xref: /freebsd/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_dir.c (revision 7a7741af18d6c8a804cc643cb7ecda9d730c6aa6)
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 https://opensource.org/licenses/CDDL-1.0.
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/condvar.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/dmu_objset.h>
60 #include <sys/dsl_dir.h>
61 
62 #include <sys/ccompat.h>
63 
64 /*
65  * zfs_match_find() is used by zfs_dirent_lookup() to perform zap lookups
66  * of names after deciding which is the appropriate lookup interface.
67  */
68 static int
zfs_match_find(zfsvfs_t * zfsvfs,znode_t * dzp,const char * name,matchtype_t mt,uint64_t * zoid)69 zfs_match_find(zfsvfs_t *zfsvfs, znode_t *dzp, const char *name,
70     matchtype_t mt, uint64_t *zoid)
71 {
72 	int error;
73 
74 	if (zfsvfs->z_norm) {
75 
76 		/*
77 		 * In the non-mixed case we only expect there would ever
78 		 * be one match, but we need to use the normalizing lookup.
79 		 */
80 		error = zap_lookup_norm(zfsvfs->z_os, dzp->z_id, name, 8, 1,
81 		    zoid, mt, NULL, 0, NULL);
82 	} else {
83 		error = zap_lookup(zfsvfs->z_os, dzp->z_id, name, 8, 1, zoid);
84 	}
85 	*zoid = ZFS_DIRENT_OBJ(*zoid);
86 
87 	return (error);
88 }
89 
90 /*
91  * Look up a directory entry under a locked vnode.
92  * dvp being locked gives us a guarantee that there are no concurrent
93  * modification of the directory and, thus, if a node can be found in
94  * the directory, then it must not be unlinked.
95  *
96  * Input arguments:
97  *	dzp	- znode for directory
98  *	name	- name of entry to lock
99  *	flag	- ZNEW: if the entry already exists, fail with EEXIST.
100  *		  ZEXISTS: if the entry does not exist, fail with ENOENT.
101  *		  ZXATTR: we want dzp's xattr directory
102  *
103  * Output arguments:
104  *	zpp	- pointer to the znode for the entry (NULL if there isn't one)
105  *
106  * Return value: 0 on success or errno on failure.
107  *
108  * NOTE: Always checks for, and rejects, '.' and '..'.
109  */
110 int
zfs_dirent_lookup(znode_t * dzp,const char * name,znode_t ** zpp,int flag)111 zfs_dirent_lookup(znode_t *dzp, const char *name, znode_t **zpp, int flag)
112 {
113 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
114 	znode_t		*zp;
115 	matchtype_t	mt = 0;
116 	uint64_t	zoid;
117 	int		error = 0;
118 
119 	if (zfsvfs->z_replay == B_FALSE)
120 		ASSERT_VOP_LOCKED(ZTOV(dzp), __func__);
121 
122 	*zpp = NULL;
123 
124 	/*
125 	 * Verify that we are not trying to lock '.', '..', or '.zfs'
126 	 */
127 	if (name[0] == '.' &&
128 	    (((name[1] == '\0') || (name[1] == '.' && name[2] == '\0')) ||
129 	    (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0)))
130 		return (SET_ERROR(EEXIST));
131 
132 	/*
133 	 * Case sensitivity and normalization preferences are set when
134 	 * the file system is created.  These are stored in the
135 	 * zfsvfs->z_case and zfsvfs->z_norm fields.  These choices
136 	 * affect how we perform zap lookups.
137 	 *
138 	 * When matching we may need to normalize & change case according to
139 	 * FS settings.
140 	 *
141 	 * Note that a normalized match is necessary for a case insensitive
142 	 * filesystem when the lookup request is not exact because normalization
143 	 * can fold case independent of normalizing code point sequences.
144 	 *
145 	 * See the table above zfs_dropname().
146 	 */
147 	if (zfsvfs->z_norm != 0) {
148 		mt = MT_NORMALIZE;
149 
150 		/*
151 		 * Determine if the match needs to honor the case specified in
152 		 * lookup, and if so keep track of that so that during
153 		 * normalization we don't fold case.
