1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013, 2016 by Delphix. All rights reserved.
25 * Copyright 2017 Nexenta Systems, Inc.
26 * Copyright (c) 2015, Joyent, Inc.
27 */
28
29#include <sys/types.h>
30#include <sys/param.h>
31#include <sys/time.h>
32#include <sys/systm.h>
33#include <sys/sysmacros.h>
34#include <sys/resource.h>
35#include <sys/vfs.h>
36#include <sys/vnode.h>
37#include <sys/file.h>
38#include <sys/mode.h>
39#include <sys/kmem.h>
40#include <sys/uio.h>
41#include <sys/pathname.h>
42#include <sys/cmn_err.h>
43#include <sys/errno.h>
44#include <sys/stat.h>
45#include <sys/unistd.h>
46#include <sys/sunddi.h>
47#include <sys/random.h>
48#include <sys/policy.h>
49#include <sys/zfs_dir.h>
50#include <sys/zfs_acl.h>
51#include <sys/fs/zfs.h>
52#include "fs/fs_subr.h"
53#include <sys/zap.h>
54#include <sys/dmu.h>
55#include <sys/atomic.h>
56#include <sys/zfs_ctldir.h>
57#include <sys/zfs_fuid.h>
58#include <sys/sa.h>
59#include <sys/zfs_sa.h>
60#include <sys/dnlc.h>
61#include <sys/extdirent.h>
62
63/*
64 * zfs_match_find() is used by zfs_dirent_lock() to peform zap lookups
65 * of names after deciding which is the appropriate lookup interface.
66 */
67static int
68zfs_match_find(zfsvfs_t *zfsvfs, znode_t *dzp, char *name, matchtype_t mt,
69    boolean_t update, int *deflags, pathname_t *rpnp, uint64_t *zoid)
70{
71	int error;
72
73	if (zfsvfs->z_norm) {
74		boolean_t conflict = B_FALSE;
75		size_t bufsz = 0;
76		char *buf = NULL;
77
78		if (rpnp) {
79			buf = rpnp->pn_buf;
80			bufsz = rpnp->pn_bufsize;
81		}
82
83		/*
84		 * In the non-mixed case we only expect there would ever
85		 * be one match, but we need to use the normalizing lookup.
86		 */
87		error = zap_lookup_norm(zfsvfs->z_os, dzp->z_id, name, 8, 1,
88		    zoid, mt, buf, bufsz, &conflict);
89		if (!error && deflags)
90			*deflags = conflict ? ED_CASE_CONFLICT : 0;
91	} else {
92		error = zap_lookup(zfsvfs->z_os, dzp->z_id, name, 8, 1, zoid);
93	}
94	*zoid = ZFS_DIRENT_OBJ(*zoid);
95
96	if (error == ENOENT && update)
97		dnlc_update(ZTOV(dzp), name, DNLC_NO_VNODE);
98
99	return (error);
100}
101
102/*
103 * Lock a directory entry.  A dirlock on <dzp, name> protects that name
104 * in dzp's directory zap object.  As long as you hold a dirlock, you can
105 * assume two things: (1) dzp cannot be reaped, and (2) no other thread
106 * can change the zap entry for (i.e. link or unlink) this name.
107 *
108 * Input arguments:
109 *	dzp	- znode for directory
110 *	name	- name of entry to lock
111 *	flag	- ZNEW: if the entry already exists, fail with EEXIST.
112 *		  ZEXISTS: if the entry does not exist, fail with ENOENT.
113 *		  ZSHARED: allow concurrent access with other ZSHARED callers.
114 *		  ZXATTR: we want dzp's xattr directory
115 *		  ZCILOOK: On a mixed sensitivity file system,
116 *			   this lookup should be case-insensitive.
117 *		  ZCIEXACT: On a purely case-insensitive file system,
118 *			    this lookup should be case-sensitive.
119 *		  ZRENAMING: we are locking for renaming, force narrow locks
120 *		  ZHAVELOCK: Don't grab the z_name_lock for this call. The
121 *			     current thread already holds it.
122 *
123 * Output arguments:
124 *	zpp	- pointer to the znode for the entry (NULL if there isn't one)
125 *	dlpp	- pointer to the dirlock for this entry (NULL on error)
126 *      direntflags - (case-insensitive lookup only)
127 *		flags if multiple case-sensitive matches exist in directory
128 *      realpnp     - (case-insensitive lookup only)
129 *		actual name matched within the directory
130 *
131 * Return value: 0 on success or errno on failure.
132 *
133 * NOTE: Always checks for, and rejects, '.' and '..'.
134 * NOTE: For case-insensitive file systems we take wide locks (see below),
135 *	 but return znode pointers to a single match.
