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) 2012, 2015 by Delphix. All rights reserved.
24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 * Copyright 2015 Joyent, Inc.
27 */
28
29 /* Portions Copyright 2007 Jeremy Teo */
30 /* Portions Copyright 2010 Robert Milkowski */
31
32 #include <sys/types.h>
33 #include <sys/param.h>
34 #include <sys/time.h>
35 #include <sys/systm.h>
36 #include <sys/sysmacros.h>
37 #include <sys/resource.h>
38 #include <sys/vfs.h>
39 #include <sys/vfs_opreg.h>
40 #include <sys/vnode.h>
41 #include <sys/file.h>
42 #include <sys/stat.h>
43 #include <sys/kmem.h>
44 #include <sys/taskq.h>
45 #include <sys/uio.h>
46 #include <sys/vmsystm.h>
47 #include <sys/atomic.h>
48 #include <sys/vm.h>
49 #include <vm/seg_vn.h>
50 #include <vm/pvn.h>
51 #include <vm/as.h>
52 #include <vm/kpm.h>
53 #include <vm/seg_kpm.h>
54 #include <sys/mman.h>
55 #include <sys/pathname.h>
56 #include <sys/cmn_err.h>
57 #include <sys/errno.h>
58 #include <sys/unistd.h>
59 #include <sys/zfs_dir.h>
60 #include <sys/zfs_acl.h>
61 #include <sys/zfs_ioctl.h>
62 #include <sys/fs/zfs.h>
63 #include <sys/dmu.h>
64 #include <sys/dmu_objset.h>
65 #include <sys/spa.h>
66 #include <sys/txg.h>
67 #include <sys/dbuf.h>
68 #include <sys/zap.h>
69 #include <sys/sa.h>
70 #include <sys/dirent.h>
71 #include <sys/policy.h>
72 #include <sys/sunddi.h>
73 #include <sys/filio.h>
74 #include <sys/sid.h>
75 #include "fs/fs_subr.h"
76 #include <sys/zfs_ctldir.h>
77 #include <sys/zfs_fuid.h>
78 #include <sys/zfs_sa.h>
79 #include <sys/dnlc.h>
80 #include <sys/zfs_rlock.h>
81 #include <sys/extdirent.h>
82 #include <sys/kidmap.h>
83 #include <sys/cred.h>
84 #include <sys/attr.h>
85
86 /*
87 * Programming rules.
88 *
89 * Each vnode op performs some logical unit of work. To do this, the ZPL must
90 * properly lock its in-core state, create a DMU transaction, do the work,
91 * record this work in the intent log (ZIL), commit the DMU transaction,
92 * and wait for the intent log to commit if it is a synchronous operation.
93 * Moreover, the vnode ops must work in both normal and log replay context.
94 * The ordering of events is important to avoid deadlocks and references
95 * to freed memory. The example below illustrates the following Big Rules:
96 *
97 * (1) A check must be made in each zfs thread for a mounted file system.
98 * This is done avoiding races using ZFS_ENTER(zfsvfs).
99 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
100 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
101 * can return EIO from the calling function.
102 *
103 * (2) VN_RELE() should always be the last thing except for zil_commit()
104 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
105 * First, if it's the last reference, the vnode/znode
106 * can be freed, so the zp may point to freed memory. Second, the last
107 * reference will call zfs_zinactive(), which may induce a lot of work --
108 * pushing cached pages (which acquires range locks) and syncing out
109 * cached atime changes. Third, zfs_zinactive() may require a new tx,
110 * which could deadlock the system if you were already holding one.
111 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
112 *
113 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
114 * as they can span dmu_tx_assign() calls.
115 *
116 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
117 * dmu_tx_assign(). This is critical because we don't want to block
118 * while holding locks.
119 *
120 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
121 * reduces lock contention and CPU usage when we must wait (note that if
122 * throughput is constrained by the storage, nearly every transaction
123 * must wait).
124 *
125 * Note, in particular, that if a lock is sometimes acquired before
126 * the tx assigns, and sometimes after (e.g. z_lock), then failing
127 * to use a non-blocking assign can deadlock the system. The scenario:
128 *
129 * Thread A has grabbed a lock before calling dmu_tx_assign().
130 * Thread B is in an already-assigned tx, and blocks for this lock.
131 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
132 * forever, because the previous txg can't quiesce until B's tx commits.
133 *
134 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
135 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
136 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
137 * to indicate that this operation has already called dmu_tx_wait().
138 * This will ensure that we don't retry forever, waiting a short bit
139 * each time.
140 *
141 * (5) If the operation succeeded, generate the intent log entry for it
142 * before dropping locks. This ensures that the ordering of events
143 * in the intent log matches the order in which they actually occurred.
144 * During ZIL replay the zfs_log_* functions will update the sequence
145 * number to indicate the zil transaction has replayed.
146 *
147 * (6) At the end of each vnode op, the DMU tx must always commit,
148 * regardless of whether there were any errors.
149 *
150 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
151 * to ensure that synchronous semantics are provided when necessary.
152 *
153 * In general, this is how things should be ordered in each vnode op:
154 *
155 * ZFS_ENTER(zfsvfs); // exit if unmounted
156 * top:
157 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
158 * rw_enter(...); // grab any other locks you need
159 * tx = dmu_tx_create(...); // get DMU tx
160 * dmu_tx_hold_*(); // hold each object you might modify
161 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
162 * if (error) {
163 * rw_exit(...); // drop locks
164 * zfs_dirent_unlock(dl); // unlock directory entry
165 * VN_RELE(...); // release held vnodes
166 * if (error == ERESTART) {
167 * waited = B_TRUE;
168 * dmu_tx_wait(tx);
169 * dmu_tx_abort(tx);
170 * goto top;
171 * }
172 * dmu_tx_abort(tx); // abort DMU tx
173 * ZFS_EXIT(zfsvfs); // finished in zfs
174 * return (error); // really out of space
175 * }
176 * error = do_real_work(); // do whatever this VOP does
177 * if (error == 0)
178 * zfs_log_*(...); // on success, make ZIL entry
179 * dmu_tx_commit(tx); // commit DMU tx -- error or not
180 * rw_exit(...); // drop locks
181 * zfs_dirent_unlock(dl); // unlock directory entry
182 * VN_RELE(...); // release held vnodes
183 * zil_commit(zilog, foid); // synchronous when necessary
184 * ZFS_EXIT(zfsvfs); // finished in zfs
185 * return (error); // done, report error
186 */
187
188 /* ARGSUSED */
189 static int
190 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
191 {
192 znode_t *zp = VTOZ(*vpp);
193 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
194
195 ZFS_ENTER(zfsvfs);
196 ZFS_VERIFY_ZP(zp);
197
198 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
199 ((flag & FAPPEND) == 0)) {
200 ZFS_EXIT(zfsvfs);
201 return (SET_ERROR(EPERM));
202 }
203
204 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
205 ZTOV(zp)->v_type == VREG &&
206 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
207 if (fs_vscan(*vpp, cr, 0) != 0) {
208 ZFS_EXIT(zfsvfs);
209 return (SET_ERROR(EACCES));
210 }
211 }
212
213 /* Keep a count of the synchronous opens in the znode */
214 if (flag & (FSYNC | FDSYNC))
215 atomic_inc_32(&zp->z_sync_cnt);
216
217 ZFS_EXIT(zfsvfs);
218 return (0);
219 }
220
221 /* ARGSUSED */
222 static int
223 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
224 caller_context_t *ct)
225 {
226 znode_t *zp = VTOZ(vp);
227 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
228
229 /*
230 * Clean up any locks held by this process on the vp.
231 */
232 cleanlocks(vp, ddi_get_pid(), 0);
233 cleanshares(vp, ddi_get_pid());
234
235 ZFS_ENTER(zfsvfs);
236 ZFS_VERIFY_ZP(zp);
237
238 /* Decrement the synchronous opens in the znode */
239 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
240 atomic_dec_32(&zp->z_sync_cnt);
241
242 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
243 ZTOV(zp)->v_type == VREG &&
244 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
245 VERIFY(fs_vscan(vp, cr, 1) == 0);
246
247 ZFS_EXIT(zfsvfs);
248 return (0);
249 }
250
251 /*
252 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
253 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
254 */
255 static int
256 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
257 {
258 znode_t *zp = VTOZ(vp);
259 uint64_t noff = (uint64_t)*off; /* new offset */
260 uint64_t file_sz;
261 int error;
262 boolean_t hole;
263
264 file_sz = zp->z_size;
265 if (noff >= file_sz) {
266 return (SET_ERROR(ENXIO));
267 }
268
269 if (cmd == _FIO_SEEK_HOLE)
270 hole = B_TRUE;
271 else
272 hole = B_FALSE;
273
274 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
275
276 if (error == ESRCH)
277 return (SET_ERROR(ENXIO));
278
279 /*
280 * We could find a hole that begins after the logical end-of-file,
281 * because dmu_offset_next() only works on whole blocks. If the
282 * EOF falls mid-block, then indicate that the "virtual hole"
283 * at the end of the file begins at the logical EOF, rather than
284 * at the end of the last block.
285 */
286 if (noff > file_sz) {
287 ASSERT(hole);
288 noff = file_sz;
289 }
290
291 if (noff < *off)
292 return (error);
293 *off = noff;
294 return (error);
295 }
296
297 /* ARGSUSED */
298 static int
299 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
300 int *rvalp, caller_context_t *ct)
301 {
302 offset_t off;
303 offset_t ndata;
304 dmu_object_info_t doi;
305 int error;
306 zfsvfs_t *zfsvfs;
307 znode_t *zp;
308
309 switch (com) {
310 case _FIOFFS:
311 {
312 return (zfs_sync(vp->v_vfsp, 0, cred));
313
314 /*
315 * The following two ioctls are used by bfu. Faking out,
316 * necessary to avoid bfu errors.
317 */
318 }
319 case _FIOGDIO:
320 case _FIOSDIO:
321 {
322 return (0);
323 }
324
325 case _FIO_SEEK_DATA:
326 case _FIO_SEEK_HOLE:
327 {
328 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
329 return (SET_ERROR(EFAULT));
330
331 zp = VTOZ(vp);
332 zfsvfs = zp->z_zfsvfs;
333 ZFS_ENTER(zfsvfs);
334 ZFS_VERIFY_ZP(zp);
335
336 /* offset parameter is in/out */
337 error = zfs_holey(vp, com, &off);
338 ZFS_EXIT(zfsvfs);
339 if (error)
340 return (error);
341 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
342 return (SET_ERROR(EFAULT));
343 return (0);
344 }
345 case _FIO_COUNT_FILLED:
346 {
347 /*
348 * _FIO_COUNT_FILLED adds a new ioctl command which
349 * exposes the number of filled blocks in a
350 * ZFS object.
351 */
352 zp = VTOZ(vp);
353 zfsvfs = zp->z_zfsvfs;
354 ZFS_ENTER(zfsvfs);
355 ZFS_VERIFY_ZP(zp);
356
357 /*
358 * Wait for all dirty blocks for this object
359 * to get synced out to disk, and the DMU info
360 * updated.
361 */
362 error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id);
363 if (error) {
364 ZFS_EXIT(zfsvfs);
365 return (error);
366 }
367
368 /*
369 * Retrieve fill count from DMU object.