154 		 */
155 		if (zfsvfs->z_case == ZFS_CASE_MIXED) {
156 			mt |= MT_MATCH_CASE;
157 		}
158 	}
159 
160 	/*
161 	 * Only look in or update the DNLC if we are looking for the
162 	 * name on a file system that does not require normalization
163 	 * or case folding.  We can also look there if we happen to be
164 	 * on a non-normalizing, mixed sensitivity file system IF we
165 	 * are looking for the exact name.
166 	 *
167 	 * NB: we do not need to worry about this flag for ZFS_CASE_SENSITIVE
168 	 * because in that case MT_EXACT and MT_FIRST should produce exactly
169 	 * the same result.
170 	 */
171 
172 	if (dzp->z_unlinked && !(flag & ZXATTR))
173 		return (ENOENT);
174 	if (flag & ZXATTR) {
175 		error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid,
176 		    sizeof (zoid));
177 		if (error == 0)
178 			error = (zoid == 0 ? ENOENT : 0);
179 	} else {
180 		error = zfs_match_find(zfsvfs, dzp, name, mt, &zoid);
181 	}
182 	if (error) {
183 		if (error != ENOENT || (flag & ZEXISTS)) {
184 			return (error);
185 		}
186 	} else {
187 		if (flag & ZNEW) {
188 			return (SET_ERROR(EEXIST));
189 		}
190 		error = zfs_zget(zfsvfs, zoid, &zp);
191 		if (error)
192 			return (error);
193 		ASSERT(!zp->z_unlinked);
194 		*zpp = zp;
195 	}
196 
197 	return (0);
198 }
199 
200 static int
zfs_dd_lookup(znode_t * dzp,znode_t ** zpp)201 zfs_dd_lookup(znode_t *dzp, znode_t **zpp)
202 {
203 	zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
204 	znode_t *zp;
205 	uint64_t parent;
206 	int error;
207 
208 #ifdef ZFS_DEBUG
209 	if (zfsvfs->z_replay == B_FALSE)
210 		ASSERT_VOP_LOCKED(ZTOV(dzp), __func__);
211 #endif
212 	if (dzp->z_unlinked)
213 		return (ENOENT);
214 
215 	if ((error = sa_lookup(dzp->z_sa_hdl,
216 	    SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
217 		return (error);
218 
219 	error = zfs_zget(zfsvfs, parent, &zp);
220 	if (error == 0)
221 		*zpp = zp;
222 	return (error);
223 }
224 
225 int
zfs_dirlook(znode_t * dzp,const char * name,znode_t ** zpp)226 zfs_dirlook(znode_t *dzp, const char *name, znode_t **zpp)
227 {
228 	zfsvfs_t *zfsvfs __unused = dzp->z_zfsvfs;
229 	znode_t *zp = NULL;
230 	int error = 0;
231 
232 #ifdef ZFS_DEBUG
233 	if (zfsvfs->z_replay == B_FALSE)
234 		ASSERT_VOP_LOCKED(ZTOV(dzp), __func__);
235 #endif
236 	if (dzp->z_unlinked)
237 		return (SET_ERROR(ENOENT));
238 
239 	if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
240 		*zpp = dzp;
241 	} else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
242 		error = zfs_dd_lookup(dzp, &zp);
243 		if (error == 0)
244 			*zpp = zp;
245 	} else {
246 		error = zfs_dirent_lookup(dzp, name, &zp, ZEXISTS);
247 		if (error == 0) {
248 			dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
249 			*zpp = zp;
250 		}
251 	}
252 	return (error);
253 }
254 
255 /*
256  * unlinked Set (formerly known as the "delete queue") Error Handling
257  *
258  * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
259  * don't specify the name of the entry that we will be manipulating.  We
260  * also fib and say that we won't be adding any new entries to the
261  * unlinked set, even though we might (this is to lower the minimum file
262  * size that can be deleted in a full filesystem).  So on the small
263  * chance that the nlink list is using a fat zap (ie. has more than
264  * 2000 entries), we *may* not pre-read a block that's needed.
265  * Therefore it is remotely possible for some of the assertions
266  * regarding the unlinked set below to fail due to i/o error.  On a
267  * nondebug system, this will result in the space being leaked.
268  */
269 void
zfs_unlinked_add(znode_t * zp,dmu_tx_t * tx)270 zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx)
271 {
272 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
273 
274 	ASSERT(zp->z_unlinked);
275 	ASSERT3U(zp->z_links, ==, 0);
276 
277 	VERIFY0(zap_add_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
278 
279 	dataset_kstats_update_nunlinks_kstat(&zfsvfs->z_kstat, 1);
280 }
281 
282 /*
283  * Clean up any znodes that had no links when we either crashed or
284  * (force) umounted the file system.
