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) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
111 * This is critical because we don't want to block while holding locks.
112 * Note, in particular, that if a lock is sometimes acquired before
113 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
114 * use a non-blocking assign can deadlock the system. The scenario:
115 *
116 * Thread A has grabbed a lock before calling dmu_tx_assign().
117 * Thread B is in an already-assigned tx, and blocks for this lock.
118 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
119 * forever, because the previous txg can't quiesce until B's tx commits.
120 *
121 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
122 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
123 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
124 * to indicate that this operation has already called dmu_tx_wait().
125 * This will ensure that we don't retry forever, waiting a short bit
126 * each time.
127 *
128 * (5) If the operation succeeded, generate the intent log entry for it
129 * before dropping locks. This ensures that the ordering of events
130 * in the intent log matches the order in which they actually occurred.
131 * During ZIL replay the zfs_log_* functions will update the sequence
132 * number to indicate the zil transaction has replayed.
133 *
134 * (6) At the end of each vnode op, the DMU tx must always commit,
135 * regardless of whether there were any errors.
136 *
137 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
138 * to ensure that synchronous semantics are provided when necessary.
139 *
140 * In general, this is how things should be ordered in each vnode op:
141 *
142 * ZFS_ENTER(zfsvfs); // exit if unmounted
143 * top:
144 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
145 * rw_enter(...); // grab any other locks you need
146 * tx = dmu_tx_create(...); // get DMU tx
147 * dmu_tx_hold_*(); // hold each object you might modify
148 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
149 * if (error) {
150 * rw_exit(...); // drop locks
151 * zfs_dirent_unlock(dl); // unlock directory entry
152 * VN_RELE(...); // release held vnodes
153 * if (error == ERESTART) {
154 * waited = B_TRUE;
155 * dmu_tx_wait(tx);
156 * dmu_tx_abort(tx);
157 * goto top;
158 * }
159 * dmu_tx_abort(tx); // abort DMU tx
160 * ZFS_EXIT(zfsvfs); // finished in zfs
161 * return (error); // really out of space
162 * }
163 * error = do_real_work(); // do whatever this VOP does
164 * if (error == 0)
165 * zfs_log_*(...); // on success, make ZIL entry
166 * dmu_tx_commit(tx); // commit DMU tx -- error or not
167 * rw_exit(...); // drop locks
168 * zfs_dirent_unlock(dl); // unlock directory entry
169 * VN_RELE(...); // release held vnodes
170 * zil_commit(zilog, foid); // synchronous when necessary
171 * ZFS_EXIT(zfsvfs); // finished in zfs
172 * return (error); // done, report error
173 */
174
175/* ARGSUSED */
176static int
177zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
178{
179 znode_t *zp = VTOZ(*vpp);
180 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
181
182 ZFS_ENTER(zfsvfs);
183 ZFS_VERIFY_ZP(zp);
184
185 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
186 ((flag & FAPPEND) == 0)) {
187 ZFS_EXIT(zfsvfs);
188 return (SET_ERROR(EPERM));
189 }
190
191 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
192 ZTOV(zp)->v_type == VREG &&
193 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
194 if (fs_vscan(*vpp, cr, 0) != 0) {
195 ZFS_EXIT(zfsvfs);
196 return (SET_ERROR(EACCES));
197 }
198 }
199
200 /* Keep a count of the synchronous opens in the znode */
201 if (flag & (FSYNC | FDSYNC))
202 atomic_inc_32(&zp->z_sync_cnt);
203
204 ZFS_EXIT(zfsvfs);
205 return (0);
206}
207
208/* ARGSUSED */
209static int
210zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
211 caller_context_t *ct)
212{
213 znode_t *zp = VTOZ(vp);
214 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
215
216 /*
217 * Clean up any locks held by this process on the vp.
218 */
219 cleanlocks(vp, ddi_get_pid(), 0);
220 cleanshares(vp, ddi_get_pid());
221
222 ZFS_ENTER(zfsvfs);
223 ZFS_VERIFY_ZP(zp);
224
225 /* Decrement the synchronous opens in the znode */
226 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
227 atomic_dec_32(&zp->z_sync_cnt);
228
229 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
230 ZTOV(zp)->v_type == VREG &&
231 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
232 VERIFY(fs_vscan(vp, cr, 1) == 0);
233
234 ZFS_EXIT(zfsvfs);
235 return (0);
236}
237
238/*
239 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
240 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
241 */
242static int
243zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
244{
245 znode_t *zp = VTOZ(vp);
246 uint64_t noff = (uint64_t)*off; /* new offset */
247 uint64_t file_sz;
248 int error;
249 boolean_t hole;
250
251 file_sz = zp->z_size;
252 if (noff >= file_sz) {
253 return (SET_ERROR(ENXIO));
254 }
255
256 if (cmd == _FIO_SEEK_HOLE)
257 hole = B_TRUE;
258 else
259 hole = B_FALSE;
260
261 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
262
263 /* end of file? */
264 if ((error == ESRCH) || (noff > file_sz)) {
265 /*
266 * Handle the virtual hole at the end of file.
267 */
268 if (hole) {
269 *off = file_sz;
270 return (0);
271 }
272 return (SET_ERROR(ENXIO));
273 }
274
275 if (noff < *off)
276 return (error);
277 *off = noff;
278 return (error);
279}
280
281/* ARGSUSED */
282static int
283zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
284 int *rvalp, caller_context_t *ct)
285{
286 offset_t off;
287 int error;
288 zfsvfs_t *zfsvfs;
289 znode_t *zp;
290
291 switch (com) {
292 case _FIOFFS:
293 return (0);
294
295 /*
296 * The following two ioctls are used by bfu. Faking out,
297 * necessary to avoid bfu errors.
298 */
299 case _FIOGDIO:
300 case _FIOSDIO:
301 return (0);
302
303 case _FIO_SEEK_DATA:
304 case _FIO_SEEK_HOLE:
305#ifdef sun
306 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
307 return (SET_ERROR(EFAULT));
308#else
309 off = *(offset_t *)data;
310#endif
311 zp = VTOZ(vp);
312 zfsvfs = zp->z_zfsvfs;
313 ZFS_ENTER(zfsvfs);
314 ZFS_VERIFY_ZP(zp);
315
316 /* offset parameter is in/out */
317 error = zfs_holey(vp, com, &off);
318 ZFS_EXIT(zfsvfs);
319 if (error)
320 return (error);
321#ifdef sun
322 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
323 return (SET_ERROR(EFAULT));
324#else
325 *(offset_t *)data = off;
326#endif
327 return (0);
328 }
329 return (SET_ERROR(ENOTTY));
330}
331
332static vm_page_t
333page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
334{
335 vm_object_t obj;
336 vm_page_t pp;
337 int64_t end;
338
339 /*
340 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
341 * aligned boundaries, if the range is not aligned. As a result a
342 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
343 * It may happen that all DEV_BSIZE subranges are marked clean and thus
344 * the whole page would be considred clean despite have some dirty data.
345 * For this reason we should shrink the range to DEV_BSIZE aligned
346 * boundaries before calling vm_page_clear_dirty.
347 */
348 end = rounddown2(off + nbytes, DEV_BSIZE);
349 off = roundup2(off, DEV_BSIZE);
350 nbytes = end - off;
351
352 obj = vp->v_object;
353 zfs_vmobject_assert_wlocked(obj);
354
355 for (;;) {
356 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
357 pp->valid) {
358 if (vm_page_xbusied(pp)) {
359 /*
360 * Reference the page before unlocking and
361 * sleeping so that the page daemon is less
362 * likely to reclaim it.
363 */
364 vm_page_reference(pp);
365 vm_page_lock(pp);
366 zfs_vmobject_wunlock(obj);
367 vm_page_busy_sleep(pp, "zfsmwb");
368 zfs_vmobject_wlock(obj);
369 continue;
370 }
371 vm_page_sbusy(pp);
372 } else if (pp == NULL) {
373 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
374 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
375 VM_ALLOC_SBUSY);
376 } else {
377 ASSERT(pp != NULL && !pp->valid);
378 pp = NULL;
379 }
380
381 if (pp != NULL) {
382 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
383 vm_object_pip_add(obj, 1);
384 pmap_remove_write(pp);
385 if (nbytes != 0)
386 vm_page_clear_dirty(pp, off, nbytes);
387 }
388 break;
389 }
390 return (pp);
391}
392
393static void
394page_unbusy(vm_page_t pp)
395{
396
397 vm_page_sunbusy(pp);
398 vm_object_pip_subtract(pp->object, 1);
399}
400
401static vm_page_t
402page_hold(vnode_t *vp, int64_t start)
403{
404 vm_object_t obj;
405 vm_page_t pp;
406
407 obj = vp->v_object;
408 zfs_vmobject_assert_wlocked(obj);
409
410 for (;;) {
411 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
412 pp->valid) {
413 if (vm_page_xbusied(pp)) {
414 /*
415 * Reference the page before unlocking and
416 * sleeping so that the page daemon is less
417 * likely to reclaim it.
418 */
419 vm_page_reference(pp);
420 vm_page_lock(pp);
421 zfs_vmobject_wunlock(obj);
422 vm_page_busy_sleep(pp, "zfsmwb");
423 zfs_vmobject_wlock(obj);
424 continue;
425 }
426
427 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
428 vm_page_lock(pp);
429 vm_page_hold(pp);
430 vm_page_unlock(pp);
431
432 } else
433 pp = NULL;
434 break;
435 }
436 return (pp);
437}
438
439static void
440page_unhold(vm_page_t pp)
441{
442
443 vm_page_lock(pp);
444 vm_page_unhold(pp);
445 vm_page_unlock(pp);
446}
447
448static caddr_t
449zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
450{
451
452 *sfp = sf_buf_alloc(pp, 0);
453 return ((caddr_t)sf_buf_kva(*sfp));
454}
455
456static void
457zfs_unmap_page(struct sf_buf *sf)
458{
459
460 sf_buf_free(sf);
461}
462
463/*
464 * When a file is memory mapped, we must keep the IO data synchronized
465 * between the DMU cache and the memory mapped pages. What this means:
466 *
467 * On Write: If we find a memory mapped page, we write to *both*
468 * the page and the dmu buffer.
469 */
470static void
471update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
472 int segflg, dmu_tx_t *tx)
473{
474 vm_object_t obj;
475 struct sf_buf *sf;
476 caddr_t va;
477 int off;
478
479 ASSERT(vp->v_mount != NULL);
480 obj = vp->v_object;
481 ASSERT(obj != NULL);
482
483 off = start & PAGEOFFSET;
484 zfs_vmobject_wlock(obj);
485 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
486 vm_page_t pp;
487 int nbytes = imin(PAGESIZE - off, len);
488
489 if (segflg == UIO_NOCOPY) {
490 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
491 KASSERT(pp != NULL,
492 ("zfs update_pages: NULL page in putpages case"));
493 KASSERT(off == 0,
494 ("zfs update_pages: unaligned data in putpages case"));
495 KASSERT(pp->valid == VM_PAGE_BITS_ALL,
496 ("zfs update_pages: invalid page in putpages case"));
497 KASSERT(vm_page_sbusied(pp),
498 ("zfs update_pages: unbusy page in putpages case"));
499 KASSERT(!pmap_page_is_write_mapped(pp),
500 ("zfs update_pages: writable page in putpages case"));
501 zfs_vmobject_wunlock(obj);
502
503 va = zfs_map_page(pp, &sf);
504 (void) dmu_write(os, oid, start, nbytes, va, tx);
505 zfs_unmap_page(sf);
506
507 zfs_vmobject_wlock(obj);
508 vm_page_undirty(pp);
509 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
510 zfs_vmobject_wunlock(obj);
511
512 va = zfs_map_page(pp, &sf);
513 (void) dmu_read(os, oid, start+off, nbytes,
514 va+off, DMU_READ_PREFETCH);;
515 zfs_unmap_page(sf);
516
517 zfs_vmobject_wlock(obj);
518 page_unbusy(pp);
519 }
520 len -= nbytes;
521 off = 0;
522 }
523 if (segflg != UIO_NOCOPY)
524 vm_object_pip_wakeupn(obj, 0);
525 zfs_vmobject_wunlock(obj);
526}
527
528/*
529 * Read with UIO_NOCOPY flag means that sendfile(2) requests
530 * ZFS to populate a range of page cache pages with data.
531 *
532 * NOTE: this function could be optimized to pre-allocate
533 * all pages in advance, drain exclusive busy on all of them,
534 * map them into contiguous KVA region and populate them
535 * in one single dmu_read() call.
536 */
537static int
538mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
539{
540 znode_t *zp = VTOZ(vp);
541 objset_t *os = zp->z_zfsvfs->z_os;
542 struct sf_buf *sf;
543 vm_object_t obj;
544 vm_page_t pp;
545 int64_t start;
546 caddr_t va;
547 int len = nbytes;
548 int off;
549 int error = 0;
550
551 ASSERT(uio->uio_segflg == UIO_NOCOPY);
552 ASSERT(vp->v_mount != NULL);
553 obj = vp->v_object;
554 ASSERT(obj != NULL);
555 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
556
557 zfs_vmobject_wlock(obj);
558 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
559 int bytes = MIN(PAGESIZE, len);
560
561 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY |
562 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
563 if (pp->valid == 0) {
564 zfs_vmobject_wunlock(obj);
565 va = zfs_map_page(pp, &sf);
566 error = dmu_read(os, zp->z_id, start, bytes, va,
567 DMU_READ_PREFETCH);
568 if (bytes != PAGESIZE && error == 0)
569 bzero(va + bytes, PAGESIZE - bytes);
570 zfs_unmap_page(sf);
571 zfs_vmobject_wlock(obj);
572 vm_page_sunbusy(pp);
573 vm_page_lock(pp);
574 if (error) {
575 if (pp->wire_count == 0 && pp->valid == 0 &&
576 !vm_page_busied(pp))
577 vm_page_free(pp);
578 } else {
579 pp->valid = VM_PAGE_BITS_ALL;
580 vm_page_activate(pp);
581 }
582 vm_page_unlock(pp);
583 } else
584 vm_page_sunbusy(pp);
585 if (error)
586 break;
587 uio->uio_resid -= bytes;
588 uio->uio_offset += bytes;
589 len -= bytes;
590 }
591 zfs_vmobject_wunlock(obj);
592 return (error);
593}
594
595/*
596 * When a file is memory mapped, we must keep the IO data synchronized
597 * between the DMU cache and the memory mapped pages. What this means:
598 *
599 * On Read: We "read" preferentially from memory mapped pages,
600 * else we default from the dmu buffer.
601 *
602 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
603 * the file is memory mapped.
604 */
605static int
606mappedread(vnode_t *vp, int nbytes, uio_t *uio)
607{
608 znode_t *zp = VTOZ(vp);
609 objset_t *os = zp->z_zfsvfs->z_os;
610 vm_object_t obj;
611 int64_t start;
612 caddr_t va;
613 int len = nbytes;
614 int off;
615 int error = 0;
616
617 ASSERT(vp->v_mount != NULL);
618 obj = vp->v_object;
619 ASSERT(obj != NULL);
620
621 start = uio->uio_loffset;
622 off = start & PAGEOFFSET;
623 zfs_vmobject_wlock(obj);
624 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
625 vm_page_t pp;
626 uint64_t bytes = MIN(PAGESIZE - off, len);
627
628 if (pp = page_hold(vp, start)) {
629 struct sf_buf *sf;
630 caddr_t va;
631
632 zfs_vmobject_wunlock(obj);
633 va = zfs_map_page(pp, &sf);
634 error = uiomove(va + off, bytes, UIO_READ, uio);
635 zfs_unmap_page(sf);
636 zfs_vmobject_wlock(obj);
637 page_unhold(pp);
638 } else {
639 zfs_vmobject_wunlock(obj);
640 error = dmu_read_uio(os, zp->z_id, uio, bytes);
641 zfs_vmobject_wlock(obj);
642 }
643 len -= bytes;
644 off = 0;
645 if (error)
646 break;
647 }
648 zfs_vmobject_wunlock(obj);
649 return (error);
650}
651
652offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
653
654/*
655 * Read bytes from specified file into supplied buffer.
656 *
657 * IN: vp - vnode of file to be read from.
658 * uio - structure supplying read location, range info,
659 * and return buffer.
660 * ioflag - SYNC flags; used to provide FRSYNC semantics.
661 * cr - credentials of caller.
662 * ct - caller context
663 *
664 * OUT: uio - updated offset and range, buffer filled.
665 *
666 * RETURN: 0 on success, error code on failure.
667 *
668 * Side Effects:
669 * vp - atime updated if byte count > 0
670 */
671/* ARGSUSED */
672static int
673zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
674{
675 znode_t *zp = VTOZ(vp);
676 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
677 objset_t *os;
678 ssize_t n, nbytes;
679 int error = 0;
680 rl_t *rl;
681 xuio_t *xuio = NULL;
682
683 ZFS_ENTER(zfsvfs);
684 ZFS_VERIFY_ZP(zp);
685 os = zfsvfs->z_os;
686
687 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
688 ZFS_EXIT(zfsvfs);
689 return (SET_ERROR(EACCES));
690 }
691
692 /*
693 * Validate file offset
694 */
695 if (uio->uio_loffset < (offset_t)0) {
696 ZFS_EXIT(zfsvfs);
697 return (SET_ERROR(EINVAL));
698 }
699
700 /*
701 * Fasttrack empty reads
702 */
703 if (uio->uio_resid == 0) {
704 ZFS_EXIT(zfsvfs);
705 return (0);
706 }
707
708 /*
709 * Check for mandatory locks
710 */
711 if (MANDMODE(zp->z_mode)) {
712 if (error = chklock(vp, FREAD,
713 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
714 ZFS_EXIT(zfsvfs);
715 return (error);
716 }
717 }
718
719 /*
720 * If we're in FRSYNC mode, sync out this znode before reading it.
721 */
722 if (zfsvfs->z_log &&
723 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
724 zil_commit(zfsvfs->z_log, zp->z_id);
725
726 /*
727 * Lock the range against changes.
728 */
729 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
730
731 /*
732 * If we are reading past end-of-file we can skip
733 * to the end; but we might still need to set atime.
734 */
735 if (uio->uio_loffset >= zp->z_size) {
736 error = 0;
737 goto out;
738 }
739
740 ASSERT(uio->uio_loffset < zp->z_size);
741 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
742
743#ifdef sun
744 if ((uio->uio_extflg == UIO_XUIO) &&
745 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
746 int nblk;
747 int blksz = zp->z_blksz;
748 uint64_t offset = uio->uio_loffset;
749
750 xuio = (xuio_t *)uio;
751 if ((ISP2(blksz))) {
752 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
753 blksz)) / blksz;
754 } else {
755 ASSERT(offset + n <= blksz);
756 nblk = 1;
757 }
758 (void) dmu_xuio_init(xuio, nblk);
759
760 if (vn_has_cached_data(vp)) {
761 /*
762 * For simplicity, we always allocate a full buffer
763 * even if we only expect to read a portion of a block.
764 */
765 while (--nblk >= 0) {
766 (void) dmu_xuio_add(xuio,
767 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
768 blksz), 0, blksz);
769 }
770 }
771 }
772#endif /* sun */
773
774 while (n > 0) {
775 nbytes = MIN(n, zfs_read_chunk_size -
776 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
777
778#ifdef __FreeBSD__
779 if (uio->uio_segflg == UIO_NOCOPY)
780 error = mappedread_sf(vp, nbytes, uio);
781 else
782#endif /* __FreeBSD__ */
783 if (vn_has_cached_data(vp))
784 error = mappedread(vp, nbytes, uio);
785 else
786 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
787 if (error) {
788 /* convert checksum errors into IO errors */
789 if (error == ECKSUM)
790 error = SET_ERROR(EIO);
791 break;
792 }
793
794 n -= nbytes;
795 }
796out:
797 zfs_range_unlock(rl);
798
799 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
800 ZFS_EXIT(zfsvfs);
801 return (error);
802}
803
804/*
805 * Write the bytes to a file.
806 *
807 * IN: vp - vnode of file to be written to.
808 * uio - structure supplying write location, range info,
809 * and data buffer.
810 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
811 * set if in append mode.
812 * cr - credentials of caller.
813 * ct - caller context (NFS/CIFS fem monitor only)
814 *
815 * OUT: uio - updated offset and range.
816 *
817 * RETURN: 0 on success, error code on failure.
818 *
819 * Timestamps:
820 * vp - ctime|mtime updated if byte count > 0
821 */
822
823/* ARGSUSED */
824static int
825zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
826{
827 znode_t *zp = VTOZ(vp);
828 rlim64_t limit = MAXOFFSET_T;
829 ssize_t start_resid = uio->uio_resid;
830 ssize_t tx_bytes;
831 uint64_t end_size;
832 dmu_tx_t *tx;
833 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
834 zilog_t *zilog;
835 offset_t woff;
836 ssize_t n, nbytes;
837 rl_t *rl;
838 int max_blksz = zfsvfs->z_max_blksz;
839 int error = 0;
840 arc_buf_t *abuf;
841 iovec_t *aiov = NULL;
842 xuio_t *xuio = NULL;
843 int i_iov = 0;
844 int iovcnt = uio->uio_iovcnt;
845 iovec_t *iovp = uio->uio_iov;
846 int write_eof;
847 int count = 0;
848 sa_bulk_attr_t bulk[4];
849 uint64_t mtime[2], ctime[2];
850
851 /*
852 * Fasttrack empty write
853 */
854 n = start_resid;
855 if (n == 0)
856 return (0);
857
858 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
859 limit = MAXOFFSET_T;
860
861 ZFS_ENTER(zfsvfs);
862 ZFS_VERIFY_ZP(zp);
863
864 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
865 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
866 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
867 &zp->z_size, 8);
868 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
869 &zp->z_pflags, 8);
870
871 /*
872 * If immutable or not appending then return EPERM
873 */
874 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
875 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
876 (uio->uio_loffset < zp->z_size))) {
877 ZFS_EXIT(zfsvfs);
878 return (SET_ERROR(EPERM));
879 }
880
881 zilog = zfsvfs->z_log;
882
883 /*
884 * Validate file offset
885 */
886 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
887 if (woff < 0) {
888 ZFS_EXIT(zfsvfs);
889 return (SET_ERROR(EINVAL));
890 }
891
892 /*
893 * Check for mandatory locks before calling zfs_range_lock()
894 * in order to prevent a deadlock with locks set via fcntl().
895 */
896 if (MANDMODE((mode_t)zp->z_mode) &&
897 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
898 ZFS_EXIT(zfsvfs);
899 return (error);
900 }
901
902#ifdef sun
903 /*
904 * Pre-fault the pages to ensure slow (eg NFS) pages
905 * don't hold up txg.
906 * Skip this if uio contains loaned arc_buf.
907 */
908 if ((uio->uio_extflg == UIO_XUIO) &&
909 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
910 xuio = (xuio_t *)uio;
911 else
912 uio_prefaultpages(MIN(n, max_blksz), uio);
913#endif /* sun */
914
915 /*
916 * If in append mode, set the io offset pointer to eof.
917 */
918 if (ioflag & FAPPEND) {
919 /*
920 * Obtain an appending range lock to guarantee file append
921 * semantics. We reset the write offset once we have the lock.
922 */
923 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
924 woff = rl->r_off;
925 if (rl->r_len == UINT64_MAX) {
926 /*
927 * We overlocked the file because this write will cause
928 * the file block size to increase.
929 * Note that zp_size cannot change with this lock held.
930 */
931 woff = zp->z_size;
932 }
933 uio->uio_loffset = woff;
934 } else {
935 /*
936 * Note that if the file block size will change as a result of
937 * this write, then this range lock will lock the entire file
938 * so that we can re-write the block safely.
939 */
940 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
941 }
942
943 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
944 zfs_range_unlock(rl);
945 ZFS_EXIT(zfsvfs);
946 return (EFBIG);
947 }
948
949 if (woff >= limit) {
950 zfs_range_unlock(rl);
951 ZFS_EXIT(zfsvfs);
952 return (SET_ERROR(EFBIG));
953 }
954
955 if ((woff + n) > limit || woff > (limit - n))
956 n = limit - woff;
957
958 /* Will this write extend the file length? */
959 write_eof = (woff + n > zp->z_size);
960
961 end_size = MAX(zp->z_size, woff + n);
962
963 /*
964 * Write the file in reasonable size chunks. Each chunk is written
965 * in a separate transaction; this keeps the intent log records small
966 * and allows us to do more fine-grained space accounting.
967 */
968 while (n > 0) {
969 abuf = NULL;
970 woff = uio->uio_loffset;
971again:
972 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
973 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
974 if (abuf != NULL)
975 dmu_return_arcbuf(abuf);
976 error = SET_ERROR(EDQUOT);
977 break;
978 }
979
980 if (xuio && abuf == NULL) {
981 ASSERT(i_iov < iovcnt);
982 aiov = &iovp[i_iov];
983 abuf = dmu_xuio_arcbuf(xuio, i_iov);
984 dmu_xuio_clear(xuio, i_iov);
985 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
986 iovec_t *, aiov, arc_buf_t *, abuf);
987 ASSERT((aiov->iov_base == abuf->b_data) ||
988 ((char *)aiov->iov_base - (char *)abuf->b_data +
989 aiov->iov_len == arc_buf_size(abuf)));
990 i_iov++;
991 } else if (abuf == NULL && n >= max_blksz &&
992 woff >= zp->z_size &&
993 P2PHASE(woff, max_blksz) == 0 &&
994 zp->z_blksz == max_blksz) {
995 /*
996 * This write covers a full block. "Borrow" a buffer
997 * from the dmu so that we can fill it before we enter
998 * a transaction. This avoids the possibility of
999 * holding up the transaction if the data copy hangs
1000 * up on a pagefault (e.g., from an NFS server mapping).
1001 */
1002 size_t cbytes;
1003
1004 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
1005 max_blksz);
1006 ASSERT(abuf != NULL);
1007 ASSERT(arc_buf_size(abuf) == max_blksz);
1008 if (error = uiocopy(abuf->b_data, max_blksz,
1009 UIO_WRITE, uio, &cbytes)) {
1010 dmu_return_arcbuf(abuf);
1011 break;
1012 }
1013 ASSERT(cbytes == max_blksz);
1014 }
1015
1016 /*
1017 * Start a transaction.
1018 */
1019 tx = dmu_tx_create(zfsvfs->z_os);
1020 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1021 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1022 zfs_sa_upgrade_txholds(tx, zp);
1023 error = dmu_tx_assign(tx, TXG_NOWAIT);
1024 if (error) {
1025 if (error == ERESTART) {
1026 dmu_tx_wait(tx);
1027 dmu_tx_abort(tx);
1028 goto again;
1029 }
1030 dmu_tx_abort(tx);
1031 if (abuf != NULL)
1032 dmu_return_arcbuf(abuf);
1033 break;
1034 }
1035
1036 /*
1037 * If zfs_range_lock() over-locked we grow the blocksize
1038 * and then reduce the lock range. This will only happen
1039 * on the first iteration since zfs_range_reduce() will
1040 * shrink down r_len to the appropriate size.
1041 */
1042 if (rl->r_len == UINT64_MAX) {
1043 uint64_t new_blksz;
1044
1045 if (zp->z_blksz > max_blksz) {
1046 ASSERT(!ISP2(zp->z_blksz));
1047 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1048 } else {
1049 new_blksz = MIN(end_size, max_blksz);
1050 }
1051 zfs_grow_blocksize(zp, new_blksz, tx);
1052 zfs_range_reduce(rl, woff, n);
1053 }
1054
1055 /*
1056 * XXX - should we really limit each write to z_max_blksz?
1057 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1058 */
1059 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1060
1061 if (woff + nbytes > zp->z_size)
1062 vnode_pager_setsize(vp, woff + nbytes);
1063
1064 if (abuf == NULL) {
1065 tx_bytes = uio->uio_resid;
1066 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1067 uio, nbytes, tx);
1068 tx_bytes -= uio->uio_resid;
1069 } else {
1070 tx_bytes = nbytes;
1071 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1072 /*
1073 * If this is not a full block write, but we are
1074 * extending the file past EOF and this data starts
1075 * block-aligned, use assign_arcbuf(). Otherwise,
1076 * write via dmu_write().
1077 */
1078 if (tx_bytes < max_blksz && (!write_eof ||
1079 aiov->iov_base != abuf->b_data)) {
1080 ASSERT(xuio);
1081 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1082 aiov->iov_len, aiov->iov_base, tx);
1083 dmu_return_arcbuf(abuf);
1084 xuio_stat_wbuf_copied();
1085 } else {
1086 ASSERT(xuio || tx_bytes == max_blksz);
1087 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1088 woff, abuf, tx);
1089 }
1090 ASSERT(tx_bytes <= uio->uio_resid);
1091 uioskip(uio, tx_bytes);
1092 }
1093 if (tx_bytes && vn_has_cached_data(vp)) {
1094 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1095 zp->z_id, uio->uio_segflg, tx);
1096 }
1097
1098 /*
1099 * If we made no progress, we're done. If we made even
1100 * partial progress, update the znode and ZIL accordingly.
1101 */
1102 if (tx_bytes == 0) {
1103 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1104 (void *)&zp->z_size, sizeof (uint64_t), tx);
1105 dmu_tx_commit(tx);
1106 ASSERT(error != 0);
1107 break;
1108 }
1109
1110 /*
1111 * Clear Set-UID/Set-GID bits on successful write if not
1112 * privileged and at least one of the excute bits is set.
1113 *
1114 * It would be nice to to this after all writes have
1115 * been done, but that would still expose the ISUID/ISGID
1116 * to another app after the partial write is committed.
1117 *
1118 * Note: we don't call zfs_fuid_map_id() here because
1119 * user 0 is not an ephemeral uid.
1120 */
1121 mutex_enter(&zp->z_acl_lock);
1122 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1123 (S_IXUSR >> 6))) != 0 &&
1124 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1125 secpolicy_vnode_setid_retain(vp, cr,
1126 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1127 uint64_t newmode;
1128 zp->z_mode &= ~(S_ISUID | S_ISGID);
1129 newmode = zp->z_mode;
1130 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1131 (void *)&newmode, sizeof (uint64_t), tx);
1132 }
1133 mutex_exit(&zp->z_acl_lock);
1134
1135 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1136 B_TRUE);
1137
1138 /*
1139 * Update the file size (zp_size) if it has changed;
1140 * account for possible concurrent updates.
1141 */
1142 while ((end_size = zp->z_size) < uio->uio_loffset) {
1143 (void) atomic_cas_64(&zp->z_size, end_size,
1144 uio->uio_loffset);
1145 ASSERT(error == 0);
1146 }
1147 /*
1148 * If we are replaying and eof is non zero then force
1149 * the file size to the specified eof. Note, there's no
1150 * concurrency during replay.
1151 */
1152 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1153 zp->z_size = zfsvfs->z_replay_eof;
1154
1155 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1156
1157 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1158 dmu_tx_commit(tx);
1159
1160 if (error != 0)
1161 break;
1162 ASSERT(tx_bytes == nbytes);
1163 n -= nbytes;
1164
1165#ifdef sun
1166 if (!xuio && n > 0)
1167 uio_prefaultpages(MIN(n, max_blksz), uio);
1168#endif /* sun */
1169 }
1170
1171 zfs_range_unlock(rl);
1172
1173 /*
1174 * If we're in replay mode, or we made no progress, return error.
1175 * Otherwise, it's at least a partial write, so it's successful.
1176 */
1177 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1178 ZFS_EXIT(zfsvfs);
1179 return (error);
1180 }
1181
1182 if (ioflag & (FSYNC | FDSYNC) ||
1183 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1184 zil_commit(zilog, zp->z_id);
1185
1186 ZFS_EXIT(zfsvfs);
1187 return (0);
1188}
1189
1190void
1191zfs_get_done(zgd_t *zgd, int error)
1192{
1193 znode_t *zp = zgd->zgd_private;
1194 objset_t *os = zp->z_zfsvfs->z_os;
1195
1196 if (zgd->zgd_db)
1197 dmu_buf_rele(zgd->zgd_db, zgd);
1198
1199 zfs_range_unlock(zgd->zgd_rl);
1200
1201 /*
1202 * Release the vnode asynchronously as we currently have the
1203 * txg stopped from syncing.
1204 */
1205 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1206
1207 if (error == 0 && zgd->zgd_bp)
1208 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1209
1210 kmem_free(zgd, sizeof (zgd_t));
1211}
1212
1213#ifdef DEBUG
1214static int zil_fault_io = 0;
1215#endif
1216
1217/*
1218 * Get data to generate a TX_WRITE intent log record.
1219 */
1220int
1221zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1222{
1223 zfsvfs_t *zfsvfs = arg;
1224 objset_t *os = zfsvfs->z_os;
1225 znode_t *zp;
1226 uint64_t object = lr->lr_foid;
1227 uint64_t offset = lr->lr_offset;
1228 uint64_t size = lr->lr_length;
1229 blkptr_t *bp = &lr->lr_blkptr;
1230 dmu_buf_t *db;
1231 zgd_t *zgd;
1232 int error = 0;
1233
1234 ASSERT(zio != NULL);
1235 ASSERT(size != 0);
1236
1237 /*
1238 * Nothing to do if the file has been removed
1239 */
1240 if (zfs_zget(zfsvfs, object, &zp) != 0)
1241 return (SET_ERROR(ENOENT));
1242 if (zp->z_unlinked) {
1243 /*
1244 * Release the vnode asynchronously as we currently have the
1245 * txg stopped from syncing.
1246 */
1247 VN_RELE_ASYNC(ZTOV(zp),
1248 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1249 return (SET_ERROR(ENOENT));
1250 }
1251
1252 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1253 zgd->zgd_zilog = zfsvfs->z_log;
1254 zgd->zgd_private = zp;
1255
1256 /*
1257 * Write records come in two flavors: immediate and indirect.
1258 * For small writes it's cheaper to store the data with the
1259 * log record (immediate); for large writes it's cheaper to
1260 * sync the data and get a pointer to it (indirect) so that
1261 * we don't have to write the data twice.
1262 */
1263 if (buf != NULL) { /* immediate write */
1264 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1265 /* test for truncation needs to be done while range locked */
1266 if (offset >= zp->z_size) {
1267 error = SET_ERROR(ENOENT);
1268 } else {
1269 error = dmu_read(os, object, offset, size, buf,
1270 DMU_READ_NO_PREFETCH);
1271 }
1272 ASSERT(error == 0 || error == ENOENT);
1273 } else { /* indirect write */
1274 /*
1275 * Have to lock the whole block to ensure when it's
1276 * written out and it's checksum is being calculated
1277 * that no one can change the data. We need to re-check
1278 * blocksize after we get the lock in case it's changed!
1279 */
1280 for (;;) {
1281 uint64_t blkoff;
1282 size = zp->z_blksz;
1283 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1284 offset -= blkoff;
1285 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1286 RL_READER);
1287 if (zp->z_blksz == size)
1288 break;
1289 offset += blkoff;
1290 zfs_range_unlock(zgd->zgd_rl);
1291 }
1292 /* test for truncation needs to be done while range locked */
1293 if (lr->lr_offset >= zp->z_size)
1294 error = SET_ERROR(ENOENT);
1295#ifdef DEBUG
1296 if (zil_fault_io) {
1297 error = SET_ERROR(EIO);
1298 zil_fault_io = 0;
1299 }
1300#endif
1301 if (error == 0)
1302 error = dmu_buf_hold(os, object, offset, zgd, &db,
1303 DMU_READ_NO_PREFETCH);
1304
1305 if (error == 0) {
1306 blkptr_t *obp = dmu_buf_get_blkptr(db);
1307 if (obp) {
1308 ASSERT(BP_IS_HOLE(bp));
1309 *bp = *obp;
1310 }
1311
1312 zgd->zgd_db = db;
1313 zgd->zgd_bp = bp;
1314
1315 ASSERT(db->db_offset == offset);
1316 ASSERT(db->db_size == size);
1317
1318 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1319 zfs_get_done, zgd);
1320 ASSERT(error || lr->lr_length <= zp->z_blksz);
1321
1322 /*
1323 * On success, we need to wait for the write I/O
1324 * initiated by dmu_sync() to complete before we can
1325 * release this dbuf. We will finish everything up
1326 * in the zfs_get_done() callback.
1327 */
1328 if (error == 0)
1329 return (0);
1330
1331 if (error == EALREADY) {
1332 lr->lr_common.lrc_txtype = TX_WRITE2;
1333 error = 0;
1334 }
1335 }
1336 }
1337
1338 zfs_get_done(zgd, error);
1339
1340 return (error);
1341}
1342
1343/*ARGSUSED*/
1344static int
1345zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1346 caller_context_t *ct)
1347{
1348 znode_t *zp = VTOZ(vp);
1349 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1350 int error;
1351
1352 ZFS_ENTER(zfsvfs);
1353 ZFS_VERIFY_ZP(zp);
1354
1355 if (flag & V_ACE_MASK)
1356 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1357 else
1358 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1359
1360 ZFS_EXIT(zfsvfs);
1361 return (error);
1362}
1363
1364/*
1365 * If vnode is for a device return a specfs vnode instead.
1366 */
1367static int
1368specvp_check(vnode_t **vpp, cred_t *cr)
1369{
1370 int error = 0;
1371
1372 if (IS_DEVVP(*vpp)) {
1373 struct vnode *svp;
1374
1375 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1376 VN_RELE(*vpp);
1377 if (svp == NULL)
1378 error = SET_ERROR(ENOSYS);
1379 *vpp = svp;
1380 }
1381 return (error);
1382}
1383
1384
1385/*
1386 * Lookup an entry in a directory, or an extended attribute directory.
1387 * If it exists, return a held vnode reference for it.
1388 *
1389 * IN: dvp - vnode of directory to search.
1390 * nm - name of entry to lookup.
1391 * pnp - full pathname to lookup [UNUSED].
1392 * flags - LOOKUP_XATTR set if looking for an attribute.
1393 * rdir - root directory vnode [UNUSED].
1394 * cr - credentials of caller.
1395 * ct - caller context
1396 * direntflags - directory lookup flags
1397 * realpnp - returned pathname.
1398 *
1399 * OUT: vpp - vnode of located entry, NULL if not found.
1400 *
1401 * RETURN: 0 on success, error code on failure.
1402 *
1403 * Timestamps:
1404 * NA
1405 */
1406/* ARGSUSED */
1407static int
1408zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1409 int nameiop, cred_t *cr, kthread_t *td, int flags)
1410{
1411 znode_t *zdp = VTOZ(dvp);
1412 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1413 int error = 0;
1414 int *direntflags = NULL;
1415 void *realpnp = NULL;
1416
1417 /* fast path */
1418 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1419
1420 if (dvp->v_type != VDIR) {
1421 return (SET_ERROR(ENOTDIR));
1422 } else if (zdp->z_sa_hdl == NULL) {
1423 return (SET_ERROR(EIO));
1424 }
1425
1426 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1427 error = zfs_fastaccesschk_execute(zdp, cr);
1428 if (!error) {
1429 *vpp = dvp;
1430 VN_HOLD(*vpp);
1431 return (0);
1432 }
1433 return (error);
1434 } else {
1435 vnode_t *tvp = dnlc_lookup(dvp, nm);
1436
1437 if (tvp) {
1438 error = zfs_fastaccesschk_execute(zdp, cr);
1439 if (error) {
1440 VN_RELE(tvp);
1441 return (error);
1442 }
1443 if (tvp == DNLC_NO_VNODE) {
1444 VN_RELE(tvp);
1445 return (SET_ERROR(ENOENT));
1446 } else {
1447 *vpp = tvp;
1448 return (specvp_check(vpp, cr));
1449 }
1450 }
1451 }
1452 }
1453
1454 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1455
1456 ZFS_ENTER(zfsvfs);
1457 ZFS_VERIFY_ZP(zdp);
1458
1459 *vpp = NULL;
1460
1461 if (flags & LOOKUP_XATTR) {
1462#ifdef TODO
1463 /*
1464 * If the xattr property is off, refuse the lookup request.
1465 */
1466 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1467 ZFS_EXIT(zfsvfs);
1468 return (SET_ERROR(EINVAL));
1469 }
1470#endif
1471
1472 /*
1473 * We don't allow recursive attributes..
1474 * Maybe someday we will.
1475 */
1476 if (zdp->z_pflags & ZFS_XATTR) {
1477 ZFS_EXIT(zfsvfs);
1478 return (SET_ERROR(EINVAL));
1479 }
1480
1481 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1482 ZFS_EXIT(zfsvfs);
1483 return (error);
1484 }
1485
1486 /*
1487 * Do we have permission to get into attribute directory?
1488 */
1489
1490 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1491 B_FALSE, cr)) {
1492 VN_RELE(*vpp);
1493 *vpp = NULL;
1494 }
1495
1496 ZFS_EXIT(zfsvfs);
1497 return (error);
1498 }
1499
1500 if (dvp->v_type != VDIR) {
1501 ZFS_EXIT(zfsvfs);
1502 return (SET_ERROR(ENOTDIR));
1503 }
1504
1505 /*
1506 * Check accessibility of directory.
1507 */
1508
1509 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1510 ZFS_EXIT(zfsvfs);
1511 return (error);
1512 }
1513
1514 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1515 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1516 ZFS_EXIT(zfsvfs);
1517 return (SET_ERROR(EILSEQ));
1518 }
1519
1520 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1521 if (error == 0)
1522 error = specvp_check(vpp, cr);
1523
1524 /* Translate errors and add SAVENAME when needed. */
1525 if (cnp->cn_flags & ISLASTCN) {
1526 switch (nameiop) {
1527 case CREATE:
1528 case RENAME:
1529 if (error == ENOENT) {
1530 error = EJUSTRETURN;
1531 cnp->cn_flags |= SAVENAME;
1532 break;
1533 }
1534 /* FALLTHROUGH */
1535 case DELETE:
1536 if (error == 0)
1537 cnp->cn_flags |= SAVENAME;
1538 break;
1539 }
1540 }
1541 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1542 int ltype = 0;
1543
1544 if (cnp->cn_flags & ISDOTDOT) {
1545 ltype = VOP_ISLOCKED(dvp);
1546 VOP_UNLOCK(dvp, 0);
1547 }
1548 ZFS_EXIT(zfsvfs);
1549 error = vn_lock(*vpp, cnp->cn_lkflags);
1550 if (cnp->cn_flags & ISDOTDOT)
1551 vn_lock(dvp, ltype | LK_RETRY);
1552 if (error != 0) {
1553 VN_RELE(*vpp);
1554 *vpp = NULL;
1555 return (error);
1556 }
1557 } else {
1558 ZFS_EXIT(zfsvfs);
1559 }
1560
1561#ifdef FREEBSD_NAMECACHE
1562 /*
1563 * Insert name into cache (as non-existent) if appropriate.
1564 */
1565 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1566 cache_enter(dvp, *vpp, cnp);
1567 /*
1568 * Insert name into cache if appropriate.
1569 */
1570 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1571 if (!(cnp->cn_flags & ISLASTCN) ||
1572 (nameiop != DELETE && nameiop != RENAME)) {
1573 cache_enter(dvp, *vpp, cnp);
1574 }
1575 }
1576#endif
1577
1578 return (error);
1579}
1580
1581/*
1582 * Attempt to create a new entry in a directory. If the entry
1583 * already exists, truncate the file if permissible, else return
1584 * an error. Return the vp of the created or trunc'd file.
1585 *
1586 * IN: dvp - vnode of directory to put new file entry in.
1587 * name - name of new file entry.
1588 * vap - attributes of new file.
1589 * excl - flag indicating exclusive or non-exclusive mode.
1590 * mode - mode to open file with.
