xref: /freebsd/sys/contrib/openzfs/module/os/linux/zfs/zfs_dir.c (revision 61145dc2b94f12f6a47344fb9aac702321880e43)
1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3  * CDDL HEADER START
4  *
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or https://opensource.org/licenses/CDDL-1.0.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 
23 /*
24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2013, 2016 by Delphix. All rights reserved.
26  * Copyright 2017 Nexenta Systems, Inc.
27  */
28 
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/time.h>
32 #include <sys/sysmacros.h>
33 #include <sys/vfs.h>
34 #include <sys/vnode.h>
35 #include <sys/file.h>
36 #include <sys/kmem.h>
37 #include <sys/uio.h>
38 #include <sys/pathname.h>
39 #include <sys/cmn_err.h>
40 #include <sys/errno.h>
41 #include <sys/stat.h>
42 #include <sys/sunddi.h>
43 #include <sys/random.h>
44 #include <sys/policy.h>
45 #include <sys/zfs_dir.h>
46 #include <sys/zfs_acl.h>
47 #include <sys/zfs_vnops.h>
48 #include <sys/fs/zfs.h>
49 #include <sys/zap.h>
50 #include <sys/dmu.h>
51 #include <sys/atomic.h>
52 #include <sys/zfs_ctldir.h>
53 #include <sys/zfs_fuid.h>
54 #include <sys/sa.h>
55 #include <sys/zfs_sa.h>
56 #include <sys/dmu_objset.h>
57 #include <sys/dsl_dir.h>
58 
59 /*
60  * zfs_match_find() is used by zfs_dirent_lock() to perform zap lookups
61  * of names after deciding which is the appropriate lookup interface.
62  */
63 static int
zfs_match_find(zfsvfs_t * zfsvfs,znode_t * dzp,const char * name,matchtype_t mt,boolean_t update,int * deflags,pathname_t * rpnp,uint64_t * zoid)64 zfs_match_find(zfsvfs_t *zfsvfs, znode_t *dzp, const char *name,
65     matchtype_t mt, boolean_t update, int *deflags, pathname_t *rpnp,
66     uint64_t *zoid)
67 {
68 	boolean_t conflict = B_FALSE;
69 	int error;
70 
71 	if (zfsvfs->z_norm) {
72 		size_t bufsz = 0;
73 		char *buf = NULL;
74 
75 		if (rpnp) {
76 			buf = rpnp->pn_buf;
77 			bufsz = rpnp->pn_bufsize;
78 		}
79 
80 		/*
81 		 * In the non-mixed case we only expect there would ever
82 		 * be one match, but we need to use the normalizing lookup.
83 		 */
84 		error = zap_lookup_norm(zfsvfs->z_os, dzp->z_id, name, 8, 1,
85 		    zoid, mt, buf, bufsz, &conflict);
86 	} else {
87 		error = zap_lookup(zfsvfs->z_os, dzp->z_id, name, 8, 1, zoid);
88 	}
89 
90 	/*
91 	 * Allow multiple entries provided the first entry is
92 	 * the object id.  Non-zpl consumers may safely make
93 	 * use of the additional space.
94 	 *
95 	 * XXX: This should be a feature flag for compatibility
96 	 */
97 	if (error == EOVERFLOW)
98 		error = 0;
99 
100 	if (zfsvfs->z_norm && !error && deflags)
101 		*deflags = conflict ? ED_CASE_CONFLICT : 0;
102 
103 	*zoid = ZFS_DIRENT_OBJ(*zoid);
104 
105 	return (error);
106 }
107 
108 /*
109  * Lock a directory entry.  A dirlock on <dzp, name> protects that name
110  * in dzp's directory zap object.  As long as you hold a dirlock, you can
111  * assume two things: (1) dzp cannot be reaped, and (2) no other thread
112  * can change the zap entry for (i.e. link or unlink) this name.
113  *
114  * Input arguments:
115  *	dzp	- znode for directory
116  *	name	- name of entry to lock
117  *	flag	- ZNEW: if the entry already exists, fail with EEXIST.
118  *		  ZEXISTS: if the entry does not exist, fail with ENOENT.
119  *		  ZSHARED: allow concurrent access with other ZSHARED callers.
120  *		  ZXATTR: we want dzp's xattr directory
121  *		  ZCILOOK: On a mixed sensitivity file system,
122  *			   this lookup should be case-insensitive.
123  *		  ZCIEXACT: On a purely case-insensitive file system,
124  *			    this lookup should be case-sensitive.
125  *		  ZRENAMING: we are locking for renaming, force narrow locks
126  *		  ZHAVELOCK: Don't grab the z_name_lock for this call. The
127  *			     current thread already holds it.
128  *
129  * Output arguments:
130  *	zpp	- pointer to the znode for the entry (NULL if there isn't one)
131  *	dlpp	- pointer to the dirlock for this entry (NULL on error)
132  *      direntflags - (case-insensitive lookup only)
133  *		flags if multiple case-sensitive matches exist in directory
134  *      realpnp     - (case-insensitive lookup only)
135  *		actual name matched within the directory
136  *
137  * Return value: 0 on success or errno on failure.
138  *
139  * NOTE: Always checks for, and rejects, '.' and '..'.
140  * NOTE: For case-insensitive file systems we take wide locks (see below),
141  *	 but return znode pointers to a single match.
142  */
143 int
zfs_dirent_lock(zfs_dirlock_t ** dlpp,znode_t * dzp,char * name,znode_t ** zpp,int flag,int * direntflags,pathname_t * realpnp)144 zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name,
145     znode_t **zpp, int flag, int *direntflags, pathname_t *realpnp)
146 {
147 	zfsvfs_t	*zfsvfs = ZTOZSB(dzp);
148 	zfs_dirlock_t	*dl;
149 	boolean_t	update;
150 	matchtype_t	mt = 0;
151 	uint64_t	zoid;
152 	int		error = 0;
153 	int		cmpflags;
154 
155 	*zpp = NULL;
156 	*dlpp = NULL;
157 
158 	/*
159 	 * Verify that we are not trying to lock '.', '..', or '.zfs'
160 	 */
161 	if ((name[0] == '.' &&
162 	    (name[1] == '\0' || (name[1] == '.' && name[2] == '\0'))) ||
163 	    (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0))
164 		return (SET_ERROR(EEXIST));
165 
166 	/*
167 	 * Case sensitivity and normalization preferences are set when
168 	 * the file system is created.  These are stored in the
169 	 * zfsvfs->z_case and zfsvfs->z_norm fields.  These choices
170 	 * affect what vnodes can be cached in the DNLC, how we
171 	 * perform zap lookups, and the "width" of our dirlocks.
