]>
Commit | Line | Data |
---|---|---|
ebe7e575 BB |
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 | * | |
23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
24 | * Copyright (C) 2011 Lawrence Livermore National Security, LLC. | |
25 | * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER). | |
26 | * LLNL-CODE-403049. | |
27 | * Rewritten for Linux by: | |
28 | * Rohan Puri <rohan.puri15@gmail.com> | |
29 | * Brian Behlendorf <behlendorf1@llnl.gov> | |
2e528b49 | 30 | * Copyright (c) 2013 by Delphix. All rights reserved. |
8adb798a | 31 | * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved. |
ebe7e575 BB |
32 | */ |
33 | ||
34 | /* | |
35 | * ZFS control directory (a.k.a. ".zfs") | |
36 | * | |
37 | * This directory provides a common location for all ZFS meta-objects. | |
38 | * Currently, this is only the 'snapshot' and 'shares' directory, but this may | |
39 | * expand in the future. The elements are built dynamically, as the hierarchy | |
40 | * does not actually exist on disk. | |
41 | * | |
42 | * For 'snapshot', we don't want to have all snapshots always mounted, because | |
43 | * this would take up a huge amount of space in /etc/mnttab. We have three | |
44 | * types of objects: | |
45 | * | |
46 | * ctldir ------> snapshotdir -------> snapshot | |
47 | * | | |
48 | * | | |
49 | * V | |
50 | * mounted fs | |
51 | * | |
52 | * The 'snapshot' node contains just enough information to lookup '..' and act | |
53 | * as a mountpoint for the snapshot. Whenever we lookup a specific snapshot, we | |
54 | * perform an automount of the underlying filesystem and return the | |
55 | * corresponding inode. | |
56 | * | |
57 | * All mounts are handled automatically by an user mode helper which invokes | |
58 | * the mount mount procedure. Unmounts are handled by allowing the mount | |
59 | * point to expire so the kernel may automatically unmount it. | |
60 | * | |
61 | * The '.zfs', '.zfs/snapshot', and all directories created under | |
62 | * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') all share the same | |
0037b49e | 63 | * share the same zfsvfs_t as the head filesystem (what '.zfs' lives under). |
ebe7e575 BB |
64 | * |
65 | * File systems mounted on top of the '.zfs/snapshot/<snapname>' paths | |
66 | * (ie: snapshots) are complete ZFS filesystems and have their own unique | |
0037b49e BB |
67 | * zfsvfs_t. However, the fsid reported by these mounts will be the same |
68 | * as that used by the parent zfsvfs_t to make NFS happy. | |
ebe7e575 BB |
69 | */ |
70 | ||
71 | #include <sys/types.h> | |
72 | #include <sys/param.h> | |
73 | #include <sys/time.h> | |
ebe7e575 BB |
74 | #include <sys/sysmacros.h> |
75 | #include <sys/pathname.h> | |
76 | #include <sys/vfs.h> | |
ebe7e575 BB |
77 | #include <sys/zfs_ctldir.h> |
78 | #include <sys/zfs_ioctl.h> | |
79 | #include <sys/zfs_vfsops.h> | |
80 | #include <sys/zfs_vnops.h> | |
81 | #include <sys/stat.h> | |
82 | #include <sys/dmu.h> | |
24ef51f6 | 83 | #include <sys/dmu_objset.h> |
13fe0198 | 84 | #include <sys/dsl_destroy.h> |
ebe7e575 | 85 | #include <sys/dsl_deleg.h> |
ebe7e575 BB |
86 | #include <sys/zpl.h> |
87 | #include "zfs_namecheck.h" | |
88 | ||
278bee93 BB |
89 | /* |
90 | * Two AVL trees are maintained which contain all currently automounted | |
91 | * snapshots. Every automounted snapshots maps to a single zfs_snapentry_t | |
92 | * entry which MUST: | |
93 | * | |
94 | * - be attached to both trees, and | |
95 | * - be unique, no duplicate entries are allowed. | |
96 | * | |
97 | * The zfs_snapshots_by_name tree is indexed by the full dataset name | |
98 | * while the zfs_snapshots_by_objsetid tree is indexed by the unique | |
99 | * objsetid. This allows for fast lookups either by name or objsetid. | |
100 | */ | |
101 | static avl_tree_t zfs_snapshots_by_name; | |
102 | static avl_tree_t zfs_snapshots_by_objsetid; | |
5ed27c57 | 103 | static krwlock_t zfs_snapshot_lock; |
278bee93 | 104 | |
ebe7e575 BB |
105 | /* |
106 | * Control Directory Tunables (.zfs) | |
107 | */ | |
108 | int zfs_expire_snapshot = ZFSCTL_EXPIRE_SNAPSHOT; | |
27ca030f | 109 | int zfs_admin_snapshot = 1; |
ebe7e575 | 110 | |
278bee93 BB |
111 | typedef struct { |
112 | char *se_name; /* full snapshot name */ | |
113 | char *se_path; /* full mount path */ | |
24ef51f6 | 114 | spa_t *se_spa; /* pool spa */ |
278bee93 BB |
115 | uint64_t se_objsetid; /* snapshot objset id */ |
116 | struct dentry *se_root_dentry; /* snapshot root dentry */ | |
117 | taskqid_t se_taskqid; /* scheduled unmount taskqid */ | |
118 | avl_node_t se_node_name; /* zfs_snapshots_by_name link */ | |
119 | avl_node_t se_node_objsetid; /* zfs_snapshots_by_objsetid link */ | |
c13060e4 | 120 | zfs_refcount_t se_refcount; /* reference count */ |
278bee93 BB |
121 | } zfs_snapentry_t; |
122 | ||
123 | static void zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t *se, int delay); | |
124 | ||
125 | /* | |
126 | * Allocate a new zfs_snapentry_t being careful to make a copy of the | |
127 | * the snapshot name and provided mount point. No reference is taken. | |
128 | */ | |
ebe7e575 | 129 | static zfs_snapentry_t * |
24ef51f6 CC |
130 | zfsctl_snapshot_alloc(char *full_name, char *full_path, spa_t *spa, |
131 | uint64_t objsetid, struct dentry *root_dentry) | |
ebe7e575 | 132 | { |
278bee93 BB |
133 | zfs_snapentry_t *se; |
134 | ||
135 | se = kmem_zalloc(sizeof (zfs_snapentry_t), KM_SLEEP); | |
136 | ||
137 | se->se_name = strdup(full_name); | |
138 | se->se_path = strdup(full_path); | |
24ef51f6 | 139 | se->se_spa = spa; |
278bee93 BB |
140 | se->se_objsetid = objsetid; |
141 | se->se_root_dentry = root_dentry; | |
48d3eb40 | 142 | se->se_taskqid = TASKQID_INVALID; |
278bee93 | 143 | |
424fd7c3 | 144 | zfs_refcount_create(&se->se_refcount); |
278bee93 BB |
145 | |
146 | return (se); | |
ebe7e575 BB |
147 | } |
148 | ||
278bee93 BB |
149 | /* |
150 | * Free a zfs_snapentry_t the called must ensure there are no active | |
151 | * references. | |
152 | */ | |
153 | static void | |
