1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
25 */
26
27/* Portions Copyright 2007 Jeremy Teo */
28/* Portions Copyright 2010 Robert Milkowski */
29
30#include <sys/types.h>
31#include <sys/param.h>
32#include <sys/time.h>
33#include <sys/systm.h>
34#include <sys/sysmacros.h>
35#include <sys/resource.h>
36#include <sys/vfs.h>
37#include <sys/vm.h>
38#include <sys/vnode.h>
39#include <sys/file.h>
40#include <sys/stat.h>
41#include <sys/kmem.h>
42#include <sys/taskq.h>
43#include <sys/uio.h>
44#include <sys/atomic.h>
45#include <sys/namei.h>
46#include <sys/mman.h>
47#include <sys/cmn_err.h>
48#include <sys/errno.h>
49#include <sys/unistd.h>
50#include <sys/zfs_dir.h>
51#include <sys/zfs_ioctl.h>
52#include <sys/fs/zfs.h>
53#include <sys/dmu.h>
54#include <sys/dmu_objset.h>
55#include <sys/spa.h>
56#include <sys/txg.h>
57#include <sys/dbuf.h>
58#include <sys/zap.h>
59#include <sys/sa.h>
60#include <sys/dirent.h>
61#include <sys/policy.h>
62#include <sys/sunddi.h>
63#include <sys/filio.h>
64#include <sys/sid.h>
65#include <sys/zfs_ctldir.h>
66#include <sys/zfs_fuid.h>
67#include <sys/zfs_sa.h>
68#include <sys/dnlc.h>
69#include <sys/zfs_rlock.h>
70#include <sys/extdirent.h>
71#include <sys/kidmap.h>
72#include <sys/bio.h>
73#include <sys/buf.h>
74#include <sys/sf_buf.h>
75#include <sys/sched.h>
76#include <sys/acl.h>
77#include <vm/vm_param.h>
78#include <vm/vm_pageout.h>
79
80/*
81 * Programming rules.
82 *
83 * Each vnode op performs some logical unit of work. To do this, the ZPL must
84 * properly lock its in-core state, create a DMU transaction, do the work,
85 * record this work in the intent log (ZIL), commit the DMU transaction,
86 * and wait for the intent log to commit if it is a synchronous operation.
87 * Moreover, the vnode ops must work in both normal and log replay context.
88 * The ordering of events is important to avoid deadlocks and references
89 * to freed memory. The example below illustrates the following Big Rules:
90 *
91 * (1) A check must be made in each zfs thread for a mounted file system.
92 * This is done avoiding races using ZFS_ENTER(zfsvfs).
93 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
94 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
95 * can return EIO from the calling function.
96 *
97 * (2) VN_RELE() should always be the last thing except for zil_commit()
98 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
99 * First, if it's the last reference, the vnode/znode
100 * can be freed, so the zp may point to freed memory. Second, the last
101 * reference will call zfs_zinactive(), which may induce a lot of work --
102 * pushing cached pages (which acquires range locks) and syncing out
103 * cached atime changes. Third, zfs_zinactive() may require a new tx,
104 * which could deadlock the system if you were already holding one.
105 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
106 *
107 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
108 * as they can span dmu_tx_assign() calls.
109 *
110 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
111 * dmu_tx_assign(). This is critical because we don't want to block
112 * while holding locks.
113 *
114 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
115 * reduces lock contention and CPU usage when we must wait (note that if
116 * throughput is constrained by the storage, nearly every transaction
117 * must wait).
118 *
119 * Note, in particular, that if a lock is sometimes acquired before
120 * the tx assigns, and sometimes after (e.g. z_lock), then failing
121 * to use a non-blocking assign can deadlock the system. The scenario:
122 *
123 * Thread A has grabbed a lock before calling dmu_tx_assign().
124 * Thread B is in an already-assigned tx, and blocks for this lock.
125 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
126 * forever, because the previous txg can't quiesce until B's tx commits.
127 *
128 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
129 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
130 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
131 * to indicate that this operation has already called dmu_tx_wait().
132 * This will ensure that we don't retry forever, waiting a short bit
133 * each time.
134 *
135 * (5) If the operation succeeded, generate the intent log entry for it
136 * before dropping locks. This ensures that the ordering of events
137 * in the intent log matches the order in which they actually occurred.
138 * During ZIL replay the zfs_log_* functions will update the sequence
139 * number to indicate the zil transaction has replayed.
140 *
141 * (6) At the end of each vnode op, the DMU tx must always commit,
142 * regardless of whether there were any errors.
143 *
144 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
145 * to ensure that synchronous semantics are provided when necessary.
146 *
147 * In general, this is how things should be ordered in each vnode op:
148 *
149 * ZFS_ENTER(zfsvfs); // exit if unmounted
150 * top:
151 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
152 * rw_enter(...); // grab any other locks you need
153 * tx = dmu_tx_create(...); // get DMU tx
154 * dmu_tx_hold_*(); // hold each object you might modify
155 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
156 * if (error) {
157 * rw_exit(...); // drop locks
158 * zfs_dirent_unlock(dl); // unlock directory entry
159 * VN_RELE(...); // release held vnodes
160 * if (error == ERESTART) {
161 * waited = B_TRUE;
162 * dmu_tx_wait(tx);
163 * dmu_tx_abort(tx);
164 * goto top;
165 * }
166 * dmu_tx_abort(tx); // abort DMU tx
167 * ZFS_EXIT(zfsvfs); // finished in zfs
168 * return (error); // really out of space
169 * }
170 * error = do_real_work(); // do whatever this VOP does
171 * if (error == 0)
172 * zfs_log_*(...); // on success, make ZIL entry
173 * dmu_tx_commit(tx); // commit DMU tx -- error or not
174 * rw_exit(...); // drop locks
175 * zfs_dirent_unlock(dl); // unlock directory entry
176 * VN_RELE(...); // release held vnodes
177 * zil_commit(zilog, foid); // synchronous when necessary
178 * ZFS_EXIT(zfsvfs); // finished in zfs
179 * return (error); // done, report error
180 */
181
182/* ARGSUSED */
183static int
184zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
185{
186 znode_t *zp = VTOZ(*vpp);
187 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
188
189 ZFS_ENTER(zfsvfs);
190 ZFS_VERIFY_ZP(zp);
191
192 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
193 ((flag & FAPPEND) == 0)) {
194 ZFS_EXIT(zfsvfs);
195 return (SET_ERROR(EPERM));
196 }
197
198 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
199 ZTOV(zp)->v_type == VREG &&
200 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
201 if (fs_vscan(*vpp, cr, 0) != 0) {
202 ZFS_EXIT(zfsvfs);
203 return (SET_ERROR(EACCES));
204 }
205 }
206
207 /* Keep a count of the synchronous opens in the znode */
208 if (flag & (FSYNC | FDSYNC))
209 atomic_inc_32(&zp->z_sync_cnt);
210
211 ZFS_EXIT(zfsvfs);
212 return (0);
213}
214
215/* ARGSUSED */
216static int
217zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
218 caller_context_t *ct)
219{
220 znode_t *zp = VTOZ(vp);
221 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
222
223 /*
224 * Clean up any locks held by this process on the vp.
225 */
226 cleanlocks(vp, ddi_get_pid(), 0);
227 cleanshares(vp, ddi_get_pid());
228
229 ZFS_ENTER(zfsvfs);
230 ZFS_VERIFY_ZP(zp);
231
232 /* Decrement the synchronous opens in the znode */
233 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
234 atomic_dec_32(&zp->z_sync_cnt);
235
236 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
237 ZTOV(zp)->v_type == VREG &&
238 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
239 VERIFY(fs_vscan(vp, cr, 1) == 0);
240
241 ZFS_EXIT(zfsvfs);
242 return (0);
243}
244
245/*
246 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
247 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
248 */
249static int
250zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
251{
252 znode_t *zp = VTOZ(vp);
253 uint64_t noff = (uint64_t)*off; /* new offset */
254 uint64_t file_sz;
255 int error;
256 boolean_t hole;
257
258 file_sz = zp->z_size;
259 if (noff >= file_sz) {
260 return (SET_ERROR(ENXIO));
261 }
262
263 if (cmd == _FIO_SEEK_HOLE)
264 hole = B_TRUE;
265 else
266 hole = B_FALSE;
267
268 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
269
270 /* end of file? */
271 if ((error == ESRCH) || (noff > file_sz)) {
272 /*
273 * Handle the virtual hole at the end of file.
274 */
275 if (hole) {
276 *off = file_sz;
277 return (0);
278 }
279 return (SET_ERROR(ENXIO));
280 }
281
282 if (noff < *off)
283 return (error);
284 *off = noff;
285 return (error);
286}
287
288/* ARGSUSED */
289static int
290zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
291 int *rvalp, caller_context_t *ct)
292{
293 offset_t off;
294 int error;
295 zfsvfs_t *zfsvfs;
296 znode_t *zp;
297
298 switch (com) {
299 case _FIOFFS:
300 return (0);
301
302 /*
303 * The following two ioctls are used by bfu. Faking out,
304 * necessary to avoid bfu errors.
305 */
306 case _FIOGDIO:
307 case _FIOSDIO:
308 return (0);
309
310 case _FIO_SEEK_DATA:
311 case _FIO_SEEK_HOLE:
312#ifdef sun
313 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
314 return (SET_ERROR(EFAULT));
315#else
316 off = *(offset_t *)data;
317#endif
318 zp = VTOZ(vp);
319 zfsvfs = zp->z_zfsvfs;
320 ZFS_ENTER(zfsvfs);
321 ZFS_VERIFY_ZP(zp);
322
323 /* offset parameter is in/out */
324 error = zfs_holey(vp, com, &off);
325 ZFS_EXIT(zfsvfs);
326 if (error)
327 return (error);
328#ifdef sun
329 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
330 return (SET_ERROR(EFAULT));
331#else
332 *(offset_t *)data = off;
333#endif
334 return (0);
335 }
336 return (SET_ERROR(ENOTTY));
337}
338
339static vm_page_t
340page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
341{
342 vm_object_t obj;
343 vm_page_t pp;
344 int64_t end;
345
346 /*
347 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
348 * aligned boundaries, if the range is not aligned. As a result a
349 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
350 * It may happen that all DEV_BSIZE subranges are marked clean and thus
351 * the whole page would be considred clean despite have some dirty data.
352 * For this reason we should shrink the range to DEV_BSIZE aligned
353 * boundaries before calling vm_page_clear_dirty.
354 */
355 end = rounddown2(off + nbytes, DEV_BSIZE);
356 off = roundup2(off, DEV_BSIZE);
357 nbytes = end - off;
358
359 obj = vp->v_object;
360 zfs_vmobject_assert_wlocked(obj);
361
362 for (;;) {
363 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
364 pp->valid) {
365 if (vm_page_xbusied(pp)) {
366 /*
367 * Reference the page before unlocking and
368 * sleeping so that the page daemon is less
369 * likely to reclaim it.
370 */
371 vm_page_reference(pp);
372 vm_page_lock(pp);
373 zfs_vmobject_wunlock(obj);
374 vm_page_busy_sleep(pp, "zfsmwb");
375 zfs_vmobject_wlock(obj);
376 continue;
377 }
378 vm_page_sbusy(pp);
379 } else if (pp == NULL) {
380 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
381 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
382 VM_ALLOC_SBUSY);
383 } else {
384 ASSERT(pp != NULL && !pp->valid);
385 pp = NULL;
386 }
387
388 if (pp != NULL) {
389 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
390 vm_object_pip_add(obj, 1);
391 pmap_remove_write(pp);
392 if (nbytes != 0)
393 vm_page_clear_dirty(pp, off, nbytes);
394 }
395 break;
396 }
397 return (pp);
398}
399
400static void
401page_unbusy(vm_page_t pp)
402{
403
404 vm_page_sunbusy(pp);
405 vm_object_pip_subtract(pp->object, 1);
406}
407
408static vm_page_t
409page_hold(vnode_t *vp, int64_t start)
410{
411 vm_object_t obj;
412 vm_page_t pp;
413
414 obj = vp->v_object;
415 zfs_vmobject_assert_wlocked(obj);
416
417 for (;;) {
418 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
419 pp->valid) {
420 if (vm_page_xbusied(pp)) {
421 /*
422 * Reference the page before unlocking and
423 * sleeping so that the page daemon is less
424 * likely to reclaim it.
425 */
426 vm_page_reference(pp);
427 vm_page_lock(pp);
428 zfs_vmobject_wunlock(obj);
429 vm_page_busy_sleep(pp, "zfsmwb");
430 zfs_vmobject_wlock(obj);
431 continue;
432 }
433
434 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
435 vm_page_lock(pp);
436 vm_page_hold(pp);
437 vm_page_unlock(pp);
438
439 } else
440 pp = NULL;
441 break;
442 }
443 return (pp);
444}
445
446static void
447page_unhold(vm_page_t pp)
448{
449
450 vm_page_lock(pp);
451 vm_page_unhold(pp);
452 vm_page_unlock(pp);
453}
454
455static caddr_t
456zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
457{
458
459 *sfp = sf_buf_alloc(pp, 0);
460 return ((caddr_t)sf_buf_kva(*sfp));
461}
462
463static void
464zfs_unmap_page(struct sf_buf *sf)
465{
466
467 sf_buf_free(sf);
468}
469
470/*
471 * When a file is memory mapped, we must keep the IO data synchronized
472 * between the DMU cache and the memory mapped pages. What this means:
473 *
474 * On Write: If we find a memory mapped page, we write to *both*
475 * the page and the dmu buffer.
476 */
477static void
478update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
479 int segflg, dmu_tx_t *tx)
480{
481 vm_object_t obj;
482 struct sf_buf *sf;
483 caddr_t va;
484 int off;
485
486 ASSERT(vp->v_mount != NULL);
487 obj = vp->v_object;
488 ASSERT(obj != NULL);
489
490 off = start & PAGEOFFSET;
491 zfs_vmobject_wlock(obj);
492 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
493 vm_page_t pp;
494 int nbytes = imin(PAGESIZE - off, len);
495
496 if (segflg == UIO_NOCOPY) {
497 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
498 KASSERT(pp != NULL,
499 ("zfs update_pages: NULL page in putpages case"));
500 KASSERT(off == 0,
501 ("zfs update_pages: unaligned data in putpages case"));
502 KASSERT(pp->valid == VM_PAGE_BITS_ALL,
503 ("zfs update_pages: invalid page in putpages case"));
504 KASSERT(vm_page_sbusied(pp),
505 ("zfs update_pages: unbusy page in putpages case"));
506 KASSERT(!pmap_page_is_write_mapped(pp),
507 ("zfs update_pages: writable page in putpages case"));
508 zfs_vmobject_wunlock(obj);
509
510 va = zfs_map_page(pp, &sf);
511 (void) dmu_write(os, oid, start, nbytes, va, tx);
512 zfs_unmap_page(sf);
513
514 zfs_vmobject_wlock(obj);
515 vm_page_undirty(pp);
516 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
517 zfs_vmobject_wunlock(obj);
518
519 va = zfs_map_page(pp, &sf);
520 (void) dmu_read(os, oid, start+off, nbytes,
521 va+off, DMU_READ_PREFETCH);;
522 zfs_unmap_page(sf);
523
524 zfs_vmobject_wlock(obj);
525 page_unbusy(pp);
526 }
527 len -= nbytes;
528 off = 0;
529 }
530 if (segflg != UIO_NOCOPY)
531 vm_object_pip_wakeupn(obj, 0);
532 zfs_vmobject_wunlock(obj);
533}
534
535/*
536 * Read with UIO_NOCOPY flag means that sendfile(2) requests
537 * ZFS to populate a range of page cache pages with data.
538 *
539 * NOTE: this function could be optimized to pre-allocate
540 * all pages in advance, drain exclusive busy on all of them,
541 * map them into contiguous KVA region and populate them
542 * in one single dmu_read() call.
543 */
544static int
545mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
546{
547 znode_t *zp = VTOZ(vp);
548 objset_t *os = zp->z_zfsvfs->z_os;
549 struct sf_buf *sf;
550 vm_object_t obj;
551 vm_page_t pp;
552 int64_t start;
553 caddr_t va;
554 int len = nbytes;
555 int off;
556 int error = 0;
557
558 ASSERT(uio->uio_segflg == UIO_NOCOPY);
559 ASSERT(vp->v_mount != NULL);
560 obj = vp->v_object;
561 ASSERT(obj != NULL);
562 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
563
564 zfs_vmobject_wlock(obj);
565 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
566 int bytes = MIN(PAGESIZE, len);
567
568 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY |
569 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
570 if (pp->valid == 0) {
571 zfs_vmobject_wunlock(obj);
572 va = zfs_map_page(pp, &sf);
573 error = dmu_read(os, zp->z_id, start, bytes, va,
574 DMU_READ_PREFETCH);
575 if (bytes != PAGESIZE && error == 0)
576 bzero(va + bytes, PAGESIZE - bytes);
577 zfs_unmap_page(sf);
578 zfs_vmobject_wlock(obj);
579 vm_page_sunbusy(pp);
580 vm_page_lock(pp);
581 if (error) {
582 if (pp->wire_count == 0 && pp->valid == 0 &&
583 !vm_page_busied(pp))
584 vm_page_free(pp);
585 } else {
586 pp->valid = VM_PAGE_BITS_ALL;
587 vm_page_activate(pp);
588 }
589 vm_page_unlock(pp);
590 } else {
591 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
592 vm_page_sunbusy(pp);
593 }
594 if (error)
595 break;
596 uio->uio_resid -= bytes;
597 uio->uio_offset += bytes;
598 len -= bytes;
599 }
600 zfs_vmobject_wunlock(obj);
601 return (error);
602}
603
604/*
605 * When a file is memory mapped, we must keep the IO data synchronized
606 * between the DMU cache and the memory mapped pages. What this means:
607 *
608 * On Read: We "read" preferentially from memory mapped pages,
609 * else we default from the dmu buffer.
610 *
611 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
612 * the file is memory mapped.
613 */
614static int
615mappedread(vnode_t *vp, int nbytes, uio_t *uio)
616{
617 znode_t *zp = VTOZ(vp);
618 objset_t *os = zp->z_zfsvfs->z_os;
619 vm_object_t obj;
620 int64_t start;
621 caddr_t va;
622 int len = nbytes;
623 int off;
624 int error = 0;
625
626 ASSERT(vp->v_mount != NULL);
627 obj = vp->v_object;
628 ASSERT(obj != NULL);
629
630 start = uio->uio_loffset;
631 off = start & PAGEOFFSET;
632 zfs_vmobject_wlock(obj);
633 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
634 vm_page_t pp;
635 uint64_t bytes = MIN(PAGESIZE - off, len);
636
637 if (pp = page_hold(vp, start)) {
638 struct sf_buf *sf;
639 caddr_t va;
640
641 zfs_vmobject_wunlock(obj);
642 va = zfs_map_page(pp, &sf);
643 error = uiomove(va + off, bytes, UIO_READ, uio);
644 zfs_unmap_page(sf);
645 zfs_vmobject_wlock(obj);
646 page_unhold(pp);
647 } else {
648 zfs_vmobject_wunlock(obj);
649 error = dmu_read_uio(os, zp->z_id, uio, bytes);
650 zfs_vmobject_wlock(obj);
651 }
652 len -= bytes;
653 off = 0;
654 if (error)
655 break;
656 }
657 zfs_vmobject_wunlock(obj);
658 return (error);
659}
660
661offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
662
663/*
664 * Read bytes from specified file into supplied buffer.
665 *
666 * IN: vp - vnode of file to be read from.
667 * uio - structure supplying read location, range info,
668 * and return buffer.
669 * ioflag - SYNC flags; used to provide FRSYNC semantics.
670 * cr - credentials of caller.
671 * ct - caller context
672 *
673 * OUT: uio - updated offset and range, buffer filled.
674 *
675 * RETURN: 0 on success, error code on failure.
676 *
677 * Side Effects:
678 * vp - atime updated if byte count > 0
679 */
680/* ARGSUSED */
681static int
682zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
683{
684 znode_t *zp = VTOZ(vp);
685 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
686 objset_t *os;
687 ssize_t n, nbytes;
688 int error = 0;
689 rl_t *rl;
690 xuio_t *xuio = NULL;
691
692 ZFS_ENTER(zfsvfs);
693 ZFS_VERIFY_ZP(zp);
694 os = zfsvfs->z_os;
695
696 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
697 ZFS_EXIT(zfsvfs);
698 return (SET_ERROR(EACCES));
699 }
700
701 /*
702 * Validate file offset
703 */
704 if (uio->uio_loffset < (offset_t)0) {
705 ZFS_EXIT(zfsvfs);
706 return (SET_ERROR(EINVAL));
707 }
708
709 /*
710 * Fasttrack empty reads
711 */
712 if (uio->uio_resid == 0) {
713 ZFS_EXIT(zfsvfs);
714 return (0);
715 }
716
717 /*
718 * Check for mandatory locks
719 */
720 if (MANDMODE(zp->z_mode)) {
721 if (error = chklock(vp, FREAD,
722 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
723 ZFS_EXIT(zfsvfs);
724 return (error);
725 }
726 }
727
728 /*
729 * If we're in FRSYNC mode, sync out this znode before reading it.
730 */
731 if (zfsvfs->z_log &&
732 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
733 zil_commit(zfsvfs->z_log, zp->z_id);
734
735 /*
736 * Lock the range against changes.
737 */
738 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
739
740 /*
741 * If we are reading past end-of-file we can skip
742 * to the end; but we might still need to set atime.
743 */
744 if (uio->uio_loffset >= zp->z_size) {
745 error = 0;
746 goto out;
747 }
748
749 ASSERT(uio->uio_loffset < zp->z_size);
750 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
751
752#ifdef sun
753 if ((uio->uio_extflg == UIO_XUIO) &&
754 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
755 int nblk;
756 int blksz = zp->z_blksz;
757 uint64_t offset = uio->uio_loffset;
758
759 xuio = (xuio_t *)uio;
760 if ((ISP2(blksz))) {
761 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
762 blksz)) / blksz;
763 } else {
764 ASSERT(offset + n <= blksz);
765 nblk = 1;
766 }
767 (void) dmu_xuio_init(xuio, nblk);
768
769 if (vn_has_cached_data(vp)) {
770 /*
771 * For simplicity, we always allocate a full buffer
772 * even if we only expect to read a portion of a block.
773 */
774 while (--nblk >= 0) {
775 (void) dmu_xuio_add(xuio,
776 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
777 blksz), 0, blksz);
778 }
779 }
780 }
781#endif /* sun */
782
783 while (n > 0) {
784 nbytes = MIN(n, zfs_read_chunk_size -
785 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
786
787#ifdef __FreeBSD__
788 if (uio->uio_segflg == UIO_NOCOPY)
789 error = mappedread_sf(vp, nbytes, uio);
790 else
791#endif /* __FreeBSD__ */
792 if (vn_has_cached_data(vp))
793 error = mappedread(vp, nbytes, uio);
794 else
795 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
796 if (error) {
797 /* convert checksum errors into IO errors */
798 if (error == ECKSUM)
799 error = SET_ERROR(EIO);
800 break;
801 }
802
803 n -= nbytes;
804 }
805out:
806 zfs_range_unlock(rl);
807
808 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
809 ZFS_EXIT(zfsvfs);
810 return (error);
811}
812
813/*
814 * Write the bytes to a file.
815 *
816 * IN: vp - vnode of file to be written to.
817 * uio - structure supplying write location, range info,
818 * and data buffer.
819 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
820 * set if in append mode.
821 * cr - credentials of caller.
822 * ct - caller context (NFS/CIFS fem monitor only)
823 *
824 * OUT: uio - updated offset and range.
825 *
826 * RETURN: 0 on success, error code on failure.
827 *
828 * Timestamps:
829 * vp - ctime|mtime updated if byte count > 0
830 */
831
832/* ARGSUSED */
833static int
834zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
835{
836 znode_t *zp = VTOZ(vp);
837 rlim64_t limit = MAXOFFSET_T;
838 ssize_t start_resid = uio->uio_resid;
839 ssize_t tx_bytes;
840 uint64_t end_size;
841 dmu_tx_t *tx;
842 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
843 zilog_t *zilog;
844 offset_t woff;
845 ssize_t n, nbytes;
846 rl_t *rl;
847 int max_blksz = zfsvfs->z_max_blksz;
848 int error = 0;
849 arc_buf_t *abuf;
850 iovec_t *aiov = NULL;
851 xuio_t *xuio = NULL;
852 int i_iov = 0;
853 int iovcnt = uio->uio_iovcnt;
854 iovec_t *iovp = uio->uio_iov;
855 int write_eof;
856 int count = 0;
857 sa_bulk_attr_t bulk[4];
858 uint64_t mtime[2], ctime[2];
859
860 /*
861 * Fasttrack empty write
862 */
863 n = start_resid;
864 if (n == 0)
865 return (0);
866
867 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
868 limit = MAXOFFSET_T;
869
870 ZFS_ENTER(zfsvfs);
871 ZFS_VERIFY_ZP(zp);
872
873 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
874 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
875 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
876 &zp->z_size, 8);
877 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
878 &zp->z_pflags, 8);
879
880 /*
881 * If immutable or not appending then return EPERM
882 */
883 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
884 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
885 (uio->uio_loffset < zp->z_size))) {
886 ZFS_EXIT(zfsvfs);
887 return (SET_ERROR(EPERM));
888 }
889
890 zilog = zfsvfs->z_log;
891
892 /*
893 * Validate file offset
894 */
895 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
896 if (woff < 0) {
897 ZFS_EXIT(zfsvfs);
898 return (SET_ERROR(EINVAL));
899 }
900
901 /*
902 * Check for mandatory locks before calling zfs_range_lock()
903 * in order to prevent a deadlock with locks set via fcntl().