172 	 *
173 	 * A normal dirlock locks a single name.  Note that with
174 	 * normalization a name can be composed multiple ways, but
175 	 * when normalized, these names all compare equal.  A wide
176 	 * dirlock locks multiple names.  We need these when the file
177 	 * system is supporting mixed-mode access.  It is sometimes
178 	 * necessary to lock all case permutations of file name at
179 	 * once so that simultaneous case-insensitive/case-sensitive
180 	 * behaves as rationally as possible.
181 	 */
182 
183 	/*
184 	 * When matching we may need to normalize & change case according to
185 	 * FS settings.
186 	 *
187 	 * Note that a normalized match is necessary for a case insensitive
188 	 * filesystem when the lookup request is not exact because normalization
189 	 * can fold case independent of normalizing code point sequences.
190 	 *
191 	 * See the table above zfs_dropname().
192 	 */
193 	if (zfsvfs->z_norm != 0) {
194 		mt = MT_NORMALIZE;
195 
196 		/*
197 		 * Determine if the match needs to honor the case specified in
198 		 * lookup, and if so keep track of that so that during
199 		 * normalization we don't fold case.
200 		 */
201 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE &&
202 		    (flag & ZCIEXACT)) ||
203 		    (zfsvfs->z_case == ZFS_CASE_MIXED && !(flag & ZCILOOK))) {
204 			mt |= MT_MATCH_CASE;
205 		}
206 	}
207 
208 	/*
209 	 * Only look in or update the DNLC if we are looking for the
210 	 * name on a file system that does not require normalization
211 	 * or case folding.  We can also look there if we happen to be
212 	 * on a non-normalizing, mixed sensitivity file system IF we
213 	 * are looking for the exact name.
214 	 *
215 	 * Maybe can add TO-UPPERed version of name to dnlc in ci-only
216 	 * case for performance improvement?
217 	 */
218 	update = !zfsvfs->z_norm ||
219 	    (zfsvfs->z_case == ZFS_CASE_MIXED &&
220 	    !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER) && !(flag & ZCILOOK));
221 
222 	/*
223 	 * ZRENAMING indicates we are in a situation where we should
224 	 * take narrow locks regardless of the file system's
225 	 * preferences for normalizing and case folding.  This will
226 	 * prevent us deadlocking trying to grab the same wide lock
227 	 * twice if the two names happen to be case-insensitive
228 	 * matches.
229 	 */
230 	if (flag & ZRENAMING)
231 		cmpflags = 0;
232 	else
233 		cmpflags = zfsvfs->z_norm;
234 
235 	/*
236 	 * Wait until there are no locks on this name.
237 	 *
238 	 * Don't grab the lock if it is already held. However, cannot
239 	 * have both ZSHARED and ZHAVELOCK together.
240 	 */
241 	ASSERT(!(flag & ZSHARED) || !(flag & ZHAVELOCK));
242 	if (!(flag & ZHAVELOCK))
243 		rw_enter(&dzp->z_name_lock, RW_READER);
244 
245 	mutex_enter(&dzp->z_lock);
246 	for (;;) {
247 		if (dzp->z_unlinked && !(flag & ZXATTR)) {
248 			mutex_exit(&dzp->z_lock);
249 			if (!(flag & ZHAVELOCK))
250 				rw_exit(&dzp->z_name_lock);
251 			return (SET_ERROR(ENOENT));
252 		}
253 		for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next) {
254 			if ((u8_strcmp(name, dl->dl_name, 0, cmpflags,
255 			    U8_UNICODE_LATEST, &error) == 0) || error != 0)
256 				break;
257 		}
258 		if (error != 0) {
259 			mutex_exit(&dzp->z_lock);
260 			if (!(flag & ZHAVELOCK))
261 				rw_exit(&dzp->z_name_lock);
262 			return (SET_ERROR(ENOENT));
263 		}
264 		if (dl == NULL)	{
265 			/*
266 			 * Allocate a new dirlock and add it to the list.
267 			 */
268 			dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP);
269 			cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL);
270 			dl->dl_name = name;
271 			dl->dl_sharecnt = 0;
272 			dl->dl_namelock = 0;
273 			dl->dl_namesize = 0;
274 			dl->dl_dzp = dzp;
275 			dl->dl_next = dzp->z_dirlocks;
276 			dzp->z_dirlocks = dl;
277 			break;
278 		}
279 		if ((flag & ZSHARED) && dl->dl_sharecnt != 0)
280 			break;
281 		cv_wait(&dl->dl_cv, &dzp->z_lock);
282 	}
283 
284 	/*
285 	 * If the z_name_lock was NOT held for this dirlock record it.
286 	 */
287 	if (flag & ZHAVELOCK)
288 		dl->dl_namelock = 1;
289 
290 	if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) {
291 		/*
292 		 * We're the second shared reference to dl.  Make a copy of
293 		 * dl_name in case the first thread goes away before we do.
294 		 * Note that we initialize the new name before storing its
295 		 * pointer into dl_name, because the first thread may load
296 		 * dl->dl_name at any time.  It'll either see the old value,
297 		 * which belongs to it, or the new shared copy; either is OK.
298 		 */
299 		dl->dl_namesize = strlen(dl->dl_name) + 1;
300 		name = kmem_alloc(dl->dl_namesize, KM_SLEEP);
301 		memcpy(name, dl->dl_name, dl->dl_namesize);
302 		dl->dl_name = name;
303 	}
304 
305 	mutex_exit(&dzp->z_lock);
306 
307 	/*
308 	 * We have a dirlock on the name.  (Note that it is the dirlock,
309 	 * not the dzp's z_lock, that protects the name in the zap object.)
310 	 * See if there's an object by this name; if so, put a hold on it.