1591 * cr - credentials of caller.
1592 * flag - large file flag [UNUSED].
1593 * ct - caller context
1594 * vsecp - ACL to be set
1595 *
1596 * OUT: vpp - vnode of created or trunc'd entry.
1597 *
1598 * RETURN: 0 on success, error code on failure.
1599 *
1600 * Timestamps:
1601 * dvp - ctime|mtime updated if new entry created
1602 * vp - ctime|mtime always, atime if new
1603 */
1604
1605/* ARGSUSED */
1606static int
1607zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1608 vnode_t **vpp, cred_t *cr, kthread_t *td)
1609{
1610 znode_t *zp, *dzp = VTOZ(dvp);
1611 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1612 zilog_t *zilog;
1613 objset_t *os;
1614 zfs_dirlock_t *dl;
1615 dmu_tx_t *tx;
1616 int error;
1617 ksid_t *ksid;
1618 uid_t uid;
1619 gid_t gid = crgetgid(cr);
1620 zfs_acl_ids_t acl_ids;
1621 boolean_t fuid_dirtied;
1622 boolean_t have_acl = B_FALSE;
1623 boolean_t waited = B_FALSE;
1624 void *vsecp = NULL;
1625 int flag = 0;
1626
1627 /*
1628 * If we have an ephemeral id, ACL, or XVATTR then
1629 * make sure file system is at proper version
1630 */
1631
1632 ksid = crgetsid(cr, KSID_OWNER);
1633 if (ksid)
1634 uid = ksid_getid(ksid);
1635 else
1636 uid = crgetuid(cr);
1637
1638 if (zfsvfs->z_use_fuids == B_FALSE &&
1639 (vsecp || (vap->va_mask & AT_XVATTR) ||
1640 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1641 return (SET_ERROR(EINVAL));
1642
1643 ZFS_ENTER(zfsvfs);
1644 ZFS_VERIFY_ZP(dzp);
1645 os = zfsvfs->z_os;
1646 zilog = zfsvfs->z_log;
1647
1648 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1649 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1650 ZFS_EXIT(zfsvfs);
1651 return (SET_ERROR(EILSEQ));
1652 }
1653
1654 if (vap->va_mask & AT_XVATTR) {
1655 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1656 crgetuid(cr), cr, vap->va_type)) != 0) {
1657 ZFS_EXIT(zfsvfs);
1658 return (error);
1659 }
1660 }
1661top:
1662 *vpp = NULL;
1663
1664 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1665 vap->va_mode &= ~S_ISVTX;
1666
1667 if (*name == '\0') {
1668 /*
1669 * Null component name refers to the directory itself.
1670 */
1671 VN_HOLD(dvp);
1672 zp = dzp;
1673 dl = NULL;
1674 error = 0;
1675 } else {
1676 /* possible VN_HOLD(zp) */
1677 int zflg = 0;
1678
1679 if (flag & FIGNORECASE)
1680 zflg |= ZCILOOK;
1681
1682 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1683 NULL, NULL);
1684 if (error) {
1685 if (have_acl)
1686 zfs_acl_ids_free(&acl_ids);
1687 if (strcmp(name, "..") == 0)
1688 error = SET_ERROR(EISDIR);
1689 ZFS_EXIT(zfsvfs);
1690 return (error);
1691 }
1692 }
1693
1694 if (zp == NULL) {
1695 uint64_t txtype;
1696
1697 /*
1698 * Create a new file object and update the directory
1699 * to reference it.
1700 */
1701 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1702 if (have_acl)
1703 zfs_acl_ids_free(&acl_ids);
1704 goto out;
1705 }
1706
1707 /*
1708 * We only support the creation of regular files in
1709 * extended attribute directories.
1710 */
1711
1712 if ((dzp->z_pflags & ZFS_XATTR) &&
1713 (vap->va_type != VREG)) {
1714 if (have_acl)
1715 zfs_acl_ids_free(&acl_ids);
1716 error = SET_ERROR(EINVAL);
1717 goto out;
1718 }
1719
1720 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1721 cr, vsecp, &acl_ids)) != 0)
1722 goto out;
1723 have_acl = B_TRUE;
1724
1725 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1726 zfs_acl_ids_free(&acl_ids);
1727 error = SET_ERROR(EDQUOT);
1728 goto out;
1729 }
1730
1731 tx = dmu_tx_create(os);
1732
1733 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1734 ZFS_SA_BASE_ATTR_SIZE);
1735
1736 fuid_dirtied = zfsvfs->z_fuid_dirty;
1737 if (fuid_dirtied)
1738 zfs_fuid_txhold(zfsvfs, tx);
1739 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1740 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1741 if (!zfsvfs->z_use_sa &&
1742 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1743 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1744 0, acl_ids.z_aclp->z_acl_bytes);
1745 }
1746 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1747 if (error) {
1748 zfs_dirent_unlock(dl);
1749 if (error == ERESTART) {
1750 waited = B_TRUE;
1751 dmu_tx_wait(tx);
1752 dmu_tx_abort(tx);
1753 goto top;
1754 }
1755 zfs_acl_ids_free(&acl_ids);
1756 dmu_tx_abort(tx);
1757 ZFS_EXIT(zfsvfs);
1758 return (error);
1759 }
1760 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1761
1762 if (fuid_dirtied)
1763 zfs_fuid_sync(zfsvfs, tx);
1764
1765 (void) zfs_link_create(dl, zp, tx, ZNEW);
1766 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1767 if (flag & FIGNORECASE)
1768 txtype |= TX_CI;
1769 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1770 vsecp, acl_ids.z_fuidp, vap);
1771 zfs_acl_ids_free(&acl_ids);
1772 dmu_tx_commit(tx);
1773 } else {
1774 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1775
1776 if (have_acl)
1777 zfs_acl_ids_free(&acl_ids);
1778 have_acl = B_FALSE;
1779
1780 /*
1781 * A directory entry already exists for this name.
1782 */
1783 /*
1784 * Can't truncate an existing file if in exclusive mode.
1785 */
1786 if (excl == EXCL) {
1787 error = SET_ERROR(EEXIST);
1788 goto out;
1789 }
1790 /*
1791 * Can't open a directory for writing.
1792 */
1793 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1794 error = SET_ERROR(EISDIR);
1795 goto out;
1796 }
1797 /*
1798 * Verify requested access to file.
1799 */
1800 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1801 goto out;
1802 }
1803
1804 mutex_enter(&dzp->z_lock);
1805 dzp->z_seq++;
1806 mutex_exit(&dzp->z_lock);
1807
1808 /*
1809 * Truncate regular files if requested.
1810 */
1811 if ((ZTOV(zp)->v_type == VREG) &&
1812 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1813 /* we can't hold any locks when calling zfs_freesp() */
1814 zfs_dirent_unlock(dl);
1815 dl = NULL;
1816 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1817 if (error == 0) {
1818 vnevent_create(ZTOV(zp), ct);
1819 }
1820 }
1821 }
1822out:
1823 if (dl)
1824 zfs_dirent_unlock(dl);
1825
1826 if (error) {
1827 if (zp)
1828 VN_RELE(ZTOV(zp));
1829 } else {
1830 *vpp = ZTOV(zp);
1831 error = specvp_check(vpp, cr);
1832 }
1833
1834 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1835 zil_commit(zilog, 0);
1836
1837 ZFS_EXIT(zfsvfs);
1838 return (error);
1839}
1840
1841/*
1842 * Remove an entry from a directory.
1843 *
1844 * IN: dvp - vnode of directory to remove entry from.
1845 * name - name of entry to remove.
1846 * cr - credentials of caller.
1847 * ct - caller context
1848 * flags - case flags
1849 *
1850 * RETURN: 0 on success, error code on failure.
1851 *
1852 * Timestamps:
1853 * dvp - ctime|mtime
1854 * vp - ctime (if nlink > 0)
1855 */
1856
1857uint64_t null_xattr = 0;
1858
1859/*ARGSUSED*/
1860static int
1861zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1862 int flags)
1863{
1864 znode_t *zp, *dzp = VTOZ(dvp);
1865 znode_t *xzp;
1866 vnode_t *vp;
1867 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1868 zilog_t *zilog;
1869 uint64_t acl_obj, xattr_obj;
1870 uint64_t xattr_obj_unlinked = 0;
1871 uint64_t obj = 0;
1872 zfs_dirlock_t *dl;
1873 dmu_tx_t *tx;
1874 boolean_t may_delete_now, delete_now = FALSE;
1875 boolean_t unlinked, toobig = FALSE;
1876 uint64_t txtype;
1877 pathname_t *realnmp = NULL;
1878 pathname_t realnm;
1879 int error;
1880 int zflg = ZEXISTS;
1881 boolean_t waited = B_FALSE;
1882
1883 ZFS_ENTER(zfsvfs);
1884 ZFS_VERIFY_ZP(dzp);
1885 zilog = zfsvfs->z_log;
1886
1887 if (flags & FIGNORECASE) {
1888 zflg |= ZCILOOK;
1889 pn_alloc(&realnm);
1890 realnmp = &realnm;
1891 }
1892
1893top:
1894 xattr_obj = 0;
1895 xzp = NULL;
1896 /*
1897 * Attempt to lock directory; fail if entry doesn't exist.
1898 */
1899 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1900 NULL, realnmp)) {
1901 if (realnmp)
1902 pn_free(realnmp);
1903 ZFS_EXIT(zfsvfs);
1904 return (error);
1905 }
1906
1907 vp = ZTOV(zp);
1908
1909 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1910 goto out;
1911 }
1912
1913 /*
1914 * Need to use rmdir for removing directories.
1915 */
1916 if (vp->v_type == VDIR) {
1917 error = SET_ERROR(EPERM);
1918 goto out;
1919 }
1920
1921 vnevent_remove(vp, dvp, name, ct);
1922
1923 if (realnmp)
1924 dnlc_remove(dvp, realnmp->pn_buf);
1925 else
1926 dnlc_remove(dvp, name);
1927
1928 VI_LOCK(vp);
1929 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1930 VI_UNLOCK(vp);
1931
1932 /*
1933 * We may delete the znode now, or we may put it in the unlinked set;
1934 * it depends on whether we're the last link, and on whether there are
1935 * other holds on the vnode. So we dmu_tx_hold() the right things to
1936 * allow for either case.
1937 */
1938 obj = zp->z_id;
1939 tx = dmu_tx_create(zfsvfs->z_os);
1940 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1941 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1942 zfs_sa_upgrade_txholds(tx, zp);
1943 zfs_sa_upgrade_txholds(tx, dzp);
1944 if (may_delete_now) {
1945 toobig =
1946 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1947 /* if the file is too big, only hold_free a token amount */
1948 dmu_tx_hold_free(tx, zp->z_id, 0,
1949 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1950 }
1951
1952 /* are there any extended attributes? */
1953 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1954 &xattr_obj, sizeof (xattr_obj));
1955 if (error == 0 && xattr_obj) {
1956 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1957 ASSERT0(error);
1958 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1959 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1960 }
1961
1962 mutex_enter(&zp->z_lock);
1963 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1964 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1965 mutex_exit(&zp->z_lock);
1966
1967 /* charge as an update -- would be nice not to charge at all */
1968 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1969
1970 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1971 if (error) {
1972 zfs_dirent_unlock(dl);
1973 VN_RELE(vp);
1974 if (xzp)
1975 VN_RELE(ZTOV(xzp));
1976 if (error == ERESTART) {
1977 waited = B_TRUE;
1978 dmu_tx_wait(tx);
1979 dmu_tx_abort(tx);
1980 goto top;
1981 }
1982 if (realnmp)
1983 pn_free(realnmp);
1984 dmu_tx_abort(tx);
1985 ZFS_EXIT(zfsvfs);
1986 return (error);
1987 }
1988
1989 /*
1990 * Remove the directory entry.
1991 */
1992 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1993
1994 if (error) {
1995 dmu_tx_commit(tx);
1996 goto out;
1997 }
1998
1999 if (unlinked) {
2000
2001 /*
2002 * Hold z_lock so that we can make sure that the ACL obj
2003 * hasn't changed. Could have been deleted due to
2004 * zfs_sa_upgrade().
2005 */
2006 mutex_enter(&zp->z_lock);
2007 VI_LOCK(vp);
2008 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2009 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2010 delete_now = may_delete_now && !toobig &&
2011 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2012 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2013 acl_obj;
2014 VI_UNLOCK(vp);
2015 }
2016
2017 if (delete_now) {
2018#ifdef __FreeBSD__
2019 panic("zfs_remove: delete_now branch taken");
2020#endif
2021 if (xattr_obj_unlinked) {
2022 ASSERT3U(xzp->z_links, ==, 2);
2023 mutex_enter(&xzp->z_lock);
2024 xzp->z_unlinked = 1;
2025 xzp->z_links = 0;
2026 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2027 &xzp->z_links, sizeof (xzp->z_links), tx);
2028 ASSERT3U(error, ==, 0);
2029 mutex_exit(&xzp->z_lock);
2030 zfs_unlinked_add(xzp, tx);
2031
2032 if (zp->z_is_sa)
2033 error = sa_remove(zp->z_sa_hdl,
2034 SA_ZPL_XATTR(zfsvfs), tx);
2035 else
2036 error = sa_update(zp->z_sa_hdl,
2037 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2038 sizeof (uint64_t), tx);
2039 ASSERT0(error);
2040 }
2041 VI_LOCK(vp);
2042 vp->v_count--;
2043 ASSERT0(vp->v_count);
2044 VI_UNLOCK(vp);
2045 mutex_exit(&zp->z_lock);
2046 zfs_znode_delete(zp, tx);
2047 } else if (unlinked) {
2048 mutex_exit(&zp->z_lock);
2049 zfs_unlinked_add(zp, tx);
2050#ifdef __FreeBSD__
2051 vp->v_vflag |= VV_NOSYNC;
2052#endif
2053 }
2054
2055 txtype = TX_REMOVE;
2056 if (flags & FIGNORECASE)
2057 txtype |= TX_CI;
2058 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2059
2060 dmu_tx_commit(tx);
2061out:
2062 if (realnmp)
2063 pn_free(realnmp);
2064
2065 zfs_dirent_unlock(dl);
2066
2067 if (!delete_now)
2068 VN_RELE(vp);
2069 if (xzp)
2070 VN_RELE(ZTOV(xzp));
2071
2072 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2073 zil_commit(zilog, 0);
2074
2075 ZFS_EXIT(zfsvfs);
2076 return (error);
2077}
2078
2079/*
2080 * Create a new directory and insert it into dvp using the name
2081 * provided. Return a pointer to the inserted directory.
2082 *
2083 * IN: dvp - vnode of directory to add subdir to.
2084 * dirname - name of new directory.
2085 * vap - attributes of new directory.
2086 * cr - credentials of caller.
2087 * ct - caller context
2088 * flags - case flags
2089 * vsecp - ACL to be set
2090 *
2091 * OUT: vpp - vnode of created directory.
2092 *
2093 * RETURN: 0 on success, error code on failure.
2094 *
2095 * Timestamps:
2096 * dvp - ctime|mtime updated
2097 * vp - ctime|mtime|atime updated
2098 */
2099/*ARGSUSED*/
2100static int
2101zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2102 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2103{
2104 znode_t *zp, *dzp = VTOZ(dvp);
2105 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2106 zilog_t *zilog;
2107 zfs_dirlock_t *dl;
2108 uint64_t txtype;
2109 dmu_tx_t *tx;
2110 int error;
2111 int zf = ZNEW;
2112 ksid_t *ksid;
2113 uid_t uid;
2114 gid_t gid = crgetgid(cr);
2115 zfs_acl_ids_t acl_ids;
2116 boolean_t fuid_dirtied;
2117 boolean_t waited = B_FALSE;
2118
2119 ASSERT(vap->va_type == VDIR);
2120
2121 /*
2122 * If we have an ephemeral id, ACL, or XVATTR then
2123 * make sure file system is at proper version
2124 */
2125
2126 ksid = crgetsid(cr, KSID_OWNER);
2127 if (ksid)
2128 uid = ksid_getid(ksid);
2129 else
2130 uid = crgetuid(cr);
2131 if (zfsvfs->z_use_fuids == B_FALSE &&
2132 (vsecp || (vap->va_mask & AT_XVATTR) ||
2133 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2134 return (SET_ERROR(EINVAL));
2135
2136 ZFS_ENTER(zfsvfs);
2137 ZFS_VERIFY_ZP(dzp);
2138 zilog = zfsvfs->z_log;
2139
2140 if (dzp->z_pflags & ZFS_XATTR) {
2141 ZFS_EXIT(zfsvfs);
2142 return (SET_ERROR(EINVAL));
2143 }
2144
2145 if (zfsvfs->z_utf8 && u8_validate(dirname,
2146 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2147 ZFS_EXIT(zfsvfs);
2148 return (SET_ERROR(EILSEQ));
2149 }
2150 if (flags & FIGNORECASE)
2151 zf |= ZCILOOK;
2152
2153 if (vap->va_mask & AT_XVATTR) {
2154 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2155 crgetuid(cr), cr, vap->va_type)) != 0) {
2156 ZFS_EXIT(zfsvfs);
2157 return (error);
2158 }
2159 }
2160
2161 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2162 vsecp, &acl_ids)) != 0) {
2163 ZFS_EXIT(zfsvfs);
2164 return (error);
2165 }
2166 /*
2167 * First make sure the new directory doesn't exist.
2168 *
2169 * Existence is checked first to make sure we don't return
2170 * EACCES instead of EEXIST which can cause some applications
2171 * to fail.
2172 */
2173top:
2174 *vpp = NULL;
2175
2176 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2177 NULL, NULL)) {
2178 zfs_acl_ids_free(&acl_ids);
2179 ZFS_EXIT(zfsvfs);
2180 return (error);
2181 }
2182
2183 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2184 zfs_acl_ids_free(&acl_ids);
2185 zfs_dirent_unlock(dl);
2186 ZFS_EXIT(zfsvfs);
2187 return (error);
2188 }
2189
2190 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2191 zfs_acl_ids_free(&acl_ids);
2192 zfs_dirent_unlock(dl);
2193 ZFS_EXIT(zfsvfs);
2194 return (SET_ERROR(EDQUOT));
2195 }
2196
2197 /*
2198 * Add a new entry to the directory.
2199 */
2200 tx = dmu_tx_create(zfsvfs->z_os);
2201 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2202 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2203 fuid_dirtied = zfsvfs->z_fuid_dirty;
2204 if (fuid_dirtied)
2205 zfs_fuid_txhold(zfsvfs, tx);
2206 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2207 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2208 acl_ids.z_aclp->z_acl_bytes);
2209 }
2210
2211 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2212 ZFS_SA_BASE_ATTR_SIZE);
2213
2214 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2215 if (error) {
2216 zfs_dirent_unlock(dl);
2217 if (error == ERESTART) {
2218 waited = B_TRUE;
2219 dmu_tx_wait(tx);
2220 dmu_tx_abort(tx);
2221 goto top;
2222 }
2223 zfs_acl_ids_free(&acl_ids);
2224 dmu_tx_abort(tx);
2225 ZFS_EXIT(zfsvfs);
2226 return (error);
2227 }
2228
2229 /*
2230 * Create new node.
2231 */
2232 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2233
2234 if (fuid_dirtied)
2235 zfs_fuid_sync(zfsvfs, tx);
2236
2237 /*
2238 * Now put new name in parent dir.
2239 */
2240 (void) zfs_link_create(dl, zp, tx, ZNEW);
2241
2242 *vpp = ZTOV(zp);
2243
2244 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2245 if (flags & FIGNORECASE)
2246 txtype |= TX_CI;
2247 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2248 acl_ids.z_fuidp, vap);
2249
2250 zfs_acl_ids_free(&acl_ids);
2251
2252 dmu_tx_commit(tx);
2253
2254 zfs_dirent_unlock(dl);
2255
2256 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2257 zil_commit(zilog, 0);
2258
2259 ZFS_EXIT(zfsvfs);
2260 return (0);
2261}
2262
2263/*
2264 * Remove a directory subdir entry. If the current working
2265 * directory is the same as the subdir to be removed, the
2266 * remove will fail.
2267 *
2268 * IN: dvp - vnode of directory to remove from.
2269 * name - name of directory to be removed.
2270 * cwd - vnode of current working directory.
2271 * cr - credentials of caller.
2272 * ct - caller context
2273 * flags - case flags
2274 *
2275 * RETURN: 0 on success, error code on failure.
2276 *
2277 * Timestamps:
2278 * dvp - ctime|mtime updated
2279 */
2280/*ARGSUSED*/
2281static int
2282zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2283 caller_context_t *ct, int flags)
2284{
2285 znode_t *dzp = VTOZ(dvp);
2286 znode_t *zp;
2287 vnode_t *vp;
2288 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2289 zilog_t *zilog;
2290 zfs_dirlock_t *dl;
2291 dmu_tx_t *tx;
2292 int error;
2293 int zflg = ZEXISTS;
2294 boolean_t waited = B_FALSE;
2295
2296 ZFS_ENTER(zfsvfs);
2297 ZFS_VERIFY_ZP(dzp);
2298 zilog = zfsvfs->z_log;
2299
2300 if (flags & FIGNORECASE)
2301 zflg |= ZCILOOK;
2302top:
2303 zp = NULL;
2304
2305 /*
2306 * Attempt to lock directory; fail if entry doesn't exist.
2307 */
2308 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2309 NULL, NULL)) {
2310 ZFS_EXIT(zfsvfs);
2311 return (error);
2312 }
2313
2314 vp = ZTOV(zp);
2315
2316 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2317 goto out;
2318 }
2319
2320 if (vp->v_type != VDIR) {
2321 error = SET_ERROR(ENOTDIR);
2322 goto out;
2323 }
2324
2325 if (vp == cwd) {
2326 error = SET_ERROR(EINVAL);
2327 goto out;
2328 }
2329
2330 vnevent_rmdir(vp, dvp, name, ct);
2331
2332 /*
2333 * Grab a lock on the directory to make sure that noone is
2334 * trying to add (or lookup) entries while we are removing it.
2335 */
2336 rw_enter(&zp->z_name_lock, RW_WRITER);
2337
2338 /*
2339 * Grab a lock on the parent pointer to make sure we play well
2340 * with the treewalk and directory rename code.
2341 */
2342 rw_enter(&zp->z_parent_lock, RW_WRITER);
2343
2344 tx = dmu_tx_create(zfsvfs->z_os);
2345 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2346 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2347 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2348 zfs_sa_upgrade_txholds(tx, zp);
2349 zfs_sa_upgrade_txholds(tx, dzp);
2350 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2351 if (error) {
2352 rw_exit(&zp->z_parent_lock);
2353 rw_exit(&zp->z_name_lock);
2354 zfs_dirent_unlock(dl);
2355 VN_RELE(vp);
2356 if (error == ERESTART) {
2357 waited = B_TRUE;
2358 dmu_tx_wait(tx);
2359 dmu_tx_abort(tx);
2360 goto top;
2361 }
2362 dmu_tx_abort(tx);
2363 ZFS_EXIT(zfsvfs);
2364 return (error);
2365 }
2366
2367#ifdef FREEBSD_NAMECACHE
2368 cache_purge(dvp);
2369#endif
2370
2371 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2372
2373 if (error == 0) {
2374 uint64_t txtype = TX_RMDIR;
2375 if (flags & FIGNORECASE)
2376 txtype |= TX_CI;
2377 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2378 }
2379
2380 dmu_tx_commit(tx);
2381
2382 rw_exit(&zp->z_parent_lock);
2383 rw_exit(&zp->z_name_lock);
2384#ifdef FREEBSD_NAMECACHE
2385 cache_purge(vp);
2386#endif
2387out:
2388 zfs_dirent_unlock(dl);
2389
2390 VN_RELE(vp);
2391
2392 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2393 zil_commit(zilog, 0);
2394
2395 ZFS_EXIT(zfsvfs);
2396 return (error);
2397}
2398
2399/*
2400 * Read as many directory entries as will fit into the provided
2401 * buffer from the given directory cursor position (specified in
2402 * the uio structure).
2403 *
2404 * IN: vp - vnode of directory to read.
2405 * uio - structure supplying read location, range info,
2406 * and return buffer.
2407 * cr - credentials of caller.
2408 * ct - caller context
2409 * flags - case flags
2410 *
2411 * OUT: uio - updated offset and range, buffer filled.
2412 * eofp - set to true if end-of-file detected.
2413 *
2414 * RETURN: 0 on success, error code on failure.
2415 *
2416 * Timestamps:
2417 * vp - atime updated
2418 *
2419 * Note that the low 4 bits of the cookie returned by zap is always zero.
2420 * This allows us to use the low range for "special" directory entries:
2421 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2422 * we use the offset 2 for the '.zfs' directory.
2423 */
2424/* ARGSUSED */
2425static int
2426zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2427{
2428 znode_t *zp = VTOZ(vp);
2429 iovec_t *iovp;
2430 edirent_t *eodp;
2431 dirent64_t *odp;
2432 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2433 objset_t *os;
2434 caddr_t outbuf;
2435 size_t bufsize;
2436 zap_cursor_t zc;
2437 zap_attribute_t zap;
2438 uint_t bytes_wanted;
2439 uint64_t offset; /* must be unsigned; checks for < 1 */
2440 uint64_t parent;
2441 int local_eof;
2442 int outcount;
2443 int error;
2444 uint8_t prefetch;
2445 boolean_t check_sysattrs;
2446 uint8_t type;
2447 int ncooks;
2448 u_long *cooks = NULL;
2449 int flags = 0;
2450
2451 ZFS_ENTER(zfsvfs);
2452 ZFS_VERIFY_ZP(zp);
2453
2454 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2455 &parent, sizeof (parent))) != 0) {
2456 ZFS_EXIT(zfsvfs);
2457 return (error);
2458 }
2459
2460 /*
2461 * If we are not given an eof variable,
2462 * use a local one.
2463 */
2464 if (eofp == NULL)
2465 eofp = &local_eof;
2466
2467 /*
2468 * Check for valid iov_len.
2469 */
2470 if (uio->uio_iov->iov_len <= 0) {
2471 ZFS_EXIT(zfsvfs);
2472 return (SET_ERROR(EINVAL));
2473 }
2474
2475 /*
2476 * Quit if directory has been removed (posix)
2477 */
2478 if ((*eofp = zp->z_unlinked) != 0) {
2479 ZFS_EXIT(zfsvfs);
2480 return (0);
2481 }
2482
2483 error = 0;
2484 os = zfsvfs->z_os;
2485 offset = uio->uio_loffset;
2486 prefetch = zp->z_zn_prefetch;
2487
2488 /*
2489 * Initialize the iterator cursor.
2490 */
2491 if (offset <= 3) {
2492 /*
2493 * Start iteration from the beginning of the directory.
2494 */
2495 zap_cursor_init(&zc, os, zp->z_id);
2496 } else {
2497 /*
2498 * The offset is a serialized cursor.
2499 */
2500 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2501 }
2502
2503 /*
2504 * Get space to change directory entries into fs independent format.
2505 */
2506 iovp = uio->uio_iov;
2507 bytes_wanted = iovp->iov_len;
2508 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2509 bufsize = bytes_wanted;
2510 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2511 odp = (struct dirent64 *)outbuf;
2512 } else {
2513 bufsize = bytes_wanted;
2514 outbuf = NULL;
2515 odp = (struct dirent64 *)iovp->iov_base;
2516 }
2517 eodp = (struct edirent *)odp;
2518
2519 if (ncookies != NULL) {
2520 /*
2521 * Minimum entry size is dirent size and 1 byte for a file name.
2522 */
2523 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2524 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2525 *cookies = cooks;
2526 *ncookies = ncooks;
2527 }
2528 /*
2529 * If this VFS supports the system attribute view interface; and
2530 * we're looking at an extended attribute directory; and we care
2531 * about normalization conflicts on this vfs; then we must check
2532 * for normalization conflicts with the sysattr name space.
2533 */
2534#ifdef TODO
2535 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2536 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2537 (flags & V_RDDIR_ENTFLAGS);
2538#else
2539 check_sysattrs = 0;
2540#endif
2541
2542 /*
2543 * Transform to file-system independent format
2544 */
2545 outcount = 0;
2546 while (outcount < bytes_wanted) {
2547 ino64_t objnum;
2548 ushort_t reclen;
2549 off64_t *next = NULL;
2550
2551 /*
2552 * Special case `.', `..', and `.zfs'.
2553 */
2554 if (offset == 0) {
2555 (void) strcpy(zap.za_name, ".");
2556 zap.za_normalization_conflict = 0;
2557 objnum = zp->z_id;
2558 type = DT_DIR;
2559 } else if (offset == 1) {
2560 (void) strcpy(zap.za_name, "..");
2561 zap.za_normalization_conflict = 0;
2562 objnum = parent;
2563 type = DT_DIR;
2564 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2565 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2566 zap.za_normalization_conflict = 0;
2567 objnum = ZFSCTL_INO_ROOT;
2568 type = DT_DIR;
2569 } else {
2570 /*
2571 * Grab next entry.
2572 */
2573 if (error = zap_cursor_retrieve(&zc, &zap)) {
2574 if ((*eofp = (error == ENOENT)) != 0)
2575 break;
2576 else
2577 goto update;
2578 }
2579
2580 if (zap.za_integer_length != 8 ||
2581 zap.za_num_integers != 1) {
2582 cmn_err(CE_WARN, "zap_readdir: bad directory "
2583 "entry, obj = %lld, offset = %lld\n",
2584 (u_longlong_t)zp->z_id,
2585 (u_longlong_t)offset);
2586 error = SET_ERROR(ENXIO);
2587 goto update;
2588 }
2589
2590 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2591 /*
2592 * MacOS X can extract the object type here such as:
2593 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2594 */
2595 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2596
2597 if (check_sysattrs && !zap.za_normalization_conflict) {
2598#ifdef TODO
2599 zap.za_normalization_conflict =
2600 xattr_sysattr_casechk(zap.za_name);
2601#else
2602 panic("%s:%u: TODO", __func__, __LINE__);
2603#endif
2604 }
2605 }
2606
2607 if (flags & V_RDDIR_ACCFILTER) {
2608 /*
2609 * If we have no access at all, don't include
2610 * this entry in the returned information
2611 */
2612 znode_t *ezp;
2613 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2614 goto skip_entry;
2615 if (!zfs_has_access(ezp, cr)) {
2616 VN_RELE(ZTOV(ezp));
2617 goto skip_entry;
2618 }
2619 VN_RELE(ZTOV(ezp));
2620 }
2621
2622 if (flags & V_RDDIR_ENTFLAGS)
2623 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2624 else
2625 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2626
2627 /*
2628 * Will this entry fit in the buffer?
2629 */
2630 if (outcount + reclen > bufsize) {
2631 /*
2632 * Did we manage to fit anything in the buffer?
2633 */
2634 if (!outcount) {
2635 error = SET_ERROR(EINVAL);
2636 goto update;
2637 }
2638 break;
2639 }
2640 if (flags & V_RDDIR_ENTFLAGS) {
2641 /*
2642 * Add extended flag entry:
2643 */
2644 eodp->ed_ino = objnum;
2645 eodp->ed_reclen = reclen;
2646 /* NOTE: ed_off is the offset for the *next* entry */
2647 next = &(eodp->ed_off);
2648 eodp->ed_eflags = zap.za_normalization_conflict ?
2649 ED_CASE_CONFLICT : 0;
2650 (void) strncpy(eodp->ed_name, zap.za_name,
2651 EDIRENT_NAMELEN(reclen));
2652 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2653 } else {
2654 /*
2655 * Add normal entry:
2656 */
2657 odp->d_ino = objnum;
2658 odp->d_reclen = reclen;
2659 odp->d_namlen = strlen(zap.za_name);
2660 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2661 odp->d_type = type;
2662 odp = (dirent64_t *)((intptr_t)odp + reclen);
2663 }
2664 outcount += reclen;
2665
2666 ASSERT(outcount <= bufsize);
2667
2668 /* Prefetch znode */
2669 if (prefetch)
2670 dmu_prefetch(os, objnum, 0, 0);
2671
2672 skip_entry:
2673 /*
2674 * Move to the next entry, fill in the previous offset.
2675 */
2676 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2677 zap_cursor_advance(&zc);
2678 offset = zap_cursor_serialize(&zc);
2679 } else {
2680 offset += 1;
2681 }
2682
2683 if (cooks != NULL) {
2684 *cooks++ = offset;
2685 ncooks--;
2686 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2687 }
2688 }
2689 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2690
2691 /* Subtract unused cookies */
2692 if (ncookies != NULL)
2693 *ncookies -= ncooks;
2694
2695 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2696 iovp->iov_base += outcount;
2697 iovp->iov_len -= outcount;
2698 uio->uio_resid -= outcount;
2699 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2700 /*
2701 * Reset the pointer.
2702 */
2703 offset = uio->uio_loffset;
2704 }
2705
2706update:
2707 zap_cursor_fini(&zc);
2708 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2709 kmem_free(outbuf, bufsize);
2710
2711 if (error == ENOENT)
2712 error = 0;
2713
2714 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2715
2716 uio->uio_loffset = offset;
2717 ZFS_EXIT(zfsvfs);
2718 if (error != 0 && cookies != NULL) {
2719 free(*cookies, M_TEMP);
2720 *cookies = NULL;
2721 *ncookies = 0;
2722 }
2723 return (error);
2724}
2725
2726ulong_t zfs_fsync_sync_cnt = 4;
2727
2728static int
2729zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2730{
2731 znode_t *zp = VTOZ(vp);
2732 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2733
2734 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2735
2736 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2737 ZFS_ENTER(zfsvfs);
2738 ZFS_VERIFY_ZP(zp);
2739 zil_commit(zfsvfs->z_log, zp->z_id);
2740 ZFS_EXIT(zfsvfs);
2741 }
2742 return (0);
2743}
2744
2745
2746/*
2747 * Get the requested file attributes and place them in the provided
2748 * vattr structure.
2749 *
2750 * IN: vp - vnode of file.
2751 * vap - va_mask identifies requested attributes.
2752 * If AT_XVATTR set, then optional attrs are requested
2753 * flags - ATTR_NOACLCHECK (CIFS server context)
2754 * cr - credentials of caller.
2755 * ct - caller context
2756 *
2757 * OUT: vap - attribute values.
2758 *
2759 * RETURN: 0 (always succeeds).
2760 */
2761/* ARGSUSED */
2762static int
2763zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2764 caller_context_t *ct)
2765{
2766 znode_t *zp = VTOZ(vp);
2767 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2768 int error = 0;
2769 uint32_t blksize;
2770 u_longlong_t nblocks;
2771 uint64_t links;
2772 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2773 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2774 xoptattr_t *xoap = NULL;
2775 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2776 sa_bulk_attr_t bulk[4];
2777 int count = 0;
2778
2779 ZFS_ENTER(zfsvfs);
2780 ZFS_VERIFY_ZP(zp);
2781
2782 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2783
2784 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2785 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2786 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2787 if (vp->v_type == VBLK || vp->v_type == VCHR)
2788 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2789 &rdev, 8);
2790
2791 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2792 ZFS_EXIT(zfsvfs);
2793 return (error);
2794 }
2795
2796 /*
2797 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2798 * Also, if we are the owner don't bother, since owner should
2799 * always be allowed to read basic attributes of file.
2800 */
2801 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2802 (vap->va_uid != crgetuid(cr))) {
2803 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2804 skipaclchk, cr)) {
2805 ZFS_EXIT(zfsvfs);
2806 return (error);
2807 }
2808 }
2809
2810 /*
2811 * Return all attributes. It's cheaper to provide the answer
2812 * than to determine whether we were asked the question.
2813 */
2814
2815 mutex_enter(&zp->z_lock);
2816 vap->va_type = IFTOVT(zp->z_mode);
2817 vap->va_mode = zp->z_mode & ~S_IFMT;
2818#ifdef sun
2819 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2820#else
2821 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2822#endif
2823 vap->va_nodeid = zp->z_id;
2824 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2825 links = zp->z_links + 1;
2826 else
2827 links = zp->z_links;
2828 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2829 vap->va_size = zp->z_size;
2830#ifdef sun
2831 vap->va_rdev = vp->v_rdev;
2832#else
2833 if (vp->v_type == VBLK || vp->v_type == VCHR)
2834 vap->va_rdev = zfs_cmpldev(rdev);
2835#endif
2836 vap->va_seq = zp->z_seq;
2837 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2838
2839 /*
2840 * Add in any requested optional attributes and the create time.
2841 * Also set the corresponding bits in the returned attribute bitmap.
2842 */
2843 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2844 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2845 xoap->xoa_archive =
2846 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2847 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2848 }
2849
2850 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2851 xoap->xoa_readonly =
2852 ((zp->z_pflags & ZFS_READONLY) != 0);
2853 XVA_SET_RTN(xvap, XAT_READONLY);
2854 }
2855
2856 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2857 xoap->xoa_system =
2858 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2859 XVA_SET_RTN(xvap, XAT_SYSTEM);
2860 }
2861
2862 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2863 xoap->xoa_hidden =
2864 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2865 XVA_SET_RTN(xvap, XAT_HIDDEN);
2866 }
2867
2868 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2869 xoap->xoa_nounlink =
2870 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2871 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2872 }
2873
2874 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2875 xoap->xoa_immutable =
2876 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2877 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2878 }
2879
2880 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2881 xoap->xoa_appendonly =
2882 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2883 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2884 }
2885
2886 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2887 xoap->xoa_nodump =
2888 ((zp->z_pflags & ZFS_NODUMP) != 0);
2889 XVA_SET_RTN(xvap, XAT_NODUMP);
2890 }
2891
2892 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2893 xoap->xoa_opaque =
2894 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2895 XVA_SET_RTN(xvap, XAT_OPAQUE);
2896 }
2897
2898 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2899 xoap->xoa_av_quarantined =
2900 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2901 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2902 }
2903
2904 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2905 xoap->xoa_av_modified =
2906 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2907 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2908 }
2909
2910 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2911 vp->v_type == VREG) {
2912 zfs_sa_get_scanstamp(zp, xvap);
2913 }
2914
2915 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2916 uint64_t times[2];
2917
2918 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2919 times, sizeof (times));
2920 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2921 XVA_SET_RTN(xvap, XAT_CREATETIME);
2922 }
2923
2924 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2925 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2926 XVA_SET_RTN(xvap, XAT_REPARSE);
2927 }
2928 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2929 xoap->xoa_generation = zp->z_gen;
2930 XVA_SET_RTN(xvap, XAT_GEN);
2931 }
2932
2933 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2934 xoap->xoa_offline =
2935 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2936 XVA_SET_RTN(xvap, XAT_OFFLINE);
2937 }
2938
2939 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2940 xoap->xoa_sparse =
2941 ((zp->z_pflags & ZFS_SPARSE) != 0);
2942 XVA_SET_RTN(xvap, XAT_SPARSE);
2943 }
2944 }
2945
2946 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2947 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2948 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2949 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2950
2951 mutex_exit(&zp->z_lock);
2952
2953 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2954 vap->va_blksize = blksize;
2955 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2956
2957 if (zp->z_blksz == 0) {
2958 /*
2959 * Block size hasn't been set; suggest maximal I/O transfers.
2960 */
2961 vap->va_blksize = zfsvfs->z_max_blksz;
2962 }
2963
2964 ZFS_EXIT(zfsvfs);
2965 return (0);
2966}
2967
2968/*
2969 * Set the file attributes to the values contained in the
2970 * vattr structure.
2971 *
2972 * IN: vp - vnode of file to be modified.
2973 * vap - new attribute values.
2974 * If AT_XVATTR set, then optional attrs are being set
2975 * flags - ATTR_UTIME set if non-default time values provided.
2976 * - ATTR_NOACLCHECK (CIFS context only).
2977 * cr - credentials of caller.
2978 * ct - caller context
2979 *
2980 * RETURN: 0 on success, error code on failure.
2981 *
2982 * Timestamps:
2983 * vp - ctime updated, mtime updated if size changed.
2984 */
2985/* ARGSUSED */
2986static int
2987zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2988 caller_context_t *ct)
2989{
2990 znode_t *zp = VTOZ(vp);
2991 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2992 zilog_t *zilog;
2993 dmu_tx_t *tx;
2994 vattr_t oldva;
2995 xvattr_t tmpxvattr;
2996 uint_t mask = vap->va_mask;
2997 uint_t saved_mask = 0;
2998 uint64_t saved_mode;
2999 int trim_mask = 0;
3000 uint64_t new_mode;
3001 uint64_t new_uid, new_gid;
3002 uint64_t xattr_obj;
3003 uint64_t mtime[2], ctime[2];
3004 znode_t *attrzp;
3005 int need_policy = FALSE;
3006 int err, err2;
3007 zfs_fuid_info_t *fuidp = NULL;
3008 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3009 xoptattr_t *xoap;
3010 zfs_acl_t *aclp;
3011 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3012 boolean_t fuid_dirtied = B_FALSE;
3013 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3014 int count = 0, xattr_count = 0;
3015
3016 if (mask == 0)
3017 return (0);
3018
3019 if (mask & AT_NOSET)
3020 return (SET_ERROR(EINVAL));
3021
3022 ZFS_ENTER(zfsvfs);
3023 ZFS_VERIFY_ZP(zp);
3024
3025 zilog = zfsvfs->z_log;
3026
3027 /*
3028 * Make sure that if we have ephemeral uid/gid or xvattr specified
3029 * that file system is at proper version level
3030 */
3031
3032 if (zfsvfs->z_use_fuids == B_FALSE &&
3033 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3034 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3035 (mask & AT_XVATTR))) {
3036 ZFS_EXIT(zfsvfs);
3037 return (SET_ERROR(EINVAL));
3038 }
3039
3040 if (mask & AT_SIZE && vp->v_type == VDIR) {
3041 ZFS_EXIT(zfsvfs);
3042 return (SET_ERROR(EISDIR));
3043 }
3044
3045 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3046 ZFS_EXIT(zfsvfs);
3047 return (SET_ERROR(EINVAL));
3048 }
3049
3050 /*
3051 * If this is an xvattr_t, then get a pointer to the structure of
3052 * optional attributes. If this is NULL, then we have a vattr_t.
3053 */
3054 xoap = xva_getxoptattr(xvap);
3055
3056 xva_init(&tmpxvattr);
3057
3058 /*
3059 * Immutable files can only alter immutable bit and atime
3060 */
3061 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3062 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3063 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3064 ZFS_EXIT(zfsvfs);
3065 return (SET_ERROR(EPERM));
3066 }
3067
3068 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3069 ZFS_EXIT(zfsvfs);
3070 return (SET_ERROR(EPERM));
3071 }
3072
3073 /*
3074 * Verify timestamps doesn't overflow 32 bits.
3075 * ZFS can handle large timestamps, but 32bit syscalls can't
3076 * handle times greater than 2039. This check should be removed
3077 * once large timestamps are fully supported.
3078 */
3079 if (mask & (AT_ATIME | AT_MTIME)) {
3080 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3081 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3082 ZFS_EXIT(zfsvfs);
3083 return (SET_ERROR(EOVERFLOW));
3084 }
3085 }
3086
3087top:
3088 attrzp = NULL;
3089 aclp = NULL;
3090
3091 /* Can this be moved to before the top label? */
3092 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3093 ZFS_EXIT(zfsvfs);
3094 return (SET_ERROR(EROFS));
3095 }
3096
3097 /*
3098 * First validate permissions
3099 */
3100
3101 if (mask & AT_SIZE) {
3102 /*
3103 * XXX - Note, we are not providing any open
3104 * mode flags here (like FNDELAY), so we may
3105 * block if there are locks present... this
3106 * should be addressed in openat().