154 | zfsctl_snapshot_free(zfs_snapentry_t *se) | |
ebe7e575 | 155 | { |
424fd7c3 | 156 | zfs_refcount_destroy(&se->se_refcount); |
278bee93 BB |
157 | strfree(se->se_name); |
158 | strfree(se->se_path); | |
159 | ||
160 | kmem_free(se, sizeof (zfs_snapentry_t)); | |
ebe7e575 BB |
161 | } |
162 | ||
163 | /* | |
278bee93 | 164 | * Hold a reference on the zfs_snapentry_t. |
ebe7e575 BB |
165 | */ |
166 | static void | |
278bee93 | 167 | zfsctl_snapshot_hold(zfs_snapentry_t *se) |
ebe7e575 | 168 | { |
c13060e4 | 169 | zfs_refcount_add(&se->se_refcount, NULL); |
278bee93 BB |
170 | } |
171 | ||
172 | /* | |
173 | * Release a reference on the zfs_snapentry_t. When the number of | |
174 | * references drops to zero the structure will be freed. | |
175 | */ | |
176 | static void | |
177 | zfsctl_snapshot_rele(zfs_snapentry_t *se) | |
178 | { | |
424fd7c3 | 179 | if (zfs_refcount_remove(&se->se_refcount, NULL) == 0) |
278bee93 BB |
180 | zfsctl_snapshot_free(se); |
181 | } | |
ebe7e575 | 182 | |
278bee93 BB |
183 | /* |
184 | * Add a zfs_snapentry_t to both the zfs_snapshots_by_name and | |
185 | * zfs_snapshots_by_objsetid trees. While the zfs_snapentry_t is part | |
186 | * of the trees a reference is held. | |
187 | */ | |
188 | static void | |
189 | zfsctl_snapshot_add(zfs_snapentry_t *se) | |
190 | { | |
5ed27c57 | 191 | ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock)); |
c13060e4 | 192 | zfs_refcount_add(&se->se_refcount, NULL); |
278bee93 BB |
193 | avl_add(&zfs_snapshots_by_name, se); |
194 | avl_add(&zfs_snapshots_by_objsetid, se); | |
ebe7e575 BB |
195 | } |
196 | ||
278bee93 BB |
197 | /* |
198 | * Remove a zfs_snapentry_t from both the zfs_snapshots_by_name and | |
199 | * zfs_snapshots_by_objsetid trees. Upon removal a reference is dropped, | |
200 | * this can result in the structure being freed if that was the last | |
201 | * remaining reference. | |
202 | */ | |
203 | static void | |
204 | zfsctl_snapshot_remove(zfs_snapentry_t *se) | |
205 | { | |
5ed27c57 | 206 | ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock)); |
278bee93 BB |
207 | avl_remove(&zfs_snapshots_by_name, se); |
208 | avl_remove(&zfs_snapshots_by_objsetid, se); | |
209 | zfsctl_snapshot_rele(se); | |
210 | } | |
211 | ||
212 | /* | |
213 | * Snapshot name comparison function for the zfs_snapshots_by_name. | |
214 | */ | |
215 | static int | |
216 | snapentry_compare_by_name(const void *a, const void *b) | |
ebe7e575 | 217 | { |
278bee93 BB |
218 | const zfs_snapentry_t *se_a = a; |
219 | const zfs_snapentry_t *se_b = b; | |
220 | int ret; | |
221 | ||
222 | ret = strcmp(se_a->se_name, se_b->se_name); | |
ebe7e575 BB |
223 | |
224 | if (ret < 0) | |
225 | return (-1); | |
226 | else if (ret > 0) | |
227 | return (1); | |
228 | else | |
229 | return (0); | |
230 | } | |
231 | ||
278bee93 BB |
232 | /* |
233 | * Snapshot name comparison function for the zfs_snapshots_by_objsetid. | |
234 | */ | |
235 | static int | |
236 | snapentry_compare_by_objsetid(const void *a, const void *b) | |
237 | { | |
238 | const zfs_snapentry_t *se_a = a; | |
239 | const zfs_snapentry_t *se_b = b; | |
240 | ||
24ef51f6 CC |
241 | if (se_a->se_spa != se_b->se_spa) |
242 | return ((ulong_t)se_a->se_spa < (ulong_t)se_b->se_spa ? -1 : 1); | |
243 | ||
278bee93 BB |
244 | if (se_a->se_objsetid < se_b->se_objsetid) |
245 | return (-1); | |
246 | else if (se_a->se_objsetid > se_b->se_objsetid) | |
247 | return (1); | |
248 | else | |
249 | return (0); | |
250 | } | |
251 | ||
252 | /* | |
253 | * Find a zfs_snapentry_t in zfs_snapshots_by_name. If the snapname | |
254 | * is found a pointer to the zfs_snapentry_t is returned and a reference | |
255 | * taken on the structure. The caller is responsible for dropping the | |
256 | * reference with zfsctl_snapshot_rele(). If the snapname is not found | |
257 | * NULL will be returned. | |
258 | */ | |
259 | static zfs_snapentry_t * | |
260 | zfsctl_snapshot_find_by_name(char *snapname) | |
261 | { | |
262 | zfs_snapentry_t *se, search; | |
263 | ||
5ed27c57 | 264 | ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock)); |
278bee93 BB |
265 | |
266 | search.se_name = snapname; | |
267 | se = avl_find(&zfs_snapshots_by_name, &search, NULL); | |
268 | if (se) | |
c13060e4 | 269 | zfs_refcount_add(&se->se_refcount, NULL); |
278bee93 BB |
270 | |
271 | return (se); | |
272 | } | |
273 | ||
274 | /* | |
275 | * Find a zfs_snapentry_t in zfs_snapshots_by_objsetid given the objset id | |
276 | * rather than the snapname. In all other respects it behaves the same | |
277 | * as zfsctl_snapshot_find_by_name(). | |
278 | */ | |
279 | static zfs_snapentry_t * | |
24ef51f6 | 280 | zfsctl_snapshot_find_by_objsetid(spa_t *spa, uint64_t objsetid) |
278bee93 BB |
281 | { |
282 | zfs_snapentry_t *se, search; | |
283 | ||
5ed27c57 | 284 | ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock)); |
278bee93 | 285 | |
24ef51f6 | 286 | search.se_spa = spa; |
278bee93 BB |
287 | search.se_objsetid = objsetid; |
288 | se = avl_find(&zfs_snapshots_by_objsetid, &search, NULL); | |
289 | if (se) | |
c13060e4 | 290 | zfs_refcount_add(&se->se_refcount, NULL); |
278bee93 BB |
291 | |
292 | return (se); | |
293 | } | |
294 | ||
295 | /* | |
296 | * Rename a zfs_snapentry_t in the zfs_snapshots_by_name. The structure is | |
297 | * removed, renamed, and added back to the new correct location in the tree. | |
298 | */ | |
299 | static int | |
300 | zfsctl_snapshot_rename(char *old_snapname, char *new_snapname) | |
301 | { | |
302 | zfs_snapentry_t *se; | |
303 | ||
5ed27c57 | 304 | ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock)); |
278bee93 BB |
305 | |
306 | se = zfsctl_snapshot_find_by_name(old_snapname); | |
307 | if (se == NULL) | |
ecb2b7dc | 308 | return (SET_ERROR(ENOENT)); |
278bee93 BB |
309 | |
310 | zfsctl_snapshot_remove(se); | |
311 | strfree(se->se_name); | |
312 | se->se_name = strdup(new_snapname); | |
313 | zfsctl_snapshot_add(se); | |
314 | zfsctl_snapshot_rele(se); | |
315 | ||
316 | return (0); | |
317 | } | |
318 | ||
319 | /* | |
