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) 2013, 2016 by Delphix. All rights reserved.
25 * Copyright 2017 Nexenta Systems, Inc.
26 */
27
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/time.h>
31 #include <sys/systm.h>
32 #include <sys/sysmacros.h>
33 #include <sys/resource.h>
34 #include <sys/vfs.h>
35 #include <sys/vnode.h>
36 #include <sys/file.h>
37 #include <sys/kmem.h>
38 #include <sys/uio.h>
39 #include <sys/cmn_err.h>
40 #include <sys/errno.h>
41 #include <sys/stat.h>
42 #include <sys/unistd.h>
43 #include <sys/sunddi.h>
44 #include <sys/random.h>
45 #include <sys/policy.h>
46 #include <sys/kcondvar.h>
47 #include <sys/callb.h>
48 #include <sys/smp.h>
49 #include <sys/zfs_dir.h>
50 #include <sys/zfs_acl.h>
51 #include <sys/fs/zfs.h>
52 #include <sys/zap.h>
53 #include <sys/dmu.h>
54 #include <sys/atomic.h>
55 #include <sys/zfs_ctldir.h>
56 #include <sys/zfs_fuid.h>
57 #include <sys/sa.h>
58 #include <sys/zfs_sa.h>
59 #include <sys/dnlc.h>
60 #include <sys/extdirent.h>
61
62 /*
63 * zfs_match_find() is used by zfs_dirent_lookup() to peform zap lookups
64 * of names after deciding which is the appropriate lookup interface.
65 */
66 static int
zfs_match_find(zfsvfs_t * zfsvfs,znode_t * dzp,const char * name,matchtype_t mt,uint64_t * zoid)67 zfs_match_find(zfsvfs_t *zfsvfs, znode_t *dzp, const char *name,
68 matchtype_t mt, uint64_t *zoid)
69 {
70 int error;
71
72 if (zfsvfs->z_norm) {
73
74 /*
75 * In the non-mixed case we only expect there would ever
76 * be one match, but we need to use the normalizing lookup.
77 */
78 error = zap_lookup_norm(zfsvfs->z_os, dzp->z_id, name, 8, 1,
79 zoid, mt, NULL, 0, NULL);
80 } else {
81 error = zap_lookup(zfsvfs->z_os, dzp->z_id, name, 8, 1, zoid);
82 }
83 *zoid = ZFS_DIRENT_OBJ(*zoid);
84
85 return (error);
86 }
87
88 /*
89 * Look up a directory entry under a locked vnode.
90 * dvp being locked gives us a guarantee that there are no concurrent
91 * modification of the directory and, thus, if a node can be found in
92 * the directory, then it must not be unlinked.
93 *
94 * Input arguments:
95 * dzp - znode for directory
96 * name - name of entry to lock
97 * flag - ZNEW: if the entry already exists, fail with EEXIST.
98 * ZEXISTS: if the entry does not exist, fail with ENOENT.
99 * ZXATTR: we want dzp's xattr directory
100 *
101 * Output arguments:
102 * zpp - pointer to the znode for the entry (NULL if there isn't one)
103 *
104 * Return value: 0 on success or errno on failure.
105 *
106 * NOTE: Always checks for, and rejects, '.' and '..'.
107 */
108 int
zfs_dirent_lookup(znode_t * dzp,const char * name,znode_t ** zpp,int flag)109 zfs_dirent_lookup(znode_t *dzp, const char *name, znode_t **zpp, int flag)
110 {
111 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
112 matchtype_t mt = 0;
113 uint64_t zoid;
114 vnode_t *vp = NULL;
115 int error = 0;
116
117 ASSERT_VOP_LOCKED(ZTOV(dzp), __func__);
118
119 *zpp = NULL;
120
121 /*
122 * Verify that we are not trying to lock '.', '..', or '.zfs'
123 */
124 if (name[0] == '.' &&
125 (name[1] == '\0' || (name[1] == '.' && name[2] == '\0')) ||
126 zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0)
127 return (SET_ERROR(EEXIST));
128
129 /*
130 * Case sensitivity and normalization preferences are set when
131 * the file system is created. These are stored in the
132 * zfsvfs->z_case and zfsvfs->z_norm fields. These choices
133 * affect how we perform zap lookups.
