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) 2012, 2015 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 * Copyright 2017 Nexenta Systems, Inc.
27 */
28
29 /* Portions Copyright 2007 Jeremy Teo */
30 /* Portions Copyright 2010 Robert Milkowski */
31
32 #include <sys/param.h>
33 #include <sys/time.h>
34 #include <sys/systm.h>
35 #include <sys/sysmacros.h>
36 #include <sys/resource.h>
37 #include <security/mac/mac_framework.h>
38 #include <sys/vfs.h>
39 #include <sys/endian.h>
40 #include <sys/vm.h>
41 #include <sys/vnode.h>
42 #if __FreeBSD_version >= 1300102
43 #include <sys/smr.h>
44 #endif
45 #include <sys/dirent.h>
46 #include <sys/file.h>
47 #include <sys/stat.h>
48 #include <sys/kmem.h>
49 #include <sys/taskq.h>
50 #include <sys/uio.h>
51 #include <sys/atomic.h>
52 #include <sys/namei.h>
53 #include <sys/mman.h>
54 #include <sys/cmn_err.h>
55 #include <sys/kdb.h>
56 #include <sys/sysproto.h>
57 #include <sys/errno.h>
58 #include <sys/unistd.h>
59 #include <sys/zfs_dir.h>
60 #include <sys/zfs_ioctl.h>
61 #include <sys/fs/zfs.h>
62 #include <sys/dmu.h>
63 #include <sys/dmu_objset.h>
64 #include <sys/spa.h>
65 #include <sys/txg.h>
66 #include <sys/dbuf.h>
67 #include <sys/zap.h>
68 #include <sys/sa.h>
69 #include <sys/policy.h>
70 #include <sys/sunddi.h>
71 #include <sys/filio.h>
72 #include <sys/sid.h>
73 #include <sys/zfs_ctldir.h>
74 #include <sys/zfs_fuid.h>
75 #include <sys/zfs_quota.h>
76 #include <sys/zfs_sa.h>
77 #include <sys/zfs_rlock.h>
78 #include <sys/bio.h>
79 #include <sys/buf.h>
80 #include <sys/sched.h>
81 #include <sys/acl.h>
82 #include <sys/vmmeter.h>
83 #include <vm/vm_param.h>
84 #include <sys/zil.h>
85 #include <sys/zfs_vnops.h>
86 #include <sys/module.h>
87 #include <sys/sysent.h>
88 #include <sys/dmu_impl.h>
89 #include <sys/brt.h>
90 #include <sys/zfeature.h>
91
92 #include <vm/vm_object.h>
93
94 #include <sys/extattr.h>
95 #include <sys/priv.h>
96
97 #ifndef VN_OPEN_INVFS
98 #define VN_OPEN_INVFS 0x0
99 #endif
100
101 VFS_SMR_DECLARE;
102
103 #if __FreeBSD_version < 1300103
104 #define NDFREE_PNBUF(ndp) NDFREE((ndp), NDF_ONLY_PNBUF)
105 #endif
106
107 #if __FreeBSD_version >= 1300047
108 #define vm_page_wire_lock(pp)
109 #define vm_page_wire_unlock(pp)
110 #else
111 #define vm_page_wire_lock(pp) vm_page_lock(pp)
112 #define vm_page_wire_unlock(pp) vm_page_unlock(pp)
113 #endif
114
115 #ifdef DEBUG_VFS_LOCKS
116 #define VNCHECKREF(vp) \
117 VNASSERT((vp)->v_holdcnt > 0 && (vp)->v_usecount > 0, vp, \
118 ("%s: wrong ref counts", __func__));
119 #else
120 #define VNCHECKREF(vp)
121 #endif
122
123 #if __FreeBSD_version >= 1400045
124 typedef uint64_t cookie_t;
125 #else
126 typedef ulong_t cookie_t;
127 #endif
128
129 /*
130 * Programming rules.
131 *
132 * Each vnode op performs some logical unit of work. To do this, the ZPL must
133 * properly lock its in-core state, create a DMU transaction, do the work,
134 * record this work in the intent log (ZIL), commit the DMU transaction,
135 * and wait for the intent log to commit if it is a synchronous operation.
136 * Moreover, the vnode ops must work in both normal and log replay context.
137 * The ordering of events is important to avoid deadlocks and references
138 * to freed memory. The example below illustrates the following Big Rules:
139 *
140 * (1) A check must be made in each zfs thread for a mounted file system.
141 * This is done avoiding races using zfs_enter(zfsvfs).
142 * A zfs_exit(zfsvfs) is needed before all returns. Any znodes
143 * must be checked with zfs_verify_zp(zp). Both of these macros
144 * can return EIO from the calling function.
145 *
146 * (2) VN_RELE() should always be the last thing except for zil_commit()
147 * (if necessary) and zfs_exit(). This is for 3 reasons:
148 * First, if it's the last reference, the vnode/znode
149 * can be freed, so the zp may point to freed memory. Second, the last
150 * reference will call zfs_zinactive(), which may induce a lot of work --
151 * pushing cached pages (which acquires range locks) and syncing out
152 * cached atime changes. Third, zfs_zinactive() may require a new tx,
153 * which could deadlock the system if you were already holding one.
154 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
155 *
156 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
157 * as they can span dmu_tx_assign() calls.
158 *
159 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
160 * dmu_tx_assign(). This is critical because we don't want to block
161 * while holding locks.
162 *
163 * If no ZPL locks are held (aside from zfs_enter()), use TXG_WAIT. This
164 * reduces lock contention and CPU usage when we must wait (note that if
165 * throughput is constrained by the storage, nearly every transaction
166 * must wait).
167 *
168 * Note, in particular, that if a lock is sometimes acquired before
169 * the tx assigns, and sometimes after (e.g. z_lock), then failing
170 * to use a non-blocking assign can deadlock the system. The scenario:
171 *
172 * Thread A has grabbed a lock before calling dmu_tx_assign().
173 * Thread B is in an already-assigned tx, and blocks for this lock.
174 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
175 * forever, because the previous txg can't quiesce until B's tx commits.
176 *
177 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
178 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
179 * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT,
180 * to indicate that this operation has already called dmu_tx_wait().
181 * This will ensure that we don't retry forever, waiting a short bit
182 * each time.
183 *
184 * (5) If the operation succeeded, generate the intent log entry for it
185 * before dropping locks. This ensures that the ordering of events
186 * in the intent log matches the order in which they actually occurred.
187 * During ZIL replay the zfs_log_* functions will update the sequence
188 * number to indicate the zil transaction has replayed.
189 *
190 * (6) At the end of each vnode op, the DMU tx must always commit,
191 * regardless of whether there were any errors.
192 *
193 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
194 * to ensure that synchronous semantics are provided when necessary.
195 *
196 * In general, this is how things should be ordered in each vnode op:
197 *
198 * zfs_enter(zfsvfs); // exit if unmounted
199 * top:
200 * zfs_dirent_lookup(&dl, ...) // lock directory entry (may VN_HOLD())
201 * rw_enter(...); // grab any other locks you need
202 * tx = dmu_tx_create(...); // get DMU tx
203 * dmu_tx_hold_*(); // hold each object you might modify
204 * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
205 * if (error) {
206 * rw_exit(...); // drop locks
207 * zfs_dirent_unlock(dl); // unlock directory entry
208 * VN_RELE(...); // release held vnodes
209 * if (error == ERESTART) {
210 * waited = B_TRUE;
211 * dmu_tx_wait(tx);
212 * dmu_tx_abort(tx);
213 * goto top;
214 * }
215 * dmu_tx_abort(tx); // abort DMU tx
216 * zfs_exit(zfsvfs); // finished in zfs
217 * return (error); // really out of space
218 * }
219 * error = do_real_work(); // do whatever this VOP does
220 * if (error == 0)
221 * zfs_log_*(...); // on success, make ZIL entry
222 * dmu_tx_commit(tx); // commit DMU tx -- error or not
223 * rw_exit(...); // drop locks
224 * zfs_dirent_unlock(dl); // unlock directory entry
225 * VN_RELE(...); // release held vnodes
226 * zil_commit(zilog, foid); // synchronous when necessary
227 * zfs_exit(zfsvfs); // finished in zfs
228 * return (error); // done, report error
229 */
230 static int
zfs_open(vnode_t ** vpp,int flag,cred_t * cr)231 zfs_open(vnode_t **vpp, int flag, cred_t *cr)
232 {
233 (void) cr;
234 znode_t *zp = VTOZ(*vpp);
235 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
236 int error;
237
238 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
239 return (error);
240
241 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
242 ((flag & FAPPEND) == 0)) {
243 zfs_exit(zfsvfs, FTAG);
244 return (SET_ERROR(EPERM));
245 }
246
247 /*
248 * Keep a count of the synchronous opens in the znode. On first
249 * synchronous open we must convert all previous async transactions
250 * into sync to keep correct ordering.
251 */
252 if (flag & O_SYNC) {
253 if (atomic_inc_32_nv(&zp->z_sync_cnt) == 1)
254 zil_async_to_sync(zfsvfs->z_log, zp->z_id);
255 }
256
257 zfs_exit(zfsvfs, FTAG);
258 return (0);
259 }
260
261 static int
zfs_close(vnode_t * vp,int flag,int count,offset_t offset,cred_t * cr)262 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr)
263 {
264 (void) offset, (void) cr;
265 znode_t *zp = VTOZ(vp);
266 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
267 int error;
268
269 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
270 return (error);
271
272 /* Decrement the synchronous opens in the znode */
273 if ((flag & O_SYNC) && (count == 1))
274 atomic_dec_32(&zp->z_sync_cnt);
275
276 zfs_exit(zfsvfs, FTAG);
277 return (0);
278 }
279
280 static int
zfs_ioctl(vnode_t * vp,ulong_t com,intptr_t data,int flag,cred_t * cred,int * rvalp)281 zfs_ioctl(vnode_t *vp, ulong_t com, intptr_t data, int flag, cred_t *cred,
282 int *rvalp)
283 {
284 (void) flag, (void) cred, (void) rvalp;
285 loff_t off;
286 int error;
287
288 switch (com) {
289 case _FIOFFS:
290 {
291 return (0);
292
293 /*
294 * The following two ioctls are used by bfu. Faking out,
295 * necessary to avoid bfu errors.
296 */
297 }
298 case _FIOGDIO:
299 case _FIOSDIO:
300 {
301 return (0);
302 }
303
304 case F_SEEK_DATA:
305 case F_SEEK_HOLE:
306 {
307 off = *(offset_t *)data;
308 /* offset parameter is in/out */
309 error = zfs_holey(VTOZ(vp), com, &off);
310 if (error)
311 return (error);
312 *(offset_t *)data = off;
313 return (0);
314 }
315 }
316 return (SET_ERROR(ENOTTY));
317 }
318
319 static vm_page_t
page_busy(vnode_t * vp,int64_t start,int64_t off,int64_t nbytes)320 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
321 {
322 vm_object_t obj;
323 vm_page_t pp;
324 int64_t end;
325
326 /*
327 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
328 * aligned boundaries, if the range is not aligned. As a result a
329 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
330 * It may happen that all DEV_BSIZE subranges are marked clean and thus
331 * the whole page would be considered clean despite have some
332 * dirty data.
333 * For this reason we should shrink the range to DEV_BSIZE aligned
334 * boundaries before calling vm_page_clear_dirty.
335 */
336 end = rounddown2(off + nbytes, DEV_BSIZE);
337 off = roundup2(off, DEV_BSIZE);
338 nbytes = end - off;
339
340 obj = vp->v_object;
341 zfs_vmobject_assert_wlocked_12(obj);
342 #if __FreeBSD_version < 1300050
343 for (;;) {
344 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
345 pp->valid) {
346 if (vm_page_xbusied(pp)) {
347 /*
348 * Reference the page before unlocking and
349 * sleeping so that the page daemon is less
350 * likely to reclaim it.
351 */
352 vm_page_reference(pp);
353 vm_page_lock(pp);
354 zfs_vmobject_wunlock(obj);
355 vm_page_busy_sleep(pp, "zfsmwb", true);
356 zfs_vmobject_wlock(obj);
357 continue;
358 }
359 vm_page_sbusy(pp);
360 } else if (pp != NULL) {
361 ASSERT(!pp->valid);
362 pp = NULL;
363 }
364 if (pp != NULL) {
365 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
366 vm_object_pip_add(obj, 1);
367 pmap_remove_write(pp);
368 if (nbytes != 0)
369 vm_page_clear_dirty(pp, off, nbytes);
370 }
371 break;
372 }
373 #else
374 vm_page_grab_valid_unlocked(&pp, obj, OFF_TO_IDX(start),
375 VM_ALLOC_NOCREAT | VM_ALLOC_SBUSY | VM_ALLOC_NORMAL |
376 VM_ALLOC_IGN_SBUSY);
377 if (pp != NULL) {
378 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
379 vm_object_pip_add(obj, 1);
380 pmap_remove_write(pp);
381 if (nbytes != 0)
382 vm_page_clear_dirty(pp, off, nbytes);
383 }
384 #endif
385 return (pp);
386 }
387
388 static void
page_unbusy(vm_page_t pp)389 page_unbusy(vm_page_t pp)
390 {
391
392 vm_page_sunbusy(pp);
393 #if __FreeBSD_version >= 1300041
394 vm_object_pip_wakeup(pp->object);
395 #else
396 vm_object_pip_subtract(pp->object, 1);
397 #endif
398 }
399
400 #if __FreeBSD_version > 1300051
401 static vm_page_t
page_hold(vnode_t * vp,int64_t start)402 page_hold(vnode_t *vp, int64_t start)
403 {
404 vm_object_t obj;
405 vm_page_t m;
406
407 obj = vp->v_object;
408 vm_page_grab_valid_unlocked(&m, obj, OFF_TO_IDX(start),
409 VM_ALLOC_NOCREAT | VM_ALLOC_WIRED | VM_ALLOC_IGN_SBUSY |
410 VM_ALLOC_NOBUSY);
411 return (m);
412 }
413 #else
414 static vm_page_t
page_hold(vnode_t * vp,int64_t start)415 page_hold(vnode_t *vp, int64_t start)
416 {
417 vm_object_t obj;
418 vm_page_t pp;
419
420 obj = vp->v_object;
421 zfs_vmobject_assert_wlocked(obj);
422
423 for (;;) {
424 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
425 pp->valid) {
426 if (vm_page_xbusied(pp)) {
427 /*
428 * Reference the page before unlocking and
429 * sleeping so that the page daemon is less
430 * likely to reclaim it.
431 */
432 vm_page_reference(pp);
433 vm_page_lock(pp);
434 zfs_vmobject_wunlock(obj);
435 vm_page_busy_sleep(pp, "zfsmwb", true);
436 zfs_vmobject_wlock(obj);
437 continue;
438 }
439
440 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
441 vm_page_wire_lock(pp);
442 vm_page_hold(pp);
443 vm_page_wire_unlock(pp);
444
445 } else
446 pp = NULL;
447 break;
448 }
449 return (pp);
450 }
451 #endif
452
453 static void
page_unhold(vm_page_t pp)454 page_unhold(vm_page_t pp)
455 {
456
457 vm_page_wire_lock(pp);
458 #if __FreeBSD_version >= 1300035
459 vm_page_unwire(pp, PQ_ACTIVE);
460 #else
461 vm_page_unhold(pp);
462 #endif
463 vm_page_wire_unlock(pp);
464 }
465
466 /*
467 * When a file is memory mapped, we must keep the IO data synchronized
468 * between the DMU cache and the memory mapped pages. What this means:
469 *
470 * On Write: If we find a memory mapped page, we write to *both*
471 * the page and the dmu buffer.
472 */
473 void
update_pages(znode_t * zp,int64_t start,int len,objset_t * os)474 update_pages(znode_t *zp, int64_t start, int len, objset_t *os)
475 {
476 vm_object_t obj;
477 struct sf_buf *sf;
478 vnode_t *vp = ZTOV(zp);
479 caddr_t va;
480 int off;
481
482 ASSERT3P(vp->v_mount, !=, NULL);
483 obj = vp->v_object;
484 ASSERT3P(obj, !=, NULL);
485
486 off = start & PAGEOFFSET;
487 zfs_vmobject_wlock_12(obj);
488 #if __FreeBSD_version >= 1300041
489 vm_object_pip_add(obj, 1);
490 #endif
491 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
492 vm_page_t pp;
493 int nbytes = imin(PAGESIZE - off, len);
494
495 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
496 zfs_vmobject_wunlock_12(obj);
497
498 va = zfs_map_page(pp, &sf);
499 (void) dmu_read(os, zp->z_id, start + off, nbytes,
500 va + off, DMU_READ_PREFETCH);
501 zfs_unmap_page(sf);
502
503 zfs_vmobject_wlock_12(obj);
504 page_unbusy(pp);
505 }
506 len -= nbytes;
507 off = 0;
508 }
509 #if __FreeBSD_version >= 1300041
510 vm_object_pip_wakeup(obj);
511 #else
512 vm_object_pip_wakeupn(obj, 0);
513 #endif
514 zfs_vmobject_wunlock_12(obj);
515 }
516
517 /*
518 * Read with UIO_NOCOPY flag means that sendfile(2) requests
519 * ZFS to populate a range of page cache pages with data.
520 *
521 * NOTE: this function could be optimized to pre-allocate
522 * all pages in advance, drain exclusive busy on all of them,
523 * map them into contiguous KVA region and populate them
524 * in one single dmu_read() call.
525 */
526 int
mappedread_sf(znode_t * zp,int nbytes,zfs_uio_t * uio)527 mappedread_sf(znode_t *zp, int nbytes, zfs_uio_t *uio)
528 {
529 vnode_t *vp = ZTOV(zp);
530 objset_t *os = zp->z_zfsvfs->z_os;
531 struct sf_buf *sf;
532 vm_object_t obj;
533 vm_page_t pp;
534 int64_t start;
535 caddr_t va;
536 int len = nbytes;
537 int error = 0;
538
539 ASSERT3U(zfs_uio_segflg(uio), ==, UIO_NOCOPY);
540 ASSERT3P(vp->v_mount, !=, NULL);
541 obj = vp->v_object;
542 ASSERT3P(obj, !=, NULL);
543 ASSERT0(zfs_uio_offset(uio) & PAGEOFFSET);
544
545 zfs_vmobject_wlock_12(obj);
546 for (start = zfs_uio_offset(uio); len > 0; start += PAGESIZE) {
547 int bytes = MIN(PAGESIZE, len);
548
549 pp = vm_page_grab_unlocked(obj, OFF_TO_IDX(start),
550 VM_ALLOC_SBUSY | VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
551 if (vm_page_none_valid(pp)) {
552 zfs_vmobject_wunlock_12(obj);
553 va = zfs_map_page(pp, &sf);
554 error = dmu_read(os, zp->z_id, start, bytes, va,
555 DMU_READ_PREFETCH);
556 if (bytes != PAGESIZE && error == 0)
557 memset(va + bytes, 0, PAGESIZE - bytes);
558 zfs_unmap_page(sf);
559 zfs_vmobject_wlock_12(obj);
560 #if __FreeBSD_version >= 1300081
561 if (error == 0) {
562 vm_page_valid(pp);
563 vm_page_activate(pp);
564 vm_page_do_sunbusy(pp);
565 } else {
566 zfs_vmobject_wlock(obj);
567 if (!vm_page_wired(pp) && pp->valid == 0 &&
568 vm_page_busy_tryupgrade(pp))
569 vm_page_free(pp);
570 else
571 vm_page_sunbusy(pp);
572 zfs_vmobject_wunlock(obj);
573 }
574 #else
575 vm_page_do_sunbusy(pp);
576 vm_page_lock(pp);
577 if (error) {
578 if (pp->wire_count == 0 && pp->valid == 0 &&
579 !vm_page_busied(pp))
580 vm_page_free(pp);
581 } else {
582 pp->valid = VM_PAGE_BITS_ALL;
583 vm_page_activate(pp);
584 }
585 vm_page_unlock(pp);
586 #endif
587 } else {
588 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
589 vm_page_do_sunbusy(pp);
590 }
591 if (error)
592 break;
593 zfs_uio_advance(uio, bytes);
594 len -= bytes;
595 }
596 zfs_vmobject_wunlock_12(obj);
597 return (error);
598 }
599
600 /*
601 * When a file is memory mapped, we must keep the IO data synchronized
602 * between the DMU cache and the memory mapped pages. What this means:
603 *
604 * On Read: We "read" preferentially from memory mapped pages,
605 * else we default from the dmu buffer.
606 *
607 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
608 * the file is memory mapped.
609 */
610 int
mappedread(znode_t * zp,int nbytes,zfs_uio_t * uio)611 mappedread(znode_t *zp, int nbytes, zfs_uio_t *uio)
612 {
613 vnode_t *vp = ZTOV(zp);
614 vm_object_t obj;
615 int64_t start;
616 int len = nbytes;
617 int off;
618 int error = 0;
619
620 ASSERT3P(vp->v_mount, !=, NULL);
621 obj = vp->v_object;
622 ASSERT3P(obj, !=, NULL);
623
624 start = zfs_uio_offset(uio);
625 off = start & PAGEOFFSET;
626 zfs_vmobject_wlock_12(obj);
627 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
628 vm_page_t pp;
629 uint64_t bytes = MIN(PAGESIZE - off, len);
630
631 if ((pp = page_hold(vp, start))) {
632 struct sf_buf *sf;
633 caddr_t va;
634
635 zfs_vmobject_wunlock_12(obj);
636 va = zfs_map_page(pp, &sf);
637 error = vn_io_fault_uiomove(va + off, bytes,
638 GET_UIO_STRUCT(uio));
639 zfs_unmap_page(sf);
640 zfs_vmobject_wlock_12(obj);
641 page_unhold(pp);
642 } else {
643 zfs_vmobject_wunlock_12(obj);
644 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
645 uio, bytes);
646 zfs_vmobject_wlock_12(obj);
647 }
648 len -= bytes;
649 off = 0;
650 if (error)
651 break;
652 }
653 zfs_vmobject_wunlock_12(obj);
654 return (error);
655 }
656
657 int
zfs_write_simple(znode_t * zp,const void * data,size_t len,loff_t pos,size_t * presid)658 zfs_write_simple(znode_t *zp, const void *data, size_t len,
659 loff_t pos, size_t *presid)
660 {
661 int error = 0;
662 ssize_t resid;
663
664 error = vn_rdwr(UIO_WRITE, ZTOV(zp), __DECONST(void *, data), len, pos,
665 UIO_SYSSPACE, IO_SYNC, kcred, NOCRED, &resid, curthread);
666
667 if (error) {
668 return (SET_ERROR(error));
669 } else if (presid == NULL) {
670 if (resid != 0) {
671 error = SET_ERROR(EIO);
672 }
673 } else {
674 *presid = resid;
675 }
676 return (error);
677 }
678
679 void
zfs_zrele_async(znode_t * zp)680 zfs_zrele_async(znode_t *zp)
681 {
682 vnode_t *vp = ZTOV(zp);
683 objset_t *os = ITOZSB(vp)->z_os;
684
685 VN_RELE_ASYNC(vp, dsl_pool_zrele_taskq(dmu_objset_pool(os)));
686 }
687
688 static int
zfs_dd_callback(struct mount * mp,void * arg,int lkflags,struct vnode ** vpp)689 zfs_dd_callback(struct mount *mp, void *arg, int lkflags, struct vnode **vpp)
690 {
691 int error;
692
693 *vpp = arg;
694 error = vn_lock(*vpp, lkflags);
695 if (error != 0)
696 vrele(*vpp);
697 return (error);
698 }
699
700 static int
zfs_lookup_lock(vnode_t * dvp,vnode_t * vp,const char * name,int lkflags)701 zfs_lookup_lock(vnode_t *dvp, vnode_t *vp, const char *name, int lkflags)
702 {
703 znode_t *zdp = VTOZ(dvp);
704 zfsvfs_t *zfsvfs __unused = zdp->z_zfsvfs;
705 int error;
706 int ltype;
707
708 if (zfsvfs->z_replay == B_FALSE)
709 ASSERT_VOP_LOCKED(dvp, __func__);
710
711 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
712 ASSERT3P(dvp, ==, vp);
713 vref(dvp);
714 ltype = lkflags & LK_TYPE_MASK;
715 if (ltype != VOP_ISLOCKED(dvp)) {
716 if (ltype == LK_EXCLUSIVE)
717 vn_lock(dvp, LK_UPGRADE | LK_RETRY);
718 else /* if (ltype == LK_SHARED) */
719 vn_lock(dvp, LK_DOWNGRADE | LK_RETRY);
720
721 /*
722 * Relock for the "." case could leave us with
723 * reclaimed vnode.
724 */
725 if (VN_IS_DOOMED(dvp)) {
726 vrele(dvp);
727 return (SET_ERROR(ENOENT));
728 }
729 }
730 return (0);
731 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
732 /*
733 * Note that in this case, dvp is the child vnode, and we
734 * are looking up the parent vnode - exactly reverse from
735 * normal operation. Unlocking dvp requires some rather
736 * tricky unlock/relock dance to prevent mp from being freed;
737 * use vn_vget_ino_gen() which takes care of all that.
738 *
739 * XXX Note that there is a time window when both vnodes are
740 * unlocked. It is possible, although highly unlikely, that
741 * during that window the parent-child relationship between
742 * the vnodes may change, for example, get reversed.
743 * In that case we would have a wrong lock order for the vnodes.
744 * All other filesystems seem to ignore this problem, so we
745 * do the same here.
746 * A potential solution could be implemented as follows:
747 * - using LK_NOWAIT when locking the second vnode and retrying
748 * if necessary
749 * - checking that the parent-child relationship still holds
750 * after locking both vnodes and retrying if it doesn't
751 */
752 error = vn_vget_ino_gen(dvp, zfs_dd_callback, vp, lkflags, &vp);
753 return (error);
754 } else {
755 error = vn_lock(vp, lkflags);
756 if (error != 0)
757 vrele(vp);
758 return (error);
759 }
760 }
761
762 /*
763 * Lookup an entry in a directory, or an extended attribute directory.
764 * If it exists, return a held vnode reference for it.
765 *
766 * IN: dvp - vnode of directory to search.
767 * nm - name of entry to lookup.
768 * pnp - full pathname to lookup [UNUSED].
769 * flags - LOOKUP_XATTR set if looking for an attribute.
770 * rdir - root directory vnode [UNUSED].
771 * cr - credentials of caller.
772 * ct - caller context
773 *
774 * OUT: vpp - vnode of located entry, NULL if not found.
775 *
776 * RETURN: 0 on success, error code on failure.
777 *
778 * Timestamps:
779 * NA
780 */
781 static int
zfs_lookup(vnode_t * dvp,const char * nm,vnode_t ** vpp,struct componentname * cnp,int nameiop,cred_t * cr,int flags,boolean_t cached)782 zfs_lookup(vnode_t *dvp, const char *nm, vnode_t **vpp,
783 struct componentname *cnp, int nameiop, cred_t *cr, int flags,
784 boolean_t cached)
785 {
786 znode_t *zdp = VTOZ(dvp);
787 znode_t *zp;
788 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
789 #if __FreeBSD_version > 1300124
790 seqc_t dvp_seqc;
791 #endif
792 int error = 0;
793
794 /*
795 * Fast path lookup, however we must skip DNLC lookup
796 * for case folding or normalizing lookups because the
797 * DNLC code only stores the passed in name. This means
798 * creating 'a' and removing 'A' on a case insensitive
799 * file system would work, but DNLC still thinks 'a'
800 * exists and won't let you create it again on the next
801 * pass through fast path.