285  */
286 void
zfs_unlinked_drain(zfsvfs_t * zfsvfs)287 zfs_unlinked_drain(zfsvfs_t *zfsvfs)
288 {
289 	zap_cursor_t	zc;
290 	zap_attribute_t *zap;
291 	dmu_object_info_t doi;
292 	znode_t		*zp;
293 	dmu_tx_t	*tx;
294 	int		error;
295 
296 	/*
297 	 * Iterate over the contents of the unlinked set.
298 	 */
299 	zap = zap_attribute_alloc();
300 	for (zap_cursor_init(&zc, zfsvfs->z_os, zfsvfs->z_unlinkedobj);
301 	    zap_cursor_retrieve(&zc, zap) == 0;
302 	    zap_cursor_advance(&zc)) {
303 
304 		/*
305 		 * See what kind of object we have in list
306 		 */
307 
308 		error = dmu_object_info(zfsvfs->z_os,
309 		    zap->za_first_integer, &doi);
310 		if (error != 0)
311 			continue;
312 
313 		ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) ||
314 		    (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS));
315 		/*
316 		 * We need to re-mark these list entries for deletion,
317 		 * so we pull them back into core and set zp->z_unlinked.
318 		 */
319 		error = zfs_zget(zfsvfs, zap->za_first_integer, &zp);
320 
321 		/*
322 		 * We may pick up znodes that are already marked for deletion.
323 		 * This could happen during the purge of an extended attribute
324 		 * directory.  All we need to do is skip over them, since they
325 		 * are already in the system marked z_unlinked.
326 		 */
327 		if (error != 0)
328 			continue;
329 
330 		vn_lock(ZTOV(zp), LK_EXCLUSIVE | LK_RETRY);
331 
332 		/*
333 		 * Due to changes in zfs_rmnode we need to make sure the
334 		 * link count is set to zero here.
335 		 */
336 		if (zp->z_links != 0) {
337 			tx = dmu_tx_create(zfsvfs->z_os);
338 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
339 			error = dmu_tx_assign(tx, TXG_WAIT);
340 			if (error != 0) {
341 				dmu_tx_abort(tx);
342 				vput(ZTOV(zp));
343 				continue;
344 			}
345 			zp->z_links = 0;
346 			VERIFY0(sa_update(zp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
347 			    &zp->z_links, sizeof (zp->z_links), tx));
348 			dmu_tx_commit(tx);
349 		}
350 
351 		zp->z_unlinked = B_TRUE;
352 		vput(ZTOV(zp));
353 	}
354 	zap_cursor_fini(&zc);
355 	zap_attribute_free(zap);
356 }
357 
358 /*
359  * Delete the entire contents of a directory.  Return a count
360  * of the number of entries that could not be deleted. If we encounter
361  * an error, return a count of at least one so that the directory stays
362  * in the unlinked set.
363  *
364  * NOTE: this function assumes that the directory is inactive,
365  *	so there is no need to lock its entries before deletion.
366  *	Also, it assumes the directory contents is *only* regular
367  *	files.