134 *
135 * When matching we may need to normalize & change case according to
136 * FS settings.
137 *
138 * Note that a normalized match is necessary for a case insensitive
139 * filesystem when the lookup request is not exact because normalization
140 * can fold case independent of normalizing code point sequences.
141 *
142 * See the table above zfs_dropname().
143 */
144 if (zfsvfs->z_norm != 0) {
145 mt = MT_NORMALIZE;
146
147 /*
148 * Determine if the match needs to honor the case specified in
149 * lookup, and if so keep track of that so that during
150 * normalization we don't fold case.
151 */
152 if (zfsvfs->z_case == ZFS_CASE_MIXED) {
153 mt |= MT_MATCH_CASE;
154 }
155 }
156
157 /*
158 * Only look in or update the DNLC if we are looking for the
159 * name on a file system that does not require normalization
160 * or case folding. We can also look there if we happen to be
161 * on a non-normalizing, mixed sensitivity file system IF we
162 * are looking for the exact name.
163 *
164 * NB: we do not need to worry about this flag for ZFS_CASE_SENSITIVE
165 * because in that case MT_EXACT and MT_FIRST should produce exactly
166 * the same result.
167 */
168
169 if (dzp->z_unlinked && !(flag & ZXATTR))
170 return (ENOENT);
171 if (flag & ZXATTR) {
172 error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid,
173 sizeof (zoid));
174 if (error == 0)
175 error = (zoid == 0 ? ENOENT : 0);
176 } else {
177 error = zfs_match_find(zfsvfs, dzp, name, mt, &zoid);
178 }
179 if (error) {
180 if (error != ENOENT || (flag & ZEXISTS)) {
181 return (error);
182 }
183 } else {
184 if (flag & ZNEW) {
185 return (SET_ERROR(EEXIST));
186 }
187 error = zfs_zget(zfsvfs, zoid, zpp);
188 if (error)
189 return (error);
190 ASSERT(!(*zpp)->z_unlinked);
191 }
192
193 return (0);
194 }
195
196 static int
zfs_dd_lookup(znode_t * dzp,znode_t ** zpp)197 zfs_dd_lookup(znode_t *dzp, znode_t **zpp)
198 {
199 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
200 znode_t *zp;
201 uint64_t parent;
202 int error;
203
204 ASSERT_VOP_LOCKED(ZTOV(dzp), __func__);
205 ASSERT(RRM_READ_HELD(&zfsvfs->z_teardown_lock));
206
207 if (dzp->z_unlinked)
208 return (ENOENT);
209
210 if ((error = sa_lookup(dzp->z_sa_hdl,
211 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
212 return (error);
213
214 error = zfs_zget(zfsvfs, parent, &zp);
215 if (error == 0)
216 *zpp = zp;
217 return (error);
218 }
219
220 int
zfs_dirlook(znode_t * dzp,const char * name,znode_t ** zpp)221 zfs_dirlook(znode_t *dzp, const char *name, znode_t **zpp)
222 {
223 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
224 znode_t *zp;
225 int error = 0;
226
227 ASSERT_VOP_LOCKED(ZTOV(dzp), __func__);
228 ASSERT(RRM_READ_HELD(&zfsvfs->z_teardown_lock));
229
230 if (dzp->z_unlinked)
231 return (SET_ERROR(ENOENT));
232
233 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
234 *zpp = dzp;
235 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
236 error = zfs_dd_lookup(dzp, zpp);
237 } else {
238 error = zfs_dirent_lookup(dzp, name, &zp, ZEXISTS);
239 if (error == 0) {
240 dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
241 *zpp = zp;
242 }
243 }
244 return (error);
245 }
246
247 /*
248 * unlinked Set (formerly known as the "delete queue") Error Handling
249 *
250 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
251 * don't specify the name of the entry that we will be manipulating. We
252 * also fib and say that we won't be adding any new entries to the
253 * unlinked set, even though we might (this is to lower the minimum file
254 * size that can be deleted in a full filesystem). So on the small
255 * chance that the nlink list is using a fat zap (ie. has more than
256 * 2000 entries), we *may* not pre-read a block that's needed.