370 */
371 error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi);
372 if (error) {
373 ZFS_EXIT(zfsvfs);
374 return (error);
375 }
376
377 ndata = doi.doi_fill_count;
378
379 ZFS_EXIT(zfsvfs);
380 if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag))
381 return (SET_ERROR(EFAULT));
382 return (0);
383 }
384 }
385 return (SET_ERROR(ENOTTY));
386 }
387
388 /*
389 * Utility functions to map and unmap a single physical page. These
390 * are used to manage the mappable copies of ZFS file data, and therefore
391 * do not update ref/mod bits.
392 */
393 caddr_t
394 zfs_map_page(page_t *pp, enum seg_rw rw)
395 {
396 if (kpm_enable)
397 return (hat_kpm_mapin(pp, 0));
398 ASSERT(rw == S_READ || rw == S_WRITE);
399 return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
400 (caddr_t)-1));
401 }
402
403 void
404 zfs_unmap_page(page_t *pp, caddr_t addr)
405 {
406 if (kpm_enable) {
407 hat_kpm_mapout(pp, 0, addr);
408 } else {
409 ppmapout(addr);
410 }
411 }
412
413 /*
414 * When a file is memory mapped, we must keep the IO data synchronized
415 * between the DMU cache and the memory mapped pages. What this means:
416 *
417 * On Write: If we find a memory mapped page, we write to *both*
418 * the page and the dmu buffer.
419 */
420 static void
421 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
422 {
423 int64_t off;
424
425 off = start & PAGEOFFSET;
426 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
427 page_t *pp;
428 uint64_t nbytes = MIN(PAGESIZE - off, len);
429
430 if (pp = page_lookup(vp, start, SE_SHARED)) {
431 caddr_t va;
432
433 va = zfs_map_page(pp, S_WRITE);
434 (void) dmu_read(os, oid, start+off, nbytes, va+off,
435 DMU_READ_PREFETCH);
436 zfs_unmap_page(pp, va);
437 page_unlock(pp);
438 }
439 len -= nbytes;
440 off = 0;
441 }
442 }
443
444 /*
445 * When a file is memory mapped, we must keep the IO data synchronized
446 * between the DMU cache and the memory mapped pages. What this means:
447 *
448 * On Read: We "read" preferentially from memory mapped pages,
449 * else we default from the dmu buffer.
450 *
451 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
452 * the file is memory mapped.
453 */
454 static int
455 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
456 {
457 znode_t *zp = VTOZ(vp);
458 int64_t start, off;
459 int len = nbytes;
460 int error = 0;
461
462 start = uio->uio_loffset;
463 off = start & PAGEOFFSET;
464 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
465 page_t *pp;
466 uint64_t bytes = MIN(PAGESIZE - off, len);
467
468 if (pp = page_lookup(vp, start, SE_SHARED)) {
469 caddr_t va;
470
471 va = zfs_map_page(pp, S_READ);
472 error = uiomove(va + off, bytes, UIO_READ, uio);
473 zfs_unmap_page(pp, va);
474 page_unlock(pp);
475 } else {
476 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
477 uio, bytes);
478 }
479 len -= bytes;
480 off = 0;
481 if (error)
482 break;
483 }
484 return (error);
485 }
486
487 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
488
489 /*
490 * Read bytes from specified file into supplied buffer.
491 *
492 * IN: vp - vnode of file to be read from.
493 * uio - structure supplying read location, range info,
494 * and return buffer.
495 * ioflag - SYNC flags; used to provide FRSYNC semantics.
496 * cr - credentials of caller.
497 * ct - caller context
498 *
499 * OUT: uio - updated offset and range, buffer filled.
500 *
501 * RETURN: 0 on success, error code on failure.
502 *
503 * Side Effects:
504 * vp - atime updated if byte count > 0
505 */
506 /* ARGSUSED */
507 static int
508 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
509 {
510 znode_t *zp = VTOZ(vp);
511 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
512 ssize_t n, nbytes;
513 int error = 0;
514 rl_t *rl;
515 xuio_t *xuio = NULL;
516
517 ZFS_ENTER(zfsvfs);
518 ZFS_VERIFY_ZP(zp);
519
520 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
521 ZFS_EXIT(zfsvfs);
522 return (SET_ERROR(EACCES));
523 }
524
525 /*
526 * Validate file offset
527 */
528 if (uio->uio_loffset < (offset_t)0) {
529 ZFS_EXIT(zfsvfs);
530 return (SET_ERROR(EINVAL));
531 }
532
533 /*
534 * Fasttrack empty reads
535 */
536 if (uio->uio_resid == 0) {
537 ZFS_EXIT(zfsvfs);
538 return (0);
539 }
540
541 /*
542 * Check for mandatory locks
543 */
544 if (MANDMODE(zp->z_mode)) {
545 if (error = chklock(vp, FREAD,
546 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
547 ZFS_EXIT(zfsvfs);
548 return (error);
549 }
550 }
551
552 /*
553 * If we're in FRSYNC mode, sync out this znode before reading it.
554 */
555 if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
556 zil_commit(zfsvfs->z_log, zp->z_id);
557
558 /*
559 * Lock the range against changes.
560 */
561 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
562
563 /*
564 * If we are reading past end-of-file we can skip
565 * to the end; but we might still need to set atime.
566 */
567 if (uio->uio_loffset >= zp->z_size) {
568 error = 0;
569 goto out;
570 }
571
572 ASSERT(uio->uio_loffset < zp->z_size);
573 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
574
575 if ((uio->uio_extflg == UIO_XUIO) &&
576 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
577 int nblk;
578 int blksz = zp->z_blksz;
579 uint64_t offset = uio->uio_loffset;
580
581 xuio = (xuio_t *)uio;
582 if ((ISP2(blksz))) {
583 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
584 blksz)) / blksz;
585 } else {
586 ASSERT(offset + n <= blksz);
587 nblk = 1;
588 }
589 (void) dmu_xuio_init(xuio, nblk);
590
591 if (vn_has_cached_data(vp)) {
592 /*
593 * For simplicity, we always allocate a full buffer
594 * even if we only expect to read a portion of a block.
595 */
596 while (--nblk >= 0) {
597 (void) dmu_xuio_add(xuio,
598 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
599 blksz), 0, blksz);
600 }
601 }
602 }
603
604 while (n > 0) {
605 nbytes = MIN(n, zfs_read_chunk_size -
606 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
607
608 if (vn_has_cached_data(vp)) {
609 error = mappedread(vp, nbytes, uio);
610 } else {
611 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
612 uio, nbytes);
613 }
614 if (error) {
615 /* convert checksum errors into IO errors */
616 if (error == ECKSUM)
617 error = SET_ERROR(EIO);
618 break;
619 }
620
621 n -= nbytes;
622 }
623 out:
624 zfs_range_unlock(rl);
625
626 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
627 ZFS_EXIT(zfsvfs);
628 return (error);
629 }
630
631 /*
632 * Write the bytes to a file.
633 *
634 * IN: vp - vnode of file to be written to.
635 * uio - structure supplying write location, range info,
636 * and data buffer.
637 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
638 * set if in append mode.
639 * cr - credentials of caller.
640 * ct - caller context (NFS/CIFS fem monitor only)
641 *
642 * OUT: uio - updated offset and range.
643 *
644 * RETURN: 0 on success, error code on failure.
645 *
646 * Timestamps:
647 * vp - ctime|mtime updated if byte count > 0
648 */
649
650 /* ARGSUSED */
651 static int
652 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
653 {
654 znode_t *zp = VTOZ(vp);
655 rlim64_t limit = uio->uio_llimit;
656 ssize_t start_resid = uio->uio_resid;
657 ssize_t tx_bytes;
658 uint64_t end_size;
659 dmu_tx_t *tx;
660 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
661 zilog_t *zilog;
662 offset_t woff;
663 ssize_t n, nbytes;
664 rl_t *rl;
665 int max_blksz = zfsvfs->z_max_blksz;
666 int error = 0;
667 arc_buf_t *abuf;
668 iovec_t *aiov = NULL;
669 xuio_t *xuio = NULL;
670 int i_iov = 0;
671 int iovcnt = uio->uio_iovcnt;
672 iovec_t *iovp = uio->uio_iov;
673 int write_eof;
674 int count = 0;
675 sa_bulk_attr_t bulk[4];
676 uint64_t mtime[2], ctime[2];
677
678 /*
679 * Fasttrack empty write
680 */
681 n = start_resid;
682 if (n == 0)
683 return (0);
684
685 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
686 limit = MAXOFFSET_T;
687
688 ZFS_ENTER(zfsvfs);
689 ZFS_VERIFY_ZP(zp);
690
691 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
692 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
693 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
694 &zp->z_size, 8);
695 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
696 &zp->z_pflags, 8);
697
698 /*
699 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
700 * callers might not be able to detect properly that we are read-only,
701 * so check it explicitly here.
702 */
703 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
704 ZFS_EXIT(zfsvfs);
705 return (SET_ERROR(EROFS));
706 }
707
708 /*
709 * If immutable or not appending then return EPERM
710 */
711 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
712 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
713 (uio->uio_loffset < zp->z_size))) {
714 ZFS_EXIT(zfsvfs);
715 return (SET_ERROR(EPERM));
716 }
717
718 zilog = zfsvfs->z_log;
719
720 /*
721 * Validate file offset
722 */
723 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
724 if (woff < 0) {
725 ZFS_EXIT(zfsvfs);
726 return (SET_ERROR(EINVAL));
727 }
728
729 /*
730 * Check for mandatory locks before calling zfs_range_lock()
731 * in order to prevent a deadlock with locks set via fcntl().
732 */
733 if (MANDMODE((mode_t)zp->z_mode) &&
734 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
735 ZFS_EXIT(zfsvfs);
736 return (error);
737 }
738
739 /*
740 * Pre-fault the pages to ensure slow (eg NFS) pages
741 * don't hold up txg.
742 * Skip this if uio contains loaned arc_buf.
743 */
744 if ((uio->uio_extflg == UIO_XUIO) &&
745 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
746 xuio = (xuio_t *)uio;
747 else
748 uio_prefaultpages(MIN(n, max_blksz), uio);
749
750 /*
751 * If in append mode, set the io offset pointer to eof.
752 */
753 if (ioflag & FAPPEND) {
754 /*
755 * Obtain an appending range lock to guarantee file append
756 * semantics. We reset the write offset once we have the lock.
757 */
758 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
759 woff = rl->r_off;
760 if (rl->r_len == UINT64_MAX) {
761 /*
762 * We overlocked the file because this write will cause
763 * the file block size to increase.
764 * Note that zp_size cannot change with this lock held.
765 */
766 woff = zp->z_size;
767 }
768 uio->uio_loffset = woff;
769 } else {
770 /*
771 * Note that if the file block size will change as a result of
772 * this write, then this range lock will lock the entire file
773 * so that we can re-write the block safely.
774 */
775 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
776 }
777
778 if (woff >= limit) {
779 zfs_range_unlock(rl);
780 ZFS_EXIT(zfsvfs);
781 return (SET_ERROR(EFBIG));
782 }
783
784 if ((woff + n) > limit || woff > (limit - n))
785 n = limit - woff;
786
787 /* Will this write extend the file length? */
788 write_eof = (woff + n > zp->z_size);
789
790 end_size = MAX(zp->z_size, woff + n);
791
792 /*
793 * Write the file in reasonable size chunks. Each chunk is written
794 * in a separate transaction; this keeps the intent log records small
795 * and allows us to do more fine-grained space accounting.