904 */
905 if (MANDMODE((mode_t)zp->z_mode) &&
906 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
907 ZFS_EXIT(zfsvfs);
908 return (error);
909 }
910
911#ifdef sun
912 /*
913 * Pre-fault the pages to ensure slow (eg NFS) pages
914 * don't hold up txg.
915 * Skip this if uio contains loaned arc_buf.
916 */
917 if ((uio->uio_extflg == UIO_XUIO) &&
918 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
919 xuio = (xuio_t *)uio;
920 else
921 uio_prefaultpages(MIN(n, max_blksz), uio);
922#endif /* sun */
923
924 /*
925 * If in append mode, set the io offset pointer to eof.
926 */
927 if (ioflag & FAPPEND) {
928 /*
929 * Obtain an appending range lock to guarantee file append
930 * semantics. We reset the write offset once we have the lock.
931 */
932 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
933 woff = rl->r_off;
934 if (rl->r_len == UINT64_MAX) {
935 /*
936 * We overlocked the file because this write will cause
937 * the file block size to increase.
938 * Note that zp_size cannot change with this lock held.
939 */
940 woff = zp->z_size;
941 }
942 uio->uio_loffset = woff;
943 } else {
944 /*
945 * Note that if the file block size will change as a result of
946 * this write, then this range lock will lock the entire file
947 * so that we can re-write the block safely.
948 */
949 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
950 }
951
952 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
953 zfs_range_unlock(rl);
954 ZFS_EXIT(zfsvfs);
955 return (EFBIG);
956 }
957
958 if (woff >= limit) {
959 zfs_range_unlock(rl);
960 ZFS_EXIT(zfsvfs);
961 return (SET_ERROR(EFBIG));
962 }
963
964 if ((woff + n) > limit || woff > (limit - n))
965 n = limit - woff;
966
967 /* Will this write extend the file length? */
968 write_eof = (woff + n > zp->z_size);
969
970 end_size = MAX(zp->z_size, woff + n);
971
972 /*
973 * Write the file in reasonable size chunks. Each chunk is written
974 * in a separate transaction; this keeps the intent log records small
975 * and allows us to do more fine-grained space accounting.
976 */
977 while (n > 0) {
978 abuf = NULL;
979 woff = uio->uio_loffset;
980 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
981 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
982 if (abuf != NULL)
983 dmu_return_arcbuf(abuf);
984 error = SET_ERROR(EDQUOT);
985 break;
986 }
987
988 if (xuio && abuf == NULL) {
989 ASSERT(i_iov < iovcnt);
990 aiov = &iovp[i_iov];
991 abuf = dmu_xuio_arcbuf(xuio, i_iov);
992 dmu_xuio_clear(xuio, i_iov);
993 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
994 iovec_t *, aiov, arc_buf_t *, abuf);
995 ASSERT((aiov->iov_base == abuf->b_data) ||
996 ((char *)aiov->iov_base - (char *)abuf->b_data +
997 aiov->iov_len == arc_buf_size(abuf)));
998 i_iov++;
999 } else if (abuf == NULL && n >= max_blksz &&
1000 woff >= zp->z_size &&
1001 P2PHASE(woff, max_blksz) == 0 &&
1002 zp->z_blksz == max_blksz) {
1003 /*
1004 * This write covers a full block. "Borrow" a buffer
1005 * from the dmu so that we can fill it before we enter
1006 * a transaction. This avoids the possibility of
1007 * holding up the transaction if the data copy hangs
1008 * up on a pagefault (e.g., from an NFS server mapping).
1009 */
1010 size_t cbytes;
1011
1012 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
1013 max_blksz);
1014 ASSERT(abuf != NULL);
1015 ASSERT(arc_buf_size(abuf) == max_blksz);
1016 if (error = uiocopy(abuf->b_data, max_blksz,
1017 UIO_WRITE, uio, &cbytes)) {
1018 dmu_return_arcbuf(abuf);
1019 break;
1020 }
1021 ASSERT(cbytes == max_blksz);
1022 }
1023
1024 /*
1025 * Start a transaction.
1026 */
1027 tx = dmu_tx_create(zfsvfs->z_os);
1028 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1029 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1030 zfs_sa_upgrade_txholds(tx, zp);
1031 error = dmu_tx_assign(tx, TXG_WAIT);
1032 if (error) {
1033 dmu_tx_abort(tx);
1034 if (abuf != NULL)
1035 dmu_return_arcbuf(abuf);
1036 break;
1037 }
1038
1039 /*
1040 * If zfs_range_lock() over-locked we grow the blocksize
1041 * and then reduce the lock range. This will only happen
1042 * on the first iteration since zfs_range_reduce() will
1043 * shrink down r_len to the appropriate size.
1044 */
1045 if (rl->r_len == UINT64_MAX) {
1046 uint64_t new_blksz;
1047
1048 if (zp->z_blksz > max_blksz) {
1049 ASSERT(!ISP2(zp->z_blksz));
1050 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1051 } else {
1052 new_blksz = MIN(end_size, max_blksz);
1053 }
1054 zfs_grow_blocksize(zp, new_blksz, tx);
1055 zfs_range_reduce(rl, woff, n);
1056 }
1057
1058 /*
1059 * XXX - should we really limit each write to z_max_blksz?
1060 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1061 */
1062 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1063
1064 if (woff + nbytes > zp->z_size)
1065 vnode_pager_setsize(vp, woff + nbytes);
1066
1067 if (abuf == NULL) {
1068 tx_bytes = uio->uio_resid;
1069 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1070 uio, nbytes, tx);
1071 tx_bytes -= uio->uio_resid;
1072 } else {
1073 tx_bytes = nbytes;
1074 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1075 /*
1076 * If this is not a full block write, but we are
1077 * extending the file past EOF and this data starts
1078 * block-aligned, use assign_arcbuf(). Otherwise,
1079 * write via dmu_write().
1080 */
1081 if (tx_bytes < max_blksz && (!write_eof ||
1082 aiov->iov_base != abuf->b_data)) {
1083 ASSERT(xuio);
1084 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1085 aiov->iov_len, aiov->iov_base, tx);
1086 dmu_return_arcbuf(abuf);
1087 xuio_stat_wbuf_copied();
1088 } else {
1089 ASSERT(xuio || tx_bytes == max_blksz);
1090 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1091 woff, abuf, tx);
1092 }
1093 ASSERT(tx_bytes <= uio->uio_resid);
1094 uioskip(uio, tx_bytes);
1095 }
1096 if (tx_bytes && vn_has_cached_data(vp)) {
1097 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1098 zp->z_id, uio->uio_segflg, tx);
1099 }
1100
1101 /*
1102 * If we made no progress, we're done. If we made even
1103 * partial progress, update the znode and ZIL accordingly.
1104 */
1105 if (tx_bytes == 0) {
1106 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1107 (void *)&zp->z_size, sizeof (uint64_t), tx);
1108 dmu_tx_commit(tx);
1109 ASSERT(error != 0);
1110 break;
1111 }
1112
1113 /*
1114 * Clear Set-UID/Set-GID bits on successful write if not
1115 * privileged and at least one of the excute bits is set.
1116 *
1117 * It would be nice to to this after all writes have
1118 * been done, but that would still expose the ISUID/ISGID
1119 * to another app after the partial write is committed.
1120 *
1121 * Note: we don't call zfs_fuid_map_id() here because
1122 * user 0 is not an ephemeral uid.
1123 */
1124 mutex_enter(&zp->z_acl_lock);
1125 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1126 (S_IXUSR >> 6))) != 0 &&
1127 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1128 secpolicy_vnode_setid_retain(vp, cr,
1129 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1130 uint64_t newmode;
1131 zp->z_mode &= ~(S_ISUID | S_ISGID);
1132 newmode = zp->z_mode;
1133 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1134 (void *)&newmode, sizeof (uint64_t), tx);
1135 }
1136 mutex_exit(&zp->z_acl_lock);
1137
1138 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1139 B_TRUE);
1140
1141 /*
1142 * Update the file size (zp_size) if it has changed;
1143 * account for possible concurrent updates.
1144 */
1145 while ((end_size = zp->z_size) < uio->uio_loffset) {
1146 (void) atomic_cas_64(&zp->z_size, end_size,
1147 uio->uio_loffset);
1148 ASSERT(error == 0);
1149 }
1150 /*
1151 * If we are replaying and eof is non zero then force
1152 * the file size to the specified eof. Note, there's no
1153 * concurrency during replay.
1154 */
1155 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1156 zp->z_size = zfsvfs->z_replay_eof;
1157
1158 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1159
1160 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1161 dmu_tx_commit(tx);
1162
1163 if (error != 0)
1164 break;
1165 ASSERT(tx_bytes == nbytes);
1166 n -= nbytes;
1167
1168#ifdef sun
1169 if (!xuio && n > 0)
1170 uio_prefaultpages(MIN(n, max_blksz), uio);
1171#endif /* sun */
1172 }
1173
1174 zfs_range_unlock(rl);
1175
1176 /*
1177 * If we're in replay mode, or we made no progress, return error.
1178 * Otherwise, it's at least a partial write, so it's successful.
1179 */
1180 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1181 ZFS_EXIT(zfsvfs);
1182 return (error);
1183 }
1184
1185 if (ioflag & (FSYNC | FDSYNC) ||
1186 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1187 zil_commit(zilog, zp->z_id);
1188
1189 ZFS_EXIT(zfsvfs);
1190 return (0);
1191}
1192
1193void
1194zfs_get_done(zgd_t *zgd, int error)
1195{
1196 znode_t *zp = zgd->zgd_private;
1197 objset_t *os = zp->z_zfsvfs->z_os;
1198
1199 if (zgd->zgd_db)
1200 dmu_buf_rele(zgd->zgd_db, zgd);
1201
1202 zfs_range_unlock(zgd->zgd_rl);
1203
1204 /*
1205 * Release the vnode asynchronously as we currently have the
1206 * txg stopped from syncing.
1207 */
1208 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1209
1210 if (error == 0 && zgd->zgd_bp)
1211 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1212
1213 kmem_free(zgd, sizeof (zgd_t));
1214}
1215
1216#ifdef DEBUG
1217static int zil_fault_io = 0;
1218#endif
1219
1220/*
1221 * Get data to generate a TX_WRITE intent log record.
1222 */
1223int
1224zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1225{
1226 zfsvfs_t *zfsvfs = arg;
1227 objset_t *os = zfsvfs->z_os;
1228 znode_t *zp;
1229 uint64_t object = lr->lr_foid;
1230 uint64_t offset = lr->lr_offset;
1231 uint64_t size = lr->lr_length;
1232 blkptr_t *bp = &lr->lr_blkptr;
1233 dmu_buf_t *db;
1234 zgd_t *zgd;
1235 int error = 0;
1236
1237 ASSERT(zio != NULL);
1238 ASSERT(size != 0);
1239
1240 /*
1241 * Nothing to do if the file has been removed
1242 */
1243 if (zfs_zget(zfsvfs, object, &zp) != 0)
1244 return (SET_ERROR(ENOENT));
1245 if (zp->z_unlinked) {
1246 /*
1247 * Release the vnode asynchronously as we currently have the
1248 * txg stopped from syncing.
1249 */
1250 VN_RELE_ASYNC(ZTOV(zp),
1251 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1252 return (SET_ERROR(ENOENT));
1253 }
1254
1255 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1256 zgd->zgd_zilog = zfsvfs->z_log;
1257 zgd->zgd_private = zp;
1258
1259 /*
1260 * Write records come in two flavors: immediate and indirect.
1261 * For small writes it's cheaper to store the data with the
1262 * log record (immediate); for large writes it's cheaper to
1263 * sync the data and get a pointer to it (indirect) so that
1264 * we don't have to write the data twice.
1265 */
1266 if (buf != NULL) { /* immediate write */
1267 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1268 /* test for truncation needs to be done while range locked */
1269 if (offset >= zp->z_size) {
1270 error = SET_ERROR(ENOENT);
1271 } else {
1272 error = dmu_read(os, object, offset, size, buf,
1273 DMU_READ_NO_PREFETCH);
1274 }
1275 ASSERT(error == 0 || error == ENOENT);
1276 } else { /* indirect write */
1277 /*
1278 * Have to lock the whole block to ensure when it's
1279 * written out and it's checksum is being calculated
1280 * that no one can change the data. We need to re-check
1281 * blocksize after we get the lock in case it's changed!
1282 */
1283 for (;;) {
1284 uint64_t blkoff;
1285 size = zp->z_blksz;
1286 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1287 offset -= blkoff;
1288 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1289 RL_READER);
1290 if (zp->z_blksz == size)
1291 break;
1292 offset += blkoff;
1293 zfs_range_unlock(zgd->zgd_rl);
1294 }
1295 /* test for truncation needs to be done while range locked */
1296 if (lr->lr_offset >= zp->z_size)
1297 error = SET_ERROR(ENOENT);
1298#ifdef DEBUG
1299 if (zil_fault_io) {
1300 error = SET_ERROR(EIO);
1301 zil_fault_io = 0;
1302 }
1303#endif
1304 if (error == 0)
1305 error = dmu_buf_hold(os, object, offset, zgd, &db,
1306 DMU_READ_NO_PREFETCH);
1307
1308 if (error == 0) {
1309 blkptr_t *obp = dmu_buf_get_blkptr(db);
1310 if (obp) {
1311 ASSERT(BP_IS_HOLE(bp));
1312 *bp = *obp;
1313 }
1314
1315 zgd->zgd_db = db;
1316 zgd->zgd_bp = bp;
1317
1318 ASSERT(db->db_offset == offset);
1319 ASSERT(db->db_size == size);
1320
1321 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1322 zfs_get_done, zgd);
1323 ASSERT(error || lr->lr_length <= zp->z_blksz);
1324
1325 /*
1326 * On success, we need to wait for the write I/O
1327 * initiated by dmu_sync() to complete before we can
1328 * release this dbuf. We will finish everything up
1329 * in the zfs_get_done() callback.
1330 */
1331 if (error == 0)
1332 return (0);
1333
1334 if (error == EALREADY) {
1335 lr->lr_common.lrc_txtype = TX_WRITE2;
1336 error = 0;
1337 }
1338 }
1339 }
1340
1341 zfs_get_done(zgd, error);
1342
1343 return (error);
1344}
1345
1346/*ARGSUSED*/
1347static int
1348zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1349 caller_context_t *ct)
1350{
1351 znode_t *zp = VTOZ(vp);
1352 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1353 int error;
1354
1355 ZFS_ENTER(zfsvfs);
1356 ZFS_VERIFY_ZP(zp);
1357
1358 if (flag & V_ACE_MASK)
1359 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1360 else
1361 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1362
1363 ZFS_EXIT(zfsvfs);
1364 return (error);
1365}
1366
1367/*
1368 * If vnode is for a device return a specfs vnode instead.
1369 */
1370static int
1371specvp_check(vnode_t **vpp, cred_t *cr)
1372{
1373 int error = 0;
1374
1375 if (IS_DEVVP(*vpp)) {
1376 struct vnode *svp;
1377
1378 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1379 VN_RELE(*vpp);
1380 if (svp == NULL)
1381 error = SET_ERROR(ENOSYS);
1382 *vpp = svp;
1383 }
1384 return (error);
1385}
1386
1387
1388/*
1389 * Lookup an entry in a directory, or an extended attribute directory.
1390 * If it exists, return a held vnode reference for it.
1391 *
1392 * IN: dvp - vnode of directory to search.
1393 * nm - name of entry to lookup.
1394 * pnp - full pathname to lookup [UNUSED].
1395 * flags - LOOKUP_XATTR set if looking for an attribute.
1396 * rdir - root directory vnode [UNUSED].
1397 * cr - credentials of caller.
1398 * ct - caller context
1399 * direntflags - directory lookup flags
1400 * realpnp - returned pathname.
1401 *
1402 * OUT: vpp - vnode of located entry, NULL if not found.
1403 *
1404 * RETURN: 0 on success, error code on failure.
1405 *
1406 * Timestamps:
1407 * NA
1408 */
1409/* ARGSUSED */
1410static int
1411zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1412 int nameiop, cred_t *cr, kthread_t *td, int flags)
1413{
1414 znode_t *zdp = VTOZ(dvp);
1415 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1416 int error = 0;
1417 int *direntflags = NULL;
1418 void *realpnp = NULL;
1419
1420 /* fast path */
1421 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1422
1423 if (dvp->v_type != VDIR) {
1424 return (SET_ERROR(ENOTDIR));
1425 } else if (zdp->z_sa_hdl == NULL) {
1426 return (SET_ERROR(EIO));
1427 }
1428
1429 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1430 error = zfs_fastaccesschk_execute(zdp, cr);
1431 if (!error) {
1432 *vpp = dvp;
1433 VN_HOLD(*vpp);
1434 return (0);
1435 }
1436 return (error);
1437 } else {
1438 vnode_t *tvp = dnlc_lookup(dvp, nm);
1439
1440 if (tvp) {
1441 error = zfs_fastaccesschk_execute(zdp, cr);
1442 if (error) {
1443 VN_RELE(tvp);
1444 return (error);
1445 }
1446 if (tvp == DNLC_NO_VNODE) {
1447 VN_RELE(tvp);
1448 return (SET_ERROR(ENOENT));
1449 } else {
1450 *vpp = tvp;
1451 return (specvp_check(vpp, cr));
1452 }
1453 }
1454 }
1455 }
1456
1457 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1458
1459 ZFS_ENTER(zfsvfs);
1460 ZFS_VERIFY_ZP(zdp);
1461
1462 *vpp = NULL;
1463
1464 if (flags & LOOKUP_XATTR) {
1465#ifdef TODO
1466 /*
1467 * If the xattr property is off, refuse the lookup request.
1468 */
1469 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1470 ZFS_EXIT(zfsvfs);
1471 return (SET_ERROR(EINVAL));
1472 }
1473#endif
1474
1475 /*
1476 * We don't allow recursive attributes..
1477 * Maybe someday we will.
1478 */
1479 if (zdp->z_pflags & ZFS_XATTR) {
1480 ZFS_EXIT(zfsvfs);
1481 return (SET_ERROR(EINVAL));
1482 }
1483
1484 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1485 ZFS_EXIT(zfsvfs);
1486 return (error);
1487 }
1488
1489 /*
1490 * Do we have permission to get into attribute directory?
1491 */
1492
1493 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1494 B_FALSE, cr)) {
1495 VN_RELE(*vpp);
1496 *vpp = NULL;
1497 }
1498
1499 ZFS_EXIT(zfsvfs);
1500 return (error);
1501 }
1502
1503 if (dvp->v_type != VDIR) {
1504 ZFS_EXIT(zfsvfs);
1505 return (SET_ERROR(ENOTDIR));
1506 }
1507
1508 /*
1509 * Check accessibility of directory.
1510 */
1511
1512 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1513 ZFS_EXIT(zfsvfs);
1514 return (error);
1515 }
1516
1517 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1518 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1519 ZFS_EXIT(zfsvfs);
1520 return (SET_ERROR(EILSEQ));
1521 }
1522
1523 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1524 if (error == 0)
1525 error = specvp_check(vpp, cr);
1526
1527 /* Translate errors and add SAVENAME when needed. */
1528 if (cnp->cn_flags & ISLASTCN) {
1529 switch (nameiop) {
1530 case CREATE:
1531 case RENAME:
1532 if (error == ENOENT) {
1533 error = EJUSTRETURN;
1534 cnp->cn_flags |= SAVENAME;
1535 break;
1536 }
1537 /* FALLTHROUGH */
1538 case DELETE:
1539 if (error == 0)
1540 cnp->cn_flags |= SAVENAME;
1541 break;
1542 }
1543 }
1544 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1545 int ltype = 0;
1546
1547 if (cnp->cn_flags & ISDOTDOT) {
1548 ltype = VOP_ISLOCKED(dvp);
1549 VOP_UNLOCK(dvp, 0);
1550 }
1551 ZFS_EXIT(zfsvfs);
1552 error = vn_lock(*vpp, cnp->cn_lkflags);
1553 if (cnp->cn_flags & ISDOTDOT)
1554 vn_lock(dvp, ltype | LK_RETRY);
1555 if (error != 0) {
1556 VN_RELE(*vpp);
1557 *vpp = NULL;
1558 return (error);
1559 }
1560 } else {
1561 ZFS_EXIT(zfsvfs);
1562 }
1563
1564#ifdef FREEBSD_NAMECACHE
1565 /*
1566 * Insert name into cache (as non-existent) if appropriate.
1567 */
1568 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1569 cache_enter(dvp, *vpp, cnp);
1570 /*
1571 * Insert name into cache if appropriate.
1572 */
1573 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1574 if (!(cnp->cn_flags & ISLASTCN) ||
1575 (nameiop != DELETE && nameiop != RENAME)) {
1576 cache_enter(dvp, *vpp, cnp);
1577 }
1578 }
1579#endif
1580
1581 return (error);
1582}
1583
1584/*
1585 * Attempt to create a new entry in a directory. If the entry
1586 * already exists, truncate the file if permissible, else return
1587 * an error. Return the vp of the created or trunc'd file.
1588 *
1589 * IN: dvp - vnode of directory to put new file entry in.
1590 * name - name of new file entry.
1591 * vap - attributes of new file.
1592 * excl - flag indicating exclusive or non-exclusive mode.
1593 * mode - mode to open file with.
1594 * cr - credentials of caller.
1595 * flag - large file flag [UNUSED].
1596 * ct - caller context
1597 * vsecp - ACL to be set
1598 *
1599 * OUT: vpp - vnode of created or trunc'd entry.
1600 *
1601 * RETURN: 0 on success, error code on failure.
1602 *
1603 * Timestamps:
1604 * dvp - ctime|mtime updated if new entry created
1605 * vp - ctime|mtime always, atime if new
1606 */
1607
1608/* ARGSUSED */
1609static int
1610zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1611 vnode_t **vpp, cred_t *cr, kthread_t *td)
1612{
1613 znode_t *zp, *dzp = VTOZ(dvp);
1614 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1615 zilog_t *zilog;
1616 objset_t *os;
1617 zfs_dirlock_t *dl;
1618 dmu_tx_t *tx;
1619 int error;
1620 ksid_t *ksid;
1621 uid_t uid;
1622 gid_t gid = crgetgid(cr);
1623 zfs_acl_ids_t acl_ids;
1624 boolean_t fuid_dirtied;
1625 boolean_t have_acl = B_FALSE;
1626 boolean_t waited = B_FALSE;
1627 void *vsecp = NULL;
1628 int flag = 0;
1629
1630 /*
1631 * If we have an ephemeral id, ACL, or XVATTR then
1632 * make sure file system is at proper version
1633 */
1634
1635 ksid = crgetsid(cr, KSID_OWNER);
1636 if (ksid)
1637 uid = ksid_getid(ksid);
1638 else
1639 uid = crgetuid(cr);
1640
1641 if (zfsvfs->z_use_fuids == B_FALSE &&
1642 (vsecp || (vap->va_mask & AT_XVATTR) ||
1643 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1644 return (SET_ERROR(EINVAL));
1645
1646 ZFS_ENTER(zfsvfs);
1647 ZFS_VERIFY_ZP(dzp);
1648 os = zfsvfs->z_os;
1649 zilog = zfsvfs->z_log;
1650
1651 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1652 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1653 ZFS_EXIT(zfsvfs);
1654 return (SET_ERROR(EILSEQ));
1655 }
1656
1657 if (vap->va_mask & AT_XVATTR) {
1658 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1659 crgetuid(cr), cr, vap->va_type)) != 0) {
1660 ZFS_EXIT(zfsvfs);
1661 return (error);
1662 }
1663 }
1664top:
1665 *vpp = NULL;
1666
1667 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1668 vap->va_mode &= ~S_ISVTX;
1669
1670 if (*name == '\0') {
1671 /*
1672 * Null component name refers to the directory itself.
1673 */
1674 VN_HOLD(dvp);
1675 zp = dzp;
1676 dl = NULL;
1677 error = 0;
1678 } else {
1679 /* possible VN_HOLD(zp) */
1680 int zflg = 0;
1681
1682 if (flag & FIGNORECASE)
1683 zflg |= ZCILOOK;
1684
1685 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1686 NULL, NULL);
1687 if (error) {
1688 if (have_acl)
1689 zfs_acl_ids_free(&acl_ids);
1690 if (strcmp(name, "..") == 0)
1691 error = SET_ERROR(EISDIR);
1692 ZFS_EXIT(zfsvfs);
1693 return (error);
1694 }
1695 }
1696
1697 if (zp == NULL) {
1698 uint64_t txtype;
1699
1700 /*
1701 * Create a new file object and update the directory
1702 * to reference it.
1703 */
1704 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1705 if (have_acl)
1706 zfs_acl_ids_free(&acl_ids);
1707 goto out;
1708 }
1709
1710 /*
1711 * We only support the creation of regular files in
1712 * extended attribute directories.