311 	 */
312 	if (flag & ZXATTR) {
313 		error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid,
314 		    sizeof (zoid));
315 		if (error == 0)
316 			error = (zoid == 0 ? SET_ERROR(ENOENT) : 0);
317 	} else {
318 		error = zfs_match_find(zfsvfs, dzp, name, mt,
319 		    update, direntflags, realpnp, &zoid);
320 	}
321 	if (error) {
322 		if (error != ENOENT || (flag & ZEXISTS)) {
323 			zfs_dirent_unlock(dl);
324 			return (error);
325 		}
326 	} else {
327 		if (flag & ZNEW) {
328 			zfs_dirent_unlock(dl);
329 			return (SET_ERROR(EEXIST));
330 		}
331 		error = zfs_zget(zfsvfs, zoid, zpp);
332 		if (error) {
333 			zfs_dirent_unlock(dl);
334 			return (error);
335 		}
336 	}
337 
338 	*dlpp = dl;
339 
340 	return (0);
341 }
342 
343 /*
344  * Unlock this directory entry and wake anyone who was waiting for it.
345  */
346 void
zfs_dirent_unlock(zfs_dirlock_t * dl)347 zfs_dirent_unlock(zfs_dirlock_t *dl)
348 {
349 	znode_t *dzp = dl->dl_dzp;
350 	zfs_dirlock_t **prev_dl, *cur_dl;
351 
352 	mutex_enter(&dzp->z_lock);
353 
354 	if (!dl->dl_namelock)
355 		rw_exit(&dzp->z_name_lock);
356 
357 	if (dl->dl_sharecnt > 1) {
358 		dl->dl_sharecnt--;
359 		mutex_exit(&dzp->z_lock);
360 		return;
361 	}
362 	prev_dl = &dzp->z_dirlocks;
363 	while ((cur_dl = *prev_dl) != dl)
364 		prev_dl = &cur_dl->dl_next;
365 	*prev_dl = dl->dl_next;
366 	cv_broadcast(&dl->dl_cv);
367 	mutex_exit(&dzp->z_lock);
368 
369 	if (dl->dl_namesize != 0)
370 		kmem_free(dl->dl_name, dl->dl_namesize);
371 	cv_destroy(&dl->dl_cv);
372 	kmem_free(dl, sizeof (*dl));
373 }
374 
375 /*
376  * Look up an entry in a directory.
377  *
378  * NOTE: '.' and '..' are handled as special cases because
379  *	no directory entries are actually stored for them.  If this is
380  *	the root of a filesystem, then '.zfs' is also treated as a
381  *	special pseudo-directory.
382  */
383 int
zfs_dirlook(znode_t * dzp,char * name,znode_t ** zpp,int flags,int * deflg,pathname_t * rpnp)384 zfs_dirlook(znode_t *dzp, char *name, znode_t **zpp, int flags,
385     int *deflg, pathname_t *rpnp)
386 {
387 	zfs_dirlock_t *dl;
388 	znode_t *zp;
389 	struct inode *ip;
390 	int error = 0;
391 	uint64_t parent;
392 
393 	if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
394 		*zpp = dzp;
395 		zhold(*zpp);
396 	} else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
397 		zfsvfs_t *zfsvfs = ZTOZSB(dzp);
398 
399 		/*
400 		 * If we are a snapshot mounted under .zfs, return
401 		 * the inode pointer for the snapshot directory.
402 		 */
403 		if ((error = sa_lookup(dzp->z_sa_hdl,
404 		    SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
405 			return (error);
406 
407 		if (parent == dzp->z_id && zfsvfs->z_parent != zfsvfs) {
408 			error = zfsctl_root_lookup(zfsvfs->z_parent->z_ctldir,
409 			    "snapshot", &ip, 0, kcred, NULL, NULL);
410 			*zpp = ITOZ(ip);
411 			return (error);
412 		}
413 		rw_enter(&dzp->z_parent_lock, RW_READER);
414 		error = zfs_zget(zfsvfs, parent, &zp);
415 		if (error == 0)
416 			*zpp = zp;
417 		rw_exit(&dzp->z_parent_lock);
418 	} else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
419 		if (ZTOZSB(dzp)->z_show_ctldir == ZFS_SNAPDIR_DISABLED) {
420 			return (SET_ERROR(ENOENT));
421 		}
422 		ip = zfsctl_root(dzp);
423 		*zpp = ITOZ(ip);
424 	} else {
425 		int zf;
426 
427 		zf = ZEXISTS | ZSHARED;
428 		if (flags & FIGNORECASE)
429 			zf |= ZCILOOK;
430 
431 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp);
432 		if (error == 0) {
433 			*zpp = zp;
434 			zfs_dirent_unlock(dl);
435 			dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
436 		}
437 		rpnp = NULL;
438 	}
439 
440 	if ((flags & FIGNORECASE) && rpnp && !error)
441 		(void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize);
442 
443 	return (error);
444 }
445 
446 /*
447  * unlinked Set (formerly known as the "delete queue") Error Handling
448  *
449  * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
450  * don't specify the name of the entry that we will be manipulating.  We
451  * also fib and say that we won't be adding any new entries to the
452  * unlinked set, even though we might (this is to lower the minimum file
453  * size that can be deleted in a full filesystem).  So on the small
454  * chance that the nlink list is using a fat zap (ie. has more than
455  * 2000 entries), we *may* not pre-read a block that's needed.
456  * Therefore it is remotely possible for some of the assertions
457  * regarding the unlinked set below to fail due to i/o error.  On a
458  * nondebug system, this will result in the space being leaked.
459  */
460 void
zfs_unlinked_add(znode_t * zp,dmu_tx_t * tx)461 zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx)
462 {
463 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
464 
465 	ASSERT(zp->z_unlinked);
466 	ASSERT(ZTOI(zp)->i_nlink == 0);
467 
468 	VERIFY3U(0, ==,
469 	    zap_add_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
470 
471 	dataset_kstats_update_nunlinks_kstat(&zfsvfs->z_kstat, 1);
472 }
473 
474 /*
475  * Clean up any znodes that had no links when we either crashed or
476  * (force) umounted the file system.
477  */
478 static void
zfs_unlinked_drain_task(void * arg)479 zfs_unlinked_drain_task(void *arg)
480 {
481 	zfsvfs_t *zfsvfs = arg;
482 	zap_cursor_t	zc;
483 	zap_attribute_t *zap = zap_attribute_alloc();
484 	dmu_object_info_t doi;
485 	znode_t		*zp;
486 	int		error;
487 
488 	ASSERT3B(zfsvfs->z_draining, ==, B_TRUE);
489 
490 	/*
491 	 * Iterate over the contents of the unlinked set.