257 * Therefore it is remotely possible for some of the assertions
258 * regarding the unlinked set below to fail due to i/o error. On a
259 * nondebug system, this will result in the space being leaked.
260 */
261 void
zfs_unlinked_add(znode_t * zp,dmu_tx_t * tx)262 zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx)
263 {
264 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
265
266 ASSERT(zp->z_unlinked);
267 ASSERT(zp->z_links == 0);
268
269 VERIFY3U(0, ==,
270 zap_add_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
271 }
272
273 /*
274 * Clean up any znodes that had no links when we either crashed or
275 * (force) umounted the file system.
276 */
277 void
zfs_unlinked_drain(zfsvfs_t * zfsvfs)278 zfs_unlinked_drain(zfsvfs_t *zfsvfs)
279 {
280 zap_cursor_t zc;
281 zap_attribute_t zap;
282 dmu_object_info_t doi;
283 znode_t *zp;
284 dmu_tx_t *tx;
285 int error;
286
287 /*
288 * Interate over the contents of the unlinked set.
289 */
290 for (zap_cursor_init(&zc, zfsvfs->z_os, zfsvfs->z_unlinkedobj);
291 zap_cursor_retrieve(&zc, &zap) == 0;
292 zap_cursor_advance(&zc)) {
293
294 /*
295 * See what kind of object we have in list
296 */
297
298 error = dmu_object_info(zfsvfs->z_os,
299 zap.za_first_integer, &doi);
300 if (error != 0)
301 continue;
302
303 ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) ||
304 (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS));
305 /*
306 * We need to re-mark these list entries for deletion,
307 * so we pull them back into core and set zp->z_unlinked.
308 */
309 error = zfs_zget(zfsvfs, zap.za_first_integer, &zp);
310
311 /*
312 * We may pick up znodes that are already marked for deletion.
313 * This could happen during the purge of an extended attribute
314 * directory. All we need to do is skip over them, since they
315 * are already in the system marked z_unlinked.
316 */
317 if (error != 0)
318 continue;
319
320 vn_lock(ZTOV(zp), LK_EXCLUSIVE | LK_RETRY);
321 #if defined(__FreeBSD__)
322 /*
323 * Due to changes in zfs_rmnode we need to make sure the
324 * link count is set to zero here.
325 */
326 if (zp->z_links != 0) {
327 tx = dmu_tx_create(zfsvfs->z_os);
328 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
329 error = dmu_tx_assign(tx, TXG_WAIT);
330 if (error != 0) {
331 dmu_tx_abort(tx);
332 vput(ZTOV(zp));
333 continue;
334 }
335 zp->z_links = 0;
336 VERIFY0(sa_update(zp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
337 &zp->z_links, sizeof (zp->z_links), tx));
338 dmu_tx_commit(tx);
339 }
340 #endif
341 zp->z_unlinked = B_TRUE;
342 vput(ZTOV(zp));
343 }
344 zap_cursor_fini(&zc);
345 }
346
347 /*
348 * Delete the entire contents of a directory. Return a count
349 * of the number of entries that could not be deleted. If we encounter
350 * an error, return a count of at least one so that the directory stays
351 * in the unlinked set.
352 *
353 * NOTE: this function assumes that the directory is inactive,
354 * so there is no need to lock its entries before deletion.
355 * Also, it assumes the directory contents is *only* regular
356 * files.