136 */
137int
138zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, znode_t **zpp,
139    int flag, int *direntflags, pathname_t *realpnp)
140{
141	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
142	zfs_dirlock_t	*dl;
143	boolean_t	update;
144	matchtype_t	mt = 0;
145	uint64_t	zoid;
146	vnode_t		*vp = NULL;
147	int		error = 0;
148	int		cmpflags;
149
150	*zpp = NULL;
151	*dlpp = NULL;
152
153	/*
154	 * Verify that we are not trying to lock '.', '..', or '.zfs'
155	 */
156	if (name[0] == '.' &&
157	    (name[1] == '\0' || (name[1] == '.' && name[2] == '\0')) ||
158	    zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0)
159		return (SET_ERROR(EEXIST));
160
161	/*
162	 * Case sensitivity and normalization preferences are set when
163	 * the file system is created.  These are stored in the
164	 * zfsvfs->z_case and zfsvfs->z_norm fields.  These choices
165	 * affect what vnodes can be cached in the DNLC, how we
166	 * perform zap lookups, and the "width" of our dirlocks.
167	 *
168	 * A normal dirlock locks a single name.  Note that with
169	 * normalization a name can be composed multiple ways, but
170	 * when normalized, these names all compare equal.  A wide
171	 * dirlock locks multiple names.  We need these when the file
172	 * system is supporting mixed-mode access.  It is sometimes
173	 * necessary to lock all case permutations of file name at
174	 * once so that simultaneous case-insensitive/case-sensitive
175	 * behaves as rationally as possible.
176	 */
177
178	/*
179	 * When matching we may need to normalize & change case according to
180	 * FS settings.
181	 *
182	 * Note that a normalized match is necessary for a case insensitive
183	 * filesystem when the lookup request is not exact because normalization
184	 * can fold case independent of normalizing code point sequences.
185	 *
186	 * See the table above zfs_dropname().
187	 */
188	if (zfsvfs->z_norm != 0) {
189		mt = MT_NORMALIZE;
190
191		/*
192		 * Determine if the match needs to honor the case specified in
193		 * lookup, and if so keep track of that so that during
194		 * normalization we don't fold case.
195		 */
196		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE &&
197		    (flag & ZCIEXACT)) ||
198		    (zfsvfs->z_case == ZFS_CASE_MIXED && !(flag & ZCILOOK))) {
199			mt |= MT_MATCH_CASE;
200		}
201	}
202
203	/*
204	 * Only look in or update the DNLC if we are looking for the
205	 * name on a file system that does not require normalization
206	 * or case folding.  We can also look there if we happen to be
207	 * on a non-normalizing, mixed sensitivity file system IF we
208	 * are looking for the exact name.
209	 *
210	 * Maybe can add TO-UPPERed version of name to dnlc in ci-only
211	 * case for performance improvement?
212	 */
213	update = !zfsvfs->z_norm ||
214	    (zfsvfs->z_case == ZFS_CASE_MIXED &&
215	    !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER) && !(flag & ZCILOOK));
216
217	/*
218	 * ZRENAMING indicates we are in a situation where we should
219	 * take narrow locks regardless of the file system's
220	 * preferences for normalizing and case folding.  This will
221	 * prevent us deadlocking trying to grab the same wide lock
222	 * twice if the two names happen to be case-insensitive
223	 * matches.
224	 */
225	if (flag & ZRENAMING)
226		cmpflags = 0;
227	else
228		cmpflags = zfsvfs->z_norm;
229
230	/*
231	 * Wait until there are no locks on this name.
232	 *
233	 * Don't grab the the lock if it is already held. However, cannot
234	 * have both ZSHARED and ZHAVELOCK together.
235	 */
236	ASSERT(!(flag & ZSHARED) || !(flag & ZHAVELOCK));
237	if (!(flag & ZHAVELOCK))
238		rw_enter(&dzp->z_name_lock, RW_READER);
239
240	mutex_enter(&dzp->z_lock);
241	for (;;) {
242		if (dzp->z_unlinked) {
243			mutex_exit(&dzp->z_lock);
244			if (!(flag & ZHAVELOCK))
245				rw_exit(&dzp->z_name_lock);
246			return (SET_ERROR(ENOENT));
247		}
248		for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next) {
249			if ((u8_strcmp(name, dl->dl_name, 0, cmpflags,
250			    U8_UNICODE_LATEST, &error) == 0) || error != 0)
251				break;
252		}
253		if (error != 0) {
254			mutex_exit(&dzp->z_lock);
255			if (!(flag & ZHAVELOCK))
256				rw_exit(&dzp->z_name_lock);
257			return (SET_ERROR(ENOENT));
258		}
259		if (dl == NULL)	{
260			/*
261			 * Allocate a new dirlock and add it to the list.
262			 */
263			dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP);
264			cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL);
265			dl->dl_name = name;
266			dl->dl_sharecnt = 0;
267			dl->dl_namelock = 0;
268			dl->dl_namesize = 0;
269			dl->dl_dzp = dzp;
270			dl->dl_next = dzp->z_dirlocks;
271			dzp->z_dirlocks = dl;
272			break;
273		}
274		if ((flag & ZSHARED) && dl->dl_sharecnt != 0)
275			break;
276		cv_wait(&dl->dl_cv, &dzp->z_lock);
277	}
278
279	/*
280	 * If the z_name_lock was NOT held for this dirlock record it.