796 */
797 while (n > 0) {
798 abuf = NULL;
799 woff = uio->uio_loffset;
800 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
801 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
802 if (abuf != NULL)
803 dmu_return_arcbuf(abuf);
804 error = SET_ERROR(EDQUOT);
805 break;
806 }
807
808 if (xuio && abuf == NULL) {
809 ASSERT(i_iov < iovcnt);
810 aiov = &iovp[i_iov];
811 abuf = dmu_xuio_arcbuf(xuio, i_iov);
812 dmu_xuio_clear(xuio, i_iov);
813 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
814 iovec_t *, aiov, arc_buf_t *, abuf);
815 ASSERT((aiov->iov_base == abuf->b_data) ||
816 ((char *)aiov->iov_base - (char *)abuf->b_data +
817 aiov->iov_len == arc_buf_size(abuf)));
818 i_iov++;
819 } else if (abuf == NULL && n >= max_blksz &&
820 woff >= zp->z_size &&
821 P2PHASE(woff, max_blksz) == 0 &&
822 zp->z_blksz == max_blksz) {
823 /*
824 * This write covers a full block. "Borrow" a buffer
825 * from the dmu so that we can fill it before we enter
826 * a transaction. This avoids the possibility of
827 * holding up the transaction if the data copy hangs
828 * up on a pagefault (e.g., from an NFS server mapping).
829 */
830 size_t cbytes;
831
832 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
833 max_blksz);
834 ASSERT(abuf != NULL);
835 ASSERT(arc_buf_size(abuf) == max_blksz);
836 if (error = uiocopy(abuf->b_data, max_blksz,
837 UIO_WRITE, uio, &cbytes)) {
838 dmu_return_arcbuf(abuf);
839 break;
840 }
841 ASSERT(cbytes == max_blksz);
842 }
843
844 /*
845 * Start a transaction.
846 */
847 tx = dmu_tx_create(zfsvfs->z_os);
848 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
849 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
850 zfs_sa_upgrade_txholds(tx, zp);
851 error = dmu_tx_assign(tx, TXG_WAIT);
852 if (error) {
853 dmu_tx_abort(tx);
854 if (abuf != NULL)
855 dmu_return_arcbuf(abuf);
856 break;
857 }
858
859 /*
860 * If zfs_range_lock() over-locked we grow the blocksize
861 * and then reduce the lock range. This will only happen
862 * on the first iteration since zfs_range_reduce() will
863 * shrink down r_len to the appropriate size.
864 */
865 if (rl->r_len == UINT64_MAX) {
866 uint64_t new_blksz;
867
868 if (zp->z_blksz > max_blksz) {
869 /*
870 * File's blocksize is already larger than the
871 * "recordsize" property. Only let it grow to
872 * the next power of 2.
873 */
874 ASSERT(!ISP2(zp->z_blksz));
875 new_blksz = MIN(end_size,
876 1 << highbit64(zp->z_blksz));
877 } else {
878 new_blksz = MIN(end_size, max_blksz);
879 }
880 zfs_grow_blocksize(zp, new_blksz, tx);
881 zfs_range_reduce(rl, woff, n);
882 }
883
884 /*
885 * XXX - should we really limit each write to z_max_blksz?
886 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
887 */
888 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
889
890 if (abuf == NULL) {
891 tx_bytes = uio->uio_resid;
892 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
893 uio, nbytes, tx);
894 tx_bytes -= uio->uio_resid;
895 } else {
896 tx_bytes = nbytes;
897 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
898 /*
899 * If this is not a full block write, but we are
900 * extending the file past EOF and this data starts
901 * block-aligned, use assign_arcbuf(). Otherwise,
902 * write via dmu_write().
903 */
904 if (tx_bytes < max_blksz && (!write_eof ||
905 aiov->iov_base != abuf->b_data)) {
906 ASSERT(xuio);
907 dmu_write(zfsvfs->z_os, zp->z_id, woff,
908 aiov->iov_len, aiov->iov_base, tx);
909 dmu_return_arcbuf(abuf);
910 xuio_stat_wbuf_copied();
911 } else {
912 ASSERT(xuio || tx_bytes == max_blksz);
913 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
914 woff, abuf, tx);
915 }
916 ASSERT(tx_bytes <= uio->uio_resid);
917 uioskip(uio, tx_bytes);
918 }
919 if (tx_bytes && vn_has_cached_data(vp)) {
920 update_pages(vp, woff,
921 tx_bytes, zfsvfs->z_os, zp->z_id);
922 }
923
924 /*
925 * If we made no progress, we're done. If we made even
926 * partial progress, update the znode and ZIL accordingly.
927 */
928 if (tx_bytes == 0) {
929 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
930 (void *)&zp->z_size, sizeof (uint64_t), tx);
931 dmu_tx_commit(tx);
932 ASSERT(error != 0);
933 break;
934 }
935
936 /*
937 * Clear Set-UID/Set-GID bits on successful write if not
938 * privileged and at least one of the excute bits is set.
939 *
940 * It would be nice to to this after all writes have
941 * been done, but that would still expose the ISUID/ISGID
942 * to another app after the partial write is committed.
943 *
944 * Note: we don't call zfs_fuid_map_id() here because
945 * user 0 is not an ephemeral uid.
946 */
947 mutex_enter(&zp->z_acl_lock);
948 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
949 (S_IXUSR >> 6))) != 0 &&
950 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
951 secpolicy_vnode_setid_retain(cr,
952 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
953 uint64_t newmode;
954 zp->z_mode &= ~(S_ISUID | S_ISGID);
955 newmode = zp->z_mode;
956 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
957 (void *)&newmode, sizeof (uint64_t), tx);
958 }
959 mutex_exit(&zp->z_acl_lock);
960
961 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
962 B_TRUE);
963
964 /*
965 * Update the file size (zp_size) if it has changed;
966 * account for possible concurrent updates.
967 */
968 while ((end_size = zp->z_size) < uio->uio_loffset) {
969 (void) atomic_cas_64(&zp->z_size, end_size,
970 uio->uio_loffset);
971 ASSERT(error == 0);
972 }
973 /*
974 * If we are replaying and eof is non zero then force
975 * the file size to the specified eof. Note, there's no
976 * concurrency during replay.
977 */
978 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
979 zp->z_size = zfsvfs->z_replay_eof;
980
981 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
982
983 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
984 dmu_tx_commit(tx);
985
986 if (error != 0)
987 break;
988 ASSERT(tx_bytes == nbytes);
989 n -= nbytes;
990
991 if (!xuio && n > 0)
992 uio_prefaultpages(MIN(n, max_blksz), uio);
993 }
994
995 zfs_range_unlock(rl);
996
997 /*
998 * If we're in replay mode, or we made no progress, return error.
999 * Otherwise, it's at least a partial write, so it's successful.
1000 */
1001 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1002 ZFS_EXIT(zfsvfs);
1003 return (error);
1004 }
1005
1006 if (ioflag & (FSYNC | FDSYNC) ||
1007 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1008 zil_commit(zilog, zp->z_id);
1009
1010 ZFS_EXIT(zfsvfs);
1011 return (0);
1012 }
1013
1014 void
1015 zfs_get_done(zgd_t *zgd, int error)
1016 {
1017 znode_t *zp = zgd->zgd_private;
1018 objset_t *os = zp->z_zfsvfs->z_os;
1019
1020 if (zgd->zgd_db)
1021 dmu_buf_rele(zgd->zgd_db, zgd);
1022
1023 zfs_range_unlock(zgd->zgd_rl);
1024
1025 /*
1026 * Release the vnode asynchronously as we currently have the
1027 * txg stopped from syncing.
1028 */
1029 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1030
1031 if (error == 0 && zgd->zgd_bp)
1032 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1033
1034 kmem_free(zgd, sizeof (zgd_t));
1035 }
1036
1037 #ifdef DEBUG
1038 static int zil_fault_io = 0;
1039 #endif
1040
1041 /*
1042 * Get data to generate a TX_WRITE intent log record.
1043 */
1044 int
1045 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1046 {
1047 zfsvfs_t *zfsvfs = arg;
1048 objset_t *os = zfsvfs->z_os;
1049 znode_t *zp;
1050 uint64_t object = lr->lr_foid;
1051 uint64_t offset = lr->lr_offset;
1052 uint64_t size = lr->lr_length;
1053 blkptr_t *bp = &lr->lr_blkptr;
1054 dmu_buf_t *db;
1055 zgd_t *zgd;
1056 int error = 0;
1057
1058 ASSERT(zio != NULL);
1059 ASSERT(size != 0);
1060
1061 /*
1062 * Nothing to do if the file has been removed
1063 */
1064 if (zfs_zget(zfsvfs, object, &zp) != 0)
1065 return (SET_ERROR(ENOENT));
1066 if (zp->z_unlinked) {
1067 /*
1068 * Release the vnode asynchronously as we currently have the
1069 * txg stopped from syncing.
1070 */
1071 VN_RELE_ASYNC(ZTOV(zp),
1072 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1073 return (SET_ERROR(ENOENT));
1074 }
1075
1076 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1077 zgd->zgd_zilog = zfsvfs->z_log;
1078 zgd->zgd_private = zp;
1079
1080 /*
1081 * Write records come in two flavors: immediate and indirect.
1082 * For small writes it's cheaper to store the data with the
1083 * log record (immediate); for large writes it's cheaper to
1084 * sync the data and get a pointer to it (indirect) so that
1085 * we don't have to write the data twice.
1086 */
1087 if (buf != NULL) { /* immediate write */
1088 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1089 /* test for truncation needs to be done while range locked */
1090 if (offset >= zp->z_size) {
1091 error = SET_ERROR(ENOENT);
1092 } else {
1093 error = dmu_read(os, object, offset, size, buf,
1094 DMU_READ_NO_PREFETCH);
1095 }
1096 ASSERT(error == 0 || error == ENOENT);
1097 } else { /* indirect write */
1098 /*
1099 * Have to lock the whole block to ensure when it's
1100 * written out and it's checksum is being calculated
1101 * that no one can change the data. We need to re-check
1102 * blocksize after we get the lock in case it's changed!
1103 */
1104 for (;;) {
1105 uint64_t blkoff;
1106 size = zp->z_blksz;
1107 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1108 offset -= blkoff;
1109 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1110 RL_READER);
1111 if (zp->z_blksz == size)
1112 break;
1113 offset += blkoff;
1114 zfs_range_unlock(zgd->zgd_rl);
1115 }
1116 /* test for truncation needs to be done while range locked */
1117 if (lr->lr_offset >= zp->z_size)
1118 error = SET_ERROR(ENOENT);
1119 #ifdef DEBUG
1120 if (zil_fault_io) {
1121 error = SET_ERROR(EIO);
1122 zil_fault_io = 0;
1123 }
1124 #endif
1125 if (error == 0)
1126 error = dmu_buf_hold(os, object, offset, zgd, &db,
1127 DMU_READ_NO_PREFETCH);
1128
1129 if (error == 0) {
1130 blkptr_t *obp = dmu_buf_get_blkptr(db);
1131 if (obp) {
1132 ASSERT(BP_IS_HOLE(bp));
1133 *bp = *obp;
1134 }
1135
1136 zgd->zgd_db = db;
1137 zgd->zgd_bp = bp;
1138
1139 ASSERT(db->db_offset == offset);
1140 ASSERT(db->db_size == size);
1141
1142 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1143 zfs_get_done, zgd);
1144 ASSERT(error || lr->lr_length <= zp->z_blksz);
1145
1146 /*
1147 * On success, we need to wait for the write I/O
1148 * initiated by dmu_sync() to complete before we can
1149 * release this dbuf. We will finish everything up
1150 * in the zfs_get_done() callback.