320 | * Delayed task responsible for unmounting an expired automounted snapshot. | |
321 | */ | |
322 | static void | |
323 | snapentry_expire(void *data) | |
324 | { | |
325 | zfs_snapentry_t *se = (zfs_snapentry_t *)data; | |
24ef51f6 | 326 | spa_t *spa = se->se_spa; |
278bee93 BB |
327 | uint64_t objsetid = se->se_objsetid; |
328 | ||
5e94284f BB |
329 | if (zfs_expire_snapshot <= 0) { |
330 | zfsctl_snapshot_rele(se); | |
331 | return; | |
332 | } | |
333 | ||
48d3eb40 | 334 | se->se_taskqid = TASKQID_INVALID; |
278bee93 BB |
335 | (void) zfsctl_snapshot_unmount(se->se_name, MNT_EXPIRE); |
336 | zfsctl_snapshot_rele(se); | |
337 | ||
338 | /* | |
339 | * Reschedule the unmount if the zfs_snapentry_t wasn't removed. | |
340 | * This can occur when the snapshot is busy. | |
341 | */ | |
5ed27c57 | 342 | rw_enter(&zfs_snapshot_lock, RW_READER); |
24ef51f6 | 343 | if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid)) != NULL) { |
278bee93 BB |
344 | zfsctl_snapshot_unmount_delay_impl(se, zfs_expire_snapshot); |
345 | zfsctl_snapshot_rele(se); | |
346 | } | |
5ed27c57 | 347 | rw_exit(&zfs_snapshot_lock); |
278bee93 BB |
348 | } |
349 | ||
350 | /* | |
351 | * Cancel an automatic unmount of a snapname. This callback is responsible | |
352 | * for dropping the reference on the zfs_snapentry_t which was taken when | |
353 | * during dispatch. | |
354 | */ | |
355 | static void | |
356 | zfsctl_snapshot_unmount_cancel(zfs_snapentry_t *se) | |
357 | { | |
57ddcda1 | 358 | if (taskq_cancel_id(system_delay_taskq, se->se_taskqid) == 0) { |
48d3eb40 | 359 | se->se_taskqid = TASKQID_INVALID; |
278bee93 BB |
360 | zfsctl_snapshot_rele(se); |
361 | } | |
362 | } | |
363 | ||
364 | /* | |
365 | * Dispatch the unmount task for delayed handling with a hold protecting it. | |
366 | */ | |
367 | static void | |
368 | zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t *se, int delay) | |
369 | { | |
48d3eb40 | 370 | ASSERT3S(se->se_taskqid, ==, TASKQID_INVALID); |
278bee93 | 371 | |
5e94284f BB |
372 | if (delay <= 0) |
373 | return; | |
374 | ||
245b7ab3 | 375 | zfsctl_snapshot_hold(se); |
57ddcda1 | 376 | se->se_taskqid = taskq_dispatch_delay(system_delay_taskq, |
278bee93 | 377 | snapentry_expire, se, TQ_SLEEP, ddi_get_lbolt() + delay * HZ); |
278bee93 BB |
378 | } |
379 | ||
380 | /* | |
381 | * Schedule an automatic unmount of objset id to occur in delay seconds from | |
382 | * now. Any previous delayed unmount will be cancelled in favor of the | |
383 | * updated deadline. A reference is taken by zfsctl_snapshot_find_by_name() | |
384 | * and held until the outstanding task is handled or cancelled. | |
385 | */ | |
386 | int | |
24ef51f6 | 387 | zfsctl_snapshot_unmount_delay(spa_t *spa, uint64_t objsetid, int delay) |
278bee93 BB |
388 | { |
389 | zfs_snapentry_t *se; | |
390 | int error = ENOENT; | |
391 | ||
5ed27c57 | 392 | rw_enter(&zfs_snapshot_lock, RW_READER); |
24ef51f6 | 393 | if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid)) != NULL) { |
278bee93 BB |
394 | zfsctl_snapshot_unmount_cancel(se); |
395 | zfsctl_snapshot_unmount_delay_impl(se, delay); | |
396 | zfsctl_snapshot_rele(se); | |
397 | error = 0; | |
398 | } | |
5ed27c57 | 399 | rw_exit(&zfs_snapshot_lock); |
278bee93 BB |
400 | |
401 | return (error); | |
402 | } | |
403 | ||
404 | /* | |
405 | * Check if snapname is currently mounted. Returned non-zero when mounted | |
406 | * and zero when unmounted. | |
407 | */ | |
408 | static boolean_t | |
409 | zfsctl_snapshot_ismounted(char *snapname) | |
410 | { | |
411 | zfs_snapentry_t *se; | |
412 | boolean_t ismounted = B_FALSE; | |
413 | ||
5ed27c57 | 414 | rw_enter(&zfs_snapshot_lock, RW_READER); |
278bee93 BB |
415 | if ((se = zfsctl_snapshot_find_by_name(snapname)) != NULL) { |
416 | zfsctl_snapshot_rele(se); | |
417 | ismounted = B_TRUE; | |
418 | } | |
5ed27c57 | 419 | rw_exit(&zfs_snapshot_lock); |
278bee93 BB |
420 | |
421 | return (ismounted); | |
422 | } | |
423 | ||
424 | /* | |
425 | * Check if the given inode is a part of the virtual .zfs directory. | |
426 | */ | |
ebe7e575 BB |
427 | boolean_t |
428 | zfsctl_is_node(struct inode *ip) | |
429 | { | |
430 | return (ITOZ(ip)->z_is_ctldir); | |
431 | } | |
432 | ||
278bee93 BB |
433 | /* |
434 | * Check if the given inode is a .zfs/snapshots/snapname directory. | |
435 | */ | |
ebe7e575 BB |
436 | boolean_t |
437 | zfsctl_is_snapdir(struct inode *ip) | |
438 | { | |
439 | return (zfsctl_is_node(ip) && (ip->i_ino <= ZFSCTL_INO_SNAPDIRS)); | |
440 | } | |
441 | ||
442 | /* | |
443 | * Allocate a new inode with the passed id and ops. | |
444 | */ | |
445 | static struct inode * | |
0037b49e | 446 | zfsctl_inode_alloc(zfsvfs_t *zfsvfs, uint64_t id, |
ebe7e575 BB |
447 | const struct file_operations *fops, const struct inode_operations *ops) |
448 | { | |
6413c95f | 449 | inode_timespec_t now; |
ebe7e575 BB |
450 | struct inode *ip; |
451 | znode_t *zp; | |
452 | ||
0037b49e | 453 | ip = new_inode(zfsvfs->z_sb); |
ebe7e575 BB |
454 | if (ip == NULL) |
455 | return (NULL); | |
456 | ||
2946a1a1 | 457 | now = current_time(ip); |
ebe7e575 BB |
458 | zp = ITOZ(ip); |
459 | ASSERT3P(zp->z_dirlocks, ==, NULL); | |
460 | ASSERT3P(zp->z_acl_cached, ==, NULL); | |
461 | ASSERT3P(zp->z_xattr_cached, ==, NULL); | |
462 | zp->z_id = id; | |
463 | zp->z_unlinked = 0; | |
464 | zp->z_atime_dirty = 0; | |
465 | zp->z_zn_prefetch = 0; | |
466 | zp->z_moved = 0; | |
467 | zp->z_sa_hdl = NULL; | |
468 | zp->z_blksz = 0; | |
469 | zp->z_seq = 0; | |
470 | zp->z_mapcnt = 0; | |
ebe7e575 | 471 | zp->z_size = 0; |
ebe7e575 | 472 | zp->z_pflags = 0; |
ebe7e575 BB |
473 | zp->z_mode = 0; |
474 | zp->z_sync_cnt = 0; | |
ebe7e575 BB |
475 | zp->z_is_mapped = B_FALSE; |
476 | zp->z_is_ctldir = B_TRUE; | |
477 | zp->z_is_sa = B_FALSE; | |
7b3e34ba | 478 | zp->z_is_stale = B_FALSE; |
278f2236 | 479 | ip->i_generation = 0; |
ebe7e575 | 480 | ip->i_ino = id; |
f74b821a | 481 | ip->i_mode = (S_IFDIR | S_IRWXUGO); |
570d6edf RY |
482 | ip->i_uid = SUID_TO_KUID(0); |
483 | ip->i_gid = SGID_TO_KGID(0); | |