357 */
358 static int
zfs_purgedir(znode_t * dzp)359 zfs_purgedir(znode_t *dzp)
360 {
361 zap_cursor_t zc;
362 zap_attribute_t zap;
363 znode_t *xzp;
364 dmu_tx_t *tx;
365 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
366 int skipped = 0;
367 int error;
368
369 for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id);
370 (error = zap_cursor_retrieve(&zc, &zap)) == 0;
371 zap_cursor_advance(&zc)) {
372 error = zfs_zget(zfsvfs,
373 ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp);
374 if (error) {
375 skipped += 1;
376 continue;
377 }
378
379 vn_lock(ZTOV(xzp), LK_EXCLUSIVE | LK_RETRY);
380 ASSERT((ZTOV(xzp)->v_type == VREG) ||
381 (ZTOV(xzp)->v_type == VLNK));
382
383 tx = dmu_tx_create(zfsvfs->z_os);
384 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
385 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name);
386 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
387 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
388 /* Is this really needed ? */
389 zfs_sa_upgrade_txholds(tx, xzp);
390 dmu_tx_mark_netfree(tx);
391 error = dmu_tx_assign(tx, TXG_WAIT);
392 if (error) {
393 dmu_tx_abort(tx);
394 vput(ZTOV(xzp));
395 skipped += 1;
396 continue;
397 }
398
399 error = zfs_link_destroy(dzp, zap.za_name, xzp, tx, 0, NULL);
400 if (error)
401 skipped += 1;
402 dmu_tx_commit(tx);
403
404 vput(ZTOV(xzp));
405 }
406 zap_cursor_fini(&zc);
407 if (error != ENOENT)
408 skipped += 1;
409 return (skipped);
410 }
411
412 #if defined(__FreeBSD__)
413 extern taskq_t *zfsvfs_taskq;
414 #endif
415
416 void
zfs_rmnode(znode_t * zp)417 zfs_rmnode(znode_t *zp)
418 {
419 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
420 objset_t *os = zfsvfs->z_os;
421 dmu_tx_t *tx;
422 uint64_t acl_obj;
423 uint64_t xattr_obj;
424 int error;
425
426 ASSERT(zp->z_links == 0);
427 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
428
429 /*
430 * If this is an attribute directory, purge its contents.
431 */
432 if (ZTOV(zp) != NULL && ZTOV(zp)->v_type == VDIR &&
433 (zp->z_pflags & ZFS_XATTR)) {
434 if (zfs_purgedir(zp) != 0) {
435 /*
436 * Not enough space to delete some xattrs.
437 * Leave it in the unlinked set.
438 */
439 zfs_znode_dmu_fini(zp);
440 zfs_znode_free(zp);
441 return;
442 }
443 } else {
444 /*
445 * Free up all the data in the file. We don't do this for
446 * XATTR directories because we need truncate and remove to be
447 * in the same tx, like in zfs_znode_delete(). Otherwise, if
448 * we crash here we'll end up with an inconsistent truncated
449 * zap object in the delete queue. Note a truncated file is
450 * harmless since it only contains user data.
451 */
452 error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
453 if (error) {
454 /*
455 * Not enough space or we were interrupted by unmount.
456 * Leave the file in the unlinked set.
457 */
458 zfs_znode_dmu_fini(zp);
459 zfs_znode_free(zp);
460 return;
461 }
462 }
463
464 /*
465 * If the file has extended attributes, we're going to unlink
466 * the xattr dir.
467 */
468 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
469 &xattr_obj, sizeof (xattr_obj));
470 if (error)
471 xattr_obj = 0;
472
473 acl_obj = zfs_external_acl(zp);
474
475 /*
476 * Set up the final transaction.
477 */
478 tx = dmu_tx_create(os);
479 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
480 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
481 if (xattr_obj)
482 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL);
483 if (acl_obj)
484 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
485
486 zfs_sa_upgrade_txholds(tx, zp);
487 error = dmu_tx_assign(tx, TXG_WAIT);
488 if (error) {
489 /*
490 * Not enough space to delete the file. Leave it in the
491 * unlinked set, leaking it until the fs is remounted (at
492 * which point we'll call zfs_unlinked_drain() to process it).