1151 */
1152 if (error == 0)
1153 return (0);
1154
1155 if (error == EALREADY) {
1156 lr->lr_common.lrc_txtype = TX_WRITE2;
1157 error = 0;
1158 }
1159 }
1160 }
1161
1162 zfs_get_done(zgd, error);
1163
1164 return (error);
1165 }
1166
1167 /*ARGSUSED*/
1168 static int
1169 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1170 caller_context_t *ct)
1171 {
1172 znode_t *zp = VTOZ(vp);
1173 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1174 int error;
1175
1176 ZFS_ENTER(zfsvfs);
1177 ZFS_VERIFY_ZP(zp);
1178
1179 if (flag & V_ACE_MASK)
1180 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1181 else
1182 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1183
1184 ZFS_EXIT(zfsvfs);
1185 return (error);
1186 }
1187
1188 /*
1189 * If vnode is for a device return a specfs vnode instead.
1190 */
1191 static int
1192 specvp_check(vnode_t **vpp, cred_t *cr)
1193 {
1194 int error = 0;
1195
1196 if (IS_DEVVP(*vpp)) {
1197 struct vnode *svp;
1198
1199 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1200 VN_RELE(*vpp);
1201 if (svp == NULL)
1202 error = SET_ERROR(ENOSYS);
1203 *vpp = svp;
1204 }
1205 return (error);
1206 }
1207
1208
1209 /*
1210 * Lookup an entry in a directory, or an extended attribute directory.
1211 * If it exists, return a held vnode reference for it.
1212 *
1213 * IN: dvp - vnode of directory to search.
1214 * nm - name of entry to lookup.
1215 * pnp - full pathname to lookup [UNUSED].
1216 * flags - LOOKUP_XATTR set if looking for an attribute.
1217 * rdir - root directory vnode [UNUSED].
1218 * cr - credentials of caller.
1219 * ct - caller context
1220 * direntflags - directory lookup flags
1221 * realpnp - returned pathname.
1222 *
1223 * OUT: vpp - vnode of located entry, NULL if not found.
1224 *
1225 * RETURN: 0 on success, error code on failure.
1226 *
1227 * Timestamps:
1228 * NA
1229 */
1230 /* ARGSUSED */
1231 static int
1232 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1233 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
1234 int *direntflags, pathname_t *realpnp)
1235 {
1236 znode_t *zdp = VTOZ(dvp);
1237 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1238 int error = 0;
1239
1240 /* fast path */
1241 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1242
1243 if (dvp->v_type != VDIR) {
1244 return (SET_ERROR(ENOTDIR));
1245 } else if (zdp->z_sa_hdl == NULL) {
1246 return (SET_ERROR(EIO));
1247 }
1248
1249 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1250 error = zfs_fastaccesschk_execute(zdp, cr);
1251 if (!error) {
1252 *vpp = dvp;
1253 VN_HOLD(*vpp);
1254 return (0);
1255 }
1256 return (error);
1257 } else {
1258 vnode_t *tvp = dnlc_lookup(dvp, nm);
1259
1260 if (tvp) {
1261 error = zfs_fastaccesschk_execute(zdp, cr);
1262 if (error) {
1263 VN_RELE(tvp);
1264 return (error);
1265 }
1266 if (tvp == DNLC_NO_VNODE) {
1267 VN_RELE(tvp);
1268 return (SET_ERROR(ENOENT));
1269 } else {
1270 *vpp = tvp;
1271 return (specvp_check(vpp, cr));
1272 }
1273 }
1274 }
1275 }
1276
1277 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1278
1279 ZFS_ENTER(zfsvfs);
1280 ZFS_VERIFY_ZP(zdp);
1281
1282 *vpp = NULL;
1283
1284 if (flags & LOOKUP_XATTR) {
1285 /*
1286 * If the xattr property is off, refuse the lookup request.
1287 */
1288 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1289 ZFS_EXIT(zfsvfs);
1290 return (SET_ERROR(EINVAL));
1291 }
1292
1293 /*
1294 * We don't allow recursive attributes..
1295 * Maybe someday we will.
1296 */
1297 if (zdp->z_pflags & ZFS_XATTR) {
1298 ZFS_EXIT(zfsvfs);
1299 return (SET_ERROR(EINVAL));
1300 }
1301
1302 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1303 ZFS_EXIT(zfsvfs);
1304 return (error);
1305 }
1306
1307 /*
1308 * Do we have permission to get into attribute directory?
1309 */
1310
1311 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1312 B_FALSE, cr)) {
1313 VN_RELE(*vpp);
1314 *vpp = NULL;
1315 }
1316
1317 ZFS_EXIT(zfsvfs);
1318 return (error);
1319 }
1320
1321 if (dvp->v_type != VDIR) {
1322 ZFS_EXIT(zfsvfs);
1323 return (SET_ERROR(ENOTDIR));
1324 }
1325
1326 /*
1327 * Check accessibility of directory.
1328 */
1329
1330 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1331 ZFS_EXIT(zfsvfs);
1332 return (error);
1333 }
1334
1335 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1336 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1337 ZFS_EXIT(zfsvfs);
1338 return (SET_ERROR(EILSEQ));
1339 }
1340
1341 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1342 if (error == 0)
1343 error = specvp_check(vpp, cr);
1344
1345 ZFS_EXIT(zfsvfs);
1346 return (error);
1347 }
1348
1349 /*
1350 * Attempt to create a new entry in a directory. If the entry
1351 * already exists, truncate the file if permissible, else return
1352 * an error. Return the vp of the created or trunc'd file.
1353 *
1354 * IN: dvp - vnode of directory to put new file entry in.
1355 * name - name of new file entry.
1356 * vap - attributes of new file.
1357 * excl - flag indicating exclusive or non-exclusive mode.
1358 * mode - mode to open file with.
1359 * cr - credentials of caller.
1360 * flag - large file flag [UNUSED].
1361 * ct - caller context
1362 * vsecp - ACL to be set
1363 *
1364 * OUT: vpp - vnode of created or trunc'd entry.
1365 *
1366 * RETURN: 0 on success, error code on failure.
1367 *
1368 * Timestamps:
1369 * dvp - ctime|mtime updated if new entry created
1370 * vp - ctime|mtime always, atime if new
1371 */
1372
1373 /* ARGSUSED */
1374 static int
1375 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1376 int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1377 vsecattr_t *vsecp)
1378 {
1379 znode_t *zp, *dzp = VTOZ(dvp);
1380 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1381 zilog_t *zilog;
1382 objset_t *os;
1383 zfs_dirlock_t *dl;
1384 dmu_tx_t *tx;
1385 int error;
1386 ksid_t *ksid;
1387 uid_t uid;
1388 gid_t gid = crgetgid(cr);
1389 zfs_acl_ids_t acl_ids;
1390 boolean_t fuid_dirtied;
1391 boolean_t have_acl = B_FALSE;
1392 boolean_t waited = B_FALSE;
1393
1394 /*
1395 * If we have an ephemeral id, ACL, or XVATTR then
1396 * make sure file system is at proper version
1397 */
1398
1399 ksid = crgetsid(cr, KSID_OWNER);
1400 if (ksid)
1401 uid = ksid_getid(ksid);
1402 else
1403 uid = crgetuid(cr);
1404
1405 if (zfsvfs->z_use_fuids == B_FALSE &&
1406 (vsecp || (vap->va_mask & AT_XVATTR) ||
1407 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1408 return (SET_ERROR(EINVAL));
1409
1410 ZFS_ENTER(zfsvfs);
1411 ZFS_VERIFY_ZP(dzp);
1412 os = zfsvfs->z_os;
1413 zilog = zfsvfs->z_log;
1414
1415 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1416 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1417 ZFS_EXIT(zfsvfs);
1418 return (SET_ERROR(EILSEQ));
1419 }
1420
1421 if (vap->va_mask & AT_XVATTR) {
1422 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1423 crgetuid(cr), cr, vap->va_type)) != 0) {
1424 ZFS_EXIT(zfsvfs);
1425 return (error);
1426 }
1427 }
1428 top:
1429 *vpp = NULL;
1430
1431 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1432 vap->va_mode &= ~VSVTX;
1433
1434 if (*name == '\0') {
1435 /*
1436 * Null component name refers to the directory itself.
1437 */
1438 VN_HOLD(dvp);
1439 zp = dzp;
1440 dl = NULL;
1441 error = 0;
1442 } else {
1443 /* possible VN_HOLD(zp) */
1444 int zflg = 0;
1445
1446 if (flag & FIGNORECASE)
1447 zflg |= ZCILOOK;
1448
1449 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1450 NULL, NULL);
1451 if (error) {
1452 if (have_acl)
1453 zfs_acl_ids_free(&acl_ids);
1454 if (strcmp(name, "..") == 0)
1455 error = SET_ERROR(EISDIR);
1456 ZFS_EXIT(zfsvfs);
1457 return (error);
1458 }
1459 }
1460
1461 if (zp == NULL) {
1462 uint64_t txtype;
1463
1464 /*
1465 * Create a new file object and update the directory
1466 * to reference it.
1467 */
1468 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1469 if (have_acl)
1470 zfs_acl_ids_free(&acl_ids);
1471 goto out;
1472 }
1473
1474 /*
1475 * We only support the creation of regular files in
1476 * extended attribute directories.
1477 */
1478
1479 if ((dzp->z_pflags & ZFS_XATTR) &&
1480 (vap->va_type != VREG)) {
1481 if (have_acl)
1482 zfs_acl_ids_free(&acl_ids);
1483 error = SET_ERROR(EINVAL);
1484 goto out;
1485 }
1486
1487 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1488 cr, vsecp, &acl_ids)) != 0)
1489 goto out;
1490 have_acl = B_TRUE;
1491
1492 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1493 zfs_acl_ids_free(&acl_ids);
1494 error = SET_ERROR(EDQUOT);
1495 goto out;
1496 }
1497
1498 tx = dmu_tx_create(os);
1499
1500 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1501 ZFS_SA_BASE_ATTR_SIZE);
1502
1503 fuid_dirtied = zfsvfs->z_fuid_dirty;
1504 if (fuid_dirtied)
1505 zfs_fuid_txhold(zfsvfs, tx);
1506 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1507 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1508 if (!zfsvfs->z_use_sa &&
1509 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1510 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1511 0, acl_ids.z_aclp->z_acl_bytes);
1512 }
1513 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1514 if (error) {
1515 zfs_dirent_unlock(dl);
1516 if (error == ERESTART) {
1517 waited = B_TRUE;
1518 dmu_tx_wait(tx);
1519 dmu_tx_abort(tx);
1520 goto top;
1521 }
1522 zfs_acl_ids_free(&acl_ids);
1523 dmu_tx_abort(tx);
1524 ZFS_EXIT(zfsvfs);
1525 return (error);
1526 }
1527 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1528
1529 if (fuid_dirtied)
1530 zfs_fuid_sync(zfsvfs, tx);
1531
1532 (void) zfs_link_create(dl, zp, tx, ZNEW);
1533 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1534 if (flag & FIGNORECASE)
1535 txtype |= TX_CI;
1536 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1537 vsecp, acl_ids.z_fuidp, vap);
1538 zfs_acl_ids_free(&acl_ids);
1539 dmu_tx_commit(tx);
1540 } else {
1541 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1542
1543 if (have_acl)
1544 zfs_acl_ids_free(&acl_ids);
1545 have_acl = B_FALSE;
1546
1547 /*
1548 * A directory entry already exists for this name.
