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, 2015 by Delphix. All rights reserved.
24 */
25
26 #include <sys/types.h>
27 #include <sys/param.h>
28 #include <sys/time.h>
29 #include <sys/systm.h>
30 #include <sys/sysmacros.h>
31 #include <sys/resource.h>
32 #include <sys/vfs.h>
33 #include <sys/vnode.h>
34 #include <sys/file.h>
35 #include <sys/kmem.h>
36 #include <sys/uio.h>
37 #include <sys/cmn_err.h>
38 #include <sys/errno.h>
39 #include <sys/stat.h>
40 #include <sys/unistd.h>
41 #include <sys/sunddi.h>
42 #include <sys/random.h>
43 #include <sys/policy.h>
44 #include <sys/kcondvar.h>
45 #include <sys/callb.h>
46 #include <sys/smp.h>
47 #include <sys/zfs_dir.h>
48 #include <sys/zfs_acl.h>
49 #include <sys/fs/zfs.h>
50 #include <sys/zap.h>
51 #include <sys/dmu.h>
52 #include <sys/atomic.h>
53 #include <sys/zfs_ctldir.h>
54 #include <sys/zfs_fuid.h>
55 #include <sys/sa.h>
56 #include <sys/zfs_sa.h>
57 #include <sys/dnlc.h>
58 #include <sys/extdirent.h>
59
60 /*
61 * zfs_match_find() is used by zfs_dirent_lock() to peform zap lookups
62 * of names after deciding which is the appropriate lookup interface.
63 */
64 static int
zfs_match_find(zfsvfs_t * zfsvfs,znode_t * dzp,char * name,boolean_t exact,boolean_t update,int * deflags,pathname_t * rpnp,uint64_t * zoid)65 zfs_match_find(zfsvfs_t *zfsvfs, znode_t *dzp, char *name, boolean_t exact,
66 boolean_t update, int *deflags, pathname_t *rpnp, uint64_t *zoid)
67 {
68 int error;
69
70 if (zfsvfs->z_norm) {
71 matchtype_t mt = MT_FIRST;
72 boolean_t conflict = B_FALSE;
73 size_t bufsz = 0;
74 char *buf = NULL;
75
76 if (rpnp) {
77 buf = rpnp->pn_buf;
78 bufsz = rpnp->pn_bufsize;
79 }
80 if (exact)
81 mt = MT_EXACT;
82 /*
83 * In the non-mixed case we only expect there would ever
84 * be one match, but we need to use the normalizing lookup.
85 */
86 error = zap_lookup_norm(zfsvfs->z_os, dzp->z_id, name, 8, 1,
87 zoid, mt, buf, bufsz, &conflict);
88 if (!error && deflags)
89 *deflags = conflict ? ED_CASE_CONFLICT : 0;
90 } else {
91 error = zap_lookup(zfsvfs->z_os, dzp->z_id, name, 8, 1, zoid);
92 }
93 *zoid = ZFS_DIRENT_OBJ(*zoid);
94
95 if (error == ENOENT && update)
96 dnlc_update(ZTOV(dzp), name, DNLC_NO_VNODE);
97
98 return (error);
99 }
100
101 /*
102 * Lock a directory entry. A dirlock on <dzp, name> protects that name
103 * in dzp's directory zap object. As long as you hold a dirlock, you can
104 * assume two things: (1) dzp cannot be reaped, and (2) no other thread
105 * can change the zap entry for (i.e. link or unlink) this name.
106 *
107 * Input arguments:
108 * dzp - znode for directory
109 * name - name of entry to lock
110 * flag - ZNEW: if the entry already exists, fail with EEXIST.
111 * ZEXISTS: if the entry does not exist, fail with ENOENT.
112 * ZSHARED: allow concurrent access with other ZSHARED callers.
113 * ZXATTR: we want dzp's xattr directory
114 * ZCILOOK: On a mixed sensitivity file system,
115 * this lookup should be case-insensitive.
116 * ZCIEXACT: On a purely case-insensitive file system,
117 * this lookup should be case-sensitive.
118 * ZRENAMING: we are locking for renaming, force narrow locks
119 * ZHAVELOCK: Don't grab the z_name_lock for this call. The
120 * current thread already holds it.
121 *
122 * Output arguments:
123 * zpp - pointer to the znode for the entry (NULL if there isn't one)
124 * dlpp - pointer to the dirlock for this entry (NULL on error)
125 * direntflags - (case-insensitive lookup only)
126 * flags if multiple case-sensitive matches exist in directory
127 * realpnp - (case-insensitive lookup only)
128 * actual name matched within the directory
129 *
130 * Return value: 0 on success or errno on failure.
131 *
132 * NOTE: Always checks for, and rejects, '.' and '..'.
133 * NOTE: For case-insensitive file systems we take wide locks (see below),
134 * but return znode pointers to a single match.