802 */
803 if (!(flags & LOOKUP_XATTR)) {
804 if (dvp->v_type != VDIR) {
805 return (SET_ERROR(ENOTDIR));
806 } else if (zdp->z_sa_hdl == NULL) {
807 return (SET_ERROR(EIO));
808 }
809 }
810
811 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp,
812 const char *, nm);
813
814 if ((error = zfs_enter_verify_zp(zfsvfs, zdp, FTAG)) != 0)
815 return (error);
816
817 #if __FreeBSD_version > 1300124
818 dvp_seqc = vn_seqc_read_notmodify(dvp);
819 #endif
820
821 *vpp = NULL;
822
823 if (flags & LOOKUP_XATTR) {
824 /*
825 * If the xattr property is off, refuse the lookup request.
826 */
827 if (!(zfsvfs->z_flags & ZSB_XATTR)) {
828 zfs_exit(zfsvfs, FTAG);
829 return (SET_ERROR(EOPNOTSUPP));
830 }
831
832 /*
833 * We don't allow recursive attributes..
834 * Maybe someday we will.
835 */
836 if (zdp->z_pflags & ZFS_XATTR) {
837 zfs_exit(zfsvfs, FTAG);
838 return (SET_ERROR(EINVAL));
839 }
840
841 if ((error = zfs_get_xattrdir(VTOZ(dvp), &zp, cr, flags))) {
842 zfs_exit(zfsvfs, FTAG);
843 return (error);
844 }
845 *vpp = ZTOV(zp);
846
847 /*
848 * Do we have permission to get into attribute directory?
849 */
850 error = zfs_zaccess(zp, ACE_EXECUTE, 0, B_FALSE, cr, NULL);
851 if (error) {
852 vrele(ZTOV(zp));
853 }
854
855 zfs_exit(zfsvfs, FTAG);
856 return (error);
857 }
858
859 /*
860 * Check accessibility of directory if we're not coming in via
861 * VOP_CACHEDLOOKUP.
862 */
863 if (!cached) {
864 #ifdef NOEXECCHECK
865 if ((cnp->cn_flags & NOEXECCHECK) != 0) {
866 cnp->cn_flags &= ~NOEXECCHECK;
867 } else
868 #endif
869 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr,
870 NULL))) {
871 zfs_exit(zfsvfs, FTAG);
872 return (error);
873 }
874 }
875
876 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
877 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
878 zfs_exit(zfsvfs, FTAG);
879 return (SET_ERROR(EILSEQ));
880 }
881
882
883 /*
884 * First handle the special cases.
885 */
886 if ((cnp->cn_flags & ISDOTDOT) != 0) {
887 /*
888 * If we are a snapshot mounted under .zfs, return
889 * the vp for the snapshot directory.
890 */
891 if (zdp->z_id == zfsvfs->z_root && zfsvfs->z_parent != zfsvfs) {
892 struct componentname cn;
893 vnode_t *zfsctl_vp;
894 int ltype;
895
896 zfs_exit(zfsvfs, FTAG);
897 ltype = VOP_ISLOCKED(dvp);
898 VOP_UNLOCK1(dvp);
899 error = zfsctl_root(zfsvfs->z_parent, LK_SHARED,
900 &zfsctl_vp);
901 if (error == 0) {
902 cn.cn_nameptr = "snapshot";
903 cn.cn_namelen = strlen(cn.cn_nameptr);
904 cn.cn_nameiop = cnp->cn_nameiop;
905 cn.cn_flags = cnp->cn_flags & ~ISDOTDOT;
906 cn.cn_lkflags = cnp->cn_lkflags;
907 error = VOP_LOOKUP(zfsctl_vp, vpp, &cn);
908 vput(zfsctl_vp);
909 }
910 vn_lock(dvp, ltype | LK_RETRY);
911 return (error);
912 }
913 }
914 if (zfs_has_ctldir(zdp) && strcmp(nm, ZFS_CTLDIR_NAME) == 0) {
915 zfs_exit(zfsvfs, FTAG);
916 if ((cnp->cn_flags & ISLASTCN) != 0 && nameiop != LOOKUP)
917 return (SET_ERROR(ENOTSUP));
918 error = zfsctl_root(zfsvfs, cnp->cn_lkflags, vpp);
919 return (error);
920 }
921
922 /*
923 * The loop is retry the lookup if the parent-child relationship
924 * changes during the dot-dot locking complexities.
925 */
926 for (;;) {
927 uint64_t parent;
928
929 error = zfs_dirlook(zdp, nm, &zp);
930 if (error == 0)
931 *vpp = ZTOV(zp);
932
933 zfs_exit(zfsvfs, FTAG);
934 if (error != 0)
935 break;
936
937 error = zfs_lookup_lock(dvp, *vpp, nm, cnp->cn_lkflags);
938 if (error != 0) {
939 /*
940 * If we've got a locking error, then the vnode
941 * got reclaimed because of a force unmount.
942 * We never enter doomed vnodes into the name cache.
943 */
944 *vpp = NULL;
945 return (error);
946 }
947
948 if ((cnp->cn_flags & ISDOTDOT) == 0)
949 break;
950
951 if ((error = zfs_enter(zfsvfs, FTAG)) != 0) {
952 vput(ZTOV(zp));
953 *vpp = NULL;
954 return (error);
955 }
956 if (zdp->z_sa_hdl == NULL) {
957 error = SET_ERROR(EIO);
958 } else {
959 error = sa_lookup(zdp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
960 &parent, sizeof (parent));
961 }
962 if (error != 0) {
963 zfs_exit(zfsvfs, FTAG);
964 vput(ZTOV(zp));
965 break;
966 }
967 if (zp->z_id == parent) {
968 zfs_exit(zfsvfs, FTAG);
969 break;
970 }
971 vput(ZTOV(zp));
972 }
973
974 if (error != 0)
975 *vpp = NULL;
976
977 /* Translate errors and add SAVENAME when needed. */
978 if (cnp->cn_flags & ISLASTCN) {
979 switch (nameiop) {
980 case CREATE:
981 case RENAME:
982 if (error == ENOENT) {
983 error = EJUSTRETURN;
984 #if __FreeBSD_version < 1400068
985 cnp->cn_flags |= SAVENAME;
986 #endif
987 break;
988 }
989 zfs_fallthrough;
990 case DELETE:
991 #if __FreeBSD_version < 1400068
992 if (error == 0)
993 cnp->cn_flags |= SAVENAME;
994 #endif
995 break;
996 }
997 }
998
999 #if __FreeBSD_version > 1300124
1000 if ((cnp->cn_flags & ISDOTDOT) != 0) {
1001 /*
1002 * FIXME: zfs_lookup_lock relocks vnodes and does nothing to
1003 * handle races. In particular different callers may end up
1004 * with different vnodes and will try to add conflicting
1005 * entries to the namecache.
1006 *
1007 * While finding different result may be acceptable in face
1008 * of concurrent modification, adding conflicting entries
1009 * trips over an assert in the namecache.
1010 *
1011 * Ultimately let an entry through once everything settles.
1012 */
1013 if (!vn_seqc_consistent(dvp, dvp_seqc)) {
1014 cnp->cn_flags &= ~MAKEENTRY;
1015 }
1016 }
1017 #endif
1018
1019 /* Insert name into cache (as non-existent) if appropriate. */
1020 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay &&
1021 error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0)
1022 cache_enter(dvp, NULL, cnp);
1023
1024 /* Insert name into cache if appropriate. */
1025 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay &&
1026 error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1027 if (!(cnp->cn_flags & ISLASTCN) ||
1028 (nameiop != DELETE && nameiop != RENAME)) {
1029 cache_enter(dvp, *vpp, cnp);
1030 }
1031 }
1032
1033 return (error);
1034 }
1035
1036 /*
1037 * Attempt to create a new entry in a directory. If the entry
1038 * already exists, truncate the file if permissible, else return
1039 * an error. Return the vp of the created or trunc'd file.
1040 *
1041 * IN: dvp - vnode of directory to put new file entry in.
1042 * name - name of new file entry.
1043 * vap - attributes of new file.
1044 * excl - flag indicating exclusive or non-exclusive mode.
1045 * mode - mode to open file with.
1046 * cr - credentials of caller.
1047 * flag - large file flag [UNUSED].
1048 * ct - caller context
1049 * vsecp - ACL to be set
1050 * mnt_ns - Unused on FreeBSD
1051 *
1052 * OUT: vpp - vnode of created or trunc'd entry.
1053 *
1054 * RETURN: 0 on success, error code on failure.
1055 *
1056 * Timestamps:
1057 * dvp - ctime|mtime updated if new entry created
1058 * vp - ctime|mtime always, atime if new
1059 */
1060 int
zfs_create(znode_t * dzp,const char * name,vattr_t * vap,int excl,int mode,znode_t ** zpp,cred_t * cr,int flag,vsecattr_t * vsecp,zidmap_t * mnt_ns)1061 zfs_create(znode_t *dzp, const char *name, vattr_t *vap, int excl, int mode,
1062 znode_t **zpp, cred_t *cr, int flag, vsecattr_t *vsecp, zidmap_t *mnt_ns)
1063 {
1064 (void) excl, (void) mode, (void) flag;
1065 znode_t *zp;
1066 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1067 zilog_t *zilog;
1068 objset_t *os;
1069 dmu_tx_t *tx;
1070 int error;
1071 uid_t uid = crgetuid(cr);
1072 gid_t gid = crgetgid(cr);
1073 uint64_t projid = ZFS_DEFAULT_PROJID;
1074 zfs_acl_ids_t acl_ids;
1075 boolean_t fuid_dirtied;
1076 uint64_t txtype;
1077 #ifdef DEBUG_VFS_LOCKS
1078 vnode_t *dvp = ZTOV(dzp);
1079 #endif
1080
1081 /*
1082 * If we have an ephemeral id, ACL, or XVATTR then
1083 * make sure file system is at proper version
1084 */
1085 if (zfsvfs->z_use_fuids == B_FALSE &&
1086 (vsecp || (vap->va_mask & AT_XVATTR) ||
1087 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1088 return (SET_ERROR(EINVAL));
1089
1090 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0)
1091 return (error);
1092 os = zfsvfs->z_os;
1093 zilog = zfsvfs->z_log;
1094
1095 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1096 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1097 zfs_exit(zfsvfs, FTAG);
1098 return (SET_ERROR(EILSEQ));
1099 }
1100
1101 if (vap->va_mask & AT_XVATTR) {
1102 if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap,
1103 crgetuid(cr), cr, vap->va_type)) != 0) {
1104 zfs_exit(zfsvfs, FTAG);
1105 return (error);
1106 }
1107 }
1108
1109 *zpp = NULL;
1110
1111 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1112 vap->va_mode &= ~S_ISVTX;
1113
1114 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW);
1115 if (error) {
1116 zfs_exit(zfsvfs, FTAG);
1117 return (error);
1118 }
1119 ASSERT3P(zp, ==, NULL);
1120
1121 /*
1122 * Create a new file object and update the directory
1123 * to reference it.
1124 */
1125 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) {
1126 goto out;
1127 }
1128
1129 /*
1130 * We only support the creation of regular files in
1131 * extended attribute directories.
1132 */
1133
1134 if ((dzp->z_pflags & ZFS_XATTR) &&
1135 (vap->va_type != VREG)) {
1136 error = SET_ERROR(EINVAL);
1137 goto out;
1138 }
1139
1140 if ((error = zfs_acl_ids_create(dzp, 0, vap,
1141 cr, vsecp, &acl_ids, NULL)) != 0)
1142 goto out;
1143
1144 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode))
1145 projid = zfs_inherit_projid(dzp);
1146 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) {
1147 zfs_acl_ids_free(&acl_ids);
1148 error = SET_ERROR(EDQUOT);
1149 goto out;
1150 }
1151
1152 getnewvnode_reserve_();
1153
1154 tx = dmu_tx_create(os);
1155
1156 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1157 ZFS_SA_BASE_ATTR_SIZE);
1158
1159 fuid_dirtied = zfsvfs->z_fuid_dirty;
1160 if (fuid_dirtied)
1161 zfs_fuid_txhold(zfsvfs, tx);
1162 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1163 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1164 if (!zfsvfs->z_use_sa &&
1165 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1166 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1167 0, acl_ids.z_aclp->z_acl_bytes);
1168 }
1169 error = dmu_tx_assign(tx, TXG_WAIT);
1170 if (error) {
1171 zfs_acl_ids_free(&acl_ids);
1172 dmu_tx_abort(tx);
1173 getnewvnode_drop_reserve();
1174 zfs_exit(zfsvfs, FTAG);
1175 return (error);
1176 }
1177 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1178 if (fuid_dirtied)
1179 zfs_fuid_sync(zfsvfs, tx);
1180
1181 (void) zfs_link_create(dzp, name, zp, tx, ZNEW);
1182 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1183 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1184 vsecp, acl_ids.z_fuidp, vap);
1185 zfs_acl_ids_free(&acl_ids);
1186 dmu_tx_commit(tx);
1187
1188 getnewvnode_drop_reserve();
1189
1190 out:
1191 VNCHECKREF(dvp);
1192 if (error == 0) {
1193 *zpp = zp;
1194 }
1195
1196 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1197 zil_commit(zilog, 0);
1198
1199 zfs_exit(zfsvfs, FTAG);
1200 return (error);
1201 }
1202
1203 /*
1204 * Remove an entry from a directory.
1205 *
1206 * IN: dvp - vnode of directory to remove entry from.
1207 * name - name of entry to remove.
1208 * cr - credentials of caller.
1209 * ct - caller context
1210 * flags - case flags
1211 *
1212 * RETURN: 0 on success, error code on failure.
1213 *
1214 * Timestamps:
1215 * dvp - ctime|mtime
1216 * vp - ctime (if nlink > 0)
1217 */
1218 static int
zfs_remove_(vnode_t * dvp,vnode_t * vp,const char * name,cred_t * cr)1219 zfs_remove_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr)
1220 {
1221 znode_t *dzp = VTOZ(dvp);
1222 znode_t *zp;
1223 znode_t *xzp;
1224 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1225 zilog_t *zilog;
1226 uint64_t xattr_obj;
1227 uint64_t obj = 0;
1228 dmu_tx_t *tx;
1229 boolean_t unlinked;
1230 uint64_t txtype;
1231 int error;
1232
1233
1234 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0)
1235 return (error);
1236 zp = VTOZ(vp);
1237 if ((error = zfs_verify_zp(zp)) != 0) {
1238 zfs_exit(zfsvfs, FTAG);
1239 return (error);
1240 }
1241 zilog = zfsvfs->z_log;
1242
1243 xattr_obj = 0;
1244 xzp = NULL;
1245
1246 if ((error = zfs_zaccess_delete(dzp, zp, cr, NULL))) {
1247 goto out;
1248 }
1249
1250 /*
1251 * Need to use rmdir for removing directories.
1252 */
1253 if (vp->v_type == VDIR) {
1254 error = SET_ERROR(EPERM);
1255 goto out;
1256 }
1257
1258 vnevent_remove(vp, dvp, name, ct);
1259
1260 obj = zp->z_id;
1261
1262 /* are there any extended attributes? */
1263 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1264 &xattr_obj, sizeof (xattr_obj));
1265 if (error == 0 && xattr_obj) {
1266 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1267 ASSERT0(error);
1268 }
1269
1270 /*
1271 * We may delete the znode now, or we may put it in the unlinked set;
1272 * it depends on whether we're the last link, and on whether there are
1273 * other holds on the vnode. So we dmu_tx_hold() the right things to
1274 * allow for either case.
1275 */
1276 tx = dmu_tx_create(zfsvfs->z_os);
1277 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1278 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1279 zfs_sa_upgrade_txholds(tx, zp);
1280 zfs_sa_upgrade_txholds(tx, dzp);
1281
1282 if (xzp) {
1283 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1284 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1285 }
1286
1287 /* charge as an update -- would be nice not to charge at all */
1288 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1289
1290 /*
1291 * Mark this transaction as typically resulting in a net free of space
1292 */
1293 dmu_tx_mark_netfree(tx);
1294
1295 error = dmu_tx_assign(tx, TXG_WAIT);
1296 if (error) {
1297 dmu_tx_abort(tx);
1298 zfs_exit(zfsvfs, FTAG);
1299 return (error);
1300 }
1301
1302 /*
1303 * Remove the directory entry.
1304 */
1305 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, &unlinked);
1306
1307 if (error) {
1308 dmu_tx_commit(tx);
1309 goto out;
1310 }
1311
1312 if (unlinked) {
1313 zfs_unlinked_add(zp, tx);
1314 vp->v_vflag |= VV_NOSYNC;
1315 }
1316 /* XXX check changes to linux vnops */
1317 txtype = TX_REMOVE;
1318 zfs_log_remove(zilog, tx, txtype, dzp, name, obj, unlinked);
1319
1320 dmu_tx_commit(tx);
1321 out:
1322
1323 if (xzp)
1324 vrele(ZTOV(xzp));
1325
1326 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1327 zil_commit(zilog, 0);
1328
1329
1330 zfs_exit(zfsvfs, FTAG);
1331 return (error);
1332 }
1333
1334
1335 static int
zfs_lookup_internal(znode_t * dzp,const char * name,vnode_t ** vpp,struct componentname * cnp,int nameiop)1336 zfs_lookup_internal(znode_t *dzp, const char *name, vnode_t **vpp,
1337 struct componentname *cnp, int nameiop)
1338 {
1339 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1340 int error;
1341
1342 cnp->cn_nameptr = __DECONST(char *, name);
1343 cnp->cn_namelen = strlen(name);
1344 cnp->cn_nameiop = nameiop;
1345 cnp->cn_flags = ISLASTCN;
1346 #if __FreeBSD_version < 1400068
1347 cnp->cn_flags |= SAVENAME;
1348 #endif
1349 cnp->cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
1350 cnp->cn_cred = kcred;
1351 #if __FreeBSD_version < 1400037
1352 cnp->cn_thread = curthread;
1353 #endif
1354
1355 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay) {
1356 struct vop_lookup_args a;
1357
1358 a.a_gen.a_desc = &vop_lookup_desc;
1359 a.a_dvp = ZTOV(dzp);
1360 a.a_vpp = vpp;
1361 a.a_cnp = cnp;
1362 error = vfs_cache_lookup(&a);
1363 } else {
1364 error = zfs_lookup(ZTOV(dzp), name, vpp, cnp, nameiop, kcred, 0,
1365 B_FALSE);
1366 }
1367 #ifdef ZFS_DEBUG
1368 if (error) {
1369 printf("got error %d on name %s on op %d\n", error, name,
1370 nameiop);
1371 kdb_backtrace();
1372 }
1373 #endif
1374 return (error);
1375 }
1376
1377 int
zfs_remove(znode_t * dzp,const char * name,cred_t * cr,int flags)1378 zfs_remove(znode_t *dzp, const char *name, cred_t *cr, int flags)
1379 {
1380 vnode_t *vp;
1381 int error;
1382 struct componentname cn;
1383
1384 if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE)))
1385 return (error);
1386
1387 error = zfs_remove_(ZTOV(dzp), vp, name, cr);
1388 vput(vp);
1389 return (error);
1390 }
1391 /*
1392 * Create a new directory and insert it into dvp using the name
1393 * provided. Return a pointer to the inserted directory.
1394 *
1395 * IN: dvp - vnode of directory to add subdir to.
1396 * dirname - name of new directory.
1397 * vap - attributes of new directory.
1398 * cr - credentials of caller.
1399 * ct - caller context
1400 * flags - case flags
1401 * vsecp - ACL to be set
1402 * mnt_ns - Unused on FreeBSD
1403 *
1404 * OUT: vpp - vnode of created directory.
1405 *
1406 * RETURN: 0 on success, error code on failure.
1407 *
1408 * Timestamps:
1409 * dvp - ctime|mtime updated
1410 * vp - ctime|mtime|atime updated
1411 */
1412 int
zfs_mkdir(znode_t * dzp,const char * dirname,vattr_t * vap,znode_t ** zpp,cred_t * cr,int flags,vsecattr_t * vsecp,zidmap_t * mnt_ns)1413 zfs_mkdir(znode_t *dzp, const char *dirname, vattr_t *vap, znode_t **zpp,
1414 cred_t *cr, int flags, vsecattr_t *vsecp, zidmap_t *mnt_ns)
1415 {
1416 (void) flags, (void) vsecp;
1417 znode_t *zp;
1418 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1419 zilog_t *zilog;
1420 uint64_t txtype;
1421 dmu_tx_t *tx;
1422 int error;
1423 uid_t uid = crgetuid(cr);
1424 gid_t gid = crgetgid(cr);
1425 zfs_acl_ids_t acl_ids;
1426 boolean_t fuid_dirtied;
1427
1428 ASSERT3U(vap->va_type, ==, VDIR);
1429
1430 /*
1431 * If we have an ephemeral id, ACL, or XVATTR then
1432 * make sure file system is at proper version
1433 */
1434 if (zfsvfs->z_use_fuids == B_FALSE &&
1435 ((vap->va_mask & AT_XVATTR) ||
1436 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1437 return (SET_ERROR(EINVAL));
1438
1439 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0)
1440 return (error);
1441 zilog = zfsvfs->z_log;
1442
1443 if (dzp->z_pflags & ZFS_XATTR) {
1444 zfs_exit(zfsvfs, FTAG);
1445 return (SET_ERROR(EINVAL));
1446 }
1447
1448 if (zfsvfs->z_utf8 && u8_validate(dirname,
1449 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1450 zfs_exit(zfsvfs, FTAG);
1451 return (SET_ERROR(EILSEQ));
1452 }
1453
1454 if (vap->va_mask & AT_XVATTR) {
1455 if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap,
1456 crgetuid(cr), cr, vap->va_type)) != 0) {
1457 zfs_exit(zfsvfs, FTAG);
1458 return (error);
1459 }
1460 }
1461
1462 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1463 NULL, &acl_ids, NULL)) != 0) {
1464 zfs_exit(zfsvfs, FTAG);
1465 return (error);
1466 }
1467
1468 /*
1469 * First make sure the new directory doesn't exist.
1470 *
1471 * Existence is checked first to make sure we don't return
1472 * EACCES instead of EEXIST which can cause some applications
1473 * to fail.
1474 */
1475 *zpp = NULL;
1476
1477 if ((error = zfs_dirent_lookup(dzp, dirname, &zp, ZNEW))) {
1478 zfs_acl_ids_free(&acl_ids);
1479 zfs_exit(zfsvfs, FTAG);
1480 return (error);
1481 }
1482 ASSERT3P(zp, ==, NULL);
1483
1484 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr,
1485 mnt_ns))) {
1486 zfs_acl_ids_free(&acl_ids);
1487 zfs_exit(zfsvfs, FTAG);
1488 return (error);
1489 }
1490
1491 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zfs_inherit_projid(dzp))) {
1492 zfs_acl_ids_free(&acl_ids);
1493 zfs_exit(zfsvfs, FTAG);
1494 return (SET_ERROR(EDQUOT));
1495 }
1496
1497 /*
1498 * Add a new entry to the directory.
1499 */
1500 getnewvnode_reserve_();
1501 tx = dmu_tx_create(zfsvfs->z_os);
1502 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1503 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1504 fuid_dirtied = zfsvfs->z_fuid_dirty;
1505 if (fuid_dirtied)
1506 zfs_fuid_txhold(zfsvfs, tx);
1507 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1508 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1509 acl_ids.z_aclp->z_acl_bytes);
1510 }
1511
1512 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1513 ZFS_SA_BASE_ATTR_SIZE);
1514
1515 error = dmu_tx_assign(tx, TXG_WAIT);
1516 if (error) {
1517 zfs_acl_ids_free(&acl_ids);
1518 dmu_tx_abort(tx);
1519 getnewvnode_drop_reserve();
1520 zfs_exit(zfsvfs, FTAG);
1521 return (error);
1522 }
1523
1524 /*
1525 * Create new node.
1526 */
1527 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1528
1529 if (fuid_dirtied)
1530 zfs_fuid_sync(zfsvfs, tx);
1531
1532 /*
1533 * Now put new name in parent dir.
1534 */
1535 (void) zfs_link_create(dzp, dirname, zp, tx, ZNEW);
1536
1537 *zpp = zp;
1538
1539 txtype = zfs_log_create_txtype(Z_DIR, NULL, vap);
1540 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, NULL,
1541 acl_ids.z_fuidp, vap);
1542
1543 zfs_acl_ids_free(&acl_ids);
1544
1545 dmu_tx_commit(tx);
1546
1547 getnewvnode_drop_reserve();
1548
1549 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1550 zil_commit(zilog, 0);
1551
1552 zfs_exit(zfsvfs, FTAG);
1553 return (0);
1554 }
1555
1556 #if __FreeBSD_version < 1300124
1557 static void
cache_vop_rmdir(struct vnode * dvp,struct vnode * vp)1558 cache_vop_rmdir(struct vnode *dvp, struct vnode *vp)
1559 {
1560
1561 cache_purge(dvp);
1562 cache_purge(vp);
1563 }
1564 #endif
1565
1566 /*
1567 * Remove a directory subdir entry. If the current working
1568 * directory is the same as the subdir to be removed, the
1569 * remove will fail.
1570 *
1571 * IN: dvp - vnode of directory to remove from.
1572 * name - name of directory to be removed.
1573 * cwd - vnode of current working directory.
1574 * cr - credentials of caller.
1575 * ct - caller context
1576 * flags - case flags
1577 *
1578 * RETURN: 0 on success, error code on failure.
1579 *
1580 * Timestamps:
1581 * dvp - ctime|mtime updated
1582 */
1583 static int
zfs_rmdir_(vnode_t * dvp,vnode_t * vp,const char * name,cred_t * cr)1584 zfs_rmdir_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr)
1585 {
1586 znode_t *dzp = VTOZ(dvp);
1587 znode_t *zp = VTOZ(vp);
1588 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1589 zilog_t *zilog;
1590 dmu_tx_t *tx;
1591 int error;
1592
1593 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0)
1594 return (error);
1595 if ((error = zfs_verify_zp(zp)) != 0) {
1596 zfs_exit(zfsvfs, FTAG);
1597 return (error);
1598 }
1599 zilog = zfsvfs->z_log;
1600
1601
1602 if ((error = zfs_zaccess_delete(dzp, zp, cr, NULL))) {
1603 goto out;
1604 }
1605
1606 if (vp->v_type != VDIR) {
1607 error = SET_ERROR(ENOTDIR);
1608 goto out;
1609 }
1610
1611 vnevent_rmdir(vp, dvp, name, ct);
1612
1613 tx = dmu_tx_create(zfsvfs->z_os);
1614 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1615 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1616 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1617 zfs_sa_upgrade_txholds(tx, zp);
1618 zfs_sa_upgrade_txholds(tx, dzp);
1619 dmu_tx_mark_netfree(tx);
1620 error = dmu_tx_assign(tx, TXG_WAIT);
1621 if (error) {
1622 dmu_tx_abort(tx);
1623 zfs_exit(zfsvfs, FTAG);
1624 return (error);
1625 }
1626
1627 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, NULL);
1628
1629 if (error == 0) {
1630 uint64_t txtype = TX_RMDIR;
1631 zfs_log_remove(zilog, tx, txtype, dzp, name,
1632 ZFS_NO_OBJECT, B_FALSE);
1633 }
1634
1635 dmu_tx_commit(tx);
1636
1637 if (zfsvfs->z_use_namecache)
1638 cache_vop_rmdir(dvp, vp);
1639 out:
1640 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1641 zil_commit(zilog, 0);
1642
1643 zfs_exit(zfsvfs, FTAG);
1644 return (error);
1645 }
1646
1647 int
zfs_rmdir(znode_t * dzp,const char * name,znode_t * cwd,cred_t * cr,int flags)1648 zfs_rmdir(znode_t *dzp, const char *name, znode_t *cwd, cred_t *cr, int flags)
1649 {
1650 struct componentname cn;
1651 vnode_t *vp;
1652 int error;
1653
1654 if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE)))
1655 return (error);
1656
1657 error = zfs_rmdir_(ZTOV(dzp), vp, name, cr);
1658 vput(vp);
1659 return (error);
1660 }
1661
1662 /*
1663 * Read as many directory entries as will fit into the provided
1664 * buffer from the given directory cursor position (specified in
1665 * the uio structure).