3107 */
3108 /* XXX - would it be OK to generate a log record here? */
3109 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3110 if (err) {
3111 ZFS_EXIT(zfsvfs);
3112 return (err);
3113 }
3114 }
3115
3116 if (mask & (AT_ATIME|AT_MTIME) ||
3117 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3118 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3119 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3120 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3121 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3122 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3123 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3124 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3125 skipaclchk, cr);
3126 }
3127
3128 if (mask & (AT_UID|AT_GID)) {
3129 int idmask = (mask & (AT_UID|AT_GID));
3130 int take_owner;
3131 int take_group;
3132
3133 /*
3134 * NOTE: even if a new mode is being set,
3135 * we may clear S_ISUID/S_ISGID bits.
3136 */
3137
3138 if (!(mask & AT_MODE))
3139 vap->va_mode = zp->z_mode;
3140
3141 /*
3142 * Take ownership or chgrp to group we are a member of
3143 */
3144
3145 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3146 take_group = (mask & AT_GID) &&
3147 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3148
3149 /*
3150 * If both AT_UID and AT_GID are set then take_owner and
3151 * take_group must both be set in order to allow taking
3152 * ownership.
3153 *
3154 * Otherwise, send the check through secpolicy_vnode_setattr()
3155 *
3156 */
3157
3158 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3159 ((idmask == AT_UID) && take_owner) ||
3160 ((idmask == AT_GID) && take_group)) {
3161 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3162 skipaclchk, cr) == 0) {
3163 /*
3164 * Remove setuid/setgid for non-privileged users
3165 */
3166 secpolicy_setid_clear(vap, vp, cr);
3167 trim_mask = (mask & (AT_UID|AT_GID));
3168 } else {
3169 need_policy = TRUE;
3170 }
3171 } else {
3172 need_policy = TRUE;
3173 }
3174 }
3175
3176 mutex_enter(&zp->z_lock);
3177 oldva.va_mode = zp->z_mode;
3178 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3179 if (mask & AT_XVATTR) {
3180 /*
3181 * Update xvattr mask to include only those attributes
3182 * that are actually changing.
3183 *
3184 * the bits will be restored prior to actually setting
3185 * the attributes so the caller thinks they were set.
3186 */
3187 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3188 if (xoap->xoa_appendonly !=
3189 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3190 need_policy = TRUE;
3191 } else {
3192 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3193 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3194 }
3195 }
3196
3197 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3198 if (xoap->xoa_nounlink !=
3199 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3200 need_policy = TRUE;
3201 } else {
3202 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3203 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3204 }
3205 }
3206
3207 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3208 if (xoap->xoa_immutable !=
3209 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3210 need_policy = TRUE;
3211 } else {
3212 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3213 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3214 }
3215 }
3216
3217 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3218 if (xoap->xoa_nodump !=
3219 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3220 need_policy = TRUE;
3221 } else {
3222 XVA_CLR_REQ(xvap, XAT_NODUMP);
3223 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3224 }
3225 }
3226
3227 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3228 if (xoap->xoa_av_modified !=
3229 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3230 need_policy = TRUE;
3231 } else {
3232 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3233 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3234 }
3235 }
3236
3237 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3238 if ((vp->v_type != VREG &&
3239 xoap->xoa_av_quarantined) ||
3240 xoap->xoa_av_quarantined !=
3241 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3242 need_policy = TRUE;
3243 } else {
3244 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3245 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3246 }
3247 }
3248
3249 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3250 mutex_exit(&zp->z_lock);
3251 ZFS_EXIT(zfsvfs);
3252 return (SET_ERROR(EPERM));
3253 }
3254
3255 if (need_policy == FALSE &&
3256 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3257 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3258 need_policy = TRUE;
3259 }
3260 }
3261
3262 mutex_exit(&zp->z_lock);
3263
3264 if (mask & AT_MODE) {
3265 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3266 err = secpolicy_setid_setsticky_clear(vp, vap,
3267 &oldva, cr);
3268 if (err) {
3269 ZFS_EXIT(zfsvfs);
3270 return (err);
3271 }
3272 trim_mask |= AT_MODE;
3273 } else {
3274 need_policy = TRUE;
3275 }
3276 }
3277
3278 if (need_policy) {
3279 /*
3280 * If trim_mask is set then take ownership
3281 * has been granted or write_acl is present and user
3282 * has the ability to modify mode. In that case remove
3283 * UID|GID and or MODE from mask so that
3284 * secpolicy_vnode_setattr() doesn't revoke it.
3285 */
3286
3287 if (trim_mask) {
3288 saved_mask = vap->va_mask;
3289 vap->va_mask &= ~trim_mask;
3290 if (trim_mask & AT_MODE) {
3291 /*
3292 * Save the mode, as secpolicy_vnode_setattr()
3293 * will overwrite it with ova.va_mode.
3294 */
3295 saved_mode = vap->va_mode;
3296 }
3297 }
3298 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3299 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3300 if (err) {
3301 ZFS_EXIT(zfsvfs);
3302 return (err);
3303 }
3304
3305 if (trim_mask) {
3306 vap->va_mask |= saved_mask;
3307 if (trim_mask & AT_MODE) {
3308 /*
3309 * Recover the mode after
3310 * secpolicy_vnode_setattr().
3311 */
3312 vap->va_mode = saved_mode;
3313 }
3314 }
3315 }
3316
3317 /*
3318 * secpolicy_vnode_setattr, or take ownership may have
3319 * changed va_mask
3320 */
3321 mask = vap->va_mask;
3322
3323 if ((mask & (AT_UID | AT_GID))) {
3324 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3325 &xattr_obj, sizeof (xattr_obj));
3326
3327 if (err == 0 && xattr_obj) {
3328 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3329 if (err)
3330 goto out2;
3331 }
3332 if (mask & AT_UID) {
3333 new_uid = zfs_fuid_create(zfsvfs,
3334 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3335 if (new_uid != zp->z_uid &&
3336 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3337 if (attrzp)
3338 VN_RELE(ZTOV(attrzp));
3339 err = SET_ERROR(EDQUOT);
3340 goto out2;
3341 }
3342 }
3343
3344 if (mask & AT_GID) {
3345 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3346 cr, ZFS_GROUP, &fuidp);
3347 if (new_gid != zp->z_gid &&
3348 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3349 if (attrzp)
3350 VN_RELE(ZTOV(attrzp));
3351 err = SET_ERROR(EDQUOT);
3352 goto out2;
3353 }
3354 }
3355 }
3356 tx = dmu_tx_create(zfsvfs->z_os);
3357
3358 if (mask & AT_MODE) {
3359 uint64_t pmode = zp->z_mode;
3360 uint64_t acl_obj;
3361 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3362
3363 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3364 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3365 err = SET_ERROR(EPERM);
3366 goto out;
3367 }
3368
3369 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3370 goto out;
3371
3372 mutex_enter(&zp->z_lock);
3373 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3374 /*
3375 * Are we upgrading ACL from old V0 format
3376 * to V1 format?
3377 */
3378 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3379 zfs_znode_acl_version(zp) ==
3380 ZFS_ACL_VERSION_INITIAL) {
3381 dmu_tx_hold_free(tx, acl_obj, 0,
3382 DMU_OBJECT_END);
3383 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3384 0, aclp->z_acl_bytes);
3385 } else {
3386 dmu_tx_hold_write(tx, acl_obj, 0,
3387 aclp->z_acl_bytes);
3388 }
3389 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3390 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3391 0, aclp->z_acl_bytes);
3392 }
3393 mutex_exit(&zp->z_lock);
3394 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3395 } else {
3396 if ((mask & AT_XVATTR) &&
3397 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3398 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3399 else
3400 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3401 }
3402
3403 if (attrzp) {
3404 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3405 }
3406
3407 fuid_dirtied = zfsvfs->z_fuid_dirty;
3408 if (fuid_dirtied)
3409 zfs_fuid_txhold(zfsvfs, tx);
3410
3411 zfs_sa_upgrade_txholds(tx, zp);
3412
3413 err = dmu_tx_assign(tx, TXG_NOWAIT);
3414 if (err) {
3415 if (err == ERESTART)
3416 dmu_tx_wait(tx);
3417 goto out;
3418 }
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 if (szp->z_pflags & (ZFS_APPENDONLY | ZFS_IMMUTABLE | ZFS_READONLY)) {
4339 ZFS_EXIT(zfsvfs);
4340 return (SET_ERROR(EPERM));
4341 }
4342
4343 /*
4344 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4345 * ctldir appear to have the same v_vfsp.
4346 */
4347 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4348 ZFS_EXIT(zfsvfs);
4349 return (SET_ERROR(EXDEV));
4350 }
4351
4352 /* Prevent links to .zfs/shares files */
4353
4354 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4355 &parent, sizeof (uint64_t))) != 0) {
4356 ZFS_EXIT(zfsvfs);
4357 return (error);
4358 }
4359 if (parent == zfsvfs->z_shares_dir) {
4360 ZFS_EXIT(zfsvfs);
4361 return (SET_ERROR(EPERM));
4362 }
4363
4364 if (zfsvfs->z_utf8 && u8_validate(name,
4365 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4366 ZFS_EXIT(zfsvfs);
4367 return (SET_ERROR(EILSEQ));
4368 }
4369 if (flags & FIGNORECASE)
4370 zf |= ZCILOOK;
4371
4372 /*
4373 * We do not support links between attributes and non-attributes
4374 * because of the potential security risk of creating links
4375 * into "normal" file space in order to circumvent restrictions
4376 * imposed in attribute space.
4377 */
4378 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4379 ZFS_EXIT(zfsvfs);
4380 return (SET_ERROR(EINVAL));
4381 }
4382
4383
4384 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4385 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4386 ZFS_EXIT(zfsvfs);
4387 return (SET_ERROR(EPERM));
4388 }
4389
4390 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4391 ZFS_EXIT(zfsvfs);
4392 return (error);
4393 }
4394
4395top:
4396 /*
4397 * Attempt to lock directory; fail if entry already exists.
4398 */
4399 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4400 if (error) {
4401 ZFS_EXIT(zfsvfs);
4402 return (error);
4403 }
4404
4405 tx = dmu_tx_create(zfsvfs->z_os);
4406 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4407 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4408 zfs_sa_upgrade_txholds(tx, szp);
4409 zfs_sa_upgrade_txholds(tx, dzp);
4410 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4411 if (error) {
4412 zfs_dirent_unlock(dl);
4413 if (error == ERESTART) {
4414 waited = B_TRUE;
4415 dmu_tx_wait(tx);
4416 dmu_tx_abort(tx);
4417 goto top;
4418 }
4419 dmu_tx_abort(tx);
4420 ZFS_EXIT(zfsvfs);
4421 return (error);
4422 }
4423
4424 error = zfs_link_create(dl, szp, tx, 0);
4425
4426 if (error == 0) {
4427 uint64_t txtype = TX_LINK;
4428 if (flags & FIGNORECASE)
4429 txtype |= TX_CI;
4430 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4431 }
4432
4433 dmu_tx_commit(tx);
4434
4435 zfs_dirent_unlock(dl);
4436
4437 if (error == 0) {
4438 vnevent_link(svp, ct);
4439 }
4440
4441 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4442 zil_commit(zilog, 0);
4443
4444 ZFS_EXIT(zfsvfs);
4445 return (error);
4446}
4447
4448#ifdef sun
4449/*
4450 * zfs_null_putapage() is used when the file system has been force
4451 * unmounted. It just drops the pages.
4452 */
4453/* ARGSUSED */
4454static int
4455zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4456 size_t *lenp, int flags, cred_t *cr)
4457{
4458 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4459 return (0);
4460}
4461
4462/*
4463 * Push a page out to disk, klustering if possible.
4464 *
4465 * IN: vp - file to push page to.
4466 * pp - page to push.
4467 * flags - additional flags.
4468 * cr - credentials of caller.
4469 *
4470 * OUT: offp - start of range pushed.
4471 * lenp - len of range pushed.
4472 *
4473 * RETURN: 0 on success, error code on failure.
4474 *
4475 * NOTE: callers must have locked the page to be pushed. On
4476 * exit, the page (and all other pages in the kluster) must be
4477 * unlocked.
4478 */
4479/* ARGSUSED */
4480static int
4481zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4482 size_t *lenp, int flags, cred_t *cr)
4483{
4484 znode_t *zp = VTOZ(vp);
4485 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4486 dmu_tx_t *tx;
4487 u_offset_t off, koff;
4488 size_t len, klen;
4489 int err;
4490
4491 off = pp->p_offset;
4492 len = PAGESIZE;
4493 /*
4494 * If our blocksize is bigger than the page size, try to kluster
4495 * multiple pages so that we write a full block (thus avoiding
4496 * a read-modify-write).
4497 */
4498 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4499 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4500 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4501 ASSERT(koff <= zp->z_size);
4502 if (koff + klen > zp->z_size)
4503 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4504 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4505 }
4506 ASSERT3U(btop(len), ==, btopr(len));
4507
4508 /*
4509 * Can't push pages past end-of-file.
4510 */
4511 if (off >= zp->z_size) {
4512 /* ignore all pages */
4513 err = 0;
4514 goto out;
4515 } else if (off + len > zp->z_size) {
4516 int npages = btopr(zp->z_size - off);
4517 page_t *trunc;
4518
4519 page_list_break(&pp, &trunc, npages);
4520 /* ignore pages past end of file */
4521 if (trunc)
4522 pvn_write_done(trunc, flags);
4523 len = zp->z_size - off;
4524 }
4525
4526 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4527 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4528 err = SET_ERROR(EDQUOT);
4529 goto out;
4530 }
4531top:
4532 tx = dmu_tx_create(zfsvfs->z_os);
4533 dmu_tx_hold_write(tx, zp->z_id, off, len);
4534
4535 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4536 zfs_sa_upgrade_txholds(tx, zp);
4537 err = dmu_tx_assign(tx, TXG_NOWAIT);
4538 if (err != 0) {
4539 if (err == ERESTART) {
4540 dmu_tx_wait(tx);
4541 dmu_tx_abort(tx);
4542 goto top;
4543 }
4544 dmu_tx_abort(tx);
4545 goto out;
4546 }
4547
4548 if (zp->z_blksz <= PAGESIZE) {
4549 caddr_t va = zfs_map_page(pp, S_READ);
4550 ASSERT3U(len, <=, PAGESIZE);
4551 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4552 zfs_unmap_page(pp, va);
4553 } else {
4554 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4555 }
4556
4557 if (err == 0) {
4558 uint64_t mtime[2], ctime[2];
4559 sa_bulk_attr_t bulk[3];
4560 int count = 0;
4561
4562 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4563 &mtime, 16);
4564 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4565 &ctime, 16);
4566 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4567 &zp->z_pflags, 8);
4568 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4569 B_TRUE);
4570 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4571 }
4572 dmu_tx_commit(tx);
4573
4574out:
4575 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4576 if (offp)
4577 *offp = off;
4578 if (lenp)
4579 *lenp = len;
4580
4581 return (err);
4582}
4583
4584/*
4585 * Copy the portion of the file indicated from pages into the file.
4586 * The pages are stored in a page list attached to the files vnode.
4587 *
4588 * IN: vp - vnode of file to push page data to.
4589 * off - position in file to put data.
4590 * len - amount of data to write.
4591 * flags - flags to control the operation.
4592 * cr - credentials of caller.
4593 * ct - caller context.
4594 *
4595 * RETURN: 0 on success, error code on failure.
4596 *
4597 * Timestamps:
4598 * vp - ctime|mtime updated
4599 */
4600/*ARGSUSED*/
4601static int
4602zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4603 caller_context_t *ct)
4604{
4605 znode_t *zp = VTOZ(vp);
4606 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4607 page_t *pp;
4608 size_t io_len;
4609 u_offset_t io_off;
4610 uint_t blksz;
4611 rl_t *rl;
4612 int error = 0;
4613
4614 ZFS_ENTER(zfsvfs);
4615 ZFS_VERIFY_ZP(zp);
4616
4617 /*
4618 * Align this request to the file block size in case we kluster.
4619 * XXX - this can result in pretty aggresive locking, which can
4620 * impact simultanious read/write access. One option might be
4621 * to break up long requests (len == 0) into block-by-block
4622 * operations to get narrower locking.
4623 */
4624 blksz = zp->z_blksz;
4625 if (ISP2(blksz))
4626 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4627 else
4628 io_off = 0;
4629 if (len > 0 && ISP2(blksz))
4630 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4631 else
4632 io_len = 0;
4633
4634 if (io_len == 0) {
4635 /*
4636 * Search the entire vp list for pages >= io_off.
4637 */
4638 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4639 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4640 goto out;
4641 }
4642 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4643
4644 if (off > zp->z_size) {
4645 /* past end of file */
4646 zfs_range_unlock(rl);
4647 ZFS_EXIT(zfsvfs);
4648 return (0);
4649 }
4650
4651 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4652
4653 for (off = io_off; io_off < off + len; io_off += io_len) {
4654 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4655 pp = page_lookup(vp, io_off,
4656 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4657 } else {
4658 pp = page_lookup_nowait(vp, io_off,
4659 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4660 }
4661
4662 if (pp != NULL && pvn_getdirty(pp, flags)) {
4663 int err;
4664
4665 /*
4666 * Found a dirty page to push
4667 */
4668 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4669 if (err)
4670 error = err;
4671 } else {
4672 io_len = PAGESIZE;
4673 }
4674 }
4675out:
4676 zfs_range_unlock(rl);
4677 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4678 zil_commit(zfsvfs->z_log, zp->z_id);
4679 ZFS_EXIT(zfsvfs);
4680 return (error);
4681}
4682#endif /* sun */
4683
4684/*ARGSUSED*/
4685void
4686zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4687{
4688 znode_t *zp = VTOZ(vp);
4689 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4690 int error;
4691
4692 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4693 if (zp->z_sa_hdl == NULL) {
4694 /*
4695 * The fs has been unmounted, or we did a
4696 * suspend/resume and this file no longer exists.
4697 */
4698 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4699 vrecycle(vp);
4700 return;
4701 }
4702
4703 mutex_enter(&zp->z_lock);
4704 if (zp->z_unlinked) {
4705 /*
4706 * Fast path to recycle a vnode of a removed file.
4707 */
4708 mutex_exit(&zp->z_lock);
4709 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4710 vrecycle(vp);
4711 return;
4712 }
4713 mutex_exit(&zp->z_lock);
4714
4715 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4716 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4717
4718 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4719 zfs_sa_upgrade_txholds(tx, zp);
4720 error = dmu_tx_assign(tx, TXG_WAIT);
4721 if (error) {
4722 dmu_tx_abort(tx);
4723 } else {
4724 mutex_enter(&zp->z_lock);
4725 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4726 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4727 zp->z_atime_dirty = 0;
4728 mutex_exit(&zp->z_lock);
4729 dmu_tx_commit(tx);
4730 }
4731 }
4732 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4733}
4734
4735#ifdef sun
4736/*
4737 * Bounds-check the seek operation.
4738 *
4739 * IN: vp - vnode seeking within
4740 * ooff - old file offset
4741 * noffp - pointer to new file offset
4742 * ct - caller context
4743 *
4744 * RETURN: 0 on success, EINVAL if new offset invalid.
4745 */
4746/* ARGSUSED */
4747static int
4748zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4749 caller_context_t *ct)
4750{
4751 if (vp->v_type == VDIR)
4752 return (0);
4753 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4754}
4755
4756/*
4757 * Pre-filter the generic locking function to trap attempts to place
4758 * a mandatory lock on a memory mapped file.
4759 */
4760static int
4761zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4762 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4763{
4764 znode_t *zp = VTOZ(vp);
4765 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4766
4767 ZFS_ENTER(zfsvfs);
4768 ZFS_VERIFY_ZP(zp);
4769
4770 /*
4771 * We are following the UFS semantics with respect to mapcnt
4772 * here: If we see that the file is mapped already, then we will
4773 * return an error, but we don't worry about races between this
4774 * function and zfs_map().
4775 */
4776 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4777 ZFS_EXIT(zfsvfs);
4778 return (SET_ERROR(EAGAIN));
4779 }
4780 ZFS_EXIT(zfsvfs);
4781 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4782}
4783
4784/*
4785 * If we can't find a page in the cache, we will create a new page
4786 * and fill it with file data. For efficiency, we may try to fill
4787 * multiple pages at once (klustering) to fill up the supplied page
4788 * list. Note that the pages to be filled are held with an exclusive
4789 * lock to prevent access by other threads while they are being filled.
4790 */
4791static int
4792zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4793 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4794{
4795 znode_t *zp = VTOZ(vp);
4796 page_t *pp, *cur_pp;
4797 objset_t *os = zp->z_zfsvfs->z_os;
4798 u_offset_t io_off, total;
4799 size_t io_len;
4800 int err;
4801
4802 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4803 /*
4804 * We only have a single page, don't bother klustering
4805 */
4806 io_off = off;
4807 io_len = PAGESIZE;
4808 pp = page_create_va(vp, io_off, io_len,
4809 PG_EXCL | PG_WAIT, seg, addr);
4810 } else {
4811 /*
4812 * Try to find enough pages to fill the page list
4813 */
4814 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4815 &io_len, off, plsz, 0);
4816 }
4817 if (pp == NULL) {
4818 /*
4819 * The page already exists, nothing to do here.
4820 */
4821 *pl = NULL;
4822 return (0);
4823 }
4824
4825 /*
4826 * Fill the pages in the kluster.
4827 */
4828 cur_pp = pp;
4829 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4830 caddr_t va;
4831
4832 ASSERT3U(io_off, ==, cur_pp->p_offset);
4833 va = zfs_map_page(cur_pp, S_WRITE);
4834 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4835 DMU_READ_PREFETCH);
4836 zfs_unmap_page(cur_pp, va);
4837 if (err) {
4838 /* On error, toss the entire kluster */
4839 pvn_read_done(pp, B_ERROR);
4840 /* convert checksum errors into IO errors */
4841 if (err == ECKSUM)
4842 err = SET_ERROR(EIO);
4843 return (err);
4844 }
4845 cur_pp = cur_pp->p_next;
4846 }
4847
4848 /*
4849 * Fill in the page list array from the kluster starting
4850 * from the desired offset `off'.
4851 * NOTE: the page list will always be null terminated.
4852 */
4853 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4854 ASSERT(pl == NULL || (*pl)->p_offset == off);
4855
4856 return (0);
4857}
4858
4859/*
4860 * Return pointers to the pages for the file region [off, off + len]
4861 * in the pl array. If plsz is greater than len, this function may
4862 * also return page pointers from after the specified region
4863 * (i.e. the region [off, off + plsz]). These additional pages are
4864 * only returned if they are already in the cache, or were created as
4865 * part of a klustered read.
4866 *
4867 * IN: vp - vnode of file to get data from.
4868 * off - position in file to get data from.
4869 * len - amount of data to retrieve.
4870 * plsz - length of provided page list.
4871 * seg - segment to obtain pages for.
4872 * addr - virtual address of fault.
4873 * rw - mode of created pages.
4874 * cr - credentials of caller.
4875 * ct - caller context.
4876 *
4877 * OUT: protp - protection mode of created pages.
4878 * pl - list of pages created.
4879 *
4880 * RETURN: 0 on success, error code on failure.
4881 *
4882 * Timestamps:
4883 * vp - atime updated
4884 */
4885/* ARGSUSED */
4886static int
4887zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4888 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4889 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4890{
4891 znode_t *zp = VTOZ(vp);
4892 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4893 page_t **pl0 = pl;
4894 int err = 0;
4895
4896 /* we do our own caching, faultahead is unnecessary */
4897 if (pl == NULL)
4898 return (0);
4899 else if (len > plsz)
4900 len = plsz;
4901 else
4902 len = P2ROUNDUP(len, PAGESIZE);
4903 ASSERT(plsz >= len);
4904
4905 ZFS_ENTER(zfsvfs);
4906 ZFS_VERIFY_ZP(zp);
4907
4908 if (protp)
4909 *protp = PROT_ALL;
4910
4911 /*
4912 * Loop through the requested range [off, off + len) looking
4913 * for pages. If we don't find a page, we will need to create
4914 * a new page and fill it with data from the file.
4915 */
4916 while (len > 0) {
4917 if (*pl = page_lookup(vp, off, SE_SHARED))
4918 *(pl+1) = NULL;
4919 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4920 goto out;
4921 while (*pl) {
4922 ASSERT3U((*pl)->p_offset, ==, off);
4923 off += PAGESIZE;
4924 addr += PAGESIZE;
4925 if (len > 0) {
4926 ASSERT3U(len, >=, PAGESIZE);
4927 len -= PAGESIZE;
4928 }
4929 ASSERT3U(plsz, >=, PAGESIZE);
4930 plsz -= PAGESIZE;
4931 pl++;
4932 }
4933 }
4934
4935 /*
4936 * Fill out the page array with any pages already in the cache.
4937 */
4938 while (plsz > 0 &&
4939 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4940 off += PAGESIZE;
4941 plsz -= PAGESIZE;
4942 }
4943out:
4944 if (err) {
4945 /*
4946 * Release any pages we have previously locked.
4947 */
4948 while (pl > pl0)
4949 page_unlock(*--pl);
4950 } else {
4951 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4952 }
4953
4954 *pl = NULL;
4955
4956 ZFS_EXIT(zfsvfs);
4957 return (err);
4958}
4959
4960/*
4961 * Request a memory map for a section of a file. This code interacts
4962 * with common code and the VM system as follows:
4963 *
4964 * - common code calls mmap(), which ends up in smmap_common()
4965 * - this calls VOP_MAP(), which takes you into (say) zfs
4966 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4967 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4968 * - zfs_addmap() updates z_mapcnt
4969 */
4970/*ARGSUSED*/
4971static int
4972zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4973 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4974 caller_context_t *ct)
4975{
4976 znode_t *zp = VTOZ(vp);
4977 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4978 segvn_crargs_t vn_a;
4979 int error;
4980
4981 ZFS_ENTER(zfsvfs);
4982 ZFS_VERIFY_ZP(zp);
4983
4984 if ((prot & PROT_WRITE) && (zp->z_pflags &
4985 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4986 ZFS_EXIT(zfsvfs);
4987 return (SET_ERROR(EPERM));
4988 }
4989
4990 if ((prot & (PROT_READ | PROT_EXEC)) &&
4991 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4992 ZFS_EXIT(zfsvfs);
4993 return (SET_ERROR(EACCES));
4994 }
4995
4996 if (vp->v_flag & VNOMAP) {
4997 ZFS_EXIT(zfsvfs);
4998 return (SET_ERROR(ENOSYS));
4999 }
5000
5001 if (off < 0 || len > MAXOFFSET_T - off) {
5002 ZFS_EXIT(zfsvfs);
5003 return (SET_ERROR(ENXIO));
5004 }
5005
5006 if (vp->v_type != VREG) {
5007 ZFS_EXIT(zfsvfs);
5008 return (SET_ERROR(ENODEV));
5009 }
5010
5011 /*
5012 * If file is locked, disallow mapping.
5013 */
5014 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
5015 ZFS_EXIT(zfsvfs);
5016 return (SET_ERROR(EAGAIN));
5017 }
5018
5019 as_rangelock(as);
5020 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5021 if (error != 0) {
5022 as_rangeunlock(as);
5023 ZFS_EXIT(zfsvfs);
5024 return (error);
5025 }
5026
5027 vn_a.vp = vp;
5028 vn_a.offset = (u_offset_t)off;
5029 vn_a.type = flags & MAP_TYPE;
5030 vn_a.prot = prot;
5031 vn_a.maxprot = maxprot;
5032 vn_a.cred = cr;
5033 vn_a.amp = NULL;
5034 vn_a.flags = flags & ~MAP_TYPE;
5035 vn_a.szc = 0;
5036 vn_a.lgrp_mem_policy_flags = 0;
5037
5038 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5039
5040 as_rangeunlock(as);
5041 ZFS_EXIT(zfsvfs);
5042 return (error);
5043}
5044
5045/* ARGSUSED */
5046static int
5047zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5048 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5049 caller_context_t *ct)
5050{
5051 uint64_t pages = btopr(len);
5052
5053 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5054 return (0);
5055}
5056
5057/*
5058 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5059 * more accurate mtime for the associated file. Since we don't have a way of
5060 * detecting when the data was actually modified, we have to resort to
5061 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5062 * last page is pushed. The problem occurs when the msync() call is omitted,
5063 * which by far the most common case:
5064 *
5065 * open()
5066 * mmap()
5067 * <modify memory>
5068 * munmap()
5069 * close()
5070 * <time lapse>
5071 * putpage() via fsflush
5072 *
5073 * If we wait until fsflush to come along, we can have a modification time that
5074 * is some arbitrary point in the future. In order to prevent this in the
5075 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5076 * torn down.
5077 */
5078/* ARGSUSED */
5079static int
5080zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5081 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5082 caller_context_t *ct)
5083{
5084 uint64_t pages = btopr(len);
5085
5086 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5087 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5088
5089 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5090 vn_has_cached_data(vp))
5091 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5092
5093 return (0);
5094}
5095
5096/*
5097 * Free or allocate space in a file. Currently, this function only
5098 * supports the `F_FREESP' command. However, this command is somewhat
5099 * misnamed, as its functionality includes the ability to allocate as
5100 * well as free space.
5101 *
5102 * IN: vp - vnode of file to free data in.
5103 * cmd - action to take (only F_FREESP supported).
5104 * bfp - section of file to free/alloc.
5105 * flag - current file open mode flags.
5106 * offset - current file offset.
5107 * cr - credentials of caller [UNUSED].
5108 * ct - caller context.
5109 *
5110 * RETURN: 0 on success, error code on failure.
5111 *
5112 * Timestamps:
5113 * vp - ctime|mtime updated
5114 */
5115/* ARGSUSED */
5116static int
5117zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5118 offset_t offset, cred_t *cr, caller_context_t *ct)
5119{
5120 znode_t *zp = VTOZ(vp);
5121 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5122 uint64_t off, len;
5123 int error;
5124
5125 ZFS_ENTER(zfsvfs);
5126 ZFS_VERIFY_ZP(zp);
5127
5128 if (cmd != F_FREESP) {
5129 ZFS_EXIT(zfsvfs);
5130 return (SET_ERROR(EINVAL));
5131 }
5132
5133 if (error = convoff(vp, bfp, 0, offset)) {
5134 ZFS_EXIT(zfsvfs);
5135 return (error);
5136 }
5137
5138 if (bfp->l_len < 0) {
5139 ZFS_EXIT(zfsvfs);
5140 return (SET_ERROR(EINVAL));
5141 }
5142
5143 off = bfp->l_start;
5144 len = bfp->l_len; /* 0 means from off to end of file */
5145
5146 error = zfs_freesp(zp, off, len, flag, TRUE);
5147
5148 ZFS_EXIT(zfsvfs);
5149 return (error);
5150}
5151#endif /* sun */
5152
5153CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5154CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5155
5156/*ARGSUSED*/
5157static int
5158zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5159{
5160 znode_t *zp = VTOZ(vp);
5161 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5162 uint32_t gen;
5163 uint64_t gen64;
5164 uint64_t object = zp->z_id;
5165 zfid_short_t *zfid;
5166 int size, i, error;
5167
5168 ZFS_ENTER(zfsvfs);
5169 ZFS_VERIFY_ZP(zp);
5170
5171 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5172 &gen64, sizeof (uint64_t))) != 0) {
5173 ZFS_EXIT(zfsvfs);
5174 return (error);
5175 }
5176
5177 gen = (uint32_t)gen64;
5178
5179 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5180
5181#ifdef illumos
5182 if (fidp->fid_len < size) {
5183 fidp->fid_len = size;
5184 ZFS_EXIT(zfsvfs);
5185 return (SET_ERROR(ENOSPC));
5186 }
5187#else
5188 fidp->fid_len = size;
5189#endif
5190
5191 zfid = (zfid_short_t *)fidp;
5192
5193 zfid->zf_len = size;
5194
5195 for (i = 0; i < sizeof (zfid->zf_object); i++)
5196 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5197
5198 /* Must have a non-zero generation number to distinguish from .zfs */
5199 if (gen == 0)
5200 gen = 1;
5201 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5202 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5203
5204 if (size == LONG_FID_LEN) {
5205 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5206 zfid_long_t *zlfid;
5207
5208 zlfid = (zfid_long_t *)fidp;
5209
5210 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5211 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5212
5213 /* XXX - this should be the generation number for the objset */
5214 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5215 zlfid->zf_setgen[i] = 0;
5216 }
5217
5218 ZFS_EXIT(zfsvfs);
5219 return (0);
5220}
5221
5222static int
5223zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5224 caller_context_t *ct)
5225{
5226 znode_t *zp, *xzp;
5227 zfsvfs_t *zfsvfs;
5228 zfs_dirlock_t *dl;
5229 int error;
5230
5231 switch (cmd) {
5232 case _PC_LINK_MAX:
5233 *valp = INT_MAX;
5234 return (0);
5235
5236 case _PC_FILESIZEBITS:
5237 *valp = 64;
5238 return (0);
5239#ifdef sun
5240 case _PC_XATTR_EXISTS:
5241 zp = VTOZ(vp);
5242 zfsvfs = zp->z_zfsvfs;
5243 ZFS_ENTER(zfsvfs);
5244 ZFS_VERIFY_ZP(zp);
5245 *valp = 0;
5246 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5247 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5248 if (error == 0) {
5249 zfs_dirent_unlock(dl);
5250 if (!zfs_dirempty(xzp))
5251 *valp = 1;
5252 VN_RELE(ZTOV(xzp));
5253 } else if (error == ENOENT) {
5254 /*
5255 * If there aren't extended attributes, it's the
5256 * same as having zero of them.
5257 */
5258 error = 0;
5259 }
5260 ZFS_EXIT(zfsvfs);
5261 return (error);
5262
5263 case _PC_SATTR_ENABLED:
5264 case _PC_SATTR_EXISTS:
5265 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5266 (vp->v_type == VREG || vp->v_type == VDIR);
5267 return (0);
5268
5269 case _PC_ACCESS_FILTERING:
5270 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5271 vp->v_type == VDIR;
5272 return (0);
5273
5274 case _PC_ACL_ENABLED:
5275 *valp = _ACL_ACE_ENABLED;
5276 return (0);
5277#endif /* sun */
5278 case _PC_MIN_HOLE_SIZE:
5279 *valp = (int)SPA_MINBLOCKSIZE;
5280 return (0);
5281#ifdef sun
5282 case _PC_TIMESTAMP_RESOLUTION:
5283 /* nanosecond timestamp resolution */
5284 *valp = 1L;
5285 return (0);
5286#endif /* sun */
5287 case _PC_ACL_EXTENDED:
5288 *valp = 0;
5289 return (0);
5290
5291 case _PC_ACL_NFS4:
5292 *valp = 1;
5293 return (0);
5294
5295 case _PC_ACL_PATH_MAX:
5296 *valp = ACL_MAX_ENTRIES;
5297 return (0);
5298
5299 default:
5300 return (EOPNOTSUPP);
5301 }
5302}
5303
5304/*ARGSUSED*/
5305static int
5306zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5307 caller_context_t *ct)
5308{
5309 znode_t *zp = VTOZ(vp);
5310 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5311 int error;
5312 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5313
5314 ZFS_ENTER(zfsvfs);
5315 ZFS_VERIFY_ZP(zp);
5316 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5317 ZFS_EXIT(zfsvfs);
5318
5319 return (error);
5320}
5321
5322/*ARGSUSED*/
5323int
5324zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5325 caller_context_t *ct)
5326{
5327 znode_t *zp = VTOZ(vp);
5328 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5329 int error;
5330 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5331 zilog_t *zilog = zfsvfs->z_log;
5332
5333 ZFS_ENTER(zfsvfs);
5334 ZFS_VERIFY_ZP(zp);
5335
5336 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5337
5338 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5339 zil_commit(zilog, 0);
5340
5341 ZFS_EXIT(zfsvfs);
5342 return (error);
5343}
5344
5345#ifdef sun
5346/*
5347 * The smallest read we may consider to loan out an arcbuf.
5348 * This must be a power of 2.
5349 */
5350int zcr_blksz_min = (1 << 10); /* 1K */
5351/*
5352 * If set to less than the file block size, allow loaning out of an
5353 * arcbuf for a partial block read. This must be a power of 2.
5354 */
5355int zcr_blksz_max = (1 << 17); /* 128K */
5356
5357/*ARGSUSED*/
5358static int
5359zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5360 caller_context_t *ct)
5361{
5362 znode_t *zp = VTOZ(vp);
5363 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5364 int max_blksz = zfsvfs->z_max_blksz;
5365 uio_t *uio = &xuio->xu_uio;
5366 ssize_t size = uio->uio_resid;
5367 offset_t offset = uio->uio_loffset;
5368 int blksz;
5369 int fullblk, i;
5370 arc_buf_t *abuf;
5371 ssize_t maxsize;
5372 int preamble, postamble;
5373
5374 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5375 return (SET_ERROR(EINVAL));
5376
5377 ZFS_ENTER(zfsvfs);
5378 ZFS_VERIFY_ZP(zp);
5379 switch (ioflag) {
5380 case UIO_WRITE:
5381 /*
5382 * Loan out an arc_buf for write if write size is bigger than
5383 * max_blksz, and the file's block size is also max_blksz.
5384 */
5385 blksz = max_blksz;
5386 if (size < blksz || zp->z_blksz != blksz) {
5387 ZFS_EXIT(zfsvfs);
5388 return (SET_ERROR(EINVAL));
5389 }
5390 /*
5391 * Caller requests buffers for write before knowing where the
5392 * write offset might be (e.g. NFS TCP write).
5393 */
5394 if (offset == -1) {
5395 preamble = 0;
5396 } else {
5397 preamble = P2PHASE(offset, blksz);
5398 if (preamble) {
5399 preamble = blksz - preamble;
5400 size -= preamble;
5401 }
5402 }
5403
5404 postamble = P2PHASE(size, blksz);
5405 size -= postamble;
5406
5407 fullblk = size / blksz;
5408 (void) dmu_xuio_init(xuio,
5409 (preamble != 0) + fullblk + (postamble != 0));
5410 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5411 int, postamble, int,
5412 (preamble != 0) + fullblk + (postamble != 0));
5413
5414 /*
5415 * Have to fix iov base/len for partial buffers. They
5416 * currently represent full arc_buf's.
5417 */
5418 if (preamble) {
5419 /* data begins in the middle of the arc_buf */
5420 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5421 blksz);
5422 ASSERT(abuf);
5423 (void) dmu_xuio_add(xuio, abuf,
5424 blksz - preamble, preamble);
5425 }
5426
5427 for (i = 0; i < fullblk; i++) {
5428 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5429 blksz);
5430 ASSERT(abuf);
5431 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5432 }
5433
5434 if (postamble) {
5435 /* data ends in the middle of the arc_buf */
5436 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5437 blksz);
5438 ASSERT(abuf);
5439 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5440 }
5441 break;
5442 case UIO_READ:
5443 /*
5444 * Loan out an arc_buf for read if the read size is larger than
5445 * the current file block size. Block alignment is not
5446 * considered. Partial arc_buf will be loaned out for read.
5447 */
5448 blksz = zp->z_blksz;
5449 if (blksz < zcr_blksz_min)
5450 blksz = zcr_blksz_min;
5451 if (blksz > zcr_blksz_max)
5452 blksz = zcr_blksz_max;
5453 /* avoid potential complexity of dealing with it */
5454 if (blksz > max_blksz) {
5455 ZFS_EXIT(zfsvfs);
5456 return (SET_ERROR(EINVAL));
5457 }
5458
5459 maxsize = zp->z_size - uio->uio_loffset;
5460 if (size > maxsize)
5461 size = maxsize;
5462
5463 if (size < blksz || vn_has_cached_data(vp)) {
5464 ZFS_EXIT(zfsvfs);
5465 return (SET_ERROR(EINVAL));
5466 }
5467 break;
5468 default:
5469 ZFS_EXIT(zfsvfs);
5470 return (SET_ERROR(EINVAL));
5471 }
5472
5473 uio->uio_extflg = UIO_XUIO;
5474 XUIO_XUZC_RW(xuio) = ioflag;
5475 ZFS_EXIT(zfsvfs);
5476 return (0);
5477}
5478
5479/*ARGSUSED*/
5480static int
5481zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5482{
5483 int i;
5484 arc_buf_t *abuf;
5485 int ioflag = XUIO_XUZC_RW(xuio);
5486
5487 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5488
5489 i = dmu_xuio_cnt(xuio);
5490 while (i-- > 0) {
5491 abuf = dmu_xuio_arcbuf(xuio, i);
5492 /*
5493 * if abuf == NULL, it must be a write buffer
5494 * that has been returned in zfs_write().
5495 */
5496 if (abuf)
5497 dmu_return_arcbuf(abuf);
5498 ASSERT(abuf || ioflag == UIO_WRITE);
5499 }
5500
5501 dmu_xuio_fini(xuio);
5502 return (0);
5503}
5504
5505/*
5506 * Predeclare these here so that the compiler assumes that
5507 * this is an "old style" function declaration that does
5508 * not include arguments => we won't get type mismatch errors
5509 * in the initializations that follow.