135 */
136 int
zfs_dirent_lock(zfs_dirlock_t ** dlpp,znode_t * dzp,char * name,znode_t ** zpp,int flag,int * direntflags,pathname_t * realpnp)137 zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, znode_t **zpp,
138 int flag, int *direntflags, pathname_t *realpnp)
139 {
140 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
141 zfs_dirlock_t *dl;
142 boolean_t update;
143 boolean_t exact;
144 uint64_t zoid;
145 vnode_t *vp = NULL;
146 int error = 0;
147 int cmpflags;
148
149 *zpp = NULL;
150 *dlpp = NULL;
151
152 /*
153 * Verify that we are not trying to lock '.', '..', or '.zfs'
154 */
155 if (name[0] == '.' &&
156 (name[1] == '\0' || (name[1] == '.' && name[2] == '\0')) ||
157 zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0)
158 return (SET_ERROR(EEXIST));
159
160 /*
161 * Case sensitivity and normalization preferences are set when
162 * the file system is created. These are stored in the
163 * zfsvfs->z_case and zfsvfs->z_norm fields. These choices
164 * affect what vnodes can be cached in the DNLC, how we
165 * perform zap lookups, and the "width" of our dirlocks.
166 *
167 * A normal dirlock locks a single name. Note that with
168 * normalization a name can be composed multiple ways, but
169 * when normalized, these names all compare equal. A wide
170 * dirlock locks multiple names. We need these when the file
171 * system is supporting mixed-mode access. It is sometimes
172 * necessary to lock all case permutations of file name at
173 * once so that simultaneous case-insensitive/case-sensitive
174 * behaves as rationally as possible.
175 */
176
177 /*
178 * Decide if exact matches should be requested when performing
179 * a zap lookup on file systems supporting case-insensitive
180 * access.
181 */
182 exact =
183 ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE) && (flag & ZCIEXACT)) ||
184 ((zfsvfs->z_case == ZFS_CASE_MIXED) && !(flag & ZCILOOK));
185
186 /*
187 * Only look in or update the DNLC if we are looking for the
188 * name on a file system that does not require normalization
189 * or case folding. We can also look there if we happen to be
190 * on a non-normalizing, mixed sensitivity file system IF we
191 * are looking for the exact name.
192 *
193 * Maybe can add TO-UPPERed version of name to dnlc in ci-only
194 * case for performance improvement?
195 */
196 update = !zfsvfs->z_norm ||
197 ((zfsvfs->z_case == ZFS_CASE_MIXED) &&
198 !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER) && !(flag & ZCILOOK));
199
200 /*
201 * ZRENAMING indicates we are in a situation where we should
202 * take narrow locks regardless of the file system's
203 * preferences for normalizing and case folding. This will
204 * prevent us deadlocking trying to grab the same wide lock
205 * twice if the two names happen to be case-insensitive
206 * matches.
207 */
208 if (flag & ZRENAMING)
209 cmpflags = 0;
210 else
211 cmpflags = zfsvfs->z_norm;
212
213 /*
214 * Wait until there are no locks on this name.
215 *
216 * Don't grab the the lock if it is already held. However, cannot
217 * have both ZSHARED and ZHAVELOCK together.
218 */
219 ASSERT(!(flag & ZSHARED) || !(flag & ZHAVELOCK));
220 if (!(flag & ZHAVELOCK))
221 rw_enter(&dzp->z_name_lock, RW_READER);
222
223 mutex_enter(&dzp->z_lock);
224 for (;;) {
225 if (dzp->z_unlinked && !(flag & ZXATTR)) {
226 mutex_exit(&dzp->z_lock);
227 if (!(flag & ZHAVELOCK))
228 rw_exit(&dzp->z_name_lock);
229 return (SET_ERROR(ENOENT));
230 }
231 for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next) {
232 if ((u8_strcmp(name, dl->dl_name, 0, cmpflags,
233 U8_UNICODE_LATEST, &error) == 0) || error != 0)
234 break;
235 }
236 if (error != 0) {
237 mutex_exit(&dzp->z_lock);
238 if (!(flag & ZHAVELOCK))
239 rw_exit(&dzp->z_name_lock);
240 return (SET_ERROR(ENOENT));
241 }
242 if (dl == NULL) {
243 size_t namesize;
244
245 /*
246 * Allocate a new dirlock and add it to the list.