1713 */
1714
1715 if ((dzp->z_pflags & ZFS_XATTR) &&
1716 (vap->va_type != VREG)) {
1717 if (have_acl)
1718 zfs_acl_ids_free(&acl_ids);
1719 error = SET_ERROR(EINVAL);
1720 goto out;
1721 }
1722
1723 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1724 cr, vsecp, &acl_ids)) != 0)
1725 goto out;
1726 have_acl = B_TRUE;
1727
1728 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1729 zfs_acl_ids_free(&acl_ids);
1730 error = SET_ERROR(EDQUOT);
1731 goto out;
1732 }
1733
1734 tx = dmu_tx_create(os);
1735
1736 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1737 ZFS_SA_BASE_ATTR_SIZE);
1738
1739 fuid_dirtied = zfsvfs->z_fuid_dirty;
1740 if (fuid_dirtied)
1741 zfs_fuid_txhold(zfsvfs, tx);
1742 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1743 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1744 if (!zfsvfs->z_use_sa &&
1745 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1746 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1747 0, acl_ids.z_aclp->z_acl_bytes);
1748 }
1749 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1750 if (error) {
1751 zfs_dirent_unlock(dl);
1752 if (error == ERESTART) {
1753 waited = B_TRUE;
1754 dmu_tx_wait(tx);
1755 dmu_tx_abort(tx);
1756 goto top;
1757 }
1758 zfs_acl_ids_free(&acl_ids);
1759 dmu_tx_abort(tx);
1760 ZFS_EXIT(zfsvfs);
1761 return (error);
1762 }
1763 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1764
1765 if (fuid_dirtied)
1766 zfs_fuid_sync(zfsvfs, tx);
1767
1768 (void) zfs_link_create(dl, zp, tx, ZNEW);
1769 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1770 if (flag & FIGNORECASE)
1771 txtype |= TX_CI;
1772 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1773 vsecp, acl_ids.z_fuidp, vap);
1774 zfs_acl_ids_free(&acl_ids);
1775 dmu_tx_commit(tx);
1776 } else {
1777 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1778
1779 if (have_acl)
1780 zfs_acl_ids_free(&acl_ids);
1781 have_acl = B_FALSE;
1782
1783 /*
1784 * A directory entry already exists for this name.
1785 */
1786 /*
1787 * Can't truncate an existing file if in exclusive mode.
1788 */
1789 if (excl == EXCL) {
1790 error = SET_ERROR(EEXIST);
1791 goto out;
1792 }
1793 /*
1794 * Can't open a directory for writing.
1795 */
1796 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1797 error = SET_ERROR(EISDIR);
1798 goto out;
1799 }
1800 /*
1801 * Verify requested access to file.
1802 */
1803 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1804 goto out;
1805 }
1806
1807 mutex_enter(&dzp->z_lock);
1808 dzp->z_seq++;
1809 mutex_exit(&dzp->z_lock);
1810
1811 /*
1812 * Truncate regular files if requested.
1813 */
1814 if ((ZTOV(zp)->v_type == VREG) &&
1815 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1816 /* we can't hold any locks when calling zfs_freesp() */
1817 zfs_dirent_unlock(dl);
1818 dl = NULL;
1819 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1820 if (error == 0) {
1821 vnevent_create(ZTOV(zp), ct);
1822 }
1823 }
1824 }
1825out:
1826 if (dl)
1827 zfs_dirent_unlock(dl);
1828
1829 if (error) {
1830 if (zp)
1831 VN_RELE(ZTOV(zp));
1832 } else {
1833 *vpp = ZTOV(zp);
1834 error = specvp_check(vpp, cr);
1835 }
1836
1837 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1838 zil_commit(zilog, 0);
1839
1840 ZFS_EXIT(zfsvfs);
1841 return (error);
1842}
1843
1844/*
1845 * Remove an entry from a directory.
1846 *
1847 * IN: dvp - vnode of directory to remove entry from.
1848 * name - name of entry to remove.
1849 * cr - credentials of caller.
1850 * ct - caller context
1851 * flags - case flags
1852 *
1853 * RETURN: 0 on success, error code on failure.
1854 *
1855 * Timestamps:
1856 * dvp - ctime|mtime
1857 * vp - ctime (if nlink > 0)
1858 */
1859
1860uint64_t null_xattr = 0;
1861
1862/*ARGSUSED*/
1863static int
1864zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1865 int flags)
1866{
1867 znode_t *zp, *dzp = VTOZ(dvp);
1868 znode_t *xzp;
1869 vnode_t *vp;
1870 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1871 zilog_t *zilog;
1872 uint64_t acl_obj, xattr_obj;
1873 uint64_t xattr_obj_unlinked = 0;
1874 uint64_t obj = 0;
1875 zfs_dirlock_t *dl;
1876 dmu_tx_t *tx;
1877 boolean_t may_delete_now, delete_now = FALSE;
1878 boolean_t unlinked, toobig = FALSE;
1879 uint64_t txtype;
1880 pathname_t *realnmp = NULL;
1881 pathname_t realnm;
1882 int error;
1883 int zflg = ZEXISTS;
1884 boolean_t waited = B_FALSE;
1885
1886 ZFS_ENTER(zfsvfs);
1887 ZFS_VERIFY_ZP(dzp);
1888 zilog = zfsvfs->z_log;
1889
1890 if (flags & FIGNORECASE) {
1891 zflg |= ZCILOOK;
1892 pn_alloc(&realnm);
1893 realnmp = &realnm;
1894 }
1895
1896top:
1897 xattr_obj = 0;
1898 xzp = NULL;
1899 /*
1900 * Attempt to lock directory; fail if entry doesn't exist.
1901 */
1902 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1903 NULL, realnmp)) {
1904 if (realnmp)
1905 pn_free(realnmp);
1906 ZFS_EXIT(zfsvfs);
1907 return (error);
1908 }
1909
1910 vp = ZTOV(zp);
1911
1912 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1913 goto out;
1914 }
1915
1916 /*
1917 * Need to use rmdir for removing directories.
1918 */
1919 if (vp->v_type == VDIR) {
1920 error = SET_ERROR(EPERM);
1921 goto out;
1922 }
1923
1924 vnevent_remove(vp, dvp, name, ct);
1925
1926 if (realnmp)
1927 dnlc_remove(dvp, realnmp->pn_buf);
1928 else
1929 dnlc_remove(dvp, name);
1930
1931 VI_LOCK(vp);
1932 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1933 VI_UNLOCK(vp);
1934
1935 /*
1936 * We may delete the znode now, or we may put it in the unlinked set;
1937 * it depends on whether we're the last link, and on whether there are
1938 * other holds on the vnode. So we dmu_tx_hold() the right things to
1939 * allow for either case.
1940 */
1941 obj = zp->z_id;
1942 tx = dmu_tx_create(zfsvfs->z_os);
1943 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1944 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1945 zfs_sa_upgrade_txholds(tx, zp);
1946 zfs_sa_upgrade_txholds(tx, dzp);
1947 if (may_delete_now) {
1948 toobig =
1949 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1950 /* if the file is too big, only hold_free a token amount */
1951 dmu_tx_hold_free(tx, zp->z_id, 0,
1952 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1953 }
1954
1955 /* are there any extended attributes? */
1956 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1957 &xattr_obj, sizeof (xattr_obj));
1958 if (error == 0 && xattr_obj) {
1959 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1960 ASSERT0(error);
1961 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1962 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1963 }
1964
1965 mutex_enter(&zp->z_lock);
1966 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1967 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1968 mutex_exit(&zp->z_lock);
1969
1970 /* charge as an update -- would be nice not to charge at all */
1971 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1972
1973 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1974 if (error) {
1975 zfs_dirent_unlock(dl);
1976 VN_RELE(vp);
1977 if (xzp)
1978 VN_RELE(ZTOV(xzp));
1979 if (error == ERESTART) {
1980 waited = B_TRUE;
1981 dmu_tx_wait(tx);
1982 dmu_tx_abort(tx);
1983 goto top;
1984 }
1985 if (realnmp)
1986 pn_free(realnmp);
1987 dmu_tx_abort(tx);
1988 ZFS_EXIT(zfsvfs);
1989 return (error);
1990 }
1991
1992 /*
1993 * Remove the directory entry.
1994 */
1995 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1996
1997 if (error) {
1998 dmu_tx_commit(tx);
1999 goto out;
2000 }
2001
2002 if (unlinked) {
2003
2004 /*
2005 * Hold z_lock so that we can make sure that the ACL obj
2006 * hasn't changed. Could have been deleted due to
2007 * zfs_sa_upgrade().
2008 */
2009 mutex_enter(&zp->z_lock);
2010 VI_LOCK(vp);
2011 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2012 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2013 delete_now = may_delete_now && !toobig &&
2014 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2015 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2016 acl_obj;
2017 VI_UNLOCK(vp);
2018 }
2019
2020 if (delete_now) {
2021#ifdef __FreeBSD__
2022 panic("zfs_remove: delete_now branch taken");
2023#endif
2024 if (xattr_obj_unlinked) {
2025 ASSERT3U(xzp->z_links, ==, 2);
2026 mutex_enter(&xzp->z_lock);
2027 xzp->z_unlinked = 1;
2028 xzp->z_links = 0;
2029 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2030 &xzp->z_links, sizeof (xzp->z_links), tx);
2031 ASSERT3U(error, ==, 0);
2032 mutex_exit(&xzp->z_lock);
2033 zfs_unlinked_add(xzp, tx);
2034
2035 if (zp->z_is_sa)
2036 error = sa_remove(zp->z_sa_hdl,
2037 SA_ZPL_XATTR(zfsvfs), tx);
2038 else
2039 error = sa_update(zp->z_sa_hdl,
2040 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2041 sizeof (uint64_t), tx);
2042 ASSERT0(error);
2043 }
2044 VI_LOCK(vp);
2045 vp->v_count--;
2046 ASSERT0(vp->v_count);
2047 VI_UNLOCK(vp);
2048 mutex_exit(&zp->z_lock);
2049 zfs_znode_delete(zp, tx);
2050 } else if (unlinked) {
2051 mutex_exit(&zp->z_lock);
2052 zfs_unlinked_add(zp, tx);
2053#ifdef __FreeBSD__
2054 vp->v_vflag |= VV_NOSYNC;
2055#endif
2056 }
2057
2058 txtype = TX_REMOVE;
2059 if (flags & FIGNORECASE)
2060 txtype |= TX_CI;
2061 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2062
2063 dmu_tx_commit(tx);
2064out:
2065 if (realnmp)
2066 pn_free(realnmp);
2067
2068 zfs_dirent_unlock(dl);
2069
2070 if (!delete_now)
2071 VN_RELE(vp);
2072 if (xzp)
2073 VN_RELE(ZTOV(xzp));
2074
2075 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2076 zil_commit(zilog, 0);
2077
2078 ZFS_EXIT(zfsvfs);
2079 return (error);
2080}
2081
2082/*
2083 * Create a new directory and insert it into dvp using the name
2084 * provided. Return a pointer to the inserted directory.
2085 *
2086 * IN: dvp - vnode of directory to add subdir to.
2087 * dirname - name of new directory.
2088 * vap - attributes of new directory.
2089 * cr - credentials of caller.
2090 * ct - caller context
2091 * flags - case flags
2092 * vsecp - ACL to be set
2093 *
2094 * OUT: vpp - vnode of created directory.
2095 *
2096 * RETURN: 0 on success, error code on failure.
2097 *
2098 * Timestamps:
2099 * dvp - ctime|mtime updated
2100 * vp - ctime|mtime|atime updated
2101 */
2102/*ARGSUSED*/
2103static int
2104zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2105 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2106{
2107 znode_t *zp, *dzp = VTOZ(dvp);
2108 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2109 zilog_t *zilog;
2110 zfs_dirlock_t *dl;
2111 uint64_t txtype;
2112 dmu_tx_t *tx;
2113 int error;
2114 int zf = ZNEW;
2115 ksid_t *ksid;
2116 uid_t uid;
2117 gid_t gid = crgetgid(cr);
2118 zfs_acl_ids_t acl_ids;
2119 boolean_t fuid_dirtied;
2120 boolean_t waited = B_FALSE;
2121
2122 ASSERT(vap->va_type == VDIR);
2123
2124 /*
2125 * If we have an ephemeral id, ACL, or XVATTR then
2126 * make sure file system is at proper version
2127 */
2128
2129 ksid = crgetsid(cr, KSID_OWNER);
2130 if (ksid)
2131 uid = ksid_getid(ksid);
2132 else
2133 uid = crgetuid(cr);
2134 if (zfsvfs->z_use_fuids == B_FALSE &&
2135 (vsecp || (vap->va_mask & AT_XVATTR) ||
2136 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2137 return (SET_ERROR(EINVAL));
2138
2139 ZFS_ENTER(zfsvfs);
2140 ZFS_VERIFY_ZP(dzp);
2141 zilog = zfsvfs->z_log;
2142
2143 if (dzp->z_pflags & ZFS_XATTR) {
2144 ZFS_EXIT(zfsvfs);
2145 return (SET_ERROR(EINVAL));
2146 }
2147
2148 if (zfsvfs->z_utf8 && u8_validate(dirname,
2149 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2150 ZFS_EXIT(zfsvfs);
2151 return (SET_ERROR(EILSEQ));
2152 }
2153 if (flags & FIGNORECASE)
2154 zf |= ZCILOOK;
2155
2156 if (vap->va_mask & AT_XVATTR) {
2157 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2158 crgetuid(cr), cr, vap->va_type)) != 0) {
2159 ZFS_EXIT(zfsvfs);
2160 return (error);
2161 }
2162 }
2163
2164 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2165 vsecp, &acl_ids)) != 0) {
2166 ZFS_EXIT(zfsvfs);
2167 return (error);
2168 }
2169 /*
2170 * First make sure the new directory doesn't exist.
2171 *
2172 * Existence is checked first to make sure we don't return
2173 * EACCES instead of EEXIST which can cause some applications
2174 * to fail.
2175 */
2176top:
2177 *vpp = NULL;
2178
2179 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2180 NULL, NULL)) {
2181 zfs_acl_ids_free(&acl_ids);
2182 ZFS_EXIT(zfsvfs);
2183 return (error);
2184 }
2185
2186 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2187 zfs_acl_ids_free(&acl_ids);
2188 zfs_dirent_unlock(dl);
2189 ZFS_EXIT(zfsvfs);
2190 return (error);
2191 }
2192
2193 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2194 zfs_acl_ids_free(&acl_ids);
2195 zfs_dirent_unlock(dl);
2196 ZFS_EXIT(zfsvfs);
2197 return (SET_ERROR(EDQUOT));
2198 }
2199
2200 /*
2201 * Add a new entry to the directory.
2202 */
2203 tx = dmu_tx_create(zfsvfs->z_os);
2204 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2205 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2206 fuid_dirtied = zfsvfs->z_fuid_dirty;
2207 if (fuid_dirtied)
2208 zfs_fuid_txhold(zfsvfs, tx);
2209 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2210 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2211 acl_ids.z_aclp->z_acl_bytes);
2212 }
2213
2214 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2215 ZFS_SA_BASE_ATTR_SIZE);
2216
2217 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2218 if (error) {
2219 zfs_dirent_unlock(dl);
2220 if (error == ERESTART) {
2221 waited = B_TRUE;
2222 dmu_tx_wait(tx);
2223 dmu_tx_abort(tx);
2224 goto top;
2225 }
2226 zfs_acl_ids_free(&acl_ids);
2227 dmu_tx_abort(tx);
2228 ZFS_EXIT(zfsvfs);
2229 return (error);
2230 }
2231
2232 /*
2233 * Create new node.
2234 */
2235 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2236
2237 if (fuid_dirtied)
2238 zfs_fuid_sync(zfsvfs, tx);
2239
2240 /*
2241 * Now put new name in parent dir.
2242 */
2243 (void) zfs_link_create(dl, zp, tx, ZNEW);
2244
2245 *vpp = ZTOV(zp);
2246
2247 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2248 if (flags & FIGNORECASE)
2249 txtype |= TX_CI;
2250 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2251 acl_ids.z_fuidp, vap);
2252
2253 zfs_acl_ids_free(&acl_ids);
2254
2255 dmu_tx_commit(tx);
2256
2257 zfs_dirent_unlock(dl);
2258
2259 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2260 zil_commit(zilog, 0);
2261
2262 ZFS_EXIT(zfsvfs);
2263 return (0);
2264}
2265
2266/*
2267 * Remove a directory subdir entry. If the current working
2268 * directory is the same as the subdir to be removed, the
2269 * remove will fail.
2270 *
2271 * IN: dvp - vnode of directory to remove from.
2272 * name - name of directory to be removed.
2273 * cwd - vnode of current working directory.
2274 * cr - credentials of caller.
2275 * ct - caller context
2276 * flags - case flags
2277 *
2278 * RETURN: 0 on success, error code on failure.
2279 *
2280 * Timestamps:
2281 * dvp - ctime|mtime updated
2282 */
2283/*ARGSUSED*/
2284static int
2285zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2286 caller_context_t *ct, int flags)
2287{
2288 znode_t *dzp = VTOZ(dvp);
2289 znode_t *zp;
2290 vnode_t *vp;
2291 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2292 zilog_t *zilog;
2293 zfs_dirlock_t *dl;
2294 dmu_tx_t *tx;
2295 int error;
2296 int zflg = ZEXISTS;
2297 boolean_t waited = B_FALSE;
2298
2299 ZFS_ENTER(zfsvfs);
2300 ZFS_VERIFY_ZP(dzp);
2301 zilog = zfsvfs->z_log;
2302
2303 if (flags & FIGNORECASE)
2304 zflg |= ZCILOOK;
2305top:
2306 zp = NULL;
2307
2308 /*
2309 * Attempt to lock directory; fail if entry doesn't exist.
2310 */
2311 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2312 NULL, NULL)) {
2313 ZFS_EXIT(zfsvfs);
2314 return (error);
2315 }
2316
2317 vp = ZTOV(zp);
2318
2319 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2320 goto out;
2321 }
2322
2323 if (vp->v_type != VDIR) {
2324 error = SET_ERROR(ENOTDIR);
2325 goto out;
2326 }
2327
2328 if (vp == cwd) {
2329 error = SET_ERROR(EINVAL);
2330 goto out;
2331 }
2332
2333 vnevent_rmdir(vp, dvp, name, ct);
2334
2335 /*
2336 * Grab a lock on the directory to make sure that noone is
2337 * trying to add (or lookup) entries while we are removing it.
2338 */
2339 rw_enter(&zp->z_name_lock, RW_WRITER);
2340
2341 /*
2342 * Grab a lock on the parent pointer to make sure we play well
2343 * with the treewalk and directory rename code.
2344 */
2345 rw_enter(&zp->z_parent_lock, RW_WRITER);
2346
2347 tx = dmu_tx_create(zfsvfs->z_os);
2348 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2349 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2350 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2351 zfs_sa_upgrade_txholds(tx, zp);
2352 zfs_sa_upgrade_txholds(tx, dzp);
2353 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2354 if (error) {
2355 rw_exit(&zp->z_parent_lock);
2356 rw_exit(&zp->z_name_lock);
2357 zfs_dirent_unlock(dl);
2358 VN_RELE(vp);
2359 if (error == ERESTART) {
2360 waited = B_TRUE;
2361 dmu_tx_wait(tx);
2362 dmu_tx_abort(tx);
2363 goto top;
2364 }
2365 dmu_tx_abort(tx);
2366 ZFS_EXIT(zfsvfs);
2367 return (error);
2368 }
2369
2370#ifdef FREEBSD_NAMECACHE
2371 cache_purge(dvp);
2372#endif
2373
2374 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2375
2376 if (error == 0) {
2377 uint64_t txtype = TX_RMDIR;
2378 if (flags & FIGNORECASE)
2379 txtype |= TX_CI;
2380 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2381 }
2382
2383 dmu_tx_commit(tx);
2384
2385 rw_exit(&zp->z_parent_lock);
2386 rw_exit(&zp->z_name_lock);
2387#ifdef FREEBSD_NAMECACHE
2388 cache_purge(vp);
2389#endif
2390out:
2391 zfs_dirent_unlock(dl);
2392
2393 VN_RELE(vp);
2394
2395 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2396 zil_commit(zilog, 0);
2397
2398 ZFS_EXIT(zfsvfs);
2399 return (error);
2400}
2401
2402/*
2403 * Read as many directory entries as will fit into the provided
2404 * buffer from the given directory cursor position (specified in
2405 * the uio structure).
2406 *
2407 * IN: vp - vnode of directory to read.
2408 * uio - structure supplying read location, range info,
2409 * and return buffer.
2410 * cr - credentials of caller.
2411 * ct - caller context
2412 * flags - case flags
2413 *
2414 * OUT: uio - updated offset and range, buffer filled.
2415 * eofp - set to true if end-of-file detected.
2416 *
2417 * RETURN: 0 on success, error code on failure.
2418 *
2419 * Timestamps:
2420 * vp - atime updated
2421 *
2422 * Note that the low 4 bits of the cookie returned by zap is always zero.
2423 * This allows us to use the low range for "special" directory entries:
2424 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2425 * we use the offset 2 for the '.zfs' directory.
2426 */
2427/* ARGSUSED */
2428static int
2429zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2430{
2431 znode_t *zp = VTOZ(vp);
2432 iovec_t *iovp;
2433 edirent_t *eodp;
2434 dirent64_t *odp;
2435 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2436 objset_t *os;
2437 caddr_t outbuf;
2438 size_t bufsize;
2439 zap_cursor_t zc;
2440 zap_attribute_t zap;
2441 uint_t bytes_wanted;
2442 uint64_t offset; /* must be unsigned; checks for < 1 */
2443 uint64_t parent;
2444 int local_eof;
2445 int outcount;
2446 int error;
2447 uint8_t prefetch;
2448 boolean_t check_sysattrs;
2449 uint8_t type;
2450 int ncooks;
2451 u_long *cooks = NULL;
2452 int flags = 0;
2453
2454 ZFS_ENTER(zfsvfs);
2455 ZFS_VERIFY_ZP(zp);
2456
2457 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2458 &parent, sizeof (parent))) != 0) {
2459 ZFS_EXIT(zfsvfs);
2460 return (error);
2461 }
2462
2463 /*
2464 * If we are not given an eof variable,
2465 * use a local one.
2466 */
2467 if (eofp == NULL)
2468 eofp = &local_eof;
2469
2470 /*
2471 * Check for valid iov_len.
2472 */
2473 if (uio->uio_iov->iov_len <= 0) {
2474 ZFS_EXIT(zfsvfs);
2475 return (SET_ERROR(EINVAL));
2476 }
2477
2478 /*
2479 * Quit if directory has been removed (posix)
2480 */
2481 if ((*eofp = zp->z_unlinked) != 0) {
2482 ZFS_EXIT(zfsvfs);
2483 return (0);
2484 }
2485
2486 error = 0;
2487 os = zfsvfs->z_os;
2488 offset = uio->uio_loffset;
2489 prefetch = zp->z_zn_prefetch;
2490
2491 /*
2492 * Initialize the iterator cursor.
2493 */
2494 if (offset <= 3) {
2495 /*
2496 * Start iteration from the beginning of the directory.
2497 */
2498 zap_cursor_init(&zc, os, zp->z_id);
2499 } else {
2500 /*
2501 * The offset is a serialized cursor.
2502 */
2503 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2504 }
2505
2506 /*
2507 * Get space to change directory entries into fs independent format.
2508 */
2509 iovp = uio->uio_iov;
2510 bytes_wanted = iovp->iov_len;
2511 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2512 bufsize = bytes_wanted;
2513 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2514 odp = (struct dirent64 *)outbuf;
2515 } else {
2516 bufsize = bytes_wanted;
2517 outbuf = NULL;
2518 odp = (struct dirent64 *)iovp->iov_base;
2519 }
2520 eodp = (struct edirent *)odp;
2521
2522 if (ncookies != NULL) {
2523 /*
2524 * Minimum entry size is dirent size and 1 byte for a file name.
2525 */
2526 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2527 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2528 *cookies = cooks;
2529 *ncookies = ncooks;
2530 }
2531 /*
2532 * If this VFS supports the system attribute view interface; and
2533 * we're looking at an extended attribute directory; and we care
2534 * about normalization conflicts on this vfs; then we must check
2535 * for normalization conflicts with the sysattr name space.
2536 */
2537#ifdef TODO
2538 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2539 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2540 (flags & V_RDDIR_ENTFLAGS);
2541#else
2542 check_sysattrs = 0;
2543#endif
2544
2545 /*
2546 * Transform to file-system independent format
2547 */
2548 outcount = 0;
2549 while (outcount < bytes_wanted) {
2550 ino64_t objnum;
2551 ushort_t reclen;
2552 off64_t *next = NULL;
2553
2554 /*
2555 * Special case `.', `..', and `.zfs'.
2556 */
2557 if (offset == 0) {
2558 (void) strcpy(zap.za_name, ".");
2559 zap.za_normalization_conflict = 0;
2560 objnum = zp->z_id;
2561 type = DT_DIR;
2562 } else if (offset == 1) {
2563 (void) strcpy(zap.za_name, "..");
2564 zap.za_normalization_conflict = 0;
2565 objnum = parent;
2566 type = DT_DIR;
2567 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2568 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2569 zap.za_normalization_conflict = 0;
2570 objnum = ZFSCTL_INO_ROOT;
2571 type = DT_DIR;
2572 } else {
2573 /*
2574 * Grab next entry.