ebe7e575 BB |
484 | ip->i_blkbits = SPA_MINBLOCKSHIFT; |
485 | ip->i_atime = now; | |
486 | ip->i_mtime = now; | |
487 | ip->i_ctime = now; | |
488 | ip->i_fop = fops; | |
489 | ip->i_op = ops; | |
9f7b066b | 490 | #if defined(IOP_XATTR) |
491 | ip->i_opflags &= ~IOP_XATTR; | |
492 | #endif | |
ebe7e575 BB |
493 | |
494 | if (insert_inode_locked(ip)) { | |
495 | unlock_new_inode(ip); | |
496 | iput(ip); | |
497 | return (NULL); | |
498 | } | |
499 | ||
0037b49e BB |
500 | mutex_enter(&zfsvfs->z_znodes_lock); |
501 | list_insert_tail(&zfsvfs->z_all_znodes, zp); | |
502 | zfsvfs->z_nr_znodes++; | |
ebe7e575 | 503 | membar_producer(); |
0037b49e | 504 | mutex_exit(&zfsvfs->z_znodes_lock); |
ebe7e575 BB |
505 | |
506 | unlock_new_inode(ip); | |
507 | ||
508 | return (ip); | |
509 | } | |
510 | ||
511 | /* | |
512 | * Lookup the inode with given id, it will be allocated if needed. | |
513 | */ | |
514 | static struct inode * | |
0037b49e | 515 | zfsctl_inode_lookup(zfsvfs_t *zfsvfs, uint64_t id, |
ebe7e575 BB |
516 | const struct file_operations *fops, const struct inode_operations *ops) |
517 | { | |
518 | struct inode *ip = NULL; | |
519 | ||
520 | while (ip == NULL) { | |
0037b49e | 521 | ip = ilookup(zfsvfs->z_sb, (unsigned long)id); |
ebe7e575 BB |
522 | if (ip) |
523 | break; | |
524 | ||
525 | /* May fail due to concurrent zfsctl_inode_alloc() */ | |
0037b49e | 526 | ip = zfsctl_inode_alloc(zfsvfs, id, fops, ops); |
ebe7e575 BB |
527 | } |
528 | ||
529 | return (ip); | |
530 | } | |
531 | ||
ebe7e575 BB |
532 | /* |
533 | * Create the '.zfs' directory. This directory is cached as part of the VFS | |
0037b49e | 534 | * structure. This results in a hold on the zfsvfs_t. The code in zfs_umount() |
ebe7e575 BB |
535 | * therefore checks against a vfs_count of 2 instead of 1. This reference |
536 | * is removed when the ctldir is destroyed in the unmount. All other entities | |
537 | * under the '.zfs' directory are created dynamically as needed. | |
fc173c85 BB |
538 | * |
539 | * Because the dynamically created '.zfs' directory entries assume the use | |
540 | * of 64-bit inode numbers this support must be disabled on 32-bit systems. | |
ebe7e575 BB |
541 | */ |
542 | int | |
0037b49e | 543 | zfsctl_create(zfsvfs_t *zfsvfs) |
ebe7e575 | 544 | { |
0037b49e | 545 | ASSERT(zfsvfs->z_ctldir == NULL); |
ebe7e575 | 546 | |
0037b49e | 547 | zfsvfs->z_ctldir = zfsctl_inode_alloc(zfsvfs, ZFSCTL_INO_ROOT, |
ebe7e575 | 548 | &zpl_fops_root, &zpl_ops_root); |
0037b49e | 549 | if (zfsvfs->z_ctldir == NULL) |
2e528b49 | 550 | return (SET_ERROR(ENOENT)); |
ebe7e575 BB |
551 | |
552 | return (0); | |
553 | } | |
554 | ||
555 | /* | |
278bee93 BB |
556 | * Destroy the '.zfs' directory or remove a snapshot from zfs_snapshots_by_name. |
557 | * Only called when the filesystem is unmounted. | |
ebe7e575 BB |
558 | */ |
559 | void | |
0037b49e | 560 | zfsctl_destroy(zfsvfs_t *zfsvfs) |
ebe7e575 | 561 | { |
0037b49e | 562 | if (zfsvfs->z_issnap) { |
278bee93 | 563 | zfs_snapentry_t *se; |
0037b49e BB |
564 | spa_t *spa = zfsvfs->z_os->os_spa; |
565 | uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); | |
278bee93 | 566 | |
5ed27c57 | 567 | rw_enter(&zfs_snapshot_lock, RW_WRITER); |
fd7265c6 RP |
568 | se = zfsctl_snapshot_find_by_objsetid(spa, objsetid); |
569 | if (se != NULL) | |
278bee93 | 570 | zfsctl_snapshot_remove(se); |
fd7265c6 RP |
571 | rw_exit(&zfs_snapshot_lock); |
572 | if (se != NULL) { | |
573 | zfsctl_snapshot_unmount_cancel(se); | |
278bee93 BB |
574 | zfsctl_snapshot_rele(se); |
575 | } | |
0037b49e BB |
576 | } else if (zfsvfs->z_ctldir) { |
577 | iput(zfsvfs->z_ctldir); | |
578 | zfsvfs->z_ctldir = NULL; | |
278bee93 | 579 | } |
ebe7e575 BB |
580 | } |
581 | ||
582 | /* | |
583 | * Given a root znode, retrieve the associated .zfs directory. | |
584 | * Add a hold to the vnode and return it. | |
585 | */ | |
586 | struct inode * | |
587 | zfsctl_root(znode_t *zp) | |
588 | { | |
589 | ASSERT(zfs_has_ctldir(zp)); | |
590 | igrab(ZTOZSB(zp)->z_ctldir); | |
591 | return (ZTOZSB(zp)->z_ctldir); | |
592 | } | |
9b77d1c9 | 593 | |
0500e835 | 594 | /* |
9b77d1c9 CC |
595 | * Generate a long fid to indicate a snapdir. We encode whether snapdir is |
596 | * already monunted in gen field. We do this because nfsd lookup will not | |
597 | * trigger automount. Next time the nfsd does fh_to_dentry, we will notice | |
598 | * this and do automount and return ESTALE to force nfsd revalidate and follow | |
599 | * mount. | |
0500e835 BB |
600 | */ |
601 | static int | |
602 | zfsctl_snapdir_fid(struct inode *ip, fid_t *fidp) | |
603 | { | |
0500e835 BB |
604 | zfid_short_t *zfid = (zfid_short_t *)fidp; |
605 | zfid_long_t *zlfid = (zfid_long_t *)fidp; | |
606 | uint32_t gen = 0; | |
607 | uint64_t object; | |
608 | uint64_t objsetid; | |
609 | int i; | |
9b77d1c9 CC |
610 | struct dentry *dentry; |
611 | ||
612 | if (fidp->fid_len < LONG_FID_LEN) { | |
613 | fidp->fid_len = LONG_FID_LEN; | |
614 | return (SET_ERROR(ENOSPC)); | |
615 | } | |
0500e835 | 616 | |
9b77d1c9 | 617 | object = ip->i_ino; |
0500e835 BB |
618 | objsetid = ZFSCTL_INO_SNAPDIRS - ip->i_ino; |
619 | zfid->zf_len = LONG_FID_LEN; | |
620 | ||
9b77d1c9 CC |
621 | dentry = d_obtain_alias(igrab(ip)); |
622 | if (!IS_ERR(dentry)) { | |
623 | gen = !!d_mountpoint(dentry); | |
624 | dput(dentry); | |
625 | } | |
626 | ||
0500e835 BB |
627 | for (i = 0; i < sizeof (zfid->zf_object); i++) |
628 | zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); | |
629 | ||
630 | for (i = 0; i < sizeof (zfid->zf_gen); i++) | |
631 | zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); | |
632 | ||
633 | for (i = 0; i < sizeof (zlfid->zf_setid); i++) | |
634 | zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); | |
635 | ||
636 | for (i = 0; i < sizeof (zlfid->zf_setgen); i++) | |
637 | zlfid->zf_setgen[i] = 0; | |
638 | ||
639 | return (0); | |
640 | } | |
ebe7e575 | 641 | |
0500e835 BB |
642 | /* |
643 | * Generate an appropriate fid for an entry in the .zfs directory. | |
644 | */ | |