368  */
369 static int
zfs_purgedir(znode_t * dzp)370 zfs_purgedir(znode_t *dzp)
371 {
372 	zap_cursor_t	zc;
373 	zap_attribute_t	*zap;
374 	znode_t		*xzp;
375 	dmu_tx_t	*tx;
376 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
377 	int skipped = 0;
378 	int error;
379 
380 	zap = zap_attribute_alloc();
381 	for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id);
382 	    (error = zap_cursor_retrieve(&zc, zap)) == 0;
383 	    zap_cursor_advance(&zc)) {
384 		error = zfs_zget(zfsvfs,
385 		    ZFS_DIRENT_OBJ(zap->za_first_integer), &xzp);
386 		if (error) {
387 			skipped += 1;
388 			continue;
389 		}
390 
391 		vn_lock(ZTOV(xzp), LK_EXCLUSIVE | LK_RETRY);
392 		ASSERT((ZTOV(xzp)->v_type == VREG) ||
393 		    (ZTOV(xzp)->v_type == VLNK));
394 
395 		tx = dmu_tx_create(zfsvfs->z_os);
396 		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
397 		dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap->za_name);
398 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
399 		dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
400 		/* Is this really needed ? */
401 		zfs_sa_upgrade_txholds(tx, xzp);
402 		dmu_tx_mark_netfree(tx);
403 		error = dmu_tx_assign(tx, TXG_WAIT);
404 		if (error) {
405 			dmu_tx_abort(tx);
406 			vput(ZTOV(xzp));
407 			skipped += 1;
408 			continue;
409 		}
410 
411 		error = zfs_link_destroy(dzp, zap->za_name, xzp, tx, 0, NULL);
412 		if (error)
413 			skipped += 1;
414 		dmu_tx_commit(tx);
415 
416 		vput(ZTOV(xzp));
417 	}
418 	zap_cursor_fini(&zc);
419 	zap_attribute_free(zap);
420 	if (error != ENOENT)
421 		skipped += 1;
422 	return (skipped);
423 }
424 
425 extern taskq_t *zfsvfs_taskq;
426 
427 void
zfs_rmnode(znode_t * zp)428 zfs_rmnode(znode_t *zp)
429 {
430 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
431 	objset_t	*os = zfsvfs->z_os;
432 	dmu_tx_t	*tx;
433 	uint64_t	z_id = zp->z_id;
434 	uint64_t	acl_obj;
435 	uint64_t	xattr_obj;
436 	uint64_t	count;
437 	int		error;
438 
439 	ASSERT3U(zp->z_links, ==, 0);
440 	if (zfsvfs->z_replay == B_FALSE)
441 		ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
442 
443 	/*
444 	 * If this is an attribute directory, purge its contents.
445 	 */
446 	if (ZTOV(zp) != NULL && ZTOV(zp)->v_type == VDIR &&
447 	    (zp->z_pflags & ZFS_XATTR)) {
448 		if (zfs_purgedir(zp) != 0) {
449 			/*
450 			 * Not enough space to delete some xattrs.
451 			 * Leave it in the unlinked set.
452 			 */
453 			ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
454 			zfs_znode_dmu_fini(zp);
455 			zfs_znode_free(zp);
456 			ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
457 			return;
458 		}
459 	} else {
460 		/*
461 		 * Free up all the data in the file.  We don't do this for
462 		 * XATTR directories because we need truncate and remove to be
463 		 * in the same tx, like in zfs_znode_delete(). Otherwise, if
464 		 * we crash here we'll end up with an inconsistent truncated
465 		 * zap object in the delete queue.  Note a truncated file is
466 		 * harmless since it only contains user data.
467 		 */
468 		error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
469 		if (error) {
470 			/*
471 			 * Not enough space or we were interrupted by unmount.
472 			 * Leave the file in the unlinked set.
473 			 */
474 			ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
475 			zfs_znode_dmu_fini(zp);
476 			zfs_znode_free(zp);
477 			ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
478 			return;
479 		}
480 	}
481 
482 	/*
483 	 * If the file has extended attributes, we're going to unlink
484 	 * the xattr dir.
485 	 */
486 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
487 	    &xattr_obj, sizeof (xattr_obj));
488 	if (error)
489 		xattr_obj = 0;
490 
491 	acl_obj = zfs_external_acl(zp);
492 
493 	/*
494 	 * Set up the final transaction.
495 	 */
496 	tx = dmu_tx_create(os);
497 	dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
498 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
499 	if (xattr_obj)
500 		dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL);
501 	if (acl_obj)
502 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
503 
504 	zfs_sa_upgrade_txholds(tx, zp);
505 	error = dmu_tx_assign(tx, TXG_WAIT);
506 	if (error) {
507 		/*
508 		 * Not enough space to delete the file.  Leave it in the
509 		 * unlinked set, leaking it until the fs is remounted (at
510 		 * which point we'll call zfs_unlinked_drain() to process it).
511 		 */
512 		dmu_tx_abort(tx);
513 		ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
514 		zfs_znode_dmu_fini(zp);
515 		zfs_znode_free(zp);
516 		ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
517 		return;
518 	}
519 
520 	/*
521 	 * FreeBSD's implementation of zfs_zget requires a vnode to back it.
522 	 * This means that we could end up calling into getnewvnode while
523 	 * calling zfs_rmnode as a result of a prior call to getnewvnode
524 	 * trying to clear vnodes out of the cache. If this repeats we can
525 	 * recurse enough that we overflow our stack. To avoid this, we
526 	 * avoid calling zfs_zget on the xattr znode and instead simply add
527 	 * it to the unlinked set and schedule a call to zfs_unlinked_drain.
