (0);
1488 }
1489 if (!(locked = MUTEX_HELD(&sdp->sd_lock)))
1490 mutex_enter(&sdp->sd_lock);
1491
1492 ASSERT(!vn_ismntpt(vp));
1493
1494 sep = avl_first(&sdp->sd_snaps);
1495 while (sep != NULL) {
1496 next = AVL_NEXT(&sdp->sd_snaps, sep);
1497
1498 if (sep->se_root == vp) {
1499 avl_remove(&sdp->sd_snaps, sep);
1500 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
1501 kmem_free(sep, sizeof (zfs_snapentry_t));
1502 break;
1503 }
1504 sep = next;
1505 }
1506 ASSERT(sep != NULL);
1507
1508 if (!locked)
1509 mutex_exit(&sdp->sd_lock);
1510 VN_RELE(dvp);
1511
1512 /*
1513 * Dispose of the vnode for the snapshot mount point.
1514 * This is safe to do because once this entry has been removed
1515 * from the AVL tree, it can't be found again, so cannot become
1516 * "active". If we lookup the same name again we will end up
1517 * creating a new vnode.
1518 */
1519 iap.a_vp = vp;
1520 gfs_vop_reclaim(&iap);
1521 return (0);
1522
1523 }
1524
1525 static int
zfsctl_traverse_begin(vnode_t ** vpp,int lktype)1526 zfsctl_traverse_begin(vnode_t **vpp, int lktype)
1527 {
1528
1529 VN_HOLD(*vpp);
1530 /* Snapshot should be already mounted, but just in case. */
1531 if (vn_mountedvfs(*vpp) == NULL)
1532 return (ENOENT);
1533 return (traverse(vpp, lktype));
1534 }
1535
1536 static void
zfsctl_traverse_end(vnode_t * vp,int err)1537 zfsctl_traverse_end(vnode_t *vp, int err)
1538 {
1539
1540 if (err == 0)
1541 vput(vp);
1542 else
1543 VN_RELE(vp);
1544 }
1545
1546 static int
zfsctl_snapshot_getattr(ap)1547 zfsctl_snapshot_getattr(ap)
1548 struct vop_getattr_args /* {
1549 struct vnode *a_vp;
1550 struct vattr *a_vap;
1551 struct ucred *a_cred;
1552 } */ *ap;
1553 {
1554 vnode_t *vp = ap->a_vp;
1555 int err;
1556
1557 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY);
1558 if (err == 0)
1559 err = VOP_GETATTR(vp, ap->a_vap, ap->a_cred);
1560 zfsctl_traverse_end(vp, err);
1561 return (err);
1562 }
1563
1564 static int
zfsctl_snapshot_fid(ap)1565 zfsctl_snapshot_fid(ap)
1566 struct vop_fid_args /* {
1567 struct vnode *a_vp;
1568 struct fid *a_fid;
1569 } */ *ap;
1570 {
1571 vnode_t *vp = ap->a_vp;
1572 int err;
1573
1574 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY);
1575 if (err == 0)
1576 err = VOP_VPTOFH(vp, (void *)ap->a_fid);
1577 zfsctl_traverse_end(vp, err);
1578 return (err);
1579 }
1580
1581 static int
zfsctl_snapshot_lookup(ap)1582 zfsctl_snapshot_lookup(ap)
1583 struct vop_lookup_args /* {
1584 struct vnode *a_dvp;
1585 struct vnode **a_vpp;
1586 struct componentname *a_cnp;
1587 } */ *ap;
1588 {
1589 vnode_t *dvp = ap->a_dvp;
1590 vnode_t **vpp = ap->a_vpp;
1591 struct componentname *cnp = ap->a_cnp;
1592 cred_t *cr = ap->a_cnp->cn_cred;
1593 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
1594 int error;
1595
1596 if (cnp->cn_namelen != 2 || cnp->cn_nameptr[0] != '.' ||
1597 cnp->cn_nameptr[1] != '.') {
1598 return (ENOENT);
1599 }
1600
1601 ASSERT(dvp->v_type == VDIR);
1602 ASSERT(zfsvfs->z_ctldir != NULL);
1603
1604 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", vpp,
1605 NULL, 0, NULL, cr, NULL, NULL, NULL);
1606 if (error == 0) {
1607 int ltype = VOP_ISLOCKED(dvp);
1608 VN_HOLD(*vpp);
1609 VOP_UNLOCK(dvp, 0);
1610 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
1611 VN_RELE(*vpp);
1612 vn_lock(dvp, ltype | LK_RETRY);
1613 }
1614
1615 return (error);
1616 }
1617
1618 static int
zfsctl_snapshot_vptocnp(struct vop_vptocnp_args * ap)1619 zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap)
1620 {
1621 zfsvfs_t *zfsvfs = ap->a_vp->v_vfsp->vfs_data;
1622 vnode_t *dvp, *vp;
1623 zfsctl_snapdir_t *sdp;
1624 zfs_snapentry_t *sep;
1625 int error;
1626
1627 ASSERT(zfsvfs->z_ctldir != NULL);
1628 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1629 NULL, 0, NULL, kcred, NULL, NULL, NULL);
1630 if (error != 0)
1631 return (error);
1632 sdp = dvp->v_data;
1633
1634 mutex_enter(&sdp->sd_lock);
1635 sep = avl_first(&sdp->sd_snaps);
1636 while (sep != NULL) {
1637 vp = sep->se_root;
1638 if (vp == ap->a_vp)
1639 break;
1640 sep = AVL_NEXT(&sdp->sd_snaps, sep);
1641 }
1642 if (sep == NULL) {
1643 mutex_exit(&sdp->sd_lock);
1644 error = ENOENT;
1645 } else {
1646 size_t len;
1647
1648 len = strlen(sep->se_name);
1649 *ap->a_buflen -= len;
1650 bcopy(sep->se_name, ap->a_buf + *ap->a_buflen, len);
1651 mutex_exit(&sdp->sd_lock);
1652 vref(dvp);
1653 *ap->a_vpp = dvp;
1654 }
1655 VN_RELE(dvp);
1656
1657 return (error);
1658 }
1659
1660 /*
1661 * These VP's should never see the light of day. They should always
1662 * be covered.