1666 *
1667 * IN: vp - vnode of directory to read.
1668 * uio - structure supplying read location, range info,
1669 * and return buffer.
1670 * cr - credentials of caller.
1671 * ct - caller context
1672 *
1673 * OUT: uio - updated offset and range, buffer filled.
1674 * eofp - set to true if end-of-file detected.
1675 * ncookies- number of entries in cookies
1676 * cookies - offsets to directory entries
1677 *
1678 * RETURN: 0 on success, error code on failure.
1679 *
1680 * Timestamps:
1681 * vp - atime updated
1682 *
1683 * Note that the low 4 bits of the cookie returned by zap is always zero.
1684 * This allows us to use the low range for "special" directory entries:
1685 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
1686 * we use the offset 2 for the '.zfs' directory.
1687 */
1688 static int
zfs_readdir(vnode_t * vp,zfs_uio_t * uio,cred_t * cr,int * eofp,int * ncookies,cookie_t ** cookies)1689 zfs_readdir(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, int *eofp,
1690 int *ncookies, cookie_t **cookies)
1691 {
1692 znode_t *zp = VTOZ(vp);
1693 iovec_t *iovp;
1694 dirent64_t *odp;
1695 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1696 objset_t *os;
1697 caddr_t outbuf;
1698 size_t bufsize;
1699 zap_cursor_t zc;
1700 zap_attribute_t zap;
1701 uint_t bytes_wanted;
1702 uint64_t offset; /* must be unsigned; checks for < 1 */
1703 uint64_t parent;
1704 int local_eof;
1705 int outcount;
1706 int error;
1707 uint8_t prefetch;
1708 uint8_t type;
1709 int ncooks;
1710 cookie_t *cooks = NULL;
1711
1712 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
1713 return (error);
1714
1715 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
1716 &parent, sizeof (parent))) != 0) {
1717 zfs_exit(zfsvfs, FTAG);
1718 return (error);
1719 }
1720
1721 /*
1722 * If we are not given an eof variable,
1723 * use a local one.
1724 */
1725 if (eofp == NULL)
1726 eofp = &local_eof;
1727
1728 /*
1729 * Check for valid iov_len.
1730 */
1731 if (GET_UIO_STRUCT(uio)->uio_iov->iov_len <= 0) {
1732 zfs_exit(zfsvfs, FTAG);
1733 return (SET_ERROR(EINVAL));
1734 }
1735
1736 /*
1737 * Quit if directory has been removed (posix)
1738 */
1739 if ((*eofp = zp->z_unlinked) != 0) {
1740 zfs_exit(zfsvfs, FTAG);
1741 return (0);
1742 }
1743
1744 error = 0;
1745 os = zfsvfs->z_os;
1746 offset = zfs_uio_offset(uio);
1747 prefetch = zp->z_zn_prefetch;
1748
1749 /*
1750 * Initialize the iterator cursor.
1751 */
1752 if (offset <= 3) {
1753 /*
1754 * Start iteration from the beginning of the directory.
1755 */
1756 zap_cursor_init(&zc, os, zp->z_id);
1757 } else {
1758 /*
1759 * The offset is a serialized cursor.
1760 */
1761 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1762 }
1763
1764 /*
1765 * Get space to change directory entries into fs independent format.
1766 */
1767 iovp = GET_UIO_STRUCT(uio)->uio_iov;
1768 bytes_wanted = iovp->iov_len;
1769 if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1) {
1770 bufsize = bytes_wanted;
1771 outbuf = kmem_alloc(bufsize, KM_SLEEP);
1772 odp = (struct dirent64 *)outbuf;
1773 } else {
1774 bufsize = bytes_wanted;
1775 outbuf = NULL;
1776 odp = (struct dirent64 *)iovp->iov_base;
1777 }
1778
1779 if (ncookies != NULL) {
1780 /*
1781 * Minimum entry size is dirent size and 1 byte for a file name.
1782 */
1783 ncooks = zfs_uio_resid(uio) / (sizeof (struct dirent) -
1784 sizeof (((struct dirent *)NULL)->d_name) + 1);
1785 cooks = malloc(ncooks * sizeof (*cooks), M_TEMP, M_WAITOK);
1786 *cookies = cooks;
1787 *ncookies = ncooks;
1788 }
1789
1790 /*
1791 * Transform to file-system independent format
1792 */
1793 outcount = 0;
1794 while (outcount < bytes_wanted) {
1795 ino64_t objnum;
1796 ushort_t reclen;
1797 off64_t *next = NULL;
1798
1799 /*
1800 * Special case `.', `..', and `.zfs'.
1801 */
1802 if (offset == 0) {
1803 (void) strcpy(zap.za_name, ".");
1804 zap.za_normalization_conflict = 0;
1805 objnum = zp->z_id;
1806 type = DT_DIR;
1807 } else if (offset == 1) {
1808 (void) strcpy(zap.za_name, "..");
1809 zap.za_normalization_conflict = 0;
1810 objnum = parent;
1811 type = DT_DIR;
1812 } else if (offset == 2 && zfs_show_ctldir(zp)) {
1813 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
1814 zap.za_normalization_conflict = 0;
1815 objnum = ZFSCTL_INO_ROOT;
1816 type = DT_DIR;
1817 } else {
1818 /*
1819 * Grab next entry.
1820 */
1821 if ((error = zap_cursor_retrieve(&zc, &zap))) {
1822 if ((*eofp = (error == ENOENT)) != 0)
1823 break;
1824 else
1825 goto update;
1826 }
1827
1828 if (zap.za_integer_length != 8 ||
1829 zap.za_num_integers != 1) {
1830 cmn_err(CE_WARN, "zap_readdir: bad directory "
1831 "entry, obj = %lld, offset = %lld\n",
1832 (u_longlong_t)zp->z_id,
1833 (u_longlong_t)offset);
1834 error = SET_ERROR(ENXIO);
1835 goto update;
1836 }
1837
1838 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
1839 /*
1840 * MacOS X can extract the object type here such as:
1841 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
1842 */
1843 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
1844 }
1845
1846 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
1847
1848 /*
1849 * Will this entry fit in the buffer?
1850 */
1851 if (outcount + reclen > bufsize) {
1852 /*
1853 * Did we manage to fit anything in the buffer?
1854 */
1855 if (!outcount) {
1856 error = SET_ERROR(EINVAL);
1857 goto update;
1858 }
1859 break;
1860 }
1861 /*
1862 * Add normal entry:
1863 */
1864 odp->d_ino = objnum;
1865 odp->d_reclen = reclen;
1866 odp->d_namlen = strlen(zap.za_name);
1867 /* NOTE: d_off is the offset for the *next* entry. */
1868 next = &odp->d_off;
1869 strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
1870 odp->d_type = type;
1871 dirent_terminate(odp);
1872 odp = (dirent64_t *)((intptr_t)odp + reclen);
1873
1874 outcount += reclen;
1875
1876 ASSERT3S(outcount, <=, bufsize);
1877
1878 if (prefetch)
1879 dmu_prefetch_dnode(os, objnum, ZIO_PRIORITY_SYNC_READ);
1880
1881 /*
1882 * Move to the next entry, fill in the previous offset.
1883 */
1884 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
1885 zap_cursor_advance(&zc);
1886 offset = zap_cursor_serialize(&zc);
1887 } else {
1888 offset += 1;
1889 }
1890
1891 /* Fill the offset right after advancing the cursor. */
1892 if (next != NULL)
1893 *next = offset;
1894 if (cooks != NULL) {
1895 *cooks++ = offset;
1896 ncooks--;
1897 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
1898 }
1899 }
1900 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
1901
1902 /* Subtract unused cookies */
1903 if (ncookies != NULL)
1904 *ncookies -= ncooks;
1905
1906 if (zfs_uio_segflg(uio) == UIO_SYSSPACE && zfs_uio_iovcnt(uio) == 1) {
1907 iovp->iov_base += outcount;
1908 iovp->iov_len -= outcount;
1909 zfs_uio_resid(uio) -= outcount;
1910 } else if ((error =
1911 zfs_uiomove(outbuf, (long)outcount, UIO_READ, uio))) {
1912 /*
1913 * Reset the pointer.
1914 */
1915 offset = zfs_uio_offset(uio);
1916 }
1917
1918 update:
1919 zap_cursor_fini(&zc);
1920 if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1)
1921 kmem_free(outbuf, bufsize);
1922
1923 if (error == ENOENT)
1924 error = 0;
1925
1926 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
1927
1928 zfs_uio_setoffset(uio, offset);
1929 zfs_exit(zfsvfs, FTAG);
1930 if (error != 0 && cookies != NULL) {
1931 free(*cookies, M_TEMP);
1932 *cookies = NULL;
1933 *ncookies = 0;
1934 }
1935 return (error);
1936 }
1937
1938 /*
1939 * Get the requested file attributes and place them in the provided
1940 * vattr structure.
1941 *
1942 * IN: vp - vnode of file.
1943 * vap - va_mask identifies requested attributes.
1944 * If AT_XVATTR set, then optional attrs are requested
1945 * flags - ATTR_NOACLCHECK (CIFS server context)
1946 * cr - credentials of caller.
1947 *
1948 * OUT: vap - attribute values.
1949 *
1950 * RETURN: 0 (always succeeds).
1951 */
1952 static int
zfs_getattr(vnode_t * vp,vattr_t * vap,int flags,cred_t * cr)1953 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr)
1954 {
1955 znode_t *zp = VTOZ(vp);
1956 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1957 int error = 0;
1958 uint32_t blksize;
1959 u_longlong_t nblocks;
1960 uint64_t mtime[2], ctime[2], crtime[2], rdev;
1961 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
1962 xoptattr_t *xoap = NULL;
1963 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
1964 sa_bulk_attr_t bulk[4];
1965 int count = 0;
1966
1967 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
1968 return (error);
1969
1970 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
1971
1972 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
1973 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
1974 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
1975 if (vp->v_type == VBLK || vp->v_type == VCHR)
1976 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
1977 &rdev, 8);
1978
1979 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
1980 zfs_exit(zfsvfs, FTAG);
1981 return (error);
1982 }
1983
1984 /*
1985 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
1986 * Also, if we are the owner don't bother, since owner should
1987 * always be allowed to read basic attributes of file.
1988 */
1989 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
1990 (vap->va_uid != crgetuid(cr))) {
1991 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
1992 skipaclchk, cr, NULL))) {
1993 zfs_exit(zfsvfs, FTAG);
1994 return (error);
1995 }
1996 }
1997
1998 /*
1999 * Return all attributes. It's cheaper to provide the answer
2000 * than to determine whether we were asked the question.
2001 */
2002
2003 vap->va_type = IFTOVT(zp->z_mode);
2004 vap->va_mode = zp->z_mode & ~S_IFMT;
2005 vn_fsid(vp, vap);
2006 vap->va_nodeid = zp->z_id;
2007 vap->va_nlink = zp->z_links;
2008 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp) &&
2009 zp->z_links < ZFS_LINK_MAX)
2010 vap->va_nlink++;
2011 vap->va_size = zp->z_size;
2012 if (vp->v_type == VBLK || vp->v_type == VCHR)
2013 vap->va_rdev = zfs_cmpldev(rdev);
2014 else
2015 vap->va_rdev = 0;
2016 vap->va_gen = zp->z_gen;
2017 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2018 vap->va_filerev = zp->z_seq;
2019
2020 /*
2021 * Add in any requested optional attributes and the create time.
2022 * Also set the corresponding bits in the returned attribute bitmap.
2023 */
2024 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2025 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2026 xoap->xoa_archive =
2027 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2028 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2029 }
2030
2031 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2032 xoap->xoa_readonly =
2033 ((zp->z_pflags & ZFS_READONLY) != 0);
2034 XVA_SET_RTN(xvap, XAT_READONLY);
2035 }
2036
2037 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2038 xoap->xoa_system =
2039 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2040 XVA_SET_RTN(xvap, XAT_SYSTEM);
2041 }
2042
2043 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2044 xoap->xoa_hidden =
2045 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2046 XVA_SET_RTN(xvap, XAT_HIDDEN);
2047 }
2048
2049 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2050 xoap->xoa_nounlink =
2051 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2052 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2053 }
2054
2055 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2056 xoap->xoa_immutable =
2057 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2058 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2059 }
2060
2061 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2062 xoap->xoa_appendonly =
2063 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2064 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2065 }
2066
2067 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2068 xoap->xoa_nodump =
2069 ((zp->z_pflags & ZFS_NODUMP) != 0);
2070 XVA_SET_RTN(xvap, XAT_NODUMP);
2071 }
2072
2073 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2074 xoap->xoa_opaque =
2075 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2076 XVA_SET_RTN(xvap, XAT_OPAQUE);
2077 }
2078
2079 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2080 xoap->xoa_av_quarantined =
2081 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2082 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2083 }
2084
2085 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2086 xoap->xoa_av_modified =
2087 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2088 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2089 }
2090
2091 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2092 vp->v_type == VREG) {
2093 zfs_sa_get_scanstamp(zp, xvap);
2094 }
2095
2096 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2097 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2098 XVA_SET_RTN(xvap, XAT_REPARSE);
2099 }
2100 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2101 xoap->xoa_generation = zp->z_gen;
2102 XVA_SET_RTN(xvap, XAT_GEN);
2103 }
2104
2105 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2106 xoap->xoa_offline =
2107 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2108 XVA_SET_RTN(xvap, XAT_OFFLINE);
2109 }
2110
2111 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2112 xoap->xoa_sparse =
2113 ((zp->z_pflags & ZFS_SPARSE) != 0);
2114 XVA_SET_RTN(xvap, XAT_SPARSE);
2115 }
2116
2117 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
2118 xoap->xoa_projinherit =
2119 ((zp->z_pflags & ZFS_PROJINHERIT) != 0);
2120 XVA_SET_RTN(xvap, XAT_PROJINHERIT);
2121 }
2122
2123 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
2124 xoap->xoa_projid = zp->z_projid;
2125 XVA_SET_RTN(xvap, XAT_PROJID);
2126 }
2127 }
2128
2129 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2130 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2131 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2132 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2133
2134
2135 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2136 vap->va_blksize = blksize;
2137 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2138
2139 if (zp->z_blksz == 0) {
2140 /*
2141 * Block size hasn't been set; suggest maximal I/O transfers.
2142 */
2143 vap->va_blksize = zfsvfs->z_max_blksz;
2144 }
2145
2146 zfs_exit(zfsvfs, FTAG);
2147 return (0);
2148 }
2149
2150 /*
2151 * Set the file attributes to the values contained in the
2152 * vattr structure.
2153 *
2154 * IN: zp - znode of file to be modified.
2155 * vap - new attribute values.
2156 * If AT_XVATTR set, then optional attrs are being set
2157 * flags - ATTR_UTIME set if non-default time values provided.
2158 * - ATTR_NOACLCHECK (CIFS context only).
2159 * cr - credentials of caller.
2160 * mnt_ns - Unused on FreeBSD
2161 *
2162 * RETURN: 0 on success, error code on failure.
2163 *
2164 * Timestamps:
2165 * vp - ctime updated, mtime updated if size changed.
2166 */
2167 int
zfs_setattr(znode_t * zp,vattr_t * vap,int flags,cred_t * cr,zidmap_t * mnt_ns)2168 zfs_setattr(znode_t *zp, vattr_t *vap, int flags, cred_t *cr, zidmap_t *mnt_ns)
2169 {
2170 vnode_t *vp = ZTOV(zp);
2171 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2172 objset_t *os;
2173 zilog_t *zilog;
2174 dmu_tx_t *tx;
2175 vattr_t oldva;
2176 xvattr_t tmpxvattr;
2177 uint_t mask = vap->va_mask;
2178 uint_t saved_mask = 0;
2179 uint64_t saved_mode;
2180 int trim_mask = 0;
2181 uint64_t new_mode;
2182 uint64_t new_uid, new_gid;
2183 uint64_t xattr_obj;
2184 uint64_t mtime[2], ctime[2];
2185 uint64_t projid = ZFS_INVALID_PROJID;
2186 znode_t *attrzp;
2187 int need_policy = FALSE;
2188 int err, err2;
2189 zfs_fuid_info_t *fuidp = NULL;
2190 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2191 xoptattr_t *xoap;
2192 zfs_acl_t *aclp;
2193 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2194 boolean_t fuid_dirtied = B_FALSE;
2195 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2196 int count = 0, xattr_count = 0;
2197
2198 if (mask == 0)
2199 return (0);
2200
2201 if (mask & AT_NOSET)
2202 return (SET_ERROR(EINVAL));
2203
2204 if ((err = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
2205 return (err);
2206
2207 os = zfsvfs->z_os;
2208 zilog = zfsvfs->z_log;
2209
2210 /*
2211 * Make sure that if we have ephemeral uid/gid or xvattr specified
2212 * that file system is at proper version level
2213 */
2214
2215 if (zfsvfs->z_use_fuids == B_FALSE &&
2216 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2217 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2218 (mask & AT_XVATTR))) {
2219 zfs_exit(zfsvfs, FTAG);
2220 return (SET_ERROR(EINVAL));
2221 }
2222
2223 if (mask & AT_SIZE && vp->v_type == VDIR) {
2224 zfs_exit(zfsvfs, FTAG);
2225 return (SET_ERROR(EISDIR));
2226 }
2227
2228 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2229 zfs_exit(zfsvfs, FTAG);
2230 return (SET_ERROR(EINVAL));
2231 }
2232
2233 /*
2234 * If this is an xvattr_t, then get a pointer to the structure of
2235 * optional attributes. If this is NULL, then we have a vattr_t.
2236 */
2237 xoap = xva_getxoptattr(xvap);
2238
2239 xva_init(&tmpxvattr);
2240
2241 /*
2242 * Immutable files can only alter immutable bit and atime
2243 */
2244 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2245 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2246 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2247 zfs_exit(zfsvfs, FTAG);
2248 return (SET_ERROR(EPERM));
2249 }
2250
2251 /*
2252 * Note: ZFS_READONLY is handled in zfs_zaccess_common.
2253 */
2254
2255 /*
2256 * Verify timestamps doesn't overflow 32 bits.
2257 * ZFS can handle large timestamps, but 32bit syscalls can't
2258 * handle times greater than 2039. This check should be removed
2259 * once large timestamps are fully supported.
2260 */
2261 if (mask & (AT_ATIME | AT_MTIME)) {
2262 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2263 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2264 zfs_exit(zfsvfs, FTAG);
2265 return (SET_ERROR(EOVERFLOW));
2266 }
2267 }
2268 if (xoap != NULL && (mask & AT_XVATTR)) {
2269 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME) &&
2270 TIMESPEC_OVERFLOW(&vap->va_birthtime)) {
2271 zfs_exit(zfsvfs, FTAG);
2272 return (SET_ERROR(EOVERFLOW));
2273 }
2274
2275 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
2276 if (!dmu_objset_projectquota_enabled(os) ||
2277 (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode))) {
2278 zfs_exit(zfsvfs, FTAG);
2279 return (SET_ERROR(EOPNOTSUPP));
2280 }
2281
2282 projid = xoap->xoa_projid;
2283 if (unlikely(projid == ZFS_INVALID_PROJID)) {
2284 zfs_exit(zfsvfs, FTAG);
2285 return (SET_ERROR(EINVAL));
2286 }
2287
2288 if (projid == zp->z_projid && zp->z_pflags & ZFS_PROJID)
2289 projid = ZFS_INVALID_PROJID;
2290 else
2291 need_policy = TRUE;
2292 }
2293
2294 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT) &&
2295 (xoap->xoa_projinherit !=
2296 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) &&
2297 (!dmu_objset_projectquota_enabled(os) ||
2298 (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode)))) {
2299 zfs_exit(zfsvfs, FTAG);
2300 return (SET_ERROR(EOPNOTSUPP));
2301 }
2302 }
2303
2304 attrzp = NULL;
2305 aclp = NULL;
2306
2307 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2308 zfs_exit(zfsvfs, FTAG);
2309 return (SET_ERROR(EROFS));
2310 }
2311
2312 /*
2313 * First validate permissions
2314 */
2315
2316 if (mask & AT_SIZE) {
2317 /*
2318 * XXX - Note, we are not providing any open
2319 * mode flags here (like FNDELAY), so we may
2320 * block if there are locks present... this
2321 * should be addressed in openat().
2322 */
2323 /* XXX - would it be OK to generate a log record here? */
2324 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2325 if (err) {
2326 zfs_exit(zfsvfs, FTAG);
2327 return (err);
2328 }
2329 }
2330
2331 if (mask & (AT_ATIME|AT_MTIME) ||
2332 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2333 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2334 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2335 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2336 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2337 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2338 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2339 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2340 skipaclchk, cr, mnt_ns);
2341 }
2342
2343 if (mask & (AT_UID|AT_GID)) {
2344 int idmask = (mask & (AT_UID|AT_GID));
2345 int take_owner;
2346 int take_group;
2347
2348 /*
2349 * NOTE: even if a new mode is being set,
2350 * we may clear S_ISUID/S_ISGID bits.
2351 */
2352
2353 if (!(mask & AT_MODE))
2354 vap->va_mode = zp->z_mode;
2355
2356 /*
2357 * Take ownership or chgrp to group we are a member of
2358 */
2359
2360 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2361 take_group = (mask & AT_GID) &&
2362 zfs_groupmember(zfsvfs, vap->va_gid, cr);
2363
2364 /*
2365 * If both AT_UID and AT_GID are set then take_owner and
2366 * take_group must both be set in order to allow taking
2367 * ownership.
2368 *
2369 * Otherwise, send the check through secpolicy_vnode_setattr()
2370 *
2371 */
2372
2373 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2374 ((idmask == AT_UID) && take_owner) ||
2375 ((idmask == AT_GID) && take_group)) {
2376 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2377 skipaclchk, cr, mnt_ns) == 0) {
2378 /*
2379 * Remove setuid/setgid for non-privileged users
2380 */
2381 secpolicy_setid_clear(vap, vp, cr);
2382 trim_mask = (mask & (AT_UID|AT_GID));
2383 } else {
2384 need_policy = TRUE;
2385 }
2386 } else {
2387 need_policy = TRUE;
2388 }
2389 }
2390
2391 oldva.va_mode = zp->z_mode;
2392 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2393 if (mask & AT_XVATTR) {
2394 /*
2395 * Update xvattr mask to include only those attributes
2396 * that are actually changing.
2397 *
2398 * the bits will be restored prior to actually setting
2399 * the attributes so the caller thinks they were set.
2400 */
2401 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2402 if (xoap->xoa_appendonly !=
2403 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2404 need_policy = TRUE;
2405 } else {
2406 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2407 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2408 }
2409 }
2410
2411 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
2412 if (xoap->xoa_projinherit !=
2413 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) {
2414 need_policy = TRUE;
2415 } else {
2416 XVA_CLR_REQ(xvap, XAT_PROJINHERIT);
2417 XVA_SET_REQ(&tmpxvattr, XAT_PROJINHERIT);
2418 }
2419 }
2420
2421 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2422 if (xoap->xoa_nounlink !=
2423 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2424 need_policy = TRUE;
2425 } else {
2426 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2427 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2428 }
2429 }
2430
2431 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2432 if (xoap->xoa_immutable !=
2433 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2434 need_policy = TRUE;
2435 } else {
2436 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2437 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2438 }
2439 }
2440
2441 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2442 if (xoap->xoa_nodump !=
2443 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2444 need_policy = TRUE;
2445 } else {
2446 XVA_CLR_REQ(xvap, XAT_NODUMP);
2447 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2448 }
2449 }
2450
2451 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2452 if (xoap->xoa_av_modified !=
2453 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2454 need_policy = TRUE;
2455 } else {
2456 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2457 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2458 }
2459 }
2460
2461 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2462 if ((vp->v_type != VREG &&
2463 xoap->xoa_av_quarantined) ||
2464 xoap->xoa_av_quarantined !=
2465 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2466 need_policy = TRUE;
2467 } else {
2468 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2469 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2470 }
2471 }
2472
2473 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2474 zfs_exit(zfsvfs, FTAG);
2475 return (SET_ERROR(EPERM));
2476 }
2477
2478 if (need_policy == FALSE &&
2479 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2480 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2481 need_policy = TRUE;
2482 }
2483 }
2484
2485 if (mask & AT_MODE) {
2486 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr,
2487 mnt_ns) == 0) {
2488 err = secpolicy_setid_setsticky_clear(vp, vap,
2489 &oldva, cr);
2490 if (err) {
2491 zfs_exit(zfsvfs, FTAG);
2492 return (err);
2493 }
2494 trim_mask |= AT_MODE;
2495 } else {
2496 need_policy = TRUE;
2497 }
2498 }
2499
2500 if (need_policy) {
2501 /*
2502 * If trim_mask is set then take ownership
2503 * has been granted or write_acl is present and user
2504 * has the ability to modify mode. In that case remove
2505 * UID|GID and or MODE from mask so that
2506 * secpolicy_vnode_setattr() doesn't revoke it.
2507 */
2508
2509 if (trim_mask) {
2510 saved_mask = vap->va_mask;
2511 vap->va_mask &= ~trim_mask;
2512 if (trim_mask & AT_MODE) {
2513 /*
2514 * Save the mode, as secpolicy_vnode_setattr()
2515 * will overwrite it with ova.va_mode.
2516 */
2517 saved_mode = vap->va_mode;
2518 }
2519 }
2520 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2521 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2522 if (err) {
2523 zfs_exit(zfsvfs, FTAG);
2524 return (err);
2525 }
2526
2527 if (trim_mask) {
2528 vap->va_mask |= saved_mask;
2529 if (trim_mask & AT_MODE) {
2530 /*
2531 * Recover the mode after
2532 * secpolicy_vnode_setattr().