247 */
248 namesize = strlen(name) + 1;
249 dl = kmem_alloc(sizeof (zfs_dirlock_t) + namesize,
250 KM_SLEEP);
251 cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL);
252 dl->dl_name = (char *)(dl + 1);
253 bcopy(name, dl->dl_name, namesize);
254 dl->dl_sharecnt = 0;
255 dl->dl_namelock = 0;
256 dl->dl_namesize = namesize;
257 dl->dl_dzp = dzp;
258 dl->dl_next = dzp->z_dirlocks;
259 dzp->z_dirlocks = dl;
260 break;
261 }
262 if ((flag & ZSHARED) && dl->dl_sharecnt != 0)
263 break;
264 cv_wait(&dl->dl_cv, &dzp->z_lock);
265 }
266
267 /*
268 * If the z_name_lock was NOT held for this dirlock record it.
269 */
270 if (flag & ZHAVELOCK)
271 dl->dl_namelock = 1;
272
273 if (flag & ZSHARED)
274 dl->dl_sharecnt++;
275
276 mutex_exit(&dzp->z_lock);
277
278 /*
279 * We have a dirlock on the name. (Note that it is the dirlock,
280 * not the dzp's z_lock, that protects the name in the zap object.)
281 * See if there's an object by this name; if so, put a hold on it.
282 */
283 if (flag & ZXATTR) {
284 error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid,
285 sizeof (zoid));
286 if (error == 0)
287 error = (zoid == 0 ? ENOENT : 0);
288 } else {
289 if (update)
290 vp = dnlc_lookup(ZTOV(dzp), name);
291 if (vp == DNLC_NO_VNODE) {
292 VN_RELE(vp);
293 error = SET_ERROR(ENOENT);
294 } else if (vp) {
295 if (flag & ZNEW) {
296 zfs_dirent_unlock(dl);
297 VN_RELE(vp);
298 return (SET_ERROR(EEXIST));
299 }
300 *dlpp = dl;
301 *zpp = VTOZ(vp);
302 return (0);
303 } else {
304 error = zfs_match_find(zfsvfs, dzp, name, exact,
305 update, direntflags, realpnp, &zoid);
306 }
307 }
308 if (error) {
309 if (error != ENOENT || (flag & ZEXISTS)) {
310 zfs_dirent_unlock(dl);
311 return (error);
312 }
313 } else {
314 if (flag & ZNEW) {
315 zfs_dirent_unlock(dl);
316 return (SET_ERROR(EEXIST));
317 }
318 error = zfs_zget(zfsvfs, zoid, zpp);
319 if (error) {
320 zfs_dirent_unlock(dl);
321 return (error);
322 }
323 if (!(flag & ZXATTR) && update)
324 dnlc_update(ZTOV(dzp), name, ZTOV(*zpp));
325 }
326
327 *dlpp = dl;
328
329 return (0);
330 }
331
332 /*
333 * Unlock this directory entry and wake anyone who was waiting for it.
334 */
335 void
zfs_dirent_unlock(zfs_dirlock_t * dl)336 zfs_dirent_unlock(zfs_dirlock_t *dl)
337 {
338 znode_t *dzp = dl->dl_dzp;
339 zfs_dirlock_t **prev_dl, *cur_dl;
340
341 mutex_enter(&dzp->z_lock);
342
343 if (!dl->dl_namelock)
344 rw_exit(&dzp->z_name_lock);
345
346 if (dl->dl_sharecnt > 1) {
347 dl->dl_sharecnt--;
348 mutex_exit(&dzp->z_lock);
349 return;
350 }
351 prev_dl = &dzp->z_dirlocks;
352 while ((cur_dl = *prev_dl) != dl)
353 prev_dl = &cur_dl->dl_next;
354 *prev_dl = dl->dl_next;
355 cv_broadcast(&dl->dl_cv);
356 mutex_exit(&dzp->z_lock);
357
358 cv_destroy(&dl->dl_cv);
359 kmem_free(dl, sizeof (*dl) + dl->dl_namesize);
360 }
361
362 /*
363 * Look up an entry in a directory.
364 *
365 * NOTE: '.' and '..' are handled as special cases because
366 * no directory entries are actually stored for them. If this is
367 * the root of a filesystem, then '.zfs' is also treated as a
368 * special pseudo-directory.
369 */
370 int
zfs_dirlook(znode_t * dzp,char * name,vnode_t ** vpp,int flags,int * deflg,pathname_t * rpnp)371 zfs_dirlook(znode_t *dzp, char *name, vnode_t **vpp, int flags,
372 int *deflg, pathname_t *rpnp)
373 {
374 zfs_dirlock_t *dl;
375 znode_t *zp;
376 int error = 0;
377 uint64_t parent;
378 int unlinked;
379
380 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
381 mutex_enter(&dzp->z_lock);
382 unlinked = dzp->z_unlinked;
383 mutex_exit(&dzp->z_lock);
384 if (unlinked)
385 return (ENOENT);
386
387 *vpp = ZTOV(dzp);
388 VN_HOLD(*vpp);
389 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
390 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
391
392 /*
393 * If we are a snapshot mounted under .zfs, return
394 * the vp for the snapshot directory.