528 	 */
529 	if (xattr_obj) {
530 		/* Add extended attribute directory to the unlinked set. */
531 		VERIFY3U(0, ==,
532 		    zap_add_int(os, zfsvfs->z_unlinkedobj, xattr_obj, tx));
533 	}
534 
535 	mutex_enter(&os->os_dsl_dataset->ds_dir->dd_activity_lock);
536 
537 	/* Remove this znode from the unlinked set */
538 	VERIFY3U(0, ==,
539 	    zap_remove_int(os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
540 
541 	if (zap_count(os, zfsvfs->z_unlinkedobj, &count) == 0 && count == 0) {
542 		cv_broadcast(&os->os_dsl_dataset->ds_dir->dd_activity_cv);
543 	}
544 
545 	mutex_exit(&os->os_dsl_dataset->ds_dir->dd_activity_lock);
546 
547 	dataset_kstats_update_nunlinked_kstat(&zfsvfs->z_kstat, 1);
548 
549 	zfs_znode_delete(zp, tx);
550 	zfs_znode_free(zp);
551 
552 	dmu_tx_commit(tx);
553 
554 	if (xattr_obj) {
555 		/*
556 		 * We're using the FreeBSD taskqueue API here instead of
557 		 * the Solaris taskq API since the FreeBSD API allows for a
558 		 * task to be enqueued multiple times but executed once.
559 		 */
560 		taskqueue_enqueue(zfsvfs_taskq->tq_queue,
561 		    &zfsvfs->z_unlinked_drain_task);
562 	}
563 }
564 
565 static uint64_t
zfs_dirent(znode_t * zp,uint64_t mode)566 zfs_dirent(znode_t *zp, uint64_t mode)
567 {
568 	uint64_t de = zp->z_id;
569 
570 	if (zp->z_zfsvfs->z_version >= ZPL_VERSION_DIRENT_TYPE)
571 		de |= IFTODT(mode) << 60;
572 	return (de);
573 }
574 
575 /*
576  * Link zp into dzp.  Can only fail if zp has been unlinked.
577  */
578 int
zfs_link_create(znode_t * dzp,const char * name,znode_t * zp,dmu_tx_t * tx,int flag)579 zfs_link_create(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx,
580     int flag)
581 {
582 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
583 	vnode_t *vp = ZTOV(zp);
584 	uint64_t value;
585 	int zp_is_dir = (vp->v_type == VDIR);
586 	sa_bulk_attr_t bulk[5];
587 	uint64_t mtime[2], ctime[2];
588 	int count = 0;
589 	int error;
590 
591 	if (zfsvfs->z_replay == B_FALSE) {
592 		ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__);
593 		ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
594 	}
595 	if (zp_is_dir) {
596 		if (dzp->z_links >= ZFS_LINK_MAX)
597 			return (SET_ERROR(EMLINK));
598 	}
599 	if (!(flag & ZRENAMING)) {
600 		if (zp->z_unlinked) {	/* no new links to unlinked zp */
601 			ASSERT(!(flag & (ZNEW | ZEXISTS)));
602 			return (SET_ERROR(ENOENT));
603 		}
604 		if (zp->z_links >= ZFS_LINK_MAX - zp_is_dir) {
605 			return (SET_ERROR(EMLINK));
606 		}
607 		zp->z_links++;
608 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
609 		    &zp->z_links, sizeof (zp->z_links));
610 
611 	} else {
612 		ASSERT(!zp->z_unlinked);
613 	}
614 	value = zfs_dirent(zp, zp->z_mode);
615 	error = zap_add(zp->z_zfsvfs->z_os, dzp->z_id, name,
616 	    8, 1, &value, tx);
617 
618 	/*
619 	 * zap_add could fail to add the entry if it exceeds the capacity of the
620 	 * leaf-block and zap_leaf_split() failed to help.
621 	 * The caller of this routine is responsible for failing the transaction
622 	 * which will rollback the SA updates done above.
623 	 */
624 	if (error != 0) {
625 		if (!(flag & ZRENAMING) && !(flag & ZNEW))
626 			zp->z_links--;
627 		return (error);
628 	}
629 
630 	/*
631 	 * If we added a longname activate the SPA_FEATURE_LONGNAME.