2533 */
2534 vap->va_mode = saved_mode;
2535 }
2536 }
2537 }
2538
2539 /*
2540 * secpolicy_vnode_setattr, or take ownership may have
2541 * changed va_mask
2542 */
2543 mask = vap->va_mask;
2544
2545 if ((mask & (AT_UID | AT_GID)) || projid != ZFS_INVALID_PROJID) {
2546 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2547 &xattr_obj, sizeof (xattr_obj));
2548
2549 if (err == 0 && xattr_obj) {
2550 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
2551 if (err == 0) {
2552 err = vn_lock(ZTOV(attrzp), LK_EXCLUSIVE);
2553 if (err != 0)
2554 vrele(ZTOV(attrzp));
2555 }
2556 if (err)
2557 goto out2;
2558 }
2559 if (mask & AT_UID) {
2560 new_uid = zfs_fuid_create(zfsvfs,
2561 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2562 if (new_uid != zp->z_uid &&
2563 zfs_id_overquota(zfsvfs, DMU_USERUSED_OBJECT,
2564 new_uid)) {
2565 if (attrzp)
2566 vput(ZTOV(attrzp));
2567 err = SET_ERROR(EDQUOT);
2568 goto out2;
2569 }
2570 }
2571
2572 if (mask & AT_GID) {
2573 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
2574 cr, ZFS_GROUP, &fuidp);
2575 if (new_gid != zp->z_gid &&
2576 zfs_id_overquota(zfsvfs, DMU_GROUPUSED_OBJECT,
2577 new_gid)) {
2578 if (attrzp)
2579 vput(ZTOV(attrzp));
2580 err = SET_ERROR(EDQUOT);
2581 goto out2;
2582 }
2583 }
2584
2585 if (projid != ZFS_INVALID_PROJID &&
2586 zfs_id_overquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid)) {
2587 if (attrzp)
2588 vput(ZTOV(attrzp));
2589 err = SET_ERROR(EDQUOT);
2590 goto out2;
2591 }
2592 }
2593 tx = dmu_tx_create(os);
2594
2595 if (mask & AT_MODE) {
2596 uint64_t pmode = zp->z_mode;
2597 uint64_t acl_obj;
2598 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2599
2600 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
2601 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
2602 err = SET_ERROR(EPERM);
2603 goto out;
2604 }
2605
2606 if ((err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)))
2607 goto out;
2608
2609 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2610 /*
2611 * Are we upgrading ACL from old V0 format
2612 * to V1 format?
2613 */
2614 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
2615 zfs_znode_acl_version(zp) ==
2616 ZFS_ACL_VERSION_INITIAL) {
2617 dmu_tx_hold_free(tx, acl_obj, 0,
2618 DMU_OBJECT_END);
2619 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2620 0, aclp->z_acl_bytes);
2621 } else {
2622 dmu_tx_hold_write(tx, acl_obj, 0,
2623 aclp->z_acl_bytes);
2624 }
2625 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2626 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2627 0, aclp->z_acl_bytes);
2628 }
2629 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2630 } else {
2631 if (((mask & AT_XVATTR) &&
2632 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) ||
2633 (projid != ZFS_INVALID_PROJID &&
2634 !(zp->z_pflags & ZFS_PROJID)))
2635 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2636 else
2637 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2638 }
2639
2640 if (attrzp) {
2641 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
2642 }
2643
2644 fuid_dirtied = zfsvfs->z_fuid_dirty;
2645 if (fuid_dirtied)
2646 zfs_fuid_txhold(zfsvfs, tx);
2647
2648 zfs_sa_upgrade_txholds(tx, zp);
2649
2650 err = dmu_tx_assign(tx, TXG_WAIT);
2651 if (err)
2652 goto out;
2653
2654 count = 0;
2655 /*
2656 * Set each attribute requested.
2657 * We group settings according to the locks they need to acquire.
2658 *
2659 * Note: you cannot set ctime directly, although it will be
2660 * updated as a side-effect of calling this function.
2661 */
2662
2663 if (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID)) {
2664 /*
2665 * For the existed object that is upgraded from old system,
2666 * its on-disk layout has no slot for the project ID attribute.
2667 * But quota accounting logic needs to access related slots by
2668 * offset directly. So we need to adjust old objects' layout
2669 * to make the project ID to some unified and fixed offset.
2670 */
2671 if (attrzp)
2672 err = sa_add_projid(attrzp->z_sa_hdl, tx, projid);
2673 if (err == 0)
2674 err = sa_add_projid(zp->z_sa_hdl, tx, projid);
2675
2676 if (unlikely(err == EEXIST))
2677 err = 0;
2678 else if (err != 0)
2679 goto out;
2680 else
2681 projid = ZFS_INVALID_PROJID;
2682 }
2683
2684 if (mask & (AT_UID|AT_GID|AT_MODE))
2685 mutex_enter(&zp->z_acl_lock);
2686
2687 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
2688 &zp->z_pflags, sizeof (zp->z_pflags));
2689
2690 if (attrzp) {
2691 if (mask & (AT_UID|AT_GID|AT_MODE))
2692 mutex_enter(&attrzp->z_acl_lock);
2693 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2694 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
2695 sizeof (attrzp->z_pflags));
2696 if (projid != ZFS_INVALID_PROJID) {
2697 attrzp->z_projid = projid;
2698 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2699 SA_ZPL_PROJID(zfsvfs), NULL, &attrzp->z_projid,
2700 sizeof (attrzp->z_projid));
2701 }
2702 }
2703
2704 if (mask & (AT_UID|AT_GID)) {
2705
2706 if (mask & AT_UID) {
2707 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
2708 &new_uid, sizeof (new_uid));
2709 zp->z_uid = new_uid;
2710 if (attrzp) {
2711 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2712 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
2713 sizeof (new_uid));
2714 attrzp->z_uid = new_uid;
2715 }
2716 }
2717
2718 if (mask & AT_GID) {
2719 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
2720 NULL, &new_gid, sizeof (new_gid));
2721 zp->z_gid = new_gid;
2722 if (attrzp) {
2723 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2724 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
2725 sizeof (new_gid));
2726 attrzp->z_gid = new_gid;
2727 }
2728 }
2729 if (!(mask & AT_MODE)) {
2730 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
2731 NULL, &new_mode, sizeof (new_mode));
2732 new_mode = zp->z_mode;
2733 }
2734 err = zfs_acl_chown_setattr(zp);
2735 ASSERT0(err);
2736 if (attrzp) {
2737 vn_seqc_write_begin(ZTOV(attrzp));
2738 err = zfs_acl_chown_setattr(attrzp);
2739 vn_seqc_write_end(ZTOV(attrzp));
2740 ASSERT0(err);
2741 }
2742 }
2743
2744 if (mask & AT_MODE) {
2745 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
2746 &new_mode, sizeof (new_mode));
2747 zp->z_mode = new_mode;
2748 ASSERT3P(aclp, !=, NULL);
2749 err = zfs_aclset_common(zp, aclp, cr, tx);
2750 ASSERT0(err);
2751 if (zp->z_acl_cached)
2752 zfs_acl_free(zp->z_acl_cached);
2753 zp->z_acl_cached = aclp;
2754 aclp = NULL;
2755 }
2756
2757
2758 if (mask & AT_ATIME) {
2759 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
2760 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
2761 &zp->z_atime, sizeof (zp->z_atime));
2762 }
2763
2764 if (mask & AT_MTIME) {
2765 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
2766 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
2767 mtime, sizeof (mtime));
2768 }
2769
2770 if (projid != ZFS_INVALID_PROJID) {
2771 zp->z_projid = projid;
2772 SA_ADD_BULK_ATTR(bulk, count,
2773 SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid,
2774 sizeof (zp->z_projid));
2775 }
2776
2777 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2778 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
2779 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
2780 NULL, mtime, sizeof (mtime));
2781 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
2782 &ctime, sizeof (ctime));
2783 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
2784 } else if (mask != 0) {
2785 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
2786 &ctime, sizeof (ctime));
2787 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime);
2788 if (attrzp) {
2789 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2790 SA_ZPL_CTIME(zfsvfs), NULL,
2791 &ctime, sizeof (ctime));
2792 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
2793 mtime, ctime);
2794 }
2795 }
2796
2797 /*
2798 * Do this after setting timestamps to prevent timestamp
2799 * update from toggling bit
2800 */
2801
2802 if (xoap && (mask & AT_XVATTR)) {
2803
2804 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME))
2805 xoap->xoa_createtime = vap->va_birthtime;
2806 /*
2807 * restore trimmed off masks
2808 * so that return masks can be set for caller.
2809 */
2810
2811 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
2812 XVA_SET_REQ(xvap, XAT_APPENDONLY);
2813 }
2814 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
2815 XVA_SET_REQ(xvap, XAT_NOUNLINK);
2816 }
2817 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
2818 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
2819 }
2820 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
2821 XVA_SET_REQ(xvap, XAT_NODUMP);
2822 }
2823 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
2824 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
2825 }
2826 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
2827 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
2828 }
2829 if (XVA_ISSET_REQ(&tmpxvattr, XAT_PROJINHERIT)) {
2830 XVA_SET_REQ(xvap, XAT_PROJINHERIT);
2831 }
2832
2833 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2834 ASSERT3S(vp->v_type, ==, VREG);
2835
2836 zfs_xvattr_set(zp, xvap, tx);
2837 }
2838
2839 if (fuid_dirtied)
2840 zfs_fuid_sync(zfsvfs, tx);
2841
2842 if (mask != 0)
2843 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2844
2845 if (mask & (AT_UID|AT_GID|AT_MODE))
2846 mutex_exit(&zp->z_acl_lock);
2847
2848 if (attrzp) {
2849 if (mask & (AT_UID|AT_GID|AT_MODE))
2850 mutex_exit(&attrzp->z_acl_lock);
2851 }
2852 out:
2853 if (err == 0 && attrzp) {
2854 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
2855 xattr_count, tx);
2856 ASSERT0(err2);
2857 }
2858
2859 if (attrzp)
2860 vput(ZTOV(attrzp));
2861
2862 if (aclp)
2863 zfs_acl_free(aclp);
2864
2865 if (fuidp) {
2866 zfs_fuid_info_free(fuidp);
2867 fuidp = NULL;
2868 }
2869
2870 if (err) {
2871 dmu_tx_abort(tx);
2872 } else {
2873 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
2874 dmu_tx_commit(tx);
2875 }
2876
2877 out2:
2878 if (os->os_sync == ZFS_SYNC_ALWAYS)
2879 zil_commit(zilog, 0);
2880
2881 zfs_exit(zfsvfs, FTAG);
2882 return (err);
2883 }
2884
2885 /*
2886 * Look up the directory entries corresponding to the source and target
2887 * directory/name pairs.
2888 */
2889 static int
zfs_rename_relock_lookup(znode_t * sdzp,const struct componentname * scnp,znode_t ** szpp,znode_t * tdzp,const struct componentname * tcnp,znode_t ** tzpp)2890 zfs_rename_relock_lookup(znode_t *sdzp, const struct componentname *scnp,
2891 znode_t **szpp, znode_t *tdzp, const struct componentname *tcnp,
2892 znode_t **tzpp)
2893 {
2894 zfsvfs_t *zfsvfs;
2895 znode_t *szp, *tzp;
2896 int error;
2897
2898 /*
2899 * Before using sdzp and tdzp we must ensure that they are live.
2900 * As a porting legacy from illumos we have two things to worry
2901 * about. One is typical for FreeBSD and it is that the vnode is
2902 * not reclaimed (doomed). The other is that the znode is live.
2903 * The current code can invalidate the znode without acquiring the
2904 * corresponding vnode lock if the object represented by the znode
2905 * and vnode is no longer valid after a rollback or receive operation.
2906 * z_teardown_lock hidden behind zfs_enter and zfs_exit is the lock
2907 * that protects the znodes from the invalidation.
2908 */
2909 zfsvfs = sdzp->z_zfsvfs;
2910 ASSERT3P(zfsvfs, ==, tdzp->z_zfsvfs);
2911 if ((error = zfs_enter_verify_zp(zfsvfs, sdzp, FTAG)) != 0)
2912 return (error);
2913 if ((error = zfs_verify_zp(tdzp)) != 0) {
2914 zfs_exit(zfsvfs, FTAG);
2915 return (error);
2916 }
2917
2918 /*
2919 * Re-resolve svp to be certain it still exists and fetch the
2920 * correct vnode.
2921 */
2922 error = zfs_dirent_lookup(sdzp, scnp->cn_nameptr, &szp, ZEXISTS);
2923 if (error != 0) {
2924 /* Source entry invalid or not there. */
2925 if ((scnp->cn_flags & ISDOTDOT) != 0 ||
2926 (scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.'))
2927 error = SET_ERROR(EINVAL);
2928 goto out;
2929 }
2930 *szpp = szp;
2931
2932 /*
2933 * Re-resolve tvp, if it disappeared we just carry on.
2934 */
2935 error = zfs_dirent_lookup(tdzp, tcnp->cn_nameptr, &tzp, 0);
2936 if (error != 0) {
2937 vrele(ZTOV(szp));
2938 if ((tcnp->cn_flags & ISDOTDOT) != 0)
2939 error = SET_ERROR(EINVAL);
2940 goto out;
2941 }
2942 *tzpp = tzp;
2943 out:
2944 zfs_exit(zfsvfs, FTAG);
2945 return (error);
2946 }
2947
2948 /*
2949 * We acquire all but fdvp locks using non-blocking acquisitions. If we
2950 * fail to acquire any lock in the path we will drop all held locks,
2951 * acquire the new lock in a blocking fashion, and then release it and
2952 * restart the rename. This acquire/release step ensures that we do not
2953 * spin on a lock waiting for release. On error release all vnode locks
2954 * and decrement references the way tmpfs_rename() would do.
2955 */
2956 static int
zfs_rename_relock(struct vnode * sdvp,struct vnode ** svpp,struct vnode * tdvp,struct vnode ** tvpp,const struct componentname * scnp,const struct componentname * tcnp)2957 zfs_rename_relock(struct vnode *sdvp, struct vnode **svpp,
2958 struct vnode *tdvp, struct vnode **tvpp,
2959 const struct componentname *scnp, const struct componentname *tcnp)
2960 {
2961 struct vnode *nvp, *svp, *tvp;
2962 znode_t *sdzp, *tdzp, *szp, *tzp;
2963 int error;
2964
2965 VOP_UNLOCK1(tdvp);
2966 if (*tvpp != NULL && *tvpp != tdvp)
2967 VOP_UNLOCK1(*tvpp);
2968
2969 relock:
2970 error = vn_lock(sdvp, LK_EXCLUSIVE);
2971 if (error)
2972 goto out;
2973 error = vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT);
2974 if (error != 0) {
2975 VOP_UNLOCK1(sdvp);
2976 if (error != EBUSY)
2977 goto out;
2978 error = vn_lock(tdvp, LK_EXCLUSIVE);
2979 if (error)
2980 goto out;
2981 VOP_UNLOCK1(tdvp);
2982 goto relock;
2983 }
2984 tdzp = VTOZ(tdvp);
2985 sdzp = VTOZ(sdvp);
2986
2987 error = zfs_rename_relock_lookup(sdzp, scnp, &szp, tdzp, tcnp, &tzp);
2988 if (error != 0) {
2989 VOP_UNLOCK1(sdvp);
2990 VOP_UNLOCK1(tdvp);
2991 goto out;
2992 }
2993 svp = ZTOV(szp);
2994 tvp = tzp != NULL ? ZTOV(tzp) : NULL;
2995
2996 /*
2997 * Now try acquire locks on svp and tvp.
2998 */
2999 nvp = svp;
3000 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT);
3001 if (error != 0) {
3002 VOP_UNLOCK1(sdvp);
3003 VOP_UNLOCK1(tdvp);
3004 if (tvp != NULL)
3005 vrele(tvp);
3006 if (error != EBUSY) {
3007 vrele(nvp);
3008 goto out;
3009 }
3010 error = vn_lock(nvp, LK_EXCLUSIVE);
3011 if (error != 0) {
3012 vrele(nvp);
3013 goto out;
3014 }
3015 VOP_UNLOCK1(nvp);
3016 /*
3017 * Concurrent rename race.
3018 * XXX ?
3019 */
3020 if (nvp == tdvp) {
3021 vrele(nvp);
3022 error = SET_ERROR(EINVAL);
3023 goto out;
3024 }
3025 vrele(*svpp);
3026 *svpp = nvp;
3027 goto relock;
3028 }
3029 vrele(*svpp);
3030 *svpp = nvp;
3031
3032 if (*tvpp != NULL)
3033 vrele(*tvpp);
3034 *tvpp = NULL;
3035 if (tvp != NULL) {
3036 nvp = tvp;
3037 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT);
3038 if (error != 0) {
3039 VOP_UNLOCK1(sdvp);
3040 VOP_UNLOCK1(tdvp);
3041 VOP_UNLOCK1(*svpp);
3042 if (error != EBUSY) {
3043 vrele(nvp);
3044 goto out;
3045 }
3046 error = vn_lock(nvp, LK_EXCLUSIVE);
3047 if (error != 0) {
3048 vrele(nvp);
3049 goto out;
3050 }
3051 vput(nvp);
3052 goto relock;
3053 }
3054 *tvpp = nvp;
3055 }
3056
3057 return (0);
3058
3059 out:
3060 return (error);
3061 }
3062
3063 /*
3064 * Note that we must use VRELE_ASYNC in this function as it walks
3065 * up the directory tree and vrele may need to acquire an exclusive
3066 * lock if a last reference to a vnode is dropped.
3067 */
3068 static int
zfs_rename_check(znode_t * szp,znode_t * sdzp,znode_t * tdzp)3069 zfs_rename_check(znode_t *szp, znode_t *sdzp, znode_t *tdzp)
3070 {
3071 zfsvfs_t *zfsvfs;
3072 znode_t *zp, *zp1;
3073 uint64_t parent;
3074 int error;
3075
3076 zfsvfs = tdzp->z_zfsvfs;
3077 if (tdzp == szp)
3078 return (SET_ERROR(EINVAL));
3079 if (tdzp == sdzp)
3080 return (0);
3081 if (tdzp->z_id == zfsvfs->z_root)
3082 return (0);
3083 zp = tdzp;
3084 for (;;) {
3085 ASSERT(!zp->z_unlinked);
3086 if ((error = sa_lookup(zp->z_sa_hdl,
3087 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
3088 break;
3089
3090 if (parent == szp->z_id) {
3091 error = SET_ERROR(EINVAL);
3092 break;
3093 }
3094 if (parent == zfsvfs->z_root)
3095 break;
3096 if (parent == sdzp->z_id)
3097 break;
3098
3099 error = zfs_zget(zfsvfs, parent, &zp1);
3100 if (error != 0)
3101 break;
3102
3103 if (zp != tdzp)
3104 VN_RELE_ASYNC(ZTOV(zp),
3105 dsl_pool_zrele_taskq(
3106 dmu_objset_pool(zfsvfs->z_os)));
3107 zp = zp1;
3108 }
3109
3110 if (error == ENOTDIR)
3111 panic("checkpath: .. not a directory\n");
3112 if (zp != tdzp)
3113 VN_RELE_ASYNC(ZTOV(zp),
3114 dsl_pool_zrele_taskq(dmu_objset_pool(zfsvfs->z_os)));
3115 return (error);
3116 }
3117
3118 #if __FreeBSD_version < 1300124
3119 static void
cache_vop_rename(struct vnode * fdvp,struct vnode * fvp,struct vnode * tdvp,struct vnode * tvp,struct componentname * fcnp,struct componentname * tcnp)3120 cache_vop_rename(struct vnode *fdvp, struct vnode *fvp, struct vnode *tdvp,
3121 struct vnode *tvp, struct componentname *fcnp, struct componentname *tcnp)
3122 {
3123
3124 cache_purge(fvp);
3125 if (tvp != NULL)
3126 cache_purge(tvp);
3127 cache_purge_negative(tdvp);
3128 }
3129 #endif
3130
3131 static int
3132 zfs_do_rename_impl(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp,
3133 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp,
3134 cred_t *cr);
3135
3136 /*
3137 * Move an entry from the provided source directory to the target
3138 * directory. Change the entry name as indicated.
3139 *
3140 * IN: sdvp - Source directory containing the "old entry".
3141 * scnp - Old entry name.
3142 * tdvp - Target directory to contain the "new entry".
3143 * tcnp - New entry name.
3144 * cr - credentials of caller.
3145 * INOUT: svpp - Source file
3146 * tvpp - Target file, may point to NULL initially
3147 *
3148 * RETURN: 0 on success, error code on failure.
3149 *
3150 * Timestamps:
3151 * sdvp,tdvp - ctime|mtime updated
3152 */
3153 static int
zfs_do_rename(vnode_t * sdvp,vnode_t ** svpp,struct componentname * scnp,vnode_t * tdvp,vnode_t ** tvpp,struct componentname * tcnp,cred_t * cr)3154 zfs_do_rename(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp,
3155 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp,
3156 cred_t *cr)
3157 {
3158 int error;
3159
3160 ASSERT_VOP_ELOCKED(tdvp, __func__);
3161 if (*tvpp != NULL)
3162 ASSERT_VOP_ELOCKED(*tvpp, __func__);
3163
3164 /* Reject renames across filesystems. */
3165 if ((*svpp)->v_mount != tdvp->v_mount ||
3166 ((*tvpp) != NULL && (*svpp)->v_mount != (*tvpp)->v_mount)) {
3167 error = SET_ERROR(EXDEV);
3168 goto out;
3169 }
3170
3171 if (zfsctl_is_node(tdvp)) {
3172 error = SET_ERROR(EXDEV);
3173 goto out;
3174 }
3175
3176 /*
3177 * Lock all four vnodes to ensure safety and semantics of renaming.
3178 */
3179 error = zfs_rename_relock(sdvp, svpp, tdvp, tvpp, scnp, tcnp);
3180 if (error != 0) {
3181 /* no vnodes are locked in the case of error here */
3182 return (error);
3183 }
3184
3185 error = zfs_do_rename_impl(sdvp, svpp, scnp, tdvp, tvpp, tcnp, cr);
3186 VOP_UNLOCK1(sdvp);
3187 VOP_UNLOCK1(*svpp);
3188 out:
3189 if (*tvpp != NULL)
3190 VOP_UNLOCK1(*tvpp);
3191 if (tdvp != *tvpp)
3192 VOP_UNLOCK1(tdvp);
3193
3194 return (error);
3195 }
3196
3197 static int
zfs_do_rename_impl(vnode_t * sdvp,vnode_t ** svpp,struct componentname * scnp,vnode_t * tdvp,vnode_t ** tvpp,struct componentname * tcnp,cred_t * cr)3198 zfs_do_rename_impl(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp,
3199 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp,
3200 cred_t *cr)
3201 {
3202 dmu_tx_t *tx;
3203 zfsvfs_t *zfsvfs;
3204 zilog_t *zilog;
3205 znode_t *tdzp, *sdzp, *tzp, *szp;
3206 const char *snm = scnp->cn_nameptr;
3207 const char *tnm = tcnp->cn_nameptr;
3208 int error;
3209
3210 tdzp = VTOZ(tdvp);
3211 sdzp = VTOZ(sdvp);
3212 zfsvfs = tdzp->z_zfsvfs;
3213
3214 if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0)
3215 return (error);
3216 if ((error = zfs_verify_zp(sdzp)) != 0) {
3217 zfs_exit(zfsvfs, FTAG);
3218 return (error);
3219 }
3220 zilog = zfsvfs->z_log;
3221
3222 if (zfsvfs->z_utf8 && u8_validate(tnm,
3223 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3224 error = SET_ERROR(EILSEQ);
3225 goto out;
3226 }
3227
3228 /* If source and target are the same file, there is nothing to do. */
3229 if ((*svpp) == (*tvpp)) {
3230 error = 0;
3231 goto out;
3232 }
3233
3234 if (((*svpp)->v_type == VDIR && (*svpp)->v_mountedhere != NULL) ||
3235 ((*tvpp) != NULL && (*tvpp)->v_type == VDIR &&
3236 (*tvpp)->v_mountedhere != NULL)) {
3237 error = SET_ERROR(EXDEV);
3238 goto out;
3239 }
3240
3241 szp = VTOZ(*svpp);
3242 if ((error = zfs_verify_zp(szp)) != 0) {
3243 zfs_exit(zfsvfs, FTAG);
3244 return (error);
3245 }
3246 tzp = *tvpp == NULL ? NULL : VTOZ(*tvpp);
3247 if (tzp != NULL) {
3248 if ((error = zfs_verify_zp(tzp)) != 0) {
3249 zfs_exit(zfsvfs, FTAG);
3250 return (error);
3251 }
3252 }
3253
3254 /*
3255 * This is to prevent the creation of links into attribute space
3256 * by renaming a linked file into/outof an attribute directory.
3257 * See the comment in zfs_link() for why this is considered bad.
3258 */
3259 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3260 error = SET_ERROR(EINVAL);
3261 goto out;
3262 }
3263
3264 /*
3265 * If we are using project inheritance, means if the directory has
3266 * ZFS_PROJINHERIT set, then its descendant directories will inherit
3267 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
3268 * such case, we only allow renames into our tree when the project
3269 * IDs are the same.
3270 */
3271 if (tdzp->z_pflags & ZFS_PROJINHERIT &&
3272 tdzp->z_projid != szp->z_projid) {
3273 error = SET_ERROR(EXDEV);
3274 goto out;
3275 }
3276
3277 /*
3278 * Must have write access at the source to remove the old entry
3279 * and write access at the target to create the new entry.
3280 * Note that if target and source are the same, this can be
3281 * done in a single check.
3282 */
3283 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr, NULL)))
3284 goto out;
3285
3286 if ((*svpp)->v_type == VDIR) {
3287 /*
3288 * Avoid ".", "..", and aliases of "." for obvious reasons.
3289 */
3290 if ((scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.') ||
3291 sdzp == szp ||
3292 (scnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) {
3293 error = EINVAL;
3294 goto out;
3295 }
3296
3297 /*
3298 * Check to make sure rename is valid.
3299 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3300 */
3301 if ((error = zfs_rename_check(szp, sdzp, tdzp)))
3302 goto out;
3303 }
3304
3305 /*
3306 * Does target exist?
3307 */
3308 if (tzp) {
3309 /*
3310 * Source and target must be the same type.
3311 */
3312 if ((*svpp)->v_type == VDIR) {
3313 if ((*tvpp)->v_type != VDIR) {
3314 error = SET_ERROR(ENOTDIR);
3315 goto out;
3316 } else {
3317 cache_purge(tdvp);
3318 if (sdvp != tdvp)
3319 cache_purge(sdvp);
3320 }
3321 } else {
3322 if ((*tvpp)->v_type == VDIR) {
3323 error = SET_ERROR(EISDIR);
3324 goto out;
3325 }
3326 }
3327 }
3328
3329 vn_seqc_write_begin(*svpp);
3330 vn_seqc_write_begin(sdvp);
3331 if (*tvpp != NULL)
3332 vn_seqc_write_begin(*tvpp);
3333 if (tdvp != *tvpp)
3334 vn_seqc_write_begin(tdvp);
3335
3336 vnevent_rename_src(*svpp, sdvp, scnp->cn_nameptr, ct);
3337 if (tzp)
3338 vnevent_rename_dest(*tvpp, tdvp, tnm, ct);
3339
3340 /*
3341 * notify the target directory if it is not the same
3342 * as source directory.