395 */
396 if ((error = sa_lookup(dzp->z_sa_hdl,
397 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
398 return (error);
399 if (parent == dzp->z_id && zfsvfs->z_parent != zfsvfs) {
400 error = zfsctl_root_lookup(zfsvfs->z_parent->z_ctldir,
401 "snapshot", vpp, NULL, 0, NULL, kcred,
402 NULL, NULL, NULL);
403 return (error);
404 }
405
406 mutex_enter(&dzp->z_lock);
407 unlinked = dzp->z_unlinked;
408 mutex_exit(&dzp->z_lock);
409 if (unlinked)
410 return (ENOENT);
411
412 rw_enter(&dzp->z_parent_lock, RW_READER);
413 error = zfs_zget(zfsvfs, parent, &zp);
414 if (error == 0)
415 *vpp = ZTOV(zp);
416 rw_exit(&dzp->z_parent_lock);
417 } else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
418 *vpp = zfsctl_root(dzp);
419 } else {
420 int zf;
421
422 zf = ZEXISTS | ZSHARED;
423 if (flags & FIGNORECASE)
424 zf |= ZCILOOK;
425
426 error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp);
427 if (error == 0) {
428 *vpp = ZTOV(zp);
429 zfs_dirent_unlock(dl);
430 dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
431 }
432 rpnp = NULL;
433 }
434
435 if ((flags & FIGNORECASE) && rpnp && !error)
436 (void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize);
437
438 return (error);
439 }
440
441 /*
442 * unlinked Set (formerly known as the "delete queue") Error Handling
443 *
444 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
445 * don't specify the name of the entry that we will be manipulating. We
446 * also fib and say that we won't be adding any new entries to the
447 * unlinked set, even though we might (this is to lower the minimum file
448 * size that can be deleted in a full filesystem). So on the small
449 * chance that the nlink list is using a fat zap (ie. has more than
450 * 2000 entries), we *may* not pre-read a block that's needed.
451 * Therefore it is remotely possible for some of the assertions
452 * regarding the unlinked set below to fail due to i/o error. On a
453 * nondebug system, this will result in the space being leaked.
454 */
455 void
zfs_unlinked_add(znode_t * zp,dmu_tx_t * tx)456 zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx)
457 {
458 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
459
460 ASSERT(zp->z_unlinked);
461 ASSERT(zp->z_links == 0);
462
463 VERIFY3U(0, ==,
464 zap_add_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
465 }
466
467 /*
468 * Clean up any znodes that had no links when we either crashed or
469 * (force) umounted the file system.
470 */
471 void
zfs_unlinked_drain(zfsvfs_t * zfsvfs)472 zfs_unlinked_drain(zfsvfs_t *zfsvfs)
473 {
474 zap_cursor_t zc;
475 zap_attribute_t zap;
476 dmu_object_info_t doi;
477 znode_t *zp;
478 int error;
479
480 /*
481 * Interate over the contents of the unlinked set.
482 */
483 for (zap_cursor_init(&zc, zfsvfs->z_os, zfsvfs->z_unlinkedobj);
484 zap_cursor_retrieve(&zc, &zap) == 0;
485 zap_cursor_advance(&zc)) {
486
487 /*
488 * See what kind of object we have in list
489 */
490
491 error = dmu_object_info(zfsvfs->z_os,
492 zap.za_first_integer, &doi);
493 if (error != 0)
494 continue;
495
496 ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) ||
497 (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS));
498 /*
499 * We need to re-mark these list entries for deletion,
500 * so we pull them back into core and set zp->z_unlinked.
501 */
502 error = zfs_zget(zfsvfs, zap.za_first_integer, &zp);
503
504 /*
505 * We may pick up znodes that are already marked for deletion.
506 * This could happen during the purge of an extended attribute
507 * directory. All we need to do is skip over them, since they
508 * are already in the system marked z_unlinked.
509 */
510 if (error != 0)
511 continue;
512
513 zp->z_unlinked = B_TRUE;
514 VN_RELE(ZTOV(zp));
515 }
516 zap_cursor_fini(&zc);
517 }
518
519 /*
520 * Delete the entire contents of a directory. Return a count
521 * of the number of entries that could not be deleted. If we encounter
522 * an error, return a count of at least one so that the directory stays
523 * in the unlinked set.
524 *
525 * NOTE: this function assumes that the directory is inactive,
526 * so there is no need to lock its entries before deletion.
527 * Also, it assumes the directory contents is *only* regular
528 * files.