632 	 */
633 	if (strlen(name) >= ZAP_MAXNAMELEN) {
634 		dsl_dataset_t *ds = dmu_objset_ds(zfsvfs->z_os);
635 		ds->ds_feature_activation[SPA_FEATURE_LONGNAME] =
636 		    (void *)B_TRUE;
637 	}
638 
639 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
640 	    &dzp->z_id, sizeof (dzp->z_id));
641 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
642 	    &zp->z_pflags, sizeof (zp->z_pflags));
643 
644 	if (!(flag & ZNEW)) {
645 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
646 		    ctime, sizeof (ctime));
647 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
648 		    ctime);
649 	}
650 	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
651 	ASSERT0(error);
652 
653 	dzp->z_size++;
654 	dzp->z_links += zp_is_dir;
655 	count = 0;
656 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
657 	    &dzp->z_size, sizeof (dzp->z_size));
658 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
659 	    &dzp->z_links, sizeof (dzp->z_links));
660 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
661 	    mtime, sizeof (mtime));
662 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
663 	    ctime, sizeof (ctime));
664 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
665 	    &dzp->z_pflags, sizeof (dzp->z_pflags));
666 	zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime);
667 	error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
668 	ASSERT0(error);
669 	return (0);
670 }
671 
672 /*
673  * The match type in the code for this function should conform to:
674  *
675  * ------------------------------------------------------------------------
676  * fs type  | z_norm      | lookup type | match type
677  * ---------|-------------|-------------|----------------------------------
678  * CS !norm | 0           |           0 | 0 (exact)
679  * CS  norm | formX       |           0 | MT_NORMALIZE
680  * CI !norm | upper       |   !ZCIEXACT | MT_NORMALIZE
681  * CI !norm | upper       |    ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
682  * CI  norm | upper|formX |   !ZCIEXACT | MT_NORMALIZE
683  * CI  norm | upper|formX |    ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
684  * CM !norm | upper       |    !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
685  * CM !norm | upper       |     ZCILOOK | MT_NORMALIZE
686  * CM  norm | upper|formX |    !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
687  * CM  norm | upper|formX |     ZCILOOK | MT_NORMALIZE
688  *
689  * Abbreviations:
690  *    CS = Case Sensitive, CI = Case Insensitive, CM = Case Mixed
691  *    upper = case folding set by fs type on creation (U8_TEXTPREP_TOUPPER)
692  *    formX = unicode normalization form set on fs creation
693  */
694 static int
zfs_dropname(znode_t * dzp,const char * name,znode_t * zp,dmu_tx_t * tx,int flag)695 zfs_dropname(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx,
696     int flag)
697 {
698 	int error;
699 
700 	if (zp->z_zfsvfs->z_norm) {
701 		matchtype_t mt = MT_NORMALIZE;
702 
703 		if (zp->z_zfsvfs->z_case == ZFS_CASE_MIXED) {
704 			mt |= MT_MATCH_CASE;
705 		}
706 
707 		error = zap_remove_norm(zp->z_zfsvfs->z_os, dzp->z_id,
708 		    name, mt, tx);
709 	} else {
710 		error = zap_remove(zp->z_zfsvfs->z_os, dzp->z_id, name, tx);
711 	}
712 
713 	return (error);
714 }
715 
716 /*
717  * Unlink zp from dzp, and mark zp for deletion if this was the last link.
718  * Can fail if zp is a mount point (EBUSY) or a non-empty directory (EEXIST).
719  * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
720  * If it's non-NULL, we use it to indicate whether the znode needs deletion,
721  * and it's the caller's job to do it.
722  */
723 int
zfs_link_destroy(znode_t * dzp,const char * name,znode_t * zp,dmu_tx_t * tx,int flag,boolean_t * unlinkedp)724 zfs_link_destroy(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx,
725     int flag, boolean_t *unlinkedp)
726 {
727 	zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
728 	vnode_t *vp = ZTOV(zp);
729 	int zp_is_dir = (vp->v_type == VDIR);
730 	boolean_t unlinked = B_FALSE;
731 	sa_bulk_attr_t bulk[5];
732 	uint64_t mtime[2], ctime[2];
733 	int count = 0;
734 	int error;
735 
736 	if (zfsvfs->z_replay == B_FALSE) {
737 		ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__);
738 		ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
739 	}
740 	if (!(flag & ZRENAMING)) {
741 
742 		if (zp_is_dir && !zfs_dirempty(zp))
743 			return (SET_ERROR(ENOTEMPTY));
744 
745 		/*
746 		 * If we get here, we are going to try to remove the object.