2575 */
2576 if (error = zap_cursor_retrieve(&zc, &zap)) {
2577 if ((*eofp = (error == ENOENT)) != 0)
2578 break;
2579 else
2580 goto update;
2581 }
2582
2583 if (zap.za_integer_length != 8 ||
2584 zap.za_num_integers != 1) {
2585 cmn_err(CE_WARN, "zap_readdir: bad directory "
2586 "entry, obj = %lld, offset = %lld\n",
2587 (u_longlong_t)zp->z_id,
2588 (u_longlong_t)offset);
2589 error = SET_ERROR(ENXIO);
2590 goto update;
2591 }
2592
2593 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2594 /*
2595 * MacOS X can extract the object type here such as:
2596 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2597 */
2598 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2599
2600 if (check_sysattrs && !zap.za_normalization_conflict) {
2601#ifdef TODO
2602 zap.za_normalization_conflict =
2603 xattr_sysattr_casechk(zap.za_name);
2604#else
2605 panic("%s:%u: TODO", __func__, __LINE__);
2606#endif
2607 }
2608 }
2609
2610 if (flags & V_RDDIR_ACCFILTER) {
2611 /*
2612 * If we have no access at all, don't include
2613 * this entry in the returned information
2614 */
2615 znode_t *ezp;
2616 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2617 goto skip_entry;
2618 if (!zfs_has_access(ezp, cr)) {
2619 VN_RELE(ZTOV(ezp));
2620 goto skip_entry;
2621 }
2622 VN_RELE(ZTOV(ezp));
2623 }
2624
2625 if (flags & V_RDDIR_ENTFLAGS)
2626 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2627 else
2628 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2629
2630 /*
2631 * Will this entry fit in the buffer?
2632 */
2633 if (outcount + reclen > bufsize) {
2634 /*
2635 * Did we manage to fit anything in the buffer?
2636 */
2637 if (!outcount) {
2638 error = SET_ERROR(EINVAL);
2639 goto update;
2640 }
2641 break;
2642 }
2643 if (flags & V_RDDIR_ENTFLAGS) {
2644 /*
2645 * Add extended flag entry:
2646 */
2647 eodp->ed_ino = objnum;
2648 eodp->ed_reclen = reclen;
2649 /* NOTE: ed_off is the offset for the *next* entry */
2650 next = &(eodp->ed_off);
2651 eodp->ed_eflags = zap.za_normalization_conflict ?
2652 ED_CASE_CONFLICT : 0;
2653 (void) strncpy(eodp->ed_name, zap.za_name,
2654 EDIRENT_NAMELEN(reclen));
2655 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2656 } else {
2657 /*
2658 * Add normal entry:
2659 */
2660 odp->d_ino = objnum;
2661 odp->d_reclen = reclen;
2662 odp->d_namlen = strlen(zap.za_name);
2663 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2664 odp->d_type = type;
2665 odp = (dirent64_t *)((intptr_t)odp + reclen);
2666 }
2667 outcount += reclen;
2668
2669 ASSERT(outcount <= bufsize);
2670
2671 /* Prefetch znode */
2672 if (prefetch)
2673 dmu_prefetch(os, objnum, 0, 0);
2674
2675 skip_entry:
2676 /*
2677 * Move to the next entry, fill in the previous offset.
2678 */
2679 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2680 zap_cursor_advance(&zc);
2681 offset = zap_cursor_serialize(&zc);
2682 } else {
2683 offset += 1;
2684 }
2685
2686 if (cooks != NULL) {
2687 *cooks++ = offset;
2688 ncooks--;
2689 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2690 }
2691 }
2692 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2693
2694 /* Subtract unused cookies */
2695 if (ncookies != NULL)
2696 *ncookies -= ncooks;
2697
2698 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2699 iovp->iov_base += outcount;
2700 iovp->iov_len -= outcount;
2701 uio->uio_resid -= outcount;
2702 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2703 /*
2704 * Reset the pointer.
2705 */
2706 offset = uio->uio_loffset;
2707 }
2708
2709update:
2710 zap_cursor_fini(&zc);
2711 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2712 kmem_free(outbuf, bufsize);
2713
2714 if (error == ENOENT)
2715 error = 0;
2716
2717 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2718
2719 uio->uio_loffset = offset;
2720 ZFS_EXIT(zfsvfs);
2721 if (error != 0 && cookies != NULL) {
2722 free(*cookies, M_TEMP);
2723 *cookies = NULL;
2724 *ncookies = 0;
2725 }
2726 return (error);
2727}
2728
2729ulong_t zfs_fsync_sync_cnt = 4;
2730
2731static int
2732zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2733{
2734 znode_t *zp = VTOZ(vp);
2735 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2736
2737 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2738
2739 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2740 ZFS_ENTER(zfsvfs);
2741 ZFS_VERIFY_ZP(zp);
2742 zil_commit(zfsvfs->z_log, zp->z_id);
2743 ZFS_EXIT(zfsvfs);
2744 }
2745 return (0);
2746}
2747
2748
2749/*
2750 * Get the requested file attributes and place them in the provided
2751 * vattr structure.
2752 *
2753 * IN: vp - vnode of file.
2754 * vap - va_mask identifies requested attributes.
2755 * If AT_XVATTR set, then optional attrs are requested
2756 * flags - ATTR_NOACLCHECK (CIFS server context)
2757 * cr - credentials of caller.
2758 * ct - caller context
2759 *
2760 * OUT: vap - attribute values.
2761 *
2762 * RETURN: 0 (always succeeds).
2763 */
2764/* ARGSUSED */
2765static int
2766zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2767 caller_context_t *ct)
2768{
2769 znode_t *zp = VTOZ(vp);
2770 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2771 int error = 0;
2772 uint32_t blksize;
2773 u_longlong_t nblocks;
2774 uint64_t links;
2775 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2776 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2777 xoptattr_t *xoap = NULL;
2778 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2779 sa_bulk_attr_t bulk[4];
2780 int count = 0;
2781
2782 ZFS_ENTER(zfsvfs);
2783 ZFS_VERIFY_ZP(zp);
2784
2785 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2786
2787 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2788 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2789 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2790 if (vp->v_type == VBLK || vp->v_type == VCHR)
2791 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2792 &rdev, 8);
2793
2794 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2795 ZFS_EXIT(zfsvfs);
2796 return (error);
2797 }
2798
2799 /*
2800 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2801 * Also, if we are the owner don't bother, since owner should
2802 * always be allowed to read basic attributes of file.
2803 */
2804 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2805 (vap->va_uid != crgetuid(cr))) {
2806 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2807 skipaclchk, cr)) {
2808 ZFS_EXIT(zfsvfs);
2809 return (error);
2810 }
2811 }
2812
2813 /*
2814 * Return all attributes. It's cheaper to provide the answer
2815 * than to determine whether we were asked the question.
2816 */
2817
2818 mutex_enter(&zp->z_lock);
2819 vap->va_type = IFTOVT(zp->z_mode);
2820 vap->va_mode = zp->z_mode & ~S_IFMT;
2821#ifdef sun
2822 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2823#else
2824 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2825#endif
2826 vap->va_nodeid = zp->z_id;
2827 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2828 links = zp->z_links + 1;
2829 else
2830 links = zp->z_links;
2831 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2832 vap->va_size = zp->z_size;
2833#ifdef sun
2834 vap->va_rdev = vp->v_rdev;
2835#else
2836 if (vp->v_type == VBLK || vp->v_type == VCHR)
2837 vap->va_rdev = zfs_cmpldev(rdev);
2838#endif
2839 vap->va_seq = zp->z_seq;
2840 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2841
2842 /*
2843 * Add in any requested optional attributes and the create time.
2844 * Also set the corresponding bits in the returned attribute bitmap.
2845 */
2846 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2847 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2848 xoap->xoa_archive =
2849 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2850 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2851 }
2852
2853 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2854 xoap->xoa_readonly =
2855 ((zp->z_pflags & ZFS_READONLY) != 0);
2856 XVA_SET_RTN(xvap, XAT_READONLY);
2857 }
2858
2859 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2860 xoap->xoa_system =
2861 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2862 XVA_SET_RTN(xvap, XAT_SYSTEM);
2863 }
2864
2865 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2866 xoap->xoa_hidden =
2867 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2868 XVA_SET_RTN(xvap, XAT_HIDDEN);
2869 }
2870
2871 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2872 xoap->xoa_nounlink =
2873 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2874 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2875 }
2876
2877 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2878 xoap->xoa_immutable =
2879 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2880 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2881 }
2882
2883 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2884 xoap->xoa_appendonly =
2885 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2886 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2887 }
2888
2889 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2890 xoap->xoa_nodump =
2891 ((zp->z_pflags & ZFS_NODUMP) != 0);
2892 XVA_SET_RTN(xvap, XAT_NODUMP);
2893 }
2894
2895 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2896 xoap->xoa_opaque =
2897 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2898 XVA_SET_RTN(xvap, XAT_OPAQUE);
2899 }
2900
2901 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2902 xoap->xoa_av_quarantined =
2903 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2904 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2905 }
2906
2907 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2908 xoap->xoa_av_modified =
2909 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2910 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2911 }
2912
2913 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2914 vp->v_type == VREG) {
2915 zfs_sa_get_scanstamp(zp, xvap);
2916 }
2917
2918 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2919 uint64_t times[2];
2920
2921 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2922 times, sizeof (times));
2923 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2924 XVA_SET_RTN(xvap, XAT_CREATETIME);
2925 }
2926
2927 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2928 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2929 XVA_SET_RTN(xvap, XAT_REPARSE);
2930 }
2931 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2932 xoap->xoa_generation = zp->z_gen;
2933 XVA_SET_RTN(xvap, XAT_GEN);
2934 }
2935
2936 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2937 xoap->xoa_offline =
2938 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2939 XVA_SET_RTN(xvap, XAT_OFFLINE);
2940 }
2941
2942 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2943 xoap->xoa_sparse =
2944 ((zp->z_pflags & ZFS_SPARSE) != 0);
2945 XVA_SET_RTN(xvap, XAT_SPARSE);
2946 }
2947 }
2948
2949 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2950 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2951 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2952 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2953
2954 mutex_exit(&zp->z_lock);
2955
2956 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2957 vap->va_blksize = blksize;
2958 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2959
2960 if (zp->z_blksz == 0) {
2961 /*
2962 * Block size hasn't been set; suggest maximal I/O transfers.
2963 */
2964 vap->va_blksize = zfsvfs->z_max_blksz;
2965 }
2966
2967 ZFS_EXIT(zfsvfs);
2968 return (0);
2969}
2970
2971/*
2972 * Set the file attributes to the values contained in the
2973 * vattr structure.
2974 *
2975 * IN: vp - vnode of file to be modified.
2976 * vap - new attribute values.
2977 * If AT_XVATTR set, then optional attrs are being set
2978 * flags - ATTR_UTIME set if non-default time values provided.
2979 * - ATTR_NOACLCHECK (CIFS context only).
2980 * cr - credentials of caller.
2981 * ct - caller context
2982 *
2983 * RETURN: 0 on success, error code on failure.
2984 *
2985 * Timestamps:
2986 * vp - ctime updated, mtime updated if size changed.
2987 */
2988/* ARGSUSED */
2989static int
2990zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2991 caller_context_t *ct)
2992{
2993 znode_t *zp = VTOZ(vp);
2994 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2995 zilog_t *zilog;
2996 dmu_tx_t *tx;
2997 vattr_t oldva;
2998 xvattr_t tmpxvattr;
2999 uint_t mask = vap->va_mask;
3000 uint_t saved_mask = 0;
3001 uint64_t saved_mode;
3002 int trim_mask = 0;
3003 uint64_t new_mode;
3004 uint64_t new_uid, new_gid;
3005 uint64_t xattr_obj;
3006 uint64_t mtime[2], ctime[2];
3007 znode_t *attrzp;
3008 int need_policy = FALSE;
3009 int err, err2;
3010 zfs_fuid_info_t *fuidp = NULL;
3011 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3012 xoptattr_t *xoap;
3013 zfs_acl_t *aclp;
3014 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3015 boolean_t fuid_dirtied = B_FALSE;
3016 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3017 int count = 0, xattr_count = 0;
3018
3019 if (mask == 0)
3020 return (0);
3021
3022 if (mask & AT_NOSET)
3023 return (SET_ERROR(EINVAL));
3024
3025 ZFS_ENTER(zfsvfs);
3026 ZFS_VERIFY_ZP(zp);
3027
3028 zilog = zfsvfs->z_log;
3029
3030 /*
3031 * Make sure that if we have ephemeral uid/gid or xvattr specified
3032 * that file system is at proper version level
3033 */
3034
3035 if (zfsvfs->z_use_fuids == B_FALSE &&
3036 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3037 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3038 (mask & AT_XVATTR))) {
3039 ZFS_EXIT(zfsvfs);
3040 return (SET_ERROR(EINVAL));
3041 }
3042
3043 if (mask & AT_SIZE && vp->v_type == VDIR) {
3044 ZFS_EXIT(zfsvfs);
3045 return (SET_ERROR(EISDIR));
3046 }
3047
3048 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3049 ZFS_EXIT(zfsvfs);
3050 return (SET_ERROR(EINVAL));
3051 }
3052
3053 /*
3054 * If this is an xvattr_t, then get a pointer to the structure of
3055 * optional attributes. If this is NULL, then we have a vattr_t.
3056 */
3057 xoap = xva_getxoptattr(xvap);
3058
3059 xva_init(&tmpxvattr);
3060
3061 /*
3062 * Immutable files can only alter immutable bit and atime
3063 */
3064 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3065 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3066 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3067 ZFS_EXIT(zfsvfs);
3068 return (SET_ERROR(EPERM));
3069 }
3070
3071 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3072 ZFS_EXIT(zfsvfs);
3073 return (SET_ERROR(EPERM));
3074 }
3075
3076 /*
3077 * Verify timestamps doesn't overflow 32 bits.
3078 * ZFS can handle large timestamps, but 32bit syscalls can't
3079 * handle times greater than 2039. This check should be removed
3080 * once large timestamps are fully supported.
3081 */
3082 if (mask & (AT_ATIME | AT_MTIME)) {
3083 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3084 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3085 ZFS_EXIT(zfsvfs);
3086 return (SET_ERROR(EOVERFLOW));
3087 }
3088 }
3089
3090top:
3091 attrzp = NULL;
3092 aclp = NULL;
3093
3094 /* Can this be moved to before the top label? */
3095 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3096 ZFS_EXIT(zfsvfs);
3097 return (SET_ERROR(EROFS));
3098 }
3099
3100 /*
3101 * First validate permissions
3102 */
3103
3104 if (mask & AT_SIZE) {
3105 /*
3106 * XXX - Note, we are not providing any open
3107 * mode flags here (like FNDELAY), so we may
3108 * block if there are locks present... this
3109 * should be addressed in openat().
3110 */
3111 /* XXX - would it be OK to generate a log record here? */
3112 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3113 if (err) {
3114 ZFS_EXIT(zfsvfs);
3115 return (err);
3116 }
3117 }
3118
3119 if (mask & (AT_ATIME|AT_MTIME) ||
3120 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3121 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3122 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3123 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3124 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3125 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3126 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3127 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3128 skipaclchk, cr);
3129 }
3130
3131 if (mask & (AT_UID|AT_GID)) {
3132 int idmask = (mask & (AT_UID|AT_GID));
3133 int take_owner;
3134 int take_group;
3135
3136 /*
3137 * NOTE: even if a new mode is being set,
3138 * we may clear S_ISUID/S_ISGID bits.
3139 */
3140
3141 if (!(mask & AT_MODE))
3142 vap->va_mode = zp->z_mode;
3143
3144 /*
3145 * Take ownership or chgrp to group we are a member of
3146 */
3147
3148 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3149 take_group = (mask & AT_GID) &&
3150 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3151
3152 /*
3153 * If both AT_UID and AT_GID are set then take_owner and
3154 * take_group must both be set in order to allow taking
3155 * ownership.
3156 *
3157 * Otherwise, send the check through secpolicy_vnode_setattr()
3158 *
3159 */
3160
3161 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3162 ((idmask == AT_UID) && take_owner) ||
3163 ((idmask == AT_GID) && take_group)) {
3164 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3165 skipaclchk, cr) == 0) {
3166 /*
3167 * Remove setuid/setgid for non-privileged users
3168 */
3169 secpolicy_setid_clear(vap, vp, cr);
3170 trim_mask = (mask & (AT_UID|AT_GID));
3171 } else {
3172 need_policy = TRUE;
3173 }
3174 } else {
3175 need_policy = TRUE;
3176 }
3177 }
3178
3179 mutex_enter(&zp->z_lock);
3180 oldva.va_mode = zp->z_mode;
3181 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3182 if (mask & AT_XVATTR) {
3183 /*
3184 * Update xvattr mask to include only those attributes
3185 * that are actually changing.
3186 *
3187 * the bits will be restored prior to actually setting
3188 * the attributes so the caller thinks they were set.
3189 */
3190 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3191 if (xoap->xoa_appendonly !=
3192 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3193 need_policy = TRUE;
3194 } else {
3195 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3196 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3197 }
3198 }
3199
3200 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3201 if (xoap->xoa_nounlink !=
3202 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3203 need_policy = TRUE;
3204 } else {
3205 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3206 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3207 }
3208 }
3209
3210 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3211 if (xoap->xoa_immutable !=
3212 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3213 need_policy = TRUE;
3214 } else {
3215 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3216 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3217 }
3218 }
3219
3220 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3221 if (xoap->xoa_nodump !=
3222 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3223 need_policy = TRUE;
3224 } else {
3225 XVA_CLR_REQ(xvap, XAT_NODUMP);
3226 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3227 }
3228 }
3229
3230 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3231 if (xoap->xoa_av_modified !=
3232 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3233 need_policy = TRUE;
3234 } else {
3235 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3236 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3237 }
3238 }
3239
3240 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3241 if ((vp->v_type != VREG &&
3242 xoap->xoa_av_quarantined) ||
3243 xoap->xoa_av_quarantined !=
3244 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3245 need_policy = TRUE;
3246 } else {
3247 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3248 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3249 }
3250 }
3251
3252 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3253 mutex_exit(&zp->z_lock);
3254 ZFS_EXIT(zfsvfs);
3255 return (SET_ERROR(EPERM));
3256 }
3257
3258 if (need_policy == FALSE &&
3259 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3260 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3261 need_policy = TRUE;
3262 }
3263 }
3264
3265 mutex_exit(&zp->z_lock);
3266
3267 if (mask & AT_MODE) {
3268 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3269 err = secpolicy_setid_setsticky_clear(vp, vap,
3270 &oldva, cr);
3271 if (err) {
3272 ZFS_EXIT(zfsvfs);
3273 return (err);
3274 }
3275 trim_mask |= AT_MODE;
3276 } else {
3277 need_policy = TRUE;
3278 }
3279 }
3280
3281 if (need_policy) {
3282 /*
3283 * If trim_mask is set then take ownership
3284 * has been granted or write_acl is present and user
3285 * has the ability to modify mode. In that case remove
3286 * UID|GID and or MODE from mask so that
3287 * secpolicy_vnode_setattr() doesn't revoke it.
3288 */
3289
3290 if (trim_mask) {
3291 saved_mask = vap->va_mask;
3292 vap->va_mask &= ~trim_mask;
3293 if (trim_mask & AT_MODE) {
3294 /*
3295 * Save the mode, as secpolicy_vnode_setattr()
3296 * will overwrite it with ova.va_mode.
3297 */
3298 saved_mode = vap->va_mode;
3299 }
3300 }
3301 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3302 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3303 if (err) {
3304 ZFS_EXIT(zfsvfs);
3305 return (err);
3306 }
3307
3308 if (trim_mask) {
3309 vap->va_mask |= saved_mask;
3310 if (trim_mask & AT_MODE) {
3311 /*
3312 * Recover the mode after
3313 * secpolicy_vnode_setattr().
3314 */
3315 vap->va_mode = saved_mode;
3316 }
3317 }
3318 }
3319
3320 /*
3321 * secpolicy_vnode_setattr, or take ownership may have
3322 * changed va_mask
3323 */
3324 mask = vap->va_mask;
3325
3326 if ((mask & (AT_UID | AT_GID))) {
3327 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3328 &xattr_obj, sizeof (xattr_obj));
3329
3330 if (err == 0 && xattr_obj) {
3331 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3332 if (err)
3333 goto out2;
3334 }
3335 if (mask & AT_UID) {
3336 new_uid = zfs_fuid_create(zfsvfs,
3337 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3338 if (new_uid != zp->z_uid &&
3339 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3340 if (attrzp)
3341 VN_RELE(ZTOV(attrzp));
3342 err = SET_ERROR(EDQUOT);
3343 goto out2;
3344 }
3345 }
3346
3347 if (mask & AT_GID) {
3348 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3349 cr, ZFS_GROUP, &fuidp);
3350 if (new_gid != zp->z_gid &&
3351 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3352 if (attrzp)
3353 VN_RELE(ZTOV(attrzp));
3354 err = SET_ERROR(EDQUOT);
3355 goto out2;
3356 }
3357 }
3358 }
3359 tx = dmu_tx_create(zfsvfs->z_os);
3360
3361 if (mask & AT_MODE) {
3362 uint64_t pmode = zp->z_mode;
3363 uint64_t acl_obj;
3364 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3365
3366 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3367 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3368 err = SET_ERROR(EPERM);
3369 goto out;
3370 }
3371
3372 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3373 goto out;
3374
3375 mutex_enter(&zp->z_lock);
3376 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3377 /*
3378 * Are we upgrading ACL from old V0 format
3379 * to V1 format?
3380 */
3381 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3382 zfs_znode_acl_version(zp) ==
3383 ZFS_ACL_VERSION_INITIAL) {
3384 dmu_tx_hold_free(tx, acl_obj, 0,
3385 DMU_OBJECT_END);
3386 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3387 0, aclp->z_acl_bytes);
3388 } else {
3389 dmu_tx_hold_write(tx, acl_obj, 0,
3390 aclp->z_acl_bytes);
3391 }
3392 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3393 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3394 0, aclp->z_acl_bytes);
3395 }
3396 mutex_exit(&zp->z_lock);
3397 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3398 } else {
3399 if ((mask & AT_XVATTR) &&
3400 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3401 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3402 else
3403 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3404 }
3405
3406 if (attrzp) {
3407 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3408 }
3409
3410 fuid_dirtied = zfsvfs->z_fuid_dirty;
3411 if (fuid_dirtied)
3412 zfs_fuid_txhold(zfsvfs, tx);
3413
3414 zfs_sa_upgrade_txholds(tx, zp);
3415
3416 err = dmu_tx_assign(tx, TXG_WAIT);
3417 if (err)
3418 goto out;
3419
3420 count = 0;
3421 /*
3422 * Set each attribute requested.
3423 * We group settings according to the locks they need to acquire.
3424 *
3425 * Note: you cannot set ctime directly, although it will be
3426 * updated as a side-effect of calling this function.
3427 */
3428
3429
3430 if (mask & (AT_UID|AT_GID|AT_MODE))
3431 mutex_enter(&zp->z_acl_lock);
3432 mutex_enter(&zp->z_lock);
3433
3434 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3435 &zp->z_pflags, sizeof (zp->z_pflags));
3436
3437 if (attrzp) {
3438 if (mask & (AT_UID|AT_GID|AT_MODE))
3439 mutex_enter(&attrzp->z_acl_lock);
3440 mutex_enter(&attrzp->z_lock);
3441 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3442 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3443 sizeof (attrzp->z_pflags));
3444 }
3445
3446 if (mask & (AT_UID|AT_GID)) {
3447
3448 if (mask & AT_UID) {
3449 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3450 &new_uid, sizeof (new_uid));
3451 zp->z_uid = new_uid;
3452 if (attrzp) {
3453 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3454 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3455 sizeof (new_uid));
3456 attrzp->z_uid = new_uid;
3457 }
3458 }
3459
3460 if (mask & AT_GID) {
3461 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3462 NULL, &new_gid, sizeof (new_gid));
3463 zp->z_gid = new_gid;
3464 if (attrzp) {
3465 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3466 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3467 sizeof (new_gid));
3468 attrzp->z_gid = new_gid;
3469 }
3470 }
3471 if (!(mask & AT_MODE)) {
3472 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3473 NULL, &new_mode, sizeof (new_mode));
3474 new_mode = zp->z_mode;
3475 }
3476 err = zfs_acl_chown_setattr(zp);
3477 ASSERT(err == 0);
3478 if (attrzp) {
3479 err = zfs_acl_chown_setattr(attrzp);
3480 ASSERT(err == 0);
3481 }
3482 }
3483
3484 if (mask & AT_MODE) {
3485 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3486 &new_mode, sizeof (new_mode));
3487 zp->z_mode = new_mode;
3488 ASSERT3U((uintptr_t)aclp, !=, 0);
3489 err = zfs_aclset_common(zp, aclp, cr, tx);
3490 ASSERT0(err);
3491 if (zp->z_acl_cached)
3492 zfs_acl_free(zp->z_acl_cached);
3493 zp->z_acl_cached = aclp;
3494 aclp = NULL;
3495 }
3496
3497
3498 if (mask & AT_ATIME) {
3499 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3500 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3501 &zp->z_atime, sizeof (zp->z_atime));
3502 }
3503
3504 if (mask & AT_MTIME) {
3505 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3506 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3507 mtime, sizeof (mtime));
3508 }
3509
3510 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3511 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3512 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3513 NULL, mtime, sizeof (mtime));
3514 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3515 &ctime, sizeof (ctime));
3516 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3517 B_TRUE);
3518 } else if (mask != 0) {
3519 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3520 &ctime, sizeof (ctime));
3521 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3522 B_TRUE);
3523 if (attrzp) {
3524 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3525 SA_ZPL_CTIME(zfsvfs), NULL,
3526 &ctime, sizeof (ctime));
3527 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3528 mtime, ctime, B_TRUE);
3529 }
3530 }
3531 /*
3532 * Do this after setting timestamps to prevent timestamp
3533 * update from toggling bit
3534 */
3535
3536 if (xoap && (mask & AT_XVATTR)) {
3537
3538 /*
3539 * restore trimmed off masks
3540 * so that return masks can be set for caller.