1549 */
1550 /*
1551 * Can't truncate an existing file if in exclusive mode.
1552 */
1553 if (excl == EXCL) {
1554 error = SET_ERROR(EEXIST);
1555 goto out;
1556 }
1557 /*
1558 * Can't open a directory for writing.
1559 */
1560 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1561 error = SET_ERROR(EISDIR);
1562 goto out;
1563 }
1564 /*
1565 * Verify requested access to file.
1566 */
1567 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1568 goto out;
1569 }
1570
1571 mutex_enter(&dzp->z_lock);
1572 dzp->z_seq++;
1573 mutex_exit(&dzp->z_lock);
1574
1575 /*
1576 * Truncate regular files if requested.
1577 */
1578 if ((ZTOV(zp)->v_type == VREG) &&
1579 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1580 /* we can't hold any locks when calling zfs_freesp() */
1581 zfs_dirent_unlock(dl);
1582 dl = NULL;
1583 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1584 if (error == 0) {
1585 vnevent_create(ZTOV(zp), ct);
1586 }
1587 }
1588 }
1589 out:
1590
1591 if (dl)
1592 zfs_dirent_unlock(dl);
1593
1594 if (error) {
1595 if (zp)
1596 VN_RELE(ZTOV(zp));
1597 } else {
1598 *vpp = ZTOV(zp);
1599 error = specvp_check(vpp, cr);
1600 }
1601
1602 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1603 zil_commit(zilog, 0);
1604
1605 ZFS_EXIT(zfsvfs);
1606 return (error);
1607 }
1608
1609 /*
1610 * Remove an entry from a directory.
1611 *
1612 * IN: dvp - vnode of directory to remove entry from.
1613 * name - name of entry to remove.
1614 * cr - credentials of caller.
1615 * ct - caller context
1616 * flags - case flags
1617 *
1618 * RETURN: 0 on success, error code on failure.
1619 *
1620 * Timestamps:
1621 * dvp - ctime|mtime
1622 * vp - ctime (if nlink > 0)
1623 */
1624
1625 uint64_t null_xattr = 0;
1626
1627 /*ARGSUSED*/
1628 static int
1629 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1630 int flags)
1631 {
1632 znode_t *zp, *dzp = VTOZ(dvp);
1633 znode_t *xzp;
1634 vnode_t *vp;
1635 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1636 zilog_t *zilog;
1637 uint64_t acl_obj, xattr_obj;
1638 uint64_t xattr_obj_unlinked = 0;
1639 uint64_t obj = 0;
1640 zfs_dirlock_t *dl;
1641 dmu_tx_t *tx;
1642 boolean_t may_delete_now, delete_now = FALSE;
1643 boolean_t unlinked, toobig = FALSE;
1644 uint64_t txtype;
1645 pathname_t *realnmp = NULL;
1646 pathname_t realnm;
1647 int error;
1648 int zflg = ZEXISTS;
1649 boolean_t waited = B_FALSE;
1650
1651 ZFS_ENTER(zfsvfs);
1652 ZFS_VERIFY_ZP(dzp);
1653 zilog = zfsvfs->z_log;
1654
1655 if (flags & FIGNORECASE) {
1656 zflg |= ZCILOOK;
1657 pn_alloc(&realnm);
1658 realnmp = &realnm;
1659 }
1660
1661 top:
1662 xattr_obj = 0;
1663 xzp = NULL;
1664 /*
1665 * Attempt to lock directory; fail if entry doesn't exist.
1666 */
1667 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1668 NULL, realnmp)) {
1669 if (realnmp)
1670 pn_free(realnmp);
1671 ZFS_EXIT(zfsvfs);
1672 return (error);
1673 }
1674
1675 vp = ZTOV(zp);
1676
1677 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1678 goto out;
1679 }
1680
1681 /*
1682 * Need to use rmdir for removing directories.
1683 */
1684 if (vp->v_type == VDIR) {
1685 error = SET_ERROR(EPERM);
1686 goto out;
1687 }
1688
1689 vnevent_remove(vp, dvp, name, ct);
1690
1691 if (realnmp)
1692 dnlc_remove(dvp, realnmp->pn_buf);
1693 else
1694 dnlc_remove(dvp, name);
1695
1696 mutex_enter(&vp->v_lock);
1697 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1698 mutex_exit(&vp->v_lock);
1699
1700 /*
1701 * We may delete the znode now, or we may put it in the unlinked set;
1702 * it depends on whether we're the last link, and on whether there are
1703 * other holds on the vnode. So we dmu_tx_hold() the right things to
1704 * allow for either case.
1705 */
1706 obj = zp->z_id;
1707 tx = dmu_tx_create(zfsvfs->z_os);
1708 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1709 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1710 zfs_sa_upgrade_txholds(tx, zp);
1711 zfs_sa_upgrade_txholds(tx, dzp);
1712 if (may_delete_now) {
1713 toobig =
1714 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1715 /* if the file is too big, only hold_free a token amount */
1716 dmu_tx_hold_free(tx, zp->z_id, 0,
1717 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1718 }
1719
1720 /* are there any extended attributes? */
1721 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1722 &xattr_obj, sizeof (xattr_obj));
1723 if (error == 0 && xattr_obj) {
1724 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1725 ASSERT0(error);
1726 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1727 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1728 }
1729
1730 mutex_enter(&zp->z_lock);
1731 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1732 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1733 mutex_exit(&zp->z_lock);
1734
1735 /* charge as an update -- would be nice not to charge at all */
1736 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1737
1738 /*
1739 * Mark this transaction as typically resulting in a net free of space
1740 */
1741 dmu_tx_mark_netfree(tx);
1742
1743 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1744 if (error) {
1745 zfs_dirent_unlock(dl);
1746 VN_RELE(vp);
1747 if (xzp)
1748 VN_RELE(ZTOV(xzp));
1749 if (error == ERESTART) {
1750 waited = B_TRUE;
1751 dmu_tx_wait(tx);
1752 dmu_tx_abort(tx);
1753 goto top;
1754 }
1755 if (realnmp)
1756 pn_free(realnmp);
1757 dmu_tx_abort(tx);
1758 ZFS_EXIT(zfsvfs);
1759 return (error);
1760 }
1761
1762 /*
1763 * Remove the directory entry.
1764 */
1765 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1766
1767 if (error) {
1768 dmu_tx_commit(tx);
1769 goto out;
1770 }
1771
1772 if (unlinked) {
1773 /*
1774 * Hold z_lock so that we can make sure that the ACL obj
1775 * hasn't changed. Could have been deleted due to
1776 * zfs_sa_upgrade().
1777 */
1778 mutex_enter(&zp->z_lock);
1779 mutex_enter(&vp->v_lock);
1780 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1781 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1782 delete_now = may_delete_now && !toobig &&
1783 vp->v_count == 1 && !vn_has_cached_data(vp) &&
1784 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1785 acl_obj;
1786 mutex_exit(&vp->v_lock);
1787 }
1788
1789 if (delete_now) {
1790 if (xattr_obj_unlinked) {
1791 ASSERT3U(xzp->z_links, ==, 2);
1792 mutex_enter(&xzp->z_lock);
1793 xzp->z_unlinked = 1;
1794 xzp->z_links = 0;
1795 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1796 &xzp->z_links, sizeof (xzp->z_links), tx);
1797 ASSERT3U(error, ==, 0);
1798 mutex_exit(&xzp->z_lock);
1799 zfs_unlinked_add(xzp, tx);
1800
1801 if (zp->z_is_sa)
1802 error = sa_remove(zp->z_sa_hdl,
1803 SA_ZPL_XATTR(zfsvfs), tx);
1804 else
1805 error = sa_update(zp->z_sa_hdl,
1806 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1807 sizeof (uint64_t), tx);
1808 ASSERT0(error);
1809 }
1810 mutex_enter(&vp->v_lock);
1811 vp->v_count--;
1812 ASSERT0(vp->v_count);
1813 mutex_exit(&vp->v_lock);
1814 mutex_exit(&zp->z_lock);
1815 zfs_znode_delete(zp, tx);
1816 } else if (unlinked) {
1817 mutex_exit(&zp->z_lock);
1818 zfs_unlinked_add(zp, tx);
1819 }
1820
1821 txtype = TX_REMOVE;
1822 if (flags & FIGNORECASE)
1823 txtype |= TX_CI;
1824 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1825
1826 dmu_tx_commit(tx);
1827 out:
1828 if (realnmp)
1829 pn_free(realnmp);
1830
1831 zfs_dirent_unlock(dl);
1832
1833 if (!delete_now)
1834 VN_RELE(vp);
1835 if (xzp)
1836 VN_RELE(ZTOV(xzp));
1837
1838 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1839 zil_commit(zilog, 0);
1840
1841 ZFS_EXIT(zfsvfs);
1842 return (error);
1843 }
1844
1845 /*
1846 * Create a new directory and insert it into dvp using the name
1847 * provided. Return a pointer to the inserted directory.
1848 *
1849 * IN: dvp - vnode of directory to add subdir to.
1850 * dirname - name of new directory.
1851 * vap - attributes of new directory.
1852 * cr - credentials of caller.
1853 * ct - caller context
1854 * flags - case flags
1855 * vsecp - ACL to be set
1856 *
1857 * OUT: vpp - vnode of created directory.
1858 *
1859 * RETURN: 0 on success, error code on failure.
1860 *
1861 * Timestamps:
1862 * dvp - ctime|mtime updated
1863 * vp - ctime|mtime|atime updated
1864 */
1865 /*ARGSUSED*/
1866 static int
1867 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1868 caller_context_t *ct, int flags, vsecattr_t *vsecp)
1869 {
1870 znode_t *zp, *dzp = VTOZ(dvp);
1871 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1872 zilog_t *zilog;
1873 zfs_dirlock_t *dl;
1874 uint64_t txtype;
1875 dmu_tx_t *tx;
1876 int error;
1877 int zf = ZNEW;
1878 ksid_t *ksid;
1879 uid_t uid;
1880 gid_t gid = crgetgid(cr);
1881 zfs_acl_ids_t acl_ids;
1882 boolean_t fuid_dirtied;
1883 boolean_t waited = B_FALSE;
1884
1885 ASSERT(vap->va_type == VDIR);
1886
1887 /*
1888 * If we have an ephemeral id, ACL, or XVATTR then
1889 * make sure file system is at proper version
1890 */
1891
1892 ksid = crgetsid(cr, KSID_OWNER);
1893 if (ksid)
1894 uid = ksid_getid(ksid);
1895 else
1896 uid = crgetuid(cr);
1897 if (zfsvfs->z_use_fuids == B_FALSE &&
1898 (vsecp || (vap->va_mask & AT_XVATTR) ||
1899 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1900 return (SET_ERROR(EINVAL));
1901
1902 ZFS_ENTER(zfsvfs);
1903 ZFS_VERIFY_ZP(dzp);
1904 zilog = zfsvfs->z_log;
1905
1906 if (dzp->z_pflags & ZFS_XATTR) {
1907 ZFS_EXIT(zfsvfs);
1908 return (SET_ERROR(EINVAL));
1909 }
1910
1911 if (zfsvfs->z_utf8 && u8_validate(dirname,
1912 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1913 ZFS_EXIT(zfsvfs);
1914 return (SET_ERROR(EILSEQ));
1915 }
1916 if (flags & FIGNORECASE)
1917 zf |= ZCILOOK;
1918
1919 if (vap->va_mask & AT_XVATTR) {
1920 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1921 crgetuid(cr), cr, vap->va_type)) != 0) {
1922 ZFS_EXIT(zfsvfs);
1923 return (error);
1924 }
1925 }
1926
1927 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1928 vsecp, &acl_ids)) != 0) {
1929 ZFS_EXIT(zfsvfs);
1930 return (error);
1931 }
1932 /*
1933 * First make sure the new directory doesn't exist.