493 */
494 dmu_tx_abort(tx);
495 zfs_znode_dmu_fini(zp);
496 zfs_znode_free(zp);
497 return;
498 }
499
500 #if defined(__FreeBSD__)
501 /*
502 * FreeBSD's implemention of zfs_zget requires a vnode to back it.
503 * This means that we could end up calling into getnewvnode while
504 * calling zfs_rmnode as a result of a prior call to getnewvnode
505 * trying to clear vnodes out of the cache. If this repeats we can
506 * recurse enough that we overflow our stack. To avoid this, we
507 * avoid calling zfs_zget on the xattr znode and instead simply add
508 * it to the unlinked set and schedule a call to zfs_unlinked_drain.
509 */
510 if (xattr_obj) {
511 /* Add extended attribute directory to the unlinked set. */
512 VERIFY3U(0, ==,
513 zap_add_int(os, zfsvfs->z_unlinkedobj, xattr_obj, tx));
514 }
515 #else
516 if (xzp) {
517 ASSERT(error == 0);
518 xzp->z_unlinked = B_TRUE; /* mark xzp for deletion */
519 xzp->z_links = 0; /* no more links to it */
520 VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
521 &xzp->z_links, sizeof (xzp->z_links), tx));
522 zfs_unlinked_add(xzp, tx);
523 }
524 #endif
525
526 /* Remove this znode from the unlinked set */
527 VERIFY3U(0, ==,
528 zap_remove_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
529
530 zfs_znode_delete(zp, tx);
531
532 dmu_tx_commit(tx);
533
534 #if defined(__FreeBSD__)
535 if (xattr_obj) {
536 /*
537 * We're using the FreeBSD taskqueue API here instead of
538 * the Solaris taskq API since the FreeBSD API allows for a
539 * task to be enqueued multiple times but executed once.
540 */
541 taskqueue_enqueue(zfsvfs_taskq->tq_queue,
542 &zfsvfs->z_unlinked_drain_task);
543 }
544 #endif
545 }
546
547 static uint64_t
zfs_dirent(znode_t * zp,uint64_t mode)548 zfs_dirent(znode_t *zp, uint64_t mode)
549 {
550 uint64_t de = zp->z_id;
551
552 if (zp->z_zfsvfs->z_version >= ZPL_VERSION_DIRENT_TYPE)
553 de |= IFTODT(mode) << 60;
554 return (de);
555 }
556
557 /*
558 * Link zp into dzp. Can only fail if zp has been unlinked.
559 */
560 int
zfs_link_create(znode_t * dzp,const char * name,znode_t * zp,dmu_tx_t * tx,int flag)561 zfs_link_create(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx,
562 int flag)
563 {
564 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
565 vnode_t *vp = ZTOV(zp);
566 uint64_t value;
567 int zp_is_dir = (vp->v_type == VDIR);
568 sa_bulk_attr_t bulk[5];
569 uint64_t mtime[2], ctime[2];
570 int count = 0;
571 int error;
572
573 ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__);
574 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
575 #if 0
576 if (zp_is_dir) {
577 error = 0;
578 if (dzp->z_links >= LINK_MAX)
579 error = SET_ERROR(EMLINK);
580 return (error);
581 }
582 #endif
583 if (!(flag & ZRENAMING)) {
584 if (zp->z_unlinked) { /* no new links to unlinked zp */
585 ASSERT(!(flag & (ZNEW | ZEXISTS)));
586 return (SET_ERROR(ENOENT));
587 }
588 #if 0
589 if (zp->z_links >= LINK_MAX) {
590 return (SET_ERROR(EMLINK));
591 }
592 #endif
593 zp->z_links++;
594 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
595 &zp->z_links, sizeof (zp->z_links));
596
597 } else {
598 ASSERT(zp->z_unlinked == 0);
599 }
600 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
601 &dzp->z_id, sizeof (dzp->z_id));
602 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
603 &zp->z_pflags, sizeof (zp->z_pflags));
604
605 if (!(flag & ZNEW)) {
606 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
607 ctime, sizeof (ctime));
608 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
609 ctime, B_TRUE);
610 }
611 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
612 ASSERT0(error);
613
614 dzp->z_size++;
615 dzp->z_links += zp_is_dir;
616 count = 0;