ebe7e575 BB |
645 | int |
646 | zfsctl_fid(struct inode *ip, fid_t *fidp) | |
647 | { | |
648 | znode_t *zp = ITOZ(ip); | |
0037b49e | 649 | zfsvfs_t *zfsvfs = ITOZSB(ip); |
ebe7e575 BB |
650 | uint64_t object = zp->z_id; |
651 | zfid_short_t *zfid; | |
652 | int i; | |
653 | ||
0037b49e | 654 | ZFS_ENTER(zfsvfs); |
ebe7e575 | 655 | |
9b77d1c9 | 656 | if (zfsctl_is_snapdir(ip)) { |
0037b49e | 657 | ZFS_EXIT(zfsvfs); |
9b77d1c9 | 658 | return (zfsctl_snapdir_fid(ip, fidp)); |
ebe7e575 BB |
659 | } |
660 | ||
9b77d1c9 CC |
661 | if (fidp->fid_len < SHORT_FID_LEN) { |
662 | fidp->fid_len = SHORT_FID_LEN; | |
0037b49e | 663 | ZFS_EXIT(zfsvfs); |
9b77d1c9 | 664 | return (SET_ERROR(ENOSPC)); |
0500e835 BB |
665 | } |
666 | ||
ebe7e575 BB |
667 | zfid = (zfid_short_t *)fidp; |
668 | ||
669 | zfid->zf_len = SHORT_FID_LEN; | |
670 | ||
671 | for (i = 0; i < sizeof (zfid->zf_object); i++) | |
672 | zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); | |
673 | ||
674 | /* .zfs znodes always have a generation number of 0 */ | |
675 | for (i = 0; i < sizeof (zfid->zf_gen); i++) | |
676 | zfid->zf_gen[i] = 0; | |
677 | ||
0037b49e | 678 | ZFS_EXIT(zfsvfs); |
ebe7e575 BB |
679 | return (0); |
680 | } | |
681 | ||
278bee93 BB |
682 | /* |
683 | * Construct a full dataset name in full_name: "pool/dataset@snap_name" | |
684 | */ | |
ebe7e575 | 685 | static int |
0037b49e | 686 | zfsctl_snapshot_name(zfsvfs_t *zfsvfs, const char *snap_name, int len, |
278bee93 | 687 | char *full_name) |
ebe7e575 | 688 | { |
0037b49e | 689 | objset_t *os = zfsvfs->z_os; |
ebe7e575 | 690 | |
278bee93 | 691 | if (zfs_component_namecheck(snap_name, NULL, NULL) != 0) |
2e528b49 | 692 | return (SET_ERROR(EILSEQ)); |
ebe7e575 | 693 | |
278bee93 BB |
694 | dmu_objset_name(os, full_name); |
695 | if ((strlen(full_name) + 1 + strlen(snap_name)) >= len) | |
2e528b49 | 696 | return (SET_ERROR(ENAMETOOLONG)); |
ebe7e575 | 697 | |
278bee93 BB |
698 | (void) strcat(full_name, "@"); |
699 | (void) strcat(full_name, snap_name); | |
ebe7e575 BB |
700 | |
701 | return (0); | |
702 | } | |
703 | ||
e49f1e20 | 704 | /* |
278bee93 | 705 | * Returns full path in full_path: "/pool/dataset/.zfs/snapshot/snap_name/" |
e49f1e20 | 706 | */ |
ebe7e575 | 707 | static int |
278bee93 | 708 | zfsctl_snapshot_path(struct path *path, int len, char *full_path) |
ebe7e575 BB |
709 | { |
710 | char *path_buffer, *path_ptr; | |
711 | int path_len, error = 0; | |
712 | ||
713 | path_buffer = kmem_alloc(len, KM_SLEEP); | |
714 | ||
715 | path_ptr = d_path(path, path_buffer, len); | |
716 | if (IS_ERR(path_ptr)) { | |
717 | error = -PTR_ERR(path_ptr); | |
718 | goto out; | |
719 | } | |
720 | ||
721 | path_len = path_buffer + len - 1 - path_ptr; | |
722 | if (path_len > len) { | |
2e528b49 | 723 | error = SET_ERROR(EFAULT); |
ebe7e575 BB |
724 | goto out; |
725 | } | |
726 | ||
278bee93 BB |
727 | memcpy(full_path, path_ptr, path_len); |
728 | full_path[path_len] = '\0'; | |
ebe7e575 BB |
729 | out: |
730 | kmem_free(path_buffer, len); | |
731 | ||
732 | return (error); | |
733 | } | |
734 | ||
0500e835 BB |
735 | /* |
736 | * Returns full path in full_path: "/pool/dataset/.zfs/snapshot/snap_name/" | |
737 | */ | |
738 | static int | |
0037b49e | 739 | zfsctl_snapshot_path_objset(zfsvfs_t *zfsvfs, uint64_t objsetid, |
0500e835 BB |
740 | int path_len, char *full_path) |
741 | { | |
0037b49e | 742 | objset_t *os = zfsvfs->z_os; |
0500e835 BB |
743 | fstrans_cookie_t cookie; |
744 | char *snapname; | |
745 | boolean_t case_conflict; | |
746 | uint64_t id, pos = 0; | |
747 | int error = 0; | |
748 | ||
1c2555ef | 749 | if (zfsvfs->z_vfs->vfs_mntpoint == NULL) |
ecb2b7dc | 750 | return (SET_ERROR(ENOENT)); |
0500e835 BB |
751 | |
752 | cookie = spl_fstrans_mark(); | |
eca7b760 | 753 | snapname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); |
0500e835 BB |
754 | |
755 | while (error == 0) { | |
756 | dsl_pool_config_enter(dmu_objset_pool(os), FTAG); | |
0037b49e | 757 | error = dmu_snapshot_list_next(zfsvfs->z_os, |
eca7b760 IK |
758 | ZFS_MAX_DATASET_NAME_LEN, snapname, &id, &pos, |
759 | &case_conflict); | |
0500e835 BB |
760 | dsl_pool_config_exit(dmu_objset_pool(os), FTAG); |
761 | if (error) | |
762 | goto out; | |
763 | ||
764 | if (id == objsetid) | |
765 | break; | |
766 | } | |
767 | ||
768 | memset(full_path, 0, path_len); | |
769 | snprintf(full_path, path_len - 1, "%s/.zfs/snapshot/%s", | |
1c2555ef | 770 | zfsvfs->z_vfs->vfs_mntpoint, snapname); |
0500e835 | 771 | out: |
eca7b760 | 772 | kmem_free(snapname, ZFS_MAX_DATASET_NAME_LEN); |
0500e835 BB |
773 | spl_fstrans_unmark(cookie); |
774 | ||
775 | return (error); | |
776 | } | |
777 | ||
ebe7e575 BB |
778 | /* |
779 | * Special case the handling of "..". | |
780 | */ | |
ebe7e575 BB |
781 | int |
782 | zfsctl_root_lookup(struct inode *dip, char *name, struct inode **ipp, | |
783 | int flags, cred_t *cr, int *direntflags, pathname_t *realpnp) | |
784 | { | |
0037b49e | 785 | zfsvfs_t *zfsvfs = ITOZSB(dip); |
ebe7e575 BB |
786 | int error = 0; |
787 | ||
0037b49e | 788 | ZFS_ENTER(zfsvfs); |
ebe7e575 BB |
789 | |
790 | if (strcmp(name, "..") == 0) { | |
791 | *ipp = dip->i_sb->s_root->d_inode; | |
792 | } else if (strcmp(name, ZFS_SNAPDIR_NAME) == 0) { | |
0037b49e | 793 | *ipp = zfsctl_inode_lookup(zfsvfs, ZFSCTL_INO_SNAPDIR, |
ebe7e575 BB |
794 | &zpl_fops_snapdir, &zpl_ops_snapdir); |
795 | } else if (strcmp(name, ZFS_SHAREDIR_NAME) == 0) { | |
0037b49e | 796 | *ipp = zfsctl_inode_lookup(zfsvfs, ZFSCTL_INO_SHARES, |
ebe7e575 BB |
797 | &zpl_fops_shares, &zpl_ops_shares); |
798 | } else { | |
799 | *ipp = NULL; | |
800 | } | |
801 | ||
802 | if (*ipp == NULL) | |
2e528b49 | 803 | error = SET_ERROR(ENOENT); |
ebe7e575 | 804 | |
0037b49e | 805 | ZFS_EXIT(zfsvfs); |
ebe7e575 BB |
806 | |
807 | return (error); | |
808 | } | |
809 | ||
810 | /* | |
811 | * Lookup entry point for the 'snapshot' directory. Try to open the | |
812 | * snapshot if it exist, creating the pseudo filesystem inode as necessary. | |