5510 */
5511static int zfs_inval();
5512static int zfs_isdir();
5513
5514static int
5515zfs_inval()
5516{
5517 return (SET_ERROR(EINVAL));
5518}
5519
5520static int
5521zfs_isdir()
5522{
5523 return (SET_ERROR(EISDIR));
5524}
5525/*
5526 * Directory vnode operations template
5527 */
5528vnodeops_t *zfs_dvnodeops;
5529const fs_operation_def_t zfs_dvnodeops_template[] = {
5530 VOPNAME_OPEN, { .vop_open = zfs_open },
5531 VOPNAME_CLOSE, { .vop_close = zfs_close },
5532 VOPNAME_READ, { .error = zfs_isdir },
5533 VOPNAME_WRITE, { .error = zfs_isdir },
5534 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5535 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5536 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5537 VOPNAME_ACCESS, { .vop_access = zfs_access },
5538 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5539 VOPNAME_CREATE, { .vop_create = zfs_create },
5540 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5541 VOPNAME_LINK, { .vop_link = zfs_link },
5542 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5543 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5544 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5545 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5546 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5547 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5548 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5549 VOPNAME_FID, { .vop_fid = zfs_fid },
5550 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5551 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5552 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5553 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5554 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5555 NULL, NULL
5556};
5557
5558/*
5559 * Regular file vnode operations template
5560 */
5561vnodeops_t *zfs_fvnodeops;
5562const fs_operation_def_t zfs_fvnodeops_template[] = {
5563 VOPNAME_OPEN, { .vop_open = zfs_open },
5564 VOPNAME_CLOSE, { .vop_close = zfs_close },
5565 VOPNAME_READ, { .vop_read = zfs_read },
5566 VOPNAME_WRITE, { .vop_write = zfs_write },
5567 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5568 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5569 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5570 VOPNAME_ACCESS, { .vop_access = zfs_access },
5571 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5572 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5573 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5574 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5575 VOPNAME_FID, { .vop_fid = zfs_fid },
5576 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5577 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5578 VOPNAME_SPACE, { .vop_space = zfs_space },
5579 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5580 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5581 VOPNAME_MAP, { .vop_map = zfs_map },
5582 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5583 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5584 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5585 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5586 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5587 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5588 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5589 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5590 NULL, NULL
5591};
5592
5593/*
5594 * Symbolic link vnode operations template
5595 */
5596vnodeops_t *zfs_symvnodeops;
5597const fs_operation_def_t zfs_symvnodeops_template[] = {
5598 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5599 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5600 VOPNAME_ACCESS, { .vop_access = zfs_access },
5601 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5602 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5603 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5604 VOPNAME_FID, { .vop_fid = zfs_fid },
5605 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5606 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5607 NULL, NULL
5608};
5609
5610/*
5611 * special share hidden files vnode operations template
5612 */
5613vnodeops_t *zfs_sharevnodeops;
5614const fs_operation_def_t zfs_sharevnodeops_template[] = {
5615 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5616 VOPNAME_ACCESS, { .vop_access = zfs_access },
5617 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5618 VOPNAME_FID, { .vop_fid = zfs_fid },
5619 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5620 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5621 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5622 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5623 NULL, NULL
5624};
5625
5626/*
5627 * Extended attribute directory vnode operations template
5628 *
5629 * This template is identical to the directory vnodes
5630 * operation template except for restricted operations:
5631 * VOP_MKDIR()
5632 * VOP_SYMLINK()
5633 *
5634 * Note that there are other restrictions embedded in:
5635 * zfs_create() - restrict type to VREG
5636 * zfs_link() - no links into/out of attribute space
5637 * zfs_rename() - no moves into/out of attribute space
5638 */
5639vnodeops_t *zfs_xdvnodeops;
5640const fs_operation_def_t zfs_xdvnodeops_template[] = {
5641 VOPNAME_OPEN, { .vop_open = zfs_open },
5642 VOPNAME_CLOSE, { .vop_close = zfs_close },
5643 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5644 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5645 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5646 VOPNAME_ACCESS, { .vop_access = zfs_access },
5647 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5648 VOPNAME_CREATE, { .vop_create = zfs_create },
5649 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5650 VOPNAME_LINK, { .vop_link = zfs_link },
5651 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5652 VOPNAME_MKDIR, { .error = zfs_inval },
5653 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5654 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5655 VOPNAME_SYMLINK, { .error = zfs_inval },
5656 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5657 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5658 VOPNAME_FID, { .vop_fid = zfs_fid },
5659 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5660 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5661 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5662 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5663 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5664 NULL, NULL
5665};
5666
5667/*
5668 * Error vnode operations template
5669 */
5670vnodeops_t *zfs_evnodeops;
5671const fs_operation_def_t zfs_evnodeops_template[] = {
5672 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5673 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5674 NULL, NULL
5675};
5676#endif /* sun */
5677
5678static int
5679ioflags(int ioflags)
5680{
5681 int flags = 0;
5682
5683 if (ioflags & IO_APPEND)
5684 flags |= FAPPEND;
5685 if (ioflags & IO_NDELAY)
5686 flags |= FNONBLOCK;
5687 if (ioflags & IO_SYNC)
5688 flags |= (FSYNC | FDSYNC | FRSYNC);
5689
5690 return (flags);
5691}
5692
5693static int
5694zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5695{
5696 znode_t *zp = VTOZ(vp);
5697 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5698 objset_t *os = zp->z_zfsvfs->z_os;
5699 vm_page_t mfirst, mlast, mreq;
5700 vm_object_t object;
5701 caddr_t va;
5702 struct sf_buf *sf;
5703 off_t startoff, endoff;
5704 int i, error;
5705 vm_pindex_t reqstart, reqend;
5706 int pcount, lsize, reqsize, size;
5707
5708 ZFS_ENTER(zfsvfs);
5709 ZFS_VERIFY_ZP(zp);
5710
5711 pcount = OFF_TO_IDX(round_page(count));
5712 mreq = m[reqpage];
5713 object = mreq->object;
5714 error = 0;
5715
5716 KASSERT(vp->v_object == object, ("mismatching object"));
5717
5718 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5719 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5720 reqstart = OFF_TO_IDX(round_page(startoff));
5721 if (reqstart < m[0]->pindex)
5722 reqstart = 0;
5723 else
5724 reqstart = reqstart - m[0]->pindex;
5725 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5726 zp->z_blksz);
5727 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5728 if (reqend > m[pcount - 1]->pindex)
5729 reqend = m[pcount - 1]->pindex;
5730 reqsize = reqend - m[reqstart]->pindex + 1;
5731 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5732 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5733 } else {
5734 reqstart = reqpage;
5735 reqsize = 1;
5736 }
5737 mfirst = m[reqstart];
5738 mlast = m[reqstart + reqsize - 1];
5739
5740 zfs_vmobject_wlock(object);
5741
5742 for (i = 0; i < reqstart; i++) {
5743 vm_page_lock(m[i]);
5744 vm_page_free(m[i]);
5745 vm_page_unlock(m[i]);
5746 }
5747 for (i = reqstart + reqsize; i < pcount; i++) {
5748 vm_page_lock(m[i]);
5749 vm_page_free(m[i]);
5750 vm_page_unlock(m[i]);
5751 }
5752
5753 if (mreq->valid && reqsize == 1) {
5754 if (mreq->valid != VM_PAGE_BITS_ALL)
5755 vm_page_zero_invalid(mreq, TRUE);
5756 zfs_vmobject_wunlock(object);
5757 ZFS_EXIT(zfsvfs);
5758 return (zfs_vm_pagerret_ok);
5759 }
5760
5761 PCPU_INC(cnt.v_vnodein);
5762 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5763
5764 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5765 for (i = reqstart; i < reqstart + reqsize; i++) {
5766 if (i != reqpage) {
5767 vm_page_lock(m[i]);
5768 vm_page_free(m[i]);
5769 vm_page_unlock(m[i]);
5770 }
5771 }
5772 zfs_vmobject_wunlock(object);
5773 ZFS_EXIT(zfsvfs);
5774 return (zfs_vm_pagerret_bad);
5775 }
5776
5777 lsize = PAGE_SIZE;
5778 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5779 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5780
5781 zfs_vmobject_wunlock(object);
5782
5783 for (i = reqstart; i < reqstart + reqsize; i++) {
5784 size = PAGE_SIZE;
5785 if (i == (reqstart + reqsize - 1))
5786 size = lsize;
5787 va = zfs_map_page(m[i], &sf);
5788 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5789 size, va, DMU_READ_PREFETCH);
5790 if (size != PAGE_SIZE)
5791 bzero(va + size, PAGE_SIZE - size);
5792 zfs_unmap_page(sf);
5793 if (error != 0)
5794 break;
5795 }
5796
5797 zfs_vmobject_wlock(object);
5798
5799 for (i = reqstart; i < reqstart + reqsize; i++) {
5800 if (!error)
5801 m[i]->valid = VM_PAGE_BITS_ALL;
5802 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5803 if (i != reqpage)
5804 vm_page_readahead_finish(m[i]);
5805 }
5806
5807 zfs_vmobject_wunlock(object);
5808
5809 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5810 ZFS_EXIT(zfsvfs);
5811 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok);
5812}
5813
5814static int
5815zfs_freebsd_getpages(ap)
5816 struct vop_getpages_args /* {
5817 struct vnode *a_vp;
5818 vm_page_t *a_m;
5819 int a_count;
5820 int a_reqpage;
5821 vm_ooffset_t a_offset;
5822 } */ *ap;
5823{
5824
5825 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5826}
5827
5828static int
5829zfs_freebsd_bmap(ap)
5830 struct vop_bmap_args /* {
5831 struct vnode *a_vp;
5832 daddr_t a_bn;
5833 struct bufobj **a_bop;
5834 daddr_t *a_bnp;
5835 int *a_runp;
5836 int *a_runb;
5837 } */ *ap;
5838{
5839
5840 if (ap->a_bop != NULL)
5841 *ap->a_bop = &ap->a_vp->v_bufobj;
5842 if (ap->a_bnp != NULL)
5843 *ap->a_bnp = ap->a_bn;
5844 if (ap->a_runp != NULL)
5845 *ap->a_runp = 0;
5846 if (ap->a_runb != NULL)
5847 *ap->a_runb = 0;
5848
5849 return (0);
5850}
5851
5852static int
5853zfs_freebsd_open(ap)
5854 struct vop_open_args /* {
5855 struct vnode *a_vp;
5856 int a_mode;
5857 struct ucred *a_cred;
5858 struct thread *a_td;
5859 } */ *ap;
5860{
5861 vnode_t *vp = ap->a_vp;
5862 znode_t *zp = VTOZ(vp);
5863 int error;
5864
5865 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5866 if (error == 0)
5867 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5868 return (error);
5869}
5870
5871static int
5872zfs_freebsd_close(ap)
5873 struct vop_close_args /* {
5874 struct vnode *a_vp;
5875 int a_fflag;
5876 struct ucred *a_cred;
5877 struct thread *a_td;
5878 } */ *ap;
5879{
5880
5881 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5882}
5883
5884static int
5885zfs_freebsd_ioctl(ap)
5886 struct vop_ioctl_args /* {
5887 struct vnode *a_vp;
5888 u_long a_command;
5889 caddr_t a_data;
5890 int a_fflag;
5891 struct ucred *cred;
5892 struct thread *td;
5893 } */ *ap;
5894{
5895
5896 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5897 ap->a_fflag, ap->a_cred, NULL, NULL));
5898}
5899
5900static int
5901zfs_freebsd_read(ap)
5902 struct vop_read_args /* {
5903 struct vnode *a_vp;
5904 struct uio *a_uio;
5905 int a_ioflag;
5906 struct ucred *a_cred;
5907 } */ *ap;
5908{
5909
5910 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5911 ap->a_cred, NULL));
5912}
5913
5914static int
5915zfs_freebsd_write(ap)
5916 struct vop_write_args /* {
5917 struct vnode *a_vp;
5918 struct uio *a_uio;
5919 int a_ioflag;
5920 struct ucred *a_cred;
5921 } */ *ap;
5922{
5923
5924 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5925 ap->a_cred, NULL));
5926}
5927
5928static int
5929zfs_freebsd_access(ap)
5930 struct vop_access_args /* {
5931 struct vnode *a_vp;
5932 accmode_t a_accmode;
5933 struct ucred *a_cred;
5934 struct thread *a_td;
5935 } */ *ap;
5936{
5937 vnode_t *vp = ap->a_vp;
5938 znode_t *zp = VTOZ(vp);
5939 accmode_t accmode;
5940 int error = 0;
5941
5942 /*
5943 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5944 */
5945 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5946 if (accmode != 0)
5947 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5948
5949 /*
5950 * VADMIN has to be handled by vaccess().
5951 */
5952 if (error == 0) {
5953 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5954 if (accmode != 0) {
5955 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5956 zp->z_gid, accmode, ap->a_cred, NULL);
5957 }
5958 }
5959
5960 /*
5961 * For VEXEC, ensure that at least one execute bit is set for
5962 * non-directories.
5963 */
5964 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
5965 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
5966 error = EACCES;
5967 }
5968
5969 return (error);
5970}
5971
5972static int
5973zfs_freebsd_lookup(ap)
5974 struct vop_lookup_args /* {
5975 struct vnode *a_dvp;
5976 struct vnode **a_vpp;
5977 struct componentname *a_cnp;
5978 } */ *ap;
5979{
5980 struct componentname *cnp = ap->a_cnp;
5981 char nm[NAME_MAX + 1];
5982
5983 ASSERT(cnp->cn_namelen < sizeof(nm));
5984 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5985
5986 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5987 cnp->cn_cred, cnp->cn_thread, 0));
5988}
5989
5990static int
5991zfs_freebsd_create(ap)
5992 struct vop_create_args /* {
5993 struct vnode *a_dvp;
5994 struct vnode **a_vpp;
5995 struct componentname *a_cnp;
5996 struct vattr *a_vap;
5997 } */ *ap;
5998{
5999 struct componentname *cnp = ap->a_cnp;
6000 vattr_t *vap = ap->a_vap;
6001 int mode;
6002
6003 ASSERT(cnp->cn_flags & SAVENAME);
6004
6005 vattr_init_mask(vap);
6006 mode = vap->va_mode & ALLPERMS;
6007
6008 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
6009 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
6010}
6011
6012static int
6013zfs_freebsd_remove(ap)
6014 struct vop_remove_args /* {
6015 struct vnode *a_dvp;
6016 struct vnode *a_vp;
6017 struct componentname *a_cnp;
6018 } */ *ap;
6019{
6020
6021 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6022
6023 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6024 ap->a_cnp->cn_cred, NULL, 0));
6025}
6026
6027static int
6028zfs_freebsd_mkdir(ap)
6029 struct vop_mkdir_args /* {
6030 struct vnode *a_dvp;
6031 struct vnode **a_vpp;
6032 struct componentname *a_cnp;
6033 struct vattr *a_vap;
6034 } */ *ap;
6035{
6036 vattr_t *vap = ap->a_vap;
6037
6038 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6039
6040 vattr_init_mask(vap);
6041
6042 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6043 ap->a_cnp->cn_cred, NULL, 0, NULL));
6044}
6045
6046static int
6047zfs_freebsd_rmdir(ap)
6048 struct vop_rmdir_args /* {
6049 struct vnode *a_dvp;
6050 struct vnode *a_vp;
6051 struct componentname *a_cnp;
6052 } */ *ap;
6053{
6054 struct componentname *cnp = ap->a_cnp;
6055
6056 ASSERT(cnp->cn_flags & SAVENAME);
6057
6058 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6059}
6060
6061static int
6062zfs_freebsd_readdir(ap)
6063 struct vop_readdir_args /* {
6064 struct vnode *a_vp;
6065 struct uio *a_uio;
6066 struct ucred *a_cred;
6067 int *a_eofflag;
6068 int *a_ncookies;
6069 u_long **a_cookies;
6070 } */ *ap;
6071{
6072
6073 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6074 ap->a_ncookies, ap->a_cookies));
6075}
6076
6077static int
6078zfs_freebsd_fsync(ap)
6079 struct vop_fsync_args /* {
6080 struct vnode *a_vp;
6081 int a_waitfor;
6082 struct thread *a_td;
6083 } */ *ap;
6084{
6085
6086 vop_stdfsync(ap);
6087 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6088}
6089
6090static int
6091zfs_freebsd_getattr(ap)
6092 struct vop_getattr_args /* {
6093 struct vnode *a_vp;
6094 struct vattr *a_vap;
6095 struct ucred *a_cred;
6096 } */ *ap;
6097{
6098 vattr_t *vap = ap->a_vap;
6099 xvattr_t xvap;
6100 u_long fflags = 0;
6101 int error;
6102
6103 xva_init(&xvap);
6104 xvap.xva_vattr = *vap;
6105 xvap.xva_vattr.va_mask |= AT_XVATTR;
6106
6107 /* Convert chflags into ZFS-type flags. */
6108 /* XXX: what about SF_SETTABLE?. */
6109 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6110 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6111 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6112 XVA_SET_REQ(&xvap, XAT_NODUMP);
6113 XVA_SET_REQ(&xvap, XAT_READONLY);
6114 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
6115 XVA_SET_REQ(&xvap, XAT_SYSTEM);
6116 XVA_SET_REQ(&xvap, XAT_HIDDEN);
6117 XVA_SET_REQ(&xvap, XAT_REPARSE);
6118 XVA_SET_REQ(&xvap, XAT_OFFLINE);
6119 XVA_SET_REQ(&xvap, XAT_SPARSE);
6120
6121 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6122 if (error != 0)
6123 return (error);
6124
6125 /* Convert ZFS xattr into chflags. */
6126#define FLAG_CHECK(fflag, xflag, xfield) do { \
6127 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6128 fflags |= (fflag); \
6129} while (0)
6130 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6131 xvap.xva_xoptattrs.xoa_immutable);
6132 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6133 xvap.xva_xoptattrs.xoa_appendonly);
6134 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6135 xvap.xva_xoptattrs.xoa_nounlink);
6136 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
6137 xvap.xva_xoptattrs.xoa_archive);
6138 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6139 xvap.xva_xoptattrs.xoa_nodump);
6140 FLAG_CHECK(UF_READONLY, XAT_READONLY,
6141 xvap.xva_xoptattrs.xoa_readonly);
6142 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
6143 xvap.xva_xoptattrs.xoa_system);
6144 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
6145 xvap.xva_xoptattrs.xoa_hidden);
6146 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
6147 xvap.xva_xoptattrs.xoa_reparse);
6148 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
6149 xvap.xva_xoptattrs.xoa_offline);
6150 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
6151 xvap.xva_xoptattrs.xoa_sparse);
6152
6153#undef FLAG_CHECK
6154 *vap = xvap.xva_vattr;
6155 vap->va_flags = fflags;
6156 return (0);
6157}
6158
6159static int
6160zfs_freebsd_setattr(ap)
6161 struct vop_setattr_args /* {
6162 struct vnode *a_vp;
6163 struct vattr *a_vap;
6164 struct ucred *a_cred;
6165 } */ *ap;
6166{
6167 vnode_t *vp = ap->a_vp;
6168 vattr_t *vap = ap->a_vap;
6169 cred_t *cred = ap->a_cred;
6170 xvattr_t xvap;
6171 u_long fflags;
6172 uint64_t zflags;
6173
6174 vattr_init_mask(vap);
6175 vap->va_mask &= ~AT_NOSET;
6176
6177 xva_init(&xvap);
6178 xvap.xva_vattr = *vap;
6179
6180 zflags = VTOZ(vp)->z_pflags;
6181
6182 if (vap->va_flags != VNOVAL) {
6183 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6184 int error;
6185
6186 if (zfsvfs->z_use_fuids == B_FALSE)
6187 return (EOPNOTSUPP);
6188
6189 fflags = vap->va_flags;
6190 /*
6191 * XXX KDM
6192 * We need to figure out whether it makes sense to allow
6193 * UF_REPARSE through, since we don't really have other
6194 * facilities to handle reparse points and zfs_setattr()
6195 * doesn't currently allow setting that attribute anyway.
6196 */
6197 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
6198 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
6199 UF_OFFLINE|UF_SPARSE)) != 0)
6200 return (EOPNOTSUPP);
6201 /*
6202 * Unprivileged processes are not permitted to unset system
6203 * flags, or modify flags if any system flags are set.
6204 * Privileged non-jail processes may not modify system flags
6205 * if securelevel > 0 and any existing system flags are set.
6206 * Privileged jail processes behave like privileged non-jail
6207 * processes if the security.jail.chflags_allowed sysctl is
6208 * is non-zero; otherwise, they behave like unprivileged
6209 * processes.
6210 */
6211 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6212 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6213 if (zflags &
6214 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6215 error = securelevel_gt(cred, 0);
6216 if (error != 0)
6217 return (error);
6218 }
6219 } else {
6220 /*
6221 * Callers may only modify the file flags on objects they
6222 * have VADMIN rights for.
6223 */
6224 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6225 return (error);
6226 if (zflags &
6227 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6228 return (EPERM);
6229 }
6230 if (fflags &
6231 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6232 return (EPERM);
6233 }
6234 }
6235
6236#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6237 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6238 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6239 XVA_SET_REQ(&xvap, (xflag)); \
6240 (xfield) = ((fflags & (fflag)) != 0); \
6241 } \
6242} while (0)
6243 /* Convert chflags into ZFS-type flags. */
6244 /* XXX: what about SF_SETTABLE?. */
6245 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6246 xvap.xva_xoptattrs.xoa_immutable);
6247 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6248 xvap.xva_xoptattrs.xoa_appendonly);
6249 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6250 xvap.xva_xoptattrs.xoa_nounlink);
6251 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
6252 xvap.xva_xoptattrs.xoa_archive);
6253 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6254 xvap.xva_xoptattrs.xoa_nodump);
6255 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
6256 xvap.xva_xoptattrs.xoa_readonly);
6257 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
6258 xvap.xva_xoptattrs.xoa_system);
6259 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
6260 xvap.xva_xoptattrs.xoa_hidden);
6261 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
6262 xvap.xva_xoptattrs.xoa_hidden);
6263 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
6264 xvap.xva_xoptattrs.xoa_offline);
6265 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
6266 xvap.xva_xoptattrs.xoa_sparse);
6267#undef FLAG_CHANGE
6268 }
6269 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6270}
6271
6272static int
6273zfs_freebsd_rename(ap)
6274 struct vop_rename_args /* {
6275 struct vnode *a_fdvp;
6276 struct vnode *a_fvp;
6277 struct componentname *a_fcnp;
6278 struct vnode *a_tdvp;
6279 struct vnode *a_tvp;
6280 struct componentname *a_tcnp;
6281 } */ *ap;
6282{
6283 vnode_t *fdvp = ap->a_fdvp;
6284 vnode_t *fvp = ap->a_fvp;
6285 vnode_t *tdvp = ap->a_tdvp;
6286 vnode_t *tvp = ap->a_tvp;
6287 int error;
6288
6289 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6290 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6291
6292 /*
6293 * Check for cross-device rename.
6294 */
6295 if ((fdvp->v_mount != tdvp->v_mount) ||
6296 (tvp && (fdvp->v_mount != tvp->v_mount)))
6297 error = EXDEV;
6298 else
6299 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6300 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6301 if (tdvp == tvp)
6302 VN_RELE(tdvp);
6303 else
6304 VN_URELE(tdvp);
6305 if (tvp)
6306 VN_URELE(tvp);
6307 VN_RELE(fdvp);
6308 VN_RELE(fvp);
6309
6310 return (error);
6311}
6312
6313static int
6314zfs_freebsd_symlink(ap)
6315 struct vop_symlink_args /* {
6316 struct vnode *a_dvp;
6317 struct vnode **a_vpp;
6318 struct componentname *a_cnp;
6319 struct vattr *a_vap;
6320 char *a_target;
6321 } */ *ap;
6322{
6323 struct componentname *cnp = ap->a_cnp;
6324 vattr_t *vap = ap->a_vap;
6325
6326 ASSERT(cnp->cn_flags & SAVENAME);
6327
6328 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6329 vattr_init_mask(vap);
6330
6331 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6332 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6333}
6334
6335static int
6336zfs_freebsd_readlink(ap)
6337 struct vop_readlink_args /* {
6338 struct vnode *a_vp;
6339 struct uio *a_uio;
6340 struct ucred *a_cred;
6341 } */ *ap;
6342{
6343
6344 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6345}
6346
6347static int
6348zfs_freebsd_link(ap)
6349 struct vop_link_args /* {
6350 struct vnode *a_tdvp;
6351 struct vnode *a_vp;
6352 struct componentname *a_cnp;
6353 } */ *ap;
6354{
6355 struct componentname *cnp = ap->a_cnp;
6356 vnode_t *vp = ap->a_vp;
6357 vnode_t *tdvp = ap->a_tdvp;
6358
6359 if (tdvp->v_mount != vp->v_mount)
6360 return (EXDEV);
6361
6362 ASSERT(cnp->cn_flags & SAVENAME);
6363
6364 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6365}
6366
6367static int
6368zfs_freebsd_inactive(ap)
6369 struct vop_inactive_args /* {
6370 struct vnode *a_vp;
6371 struct thread *a_td;
6372 } */ *ap;
6373{
6374 vnode_t *vp = ap->a_vp;
6375
6376 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6377 return (0);
6378}
6379
6380static int
6381zfs_freebsd_reclaim(ap)
6382 struct vop_reclaim_args /* {
6383 struct vnode *a_vp;
6384 struct thread *a_td;
6385 } */ *ap;
6386{
6387 vnode_t *vp = ap->a_vp;
6388 znode_t *zp = VTOZ(vp);
6389 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6390
6391 ASSERT(zp != NULL);
6392
6393 /* Destroy the vm object and flush associated pages. */
6394 vnode_destroy_vobject(vp);
6395
6396 /*
6397 * z_teardown_inactive_lock protects from a race with
6398 * zfs_znode_dmu_fini in zfsvfs_teardown during
6399 * force unmount.
6400 */
6401 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6402 if (zp->z_sa_hdl == NULL)
6403 zfs_znode_free(zp);
6404 else
6405 zfs_zinactive(zp);
6406 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6407
6408 vp->v_data = NULL;
6409 return (0);
6410}
6411
6412static int
6413zfs_freebsd_fid(ap)
6414 struct vop_fid_args /* {
6415 struct vnode *a_vp;
6416 struct fid *a_fid;
6417 } */ *ap;
6418{
6419
6420 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6421}
6422
6423static int
6424zfs_freebsd_pathconf(ap)
6425 struct vop_pathconf_args /* {
6426 struct vnode *a_vp;
6427 int a_name;
6428 register_t *a_retval;
6429 } */ *ap;
6430{
6431 ulong_t val;
6432 int error;
6433
6434 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6435 if (error == 0)
6436 *ap->a_retval = val;
6437 else if (error == EOPNOTSUPP)
6438 error = vop_stdpathconf(ap);
6439 return (error);
6440}
6441
6442static int
6443zfs_freebsd_fifo_pathconf(ap)
6444 struct vop_pathconf_args /* {
6445 struct vnode *a_vp;
6446 int a_name;
6447 register_t *a_retval;
6448 } */ *ap;
6449{
6450
6451 switch (ap->a_name) {
6452 case _PC_ACL_EXTENDED:
6453 case _PC_ACL_NFS4:
6454 case _PC_ACL_PATH_MAX:
6455 case _PC_MAC_PRESENT:
6456 return (zfs_freebsd_pathconf(ap));
6457 default:
6458 return (fifo_specops.vop_pathconf(ap));
6459 }
6460}
6461
6462/*
6463 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6464 * extended attribute name:
6465 *
6466 * NAMESPACE PREFIX
6467 * system freebsd:system:
6468 * user (none, can be used to access ZFS fsattr(5) attributes
6469 * created on Solaris)
6470 */
6471static int
6472zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6473 size_t size)
6474{
6475 const char *namespace, *prefix, *suffix;
6476
6477 /* We don't allow '/' character in attribute name. */
6478 if (strchr(name, '/') != NULL)
6479 return (EINVAL);
6480 /* We don't allow attribute names that start with "freebsd:" string. */
6481 if (strncmp(name, "freebsd:", 8) == 0)
6482 return (EINVAL);
6483
6484 bzero(attrname, size);
6485
6486 switch (attrnamespace) {
6487 case EXTATTR_NAMESPACE_USER:
6488#if 0
6489 prefix = "freebsd:";
6490 namespace = EXTATTR_NAMESPACE_USER_STRING;
6491 suffix = ":";
6492#else
6493 /*
6494 * This is the default namespace by which we can access all
6495 * attributes created on Solaris.
6496 */
6497 prefix = namespace = suffix = "";
6498#endif
6499 break;
6500 case EXTATTR_NAMESPACE_SYSTEM:
6501 prefix = "freebsd:";
6502 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6503 suffix = ":";
6504 break;
6505 case EXTATTR_NAMESPACE_EMPTY:
6506 default:
6507 return (EINVAL);
6508 }
6509 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6510 name) >= size) {
6511 return (ENAMETOOLONG);
6512 }
6513 return (0);
6514}
6515
6516/*
6517 * Vnode operating to retrieve a named extended attribute.
6518 */
6519static int
6520zfs_getextattr(struct vop_getextattr_args *ap)
6521/*
6522vop_getextattr {
6523 IN struct vnode *a_vp;
6524 IN int a_attrnamespace;
6525 IN const char *a_name;
6526 INOUT struct uio *a_uio;
6527 OUT size_t *a_size;
6528 IN struct ucred *a_cred;
6529 IN struct thread *a_td;
6530};
6531*/
6532{
6533 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6534 struct thread *td = ap->a_td;
6535 struct nameidata nd;
6536 char attrname[255];
6537 struct vattr va;
6538 vnode_t *xvp = NULL, *vp;
6539 int error, flags;
6540
6541 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6542 ap->a_cred, ap->a_td, VREAD);
6543 if (error != 0)
6544 return (error);
6545
6546 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6547 sizeof(attrname));
6548 if (error != 0)
6549 return (error);
6550
6551 ZFS_ENTER(zfsvfs);
6552
6553 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6554 LOOKUP_XATTR);
6555 if (error != 0) {
6556 ZFS_EXIT(zfsvfs);
6557 return (error);
6558 }
6559
6560 flags = FREAD;
6561 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6562 xvp, td);
6563 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6564 vp = nd.ni_vp;
6565 NDFREE(&nd, NDF_ONLY_PNBUF);
6566 if (error != 0) {
6567 ZFS_EXIT(zfsvfs);
6568 if (error == ENOENT)
6569 error = ENOATTR;
6570 return (error);
6571 }
6572
6573 if (ap->a_size != NULL) {
6574 error = VOP_GETATTR(vp, &va, ap->a_cred);
6575 if (error == 0)
6576 *ap->a_size = (size_t)va.va_size;
6577 } else if (ap->a_uio != NULL)
6578 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6579
6580 VOP_UNLOCK(vp, 0);
6581 vn_close(vp, flags, ap->a_cred, td);
6582 ZFS_EXIT(zfsvfs);
6583
6584 return (error);
6585}
6586
6587/*
6588 * Vnode operation to remove a named attribute.
6589 */
6590int
6591zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6592/*
6593vop_deleteextattr {
6594 IN struct vnode *a_vp;
6595 IN int a_attrnamespace;
6596 IN const char *a_name;
6597 IN struct ucred *a_cred;
6598 IN struct thread *a_td;
6599};
6600*/
6601{
6602 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6603 struct thread *td = ap->a_td;
6604 struct nameidata nd;
6605 char attrname[255];
6606 struct vattr va;
6607 vnode_t *xvp = NULL, *vp;
6608 int error, flags;
6609
6610 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6611 ap->a_cred, ap->a_td, VWRITE);
6612 if (error != 0)
6613 return (error);
6614
6615 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6616 sizeof(attrname));
6617 if (error != 0)
6618 return (error);
6619
6620 ZFS_ENTER(zfsvfs);
6621
6622 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6623 LOOKUP_XATTR);
6624 if (error != 0) {
6625 ZFS_EXIT(zfsvfs);
6626 return (error);
6627 }
6628
6629 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6630 UIO_SYSSPACE, attrname, xvp, td);
6631 error = namei(&nd);
6632 vp = nd.ni_vp;
6633 NDFREE(&nd, NDF_ONLY_PNBUF);
6634 if (error != 0) {
6635 ZFS_EXIT(zfsvfs);
6636 if (error == ENOENT)
6637 error = ENOATTR;
6638 return (error);
6639 }
6640 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6641
6642 vput(nd.ni_dvp);
6643 if (vp == nd.ni_dvp)
6644 vrele(vp);
6645 else
6646 vput(vp);
6647 ZFS_EXIT(zfsvfs);
6648
6649 return (error);
6650}
6651
6652/*
6653 * Vnode operation to set a named attribute.
6654 */
6655static int
6656zfs_setextattr(struct vop_setextattr_args *ap)
6657/*
6658vop_setextattr {
6659 IN struct vnode *a_vp;
6660 IN int a_attrnamespace;
6661 IN const char *a_name;
6662 INOUT struct uio *a_uio;
6663 IN struct ucred *a_cred;
6664 IN struct thread *a_td;
6665};
6666*/
6667{
6668 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6669 struct thread *td = ap->a_td;
6670 struct nameidata nd;
6671 char attrname[255];
6672 struct vattr va;
6673 vnode_t *xvp = NULL, *vp;
6674 int error, flags;
6675
6676 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6677 ap->a_cred, ap->a_td, VWRITE);
6678 if (error != 0)
6679 return (error);
6680
6681 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6682 sizeof(attrname));
6683 if (error != 0)
6684 return (error);
6685
6686 ZFS_ENTER(zfsvfs);
6687
6688 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6689 LOOKUP_XATTR | CREATE_XATTR_DIR);
6690 if (error != 0) {
6691 ZFS_EXIT(zfsvfs);
6692 return (error);
6693 }
6694
6695 flags = FFLAGS(O_WRONLY | O_CREAT);
6696 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6697 xvp, td);
6698 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6699 vp = nd.ni_vp;
6700 NDFREE(&nd, NDF_ONLY_PNBUF);
6701 if (error != 0) {
6702 ZFS_EXIT(zfsvfs);
6703 return (error);
6704 }
6705
6706 VATTR_NULL(&va);
6707 va.va_size = 0;
6708 error = VOP_SETATTR(vp, &va, ap->a_cred);
6709 if (error == 0)
6710 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6711
6712 VOP_UNLOCK(vp, 0);
6713 vn_close(vp, flags, ap->a_cred, td);
6714 ZFS_EXIT(zfsvfs);
6715
6716 return (error);
6717}
6718
6719/*
6720 * Vnode operation to retrieve extended attributes on a vnode.
6721 */
6722static int
6723zfs_listextattr(struct vop_listextattr_args *ap)
6724/*
6725vop_listextattr {
6726 IN struct vnode *a_vp;
6727 IN int a_attrnamespace;
6728 INOUT struct uio *a_uio;
6729 OUT size_t *a_size;
6730 IN struct ucred *a_cred;
6731 IN struct thread *a_td;
6732};
6733*/
6734{
6735 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6736 struct thread *td = ap->a_td;
6737 struct nameidata nd;
6738 char attrprefix[16];
6739 u_char dirbuf[sizeof(struct dirent)];
6740 struct dirent *dp;
6741 struct iovec aiov;
6742 struct uio auio, *uio = ap->a_uio;
6743 size_t *sizep = ap->a_size;
6744 size_t plen;
6745 vnode_t *xvp = NULL, *vp;
6746 int done, error, eof, pos;
6747
6748 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6749 ap->a_cred, ap->a_td, VREAD);
6750 if (error != 0)
6751 return (error);
6752
6753 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6754 sizeof(attrprefix));
6755 if (error != 0)
6756 return (error);
6757 plen = strlen(attrprefix);
6758
6759 ZFS_ENTER(zfsvfs);
6760
6761 if (sizep != NULL)
6762 *sizep = 0;
6763
6764 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6765 LOOKUP_XATTR);
6766 if (error != 0) {
6767 ZFS_EXIT(zfsvfs);
6768 /*
6769 * ENOATTR means that the EA directory does not yet exist,
6770 * i.e. there are no extended attributes there.
6771 */
6772 if (error == ENOATTR)
6773 error = 0;
6774 return (error);
6775 }
6776
6777 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6778 UIO_SYSSPACE, ".", xvp, td);
6779 error = namei(&nd);
6780 vp = nd.ni_vp;
6781 NDFREE(&nd, NDF_ONLY_PNBUF);
6782 if (error != 0) {
6783 ZFS_EXIT(zfsvfs);
6784 return (error);
6785 }
6786
6787 auio.uio_iov = &aiov;
6788 auio.uio_iovcnt = 1;
6789 auio.uio_segflg = UIO_SYSSPACE;
6790 auio.uio_td = td;
6791 auio.uio_rw = UIO_READ;
6792 auio.uio_offset = 0;
6793
6794 do {
6795 u_char nlen;
6796
6797 aiov.iov_base = (void *)dirbuf;
6798 aiov.iov_len = sizeof(dirbuf);
6799 auio.uio_resid = sizeof(dirbuf);
6800 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6801 done = sizeof(dirbuf) - auio.uio_resid;
6802 if (error != 0)
6803 break;
6804 for (pos = 0; pos < done;) {
6805 dp = (struct dirent *)(dirbuf + pos);
6806 pos += dp->d_reclen;
6807 /*
6808 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6809 * is what we get when attribute was created on Solaris.
6810 */
6811 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6812 continue;
6813 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6814 continue;
6815 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6816 continue;
6817 nlen = dp->d_namlen - plen;
6818 if (sizep != NULL)
6819 *sizep += 1 + nlen;
6820 else if (uio != NULL) {
6821 /*
6822 * Format of extattr name entry is one byte for
6823 * length and the rest for name.
6824 */
6825 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6826 if (error == 0) {
6827 error = uiomove(dp->d_name + plen, nlen,
6828 uio->uio_rw, uio);
6829 }
6830 if (error != 0)
6831 break;
6832 }
6833 }
6834 } while (!eof && error == 0);
6835
6836 vput(vp);
6837 ZFS_EXIT(zfsvfs);
6838
6839 return (error);
6840}
6841
6842int
6843zfs_freebsd_getacl(ap)
6844 struct vop_getacl_args /* {
6845 struct vnode *vp;
6846 acl_type_t type;
6847 struct acl *aclp;
6848 struct ucred *cred;
6849 struct thread *td;
6850 } */ *ap;
6851{
6852 int error;
6853 vsecattr_t vsecattr;
6854
6855 if (ap->a_type != ACL_TYPE_NFS4)
6856 return (EINVAL);
6857
6858 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6859 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6860 return (error);
6861
6862 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6863 if (vsecattr.vsa_aclentp != NULL)
6864 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6865
6866 return (error);
6867}
6868
6869int
6870zfs_freebsd_setacl(ap)
6871 struct vop_setacl_args /* {
6872 struct vnode *vp;
6873 acl_type_t type;
6874 struct acl *aclp;
6875 struct ucred *cred;
6876 struct thread *td;
6877 } */ *ap;
6878{
6879 int error;
6880 vsecattr_t vsecattr;
6881 int aclbsize; /* size of acl list in bytes */
6882 aclent_t *aaclp;
6883
6884 if (ap->a_type != ACL_TYPE_NFS4)
6885 return (EINVAL);
6886
6887 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6888 return (EINVAL);
6889
6890 /*
6891 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6892 * splitting every entry into two and appending "canonical six"
6893 * entries at the end. Don't allow for setting an ACL that would
6894 * cause chmod(2) to run out of ACL entries.
6895 */
6896 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6897 return (ENOSPC);
6898
6899 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6900 if (error != 0)
6901 return (error);
6902
6903 vsecattr.vsa_mask = VSA_ACE;
6904 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6905 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6906 aaclp = vsecattr.vsa_aclentp;
6907 vsecattr.vsa_aclentsz = aclbsize;
6908
6909 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6910 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6911 kmem_free(aaclp, aclbsize);
6912
6913 return (error);
6914}
6915
6916int
6917zfs_freebsd_aclcheck(ap)
6918 struct vop_aclcheck_args /* {
6919 struct vnode *vp;
6920 acl_type_t type;
6921 struct acl *aclp;
6922 struct ucred *cred;
6923 struct thread *td;
6924 } */ *ap;
6925{
6926
6927 return (EOPNOTSUPP);
6928}
6929
6930struct vop_vector zfs_vnodeops;
6931struct vop_vector zfs_fifoops;
6932struct vop_vector zfs_shareops;
6933
6934struct vop_vector zfs_vnodeops = {
6935 .vop_default = &default_vnodeops,
6936 .vop_inactive = zfs_freebsd_inactive,
6937 .vop_reclaim = zfs_freebsd_reclaim,
6938 .vop_access = zfs_freebsd_access,
6939#ifdef FREEBSD_NAMECACHE
6940 .vop_lookup = vfs_cache_lookup,
6941 .vop_cachedlookup = zfs_freebsd_lookup,
6942#else
6943 .vop_lookup = zfs_freebsd_lookup,
6944#endif
6945 .vop_getattr = zfs_freebsd_getattr,
6946 .vop_setattr = zfs_freebsd_setattr,
6947 .vop_create = zfs_freebsd_create,
6948 .vop_mknod = zfs_freebsd_create,
6949 .vop_mkdir = zfs_freebsd_mkdir,
6950 .vop_readdir = zfs_freebsd_readdir,
6951 .vop_fsync = zfs_freebsd_fsync,
6952 .vop_open = zfs_freebsd_open,
6953 .vop_close = zfs_freebsd_close,
6954 .vop_rmdir = zfs_freebsd_rmdir,
6955 .vop_ioctl = zfs_freebsd_ioctl,
6956 .vop_link = zfs_freebsd_link,
6957 .vop_symlink = zfs_freebsd_symlink,
6958 .vop_readlink = zfs_freebsd_readlink,
6959 .vop_read = zfs_freebsd_read,
6960 .vop_write = zfs_freebsd_write,
6961 .vop_remove = zfs_freebsd_remove,
6962 .vop_rename = zfs_freebsd_rename,
6963 .vop_pathconf = zfs_freebsd_pathconf,
6964 .vop_bmap = zfs_freebsd_bmap,
6965 .vop_fid = zfs_freebsd_fid,
6966 .vop_getextattr = zfs_getextattr,
6967 .vop_deleteextattr = zfs_deleteextattr,
6968 .vop_setextattr = zfs_setextattr,
6969 .vop_listextattr = zfs_listextattr,
6970 .vop_getacl = zfs_freebsd_getacl,
6971 .vop_setacl = zfs_freebsd_setacl,
6972 .vop_aclcheck = zfs_freebsd_aclcheck,
6973 .vop_getpages = zfs_freebsd_getpages,
6974};
6975
6976struct vop_vector zfs_fifoops = {
6977 .vop_default = &fifo_specops,
6978 .vop_fsync = zfs_freebsd_fsync,
6979 .vop_access = zfs_freebsd_access,
6980 .vop_getattr = zfs_freebsd_getattr,
6981 .vop_inactive = zfs_freebsd_inactive,
6982 .vop_read = VOP_PANIC,
6983 .vop_reclaim = zfs_freebsd_reclaim,
6984 .vop_setattr = zfs_freebsd_setattr,
6985 .vop_write = VOP_PANIC,
6986 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6987 .vop_fid = zfs_freebsd_fid,
6988 .vop_getacl = zfs_freebsd_getacl,
6989 .vop_setacl = zfs_freebsd_setacl,
6990 .vop_aclcheck = zfs_freebsd_aclcheck,
6991};
6992
6993/*
6994 * special share hidden files vnode operations template
6995 */
6996struct vop_vector zfs_shareops = {
6997 .vop_default = &default_vnodeops,
6998 .vop_access = zfs_freebsd_access,
6999 .vop_inactive = zfs_freebsd_inactive,
7000 .vop_reclaim = zfs_freebsd_reclaim,
7001 .vop_fid = zfs_freebsd_fid,
7002 .vop_pathconf = zfs_freebsd_pathconf,
7003};
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) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
111 * This is critical because we don't want to block while holding locks.
112 * Note, in particular, that if a lock is sometimes acquired before
113 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
114 * use a non-blocking assign can deadlock the system. The scenario:
115 *
116 * Thread A has grabbed a lock before calling dmu_tx_assign().
117 * Thread B is in an already-assigned tx, and blocks for this lock.
118 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
119 * forever, because the previous txg can't quiesce until B's tx commits.
120 *
121 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
122 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
123 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
124 * to indicate that this operation has already called dmu_tx_wait().
125 * This will ensure that we don't retry forever, waiting a short bit
126 * each time.
127 *
128 * (5) If the operation succeeded, generate the intent log entry for it
129 * before dropping locks. This ensures that the ordering of events
130 * in the intent log matches the order in which they actually occurred.
131 * During ZIL replay the zfs_log_* functions will update the sequence
132 * number to indicate the zil transaction has replayed.
133 *
134 * (6) At the end of each vnode op, the DMU tx must always commit,
135 * regardless of whether there were any errors.
136 *
137 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
138 * to ensure that synchronous semantics are provided when necessary.