281	 */
282	if (flag & ZHAVELOCK)
283		dl->dl_namelock = 1;
284
285	if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) {
286		/*
287		 * We're the second shared reference to dl.  Make a copy of
288		 * dl_name in case the first thread goes away before we do.
289		 * Note that we initialize the new name before storing its
290		 * pointer into dl_name, because the first thread may load
291		 * dl->dl_name at any time.  It'll either see the old value,
292		 * which belongs to it, or the new shared copy; either is OK.
293		 */
294		dl->dl_namesize = strlen(dl->dl_name) + 1;
295		name = kmem_alloc(dl->dl_namesize, KM_SLEEP);
296		bcopy(dl->dl_name, name, dl->dl_namesize);
297		dl->dl_name = name;
298	}
299
300	mutex_exit(&dzp->z_lock);
301
302	/*
303	 * We have a dirlock on the name.  (Note that it is the dirlock,
304	 * not the dzp's z_lock, that protects the name in the zap object.)
305	 * See if there's an object by this name; if so, put a hold on it.
306	 */
307	if (flag & ZXATTR) {
308		error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid,
309		    sizeof (zoid));
310		if (error == 0)
311			error = (zoid == 0 ? ENOENT : 0);
312	} else {
313		if (update)
314			vp = dnlc_lookup(ZTOV(dzp), name);
315		if (vp == DNLC_NO_VNODE) {
316			VN_RELE(vp);
317			error = SET_ERROR(ENOENT);
318		} else if (vp) {
319			if (flag & ZNEW) {
320				zfs_dirent_unlock(dl);
321				VN_RELE(vp);
322				return (SET_ERROR(EEXIST));
323			}
324			*dlpp = dl;
325			*zpp = VTOZ(vp);
326			return (0);
327		} else {
328			error = zfs_match_find(zfsvfs, dzp, name, mt,
329			    update, direntflags, realpnp, &zoid);
330		}
331	}
332	if (error) {
333		if (error != ENOENT || (flag & ZEXISTS)) {
334			zfs_dirent_unlock(dl);
335			return (error);
336		}
337	} else {
338		if (flag & ZNEW) {
339			zfs_dirent_unlock(dl);
340			return (SET_ERROR(EEXIST));
341		}
342		error = zfs_zget(zfsvfs, zoid, zpp);
343		if (error) {
344			zfs_dirent_unlock(dl);
345			return (error);
346		}
347		if (!(flag & ZXATTR) && update)
348			dnlc_update(ZTOV(dzp), name, ZTOV(*zpp));
349	}
350
351	*dlpp = dl;
352
353	return (0);
354}
355
356/*
357 * Unlock this directory entry and wake anyone who was waiting for it.
358 */
359void
360zfs_dirent_unlock(zfs_dirlock_t *dl)
361{
362	znode_t *dzp = dl->dl_dzp;
363	zfs_dirlock_t **prev_dl, *cur_dl;
364
365	mutex_enter(&dzp->z_lock);
366
367	if (!dl->dl_namelock)
368		rw_exit(&dzp->z_name_lock);
369
370	if (dl->dl_sharecnt > 1) {
371		dl->dl_sharecnt--;
372		mutex_exit(&dzp->z_lock);
373		return;
374	}
375	prev_dl = &dzp->z_dirlocks;
376	while ((cur_dl = *prev_dl) != dl)
377		prev_dl = &cur_dl->dl_next;
378	*prev_dl = dl->dl_next;
379	cv_broadcast(&dl->dl_cv);
380	mutex_exit(&dzp->z_lock);
381
382	if (dl->dl_namesize != 0)
383		kmem_free(dl->dl_name, dl->dl_namesize);
384	cv_destroy(&dl->dl_cv);
385	kmem_free(dl, sizeof (*dl));
386}
387
388/*
389 * Look up an entry in a directory.
390 *
391 * NOTE: '.' and '..' are handled as special cases because
392 *	no directory entries are actually stored for them.  If this is
393 *	the root of a filesystem, then '.zfs' is also treated as a
394 *	special pseudo-directory.
395 */
396int
397zfs_dirlook(znode_t *dzp, char *name, vnode_t **vpp, int flags,
398    int *deflg, pathname_t *rpnp)
399{
400	zfs_dirlock_t *dl;
401	znode_t *zp;
402	int error = 0;
403	uint64_t parent;
404
405	if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
406		*vpp = ZTOV(dzp);
407		VN_HOLD(*vpp);
408	} else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
409		zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
410
411		/*
412		 * If we are a snapshot mounted under .zfs, return
413		 * the vp for the snapshot directory.