617 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
618 &dzp->z_size, sizeof (dzp->z_size));
619 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
620 &dzp->z_links, sizeof (dzp->z_links));
621 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
622 mtime, sizeof (mtime));
623 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
624 ctime, sizeof (ctime));
625 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
626 &dzp->z_pflags, sizeof (dzp->z_pflags));
627 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
628 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
629 ASSERT0(error);
630
631 value = zfs_dirent(zp, zp->z_mode);
632 error = zap_add(zp->z_zfsvfs->z_os, dzp->z_id, name,
633 8, 1, &value, tx);
634 VERIFY0(error);
635
636 return (0);
637 }
638
639 /*
640 * The match type in the code for this function should conform to:
641 *
642 * ------------------------------------------------------------------------
643 * fs type | z_norm | lookup type | match type
644 * ---------|-------------|-------------|----------------------------------
645 * CS !norm | 0 | 0 | 0 (exact)
646 * CS norm | formX | 0 | MT_NORMALIZE
647 * CI !norm | upper | !ZCIEXACT | MT_NORMALIZE
648 * CI !norm | upper | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
649 * CI norm | upper|formX | !ZCIEXACT | MT_NORMALIZE
650 * CI norm | upper|formX | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
651 * CM !norm | upper | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
652 * CM !norm | upper | ZCILOOK | MT_NORMALIZE
653 * CM norm | upper|formX | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
654 * CM norm | upper|formX | ZCILOOK | MT_NORMALIZE
655 *
656 * Abbreviations:
657 * CS = Case Sensitive, CI = Case Insensitive, CM = Case Mixed
658 * upper = case folding set by fs type on creation (U8_TEXTPREP_TOUPPER)
659 * formX = unicode normalization form set on fs creation
660 */
661 static int
zfs_dropname(znode_t * dzp,const char * name,znode_t * zp,dmu_tx_t * tx,int flag)662 zfs_dropname(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx,
663 int flag)
664 {
665 int error;
666
667 if (zp->z_zfsvfs->z_norm) {
668 matchtype_t mt = MT_NORMALIZE;
669
670 if (zp->z_zfsvfs->z_case == ZFS_CASE_MIXED) {
671 mt |= MT_MATCH_CASE;
672 }
673
674 error = zap_remove_norm(zp->z_zfsvfs->z_os, dzp->z_id,
675 name, mt, tx);
676 } else {
677 error = zap_remove(zp->z_zfsvfs->z_os, dzp->z_id, name, tx);
678 }
679
680 return (error);
681 }
682
683 /*
684 * Unlink zp from dzp, and mark zp for deletion if this was the last link.
685 * Can fail if zp is a mount point (EBUSY) or a non-empty directory (EEXIST).
686 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
687 * If it's non-NULL, we use it to indicate whether the znode needs deletion,
688 * and it's the caller's job to do it.
689 */
690 int
zfs_link_destroy(znode_t * dzp,const char * name,znode_t * zp,dmu_tx_t * tx,int flag,boolean_t * unlinkedp)691 zfs_link_destroy(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx,
692 int flag, boolean_t *unlinkedp)
693 {
694 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
695 vnode_t *vp = ZTOV(zp);
696 int zp_is_dir = (vp->v_type == VDIR);
697 boolean_t unlinked = B_FALSE;
698 sa_bulk_attr_t bulk[5];
699 uint64_t mtime[2], ctime[2];
700 int count = 0;
701 int error;
702
703 ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__);
704 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
705
706 if (!(flag & ZRENAMING)) {
707
708 if (zp_is_dir && !zfs_dirempty(zp)) {
709 #ifdef illumos
710 return (SET_ERROR(EEXIST));
711 #else
712 return (SET_ERROR(ENOTEMPTY));
713 #endif
714 }
715
716 /*
717 * If we get here, we are going to try to remove the object.