139 *
140 * In general, this is how things should be ordered in each vnode op:
141 *
142 * ZFS_ENTER(zfsvfs); // exit if unmounted
143 * top:
144 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
145 * rw_enter(...); // grab any other locks you need
146 * tx = dmu_tx_create(...); // get DMU tx
147 * dmu_tx_hold_*(); // hold each object you might modify
148 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
149 * if (error) {
150 * rw_exit(...); // drop locks
151 * zfs_dirent_unlock(dl); // unlock directory entry
152 * VN_RELE(...); // release held vnodes
153 * if (error == ERESTART) {
154 * waited = B_TRUE;
155 * dmu_tx_wait(tx);
156 * dmu_tx_abort(tx);
157 * goto top;
158 * }
159 * dmu_tx_abort(tx); // abort DMU tx
160 * ZFS_EXIT(zfsvfs); // finished in zfs
161 * return (error); // really out of space
162 * }
163 * error = do_real_work(); // do whatever this VOP does
164 * if (error == 0)
165 * zfs_log_*(...); // on success, make ZIL entry
166 * dmu_tx_commit(tx); // commit DMU tx -- error or not
167 * rw_exit(...); // drop locks
168 * zfs_dirent_unlock(dl); // unlock directory entry
169 * VN_RELE(...); // release held vnodes
170 * zil_commit(zilog, foid); // synchronous when necessary
171 * ZFS_EXIT(zfsvfs); // finished in zfs
172 * return (error); // done, report error
173 */
174
175/* ARGSUSED */
176static int
177zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
178{
179 znode_t *zp = VTOZ(*vpp);
180 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
181
182 ZFS_ENTER(zfsvfs);
183 ZFS_VERIFY_ZP(zp);
184
185 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
186 ((flag & FAPPEND) == 0)) {
187 ZFS_EXIT(zfsvfs);
188 return (SET_ERROR(EPERM));
189 }
190
191 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
192 ZTOV(zp)->v_type == VREG &&
193 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
194 if (fs_vscan(*vpp, cr, 0) != 0) {
195 ZFS_EXIT(zfsvfs);
196 return (SET_ERROR(EACCES));
197 }
198 }
199
200 /* Keep a count of the synchronous opens in the znode */
201 if (flag & (FSYNC | FDSYNC))
202 atomic_inc_32(&zp->z_sync_cnt);
203
204 ZFS_EXIT(zfsvfs);
205 return (0);
206}
207
208/* ARGSUSED */
209static int
210zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
211 caller_context_t *ct)
212{
213 znode_t *zp = VTOZ(vp);
214 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
215
216 /*
217 * Clean up any locks held by this process on the vp.
218 */
219 cleanlocks(vp, ddi_get_pid(), 0);
220 cleanshares(vp, ddi_get_pid());
221
222 ZFS_ENTER(zfsvfs);
223 ZFS_VERIFY_ZP(zp);
224
225 /* Decrement the synchronous opens in the znode */
226 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
227 atomic_dec_32(&zp->z_sync_cnt);
228
229 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
230 ZTOV(zp)->v_type == VREG &&
231 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
232 VERIFY(fs_vscan(vp, cr, 1) == 0);
233
234 ZFS_EXIT(zfsvfs);
235 return (0);
236}
237
238/*
239 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
240 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
241 */
242static int
243zfs_holey(vnode_t *vp, u_long cmd, offset_t *off)
244{
245 znode_t *zp = VTOZ(vp);
246 uint64_t noff = (uint64_t)*off; /* new offset */
247 uint64_t file_sz;
248 int error;
249 boolean_t hole;
250
251 file_sz = zp->z_size;
252 if (noff >= file_sz) {
253 return (SET_ERROR(ENXIO));
254 }
255
256 if (cmd == _FIO_SEEK_HOLE)
257 hole = B_TRUE;
258 else
259 hole = B_FALSE;
260
261 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
262
263 /* end of file? */
264 if ((error == ESRCH) || (noff > file_sz)) {
265 /*
266 * Handle the virtual hole at the end of file.
267 */
268 if (hole) {
269 *off = file_sz;
270 return (0);
271 }
272 return (SET_ERROR(ENXIO));
273 }
274
275 if (noff < *off)
276 return (error);
277 *off = noff;
278 return (error);
279}
280
281/* ARGSUSED */
282static int
283zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred,
284 int *rvalp, caller_context_t *ct)
285{
286 offset_t off;
287 int error;
288 zfsvfs_t *zfsvfs;
289 znode_t *zp;
290
291 switch (com) {
292 case _FIOFFS:
293 return (0);
294
295 /*
296 * The following two ioctls are used by bfu. Faking out,
297 * necessary to avoid bfu errors.
298 */
299 case _FIOGDIO:
300 case _FIOSDIO:
301 return (0);
302
303 case _FIO_SEEK_DATA:
304 case _FIO_SEEK_HOLE:
305#ifdef sun
306 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
307 return (SET_ERROR(EFAULT));
308#else
309 off = *(offset_t *)data;
310#endif
311 zp = VTOZ(vp);
312 zfsvfs = zp->z_zfsvfs;
313 ZFS_ENTER(zfsvfs);
314 ZFS_VERIFY_ZP(zp);
315
316 /* offset parameter is in/out */
317 error = zfs_holey(vp, com, &off);
318 ZFS_EXIT(zfsvfs);
319 if (error)
320 return (error);
321#ifdef sun
322 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
323 return (SET_ERROR(EFAULT));
324#else
325 *(offset_t *)data = off;
326#endif
327 return (0);
328 }
329 return (SET_ERROR(ENOTTY));
330}
331
332static vm_page_t
333page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes)
334{
335 vm_object_t obj;
336 vm_page_t pp;
337 int64_t end;
338
339 /*
340 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE
341 * aligned boundaries, if the range is not aligned. As a result a
342 * DEV_BSIZE subrange with partially dirty data may get marked as clean.
343 * It may happen that all DEV_BSIZE subranges are marked clean and thus
344 * the whole page would be considred clean despite have some dirty data.
345 * For this reason we should shrink the range to DEV_BSIZE aligned
346 * boundaries before calling vm_page_clear_dirty.
347 */
348 end = rounddown2(off + nbytes, DEV_BSIZE);
349 off = roundup2(off, DEV_BSIZE);
350 nbytes = end - off;
351
352 obj = vp->v_object;
353 zfs_vmobject_assert_wlocked(obj);
354
355 for (;;) {
356 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
357 pp->valid) {
358 if (vm_page_xbusied(pp)) {
359 /*
360 * Reference the page before unlocking and
361 * sleeping so that the page daemon is less
362 * likely to reclaim it.
363 */
364 vm_page_reference(pp);
365 vm_page_lock(pp);
366 zfs_vmobject_wunlock(obj);
367 vm_page_busy_sleep(pp, "zfsmwb");
368 zfs_vmobject_wlock(obj);
369 continue;
370 }
371 vm_page_sbusy(pp);
372 } else if (pp == NULL) {
373 pp = vm_page_alloc(obj, OFF_TO_IDX(start),
374 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED |
375 VM_ALLOC_SBUSY);
376 } else {
377 ASSERT(pp != NULL && !pp->valid);
378 pp = NULL;
379 }
380
381 if (pp != NULL) {
382 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
383 vm_object_pip_add(obj, 1);
384 pmap_remove_write(pp);
385 if (nbytes != 0)
386 vm_page_clear_dirty(pp, off, nbytes);
387 }
388 break;
389 }
390 return (pp);
391}
392
393static void
394page_unbusy(vm_page_t pp)
395{
396
397 vm_page_sunbusy(pp);
398 vm_object_pip_subtract(pp->object, 1);
399}
400
401static vm_page_t
402page_hold(vnode_t *vp, int64_t start)
403{
404 vm_object_t obj;
405 vm_page_t pp;
406
407 obj = vp->v_object;
408 zfs_vmobject_assert_wlocked(obj);
409
410 for (;;) {
411 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL &&
412 pp->valid) {
413 if (vm_page_xbusied(pp)) {
414 /*
415 * Reference the page before unlocking and
416 * sleeping so that the page daemon is less
417 * likely to reclaim it.
418 */
419 vm_page_reference(pp);
420 vm_page_lock(pp);
421 zfs_vmobject_wunlock(obj);
422 vm_page_busy_sleep(pp, "zfsmwb");
423 zfs_vmobject_wlock(obj);
424 continue;
425 }
426
427 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL);
428 vm_page_lock(pp);
429 vm_page_hold(pp);
430 vm_page_unlock(pp);
431
432 } else
433 pp = NULL;
434 break;
435 }
436 return (pp);
437}
438
439static void
440page_unhold(vm_page_t pp)
441{
442
443 vm_page_lock(pp);
444 vm_page_unhold(pp);
445 vm_page_unlock(pp);
446}
447
448static caddr_t
449zfs_map_page(vm_page_t pp, struct sf_buf **sfp)
450{
451
452 *sfp = sf_buf_alloc(pp, 0);
453 return ((caddr_t)sf_buf_kva(*sfp));
454}
455
456static void
457zfs_unmap_page(struct sf_buf *sf)
458{
459
460 sf_buf_free(sf);
461}
462
463/*
464 * When a file is memory mapped, we must keep the IO data synchronized
465 * between the DMU cache and the memory mapped pages. What this means:
466 *
467 * On Write: If we find a memory mapped page, we write to *both*
468 * the page and the dmu buffer.
469 */
470static void
471update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid,
472 int segflg, dmu_tx_t *tx)
473{
474 vm_object_t obj;
475 struct sf_buf *sf;
476 caddr_t va;
477 int off;
478
479 ASSERT(vp->v_mount != NULL);
480 obj = vp->v_object;
481 ASSERT(obj != NULL);
482
483 off = start & PAGEOFFSET;
484 zfs_vmobject_wlock(obj);
485 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
486 vm_page_t pp;
487 int nbytes = imin(PAGESIZE - off, len);
488
489 if (segflg == UIO_NOCOPY) {
490 pp = vm_page_lookup(obj, OFF_TO_IDX(start));
491 KASSERT(pp != NULL,
492 ("zfs update_pages: NULL page in putpages case"));
493 KASSERT(off == 0,
494 ("zfs update_pages: unaligned data in putpages case"));
495 KASSERT(pp->valid == VM_PAGE_BITS_ALL,
496 ("zfs update_pages: invalid page in putpages case"));
497 KASSERT(vm_page_sbusied(pp),
498 ("zfs update_pages: unbusy page in putpages case"));
499 KASSERT(!pmap_page_is_write_mapped(pp),
500 ("zfs update_pages: writable page in putpages case"));
501 zfs_vmobject_wunlock(obj);
502
503 va = zfs_map_page(pp, &sf);
504 (void) dmu_write(os, oid, start, nbytes, va, tx);
505 zfs_unmap_page(sf);
506
507 zfs_vmobject_wlock(obj);
508 vm_page_undirty(pp);
509 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) {
510 zfs_vmobject_wunlock(obj);
511
512 va = zfs_map_page(pp, &sf);
513 (void) dmu_read(os, oid, start+off, nbytes,
514 va+off, DMU_READ_PREFETCH);;
515 zfs_unmap_page(sf);
516
517 zfs_vmobject_wlock(obj);
518 page_unbusy(pp);
519 }
520 len -= nbytes;
521 off = 0;
522 }
523 if (segflg != UIO_NOCOPY)
524 vm_object_pip_wakeupn(obj, 0);
525 zfs_vmobject_wunlock(obj);
526}
527
528/*
529 * Read with UIO_NOCOPY flag means that sendfile(2) requests
530 * ZFS to populate a range of page cache pages with data.
531 *
532 * NOTE: this function could be optimized to pre-allocate
533 * all pages in advance, drain exclusive busy on all of them,
534 * map them into contiguous KVA region and populate them
535 * in one single dmu_read() call.
536 */
537static int
538mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio)
539{
540 znode_t *zp = VTOZ(vp);
541 objset_t *os = zp->z_zfsvfs->z_os;
542 struct sf_buf *sf;
543 vm_object_t obj;
544 vm_page_t pp;
545 int64_t start;
546 caddr_t va;
547 int len = nbytes;
548 int off;
549 int error = 0;
550
551 ASSERT(uio->uio_segflg == UIO_NOCOPY);
552 ASSERT(vp->v_mount != NULL);
553 obj = vp->v_object;
554 ASSERT(obj != NULL);
555 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0);
556
557 zfs_vmobject_wlock(obj);
558 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) {
559 int bytes = MIN(PAGESIZE, len);
560
561 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY |
562 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY);
563 if (pp->valid == 0) {
564 zfs_vmobject_wunlock(obj);
565 va = zfs_map_page(pp, &sf);
566 error = dmu_read(os, zp->z_id, start, bytes, va,
567 DMU_READ_PREFETCH);
568 if (bytes != PAGESIZE && error == 0)
569 bzero(va + bytes, PAGESIZE - bytes);
570 zfs_unmap_page(sf);
571 zfs_vmobject_wlock(obj);
572 vm_page_sunbusy(pp);
573 vm_page_lock(pp);
574 if (error) {
575 if (pp->wire_count == 0 && pp->valid == 0 &&
576 !vm_page_busied(pp))
577 vm_page_free(pp);
578 } else {
579 pp->valid = VM_PAGE_BITS_ALL;
580 vm_page_activate(pp);
581 }
582 vm_page_unlock(pp);
583 } else
584 vm_page_sunbusy(pp);
585 if (error)
586 break;
587 uio->uio_resid -= bytes;
588 uio->uio_offset += bytes;
589 len -= bytes;
590 }
591 zfs_vmobject_wunlock(obj);
592 return (error);
593}
594
595/*
596 * When a file is memory mapped, we must keep the IO data synchronized
597 * between the DMU cache and the memory mapped pages. What this means:
598 *
599 * On Read: We "read" preferentially from memory mapped pages,
600 * else we default from the dmu buffer.
601 *
602 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
603 * the file is memory mapped.
604 */
605static int
606mappedread(vnode_t *vp, int nbytes, uio_t *uio)
607{
608 znode_t *zp = VTOZ(vp);
609 objset_t *os = zp->z_zfsvfs->z_os;
610 vm_object_t obj;
611 int64_t start;
612 caddr_t va;
613 int len = nbytes;
614 int off;
615 int error = 0;
616
617 ASSERT(vp->v_mount != NULL);
618 obj = vp->v_object;
619 ASSERT(obj != NULL);
620
621 start = uio->uio_loffset;
622 off = start & PAGEOFFSET;
623 zfs_vmobject_wlock(obj);
624 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
625 vm_page_t pp;
626 uint64_t bytes = MIN(PAGESIZE - off, len);
627
628 if (pp = page_hold(vp, start)) {
629 struct sf_buf *sf;
630 caddr_t va;
631
632 zfs_vmobject_wunlock(obj);
633 va = zfs_map_page(pp, &sf);
634 error = uiomove(va + off, bytes, UIO_READ, uio);
635 zfs_unmap_page(sf);
636 zfs_vmobject_wlock(obj);
637 page_unhold(pp);
638 } else {
639 zfs_vmobject_wunlock(obj);
640 error = dmu_read_uio(os, zp->z_id, uio, bytes);
641 zfs_vmobject_wlock(obj);
642 }
643 len -= bytes;
644 off = 0;
645 if (error)
646 break;
647 }
648 zfs_vmobject_wunlock(obj);
649 return (error);
650}
651
652offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
653
654/*
655 * Read bytes from specified file into supplied buffer.
656 *
657 * IN: vp - vnode of file to be read from.
658 * uio - structure supplying read location, range info,
659 * and return buffer.
660 * ioflag - SYNC flags; used to provide FRSYNC semantics.
661 * cr - credentials of caller.
662 * ct - caller context
663 *
664 * OUT: uio - updated offset and range, buffer filled.
665 *
666 * RETURN: 0 on success, error code on failure.
667 *
668 * Side Effects:
669 * vp - atime updated if byte count > 0
670 */
671/* ARGSUSED */
672static int
673zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
674{
675 znode_t *zp = VTOZ(vp);
676 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
677 objset_t *os;
678 ssize_t n, nbytes;
679 int error = 0;
680 rl_t *rl;
681 xuio_t *xuio = NULL;
682
683 ZFS_ENTER(zfsvfs);
684 ZFS_VERIFY_ZP(zp);
685 os = zfsvfs->z_os;
686
687 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
688 ZFS_EXIT(zfsvfs);
689 return (SET_ERROR(EACCES));
690 }
691
692 /*
693 * Validate file offset
694 */
695 if (uio->uio_loffset < (offset_t)0) {
696 ZFS_EXIT(zfsvfs);
697 return (SET_ERROR(EINVAL));
698 }
699
700 /*
701 * Fasttrack empty reads
702 */
703 if (uio->uio_resid == 0) {
704 ZFS_EXIT(zfsvfs);
705 return (0);
706 }
707
708 /*
709 * Check for mandatory locks
710 */
711 if (MANDMODE(zp->z_mode)) {
712 if (error = chklock(vp, FREAD,
713 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
714 ZFS_EXIT(zfsvfs);
715 return (error);
716 }
717 }
718
719 /*
720 * If we're in FRSYNC mode, sync out this znode before reading it.
721 */
722 if (zfsvfs->z_log &&
723 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS))
724 zil_commit(zfsvfs->z_log, zp->z_id);
725
726 /*
727 * Lock the range against changes.
728 */
729 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
730
731 /*
732 * If we are reading past end-of-file we can skip
733 * to the end; but we might still need to set atime.
734 */
735 if (uio->uio_loffset >= zp->z_size) {
736 error = 0;
737 goto out;
738 }
739
740 ASSERT(uio->uio_loffset < zp->z_size);
741 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
742
743#ifdef sun
744 if ((uio->uio_extflg == UIO_XUIO) &&
745 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
746 int nblk;
747 int blksz = zp->z_blksz;
748 uint64_t offset = uio->uio_loffset;
749
750 xuio = (xuio_t *)uio;
751 if ((ISP2(blksz))) {
752 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
753 blksz)) / blksz;
754 } else {
755 ASSERT(offset + n <= blksz);
756 nblk = 1;
757 }
758 (void) dmu_xuio_init(xuio, nblk);
759
760 if (vn_has_cached_data(vp)) {
761 /*
762 * For simplicity, we always allocate a full buffer
763 * even if we only expect to read a portion of a block.
764 */
765 while (--nblk >= 0) {
766 (void) dmu_xuio_add(xuio,
767 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
768 blksz), 0, blksz);
769 }
770 }
771 }
772#endif /* sun */
773
774 while (n > 0) {
775 nbytes = MIN(n, zfs_read_chunk_size -
776 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
777
778#ifdef __FreeBSD__
779 if (uio->uio_segflg == UIO_NOCOPY)
780 error = mappedread_sf(vp, nbytes, uio);
781 else
782#endif /* __FreeBSD__ */
783 if (vn_has_cached_data(vp))
784 error = mappedread(vp, nbytes, uio);
785 else
786 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
787 if (error) {
788 /* convert checksum errors into IO errors */
789 if (error == ECKSUM)
790 error = SET_ERROR(EIO);
791 break;
792 }
793
794 n -= nbytes;
795 }
796out:
797 zfs_range_unlock(rl);
798
799 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
800 ZFS_EXIT(zfsvfs);
801 return (error);
802}
803
804/*
805 * Write the bytes to a file.
806 *
807 * IN: vp - vnode of file to be written to.
808 * uio - structure supplying write location, range info,
809 * and data buffer.
810 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
811 * set if in append mode.
812 * cr - credentials of caller.
813 * ct - caller context (NFS/CIFS fem monitor only)
814 *
815 * OUT: uio - updated offset and range.
816 *
817 * RETURN: 0 on success, error code on failure.
818 *
819 * Timestamps:
820 * vp - ctime|mtime updated if byte count > 0
821 */
822
823/* ARGSUSED */
824static int
825zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
826{
827 znode_t *zp = VTOZ(vp);
828 rlim64_t limit = MAXOFFSET_T;
829 ssize_t start_resid = uio->uio_resid;
830 ssize_t tx_bytes;
831 uint64_t end_size;
832 dmu_tx_t *tx;
833 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
834 zilog_t *zilog;
835 offset_t woff;
836 ssize_t n, nbytes;
837 rl_t *rl;
838 int max_blksz = zfsvfs->z_max_blksz;
839 int error = 0;
840 arc_buf_t *abuf;
841 iovec_t *aiov = NULL;
842 xuio_t *xuio = NULL;
843 int i_iov = 0;
844 int iovcnt = uio->uio_iovcnt;
845 iovec_t *iovp = uio->uio_iov;
846 int write_eof;
847 int count = 0;
848 sa_bulk_attr_t bulk[4];
849 uint64_t mtime[2], ctime[2];
850
851 /*
852 * Fasttrack empty write
853 */
854 n = start_resid;
855 if (n == 0)
856 return (0);
857
858 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
859 limit = MAXOFFSET_T;
860
861 ZFS_ENTER(zfsvfs);
862 ZFS_VERIFY_ZP(zp);
863
864 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
865 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
866 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
867 &zp->z_size, 8);
868 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
869 &zp->z_pflags, 8);
870
871 /*
872 * If immutable or not appending then return EPERM
873 */
874 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
875 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
876 (uio->uio_loffset < zp->z_size))) {
877 ZFS_EXIT(zfsvfs);
878 return (SET_ERROR(EPERM));
879 }
880
881 zilog = zfsvfs->z_log;
882
883 /*
884 * Validate file offset
885 */
886 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
887 if (woff < 0) {
888 ZFS_EXIT(zfsvfs);
889 return (SET_ERROR(EINVAL));
890 }
891
892 /*
893 * Check for mandatory locks before calling zfs_range_lock()
894 * in order to prevent a deadlock with locks set via fcntl().
895 */
896 if (MANDMODE((mode_t)zp->z_mode) &&
897 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
898 ZFS_EXIT(zfsvfs);
899 return (error);
900 }
901
902#ifdef sun
903 /*
904 * Pre-fault the pages to ensure slow (eg NFS) pages
905 * don't hold up txg.
906 * Skip this if uio contains loaned arc_buf.
907 */
908 if ((uio->uio_extflg == UIO_XUIO) &&
909 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
910 xuio = (xuio_t *)uio;
911 else
912 uio_prefaultpages(MIN(n, max_blksz), uio);
913#endif /* sun */
914
915 /*
916 * If in append mode, set the io offset pointer to eof.
917 */
918 if (ioflag & FAPPEND) {
919 /*
920 * Obtain an appending range lock to guarantee file append
921 * semantics. We reset the write offset once we have the lock.
922 */
923 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
924 woff = rl->r_off;
925 if (rl->r_len == UINT64_MAX) {
926 /*
927 * We overlocked the file because this write will cause
928 * the file block size to increase.
929 * Note that zp_size cannot change with this lock held.
930 */
931 woff = zp->z_size;
932 }
933 uio->uio_loffset = woff;
934 } else {
935 /*
936 * Note that if the file block size will change as a result of
937 * this write, then this range lock will lock the entire file
938 * so that we can re-write the block safely.
939 */
940 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
941 }
942
943 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) {
944 zfs_range_unlock(rl);
945 ZFS_EXIT(zfsvfs);
946 return (EFBIG);
947 }
948
949 if (woff >= limit) {
950 zfs_range_unlock(rl);
951 ZFS_EXIT(zfsvfs);
952 return (SET_ERROR(EFBIG));
953 }
954
955 if ((woff + n) > limit || woff > (limit - n))
956 n = limit - woff;
957
958 /* Will this write extend the file length? */
959 write_eof = (woff + n > zp->z_size);
960
961 end_size = MAX(zp->z_size, woff + n);
962
963 /*
964 * Write the file in reasonable size chunks. Each chunk is written
965 * in a separate transaction; this keeps the intent log records small
966 * and allows us to do more fine-grained space accounting.
967 */
968 while (n > 0) {
969 abuf = NULL;
970 woff = uio->uio_loffset;
971again:
972 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
973 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
974 if (abuf != NULL)
975 dmu_return_arcbuf(abuf);
976 error = SET_ERROR(EDQUOT);
977 break;
978 }
979
980 if (xuio && abuf == NULL) {
981 ASSERT(i_iov < iovcnt);
982 aiov = &iovp[i_iov];
983 abuf = dmu_xuio_arcbuf(xuio, i_iov);
984 dmu_xuio_clear(xuio, i_iov);
985 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
986 iovec_t *, aiov, arc_buf_t *, abuf);
987 ASSERT((aiov->iov_base == abuf->b_data) ||
988 ((char *)aiov->iov_base - (char *)abuf->b_data +
989 aiov->iov_len == arc_buf_size(abuf)));
990 i_iov++;
991 } else if (abuf == NULL && n >= max_blksz &&
992 woff >= zp->z_size &&
993 P2PHASE(woff, max_blksz) == 0 &&
994 zp->z_blksz == max_blksz) {
995 /*
996 * This write covers a full block. "Borrow" a buffer
997 * from the dmu so that we can fill it before we enter
998 * a transaction. This avoids the possibility of
999 * holding up the transaction if the data copy hangs
1000 * up on a pagefault (e.g., from an NFS server mapping).
1001 */
1002 size_t cbytes;
1003
1004 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
1005 max_blksz);
1006 ASSERT(abuf != NULL);
1007 ASSERT(arc_buf_size(abuf) == max_blksz);
1008 if (error = uiocopy(abuf->b_data, max_blksz,
1009 UIO_WRITE, uio, &cbytes)) {
1010 dmu_return_arcbuf(abuf);
1011 break;
1012 }
1013 ASSERT(cbytes == max_blksz);
1014 }
1015
1016 /*
1017 * Start a transaction.
1018 */
1019 tx = dmu_tx_create(zfsvfs->z_os);
1020 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1021 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
1022 zfs_sa_upgrade_txholds(tx, zp);
1023 error = dmu_tx_assign(tx, TXG_NOWAIT);
1024 if (error) {
1025 if (error == ERESTART) {
1026 dmu_tx_wait(tx);
1027 dmu_tx_abort(tx);
1028 goto again;
1029 }
1030 dmu_tx_abort(tx);
1031 if (abuf != NULL)
1032 dmu_return_arcbuf(abuf);
1033 break;
1034 }
1035
1036 /*
1037 * If zfs_range_lock() over-locked we grow the blocksize
1038 * and then reduce the lock range. This will only happen
1039 * on the first iteration since zfs_range_reduce() will
1040 * shrink down r_len to the appropriate size.
1041 */
1042 if (rl->r_len == UINT64_MAX) {
1043 uint64_t new_blksz;
1044
1045 if (zp->z_blksz > max_blksz) {
1046 ASSERT(!ISP2(zp->z_blksz));
1047 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
1048 } else {
1049 new_blksz = MIN(end_size, max_blksz);
1050 }
1051 zfs_grow_blocksize(zp, new_blksz, tx);
1052 zfs_range_reduce(rl, woff, n);
1053 }
1054
1055 /*
1056 * XXX - should we really limit each write to z_max_blksz?
1057 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
1058 */
1059 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
1060
1061 if (woff + nbytes > zp->z_size)
1062 vnode_pager_setsize(vp, woff + nbytes);
1063
1064 if (abuf == NULL) {
1065 tx_bytes = uio->uio_resid;
1066 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
1067 uio, nbytes, tx);
1068 tx_bytes -= uio->uio_resid;
1069 } else {
1070 tx_bytes = nbytes;
1071 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
1072 /*
1073 * If this is not a full block write, but we are
1074 * extending the file past EOF and this data starts
1075 * block-aligned, use assign_arcbuf(). Otherwise,
1076 * write via dmu_write().
1077 */
1078 if (tx_bytes < max_blksz && (!write_eof ||
1079 aiov->iov_base != abuf->b_data)) {
1080 ASSERT(xuio);
1081 dmu_write(zfsvfs->z_os, zp->z_id, woff,
1082 aiov->iov_len, aiov->iov_base, tx);
1083 dmu_return_arcbuf(abuf);
1084 xuio_stat_wbuf_copied();
1085 } else {
1086 ASSERT(xuio || tx_bytes == max_blksz);
1087 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
1088 woff, abuf, tx);
1089 }
1090 ASSERT(tx_bytes <= uio->uio_resid);
1091 uioskip(uio, tx_bytes);
1092 }
1093 if (tx_bytes && vn_has_cached_data(vp)) {
1094 update_pages(vp, woff, tx_bytes, zfsvfs->z_os,
1095 zp->z_id, uio->uio_segflg, tx);
1096 }
1097
1098 /*
1099 * If we made no progress, we're done. If we made even
1100 * partial progress, update the znode and ZIL accordingly.
1101 */
1102 if (tx_bytes == 0) {
1103 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
1104 (void *)&zp->z_size, sizeof (uint64_t), tx);
1105 dmu_tx_commit(tx);
1106 ASSERT(error != 0);
1107 break;
1108 }
1109
1110 /*
1111 * Clear Set-UID/Set-GID bits on successful write if not
1112 * privileged and at least one of the excute bits is set.
1113 *
1114 * It would be nice to to this after all writes have
1115 * been done, but that would still expose the ISUID/ISGID
1116 * to another app after the partial write is committed.
1117 *
1118 * Note: we don't call zfs_fuid_map_id() here because
1119 * user 0 is not an ephemeral uid.
1120 */
1121 mutex_enter(&zp->z_acl_lock);
1122 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
1123 (S_IXUSR >> 6))) != 0 &&
1124 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
1125 secpolicy_vnode_setid_retain(vp, cr,
1126 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
1127 uint64_t newmode;
1128 zp->z_mode &= ~(S_ISUID | S_ISGID);
1129 newmode = zp->z_mode;
1130 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
1131 (void *)&newmode, sizeof (uint64_t), tx);
1132 }
1133 mutex_exit(&zp->z_acl_lock);
1134
1135 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
1136 B_TRUE);
1137
1138 /*
1139 * Update the file size (zp_size) if it has changed;
1140 * account for possible concurrent updates.
1141 */
1142 while ((end_size = zp->z_size) < uio->uio_loffset) {
1143 (void) atomic_cas_64(&zp->z_size, end_size,
1144 uio->uio_loffset);
1145 ASSERT(error == 0);
1146 }
1147 /*
1148 * If we are replaying and eof is non zero then force
1149 * the file size to the specified eof. Note, there's no
1150 * concurrency during replay.
1151 */
1152 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
1153 zp->z_size = zfsvfs->z_replay_eof;
1154
1155 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1156
1157 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
1158 dmu_tx_commit(tx);
1159
1160 if (error != 0)
1161 break;
1162 ASSERT(tx_bytes == nbytes);
1163 n -= nbytes;
1164
1165#ifdef sun
1166 if (!xuio && n > 0)
1167 uio_prefaultpages(MIN(n, max_blksz), uio);
1168#endif /* sun */
1169 }
1170
1171 zfs_range_unlock(rl);
1172
1173 /*
1174 * If we're in replay mode, or we made no progress, return error.
1175 * Otherwise, it's at least a partial write, so it's successful.
1176 */
1177 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1178 ZFS_EXIT(zfsvfs);
1179 return (error);
1180 }
1181
1182 if (ioflag & (FSYNC | FDSYNC) ||
1183 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1184 zil_commit(zilog, zp->z_id);
1185
1186 ZFS_EXIT(zfsvfs);
1187 return (0);
1188}
1189
1190void
1191zfs_get_done(zgd_t *zgd, int error)
1192{
1193 znode_t *zp = zgd->zgd_private;
1194 objset_t *os = zp->z_zfsvfs->z_os;
1195
1196 if (zgd->zgd_db)
1197 dmu_buf_rele(zgd->zgd_db, zgd);
1198
1199 zfs_range_unlock(zgd->zgd_rl);
1200
1201 /*
1202 * Release the vnode asynchronously as we currently have the
1203 * txg stopped from syncing.
1204 */
1205 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1206
1207 if (error == 0 && zgd->zgd_bp)
1208 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1209
1210 kmem_free(zgd, sizeof (zgd_t));
1211}
1212
1213#ifdef DEBUG
1214static int zil_fault_io = 0;
1215#endif
1216
1217/*
1218 * Get data to generate a TX_WRITE intent log record.
1219 */
1220int
1221zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1222{
1223 zfsvfs_t *zfsvfs = arg;
1224 objset_t *os = zfsvfs->z_os;
1225 znode_t *zp;
1226 uint64_t object = lr->lr_foid;
1227 uint64_t offset = lr->lr_offset;
1228 uint64_t size = lr->lr_length;
1229 blkptr_t *bp = &lr->lr_blkptr;
1230 dmu_buf_t *db;
1231 zgd_t *zgd;
1232 int error = 0;
1233
1234 ASSERT(zio != NULL);
1235 ASSERT(size != 0);
1236
1237 /*
1238 * Nothing to do if the file has been removed
1239 */
1240 if (zfs_zget(zfsvfs, object, &zp) != 0)
1241 return (SET_ERROR(ENOENT));
1242 if (zp->z_unlinked) {
1243 /*
1244 * Release the vnode asynchronously as we currently have the
1245 * txg stopped from syncing.
1246 */
1247 VN_RELE_ASYNC(ZTOV(zp),
1248 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1249 return (SET_ERROR(ENOENT));
1250 }
1251
1252 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1253 zgd->zgd_zilog = zfsvfs->z_log;
1254 zgd->zgd_private = zp;
1255
1256 /*
1257 * Write records come in two flavors: immediate and indirect.
1258 * For small writes it's cheaper to store the data with the
1259 * log record (immediate); for large writes it's cheaper to
1260 * sync the data and get a pointer to it (indirect) so that
1261 * we don't have to write the data twice.
1262 */
1263 if (buf != NULL) { /* immediate write */
1264 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1265 /* test for truncation needs to be done while range locked */
1266 if (offset >= zp->z_size) {
1267 error = SET_ERROR(ENOENT);
1268 } else {
1269 error = dmu_read(os, object, offset, size, buf,
1270 DMU_READ_NO_PREFETCH);
1271 }
1272 ASSERT(error == 0 || error == ENOENT);
1273 } else { /* indirect write */
1274 /*
1275 * Have to lock the whole block to ensure when it's
1276 * written out and it's checksum is being calculated
1277 * that no one can change the data. We need to re-check
1278 * blocksize after we get the lock in case it's changed!
1279 */
1280 for (;;) {
1281 uint64_t blkoff;
1282 size = zp->z_blksz;
1283 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1284 offset -= blkoff;
1285 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1286 RL_READER);
1287 if (zp->z_blksz == size)
1288 break;
1289 offset += blkoff;
1290 zfs_range_unlock(zgd->zgd_rl);
1291 }
1292 /* test for truncation needs to be done while range locked */
1293 if (lr->lr_offset >= zp->z_size)
1294 error = SET_ERROR(ENOENT);
1295#ifdef DEBUG
1296 if (zil_fault_io) {
1297 error = SET_ERROR(EIO);
1298 zil_fault_io = 0;
1299 }
1300#endif
1301 if (error == 0)
1302 error = dmu_buf_hold(os, object, offset, zgd, &db,
1303 DMU_READ_NO_PREFETCH);
1304
1305 if (error == 0) {
1306 blkptr_t *obp = dmu_buf_get_blkptr(db);
1307 if (obp) {
1308 ASSERT(BP_IS_HOLE(bp));
1309 *bp = *obp;
1310 }
1311
1312 zgd->zgd_db = db;
1313 zgd->zgd_bp = bp;
1314
1315 ASSERT(db->db_offset == offset);
1316 ASSERT(db->db_size == size);
1317
1318 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1319 zfs_get_done, zgd);
1320 ASSERT(error || lr->lr_length <= zp->z_blksz);
1321
1322 /*
1323 * On success, we need to wait for the write I/O
1324 * initiated by dmu_sync() to complete before we can
1325 * release this dbuf. We will finish everything up
1326 * in the zfs_get_done() callback.
1327 */
1328 if (error == 0)
1329 return (0);
1330
1331 if (error == EALREADY) {
1332 lr->lr_common.lrc_txtype = TX_WRITE2;
1333 error = 0;
1334 }
1335 }
1336 }
1337
1338 zfs_get_done(zgd, error);
1339
1340 return (error);
1341}
1342
1343/*ARGSUSED*/
1344static int
1345zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1346 caller_context_t *ct)
1347{
1348 znode_t *zp = VTOZ(vp);
1349 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1350 int error;
1351
1352 ZFS_ENTER(zfsvfs);
1353 ZFS_VERIFY_ZP(zp);
1354
1355 if (flag & V_ACE_MASK)
1356 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1357 else
1358 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1359
1360 ZFS_EXIT(zfsvfs);
1361 return (error);
1362}
1363
1364/*
1365 * If vnode is for a device return a specfs vnode instead.
1366 */
1367static int
1368specvp_check(vnode_t **vpp, cred_t *cr)
1369{
1370 int error = 0;
1371
1372 if (IS_DEVVP(*vpp)) {
1373 struct vnode *svp;
1374
1375 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1376 VN_RELE(*vpp);
1377 if (svp == NULL)
1378 error = SET_ERROR(ENOSYS);
1379 *vpp = svp;
1380 }
1381 return (error);
1382}
1383
1384
1385/*
1386 * Lookup an entry in a directory, or an extended attribute directory.
1387 * If it exists, return a held vnode reference for it.
1388 *
1389 * IN: dvp - vnode of directory to search.
1390 * nm - name of entry to lookup.
1391 * pnp - full pathname to lookup [UNUSED].
1392 * flags - LOOKUP_XATTR set if looking for an attribute.
1393 * rdir - root directory vnode [UNUSED].
1394 * cr - credentials of caller.
1395 * ct - caller context
1396 * direntflags - directory lookup flags
1397 * realpnp - returned pathname.
1398 *
1399 * OUT: vpp - vnode of located entry, NULL if not found.
1400 *
1401 * RETURN: 0 on success, error code on failure.
1402 *
1403 * Timestamps:
1404 * NA
1405 */
1406/* ARGSUSED */
1407static int
1408zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp,
1409 int nameiop, cred_t *cr, kthread_t *td, int flags)
1410{
1411 znode_t *zdp = VTOZ(dvp);
1412 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1413 int error = 0;
1414 int *direntflags = NULL;
1415 void *realpnp = NULL;
1416
1417 /* fast path */
1418 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1419
1420 if (dvp->v_type != VDIR) {
1421 return (SET_ERROR(ENOTDIR));
1422 } else if (zdp->z_sa_hdl == NULL) {
1423 return (SET_ERROR(EIO));
1424 }
1425
1426 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1427 error = zfs_fastaccesschk_execute(zdp, cr);
1428 if (!error) {
1429 *vpp = dvp;
1430 VN_HOLD(*vpp);
1431 return (0);
1432 }
1433 return (error);
1434 } else {
1435 vnode_t *tvp = dnlc_lookup(dvp, nm);
1436
1437 if (tvp) {
1438 error = zfs_fastaccesschk_execute(zdp, cr);
1439 if (error) {
1440 VN_RELE(tvp);
1441 return (error);
1442 }
1443 if (tvp == DNLC_NO_VNODE) {
1444 VN_RELE(tvp);
1445 return (SET_ERROR(ENOENT));
1446 } else {
1447 *vpp = tvp;
1448 return (specvp_check(vpp, cr));
1449 }
1450 }
1451 }
1452 }
1453
1454 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1455
1456 ZFS_ENTER(zfsvfs);
1457 ZFS_VERIFY_ZP(zdp);
1458
1459 *vpp = NULL;
1460
1461 if (flags & LOOKUP_XATTR) {
1462#ifdef TODO
1463 /*
1464 * If the xattr property is off, refuse the lookup request.
1465 */
1466 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1467 ZFS_EXIT(zfsvfs);
1468 return (SET_ERROR(EINVAL));
1469 }
1470#endif
1471
1472 /*
1473 * We don't allow recursive attributes..
1474 * Maybe someday we will.
1475 */
1476 if (zdp->z_pflags & ZFS_XATTR) {
1477 ZFS_EXIT(zfsvfs);
1478 return (SET_ERROR(EINVAL));
1479 }
1480
1481 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1482 ZFS_EXIT(zfsvfs);
1483 return (error);
1484 }
1485
1486 /*
1487 * Do we have permission to get into attribute directory?
1488 */
1489
1490 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1491 B_FALSE, cr)) {
1492 VN_RELE(*vpp);
1493 *vpp = NULL;
1494 }
1495
1496 ZFS_EXIT(zfsvfs);
1497 return (error);
1498 }
1499
1500 if (dvp->v_type != VDIR) {
1501 ZFS_EXIT(zfsvfs);
1502 return (SET_ERROR(ENOTDIR));
1503 }
1504
1505 /*
1506 * Check accessibility of directory.
1507 */
1508
1509 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1510 ZFS_EXIT(zfsvfs);
1511 return (error);
1512 }
1513
1514 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1515 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1516 ZFS_EXIT(zfsvfs);
1517 return (SET_ERROR(EILSEQ));
1518 }
1519
1520 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1521 if (error == 0)
1522 error = specvp_check(vpp, cr);
1523
1524 /* Translate errors and add SAVENAME when needed. */
1525 if (cnp->cn_flags & ISLASTCN) {
1526 switch (nameiop) {
1527 case CREATE:
1528 case RENAME:
1529 if (error == ENOENT) {
1530 error = EJUSTRETURN;
1531 cnp->cn_flags |= SAVENAME;
1532 break;
1533 }
1534 /* FALLTHROUGH */
1535 case DELETE:
1536 if (error == 0)
1537 cnp->cn_flags |= SAVENAME;
1538 break;
1539 }
1540 }
1541 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) {
1542 int ltype = 0;
1543
1544 if (cnp->cn_flags & ISDOTDOT) {
1545 ltype = VOP_ISLOCKED(dvp);
1546 VOP_UNLOCK(dvp, 0);
1547 }
1548 ZFS_EXIT(zfsvfs);
1549 error = vn_lock(*vpp, cnp->cn_lkflags);
1550 if (cnp->cn_flags & ISDOTDOT)
1551 vn_lock(dvp, ltype | LK_RETRY);
1552 if (error != 0) {
1553 VN_RELE(*vpp);
1554 *vpp = NULL;
1555 return (error);
1556 }
1557 } else {
1558 ZFS_EXIT(zfsvfs);
1559 }
1560
1561#ifdef FREEBSD_NAMECACHE
1562 /*
1563 * Insert name into cache (as non-existent) if appropriate.
1564 */
1565 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
1566 cache_enter(dvp, *vpp, cnp);
1567 /*
1568 * Insert name into cache if appropriate.
1569 */
1570 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) {
1571 if (!(cnp->cn_flags & ISLASTCN) ||
1572 (nameiop != DELETE && nameiop != RENAME)) {
1573 cache_enter(dvp, *vpp, cnp);
1574 }
1575 }
1576#endif
1577
1578 return (error);
1579}
1580
1581/*
1582 * Attempt to create a new entry in a directory. If the entry
1583 * already exists, truncate the file if permissible, else return
1584 * an error. Return the vp of the created or trunc'd file.
1585 *
1586 * IN: dvp - vnode of directory to put new file entry in.
1587 * name - name of new file entry.
1588 * vap - attributes of new file.
1589 * excl - flag indicating exclusive or non-exclusive mode.
1590 * mode - mode to open file with.
1591 * cr - credentials of caller.
1592 * flag - large file flag [UNUSED].
1593 * ct - caller context
1594 * vsecp - ACL to be set
1595 *
1596 * OUT: vpp - vnode of created or trunc'd entry.
1597 *
1598 * RETURN: 0 on success, error code on failure.
1599 *
1600 * Timestamps:
1601 * dvp - ctime|mtime updated if new entry created
1602 * vp - ctime|mtime always, atime if new
1603 */
1604
1605/* ARGSUSED */
1606static int
1607zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode,
1608 vnode_t **vpp, cred_t *cr, kthread_t *td)
1609{
1610 znode_t *zp, *dzp = VTOZ(dvp);
1611 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1612 zilog_t *zilog;
1613 objset_t *os;
1614 zfs_dirlock_t *dl;
1615 dmu_tx_t *tx;
1616 int error;
1617 ksid_t *ksid;
1618 uid_t uid;
1619 gid_t gid = crgetgid(cr);
1620 zfs_acl_ids_t acl_ids;
1621 boolean_t fuid_dirtied;
1622 boolean_t have_acl = B_FALSE;
1623 boolean_t waited = B_FALSE;
1624 void *vsecp = NULL;
1625 int flag = 0;
1626
1627 /*
1628 * If we have an ephemeral id, ACL, or XVATTR then
1629 * make sure file system is at proper version
1630 */
1631
1632 ksid = crgetsid(cr, KSID_OWNER);
1633 if (ksid)
1634 uid = ksid_getid(ksid);
1635 else
1636 uid = crgetuid(cr);
1637
1638 if (zfsvfs->z_use_fuids == B_FALSE &&
1639 (vsecp || (vap->va_mask & AT_XVATTR) ||
1640 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1641 return (SET_ERROR(EINVAL));
1642
1643 ZFS_ENTER(zfsvfs);
1644 ZFS_VERIFY_ZP(dzp);
1645 os = zfsvfs->z_os;
1646 zilog = zfsvfs->z_log;
1647
1648 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1649 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1650 ZFS_EXIT(zfsvfs);
1651 return (SET_ERROR(EILSEQ));
1652 }
1653
1654 if (vap->va_mask & AT_XVATTR) {
1655 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
1656 crgetuid(cr), cr, vap->va_type)) != 0) {
1657 ZFS_EXIT(zfsvfs);
1658 return (error);
1659 }
1660 }
1661top:
1662 *vpp = NULL;
1663
1664 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr))
1665 vap->va_mode &= ~S_ISVTX;
1666
1667 if (*name == '\0') {
1668 /*
1669 * Null component name refers to the directory itself.