414		 */
415		if ((error = sa_lookup(dzp->z_sa_hdl,
416		    SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
417			return (error);
418		if (parent == dzp->z_id && zfsvfs->z_parent != zfsvfs) {
419			error = zfsctl_root_lookup(zfsvfs->z_parent->z_ctldir,
420			    "snapshot", vpp, NULL, 0, NULL, kcred,
421			    NULL, NULL, NULL);
422			return (error);
423		}
424		rw_enter(&dzp->z_parent_lock, RW_READER);
425		error = zfs_zget(zfsvfs, parent, &zp);
426		if (error == 0)
427			*vpp = ZTOV(zp);
428		rw_exit(&dzp->z_parent_lock);
429	} else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
430		*vpp = zfsctl_root(dzp);
431	} else {
432		int zf;
433
434		zf = ZEXISTS | ZSHARED;
435		if (flags & FIGNORECASE)
436			zf |= ZCILOOK;
437
438		error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp);
439		if (error == 0) {
440			*vpp = ZTOV(zp);
441			zfs_dirent_unlock(dl);
442			dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
443		}
444		rpnp = NULL;
445	}
446
447	if ((flags & FIGNORECASE) && rpnp && !error)
448		(void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize);
449
450	return (error);
451}
452
453/*
454 * unlinked Set (formerly known as the "delete queue") Error Handling
455 *
456 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
457 * don't specify the name of the entry that we will be manipulating.  We
458 * also fib and say that we won't be adding any new entries to the
459 * unlinked set, even though we might (this is to lower the minimum file
460 * size that can be deleted in a full filesystem).  So on the small
461 * chance that the nlink list is using a fat zap (ie. has more than
462 * 2000 entries), we *may* not pre-read a block that's needed.
463 * Therefore it is remotely possible for some of the assertions
464 * regarding the unlinked set below to fail due to i/o error.  On a
465 * nondebug system, this will result in the space being leaked.
466 */
467void
468zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx)
469{
470	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
471
472	ASSERT(zp->z_unlinked);
473	ASSERT(zp->z_links == 0);
474
475	VERIFY3U(0, ==,
476	    zap_add_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
477}
478
479/*
480 * Clean up any znodes that had no links when we either crashed or
481 * (force) umounted the file system.
482 */
483static void
484zfs_unlinked_drain_task(void *arg)
485{
486	zfsvfs_t *zfsvfs = arg;
487	zap_cursor_t	zc;
488	zap_attribute_t zap;
489	dmu_object_info_t doi;
490	znode_t		*zp;
491	int		error;
492
493	ASSERT3B(zfsvfs->z_draining, ==, B_TRUE);
494
495	/*
496	 * Interate over the contents of the unlinked set.
497	 */
498	for (zap_cursor_init(&zc, zfsvfs->z_os, zfsvfs->z_unlinkedobj);
499	    zap_cursor_retrieve(&zc, &zap) == 0 && !zfsvfs->z_drain_cancel;
500	    zap_cursor_advance(&zc)) {
501
502		/*
503		 * See what kind of object we have in list
504		 */
505
506		error = dmu_object_info(zfsvfs->z_os,
507		    zap.za_first_integer, &doi);
508		if (error != 0)
509			continue;
510
511		ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) ||
512		    (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS));
513		/*
514		 * We need to re-mark these list entries for deletion,
515		 * so we pull them back into core and set zp->z_unlinked.
516		 */
517		error = zfs_zget(zfsvfs, zap.za_first_integer, &zp);
518
519		/*
520		 * We may pick up znodes that are already marked for deletion.
521		 * This could happen during the purge of an extended attribute
522		 * directory.  All we need to do is skip over them, since they
523		 * are already in the system marked z_unlinked.
524		 */
525		if (error != 0)
526			continue;
527
528		zp->z_unlinked = B_TRUE;
529
530		VN_RELE(ZTOV(zp));
531		ASSERT3B(zfsvfs->z_unmounted, ==, B_FALSE);
532	}
533	zap_cursor_fini(&zc);
534
535	zfsvfs->z_draining = B_FALSE;
536	zfsvfs->z_drain_task = TASKQID_INVALID;
537}
538
539/*
540 * Sets z_draining then tries to dispatch async unlinked drain.
541 * If that fails executes synchronous unlinked drain.
542 */
543void
544zfs_unlinked_drain(zfsvfs_t *zfsvfs)
545{
546	ASSERT3B(zfsvfs->z_unmounted, ==, B_FALSE);
547	ASSERT3B(zfsvfs->z_draining, ==, B_FALSE);
548
549	zfsvfs->z_draining = B_TRUE;
550	zfsvfs->z_drain_cancel = B_FALSE;
551
552	zfsvfs->z_drain_task = taskq_dispatch(
553	    dsl_pool_unlinked_drain_taskq(dmu_objset_pool(zfsvfs->z_os)),
554	    zfs_unlinked_drain_task, zfsvfs, TQ_SLEEP);
555	if (zfsvfs->z_drain_task == TASKQID_INVALID) {
556		zfs_dbgmsg("async zfs_unlinked_drain dispatch failed");
557		zfs_unlinked_drain_task(zfsvfs);
558	}
559}
560
561/*
562 * Wait for the unlinked drain taskq task to stop. This will interrupt the
563 * unlinked set processing if it is in progress.