492 	 */
493 	for (zap_cursor_init(&zc, zfsvfs->z_os, zfsvfs->z_unlinkedobj);
494 	    zap_cursor_retrieve(&zc, zap) == 0 && !zfsvfs->z_drain_cancel;
495 	    zap_cursor_advance(&zc)) {
496 
497 		/*
498 		 * See what kind of object we have in list
499 		 */
500 
501 		error = dmu_object_info(zfsvfs->z_os,
502 		    zap->za_first_integer, &doi);
503 		if (error != 0)
504 			continue;
505 
506 		ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) ||
507 		    (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS));
508 		/*
509 		 * We need to re-mark these list entries for deletion,
510 		 * so we pull them back into core and set zp->z_unlinked.
511 		 */
512 		error = zfs_zget(zfsvfs, zap->za_first_integer, &zp);
513 
514 		/*
515 		 * We may pick up znodes that are already marked for deletion.
516 		 * This could happen during the purge of an extended attribute
517 		 * directory.  All we need to do is skip over them, since they
518 		 * are already in the system marked z_unlinked.
519 		 */
520 		if (error != 0)
521 			continue;
522 
523 		zp->z_unlinked = B_TRUE;
524 
525 		/*
526 		 * zrele() decrements the znode's ref count and may cause
527 		 * it to be synchronously freed. We interrupt freeing
528 		 * of this znode by checking the return value of
529 		 * dmu_objset_zfs_unmounting() in dmu_free_long_range()
530 		 * when an unmount is requested.
531 		 */
532 		zrele(zp);
533 		ASSERT3B(zfsvfs->z_unmounted, ==, B_FALSE);
534 	}
535 	zap_cursor_fini(&zc);
536 
537 	zfsvfs->z_draining = B_FALSE;
538 	zfsvfs->z_drain_task = TASKQID_INVALID;
539 	zap_attribute_free(zap);
540 }
541 
542 /*
543  * Sets z_draining then tries to dispatch async unlinked drain.
544  * If that fails executes synchronous unlinked drain.
545  */
546 void
zfs_unlinked_drain(zfsvfs_t * zfsvfs)547 zfs_unlinked_drain(zfsvfs_t *zfsvfs)
548 {
549 	ASSERT3B(zfsvfs->z_unmounted, ==, B_FALSE);
550 	ASSERT3B(zfsvfs->z_draining, ==, B_FALSE);
551 
552 	zfsvfs->z_draining = B_TRUE;
553 	zfsvfs->z_drain_cancel = B_FALSE;
554 
555 	zfsvfs->z_drain_task = taskq_dispatch(
556 	    dsl_pool_unlinked_drain_taskq(dmu_objset_pool(zfsvfs->z_os)),
557 	    zfs_unlinked_drain_task, zfsvfs, TQ_SLEEP);
558 	if (zfsvfs->z_drain_task == TASKQID_INVALID) {
559 		zfs_dbgmsg("async zfs_unlinked_drain dispatch failed");
560 		zfs_unlinked_drain_task(zfsvfs);
561 	}
562 }
563 
564 /*
565  * Wait for the unlinked drain taskq task to stop. This will interrupt the
566  * unlinked set processing if it is in progress.
567  */
568 void
zfs_unlinked_drain_stop_wait(zfsvfs_t * zfsvfs)569 zfs_unlinked_drain_stop_wait(zfsvfs_t *zfsvfs)
570 {
571 	ASSERT3B(zfsvfs->z_unmounted, ==, B_FALSE);
572 
573 	if (zfsvfs->z_draining) {
574 		zfsvfs->z_drain_cancel = B_TRUE;
575 		taskq_cancel_id(dsl_pool_unlinked_drain_taskq(
576 		    dmu_objset_pool(zfsvfs->z_os)), zfsvfs->z_drain_task);
577 		zfsvfs->z_drain_task = TASKQID_INVALID;
578 		zfsvfs->z_draining = B_FALSE;
579 	}
580 }
581 
582 /*
583  * Delete the entire contents of a directory.  Return a count
584  * of the number of entries that could not be deleted. If we encounter
585  * an error, return a count of at least one so that the directory stays
586  * in the unlinked set.
587  *
588  * NOTE: this function assumes that the directory is inactive,
589  *	so there is no need to lock its entries before deletion.
590  *	Also, it assumes the directory contents is *only* regular
591  *	files.
592  */
593 static int
zfs_purgedir(znode_t * dzp)594 zfs_purgedir(znode_t *dzp)
595 {
596 	zap_cursor_t	zc;
597 	zap_attribute_t	*zap = zap_attribute_alloc();
598 	znode_t		*xzp;
599 	dmu_tx_t	*tx;
600 	zfsvfs_t	*zfsvfs = ZTOZSB(dzp);
601 	zfs_dirlock_t	dl;
602 	int skipped = 0;
603 	int error;
604 
605 	for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id);
606 	    (error = zap_cursor_retrieve(&zc, zap)) == 0;
607 	    zap_cursor_advance(&zc)) {
608 		error = zfs_zget(zfsvfs,
609 		    ZFS_DIRENT_OBJ(zap->za_first_integer), &xzp);
610 		if (error) {
611 			skipped += 1;
612 			continue;
613 		}
614 
615 		ASSERT(S_ISREG(ZTOI(xzp)->i_mode) ||
616 		    S_ISLNK(ZTOI(xzp)->i_mode));
617 
618 		tx = dmu_tx_create(zfsvfs->z_os);
619 		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
620 		dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap->za_name);
621 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
622 		dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
623 		/* Is this really needed ? */
624 		zfs_sa_upgrade_txholds(tx, xzp);
625 		dmu_tx_mark_netfree(tx);
626 		error = dmu_tx_assign(tx, DMU_TX_WAIT);
627 		if (error) {
628 			dmu_tx_abort(tx);
629 			zfs_zrele_async(xzp);
630 			skipped += 1;
631 			continue;
632 		}
633 		memset(&dl, 0, sizeof (dl));
634 		dl.dl_dzp = dzp;
635 		dl.dl_name = zap->za_name;
636 
637 		error = zfs_link_destroy(&dl, xzp, tx, 0, NULL);
638 		if (error)
639 			skipped += 1;
640 		dmu_tx_commit(tx);
641 
642 		zfs_zrele_async(xzp);
643 	}
644 	zap_cursor_fini(&zc);
645 	zap_attribute_free(zap);
646 	if (error != ENOENT)
647 		skipped += 1;
648 	return (skipped);
649 }
650 
651 void
zfs_rmnode(znode_t * zp)652 zfs_rmnode(znode_t *zp)
653 {
654 	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
655 	objset_t	*os = zfsvfs->z_os;
656 	znode_t		*xzp = NULL;
657 	dmu_tx_t	*tx;
658 	znode_hold_t	*zh;
659 	uint64_t	z_id = zp->z_id;
660 	uint64_t	acl_obj;
661 	uint64_t	xattr_obj;
662 	uint64_t	links;
663 	int		error;
664 
665 	ASSERT(ZTOI(zp)->i_nlink == 0);
666 	ASSERT(atomic_read(&ZTOI(zp)->i_count) == 0);
667 
668 	/*
669 	 * If this is an attribute directory, purge its contents.