1670 */
1671 VN_HOLD(dvp);
1672 zp = dzp;
1673 dl = NULL;
1674 error = 0;
1675 } else {
1676 /* possible VN_HOLD(zp) */
1677 int zflg = 0;
1678
1679 if (flag & FIGNORECASE)
1680 zflg |= ZCILOOK;
1681
1682 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1683 NULL, NULL);
1684 if (error) {
1685 if (have_acl)
1686 zfs_acl_ids_free(&acl_ids);
1687 if (strcmp(name, "..") == 0)
1688 error = SET_ERROR(EISDIR);
1689 ZFS_EXIT(zfsvfs);
1690 return (error);
1691 }
1692 }
1693
1694 if (zp == NULL) {
1695 uint64_t txtype;
1696
1697 /*
1698 * Create a new file object and update the directory
1699 * to reference it.
1700 */
1701 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1702 if (have_acl)
1703 zfs_acl_ids_free(&acl_ids);
1704 goto out;
1705 }
1706
1707 /*
1708 * We only support the creation of regular files in
1709 * extended attribute directories.
1710 */
1711
1712 if ((dzp->z_pflags & ZFS_XATTR) &&
1713 (vap->va_type != VREG)) {
1714 if (have_acl)
1715 zfs_acl_ids_free(&acl_ids);
1716 error = SET_ERROR(EINVAL);
1717 goto out;
1718 }
1719
1720 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1721 cr, vsecp, &acl_ids)) != 0)
1722 goto out;
1723 have_acl = B_TRUE;
1724
1725 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1726 zfs_acl_ids_free(&acl_ids);
1727 error = SET_ERROR(EDQUOT);
1728 goto out;
1729 }
1730
1731 tx = dmu_tx_create(os);
1732
1733 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1734 ZFS_SA_BASE_ATTR_SIZE);
1735
1736 fuid_dirtied = zfsvfs->z_fuid_dirty;
1737 if (fuid_dirtied)
1738 zfs_fuid_txhold(zfsvfs, tx);
1739 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1740 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1741 if (!zfsvfs->z_use_sa &&
1742 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1743 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1744 0, acl_ids.z_aclp->z_acl_bytes);
1745 }
1746 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1747 if (error) {
1748 zfs_dirent_unlock(dl);
1749 if (error == ERESTART) {
1750 waited = B_TRUE;
1751 dmu_tx_wait(tx);
1752 dmu_tx_abort(tx);
1753 goto top;
1754 }
1755 zfs_acl_ids_free(&acl_ids);
1756 dmu_tx_abort(tx);
1757 ZFS_EXIT(zfsvfs);
1758 return (error);
1759 }
1760 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1761
1762 if (fuid_dirtied)
1763 zfs_fuid_sync(zfsvfs, tx);
1764
1765 (void) zfs_link_create(dl, zp, tx, ZNEW);
1766 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1767 if (flag & FIGNORECASE)
1768 txtype |= TX_CI;
1769 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1770 vsecp, acl_ids.z_fuidp, vap);
1771 zfs_acl_ids_free(&acl_ids);
1772 dmu_tx_commit(tx);
1773 } else {
1774 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1775
1776 if (have_acl)
1777 zfs_acl_ids_free(&acl_ids);
1778 have_acl = B_FALSE;
1779
1780 /*
1781 * A directory entry already exists for this name.
1782 */
1783 /*
1784 * Can't truncate an existing file if in exclusive mode.
1785 */
1786 if (excl == EXCL) {
1787 error = SET_ERROR(EEXIST);
1788 goto out;
1789 }
1790 /*
1791 * Can't open a directory for writing.
1792 */
1793 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1794 error = SET_ERROR(EISDIR);
1795 goto out;
1796 }
1797 /*
1798 * Verify requested access to file.
1799 */
1800 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1801 goto out;
1802 }
1803
1804 mutex_enter(&dzp->z_lock);
1805 dzp->z_seq++;
1806 mutex_exit(&dzp->z_lock);
1807
1808 /*
1809 * Truncate regular files if requested.
1810 */
1811 if ((ZTOV(zp)->v_type == VREG) &&
1812 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1813 /* we can't hold any locks when calling zfs_freesp() */
1814 zfs_dirent_unlock(dl);
1815 dl = NULL;
1816 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1817 if (error == 0) {
1818 vnevent_create(ZTOV(zp), ct);
1819 }
1820 }
1821 }
1822out:
1823 if (dl)
1824 zfs_dirent_unlock(dl);
1825
1826 if (error) {
1827 if (zp)
1828 VN_RELE(ZTOV(zp));
1829 } else {
1830 *vpp = ZTOV(zp);
1831 error = specvp_check(vpp, cr);
1832 }
1833
1834 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1835 zil_commit(zilog, 0);
1836
1837 ZFS_EXIT(zfsvfs);
1838 return (error);
1839}
1840
1841/*
1842 * Remove an entry from a directory.
1843 *
1844 * IN: dvp - vnode of directory to remove entry from.
1845 * name - name of entry to remove.
1846 * cr - credentials of caller.
1847 * ct - caller context
1848 * flags - case flags
1849 *
1850 * RETURN: 0 on success, error code on failure.
1851 *
1852 * Timestamps:
1853 * dvp - ctime|mtime
1854 * vp - ctime (if nlink > 0)
1855 */
1856
1857uint64_t null_xattr = 0;
1858
1859/*ARGSUSED*/
1860static int
1861zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1862 int flags)
1863{
1864 znode_t *zp, *dzp = VTOZ(dvp);
1865 znode_t *xzp;
1866 vnode_t *vp;
1867 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1868 zilog_t *zilog;
1869 uint64_t acl_obj, xattr_obj;
1870 uint64_t xattr_obj_unlinked = 0;
1871 uint64_t obj = 0;
1872 zfs_dirlock_t *dl;
1873 dmu_tx_t *tx;
1874 boolean_t may_delete_now, delete_now = FALSE;
1875 boolean_t unlinked, toobig = FALSE;
1876 uint64_t txtype;
1877 pathname_t *realnmp = NULL;
1878 pathname_t realnm;
1879 int error;
1880 int zflg = ZEXISTS;
1881 boolean_t waited = B_FALSE;
1882
1883 ZFS_ENTER(zfsvfs);
1884 ZFS_VERIFY_ZP(dzp);
1885 zilog = zfsvfs->z_log;
1886
1887 if (flags & FIGNORECASE) {
1888 zflg |= ZCILOOK;
1889 pn_alloc(&realnm);
1890 realnmp = &realnm;
1891 }
1892
1893top:
1894 xattr_obj = 0;
1895 xzp = NULL;
1896 /*
1897 * Attempt to lock directory; fail if entry doesn't exist.
1898 */
1899 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1900 NULL, realnmp)) {
1901 if (realnmp)
1902 pn_free(realnmp);
1903 ZFS_EXIT(zfsvfs);
1904 return (error);
1905 }
1906
1907 vp = ZTOV(zp);
1908
1909 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1910 goto out;
1911 }
1912
1913 /*
1914 * Need to use rmdir for removing directories.
1915 */
1916 if (vp->v_type == VDIR) {
1917 error = SET_ERROR(EPERM);
1918 goto out;
1919 }
1920
1921 vnevent_remove(vp, dvp, name, ct);
1922
1923 if (realnmp)
1924 dnlc_remove(dvp, realnmp->pn_buf);
1925 else
1926 dnlc_remove(dvp, name);
1927
1928 VI_LOCK(vp);
1929 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1930 VI_UNLOCK(vp);
1931
1932 /*
1933 * We may delete the znode now, or we may put it in the unlinked set;
1934 * it depends on whether we're the last link, and on whether there are
1935 * other holds on the vnode. So we dmu_tx_hold() the right things to
1936 * allow for either case.
1937 */
1938 obj = zp->z_id;
1939 tx = dmu_tx_create(zfsvfs->z_os);
1940 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1941 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1942 zfs_sa_upgrade_txholds(tx, zp);
1943 zfs_sa_upgrade_txholds(tx, dzp);
1944 if (may_delete_now) {
1945 toobig =
1946 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1947 /* if the file is too big, only hold_free a token amount */
1948 dmu_tx_hold_free(tx, zp->z_id, 0,
1949 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1950 }
1951
1952 /* are there any extended attributes? */
1953 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1954 &xattr_obj, sizeof (xattr_obj));
1955 if (error == 0 && xattr_obj) {
1956 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1957 ASSERT0(error);
1958 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1959 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1960 }
1961
1962 mutex_enter(&zp->z_lock);
1963 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1964 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1965 mutex_exit(&zp->z_lock);
1966
1967 /* charge as an update -- would be nice not to charge at all */
1968 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1969
1970 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1971 if (error) {
1972 zfs_dirent_unlock(dl);
1973 VN_RELE(vp);
1974 if (xzp)
1975 VN_RELE(ZTOV(xzp));
1976 if (error == ERESTART) {
1977 waited = B_TRUE;
1978 dmu_tx_wait(tx);
1979 dmu_tx_abort(tx);
1980 goto top;
1981 }
1982 if (realnmp)
1983 pn_free(realnmp);
1984 dmu_tx_abort(tx);
1985 ZFS_EXIT(zfsvfs);
1986 return (error);
1987 }
1988
1989 /*
1990 * Remove the directory entry.
1991 */
1992 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1993
1994 if (error) {
1995 dmu_tx_commit(tx);
1996 goto out;
1997 }
1998
1999 if (unlinked) {
2000
2001 /*
2002 * Hold z_lock so that we can make sure that the ACL obj
2003 * hasn't changed. Could have been deleted due to
2004 * zfs_sa_upgrade().
2005 */
2006 mutex_enter(&zp->z_lock);
2007 VI_LOCK(vp);
2008 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2009 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
2010 delete_now = may_delete_now && !toobig &&
2011 vp->v_count == 1 && !vn_has_cached_data(vp) &&
2012 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
2013 acl_obj;
2014 VI_UNLOCK(vp);
2015 }
2016
2017 if (delete_now) {
2018#ifdef __FreeBSD__
2019 panic("zfs_remove: delete_now branch taken");
2020#endif
2021 if (xattr_obj_unlinked) {
2022 ASSERT3U(xzp->z_links, ==, 2);
2023 mutex_enter(&xzp->z_lock);
2024 xzp->z_unlinked = 1;
2025 xzp->z_links = 0;
2026 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
2027 &xzp->z_links, sizeof (xzp->z_links), tx);
2028 ASSERT3U(error, ==, 0);
2029 mutex_exit(&xzp->z_lock);
2030 zfs_unlinked_add(xzp, tx);
2031
2032 if (zp->z_is_sa)
2033 error = sa_remove(zp->z_sa_hdl,
2034 SA_ZPL_XATTR(zfsvfs), tx);
2035 else
2036 error = sa_update(zp->z_sa_hdl,
2037 SA_ZPL_XATTR(zfsvfs), &null_xattr,
2038 sizeof (uint64_t), tx);
2039 ASSERT0(error);
2040 }
2041 VI_LOCK(vp);
2042 vp->v_count--;
2043 ASSERT0(vp->v_count);
2044 VI_UNLOCK(vp);
2045 mutex_exit(&zp->z_lock);
2046 zfs_znode_delete(zp, tx);
2047 } else if (unlinked) {
2048 mutex_exit(&zp->z_lock);
2049 zfs_unlinked_add(zp, tx);
2050#ifdef __FreeBSD__
2051 vp->v_vflag |= VV_NOSYNC;
2052#endif
2053 }
2054
2055 txtype = TX_REMOVE;
2056 if (flags & FIGNORECASE)
2057 txtype |= TX_CI;
2058 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
2059
2060 dmu_tx_commit(tx);
2061out:
2062 if (realnmp)
2063 pn_free(realnmp);
2064
2065 zfs_dirent_unlock(dl);
2066
2067 if (!delete_now)
2068 VN_RELE(vp);
2069 if (xzp)
2070 VN_RELE(ZTOV(xzp));
2071
2072 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2073 zil_commit(zilog, 0);
2074
2075 ZFS_EXIT(zfsvfs);
2076 return (error);
2077}
2078
2079/*
2080 * Create a new directory and insert it into dvp using the name
2081 * provided. Return a pointer to the inserted directory.
2082 *
2083 * IN: dvp - vnode of directory to add subdir to.
2084 * dirname - name of new directory.
2085 * vap - attributes of new directory.
2086 * cr - credentials of caller.
2087 * ct - caller context
2088 * flags - case flags
2089 * vsecp - ACL to be set
2090 *
2091 * OUT: vpp - vnode of created directory.
2092 *
2093 * RETURN: 0 on success, error code on failure.
2094 *
2095 * Timestamps:
2096 * dvp - ctime|mtime updated
2097 * vp - ctime|mtime|atime updated
2098 */
2099/*ARGSUSED*/
2100static int
2101zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
2102 caller_context_t *ct, int flags, vsecattr_t *vsecp)
2103{
2104 znode_t *zp, *dzp = VTOZ(dvp);
2105 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2106 zilog_t *zilog;
2107 zfs_dirlock_t *dl;
2108 uint64_t txtype;
2109 dmu_tx_t *tx;
2110 int error;
2111 int zf = ZNEW;
2112 ksid_t *ksid;
2113 uid_t uid;
2114 gid_t gid = crgetgid(cr);
2115 zfs_acl_ids_t acl_ids;
2116 boolean_t fuid_dirtied;
2117 boolean_t waited = B_FALSE;
2118
2119 ASSERT(vap->va_type == VDIR);
2120
2121 /*
2122 * If we have an ephemeral id, ACL, or XVATTR then
2123 * make sure file system is at proper version
2124 */
2125
2126 ksid = crgetsid(cr, KSID_OWNER);
2127 if (ksid)
2128 uid = ksid_getid(ksid);
2129 else
2130 uid = crgetuid(cr);
2131 if (zfsvfs->z_use_fuids == B_FALSE &&
2132 (vsecp || (vap->va_mask & AT_XVATTR) ||
2133 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
2134 return (SET_ERROR(EINVAL));
2135
2136 ZFS_ENTER(zfsvfs);
2137 ZFS_VERIFY_ZP(dzp);
2138 zilog = zfsvfs->z_log;
2139
2140 if (dzp->z_pflags & ZFS_XATTR) {
2141 ZFS_EXIT(zfsvfs);
2142 return (SET_ERROR(EINVAL));
2143 }
2144
2145 if (zfsvfs->z_utf8 && u8_validate(dirname,
2146 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2147 ZFS_EXIT(zfsvfs);
2148 return (SET_ERROR(EILSEQ));
2149 }
2150 if (flags & FIGNORECASE)
2151 zf |= ZCILOOK;
2152
2153 if (vap->va_mask & AT_XVATTR) {
2154 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap,
2155 crgetuid(cr), cr, vap->va_type)) != 0) {
2156 ZFS_EXIT(zfsvfs);
2157 return (error);
2158 }
2159 }
2160
2161 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
2162 vsecp, &acl_ids)) != 0) {
2163 ZFS_EXIT(zfsvfs);
2164 return (error);
2165 }
2166 /*
2167 * First make sure the new directory doesn't exist.
2168 *
2169 * Existence is checked first to make sure we don't return
2170 * EACCES instead of EEXIST which can cause some applications
2171 * to fail.
2172 */
2173top:
2174 *vpp = NULL;
2175
2176 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
2177 NULL, NULL)) {
2178 zfs_acl_ids_free(&acl_ids);
2179 ZFS_EXIT(zfsvfs);
2180 return (error);
2181 }
2182
2183 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2184 zfs_acl_ids_free(&acl_ids);
2185 zfs_dirent_unlock(dl);
2186 ZFS_EXIT(zfsvfs);
2187 return (error);
2188 }
2189
2190 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2191 zfs_acl_ids_free(&acl_ids);
2192 zfs_dirent_unlock(dl);
2193 ZFS_EXIT(zfsvfs);
2194 return (SET_ERROR(EDQUOT));
2195 }
2196
2197 /*
2198 * Add a new entry to the directory.
2199 */
2200 tx = dmu_tx_create(zfsvfs->z_os);
2201 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2202 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2203 fuid_dirtied = zfsvfs->z_fuid_dirty;
2204 if (fuid_dirtied)
2205 zfs_fuid_txhold(zfsvfs, tx);
2206 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2207 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2208 acl_ids.z_aclp->z_acl_bytes);
2209 }
2210
2211 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2212 ZFS_SA_BASE_ATTR_SIZE);
2213
2214 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2215 if (error) {
2216 zfs_dirent_unlock(dl);
2217 if (error == ERESTART) {
2218 waited = B_TRUE;
2219 dmu_tx_wait(tx);
2220 dmu_tx_abort(tx);
2221 goto top;
2222 }
2223 zfs_acl_ids_free(&acl_ids);
2224 dmu_tx_abort(tx);
2225 ZFS_EXIT(zfsvfs);
2226 return (error);
2227 }
2228
2229 /*
2230 * Create new node.
2231 */
2232 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2233
2234 if (fuid_dirtied)
2235 zfs_fuid_sync(zfsvfs, tx);
2236
2237 /*
2238 * Now put new name in parent dir.
2239 */
2240 (void) zfs_link_create(dl, zp, tx, ZNEW);
2241
2242 *vpp = ZTOV(zp);
2243
2244 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2245 if (flags & FIGNORECASE)
2246 txtype |= TX_CI;
2247 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2248 acl_ids.z_fuidp, vap);
2249
2250 zfs_acl_ids_free(&acl_ids);
2251
2252 dmu_tx_commit(tx);
2253
2254 zfs_dirent_unlock(dl);
2255
2256 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2257 zil_commit(zilog, 0);
2258
2259 ZFS_EXIT(zfsvfs);
2260 return (0);
2261}
2262
2263/*
2264 * Remove a directory subdir entry. If the current working
2265 * directory is the same as the subdir to be removed, the
2266 * remove will fail.
2267 *
2268 * IN: dvp - vnode of directory to remove from.
2269 * name - name of directory to be removed.
2270 * cwd - vnode of current working directory.
2271 * cr - credentials of caller.
2272 * ct - caller context
2273 * flags - case flags
2274 *
2275 * RETURN: 0 on success, error code on failure.
2276 *
2277 * Timestamps:
2278 * dvp - ctime|mtime updated
2279 */
2280/*ARGSUSED*/
2281static int
2282zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2283 caller_context_t *ct, int flags)
2284{
2285 znode_t *dzp = VTOZ(dvp);
2286 znode_t *zp;
2287 vnode_t *vp;
2288 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2289 zilog_t *zilog;
2290 zfs_dirlock_t *dl;
2291 dmu_tx_t *tx;
2292 int error;
2293 int zflg = ZEXISTS;
2294 boolean_t waited = B_FALSE;
2295
2296 ZFS_ENTER(zfsvfs);
2297 ZFS_VERIFY_ZP(dzp);
2298 zilog = zfsvfs->z_log;
2299
2300 if (flags & FIGNORECASE)
2301 zflg |= ZCILOOK;
2302top:
2303 zp = NULL;
2304
2305 /*
2306 * Attempt to lock directory; fail if entry doesn't exist.
2307 */
2308 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2309 NULL, NULL)) {
2310 ZFS_EXIT(zfsvfs);
2311 return (error);
2312 }
2313
2314 vp = ZTOV(zp);
2315
2316 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2317 goto out;
2318 }
2319
2320 if (vp->v_type != VDIR) {
2321 error = SET_ERROR(ENOTDIR);
2322 goto out;
2323 }
2324
2325 if (vp == cwd) {
2326 error = SET_ERROR(EINVAL);
2327 goto out;
2328 }
2329
2330 vnevent_rmdir(vp, dvp, name, ct);
2331
2332 /*
2333 * Grab a lock on the directory to make sure that noone is
2334 * trying to add (or lookup) entries while we are removing it.
2335 */
2336 rw_enter(&zp->z_name_lock, RW_WRITER);
2337
2338 /*
2339 * Grab a lock on the parent pointer to make sure we play well
2340 * with the treewalk and directory rename code.
2341 */
2342 rw_enter(&zp->z_parent_lock, RW_WRITER);
2343
2344 tx = dmu_tx_create(zfsvfs->z_os);
2345 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2346 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2347 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2348 zfs_sa_upgrade_txholds(tx, zp);
2349 zfs_sa_upgrade_txholds(tx, dzp);
2350 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2351 if (error) {
2352 rw_exit(&zp->z_parent_lock);
2353 rw_exit(&zp->z_name_lock);
2354 zfs_dirent_unlock(dl);
2355 VN_RELE(vp);
2356 if (error == ERESTART) {
2357 waited = B_TRUE;
2358 dmu_tx_wait(tx);
2359 dmu_tx_abort(tx);
2360 goto top;
2361 }
2362 dmu_tx_abort(tx);
2363 ZFS_EXIT(zfsvfs);
2364 return (error);
2365 }
2366
2367#ifdef FREEBSD_NAMECACHE
2368 cache_purge(dvp);
2369#endif
2370
2371 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2372
2373 if (error == 0) {
2374 uint64_t txtype = TX_RMDIR;
2375 if (flags & FIGNORECASE)
2376 txtype |= TX_CI;
2377 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2378 }
2379
2380 dmu_tx_commit(tx);
2381
2382 rw_exit(&zp->z_parent_lock);
2383 rw_exit(&zp->z_name_lock);
2384#ifdef FREEBSD_NAMECACHE
2385 cache_purge(vp);
2386#endif
2387out:
2388 zfs_dirent_unlock(dl);
2389
2390 VN_RELE(vp);
2391
2392 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2393 zil_commit(zilog, 0);
2394
2395 ZFS_EXIT(zfsvfs);
2396 return (error);
2397}
2398
2399/*
2400 * Read as many directory entries as will fit into the provided
2401 * buffer from the given directory cursor position (specified in
2402 * the uio structure).
2403 *
2404 * IN: vp - vnode of directory to read.
2405 * uio - structure supplying read location, range info,
2406 * and return buffer.
2407 * cr - credentials of caller.
2408 * ct - caller context
2409 * flags - case flags
2410 *
2411 * OUT: uio - updated offset and range, buffer filled.
2412 * eofp - set to true if end-of-file detected.
2413 *
2414 * RETURN: 0 on success, error code on failure.
2415 *
2416 * Timestamps:
2417 * vp - atime updated
2418 *
2419 * Note that the low 4 bits of the cookie returned by zap is always zero.
2420 * This allows us to use the low range for "special" directory entries:
2421 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2422 * we use the offset 2 for the '.zfs' directory.
2423 */
2424/* ARGSUSED */
2425static int
2426zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies)
2427{
2428 znode_t *zp = VTOZ(vp);
2429 iovec_t *iovp;
2430 edirent_t *eodp;
2431 dirent64_t *odp;
2432 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2433 objset_t *os;
2434 caddr_t outbuf;
2435 size_t bufsize;
2436 zap_cursor_t zc;
2437 zap_attribute_t zap;
2438 uint_t bytes_wanted;
2439 uint64_t offset; /* must be unsigned; checks for < 1 */
2440 uint64_t parent;
2441 int local_eof;
2442 int outcount;
2443 int error;
2444 uint8_t prefetch;
2445 boolean_t check_sysattrs;
2446 uint8_t type;
2447 int ncooks;
2448 u_long *cooks = NULL;
2449 int flags = 0;
2450
2451 ZFS_ENTER(zfsvfs);
2452 ZFS_VERIFY_ZP(zp);
2453
2454 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2455 &parent, sizeof (parent))) != 0) {
2456 ZFS_EXIT(zfsvfs);
2457 return (error);
2458 }
2459
2460 /*
2461 * If we are not given an eof variable,
2462 * use a local one.
2463 */
2464 if (eofp == NULL)
2465 eofp = &local_eof;
2466
2467 /*
2468 * Check for valid iov_len.
2469 */
2470 if (uio->uio_iov->iov_len <= 0) {
2471 ZFS_EXIT(zfsvfs);
2472 return (SET_ERROR(EINVAL));
2473 }
2474
2475 /*
2476 * Quit if directory has been removed (posix)
2477 */
2478 if ((*eofp = zp->z_unlinked) != 0) {
2479 ZFS_EXIT(zfsvfs);
2480 return (0);
2481 }
2482
2483 error = 0;
2484 os = zfsvfs->z_os;
2485 offset = uio->uio_loffset;
2486 prefetch = zp->z_zn_prefetch;
2487
2488 /*
2489 * Initialize the iterator cursor.
2490 */
2491 if (offset <= 3) {
2492 /*
2493 * Start iteration from the beginning of the directory.
2494 */
2495 zap_cursor_init(&zc, os, zp->z_id);
2496 } else {
2497 /*
2498 * The offset is a serialized cursor.
2499 */
2500 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2501 }
2502
2503 /*
2504 * Get space to change directory entries into fs independent format.
2505 */
2506 iovp = uio->uio_iov;
2507 bytes_wanted = iovp->iov_len;
2508 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2509 bufsize = bytes_wanted;
2510 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2511 odp = (struct dirent64 *)outbuf;
2512 } else {
2513 bufsize = bytes_wanted;
2514 outbuf = NULL;
2515 odp = (struct dirent64 *)iovp->iov_base;
2516 }
2517 eodp = (struct edirent *)odp;
2518
2519 if (ncookies != NULL) {
2520 /*
2521 * Minimum entry size is dirent size and 1 byte for a file name.
2522 */
2523 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1);
2524 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK);
2525 *cookies = cooks;
2526 *ncookies = ncooks;
2527 }
2528 /*
2529 * If this VFS supports the system attribute view interface; and
2530 * we're looking at an extended attribute directory; and we care
2531 * about normalization conflicts on this vfs; then we must check
2532 * for normalization conflicts with the sysattr name space.
2533 */
2534#ifdef TODO
2535 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2536 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2537 (flags & V_RDDIR_ENTFLAGS);
2538#else
2539 check_sysattrs = 0;
2540#endif
2541
2542 /*
2543 * Transform to file-system independent format
2544 */
2545 outcount = 0;
2546 while (outcount < bytes_wanted) {
2547 ino64_t objnum;
2548 ushort_t reclen;
2549 off64_t *next = NULL;
2550
2551 /*
2552 * Special case `.', `..', and `.zfs'.
2553 */
2554 if (offset == 0) {
2555 (void) strcpy(zap.za_name, ".");
2556 zap.za_normalization_conflict = 0;
2557 objnum = zp->z_id;
2558 type = DT_DIR;
2559 } else if (offset == 1) {
2560 (void) strcpy(zap.za_name, "..");
2561 zap.za_normalization_conflict = 0;
2562 objnum = parent;
2563 type = DT_DIR;
2564 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2565 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2566 zap.za_normalization_conflict = 0;
2567 objnum = ZFSCTL_INO_ROOT;
2568 type = DT_DIR;
2569 } else {
2570 /*
2571 * Grab next entry.
2572 */
2573 if (error = zap_cursor_retrieve(&zc, &zap)) {
2574 if ((*eofp = (error == ENOENT)) != 0)
2575 break;
2576 else
2577 goto update;
2578 }
2579
2580 if (zap.za_integer_length != 8 ||
2581 zap.za_num_integers != 1) {
2582 cmn_err(CE_WARN, "zap_readdir: bad directory "
2583 "entry, obj = %lld, offset = %lld\n",
2584 (u_longlong_t)zp->z_id,
2585 (u_longlong_t)offset);
2586 error = SET_ERROR(ENXIO);
2587 goto update;
2588 }
2589
2590 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2591 /*
2592 * MacOS X can extract the object type here such as:
2593 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2594 */
2595 type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2596
2597 if (check_sysattrs && !zap.za_normalization_conflict) {
2598#ifdef TODO
2599 zap.za_normalization_conflict =
2600 xattr_sysattr_casechk(zap.za_name);
2601#else
2602 panic("%s:%u: TODO", __func__, __LINE__);
2603#endif
2604 }
2605 }
2606
2607 if (flags & V_RDDIR_ACCFILTER) {
2608 /*
2609 * If we have no access at all, don't include
2610 * this entry in the returned information
2611 */
2612 znode_t *ezp;
2613 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2614 goto skip_entry;
2615 if (!zfs_has_access(ezp, cr)) {
2616 VN_RELE(ZTOV(ezp));
2617 goto skip_entry;
2618 }
2619 VN_RELE(ZTOV(ezp));
2620 }
2621
2622 if (flags & V_RDDIR_ENTFLAGS)
2623 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2624 else
2625 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2626
2627 /*
2628 * Will this entry fit in the buffer?
2629 */
2630 if (outcount + reclen > bufsize) {
2631 /*
2632 * Did we manage to fit anything in the buffer?
2633 */
2634 if (!outcount) {
2635 error = SET_ERROR(EINVAL);
2636 goto update;
2637 }
2638 break;
2639 }
2640 if (flags & V_RDDIR_ENTFLAGS) {
2641 /*
2642 * Add extended flag entry:
2643 */
2644 eodp->ed_ino = objnum;
2645 eodp->ed_reclen = reclen;
2646 /* NOTE: ed_off is the offset for the *next* entry */
2647 next = &(eodp->ed_off);
2648 eodp->ed_eflags = zap.za_normalization_conflict ?
2649 ED_CASE_CONFLICT : 0;
2650 (void) strncpy(eodp->ed_name, zap.za_name,
2651 EDIRENT_NAMELEN(reclen));
2652 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2653 } else {
2654 /*
2655 * Add normal entry:
2656 */
2657 odp->d_ino = objnum;
2658 odp->d_reclen = reclen;
2659 odp->d_namlen = strlen(zap.za_name);
2660 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1);
2661 odp->d_type = type;
2662 odp = (dirent64_t *)((intptr_t)odp + reclen);
2663 }
2664 outcount += reclen;
2665
2666 ASSERT(outcount <= bufsize);
2667
2668 /* Prefetch znode */
2669 if (prefetch)
2670 dmu_prefetch(os, objnum, 0, 0);
2671
2672 skip_entry:
2673 /*
2674 * Move to the next entry, fill in the previous offset.
2675 */
2676 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2677 zap_cursor_advance(&zc);
2678 offset = zap_cursor_serialize(&zc);
2679 } else {
2680 offset += 1;
2681 }
2682
2683 if (cooks != NULL) {
2684 *cooks++ = offset;
2685 ncooks--;
2686 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks));
2687 }
2688 }
2689 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2690
2691 /* Subtract unused cookies */
2692 if (ncookies != NULL)
2693 *ncookies -= ncooks;
2694
2695 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2696 iovp->iov_base += outcount;
2697 iovp->iov_len -= outcount;
2698 uio->uio_resid -= outcount;
2699 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2700 /*
2701 * Reset the pointer.
2702 */
2703 offset = uio->uio_loffset;
2704 }
2705
2706update:
2707 zap_cursor_fini(&zc);
2708 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2709 kmem_free(outbuf, bufsize);
2710
2711 if (error == ENOENT)
2712 error = 0;
2713
2714 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2715
2716 uio->uio_loffset = offset;
2717 ZFS_EXIT(zfsvfs);
2718 if (error != 0 && cookies != NULL) {
2719 free(*cookies, M_TEMP);
2720 *cookies = NULL;
2721 *ncookies = 0;
2722 }
2723 return (error);
2724}
2725
2726ulong_t zfs_fsync_sync_cnt = 4;
2727
2728static int
2729zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2730{
2731 znode_t *zp = VTOZ(vp);
2732 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2733
2734 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2735
2736 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2737 ZFS_ENTER(zfsvfs);
2738 ZFS_VERIFY_ZP(zp);
2739 zil_commit(zfsvfs->z_log, zp->z_id);
2740 ZFS_EXIT(zfsvfs);
2741 }
2742 return (0);
2743}
2744
2745
2746/*
2747 * Get the requested file attributes and place them in the provided
2748 * vattr structure.
2749 *
2750 * IN: vp - vnode of file.
2751 * vap - va_mask identifies requested attributes.
2752 * If AT_XVATTR set, then optional attrs are requested
2753 * flags - ATTR_NOACLCHECK (CIFS server context)
2754 * cr - credentials of caller.
2755 * ct - caller context
2756 *
2757 * OUT: vap - attribute values.
2758 *
2759 * RETURN: 0 (always succeeds).
2760 */
2761/* ARGSUSED */
2762static int
2763zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2764 caller_context_t *ct)
2765{
2766 znode_t *zp = VTOZ(vp);
2767 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2768 int error = 0;
2769 uint32_t blksize;
2770 u_longlong_t nblocks;
2771 uint64_t links;
2772 uint64_t mtime[2], ctime[2], crtime[2], rdev;
2773 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2774 xoptattr_t *xoap = NULL;
2775 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2776 sa_bulk_attr_t bulk[4];
2777 int count = 0;
2778
2779 ZFS_ENTER(zfsvfs);
2780 ZFS_VERIFY_ZP(zp);
2781
2782 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2783
2784 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2785 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2786 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16);
2787 if (vp->v_type == VBLK || vp->v_type == VCHR)
2788 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL,
2789 &rdev, 8);
2790
2791 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2792 ZFS_EXIT(zfsvfs);
2793 return (error);
2794 }
2795
2796 /*
2797 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2798 * Also, if we are the owner don't bother, since owner should
2799 * always be allowed to read basic attributes of file.
2800 */
2801 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2802 (vap->va_uid != crgetuid(cr))) {
2803 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2804 skipaclchk, cr)) {
2805 ZFS_EXIT(zfsvfs);
2806 return (error);
2807 }
2808 }
2809
2810 /*
2811 * Return all attributes. It's cheaper to provide the answer
2812 * than to determine whether we were asked the question.
2813 */
2814
2815 mutex_enter(&zp->z_lock);
2816 vap->va_type = IFTOVT(zp->z_mode);
2817 vap->va_mode = zp->z_mode & ~S_IFMT;
2818#ifdef sun
2819 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2820#else
2821 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
2822#endif
2823 vap->va_nodeid = zp->z_id;
2824 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2825 links = zp->z_links + 1;
2826 else
2827 links = zp->z_links;
2828 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */
2829 vap->va_size = zp->z_size;
2830#ifdef sun
2831 vap->va_rdev = vp->v_rdev;
2832#else
2833 if (vp->v_type == VBLK || vp->v_type == VCHR)
2834 vap->va_rdev = zfs_cmpldev(rdev);
2835#endif
2836 vap->va_seq = zp->z_seq;
2837 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */
2838
2839 /*
2840 * Add in any requested optional attributes and the create time.
2841 * Also set the corresponding bits in the returned attribute bitmap.
2842 */
2843 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2844 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2845 xoap->xoa_archive =
2846 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2847 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2848 }
2849
2850 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2851 xoap->xoa_readonly =
2852 ((zp->z_pflags & ZFS_READONLY) != 0);
2853 XVA_SET_RTN(xvap, XAT_READONLY);
2854 }
2855
2856 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2857 xoap->xoa_system =
2858 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2859 XVA_SET_RTN(xvap, XAT_SYSTEM);
2860 }
2861
2862 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2863 xoap->xoa_hidden =
2864 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2865 XVA_SET_RTN(xvap, XAT_HIDDEN);
2866 }
2867
2868 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2869 xoap->xoa_nounlink =
2870 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2871 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2872 }
2873
2874 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2875 xoap->xoa_immutable =
2876 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2877 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2878 }
2879
2880 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2881 xoap->xoa_appendonly =
2882 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2883 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2884 }
2885
2886 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2887 xoap->xoa_nodump =
2888 ((zp->z_pflags & ZFS_NODUMP) != 0);
2889 XVA_SET_RTN(xvap, XAT_NODUMP);
2890 }
2891
2892 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2893 xoap->xoa_opaque =
2894 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2895 XVA_SET_RTN(xvap, XAT_OPAQUE);
2896 }
2897
2898 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2899 xoap->xoa_av_quarantined =
2900 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2901 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2902 }
2903
2904 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2905 xoap->xoa_av_modified =
2906 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2907 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2908 }
2909
2910 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2911 vp->v_type == VREG) {
2912 zfs_sa_get_scanstamp(zp, xvap);
2913 }
2914
2915 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2916 uint64_t times[2];
2917
2918 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2919 times, sizeof (times));
2920 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2921 XVA_SET_RTN(xvap, XAT_CREATETIME);
2922 }
2923
2924 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2925 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2926 XVA_SET_RTN(xvap, XAT_REPARSE);
2927 }
2928 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2929 xoap->xoa_generation = zp->z_gen;
2930 XVA_SET_RTN(xvap, XAT_GEN);
2931 }
2932
2933 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2934 xoap->xoa_offline =
2935 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2936 XVA_SET_RTN(xvap, XAT_OFFLINE);
2937 }
2938
2939 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2940 xoap->xoa_sparse =
2941 ((zp->z_pflags & ZFS_SPARSE) != 0);
2942 XVA_SET_RTN(xvap, XAT_SPARSE);
2943 }
2944 }
2945
2946 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2947 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2948 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2949 ZFS_TIME_DECODE(&vap->va_birthtime, crtime);
2950
2951 mutex_exit(&zp->z_lock);
2952
2953 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
2954 vap->va_blksize = blksize;
2955 vap->va_bytes = nblocks << 9; /* nblocks * 512 */
2956
2957 if (zp->z_blksz == 0) {
2958 /*
2959 * Block size hasn't been set; suggest maximal I/O transfers.
2960 */
2961 vap->va_blksize = zfsvfs->z_max_blksz;
2962 }
2963
2964 ZFS_EXIT(zfsvfs);
2965 return (0);
2966}
2967
2968/*
2969 * Set the file attributes to the values contained in the
2970 * vattr structure.
2971 *
2972 * IN: vp - vnode of file to be modified.
2973 * vap - new attribute values.
2974 * If AT_XVATTR set, then optional attrs are being set
2975 * flags - ATTR_UTIME set if non-default time values provided.
2976 * - ATTR_NOACLCHECK (CIFS context only).
2977 * cr - credentials of caller.
2978 * ct - caller context
2979 *
2980 * RETURN: 0 on success, error code on failure.
2981 *
2982 * Timestamps:
2983 * vp - ctime updated, mtime updated if size changed.
2984 */
2985/* ARGSUSED */
2986static int
2987zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2988 caller_context_t *ct)
2989{
2990 znode_t *zp = VTOZ(vp);
2991 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2992 zilog_t *zilog;
2993 dmu_tx_t *tx;
2994 vattr_t oldva;
2995 xvattr_t tmpxvattr;
2996 uint_t mask = vap->va_mask;
2997 uint_t saved_mask = 0;
2998 uint64_t saved_mode;
2999 int trim_mask = 0;
3000 uint64_t new_mode;
3001 uint64_t new_uid, new_gid;
3002 uint64_t xattr_obj;
3003 uint64_t mtime[2], ctime[2];
3004 znode_t *attrzp;
3005 int need_policy = FALSE;
3006 int err, err2;
3007 zfs_fuid_info_t *fuidp = NULL;
3008 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
3009 xoptattr_t *xoap;
3010 zfs_acl_t *aclp;
3011 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
3012 boolean_t fuid_dirtied = B_FALSE;
3013 sa_bulk_attr_t bulk[7], xattr_bulk[7];
3014 int count = 0, xattr_count = 0;
3015
3016 if (mask == 0)
3017 return (0);
3018
3019 if (mask & AT_NOSET)
3020 return (SET_ERROR(EINVAL));
3021
3022 ZFS_ENTER(zfsvfs);
3023 ZFS_VERIFY_ZP(zp);
3024
3025 zilog = zfsvfs->z_log;
3026
3027 /*
3028 * Make sure that if we have ephemeral uid/gid or xvattr specified
3029 * that file system is at proper version level
3030 */
3031
3032 if (zfsvfs->z_use_fuids == B_FALSE &&
3033 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
3034 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
3035 (mask & AT_XVATTR))) {
3036 ZFS_EXIT(zfsvfs);
3037 return (SET_ERROR(EINVAL));
3038 }
3039
3040 if (mask & AT_SIZE && vp->v_type == VDIR) {
3041 ZFS_EXIT(zfsvfs);
3042 return (SET_ERROR(EISDIR));
3043 }
3044
3045 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
3046 ZFS_EXIT(zfsvfs);
3047 return (SET_ERROR(EINVAL));
3048 }
3049
3050 /*
3051 * If this is an xvattr_t, then get a pointer to the structure of
3052 * optional attributes. If this is NULL, then we have a vattr_t.
3053 */
3054 xoap = xva_getxoptattr(xvap);
3055
3056 xva_init(&tmpxvattr);
3057
3058 /*
3059 * Immutable files can only alter immutable bit and atime
3060 */
3061 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
3062 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
3063 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
3064 ZFS_EXIT(zfsvfs);
3065 return (SET_ERROR(EPERM));
3066 }
3067
3068 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
3069 ZFS_EXIT(zfsvfs);
3070 return (SET_ERROR(EPERM));
3071 }
3072
3073 /*
3074 * Verify timestamps doesn't overflow 32 bits.
3075 * ZFS can handle large timestamps, but 32bit syscalls can't
3076 * handle times greater than 2039. This check should be removed
3077 * once large timestamps are fully supported.
3078 */
3079 if (mask & (AT_ATIME | AT_MTIME)) {
3080 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
3081 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
3082 ZFS_EXIT(zfsvfs);
3083 return (SET_ERROR(EOVERFLOW));
3084 }
3085 }
3086
3087top:
3088 attrzp = NULL;
3089 aclp = NULL;
3090
3091 /* Can this be moved to before the top label? */
3092 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
3093 ZFS_EXIT(zfsvfs);
3094 return (SET_ERROR(EROFS));
3095 }
3096
3097 /*
3098 * First validate permissions
3099 */
3100
3101 if (mask & AT_SIZE) {
3102 /*
3103 * XXX - Note, we are not providing any open
3104 * mode flags here (like FNDELAY), so we may
3105 * block if there are locks present... this
3106 * should be addressed in openat().
3107 */
3108 /* XXX - would it be OK to generate a log record here? */
3109 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
3110 if (err) {
3111 ZFS_EXIT(zfsvfs);
3112 return (err);
3113 }
3114 }
3115
3116 if (mask & (AT_ATIME|AT_MTIME) ||
3117 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
3118 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
3119 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
3120 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
3121 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
3122 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
3123 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
3124 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
3125 skipaclchk, cr);
3126 }
3127
3128 if (mask & (AT_UID|AT_GID)) {
3129 int idmask = (mask & (AT_UID|AT_GID));
3130 int take_owner;
3131 int take_group;
3132
3133 /*
3134 * NOTE: even if a new mode is being set,
3135 * we may clear S_ISUID/S_ISGID bits.