813 | * Perform a mount of the associated dataset on top of the inode. | |
814 | */ | |
ebe7e575 BB |
815 | int |
816 | zfsctl_snapdir_lookup(struct inode *dip, char *name, struct inode **ipp, | |
817 | int flags, cred_t *cr, int *direntflags, pathname_t *realpnp) | |
818 | { | |
0037b49e | 819 | zfsvfs_t *zfsvfs = ITOZSB(dip); |
ebe7e575 BB |
820 | uint64_t id; |
821 | int error; | |
822 | ||
0037b49e | 823 | ZFS_ENTER(zfsvfs); |
ebe7e575 | 824 | |
0037b49e | 825 | error = dmu_snapshot_lookup(zfsvfs->z_os, name, &id); |
ebe7e575 | 826 | if (error) { |
0037b49e | 827 | ZFS_EXIT(zfsvfs); |
ebe7e575 BB |
828 | return (error); |
829 | } | |
830 | ||
0037b49e | 831 | *ipp = zfsctl_inode_lookup(zfsvfs, ZFSCTL_INO_SNAPDIRS - id, |
ebe7e575 | 832 | &simple_dir_operations, &simple_dir_inode_operations); |
278bee93 | 833 | if (*ipp == NULL) |
2e528b49 | 834 | error = SET_ERROR(ENOENT); |
ebe7e575 | 835 | |
0037b49e | 836 | ZFS_EXIT(zfsvfs); |
ebe7e575 BB |
837 | |
838 | return (error); | |
839 | } | |
840 | ||
ebe7e575 BB |
841 | /* |
842 | * Renaming a directory under '.zfs/snapshot' will automatically trigger | |
843 | * a rename of the snapshot to the new given name. The rename is confined | |
844 | * to the '.zfs/snapshot' directory snapshots cannot be moved elsewhere. | |
845 | */ | |
ebe7e575 | 846 | int |
13fe0198 MA |
847 | zfsctl_snapdir_rename(struct inode *sdip, char *snm, |
848 | struct inode *tdip, char *tnm, cred_t *cr, int flags) | |
ebe7e575 | 849 | { |
0037b49e | 850 | zfsvfs_t *zfsvfs = ITOZSB(sdip); |
13fe0198 | 851 | char *to, *from, *real, *fsname; |
ebe7e575 BB |
852 | int error; |
853 | ||
0500e835 | 854 | if (!zfs_admin_snapshot) |
ecb2b7dc | 855 | return (SET_ERROR(EACCES)); |
0500e835 | 856 | |
0037b49e | 857 | ZFS_ENTER(zfsvfs); |
ebe7e575 | 858 | |
eca7b760 IK |
859 | to = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); |
860 | from = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); | |
861 | real = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); | |
862 | fsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); | |
ebe7e575 | 863 | |
0037b49e BB |
864 | if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { |
865 | error = dmu_snapshot_realname(zfsvfs->z_os, snm, real, | |
eca7b760 | 866 | ZFS_MAX_DATASET_NAME_LEN, NULL); |
ebe7e575 | 867 | if (error == 0) { |
13fe0198 | 868 | snm = real; |
ebe7e575 BB |
869 | } else if (error != ENOTSUP) { |
870 | goto out; | |
871 | } | |
872 | } | |
873 | ||
0037b49e | 874 | dmu_objset_name(zfsvfs->z_os, fsname); |
13fe0198 | 875 | |
eca7b760 IK |
876 | error = zfsctl_snapshot_name(ITOZSB(sdip), snm, |
877 | ZFS_MAX_DATASET_NAME_LEN, from); | |
13fe0198 | 878 | if (error == 0) |
eca7b760 | 879 | error = zfsctl_snapshot_name(ITOZSB(tdip), tnm, |
02730c33 | 880 | ZFS_MAX_DATASET_NAME_LEN, to); |
13fe0198 | 881 | if (error == 0) |
ebe7e575 | 882 | error = zfs_secpolicy_rename_perms(from, to, cr); |
13fe0198 | 883 | if (error != 0) |
ebe7e575 BB |
884 | goto out; |
885 | ||
886 | /* | |
887 | * Cannot move snapshots out of the snapdir. | |
888 | */ | |
889 | if (sdip != tdip) { | |
2e528b49 | 890 | error = SET_ERROR(EINVAL); |
ebe7e575 BB |
891 | goto out; |
892 | } | |
893 | ||
894 | /* | |
895 | * No-op when names are identical. | |
896 | */ | |
13fe0198 | 897 | if (strcmp(snm, tnm) == 0) { |
ebe7e575 BB |
898 | error = 0; |
899 | goto out; | |
900 | } | |
901 | ||
5ed27c57 | 902 | rw_enter(&zfs_snapshot_lock, RW_WRITER); |
ebe7e575 | 903 | |
13fe0198 | 904 | error = dsl_dataset_rename_snapshot(fsname, snm, tnm, B_FALSE); |
278bee93 BB |
905 | if (error == 0) |
906 | (void) zfsctl_snapshot_rename(snm, tnm); | |
ebe7e575 | 907 | |
5ed27c57 | 908 | rw_exit(&zfs_snapshot_lock); |
ebe7e575 | 909 | out: |
eca7b760 IK |
910 | kmem_free(from, ZFS_MAX_DATASET_NAME_LEN); |
911 | kmem_free(to, ZFS_MAX_DATASET_NAME_LEN); | |
912 | kmem_free(real, ZFS_MAX_DATASET_NAME_LEN); | |
913 | kmem_free(fsname, ZFS_MAX_DATASET_NAME_LEN); | |
ebe7e575 | 914 | |
0037b49e | 915 | ZFS_EXIT(zfsvfs); |
ebe7e575 BB |
916 | |
917 | return (error); | |
918 | } | |
919 | ||
920 | /* | |
921 | * Removing a directory under '.zfs/snapshot' will automatically trigger | |
922 | * the removal of the snapshot with the given name. | |
923 | */ | |
ebe7e575 BB |
924 | int |
925 | zfsctl_snapdir_remove(struct inode *dip, char *name, cred_t *cr, int flags) | |
926 | { | |
0037b49e | 927 | zfsvfs_t *zfsvfs = ITOZSB(dip); |
ebe7e575 BB |
928 | char *snapname, *real; |
929 | int error; | |
930 | ||
0500e835 | 931 | if (!zfs_admin_snapshot) |
ecb2b7dc | 932 | return (SET_ERROR(EACCES)); |
0500e835 | 933 | |
0037b49e | 934 | ZFS_ENTER(zfsvfs); |
ebe7e575 | 935 | |
eca7b760 IK |
936 | snapname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); |
937 | real = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); | |
ebe7e575 | 938 | |
0037b49e BB |
939 | if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { |
940 | error = dmu_snapshot_realname(zfsvfs->z_os, name, real, | |
eca7b760 | 941 | ZFS_MAX_DATASET_NAME_LEN, NULL); |
ebe7e575 BB |
942 | if (error == 0) { |
943 | name = real; | |
944 | } else if (error != ENOTSUP) { | |
945 | goto out; | |
946 | } | |
947 | } | |
948 | ||
eca7b760 IK |
949 | error = zfsctl_snapshot_name(ITOZSB(dip), name, |
950 | ZFS_MAX_DATASET_NAME_LEN, snapname); | |
13fe0198 | 951 | if (error == 0) |
ebe7e575 | 952 | error = zfs_secpolicy_destroy_perms(snapname, cr); |
13fe0198 | 953 | if (error != 0) |
ebe7e575 BB |
954 | goto out; |
955 | ||
278bee93 | 956 | error = zfsctl_snapshot_unmount(snapname, MNT_FORCE); |
ebe7e575 | 957 | if ((error == 0) || (error == ENOENT)) |
13fe0198 | 958 | error = dsl_destroy_snapshot(snapname, B_FALSE); |
ebe7e575 | 959 | out: |
eca7b760 IK |
960 | kmem_free(snapname, ZFS_MAX_DATASET_NAME_LEN); |
961 | kmem_free(real, ZFS_MAX_DATASET_NAME_LEN); | |
ebe7e575 | 962 | |
0037b49e | 963 | ZFS_EXIT(zfsvfs); |
ebe7e575 BB |
964 | |
965 | return (error); | |
966 | } | |
967 | ||
968 | /* | |
969 | * Creating a directory under '.zfs/snapshot' will automatically trigger | |