670 	 */
671 	if (S_ISDIR(ZTOI(zp)->i_mode) && (zp->z_pflags & ZFS_XATTR)) {
672 		if (zfs_purgedir(zp) != 0) {
673 			/*
674 			 * Not enough space to delete some xattrs.
675 			 * Leave it in the unlinked set.
676 			 */
677 			zh = zfs_znode_hold_enter(zfsvfs, z_id);
678 			zfs_znode_dmu_fini(zp);
679 			zfs_znode_hold_exit(zfsvfs, zh);
680 			return;
681 		}
682 	}
683 
684 	/*
685 	 * Free up all the data in the file.  We don't do this for directories
686 	 * because we need truncate and remove to be in the same tx, like in
687 	 * zfs_znode_delete(). Otherwise, if we crash here we'll end up with
688 	 * an inconsistent truncated zap object in the delete queue.  Note a
689 	 * truncated file is harmless since it only contains user data.
690 	 */
691 	if (S_ISREG(ZTOI(zp)->i_mode)) {
692 		error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
693 		if (error) {
694 			/*
695 			 * Not enough space or we were interrupted by unmount.
696 			 * Leave the file in the unlinked set.
697 			 */
698 			zh = zfs_znode_hold_enter(zfsvfs, z_id);
699 			zfs_znode_dmu_fini(zp);
700 			zfs_znode_hold_exit(zfsvfs, zh);
701 			return;
702 		}
703 	}
704 
705 	/*
706 	 * If the file has extended attributes, we're going to unlink
707 	 * the xattr dir.
708 	 */
709 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
710 	    &xattr_obj, sizeof (xattr_obj));
711 	if (error == 0 && xattr_obj) {
712 		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
713 		ASSERT(error == 0);
714 	}
715 
716 	acl_obj = zfs_external_acl(zp);
717 
718 	/*
719 	 * Set up the final transaction.
720 	 */
721 	tx = dmu_tx_create(os);
722 	dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
723 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
724 	if (xzp) {
725 		dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL);
726 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
727 	}
728 	if (acl_obj)
729 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
730 
731 	zfs_sa_upgrade_txholds(tx, zp);
732 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
733 	if (error) {
734 		/*
735 		 * Not enough space to delete the file.  Leave it in the
736 		 * unlinked set, leaking it until the fs is remounted (at
737 		 * which point we'll call zfs_unlinked_drain() to process it).
738 		 */
739 		dmu_tx_abort(tx);
740 		zh = zfs_znode_hold_enter(zfsvfs, z_id);
741 		zfs_znode_dmu_fini(zp);
742 		zfs_znode_hold_exit(zfsvfs, zh);
743 		goto out;
744 	}
745 
746 	if (xzp) {
747 		ASSERT(error == 0);
748 		mutex_enter(&xzp->z_lock);
749 		xzp->z_unlinked = B_TRUE;	/* mark xzp for deletion */
750 		clear_nlink(ZTOI(xzp));		/* no more links to it */
751 		links = 0;
752 		VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
753 		    &links, sizeof (links), tx));
754 		mutex_exit(&xzp->z_lock);
755 		zfs_unlinked_add(xzp, tx);
756 	}
757 
758 	mutex_enter(&os->os_dsl_dataset->ds_dir->dd_activity_lock);
759 
760 	/*
761 	 * Remove this znode from the unlinked set.  If a has rollback has
762 	 * occurred while a file is open and unlinked.  Then when the file
763 	 * is closed post rollback it will not exist in the rolled back
764 	 * version of the unlinked object.
765 	 */
766 	error = zap_remove_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj,
767 	    zp->z_id, tx);
768 	VERIFY(error == 0 || error == ENOENT);
769 
770 	uint64_t count;
771 	if (zap_count(os, zfsvfs->z_unlinkedobj, &count) == 0 && count == 0) {
772 		cv_broadcast(&os->os_dsl_dataset->ds_dir->dd_activity_cv);
773 	}
774 
775 	mutex_exit(&os->os_dsl_dataset->ds_dir->dd_activity_lock);
776 
777 	dataset_kstats_update_nunlinked_kstat(&zfsvfs->z_kstat, 1);
778 
779 	zfs_znode_delete(zp, tx);
780 
781 	dmu_tx_commit(tx);
782 out:
783 	if (xzp)
784 		zfs_zrele_async(xzp);
785 }
786 
787 static uint64_t
zfs_dirent(znode_t * zp,uint64_t mode)788 zfs_dirent(znode_t *zp, uint64_t mode)
789 {
790 	uint64_t de = zp->z_id;
791 
792 	if (ZTOZSB(zp)->z_version >= ZPL_VERSION_DIRENT_TYPE)
793 		de |= IFTODT(mode) << 60;
794 	return (de);
795 }
796 
797 /*
798  * Link zp into dl.  Can fail in the following cases :
799  * - if zp has been unlinked.
800  * - if the number of entries with the same hash (aka. colliding entries)
801  *    exceed the capacity of a leaf-block of fatzap and splitting of the
802  *    leaf-block does not help.