529 */
530 static int
zfs_purgedir(znode_t * dzp)531 zfs_purgedir(znode_t *dzp)
532 {
533 zap_cursor_t zc;
534 zap_attribute_t zap;
535 znode_t *xzp;
536 dmu_tx_t *tx;
537 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
538 zfs_dirlock_t dl;
539 int skipped = 0;
540 int error;
541
542 for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id);
543 (error = zap_cursor_retrieve(&zc, &zap)) == 0;
544 zap_cursor_advance(&zc)) {
545 error = zfs_zget(zfsvfs,
546 ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp);
547 if (error) {
548 skipped += 1;
549 continue;
550 }
551
552 ASSERT((ZTOV(xzp)->v_type == VREG) ||
553 (ZTOV(xzp)->v_type == VLNK));
554
555 tx = dmu_tx_create(zfsvfs->z_os);
556 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
557 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name);
558 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
559 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
560 /* Is this really needed ? */
561 zfs_sa_upgrade_txholds(tx, xzp);
562 dmu_tx_mark_netfree(tx);
563 error = dmu_tx_assign(tx, TXG_WAIT);
564 if (error) {
565 dmu_tx_abort(tx);
566 VN_RELE(ZTOV(xzp));
567 skipped += 1;
568 continue;
569 }
570 bzero(&dl, sizeof (dl));
571 dl.dl_dzp = dzp;
572 dl.dl_name = zap.za_name;
573
574 error = zfs_link_destroy(&dl, xzp, tx, 0, NULL);
575 if (error)
576 skipped += 1;
577 dmu_tx_commit(tx);
578
579 VN_RELE(ZTOV(xzp));
580 }
581 zap_cursor_fini(&zc);
582 if (error != ENOENT)
583 skipped += 1;
584 return (skipped);
585 }
586
587 void
zfs_rmnode(znode_t * zp)588 zfs_rmnode(znode_t *zp)
589 {
590 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
591 objset_t *os = zfsvfs->z_os;
592 znode_t *xzp = NULL;
593 dmu_tx_t *tx;
594 uint64_t acl_obj;
595 uint64_t xattr_obj;
596 int error;
597
598 ASSERT(zp->z_links == 0);
599
600 /*
601 * If this is an attribute directory, purge its contents.
602 */
603 if (ZTOV(zp) != NULL && ZTOV(zp)->v_type == VDIR &&
604 (zp->z_pflags & ZFS_XATTR)) {
605 if (zfs_purgedir(zp) != 0) {
606 /*
607 * Not enough space to delete some xattrs.
608 * Leave it in the unlinked set.
609 */
610 zfs_znode_dmu_fini(zp);
611 zfs_znode_free(zp);
612 return;
613 }
614 }
615
616 /*
617 * Free up all the data in the file.
618 */
619 error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
620 if (error) {
621 /*
622 * Not enough space. Leave the file in the unlinked set.
623 */
624 zfs_znode_dmu_fini(zp);
625 zfs_znode_free(zp);
626 return;
627 }
628
629 /*
630 * If the file has extended attributes, we're going to unlink
631 * the xattr dir.
632 */
633 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
634 &xattr_obj, sizeof (xattr_obj));
635 if (error == 0 && xattr_obj) {
636 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
637 ASSERT(error == 0);
638 }
639
640 acl_obj = zfs_external_acl(zp);
641
642 /*
643 * Set up the final transaction.
644 */
645 tx = dmu_tx_create(os);
646 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
647 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
648 if (xzp) {
649 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL);
650 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
651 }
652 if (acl_obj)
653 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
654
655 zfs_sa_upgrade_txholds(tx, zp);
656 error = dmu_tx_assign(tx, TXG_WAIT);
657 if (error) {
658 /*
659 * Not enough space to delete the file. Leave it in the
660 * unlinked set, leaking it until the fs is remounted (at
661 * which point we'll call zfs_unlinked_drain() to process it).
662 */
663 dmu_tx_abort(tx);
664 zfs_znode_dmu_fini(zp);
665 zfs_znode_free(zp);
666 goto out;
667 }
668
669 if (xzp) {
670 ASSERT(error == 0);
671 mutex_enter(&xzp->z_lock);
672 xzp->z_unlinked = B_TRUE; /* mark xzp for deletion */
673 xzp->z_links = 0; /* no more links to it */
674 VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
675 &xzp->z_links, sizeof (xzp->z_links), tx));
676 mutex_exit(&xzp->z_lock);
677 zfs_unlinked_add(xzp, tx);
678 }
679
680 /* Remove this znode from the unlinked set */
681 VERIFY3U(0, ==,
682 zap_remove_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
683
684 zfs_znode_delete(zp, tx);
685
686 dmu_tx_commit(tx);
687 out:
688 if (xzp)
689 VN_RELE(ZTOV(xzp));
690 }
691
692 static uint64_t
zfs_dirent(znode_t * zp,uint64_t mode)693 zfs_dirent(znode_t *zp, uint64_t mode)
694 {
695 uint64_t de = zp->z_id;
696
697 if (zp->z_zfsvfs->z_version >= ZPL_VERSION_DIRENT_TYPE)
698 de |= IFTODT(mode) << 60;
699 return (de);
700 }
701
702 /*
703 * Link zp into dl. Can only fail if zp has been unlinked.