970 | * the creation of a new snapshot with the given name. | |
971 | */ | |
ebe7e575 BB |
972 | int |
973 | zfsctl_snapdir_mkdir(struct inode *dip, char *dirname, vattr_t *vap, | |
4ea3f864 | 974 | struct inode **ipp, cred_t *cr, int flags) |
ebe7e575 | 975 | { |
0037b49e | 976 | zfsvfs_t *zfsvfs = ITOZSB(dip); |
ebe7e575 BB |
977 | char *dsname; |
978 | int error; | |
979 | ||
0500e835 | 980 | if (!zfs_admin_snapshot) |
ecb2b7dc | 981 | return (SET_ERROR(EACCES)); |
0500e835 | 982 | |
eca7b760 | 983 | dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); |
ebe7e575 | 984 | |
da536844 | 985 | if (zfs_component_namecheck(dirname, NULL, NULL) != 0) { |
2e528b49 | 986 | error = SET_ERROR(EILSEQ); |
ebe7e575 BB |
987 | goto out; |
988 | } | |
989 | ||
0037b49e | 990 | dmu_objset_name(zfsvfs->z_os, dsname); |
ebe7e575 BB |
991 | |
992 | error = zfs_secpolicy_snapshot_perms(dsname, cr); | |
13fe0198 | 993 | if (error != 0) |
ebe7e575 BB |
994 | goto out; |
995 | ||
996 | if (error == 0) { | |
6f1ffb06 | 997 | error = dmu_objset_snapshot_one(dsname, dirname); |
13fe0198 | 998 | if (error != 0) |
ebe7e575 BB |
999 | goto out; |
1000 | ||
1001 | error = zfsctl_snapdir_lookup(dip, dirname, ipp, | |
1002 | 0, cr, NULL, NULL); | |
1003 | } | |
1004 | out: | |
eca7b760 | 1005 | kmem_free(dsname, ZFS_MAX_DATASET_NAME_LEN); |
ebe7e575 BB |
1006 | |
1007 | return (error); | |
1008 | } | |
1009 | ||
ebe7e575 BB |
1010 | /* |
1011 | * Attempt to unmount a snapshot by making a call to user space. | |
1012 | * There is no assurance that this can or will succeed, is just a | |
1013 | * best effort. In the case where it does fail, perhaps because | |
1014 | * it's in use, the unmount will fail harmlessly. | |
1015 | */ | |
278bee93 BB |
1016 | int |
1017 | zfsctl_snapshot_unmount(char *snapname, int flags) | |
ebe7e575 | 1018 | { |
5dc1ff29 SE |
1019 | char *argv[] = { "/usr/bin/env", "umount", "-t", "zfs", "-n", NULL, |
1020 | NULL }; | |
ebe7e575 | 1021 | char *envp[] = { NULL }; |
278bee93 | 1022 | zfs_snapentry_t *se; |
ebe7e575 BB |
1023 | int error; |
1024 | ||
5ed27c57 | 1025 | rw_enter(&zfs_snapshot_lock, RW_READER); |
278bee93 | 1026 | if ((se = zfsctl_snapshot_find_by_name(snapname)) == NULL) { |
5ed27c57 | 1027 | rw_exit(&zfs_snapshot_lock); |
ecb2b7dc | 1028 | return (SET_ERROR(ENOENT)); |
278bee93 | 1029 | } |
5ed27c57 | 1030 | rw_exit(&zfs_snapshot_lock); |
278bee93 | 1031 | |
5dc1ff29 SE |
1032 | if (flags & MNT_FORCE) |
1033 | argv[4] = "-fn"; | |
1034 | argv[5] = se->se_path; | |
278bee93 | 1035 | dprintf("unmount; path=%s\n", se->se_path); |
761394b3 | 1036 | error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC); |
00b65db7 | 1037 | zfsctl_snapshot_rele(se); |
ebe7e575 | 1038 | |
278bee93 | 1039 | |
ebe7e575 BB |
1040 | /* |
1041 | * The umount system utility will return 256 on error. We must | |
1042 | * assume this error is because the file system is busy so it is | |
1043 | * converted to the more sensible EBUSY. | |
1044 | */ | |
1045 | if (error) | |
2e528b49 | 1046 | error = SET_ERROR(EBUSY); |
ebe7e575 | 1047 | |
ebe7e575 BB |
1048 | return (error); |
1049 | } | |
1050 | ||
fd4f7616 | 1051 | #define MOUNT_BUSY 0x80 /* Mount failed due to EBUSY (from mntent.h) */ |
ebe7e575 BB |
1052 | |
1053 | int | |
278bee93 | 1054 | zfsctl_snapshot_mount(struct path *path, int flags) |
ebe7e575 BB |
1055 | { |
1056 | struct dentry *dentry = path->dentry; | |
1057 | struct inode *ip = dentry->d_inode; | |
0037b49e BB |
1058 | zfsvfs_t *zfsvfs; |
1059 | zfsvfs_t *snap_zfsvfs; | |
278bee93 | 1060 | zfs_snapentry_t *se; |
ebe7e575 | 1061 | char *full_name, *full_path; |
5dc1ff29 SE |
1062 | char *argv[] = { "/usr/bin/env", "mount", "-t", "zfs", "-n", NULL, NULL, |
1063 | NULL }; | |
ebe7e575 BB |
1064 | char *envp[] = { NULL }; |
1065 | int error; | |
d287880a | 1066 | struct path spath; |
ebe7e575 | 1067 | |
278bee93 | 1068 | if (ip == NULL) |
ecb2b7dc | 1069 | return (SET_ERROR(EISDIR)); |
278bee93 | 1070 | |
0037b49e BB |
1071 | zfsvfs = ITOZSB(ip); |
1072 | ZFS_ENTER(zfsvfs); | |
ebe7e575 | 1073 | |
eca7b760 | 1074 | full_name = kmem_zalloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); |
278bee93 | 1075 | full_path = kmem_zalloc(MAXPATHLEN, KM_SLEEP); |
ebe7e575 | 1076 | |
0037b49e | 1077 | error = zfsctl_snapshot_name(zfsvfs, dname(dentry), |
eca7b760 | 1078 | ZFS_MAX_DATASET_NAME_LEN, full_name); |
ebe7e575 BB |
1079 | if (error) |
1080 | goto error; | |
1081 | ||
278bee93 | 1082 | error = zfsctl_snapshot_path(path, MAXPATHLEN, full_path); |
ebe7e575 BB |
1083 | if (error) |
1084 | goto error; | |
1085 | ||
278bee93 BB |
1086 | /* |
1087 | * Multiple concurrent automounts of a snapshot are never allowed. | |
1088 | * The snapshot may be manually mounted as many times as desired. | |
1089 | */ | |
1090 | if (zfsctl_snapshot_ismounted(full_name)) { | |
19976601 | 1091 | error = 0; |
278bee93 BB |
1092 | goto error; |
1093 | } | |
1094 | ||
ebe7e575 BB |
1095 | /* |
1096 | * Attempt to mount the snapshot from user space. Normally this | |
1097 | * would be done using the vfs_kern_mount() function, however that | |
1098 | * function is marked GPL-only and cannot be used. On error we | |
1099 | * careful to log the real error to the console and return EISDIR | |
1100 | * to safely abort the automount. This should be very rare. | |
fd4f7616 TC |
1101 | * |
1102 | * If the user mode helper happens to return EBUSY, a concurrent | |
1103 | * mount is already in progress in which case the error is ignored. | |
1104 | * Take note that if the program was executed successfully the return | |
1105 | * value from call_usermodehelper() will be (exitcode << 8 + signal). | |
ebe7e575 | 1106 | */ |
278bee93 | 1107 | dprintf("mount; name=%s path=%s\n", full_name, full_path); |
5dc1ff29 SE |
1108 | argv[5] = full_name; |
1109 | argv[6] = full_path; | |
761394b3 | 1110 | error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC); |
d287880a CC |
1111 | if (error) { |
1112 | if (!(error & MOUNT_BUSY << 8)) { | |
1113 | cmn_err(CE_WARN, "Unable to automount %s/%s: %d", | |