3541 */
3542
3543 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3544 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3545 }
3546 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3547 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3548 }
3549 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3550 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3551 }
3552 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3553 XVA_SET_REQ(xvap, XAT_NODUMP);
3554 }
3555 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3556 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3557 }
3558 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3559 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3560 }
3561
3562 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3563 ASSERT(vp->v_type == VREG);
3564
3565 zfs_xvattr_set(zp, xvap, tx);
3566 }
3567
3568 if (fuid_dirtied)
3569 zfs_fuid_sync(zfsvfs, tx);
3570
3571 if (mask != 0)
3572 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3573
3574 mutex_exit(&zp->z_lock);
3575 if (mask & (AT_UID|AT_GID|AT_MODE))
3576 mutex_exit(&zp->z_acl_lock);
3577
3578 if (attrzp) {
3579 if (mask & (AT_UID|AT_GID|AT_MODE))
3580 mutex_exit(&attrzp->z_acl_lock);
3581 mutex_exit(&attrzp->z_lock);
3582 }
3583out:
3584 if (err == 0 && attrzp) {
3585 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3586 xattr_count, tx);
3587 ASSERT(err2 == 0);
3588 }
3589
3590 if (attrzp)
3591 VN_RELE(ZTOV(attrzp));
3592
3593 if (aclp)
3594 zfs_acl_free(aclp);
3595
3596 if (fuidp) {
3597 zfs_fuid_info_free(fuidp);
3598 fuidp = NULL;
3599 }
3600
3601 if (err) {
3602 dmu_tx_abort(tx);
3603 if (err == ERESTART)
3604 goto top;
3605 } else {
3606 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3607 dmu_tx_commit(tx);
3608 }
3609
3610out2:
3611 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3612 zil_commit(zilog, 0);
3613
3614 ZFS_EXIT(zfsvfs);
3615 return (err);
3616}
3617
3618typedef struct zfs_zlock {
3619 krwlock_t *zl_rwlock; /* lock we acquired */
3620 znode_t *zl_znode; /* znode we held */
3621 struct zfs_zlock *zl_next; /* next in list */
3622} zfs_zlock_t;
3623
3624/*
3625 * Drop locks and release vnodes that were held by zfs_rename_lock().
3626 */
3627static void
3628zfs_rename_unlock(zfs_zlock_t **zlpp)
3629{
3630 zfs_zlock_t *zl;
3631
3632 while ((zl = *zlpp) != NULL) {
3633 if (zl->zl_znode != NULL)
3634 VN_RELE(ZTOV(zl->zl_znode));
3635 rw_exit(zl->zl_rwlock);
3636 *zlpp = zl->zl_next;
3637 kmem_free(zl, sizeof (*zl));
3638 }
3639}
3640
3641/*
3642 * Search back through the directory tree, using the ".." entries.
3643 * Lock each directory in the chain to prevent concurrent renames.
3644 * Fail any attempt to move a directory into one of its own descendants.
3645 * XXX - z_parent_lock can overlap with map or grow locks
3646 */
3647static int
3648zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3649{
3650 zfs_zlock_t *zl;
3651 znode_t *zp = tdzp;
3652 uint64_t rootid = zp->z_zfsvfs->z_root;
3653 uint64_t oidp = zp->z_id;
3654 krwlock_t *rwlp = &szp->z_parent_lock;
3655 krw_t rw = RW_WRITER;
3656
3657 /*
3658 * First pass write-locks szp and compares to zp->z_id.
3659 * Later passes read-lock zp and compare to zp->z_parent.
3660 */
3661 do {
3662 if (!rw_tryenter(rwlp, rw)) {
3663 /*
3664 * Another thread is renaming in this path.
3665 * Note that if we are a WRITER, we don't have any
3666 * parent_locks held yet.
3667 */
3668 if (rw == RW_READER && zp->z_id > szp->z_id) {
3669 /*
3670 * Drop our locks and restart
3671 */
3672 zfs_rename_unlock(&zl);
3673 *zlpp = NULL;
3674 zp = tdzp;
3675 oidp = zp->z_id;
3676 rwlp = &szp->z_parent_lock;
3677 rw = RW_WRITER;
3678 continue;
3679 } else {
3680 /*
3681 * Wait for other thread to drop its locks
3682 */
3683 rw_enter(rwlp, rw);
3684 }
3685 }
3686
3687 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3688 zl->zl_rwlock = rwlp;
3689 zl->zl_znode = NULL;
3690 zl->zl_next = *zlpp;
3691 *zlpp = zl;
3692
3693 if (oidp == szp->z_id) /* We're a descendant of szp */
3694 return (SET_ERROR(EINVAL));
3695
3696 if (oidp == rootid) /* We've hit the top */
3697 return (0);
3698
3699 if (rw == RW_READER) { /* i.e. not the first pass */
3700 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3701 if (error)
3702 return (error);
3703 zl->zl_znode = zp;
3704 }
3705 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3706 &oidp, sizeof (oidp));
3707 rwlp = &zp->z_parent_lock;
3708 rw = RW_READER;
3709
3710 } while (zp->z_id != sdzp->z_id);
3711
3712 return (0);
3713}
3714
3715/*
3716 * Move an entry from the provided source directory to the target
3717 * directory. Change the entry name as indicated.
3718 *
3719 * IN: sdvp - Source directory containing the "old entry".
3720 * snm - Old entry name.
3721 * tdvp - Target directory to contain the "new entry".
3722 * tnm - New entry name.
3723 * cr - credentials of caller.
3724 * ct - caller context
3725 * flags - case flags
3726 *
3727 * RETURN: 0 on success, error code on failure.
3728 *
3729 * Timestamps:
3730 * sdvp,tdvp - ctime|mtime updated
3731 */
3732/*ARGSUSED*/
3733static int
3734zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3735 caller_context_t *ct, int flags)
3736{
3737 znode_t *tdzp, *szp, *tzp;
3738 znode_t *sdzp = VTOZ(sdvp);
3739 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3740 zilog_t *zilog;
3741 vnode_t *realvp;
3742 zfs_dirlock_t *sdl, *tdl;
3743 dmu_tx_t *tx;
3744 zfs_zlock_t *zl;
3745 int cmp, serr, terr;
3746 int error = 0;
3747 int zflg = 0;
3748 boolean_t waited = B_FALSE;
3749
3750 ZFS_ENTER(zfsvfs);
3751 ZFS_VERIFY_ZP(sdzp);
3752 zilog = zfsvfs->z_log;
3753
3754 /*
3755 * Make sure we have the real vp for the target directory.
3756 */
3757 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3758 tdvp = realvp;
3759
3760 tdzp = VTOZ(tdvp);
3761 ZFS_VERIFY_ZP(tdzp);
3762
3763 /*
3764 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3765 * ctldir appear to have the same v_vfsp.
3766 */
3767 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3768 ZFS_EXIT(zfsvfs);
3769 return (SET_ERROR(EXDEV));
3770 }
3771
3772 if (zfsvfs->z_utf8 && u8_validate(tnm,
3773 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3774 ZFS_EXIT(zfsvfs);
3775 return (SET_ERROR(EILSEQ));
3776 }
3777
3778 if (flags & FIGNORECASE)
3779 zflg |= ZCILOOK;
3780
3781top:
3782 szp = NULL;
3783 tzp = NULL;
3784 zl = NULL;
3785
3786 /*
3787 * This is to prevent the creation of links into attribute space
3788 * by renaming a linked file into/outof an attribute directory.
3789 * See the comment in zfs_link() for why this is considered bad.
3790 */
3791 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3792 ZFS_EXIT(zfsvfs);
3793 return (SET_ERROR(EINVAL));
3794 }
3795
3796 /*
3797 * Lock source and target directory entries. To prevent deadlock,
3798 * a lock ordering must be defined. We lock the directory with
3799 * the smallest object id first, or if it's a tie, the one with
3800 * the lexically first name.
3801 */
3802 if (sdzp->z_id < tdzp->z_id) {
3803 cmp = -1;
3804 } else if (sdzp->z_id > tdzp->z_id) {
3805 cmp = 1;
3806 } else {
3807 /*
3808 * First compare the two name arguments without
3809 * considering any case folding.
3810 */
3811 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3812
3813 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3814 ASSERT(error == 0 || !zfsvfs->z_utf8);
3815 if (cmp == 0) {
3816 /*
3817 * POSIX: "If the old argument and the new argument
3818 * both refer to links to the same existing file,
3819 * the rename() function shall return successfully
3820 * and perform no other action."
3821 */
3822 ZFS_EXIT(zfsvfs);
3823 return (0);
3824 }
3825 /*
3826 * If the file system is case-folding, then we may
3827 * have some more checking to do. A case-folding file
3828 * system is either supporting mixed case sensitivity
3829 * access or is completely case-insensitive. Note
3830 * that the file system is always case preserving.
3831 *
3832 * In mixed sensitivity mode case sensitive behavior
3833 * is the default. FIGNORECASE must be used to
3834 * explicitly request case insensitive behavior.
3835 *
3836 * If the source and target names provided differ only
3837 * by case (e.g., a request to rename 'tim' to 'Tim'),
3838 * we will treat this as a special case in the
3839 * case-insensitive mode: as long as the source name
3840 * is an exact match, we will allow this to proceed as
3841 * a name-change request.
3842 */
3843 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3844 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3845 flags & FIGNORECASE)) &&
3846 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3847 &error) == 0) {
3848 /*
3849 * case preserving rename request, require exact
3850 * name matches
3851 */
3852 zflg |= ZCIEXACT;
3853 zflg &= ~ZCILOOK;
3854 }
3855 }
3856
3857 /*
3858 * If the source and destination directories are the same, we should
3859 * grab the z_name_lock of that directory only once.
3860 */
3861 if (sdzp == tdzp) {
3862 zflg |= ZHAVELOCK;
3863 rw_enter(&sdzp->z_name_lock, RW_READER);
3864 }
3865
3866 if (cmp < 0) {
3867 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3868 ZEXISTS | zflg, NULL, NULL);
3869 terr = zfs_dirent_lock(&tdl,
3870 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3871 } else {
3872 terr = zfs_dirent_lock(&tdl,
3873 tdzp, tnm, &tzp, zflg, NULL, NULL);
3874 serr = zfs_dirent_lock(&sdl,
3875 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3876 NULL, NULL);
3877 }
3878
3879 if (serr) {
3880 /*
3881 * Source entry invalid or not there.
3882 */
3883 if (!terr) {
3884 zfs_dirent_unlock(tdl);
3885 if (tzp)
3886 VN_RELE(ZTOV(tzp));
3887 }
3888
3889 if (sdzp == tdzp)
3890 rw_exit(&sdzp->z_name_lock);
3891
3892 /*
3893 * FreeBSD: In OpenSolaris they only check if rename source is
3894 * ".." here, because "." is handled in their lookup. This is
3895 * not the case for FreeBSD, so we check for "." explicitly.
3896 */
3897 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0)
3898 serr = SET_ERROR(EINVAL);
3899 ZFS_EXIT(zfsvfs);
3900 return (serr);
3901 }
3902 if (terr) {
3903 zfs_dirent_unlock(sdl);
3904 VN_RELE(ZTOV(szp));
3905
3906 if (sdzp == tdzp)
3907 rw_exit(&sdzp->z_name_lock);
3908
3909 if (strcmp(tnm, "..") == 0)
3910 terr = SET_ERROR(EINVAL);
3911 ZFS_EXIT(zfsvfs);
3912 return (terr);
3913 }
3914
3915 /*
3916 * Must have write access at the source to remove the old entry
3917 * and write access at the target to create the new entry.
3918 * Note that if target and source are the same, this can be
3919 * done in a single check.
3920 */
3921
3922 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3923 goto out;
3924
3925 if (ZTOV(szp)->v_type == VDIR) {
3926 /*
3927 * Check to make sure rename is valid.
3928 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3929 */
3930 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3931 goto out;
3932 }
3933
3934 /*
3935 * Does target exist?
3936 */
3937 if (tzp) {
3938 /*
3939 * Source and target must be the same type.
3940 */
3941 if (ZTOV(szp)->v_type == VDIR) {
3942 if (ZTOV(tzp)->v_type != VDIR) {
3943 error = SET_ERROR(ENOTDIR);
3944 goto out;
3945 }
3946 } else {
3947 if (ZTOV(tzp)->v_type == VDIR) {
3948 error = SET_ERROR(EISDIR);
3949 goto out;
3950 }
3951 }
3952 /*
3953 * POSIX dictates that when the source and target
3954 * entries refer to the same file object, rename
3955 * must do nothing and exit without error.
3956 */
3957 if (szp->z_id == tzp->z_id) {
3958 error = 0;
3959 goto out;
3960 }
3961 }
3962
3963 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3964 if (tzp)
3965 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3966
3967 /*
3968 * notify the target directory if it is not the same
3969 * as source directory.
3970 */
3971 if (tdvp != sdvp) {
3972 vnevent_rename_dest_dir(tdvp, ct);
3973 }
3974
3975 tx = dmu_tx_create(zfsvfs->z_os);
3976 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3977 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3978 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3979 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3980 if (sdzp != tdzp) {
3981 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3982 zfs_sa_upgrade_txholds(tx, tdzp);
3983 }
3984 if (tzp) {
3985 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3986 zfs_sa_upgrade_txholds(tx, tzp);
3987 }
3988
3989 zfs_sa_upgrade_txholds(tx, szp);
3990 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3991 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3992 if (error) {
3993 if (zl != NULL)
3994 zfs_rename_unlock(&zl);
3995 zfs_dirent_unlock(sdl);
3996 zfs_dirent_unlock(tdl);
3997
3998 if (sdzp == tdzp)
3999 rw_exit(&sdzp->z_name_lock);
4000
4001 VN_RELE(ZTOV(szp));
4002 if (tzp)
4003 VN_RELE(ZTOV(tzp));
4004 if (error == ERESTART) {
4005 waited = B_TRUE;
4006 dmu_tx_wait(tx);
4007 dmu_tx_abort(tx);
4008 goto top;
4009 }
4010 dmu_tx_abort(tx);
4011 ZFS_EXIT(zfsvfs);
4012 return (error);
4013 }
4014
4015 if (tzp) /* Attempt to remove the existing target */
4016 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
4017
4018 if (error == 0) {
4019 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
4020 if (error == 0) {
4021 szp->z_pflags |= ZFS_AV_MODIFIED;
4022
4023 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
4024 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
4025 ASSERT0(error);
4026
4027 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
4028 if (error == 0) {
4029 zfs_log_rename(zilog, tx, TX_RENAME |
4030 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
4031 sdl->dl_name, tdzp, tdl->dl_name, szp);
4032
4033 /*
4034 * Update path information for the target vnode
4035 */
4036 vn_renamepath(tdvp, ZTOV(szp), tnm,
4037 strlen(tnm));
4038 } else {
4039 /*
4040 * At this point, we have successfully created
4041 * the target name, but have failed to remove
4042 * the source name. Since the create was done
4043 * with the ZRENAMING flag, there are
4044 * complications; for one, the link count is
4045 * wrong. The easiest way to deal with this
4046 * is to remove the newly created target, and
4047 * return the original error. This must
4048 * succeed; fortunately, it is very unlikely to
4049 * fail, since we just created it.
4050 */
4051 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
4052 ZRENAMING, NULL), ==, 0);
4053 }
4054 }
4055#ifdef FREEBSD_NAMECACHE
4056 if (error == 0) {
4057 cache_purge(sdvp);
4058 cache_purge(tdvp);
4059 cache_purge(ZTOV(szp));
4060 if (tzp)
4061 cache_purge(ZTOV(tzp));
4062 }
4063#endif
4064 }
4065
4066 dmu_tx_commit(tx);
4067out:
4068 if (zl != NULL)
4069 zfs_rename_unlock(&zl);
4070
4071 zfs_dirent_unlock(sdl);
4072 zfs_dirent_unlock(tdl);
4073
4074 if (sdzp == tdzp)
4075 rw_exit(&sdzp->z_name_lock);
4076
4077
4078 VN_RELE(ZTOV(szp));
4079 if (tzp)
4080 VN_RELE(ZTOV(tzp));
4081
4082 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4083 zil_commit(zilog, 0);
4084
4085 ZFS_EXIT(zfsvfs);
4086
4087 return (error);
4088}
4089
4090/*
4091 * Insert the indicated symbolic reference entry into the directory.
4092 *
4093 * IN: dvp - Directory to contain new symbolic link.
4094 * link - Name for new symlink entry.
4095 * vap - Attributes of new entry.
4096 * cr - credentials of caller.
4097 * ct - caller context
4098 * flags - case flags
4099 *
4100 * RETURN: 0 on success, error code on failure.
4101 *
4102 * Timestamps:
4103 * dvp - ctime|mtime updated
4104 */
4105/*ARGSUSED*/
4106static int
4107zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link,
4108 cred_t *cr, kthread_t *td)
4109{
4110 znode_t *zp, *dzp = VTOZ(dvp);
4111 zfs_dirlock_t *dl;
4112 dmu_tx_t *tx;
4113 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4114 zilog_t *zilog;
4115 uint64_t len = strlen(link);
4116 int error;
4117 int zflg = ZNEW;
4118 zfs_acl_ids_t acl_ids;
4119 boolean_t fuid_dirtied;
4120 uint64_t txtype = TX_SYMLINK;
4121 boolean_t waited = B_FALSE;
4122 int flags = 0;
4123
4124 ASSERT(vap->va_type == VLNK);
4125
4126 ZFS_ENTER(zfsvfs);
4127 ZFS_VERIFY_ZP(dzp);
4128 zilog = zfsvfs->z_log;
4129
4130 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
4131 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4132 ZFS_EXIT(zfsvfs);
4133 return (SET_ERROR(EILSEQ));
4134 }
4135 if (flags & FIGNORECASE)
4136 zflg |= ZCILOOK;
4137
4138 if (len > MAXPATHLEN) {
4139 ZFS_EXIT(zfsvfs);
4140 return (SET_ERROR(ENAMETOOLONG));
4141 }
4142
4143 if ((error = zfs_acl_ids_create(dzp, 0,
4144 vap, cr, NULL, &acl_ids)) != 0) {
4145 ZFS_EXIT(zfsvfs);
4146 return (error);
4147 }
4148top:
4149 /*
4150 * Attempt to lock directory; fail if entry already exists.
4151 */
4152 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
4153 if (error) {
4154 zfs_acl_ids_free(&acl_ids);
4155 ZFS_EXIT(zfsvfs);
4156 return (error);
4157 }
4158
4159 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4160 zfs_acl_ids_free(&acl_ids);
4161 zfs_dirent_unlock(dl);
4162 ZFS_EXIT(zfsvfs);
4163 return (error);
4164 }
4165
4166 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
4167 zfs_acl_ids_free(&acl_ids);
4168 zfs_dirent_unlock(dl);
4169 ZFS_EXIT(zfsvfs);
4170 return (SET_ERROR(EDQUOT));
4171 }
4172 tx = dmu_tx_create(zfsvfs->z_os);
4173 fuid_dirtied = zfsvfs->z_fuid_dirty;
4174 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
4175 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4176 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
4177 ZFS_SA_BASE_ATTR_SIZE + len);
4178 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
4179 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
4180 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
4181 acl_ids.z_aclp->z_acl_bytes);
4182 }
4183 if (fuid_dirtied)
4184 zfs_fuid_txhold(zfsvfs, tx);
4185 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4186 if (error) {
4187 zfs_dirent_unlock(dl);
4188 if (error == ERESTART) {
4189 waited = B_TRUE;
4190 dmu_tx_wait(tx);
4191 dmu_tx_abort(tx);
4192 goto top;
4193 }
4194 zfs_acl_ids_free(&acl_ids);
4195 dmu_tx_abort(tx);
4196 ZFS_EXIT(zfsvfs);
4197 return (error);
4198 }
4199
4200 /*
4201 * Create a new object for the symlink.
4202 * for version 4 ZPL datsets the symlink will be an SA attribute
4203 */
4204 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
4205
4206 if (fuid_dirtied)
4207 zfs_fuid_sync(zfsvfs, tx);
4208
4209 mutex_enter(&zp->z_lock);
4210 if (zp->z_is_sa)
4211 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
4212 link, len, tx);
4213 else
4214 zfs_sa_symlink(zp, link, len, tx);
4215 mutex_exit(&zp->z_lock);
4216
4217 zp->z_size = len;
4218 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
4219 &zp->z_size, sizeof (zp->z_size), tx);
4220 /*
4221 * Insert the new object into the directory.
4222 */
4223 (void) zfs_link_create(dl, zp, tx, ZNEW);
4224
4225 if (flags & FIGNORECASE)
4226 txtype |= TX_CI;
4227 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
4228 *vpp = ZTOV(zp);
4229
4230 zfs_acl_ids_free(&acl_ids);
4231
4232 dmu_tx_commit(tx);
4233
4234 zfs_dirent_unlock(dl);
4235
4236 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4237 zil_commit(zilog, 0);
4238
4239 ZFS_EXIT(zfsvfs);
4240 return (error);
4241}
4242
4243/*
4244 * Return, in the buffer contained in the provided uio structure,
4245 * the symbolic path referred to by vp.
4246 *
4247 * IN: vp - vnode of symbolic link.
4248 * uio - structure to contain the link path.
4249 * cr - credentials of caller.
4250 * ct - caller context
4251 *
4252 * OUT: uio - structure containing the link path.
4253 *
4254 * RETURN: 0 on success, error code on failure.
4255 *
4256 * Timestamps:
4257 * vp - atime updated
4258 */
4259/* ARGSUSED */
4260static int
4261zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4262{
4263 znode_t *zp = VTOZ(vp);
4264 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4265 int error;
4266
4267 ZFS_ENTER(zfsvfs);
4268 ZFS_VERIFY_ZP(zp);
4269
4270 mutex_enter(&zp->z_lock);
4271 if (zp->z_is_sa)
4272 error = sa_lookup_uio(zp->z_sa_hdl,
4273 SA_ZPL_SYMLINK(zfsvfs), uio);
4274 else
4275 error = zfs_sa_readlink(zp, uio);
4276 mutex_exit(&zp->z_lock);
4277
4278 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4279
4280 ZFS_EXIT(zfsvfs);
4281 return (error);
4282}
4283
4284/*
4285 * Insert a new entry into directory tdvp referencing svp.
4286 *
4287 * IN: tdvp - Directory to contain new entry.
4288 * svp - vnode of new entry.
4289 * name - name of new entry.
4290 * cr - credentials of caller.
4291 * ct - caller context
4292 *
4293 * RETURN: 0 on success, error code on failure.
4294 *
4295 * Timestamps:
4296 * tdvp - ctime|mtime updated
4297 * svp - ctime updated
4298 */
4299/* ARGSUSED */
4300static int
4301zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4302 caller_context_t *ct, int flags)
4303{
4304 znode_t *dzp = VTOZ(tdvp);
4305 znode_t *tzp, *szp;
4306 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4307 zilog_t *zilog;
4308 zfs_dirlock_t *dl;
4309 dmu_tx_t *tx;
4310 vnode_t *realvp;
4311 int error;
4312 int zf = ZNEW;
4313 uint64_t parent;
4314 uid_t owner;
4315 boolean_t waited = B_FALSE;
4316
4317 ASSERT(tdvp->v_type == VDIR);
4318
4319 ZFS_ENTER(zfsvfs);
4320 ZFS_VERIFY_ZP(dzp);
4321 zilog = zfsvfs->z_log;
4322
4323 if (VOP_REALVP(svp, &realvp, ct) == 0)
4324 svp = realvp;
4325
4326 /*
4327 * POSIX dictates that we return EPERM here.
4328 * Better choices include ENOTSUP or EISDIR.
4329 */
4330 if (svp->v_type == VDIR) {
4331 ZFS_EXIT(zfsvfs);
4332 return (SET_ERROR(EPERM));
4333 }
4334
4335 szp = VTOZ(svp);
4336 ZFS_VERIFY_ZP(szp);
4337
4338 /*
4339 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4340 * ctldir appear to have the same v_vfsp.
4341 */
4342 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4343 ZFS_EXIT(zfsvfs);
4344 return (SET_ERROR(EXDEV));
4345 }
4346
4347 /* Prevent links to .zfs/shares files */
4348
4349 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4350 &parent, sizeof (uint64_t))) != 0) {
4351 ZFS_EXIT(zfsvfs);
4352 return (error);
4353 }
4354 if (parent == zfsvfs->z_shares_dir) {
4355 ZFS_EXIT(zfsvfs);
4356 return (SET_ERROR(EPERM));
4357 }
4358
4359 if (zfsvfs->z_utf8 && u8_validate(name,
4360 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4361 ZFS_EXIT(zfsvfs);
4362 return (SET_ERROR(EILSEQ));
4363 }
4364 if (flags & FIGNORECASE)
4365 zf |= ZCILOOK;
4366
4367 /*
4368 * We do not support links between attributes and non-attributes
4369 * because of the potential security risk of creating links
4370 * into "normal" file space in order to circumvent restrictions
4371 * imposed in attribute space.