704 */
705 int
zfs_link_create(zfs_dirlock_t * dl,znode_t * zp,dmu_tx_t * tx,int flag)706 zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag)
707 {
708 znode_t *dzp = dl->dl_dzp;
709 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
710 vnode_t *vp = ZTOV(zp);
711 uint64_t value;
712 int zp_is_dir = (vp->v_type == VDIR);
713 sa_bulk_attr_t bulk[5];
714 uint64_t mtime[2], ctime[2];
715 int count = 0;
716 int error;
717
718 mutex_enter(&zp->z_lock);
719
720 if (!(flag & ZRENAMING)) {
721 if (zp->z_unlinked) { /* no new links to unlinked zp */
722 ASSERT(!(flag & (ZNEW | ZEXISTS)));
723 mutex_exit(&zp->z_lock);
724 return (SET_ERROR(ENOENT));
725 }
726 zp->z_links++;
727 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
728 &zp->z_links, sizeof (zp->z_links));
729
730 }
731 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
732 &dzp->z_id, sizeof (dzp->z_id));
733 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
734 &zp->z_pflags, sizeof (zp->z_pflags));
735
736 if (!(flag & ZNEW)) {
737 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
738 ctime, sizeof (ctime));
739 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
740 ctime, B_TRUE);
741 }
742 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
743 ASSERT(error == 0);
744
745 mutex_exit(&zp->z_lock);
746
747 mutex_enter(&dzp->z_lock);
748 dzp->z_size++;
749 dzp->z_links += zp_is_dir;
750 count = 0;
751 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
752 &dzp->z_size, sizeof (dzp->z_size));
753 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
754 &dzp->z_links, sizeof (dzp->z_links));
755 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
756 mtime, sizeof (mtime));
757 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
758 ctime, sizeof (ctime));
759 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
760 &dzp->z_pflags, sizeof (dzp->z_pflags));
761 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
762 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
763 ASSERT(error == 0);
764 mutex_exit(&dzp->z_lock);
765
766 value = zfs_dirent(zp, zp->z_mode);
767 error = zap_add(zp->z_zfsvfs->z_os, dzp->z_id, dl->dl_name,
768 8, 1, &value, tx);
769 ASSERT(error == 0);
770
771 dnlc_update(ZTOV(dzp), dl->dl_name, vp);
772
773 return (0);
774 }
775
776 static int
zfs_dropname(zfs_dirlock_t * dl,znode_t * zp,znode_t * dzp,dmu_tx_t * tx,int flag)777 zfs_dropname(zfs_dirlock_t *dl, znode_t *zp, znode_t *dzp, dmu_tx_t *tx,
778 int flag)
779 {
780 int error;
781
782 if (zp->z_zfsvfs->z_norm) {
783 if (((zp->z_zfsvfs->z_case == ZFS_CASE_INSENSITIVE) &&
784 (flag & ZCIEXACT)) ||
785 ((zp->z_zfsvfs->z_case == ZFS_CASE_MIXED) &&
786 !(flag & ZCILOOK)))
787 error = zap_remove_norm(zp->z_zfsvfs->z_os,
788 dzp->z_id, dl->dl_name, MT_EXACT, tx);
789 else
790 error = zap_remove_norm(zp->z_zfsvfs->z_os,
791 dzp->z_id, dl->dl_name, MT_FIRST, tx);
792 } else {
793 error = zap_remove(zp->z_zfsvfs->z_os,
794 dzp->z_id, dl->dl_name, tx);
795 }
796
797 return (error);
798 }
799
800 /*
801 * Unlink zp from dl, and mark zp for deletion if this was the last link.
802 * Can fail if zp is a mount point (EBUSY) or a non-empty directory (EEXIST).
803 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
804 * If it's non-NULL, we use it to indicate whether the znode needs deletion,
805 * and it's the caller's job to do it.
806 */
807 int
zfs_link_destroy(zfs_dirlock_t * dl,znode_t * zp,dmu_tx_t * tx,int flag,boolean_t * unlinkedp)808 zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag,
809 boolean_t *unlinkedp)
810 {
811 znode_t *dzp = dl->dl_dzp;
812 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
813 vnode_t *vp = ZTOV(zp);
814 int zp_is_dir = (vp->v_type == VDIR);
815 boolean_t unlinked = B_FALSE;
816 sa_bulk_attr_t bulk[5];
817 uint64_t mtime[2], ctime[2];
818 int count = 0;
819 int error;
820
821 dnlc_remove(ZTOV(dzp), dl->dl_name);
822
823 if (!(flag & ZRENAMING)) {
824 if (vn_vfswlock(vp)) /* prevent new mounts on zp */
825 return (SET_ERROR(EBUSY));
826
827 if (vn_ismntpt(vp)) { /* don't remove mount point */
828 vn_vfsunlock(vp);
829 return (SET_ERROR(EBUSY));
830 }
831
832 mutex_enter(&zp->z_lock);
833
834 if (zp_is_dir && !zfs_dirempty(zp)) {
835 mutex_exit(&zp->z_lock);
836 vn_vfsunlock(vp);
837 #ifdef illumos
838 return (SET_ERROR(EEXIST));
839 #else
840 return (SET_ERROR(ENOTEMPTY));
841 #endif
842 }
843
844 /*
845 * If we get here, we are going to try to remove the object.