747 		 * First try removing the name from the directory; if that
748 		 * fails, return the error.
749 		 */
750 		error = zfs_dropname(dzp, name, zp, tx, flag);
751 		if (error != 0) {
752 			return (error);
753 		}
754 
755 		if (zp->z_links <= zp_is_dir) {
756 			zfs_panic_recover("zfs: link count on vnode %p is %u, "
757 			    "should be at least %u", zp->z_vnode,
758 			    (int)zp->z_links,
759 			    zp_is_dir + 1);
760 			zp->z_links = zp_is_dir + 1;
761 		}
762 		if (--zp->z_links == zp_is_dir) {
763 			zp->z_unlinked = B_TRUE;
764 			zp->z_links = 0;
765 			unlinked = B_TRUE;
766 		} else {
767 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
768 			    NULL, &ctime, sizeof (ctime));
769 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
770 			    NULL, &zp->z_pflags, sizeof (zp->z_pflags));
771 			zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
772 			    ctime);
773 		}
774 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
775 		    NULL, &zp->z_links, sizeof (zp->z_links));
776 		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
777 		count = 0;
778 		ASSERT0(error);
779 	} else {
780 		ASSERT(!zp->z_unlinked);
781 		error = zfs_dropname(dzp, name, zp, tx, flag);
782 		if (error != 0)
783 			return (error);
784 	}
785 
786 	dzp->z_size--;		/* one dirent removed */
787 	dzp->z_links -= zp_is_dir;	/* ".." link from zp */
788 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
789 	    NULL, &dzp->z_links, sizeof (dzp->z_links));
790 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
791 	    NULL, &dzp->z_size, sizeof (dzp->z_size));
792 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
793 	    NULL, ctime, sizeof (ctime));
794 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
795 	    NULL, mtime, sizeof (mtime));
796 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
797 	    NULL, &dzp->z_pflags, sizeof (dzp->z_pflags));
798 	zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime);
799 	error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
800 	ASSERT0(error);
801 
802 	if (unlinkedp != NULL)
803 		*unlinkedp = unlinked;
804 	else if (unlinked)
805 		zfs_unlinked_add(zp, tx);
806 
807 	return (0);
808 }
809 
810 /*
811  * Indicate whether the directory is empty.
812  */
813 boolean_t
zfs_dirempty(znode_t * dzp)814 zfs_dirempty(znode_t *dzp)
815 {
816 	return (dzp->z_size == 2);
817 }
818 
819 int
zfs_make_xattrdir(znode_t * zp,vattr_t * vap,znode_t ** xvpp,cred_t * cr)820 zfs_make_xattrdir(znode_t *zp, vattr_t *vap, znode_t **xvpp, cred_t *cr)
821 {
822 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
823 	znode_t *xzp;
824 	dmu_tx_t *tx;
825 	int error;
826 	zfs_acl_ids_t acl_ids;
827 	boolean_t fuid_dirtied;
828 	uint64_t parent __maybe_unused;
829 
830 	*xvpp = NULL;
831 
832 	if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
833 	    &acl_ids, NULL)) != 0)
834 		return (error);
835 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, 0)) {
836 		zfs_acl_ids_free(&acl_ids);
837 		return (SET_ERROR(EDQUOT));
838 	}
839 
840 	getnewvnode_reserve();
841 
842 	tx = dmu_tx_create(zfsvfs->z_os);
843 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
844 	    ZFS_SA_BASE_ATTR_SIZE);
845 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
846 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
847 	fuid_dirtied = zfsvfs->z_fuid_dirty;
848 	if (fuid_dirtied)
849 		zfs_fuid_txhold(zfsvfs, tx);
850 	error = dmu_tx_assign(tx, TXG_WAIT);
851 	if (error) {
852 		zfs_acl_ids_free(&acl_ids);
853 		dmu_tx_abort(tx);
854 		getnewvnode_drop_reserve();
855 		return (error);
856 	}
857 	zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);
858 
859 	if (fuid_dirtied)
860 		zfs_fuid_sync(zfsvfs, tx);
861 
862 	ASSERT0(sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), &parent,
863 	    sizeof (parent)));
864 	ASSERT3U(parent, ==, zp->z_id);
865 
866 	VERIFY0(sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id,
867 	    sizeof (xzp->z_id), tx));
868 
869 	zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp, xzp, "", NULL,
870 	    acl_ids.z_fuidp, vap);
871 
872 	zfs_acl_ids_free(&acl_ids);
873 	dmu_tx_commit(tx);
874 
875 	getnewvnode_drop_reserve();
876 
877 	*xvpp = xzp;
878 
879 	return (0);
880 }
881 
882 /*
883  * Return a znode for the extended attribute directory for zp.