3136 */
3137
3138 if (!(mask & AT_MODE))
3139 vap->va_mode = zp->z_mode;
3140
3141 /*
3142 * Take ownership or chgrp to group we are a member of
3143 */
3144
3145 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
3146 take_group = (mask & AT_GID) &&
3147 zfs_groupmember(zfsvfs, vap->va_gid, cr);
3148
3149 /*
3150 * If both AT_UID and AT_GID are set then take_owner and
3151 * take_group must both be set in order to allow taking
3152 * ownership.
3153 *
3154 * Otherwise, send the check through secpolicy_vnode_setattr()
3155 *
3156 */
3157
3158 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
3159 ((idmask == AT_UID) && take_owner) ||
3160 ((idmask == AT_GID) && take_group)) {
3161 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
3162 skipaclchk, cr) == 0) {
3163 /*
3164 * Remove setuid/setgid for non-privileged users
3165 */
3166 secpolicy_setid_clear(vap, vp, cr);
3167 trim_mask = (mask & (AT_UID|AT_GID));
3168 } else {
3169 need_policy = TRUE;
3170 }
3171 } else {
3172 need_policy = TRUE;
3173 }
3174 }
3175
3176 mutex_enter(&zp->z_lock);
3177 oldva.va_mode = zp->z_mode;
3178 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
3179 if (mask & AT_XVATTR) {
3180 /*
3181 * Update xvattr mask to include only those attributes
3182 * that are actually changing.
3183 *
3184 * the bits will be restored prior to actually setting
3185 * the attributes so the caller thinks they were set.
3186 */
3187 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
3188 if (xoap->xoa_appendonly !=
3189 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
3190 need_policy = TRUE;
3191 } else {
3192 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
3193 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
3194 }
3195 }
3196
3197 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
3198 if (xoap->xoa_nounlink !=
3199 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
3200 need_policy = TRUE;
3201 } else {
3202 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
3203 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
3204 }
3205 }
3206
3207 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
3208 if (xoap->xoa_immutable !=
3209 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
3210 need_policy = TRUE;
3211 } else {
3212 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
3213 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
3214 }
3215 }
3216
3217 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
3218 if (xoap->xoa_nodump !=
3219 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
3220 need_policy = TRUE;
3221 } else {
3222 XVA_CLR_REQ(xvap, XAT_NODUMP);
3223 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3224 }
3225 }
3226
3227 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3228 if (xoap->xoa_av_modified !=
3229 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3230 need_policy = TRUE;
3231 } else {
3232 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3233 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3234 }
3235 }
3236
3237 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3238 if ((vp->v_type != VREG &&
3239 xoap->xoa_av_quarantined) ||
3240 xoap->xoa_av_quarantined !=
3241 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3242 need_policy = TRUE;
3243 } else {
3244 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3245 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3246 }
3247 }
3248
3249 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3250 mutex_exit(&zp->z_lock);
3251 ZFS_EXIT(zfsvfs);
3252 return (SET_ERROR(EPERM));
3253 }
3254
3255 if (need_policy == FALSE &&
3256 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3257 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3258 need_policy = TRUE;
3259 }
3260 }
3261
3262 mutex_exit(&zp->z_lock);
3263
3264 if (mask & AT_MODE) {
3265 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3266 err = secpolicy_setid_setsticky_clear(vp, vap,
3267 &oldva, cr);
3268 if (err) {
3269 ZFS_EXIT(zfsvfs);
3270 return (err);
3271 }
3272 trim_mask |= AT_MODE;
3273 } else {
3274 need_policy = TRUE;
3275 }
3276 }
3277
3278 if (need_policy) {
3279 /*
3280 * If trim_mask is set then take ownership
3281 * has been granted or write_acl is present and user
3282 * has the ability to modify mode. In that case remove
3283 * UID|GID and or MODE from mask so that
3284 * secpolicy_vnode_setattr() doesn't revoke it.
3285 */
3286
3287 if (trim_mask) {
3288 saved_mask = vap->va_mask;
3289 vap->va_mask &= ~trim_mask;
3290 if (trim_mask & AT_MODE) {
3291 /*
3292 * Save the mode, as secpolicy_vnode_setattr()
3293 * will overwrite it with ova.va_mode.
3294 */
3295 saved_mode = vap->va_mode;
3296 }
3297 }
3298 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3299 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3300 if (err) {
3301 ZFS_EXIT(zfsvfs);
3302 return (err);
3303 }
3304
3305 if (trim_mask) {
3306 vap->va_mask |= saved_mask;
3307 if (trim_mask & AT_MODE) {
3308 /*
3309 * Recover the mode after
3310 * secpolicy_vnode_setattr().
3311 */
3312 vap->va_mode = saved_mode;
3313 }
3314 }
3315 }
3316
3317 /*
3318 * secpolicy_vnode_setattr, or take ownership may have
3319 * changed va_mask
3320 */
3321 mask = vap->va_mask;
3322
3323 if ((mask & (AT_UID | AT_GID))) {
3324 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3325 &xattr_obj, sizeof (xattr_obj));
3326
3327 if (err == 0 && xattr_obj) {
3328 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3329 if (err)
3330 goto out2;
3331 }
3332 if (mask & AT_UID) {
3333 new_uid = zfs_fuid_create(zfsvfs,
3334 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3335 if (new_uid != zp->z_uid &&
3336 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3337 if (attrzp)
3338 VN_RELE(ZTOV(attrzp));
3339 err = SET_ERROR(EDQUOT);
3340 goto out2;
3341 }
3342 }
3343
3344 if (mask & AT_GID) {
3345 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3346 cr, ZFS_GROUP, &fuidp);
3347 if (new_gid != zp->z_gid &&
3348 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3349 if (attrzp)
3350 VN_RELE(ZTOV(attrzp));
3351 err = SET_ERROR(EDQUOT);
3352 goto out2;
3353 }
3354 }
3355 }
3356 tx = dmu_tx_create(zfsvfs->z_os);
3357
3358 if (mask & AT_MODE) {
3359 uint64_t pmode = zp->z_mode;
3360 uint64_t acl_obj;
3361 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3362
3363 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3364 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3365 err = SET_ERROR(EPERM);
3366 goto out;
3367 }
3368
3369 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3370 goto out;
3371
3372 mutex_enter(&zp->z_lock);
3373 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3374 /*
3375 * Are we upgrading ACL from old V0 format
3376 * to V1 format?
3377 */
3378 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3379 zfs_znode_acl_version(zp) ==
3380 ZFS_ACL_VERSION_INITIAL) {
3381 dmu_tx_hold_free(tx, acl_obj, 0,
3382 DMU_OBJECT_END);
3383 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3384 0, aclp->z_acl_bytes);
3385 } else {
3386 dmu_tx_hold_write(tx, acl_obj, 0,
3387 aclp->z_acl_bytes);
3388 }
3389 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3390 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3391 0, aclp->z_acl_bytes);
3392 }
3393 mutex_exit(&zp->z_lock);
3394 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3395 } else {
3396 if ((mask & AT_XVATTR) &&
3397 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3398 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3399 else
3400 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3401 }
3402
3403 if (attrzp) {
3404 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3405 }
3406
3407 fuid_dirtied = zfsvfs->z_fuid_dirty;
3408 if (fuid_dirtied)
3409 zfs_fuid_txhold(zfsvfs, tx);
3410
3411 zfs_sa_upgrade_txholds(tx, zp);
3412
3413 err = dmu_tx_assign(tx, TXG_NOWAIT);
3414 if (err) {
3415 if (err == ERESTART)
3416 dmu_tx_wait(tx);
3417 goto out;
3418 }
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 if (szp->z_pflags & (ZFS_APPENDONLY | ZFS_IMMUTABLE | ZFS_READONLY)) {
4339 ZFS_EXIT(zfsvfs);
4340 return (SET_ERROR(EPERM));
4341 }
4342
4343 /*
4344 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4345 * ctldir appear to have the same v_vfsp.
4346 */
4347 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4348 ZFS_EXIT(zfsvfs);
4349 return (SET_ERROR(EXDEV));
4350 }
4351
4352 /* Prevent links to .zfs/shares files */
4353
4354 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4355 &parent, sizeof (uint64_t))) != 0) {
4356 ZFS_EXIT(zfsvfs);
4357 return (error);
4358 }
4359 if (parent == zfsvfs->z_shares_dir) {
4360 ZFS_EXIT(zfsvfs);
4361 return (SET_ERROR(EPERM));
4362 }
4363
4364 if (zfsvfs->z_utf8 && u8_validate(name,
4365 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4366 ZFS_EXIT(zfsvfs);
4367 return (SET_ERROR(EILSEQ));
4368 }
4369 if (flags & FIGNORECASE)
4370 zf |= ZCILOOK;
4371
4372 /*
4373 * We do not support links between attributes and non-attributes
4374 * because of the potential security risk of creating links
4375 * into "normal" file space in order to circumvent restrictions
4376 * imposed in attribute space.
4377 */
4378 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4379 ZFS_EXIT(zfsvfs);
4380 return (SET_ERROR(EINVAL));
4381 }
4382
4383
4384 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4385 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) {
4386 ZFS_EXIT(zfsvfs);
4387 return (SET_ERROR(EPERM));
4388 }
4389
4390 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4391 ZFS_EXIT(zfsvfs);
4392 return (error);
4393 }
4394
4395top:
4396 /*
4397 * Attempt to lock directory; fail if entry already exists.
4398 */
4399 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4400 if (error) {
4401 ZFS_EXIT(zfsvfs);
4402 return (error);
4403 }
4404
4405 tx = dmu_tx_create(zfsvfs->z_os);
4406 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4407 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4408 zfs_sa_upgrade_txholds(tx, szp);
4409 zfs_sa_upgrade_txholds(tx, dzp);
4410 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4411 if (error) {
4412 zfs_dirent_unlock(dl);
4413 if (error == ERESTART) {
4414 waited = B_TRUE;
4415 dmu_tx_wait(tx);
4416 dmu_tx_abort(tx);
4417 goto top;
4418 }
4419 dmu_tx_abort(tx);
4420 ZFS_EXIT(zfsvfs);
4421 return (error);
4422 }
4423
4424 error = zfs_link_create(dl, szp, tx, 0);
4425
4426 if (error == 0) {
4427 uint64_t txtype = TX_LINK;
4428 if (flags & FIGNORECASE)
4429 txtype |= TX_CI;
4430 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4431 }
4432
4433 dmu_tx_commit(tx);
4434
4435 zfs_dirent_unlock(dl);
4436
4437 if (error == 0) {
4438 vnevent_link(svp, ct);
4439 }
4440
4441 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4442 zil_commit(zilog, 0);
4443
4444 ZFS_EXIT(zfsvfs);
4445 return (error);
4446}
4447
4448#ifdef sun
4449/*
4450 * zfs_null_putapage() is used when the file system has been force
4451 * unmounted. It just drops the pages.
4452 */
4453/* ARGSUSED */
4454static int
4455zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4456 size_t *lenp, int flags, cred_t *cr)
4457{
4458 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4459 return (0);
4460}
4461
4462/*
4463 * Push a page out to disk, klustering if possible.
4464 *
4465 * IN: vp - file to push page to.
4466 * pp - page to push.
4467 * flags - additional flags.
4468 * cr - credentials of caller.
4469 *
4470 * OUT: offp - start of range pushed.
4471 * lenp - len of range pushed.
4472 *
4473 * RETURN: 0 on success, error code on failure.
4474 *
4475 * NOTE: callers must have locked the page to be pushed. On
4476 * exit, the page (and all other pages in the kluster) must be
4477 * unlocked.
4478 */
4479/* ARGSUSED */
4480static int
4481zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4482 size_t *lenp, int flags, cred_t *cr)
4483{
4484 znode_t *zp = VTOZ(vp);
4485 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4486 dmu_tx_t *tx;
4487 u_offset_t off, koff;
4488 size_t len, klen;
4489 int err;
4490
4491 off = pp->p_offset;
4492 len = PAGESIZE;
4493 /*
4494 * If our blocksize is bigger than the page size, try to kluster
4495 * multiple pages so that we write a full block (thus avoiding
4496 * a read-modify-write).
4497 */
4498 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4499 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4500 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4501 ASSERT(koff <= zp->z_size);
4502 if (koff + klen > zp->z_size)
4503 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4504 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4505 }
4506 ASSERT3U(btop(len), ==, btopr(len));
4507
4508 /*
4509 * Can't push pages past end-of-file.
4510 */
4511 if (off >= zp->z_size) {
4512 /* ignore all pages */
4513 err = 0;
4514 goto out;
4515 } else if (off + len > zp->z_size) {
4516 int npages = btopr(zp->z_size - off);
4517 page_t *trunc;
4518
4519 page_list_break(&pp, &trunc, npages);
4520 /* ignore pages past end of file */
4521 if (trunc)
4522 pvn_write_done(trunc, flags);
4523 len = zp->z_size - off;
4524 }
4525
4526 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4527 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4528 err = SET_ERROR(EDQUOT);
4529 goto out;
4530 }
4531top:
4532 tx = dmu_tx_create(zfsvfs->z_os);
4533 dmu_tx_hold_write(tx, zp->z_id, off, len);
4534
4535 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4536 zfs_sa_upgrade_txholds(tx, zp);
4537 err = dmu_tx_assign(tx, TXG_NOWAIT);
4538 if (err != 0) {
4539 if (err == ERESTART) {
4540 dmu_tx_wait(tx);
4541 dmu_tx_abort(tx);
4542 goto top;
4543 }
4544 dmu_tx_abort(tx);
4545 goto out;
4546 }
4547
4548 if (zp->z_blksz <= PAGESIZE) {
4549 caddr_t va = zfs_map_page(pp, S_READ);
4550 ASSERT3U(len, <=, PAGESIZE);
4551 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4552 zfs_unmap_page(pp, va);
4553 } else {
4554 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4555 }
4556
4557 if (err == 0) {
4558 uint64_t mtime[2], ctime[2];
4559 sa_bulk_attr_t bulk[3];
4560 int count = 0;
4561
4562 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4563 &mtime, 16);
4564 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4565 &ctime, 16);
4566 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4567 &zp->z_pflags, 8);
4568 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4569 B_TRUE);
4570 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4571 }
4572 dmu_tx_commit(tx);
4573
4574out:
4575 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4576 if (offp)
4577 *offp = off;
4578 if (lenp)
4579 *lenp = len;
4580
4581 return (err);
4582}
4583
4584/*
4585 * Copy the portion of the file indicated from pages into the file.
4586 * The pages are stored in a page list attached to the files vnode.
4587 *
4588 * IN: vp - vnode of file to push page data to.
4589 * off - position in file to put data.
4590 * len - amount of data to write.
4591 * flags - flags to control the operation.
4592 * cr - credentials of caller.
4593 * ct - caller context.
4594 *
4595 * RETURN: 0 on success, error code on failure.
4596 *
4597 * Timestamps:
4598 * vp - ctime|mtime updated
4599 */
4600/*ARGSUSED*/
4601static int
4602zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4603 caller_context_t *ct)
4604{
4605 znode_t *zp = VTOZ(vp);
4606 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4607 page_t *pp;
4608 size_t io_len;
4609 u_offset_t io_off;
4610 uint_t blksz;
4611 rl_t *rl;
4612 int error = 0;
4613
4614 ZFS_ENTER(zfsvfs);
4615 ZFS_VERIFY_ZP(zp);
4616
4617 /*
4618 * Align this request to the file block size in case we kluster.
4619 * XXX - this can result in pretty aggresive locking, which can
4620 * impact simultanious read/write access. One option might be
4621 * to break up long requests (len == 0) into block-by-block
4622 * operations to get narrower locking.
4623 */
4624 blksz = zp->z_blksz;
4625 if (ISP2(blksz))
4626 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4627 else
4628 io_off = 0;
4629 if (len > 0 && ISP2(blksz))
4630 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4631 else
4632 io_len = 0;
4633
4634 if (io_len == 0) {
4635 /*
4636 * Search the entire vp list for pages >= io_off.
4637 */
4638 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4639 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4640 goto out;
4641 }
4642 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4643
4644 if (off > zp->z_size) {
4645 /* past end of file */
4646 zfs_range_unlock(rl);
4647 ZFS_EXIT(zfsvfs);
4648 return (0);
4649 }
4650
4651 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4652
4653 for (off = io_off; io_off < off + len; io_off += io_len) {
4654 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4655 pp = page_lookup(vp, io_off,
4656 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4657 } else {
4658 pp = page_lookup_nowait(vp, io_off,
4659 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4660 }
4661
4662 if (pp != NULL && pvn_getdirty(pp, flags)) {
4663 int err;
4664
4665 /*
4666 * Found a dirty page to push
4667 */
4668 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4669 if (err)
4670 error = err;
4671 } else {
4672 io_len = PAGESIZE;
4673 }
4674 }
4675out:
4676 zfs_range_unlock(rl);
4677 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4678 zil_commit(zfsvfs->z_log, zp->z_id);
4679 ZFS_EXIT(zfsvfs);
4680 return (error);
4681}
4682#endif /* sun */
4683
4684/*ARGSUSED*/
4685void
4686zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4687{
4688 znode_t *zp = VTOZ(vp);
4689 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4690 int error;
4691
4692 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4693 if (zp->z_sa_hdl == NULL) {
4694 /*
4695 * The fs has been unmounted, or we did a
4696 * suspend/resume and this file no longer exists.
4697 */
4698 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4699 vrecycle(vp);
4700 return;
4701 }
4702
4703 mutex_enter(&zp->z_lock);
4704 if (zp->z_unlinked) {
4705 /*
4706 * Fast path to recycle a vnode of a removed file.
4707 */
4708 mutex_exit(&zp->z_lock);
4709 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4710 vrecycle(vp);
4711 return;
4712 }
4713 mutex_exit(&zp->z_lock);
4714
4715 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4716 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4717
4718 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4719 zfs_sa_upgrade_txholds(tx, zp);
4720 error = dmu_tx_assign(tx, TXG_WAIT);
4721 if (error) {
4722 dmu_tx_abort(tx);
4723 } else {
4724 mutex_enter(&zp->z_lock);
4725 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4726 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4727 zp->z_atime_dirty = 0;
4728 mutex_exit(&zp->z_lock);
4729 dmu_tx_commit(tx);
4730 }
4731 }
4732 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4733}
4734
4735#ifdef sun
4736/*
4737 * Bounds-check the seek operation.
4738 *
4739 * IN: vp - vnode seeking within
4740 * ooff - old file offset
4741 * noffp - pointer to new file offset
4742 * ct - caller context
4743 *
4744 * RETURN: 0 on success, EINVAL if new offset invalid.
4745 */
4746/* ARGSUSED */
4747static int
4748zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4749 caller_context_t *ct)
4750{
4751 if (vp->v_type == VDIR)
4752 return (0);
4753 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4754}
4755
4756/*
4757 * Pre-filter the generic locking function to trap attempts to place
4758 * a mandatory lock on a memory mapped file.
4759 */
4760static int
4761zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4762 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4763{
4764 znode_t *zp = VTOZ(vp);
4765 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4766
4767 ZFS_ENTER(zfsvfs);
4768 ZFS_VERIFY_ZP(zp);
4769
4770 /*
4771 * We are following the UFS semantics with respect to mapcnt
4772 * here: If we see that the file is mapped already, then we will
4773 * return an error, but we don't worry about races between this
4774 * function and zfs_map().
4775 */
4776 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4777 ZFS_EXIT(zfsvfs);
4778 return (SET_ERROR(EAGAIN));
4779 }
4780 ZFS_EXIT(zfsvfs);
4781 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4782}
4783
4784/*
4785 * If we can't find a page in the cache, we will create a new page
4786 * and fill it with file data. For efficiency, we may try to fill
4787 * multiple pages at once (klustering) to fill up the supplied page
4788 * list. Note that the pages to be filled are held with an exclusive
4789 * lock to prevent access by other threads while they are being filled.
4790 */
4791static int
4792zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4793 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4794{
4795 znode_t *zp = VTOZ(vp);
4796 page_t *pp, *cur_pp;
4797 objset_t *os = zp->z_zfsvfs->z_os;
4798 u_offset_t io_off, total;
4799 size_t io_len;
4800 int err;
4801
4802 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4803 /*
4804 * We only have a single page, don't bother klustering
4805 */
4806 io_off = off;
4807 io_len = PAGESIZE;
4808 pp = page_create_va(vp, io_off, io_len,
4809 PG_EXCL | PG_WAIT, seg, addr);
4810 } else {
4811 /*
4812 * Try to find enough pages to fill the page list
4813 */
4814 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4815 &io_len, off, plsz, 0);
4816 }
4817 if (pp == NULL) {
4818 /*
4819 * The page already exists, nothing to do here.
4820 */
4821 *pl = NULL;
4822 return (0);
4823 }
4824
4825 /*
4826 * Fill the pages in the kluster.
4827 */
4828 cur_pp = pp;
4829 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4830 caddr_t va;
4831
4832 ASSERT3U(io_off, ==, cur_pp->p_offset);
4833 va = zfs_map_page(cur_pp, S_WRITE);
4834 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4835 DMU_READ_PREFETCH);
4836 zfs_unmap_page(cur_pp, va);
4837 if (err) {
4838 /* On error, toss the entire kluster */
4839 pvn_read_done(pp, B_ERROR);
4840 /* convert checksum errors into IO errors */
4841 if (err == ECKSUM)
4842 err = SET_ERROR(EIO);
4843 return (err);
4844 }
4845 cur_pp = cur_pp->p_next;
4846 }
4847
4848 /*
4849 * Fill in the page list array from the kluster starting
4850 * from the desired offset `off'.
4851 * NOTE: the page list will always be null terminated.
4852 */
4853 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4854 ASSERT(pl == NULL || (*pl)->p_offset == off);
4855
4856 return (0);
4857}
4858
4859/*
4860 * Return pointers to the pages for the file region [off, off + len]
4861 * in the pl array. If plsz is greater than len, this function may
4862 * also return page pointers from after the specified region
4863 * (i.e. the region [off, off + plsz]). These additional pages are
4864 * only returned if they are already in the cache, or were created as
4865 * part of a klustered read.
4866 *
4867 * IN: vp - vnode of file to get data from.
4868 * off - position in file to get data from.
4869 * len - amount of data to retrieve.
4870 * plsz - length of provided page list.
4871 * seg - segment to obtain pages for.
4872 * addr - virtual address of fault.
4873 * rw - mode of created pages.
4874 * cr - credentials of caller.
4875 * ct - caller context.
4876 *
4877 * OUT: protp - protection mode of created pages.
4878 * pl - list of pages created.
4879 *
4880 * RETURN: 0 on success, error code on failure.
4881 *
4882 * Timestamps:
4883 * vp - atime updated
4884 */
4885/* ARGSUSED */
4886static int
4887zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4888 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4889 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4890{
4891 znode_t *zp = VTOZ(vp);
4892 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4893 page_t **pl0 = pl;
4894 int err = 0;
4895
4896 /* we do our own caching, faultahead is unnecessary */
4897 if (pl == NULL)
4898 return (0);
4899 else if (len > plsz)
4900 len = plsz;
4901 else
4902 len = P2ROUNDUP(len, PAGESIZE);
4903 ASSERT(plsz >= len);
4904
4905 ZFS_ENTER(zfsvfs);
4906 ZFS_VERIFY_ZP(zp);
4907
4908 if (protp)
4909 *protp = PROT_ALL;
4910
4911 /*
4912 * Loop through the requested range [off, off + len) looking
4913 * for pages. If we don't find a page, we will need to create
4914 * a new page and fill it with data from the file.
4915 */
4916 while (len > 0) {
4917 if (*pl = page_lookup(vp, off, SE_SHARED))
4918 *(pl+1) = NULL;
4919 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4920 goto out;
4921 while (*pl) {
4922 ASSERT3U((*pl)->p_offset, ==, off);
4923 off += PAGESIZE;
4924 addr += PAGESIZE;
4925 if (len > 0) {
4926 ASSERT3U(len, >=, PAGESIZE);
4927 len -= PAGESIZE;
4928 }
4929 ASSERT3U(plsz, >=, PAGESIZE);
4930 plsz -= PAGESIZE;
4931 pl++;
4932 }
4933 }
4934
4935 /*
4936 * Fill out the page array with any pages already in the cache.
4937 */
4938 while (plsz > 0 &&
4939 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4940 off += PAGESIZE;
4941 plsz -= PAGESIZE;
4942 }
4943out:
4944 if (err) {
4945 /*
4946 * Release any pages we have previously locked.
4947 */
4948 while (pl > pl0)
4949 page_unlock(*--pl);
4950 } else {
4951 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4952 }
4953
4954 *pl = NULL;
4955
4956 ZFS_EXIT(zfsvfs);
4957 return (err);
4958}
4959
4960/*
4961 * Request a memory map for a section of a file. This code interacts
4962 * with common code and the VM system as follows:
4963 *
4964 * - common code calls mmap(), which ends up in smmap_common()
4965 * - this calls VOP_MAP(), which takes you into (say) zfs
4966 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4967 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4968 * - zfs_addmap() updates z_mapcnt
4969 */
4970/*ARGSUSED*/
4971static int
4972zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4973 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4974 caller_context_t *ct)
4975{
4976 znode_t *zp = VTOZ(vp);
4977 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4978 segvn_crargs_t vn_a;
4979 int error;
4980
4981 ZFS_ENTER(zfsvfs);
4982 ZFS_VERIFY_ZP(zp);
4983
4984 if ((prot & PROT_WRITE) && (zp->z_pflags &
4985 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4986 ZFS_EXIT(zfsvfs);
4987 return (SET_ERROR(EPERM));
4988 }
4989
4990 if ((prot & (PROT_READ | PROT_EXEC)) &&
4991 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4992 ZFS_EXIT(zfsvfs);
4993 return (SET_ERROR(EACCES));
4994 }
4995
4996 if (vp->v_flag & VNOMAP) {
4997 ZFS_EXIT(zfsvfs);
4998 return (SET_ERROR(ENOSYS));
4999 }
5000
5001 if (off < 0 || len > MAXOFFSET_T - off) {
5002 ZFS_EXIT(zfsvfs);
5003 return (SET_ERROR(ENXIO));
5004 }
5005
5006 if (vp->v_type != VREG) {
5007 ZFS_EXIT(zfsvfs);
5008 return (SET_ERROR(ENODEV));
5009 }
5010
5011 /*
5012 * If file is locked, disallow mapping.
5013 */
5014 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
5015 ZFS_EXIT(zfsvfs);
5016 return (SET_ERROR(EAGAIN));
5017 }
5018
5019 as_rangelock(as);
5020 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5021 if (error != 0) {
5022 as_rangeunlock(as);
5023 ZFS_EXIT(zfsvfs);
5024 return (error);
5025 }
5026
5027 vn_a.vp = vp;
5028 vn_a.offset = (u_offset_t)off;
5029 vn_a.type = flags & MAP_TYPE;
5030 vn_a.prot = prot;
5031 vn_a.maxprot = maxprot;
5032 vn_a.cred = cr;
5033 vn_a.amp = NULL;
5034 vn_a.flags = flags & ~MAP_TYPE;
5035 vn_a.szc = 0;
5036 vn_a.lgrp_mem_policy_flags = 0;
5037
5038 error = as_map(as, *addrp, len, segvn_create, &vn_a);
5039
5040 as_rangeunlock(as);
5041 ZFS_EXIT(zfsvfs);
5042 return (error);
5043}
5044
5045/* ARGSUSED */
5046static int
5047zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5048 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
5049 caller_context_t *ct)
5050{
5051 uint64_t pages = btopr(len);
5052
5053 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
5054 return (0);
5055}
5056
5057/*
5058 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
5059 * more accurate mtime for the associated file. Since we don't have a way of
5060 * detecting when the data was actually modified, we have to resort to
5061 * heuristics. If an explicit msync() is done, then we mark the mtime when the
5062 * last page is pushed. The problem occurs when the msync() call is omitted,
5063 * which by far the most common case:
5064 *
5065 * open()
5066 * mmap()
5067 * <modify memory>
5068 * munmap()
5069 * close()
5070 * <time lapse>
5071 * putpage() via fsflush
5072 *
5073 * If we wait until fsflush to come along, we can have a modification time that
5074 * is some arbitrary point in the future. In order to prevent this in the
5075 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
5076 * torn down.
5077 */
5078/* ARGSUSED */
5079static int
5080zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
5081 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
5082 caller_context_t *ct)
5083{
5084 uint64_t pages = btopr(len);
5085
5086 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
5087 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
5088
5089 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
5090 vn_has_cached_data(vp))
5091 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
5092
5093 return (0);
5094}
5095
5096/*
5097 * Free or allocate space in a file. Currently, this function only
5098 * supports the `F_FREESP' command. However, this command is somewhat
5099 * misnamed, as its functionality includes the ability to allocate as
5100 * well as free space.
5101 *
5102 * IN: vp - vnode of file to free data in.
5103 * cmd - action to take (only F_FREESP supported).
5104 * bfp - section of file to free/alloc.
5105 * flag - current file open mode flags.
5106 * offset - current file offset.
5107 * cr - credentials of caller [UNUSED].
5108 * ct - caller context.
5109 *
5110 * RETURN: 0 on success, error code on failure.
5111 *
5112 * Timestamps:
5113 * vp - ctime|mtime updated
5114 */
5115/* ARGSUSED */
5116static int
5117zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
5118 offset_t offset, cred_t *cr, caller_context_t *ct)
5119{
5120 znode_t *zp = VTOZ(vp);
5121 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5122 uint64_t off, len;
5123 int error;
5124
5125 ZFS_ENTER(zfsvfs);
5126 ZFS_VERIFY_ZP(zp);
5127
5128 if (cmd != F_FREESP) {
5129 ZFS_EXIT(zfsvfs);
5130 return (SET_ERROR(EINVAL));
5131 }
5132
5133 if (error = convoff(vp, bfp, 0, offset)) {
5134 ZFS_EXIT(zfsvfs);
5135 return (error);
5136 }
5137
5138 if (bfp->l_len < 0) {
5139 ZFS_EXIT(zfsvfs);
5140 return (SET_ERROR(EINVAL));
5141 }
5142
5143 off = bfp->l_start;
5144 len = bfp->l_len; /* 0 means from off to end of file */
5145
5146 error = zfs_freesp(zp, off, len, flag, TRUE);
5147
5148 ZFS_EXIT(zfsvfs);
5149 return (error);
5150}
5151#endif /* sun */
5152
5153CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid));
5154CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid));
5155
5156/*ARGSUSED*/
5157static int
5158zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
5159{
5160 znode_t *zp = VTOZ(vp);
5161 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5162 uint32_t gen;
5163 uint64_t gen64;
5164 uint64_t object = zp->z_id;
5165 zfid_short_t *zfid;
5166 int size, i, error;
5167
5168 ZFS_ENTER(zfsvfs);
5169 ZFS_VERIFY_ZP(zp);
5170
5171 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
5172 &gen64, sizeof (uint64_t))) != 0) {
5173 ZFS_EXIT(zfsvfs);
5174 return (error);
5175 }
5176
5177 gen = (uint32_t)gen64;
5178
5179 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
5180
5181#ifdef illumos
5182 if (fidp->fid_len < size) {
5183 fidp->fid_len = size;
5184 ZFS_EXIT(zfsvfs);
5185 return (SET_ERROR(ENOSPC));
5186 }
5187#else
5188 fidp->fid_len = size;
5189#endif
5190
5191 zfid = (zfid_short_t *)fidp;
5192
5193 zfid->zf_len = size;
5194
5195 for (i = 0; i < sizeof (zfid->zf_object); i++)
5196 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
5197
5198 /* Must have a non-zero generation number to distinguish from .zfs */
5199 if (gen == 0)
5200 gen = 1;
5201 for (i = 0; i < sizeof (zfid->zf_gen); i++)
5202 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
5203
5204 if (size == LONG_FID_LEN) {
5205 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
5206 zfid_long_t *zlfid;
5207
5208 zlfid = (zfid_long_t *)fidp;
5209
5210 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
5211 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
5212
5213 /* XXX - this should be the generation number for the objset */
5214 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
5215 zlfid->zf_setgen[i] = 0;
5216 }
5217
5218 ZFS_EXIT(zfsvfs);
5219 return (0);
5220}
5221
5222static int
5223zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
5224 caller_context_t *ct)
5225{
5226 znode_t *zp, *xzp;
5227 zfsvfs_t *zfsvfs;
5228 zfs_dirlock_t *dl;
5229 int error;
5230
5231 switch (cmd) {
5232 case _PC_LINK_MAX:
5233 *valp = INT_MAX;
5234 return (0);
5235
5236 case _PC_FILESIZEBITS:
5237 *valp = 64;
5238 return (0);
5239#ifdef sun
5240 case _PC_XATTR_EXISTS:
5241 zp = VTOZ(vp);
5242 zfsvfs = zp->z_zfsvfs;
5243 ZFS_ENTER(zfsvfs);
5244 ZFS_VERIFY_ZP(zp);
5245 *valp = 0;
5246 error = zfs_dirent_lock(&dl, zp, "", &xzp,
5247 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
5248 if (error == 0) {
5249 zfs_dirent_unlock(dl);
5250 if (!zfs_dirempty(xzp))
5251 *valp = 1;
5252 VN_RELE(ZTOV(xzp));
5253 } else if (error == ENOENT) {
5254 /*
5255 * If there aren't extended attributes, it's the
5256 * same as having zero of them.
5257 */
5258 error = 0;
5259 }
5260 ZFS_EXIT(zfsvfs);
5261 return (error);
5262
5263 case _PC_SATTR_ENABLED:
5264 case _PC_SATTR_EXISTS:
5265 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5266 (vp->v_type == VREG || vp->v_type == VDIR);
5267 return (0);
5268
5269 case _PC_ACCESS_FILTERING:
5270 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5271 vp->v_type == VDIR;
5272 return (0);
5273
5274 case _PC_ACL_ENABLED:
5275 *valp = _ACL_ACE_ENABLED;
5276 return (0);
5277#endif /* sun */
5278 case _PC_MIN_HOLE_SIZE:
5279 *valp = (int)SPA_MINBLOCKSIZE;
5280 return (0);
5281#ifdef sun
5282 case _PC_TIMESTAMP_RESOLUTION:
5283 /* nanosecond timestamp resolution */
5284 *valp = 1L;
5285 return (0);
5286#endif /* sun */
5287 case _PC_ACL_EXTENDED:
5288 *valp = 0;
5289 return (0);
5290
5291 case _PC_ACL_NFS4:
5292 *valp = 1;
5293 return (0);
5294
5295 case _PC_ACL_PATH_MAX:
5296 *valp = ACL_MAX_ENTRIES;
5297 return (0);
5298
5299 default:
5300 return (EOPNOTSUPP);
5301 }
5302}
5303
5304/*ARGSUSED*/
5305static int
5306zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5307 caller_context_t *ct)
5308{
5309 znode_t *zp = VTOZ(vp);
5310 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5311 int error;
5312 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5313
5314 ZFS_ENTER(zfsvfs);
5315 ZFS_VERIFY_ZP(zp);
5316 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5317 ZFS_EXIT(zfsvfs);
5318
5319 return (error);
5320}
5321
5322/*ARGSUSED*/
5323int
5324zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5325 caller_context_t *ct)
5326{
5327 znode_t *zp = VTOZ(vp);
5328 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5329 int error;
5330 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5331 zilog_t *zilog = zfsvfs->z_log;
5332
5333 ZFS_ENTER(zfsvfs);
5334 ZFS_VERIFY_ZP(zp);
5335
5336 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5337
5338 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5339 zil_commit(zilog, 0);
5340
5341 ZFS_EXIT(zfsvfs);
5342 return (error);
5343}
5344
5345#ifdef sun
5346/*
5347 * The smallest read we may consider to loan out an arcbuf.
5348 * This must be a power of 2.
5349 */
5350int zcr_blksz_min = (1 << 10); /* 1K */
5351/*
5352 * If set to less than the file block size, allow loaning out of an
5353 * arcbuf for a partial block read. This must be a power of 2.
5354 */
5355int zcr_blksz_max = (1 << 17); /* 128K */
5356
5357/*ARGSUSED*/
5358static int
5359zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5360 caller_context_t *ct)
5361{
5362 znode_t *zp = VTOZ(vp);
5363 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5364 int max_blksz = zfsvfs->z_max_blksz;
5365 uio_t *uio = &xuio->xu_uio;
5366 ssize_t size = uio->uio_resid;
5367 offset_t offset = uio->uio_loffset;
5368 int blksz;
5369 int fullblk, i;
5370 arc_buf_t *abuf;
5371 ssize_t maxsize;
5372 int preamble, postamble;
5373
5374 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5375 return (SET_ERROR(EINVAL));
5376
5377 ZFS_ENTER(zfsvfs);
5378 ZFS_VERIFY_ZP(zp);
5379 switch (ioflag) {
5380 case UIO_WRITE:
5381 /*
5382 * Loan out an arc_buf for write if write size is bigger than
5383 * max_blksz, and the file's block size is also max_blksz.
5384 */
5385 blksz = max_blksz;
5386 if (size < blksz || zp->z_blksz != blksz) {
5387 ZFS_EXIT(zfsvfs);
5388 return (SET_ERROR(EINVAL));
5389 }
5390 /*
5391 * Caller requests buffers for write before knowing where the
5392 * write offset might be (e.g. NFS TCP write).
5393 */
5394 if (offset == -1) {
5395 preamble = 0;
5396 } else {
5397 preamble = P2PHASE(offset, blksz);
5398 if (preamble) {
5399 preamble = blksz - preamble;
5400 size -= preamble;
5401 }
5402 }
5403
5404 postamble = P2PHASE(size, blksz);
5405 size -= postamble;
5406
5407 fullblk = size / blksz;
5408 (void) dmu_xuio_init(xuio,
5409 (preamble != 0) + fullblk + (postamble != 0));
5410 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5411 int, postamble, int,
5412 (preamble != 0) + fullblk + (postamble != 0));
5413
5414 /*
5415 * Have to fix iov base/len for partial buffers. They
5416 * currently represent full arc_buf's.
5417 */
5418 if (preamble) {
5419 /* data begins in the middle of the arc_buf */
5420 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5421 blksz);
5422 ASSERT(abuf);
5423 (void) dmu_xuio_add(xuio, abuf,
5424 blksz - preamble, preamble);
5425 }
5426
5427 for (i = 0; i < fullblk; i++) {
5428 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5429 blksz);
5430 ASSERT(abuf);
5431 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5432 }
5433
5434 if (postamble) {
5435 /* data ends in the middle of the arc_buf */
5436 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5437 blksz);
5438 ASSERT(abuf);
5439 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5440 }
5441 break;
5442 case UIO_READ:
5443 /*
5444 * Loan out an arc_buf for read if the read size is larger than
5445 * the current file block size. Block alignment is not
5446 * considered. Partial arc_buf will be loaned out for read.
5447 */
5448 blksz = zp->z_blksz;
5449 if (blksz < zcr_blksz_min)
5450 blksz = zcr_blksz_min;
5451 if (blksz > zcr_blksz_max)
5452 blksz = zcr_blksz_max;
5453 /* avoid potential complexity of dealing with it */
5454 if (blksz > max_blksz) {
5455 ZFS_EXIT(zfsvfs);
5456 return (SET_ERROR(EINVAL));
5457 }
5458
5459 maxsize = zp->z_size - uio->uio_loffset;
5460 if (size > maxsize)
5461 size = maxsize;
5462
5463 if (size < blksz || vn_has_cached_data(vp)) {
5464 ZFS_EXIT(zfsvfs);
5465 return (SET_ERROR(EINVAL));
5466 }
5467 break;
5468 default:
5469 ZFS_EXIT(zfsvfs);
5470 return (SET_ERROR(EINVAL));
5471 }
5472
5473 uio->uio_extflg = UIO_XUIO;
5474 XUIO_XUZC_RW(xuio) = ioflag;
5475 ZFS_EXIT(zfsvfs);
5476 return (0);
5477}
5478
5479/*ARGSUSED*/
5480static int
5481zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5482{
5483 int i;
5484 arc_buf_t *abuf;
5485 int ioflag = XUIO_XUZC_RW(xuio);
5486
5487 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5488
5489 i = dmu_xuio_cnt(xuio);
5490 while (i-- > 0) {
5491 abuf = dmu_xuio_arcbuf(xuio, i);
5492 /*
5493 * if abuf == NULL, it must be a write buffer
5494 * that has been returned in zfs_write().
5495 */
5496 if (abuf)
5497 dmu_return_arcbuf(abuf);
5498 ASSERT(abuf || ioflag == UIO_WRITE);
5499 }
5500
5501 dmu_xuio_fini(xuio);
5502 return (0);
5503}
5504
5505/*
5506 * Predeclare these here so that the compiler assumes that
5507 * this is an "old style" function declaration that does
5508 * not include arguments => we won't get type mismatch errors
5509 * in the initializations that follow.