3343 */
3344 if (tdvp != sdvp) {
3345 vnevent_rename_dest_dir(tdvp, ct);
3346 }
3347
3348 tx = dmu_tx_create(zfsvfs->z_os);
3349 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3350 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3351 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3352 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3353 if (sdzp != tdzp) {
3354 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3355 zfs_sa_upgrade_txholds(tx, tdzp);
3356 }
3357 if (tzp) {
3358 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3359 zfs_sa_upgrade_txholds(tx, tzp);
3360 }
3361
3362 zfs_sa_upgrade_txholds(tx, szp);
3363 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3364 error = dmu_tx_assign(tx, TXG_WAIT);
3365 if (error) {
3366 dmu_tx_abort(tx);
3367 goto out_seq;
3368 }
3369
3370 if (tzp) /* Attempt to remove the existing target */
3371 error = zfs_link_destroy(tdzp, tnm, tzp, tx, 0, NULL);
3372
3373 if (error == 0) {
3374 error = zfs_link_create(tdzp, tnm, szp, tx, ZRENAMING);
3375 if (error == 0) {
3376 szp->z_pflags |= ZFS_AV_MODIFIED;
3377
3378 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3379 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3380 ASSERT0(error);
3381
3382 error = zfs_link_destroy(sdzp, snm, szp, tx, ZRENAMING,
3383 NULL);
3384 if (error == 0) {
3385 zfs_log_rename(zilog, tx, TX_RENAME, sdzp,
3386 snm, tdzp, tnm, szp);
3387 } else {
3388 /*
3389 * At this point, we have successfully created
3390 * the target name, but have failed to remove
3391 * the source name. Since the create was done
3392 * with the ZRENAMING flag, there are
3393 * complications; for one, the link count is
3394 * wrong. The easiest way to deal with this
3395 * is to remove the newly created target, and
3396 * return the original error. This must
3397 * succeed; fortunately, it is very unlikely to
3398 * fail, since we just created it.
3399 */
3400 VERIFY0(zfs_link_destroy(tdzp, tnm, szp, tx,
3401 ZRENAMING, NULL));
3402 }
3403 }
3404 if (error == 0) {
3405 cache_vop_rename(sdvp, *svpp, tdvp, *tvpp, scnp, tcnp);
3406 }
3407 }
3408
3409 dmu_tx_commit(tx);
3410
3411 out_seq:
3412 vn_seqc_write_end(*svpp);
3413 vn_seqc_write_end(sdvp);
3414 if (*tvpp != NULL)
3415 vn_seqc_write_end(*tvpp);
3416 if (tdvp != *tvpp)
3417 vn_seqc_write_end(tdvp);
3418
3419 out:
3420 if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3421 zil_commit(zilog, 0);
3422 zfs_exit(zfsvfs, FTAG);
3423
3424 return (error);
3425 }
3426
3427 int
zfs_rename(znode_t * sdzp,const char * sname,znode_t * tdzp,const char * tname,cred_t * cr,int flags,uint64_t rflags,vattr_t * wo_vap,zidmap_t * mnt_ns)3428 zfs_rename(znode_t *sdzp, const char *sname, znode_t *tdzp, const char *tname,
3429 cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap, zidmap_t *mnt_ns)
3430 {
3431 struct componentname scn, tcn;
3432 vnode_t *sdvp, *tdvp;
3433 vnode_t *svp, *tvp;
3434 int error;
3435 svp = tvp = NULL;
3436
3437 if (rflags != 0 || wo_vap != NULL)
3438 return (SET_ERROR(EINVAL));
3439
3440 sdvp = ZTOV(sdzp);
3441 tdvp = ZTOV(tdzp);
3442 error = zfs_lookup_internal(sdzp, sname, &svp, &scn, DELETE);
3443 if (sdzp->z_zfsvfs->z_replay == B_FALSE)
3444 VOP_UNLOCK1(sdvp);
3445 if (error != 0)
3446 goto fail;
3447 VOP_UNLOCK1(svp);
3448
3449 vn_lock(tdvp, LK_EXCLUSIVE | LK_RETRY);
3450 error = zfs_lookup_internal(tdzp, tname, &tvp, &tcn, RENAME);
3451 if (error == EJUSTRETURN)
3452 tvp = NULL;
3453 else if (error != 0) {
3454 VOP_UNLOCK1(tdvp);
3455 goto fail;
3456 }
3457
3458 error = zfs_do_rename(sdvp, &svp, &scn, tdvp, &tvp, &tcn, cr);
3459 fail:
3460 if (svp != NULL)
3461 vrele(svp);
3462 if (tvp != NULL)
3463 vrele(tvp);
3464
3465 return (error);
3466 }
3467
3468 /*
3469 * Insert the indicated symbolic reference entry into the directory.
3470 *
3471 * IN: dvp - Directory to contain new symbolic link.
3472 * link - Name for new symlink entry.
3473 * vap - Attributes of new entry.
3474 * cr - credentials of caller.
3475 * ct - caller context
3476 * flags - case flags
3477 * mnt_ns - Unused on FreeBSD
3478 *
3479 * RETURN: 0 on success, error code on failure.
3480 *
3481 * Timestamps:
3482 * dvp - ctime|mtime updated
3483 */
3484 int
zfs_symlink(znode_t * dzp,const char * name,vattr_t * vap,const char * link,znode_t ** zpp,cred_t * cr,int flags,zidmap_t * mnt_ns)3485 zfs_symlink(znode_t *dzp, const char *name, vattr_t *vap,
3486 const char *link, znode_t **zpp, cred_t *cr, int flags, zidmap_t *mnt_ns)
3487 {
3488 (void) flags;
3489 znode_t *zp;
3490 dmu_tx_t *tx;
3491 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
3492 zilog_t *zilog;
3493 uint64_t len = strlen(link);
3494 int error;
3495 zfs_acl_ids_t acl_ids;
3496 boolean_t fuid_dirtied;
3497 uint64_t txtype = TX_SYMLINK;
3498
3499 ASSERT3S(vap->va_type, ==, VLNK);
3500
3501 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0)
3502 return (error);
3503 zilog = zfsvfs->z_log;
3504
3505 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3506 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3507 zfs_exit(zfsvfs, FTAG);
3508 return (SET_ERROR(EILSEQ));
3509 }
3510
3511 if (len > MAXPATHLEN) {
3512 zfs_exit(zfsvfs, FTAG);
3513 return (SET_ERROR(ENAMETOOLONG));
3514 }
3515
3516 if ((error = zfs_acl_ids_create(dzp, 0,
3517 vap, cr, NULL, &acl_ids, NULL)) != 0) {
3518 zfs_exit(zfsvfs, FTAG);
3519 return (error);
3520 }
3521
3522 /*
3523 * Attempt to lock directory; fail if entry already exists.
3524 */
3525 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW);
3526 if (error) {
3527 zfs_acl_ids_free(&acl_ids);
3528 zfs_exit(zfsvfs, FTAG);
3529 return (error);
3530 }
3531
3532 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) {
3533 zfs_acl_ids_free(&acl_ids);
3534 zfs_exit(zfsvfs, FTAG);
3535 return (error);
3536 }
3537
3538 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids,
3539 0 /* projid */)) {
3540 zfs_acl_ids_free(&acl_ids);
3541 zfs_exit(zfsvfs, FTAG);
3542 return (SET_ERROR(EDQUOT));
3543 }
3544
3545 getnewvnode_reserve_();
3546 tx = dmu_tx_create(zfsvfs->z_os);
3547 fuid_dirtied = zfsvfs->z_fuid_dirty;
3548 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3549 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3550 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3551 ZFS_SA_BASE_ATTR_SIZE + len);
3552 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3553 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3554 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3555 acl_ids.z_aclp->z_acl_bytes);
3556 }
3557 if (fuid_dirtied)
3558 zfs_fuid_txhold(zfsvfs, tx);
3559 error = dmu_tx_assign(tx, TXG_WAIT);
3560 if (error) {
3561 zfs_acl_ids_free(&acl_ids);
3562 dmu_tx_abort(tx);
3563 getnewvnode_drop_reserve();
3564 zfs_exit(zfsvfs, FTAG);
3565 return (error);
3566 }
3567
3568 /*
3569 * Create a new object for the symlink.
3570 * for version 4 ZPL datasets the symlink will be an SA attribute
3571 */
3572 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3573
3574 if (fuid_dirtied)
3575 zfs_fuid_sync(zfsvfs, tx);
3576
3577 if (zp->z_is_sa)
3578 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3579 __DECONST(void *, link), len, tx);
3580 else
3581 zfs_sa_symlink(zp, __DECONST(char *, link), len, tx);
3582
3583 zp->z_size = len;
3584 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3585 &zp->z_size, sizeof (zp->z_size), tx);
3586 /*
3587 * Insert the new object into the directory.
3588 */
3589 (void) zfs_link_create(dzp, name, zp, tx, ZNEW);
3590
3591 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3592 *zpp = zp;
3593
3594 zfs_acl_ids_free(&acl_ids);
3595
3596 dmu_tx_commit(tx);
3597
3598 getnewvnode_drop_reserve();
3599
3600 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3601 zil_commit(zilog, 0);
3602
3603 zfs_exit(zfsvfs, FTAG);
3604 return (error);
3605 }
3606
3607 /*
3608 * Return, in the buffer contained in the provided uio structure,
3609 * the symbolic path referred to by vp.
3610 *
3611 * IN: vp - vnode of symbolic link.
3612 * uio - structure to contain the link path.
3613 * cr - credentials of caller.
3614 * ct - caller context
3615 *
3616 * OUT: uio - structure containing the link path.
3617 *
3618 * RETURN: 0 on success, error code on failure.
3619 *
3620 * Timestamps:
3621 * vp - atime updated
3622 */
3623 static int
zfs_readlink(vnode_t * vp,zfs_uio_t * uio,cred_t * cr,caller_context_t * ct)3624 zfs_readlink(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, caller_context_t *ct)
3625 {
3626 (void) cr, (void) ct;
3627 znode_t *zp = VTOZ(vp);
3628 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3629 int error;
3630
3631 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
3632 return (error);
3633
3634 if (zp->z_is_sa)
3635 error = sa_lookup_uio(zp->z_sa_hdl,
3636 SA_ZPL_SYMLINK(zfsvfs), uio);
3637 else
3638 error = zfs_sa_readlink(zp, uio);
3639
3640 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3641
3642 zfs_exit(zfsvfs, FTAG);
3643 return (error);
3644 }
3645
3646 /*
3647 * Insert a new entry into directory tdvp referencing svp.
3648 *
3649 * IN: tdvp - Directory to contain new entry.
3650 * svp - vnode of new entry.
3651 * name - name of new entry.
3652 * cr - credentials of caller.
3653 *
3654 * RETURN: 0 on success, error code on failure.
3655 *
3656 * Timestamps:
3657 * tdvp - ctime|mtime updated
3658 * svp - ctime updated
3659 */
3660 int
zfs_link(znode_t * tdzp,znode_t * szp,const char * name,cred_t * cr,int flags)3661 zfs_link(znode_t *tdzp, znode_t *szp, const char *name, cred_t *cr,
3662 int flags)
3663 {
3664 (void) flags;
3665 znode_t *tzp;
3666 zfsvfs_t *zfsvfs = tdzp->z_zfsvfs;
3667 zilog_t *zilog;
3668 dmu_tx_t *tx;
3669 int error;
3670 uint64_t parent;
3671 uid_t owner;
3672
3673 ASSERT3S(ZTOV(tdzp)->v_type, ==, VDIR);
3674
3675 if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0)
3676 return (error);
3677 zilog = zfsvfs->z_log;
3678
3679 /*
3680 * POSIX dictates that we return EPERM here.
3681 * Better choices include ENOTSUP or EISDIR.
3682 */
3683 if (ZTOV(szp)->v_type == VDIR) {
3684 zfs_exit(zfsvfs, FTAG);
3685 return (SET_ERROR(EPERM));
3686 }
3687
3688 if ((error = zfs_verify_zp(szp)) != 0) {
3689 zfs_exit(zfsvfs, FTAG);
3690 return (error);
3691 }
3692
3693 /*
3694 * If we are using project inheritance, means if the directory has
3695 * ZFS_PROJINHERIT set, then its descendant directories will inherit
3696 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
3697 * such case, we only allow hard link creation in our tree when the
3698 * project IDs are the same.
3699 */
3700 if (tdzp->z_pflags & ZFS_PROJINHERIT &&
3701 tdzp->z_projid != szp->z_projid) {
3702 zfs_exit(zfsvfs, FTAG);
3703 return (SET_ERROR(EXDEV));
3704 }
3705
3706 if (szp->z_pflags & (ZFS_APPENDONLY |
3707 ZFS_IMMUTABLE | ZFS_READONLY)) {
3708 zfs_exit(zfsvfs, FTAG);
3709 return (SET_ERROR(EPERM));
3710 }
3711
3712 /* Prevent links to .zfs/shares files */
3713
3714 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
3715 &parent, sizeof (uint64_t))) != 0) {
3716 zfs_exit(zfsvfs, FTAG);
3717 return (error);
3718 }
3719 if (parent == zfsvfs->z_shares_dir) {
3720 zfs_exit(zfsvfs, FTAG);
3721 return (SET_ERROR(EPERM));
3722 }
3723
3724 if (zfsvfs->z_utf8 && u8_validate(name,
3725 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3726 zfs_exit(zfsvfs, FTAG);
3727 return (SET_ERROR(EILSEQ));
3728 }
3729
3730 /*
3731 * We do not support links between attributes and non-attributes
3732 * because of the potential security risk of creating links
3733 * into "normal" file space in order to circumvent restrictions
3734 * imposed in attribute space.
3735 */
3736 if ((szp->z_pflags & ZFS_XATTR) != (tdzp->z_pflags & ZFS_XATTR)) {
3737 zfs_exit(zfsvfs, FTAG);
3738 return (SET_ERROR(EINVAL));
3739 }
3740
3741
3742 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
3743 if (owner != crgetuid(cr) && secpolicy_basic_link(ZTOV(szp), cr) != 0) {
3744 zfs_exit(zfsvfs, FTAG);
3745 return (SET_ERROR(EPERM));
3746 }
3747
3748 if ((error = zfs_zaccess(tdzp, ACE_ADD_FILE, 0, B_FALSE, cr, NULL))) {
3749 zfs_exit(zfsvfs, FTAG);
3750 return (error);
3751 }
3752
3753 /*
3754 * Attempt to lock directory; fail if entry already exists.
3755 */
3756 error = zfs_dirent_lookup(tdzp, name, &tzp, ZNEW);
3757 if (error) {
3758 zfs_exit(zfsvfs, FTAG);
3759 return (error);
3760 }
3761
3762 tx = dmu_tx_create(zfsvfs->z_os);
3763 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3764 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, name);
3765 zfs_sa_upgrade_txholds(tx, szp);
3766 zfs_sa_upgrade_txholds(tx, tdzp);
3767 error = dmu_tx_assign(tx, TXG_WAIT);
3768 if (error) {
3769 dmu_tx_abort(tx);
3770 zfs_exit(zfsvfs, FTAG);
3771 return (error);
3772 }
3773
3774 error = zfs_link_create(tdzp, name, szp, tx, 0);
3775
3776 if (error == 0) {
3777 uint64_t txtype = TX_LINK;
3778 zfs_log_link(zilog, tx, txtype, tdzp, szp, name);
3779 }
3780
3781 dmu_tx_commit(tx);
3782
3783 if (error == 0) {
3784 vnevent_link(ZTOV(szp), ct);
3785 }
3786
3787 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3788 zil_commit(zilog, 0);
3789
3790 zfs_exit(zfsvfs, FTAG);
3791 return (error);
3792 }
3793
3794 /*
3795 * Free or allocate space in a file. Currently, this function only
3796 * supports the `F_FREESP' command. However, this command is somewhat
3797 * misnamed, as its functionality includes the ability to allocate as
3798 * well as free space.
3799 *
3800 * IN: ip - inode of file to free data in.
3801 * cmd - action to take (only F_FREESP supported).
3802 * bfp - section of file to free/alloc.
3803 * flag - current file open mode flags.
3804 * offset - current file offset.
3805 * cr - credentials of caller.
3806 *
3807 * RETURN: 0 on success, error code on failure.
3808 *
3809 * Timestamps:
3810 * ip - ctime|mtime updated
3811 */
3812 int
zfs_space(znode_t * zp,int cmd,flock64_t * bfp,int flag,offset_t offset,cred_t * cr)3813 zfs_space(znode_t *zp, int cmd, flock64_t *bfp, int flag,
3814 offset_t offset, cred_t *cr)
3815 {
3816 (void) offset;
3817 zfsvfs_t *zfsvfs = ZTOZSB(zp);
3818 uint64_t off, len;
3819 int error;
3820
3821 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
3822 return (error);
3823
3824 if (cmd != F_FREESP) {
3825 zfs_exit(zfsvfs, FTAG);
3826 return (SET_ERROR(EINVAL));
3827 }
3828
3829 /*
3830 * Callers might not be able to detect properly that we are read-only,
3831 * so check it explicitly here.
3832 */
3833 if (zfs_is_readonly(zfsvfs)) {
3834 zfs_exit(zfsvfs, FTAG);
3835 return (SET_ERROR(EROFS));
3836 }
3837
3838 if (bfp->l_len < 0) {
3839 zfs_exit(zfsvfs, FTAG);
3840 return (SET_ERROR(EINVAL));
3841 }
3842
3843 /*
3844 * Permissions aren't checked on Solaris because on this OS
3845 * zfs_space() can only be called with an opened file handle.
3846 * On Linux we can get here through truncate_range() which
3847 * operates directly on inodes, so we need to check access rights.
3848 */
3849 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr, NULL))) {
3850 zfs_exit(zfsvfs, FTAG);
3851 return (error);
3852 }
3853
3854 off = bfp->l_start;
3855 len = bfp->l_len; /* 0 means from off to end of file */
3856
3857 error = zfs_freesp(zp, off, len, flag, TRUE);
3858
3859 zfs_exit(zfsvfs, FTAG);
3860 return (error);
3861 }
3862
3863 static void
zfs_inactive(vnode_t * vp,cred_t * cr,caller_context_t * ct)3864 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
3865 {
3866 (void) cr, (void) ct;
3867 znode_t *zp = VTOZ(vp);
3868 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3869 int error;
3870
3871 ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs);
3872 if (zp->z_sa_hdl == NULL) {
3873 /*
3874 * The fs has been unmounted, or we did a
3875 * suspend/resume and this file no longer exists.
3876 */
3877 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
3878 vrecycle(vp);
3879 return;
3880 }
3881
3882 if (zp->z_unlinked) {
3883 /*
3884 * Fast path to recycle a vnode of a removed file.
3885 */
3886 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
3887 vrecycle(vp);
3888 return;
3889 }
3890
3891 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3892 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3893
3894 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3895 zfs_sa_upgrade_txholds(tx, zp);
3896 error = dmu_tx_assign(tx, TXG_WAIT);
3897 if (error) {
3898 dmu_tx_abort(tx);
3899 } else {
3900 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
3901 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
3902 zp->z_atime_dirty = 0;
3903 dmu_tx_commit(tx);
3904 }
3905 }
3906 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
3907 }
3908
3909
3910 _Static_assert(sizeof (struct zfid_short) <= sizeof (struct fid),
3911 "struct zfid_short bigger than struct fid");
3912 _Static_assert(sizeof (struct zfid_long) <= sizeof (struct fid),
3913 "struct zfid_long bigger than struct fid");
3914
3915 static int
zfs_fid(vnode_t * vp,fid_t * fidp,caller_context_t * ct)3916 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
3917 {
3918 (void) ct;
3919 znode_t *zp = VTOZ(vp);
3920 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3921 uint32_t gen;
3922 uint64_t gen64;
3923 uint64_t object = zp->z_id;
3924 zfid_short_t *zfid;
3925 int size, i, error;
3926
3927 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
3928 return (error);
3929
3930 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
3931 &gen64, sizeof (uint64_t))) != 0) {
3932 zfs_exit(zfsvfs, FTAG);
3933 return (error);
3934 }
3935
3936 gen = (uint32_t)gen64;
3937
3938 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
3939 fidp->fid_len = size;
3940
3941 zfid = (zfid_short_t *)fidp;
3942
3943 zfid->zf_len = size;
3944
3945 for (i = 0; i < sizeof (zfid->zf_object); i++)
3946 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
3947
3948 /* Must have a non-zero generation number to distinguish from .zfs */
3949 if (gen == 0)
3950 gen = 1;
3951 for (i = 0; i < sizeof (zfid->zf_gen); i++)
3952 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
3953
3954 if (size == LONG_FID_LEN) {
3955 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
3956 zfid_long_t *zlfid;
3957
3958 zlfid = (zfid_long_t *)fidp;
3959
3960 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
3961 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
3962
3963 /* XXX - this should be the generation number for the objset */
3964 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
3965 zlfid->zf_setgen[i] = 0;
3966 }
3967
3968 zfs_exit(zfsvfs, FTAG);
3969 return (0);
3970 }
3971
3972 static int
zfs_pathconf(vnode_t * vp,int cmd,ulong_t * valp,cred_t * cr,caller_context_t * ct)3973 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
3974 caller_context_t *ct)
3975 {
3976 znode_t *zp;
3977 zfsvfs_t *zfsvfs;
3978 int error;
3979
3980 switch (cmd) {
3981 case _PC_LINK_MAX:
3982 *valp = MIN(LONG_MAX, ZFS_LINK_MAX);
3983 return (0);
3984
3985 case _PC_FILESIZEBITS:
3986 *valp = 64;
3987 return (0);
3988 case _PC_MIN_HOLE_SIZE:
3989 *valp = (int)SPA_MINBLOCKSIZE;
3990 return (0);
3991 case _PC_ACL_EXTENDED:
3992 #if 0 /* POSIX ACLs are not implemented for ZFS on FreeBSD yet. */
3993 zp = VTOZ(vp);
3994 zfsvfs = zp->z_zfsvfs;
3995 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
3996 return (error);
3997 *valp = zfsvfs->z_acl_type == ZFSACLTYPE_POSIX ? 1 : 0;
3998 zfs_exit(zfsvfs, FTAG);
3999 #else
4000 *valp = 0;
4001 #endif
4002 return (0);
4003
4004 case _PC_ACL_NFS4:
4005 zp = VTOZ(vp);
4006 zfsvfs = zp->z_zfsvfs;
4007 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
4008 return (error);
4009 *valp = zfsvfs->z_acl_type == ZFS_ACLTYPE_NFSV4 ? 1 : 0;
4010 zfs_exit(zfsvfs, FTAG);
4011 return (0);
4012
4013 case _PC_ACL_PATH_MAX:
4014 *valp = ACL_MAX_ENTRIES;
4015 return (0);
4016
4017 default:
4018 return (EOPNOTSUPP);
4019 }
4020 }
4021
4022 static int
zfs_getpages(struct vnode * vp,vm_page_t * ma,int count,int * rbehind,int * rahead)4023 zfs_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind,
4024 int *rahead)
4025 {
4026 znode_t *zp = VTOZ(vp);
4027 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4028 zfs_locked_range_t *lr;
4029 vm_object_t object;
4030 off_t start, end, obj_size;
4031 uint_t blksz;
4032 int pgsin_b, pgsin_a;
4033 int error;
4034
4035 if (zfs_enter_verify_zp(zfsvfs, zp, FTAG) != 0)
4036 return (zfs_vm_pagerret_error);
4037
4038 start = IDX_TO_OFF(ma[0]->pindex);
4039 end = IDX_TO_OFF(ma[count - 1]->pindex + 1);
4040
4041 /*
4042 * Lock a range covering all required and optional pages.
4043 * Note that we need to handle the case of the block size growing.
4044 */
4045 for (;;) {
4046 blksz = zp->z_blksz;
4047 lr = zfs_rangelock_tryenter(&zp->z_rangelock,
4048 rounddown(start, blksz),
4049 roundup(end, blksz) - rounddown(start, blksz), RL_READER);
4050 if (lr == NULL) {
4051 if (rahead != NULL) {
4052 *rahead = 0;
4053 rahead = NULL;
4054 }
4055 if (rbehind != NULL) {
4056 *rbehind = 0;
4057 rbehind = NULL;
4058 }
4059 break;
4060 }
4061 if (blksz == zp->z_blksz)
4062 break;
4063 zfs_rangelock_exit(lr);
4064 }
4065
4066 object = ma[0]->object;
4067 zfs_vmobject_wlock(object);
4068 obj_size = object->un_pager.vnp.vnp_size;
4069 zfs_vmobject_wunlock(object);
4070 if (IDX_TO_OFF(ma[count - 1]->pindex) >= obj_size) {
4071 if (lr != NULL)
4072 zfs_rangelock_exit(lr);
4073 zfs_exit(zfsvfs, FTAG);
4074 return (zfs_vm_pagerret_bad);
4075 }
4076
4077 pgsin_b = 0;
4078 if (rbehind != NULL) {
4079 pgsin_b = OFF_TO_IDX(start - rounddown(start, blksz));
4080 pgsin_b = MIN(*rbehind, pgsin_b);
4081 }
4082
4083 pgsin_a = 0;
4084 if (rahead != NULL) {
4085 pgsin_a = OFF_TO_IDX(roundup(end, blksz) - end);
4086 if (end + IDX_TO_OFF(pgsin_a) >= obj_size)
4087 pgsin_a = OFF_TO_IDX(round_page(obj_size) - end);
4088 pgsin_a = MIN(*rahead, pgsin_a);
4089 }
4090
4091 /*
4092 * NB: we need to pass the exact byte size of the data that we expect
4093 * to read after accounting for the file size. This is required because
4094 * ZFS will panic if we request DMU to read beyond the end of the last
4095 * allocated block.
4096 */
4097 error = dmu_read_pages(zfsvfs->z_os, zp->z_id, ma, count, &pgsin_b,
4098 &pgsin_a, MIN(end, obj_size) - (end - PAGE_SIZE));
4099
4100 if (lr != NULL)
4101 zfs_rangelock_exit(lr);
4102 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4103
4104 dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, count*PAGE_SIZE);
4105
4106 zfs_exit(zfsvfs, FTAG);
4107
4108 if (error != 0)
4109 return (zfs_vm_pagerret_error);
4110
4111 VM_CNT_INC(v_vnodein);
4112 VM_CNT_ADD(v_vnodepgsin, count + pgsin_b + pgsin_a);
4113 if (rbehind != NULL)
4114 *rbehind = pgsin_b;
4115 if (rahead != NULL)
4116 *rahead = pgsin_a;
4117 return (zfs_vm_pagerret_ok);
4118 }
4119
4120 #ifndef _SYS_SYSPROTO_H_
4121 struct vop_getpages_args {
4122 struct vnode *a_vp;
4123 vm_page_t *a_m;
4124 int a_count;
4125 int *a_rbehind;
4126 int *a_rahead;
4127 };
4128 #endif
4129
4130 static int
zfs_freebsd_getpages(struct vop_getpages_args * ap)4131 zfs_freebsd_getpages(struct vop_getpages_args *ap)
4132 {
4133
4134 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind,
4135 ap->a_rahead));
4136 }
4137
4138 static int
zfs_putpages(struct vnode * vp,vm_page_t * ma,size_t len,int flags,int * rtvals)4139 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags,
4140 int *rtvals)
4141 {
4142 znode_t *zp = VTOZ(vp);
4143 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4144 zfs_locked_range_t *lr;
4145 dmu_tx_t *tx;
4146 struct sf_buf *sf;
4147 vm_object_t object;
4148 vm_page_t m;
4149 caddr_t va;
4150 size_t tocopy;
4151 size_t lo_len;
4152 vm_ooffset_t lo_off;
4153 vm_ooffset_t off;
4154 uint_t blksz;
4155 int ncount;
4156 int pcount;
4157 int err;
4158 int i;
4159
4160 object = vp->v_object;
4161 KASSERT(ma[0]->object == object, ("mismatching object"));
4162 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length"));
4163
4164 pcount = btoc(len);
4165 ncount = pcount;
4166 for (i = 0; i < pcount; i++)
4167 rtvals[i] = zfs_vm_pagerret_error;
4168
4169 if (zfs_enter_verify_zp(zfsvfs, zp, FTAG) != 0)
4170 return (zfs_vm_pagerret_error);
4171
4172 off = IDX_TO_OFF(ma[0]->pindex);
4173 blksz = zp->z_blksz;
4174 lo_off = rounddown(off, blksz);
4175 lo_len = roundup(len + (off - lo_off), blksz);
4176 lr = zfs_rangelock_enter(&zp->z_rangelock, lo_off, lo_len, RL_WRITER);
4177
4178 zfs_vmobject_wlock(object);
4179 if (len + off > object->un_pager.vnp.vnp_size) {
4180 if (object->un_pager.vnp.vnp_size > off) {
4181 int pgoff;
4182
4183 len = object->un_pager.vnp.vnp_size - off;
4184 ncount = btoc(len);
4185 if ((pgoff = (int)len & PAGE_MASK) != 0) {
4186 /*
4187 * If the object is locked and the following
4188 * conditions hold, then the page's dirty
4189 * field cannot be concurrently changed by a
4190 * pmap operation.