564 */
565void
566zfs_unlinked_drain_stop_wait(zfsvfs_t *zfsvfs)
567{
568	ASSERT3B(zfsvfs->z_unmounted, ==, B_FALSE);
569
570	while (zfsvfs->z_draining) {
571		zfsvfs->z_drain_cancel = B_TRUE;
572		taskq_wait(dsl_pool_unlinked_drain_taskq(
573		    dmu_objset_pool(zfsvfs->z_os)));
574	}
575}
576
577/*
578 * Delete the entire contents of a directory.  Return a count
579 * of the number of entries that could not be deleted. If we encounter
580 * an error, return a count of at least one so that the directory stays
581 * in the unlinked set.
582 *
583 * NOTE: this function assumes that the directory is inactive,
584 *	so there is no need to lock its entries before deletion.
585 *	Also, it assumes the directory contents is *only* regular
586 *	files.
587 */
588static int
589zfs_purgedir(znode_t *dzp)
590{
591	zap_cursor_t	zc;
592	zap_attribute_t	zap;
593	znode_t		*xzp;
594	dmu_tx_t	*tx;
595	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
596	zfs_dirlock_t	dl;
597	int skipped = 0;
598	int error;
599
600	for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id);
601	    (error = zap_cursor_retrieve(&zc, &zap)) == 0;
602	    zap_cursor_advance(&zc)) {
603		error = zfs_zget(zfsvfs,
604		    ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp);
605		if (error) {
606			skipped += 1;
607			continue;
608		}
609
610		ASSERT((ZTOV(xzp)->v_type == VREG) ||
611		    (ZTOV(xzp)->v_type == VLNK));
612
613		tx = dmu_tx_create(zfsvfs->z_os);
614		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
615		dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name);
616		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
617		dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
618		/* Is this really needed ? */
619		zfs_sa_upgrade_txholds(tx, xzp);
620		dmu_tx_mark_netfree(tx);
621		error = dmu_tx_assign(tx, TXG_WAIT);
622		if (error) {
623			dmu_tx_abort(tx);
624			VN_RELE(ZTOV(xzp));
625			skipped += 1;
626			continue;
627		}
628		bzero(&dl, sizeof (dl));
629		dl.dl_dzp = dzp;
630		dl.dl_name = zap.za_name;
631
632		error = zfs_link_destroy(&dl, xzp, tx, 0, NULL);
633		if (error)
634			skipped += 1;
635		dmu_tx_commit(tx);
636
637		VN_RELE(ZTOV(xzp));
638	}
639	zap_cursor_fini(&zc);
640	if (error != ENOENT)
641		skipped += 1;
642	return (skipped);
643}
644
645void
646zfs_rmnode(znode_t *zp)
647{
648	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
649	objset_t	*os = zfsvfs->z_os;
650	znode_t		*xzp = NULL;
651	dmu_tx_t	*tx;
652	uint64_t	acl_obj;
653	uint64_t	xattr_obj;
654	int		error;
655
656	ASSERT(zp->z_links == 0);
657	ASSERT(ZTOV(zp)->v_count == 0);
658
659	/*
660	 * If this is an attribute directory, purge its contents.
661	 */
662	if (ZTOV(zp)->v_type == VDIR && (zp->z_pflags & ZFS_XATTR)) {
663		if (zfs_purgedir(zp) != 0) {
664			/*
665			 * Not enough space to delete some xattrs.
666			 * Leave it in the unlinked set.
667			 */
668			zfs_znode_dmu_fini(zp);
669			zfs_znode_free(zp);
670			return;
671		}
672	} else {
673		/*
674		 * Free up all the data in the file.  We don't do this for
675		 * XATTR directories because we need truncate and remove to be
676		 * in the same tx, like in zfs_znode_delete(). Otherwise, if
677		 * we crash here we'll end up with an inconsistent truncated
678		 * zap object in the delete queue.  Note a truncated file is
679		 * harmless since it only contains user data.
680		 */
681		error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
682		if (error) {
683			/*
684			 * Not enough space or we were interrupted by unmount.
685			 * Leave the file in the unlinked set.
686			 */
687			zfs_znode_dmu_fini(zp);
688			zfs_znode_free(zp);
689			return;
690		}
691	}
692
693	/*
694	 * If the file has extended attributes, we're going to unlink
695	 * the xattr dir.
696	 */
697	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
698	    &xattr_obj, sizeof (xattr_obj));
699	if (error == 0 && xattr_obj) {
700		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
701		ASSERT(error == 0);
702	}
703
704	acl_obj = zfs_external_acl(zp);
705
706	/*
707	 * Set up the final transaction.
708	 */
709	tx = dmu_tx_create(os);
710	dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
711	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
712	if (xzp) {
713		dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL);
714		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
715	}
716	if (acl_obj)
717		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
718
719	zfs_sa_upgrade_txholds(tx, zp);
720	error = dmu_tx_assign(tx, TXG_WAIT);
721	if (error) {
722		/*
723		 * Not enough space to delete the file.  Leave it in the
724		 * unlinked set, leaking it until the fs is remounted (at
725		 * which point we'll call zfs_unlinked_drain() to process it).