1934 *
1935 * Existence is checked first to make sure we don't return
1936 * EACCES instead of EEXIST which can cause some applications
1937 * to fail.
1938 */
1939 top:
1940 *vpp = NULL;
1941
1942 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1943 NULL, NULL)) {
1944 zfs_acl_ids_free(&acl_ids);
1945 ZFS_EXIT(zfsvfs);
1946 return (error);
1947 }
1948
1949 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1950 zfs_acl_ids_free(&acl_ids);
1951 zfs_dirent_unlock(dl);
1952 ZFS_EXIT(zfsvfs);
1953 return (error);
1954 }
1955
1956 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1957 zfs_acl_ids_free(&acl_ids);
1958 zfs_dirent_unlock(dl);
1959 ZFS_EXIT(zfsvfs);
1960 return (SET_ERROR(EDQUOT));
1961 }
1962
1963 /*
1964 * Add a new entry to the directory.
1965 */
1966 tx = dmu_tx_create(zfsvfs->z_os);
1967 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1968 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1969 fuid_dirtied = zfsvfs->z_fuid_dirty;
1970 if (fuid_dirtied)
1971 zfs_fuid_txhold(zfsvfs, tx);
1972 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1973 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1974 acl_ids.z_aclp->z_acl_bytes);
1975 }
1976
1977 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1978 ZFS_SA_BASE_ATTR_SIZE);
1979
1980 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1981 if (error) {
1982 zfs_dirent_unlock(dl);
1983 if (error == ERESTART) {
1984 waited = B_TRUE;
1985 dmu_tx_wait(tx);
1986 dmu_tx_abort(tx);
1987 goto top;
1988 }
1989 zfs_acl_ids_free(&acl_ids);
1990 dmu_tx_abort(tx);
1991 ZFS_EXIT(zfsvfs);
1992 return (error);
1993 }
1994
1995 /*
1996 * Create new node.
1997 */
1998 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1999
2000 if (fuid_dirtied)
2001 zfs_fuid_sync(zfsvfs, tx);
2002
2003 /*
2004 * Now put new name in parent dir.
2005 */
2006 (void) zfs_link_create(dl, zp, tx, ZNEW);
2007
2008 *vpp = ZTOV(zp);
2009
2010 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2011 if (flags & FIGNORECASE)
2012 txtype |= TX_CI;
2013 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2014 acl_ids.z_fuidp, vap);
2015
2016 zfs_acl_ids_free(&acl_ids);
2017
2018 dmu_tx_commit(tx);
2019
2020 zfs_dirent_unlock(dl);
2021
2022 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2023 zil_commit(zilog, 0);
2024
2025 ZFS_EXIT(zfsvfs);
2026 return (0);
2027 }
2028
2029 /*
2030 * Remove a directory subdir entry. If the current working
2031 * directory is the same as the subdir to be removed, the
2032 * remove will fail.
2033 *
2034 * IN: dvp - vnode of directory to remove from.
2035 * name - name of directory to be removed.
2036 * cwd - vnode of current working directory.
2037 * cr - credentials of caller.
2038 * ct - caller context
2039 * flags - case flags
2040 *
2041 * RETURN: 0 on success, error code on failure.
2042 *
2043 * Timestamps:
2044 * dvp - ctime|mtime updated
2045 */
2046 /*ARGSUSED*/
2047 static int
2048 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2049 caller_context_t *ct, int flags)
2050 {
2051 znode_t *dzp = VTOZ(dvp);
2052 znode_t *zp;
2053 vnode_t *vp;
2054 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2055 zilog_t *zilog;
2056 zfs_dirlock_t *dl;
2057 dmu_tx_t *tx;
2058 int error;
2059 int zflg = ZEXISTS;
2060 boolean_t waited = B_FALSE;
2061
2062 ZFS_ENTER(zfsvfs);
2063 ZFS_VERIFY_ZP(dzp);
2064 zilog = zfsvfs->z_log;
2065
2066 if (flags & FIGNORECASE)
2067 zflg |= ZCILOOK;
2068 top:
2069 zp = NULL;
2070
2071 /*
2072 * Attempt to lock directory; fail if entry doesn't exist.
2073 */
2074 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2075 NULL, NULL)) {
2076 ZFS_EXIT(zfsvfs);
2077 return (error);
2078 }
2079
2080 vp = ZTOV(zp);
2081
2082 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2083 goto out;
2084 }
2085
2086 if (vp->v_type != VDIR) {
2087 error = SET_ERROR(ENOTDIR);
2088 goto out;
2089 }
2090
2091 if (vp == cwd) {
2092 error = SET_ERROR(EINVAL);
2093 goto out;
2094 }
2095
2096 vnevent_rmdir(vp, dvp, name, ct);
2097
2098 /*
2099 * Grab a lock on the directory to make sure that noone is
2100 * trying to add (or lookup) entries while we are removing it.
2101 */
2102 rw_enter(&zp->z_name_lock, RW_WRITER);
2103
2104 /*
2105 * Grab a lock on the parent pointer to make sure we play well
2106 * with the treewalk and directory rename code.
2107 */
2108 rw_enter(&zp->z_parent_lock, RW_WRITER);
2109
2110 tx = dmu_tx_create(zfsvfs->z_os);
2111 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2112 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2113 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2114 zfs_sa_upgrade_txholds(tx, zp);
2115 zfs_sa_upgrade_txholds(tx, dzp);
2116 dmu_tx_mark_netfree(tx);
2117 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2118 if (error) {
2119 rw_exit(&zp->z_parent_lock);
2120 rw_exit(&zp->z_name_lock);
2121 zfs_dirent_unlock(dl);
2122 VN_RELE(vp);
2123 if (error == ERESTART) {
2124 waited = B_TRUE;
2125 dmu_tx_wait(tx);
2126 dmu_tx_abort(tx);
2127 goto top;
2128 }
2129 dmu_tx_abort(tx);
2130 ZFS_EXIT(zfsvfs);
2131 return (error);
2132 }
2133
2134 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2135
2136 if (error == 0) {
2137 uint64_t txtype = TX_RMDIR;
2138 if (flags & FIGNORECASE)
2139 txtype |= TX_CI;
2140 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2141 }
2142
2143 dmu_tx_commit(tx);
2144
2145 rw_exit(&zp->z_parent_lock);
2146 rw_exit(&zp->z_name_lock);
2147 out:
2148 zfs_dirent_unlock(dl);
2149
2150 VN_RELE(vp);
2151
2152 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2153 zil_commit(zilog, 0);
2154
2155 ZFS_EXIT(zfsvfs);
2156 return (error);
2157 }
2158
2159 /*
2160 * Read as many directory entries as will fit into the provided
2161 * buffer from the given directory cursor position (specified in
2162 * the uio structure).
2163 *
2164 * IN: vp - vnode of directory to read.
2165 * uio - structure supplying read location, range info,
2166 * and return buffer.
2167 * cr - credentials of caller.
2168 * ct - caller context
2169 * flags - case flags
2170 *
2171 * OUT: uio - updated offset and range, buffer filled.
2172 * eofp - set to true if end-of-file detected.
2173 *
2174 * RETURN: 0 on success, error code on failure.
2175 *
2176 * Timestamps:
2177 * vp - atime updated
2178 *
2179 * Note that the low 4 bits of the cookie returned by zap is always zero.
2180 * This allows us to use the low range for "special" directory entries:
2181 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2182 * we use the offset 2 for the '.zfs' directory.
2183 */
2184 /* ARGSUSED */
2185 static int
2186 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2187 caller_context_t *ct, int flags)
2188 {
2189 znode_t *zp = VTOZ(vp);
2190 iovec_t *iovp;
2191 edirent_t *eodp;
2192 dirent64_t *odp;
2193 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2194 objset_t *os;
2195 caddr_t outbuf;
2196 size_t bufsize;
2197 zap_cursor_t zc;
2198 zap_attribute_t zap;
2199 uint_t bytes_wanted;
2200 uint64_t offset; /* must be unsigned; checks for < 1 */
2201 uint64_t parent;
2202 int local_eof;
2203 int outcount;
2204 int error;
2205 uint8_t prefetch;
2206 boolean_t check_sysattrs;
2207
2208 ZFS_ENTER(zfsvfs);
2209 ZFS_VERIFY_ZP(zp);
2210
2211 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2212 &parent, sizeof (parent))) != 0) {
2213 ZFS_EXIT(zfsvfs);
2214 return (error);
2215 }
2216
2217 /*
2218 * If we are not given an eof variable,
2219 * use a local one.
2220 */
2221 if (eofp == NULL)
2222 eofp = &local_eof;
2223
2224 /*
2225 * Check for valid iov_len.
2226 */
2227 if (uio->uio_iov->iov_len <= 0) {
2228 ZFS_EXIT(zfsvfs);
2229 return (SET_ERROR(EINVAL));
2230 }
2231
2232 /*
2233 * Quit if directory has been removed (posix)
2234 */
2235 if ((*eofp = zp->z_unlinked) != 0) {
2236 ZFS_EXIT(zfsvfs);
2237 return (0);
2238 }
2239
2240 error = 0;
2241 os = zfsvfs->z_os;
2242 offset = uio->uio_loffset;
2243 prefetch = zp->z_zn_prefetch;
2244
2245 /*
2246 * Initialize the iterator cursor.
2247 */
2248 if (offset <= 3) {
2249 /*
2250 * Start iteration from the beginning of the directory.
2251 */
2252 zap_cursor_init(&zc, os, zp->z_id);
2253 } else {
2254 /*
2255 * The offset is a serialized cursor.
2256 */
2257 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2258 }
2259
2260 /*
2261 * Get space to change directory entries into fs independent format.
2262 */
2263 iovp = uio->uio_iov;
2264 bytes_wanted = iovp->iov_len;
2265 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2266 bufsize = bytes_wanted;
2267 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2268 odp = (struct dirent64 *)outbuf;
2269 } else {
2270 bufsize = bytes_wanted;
2271 outbuf = NULL;
2272 odp = (struct dirent64 *)iovp->iov_base;
2273 }
2274 eodp = (struct edirent *)odp;
2275
2276 /*
2277 * If this VFS supports the system attribute view interface; and
2278 * we're looking at an extended attribute directory; and we care
2279 * about normalization conflicts on this vfs; then we must check
2280 * for normalization conflicts with the sysattr name space.
2281 */
2282 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2283 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2284 (flags & V_RDDIR_ENTFLAGS);
2285
2286 /*
2287 * Transform to file-system independent format
2288 */
2289 outcount = 0;
2290 while (outcount < bytes_wanted) {
2291 ino64_t objnum;
2292 ushort_t reclen;
2293 off64_t *next = NULL;
2294
2295 /*
2296 * Special case `.', `..', and `.zfs'.
