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