4372 */
4373 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4374 ZFS_EXIT(zfsvfs);
4375 return (SET_ERROR(EINVAL));
4376 }
4377
4378
4379 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4380 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4381 ZFS_EXIT(zfsvfs);
4382 return (SET_ERROR(EPERM));
4383 }
4384
4385 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4386 ZFS_EXIT(zfsvfs);
4387 return (error);
4388 }
4389
4390top:
4391 /*
4392 * Attempt to lock directory; fail if entry already exists.
4393 */
4394 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4395 if (error) {
4396 ZFS_EXIT(zfsvfs);
4397 return (error);
4398 }
4399
4400 tx = dmu_tx_create(zfsvfs->z_os);
4401 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4402 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4403 zfs_sa_upgrade_txholds(tx, szp);
4404 zfs_sa_upgrade_txholds(tx, dzp);
4405 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4406 if (error) {
4407 zfs_dirent_unlock(dl);
4408 if (error == ERESTART) {
4409 waited = B_TRUE;
4410 dmu_tx_wait(tx);
4411 dmu_tx_abort(tx);
4412 goto top;
4413 }
4414 dmu_tx_abort(tx);
4415 ZFS_EXIT(zfsvfs);
4416 return (error);
4417 }
4418
4419 error = zfs_link_create(dl, szp, tx, 0);
4420
4421 if (error == 0) {
4422 uint64_t txtype = TX_LINK;
4423 if (flags & FIGNORECASE)
4424 txtype |= TX_CI;
4425 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4426 }
4427
4428 dmu_tx_commit(tx);
4429
4430 zfs_dirent_unlock(dl);
4431
4432 if (error == 0) {
4433 vnevent_link(svp, ct);
4434 }
4435
4436 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4437 zil_commit(zilog, 0);
4438
4439 ZFS_EXIT(zfsvfs);
4440 return (error);
4441}
4442
4443#ifdef sun
4444/*
4445 * zfs_null_putapage() is used when the file system has been force
4446 * unmounted. It just drops the pages.
4447 */
4448/* ARGSUSED */
4449static int
4450zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4451 size_t *lenp, int flags, cred_t *cr)
4452{
4453 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4454 return (0);
4455}
4456
4457/*
4458 * Push a page out to disk, klustering if possible.
4459 *
4460 * IN: vp - file to push page to.
4461 * pp - page to push.
4462 * flags - additional flags.
4463 * cr - credentials of caller.
4464 *
4465 * OUT: offp - start of range pushed.
4466 * lenp - len of range pushed.
4467 *
4468 * RETURN: 0 on success, error code on failure.
4469 *
4470 * NOTE: callers must have locked the page to be pushed. On
4471 * exit, the page (and all other pages in the kluster) must be
4472 * unlocked.
4473 */
4474/* ARGSUSED */
4475static int
4476zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4477 size_t *lenp, int flags, cred_t *cr)
4478{
4479 znode_t *zp = VTOZ(vp);
4480 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4481 dmu_tx_t *tx;
4482 u_offset_t off, koff;
4483 size_t len, klen;
4484 int err;
4485
4486 off = pp->p_offset;
4487 len = PAGESIZE;
4488 /*
4489 * If our blocksize is bigger than the page size, try to kluster
4490 * multiple pages so that we write a full block (thus avoiding
4491 * a read-modify-write).
4492 */
4493 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4494 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4495 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4496 ASSERT(koff <= zp->z_size);
4497 if (koff + klen > zp->z_size)
4498 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4499 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4500 }
4501 ASSERT3U(btop(len), ==, btopr(len));
4502
4503 /*
4504 * Can't push pages past end-of-file.
4505 */
4506 if (off >= zp->z_size) {
4507 /* ignore all pages */
4508 err = 0;
4509 goto out;
4510 } else if (off + len > zp->z_size) {
4511 int npages = btopr(zp->z_size - off);
4512 page_t *trunc;
4513
4514 page_list_break(&pp, &trunc, npages);
4515 /* ignore pages past end of file */
4516 if (trunc)
4517 pvn_write_done(trunc, flags);
4518 len = zp->z_size - off;
4519 }
4520
4521 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4522 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4523 err = SET_ERROR(EDQUOT);
4524 goto out;
4525 }
4526 tx = dmu_tx_create(zfsvfs->z_os);
4527 dmu_tx_hold_write(tx, zp->z_id, off, len);
4528
4529 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4530 zfs_sa_upgrade_txholds(tx, zp);
4531 err = dmu_tx_assign(tx, TXG_WAIT);
4532 if (err != 0) {
4533 dmu_tx_abort(tx);
4534 goto out;
4535 }
4536
4537 if (zp->z_blksz <= PAGESIZE) {
4538 caddr_t va = zfs_map_page(pp, S_READ);
4539 ASSERT3U(len, <=, PAGESIZE);
4540 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4541 zfs_unmap_page(pp, va);
4542 } else {
4543 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4544 }
4545
4546 if (err == 0) {
4547 uint64_t mtime[2], ctime[2];
4548 sa_bulk_attr_t bulk[3];
4549 int count = 0;
4550
4551 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4552 &mtime, 16);
4553 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4554 &ctime, 16);
4555 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4556 &zp->z_pflags, 8);
4557 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4558 B_TRUE);
4559 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4560 }
4561 dmu_tx_commit(tx);
4562
4563out:
4564 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4565 if (offp)
4566 *offp = off;
4567 if (lenp)
4568 *lenp = len;
4569
4570 return (err);
4571}
4572
4573/*
4574 * Copy the portion of the file indicated from pages into the file.
4575 * The pages are stored in a page list attached to the files vnode.
4576 *
4577 * IN: vp - vnode of file to push page data to.
4578 * off - position in file to put data.
4579 * len - amount of data to write.
4580 * flags - flags to control the operation.
4581 * cr - credentials of caller.
4582 * ct - caller context.
4583 *
4584 * RETURN: 0 on success, error code on failure.
4585 *
4586 * Timestamps:
4587 * vp - ctime|mtime updated
4588 */
4589/*ARGSUSED*/
4590static int
4591zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4592 caller_context_t *ct)
4593{
4594 znode_t *zp = VTOZ(vp);
4595 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4596 page_t *pp;
4597 size_t io_len;
4598 u_offset_t io_off;
4599 uint_t blksz;
4600 rl_t *rl;
4601 int error = 0;
4602
4603 ZFS_ENTER(zfsvfs);
4604 ZFS_VERIFY_ZP(zp);
4605
4606 /*
4607 * Align this request to the file block size in case we kluster.
4608 * XXX - this can result in pretty aggresive locking, which can
4609 * impact simultanious read/write access. One option might be
4610 * to break up long requests (len == 0) into block-by-block
4611 * operations to get narrower locking.
4612 */
4613 blksz = zp->z_blksz;
4614 if (ISP2(blksz))
4615 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4616 else
4617 io_off = 0;
4618 if (len > 0 && ISP2(blksz))
4619 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4620 else
4621 io_len = 0;
4622
4623 if (io_len == 0) {
4624 /*
4625 * Search the entire vp list for pages >= io_off.
4626 */
4627 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4628 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4629 goto out;
4630 }
4631 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4632
4633 if (off > zp->z_size) {
4634 /* past end of file */
4635 zfs_range_unlock(rl);
4636 ZFS_EXIT(zfsvfs);
4637 return (0);
4638 }
4639
4640 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4641
4642 for (off = io_off; io_off < off + len; io_off += io_len) {
4643 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4644 pp = page_lookup(vp, io_off,
4645 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4646 } else {
4647 pp = page_lookup_nowait(vp, io_off,
4648 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4649 }
4650
4651 if (pp != NULL && pvn_getdirty(pp, flags)) {
4652 int err;
4653
4654 /*
4655 * Found a dirty page to push
4656 */
4657 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4658 if (err)
4659 error = err;
4660 } else {
4661 io_len = PAGESIZE;
4662 }
4663 }
4664out:
4665 zfs_range_unlock(rl);
4666 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4667 zil_commit(zfsvfs->z_log, zp->z_id);
4668 ZFS_EXIT(zfsvfs);
4669 return (error);
4670}
4671#endif /* sun */
4672
4673/*ARGSUSED*/
4674void
4675zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4676{
4677 znode_t *zp = VTOZ(vp);
4678 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4679 int error;
4680
4681 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4682 if (zp->z_sa_hdl == NULL) {
4683 /*
4684 * The fs has been unmounted, or we did a
4685 * suspend/resume and this file no longer exists.
4686 */
4687 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4688 vrecycle(vp);
4689 return;
4690 }
4691
4692 mutex_enter(&zp->z_lock);
4693 if (zp->z_unlinked) {
4694 /*
4695 * Fast path to recycle a vnode of a removed file.
4696 */
4697 mutex_exit(&zp->z_lock);
4698 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4699 vrecycle(vp);
4700 return;
4701 }
4702 mutex_exit(&zp->z_lock);
4703
4704 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4705 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4706
4707 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4708 zfs_sa_upgrade_txholds(tx, zp);
4709 error = dmu_tx_assign(tx, TXG_WAIT);
4710 if (error) {
4711 dmu_tx_abort(tx);
4712 } else {
4713 mutex_enter(&zp->z_lock);
4714 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4715 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4716 zp->z_atime_dirty = 0;
4717 mutex_exit(&zp->z_lock);
4718 dmu_tx_commit(tx);
4719 }
4720 }
4721 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4722}
4723
4724#ifdef sun
4725/*
4726 * Bounds-check the seek operation.
4727 *
4728 * IN: vp - vnode seeking within
4729 * ooff - old file offset
4730 * noffp - pointer to new file offset
4731 * ct - caller context
4732 *
4733 * RETURN: 0 on success, EINVAL if new offset invalid.
4734 */
4735/* ARGSUSED */
4736static int
4737zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4738 caller_context_t *ct)
4739{
4740 if (vp->v_type == VDIR)
4741 return (0);
4742 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4743}
4744
4745/*
4746 * Pre-filter the generic locking function to trap attempts to place
4747 * a mandatory lock on a memory mapped file.
4748 */
4749static int
4750zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4751 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4752{
4753 znode_t *zp = VTOZ(vp);
4754 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4755
4756 ZFS_ENTER(zfsvfs);
4757 ZFS_VERIFY_ZP(zp);
4758
4759 /*
4760 * We are following the UFS semantics with respect to mapcnt
4761 * here: If we see that the file is mapped already, then we will
4762 * return an error, but we don't worry about races between this
4763 * function and zfs_map().
4764 */
4765 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4766 ZFS_EXIT(zfsvfs);
4767 return (SET_ERROR(EAGAIN));
4768 }
4769 ZFS_EXIT(zfsvfs);
4770 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4771}
4772
4773/*
4774 * If we can't find a page in the cache, we will create a new page
4775 * and fill it with file data. For efficiency, we may try to fill
4776 * multiple pages at once (klustering) to fill up the supplied page
4777 * list. Note that the pages to be filled are held with an exclusive
4778 * lock to prevent access by other threads while they are being filled.
4779 */
4780static int
4781zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4782 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4783{
4784 znode_t *zp = VTOZ(vp);
4785 page_t *pp, *cur_pp;
4786 objset_t *os = zp->z_zfsvfs->z_os;
4787 u_offset_t io_off, total;
4788 size_t io_len;
4789 int err;
4790
4791 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4792 /*
4793 * We only have a single page, don't bother klustering
4794 */
4795 io_off = off;
4796 io_len = PAGESIZE;
4797 pp = page_create_va(vp, io_off, io_len,
4798 PG_EXCL | PG_WAIT, seg, addr);
4799 } else {
4800 /*
4801 * Try to find enough pages to fill the page list
4802 */
4803 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4804 &io_len, off, plsz, 0);
4805 }
4806 if (pp == NULL) {
4807 /*
4808 * The page already exists, nothing to do here.
4809 */
4810 *pl = NULL;
4811 return (0);
4812 }
4813
4814 /*
4815 * Fill the pages in the kluster.
4816 */
4817 cur_pp = pp;
4818 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4819 caddr_t va;
4820
4821 ASSERT3U(io_off, ==, cur_pp->p_offset);
4822 va = zfs_map_page(cur_pp, S_WRITE);
4823 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4824 DMU_READ_PREFETCH);
4825 zfs_unmap_page(cur_pp, va);
4826 if (err) {
4827 /* On error, toss the entire kluster */
4828 pvn_read_done(pp, B_ERROR);
4829 /* convert checksum errors into IO errors */
4830 if (err == ECKSUM)
4831 err = SET_ERROR(EIO);
4832 return (err);
4833 }
4834 cur_pp = cur_pp->p_next;
4835 }
4836
4837 /*
4838 * Fill in the page list array from the kluster starting
4839 * from the desired offset `off'.
4840 * NOTE: the page list will always be null terminated.
4841 */
4842 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4843 ASSERT(pl == NULL || (*pl)->p_offset == off);
4844
4845 return (0);
4846}
4847
4848/*
4849 * Return pointers to the pages for the file region [off, off + len]
4850 * in the pl array. If plsz is greater than len, this function may
4851 * also return page pointers from after the specified region
4852 * (i.e. the region [off, off + plsz]). These additional pages are
4853 * only returned if they are already in the cache, or were created as
4854 * part of a klustered read.
4855 *
4856 * IN: vp - vnode of file to get data from.
4857 * off - position in file to get data from.
4858 * len - amount of data to retrieve.
4859 * plsz - length of provided page list.
4860 * seg - segment to obtain pages for.
4861 * addr - virtual address of fault.
4862 * rw - mode of created pages.
4863 * cr - credentials of caller.
4864 * ct - caller context.
4865 *
4866 * OUT: protp - protection mode of created pages.
4867 * pl - list of pages created.
4868 *
4869 * RETURN: 0 on success, error code on failure.
4870 *
4871 * Timestamps:
4872 * vp - atime updated
4873 */
4874/* ARGSUSED */
4875static int
4876zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4877 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4878 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4879{
4880 znode_t *zp = VTOZ(vp);
4881 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4882 page_t **pl0 = pl;
4883 int err = 0;
4884
4885 /* we do our own caching, faultahead is unnecessary */
4886 if (pl == NULL)
4887 return (0);
4888 else if (len > plsz)
4889 len = plsz;
4890 else
4891 len = P2ROUNDUP(len, PAGESIZE);
4892 ASSERT(plsz >= len);
4893
4894 ZFS_ENTER(zfsvfs);
4895 ZFS_VERIFY_ZP(zp);
4896
4897 if (protp)
4898 *protp = PROT_ALL;
4899
4900 /*
4901 * Loop through the requested range [off, off + len) looking
4902 * for pages. If we don't find a page, we will need to create
4903 * a new page and fill it with data from the file.
4904 */
4905 while (len > 0) {
4906 if (*pl = page_lookup(vp, off, SE_SHARED))
4907 *(pl+1) = NULL;
4908 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4909 goto out;
4910 while (*pl) {
4911 ASSERT3U((*pl)->p_offset, ==, off);
4912 off += PAGESIZE;
4913 addr += PAGESIZE;
4914 if (len > 0) {
4915 ASSERT3U(len, >=, PAGESIZE);
4916 len -= PAGESIZE;
4917 }
4918 ASSERT3U(plsz, >=, PAGESIZE);
4919 plsz -= PAGESIZE;
4920 pl++;
4921 }
4922 }
4923
4924 /*
4925 * Fill out the page array with any pages already in the cache.
4926 */
4927 while (plsz > 0 &&
4928 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4929 off += PAGESIZE;
4930 plsz -= PAGESIZE;
4931 }
4932out:
4933 if (err) {
4934 /*
4935 * Release any pages we have previously locked.
4936 */
4937 while (pl > pl0)
4938 page_unlock(*--pl);
4939 } else {
4940 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4941 }
4942
4943 *pl = NULL;
4944
4945 ZFS_EXIT(zfsvfs);
4946 return (err);
4947}
4948
4949/*
4950 * Request a memory map for a section of a file. This code interacts
4951 * with common code and the VM system as follows:
4952 *
4953 * - common code calls mmap(), which ends up in smmap_common()
4954 * - this calls VOP_MAP(), which takes you into (say) zfs
4955 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4956 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4957 * - zfs_addmap() updates z_mapcnt
4958 */
4959/*ARGSUSED*/
4960static int
4961zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4962 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4963 caller_context_t *ct)
4964{
4965 znode_t *zp = VTOZ(vp);
4966 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4967 segvn_crargs_t vn_a;
4968 int error;
4969
4970 ZFS_ENTER(zfsvfs);
4971 ZFS_VERIFY_ZP(zp);
4972
4973 if ((prot & PROT_WRITE) && (zp->z_pflags &
4974 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4975 ZFS_EXIT(zfsvfs);
4976 return (SET_ERROR(EPERM));
4977 }
4978
4979 if ((prot & (PROT_READ | PROT_EXEC)) &&
4980 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4981 ZFS_EXIT(zfsvfs);
4982 return (SET_ERROR(EACCES));
4983 }
4984
4985 if (vp->v_flag & VNOMAP) {
4986 ZFS_EXIT(zfsvfs);
4987 return (SET_ERROR(ENOSYS));
4988 }
4989
4990 if (off < 0 || len > MAXOFFSET_T - off) {
4991 ZFS_EXIT(zfsvfs);
4992 return (SET_ERROR(ENXIO));
4993 }
4994
4995 if (vp->v_type != VREG) {
4996 ZFS_EXIT(zfsvfs);
4997 return (SET_ERROR(ENODEV));
4998 }
4999
5000 /*
5001 * If file is locked, disallow mapping.
5002 */
5003 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
5004 ZFS_EXIT(zfsvfs);
5005 return (SET_ERROR(EAGAIN));
5006 }
5007
5008 as_rangelock(as);
5009 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5010 if (error != 0) {
5011 as_rangeunlock(as);
5012 ZFS_EXIT(zfsvfs);
5013 return (error);
5014 }
5015
5016 vn_a.vp = vp;
5017 vn_a.offset = (u_offset_t)off;
5018 vn_a.type = flags & MAP_TYPE;
5019 vn_a.prot = prot;
5020 vn_a.maxprot = maxprot;
5021 vn_a.cred = cr;
5022 vn_a.amp = NULL;
5023 vn_a.flags = flags & ~MAP_TYPE;
5024 vn_a.szc = 0;
5025 vn_a.lgrp_mem_policy_flags = 0;
5026
5027 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5028
5029 as_rangeunlock(as);
5030 ZFS_EXIT(zfsvfs);
5031 return (error);
5032}
5033
5034/* ARGSUSED */
5035static int
5036zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5037 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5038 caller_context_t *ct)
5039{
5040 uint64_t pages = btopr(len);
5041
5042 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5043 return (0);
5044}
5045
5046/*
5047 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5048 * more accurate mtime for the associated file. Since we don't have a way of
5049 * detecting when the data was actually modified, we have to resort to
5050 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5051 * last page is pushed. The problem occurs when the msync() call is omitted,
5052 * which by far the most common case:
5053 *
5054 * open()
5055 * mmap()
5056 * <modify memory>
5057 * munmap()
5058 * close()
5059 * <time lapse>
5060 * putpage() via fsflush
5061 *
5062 * If we wait until fsflush to come along, we can have a modification time that
5063 * is some arbitrary point in the future. In order to prevent this in the
5064 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5065 * torn down.
5066 */
5067/* ARGSUSED */
5068static int
5069zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5070 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5071 caller_context_t *ct)
5072{
5073 uint64_t pages = btopr(len);
5074
5075 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5076 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5077
5078 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5079 vn_has_cached_data(vp))
5080 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5081
5082 return (0);
5083}
5084
5085/*
5086 * Free or allocate space in a file. Currently, this function only
5087 * supports the `F_FREESP' command. However, this command is somewhat
5088 * misnamed, as its functionality includes the ability to allocate as
5089 * well as free space.
5090 *
5091 * IN: vp - vnode of file to free data in.
5092 * cmd - action to take (only F_FREESP supported).
5093 * bfp - section of file to free/alloc.
5094 * flag - current file open mode flags.
5095 * offset - current file offset.
5096 * cr - credentials of caller [UNUSED].
5097 * ct - caller context.
5098 *
5099 * RETURN: 0 on success, error code on failure.
5100 *
5101 * Timestamps:
5102 * vp - ctime|mtime updated
5103 */
5104/* ARGSUSED */
5105static int
5106zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5107 offset_t offset, cred_t *cr, caller_context_t *ct)
5108{
5109 znode_t *zp = VTOZ(vp);
5110 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5111 uint64_t off, len;
5112 int error;
5113
5114 ZFS_ENTER(zfsvfs);
5115 ZFS_VERIFY_ZP(zp);
5116
5117 if (cmd != F_FREESP) {
5118 ZFS_EXIT(zfsvfs);
5119 return (SET_ERROR(EINVAL));
5120 }
5121
5122 if (error = convoff(vp, bfp, 0, offset)) {
5123 ZFS_EXIT(zfsvfs);
5124 return (error);
5125 }
5126
5127 if (bfp->l_len < 0) {
5128 ZFS_EXIT(zfsvfs);
5129 return (SET_ERROR(EINVAL));
5130 }
5131
5132 off = bfp->l_start;
5133 len = bfp->l_len; /* 0 means from off to end of file */
5134
5135 error = zfs_freesp(zp, off, len, flag, TRUE);
5136
5137 ZFS_EXIT(zfsvfs);
5138 return (error);
5139}
5140#endif /* sun */
5141
5142CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5143CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5144
5145/*ARGSUSED*/
5146static int
5147zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5148{
5149 znode_t *zp = VTOZ(vp);
5150 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5151 uint32_t gen;
5152 uint64_t gen64;
5153 uint64_t object = zp->z_id;
5154 zfid_short_t *zfid;
5155 int size, i, error;
5156
5157 ZFS_ENTER(zfsvfs);
5158 ZFS_VERIFY_ZP(zp);
5159
5160 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5161 &gen64, sizeof (uint64_t))) != 0) {
5162 ZFS_EXIT(zfsvfs);
5163 return (error);
5164 }
5165
5166 gen = (uint32_t)gen64;
5167
5168 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5169
5170#ifdef illumos
5171 if (fidp->fid_len < size) {
5172 fidp->fid_len = size;
5173 ZFS_EXIT(zfsvfs);
5174 return (SET_ERROR(ENOSPC));
5175 }
5176#else
5177 fidp->fid_len = size;
5178#endif
5179
5180 zfid = (zfid_short_t *)fidp;
5181
5182 zfid->zf_len = size;
5183
5184 for (i = 0; i < sizeof (zfid->zf_object); i++)
5185 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5186
5187 /* Must have a non-zero generation number to distinguish from .zfs */
5188 if (gen == 0)
5189 gen = 1;
5190 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5191 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5192
5193 if (size == LONG_FID_LEN) {
5194 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5195 zfid_long_t *zlfid;
5196
5197 zlfid = (zfid_long_t *)fidp;
5198
5199 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5200 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5201
5202 /* XXX - this should be the generation number for the objset */
5203 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5204 zlfid->zf_setgen[i] = 0;
5205 }
5206
5207 ZFS_EXIT(zfsvfs);
5208 return (0);
5209}
5210
5211static int
5212zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5213 caller_context_t *ct)
5214{
5215 znode_t *zp, *xzp;
5216 zfsvfs_t *zfsvfs;
5217 zfs_dirlock_t *dl;
5218 int error;
5219
5220 switch (cmd) {
5221 case _PC_LINK_MAX:
5222 *valp = INT_MAX;
5223 return (0);
5224
5225 case _PC_FILESIZEBITS:
5226 *valp = 64;
5227 return (0);
5228#ifdef sun
5229 case _PC_XATTR_EXISTS:
5230 zp = VTOZ(vp);
5231 zfsvfs = zp->z_zfsvfs;
5232 ZFS_ENTER(zfsvfs);
5233 ZFS_VERIFY_ZP(zp);
5234 *valp = 0;
5235 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5236 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5237 if (error == 0) {
5238 zfs_dirent_unlock(dl);
5239 if (!zfs_dirempty(xzp))
5240 *valp = 1;
5241 VN_RELE(ZTOV(xzp));
5242 } else if (error == ENOENT) {
5243 /*
5244 * If there aren't extended attributes, it's the
5245 * same as having zero of them.