4191 */
4192 m = ma[ncount - 1];
4193 vm_page_assert_sbusied(m);
4194 KASSERT(!pmap_page_is_write_mapped(m),
4195 ("zfs_putpages: page %p is not read-only",
4196 m));
4197 vm_page_clear_dirty(m, pgoff, PAGE_SIZE -
4198 pgoff);
4199 }
4200 } else {
4201 len = 0;
4202 ncount = 0;
4203 }
4204 if (ncount < pcount) {
4205 for (i = ncount; i < pcount; i++) {
4206 rtvals[i] = zfs_vm_pagerret_bad;
4207 }
4208 }
4209 }
4210 zfs_vmobject_wunlock(object);
4211
4212 boolean_t commit = (flags & (zfs_vm_pagerput_sync |
4213 zfs_vm_pagerput_inval)) != 0 ||
4214 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS;
4215
4216 if (ncount == 0)
4217 goto out;
4218
4219 if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT, zp->z_uid) ||
4220 zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT, zp->z_gid) ||
4221 (zp->z_projid != ZFS_DEFAULT_PROJID &&
4222 zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT,
4223 zp->z_projid))) {
4224 goto out;
4225 }
4226
4227 tx = dmu_tx_create(zfsvfs->z_os);
4228 dmu_tx_hold_write(tx, zp->z_id, off, len);
4229
4230 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4231 zfs_sa_upgrade_txholds(tx, zp);
4232 err = dmu_tx_assign(tx, TXG_WAIT);
4233 if (err != 0) {
4234 dmu_tx_abort(tx);
4235 goto out;
4236 }
4237
4238 if (zp->z_blksz < PAGE_SIZE) {
4239 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) {
4240 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len;
4241 va = zfs_map_page(ma[i], &sf);
4242 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx);
4243 zfs_unmap_page(sf);
4244 }
4245 } else {
4246 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx);
4247 }
4248
4249 if (err == 0) {
4250 uint64_t mtime[2], ctime[2];
4251 sa_bulk_attr_t bulk[3];
4252 int count = 0;
4253
4254 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4255 &mtime, 16);
4256 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4257 &ctime, 16);
4258 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4259 &zp->z_pflags, 8);
4260 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
4261 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
4262 ASSERT0(err);
4263 /*
4264 * XXX we should be passing a callback to undirty
4265 * but that would make the locking messier
4266 */
4267 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off,
4268 len, commit, NULL, NULL);
4269
4270 zfs_vmobject_wlock(object);
4271 for (i = 0; i < ncount; i++) {
4272 rtvals[i] = zfs_vm_pagerret_ok;
4273 vm_page_undirty(ma[i]);
4274 }
4275 zfs_vmobject_wunlock(object);
4276 VM_CNT_INC(v_vnodeout);
4277 VM_CNT_ADD(v_vnodepgsout, ncount);
4278 }
4279 dmu_tx_commit(tx);
4280
4281 out:
4282 zfs_rangelock_exit(lr);
4283 if (commit)
4284 zil_commit(zfsvfs->z_log, zp->z_id);
4285
4286 dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, len);
4287
4288 zfs_exit(zfsvfs, FTAG);
4289 return (rtvals[0]);
4290 }
4291
4292 #ifndef _SYS_SYSPROTO_H_
4293 struct vop_putpages_args {
4294 struct vnode *a_vp;
4295 vm_page_t *a_m;
4296 int a_count;
4297 int a_sync;
4298 int *a_rtvals;
4299 };
4300 #endif
4301
4302 static int
zfs_freebsd_putpages(struct vop_putpages_args * ap)4303 zfs_freebsd_putpages(struct vop_putpages_args *ap)
4304 {
4305
4306 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync,
4307 ap->a_rtvals));
4308 }
4309
4310 #ifndef _SYS_SYSPROTO_H_
4311 struct vop_bmap_args {
4312 struct vnode *a_vp;
4313 daddr_t a_bn;
4314 struct bufobj **a_bop;
4315 daddr_t *a_bnp;
4316 int *a_runp;
4317 int *a_runb;
4318 };
4319 #endif
4320
4321 static int
zfs_freebsd_bmap(struct vop_bmap_args * ap)4322 zfs_freebsd_bmap(struct vop_bmap_args *ap)
4323 {
4324
4325 if (ap->a_bop != NULL)
4326 *ap->a_bop = &ap->a_vp->v_bufobj;
4327 if (ap->a_bnp != NULL)
4328 *ap->a_bnp = ap->a_bn;
4329 if (ap->a_runp != NULL)
4330 *ap->a_runp = 0;
4331 if (ap->a_runb != NULL)
4332 *ap->a_runb = 0;
4333
4334 return (0);
4335 }
4336
4337 #ifndef _SYS_SYSPROTO_H_
4338 struct vop_open_args {
4339 struct vnode *a_vp;
4340 int a_mode;
4341 struct ucred *a_cred;
4342 struct thread *a_td;
4343 };
4344 #endif
4345
4346 static int
zfs_freebsd_open(struct vop_open_args * ap)4347 zfs_freebsd_open(struct vop_open_args *ap)
4348 {
4349 vnode_t *vp = ap->a_vp;
4350 znode_t *zp = VTOZ(vp);
4351 int error;
4352
4353 error = zfs_open(&vp, ap->a_mode, ap->a_cred);
4354 if (error == 0)
4355 vnode_create_vobject(vp, zp->z_size, ap->a_td);
4356 return (error);
4357 }
4358
4359 #ifndef _SYS_SYSPROTO_H_
4360 struct vop_close_args {
4361 struct vnode *a_vp;
4362 int a_fflag;
4363 struct ucred *a_cred;
4364 struct thread *a_td;
4365 };
4366 #endif
4367
4368 static int
zfs_freebsd_close(struct vop_close_args * ap)4369 zfs_freebsd_close(struct vop_close_args *ap)
4370 {
4371
4372 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred));
4373 }
4374
4375 #ifndef _SYS_SYSPROTO_H_
4376 struct vop_ioctl_args {
4377 struct vnode *a_vp;
4378 ulong_t a_command;
4379 caddr_t a_data;
4380 int a_fflag;
4381 struct ucred *cred;
4382 struct thread *td;
4383 };
4384 #endif
4385
4386 static int
zfs_freebsd_ioctl(struct vop_ioctl_args * ap)4387 zfs_freebsd_ioctl(struct vop_ioctl_args *ap)
4388 {
4389
4390 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
4391 ap->a_fflag, ap->a_cred, NULL));
4392 }
4393
4394 static int
ioflags(int ioflags)4395 ioflags(int ioflags)
4396 {
4397 int flags = 0;
4398
4399 if (ioflags & IO_APPEND)
4400 flags |= O_APPEND;
4401 if (ioflags & IO_NDELAY)
4402 flags |= O_NONBLOCK;
4403 if (ioflags & IO_SYNC)
4404 flags |= O_SYNC;
4405
4406 return (flags);
4407 }
4408
4409 #ifndef _SYS_SYSPROTO_H_
4410 struct vop_read_args {
4411 struct vnode *a_vp;
4412 struct uio *a_uio;
4413 int a_ioflag;
4414 struct ucred *a_cred;
4415 };
4416 #endif
4417
4418 static int
zfs_freebsd_read(struct vop_read_args * ap)4419 zfs_freebsd_read(struct vop_read_args *ap)
4420 {
4421 zfs_uio_t uio;
4422 zfs_uio_init(&uio, ap->a_uio);
4423 return (zfs_read(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag),
4424 ap->a_cred));
4425 }
4426
4427 #ifndef _SYS_SYSPROTO_H_
4428 struct vop_write_args {
4429 struct vnode *a_vp;
4430 struct uio *a_uio;
4431 int a_ioflag;
4432 struct ucred *a_cred;
4433 };
4434 #endif
4435
4436 static int
zfs_freebsd_write(struct vop_write_args * ap)4437 zfs_freebsd_write(struct vop_write_args *ap)
4438 {
4439 zfs_uio_t uio;
4440 zfs_uio_init(&uio, ap->a_uio);
4441 return (zfs_write(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag),
4442 ap->a_cred));
4443 }
4444
4445 #if __FreeBSD_version >= 1300102
4446 /*
4447 * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see
4448 * the comment above cache_fplookup for details.
4449 */
4450 static int
zfs_freebsd_fplookup_vexec(struct vop_fplookup_vexec_args * v)4451 zfs_freebsd_fplookup_vexec(struct vop_fplookup_vexec_args *v)
4452 {
4453 vnode_t *vp;
4454 znode_t *zp;
4455 uint64_t pflags;
4456
4457 vp = v->a_vp;
4458 zp = VTOZ_SMR(vp);
4459 if (__predict_false(zp == NULL))
4460 return (EAGAIN);
4461 pflags = atomic_load_64(&zp->z_pflags);
4462 if (pflags & ZFS_AV_QUARANTINED)
4463 return (EAGAIN);
4464 if (pflags & ZFS_XATTR)
4465 return (EAGAIN);
4466 if ((pflags & ZFS_NO_EXECS_DENIED) == 0)
4467 return (EAGAIN);
4468 return (0);
4469 }
4470 #endif
4471
4472 #if __FreeBSD_version >= 1300139
4473 static int
zfs_freebsd_fplookup_symlink(struct vop_fplookup_symlink_args * v)4474 zfs_freebsd_fplookup_symlink(struct vop_fplookup_symlink_args *v)
4475 {
4476 vnode_t *vp;
4477 znode_t *zp;
4478 char *target;
4479
4480 vp = v->a_vp;
4481 zp = VTOZ_SMR(vp);
4482 if (__predict_false(zp == NULL)) {
4483 return (EAGAIN);
4484 }
4485
4486 target = atomic_load_consume_ptr(&zp->z_cached_symlink);
4487 if (target == NULL) {
4488 return (EAGAIN);
4489 }
4490 return (cache_symlink_resolve(v->a_fpl, target, strlen(target)));
4491 }
4492 #endif
4493
4494 #ifndef _SYS_SYSPROTO_H_
4495 struct vop_access_args {
4496 struct vnode *a_vp;
4497 accmode_t a_accmode;
4498 struct ucred *a_cred;
4499 struct thread *a_td;
4500 };
4501 #endif
4502
4503 static int
zfs_freebsd_access(struct vop_access_args * ap)4504 zfs_freebsd_access(struct vop_access_args *ap)
4505 {
4506 vnode_t *vp = ap->a_vp;
4507 znode_t *zp = VTOZ(vp);
4508 accmode_t accmode;
4509 int error = 0;
4510
4511
4512 if (ap->a_accmode == VEXEC) {
4513 if (zfs_fastaccesschk_execute(zp, ap->a_cred) == 0)
4514 return (0);
4515 }
4516
4517 /*
4518 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
4519 */
4520 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
4521 if (accmode != 0)
4522 error = zfs_access(zp, accmode, 0, ap->a_cred);
4523
4524 /*
4525 * VADMIN has to be handled by vaccess().
4526 */
4527 if (error == 0) {
4528 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
4529 if (accmode != 0) {
4530 #if __FreeBSD_version >= 1300105
4531 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
4532 zp->z_gid, accmode, ap->a_cred);
4533 #else
4534 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
4535 zp->z_gid, accmode, ap->a_cred, NULL);
4536 #endif
4537 }
4538 }
4539
4540 /*
4541 * For VEXEC, ensure that at least one execute bit is set for
4542 * non-directories.
4543 */
4544 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
4545 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
4546 error = EACCES;
4547 }
4548
4549 return (error);
4550 }
4551
4552 #ifndef _SYS_SYSPROTO_H_
4553 struct vop_lookup_args {
4554 struct vnode *a_dvp;
4555 struct vnode **a_vpp;
4556 struct componentname *a_cnp;
4557 };
4558 #endif
4559
4560 static int
zfs_freebsd_lookup(struct vop_lookup_args * ap,boolean_t cached)4561 zfs_freebsd_lookup(struct vop_lookup_args *ap, boolean_t cached)
4562 {
4563 struct componentname *cnp = ap->a_cnp;
4564 char nm[NAME_MAX + 1];
4565
4566 ASSERT3U(cnp->cn_namelen, <, sizeof (nm));
4567 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof (nm)));
4568
4569 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
4570 cnp->cn_cred, 0, cached));
4571 }
4572
4573 static int
zfs_freebsd_cachedlookup(struct vop_cachedlookup_args * ap)4574 zfs_freebsd_cachedlookup(struct vop_cachedlookup_args *ap)
4575 {
4576
4577 return (zfs_freebsd_lookup((struct vop_lookup_args *)ap, B_TRUE));
4578 }
4579
4580 #ifndef _SYS_SYSPROTO_H_
4581 struct vop_lookup_args {
4582 struct vnode *a_dvp;
4583 struct vnode **a_vpp;
4584 struct componentname *a_cnp;
4585 };
4586 #endif
4587
4588 static int
zfs_cache_lookup(struct vop_lookup_args * ap)4589 zfs_cache_lookup(struct vop_lookup_args *ap)
4590 {
4591 zfsvfs_t *zfsvfs;
4592
4593 zfsvfs = ap->a_dvp->v_mount->mnt_data;
4594 if (zfsvfs->z_use_namecache)
4595 return (vfs_cache_lookup(ap));
4596 else
4597 return (zfs_freebsd_lookup(ap, B_FALSE));
4598 }
4599
4600 #ifndef _SYS_SYSPROTO_H_
4601 struct vop_create_args {
4602 struct vnode *a_dvp;
4603 struct vnode **a_vpp;
4604 struct componentname *a_cnp;
4605 struct vattr *a_vap;
4606 };
4607 #endif
4608
4609 static int
zfs_freebsd_create(struct vop_create_args * ap)4610 zfs_freebsd_create(struct vop_create_args *ap)
4611 {
4612 zfsvfs_t *zfsvfs;
4613 struct componentname *cnp = ap->a_cnp;
4614 vattr_t *vap = ap->a_vap;
4615 znode_t *zp = NULL;
4616 int rc, mode;
4617
4618 #if __FreeBSD_version < 1400068
4619 ASSERT(cnp->cn_flags & SAVENAME);
4620 #endif
4621
4622 vattr_init_mask(vap);
4623 mode = vap->va_mode & ALLPERMS;
4624 zfsvfs = ap->a_dvp->v_mount->mnt_data;
4625 *ap->a_vpp = NULL;
4626
4627 rc = zfs_create(VTOZ(ap->a_dvp), cnp->cn_nameptr, vap, 0, mode,
4628 &zp, cnp->cn_cred, 0 /* flag */, NULL /* vsecattr */, NULL);
4629 if (rc == 0)
4630 *ap->a_vpp = ZTOV(zp);
4631 if (zfsvfs->z_use_namecache &&
4632 rc == 0 && (cnp->cn_flags & MAKEENTRY) != 0)
4633 cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
4634
4635 return (rc);
4636 }
4637
4638 #ifndef _SYS_SYSPROTO_H_
4639 struct vop_remove_args {
4640 struct vnode *a_dvp;
4641 struct vnode *a_vp;
4642 struct componentname *a_cnp;
4643 };
4644 #endif
4645
4646 static int
zfs_freebsd_remove(struct vop_remove_args * ap)4647 zfs_freebsd_remove(struct vop_remove_args *ap)
4648 {
4649
4650 #if __FreeBSD_version < 1400068
4651 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
4652 #endif
4653
4654 return (zfs_remove_(ap->a_dvp, ap->a_vp, ap->a_cnp->cn_nameptr,
4655 ap->a_cnp->cn_cred));
4656 }
4657
4658 #ifndef _SYS_SYSPROTO_H_
4659 struct vop_mkdir_args {
4660 struct vnode *a_dvp;
4661 struct vnode **a_vpp;
4662 struct componentname *a_cnp;
4663 struct vattr *a_vap;
4664 };
4665 #endif
4666
4667 static int
zfs_freebsd_mkdir(struct vop_mkdir_args * ap)4668 zfs_freebsd_mkdir(struct vop_mkdir_args *ap)
4669 {
4670 vattr_t *vap = ap->a_vap;
4671 znode_t *zp = NULL;
4672 int rc;
4673
4674 #if __FreeBSD_version < 1400068
4675 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
4676 #endif
4677
4678 vattr_init_mask(vap);
4679 *ap->a_vpp = NULL;
4680
4681 rc = zfs_mkdir(VTOZ(ap->a_dvp), ap->a_cnp->cn_nameptr, vap, &zp,
4682 ap->a_cnp->cn_cred, 0, NULL, NULL);
4683
4684 if (rc == 0)
4685 *ap->a_vpp = ZTOV(zp);
4686 return (rc);
4687 }
4688
4689 #ifndef _SYS_SYSPROTO_H_
4690 struct vop_rmdir_args {
4691 struct vnode *a_dvp;
4692 struct vnode *a_vp;
4693 struct componentname *a_cnp;
4694 };
4695 #endif
4696
4697 static int
zfs_freebsd_rmdir(struct vop_rmdir_args * ap)4698 zfs_freebsd_rmdir(struct vop_rmdir_args *ap)
4699 {
4700 struct componentname *cnp = ap->a_cnp;
4701
4702 #if __FreeBSD_version < 1400068
4703 ASSERT(cnp->cn_flags & SAVENAME);
4704 #endif
4705
4706 return (zfs_rmdir_(ap->a_dvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred));
4707 }
4708
4709 #ifndef _SYS_SYSPROTO_H_
4710 struct vop_readdir_args {
4711 struct vnode *a_vp;
4712 struct uio *a_uio;
4713 struct ucred *a_cred;
4714 int *a_eofflag;
4715 int *a_ncookies;
4716 cookie_t **a_cookies;
4717 };
4718 #endif
4719
4720 static int
zfs_freebsd_readdir(struct vop_readdir_args * ap)4721 zfs_freebsd_readdir(struct vop_readdir_args *ap)
4722 {
4723 zfs_uio_t uio;
4724 zfs_uio_init(&uio, ap->a_uio);
4725 return (zfs_readdir(ap->a_vp, &uio, ap->a_cred, ap->a_eofflag,
4726 ap->a_ncookies, ap->a_cookies));
4727 }
4728
4729 #ifndef _SYS_SYSPROTO_H_
4730 struct vop_fsync_args {
4731 struct vnode *a_vp;
4732 int a_waitfor;
4733 struct thread *a_td;
4734 };
4735 #endif
4736
4737 static int
zfs_freebsd_fsync(struct vop_fsync_args * ap)4738 zfs_freebsd_fsync(struct vop_fsync_args *ap)
4739 {
4740
4741 return (zfs_fsync(VTOZ(ap->a_vp), 0, ap->a_td->td_ucred));
4742 }
4743
4744 #ifndef _SYS_SYSPROTO_H_
4745 struct vop_getattr_args {
4746 struct vnode *a_vp;
4747 struct vattr *a_vap;
4748 struct ucred *a_cred;
4749 };
4750 #endif
4751
4752 static int
zfs_freebsd_getattr(struct vop_getattr_args * ap)4753 zfs_freebsd_getattr(struct vop_getattr_args *ap)
4754 {
4755 vattr_t *vap = ap->a_vap;
4756 xvattr_t xvap;
4757 ulong_t fflags = 0;
4758 int error;
4759
4760 xva_init(&xvap);
4761 xvap.xva_vattr = *vap;
4762 xvap.xva_vattr.va_mask |= AT_XVATTR;
4763
4764 /* Convert chflags into ZFS-type flags. */
4765 /* XXX: what about SF_SETTABLE?. */
4766 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
4767 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
4768 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
4769 XVA_SET_REQ(&xvap, XAT_NODUMP);
4770 XVA_SET_REQ(&xvap, XAT_READONLY);
4771 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
4772 XVA_SET_REQ(&xvap, XAT_SYSTEM);
4773 XVA_SET_REQ(&xvap, XAT_HIDDEN);
4774 XVA_SET_REQ(&xvap, XAT_REPARSE);
4775 XVA_SET_REQ(&xvap, XAT_OFFLINE);
4776 XVA_SET_REQ(&xvap, XAT_SPARSE);
4777
4778 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred);
4779 if (error != 0)
4780 return (error);
4781
4782 /* Convert ZFS xattr into chflags. */
4783 #define FLAG_CHECK(fflag, xflag, xfield) do { \
4784 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
4785 fflags |= (fflag); \
4786 } while (0)
4787 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
4788 xvap.xva_xoptattrs.xoa_immutable);
4789 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
4790 xvap.xva_xoptattrs.xoa_appendonly);
4791 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
4792 xvap.xva_xoptattrs.xoa_nounlink);
4793 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
4794 xvap.xva_xoptattrs.xoa_archive);
4795 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
4796 xvap.xva_xoptattrs.xoa_nodump);
4797 FLAG_CHECK(UF_READONLY, XAT_READONLY,
4798 xvap.xva_xoptattrs.xoa_readonly);
4799 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
4800 xvap.xva_xoptattrs.xoa_system);
4801 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
4802 xvap.xva_xoptattrs.xoa_hidden);
4803 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
4804 xvap.xva_xoptattrs.xoa_reparse);
4805 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
4806 xvap.xva_xoptattrs.xoa_offline);
4807 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
4808 xvap.xva_xoptattrs.xoa_sparse);
4809
4810 #undef FLAG_CHECK
4811 *vap = xvap.xva_vattr;
4812 vap->va_flags = fflags;
4813 return (0);
4814 }
4815
4816 #ifndef _SYS_SYSPROTO_H_
4817 struct vop_setattr_args {
4818 struct vnode *a_vp;
4819 struct vattr *a_vap;
4820 struct ucred *a_cred;
4821 };
4822 #endif
4823
4824 static int
zfs_freebsd_setattr(struct vop_setattr_args * ap)4825 zfs_freebsd_setattr(struct vop_setattr_args *ap)
4826 {
4827 vnode_t *vp = ap->a_vp;
4828 vattr_t *vap = ap->a_vap;
4829 cred_t *cred = ap->a_cred;
4830 xvattr_t xvap;
4831 ulong_t fflags;
4832 uint64_t zflags;
4833
4834 vattr_init_mask(vap);
4835 vap->va_mask &= ~AT_NOSET;
4836
4837 xva_init(&xvap);
4838 xvap.xva_vattr = *vap;
4839
4840 zflags = VTOZ(vp)->z_pflags;
4841
4842 if (vap->va_flags != VNOVAL) {
4843 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
4844 int error;
4845
4846 if (zfsvfs->z_use_fuids == B_FALSE)
4847 return (EOPNOTSUPP);
4848
4849 fflags = vap->va_flags;
4850 /*
4851 * XXX KDM
4852 * We need to figure out whether it makes sense to allow
4853 * UF_REPARSE through, since we don't really have other
4854 * facilities to handle reparse points and zfs_setattr()
4855 * doesn't currently allow setting that attribute anyway.
4856 */
4857 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
4858 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
4859 UF_OFFLINE|UF_SPARSE)) != 0)
4860 return (EOPNOTSUPP);
4861 /*
4862 * Unprivileged processes are not permitted to unset system
4863 * flags, or modify flags if any system flags are set.
4864 * Privileged non-jail processes may not modify system flags
4865 * if securelevel > 0 and any existing system flags are set.
4866 * Privileged jail processes behave like privileged non-jail
4867 * processes if the PR_ALLOW_CHFLAGS permission bit is set;
4868 * otherwise, they behave like unprivileged processes.
4869 */
4870 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
4871 spl_priv_check_cred(cred, PRIV_VFS_SYSFLAGS) == 0) {
4872 if (zflags &
4873 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
4874 error = securelevel_gt(cred, 0);
4875 if (error != 0)
4876 return (error);
4877 }
4878 } else {
4879 /*
4880 * Callers may only modify the file flags on
4881 * objects they have VADMIN rights for.