718 * First try removing the name from the directory; if that
719 * fails, return the error.
720 */
721 error = zfs_dropname(dzp, name, zp, tx, flag);
722 if (error != 0) {
723 return (error);
724 }
725
726 if (zp->z_links <= zp_is_dir) {
727 zfs_panic_recover("zfs: link count on vnode %p is %u, "
728 "should be at least %u", zp->z_vnode,
729 (int)zp->z_links,
730 zp_is_dir + 1);
731 zp->z_links = zp_is_dir + 1;
732 }
733 if (--zp->z_links == zp_is_dir) {
734 zp->z_unlinked = B_TRUE;
735 zp->z_links = 0;
736 unlinked = B_TRUE;
737 } else {
738 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
739 NULL, &ctime, sizeof (ctime));
740 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
741 NULL, &zp->z_pflags, sizeof (zp->z_pflags));
742 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
743 B_TRUE);
744 }
745 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
746 NULL, &zp->z_links, sizeof (zp->z_links));
747 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
748 count = 0;
749 ASSERT0(error);
750 } else {
751 ASSERT(zp->z_unlinked == 0);
752 error = zfs_dropname(dzp, name, zp, tx, flag);
753 if (error != 0)
754 return (error);
755 }
756
757 dzp->z_size--; /* one dirent removed */
758 dzp->z_links -= zp_is_dir; /* ".." link from zp */
759 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
760 NULL, &dzp->z_links, sizeof (dzp->z_links));
761 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
762 NULL, &dzp->z_size, sizeof (dzp->z_size));
763 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
764 NULL, ctime, sizeof (ctime));
765 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
766 NULL, mtime, sizeof (mtime));
767 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
768 NULL, &dzp->z_pflags, sizeof (dzp->z_pflags));
769 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
770 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
771 ASSERT0(error);
772
773 if (unlinkedp != NULL)
774 *unlinkedp = unlinked;
775 else if (unlinked)
776 zfs_unlinked_add(zp, tx);
777
778 return (0);
779 }
780
781 /*
782 * Indicate whether the directory is empty.
783 */
784 boolean_t
zfs_dirempty(znode_t * dzp)785 zfs_dirempty(znode_t *dzp)
786 {
787 return (dzp->z_size == 2);
788 }
789
790 int
zfs_make_xattrdir(znode_t * zp,vattr_t * vap,vnode_t ** xvpp,cred_t * cr)791 zfs_make_xattrdir(znode_t *zp, vattr_t *vap, vnode_t **xvpp, cred_t *cr)
792 {
793 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
794 znode_t *xzp;
795 dmu_tx_t *tx;
796 int error;
797 zfs_acl_ids_t acl_ids;
798 boolean_t fuid_dirtied;
799 uint64_t parent;
800
801 *xvpp = NULL;
802
803 /*
804 * In FreeBSD, access checking for creating an EA is being done
805 * in zfs_setextattr(),
806 */
807 #ifndef __FreeBSD_kernel__
808 if (error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr))
809 return (error);
810 #endif
811
812 if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
813 &acl_ids)) != 0)
814 return (error);
815 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
816 zfs_acl_ids_free(&acl_ids);
817 return (SET_ERROR(EDQUOT));
818 }
819
820 getnewvnode_reserve(1);
821
822 tx = dmu_tx_create(zfsvfs->z_os);
823 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
824 ZFS_SA_BASE_ATTR_SIZE);
825 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
826 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
827 fuid_dirtied = zfsvfs->z_fuid_dirty;
828 if (fuid_dirtied)
829 zfs_fuid_txhold(zfsvfs, tx);
830 error = dmu_tx_assign(tx, TXG_WAIT);
831 if (error) {
832 zfs_acl_ids_free(&acl_ids);
833 dmu_tx_abort(tx);
834 return (error);
835 }
836 zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);
837
838 if (fuid_dirtied)
839 zfs_fuid_sync(zfsvfs, tx);
840
841 #ifdef DEBUG
842 error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
843 &parent, sizeof (parent));
844 ASSERT(error == 0 && parent == zp->z_id);
845 #endif
846
847 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id,
848 sizeof (xzp->z_id), tx));
849
850 (void) zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp,
851 xzp, "", NULL, acl_ids.z_fuidp, vap);
852
853 zfs_acl_ids_free(&acl_ids);
854 dmu_tx_commit(tx);
855
856 getnewvnode_drop_reserve();
857
858 *xvpp = ZTOV(xzp);
859
860 return (0);
861 }
862
863 /*
864 * Return a znode for the extended attribute directory for zp.