5246 */
5247 error = 0;
5248 }
5249 ZFS_EXIT(zfsvfs);
5250 return (error);
5251
5252 case _PC_SATTR_ENABLED:
5253 case _PC_SATTR_EXISTS:
5254 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5255 (vp->v_type == VREG || vp->v_type == VDIR);
5256 return (0);
5257
5258 case _PC_ACCESS_FILTERING:
5259 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5260 vp->v_type == VDIR;
5261 return (0);
5262
5263 case _PC_ACL_ENABLED:
5264 *valp = _ACL_ACE_ENABLED;
5265 return (0);
5266#endif /* sun */
5267 case _PC_MIN_HOLE_SIZE:
5268 *valp = (int)SPA_MINBLOCKSIZE;
5269 return (0);
5270#ifdef sun
5271 case _PC_TIMESTAMP_RESOLUTION:
5272 /* nanosecond timestamp resolution */
5273 *valp = 1L;
5274 return (0);
5275#endif /* sun */
5276 case _PC_ACL_EXTENDED:
5277 *valp = 0;
5278 return (0);
5279
5280 case _PC_ACL_NFS4:
5281 *valp = 1;
5282 return (0);
5283
5284 case _PC_ACL_PATH_MAX:
5285 *valp = ACL_MAX_ENTRIES;
5286 return (0);
5287
5288 default:
5289 return (EOPNOTSUPP);
5290 }
5291}
5292
5293/*ARGSUSED*/
5294static int
5295zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5296 caller_context_t *ct)
5297{
5298 znode_t *zp = VTOZ(vp);
5299 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5300 int error;
5301 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5302
5303 ZFS_ENTER(zfsvfs);
5304 ZFS_VERIFY_ZP(zp);
5305 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5306 ZFS_EXIT(zfsvfs);
5307
5308 return (error);
5309}
5310
5311/*ARGSUSED*/
5312int
5313zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5314 caller_context_t *ct)
5315{
5316 znode_t *zp = VTOZ(vp);
5317 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5318 int error;
5319 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5320 zilog_t *zilog = zfsvfs->z_log;
5321
5322 ZFS_ENTER(zfsvfs);
5323 ZFS_VERIFY_ZP(zp);
5324
5325 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5326
5327 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5328 zil_commit(zilog, 0);
5329
5330 ZFS_EXIT(zfsvfs);
5331 return (error);
5332}
5333
5334#ifdef sun
5335/*
5336 * The smallest read we may consider to loan out an arcbuf.
5337 * This must be a power of 2.
5338 */
5339int zcr_blksz_min = (1 << 10); /* 1K */
5340/*
5341 * If set to less than the file block size, allow loaning out of an
5342 * arcbuf for a partial block read. This must be a power of 2.
5343 */
5344int zcr_blksz_max = (1 << 17); /* 128K */
5345
5346/*ARGSUSED*/
5347static int
5348zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5349 caller_context_t *ct)
5350{
5351 znode_t *zp = VTOZ(vp);
5352 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5353 int max_blksz = zfsvfs->z_max_blksz;
5354 uio_t *uio = &xuio->xu_uio;
5355 ssize_t size = uio->uio_resid;
5356 offset_t offset = uio->uio_loffset;
5357 int blksz;
5358 int fullblk, i;
5359 arc_buf_t *abuf;
5360 ssize_t maxsize;
5361 int preamble, postamble;
5362
5363 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5364 return (SET_ERROR(EINVAL));
5365
5366 ZFS_ENTER(zfsvfs);
5367 ZFS_VERIFY_ZP(zp);
5368 switch (ioflag) {
5369 case UIO_WRITE:
5370 /*
5371 * Loan out an arc_buf for write if write size is bigger than
5372 * max_blksz, and the file's block size is also max_blksz.
5373 */
5374 blksz = max_blksz;
5375 if (size < blksz || zp->z_blksz != blksz) {
5376 ZFS_EXIT(zfsvfs);
5377 return (SET_ERROR(EINVAL));
5378 }
5379 /*
5380 * Caller requests buffers for write before knowing where the
5381 * write offset might be (e.g. NFS TCP write).
5382 */
5383 if (offset == -1) {
5384 preamble = 0;
5385 } else {
5386 preamble = P2PHASE(offset, blksz);
5387 if (preamble) {
5388 preamble = blksz - preamble;
5389 size -= preamble;
5390 }
5391 }
5392
5393 postamble = P2PHASE(size, blksz);
5394 size -= postamble;
5395
5396 fullblk = size / blksz;
5397 (void) dmu_xuio_init(xuio,
5398 (preamble != 0) + fullblk + (postamble != 0));
5399 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5400 int, postamble, int,
5401 (preamble != 0) + fullblk + (postamble != 0));
5402
5403 /*
5404 * Have to fix iov base/len for partial buffers. They
5405 * currently represent full arc_buf's.
5406 */
5407 if (preamble) {
5408 /* data begins in the middle of the arc_buf */
5409 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5410 blksz);
5411 ASSERT(abuf);
5412 (void) dmu_xuio_add(xuio, abuf,
5413 blksz - preamble, preamble);
5414 }
5415
5416 for (i = 0; i < fullblk; i++) {
5417 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5418 blksz);
5419 ASSERT(abuf);
5420 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5421 }
5422
5423 if (postamble) {
5424 /* data ends in the middle of the arc_buf */
5425 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5426 blksz);
5427 ASSERT(abuf);
5428 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5429 }
5430 break;
5431 case UIO_READ:
5432 /*
5433 * Loan out an arc_buf for read if the read size is larger than
5434 * the current file block size. Block alignment is not
5435 * considered. Partial arc_buf will be loaned out for read.
5436 */
5437 blksz = zp->z_blksz;
5438 if (blksz < zcr_blksz_min)
5439 blksz = zcr_blksz_min;
5440 if (blksz > zcr_blksz_max)
5441 blksz = zcr_blksz_max;
5442 /* avoid potential complexity of dealing with it */
5443 if (blksz > max_blksz) {
5444 ZFS_EXIT(zfsvfs);
5445 return (SET_ERROR(EINVAL));
5446 }
5447
5448 maxsize = zp->z_size - uio->uio_loffset;
5449 if (size > maxsize)
5450 size = maxsize;
5451
5452 if (size < blksz || vn_has_cached_data(vp)) {
5453 ZFS_EXIT(zfsvfs);
5454 return (SET_ERROR(EINVAL));
5455 }
5456 break;
5457 default:
5458 ZFS_EXIT(zfsvfs);
5459 return (SET_ERROR(EINVAL));
5460 }
5461
5462 uio->uio_extflg = UIO_XUIO;
5463 XUIO_XUZC_RW(xuio) = ioflag;
5464 ZFS_EXIT(zfsvfs);
5465 return (0);
5466}
5467
5468/*ARGSUSED*/
5469static int
5470zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5471{
5472 int i;
5473 arc_buf_t *abuf;
5474 int ioflag = XUIO_XUZC_RW(xuio);
5475
5476 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5477
5478 i = dmu_xuio_cnt(xuio);
5479 while (i-- > 0) {
5480 abuf = dmu_xuio_arcbuf(xuio, i);
5481 /*
5482 * if abuf == NULL, it must be a write buffer
5483 * that has been returned in zfs_write().
5484 */
5485 if (abuf)
5486 dmu_return_arcbuf(abuf);
5487 ASSERT(abuf || ioflag == UIO_WRITE);
5488 }
5489
5490 dmu_xuio_fini(xuio);
5491 return (0);
5492}
5493
5494/*
5495 * Predeclare these here so that the compiler assumes that
5496 * this is an "old style" function declaration that does
5497 * not include arguments => we won't get type mismatch errors
5498 * in the initializations that follow.
5499 */
5500static int zfs_inval();
5501static int zfs_isdir();
5502
5503static int
5504zfs_inval()
5505{
5506 return (SET_ERROR(EINVAL));
5507}
5508
5509static int
5510zfs_isdir()
5511{
5512 return (SET_ERROR(EISDIR));
5513}
5514/*
5515 * Directory vnode operations template
5516 */
5517vnodeops_t *zfs_dvnodeops;
5518const fs_operation_def_t zfs_dvnodeops_template[] = {
5519 VOPNAME_OPEN, { .vop_open = zfs_open },
5520 VOPNAME_CLOSE, { .vop_close = zfs_close },
5521 VOPNAME_READ, { .error = zfs_isdir },
5522 VOPNAME_WRITE, { .error = zfs_isdir },
5523 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5524 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5525 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5526 VOPNAME_ACCESS, { .vop_access = zfs_access },
5527 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5528 VOPNAME_CREATE, { .vop_create = zfs_create },
5529 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5530 VOPNAME_LINK, { .vop_link = zfs_link },
5531 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5532 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5533 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5534 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5535 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5536 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5537 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5538 VOPNAME_FID, { .vop_fid = zfs_fid },
5539 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5540 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5541 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5542 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5543 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5544 NULL, NULL
5545};
5546
5547/*
5548 * Regular file vnode operations template
5549 */
5550vnodeops_t *zfs_fvnodeops;
5551const fs_operation_def_t zfs_fvnodeops_template[] = {
5552 VOPNAME_OPEN, { .vop_open = zfs_open },
5553 VOPNAME_CLOSE, { .vop_close = zfs_close },
5554 VOPNAME_READ, { .vop_read = zfs_read },
5555 VOPNAME_WRITE, { .vop_write = zfs_write },
5556 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5557 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5558 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5559 VOPNAME_ACCESS, { .vop_access = zfs_access },
5560 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5561 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5562 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5563 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5564 VOPNAME_FID, { .vop_fid = zfs_fid },
5565 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5566 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5567 VOPNAME_SPACE, { .vop_space = zfs_space },
5568 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5569 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5570 VOPNAME_MAP, { .vop_map = zfs_map },
5571 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5572 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5573 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5574 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5575 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5576 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5577 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5578 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5579 NULL, NULL
5580};
5581
5582/*
5583 * Symbolic link vnode operations template
5584 */
5585vnodeops_t *zfs_symvnodeops;
5586const fs_operation_def_t zfs_symvnodeops_template[] = {
5587 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5588 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5589 VOPNAME_ACCESS, { .vop_access = zfs_access },
5590 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5591 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5592 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5593 VOPNAME_FID, { .vop_fid = zfs_fid },
5594 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5595 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5596 NULL, NULL
5597};
5598
5599/*
5600 * special share hidden files vnode operations template
5601 */
5602vnodeops_t *zfs_sharevnodeops;
5603const fs_operation_def_t zfs_sharevnodeops_template[] = {
5604 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5605 VOPNAME_ACCESS, { .vop_access = zfs_access },
5606 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5607 VOPNAME_FID, { .vop_fid = zfs_fid },
5608 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5609 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5610 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5611 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5612 NULL, NULL
5613};
5614
5615/*
5616 * Extended attribute directory vnode operations template
5617 *
5618 * This template is identical to the directory vnodes
5619 * operation template except for restricted operations:
5620 * VOP_MKDIR()
5621 * VOP_SYMLINK()
5622 *
5623 * Note that there are other restrictions embedded in:
5624 * zfs_create() - restrict type to VREG
5625 * zfs_link() - no links into/out of attribute space
5626 * zfs_rename() - no moves into/out of attribute space
5627 */
5628vnodeops_t *zfs_xdvnodeops;
5629const fs_operation_def_t zfs_xdvnodeops_template[] = {
5630 VOPNAME_OPEN, { .vop_open = zfs_open },
5631 VOPNAME_CLOSE, { .vop_close = zfs_close },
5632 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5633 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5634 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5635 VOPNAME_ACCESS, { .vop_access = zfs_access },
5636 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5637 VOPNAME_CREATE, { .vop_create = zfs_create },
5638 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5639 VOPNAME_LINK, { .vop_link = zfs_link },
5640 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5641 VOPNAME_MKDIR, { .error = zfs_inval },
5642 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5643 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5644 VOPNAME_SYMLINK, { .error = zfs_inval },
5645 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5646 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5647 VOPNAME_FID, { .vop_fid = zfs_fid },
5648 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5649 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5650 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5651 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5652 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5653 NULL, NULL
5654};
5655
5656/*
5657 * Error vnode operations template
5658 */
5659vnodeops_t *zfs_evnodeops;
5660const fs_operation_def_t zfs_evnodeops_template[] = {
5661 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5662 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5663 NULL, NULL
5664};
5665#endif /* sun */
5666
5667static int
5668ioflags(int ioflags)
5669{
5670 int flags = 0;
5671
5672 if (ioflags & IO_APPEND)
5673 flags |= FAPPEND;
5674 if (ioflags & IO_NDELAY)
5675 flags |= FNONBLOCK;
5676 if (ioflags & IO_SYNC)
5677 flags |= (FSYNC | FDSYNC | FRSYNC);
5678
5679 return (flags);
5680}
5681
5682static int
5683zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5684{
5685 znode_t *zp = VTOZ(vp);
5686 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5687 objset_t *os = zp->z_zfsvfs->z_os;
5688 vm_page_t mfirst, mlast, mreq;
5689 vm_object_t object;
5690 caddr_t va;
5691 struct sf_buf *sf;
5692 off_t startoff, endoff;
5693 int i, error;
5694 vm_pindex_t reqstart, reqend;
5695 int pcount, lsize, reqsize, size;
5696
5697 ZFS_ENTER(zfsvfs);
5698 ZFS_VERIFY_ZP(zp);
5699
5700 pcount = OFF_TO_IDX(round_page(count));
5701 mreq = m[reqpage];
5702 object = mreq->object;
5703 error = 0;
5704
5705 KASSERT(vp->v_object == object, ("mismatching object"));
5706
5707 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5708 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5709 reqstart = OFF_TO_IDX(round_page(startoff));
5710 if (reqstart < m[0]->pindex)
5711 reqstart = 0;
5712 else
5713 reqstart = reqstart - m[0]->pindex;
5714 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5715 zp->z_blksz);
5716 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5717 if (reqend > m[pcount - 1]->pindex)
5718 reqend = m[pcount - 1]->pindex;
5719 reqsize = reqend - m[reqstart]->pindex + 1;
5720 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5721 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5722 } else {
5723 reqstart = reqpage;
5724 reqsize = 1;
5725 }
5726 mfirst = m[reqstart];
5727 mlast = m[reqstart + reqsize - 1];
5728
5729 zfs_vmobject_wlock(object);
5730
5731 for (i = 0; i < reqstart; i++) {
5732 vm_page_lock(m[i]);
5733 vm_page_free(m[i]);
5734 vm_page_unlock(m[i]);
5735 }
5736 for (i = reqstart + reqsize; i < pcount; i++) {
5737 vm_page_lock(m[i]);
5738 vm_page_free(m[i]);
5739 vm_page_unlock(m[i]);
5740 }
5741
5742 if (mreq->valid && reqsize == 1) {
5743 if (mreq->valid != VM_PAGE_BITS_ALL)
5744 vm_page_zero_invalid(mreq, TRUE);
5745 zfs_vmobject_wunlock(object);
5746 ZFS_EXIT(zfsvfs);
5747 return (zfs_vm_pagerret_ok);
5748 }
5749
5750 PCPU_INC(cnt.v_vnodein);
5751 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5752
5753 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5754 for (i = reqstart; i < reqstart + reqsize; i++) {
5755 if (i != reqpage) {
5756 vm_page_lock(m[i]);
5757 vm_page_free(m[i]);
5758 vm_page_unlock(m[i]);
5759 }
5760 }
5761 zfs_vmobject_wunlock(object);
5762 ZFS_EXIT(zfsvfs);
5763 return (zfs_vm_pagerret_bad);
5764 }
5765
5766 lsize = PAGE_SIZE;
5767 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5768 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5769
5770 zfs_vmobject_wunlock(object);
5771
5772 for (i = reqstart; i < reqstart + reqsize; i++) {
5773 size = PAGE_SIZE;
5774 if (i == (reqstart + reqsize - 1))
5775 size = lsize;
5776 va = zfs_map_page(m[i], &sf);
5777 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5778 size, va, DMU_READ_PREFETCH);
5779 if (size != PAGE_SIZE)
5780 bzero(va + size, PAGE_SIZE - size);
5781 zfs_unmap_page(sf);
5782 if (error != 0)
5783 break;
5784 }
5785
5786 zfs_vmobject_wlock(object);
5787
5788 for (i = reqstart; i < reqstart + reqsize; i++) {
5789 if (!error)
5790 m[i]->valid = VM_PAGE_BITS_ALL;
5791 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5792 if (i != reqpage)
5793 vm_page_readahead_finish(m[i]);
5794 }
5795
5796 zfs_vmobject_wunlock(object);
5797
5798 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5799 ZFS_EXIT(zfsvfs);
5800 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok);
5801}
5802
5803static int
5804zfs_freebsd_getpages(ap)
5805 struct vop_getpages_args /* {
5806 struct vnode *a_vp;
5807 vm_page_t *a_m;
5808 int a_count;
5809 int a_reqpage;
5810 vm_ooffset_t a_offset;
5811 } */ *ap;
5812{
5813
5814 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5815}
5816
5817static int
5818zfs_freebsd_bmap(ap)
5819 struct vop_bmap_args /* {
5820 struct vnode *a_vp;
5821 daddr_t a_bn;
5822 struct bufobj **a_bop;
5823 daddr_t *a_bnp;
5824 int *a_runp;
5825 int *a_runb;
5826 } */ *ap;
5827{
5828
5829 if (ap->a_bop != NULL)
5830 *ap->a_bop = &ap->a_vp->v_bufobj;
5831 if (ap->a_bnp != NULL)
5832 *ap->a_bnp = ap->a_bn;
5833 if (ap->a_runp != NULL)
5834 *ap->a_runp = 0;
5835 if (ap->a_runb != NULL)
5836 *ap->a_runb = 0;
5837
5838 return (0);
5839}
5840
5841static int
5842zfs_freebsd_open(ap)
5843 struct vop_open_args /* {
5844 struct vnode *a_vp;
5845 int a_mode;
5846 struct ucred *a_cred;
5847 struct thread *a_td;
5848 } */ *ap;
5849{
5850 vnode_t *vp = ap->a_vp;
5851 znode_t *zp = VTOZ(vp);
5852 int error;
5853
5854 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5855 if (error == 0)
5856 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5857 return (error);
5858}
5859
5860static int
5861zfs_freebsd_close(ap)
5862 struct vop_close_args /* {
5863 struct vnode *a_vp;
5864 int a_fflag;
5865 struct ucred *a_cred;
5866 struct thread *a_td;
5867 } */ *ap;
5868{
5869
5870 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5871}
5872
5873static int
5874zfs_freebsd_ioctl(ap)
5875 struct vop_ioctl_args /* {
5876 struct vnode *a_vp;
5877 u_long a_command;
5878 caddr_t a_data;
5879 int a_fflag;
5880 struct ucred *cred;
5881 struct thread *td;
5882 } */ *ap;
5883{
5884
5885 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5886 ap->a_fflag, ap->a_cred, NULL, NULL));
5887}
5888
5889static int
5890zfs_freebsd_read(ap)
5891 struct vop_read_args /* {
5892 struct vnode *a_vp;
5893 struct uio *a_uio;
5894 int a_ioflag;
5895 struct ucred *a_cred;
5896 } */ *ap;
5897{
5898
5899 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5900 ap->a_cred, NULL));
5901}
5902
5903static int
5904zfs_freebsd_write(ap)
5905 struct vop_write_args /* {
5906 struct vnode *a_vp;
5907 struct uio *a_uio;
5908 int a_ioflag;
5909 struct ucred *a_cred;
5910 } */ *ap;
5911{
5912
5913 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5914 ap->a_cred, NULL));
5915}
5916
5917static int
5918zfs_freebsd_access(ap)
5919 struct vop_access_args /* {
5920 struct vnode *a_vp;
5921 accmode_t a_accmode;
5922 struct ucred *a_cred;
5923 struct thread *a_td;
5924 } */ *ap;
5925{
5926 vnode_t *vp = ap->a_vp;
5927 znode_t *zp = VTOZ(vp);
5928 accmode_t accmode;
5929 int error = 0;
5930
5931 /*
5932 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5933 */
5934 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5935 if (accmode != 0)
5936 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5937
5938 /*
5939 * VADMIN has to be handled by vaccess().
5940 */
5941 if (error == 0) {
5942 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5943 if (accmode != 0) {
5944 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5945 zp->z_gid, accmode, ap->a_cred, NULL);
5946 }
5947 }
5948
5949 /*
5950 * For VEXEC, ensure that at least one execute bit is set for
5951 * non-directories.
5952 */
5953 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
5954 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
5955 error = EACCES;
5956 }
5957
5958 return (error);
5959}
5960
5961static int
5962zfs_freebsd_lookup(ap)
5963 struct vop_lookup_args /* {
5964 struct vnode *a_dvp;
5965 struct vnode **a_vpp;
5966 struct componentname *a_cnp;
5967 } */ *ap;
5968{
5969 struct componentname *cnp = ap->a_cnp;
5970 char nm[NAME_MAX + 1];
5971
5972 ASSERT(cnp->cn_namelen < sizeof(nm));
5973 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5974
5975 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5976 cnp->cn_cred, cnp->cn_thread, 0));
5977}
5978
5979static int
5980zfs_freebsd_create(ap)
5981 struct vop_create_args /* {
5982 struct vnode *a_dvp;
5983 struct vnode **a_vpp;
5984 struct componentname *a_cnp;
5985 struct vattr *a_vap;
5986 } */ *ap;
5987{
5988 struct componentname *cnp = ap->a_cnp;
5989 vattr_t *vap = ap->a_vap;
5990 int mode;
5991
5992 ASSERT(cnp->cn_flags & SAVENAME);
5993
5994 vattr_init_mask(vap);
5995 mode = vap->va_mode & ALLPERMS;
5996
5997 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
5998 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
5999}
6000
6001static int
6002zfs_freebsd_remove(ap)
6003 struct vop_remove_args /* {
6004 struct vnode *a_dvp;
6005 struct vnode *a_vp;
6006 struct componentname *a_cnp;
6007 } */ *ap;
6008{
6009
6010 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6011
6012 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6013 ap->a_cnp->cn_cred, NULL, 0));
6014}
6015
6016static int
6017zfs_freebsd_mkdir(ap)
6018 struct vop_mkdir_args /* {
6019 struct vnode *a_dvp;
6020 struct vnode **a_vpp;
6021 struct componentname *a_cnp;
6022 struct vattr *a_vap;
6023 } */ *ap;
6024{
6025 vattr_t *vap = ap->a_vap;
6026
6027 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6028
6029 vattr_init_mask(vap);
6030
6031 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6032 ap->a_cnp->cn_cred, NULL, 0, NULL));
6033}
6034
6035static int
6036zfs_freebsd_rmdir(ap)
6037 struct vop_rmdir_args /* {
6038 struct vnode *a_dvp;
6039 struct vnode *a_vp;
6040 struct componentname *a_cnp;
6041 } */ *ap;
6042{
6043 struct componentname *cnp = ap->a_cnp;
6044
6045 ASSERT(cnp->cn_flags & SAVENAME);
6046
6047 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6048}
6049
6050static int
6051zfs_freebsd_readdir(ap)
6052 struct vop_readdir_args /* {
6053 struct vnode *a_vp;
6054 struct uio *a_uio;
6055 struct ucred *a_cred;
6056 int *a_eofflag;
6057 int *a_ncookies;
6058 u_long **a_cookies;
6059 } */ *ap;
6060{
6061
6062 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6063 ap->a_ncookies, ap->a_cookies));
6064}
6065
6066static int
6067zfs_freebsd_fsync(ap)
6068 struct vop_fsync_args /* {
6069 struct vnode *a_vp;
6070 int a_waitfor;
6071 struct thread *a_td;
6072 } */ *ap;
6073{
6074
6075 vop_stdfsync(ap);
6076 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6077}
6078
6079static int
6080zfs_freebsd_getattr(ap)
6081 struct vop_getattr_args /* {
6082 struct vnode *a_vp;
6083 struct vattr *a_vap;
6084 struct ucred *a_cred;
6085 } */ *ap;
6086{
6087 vattr_t *vap = ap->a_vap;
6088 xvattr_t xvap;
6089 u_long fflags = 0;
6090 int error;
6091
6092 xva_init(&xvap);
6093 xvap.xva_vattr = *vap;
6094 xvap.xva_vattr.va_mask |= AT_XVATTR;
6095
6096 /* Convert chflags into ZFS-type flags. */
6097 /* XXX: what about SF_SETTABLE?. */
6098 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6099 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6100 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6101 XVA_SET_REQ(&xvap, XAT_NODUMP);
6102 XVA_SET_REQ(&xvap, XAT_READONLY);
6103 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
6104 XVA_SET_REQ(&xvap, XAT_SYSTEM);
6105 XVA_SET_REQ(&xvap, XAT_HIDDEN);
6106 XVA_SET_REQ(&xvap, XAT_REPARSE);
6107 XVA_SET_REQ(&xvap, XAT_OFFLINE);
6108 XVA_SET_REQ(&xvap, XAT_SPARSE);
6109
6110 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6111 if (error != 0)
6112 return (error);
6113
6114 /* Convert ZFS xattr into chflags. */
6115#define FLAG_CHECK(fflag, xflag, xfield) do { \
6116 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6117 fflags |= (fflag); \
6118} while (0)
6119 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6120 xvap.xva_xoptattrs.xoa_immutable);
6121 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6122 xvap.xva_xoptattrs.xoa_appendonly);
6123 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6124 xvap.xva_xoptattrs.xoa_nounlink);
6125 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
6126 xvap.xva_xoptattrs.xoa_archive);
6127 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6128 xvap.xva_xoptattrs.xoa_nodump);
6129 FLAG_CHECK(UF_READONLY, XAT_READONLY,
6130 xvap.xva_xoptattrs.xoa_readonly);
6131 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
6132 xvap.xva_xoptattrs.xoa_system);
6133 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
6134 xvap.xva_xoptattrs.xoa_hidden);
6135 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
6136 xvap.xva_xoptattrs.xoa_reparse);
6137 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
6138 xvap.xva_xoptattrs.xoa_offline);
6139 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
6140 xvap.xva_xoptattrs.xoa_sparse);
6141
6142#undef FLAG_CHECK
6143 *vap = xvap.xva_vattr;
6144 vap->va_flags = fflags;
6145 return (0);
6146}
6147
6148static int
6149zfs_freebsd_setattr(ap)
6150 struct vop_setattr_args /* {
6151 struct vnode *a_vp;
6152 struct vattr *a_vap;
6153 struct ucred *a_cred;
6154 } */ *ap;
6155{
6156 vnode_t *vp = ap->a_vp;
6157 vattr_t *vap = ap->a_vap;
6158 cred_t *cred = ap->a_cred;
6159 xvattr_t xvap;
6160 u_long fflags;
6161 uint64_t zflags;
6162
6163 vattr_init_mask(vap);
6164 vap->va_mask &= ~AT_NOSET;
6165
6166 xva_init(&xvap);
6167 xvap.xva_vattr = *vap;
6168
6169 zflags = VTOZ(vp)->z_pflags;
6170
6171 if (vap->va_flags != VNOVAL) {
6172 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6173 int error;
6174
6175 if (zfsvfs->z_use_fuids == B_FALSE)
6176 return (EOPNOTSUPP);
6177
6178 fflags = vap->va_flags;
6179 /*
6180 * XXX KDM
6181 * We need to figure out whether it makes sense to allow
6182 * UF_REPARSE through, since we don't really have other
6183 * facilities to handle reparse points and zfs_setattr()
6184 * doesn't currently allow setting that attribute anyway.