4882 */
4883 if ((error = VOP_ACCESS(vp, VADMIN, cred,
4884 curthread)) != 0)
4885 return (error);
4886 if (zflags &
4887 (ZFS_IMMUTABLE | ZFS_APPENDONLY |
4888 ZFS_NOUNLINK)) {
4889 return (EPERM);
4890 }
4891 if (fflags &
4892 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
4893 return (EPERM);
4894 }
4895 }
4896
4897 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
4898 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
4899 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
4900 XVA_SET_REQ(&xvap, (xflag)); \
4901 (xfield) = ((fflags & (fflag)) != 0); \
4902 } \
4903 } while (0)
4904 /* Convert chflags into ZFS-type flags. */
4905 /* XXX: what about SF_SETTABLE?. */
4906 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
4907 xvap.xva_xoptattrs.xoa_immutable);
4908 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
4909 xvap.xva_xoptattrs.xoa_appendonly);
4910 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
4911 xvap.xva_xoptattrs.xoa_nounlink);
4912 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
4913 xvap.xva_xoptattrs.xoa_archive);
4914 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
4915 xvap.xva_xoptattrs.xoa_nodump);
4916 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
4917 xvap.xva_xoptattrs.xoa_readonly);
4918 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
4919 xvap.xva_xoptattrs.xoa_system);
4920 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
4921 xvap.xva_xoptattrs.xoa_hidden);
4922 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
4923 xvap.xva_xoptattrs.xoa_reparse);
4924 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
4925 xvap.xva_xoptattrs.xoa_offline);
4926 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
4927 xvap.xva_xoptattrs.xoa_sparse);
4928 #undef FLAG_CHANGE
4929 }
4930 if (vap->va_birthtime.tv_sec != VNOVAL) {
4931 xvap.xva_vattr.va_mask |= AT_XVATTR;
4932 XVA_SET_REQ(&xvap, XAT_CREATETIME);
4933 }
4934 return (zfs_setattr(VTOZ(vp), (vattr_t *)&xvap, 0, cred, NULL));
4935 }
4936
4937 #ifndef _SYS_SYSPROTO_H_
4938 struct vop_rename_args {
4939 struct vnode *a_fdvp;
4940 struct vnode *a_fvp;
4941 struct componentname *a_fcnp;
4942 struct vnode *a_tdvp;
4943 struct vnode *a_tvp;
4944 struct componentname *a_tcnp;
4945 };
4946 #endif
4947
4948 static int
zfs_freebsd_rename(struct vop_rename_args * ap)4949 zfs_freebsd_rename(struct vop_rename_args *ap)
4950 {
4951 vnode_t *fdvp = ap->a_fdvp;
4952 vnode_t *fvp = ap->a_fvp;
4953 vnode_t *tdvp = ap->a_tdvp;
4954 vnode_t *tvp = ap->a_tvp;
4955 int error;
4956
4957 #if __FreeBSD_version < 1400068
4958 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
4959 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
4960 #endif
4961
4962 error = zfs_do_rename(fdvp, &fvp, ap->a_fcnp, tdvp, &tvp,
4963 ap->a_tcnp, ap->a_fcnp->cn_cred);
4964
4965 vrele(fdvp);
4966 vrele(fvp);
4967 vrele(tdvp);
4968 if (tvp != NULL)
4969 vrele(tvp);
4970
4971 return (error);
4972 }
4973
4974 #ifndef _SYS_SYSPROTO_H_
4975 struct vop_symlink_args {
4976 struct vnode *a_dvp;
4977 struct vnode **a_vpp;
4978 struct componentname *a_cnp;
4979 struct vattr *a_vap;
4980 char *a_target;
4981 };
4982 #endif
4983
4984 static int
zfs_freebsd_symlink(struct vop_symlink_args * ap)4985 zfs_freebsd_symlink(struct vop_symlink_args *ap)
4986 {
4987 struct componentname *cnp = ap->a_cnp;
4988 vattr_t *vap = ap->a_vap;
4989 znode_t *zp = NULL;
4990 #if __FreeBSD_version >= 1300139
4991 char *symlink;
4992 size_t symlink_len;
4993 #endif
4994 int rc;
4995
4996 #if __FreeBSD_version < 1400068
4997 ASSERT(cnp->cn_flags & SAVENAME);
4998 #endif
4999
5000 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
5001 vattr_init_mask(vap);
5002 *ap->a_vpp = NULL;
5003
5004 rc = zfs_symlink(VTOZ(ap->a_dvp), cnp->cn_nameptr, vap,
5005 ap->a_target, &zp, cnp->cn_cred, 0 /* flags */, NULL);
5006 if (rc == 0) {
5007 *ap->a_vpp = ZTOV(zp);
5008 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
5009 #if __FreeBSD_version >= 1300139
5010 MPASS(zp->z_cached_symlink == NULL);
5011 symlink_len = strlen(ap->a_target);
5012 symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK);
5013 if (symlink != NULL) {
5014 memcpy(symlink, ap->a_target, symlink_len);
5015 symlink[symlink_len] = '\0';
5016 atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink,
5017 (uintptr_t)symlink);
5018 }
5019 #endif
5020 }
5021 return (rc);
5022 }
5023
5024 #ifndef _SYS_SYSPROTO_H_
5025 struct vop_readlink_args {
5026 struct vnode *a_vp;
5027 struct uio *a_uio;
5028 struct ucred *a_cred;
5029 };
5030 #endif
5031
5032 static int
zfs_freebsd_readlink(struct vop_readlink_args * ap)5033 zfs_freebsd_readlink(struct vop_readlink_args *ap)
5034 {
5035 zfs_uio_t uio;
5036 int error;
5037 #if __FreeBSD_version >= 1300139
5038 znode_t *zp = VTOZ(ap->a_vp);
5039 char *symlink, *base;
5040 size_t symlink_len;
5041 bool trycache;
5042 #endif
5043
5044 zfs_uio_init(&uio, ap->a_uio);
5045 #if __FreeBSD_version >= 1300139
5046 trycache = false;
5047 if (zfs_uio_segflg(&uio) == UIO_SYSSPACE &&
5048 zfs_uio_iovcnt(&uio) == 1) {
5049 base = zfs_uio_iovbase(&uio, 0);
5050 symlink_len = zfs_uio_iovlen(&uio, 0);
5051 trycache = true;
5052 }
5053 #endif
5054 error = zfs_readlink(ap->a_vp, &uio, ap->a_cred, NULL);
5055 #if __FreeBSD_version >= 1300139
5056 if (atomic_load_ptr(&zp->z_cached_symlink) != NULL ||
5057 error != 0 || !trycache) {
5058 return (error);
5059 }
5060 symlink_len -= zfs_uio_resid(&uio);
5061 symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK);
5062 if (symlink != NULL) {
5063 memcpy(symlink, base, symlink_len);
5064 symlink[symlink_len] = '\0';
5065 if (!atomic_cmpset_rel_ptr((uintptr_t *)&zp->z_cached_symlink,
5066 (uintptr_t)NULL, (uintptr_t)symlink)) {
5067 cache_symlink_free(symlink, symlink_len + 1);
5068 }
5069 }
5070 #endif
5071 return (error);
5072 }
5073
5074 #ifndef _SYS_SYSPROTO_H_
5075 struct vop_link_args {
5076 struct vnode *a_tdvp;
5077 struct vnode *a_vp;
5078 struct componentname *a_cnp;
5079 };
5080 #endif
5081
5082 static int
zfs_freebsd_link(struct vop_link_args * ap)5083 zfs_freebsd_link(struct vop_link_args *ap)
5084 {
5085 struct componentname *cnp = ap->a_cnp;
5086 vnode_t *vp = ap->a_vp;
5087 vnode_t *tdvp = ap->a_tdvp;
5088
5089 if (tdvp->v_mount != vp->v_mount)
5090 return (EXDEV);
5091
5092 #if __FreeBSD_version < 1400068
5093 ASSERT(cnp->cn_flags & SAVENAME);
5094 #endif
5095
5096 return (zfs_link(VTOZ(tdvp), VTOZ(vp),
5097 cnp->cn_nameptr, cnp->cn_cred, 0));
5098 }
5099
5100 #ifndef _SYS_SYSPROTO_H_
5101 struct vop_inactive_args {
5102 struct vnode *a_vp;
5103 struct thread *a_td;
5104 };
5105 #endif
5106
5107 static int
zfs_freebsd_inactive(struct vop_inactive_args * ap)5108 zfs_freebsd_inactive(struct vop_inactive_args *ap)
5109 {
5110 vnode_t *vp = ap->a_vp;
5111
5112 #if __FreeBSD_version >= 1300123
5113 zfs_inactive(vp, curthread->td_ucred, NULL);
5114 #else
5115 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
5116 #endif
5117 return (0);
5118 }
5119
5120 #if __FreeBSD_version >= 1300042
5121 #ifndef _SYS_SYSPROTO_H_
5122 struct vop_need_inactive_args {
5123 struct vnode *a_vp;
5124 struct thread *a_td;
5125 };
5126 #endif
5127
5128 static int
zfs_freebsd_need_inactive(struct vop_need_inactive_args * ap)5129 zfs_freebsd_need_inactive(struct vop_need_inactive_args *ap)
5130 {
5131 vnode_t *vp = ap->a_vp;
5132 znode_t *zp = VTOZ(vp);
5133 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5134 int need;
5135
5136 if (vn_need_pageq_flush(vp))
5137 return (1);
5138
5139 if (!ZFS_TEARDOWN_INACTIVE_TRY_ENTER_READ(zfsvfs))
5140 return (1);
5141 need = (zp->z_sa_hdl == NULL || zp->z_unlinked || zp->z_atime_dirty);
5142 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
5143
5144 return (need);
5145 }
5146 #endif
5147
5148 #ifndef _SYS_SYSPROTO_H_
5149 struct vop_reclaim_args {
5150 struct vnode *a_vp;
5151 struct thread *a_td;
5152 };
5153 #endif
5154
5155 static int
zfs_freebsd_reclaim(struct vop_reclaim_args * ap)5156 zfs_freebsd_reclaim(struct vop_reclaim_args *ap)
5157 {
5158 vnode_t *vp = ap->a_vp;
5159 znode_t *zp = VTOZ(vp);
5160 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5161
5162 ASSERT3P(zp, !=, NULL);
5163
5164 #if __FreeBSD_version < 1300042
5165 /* Destroy the vm object and flush associated pages. */
5166 vnode_destroy_vobject(vp);
5167 #endif
5168 /*
5169 * z_teardown_inactive_lock protects from a race with
5170 * zfs_znode_dmu_fini in zfsvfs_teardown during
5171 * force unmount.
5172 */
5173 ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs);
5174 if (zp->z_sa_hdl == NULL)
5175 zfs_znode_free(zp);
5176 else
5177 zfs_zinactive(zp);
5178 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs);
5179
5180 vp->v_data = NULL;
5181 return (0);
5182 }
5183
5184 #ifndef _SYS_SYSPROTO_H_
5185 struct vop_fid_args {
5186 struct vnode *a_vp;
5187 struct fid *a_fid;
5188 };
5189 #endif
5190
5191 static int
zfs_freebsd_fid(struct vop_fid_args * ap)5192 zfs_freebsd_fid(struct vop_fid_args *ap)
5193 {
5194
5195 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
5196 }
5197
5198
5199 #ifndef _SYS_SYSPROTO_H_
5200 struct vop_pathconf_args {
5201 struct vnode *a_vp;
5202 int a_name;
5203 register_t *a_retval;
5204 } *ap;
5205 #endif
5206
5207 static int
zfs_freebsd_pathconf(struct vop_pathconf_args * ap)5208 zfs_freebsd_pathconf(struct vop_pathconf_args *ap)
5209 {
5210 ulong_t val;
5211 int error;
5212
5213 error = zfs_pathconf(ap->a_vp, ap->a_name, &val,
5214 curthread->td_ucred, NULL);
5215 if (error == 0) {
5216 *ap->a_retval = val;
5217 return (error);
5218 }
5219 if (error != EOPNOTSUPP)
5220 return (error);
5221
5222 switch (ap->a_name) {
5223 case _PC_NAME_MAX:
5224 *ap->a_retval = NAME_MAX;
5225 return (0);
5226 #if __FreeBSD_version >= 1400032
5227 case _PC_DEALLOC_PRESENT:
5228 *ap->a_retval = 1;
5229 return (0);
5230 #endif
5231 case _PC_PIPE_BUF:
5232 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) {
5233 *ap->a_retval = PIPE_BUF;
5234 return (0);
5235 }
5236 return (EINVAL);
5237 default:
5238 return (vop_stdpathconf(ap));
5239 }
5240 }
5241
5242 static int zfs_xattr_compat = 1;
5243
5244 static int
zfs_check_attrname(const char * name)5245 zfs_check_attrname(const char *name)
5246 {
5247 /* We don't allow '/' character in attribute name. */
5248 if (strchr(name, '/') != NULL)
5249 return (SET_ERROR(EINVAL));
5250 /* We don't allow attribute names that start with a namespace prefix. */
5251 if (ZFS_XA_NS_PREFIX_FORBIDDEN(name))
5252 return (SET_ERROR(EINVAL));
5253 return (0);
5254 }
5255
5256 /*
5257 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
5258 * extended attribute name:
5259 *
5260 * NAMESPACE XATTR_COMPAT PREFIX
5261 * system * freebsd:system:
5262 * user 1 (none, can be used to access ZFS
5263 * fsattr(5) attributes created on Solaris)
5264 * user 0 user.
5265 */
5266 static int
zfs_create_attrname(int attrnamespace,const char * name,char * attrname,size_t size,boolean_t compat)5267 zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
5268 size_t size, boolean_t compat)
5269 {
5270 const char *namespace, *prefix, *suffix;
5271
5272 memset(attrname, 0, size);
5273
5274 switch (attrnamespace) {
5275 case EXTATTR_NAMESPACE_USER:
5276 if (compat) {
5277 /*
5278 * This is the default namespace by which we can access
5279 * all attributes created on Solaris.
5280 */
5281 prefix = namespace = suffix = "";
5282 } else {
5283 /*
5284 * This is compatible with the user namespace encoding
5285 * on Linux prior to xattr_compat, but nothing
5286 * else.
5287 */
5288 prefix = "";
5289 namespace = "user";
5290 suffix = ".";
5291 }
5292 break;
5293 case EXTATTR_NAMESPACE_SYSTEM:
5294 prefix = "freebsd:";
5295 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
5296 suffix = ":";
5297 break;
5298 case EXTATTR_NAMESPACE_EMPTY:
5299 default:
5300 return (SET_ERROR(EINVAL));
5301 }
5302 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
5303 name) >= size) {
5304 return (SET_ERROR(ENAMETOOLONG));
5305 }
5306 return (0);
5307 }
5308
5309 static int
zfs_ensure_xattr_cached(znode_t * zp)5310 zfs_ensure_xattr_cached(znode_t *zp)
5311 {
5312 int error = 0;
5313
5314 ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock));
5315
5316 if (zp->z_xattr_cached != NULL)
5317 return (0);
5318
5319 if (rw_write_held(&zp->z_xattr_lock))
5320 return (zfs_sa_get_xattr(zp));
5321
5322 if (!rw_tryupgrade(&zp->z_xattr_lock)) {
5323 rw_exit(&zp->z_xattr_lock);
5324 rw_enter(&zp->z_xattr_lock, RW_WRITER);
5325 }
5326 if (zp->z_xattr_cached == NULL)
5327 error = zfs_sa_get_xattr(zp);
5328 rw_downgrade(&zp->z_xattr_lock);
5329 return (error);
5330 }
5331
5332 #ifndef _SYS_SYSPROTO_H_
5333 struct vop_getextattr {
5334 IN struct vnode *a_vp;
5335 IN int a_attrnamespace;
5336 IN const char *a_name;
5337 INOUT struct uio *a_uio;
5338 OUT size_t *a_size;
5339 IN struct ucred *a_cred;
5340 IN struct thread *a_td;
5341 };
5342 #endif
5343
5344 static int
zfs_getextattr_dir(struct vop_getextattr_args * ap,const char * attrname)5345 zfs_getextattr_dir(struct vop_getextattr_args *ap, const char *attrname)
5346 {
5347 struct thread *td = ap->a_td;
5348 struct nameidata nd;
5349 struct vattr va;
5350 vnode_t *xvp = NULL, *vp;
5351 int error, flags;
5352
5353 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred,
5354 LOOKUP_XATTR, B_FALSE);
5355 if (error != 0)
5356 return (error);
5357
5358 flags = FREAD;
5359 #if __FreeBSD_version < 1400043
5360 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
5361 xvp, td);
5362 #else
5363 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp);
5364 #endif
5365 error = vn_open_cred(&nd, &flags, 0, VN_OPEN_INVFS, ap->a_cred, NULL);
5366 if (error != 0)
5367 return (SET_ERROR(error));
5368 vp = nd.ni_vp;
5369 NDFREE_PNBUF(&nd);
5370
5371 if (ap->a_size != NULL) {
5372 error = VOP_GETATTR(vp, &va, ap->a_cred);
5373 if (error == 0)
5374 *ap->a_size = (size_t)va.va_size;
5375 } else if (ap->a_uio != NULL)
5376 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
5377
5378 VOP_UNLOCK1(vp);
5379 vn_close(vp, flags, ap->a_cred, td);
5380 return (error);
5381 }
5382
5383 static int
zfs_getextattr_sa(struct vop_getextattr_args * ap,const char * attrname)5384 zfs_getextattr_sa(struct vop_getextattr_args *ap, const char *attrname)
5385 {
5386 znode_t *zp = VTOZ(ap->a_vp);
5387 uchar_t *nv_value;
5388 uint_t nv_size;
5389 int error;
5390
5391 error = zfs_ensure_xattr_cached(zp);
5392 if (error != 0)
5393 return (error);
5394
5395 ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock));
5396 ASSERT3P(zp->z_xattr_cached, !=, NULL);
5397
5398 error = nvlist_lookup_byte_array(zp->z_xattr_cached, attrname,
5399 &nv_value, &nv_size);
5400 if (error != 0)
5401 return (SET_ERROR(error));
5402
5403 if (ap->a_size != NULL)
5404 *ap->a_size = nv_size;
5405 else if (ap->a_uio != NULL)
5406 error = uiomove(nv_value, nv_size, ap->a_uio);
5407 if (error != 0)
5408 return (SET_ERROR(error));
5409
5410 return (0);
5411 }
5412
5413 static int
zfs_getextattr_impl(struct vop_getextattr_args * ap,boolean_t compat)5414 zfs_getextattr_impl(struct vop_getextattr_args *ap, boolean_t compat)
5415 {
5416 znode_t *zp = VTOZ(ap->a_vp);
5417 zfsvfs_t *zfsvfs = ZTOZSB(zp);
5418 char attrname[EXTATTR_MAXNAMELEN+1];
5419 int error;
5420
5421 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5422 sizeof (attrname), compat);
5423 if (error != 0)
5424 return (error);
5425
5426 error = ENOENT;
5427 if (zfsvfs->z_use_sa && zp->z_is_sa)
5428 error = zfs_getextattr_sa(ap, attrname);
5429 if (error == ENOENT)
5430 error = zfs_getextattr_dir(ap, attrname);
5431 return (error);
5432 }
5433
5434 /*
5435 * Vnode operation to retrieve a named extended attribute.
5436 */
5437 static int
zfs_getextattr(struct vop_getextattr_args * ap)5438 zfs_getextattr(struct vop_getextattr_args *ap)
5439 {
5440 znode_t *zp = VTOZ(ap->a_vp);
5441 zfsvfs_t *zfsvfs = ZTOZSB(zp);
5442 int error;
5443
5444 /*
5445 * If the xattr property is off, refuse the request.
5446 */
5447 if (!(zfsvfs->z_flags & ZSB_XATTR))
5448 return (SET_ERROR(EOPNOTSUPP));
5449
5450 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5451 ap->a_cred, ap->a_td, VREAD);
5452 if (error != 0)
5453 return (SET_ERROR(error));
5454
5455 error = zfs_check_attrname(ap->a_name);
5456 if (error != 0)
5457 return (error);
5458
5459 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
5460 return (error);
5461 error = ENOENT;
5462 rw_enter(&zp->z_xattr_lock, RW_READER);
5463
5464 error = zfs_getextattr_impl(ap, zfs_xattr_compat);
5465 if ((error == ENOENT || error == ENOATTR) &&
5466 ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) {
5467 /*
5468 * Fall back to the alternate namespace format if we failed to
5469 * find a user xattr.
5470 */
5471 error = zfs_getextattr_impl(ap, !zfs_xattr_compat);
5472 }
5473
5474 rw_exit(&zp->z_xattr_lock);
5475 zfs_exit(zfsvfs, FTAG);
5476 if (error == ENOENT)
5477 error = SET_ERROR(ENOATTR);
5478 return (error);
5479 }
5480
5481 #ifndef _SYS_SYSPROTO_H_
5482 struct vop_deleteextattr {
5483 IN struct vnode *a_vp;
5484 IN int a_attrnamespace;
5485 IN const char *a_name;
5486 IN struct ucred *a_cred;
5487 IN struct thread *a_td;
5488 };
5489 #endif
5490
5491 static int
zfs_deleteextattr_dir(struct vop_deleteextattr_args * ap,const char * attrname)5492 zfs_deleteextattr_dir(struct vop_deleteextattr_args *ap, const char *attrname)
5493 {
5494 struct nameidata nd;
5495 vnode_t *xvp = NULL, *vp;
5496 int error;
5497
5498 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred,
5499 LOOKUP_XATTR, B_FALSE);
5500 if (error != 0)
5501 return (error);
5502
5503 #if __FreeBSD_version < 1400043
5504 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
5505 UIO_SYSSPACE, attrname, xvp, ap->a_td);
5506 #else
5507 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
5508 UIO_SYSSPACE, attrname, xvp);
5509 #endif
5510 error = namei(&nd);
5511 if (error != 0)
5512 return (SET_ERROR(error));
5513
5514 vp = nd.ni_vp;
5515 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
5516 NDFREE_PNBUF(&nd);
5517
5518 vput(nd.ni_dvp);
5519 if (vp == nd.ni_dvp)
5520 vrele(vp);
5521 else
5522 vput(vp);
5523
5524 return (error);
5525 }
5526
5527 static int
zfs_deleteextattr_sa(struct vop_deleteextattr_args * ap,const char * attrname)5528 zfs_deleteextattr_sa(struct vop_deleteextattr_args *ap, const char *attrname)
5529 {
5530 znode_t *zp = VTOZ(ap->a_vp);
5531 nvlist_t *nvl;
5532 int error;
5533
5534 error = zfs_ensure_xattr_cached(zp);
5535 if (error != 0)
5536 return (error);
5537
5538 ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock));
5539 ASSERT3P(zp->z_xattr_cached, !=, NULL);
5540
5541 nvl = zp->z_xattr_cached;
5542 error = nvlist_remove(nvl, attrname, DATA_TYPE_BYTE_ARRAY);
5543 if (error != 0)
5544 error = SET_ERROR(error);
5545 else
5546 error = zfs_sa_set_xattr(zp, attrname, NULL, 0);
5547 if (error != 0) {
5548 zp->z_xattr_cached = NULL;
5549 nvlist_free(nvl);
5550 }
5551 return (error);
5552 }
5553
5554 static int
zfs_deleteextattr_impl(struct vop_deleteextattr_args * ap,boolean_t compat)5555 zfs_deleteextattr_impl(struct vop_deleteextattr_args *ap, boolean_t compat)
5556 {
5557 znode_t *zp = VTOZ(ap->a_vp);
5558 zfsvfs_t *zfsvfs = ZTOZSB(zp);
5559 char attrname[EXTATTR_MAXNAMELEN+1];
5560 int error;
5561
5562 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5563 sizeof (attrname), compat);
5564 if (error != 0)
5565 return (error);
5566
5567 error = ENOENT;
5568 if (zfsvfs->z_use_sa && zp->z_is_sa)
5569 error = zfs_deleteextattr_sa(ap, attrname);
5570 if (error == ENOENT)
5571 error = zfs_deleteextattr_dir(ap, attrname);
5572 return (error);
5573 }
5574
5575 /*
5576 * Vnode operation to remove a named attribute.
5577 */
5578 static int
zfs_deleteextattr(struct vop_deleteextattr_args * ap)5579 zfs_deleteextattr(struct vop_deleteextattr_args *ap)
5580 {
5581 znode_t *zp = VTOZ(ap->a_vp);
5582 zfsvfs_t *zfsvfs = ZTOZSB(zp);
5583 int error;
5584
5585 /*
5586 * If the xattr property is off, refuse the request.
5587 */
5588 if (!(zfsvfs->z_flags & ZSB_XATTR))
5589 return (SET_ERROR(EOPNOTSUPP));
5590
5591 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5592 ap->a_cred, ap->a_td, VWRITE);
5593 if (error != 0)
5594 return (SET_ERROR(error));
5595
5596 error = zfs_check_attrname(ap->a_name);
5597 if (error != 0)
5598 return (error);
5599
5600 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
5601 return (error);
5602 rw_enter(&zp->z_xattr_lock, RW_WRITER);
5603
5604 error = zfs_deleteextattr_impl(ap, zfs_xattr_compat);
5605 if ((error == ENOENT || error == ENOATTR) &&
5606 ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) {
5607 /*
5608 * Fall back to the alternate namespace format if we failed to
5609 * find a user xattr.
5610 */
5611 error = zfs_deleteextattr_impl(ap, !zfs_xattr_compat);
5612 }
5613
5614 rw_exit(&zp->z_xattr_lock);
5615 zfs_exit(zfsvfs, FTAG);
5616 if (error == ENOENT)
5617 error = SET_ERROR(ENOATTR);
5618 return (error);
5619 }
5620
5621 #ifndef _SYS_SYSPROTO_H_
5622 struct vop_setextattr {
5623 IN struct vnode *a_vp;
5624 IN int a_attrnamespace;
5625 IN const char *a_name;
5626 INOUT struct uio *a_uio;
5627 IN struct ucred *a_cred;
5628 IN struct thread *a_td;
5629 };
5630 #endif
5631
5632 static int
zfs_setextattr_dir(struct vop_setextattr_args * ap,const char * attrname)5633 zfs_setextattr_dir(struct vop_setextattr_args *ap, const char *attrname)
5634 {
5635 struct thread *td = ap->a_td;
5636 struct nameidata nd;
5637 struct vattr va;
5638 vnode_t *xvp = NULL, *vp;
5639 int error, flags;
5640
5641 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred,
5642 LOOKUP_XATTR | CREATE_XATTR_DIR, B_FALSE);
5643 if (error != 0)
5644 return (error);
5645
5646 flags = FFLAGS(O_WRONLY | O_CREAT);
5647 #if __FreeBSD_version < 1400043
5648 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp, td);
5649 #else
5650 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp);
5651 #endif
5652 error = vn_open_cred(&nd, &flags, 0600, VN_OPEN_INVFS, ap->a_cred,
5653 NULL);
5654 if (error != 0)
5655 return (SET_ERROR(error));
5656 vp = nd.ni_vp;
5657 NDFREE_PNBUF(&nd);
5658
5659 VATTR_NULL(&va);
5660 va.va_size = 0;
5661 error = VOP_SETATTR(vp, &va, ap->a_cred);
5662 if (error == 0)
5663 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred);
5664
5665 VOP_UNLOCK1(vp);
5666 vn_close(vp, flags, ap->a_cred, td);
5667 return (error);
5668 }
5669
5670 static int
zfs_setextattr_sa(struct vop_setextattr_args * ap,const char * attrname)5671 zfs_setextattr_sa(struct vop_setextattr_args *ap, const char *attrname)
5672 {
5673 znode_t *zp = VTOZ(ap->a_vp);
5674 nvlist_t *nvl;
5675 size_t sa_size;
5676 int error;
5677
5678 error = zfs_ensure_xattr_cached(zp);
5679 if (error != 0)
5680 return (error);
5681
5682 ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock));
5683 ASSERT3P(zp->z_xattr_cached, !=, NULL);
5684
5685 nvl = zp->z_xattr_cached;
5686 size_t entry_size = ap->a_uio->uio_resid;
5687 if (entry_size > DXATTR_MAX_ENTRY_SIZE)
5688 return (SET_ERROR(EFBIG));
5689 error = nvlist_size(nvl, &sa_size, NV_ENCODE_XDR);
5690 if (error != 0)
5691 return (SET_ERROR(error));
5692 if (sa_size > DXATTR_MAX_SA_SIZE)
5693 return (SET_ERROR(EFBIG));
5694 uchar_t *buf = kmem_alloc(entry_size, KM_SLEEP);
5695 error = uiomove(buf, entry_size, ap->a_uio);
5696 if (error != 0) {
5697 error = SET_ERROR(error);
5698 } else {
5699 error = nvlist_add_byte_array(nvl, attrname, buf, entry_size);
5700 if (error != 0)
5701 error = SET_ERROR(error);
5702 }
5703 if (error == 0)
5704 error = zfs_sa_set_xattr(zp, attrname, buf, entry_size);
5705 kmem_free(buf, entry_size);
5706 if (error != 0) {
5707 zp->z_xattr_cached = NULL;
5708 nvlist_free(nvl);
5709 }
5710 return (error);
5711 }
5712
5713 static int
zfs_setextattr_impl(struct vop_setextattr_args * ap,boolean_t compat)5714 zfs_setextattr_impl(struct vop_setextattr_args *ap, boolean_t compat)
5715 {
5716 znode_t *zp = VTOZ(ap->a_vp);
5717 zfsvfs_t *zfsvfs = ZTOZSB(zp);
5718 char attrname[EXTATTR_MAXNAMELEN+1];
5719 int error;
5720
5721 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
5722 sizeof (attrname), compat);
5723 if (error != 0)
5724 return (error);
5725
5726 struct vop_deleteextattr_args vda = {
5727 .a_vp = ap->a_vp,
5728 .a_attrnamespace = ap->a_attrnamespace,
5729 .a_name = ap->a_name,
5730 .a_cred = ap->a_cred,
5731 .a_td = ap->a_td,
5732 };
5733 error = ENOENT;
5734 if (zfsvfs->z_use_sa && zp->z_is_sa && zfsvfs->z_xattr_sa) {
5735 error = zfs_setextattr_sa(ap, attrname);
5736 if (error == 0) {
5737 /*
5738 * Successfully put into SA, we need to clear the one
5739 * in dir if present.