846 * First try removing the name from the directory; if that
847 * fails, return the error.
848 */
849 error = zfs_dropname(dl, zp, dzp, tx, flag);
850 if (error != 0) {
851 mutex_exit(&zp->z_lock);
852 vn_vfsunlock(vp);
853 return (error);
854 }
855
856 if (zp->z_links <= zp_is_dir) {
857 zfs_panic_recover("zfs: link count on vnode %p is %u, "
858 "should be at least %u", zp->z_vnode,
859 (int)zp->z_links,
860 zp_is_dir + 1);
861 zp->z_links = zp_is_dir + 1;
862 }
863 if (--zp->z_links == zp_is_dir) {
864 zp->z_unlinked = B_TRUE;
865 zp->z_links = 0;
866 unlinked = B_TRUE;
867 } else {
868 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
869 NULL, &ctime, sizeof (ctime));
870 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
871 NULL, &zp->z_pflags, sizeof (zp->z_pflags));
872 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
873 B_TRUE);
874 }
875 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
876 NULL, &zp->z_links, sizeof (zp->z_links));
877 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
878 count = 0;
879 ASSERT(error == 0);
880 mutex_exit(&zp->z_lock);
881 vn_vfsunlock(vp);
882 } else {
883 error = zfs_dropname(dl, zp, dzp, tx, flag);
884 if (error != 0)
885 return (error);
886 }
887
888 mutex_enter(&dzp->z_lock);
889 dzp->z_size--; /* one dirent removed */
890 dzp->z_links -= zp_is_dir; /* ".." link from zp */
891 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
892 NULL, &dzp->z_links, sizeof (dzp->z_links));
893 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
894 NULL, &dzp->z_size, sizeof (dzp->z_size));
895 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
896 NULL, ctime, sizeof (ctime));
897 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
898 NULL, mtime, sizeof (mtime));
899 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
900 NULL, &dzp->z_pflags, sizeof (dzp->z_pflags));
901 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
902 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
903 ASSERT(error == 0);
904 mutex_exit(&dzp->z_lock);
905
906 if (unlinkedp != NULL)
907 *unlinkedp = unlinked;
908 else if (unlinked)
909 zfs_unlinked_add(zp, tx);
910
911 return (0);
912 }
913
914 /*
915 * Indicate whether the directory is empty. Works with or without z_lock
916 * held, but can only be consider a hint in the latter case. Returns true
917 * if only "." and ".." remain and there's no work in progress.
918 */
919 boolean_t
zfs_dirempty(znode_t * dzp)920 zfs_dirempty(znode_t *dzp)
921 {
922 return (dzp->z_size == 2 && dzp->z_dirlocks == 0);
923 }
924
925 int
zfs_make_xattrdir(znode_t * zp,vattr_t * vap,vnode_t ** xvpp,cred_t * cr)926 zfs_make_xattrdir(znode_t *zp, vattr_t *vap, vnode_t **xvpp, cred_t *cr)
927 {
928 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
929 znode_t *xzp;
930 dmu_tx_t *tx;
931 int error;
932 zfs_acl_ids_t acl_ids;
933 boolean_t fuid_dirtied;
934 uint64_t parent;
935
936 *xvpp = NULL;
937
938 /*
939 * In FreeBSD, access checking for creating an EA is being done
940 * in zfs_setextattr(),
941 */
942 #ifndef __FreeBSD_kernel__
943 if (error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr))
944 return (error);
945 #endif
946
947 if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
948 &acl_ids)) != 0)
949 return (error);
950 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
951 zfs_acl_ids_free(&acl_ids);
952 return (SET_ERROR(EDQUOT));
953 }
954
955 getnewvnode_reserve(1);
956
957 tx = dmu_tx_create(zfsvfs->z_os);
958 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
959 ZFS_SA_BASE_ATTR_SIZE);
960 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
961 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
962 fuid_dirtied = zfsvfs->z_fuid_dirty;
963 if (fuid_dirtied)
964 zfs_fuid_txhold(zfsvfs, tx);
965 error = dmu_tx_assign(tx, TXG_WAIT);
966 if (error) {
967 zfs_acl_ids_free(&acl_ids);
968 dmu_tx_abort(tx);
969 return (error);
970 }
971 zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);
972
973 if (fuid_dirtied)
974 zfs_fuid_sync(zfsvfs, tx);
975
976 #ifdef DEBUG
977 error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
978 &parent, sizeof (parent));
979 ASSERT(error == 0 && parent == zp->z_id);
980 #endif
981
982 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id,
983 sizeof (xzp->z_id), tx));
984
985 (void) zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp,
986 xzp, "", NULL, acl_ids.z_fuidp, vap);
987
988 zfs_acl_ids_free(&acl_ids);
989 dmu_tx_commit(tx);
990
991 getnewvnode_drop_reserve();
992
993 *xvpp = ZTOV(xzp);
994
995 return (0);
996 }
997
998 /*
999 * Return a znode for the extended attribute directory for zp.