803  */
804 int
zfs_link_create(zfs_dirlock_t * dl,znode_t * zp,dmu_tx_t * tx,int flag)805 zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag)
806 {
807 	znode_t *dzp = dl->dl_dzp;
808 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
809 	uint64_t value;
810 	int zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode);
811 	sa_bulk_attr_t bulk[5];
812 	uint64_t mtime[2], ctime[2];
813 	uint64_t links;
814 	int count = 0;
815 	int error;
816 
817 	mutex_enter(&zp->z_lock);
818 
819 	if (!(flag & ZRENAMING)) {
820 		if (zp->z_unlinked) {	/* no new links to unlinked zp */
821 			ASSERT(!(flag & (ZNEW | ZEXISTS)));
822 			mutex_exit(&zp->z_lock);
823 			return (SET_ERROR(ENOENT));
824 		}
825 		if (!(flag & ZNEW)) {
826 			/*
827 			 * ZNEW nodes come from zfs_mknode() where the link
828 			 * count has already been initialised
829 			 */
830 			inc_nlink(ZTOI(zp));
831 			links = ZTOI(zp)->i_nlink;
832 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
833 			    NULL, &links, sizeof (links));
834 		}
835 	}
836 
837 	value = zfs_dirent(zp, zp->z_mode);
838 	error = zap_add(ZTOZSB(zp)->z_os, dzp->z_id, dl->dl_name, 8, 1,
839 	    &value, tx);
840 
841 	/*
842 	 * zap_add could fail to add the entry if it exceeds the capacity of the
843 	 * leaf-block and zap_leaf_split() failed to help.
844 	 * The caller of this routine is responsible for failing the transaction
845 	 * which will rollback the SA updates done above.
846 	 */
847 	if (error != 0) {
848 		if (!(flag & ZRENAMING) && !(flag & ZNEW))
849 			drop_nlink(ZTOI(zp));
850 		mutex_exit(&zp->z_lock);
851 		return (error);
852 	}
853 
854 	/*
855 	 * If we added a longname activate the SPA_FEATURE_LONGNAME.
856 	 */
857 	if (strlen(dl->dl_name) >= ZAP_MAXNAMELEN) {
858 		dsl_dataset_t *ds = dmu_objset_ds(zfsvfs->z_os);
859 		ds->ds_feature_activation[SPA_FEATURE_LONGNAME] =
860 		    (void *)B_TRUE;
861 	}
862 
863 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
864 	    &dzp->z_id, sizeof (dzp->z_id));
865 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
866 	    &zp->z_pflags, sizeof (zp->z_pflags));
867 
868 	if (!(flag & ZNEW)) {
869 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
870 		    ctime, sizeof (ctime));
871 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
872 		    ctime);
873 	}
874 	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
875 	ASSERT(error == 0);
876 
877 	mutex_exit(&zp->z_lock);
878 
879 	mutex_enter(&dzp->z_lock);
880 	dzp->z_size++;
881 	if (zp_is_dir)
882 		inc_nlink(ZTOI(dzp));
883 	links = ZTOI(dzp)->i_nlink;
884 	count = 0;
885 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
886 	    &dzp->z_size, sizeof (dzp->z_size));
887 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
888 	    &links, sizeof (links));
889 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
890 	    mtime, sizeof (mtime));
891 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
892 	    ctime, sizeof (ctime));
893 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
894 	    &dzp->z_pflags, sizeof (dzp->z_pflags));
895 	zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime);
896 	error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
897 	ASSERT(error == 0);
898 	mutex_exit(&dzp->z_lock);
899 
900 	return (0);
901 }
902 
903 /*
904  * The match type in the code for this function should conform to:
905  *
906  * ------------------------------------------------------------------------
907  * fs type  | z_norm      | lookup type | match type
908  * ---------|-------------|-------------|----------------------------------
909  * CS !norm | 0           |           0 | 0 (exact)
910  * CS  norm | formX       |           0 | MT_NORMALIZE
911  * CI !norm | upper       |   !ZCIEXACT | MT_NORMALIZE
912  * CI !norm | upper       |    ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
913  * CI  norm | upper|formX |   !ZCIEXACT | MT_NORMALIZE
914  * CI  norm | upper|formX |    ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
915  * CM !norm | upper       |    !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
916  * CM !norm | upper       |     ZCILOOK | MT_NORMALIZE
917  * CM  norm | upper|formX |    !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
918  * CM  norm | upper|formX |     ZCILOOK | MT_NORMALIZE
919  *
920  * Abbreviations:
921  *    CS = Case Sensitive, CI = Case Insensitive, CM = Case Mixed
922  *    upper = case folding set by fs type on creation (U8_TEXTPREP_TOUPPER)
923  *    formX = unicode normalization form set on fs creation
924  */
925 static int
zfs_dropname(zfs_dirlock_t * dl,znode_t * zp,znode_t * dzp,dmu_tx_t * tx,int flag)926 zfs_dropname(zfs_dirlock_t *dl, znode_t *zp, znode_t *dzp, dmu_tx_t *tx,
927     int flag)
928 {
929 	int error;
930 
931 	if (ZTOZSB(zp)->z_norm) {
932 		matchtype_t mt = MT_NORMALIZE;
933 
934 		if ((ZTOZSB(zp)->z_case == ZFS_CASE_INSENSITIVE &&
935 		    (flag & ZCIEXACT)) ||
936 		    (ZTOZSB(zp)->z_case == ZFS_CASE_MIXED &&
937 		    !(flag & ZCILOOK))) {
938 			mt |= MT_MATCH_CASE;
939 		}
940 
941 		error = zap_remove_norm(ZTOZSB(zp)->z_os, dzp->z_id,
942 		    dl->dl_name, mt, tx);
943 	} else {
944 		error = zap_remove(ZTOZSB(zp)->z_os, dzp->z_id, dl->dl_name,
945 		    tx);
946 	}
947 
948 	return (error);
949 }
950 
951 static int
zfs_drop_nlink_locked(znode_t * zp,dmu_tx_t * tx,boolean_t * unlinkedp)952 zfs_drop_nlink_locked(znode_t *zp, dmu_tx_t *tx, boolean_t *unlinkedp)
953 {
954 	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
955 	int		zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode);
956 	boolean_t	unlinked = B_FALSE;
957 	sa_bulk_attr_t	bulk[3];
958 	uint64_t	mtime[2], ctime[2];
959 	uint64_t	links;
960 	int		count = 0;
961 	int		error;
962 
963 	if (zp_is_dir && !zfs_dirempty(zp))
964 		return (SET_ERROR(ENOTEMPTY));
965 
966 	if (ZTOI(zp)->i_nlink <= zp_is_dir) {
967 		zfs_panic_recover("zfs: link count on %lu is %u, "
968 		    "should be at least %u", zp->z_id,
969 		    (int)ZTOI(zp)->i_nlink, zp_is_dir + 1);
970 		set_nlink(ZTOI(zp), zp_is_dir + 1);
971 	}
972 	drop_nlink(ZTOI(zp));
973 	if (ZTOI(zp)->i_nlink == zp_is_dir) {
974 		zp->z_unlinked = B_TRUE;
975 		clear_nlink(ZTOI(zp));
976 		unlinked = B_TRUE;
977 	} else {
978 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
979 		    NULL, &ctime, sizeof (ctime));
980 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
981 		    NULL, &zp->z_pflags, sizeof (zp->z_pflags));
982 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
983 		    ctime);
984 	}
985 	links = ZTOI(zp)->i_nlink;
986 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
987 	    NULL, &links, sizeof (links));
988 	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
989 	ASSERT3U(error, ==, 0);
990 
991 	if (unlinkedp != NULL)
992 		*unlinkedp = unlinked;
993 	else if (unlinked)
994 		zfs_unlinked_add(zp, tx);
995 
996 	return (0);
997 }
998 
999 /*
1000  * Forcefully drop an nlink reference from (zp) and mark it for deletion if it
1001  * was the last link. This *must* only be done to znodes which have already
1002  * been zfs_link_destroy()'d with ZRENAMING. This is explicitly only used in
1003  * the error path of zfs_rename(), where we have to correct the nlink count if
1004  * we failed to link the target as well as failing to re-link the original
1005  * znodes.