6185 */
6186 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
6187 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
6188 UF_OFFLINE|UF_SPARSE)) != 0)
6189 return (EOPNOTSUPP);
6190 /*
6191 * Unprivileged processes are not permitted to unset system
6192 * flags, or modify flags if any system flags are set.
6193 * Privileged non-jail processes may not modify system flags
6194 * if securelevel > 0 and any existing system flags are set.
6195 * Privileged jail processes behave like privileged non-jail
6196 * processes if the security.jail.chflags_allowed sysctl is
6197 * is non-zero; otherwise, they behave like unprivileged
6198 * processes.
6199 */
6200 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6201 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6202 if (zflags &
6203 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6204 error = securelevel_gt(cred, 0);
6205 if (error != 0)
6206 return (error);
6207 }
6208 } else {
6209 /*
6210 * Callers may only modify the file flags on objects they
6211 * have VADMIN rights for.
6212 */
6213 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6214 return (error);
6215 if (zflags &
6216 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6217 return (EPERM);
6218 }
6219 if (fflags &
6220 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6221 return (EPERM);
6222 }
6223 }
6224
6225#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6226 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6227 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6228 XVA_SET_REQ(&xvap, (xflag)); \
6229 (xfield) = ((fflags & (fflag)) != 0); \
6230 } \
6231} while (0)
6232 /* Convert chflags into ZFS-type flags. */
6233 /* XXX: what about SF_SETTABLE?. */
6234 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6235 xvap.xva_xoptattrs.xoa_immutable);
6236 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6237 xvap.xva_xoptattrs.xoa_appendonly);
6238 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6239 xvap.xva_xoptattrs.xoa_nounlink);
6240 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
6241 xvap.xva_xoptattrs.xoa_archive);
6242 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6243 xvap.xva_xoptattrs.xoa_nodump);
6244 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
6245 xvap.xva_xoptattrs.xoa_readonly);
6246 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
6247 xvap.xva_xoptattrs.xoa_system);
6248 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
6249 xvap.xva_xoptattrs.xoa_hidden);
6250 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
6251 xvap.xva_xoptattrs.xoa_hidden);
6252 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
6253 xvap.xva_xoptattrs.xoa_offline);
6254 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
6255 xvap.xva_xoptattrs.xoa_sparse);
6256#undef FLAG_CHANGE
6257 }
6258 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6259}
6260
6261static int
6262zfs_freebsd_rename(ap)
6263 struct vop_rename_args /* {
6264 struct vnode *a_fdvp;
6265 struct vnode *a_fvp;
6266 struct componentname *a_fcnp;
6267 struct vnode *a_tdvp;
6268 struct vnode *a_tvp;
6269 struct componentname *a_tcnp;
6270 } */ *ap;
6271{
6272 vnode_t *fdvp = ap->a_fdvp;
6273 vnode_t *fvp = ap->a_fvp;
6274 vnode_t *tdvp = ap->a_tdvp;
6275 vnode_t *tvp = ap->a_tvp;
6276 int error;
6277
6278 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6279 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6280
6281 /*
6282 * Check for cross-device rename.
6283 */
6284 if ((fdvp->v_mount != tdvp->v_mount) ||
6285 (tvp && (fdvp->v_mount != tvp->v_mount)))
6286 error = EXDEV;
6287 else
6288 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6289 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6290 if (tdvp == tvp)
6291 VN_RELE(tdvp);
6292 else
6293 VN_URELE(tdvp);
6294 if (tvp)
6295 VN_URELE(tvp);
6296 VN_RELE(fdvp);
6297 VN_RELE(fvp);
6298
6299 return (error);
6300}
6301
6302static int
6303zfs_freebsd_symlink(ap)
6304 struct vop_symlink_args /* {
6305 struct vnode *a_dvp;
6306 struct vnode **a_vpp;
6307 struct componentname *a_cnp;
6308 struct vattr *a_vap;
6309 char *a_target;
6310 } */ *ap;
6311{
6312 struct componentname *cnp = ap->a_cnp;
6313 vattr_t *vap = ap->a_vap;
6314
6315 ASSERT(cnp->cn_flags & SAVENAME);
6316
6317 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6318 vattr_init_mask(vap);
6319
6320 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6321 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6322}
6323
6324static int
6325zfs_freebsd_readlink(ap)
6326 struct vop_readlink_args /* {
6327 struct vnode *a_vp;
6328 struct uio *a_uio;
6329 struct ucred *a_cred;
6330 } */ *ap;
6331{
6332
6333 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6334}
6335
6336static int
6337zfs_freebsd_link(ap)
6338 struct vop_link_args /* {
6339 struct vnode *a_tdvp;
6340 struct vnode *a_vp;
6341 struct componentname *a_cnp;
6342 } */ *ap;
6343{
6344 struct componentname *cnp = ap->a_cnp;
6345 vnode_t *vp = ap->a_vp;
6346 vnode_t *tdvp = ap->a_tdvp;
6347
6348 if (tdvp->v_mount != vp->v_mount)
6349 return (EXDEV);
6350
6351 ASSERT(cnp->cn_flags & SAVENAME);
6352
6353 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6354}
6355
6356static int
6357zfs_freebsd_inactive(ap)
6358 struct vop_inactive_args /* {
6359 struct vnode *a_vp;
6360 struct thread *a_td;
6361 } */ *ap;
6362{
6363 vnode_t *vp = ap->a_vp;
6364
6365 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6366 return (0);
6367}
6368
6369static int
6370zfs_freebsd_reclaim(ap)
6371 struct vop_reclaim_args /* {
6372 struct vnode *a_vp;
6373 struct thread *a_td;
6374 } */ *ap;
6375{
6376 vnode_t *vp = ap->a_vp;
6377 znode_t *zp = VTOZ(vp);
6378 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6379
6380 ASSERT(zp != NULL);
6381
6382 /* Destroy the vm object and flush associated pages. */
6383 vnode_destroy_vobject(vp);
6384
6385 /*
6386 * z_teardown_inactive_lock protects from a race with
6387 * zfs_znode_dmu_fini in zfsvfs_teardown during
6388 * force unmount.
6389 */
6390 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6391 if (zp->z_sa_hdl == NULL)
6392 zfs_znode_free(zp);
6393 else
6394 zfs_zinactive(zp);
6395 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6396
6397 vp->v_data = NULL;
6398 return (0);
6399}
6400
6401static int
6402zfs_freebsd_fid(ap)
6403 struct vop_fid_args /* {
6404 struct vnode *a_vp;
6405 struct fid *a_fid;
6406 } */ *ap;
6407{
6408
6409 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6410}
6411
6412static int
6413zfs_freebsd_pathconf(ap)
6414 struct vop_pathconf_args /* {
6415 struct vnode *a_vp;
6416 int a_name;
6417 register_t *a_retval;
6418 } */ *ap;
6419{
6420 ulong_t val;
6421 int error;
6422
6423 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6424 if (error == 0)
6425 *ap->a_retval = val;
6426 else if (error == EOPNOTSUPP)
6427 error = vop_stdpathconf(ap);
6428 return (error);
6429}
6430
6431static int
6432zfs_freebsd_fifo_pathconf(ap)
6433 struct vop_pathconf_args /* {
6434 struct vnode *a_vp;
6435 int a_name;
6436 register_t *a_retval;
6437 } */ *ap;
6438{
6439
6440 switch (ap->a_name) {
6441 case _PC_ACL_EXTENDED:
6442 case _PC_ACL_NFS4:
6443 case _PC_ACL_PATH_MAX:
6444 case _PC_MAC_PRESENT:
6445 return (zfs_freebsd_pathconf(ap));
6446 default:
6447 return (fifo_specops.vop_pathconf(ap));
6448 }
6449}
6450
6451/*
6452 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6453 * extended attribute name:
6454 *
6455 * NAMESPACE PREFIX
6456 * system freebsd:system:
6457 * user (none, can be used to access ZFS fsattr(5) attributes
6458 * created on Solaris)
6459 */
6460static int
6461zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6462 size_t size)
6463{
6464 const char *namespace, *prefix, *suffix;
6465
6466 /* We don't allow '/' character in attribute name. */
6467 if (strchr(name, '/') != NULL)
6468 return (EINVAL);
6469 /* We don't allow attribute names that start with "freebsd:" string. */
6470 if (strncmp(name, "freebsd:", 8) == 0)
6471 return (EINVAL);
6472
6473 bzero(attrname, size);
6474
6475 switch (attrnamespace) {
6476 case EXTATTR_NAMESPACE_USER:
6477#if 0
6478 prefix = "freebsd:";
6479 namespace = EXTATTR_NAMESPACE_USER_STRING;
6480 suffix = ":";
6481#else
6482 /*
6483 * This is the default namespace by which we can access all
6484 * attributes created on Solaris.
6485 */
6486 prefix = namespace = suffix = "";
6487#endif
6488 break;
6489 case EXTATTR_NAMESPACE_SYSTEM:
6490 prefix = "freebsd:";
6491 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6492 suffix = ":";
6493 break;
6494 case EXTATTR_NAMESPACE_EMPTY:
6495 default:
6496 return (EINVAL);
6497 }
6498 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6499 name) >= size) {
6500 return (ENAMETOOLONG);
6501 }
6502 return (0);
6503}
6504
6505/*
6506 * Vnode operating to retrieve a named extended attribute.
6507 */
6508static int
6509zfs_getextattr(struct vop_getextattr_args *ap)
6510/*
6511vop_getextattr {
6512 IN struct vnode *a_vp;
6513 IN int a_attrnamespace;
6514 IN const char *a_name;
6515 INOUT struct uio *a_uio;
6516 OUT size_t *a_size;
6517 IN struct ucred *a_cred;
6518 IN struct thread *a_td;
6519};
6520*/
6521{
6522 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6523 struct thread *td = ap->a_td;
6524 struct nameidata nd;
6525 char attrname[255];
6526 struct vattr va;
6527 vnode_t *xvp = NULL, *vp;
6528 int error, flags;
6529
6530 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6531 ap->a_cred, ap->a_td, VREAD);
6532 if (error != 0)
6533 return (error);
6534
6535 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6536 sizeof(attrname));
6537 if (error != 0)
6538 return (error);
6539
6540 ZFS_ENTER(zfsvfs);
6541
6542 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6543 LOOKUP_XATTR);
6544 if (error != 0) {
6545 ZFS_EXIT(zfsvfs);
6546 return (error);
6547 }
6548
6549 flags = FREAD;
6550 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6551 xvp, td);
6552 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6553 vp = nd.ni_vp;
6554 NDFREE(&nd, NDF_ONLY_PNBUF);
6555 if (error != 0) {
6556 ZFS_EXIT(zfsvfs);
6557 if (error == ENOENT)
6558 error = ENOATTR;
6559 return (error);
6560 }
6561
6562 if (ap->a_size != NULL) {
6563 error = VOP_GETATTR(vp, &va, ap->a_cred);
6564 if (error == 0)
6565 *ap->a_size = (size_t)va.va_size;
6566 } else if (ap->a_uio != NULL)
6567 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6568
6569 VOP_UNLOCK(vp, 0);
6570 vn_close(vp, flags, ap->a_cred, td);
6571 ZFS_EXIT(zfsvfs);
6572
6573 return (error);
6574}
6575
6576/*
6577 * Vnode operation to remove a named attribute.
6578 */
6579int
6580zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6581/*
6582vop_deleteextattr {
6583 IN struct vnode *a_vp;
6584 IN int a_attrnamespace;
6585 IN const char *a_name;
6586 IN struct ucred *a_cred;
6587 IN struct thread *a_td;
6588};
6589*/
6590{
6591 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6592 struct thread *td = ap->a_td;
6593 struct nameidata nd;
6594 char attrname[255];
6595 struct vattr va;
6596 vnode_t *xvp = NULL, *vp;
6597 int error, flags;
6598
6599 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6600 ap->a_cred, ap->a_td, VWRITE);
6601 if (error != 0)
6602 return (error);
6603
6604 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6605 sizeof(attrname));
6606 if (error != 0)
6607 return (error);
6608
6609 ZFS_ENTER(zfsvfs);
6610
6611 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6612 LOOKUP_XATTR);
6613 if (error != 0) {
6614 ZFS_EXIT(zfsvfs);
6615 return (error);
6616 }
6617
6618 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6619 UIO_SYSSPACE, attrname, xvp, td);
6620 error = namei(&nd);
6621 vp = nd.ni_vp;
6622 NDFREE(&nd, NDF_ONLY_PNBUF);
6623 if (error != 0) {
6624 ZFS_EXIT(zfsvfs);
6625 if (error == ENOENT)
6626 error = ENOATTR;
6627 return (error);
6628 }
6629 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6630
6631 vput(nd.ni_dvp);
6632 if (vp == nd.ni_dvp)
6633 vrele(vp);
6634 else
6635 vput(vp);
6636 ZFS_EXIT(zfsvfs);
6637
6638 return (error);
6639}
6640
6641/*
6642 * Vnode operation to set a named attribute.
6643 */
6644static int
6645zfs_setextattr(struct vop_setextattr_args *ap)
6646/*
6647vop_setextattr {
6648 IN struct vnode *a_vp;
6649 IN int a_attrnamespace;
6650 IN const char *a_name;
6651 INOUT struct uio *a_uio;
6652 IN struct ucred *a_cred;
6653 IN struct thread *a_td;
6654};
6655*/
6656{
6657 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6658 struct thread *td = ap->a_td;
6659 struct nameidata nd;
6660 char attrname[255];
6661 struct vattr va;
6662 vnode_t *xvp = NULL, *vp;
6663 int error, flags;
6664
6665 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6666 ap->a_cred, ap->a_td, VWRITE);
6667 if (error != 0)
6668 return (error);
6669
6670 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6671 sizeof(attrname));
6672 if (error != 0)
6673 return (error);
6674
6675 ZFS_ENTER(zfsvfs);
6676
6677 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6678 LOOKUP_XATTR | CREATE_XATTR_DIR);
6679 if (error != 0) {
6680 ZFS_EXIT(zfsvfs);
6681 return (error);
6682 }
6683
6684 flags = FFLAGS(O_WRONLY | O_CREAT);
6685 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6686 xvp, td);
6687 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6688 vp = nd.ni_vp;
6689 NDFREE(&nd, NDF_ONLY_PNBUF);
6690 if (error != 0) {
6691 ZFS_EXIT(zfsvfs);
6692 return (error);
6693 }
6694
6695 VATTR_NULL(&va);
6696 va.va_size = 0;
6697 error = VOP_SETATTR(vp, &va, ap->a_cred);
6698 if (error == 0)
6699 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6700
6701 VOP_UNLOCK(vp, 0);
6702 vn_close(vp, flags, ap->a_cred, td);
6703 ZFS_EXIT(zfsvfs);
6704
6705 return (error);
6706}
6707
6708/*
6709 * Vnode operation to retrieve extended attributes on a vnode.
6710 */
6711static int
6712zfs_listextattr(struct vop_listextattr_args *ap)
6713/*
6714vop_listextattr {
6715 IN struct vnode *a_vp;
6716 IN int a_attrnamespace;
6717 INOUT struct uio *a_uio;
6718 OUT size_t *a_size;
6719 IN struct ucred *a_cred;
6720 IN struct thread *a_td;
6721};
6722*/
6723{
6724 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6725 struct thread *td = ap->a_td;
6726 struct nameidata nd;
6727 char attrprefix[16];
6728 u_char dirbuf[sizeof(struct dirent)];
6729 struct dirent *dp;
6730 struct iovec aiov;
6731 struct uio auio, *uio = ap->a_uio;
6732 size_t *sizep = ap->a_size;
6733 size_t plen;
6734 vnode_t *xvp = NULL, *vp;
6735 int done, error, eof, pos;
6736
6737 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6738 ap->a_cred, ap->a_td, VREAD);
6739 if (error != 0)
6740 return (error);
6741
6742 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6743 sizeof(attrprefix));
6744 if (error != 0)
6745 return (error);
6746 plen = strlen(attrprefix);
6747
6748 ZFS_ENTER(zfsvfs);
6749
6750 if (sizep != NULL)
6751 *sizep = 0;
6752
6753 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6754 LOOKUP_XATTR);
6755 if (error != 0) {
6756 ZFS_EXIT(zfsvfs);
6757 /*
6758 * ENOATTR means that the EA directory does not yet exist,
6759 * i.e. there are no extended attributes there.
6760 */
6761 if (error == ENOATTR)
6762 error = 0;
6763 return (error);
6764 }
6765
6766 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6767 UIO_SYSSPACE, ".", xvp, td);
6768 error = namei(&nd);
6769 vp = nd.ni_vp;
6770 NDFREE(&nd, NDF_ONLY_PNBUF);
6771 if (error != 0) {
6772 ZFS_EXIT(zfsvfs);
6773 return (error);
6774 }
6775
6776 auio.uio_iov = &aiov;
6777 auio.uio_iovcnt = 1;
6778 auio.uio_segflg = UIO_SYSSPACE;
6779 auio.uio_td = td;
6780 auio.uio_rw = UIO_READ;
6781 auio.uio_offset = 0;
6782
6783 do {
6784 u_char nlen;
6785
6786 aiov.iov_base = (void *)dirbuf;
6787 aiov.iov_len = sizeof(dirbuf);
6788 auio.uio_resid = sizeof(dirbuf);
6789 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6790 done = sizeof(dirbuf) - auio.uio_resid;
6791 if (error != 0)
6792 break;
6793 for (pos = 0; pos < done;) {
6794 dp = (struct dirent *)(dirbuf + pos);
6795 pos += dp->d_reclen;
6796 /*
6797 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6798 * is what we get when attribute was created on Solaris.
6799 */
6800 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6801 continue;
6802 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6803 continue;
6804 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6805 continue;
6806 nlen = dp->d_namlen - plen;
6807 if (sizep != NULL)
6808 *sizep += 1 + nlen;
6809 else if (uio != NULL) {
6810 /*
6811 * Format of extattr name entry is one byte for
6812 * length and the rest for name.
6813 */
6814 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6815 if (error == 0) {
6816 error = uiomove(dp->d_name + plen, nlen,
6817 uio->uio_rw, uio);
6818 }
6819 if (error != 0)
6820 break;
6821 }
6822 }
6823 } while (!eof && error == 0);
6824
6825 vput(vp);
6826 ZFS_EXIT(zfsvfs);
6827
6828 return (error);
6829}
6830
6831int
6832zfs_freebsd_getacl(ap)
6833 struct vop_getacl_args /* {
6834 struct vnode *vp;
6835 acl_type_t type;
6836 struct acl *aclp;
6837 struct ucred *cred;
6838 struct thread *td;
6839 } */ *ap;
6840{
6841 int error;
6842 vsecattr_t vsecattr;
6843
6844 if (ap->a_type != ACL_TYPE_NFS4)
6845 return (EINVAL);
6846
6847 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6848 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6849 return (error);
6850
6851 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6852 if (vsecattr.vsa_aclentp != NULL)
6853 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6854
6855 return (error);
6856}
6857
6858int
6859zfs_freebsd_setacl(ap)
6860 struct vop_setacl_args /* {
6861 struct vnode *vp;
6862 acl_type_t type;
6863 struct acl *aclp;
6864 struct ucred *cred;
6865 struct thread *td;
6866 } */ *ap;
6867{
6868 int error;
6869 vsecattr_t vsecattr;
6870 int aclbsize; /* size of acl list in bytes */
6871 aclent_t *aaclp;
6872
6873 if (ap->a_type != ACL_TYPE_NFS4)
6874 return (EINVAL);
6875
6876 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6877 return (EINVAL);
6878
6879 /*
6880 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6881 * splitting every entry into two and appending "canonical six"
6882 * entries at the end. Don't allow for setting an ACL that would
6883 * cause chmod(2) to run out of ACL entries.
6884 */
6885 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6886 return (ENOSPC);
6887
6888 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6889 if (error != 0)
6890 return (error);
6891
6892 vsecattr.vsa_mask = VSA_ACE;
6893 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6894 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6895 aaclp = vsecattr.vsa_aclentp;
6896 vsecattr.vsa_aclentsz = aclbsize;
6897
6898 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6899 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6900 kmem_free(aaclp, aclbsize);
6901
6902 return (error);
6903}
6904
6905int
6906zfs_freebsd_aclcheck(ap)
6907 struct vop_aclcheck_args /* {
6908 struct vnode *vp;
6909 acl_type_t type;
6910 struct acl *aclp;
6911 struct ucred *cred;
6912 struct thread *td;
6913 } */ *ap;
6914{
6915
6916 return (EOPNOTSUPP);
6917}
6918
6919struct vop_vector zfs_vnodeops;
6920struct vop_vector zfs_fifoops;
6921struct vop_vector zfs_shareops;
6922
6923struct vop_vector zfs_vnodeops = {
6924 .vop_default = &default_vnodeops,
6925 .vop_inactive = zfs_freebsd_inactive,
6926 .vop_reclaim = zfs_freebsd_reclaim,
6927 .vop_access = zfs_freebsd_access,
6928#ifdef FREEBSD_NAMECACHE
6929 .vop_lookup = vfs_cache_lookup,
6930 .vop_cachedlookup = zfs_freebsd_lookup,
6931#else
6932 .vop_lookup = zfs_freebsd_lookup,
6933#endif
6934 .vop_getattr = zfs_freebsd_getattr,
6935 .vop_setattr = zfs_freebsd_setattr,
6936 .vop_create = zfs_freebsd_create,
6937 .vop_mknod = zfs_freebsd_create,
6938 .vop_mkdir = zfs_freebsd_mkdir,
6939 .vop_readdir = zfs_freebsd_readdir,
6940 .vop_fsync = zfs_freebsd_fsync,
6941 .vop_open = zfs_freebsd_open,
6942 .vop_close = zfs_freebsd_close,
6943 .vop_rmdir = zfs_freebsd_rmdir,
6944 .vop_ioctl = zfs_freebsd_ioctl,
6945 .vop_link = zfs_freebsd_link,
6946 .vop_symlink = zfs_freebsd_symlink,
6947 .vop_readlink = zfs_freebsd_readlink,
6948 .vop_read = zfs_freebsd_read,
6949 .vop_write = zfs_freebsd_write,
6950 .vop_remove = zfs_freebsd_remove,
6951 .vop_rename = zfs_freebsd_rename,
6952 .vop_pathconf = zfs_freebsd_pathconf,
6953 .vop_bmap = zfs_freebsd_bmap,
6954 .vop_fid = zfs_freebsd_fid,
6955 .vop_getextattr = zfs_getextattr,
6956 .vop_deleteextattr = zfs_deleteextattr,
6957 .vop_setextattr = zfs_setextattr,
6958 .vop_listextattr = zfs_listextattr,
6959 .vop_getacl = zfs_freebsd_getacl,
6960 .vop_setacl = zfs_freebsd_setacl,
6961 .vop_aclcheck = zfs_freebsd_aclcheck,
6962 .vop_getpages = zfs_freebsd_getpages,
6963};
6964
6965struct vop_vector zfs_fifoops = {
6966 .vop_default = &fifo_specops,
6967 .vop_fsync = zfs_freebsd_fsync,
6968 .vop_access = zfs_freebsd_access,
6969 .vop_getattr = zfs_freebsd_getattr,
6970 .vop_inactive = zfs_freebsd_inactive,
6971 .vop_read = VOP_PANIC,
6972 .vop_reclaim = zfs_freebsd_reclaim,
6973 .vop_setattr = zfs_freebsd_setattr,
6974 .vop_write = VOP_PANIC,
6975 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6976 .vop_fid = zfs_freebsd_fid,
6977 .vop_getacl = zfs_freebsd_getacl,
6978 .vop_setacl = zfs_freebsd_setacl,
6979 .vop_aclcheck = zfs_freebsd_aclcheck,
6980};
6981
6982/*
6983 * special share hidden files vnode operations template
6984 */
6985struct vop_vector zfs_shareops = {
6986 .vop_default = &default_vnodeops,
6987 .vop_access = zfs_freebsd_access,
6988 .vop_inactive = zfs_freebsd_inactive,
6989 .vop_reclaim = zfs_freebsd_reclaim,
6990 .vop_fid = zfs_freebsd_fid,
6991 .vop_pathconf = zfs_freebsd_pathconf,
6992};