1000 * ** If the directory does not already exist, it is created **
1001 *
1002 * IN: zp - znode to obtain attribute directory from
1003 * cr - credentials of caller
1004 * flags - flags from the VOP_LOOKUP call
1005 *
1006 * OUT: xzpp - pointer to extended attribute znode
1007 *
1008 * RETURN: 0 on success
1009 * error number on failure
1010 */
1011 int
zfs_get_xattrdir(znode_t * zp,vnode_t ** xvpp,cred_t * cr,int flags)1012 zfs_get_xattrdir(znode_t *zp, vnode_t **xvpp, cred_t *cr, int flags)
1013 {
1014 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1015 znode_t *xzp;
1016 zfs_dirlock_t *dl;
1017 vattr_t va;
1018 int error;
1019 top:
1020 error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL);
1021 if (error)
1022 return (error);
1023
1024 if (xzp != NULL) {
1025 *xvpp = ZTOV(xzp);
1026 zfs_dirent_unlock(dl);
1027 return (0);
1028 }
1029
1030
1031 if (!(flags & CREATE_XATTR_DIR)) {
1032 zfs_dirent_unlock(dl);
1033 #ifdef illumos
1034 return (SET_ERROR(ENOENT));
1035 #else
1036 return (SET_ERROR(ENOATTR));
1037 #endif
1038 }
1039
1040 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
1041 zfs_dirent_unlock(dl);
1042 return (SET_ERROR(EROFS));
1043 }
1044
1045 /*
1046 * The ability to 'create' files in an attribute
1047 * directory comes from the write_xattr permission on the base file.
1048 *
1049 * The ability to 'search' an attribute directory requires
1050 * read_xattr permission on the base file.
1051 *
1052 * Once in a directory the ability to read/write attributes
1053 * is controlled by the permissions on the attribute file.
1054 */
1055 va.va_mask = AT_TYPE | AT_MODE | AT_UID | AT_GID;
1056 va.va_type = VDIR;
1057 va.va_mode = S_IFDIR | S_ISVTX | 0777;
1058 zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
1059
1060 error = zfs_make_xattrdir(zp, &va, xvpp, cr);
1061 zfs_dirent_unlock(dl);
1062
1063 if (error == ERESTART) {
1064 /* NB: we already did dmu_tx_wait() if necessary */
1065 goto top;
1066 }
1067 if (error == 0)
1068 VOP_UNLOCK(*xvpp, 0);
1069
1070 return (error);
1071 }
1072
1073 /*
1074 * Decide whether it is okay to remove within a sticky directory.
1075 *
1076 * In sticky directories, write access is not sufficient;
1077 * you can remove entries from a directory only if:
1078 *
1079 * you own the directory,
1080 * you own the entry,
1081 * the entry is a plain file and you have write access,
1082 * or you are privileged (checked in secpolicy...).
1083 *
1084 * The function returns 0 if remove access is granted.
1085 */
1086 int
zfs_sticky_remove_access(znode_t * zdp,znode_t * zp,cred_t * cr)1087 zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr)
1088 {
1089 uid_t uid;
1090 uid_t downer;
1091 uid_t fowner;
1092 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1093
1094 if (zdp->z_zfsvfs->z_replay)
1095 return (0);
1096
1097 if ((zdp->z_mode & S_ISVTX) == 0)
1098 return (0);
1099
1100 downer = zfs_fuid_map_id(zfsvfs, zdp->z_uid, cr, ZFS_OWNER);
1101 fowner = zfs_fuid_map_id(zfsvfs, zp->z_uid, cr, ZFS_OWNER);
1102
1103 if ((uid = crgetuid(cr)) == downer || uid == fowner ||
1104 (ZTOV(zp)->v_type == VREG &&
1105 zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr) == 0))
1106 return (0);
1107 else
1108 return (secpolicy_vnode_remove(ZTOV(zp), cr));
1109 }
1110