884  * ** If the directory does not already exist, it is created **
885  *
886  *	IN:	zp	- znode to obtain attribute directory from
887  *		cr	- credentials of caller
888  *		flags	- flags from the VOP_LOOKUP call
889  *
890  *	OUT:	xzpp	- pointer to extended attribute znode
891  *
892  *	RETURN:	0 on success
893  *		error number on failure
894  */
895 int
zfs_get_xattrdir(znode_t * zp,znode_t ** xzpp,cred_t * cr,int flags)896 zfs_get_xattrdir(znode_t *zp, znode_t **xzpp, cred_t *cr, int flags)
897 {
898 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
899 	znode_t		*xzp;
900 	vattr_t		va;
901 	int		error;
902 top:
903 	error = zfs_dirent_lookup(zp, "", &xzp, ZXATTR);
904 	if (error)
905 		return (error);
906 
907 	if (xzp != NULL) {
908 		*xzpp = xzp;
909 		return (0);
910 	}
911 
912 
913 	if (!(flags & CREATE_XATTR_DIR))
914 		return (SET_ERROR(ENOATTR));
915 
916 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
917 		return (SET_ERROR(EROFS));
918 	}
919 
920 	/*
921 	 * The ability to 'create' files in an attribute
922 	 * directory comes from the write_xattr permission on the base file.
923 	 *
924 	 * The ability to 'search' an attribute directory requires
925 	 * read_xattr permission on the base file.
926 	 *
927 	 * Once in a directory the ability to read/write attributes
928 	 * is controlled by the permissions on the attribute file.
929 	 */
930 	va.va_mask = AT_MODE | AT_UID | AT_GID;
931 	va.va_type = VDIR;
932 	va.va_mode = S_IFDIR | S_ISVTX | 0777;
933 	zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
934 
935 	error = zfs_make_xattrdir(zp, &va, xzpp, cr);
936 
937 	if (error == ERESTART) {
938 		/* NB: we already did dmu_tx_wait() if necessary */
939 		goto top;
940 	}
941 	if (error == 0)
942 		VOP_UNLOCK(ZTOV(*xzpp));
943 
944 	return (error);
945 }
946 
947 /*
948  * Decide whether it is okay to remove within a sticky directory.
949  *
950  * In sticky directories, write access is not sufficient;
951  * you can remove entries from a directory only if:
952  *
953  *	you own the directory,
954  *	you own the entry,
955  *	the entry is a plain file and you have write access,
956  *	or you are privileged (checked in secpolicy...).
957  *
958  * The function returns 0 if remove access is granted.
959  */
960 int
zfs_sticky_remove_access(znode_t * zdp,znode_t * zp,cred_t * cr)961 zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr)
962 {
963 	uid_t  		uid;
964 	uid_t		downer;
965 	uid_t		fowner;
966 	zfsvfs_t	*zfsvfs = zdp->z_zfsvfs;
967 
968 	if (zdp->z_zfsvfs->z_replay)
969 		return (0);
970 
971 	if ((zdp->z_mode & S_ISVTX) == 0)
972 		return (0);
973 
974 	downer = zfs_fuid_map_id(zfsvfs, zdp->z_uid, cr, ZFS_OWNER);
975 	fowner = zfs_fuid_map_id(zfsvfs, zp->z_uid, cr, ZFS_OWNER);
976 
977 	if ((uid = crgetuid(cr)) == downer || uid == fowner ||
978 	    (ZTOV(zp)->v_type == VREG &&
979 	    zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr, NULL) == 0))
980 		return (0);
981 	else
982 		return (secpolicy_vnode_remove(ZTOV(zp), cr));
983 }
984