726		 */
727		dmu_tx_abort(tx);
728		zfs_znode_dmu_fini(zp);
729		zfs_znode_free(zp);
730		goto out;
731	}
732
733	if (xzp) {
734		ASSERT(error == 0);
735		mutex_enter(&xzp->z_lock);
736		xzp->z_unlinked = B_TRUE;	/* mark xzp for deletion */
737		xzp->z_links = 0;	/* no more links to it */
738		VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
739		    &xzp->z_links, sizeof (xzp->z_links), tx));
740		mutex_exit(&xzp->z_lock);
741		zfs_unlinked_add(xzp, tx);
742	}
743
744	/* Remove this znode from the unlinked set */
745	VERIFY3U(0, ==,
746	    zap_remove_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
747
748	zfs_znode_delete(zp, tx);
749
750	dmu_tx_commit(tx);
751out:
752	if (xzp)
753		VN_RELE(ZTOV(xzp));
754}
755
756static uint64_t
757zfs_dirent(znode_t *zp, uint64_t mode)
758{
759	uint64_t de = zp->z_id;
760
761	if (zp->z_zfsvfs->z_version >= ZPL_VERSION_DIRENT_TYPE)
762		de |= IFTODT(mode) << 60;
763	return (de);
764}
765
766/*
767 * Link zp into dl.  Can only fail if zp has been unlinked.
768 */
769int
770zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag)
771{
772	znode_t *dzp = dl->dl_dzp;
773	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
774	vnode_t *vp = ZTOV(zp);
775	uint64_t value;
776	int zp_is_dir = (vp->v_type == VDIR);
777	sa_bulk_attr_t bulk[5];
778	uint64_t mtime[2], ctime[2];
779	int count = 0;
780	int error;
781
782	mutex_enter(&zp->z_lock);
783
784	if (!(flag & ZRENAMING)) {
785		if (zp->z_unlinked) {	/* no new links to unlinked zp */
786			ASSERT(!(flag & (ZNEW | ZEXISTS)));
787			mutex_exit(&zp->z_lock);
788			return (SET_ERROR(ENOENT));
789		}
790		zp->z_links++;
791		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
792		    &zp->z_links, sizeof (zp->z_links));
793
794	}
795	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
796	    &dzp->z_id, sizeof (dzp->z_id));
797	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
798	    &zp->z_pflags, sizeof (zp->z_pflags));
799
800	if (!(flag & ZNEW)) {
801		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
802		    ctime, sizeof (ctime));
803		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
804		    ctime, B_TRUE);
805	}
806	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
807	ASSERT(error == 0);
808
809	mutex_exit(&zp->z_lock);
810
811	mutex_enter(&dzp->z_lock);
812	dzp->z_size++;
813	dzp->z_links += zp_is_dir;
814	count = 0;
815	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
816	    &dzp->z_size, sizeof (dzp->z_size));
817	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
818	    &dzp->z_links, sizeof (dzp->z_links));
819	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
820	    mtime, sizeof (mtime));
821	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
822	    ctime, sizeof (ctime));
823	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
824	    &dzp->z_pflags, sizeof (dzp->z_pflags));
825	zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
826	error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
827	ASSERT(error == 0);
828	mutex_exit(&dzp->z_lock);
829
830	value = zfs_dirent(zp, zp->z_mode);
831	error = zap_add(zp->z_zfsvfs->z_os, dzp->z_id, dl->dl_name,
832	    8, 1, &value, tx);
833	ASSERT(error == 0);
834
835	dnlc_update(ZTOV(dzp), dl->dl_name, vp);
836
837	return (0);
838}
839
840/*
841 * The match type in the code for this function should conform to:
842 *
843 * ------------------------------------------------------------------------
844 * fs type  | z_norm      | lookup type | match type
845 * ---------|-------------|-------------|----------------------------------
846 * CS !norm | 0           |           0 | 0 (exact)
847 * CS  norm | formX       |           0 | MT_NORMALIZE
848 * CI !norm | upper       |   !ZCIEXACT | MT_NORMALIZE
849 * CI !norm | upper       |    ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
850 * CI  norm | upper|formX |   !ZCIEXACT | MT_NORMALIZE
851 * CI  norm | upper|formX |    ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
852 * CM !norm | upper       |    !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
853 * CM !norm | upper       |     ZCILOOK | MT_NORMALIZE
854 * CM  norm | upper|formX |    !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
855 * CM  norm | upper|formX |     ZCILOOK | MT_NORMALIZE
856 *
857 * Abbreviations:
858 *    CS = Case Sensitive, CI = Case Insensitive, CM = Case Mixed
859 *    upper = case folding set by fs type on creation (U8_TEXTPREP_TOUPPER)
860 *    formX = unicode normalization form set on fs creation
861 */
862static int
863zfs_dropname(zfs_dirlock_t *dl, znode_t *zp, znode_t *dzp, dmu_tx_t *tx,
864    int flag)
865{
866	int error;
867
868	if (zp->z_zfsvfs->z_norm) {
869		matchtype_t mt = MT_NORMALIZE;
870
871		if ((zp->z_zfsvfs->z_case == ZFS_CASE_INSENSITIVE &&
872		    (flag & ZCIEXACT)) ||
873		    (zp->z_zfsvfs->z_case == ZFS_CASE_MIXED &&
874		    !(flag & ZCILOOK))) {
875			mt |= MT_MATCH_CASE;
876		}
877
878		error = zap_remove_norm(zp->z_zfsvfs->z_os, dzp->z_id,
879		    dl->dl_name, mt, tx);
880	} else {
881		error = zap_remove(zp->z_zfsvfs->z_os, dzp->z_id, dl->dl_name,
882		    tx);
883	}
884
885	return (error);
886}
887
888/*
889 * Unlink zp from dl, and mark zp for deletion if this was the last link.