1006  */
1007 int
zfs_drop_nlink(znode_t * zp,dmu_tx_t * tx,boolean_t * unlinkedp)1008 zfs_drop_nlink(znode_t *zp, dmu_tx_t *tx, boolean_t *unlinkedp)
1009 {
1010 	int error;
1011 
1012 	mutex_enter(&zp->z_lock);
1013 	error = zfs_drop_nlink_locked(zp, tx, unlinkedp);
1014 	mutex_exit(&zp->z_lock);
1015 
1016 	return (error);
1017 }
1018 
1019 /*
1020  * Unlink zp from dl, and mark zp for deletion if this was the last link. Can
1021  * fail if zp is a mount point (EBUSY) or a non-empty directory (ENOTEMPTY).
1022  * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
1023  * If it's non-NULL, we use it to indicate whether the znode needs deletion,
1024  * and it's the caller's job to do it.
1025  */
1026 int
zfs_link_destroy(zfs_dirlock_t * dl,znode_t * zp,dmu_tx_t * tx,int flag,boolean_t * unlinkedp)1027 zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag,
1028     boolean_t *unlinkedp)
1029 {
1030 	znode_t *dzp = dl->dl_dzp;
1031 	zfsvfs_t *zfsvfs = ZTOZSB(dzp);
1032 	int zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode);
1033 	boolean_t unlinked = B_FALSE;
1034 	sa_bulk_attr_t bulk[5];
1035 	uint64_t mtime[2], ctime[2];
1036 	uint64_t links;
1037 	int count = 0;
1038 	int error;
1039 
1040 	if (!(flag & ZRENAMING)) {
1041 		mutex_enter(&zp->z_lock);
1042 
1043 		if (zp_is_dir && !zfs_dirempty(zp)) {
1044 			mutex_exit(&zp->z_lock);
1045 			return (SET_ERROR(ENOTEMPTY));
1046 		}
1047 
1048 		/*
1049 		 * If we get here, we are going to try to remove the object.
1050 		 * First try removing the name from the directory; if that
1051 		 * fails, return the error.
1052 		 */
1053 		error = zfs_dropname(dl, zp, dzp, tx, flag);
1054 		if (error != 0) {
1055 			mutex_exit(&zp->z_lock);
1056 			return (error);
1057 		}
1058 
1059 		/* The only error is !zfs_dirempty() and we checked earlier. */
1060 		error = zfs_drop_nlink_locked(zp, tx, &unlinked);
1061 		ASSERT3U(error, ==, 0);
1062 		mutex_exit(&zp->z_lock);
1063 	} else {
1064 		error = zfs_dropname(dl, zp, dzp, tx, flag);
1065 		if (error != 0)
1066 			return (error);
1067 	}
1068 
1069 	mutex_enter(&dzp->z_lock);
1070 	dzp->z_size--;		/* one dirent removed */
1071 	if (zp_is_dir)
1072 		drop_nlink(ZTOI(dzp));	/* ".." link from zp */
1073 	links = ZTOI(dzp)->i_nlink;
1074 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
1075 	    NULL, &links, sizeof (links));
1076 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1077 	    NULL, &dzp->z_size, sizeof (dzp->z_size));
1078 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
1079 	    NULL, ctime, sizeof (ctime));
1080 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
1081 	    NULL, mtime, sizeof (mtime));
1082 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1083 	    NULL, &dzp->z_pflags, sizeof (dzp->z_pflags));
1084 	zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime);
1085 	error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
1086 	ASSERT(error == 0);
1087 	mutex_exit(&dzp->z_lock);
1088 
1089 	if (unlinkedp != NULL)
1090 		*unlinkedp = unlinked;
1091 	else if (unlinked)
1092 		zfs_unlinked_add(zp, tx);
1093 
1094 	return (0);
1095 }
1096 
1097 /*
1098  * Indicate whether the directory is empty.  Works with or without z_lock
1099  * held, but can only be consider a hint in the latter case.  Returns true
1100  * if only "." and ".." remain and there's no work in progress.
1101  *
1102  * The internal ZAP size, rather than zp->z_size, needs to be checked since
1103  * some consumers (Lustre) do not strictly maintain an accurate SA_ZPL_SIZE.