1663 */
1664 static struct vop_vector zfsctl_ops_snapshot = {
1665 .vop_default = &default_vnodeops,
1666 .vop_inactive = VOP_NULL,
1667 .vop_lookup = zfsctl_snapshot_lookup,
1668 .vop_reclaim = zfsctl_snapshot_reclaim,
1669 .vop_getattr = zfsctl_snapshot_getattr,
1670 .vop_fid = zfsctl_snapshot_fid,
1671 .vop_vptocnp = zfsctl_snapshot_vptocnp,
1672 };
1673
1674 int
zfsctl_lookup_objset(vfs_t * vfsp,uint64_t objsetid,zfsvfs_t ** zfsvfsp)1675 zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp)
1676 {
1677 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1678 vnode_t *dvp, *vp;
1679 zfsctl_snapdir_t *sdp;
1680 zfsctl_node_t *zcp;
1681 zfs_snapentry_t *sep;
1682 int error;
1683
1684 ASSERT(zfsvfs->z_ctldir != NULL);
1685 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1686 NULL, 0, NULL, kcred, NULL, NULL, NULL);
1687 if (error != 0)
1688 return (error);
1689 sdp = dvp->v_data;
1690
1691 mutex_enter(&sdp->sd_lock);
1692 sep = avl_first(&sdp->sd_snaps);
1693 while (sep != NULL) {
1694 vp = sep->se_root;
1695 zcp = vp->v_data;
1696 if (zcp->zc_id == objsetid)
1697 break;
1698
1699 sep = AVL_NEXT(&sdp->sd_snaps, sep);
1700 }
1701
1702 if (sep != NULL) {
1703 VN_HOLD(vp);
1704 /*
1705 * Return the mounted root rather than the covered mount point.
1706 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid>
1707 * and returns the ZFS vnode mounted on top of the GFS node.
1708 * This ZFS vnode is the root of the vfs for objset 'objsetid'.
1709 */
1710 error = traverse(&vp, LK_SHARED | LK_RETRY);
1711 if (error == 0) {
1712 if (vp == sep->se_root)
1713 error = SET_ERROR(EINVAL);
1714 else
1715 *zfsvfsp = VTOZ(vp)->z_zfsvfs;
1716 }
1717 mutex_exit(&sdp->sd_lock);
1718 if (error == 0)
1719 VN_URELE(vp);
1720 else
1721 VN_RELE(vp);
1722 } else {
1723 error = SET_ERROR(EINVAL);
1724 mutex_exit(&sdp->sd_lock);
1725 }
1726
1727 VN_RELE(dvp);
1728
1729 return (error);
1730 }
1731
1732 /*
1733 * Unmount any snapshots for the given filesystem. This is called from
1734 * zfs_umount() - if we have a ctldir, then go through and unmount all the
1735 * snapshots.
1736 */
1737 int
zfsctl_umount_snapshots(vfs_t * vfsp,int fflags,cred_t * cr)1738 zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr)
1739 {
1740 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1741 vnode_t *dvp;
1742 zfsctl_snapdir_t *sdp;
1743 zfs_snapentry_t *sep, *next;
1744 int error;
1745
1746 ASSERT(zfsvfs->z_ctldir != NULL);
1747 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1748 NULL, 0, NULL, cr, NULL, NULL, NULL);
1749 if (error != 0)
1750 return (error);
1751 sdp = dvp->v_data;
1752
1753 mutex_enter(&sdp->sd_lock);
1754
1755 sep = avl_first(&sdp->sd_snaps);
1756 while (sep != NULL) {
1757 next = AVL_NEXT(&sdp->sd_snaps, sep);
1758
1759 /*
1760 * If this snapshot is not mounted, then it must
1761 * have just been unmounted by somebody else, and
1762 * will be cleaned up by zfsctl_snapdir_inactive().
1763 */
1764 if (vn_ismntpt(sep->se_root)) {
1765 error = zfsctl_unmount_snap(sep, fflags, cr);
1766 if (error) {
1767 avl_index_t where;
1768
1769 /*
1770 * Before reinserting snapshot to the tree,
1771 * check if it was actually removed. For example
1772 * when snapshot mount point is busy, we will
1773 * have an error here, but there will be no need
1774 * to reinsert snapshot.
1775 */
1776 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL)
1777 avl_insert(&sdp->sd_snaps, sep, where);
1778 break;
1779 }
1780 }
1781 sep = next;
1782 }
1783
1784 mutex_exit(&sdp->sd_lock);
1785 VN_RELE(dvp);
1786
1787 return (error);
1788 }
1789