890 * Can fail if zp is a mount point (EBUSY) or a non-empty directory (EEXIST).
891 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
892 * If it's non-NULL, we use it to indicate whether the znode needs deletion,
893 * and it's the caller's job to do it.
894 */
895int
896zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag,
897    boolean_t *unlinkedp)
898{
899	znode_t *dzp = dl->dl_dzp;
900	zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
901	vnode_t *vp = ZTOV(zp);
902	int zp_is_dir = (vp->v_type == VDIR);
903	boolean_t unlinked = B_FALSE;
904	sa_bulk_attr_t bulk[5];
905	uint64_t mtime[2], ctime[2];
906	int count = 0;
907	int error;
908
909	dnlc_remove(ZTOV(dzp), dl->dl_name);
910
911	if (!(flag & ZRENAMING)) {
912		if (vn_vfswlock(vp))		/* prevent new mounts on zp */
913			return (SET_ERROR(EBUSY));
914
915		if (vn_ismntpt(vp)) {		/* don't remove mount point */
916			vn_vfsunlock(vp);
917			return (SET_ERROR(EBUSY));
918		}
919
920		mutex_enter(&zp->z_lock);
921
922		if (zp_is_dir && !zfs_dirempty(zp)) {
923			mutex_exit(&zp->z_lock);
924			vn_vfsunlock(vp);
925			return (SET_ERROR(EEXIST));
926		}
927
928		/*
929		 * If we get here, we are going to try to remove the object.
930		 * First try removing the name from the directory; if that
931		 * fails, return the error.
932		 */
933		error = zfs_dropname(dl, zp, dzp, tx, flag);
934		if (error != 0) {
935			mutex_exit(&zp->z_lock);
936			vn_vfsunlock(vp);
937			return (error);
938		}
939
940		if (zp->z_links <= zp_is_dir) {
941			zfs_panic_recover("zfs: link count on %s is %u, "
942			    "should be at least %u",
943			    zp->z_vnode->v_path != vn_vpath_empty ?
944			    zp->z_vnode->v_path : "<unknown>",
945			    (int)zp->z_links, zp_is_dir + 1);
946			zp->z_links = zp_is_dir + 1;
947		}
948		if (--zp->z_links == zp_is_dir) {
949			zp->z_unlinked = B_TRUE;
950			zp->z_links = 0;
951			unlinked = B_TRUE;
952		} else {
953			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
954			    NULL, &ctime, sizeof (ctime));
955			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
956			    NULL, &zp->z_pflags, sizeof (zp->z_pflags));
957			zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
958			    B_TRUE);
959		}
960		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
961		    NULL, &zp->z_links, sizeof (zp->z_links));
962		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
963		count = 0;
964		ASSERT(error == 0);
965		mutex_exit(&zp->z_lock);
966		vn_vfsunlock(vp);
967	} else {
968		error = zfs_dropname(dl, zp, dzp, tx, flag);
969		if (error != 0)
970			return (error);
971	}
972
973	mutex_enter(&dzp->z_lock);
974	dzp->z_size--;		/* one dirent removed */
975	dzp->z_links -= zp_is_dir;	/* ".." link from zp */
976	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
977	    NULL, &dzp->z_links, sizeof (dzp->z_links));
978	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
979	    NULL, &dzp->z_size, sizeof (dzp->z_size));
980	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
981	    NULL, ctime, sizeof (ctime));
982	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
983	    NULL, mtime, sizeof (mtime));
984	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
985	    NULL, &dzp->z_pflags, sizeof (dzp->z_pflags));
986	zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
987	error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
988	ASSERT(error == 0);
989	mutex_exit(&dzp->z_lock);
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 * Indicate whether the directory is empty.  Works with or without z_lock
1001 * held, but can only be consider a hint in the latter case.  Returns true
1002 * if only "." and ".." remain and there's no work in progress.