5740 */
5741 zfs_deleteextattr_dir(&vda, attrname);
5742 }
5743 }
5744 if (error != 0) {
5745 error = zfs_setextattr_dir(ap, attrname);
5746 if (error == 0 && zp->z_is_sa) {
5747 /*
5748 * Successfully put into dir, we need to clear the one
5749 * in SA if present.
5750 */
5751 zfs_deleteextattr_sa(&vda, attrname);
5752 }
5753 }
5754 if (error == 0 && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) {
5755 /*
5756 * Also clear all versions of the alternate compat name.
5757 */
5758 zfs_deleteextattr_impl(&vda, !compat);
5759 }
5760 return (error);
5761 }
5762
5763 /*
5764 * Vnode operation to set a named attribute.
5765 */
5766 static int
zfs_setextattr(struct vop_setextattr_args * ap)5767 zfs_setextattr(struct vop_setextattr_args *ap)
5768 {
5769 znode_t *zp = VTOZ(ap->a_vp);
5770 zfsvfs_t *zfsvfs = ZTOZSB(zp);
5771 int error;
5772
5773 /*
5774 * If the xattr property is off, refuse the request.
5775 */
5776 if (!(zfsvfs->z_flags & ZSB_XATTR))
5777 return (SET_ERROR(EOPNOTSUPP));
5778
5779 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5780 ap->a_cred, ap->a_td, VWRITE);
5781 if (error != 0)
5782 return (SET_ERROR(error));
5783
5784 error = zfs_check_attrname(ap->a_name);
5785 if (error != 0)
5786 return (error);
5787
5788 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
5789 return (error);
5790 rw_enter(&zp->z_xattr_lock, RW_WRITER);
5791
5792 error = zfs_setextattr_impl(ap, zfs_xattr_compat);
5793
5794 rw_exit(&zp->z_xattr_lock);
5795 zfs_exit(zfsvfs, FTAG);
5796 return (error);
5797 }
5798
5799 #ifndef _SYS_SYSPROTO_H_
5800 struct vop_listextattr {
5801 IN struct vnode *a_vp;
5802 IN int a_attrnamespace;
5803 INOUT struct uio *a_uio;
5804 OUT size_t *a_size;
5805 IN struct ucred *a_cred;
5806 IN struct thread *a_td;
5807 };
5808 #endif
5809
5810 static int
zfs_listextattr_dir(struct vop_listextattr_args * ap,const char * attrprefix)5811 zfs_listextattr_dir(struct vop_listextattr_args *ap, const char *attrprefix)
5812 {
5813 struct thread *td = ap->a_td;
5814 struct nameidata nd;
5815 uint8_t dirbuf[sizeof (struct dirent)];
5816 struct iovec aiov;
5817 struct uio auio;
5818 vnode_t *xvp = NULL, *vp;
5819 int error, eof;
5820
5821 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred,
5822 LOOKUP_XATTR, B_FALSE);
5823 if (error != 0) {
5824 /*
5825 * ENOATTR means that the EA directory does not yet exist,
5826 * i.e. there are no extended attributes there.
5827 */
5828 if (error == ENOATTR)
5829 error = 0;
5830 return (error);
5831 }
5832
5833 #if __FreeBSD_version < 1400043
5834 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
5835 UIO_SYSSPACE, ".", xvp, td);
5836 #else
5837 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
5838 UIO_SYSSPACE, ".", xvp);
5839 #endif
5840 error = namei(&nd);
5841 if (error != 0)
5842 return (SET_ERROR(error));
5843 vp = nd.ni_vp;
5844 NDFREE_PNBUF(&nd);
5845
5846 auio.uio_iov = &aiov;
5847 auio.uio_iovcnt = 1;
5848 auio.uio_segflg = UIO_SYSSPACE;
5849 auio.uio_td = td;
5850 auio.uio_rw = UIO_READ;
5851 auio.uio_offset = 0;
5852
5853 size_t plen = strlen(attrprefix);
5854
5855 do {
5856 aiov.iov_base = (void *)dirbuf;
5857 aiov.iov_len = sizeof (dirbuf);
5858 auio.uio_resid = sizeof (dirbuf);
5859 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
5860 if (error != 0)
5861 break;
5862 int done = sizeof (dirbuf) - auio.uio_resid;
5863 for (int pos = 0; pos < done; ) {
5864 struct dirent *dp = (struct dirent *)(dirbuf + pos);
5865 pos += dp->d_reclen;
5866 /*
5867 * XXX: Temporarily we also accept DT_UNKNOWN, as this
5868 * is what we get when attribute was created on Solaris.
5869 */
5870 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
5871 continue;
5872 else if (plen == 0 &&
5873 ZFS_XA_NS_PREFIX_FORBIDDEN(dp->d_name))
5874 continue;
5875 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
5876 continue;
5877 uint8_t nlen = dp->d_namlen - plen;
5878 if (ap->a_size != NULL) {
5879 *ap->a_size += 1 + nlen;
5880 } else if (ap->a_uio != NULL) {
5881 /*
5882 * Format of extattr name entry is one byte for
5883 * length and the rest for name.
5884 */
5885 error = uiomove(&nlen, 1, ap->a_uio);
5886 if (error == 0) {
5887 char *namep = dp->d_name + plen;
5888 error = uiomove(namep, nlen, ap->a_uio);
5889 }
5890 if (error != 0) {
5891 error = SET_ERROR(error);
5892 break;
5893 }
5894 }
5895 }
5896 } while (!eof && error == 0);
5897
5898 vput(vp);
5899 return (error);
5900 }
5901
5902 static int
zfs_listextattr_sa(struct vop_listextattr_args * ap,const char * attrprefix)5903 zfs_listextattr_sa(struct vop_listextattr_args *ap, const char *attrprefix)
5904 {
5905 znode_t *zp = VTOZ(ap->a_vp);
5906 int error;
5907
5908 error = zfs_ensure_xattr_cached(zp);
5909 if (error != 0)
5910 return (error);
5911
5912 ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock));
5913 ASSERT3P(zp->z_xattr_cached, !=, NULL);
5914
5915 size_t plen = strlen(attrprefix);
5916 nvpair_t *nvp = NULL;
5917 while ((nvp = nvlist_next_nvpair(zp->z_xattr_cached, nvp)) != NULL) {
5918 ASSERT3U(nvpair_type(nvp), ==, DATA_TYPE_BYTE_ARRAY);
5919
5920 const char *name = nvpair_name(nvp);
5921 if (plen == 0 && ZFS_XA_NS_PREFIX_FORBIDDEN(name))
5922 continue;
5923 else if (strncmp(name, attrprefix, plen) != 0)
5924 continue;
5925 uint8_t nlen = strlen(name) - plen;
5926 if (ap->a_size != NULL) {
5927 *ap->a_size += 1 + nlen;
5928 } else if (ap->a_uio != NULL) {
5929 /*
5930 * Format of extattr name entry is one byte for
5931 * length and the rest for name.
5932 */
5933 error = uiomove(&nlen, 1, ap->a_uio);
5934 if (error == 0) {
5935 char *namep = __DECONST(char *, name) + plen;
5936 error = uiomove(namep, nlen, ap->a_uio);
5937 }
5938 if (error != 0) {
5939 error = SET_ERROR(error);
5940 break;
5941 }
5942 }
5943 }
5944
5945 return (error);
5946 }
5947
5948 static int
zfs_listextattr_impl(struct vop_listextattr_args * ap,boolean_t compat)5949 zfs_listextattr_impl(struct vop_listextattr_args *ap, boolean_t compat)
5950 {
5951 znode_t *zp = VTOZ(ap->a_vp);
5952 zfsvfs_t *zfsvfs = ZTOZSB(zp);
5953 char attrprefix[16];
5954 int error;
5955
5956 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
5957 sizeof (attrprefix), compat);
5958 if (error != 0)
5959 return (error);
5960
5961 if (zfsvfs->z_use_sa && zp->z_is_sa)
5962 error = zfs_listextattr_sa(ap, attrprefix);
5963 if (error == 0)
5964 error = zfs_listextattr_dir(ap, attrprefix);
5965 return (error);
5966 }
5967
5968 /*
5969 * Vnode operation to retrieve extended attributes on a vnode.
5970 */
5971 static int
zfs_listextattr(struct vop_listextattr_args * ap)5972 zfs_listextattr(struct vop_listextattr_args *ap)
5973 {
5974 znode_t *zp = VTOZ(ap->a_vp);
5975 zfsvfs_t *zfsvfs = ZTOZSB(zp);
5976 int error;
5977
5978 if (ap->a_size != NULL)
5979 *ap->a_size = 0;
5980
5981 /*
5982 * If the xattr property is off, refuse the request.
5983 */
5984 if (!(zfsvfs->z_flags & ZSB_XATTR))
5985 return (SET_ERROR(EOPNOTSUPP));
5986
5987 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
5988 ap->a_cred, ap->a_td, VREAD);
5989 if (error != 0)
5990 return (SET_ERROR(error));
5991
5992 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
5993 return (error);
5994 rw_enter(&zp->z_xattr_lock, RW_READER);
5995
5996 error = zfs_listextattr_impl(ap, zfs_xattr_compat);
5997 if (error == 0 && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) {
5998 /* Also list user xattrs with the alternate format. */
5999 error = zfs_listextattr_impl(ap, !zfs_xattr_compat);
6000 }
6001
6002 rw_exit(&zp->z_xattr_lock);
6003 zfs_exit(zfsvfs, FTAG);
6004 return (error);
6005 }
6006
6007 #ifndef _SYS_SYSPROTO_H_
6008 struct vop_getacl_args {
6009 struct vnode *vp;
6010 acl_type_t type;
6011 struct acl *aclp;
6012 struct ucred *cred;
6013 struct thread *td;
6014 };
6015 #endif
6016
6017 static int
zfs_freebsd_getacl(struct vop_getacl_args * ap)6018 zfs_freebsd_getacl(struct vop_getacl_args *ap)
6019 {
6020 int error;
6021 vsecattr_t vsecattr;
6022
6023 if (ap->a_type != ACL_TYPE_NFS4)
6024 return (EINVAL);
6025
6026 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6027 if ((error = zfs_getsecattr(VTOZ(ap->a_vp),
6028 &vsecattr, 0, ap->a_cred)))
6029 return (error);
6030
6031 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp,
6032 vsecattr.vsa_aclcnt);
6033 if (vsecattr.vsa_aclentp != NULL)
6034 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6035
6036 return (error);
6037 }
6038
6039 #ifndef _SYS_SYSPROTO_H_
6040 struct vop_setacl_args {
6041 struct vnode *vp;
6042 acl_type_t type;
6043 struct acl *aclp;
6044 struct ucred *cred;
6045 struct thread *td;
6046 };
6047 #endif
6048
6049 static int
zfs_freebsd_setacl(struct vop_setacl_args * ap)6050 zfs_freebsd_setacl(struct vop_setacl_args *ap)
6051 {
6052 int error;
6053 vsecattr_t vsecattr;
6054 int aclbsize; /* size of acl list in bytes */
6055 aclent_t *aaclp;
6056
6057 if (ap->a_type != ACL_TYPE_NFS4)
6058 return (EINVAL);
6059
6060 if (ap->a_aclp == NULL)
6061 return (EINVAL);
6062
6063 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6064 return (EINVAL);
6065
6066 /*
6067 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6068 * splitting every entry into two and appending "canonical six"
6069 * entries at the end. Don't allow for setting an ACL that would
6070 * cause chmod(2) to run out of ACL entries.
6071 */
6072 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6073 return (ENOSPC);
6074
6075 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6076 if (error != 0)
6077 return (error);
6078
6079 vsecattr.vsa_mask = VSA_ACE;
6080 aclbsize = ap->a_aclp->acl_cnt * sizeof (ace_t);
6081 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6082 aaclp = vsecattr.vsa_aclentp;
6083 vsecattr.vsa_aclentsz = aclbsize;
6084
6085 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6086 error = zfs_setsecattr(VTOZ(ap->a_vp), &vsecattr, 0, ap->a_cred);
6087 kmem_free(aaclp, aclbsize);
6088
6089 return (error);
6090 }
6091
6092 #ifndef _SYS_SYSPROTO_H_
6093 struct vop_aclcheck_args {
6094 struct vnode *vp;
6095 acl_type_t type;
6096 struct acl *aclp;
6097 struct ucred *cred;
6098 struct thread *td;
6099 };
6100 #endif
6101
6102 static int
zfs_freebsd_aclcheck(struct vop_aclcheck_args * ap)6103 zfs_freebsd_aclcheck(struct vop_aclcheck_args *ap)
6104 {
6105
6106 return (EOPNOTSUPP);
6107 }
6108
6109 static int
zfs_vptocnp(struct vop_vptocnp_args * ap)6110 zfs_vptocnp(struct vop_vptocnp_args *ap)
6111 {
6112 vnode_t *covered_vp;
6113 vnode_t *vp = ap->a_vp;
6114 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
6115 znode_t *zp = VTOZ(vp);
6116 int ltype;
6117 int error;
6118
6119 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
6120 return (error);
6121
6122 /*
6123 * If we are a snapshot mounted under .zfs, run the operation
6124 * on the covered vnode.
6125 */
6126 if (zp->z_id != zfsvfs->z_root || zfsvfs->z_parent == zfsvfs) {
6127 char name[MAXNAMLEN + 1];
6128 znode_t *dzp;
6129 size_t len;
6130
6131 error = zfs_znode_parent_and_name(zp, &dzp, name);
6132 if (error == 0) {
6133 len = strlen(name);
6134 if (*ap->a_buflen < len)
6135 error = SET_ERROR(ENOMEM);
6136 }
6137 if (error == 0) {
6138 *ap->a_buflen -= len;
6139 memcpy(ap->a_buf + *ap->a_buflen, name, len);
6140 *ap->a_vpp = ZTOV(dzp);
6141 }
6142 zfs_exit(zfsvfs, FTAG);
6143 return (error);
6144 }
6145 zfs_exit(zfsvfs, FTAG);
6146
6147 covered_vp = vp->v_mount->mnt_vnodecovered;
6148 #if __FreeBSD_version >= 1300045
6149 enum vgetstate vs = vget_prep(covered_vp);
6150 #else
6151 vhold(covered_vp);
6152 #endif
6153 ltype = VOP_ISLOCKED(vp);
6154 VOP_UNLOCK1(vp);
6155 #if __FreeBSD_version >= 1300045
6156 error = vget_finish(covered_vp, LK_SHARED, vs);
6157 #else
6158 error = vget(covered_vp, LK_SHARED | LK_VNHELD, curthread);
6159 #endif
6160 if (error == 0) {
6161 #if __FreeBSD_version >= 1300123
6162 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_buf,
6163 ap->a_buflen);
6164 #else
6165 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_cred,
6166 ap->a_buf, ap->a_buflen);
6167 #endif
6168 vput(covered_vp);
6169 }
6170 vn_lock(vp, ltype | LK_RETRY);
6171 if (VN_IS_DOOMED(vp))
6172 error = SET_ERROR(ENOENT);
6173 return (error);
6174 }
6175
6176 #if __FreeBSD_version >= 1400032
6177 static int
zfs_deallocate(struct vop_deallocate_args * ap)6178 zfs_deallocate(struct vop_deallocate_args *ap)
6179 {
6180 znode_t *zp = VTOZ(ap->a_vp);
6181 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6182 zilog_t *zilog;
6183 off_t off, len, file_sz;
6184 int error;
6185
6186 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0)
6187 return (error);
6188
6189 /*
6190 * Callers might not be able to detect properly that we are read-only,
6191 * so check it explicitly here.
6192 */
6193 if (zfs_is_readonly(zfsvfs)) {
6194 zfs_exit(zfsvfs, FTAG);
6195 return (SET_ERROR(EROFS));
6196 }
6197
6198 zilog = zfsvfs->z_log;
6199 off = *ap->a_offset;
6200 len = *ap->a_len;
6201 file_sz = zp->z_size;
6202 if (off + len > file_sz)
6203 len = file_sz - off;
6204 /* Fast path for out-of-range request. */
6205 if (len <= 0) {
6206 *ap->a_len = 0;
6207 zfs_exit(zfsvfs, FTAG);
6208 return (0);
6209 }
6210
6211 error = zfs_freesp(zp, off, len, O_RDWR, TRUE);
6212 if (error == 0) {
6213 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS ||
6214 (ap->a_ioflag & IO_SYNC) != 0)
6215 zil_commit(zilog, zp->z_id);
6216 *ap->a_offset = off + len;
6217 *ap->a_len = 0;
6218 }
6219
6220 zfs_exit(zfsvfs, FTAG);
6221 return (error);
6222 }
6223 #endif
6224
6225 #if __FreeBSD_version >= 1300039
6226 #ifndef _SYS_SYSPROTO_H_
6227 struct vop_copy_file_range_args {
6228 struct vnode *a_invp;
6229 off_t *a_inoffp;
6230 struct vnode *a_outvp;
6231 off_t *a_outoffp;
6232 size_t *a_lenp;
6233 unsigned int a_flags;
6234 struct ucred *a_incred;
6235 struct ucred *a_outcred;
6236 struct thread *a_fsizetd;
6237 }
6238 #endif
6239 /*
6240 * TODO: FreeBSD will only call file system-specific copy_file_range() if both
6241 * files resides under the same mountpoint. In case of ZFS we want to be called
6242 * even is files are in different datasets (but on the same pools, but we need
6243 * to check that ourselves).
6244 */
6245 static int
zfs_freebsd_copy_file_range(struct vop_copy_file_range_args * ap)6246 zfs_freebsd_copy_file_range(struct vop_copy_file_range_args *ap)
6247 {
6248 zfsvfs_t *outzfsvfs;
6249 struct vnode *invp = ap->a_invp;
6250 struct vnode *outvp = ap->a_outvp;
6251 struct mount *mp;
6252 struct uio io;
6253 int error;
6254 uint64_t len = *ap->a_lenp;
6255
6256 if (!zfs_bclone_enabled) {
6257 mp = NULL;
6258 goto bad_write_fallback;
6259 }
6260
6261 /*
6262 * TODO: If offset/length is not aligned to recordsize, use
6263 * vn_generic_copy_file_range() on this fragment.
6264 * It would be better to do this after we lock the vnodes, but then we
6265 * need something else than vn_generic_copy_file_range().
6266 */
6267
6268 vn_start_write(outvp, &mp, V_WAIT);
6269 if (__predict_true(mp == outvp->v_mount)) {
6270 outzfsvfs = (zfsvfs_t *)mp->mnt_data;
6271 if (!spa_feature_is_enabled(dmu_objset_spa(outzfsvfs->z_os),
6272 SPA_FEATURE_BLOCK_CLONING)) {
6273 goto bad_write_fallback;
6274 }
6275 }
6276 if (invp == outvp) {
6277 if (vn_lock(outvp, LK_EXCLUSIVE) != 0) {
6278 goto bad_write_fallback;
6279 }
6280 } else {
6281 #if (__FreeBSD_version >= 1302506 && __FreeBSD_version < 1400000) || \
6282 __FreeBSD_version >= 1400086
6283 vn_lock_pair(invp, false, LK_EXCLUSIVE, outvp, false,
6284 LK_EXCLUSIVE);
6285 #else
6286 vn_lock_pair(invp, false, outvp, false);
6287 #endif
6288 if (VN_IS_DOOMED(invp) || VN_IS_DOOMED(outvp)) {
6289 goto bad_locked_fallback;
6290 }
6291 }
6292
6293 #ifdef MAC
6294 error = mac_vnode_check_write(curthread->td_ucred, ap->a_outcred,
6295 outvp);
6296 if (error != 0)
6297 goto out_locked;
6298 #endif
6299
6300 io.uio_offset = *ap->a_outoffp;
6301 io.uio_resid = *ap->a_lenp;
6302 error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd);
6303 if (error != 0)
6304 goto out_locked;
6305
6306 error = zfs_clone_range(VTOZ(invp), ap->a_inoffp, VTOZ(outvp),
6307 ap->a_outoffp, &len, ap->a_outcred);
6308 if (error == EXDEV || error == EAGAIN || error == EINVAL ||
6309 error == EOPNOTSUPP)
6310 goto bad_locked_fallback;
6311 *ap->a_lenp = (size_t)len;
6312 out_locked:
6313 if (invp != outvp)
6314 VOP_UNLOCK(invp);
6315 VOP_UNLOCK(outvp);
6316 if (mp != NULL)
6317 vn_finished_write(mp);
6318 return (error);
6319
6320 bad_locked_fallback:
6321 if (invp != outvp)
6322 VOP_UNLOCK(invp);
6323 VOP_UNLOCK(outvp);
6324 bad_write_fallback:
6325 if (mp != NULL)
6326 vn_finished_write(mp);
6327 error = ENOSYS;
6328 return (error);
6329 }
6330 #endif
6331
6332 struct vop_vector zfs_vnodeops;
6333 struct vop_vector zfs_fifoops;
6334 struct vop_vector zfs_shareops;
6335
6336 struct vop_vector zfs_vnodeops = {
6337 .vop_default = &default_vnodeops,
6338 .vop_inactive = zfs_freebsd_inactive,
6339 #if __FreeBSD_version >= 1300042
6340 .vop_need_inactive = zfs_freebsd_need_inactive,
6341 #endif
6342 .vop_reclaim = zfs_freebsd_reclaim,
6343 #if __FreeBSD_version >= 1300102
6344 .vop_fplookup_vexec = zfs_freebsd_fplookup_vexec,
6345 #endif
6346 #if __FreeBSD_version >= 1300139
6347 .vop_fplookup_symlink = zfs_freebsd_fplookup_symlink,
6348 #endif
6349 .vop_access = zfs_freebsd_access,
6350 .vop_allocate = VOP_EINVAL,
6351 #if __FreeBSD_version >= 1400032
6352 .vop_deallocate = zfs_deallocate,
6353 #endif
6354 .vop_lookup = zfs_cache_lookup,
6355 .vop_cachedlookup = zfs_freebsd_cachedlookup,
6356 .vop_getattr = zfs_freebsd_getattr,
6357 .vop_setattr = zfs_freebsd_setattr,
6358 .vop_create = zfs_freebsd_create,
6359 .vop_mknod = (vop_mknod_t *)zfs_freebsd_create,
6360 .vop_mkdir = zfs_freebsd_mkdir,
6361 .vop_readdir = zfs_freebsd_readdir,
6362 .vop_fsync = zfs_freebsd_fsync,
6363 .vop_open = zfs_freebsd_open,
6364 .vop_close = zfs_freebsd_close,
6365 .vop_rmdir = zfs_freebsd_rmdir,
6366 .vop_ioctl = zfs_freebsd_ioctl,
6367 .vop_link = zfs_freebsd_link,
6368 .vop_symlink = zfs_freebsd_symlink,
6369 .vop_readlink = zfs_freebsd_readlink,
6370 .vop_read = zfs_freebsd_read,
6371 .vop_write = zfs_freebsd_write,
6372 .vop_remove = zfs_freebsd_remove,
6373 .vop_rename = zfs_freebsd_rename,
6374 .vop_pathconf = zfs_freebsd_pathconf,
6375 .vop_bmap = zfs_freebsd_bmap,
6376 .vop_fid = zfs_freebsd_fid,
6377 .vop_getextattr = zfs_getextattr,
6378 .vop_deleteextattr = zfs_deleteextattr,
6379 .vop_setextattr = zfs_setextattr,
6380 .vop_listextattr = zfs_listextattr,
6381 .vop_getacl = zfs_freebsd_getacl,
6382 .vop_setacl = zfs_freebsd_setacl,
6383 .vop_aclcheck = zfs_freebsd_aclcheck,
6384 .vop_getpages = zfs_freebsd_getpages,
6385 .vop_putpages = zfs_freebsd_putpages,
6386 .vop_vptocnp = zfs_vptocnp,
6387 #if __FreeBSD_version >= 1300064
6388 .vop_lock1 = vop_lock,
6389 .vop_unlock = vop_unlock,
6390 .vop_islocked = vop_islocked,
6391 #endif
6392 #if __FreeBSD_version >= 1400043
6393 .vop_add_writecount = vop_stdadd_writecount_nomsync,
6394 #endif
6395 #if __FreeBSD_version >= 1300039
6396 .vop_copy_file_range = zfs_freebsd_copy_file_range,
6397 #endif
6398 };
6399 VFS_VOP_VECTOR_REGISTER(zfs_vnodeops);
6400
6401 struct vop_vector zfs_fifoops = {
6402 .vop_default = &fifo_specops,
6403 .vop_fsync = zfs_freebsd_fsync,
6404 #if __FreeBSD_version >= 1300102
6405 .vop_fplookup_vexec = zfs_freebsd_fplookup_vexec,
6406 #endif
6407 #if __FreeBSD_version >= 1300139
6408 .vop_fplookup_symlink = zfs_freebsd_fplookup_symlink,
6409 #endif
6410 .vop_access = zfs_freebsd_access,
6411 .vop_getattr = zfs_freebsd_getattr,
6412 .vop_inactive = zfs_freebsd_inactive,
6413 .vop_read = VOP_PANIC,
6414 .vop_reclaim = zfs_freebsd_reclaim,
6415 .vop_setattr = zfs_freebsd_setattr,
6416 .vop_write = VOP_PANIC,
6417 .vop_pathconf = zfs_freebsd_pathconf,
6418 .vop_fid = zfs_freebsd_fid,
6419 .vop_getacl = zfs_freebsd_getacl,
6420 .vop_setacl = zfs_freebsd_setacl,
6421 .vop_aclcheck = zfs_freebsd_aclcheck,
6422 #if __FreeBSD_version >= 1400043
6423 .vop_add_writecount = vop_stdadd_writecount_nomsync,
6424 #endif
6425 };
6426 VFS_VOP_VECTOR_REGISTER(zfs_fifoops);
6427
6428 /*
6429 * special share hidden files vnode operations template
6430 */
6431 struct vop_vector zfs_shareops = {
6432 .vop_default = &default_vnodeops,
6433 #if __FreeBSD_version >= 1300121
6434 .vop_fplookup_vexec = VOP_EAGAIN,
6435 #endif
6436 #if __FreeBSD_version >= 1300139
6437 .vop_fplookup_symlink = VOP_EAGAIN,
6438 #endif
6439 .vop_access = zfs_freebsd_access,
6440 .vop_inactive = zfs_freebsd_inactive,
6441 .vop_reclaim = zfs_freebsd_reclaim,
6442 .vop_fid = zfs_freebsd_fid,
6443 .vop_pathconf = zfs_freebsd_pathconf,
6444 #if __FreeBSD_version >= 1400043
6445 .vop_add_writecount = vop_stdadd_writecount_nomsync,
6446 #endif
6447 };
6448 VFS_VOP_VECTOR_REGISTER(zfs_shareops);
6449
6450 ZFS_MODULE_PARAM(zfs, zfs_, xattr_compat, INT, ZMOD_RW,
6451 "Use legacy ZFS xattr naming for writing new user namespace xattrs");
6452