865 * ** If the directory does not already exist, it is created **
866 *
867 * IN: zp - znode to obtain attribute directory from
868 * cr - credentials of caller
869 * flags - flags from the VOP_LOOKUP call
870 *
871 * OUT: xzpp - pointer to extended attribute znode
872 *
873 * RETURN: 0 on success
874 * error number on failure
875 */
876 int
zfs_get_xattrdir(znode_t * zp,vnode_t ** xvpp,cred_t * cr,int flags)877 zfs_get_xattrdir(znode_t *zp, vnode_t **xvpp, cred_t *cr, int flags)
878 {
879 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
880 znode_t *xzp;
881 vattr_t va;
882 int error;
883 top:
884 error = zfs_dirent_lookup(zp, "", &xzp, ZXATTR);
885 if (error)
886 return (error);
887
888 if (xzp != NULL) {
889 *xvpp = ZTOV(xzp);
890 return (0);
891 }
892
893
894 if (!(flags & CREATE_XATTR_DIR)) {
895 #ifdef illumos
896 return (SET_ERROR(ENOENT));
897 #else
898 return (SET_ERROR(ENOATTR));
899 #endif
900 }
901
902 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
903 return (SET_ERROR(EROFS));
904 }
905
906 /*
907 * The ability to 'create' files in an attribute
908 * directory comes from the write_xattr permission on the base file.
909 *
910 * The ability to 'search' an attribute directory requires
911 * read_xattr permission on the base file.
912 *
913 * Once in a directory the ability to read/write attributes
914 * is controlled by the permissions on the attribute file.
915 */
916 va.va_mask = AT_TYPE | AT_MODE | AT_UID | AT_GID;
917 va.va_type = VDIR;
918 va.va_mode = S_IFDIR | S_ISVTX | 0777;
919 zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
920
921 error = zfs_make_xattrdir(zp, &va, xvpp, cr);
922
923 if (error == ERESTART) {
924 /* NB: we already did dmu_tx_wait() if necessary */
925 goto top;
926 }
927 if (error == 0)
928 VOP_UNLOCK(*xvpp, 0);
929
930 return (error);
931 }
932
933 /*
934 * Decide whether it is okay to remove within a sticky directory.
935 *
936 * In sticky directories, write access is not sufficient;
937 * you can remove entries from a directory only if:
938 *
939 * you own the directory,
940 * you own the entry,
941 * the entry is a plain file and you have write access,
942 * or you are privileged (checked in secpolicy...).
943 *
944 * The function returns 0 if remove access is granted.
945 */
946 int
zfs_sticky_remove_access(znode_t * zdp,znode_t * zp,cred_t * cr)947 zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr)
948 {
949 uid_t uid;
950 uid_t downer;
951 uid_t fowner;
952 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
953
954 if (zdp->z_zfsvfs->z_replay)
955 return (0);
956
957 if ((zdp->z_mode & S_ISVTX) == 0)
958 return (0);
959
960 downer = zfs_fuid_map_id(zfsvfs, zdp->z_uid, cr, ZFS_OWNER);
961 fowner = zfs_fuid_map_id(zfsvfs, zp->z_uid, cr, ZFS_OWNER);
962
963 if ((uid = crgetuid(cr)) == downer || uid == fowner ||
964 (ZTOV(zp)->v_type == VREG &&
965 zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr) == 0))
966 return (0);
967 else
968 return (secpolicy_vnode_remove(ZTOV(zp), cr));
969 }
970