1003 */
1004boolean_t
1005zfs_dirempty(znode_t *dzp)
1006{
1007	return (dzp->z_size == 2 && dzp->z_dirlocks == 0);
1008}
1009
1010int
1011zfs_make_xattrdir(znode_t *zp, vattr_t *vap, vnode_t **xvpp, cred_t *cr)
1012{
1013	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1014	znode_t *xzp;
1015	dmu_tx_t *tx;
1016	int error;
1017	zfs_acl_ids_t acl_ids;
1018	boolean_t fuid_dirtied;
1019	uint64_t parent;
1020
1021	*xvpp = NULL;
1022
1023	if (error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr))
1024		return (error);
1025
1026	if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
1027	    &acl_ids)) != 0)
1028		return (error);
1029	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zp->z_projid)) {
1030		zfs_acl_ids_free(&acl_ids);
1031		return (SET_ERROR(EDQUOT));
1032	}
1033
1034	tx = dmu_tx_create(zfsvfs->z_os);
1035	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1036	    ZFS_SA_BASE_ATTR_SIZE);
1037	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1038	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1039	fuid_dirtied = zfsvfs->z_fuid_dirty;
1040	if (fuid_dirtied)
1041		zfs_fuid_txhold(zfsvfs, tx);
1042	error = dmu_tx_assign(tx, TXG_WAIT);
1043	if (error) {
1044		zfs_acl_ids_free(&acl_ids);
1045		dmu_tx_abort(tx);
1046		return (error);
1047	}
1048	zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);
1049
1050	if (fuid_dirtied)
1051		zfs_fuid_sync(zfsvfs, tx);
1052
1053#ifdef DEBUG
1054	error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
1055	    &parent, sizeof (parent));
1056	ASSERT(error == 0 && parent == zp->z_id);
1057#endif
1058
1059	VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id,
1060	    sizeof (xzp->z_id), tx));
1061
1062	(void) zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp,
1063	    xzp, "", NULL, acl_ids.z_fuidp, vap);
1064
1065	zfs_acl_ids_free(&acl_ids);
1066	dmu_tx_commit(tx);
1067
1068	*xvpp = ZTOV(xzp);
1069
1070	return (0);
1071}
1072
1073/*
1074 * Return a znode for the extended attribute directory for zp.
1075 * ** If the directory does not already exist, it is created **
1076 *
1077 *	IN:	zp	- znode to obtain attribute directory from
1078 *		cr	- credentials of caller
1079 *		flags	- flags from the VOP_LOOKUP call
1080 *
1081 *	OUT:	xzpp	- pointer to extended attribute znode
1082 *
1083 *	RETURN:	0 on success
1084 *		error number on failure
1085 */
1086int
1087zfs_get_xattrdir(znode_t *zp, vnode_t **xvpp, cred_t *cr, int flags)
1088{
1089	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1090	znode_t		*xzp;
1091	zfs_dirlock_t	*dl;
1092	vattr_t		va;
1093	int		error;
1094top:
1095	error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL);
1096	if (error)
1097		return (error);
1098
1099	if (xzp != NULL) {
1100		*xvpp = ZTOV(xzp);
1101		zfs_dirent_unlock(dl);
1102		return (0);
1103	}
1104
1105
1106	if (!(flags & CREATE_XATTR_DIR)) {
1107		zfs_dirent_unlock(dl);
1108		return (SET_ERROR(ENOENT));
1109	}
1110
1111	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
1112		zfs_dirent_unlock(dl);
1113		return (SET_ERROR(EROFS));
1114	}
1115
1116	/*
1117	 * The ability to 'create' files in an attribute
1118	 * directory comes from the write_xattr permission on the base file.
1119	 *
1120	 * The ability to 'search' an attribute directory requires
1121	 * read_xattr permission on the base file.
1122	 *
1123	 * Once in a directory the ability to read/write attributes
1124	 * is controlled by the permissions on the attribute file.
1125	 */
1126	va.va_mask = AT_TYPE | AT_MODE | AT_UID | AT_GID;
1127	va.va_type = VDIR;
1128	va.va_mode = S_IFDIR | S_ISVTX | 0777;
1129	zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
1130
1131	error = zfs_make_xattrdir(zp, &va, xvpp, cr);
1132	zfs_dirent_unlock(dl);
1133
1134	if (error == ERESTART) {
1135		/* NB: we already did dmu_tx_wait() if necessary */
1136		goto top;
1137	}
1138
1139	return (error);
1140}
1141
1142/*
1143 * Decide whether it is okay to remove within a sticky directory.
1144 *
1145 * In sticky directories, write access is not sufficient;
1146 * you can remove entries from a directory only if:
1147 *
1148 *	you own the directory,
1149 *	you own the entry,
1150 *	the entry is a plain file and you have write access,
1151 *	or you are privileged (checked in secpolicy...).
1152 *
1153 * The function returns 0 if remove access is granted.
1154 */
1155int
1156zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr)
1157{
1158	uid_t		uid;
1159	uid_t		downer;
1160	uid_t		fowner;
1161	zfsvfs_t	*zfsvfs = zdp->z_zfsvfs;
1162
1163	if (zdp->z_zfsvfs->z_replay)
1164		return (0);
1165
1166	if ((zdp->z_mode & S_ISVTX) == 0)
1167		return (0);
1168
1169	downer = zfs_fuid_map_id(zfsvfs, zdp->z_uid, cr, ZFS_OWNER);
1170	fowner = zfs_fuid_map_id(zfsvfs, zp->z_uid, cr, ZFS_OWNER);
1171
1172	if ((uid = crgetuid(cr)) == downer || uid == fowner ||
1173	    (ZTOV(zp)->v_type == VREG &&
1174	    zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr) == 0))
1175		return (0);
1176	else
1177		return (secpolicy_vnode_remove(cr));
1178}
1179