2297 */
2298 if (offset == 0) {
2299 (void) strcpy(zap.za_name, ".");
2300 zap.za_normalization_conflict = 0;
2301 objnum = zp->z_id;
2302 } else if (offset == 1) {
2303 (void) strcpy(zap.za_name, "..");
2304 zap.za_normalization_conflict = 0;
2305 objnum = parent;
2306 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2307 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2308 zap.za_normalization_conflict = 0;
2309 objnum = ZFSCTL_INO_ROOT;
2310 } else {
2311 /*
2312 * Grab next entry.
2313 */
2314 if (error = zap_cursor_retrieve(&zc, &zap)) {
2315 if ((*eofp = (error == ENOENT)) != 0)
2316 break;
2317 else
2318 goto update;
2319 }
2320
2321 if (zap.za_integer_length != 8 ||
2322 zap.za_num_integers != 1) {
2323 cmn_err(CE_WARN, "zap_readdir: bad directory "
2324 "entry, obj = %lld, offset = %lld\n",
2325 (u_longlong_t)zp->z_id,
2326 (u_longlong_t)offset);
2327 error = SET_ERROR(ENXIO);
2328 goto update;
2329 }
2330
2331 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2332 /*
2333 * MacOS X can extract the object type here such as:
2334 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2335 */
2336
2337 if (check_sysattrs && !zap.za_normalization_conflict) {
2338 zap.za_normalization_conflict =
2339 xattr_sysattr_casechk(zap.za_name);
2340 }
2341 }
2342
2343 if (flags & V_RDDIR_ACCFILTER) {
2344 /*
2345 * If we have no access at all, don't include
2346 * this entry in the returned information
2347 */
2348 znode_t *ezp;
2349 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2350 goto skip_entry;
2351 if (!zfs_has_access(ezp, cr)) {
2352 VN_RELE(ZTOV(ezp));
2353 goto skip_entry;
2354 }
2355 VN_RELE(ZTOV(ezp));
2356 }
2357
2358 if (flags & V_RDDIR_ENTFLAGS)
2359 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2360 else
2361 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2362
2363 /*
2364 * Will this entry fit in the buffer?
2365 */
2366 if (outcount + reclen > bufsize) {
2367 /*
2368 * Did we manage to fit anything in the buffer?
2369 */
2370 if (!outcount) {
2371 error = SET_ERROR(EINVAL);
2372 goto update;
2373 }
2374 break;
2375 }
2376 if (flags & V_RDDIR_ENTFLAGS) {
2377 /*
2378 * Add extended flag entry:
2379 */
2380 eodp->ed_ino = objnum;
2381 eodp->ed_reclen = reclen;
2382 /* NOTE: ed_off is the offset for the *next* entry */
2383 next = &(eodp->ed_off);
2384 eodp->ed_eflags = zap.za_normalization_conflict ?
2385 ED_CASE_CONFLICT : 0;
2386 (void) strncpy(eodp->ed_name, zap.za_name,
2387 EDIRENT_NAMELEN(reclen));
2388 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2389 } else {
2390 /*
2391 * Add normal entry:
2392 */
2393 odp->d_ino = objnum;
2394 odp->d_reclen = reclen;
2395 /* NOTE: d_off is the offset for the *next* entry */
2396 next = &(odp->d_off);
2397 (void) strncpy(odp->d_name, zap.za_name,
2398 DIRENT64_NAMELEN(reclen));
2399 odp = (dirent64_t *)((intptr_t)odp + reclen);
2400 }
2401 outcount += reclen;
2402
2403 ASSERT(outcount <= bufsize);
2404
2405 /* Prefetch znode */
2406 if (prefetch)
2407 dmu_prefetch(os, objnum, 0, 0, 0,
2408 ZIO_PRIORITY_SYNC_READ);
2409
2410 skip_entry:
2411 /*
2412 * Move to the next entry, fill in the previous offset.
2413 */
2414 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2415 zap_cursor_advance(&zc);
2416 offset = zap_cursor_serialize(&zc);
2417 } else {
2418 offset += 1;
2419 }
2420 if (next)
2421 *next = offset;
2422 }
2423 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2424
2425 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2426 iovp->iov_base += outcount;
2427 iovp->iov_len -= outcount;
2428 uio->uio_resid -= outcount;
2429 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2430 /*
2431 * Reset the pointer.
2432 */
2433 offset = uio->uio_loffset;
2434 }
2435
2436 update:
2437 zap_cursor_fini(&zc);
2438 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2439 kmem_free(outbuf, bufsize);
2440
2441 if (error == ENOENT)
2442 error = 0;
2443
2444 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2445
2446 uio->uio_loffset = offset;
2447 ZFS_EXIT(zfsvfs);
2448 return (error);
2449 }
2450
2451 ulong_t zfs_fsync_sync_cnt = 4;
2452
2453 static int
2454 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2455 {
2456 znode_t *zp = VTOZ(vp);
2457 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2458
2459 /*
2460 * Regardless of whether this is required for standards conformance,
2461 * this is the logical behavior when fsync() is called on a file with
2462 * dirty pages. We use B_ASYNC since the ZIL transactions are already
2463 * going to be pushed out as part of the zil_commit().
2464 */
2465 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2466 (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2467 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2468
2469 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2470
2471 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2472 ZFS_ENTER(zfsvfs);
2473 ZFS_VERIFY_ZP(zp);
2474 zil_commit(zfsvfs->z_log, zp->z_id);
2475 ZFS_EXIT(zfsvfs);
2476 }
2477 return (0);
2478 }
2479
2480
2481 /*
2482 * Get the requested file attributes and place them in the provided
2483 * vattr structure.
2484 *
2485 * IN: vp - vnode of file.
2486 * vap - va_mask identifies requested attributes.
2487 * If AT_XVATTR set, then optional attrs are requested
2488 * flags - ATTR_NOACLCHECK (CIFS server context)
2489 * cr - credentials of caller.
2490 * ct - caller context
2491 *
2492 * OUT: vap - attribute values.
2493 *
2494 * RETURN: 0 (always succeeds).
2495 */
2496 /* ARGSUSED */
2497 static int
2498 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2499 caller_context_t *ct)
2500 {
2501 znode_t *zp = VTOZ(vp);
2502 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2503 int error = 0;
2504 uint64_t links;
2505 uint64_t mtime[2], ctime[2];
2506 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2507 xoptattr_t *xoap = NULL;
2508 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2509 sa_bulk_attr_t bulk[2];
2510 int count = 0;
2511
2512 ZFS_ENTER(zfsvfs);
2513 ZFS_VERIFY_ZP(zp);
2514
2515 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2516
2517 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2518 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2519
2520 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2521 ZFS_EXIT(zfsvfs);
2522 return (error);
2523 }
2524
2525 /*
2526 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2527 * Also, if we are the owner don't bother, since owner should
2528 * always be allowed to read basic attributes of file.
2529 */
2530 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2531 (vap->va_uid != crgetuid(cr))) {
2532 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2533 skipaclchk, cr)) {
2534 ZFS_EXIT(zfsvfs);
2535 return (error);
2536 }
2537 }
2538
2539 /*
2540 * Return all attributes. It's cheaper to provide the answer
2541 * than to determine whether we were asked the question.
2542 */
2543
2544 mutex_enter(&zp->z_lock);
2545 vap->va_type = vp->v_type;
2546 vap->va_mode = zp->z_mode & MODEMASK;
2547 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2548 vap->va_nodeid = zp->z_id;
2549 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2550 links = zp->z_links + 1;
2551 else
2552 links = zp->z_links;
2553 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */
2554 vap->va_size = zp->z_size;
2555 vap->va_rdev = vp->v_rdev;
2556 vap->va_seq = zp->z_seq;
2557
2558 /*
2559 * Add in any requested optional attributes and the create time.
2560 * Also set the corresponding bits in the returned attribute bitmap.
2561 */
2562 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2563 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2564 xoap->xoa_archive =
2565 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2566 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2567 }
2568
2569 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2570 xoap->xoa_readonly =
2571 ((zp->z_pflags & ZFS_READONLY) != 0);
2572 XVA_SET_RTN(xvap, XAT_READONLY);
2573 }
2574
2575 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2576 xoap->xoa_system =
2577 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2578 XVA_SET_RTN(xvap, XAT_SYSTEM);
2579 }
2580
2581 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2582 xoap->xoa_hidden =
2583 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2584 XVA_SET_RTN(xvap, XAT_HIDDEN);
2585 }
2586
2587 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2588 xoap->xoa_nounlink =
2589 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2590 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2591 }
2592
2593 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2594 xoap->xoa_immutable =
2595 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2596 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2597 }
2598
2599 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2600 xoap->xoa_appendonly =
2601 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2602 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2603 }
2604
2605 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2606 xoap->xoa_nodump =
2607 ((zp->z_pflags & ZFS_NODUMP) != 0);
2608 XVA_SET_RTN(xvap, XAT_NODUMP);
2609 }
2610
2611 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2612 xoap->xoa_opaque =
2613 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2614 XVA_SET_RTN(xvap, XAT_OPAQUE);
2615 }
2616
2617 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2618 xoap->xoa_av_quarantined =
2619 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2620 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2621 }
2622
2623 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2624 xoap->xoa_av_modified =
2625 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2626 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2627 }
2628
2629 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2630 vp->v_type == VREG) {
2631 zfs_sa_get_scanstamp(zp, xvap);
2632 }
2633
2634 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2635 uint64_t times[2];
2636
2637 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2638 times, sizeof (times));
2639 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2640 XVA_SET_RTN(xvap, XAT_CREATETIME);
2641 }
2642
2643 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2644 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2645 XVA_SET_RTN(xvap, XAT_REPARSE);
2646 }
2647 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2648 xoap->xoa_generation = zp->z_gen;
2649 XVA_SET_RTN(xvap, XAT_GEN);
2650 }
2651
2652 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2653 xoap->xoa_offline =
2654 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2655 XVA_SET_RTN(xvap, XAT_OFFLINE);
2656 }
2657
2658 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2659 xoap->xoa_sparse =
2660 ((zp->z_pflags & ZFS_SPARSE) != 0);
2661 XVA_SET_RTN(xvap, XAT_SPARSE);
2662 }
2663 }
2664
2665 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2666 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2667 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2668
2669 mutex_exit(&zp->z_lock);
2670
2671 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2672
2673 if (zp->z_blksz == 0) {
2674 /*
2675 * Block size hasn't been set; suggest maximal I/O transfers.
2676 */
2677 vap->va_blksize = zfsvfs->z_max_blksz;
2678 }
2679
2680 ZFS_EXIT(zfsvfs);
2681 return (0);
2682 }
2683
2684 /*
2685 * Set the file attributes to the values contained in the
2686 * vattr structure.
2687 *
2688 * IN: vp - vnode of file to be modified.
2689 * vap - new attribute values.
2690 * If AT_XVATTR set, then optional attrs are being set
2691 * flags - ATTR_UTIME set if non-default time values provided.
2692 * - ATTR_NOACLCHECK (CIFS context only).
2693 * cr - credentials of caller.
2694 * ct - caller context
2695 *
2696 * RETURN: 0 on success, error code on failure.
2697 *
2698 * Timestamps:
2699 * vp - ctime updated, mtime updated if size changed.