1114 | full_path, full_name, error); | |
1115 | error = SET_ERROR(EISDIR); | |
1116 | } else { | |
1117 | /* | |
1118 | * EBUSY, this could mean a concurrent mount, or the | |
1119 | * snapshot has already been mounted at completely | |
1120 | * different place. We return 0 so VFS will retry. For | |
1121 | * the latter case the VFS will retry several times | |
1122 | * and return ELOOP, which is probably not a very good | |
1123 | * behavior. | |
1124 | */ | |
1125 | error = 0; | |
1126 | } | |
ebe7e575 BB |
1127 | goto error; |
1128 | } | |
1129 | ||
ebe7e575 | 1130 | /* |
278bee93 BB |
1131 | * Follow down in to the mounted snapshot and set MNT_SHRINKABLE |
1132 | * to identify this as an automounted filesystem. | |
ebe7e575 | 1133 | */ |
d287880a CC |
1134 | spath = *path; |
1135 | path_get(&spath); | |
1136 | if (zpl_follow_down_one(&spath)) { | |
0037b49e BB |
1137 | snap_zfsvfs = ITOZSB(spath.dentry->d_inode); |
1138 | snap_zfsvfs->z_parent = zfsvfs; | |
d287880a CC |
1139 | dentry = spath.dentry; |
1140 | spath.mnt->mnt_flags |= MNT_SHRINKABLE; | |
ebe7e575 | 1141 | |
5ed27c57 | 1142 | rw_enter(&zfs_snapshot_lock, RW_WRITER); |
d287880a | 1143 | se = zfsctl_snapshot_alloc(full_name, full_path, |
0037b49e | 1144 | snap_zfsvfs->z_os->os_spa, dmu_objset_id(snap_zfsvfs->z_os), |
24ef51f6 | 1145 | dentry); |
d287880a CC |
1146 | zfsctl_snapshot_add(se); |
1147 | zfsctl_snapshot_unmount_delay_impl(se, zfs_expire_snapshot); | |
5ed27c57 | 1148 | rw_exit(&zfs_snapshot_lock); |
d287880a CC |
1149 | } |
1150 | path_put(&spath); | |
ebe7e575 | 1151 | error: |
eca7b760 | 1152 | kmem_free(full_name, ZFS_MAX_DATASET_NAME_LEN); |
278bee93 | 1153 | kmem_free(full_path, MAXPATHLEN); |
ebe7e575 | 1154 | |
0037b49e | 1155 | ZFS_EXIT(zfsvfs); |
ebe7e575 BB |
1156 | |
1157 | return (error); | |
1158 | } | |
1159 | ||
1160 | /* | |
9b77d1c9 | 1161 | * Get the snapdir inode from fid |
ebe7e575 | 1162 | */ |
ebe7e575 | 1163 | int |
9b77d1c9 CC |
1164 | zfsctl_snapdir_vget(struct super_block *sb, uint64_t objsetid, int gen, |
1165 | struct inode **ipp) | |
ebe7e575 | 1166 | { |
ebe7e575 | 1167 | int error; |
9b77d1c9 CC |
1168 | struct path path; |
1169 | char *mnt; | |
1170 | struct dentry *dentry; | |
d4787d55 | 1171 | |
9b77d1c9 | 1172 | mnt = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
d4787d55 | 1173 | |
9b77d1c9 CC |
1174 | error = zfsctl_snapshot_path_objset(sb->s_fs_info, objsetid, |
1175 | MAXPATHLEN, mnt); | |
1176 | if (error) | |
1177 | goto out; | |
d4787d55 | 1178 | |
9b77d1c9 | 1179 | /* Trigger automount */ |
cfa37548 | 1180 | error = -kern_path(mnt, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &path); |
9b77d1c9 CC |
1181 | if (error) |
1182 | goto out; | |
ebe7e575 | 1183 | |
9b77d1c9 | 1184 | path_put(&path); |
0500e835 | 1185 | /* |
9b77d1c9 CC |
1186 | * Get the snapdir inode. Note, we don't want to use the above |
1187 | * path because it contains the root of the snapshot rather | |
1188 | * than the snapdir. | |
0500e835 | 1189 | */ |
9b77d1c9 CC |
1190 | *ipp = ilookup(sb, ZFSCTL_INO_SNAPDIRS - objsetid); |
1191 | if (*ipp == NULL) { | |
1192 | error = SET_ERROR(ENOENT); | |
1193 | goto out; | |
0500e835 BB |
1194 | } |
1195 | ||
9b77d1c9 CC |
1196 | /* check gen, see zfsctl_snapdir_fid */ |
1197 | dentry = d_obtain_alias(igrab(*ipp)); | |
1198 | if (gen != (!IS_ERR(dentry) && d_mountpoint(dentry))) { | |
1199 | iput(*ipp); | |
1200 | *ipp = NULL; | |
1201 | error = SET_ERROR(ENOENT); | |
1202 | } | |
1203 | if (!IS_ERR(dentry)) | |
1204 | dput(dentry); | |
1205 | out: | |
1206 | kmem_free(mnt, MAXPATHLEN); | |
ebe7e575 BB |
1207 | return (error); |
1208 | } | |
1209 | ||
ebe7e575 BB |
1210 | int |
1211 | zfsctl_shares_lookup(struct inode *dip, char *name, struct inode **ipp, | |
1212 | int flags, cred_t *cr, int *direntflags, pathname_t *realpnp) | |
1213 | { | |
0037b49e | 1214 | zfsvfs_t *zfsvfs = ITOZSB(dip); |
ebe7e575 BB |
1215 | struct inode *ip; |
1216 | znode_t *dzp; | |
1217 | int error; | |
1218 | ||
0037b49e | 1219 | ZFS_ENTER(zfsvfs); |
ebe7e575 | 1220 | |
0037b49e BB |
1221 | if (zfsvfs->z_shares_dir == 0) { |
1222 | ZFS_EXIT(zfsvfs); | |
2e528b49 | 1223 | return (SET_ERROR(ENOTSUP)); |
ebe7e575 BB |
1224 | } |
1225 | ||
0037b49e | 1226 | if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { |
8adb798a DM |
1227 | error = zfs_lookup(ZTOI(dzp), name, &ip, 0, cr, NULL, NULL); |
1228 | iput(ZTOI(dzp)); | |
ebe7e575 BB |
1229 | } |
1230 | ||
0037b49e | 1231 | ZFS_EXIT(zfsvfs); |
ebe7e575 BB |
1232 | |
1233 | return (error); | |
1234 | } | |
1235 | ||
ebe7e575 BB |
1236 | /* |
1237 | * Initialize the various pieces we'll need to create and manipulate .zfs | |
1238 | * directories. Currently this is unused but available. | |
1239 | */ | |
1240 | void | |
1241 | zfsctl_init(void) | |
1242 | { | |
278bee93 BB |
1243 | avl_create(&zfs_snapshots_by_name, snapentry_compare_by_name, |
1244 | sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, | |
1245 | se_node_name)); | |
1246 | avl_create(&zfs_snapshots_by_objsetid, snapentry_compare_by_objsetid, | |
1247 | sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, | |
1248 | se_node_objsetid)); | |
5ed27c57 | 1249 | rw_init(&zfs_snapshot_lock, NULL, RW_DEFAULT, NULL); |
ebe7e575 BB |
1250 | } |
1251 | ||
1252 | /* | |
1253 | * Cleanup the various pieces we needed for .zfs directories. In particular | |
1254 | * ensure the expiry timer is canceled safely. | |
1255 | */ | |
1256 | void | |
1257 | zfsctl_fini(void) | |
1258 | { | |
278bee93 BB |
1259 | avl_destroy(&zfs_snapshots_by_name); |
1260 | avl_destroy(&zfs_snapshots_by_objsetid); | |
5ed27c57 | 1261 | rw_destroy(&zfs_snapshot_lock); |
ebe7e575 BB |
1262 | } |
1263 | ||
0500e835 BB |
1264 | module_param(zfs_admin_snapshot, int, 0644); |
1265 | MODULE_PARM_DESC(zfs_admin_snapshot, "Enable mkdir/rmdir/mv in .zfs/snapshot"); | |
1266 | ||
ebe7e575 BB |
1267 | module_param(zfs_expire_snapshot, int, 0644); |
1268 | MODULE_PARM_DESC(zfs_expire_snapshot, "Seconds to expire .zfs/snapshot"); |