5510 */
5511static int zfs_inval();
5512static int zfs_isdir();
5513
5514static int
5515zfs_inval()
5516{
5517 return (SET_ERROR(EINVAL));
5518}
5519
5520static int
5521zfs_isdir()
5522{
5523 return (SET_ERROR(EISDIR));
5524}
5525/*
5526 * Directory vnode operations template
5527 */
5528vnodeops_t *zfs_dvnodeops;
5529const fs_operation_def_t zfs_dvnodeops_template[] = {
5530 VOPNAME_OPEN, { .vop_open = zfs_open },
5531 VOPNAME_CLOSE, { .vop_close = zfs_close },
5532 VOPNAME_READ, { .error = zfs_isdir },
5533 VOPNAME_WRITE, { .error = zfs_isdir },
5534 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5535 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5536 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5537 VOPNAME_ACCESS, { .vop_access = zfs_access },
5538 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5539 VOPNAME_CREATE, { .vop_create = zfs_create },
5540 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5541 VOPNAME_LINK, { .vop_link = zfs_link },
5542 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5543 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5544 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5545 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5546 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5547 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5548 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5549 VOPNAME_FID, { .vop_fid = zfs_fid },
5550 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5551 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5552 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5553 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5554 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5555 NULL, NULL
5556};
5557
5558/*
5559 * Regular file vnode operations template
5560 */
5561vnodeops_t *zfs_fvnodeops;
5562const fs_operation_def_t zfs_fvnodeops_template[] = {
5563 VOPNAME_OPEN, { .vop_open = zfs_open },
5564 VOPNAME_CLOSE, { .vop_close = zfs_close },
5565 VOPNAME_READ, { .vop_read = zfs_read },
5566 VOPNAME_WRITE, { .vop_write = zfs_write },
5567 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5568 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5569 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5570 VOPNAME_ACCESS, { .vop_access = zfs_access },
5571 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5572 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5573 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5574 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5575 VOPNAME_FID, { .vop_fid = zfs_fid },
5576 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5577 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5578 VOPNAME_SPACE, { .vop_space = zfs_space },
5579 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5580 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5581 VOPNAME_MAP, { .vop_map = zfs_map },
5582 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5583 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5584 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5585 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5586 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5587 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5588 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5589 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5590 NULL, NULL
5591};
5592
5593/*
5594 * Symbolic link vnode operations template
5595 */
5596vnodeops_t *zfs_symvnodeops;
5597const fs_operation_def_t zfs_symvnodeops_template[] = {
5598 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5599 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5600 VOPNAME_ACCESS, { .vop_access = zfs_access },
5601 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5602 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5603 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5604 VOPNAME_FID, { .vop_fid = zfs_fid },
5605 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5606 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5607 NULL, NULL
5608};
5609
5610/*
5611 * special share hidden files vnode operations template
5612 */
5613vnodeops_t *zfs_sharevnodeops;
5614const fs_operation_def_t zfs_sharevnodeops_template[] = {
5615 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5616 VOPNAME_ACCESS, { .vop_access = zfs_access },
5617 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5618 VOPNAME_FID, { .vop_fid = zfs_fid },
5619 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5620 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5621 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5622 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5623 NULL, NULL
5624};
5625
5626/*
5627 * Extended attribute directory vnode operations template
5628 *
5629 * This template is identical to the directory vnodes
5630 * operation template except for restricted operations:
5631 * VOP_MKDIR()
5632 * VOP_SYMLINK()
5633 *
5634 * Note that there are other restrictions embedded in:
5635 * zfs_create() - restrict type to VREG
5636 * zfs_link() - no links into/out of attribute space
5637 * zfs_rename() - no moves into/out of attribute space
5638 */
5639vnodeops_t *zfs_xdvnodeops;
5640const fs_operation_def_t zfs_xdvnodeops_template[] = {
5641 VOPNAME_OPEN, { .vop_open = zfs_open },
5642 VOPNAME_CLOSE, { .vop_close = zfs_close },
5643 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5644 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5645 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5646 VOPNAME_ACCESS, { .vop_access = zfs_access },
5647 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5648 VOPNAME_CREATE, { .vop_create = zfs_create },
5649 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5650 VOPNAME_LINK, { .vop_link = zfs_link },
5651 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5652 VOPNAME_MKDIR, { .error = zfs_inval },
5653 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5654 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5655 VOPNAME_SYMLINK, { .error = zfs_inval },
5656 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5657 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5658 VOPNAME_FID, { .vop_fid = zfs_fid },
5659 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5660 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5661 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5662 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5663 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5664 NULL, NULL
5665};
5666
5667/*
5668 * Error vnode operations template
5669 */
5670vnodeops_t *zfs_evnodeops;
5671const fs_operation_def_t zfs_evnodeops_template[] = {
5672 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5673 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5674 NULL, NULL
5675};
5676#endif /* sun */
5677
5678static int
5679ioflags(int ioflags)
5680{
5681 int flags = 0;
5682
5683 if (ioflags & IO_APPEND)
5684 flags |= FAPPEND;
5685 if (ioflags & IO_NDELAY)
5686 flags |= FNONBLOCK;
5687 if (ioflags & IO_SYNC)
5688 flags |= (FSYNC | FDSYNC | FRSYNC);
5689
5690 return (flags);
5691}
5692
5693static int
5694zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage)
5695{
5696 znode_t *zp = VTOZ(vp);
5697 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5698 objset_t *os = zp->z_zfsvfs->z_os;
5699 vm_page_t mfirst, mlast, mreq;
5700 vm_object_t object;
5701 caddr_t va;
5702 struct sf_buf *sf;
5703 off_t startoff, endoff;
5704 int i, error;
5705 vm_pindex_t reqstart, reqend;
5706 int pcount, lsize, reqsize, size;
5707
5708 ZFS_ENTER(zfsvfs);
5709 ZFS_VERIFY_ZP(zp);
5710
5711 pcount = OFF_TO_IDX(round_page(count));
5712 mreq = m[reqpage];
5713 object = mreq->object;
5714 error = 0;
5715
5716 KASSERT(vp->v_object == object, ("mismatching object"));
5717
5718 if (pcount > 1 && zp->z_blksz > PAGESIZE) {
5719 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz);
5720 reqstart = OFF_TO_IDX(round_page(startoff));
5721 if (reqstart < m[0]->pindex)
5722 reqstart = 0;
5723 else
5724 reqstart = reqstart - m[0]->pindex;
5725 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE,
5726 zp->z_blksz);
5727 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1;
5728 if (reqend > m[pcount - 1]->pindex)
5729 reqend = m[pcount - 1]->pindex;
5730 reqsize = reqend - m[reqstart]->pindex + 1;
5731 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize,
5732 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds"));
5733 } else {
5734 reqstart = reqpage;
5735 reqsize = 1;
5736 }
5737 mfirst = m[reqstart];
5738 mlast = m[reqstart + reqsize - 1];
5739
5740 zfs_vmobject_wlock(object);
5741
5742 for (i = 0; i < reqstart; i++) {
5743 vm_page_lock(m[i]);
5744 vm_page_free(m[i]);
5745 vm_page_unlock(m[i]);
5746 }
5747 for (i = reqstart + reqsize; i < pcount; i++) {
5748 vm_page_lock(m[i]);
5749 vm_page_free(m[i]);
5750 vm_page_unlock(m[i]);
5751 }
5752
5753 if (mreq->valid && reqsize == 1) {
5754 if (mreq->valid != VM_PAGE_BITS_ALL)
5755 vm_page_zero_invalid(mreq, TRUE);
5756 zfs_vmobject_wunlock(object);
5757 ZFS_EXIT(zfsvfs);
5758 return (zfs_vm_pagerret_ok);
5759 }
5760
5761 PCPU_INC(cnt.v_vnodein);
5762 PCPU_ADD(cnt.v_vnodepgsin, reqsize);
5763
5764 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) {
5765 for (i = reqstart; i < reqstart + reqsize; i++) {
5766 if (i != reqpage) {
5767 vm_page_lock(m[i]);
5768 vm_page_free(m[i]);
5769 vm_page_unlock(m[i]);
5770 }
5771 }
5772 zfs_vmobject_wunlock(object);
5773 ZFS_EXIT(zfsvfs);
5774 return (zfs_vm_pagerret_bad);
5775 }
5776
5777 lsize = PAGE_SIZE;
5778 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size)
5779 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex);
5780
5781 zfs_vmobject_wunlock(object);
5782
5783 for (i = reqstart; i < reqstart + reqsize; i++) {
5784 size = PAGE_SIZE;
5785 if (i == (reqstart + reqsize - 1))
5786 size = lsize;
5787 va = zfs_map_page(m[i], &sf);
5788 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex),
5789 size, va, DMU_READ_PREFETCH);
5790 if (size != PAGE_SIZE)
5791 bzero(va + size, PAGE_SIZE - size);
5792 zfs_unmap_page(sf);
5793 if (error != 0)
5794 break;
5795 }
5796
5797 zfs_vmobject_wlock(object);
5798
5799 for (i = reqstart; i < reqstart + reqsize; i++) {
5800 if (!error)
5801 m[i]->valid = VM_PAGE_BITS_ALL;
5802 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i]));
5803 if (i != reqpage)
5804 vm_page_readahead_finish(m[i]);
5805 }
5806
5807 zfs_vmobject_wunlock(object);
5808
5809 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
5810 ZFS_EXIT(zfsvfs);
5811 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok);
5812}
5813
5814static int
5815zfs_freebsd_getpages(ap)
5816 struct vop_getpages_args /* {
5817 struct vnode *a_vp;
5818 vm_page_t *a_m;
5819 int a_count;
5820 int a_reqpage;
5821 vm_ooffset_t a_offset;
5822 } */ *ap;
5823{
5824
5825 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage));
5826}
5827
5828static int
5829zfs_freebsd_bmap(ap)
5830 struct vop_bmap_args /* {
5831 struct vnode *a_vp;
5832 daddr_t a_bn;
5833 struct bufobj **a_bop;
5834 daddr_t *a_bnp;
5835 int *a_runp;
5836 int *a_runb;
5837 } */ *ap;
5838{
5839
5840 if (ap->a_bop != NULL)
5841 *ap->a_bop = &ap->a_vp->v_bufobj;
5842 if (ap->a_bnp != NULL)
5843 *ap->a_bnp = ap->a_bn;
5844 if (ap->a_runp != NULL)
5845 *ap->a_runp = 0;
5846 if (ap->a_runb != NULL)
5847 *ap->a_runb = 0;
5848
5849 return (0);
5850}
5851
5852static int
5853zfs_freebsd_open(ap)
5854 struct vop_open_args /* {
5855 struct vnode *a_vp;
5856 int a_mode;
5857 struct ucred *a_cred;
5858 struct thread *a_td;
5859 } */ *ap;
5860{
5861 vnode_t *vp = ap->a_vp;
5862 znode_t *zp = VTOZ(vp);
5863 int error;
5864
5865 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL);
5866 if (error == 0)
5867 vnode_create_vobject(vp, zp->z_size, ap->a_td);
5868 return (error);
5869}
5870
5871static int
5872zfs_freebsd_close(ap)
5873 struct vop_close_args /* {
5874 struct vnode *a_vp;
5875 int a_fflag;
5876 struct ucred *a_cred;
5877 struct thread *a_td;
5878 } */ *ap;
5879{
5880
5881 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL));
5882}
5883
5884static int
5885zfs_freebsd_ioctl(ap)
5886 struct vop_ioctl_args /* {
5887 struct vnode *a_vp;
5888 u_long a_command;
5889 caddr_t a_data;
5890 int a_fflag;
5891 struct ucred *cred;
5892 struct thread *td;
5893 } */ *ap;
5894{
5895
5896 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data,
5897 ap->a_fflag, ap->a_cred, NULL, NULL));
5898}
5899
5900static int
5901zfs_freebsd_read(ap)
5902 struct vop_read_args /* {
5903 struct vnode *a_vp;
5904 struct uio *a_uio;
5905 int a_ioflag;
5906 struct ucred *a_cred;
5907 } */ *ap;
5908{
5909
5910 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5911 ap->a_cred, NULL));
5912}
5913
5914static int
5915zfs_freebsd_write(ap)
5916 struct vop_write_args /* {
5917 struct vnode *a_vp;
5918 struct uio *a_uio;
5919 int a_ioflag;
5920 struct ucred *a_cred;
5921 } */ *ap;
5922{
5923
5924 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag),
5925 ap->a_cred, NULL));
5926}
5927
5928static int
5929zfs_freebsd_access(ap)
5930 struct vop_access_args /* {
5931 struct vnode *a_vp;
5932 accmode_t a_accmode;
5933 struct ucred *a_cred;
5934 struct thread *a_td;
5935 } */ *ap;
5936{
5937 vnode_t *vp = ap->a_vp;
5938 znode_t *zp = VTOZ(vp);
5939 accmode_t accmode;
5940 int error = 0;
5941
5942 /*
5943 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND,
5944 */
5945 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND);
5946 if (accmode != 0)
5947 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL);
5948
5949 /*
5950 * VADMIN has to be handled by vaccess().
5951 */
5952 if (error == 0) {
5953 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND);
5954 if (accmode != 0) {
5955 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid,
5956 zp->z_gid, accmode, ap->a_cred, NULL);
5957 }
5958 }
5959
5960 /*
5961 * For VEXEC, ensure that at least one execute bit is set for
5962 * non-directories.
5963 */
5964 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR &&
5965 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) {
5966 error = EACCES;
5967 }
5968
5969 return (error);
5970}
5971
5972static int
5973zfs_freebsd_lookup(ap)
5974 struct vop_lookup_args /* {
5975 struct vnode *a_dvp;
5976 struct vnode **a_vpp;
5977 struct componentname *a_cnp;
5978 } */ *ap;
5979{
5980 struct componentname *cnp = ap->a_cnp;
5981 char nm[NAME_MAX + 1];
5982
5983 ASSERT(cnp->cn_namelen < sizeof(nm));
5984 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm)));
5985
5986 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop,
5987 cnp->cn_cred, cnp->cn_thread, 0));
5988}
5989
5990static int
5991zfs_freebsd_create(ap)
5992 struct vop_create_args /* {
5993 struct vnode *a_dvp;
5994 struct vnode **a_vpp;
5995 struct componentname *a_cnp;
5996 struct vattr *a_vap;
5997 } */ *ap;
5998{
5999 struct componentname *cnp = ap->a_cnp;
6000 vattr_t *vap = ap->a_vap;
6001 int mode;
6002
6003 ASSERT(cnp->cn_flags & SAVENAME);
6004
6005 vattr_init_mask(vap);
6006 mode = vap->va_mode & ALLPERMS;
6007
6008 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode,
6009 ap->a_vpp, cnp->cn_cred, cnp->cn_thread));
6010}
6011
6012static int
6013zfs_freebsd_remove(ap)
6014 struct vop_remove_args /* {
6015 struct vnode *a_dvp;
6016 struct vnode *a_vp;
6017 struct componentname *a_cnp;
6018 } */ *ap;
6019{
6020
6021 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6022
6023 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr,
6024 ap->a_cnp->cn_cred, NULL, 0));
6025}
6026
6027static int
6028zfs_freebsd_mkdir(ap)
6029 struct vop_mkdir_args /* {
6030 struct vnode *a_dvp;
6031 struct vnode **a_vpp;
6032 struct componentname *a_cnp;
6033 struct vattr *a_vap;
6034 } */ *ap;
6035{
6036 vattr_t *vap = ap->a_vap;
6037
6038 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
6039
6040 vattr_init_mask(vap);
6041
6042 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp,
6043 ap->a_cnp->cn_cred, NULL, 0, NULL));
6044}
6045
6046static int
6047zfs_freebsd_rmdir(ap)
6048 struct vop_rmdir_args /* {
6049 struct vnode *a_dvp;
6050 struct vnode *a_vp;
6051 struct componentname *a_cnp;
6052 } */ *ap;
6053{
6054 struct componentname *cnp = ap->a_cnp;
6055
6056 ASSERT(cnp->cn_flags & SAVENAME);
6057
6058 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0));
6059}
6060
6061static int
6062zfs_freebsd_readdir(ap)
6063 struct vop_readdir_args /* {
6064 struct vnode *a_vp;
6065 struct uio *a_uio;
6066 struct ucred *a_cred;
6067 int *a_eofflag;
6068 int *a_ncookies;
6069 u_long **a_cookies;
6070 } */ *ap;
6071{
6072
6073 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag,
6074 ap->a_ncookies, ap->a_cookies));
6075}
6076
6077static int
6078zfs_freebsd_fsync(ap)
6079 struct vop_fsync_args /* {
6080 struct vnode *a_vp;
6081 int a_waitfor;
6082 struct thread *a_td;
6083 } */ *ap;
6084{
6085
6086 vop_stdfsync(ap);
6087 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL));
6088}
6089
6090static int
6091zfs_freebsd_getattr(ap)
6092 struct vop_getattr_args /* {
6093 struct vnode *a_vp;
6094 struct vattr *a_vap;
6095 struct ucred *a_cred;
6096 } */ *ap;
6097{
6098 vattr_t *vap = ap->a_vap;
6099 xvattr_t xvap;
6100 u_long fflags = 0;
6101 int error;
6102
6103 xva_init(&xvap);
6104 xvap.xva_vattr = *vap;
6105 xvap.xva_vattr.va_mask |= AT_XVATTR;
6106
6107 /* Convert chflags into ZFS-type flags. */
6108 /* XXX: what about SF_SETTABLE?. */
6109 XVA_SET_REQ(&xvap, XAT_IMMUTABLE);
6110 XVA_SET_REQ(&xvap, XAT_APPENDONLY);
6111 XVA_SET_REQ(&xvap, XAT_NOUNLINK);
6112 XVA_SET_REQ(&xvap, XAT_NODUMP);
6113 XVA_SET_REQ(&xvap, XAT_READONLY);
6114 XVA_SET_REQ(&xvap, XAT_ARCHIVE);
6115 XVA_SET_REQ(&xvap, XAT_SYSTEM);
6116 XVA_SET_REQ(&xvap, XAT_HIDDEN);
6117 XVA_SET_REQ(&xvap, XAT_REPARSE);
6118 XVA_SET_REQ(&xvap, XAT_OFFLINE);
6119 XVA_SET_REQ(&xvap, XAT_SPARSE);
6120
6121 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL);
6122 if (error != 0)
6123 return (error);
6124
6125 /* Convert ZFS xattr into chflags. */
6126#define FLAG_CHECK(fflag, xflag, xfield) do { \
6127 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \
6128 fflags |= (fflag); \
6129} while (0)
6130 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE,
6131 xvap.xva_xoptattrs.xoa_immutable);
6132 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY,
6133 xvap.xva_xoptattrs.xoa_appendonly);
6134 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK,
6135 xvap.xva_xoptattrs.xoa_nounlink);
6136 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE,
6137 xvap.xva_xoptattrs.xoa_archive);
6138 FLAG_CHECK(UF_NODUMP, XAT_NODUMP,
6139 xvap.xva_xoptattrs.xoa_nodump);
6140 FLAG_CHECK(UF_READONLY, XAT_READONLY,
6141 xvap.xva_xoptattrs.xoa_readonly);
6142 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM,
6143 xvap.xva_xoptattrs.xoa_system);
6144 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN,
6145 xvap.xva_xoptattrs.xoa_hidden);
6146 FLAG_CHECK(UF_REPARSE, XAT_REPARSE,
6147 xvap.xva_xoptattrs.xoa_reparse);
6148 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE,
6149 xvap.xva_xoptattrs.xoa_offline);
6150 FLAG_CHECK(UF_SPARSE, XAT_SPARSE,
6151 xvap.xva_xoptattrs.xoa_sparse);
6152
6153#undef FLAG_CHECK
6154 *vap = xvap.xva_vattr;
6155 vap->va_flags = fflags;
6156 return (0);
6157}
6158
6159static int
6160zfs_freebsd_setattr(ap)
6161 struct vop_setattr_args /* {
6162 struct vnode *a_vp;
6163 struct vattr *a_vap;
6164 struct ucred *a_cred;
6165 } */ *ap;
6166{
6167 vnode_t *vp = ap->a_vp;
6168 vattr_t *vap = ap->a_vap;
6169 cred_t *cred = ap->a_cred;
6170 xvattr_t xvap;
6171 u_long fflags;
6172 uint64_t zflags;
6173
6174 vattr_init_mask(vap);
6175 vap->va_mask &= ~AT_NOSET;
6176
6177 xva_init(&xvap);
6178 xvap.xva_vattr = *vap;
6179
6180 zflags = VTOZ(vp)->z_pflags;
6181
6182 if (vap->va_flags != VNOVAL) {
6183 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs;
6184 int error;
6185
6186 if (zfsvfs->z_use_fuids == B_FALSE)
6187 return (EOPNOTSUPP);
6188
6189 fflags = vap->va_flags;
6190 /*
6191 * XXX KDM
6192 * We need to figure out whether it makes sense to allow
6193 * UF_REPARSE through, since we don't really have other
6194 * facilities to handle reparse points and zfs_setattr()
6195 * doesn't currently allow setting that attribute anyway.
6196 */
6197 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE|
6198 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE|
6199 UF_OFFLINE|UF_SPARSE)) != 0)
6200 return (EOPNOTSUPP);
6201 /*
6202 * Unprivileged processes are not permitted to unset system
6203 * flags, or modify flags if any system flags are set.
6204 * Privileged non-jail processes may not modify system flags
6205 * if securelevel > 0 and any existing system flags are set.
6206 * Privileged jail processes behave like privileged non-jail
6207 * processes if the security.jail.chflags_allowed sysctl is
6208 * is non-zero; otherwise, they behave like unprivileged
6209 * processes.
6210 */
6211 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 ||
6212 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) {
6213 if (zflags &
6214 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6215 error = securelevel_gt(cred, 0);
6216 if (error != 0)
6217 return (error);
6218 }
6219 } else {
6220 /*
6221 * Callers may only modify the file flags on objects they
6222 * have VADMIN rights for.
6223 */
6224 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0)
6225 return (error);
6226 if (zflags &
6227 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) {
6228 return (EPERM);
6229 }
6230 if (fflags &
6231 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) {
6232 return (EPERM);
6233 }
6234 }
6235
6236#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \
6237 if (((fflags & (fflag)) && !(zflags & (zflag))) || \
6238 ((zflags & (zflag)) && !(fflags & (fflag)))) { \
6239 XVA_SET_REQ(&xvap, (xflag)); \
6240 (xfield) = ((fflags & (fflag)) != 0); \
6241 } \
6242} while (0)
6243 /* Convert chflags into ZFS-type flags. */
6244 /* XXX: what about SF_SETTABLE?. */
6245 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE,
6246 xvap.xva_xoptattrs.xoa_immutable);
6247 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY,
6248 xvap.xva_xoptattrs.xoa_appendonly);
6249 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK,
6250 xvap.xva_xoptattrs.xoa_nounlink);
6251 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE,
6252 xvap.xva_xoptattrs.xoa_archive);
6253 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP,
6254 xvap.xva_xoptattrs.xoa_nodump);
6255 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY,
6256 xvap.xva_xoptattrs.xoa_readonly);
6257 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM,
6258 xvap.xva_xoptattrs.xoa_system);
6259 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN,
6260 xvap.xva_xoptattrs.xoa_hidden);
6261 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE,
6262 xvap.xva_xoptattrs.xoa_hidden);
6263 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE,
6264 xvap.xva_xoptattrs.xoa_offline);
6265 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE,
6266 xvap.xva_xoptattrs.xoa_sparse);
6267#undef FLAG_CHANGE
6268 }
6269 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL));
6270}
6271
6272static int
6273zfs_freebsd_rename(ap)
6274 struct vop_rename_args /* {
6275 struct vnode *a_fdvp;
6276 struct vnode *a_fvp;
6277 struct componentname *a_fcnp;
6278 struct vnode *a_tdvp;
6279 struct vnode *a_tvp;
6280 struct componentname *a_tcnp;
6281 } */ *ap;
6282{
6283 vnode_t *fdvp = ap->a_fdvp;
6284 vnode_t *fvp = ap->a_fvp;
6285 vnode_t *tdvp = ap->a_tdvp;
6286 vnode_t *tvp = ap->a_tvp;
6287 int error;
6288
6289 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART));
6290 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART));
6291
6292 /*
6293 * Check for cross-device rename.
6294 */
6295 if ((fdvp->v_mount != tdvp->v_mount) ||
6296 (tvp && (fdvp->v_mount != tvp->v_mount)))
6297 error = EXDEV;
6298 else
6299 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp,
6300 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0);
6301 if (tdvp == tvp)
6302 VN_RELE(tdvp);
6303 else
6304 VN_URELE(tdvp);
6305 if (tvp)
6306 VN_URELE(tvp);
6307 VN_RELE(fdvp);
6308 VN_RELE(fvp);
6309
6310 return (error);
6311}
6312
6313static int
6314zfs_freebsd_symlink(ap)
6315 struct vop_symlink_args /* {
6316 struct vnode *a_dvp;
6317 struct vnode **a_vpp;
6318 struct componentname *a_cnp;
6319 struct vattr *a_vap;
6320 char *a_target;
6321 } */ *ap;
6322{
6323 struct componentname *cnp = ap->a_cnp;
6324 vattr_t *vap = ap->a_vap;
6325
6326 ASSERT(cnp->cn_flags & SAVENAME);
6327
6328 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */
6329 vattr_init_mask(vap);
6330
6331 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap,
6332 ap->a_target, cnp->cn_cred, cnp->cn_thread));
6333}
6334
6335static int
6336zfs_freebsd_readlink(ap)
6337 struct vop_readlink_args /* {
6338 struct vnode *a_vp;
6339 struct uio *a_uio;
6340 struct ucred *a_cred;
6341 } */ *ap;
6342{
6343
6344 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL));
6345}
6346
6347static int
6348zfs_freebsd_link(ap)
6349 struct vop_link_args /* {
6350 struct vnode *a_tdvp;
6351 struct vnode *a_vp;
6352 struct componentname *a_cnp;
6353 } */ *ap;
6354{
6355 struct componentname *cnp = ap->a_cnp;
6356 vnode_t *vp = ap->a_vp;
6357 vnode_t *tdvp = ap->a_tdvp;
6358
6359 if (tdvp->v_mount != vp->v_mount)
6360 return (EXDEV);
6361
6362 ASSERT(cnp->cn_flags & SAVENAME);
6363
6364 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0));
6365}
6366
6367static int
6368zfs_freebsd_inactive(ap)
6369 struct vop_inactive_args /* {
6370 struct vnode *a_vp;
6371 struct thread *a_td;
6372 } */ *ap;
6373{
6374 vnode_t *vp = ap->a_vp;
6375
6376 zfs_inactive(vp, ap->a_td->td_ucred, NULL);
6377 return (0);
6378}
6379
6380static int
6381zfs_freebsd_reclaim(ap)
6382 struct vop_reclaim_args /* {
6383 struct vnode *a_vp;
6384 struct thread *a_td;
6385 } */ *ap;
6386{
6387 vnode_t *vp = ap->a_vp;
6388 znode_t *zp = VTOZ(vp);
6389 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
6390
6391 ASSERT(zp != NULL);
6392
6393 /* Destroy the vm object and flush associated pages. */
6394 vnode_destroy_vobject(vp);
6395
6396 /*
6397 * z_teardown_inactive_lock protects from a race with
6398 * zfs_znode_dmu_fini in zfsvfs_teardown during
6399 * force unmount.
6400 */
6401 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
6402 if (zp->z_sa_hdl == NULL)
6403 zfs_znode_free(zp);
6404 else
6405 zfs_zinactive(zp);
6406 rw_exit(&zfsvfs->z_teardown_inactive_lock);
6407
6408 vp->v_data = NULL;
6409 return (0);
6410}
6411
6412static int
6413zfs_freebsd_fid(ap)
6414 struct vop_fid_args /* {
6415 struct vnode *a_vp;
6416 struct fid *a_fid;
6417 } */ *ap;
6418{
6419
6420 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL));
6421}
6422
6423static int
6424zfs_freebsd_pathconf(ap)
6425 struct vop_pathconf_args /* {
6426 struct vnode *a_vp;
6427 int a_name;
6428 register_t *a_retval;
6429 } */ *ap;
6430{
6431 ulong_t val;
6432 int error;
6433
6434 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL);
6435 if (error == 0)
6436 *ap->a_retval = val;
6437 else if (error == EOPNOTSUPP)
6438 error = vop_stdpathconf(ap);
6439 return (error);
6440}
6441
6442static int
6443zfs_freebsd_fifo_pathconf(ap)
6444 struct vop_pathconf_args /* {
6445 struct vnode *a_vp;
6446 int a_name;
6447 register_t *a_retval;
6448 } */ *ap;
6449{
6450
6451 switch (ap->a_name) {
6452 case _PC_ACL_EXTENDED:
6453 case _PC_ACL_NFS4:
6454 case _PC_ACL_PATH_MAX:
6455 case _PC_MAC_PRESENT:
6456 return (zfs_freebsd_pathconf(ap));
6457 default:
6458 return (fifo_specops.vop_pathconf(ap));
6459 }
6460}
6461
6462/*
6463 * FreeBSD's extended attributes namespace defines file name prefix for ZFS'
6464 * extended attribute name:
6465 *
6466 * NAMESPACE PREFIX
6467 * system freebsd:system:
6468 * user (none, can be used to access ZFS fsattr(5) attributes
6469 * created on Solaris)
6470 */
6471static int
6472zfs_create_attrname(int attrnamespace, const char *name, char *attrname,
6473 size_t size)
6474{
6475 const char *namespace, *prefix, *suffix;
6476
6477 /* We don't allow '/' character in attribute name. */
6478 if (strchr(name, '/') != NULL)
6479 return (EINVAL);
6480 /* We don't allow attribute names that start with "freebsd:" string. */
6481 if (strncmp(name, "freebsd:", 8) == 0)
6482 return (EINVAL);
6483
6484 bzero(attrname, size);
6485
6486 switch (attrnamespace) {
6487 case EXTATTR_NAMESPACE_USER:
6488#if 0
6489 prefix = "freebsd:";
6490 namespace = EXTATTR_NAMESPACE_USER_STRING;
6491 suffix = ":";
6492#else
6493 /*
6494 * This is the default namespace by which we can access all
6495 * attributes created on Solaris.
6496 */
6497 prefix = namespace = suffix = "";
6498#endif
6499 break;
6500 case EXTATTR_NAMESPACE_SYSTEM:
6501 prefix = "freebsd:";
6502 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING;
6503 suffix = ":";
6504 break;
6505 case EXTATTR_NAMESPACE_EMPTY:
6506 default:
6507 return (EINVAL);
6508 }
6509 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix,
6510 name) >= size) {
6511 return (ENAMETOOLONG);
6512 }
6513 return (0);
6514}
6515
6516/*
6517 * Vnode operating to retrieve a named extended attribute.
6518 */
6519static int
6520zfs_getextattr(struct vop_getextattr_args *ap)
6521/*
6522vop_getextattr {
6523 IN struct vnode *a_vp;
6524 IN int a_attrnamespace;
6525 IN const char *a_name;
6526 INOUT struct uio *a_uio;
6527 OUT size_t *a_size;
6528 IN struct ucred *a_cred;
6529 IN struct thread *a_td;
6530};
6531*/
6532{
6533 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6534 struct thread *td = ap->a_td;
6535 struct nameidata nd;
6536 char attrname[255];
6537 struct vattr va;
6538 vnode_t *xvp = NULL, *vp;
6539 int error, flags;
6540
6541 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6542 ap->a_cred, ap->a_td, VREAD);
6543 if (error != 0)
6544 return (error);
6545
6546 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6547 sizeof(attrname));
6548 if (error != 0)
6549 return (error);
6550
6551 ZFS_ENTER(zfsvfs);
6552
6553 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6554 LOOKUP_XATTR);
6555 if (error != 0) {
6556 ZFS_EXIT(zfsvfs);
6557 return (error);
6558 }
6559
6560 flags = FREAD;
6561 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6562 xvp, td);
6563 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL);
6564 vp = nd.ni_vp;
6565 NDFREE(&nd, NDF_ONLY_PNBUF);
6566 if (error != 0) {
6567 ZFS_EXIT(zfsvfs);
6568 if (error == ENOENT)
6569 error = ENOATTR;
6570 return (error);
6571 }
6572
6573 if (ap->a_size != NULL) {
6574 error = VOP_GETATTR(vp, &va, ap->a_cred);
6575 if (error == 0)
6576 *ap->a_size = (size_t)va.va_size;
6577 } else if (ap->a_uio != NULL)
6578 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred);
6579
6580 VOP_UNLOCK(vp, 0);
6581 vn_close(vp, flags, ap->a_cred, td);
6582 ZFS_EXIT(zfsvfs);
6583
6584 return (error);
6585}
6586
6587/*
6588 * Vnode operation to remove a named attribute.
6589 */
6590int
6591zfs_deleteextattr(struct vop_deleteextattr_args *ap)
6592/*
6593vop_deleteextattr {
6594 IN struct vnode *a_vp;
6595 IN int a_attrnamespace;
6596 IN const char *a_name;
6597 IN struct ucred *a_cred;
6598 IN struct thread *a_td;
6599};
6600*/
6601{
6602 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6603 struct thread *td = ap->a_td;
6604 struct nameidata nd;
6605 char attrname[255];
6606 struct vattr va;
6607 vnode_t *xvp = NULL, *vp;
6608 int error, flags;
6609
6610 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6611 ap->a_cred, ap->a_td, VWRITE);
6612 if (error != 0)
6613 return (error);
6614
6615 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6616 sizeof(attrname));
6617 if (error != 0)
6618 return (error);
6619
6620 ZFS_ENTER(zfsvfs);
6621
6622 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6623 LOOKUP_XATTR);
6624 if (error != 0) {
6625 ZFS_EXIT(zfsvfs);
6626 return (error);
6627 }
6628
6629 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF,
6630 UIO_SYSSPACE, attrname, xvp, td);
6631 error = namei(&nd);
6632 vp = nd.ni_vp;
6633 NDFREE(&nd, NDF_ONLY_PNBUF);
6634 if (error != 0) {
6635 ZFS_EXIT(zfsvfs);
6636 if (error == ENOENT)
6637 error = ENOATTR;
6638 return (error);
6639 }
6640 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd);
6641
6642 vput(nd.ni_dvp);
6643 if (vp == nd.ni_dvp)
6644 vrele(vp);
6645 else
6646 vput(vp);
6647 ZFS_EXIT(zfsvfs);
6648
6649 return (error);
6650}
6651
6652/*
6653 * Vnode operation to set a named attribute.
6654 */
6655static int
6656zfs_setextattr(struct vop_setextattr_args *ap)
6657/*
6658vop_setextattr {
6659 IN struct vnode *a_vp;
6660 IN int a_attrnamespace;
6661 IN const char *a_name;
6662 INOUT struct uio *a_uio;
6663 IN struct ucred *a_cred;
6664 IN struct thread *a_td;
6665};
6666*/
6667{
6668 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6669 struct thread *td = ap->a_td;
6670 struct nameidata nd;
6671 char attrname[255];
6672 struct vattr va;
6673 vnode_t *xvp = NULL, *vp;
6674 int error, flags;
6675
6676 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6677 ap->a_cred, ap->a_td, VWRITE);
6678 if (error != 0)
6679 return (error);
6680
6681 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname,
6682 sizeof(attrname));
6683 if (error != 0)
6684 return (error);
6685
6686 ZFS_ENTER(zfsvfs);
6687
6688 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6689 LOOKUP_XATTR | CREATE_XATTR_DIR);
6690 if (error != 0) {
6691 ZFS_EXIT(zfsvfs);
6692 return (error);
6693 }
6694
6695 flags = FFLAGS(O_WRONLY | O_CREAT);
6696 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname,
6697 xvp, td);
6698 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL);
6699 vp = nd.ni_vp;
6700 NDFREE(&nd, NDF_ONLY_PNBUF);
6701 if (error != 0) {
6702 ZFS_EXIT(zfsvfs);
6703 return (error);
6704 }
6705
6706 VATTR_NULL(&va);
6707 va.va_size = 0;
6708 error = VOP_SETATTR(vp, &va, ap->a_cred);
6709 if (error == 0)
6710 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred);
6711
6712 VOP_UNLOCK(vp, 0);
6713 vn_close(vp, flags, ap->a_cred, td);
6714 ZFS_EXIT(zfsvfs);
6715
6716 return (error);
6717}
6718
6719/*
6720 * Vnode operation to retrieve extended attributes on a vnode.
6721 */
6722static int
6723zfs_listextattr(struct vop_listextattr_args *ap)
6724/*
6725vop_listextattr {
6726 IN struct vnode *a_vp;
6727 IN int a_attrnamespace;
6728 INOUT struct uio *a_uio;
6729 OUT size_t *a_size;
6730 IN struct ucred *a_cred;
6731 IN struct thread *a_td;
6732};
6733*/
6734{
6735 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs;
6736 struct thread *td = ap->a_td;
6737 struct nameidata nd;
6738 char attrprefix[16];
6739 u_char dirbuf[sizeof(struct dirent)];
6740 struct dirent *dp;
6741 struct iovec aiov;
6742 struct uio auio, *uio = ap->a_uio;
6743 size_t *sizep = ap->a_size;
6744 size_t plen;
6745 vnode_t *xvp = NULL, *vp;
6746 int done, error, eof, pos;
6747
6748 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace,
6749 ap->a_cred, ap->a_td, VREAD);
6750 if (error != 0)
6751 return (error);
6752
6753 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix,
6754 sizeof(attrprefix));
6755 if (error != 0)
6756 return (error);
6757 plen = strlen(attrprefix);
6758
6759 ZFS_ENTER(zfsvfs);
6760
6761 if (sizep != NULL)
6762 *sizep = 0;
6763
6764 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td,
6765 LOOKUP_XATTR);
6766 if (error != 0) {
6767 ZFS_EXIT(zfsvfs);
6768 /*
6769 * ENOATTR means that the EA directory does not yet exist,
6770 * i.e. there are no extended attributes there.
6771 */
6772 if (error == ENOATTR)
6773 error = 0;
6774 return (error);
6775 }
6776
6777 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED,
6778 UIO_SYSSPACE, ".", xvp, td);
6779 error = namei(&nd);
6780 vp = nd.ni_vp;
6781 NDFREE(&nd, NDF_ONLY_PNBUF);
6782 if (error != 0) {
6783 ZFS_EXIT(zfsvfs);
6784 return (error);
6785 }
6786
6787 auio.uio_iov = &aiov;
6788 auio.uio_iovcnt = 1;
6789 auio.uio_segflg = UIO_SYSSPACE;
6790 auio.uio_td = td;
6791 auio.uio_rw = UIO_READ;
6792 auio.uio_offset = 0;
6793
6794 do {
6795 u_char nlen;
6796
6797 aiov.iov_base = (void *)dirbuf;
6798 aiov.iov_len = sizeof(dirbuf);
6799 auio.uio_resid = sizeof(dirbuf);
6800 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL);
6801 done = sizeof(dirbuf) - auio.uio_resid;
6802 if (error != 0)
6803 break;
6804 for (pos = 0; pos < done;) {
6805 dp = (struct dirent *)(dirbuf + pos);
6806 pos += dp->d_reclen;
6807 /*
6808 * XXX: Temporarily we also accept DT_UNKNOWN, as this
6809 * is what we get when attribute was created on Solaris.
6810 */
6811 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN)
6812 continue;
6813 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0)
6814 continue;
6815 else if (strncmp(dp->d_name, attrprefix, plen) != 0)
6816 continue;
6817 nlen = dp->d_namlen - plen;
6818 if (sizep != NULL)
6819 *sizep += 1 + nlen;
6820 else if (uio != NULL) {
6821 /*
6822 * Format of extattr name entry is one byte for
6823 * length and the rest for name.
6824 */
6825 error = uiomove(&nlen, 1, uio->uio_rw, uio);
6826 if (error == 0) {
6827 error = uiomove(dp->d_name + plen, nlen,
6828 uio->uio_rw, uio);
6829 }
6830 if (error != 0)
6831 break;
6832 }
6833 }
6834 } while (!eof && error == 0);
6835
6836 vput(vp);
6837 ZFS_EXIT(zfsvfs);
6838
6839 return (error);
6840}
6841
6842int
6843zfs_freebsd_getacl(ap)
6844 struct vop_getacl_args /* {
6845 struct vnode *vp;
6846 acl_type_t type;
6847 struct acl *aclp;
6848 struct ucred *cred;
6849 struct thread *td;
6850 } */ *ap;
6851{
6852 int error;
6853 vsecattr_t vsecattr;
6854
6855 if (ap->a_type != ACL_TYPE_NFS4)
6856 return (EINVAL);
6857
6858 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT;
6859 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL))
6860 return (error);
6861
6862 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt);
6863 if (vsecattr.vsa_aclentp != NULL)
6864 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz);
6865
6866 return (error);
6867}
6868
6869int
6870zfs_freebsd_setacl(ap)
6871 struct vop_setacl_args /* {
6872 struct vnode *vp;
6873 acl_type_t type;
6874 struct acl *aclp;
6875 struct ucred *cred;
6876 struct thread *td;
6877 } */ *ap;
6878{
6879 int error;
6880 vsecattr_t vsecattr;
6881 int aclbsize; /* size of acl list in bytes */
6882 aclent_t *aaclp;
6883
6884 if (ap->a_type != ACL_TYPE_NFS4)
6885 return (EINVAL);
6886
6887 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES)
6888 return (EINVAL);
6889
6890 /*
6891 * With NFSv4 ACLs, chmod(2) may need to add additional entries,
6892 * splitting every entry into two and appending "canonical six"
6893 * entries at the end. Don't allow for setting an ACL that would
6894 * cause chmod(2) to run out of ACL entries.
6895 */
6896 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES)
6897 return (ENOSPC);
6898
6899 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR);
6900 if (error != 0)
6901 return (error);
6902
6903 vsecattr.vsa_mask = VSA_ACE;
6904 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t);
6905 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP);
6906 aaclp = vsecattr.vsa_aclentp;
6907 vsecattr.vsa_aclentsz = aclbsize;
6908
6909 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp);
6910 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL);
6911 kmem_free(aaclp, aclbsize);
6912
6913 return (error);
6914}
6915
6916int
6917zfs_freebsd_aclcheck(ap)
6918 struct vop_aclcheck_args /* {
6919 struct vnode *vp;
6920 acl_type_t type;
6921 struct acl *aclp;
6922 struct ucred *cred;
6923 struct thread *td;
6924 } */ *ap;
6925{
6926
6927 return (EOPNOTSUPP);
6928}
6929
6930struct vop_vector zfs_vnodeops;
6931struct vop_vector zfs_fifoops;
6932struct vop_vector zfs_shareops;
6933
6934struct vop_vector zfs_vnodeops = {
6935 .vop_default = &default_vnodeops,
6936 .vop_inactive = zfs_freebsd_inactive,
6937 .vop_reclaim = zfs_freebsd_reclaim,
6938 .vop_access = zfs_freebsd_access,
6939#ifdef FREEBSD_NAMECACHE
6940 .vop_lookup = vfs_cache_lookup,
6941 .vop_cachedlookup = zfs_freebsd_lookup,
6942#else
6943 .vop_lookup = zfs_freebsd_lookup,
6944#endif
6945 .vop_getattr = zfs_freebsd_getattr,
6946 .vop_setattr = zfs_freebsd_setattr,
6947 .vop_create = zfs_freebsd_create,
6948 .vop_mknod = zfs_freebsd_create,
6949 .vop_mkdir = zfs_freebsd_mkdir,
6950 .vop_readdir = zfs_freebsd_readdir,
6951 .vop_fsync = zfs_freebsd_fsync,
6952 .vop_open = zfs_freebsd_open,
6953 .vop_close = zfs_freebsd_close,
6954 .vop_rmdir = zfs_freebsd_rmdir,
6955 .vop_ioctl = zfs_freebsd_ioctl,
6956 .vop_link = zfs_freebsd_link,
6957 .vop_symlink = zfs_freebsd_symlink,
6958 .vop_readlink = zfs_freebsd_readlink,
6959 .vop_read = zfs_freebsd_read,
6960 .vop_write = zfs_freebsd_write,
6961 .vop_remove = zfs_freebsd_remove,
6962 .vop_rename = zfs_freebsd_rename,
6963 .vop_pathconf = zfs_freebsd_pathconf,
6964 .vop_bmap = zfs_freebsd_bmap,
6965 .vop_fid = zfs_freebsd_fid,
6966 .vop_getextattr = zfs_getextattr,
6967 .vop_deleteextattr = zfs_deleteextattr,
6968 .vop_setextattr = zfs_setextattr,
6969 .vop_listextattr = zfs_listextattr,
6970 .vop_getacl = zfs_freebsd_getacl,
6971 .vop_setacl = zfs_freebsd_setacl,
6972 .vop_aclcheck = zfs_freebsd_aclcheck,
6973 .vop_getpages = zfs_freebsd_getpages,
6974};
6975
6976struct vop_vector zfs_fifoops = {
6977 .vop_default = &fifo_specops,
6978 .vop_fsync = zfs_freebsd_fsync,
6979 .vop_access = zfs_freebsd_access,
6980 .vop_getattr = zfs_freebsd_getattr,
6981 .vop_inactive = zfs_freebsd_inactive,
6982 .vop_read = VOP_PANIC,
6983 .vop_reclaim = zfs_freebsd_reclaim,
6984 .vop_setattr = zfs_freebsd_setattr,
6985 .vop_write = VOP_PANIC,
6986 .vop_pathconf = zfs_freebsd_fifo_pathconf,
6987 .vop_fid = zfs_freebsd_fid,
6988 .vop_getacl = zfs_freebsd_getacl,
6989 .vop_setacl = zfs_freebsd_setacl,
6990 .vop_aclcheck = zfs_freebsd_aclcheck,
6991};
6992
6993/*
6994 * special share hidden files vnode operations template
6995 */
6996struct vop_vector zfs_shareops = {
6997 .vop_default = &default_vnodeops,
6998 .vop_access = zfs_freebsd_access,
6999 .vop_inactive = zfs_freebsd_inactive,
7000 .vop_reclaim = zfs_freebsd_reclaim,
7001 .vop_fid = zfs_freebsd_fid,
7002 .vop_pathconf = zfs_freebsd_pathconf,
7003};