1104  */
1105 boolean_t
zfs_dirempty(znode_t * dzp)1106 zfs_dirempty(znode_t *dzp)
1107 {
1108 	zfsvfs_t *zfsvfs = ZTOZSB(dzp);
1109 	uint64_t count;
1110 	int error;
1111 
1112 	if (dzp->z_dirlocks != NULL)
1113 		return (B_FALSE);
1114 
1115 	error = zap_count(zfsvfs->z_os, dzp->z_id, &count);
1116 	if (error != 0 || count != 0)
1117 		return (B_FALSE);
1118 
1119 	return (B_TRUE);
1120 }
1121 
1122 int
zfs_make_xattrdir(znode_t * zp,vattr_t * vap,znode_t ** xzpp,cred_t * cr)1123 zfs_make_xattrdir(znode_t *zp, vattr_t *vap, znode_t **xzpp, cred_t *cr)
1124 {
1125 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
1126 	znode_t *xzp;
1127 	dmu_tx_t *tx;
1128 	int error;
1129 	zfs_acl_ids_t acl_ids;
1130 	boolean_t fuid_dirtied;
1131 #ifdef ZFS_DEBUG
1132 	uint64_t parent;
1133 #endif
1134 
1135 	*xzpp = NULL;
1136 
1137 	if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
1138 	    &acl_ids, zfs_init_idmap)) != 0)
1139 		return (error);
1140 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zp->z_projid)) {
1141 		zfs_acl_ids_free(&acl_ids);
1142 		return (SET_ERROR(EDQUOT));
1143 	}
1144 
1145 	tx = dmu_tx_create(zfsvfs->z_os);
1146 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1147 	    ZFS_SA_BASE_ATTR_SIZE);
1148 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1149 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1150 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1151 	if (fuid_dirtied)
1152 		zfs_fuid_txhold(zfsvfs, tx);
1153 	error = dmu_tx_assign(tx, DMU_TX_WAIT);
1154 	if (error) {
1155 		zfs_acl_ids_free(&acl_ids);
1156 		dmu_tx_abort(tx);
1157 		return (error);
1158 	}
1159 	zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);
1160 
1161 	if (fuid_dirtied)
1162 		zfs_fuid_sync(zfsvfs, tx);
1163 
1164 #ifdef ZFS_DEBUG
1165 	error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
1166 	    &parent, sizeof (parent));
1167 	ASSERT(error == 0 && parent == zp->z_id);
1168 #endif
1169 
1170 	VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id,
1171 	    sizeof (xzp->z_id), tx));
1172 
1173 	if (!zp->z_unlinked)
1174 		zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp, xzp, "", NULL,
1175 		    acl_ids.z_fuidp, vap);
1176 
1177 	zfs_acl_ids_free(&acl_ids);
1178 	dmu_tx_commit(tx);
1179 
1180 	*xzpp = xzp;
1181 
1182 	return (0);
1183 }
1184 
1185 /*
1186  * Return a znode for the extended attribute directory for zp.
1187  * ** If the directory does not already exist, it is created **
1188  *
1189  *	IN:	zp	- znode to obtain attribute directory from
1190  *		cr	- credentials of caller
1191  *		flags	- flags from the VOP_LOOKUP call
1192  *
1193  *	OUT:	xipp	- pointer to extended attribute znode
1194  *
1195  *	RETURN:	0 on success
1196  *		error number on failure
1197  */
1198 int
zfs_get_xattrdir(znode_t * zp,znode_t ** xzpp,cred_t * cr,int flags)1199 zfs_get_xattrdir(znode_t *zp, znode_t **xzpp, cred_t *cr, int flags)
1200 {
1201 	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
1202 	znode_t		*xzp;
1203 	zfs_dirlock_t	*dl;
1204 	vattr_t		va;
1205 	int		error;
1206 top:
1207 	error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL);
1208 	if (error)
1209 		return (error);
1210 
1211 	if (xzp != NULL) {
1212 		*xzpp = xzp;
1213 		zfs_dirent_unlock(dl);
1214 		return (0);
1215 	}
1216 
1217 	if (!(flags & CREATE_XATTR_DIR)) {
1218 		zfs_dirent_unlock(dl);
1219 		return (SET_ERROR(ENOENT));
1220 	}
1221 
1222 	if (zfs_is_readonly(zfsvfs)) {
1223 		zfs_dirent_unlock(dl);
1224 		return (SET_ERROR(EROFS));
1225 	}
1226 
1227 	/*
1228 	 * The ability to 'create' files in an attribute
1229 	 * directory comes from the write_xattr permission on the base file.
1230 	 *
1231 	 * The ability to 'search' an attribute directory requires
1232 	 * read_xattr permission on the base file.
1233 	 *
1234 	 * Once in a directory the ability to read/write attributes
1235 	 * is controlled by the permissions on the attribute file.
1236 	 */
1237 	va.va_mask = ATTR_MODE | ATTR_UID | ATTR_GID;
1238 	va.va_mode = S_IFDIR | S_ISVTX | 0777;
1239 	zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
1240 
1241 	va.va_dentry = NULL;
1242 	error = zfs_make_xattrdir(zp, &va, xzpp, cr);
1243 	zfs_dirent_unlock(dl);
1244 
1245 	if (error == ERESTART) {
1246 		/* NB: we already did dmu_tx_wait() if necessary */
1247 		goto top;
1248 	}
1249 
1250 	return (error);
1251 }
1252 
1253 /*
1254  * Decide whether it is okay to remove within a sticky directory.
1255  *
1256  * In sticky directories, write access is not sufficient;
1257  * you can remove entries from a directory only if:
1258  *
1259  *	you own the directory,
1260  *	you own the entry,
1261  *	you have write access to the entry,
1262  *	or you are privileged (checked in secpolicy...).
1263  *
1264  * The function returns 0 if remove access is granted.
1265  */
1266 int
zfs_sticky_remove_access(znode_t * zdp,znode_t * zp,cred_t * cr)1267 zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr)
1268 {
1269 	uid_t		uid;
1270 	uid_t		downer;
1271 	uid_t		fowner;
1272 	zfsvfs_t	*zfsvfs = ZTOZSB(zdp);
1273 
1274 	if (zfsvfs->z_replay)
1275 		return (0);
1276 
1277 	if ((zdp->z_mode & S_ISVTX) == 0)
1278 		return (0);
1279 
1280 	downer = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(ZTOI(zdp)->i_uid),
1281 	    cr, ZFS_OWNER);
1282 	fowner = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(ZTOI(zp)->i_uid),
1283 	    cr, ZFS_OWNER);
1284 
1285 	if ((uid = crgetuid(cr)) == downer || uid == fowner ||
1286 	    zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr,
1287 	    zfs_init_idmap) == 0)
1288 		return (0);
1289 	else
1290 		return (secpolicy_vnode_remove(cr));
1291 }
1292