2700 */
2701 /* ARGSUSED */
2702 static int
2703 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2704 caller_context_t *ct)
2705 {
2706 znode_t *zp = VTOZ(vp);
2707 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2708 zilog_t *zilog;
2709 dmu_tx_t *tx;
2710 vattr_t oldva;
2711 xvattr_t tmpxvattr;
2712 uint_t mask = vap->va_mask;
2713 uint_t saved_mask = 0;
2714 int trim_mask = 0;
2715 uint64_t new_mode;
2716 uint64_t new_uid, new_gid;
2717 uint64_t xattr_obj;
2718 uint64_t mtime[2], ctime[2];
2719 znode_t *attrzp;
2720 int need_policy = FALSE;
2721 int err, err2;
2722 zfs_fuid_info_t *fuidp = NULL;
2723 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2724 xoptattr_t *xoap;
2725 zfs_acl_t *aclp;
2726 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2727 boolean_t fuid_dirtied = B_FALSE;
2728 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2729 int count = 0, xattr_count = 0;
2730
2731 if (mask == 0)
2732 return (0);
2733
2734 if (mask & AT_NOSET)
2735 return (SET_ERROR(EINVAL));
2736
2737 ZFS_ENTER(zfsvfs);
2738 ZFS_VERIFY_ZP(zp);
2739
2740 zilog = zfsvfs->z_log;
2741
2742 /*
2743 * Make sure that if we have ephemeral uid/gid or xvattr specified
2744 * that file system is at proper version level
2745 */
2746
2747 if (zfsvfs->z_use_fuids == B_FALSE &&
2748 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2749 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2750 (mask & AT_XVATTR))) {
2751 ZFS_EXIT(zfsvfs);
2752 return (SET_ERROR(EINVAL));
2753 }
2754
2755 if (mask & AT_SIZE && vp->v_type == VDIR) {
2756 ZFS_EXIT(zfsvfs);
2757 return (SET_ERROR(EISDIR));
2758 }
2759
2760 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2761 ZFS_EXIT(zfsvfs);
2762 return (SET_ERROR(EINVAL));
2763 }
2764
2765 /*
2766 * If this is an xvattr_t, then get a pointer to the structure of
2767 * optional attributes. If this is NULL, then we have a vattr_t.
2768 */
2769 xoap = xva_getxoptattr(xvap);
2770
2771 xva_init(&tmpxvattr);
2772
2773 /*
2774 * Immutable files can only alter immutable bit and atime
2775 */
2776 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2777 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2778 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2779 ZFS_EXIT(zfsvfs);
2780 return (SET_ERROR(EPERM));
2781 }
2782
2783 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2784 ZFS_EXIT(zfsvfs);
2785 return (SET_ERROR(EPERM));
2786 }
2787
2788 /*
2789 * Verify timestamps doesn't overflow 32 bits.
2790 * ZFS can handle large timestamps, but 32bit syscalls can't
2791 * handle times greater than 2039. This check should be removed
2792 * once large timestamps are fully supported.
2793 */
2794 if (mask & (AT_ATIME | AT_MTIME)) {
2795 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2796 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2797 ZFS_EXIT(zfsvfs);
2798 return (SET_ERROR(EOVERFLOW));
2799 }
2800 }
2801
2802 top:
2803 attrzp = NULL;
2804 aclp = NULL;
2805
2806 /* Can this be moved to before the top label? */
2807 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2808 ZFS_EXIT(zfsvfs);
2809 return (SET_ERROR(EROFS));
2810 }
2811
2812 /*
2813 * First validate permissions
2814 */
2815
2816 if (mask & AT_SIZE) {
2817 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2818 if (err) {
2819 ZFS_EXIT(zfsvfs);
2820 return (err);
2821 }
2822 /*
2823 * XXX - Note, we are not providing any open
2824 * mode flags here (like FNDELAY), so we may
2825 * block if there are locks present... this
2826 * should be addressed in openat().
2827 */
2828 /* XXX - would it be OK to generate a log record here? */
2829 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2830 if (err) {
2831 ZFS_EXIT(zfsvfs);
2832 return (err);
2833 }
2834
2835 if (vap->va_size == 0)
2836 vnevent_truncate(ZTOV(zp), ct);
2837 }
2838
2839 if (mask & (AT_ATIME|AT_MTIME) ||
2840 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2841 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2842 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2843 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2844 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2845 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2846 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2847 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2848 skipaclchk, cr);
2849 }
2850
2851 if (mask & (AT_UID|AT_GID)) {
2852 int idmask = (mask & (AT_UID|AT_GID));
2853 int take_owner;
2854 int take_group;
2855
2856 /*
2857 * NOTE: even if a new mode is being set,
2858 * we may clear S_ISUID/S_ISGID bits.
2859 */
2860
2861 if (!(mask & AT_MODE))
2862 vap->va_mode = zp->z_mode;
2863
2864 /*
2865 * Take ownership or chgrp to group we are a member of
2866 */
2867
2868 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2869 take_group = (mask & AT_GID) &&
2870 zfs_groupmember(zfsvfs, vap->va_gid, cr);
2871
2872 /*
2873 * If both AT_UID and AT_GID are set then take_owner and
2874 * take_group must both be set in order to allow taking
2875 * ownership.
2876 *
2877 * Otherwise, send the check through secpolicy_vnode_setattr()
2878 *
2879 */
2880
2881 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2882 ((idmask == AT_UID) && take_owner) ||
2883 ((idmask == AT_GID) && take_group)) {
2884 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2885 skipaclchk, cr) == 0) {
2886 /*
2887 * Remove setuid/setgid for non-privileged users
2888 */
2889 secpolicy_setid_clear(vap, cr);
2890 trim_mask = (mask & (AT_UID|AT_GID));
2891 } else {
2892 need_policy = TRUE;
2893 }
2894 } else {
2895 need_policy = TRUE;
2896 }
2897 }
2898
2899 mutex_enter(&zp->z_lock);
2900 oldva.va_mode = zp->z_mode;
2901 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2902 if (mask & AT_XVATTR) {
2903 /*
2904 * Update xvattr mask to include only those attributes
2905 * that are actually changing.
2906 *
2907 * the bits will be restored prior to actually setting
2908 * the attributes so the caller thinks they were set.
2909 */
2910 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2911 if (xoap->xoa_appendonly !=
2912 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2913 need_policy = TRUE;
2914 } else {
2915 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2916 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2917 }
2918 }
2919
2920 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2921 if (xoap->xoa_nounlink !=
2922 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2923 need_policy = TRUE;
2924 } else {
2925 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2926 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2927 }
2928 }
2929
2930 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2931 if (xoap->xoa_immutable !=
2932 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2933 need_policy = TRUE;
2934 } else {
2935 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2936 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2937 }
2938 }
2939
2940 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2941 if (xoap->xoa_nodump !=
2942 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2943 need_policy = TRUE;
2944 } else {
2945 XVA_CLR_REQ(xvap, XAT_NODUMP);
2946 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2947 }
2948 }
2949
2950 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2951 if (xoap->xoa_av_modified !=
2952 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2953 need_policy = TRUE;
2954 } else {
2955 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2956 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2957 }
2958 }
2959
2960 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2961 if ((vp->v_type != VREG &&
2962 xoap->xoa_av_quarantined) ||
2963 xoap->xoa_av_quarantined !=
2964 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2965 need_policy = TRUE;
2966 } else {
2967 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2968 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2969 }
2970 }
2971
2972 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2973 mutex_exit(&zp->z_lock);
2974 ZFS_EXIT(zfsvfs);
2975 return (SET_ERROR(EPERM));
2976 }
2977
2978 if (need_policy == FALSE &&
2979 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2980 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2981 need_policy = TRUE;
2982 }
2983 }
2984
2985 mutex_exit(&zp->z_lock);
2986
2987 if (mask & AT_MODE) {
2988 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2989 err = secpolicy_setid_setsticky_clear(vp, vap,
2990 &oldva, cr);
2991 if (err) {
2992 ZFS_EXIT(zfsvfs);
2993 return (err);
2994 }
2995 trim_mask |= AT_MODE;
2996 } else {
2997 need_policy = TRUE;
2998 }
2999 }
3000
3001 if (need_policy) {
3002 /*
3003 * If trim_mask is set then take ownership
3004 * has been granted or write_acl is present and user
3005 * has the ability to modify mode. In that case remove
3006 * UID|GID and or MODE from mask so that
3007 * secpolicy_vnode_setattr() doesn't revoke it.
3008 */
3009
3010 if (trim_mask) {
3011 saved_mask = vap->va_mask;
3012 vap->va_mask &= ~trim_mask;
3013 }
3014 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3015 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3016 if (err) {
3017 ZFS_EXIT(zfsvfs);
3018 return (err);
3019 }
3020
3021 if (trim_mask)
3022 vap->va_mask |= saved_mask;
3023 }
3024
3025 /*
3026 * secpolicy_vnode_setattr, or take ownership may have
3027 * changed va_mask
3028 */
3029 mask = vap->va_mask;
3030
3031 if ((mask & (AT_UID | AT_GID))) {
3032 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3033 &xattr_obj, sizeof (xattr_obj));
3034
3035 if (err == 0 && xattr_obj) {
3036 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3037 if (err)
3038 goto out2;
3039 }
3040 if (mask & AT_UID) {
3041 new_uid = zfs_fuid_create(zfsvfs,
3042 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3043 if (new_uid != zp->z_uid &&
3044 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3045 if (attrzp)
3046 VN_RELE(ZTOV(attrzp));
3047 err = SET_ERROR(EDQUOT);
3048 goto out2;
3049 }
3050 }
3051
3052 if (mask & AT_GID) {
3053 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3054 cr, ZFS_GROUP, &fuidp);
3055 if (new_gid != zp->z_gid &&
3056 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3057 if (attrzp)
3058 VN_RELE(ZTOV(attrzp));
3059 err = SET_ERROR(EDQUOT);
3060 goto out2;
3061 }
3062 }
3063 }
3064 tx = dmu_tx_create(zfsvfs->z_os);
3065
3066 if (mask & AT_MODE) {
3067 uint64_t pmode = zp->z_mode;
3068 uint64_t acl_obj;
3069 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3070
3071 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3072 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3073 err = SET_ERROR(EPERM);
3074 goto out;
3075 }
3076
3077 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3078 goto out;
3079
3080 mutex_enter(&zp->z_lock);
3081 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3082 /*
3083 * Are we upgrading ACL from old V0 format
3084 * to V1 format?
3085 */
3086 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3087 zfs_znode_acl_version(zp) ==
3088 ZFS_ACL_VERSION_INITIAL) {
3089 dmu_tx_hold_free(tx, acl_obj, 0,
3090 DMU_OBJECT_END);
3091 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3092 0, aclp->z_acl_bytes);
3093 } else {
3094 dmu_tx_hold_write(tx, acl_obj, 0,
3095 aclp->z_acl_bytes);
3096 }
3097 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3098 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3099 0, aclp->z_acl_bytes);
3100 }
3101 mutex_exit(&zp->z_lock);
3102 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3103 } else {
3104 if ((mask & AT_XVATTR) &&
3105 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3106 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3107 else
3108 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3109 }
3110
3111 if (attrzp) {
3112 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3113 }
3114
3115 fuid_dirtied = zfsvfs->z_fuid_dirty;
3116 if (fuid_dirtied)
3117 zfs_fuid_txhold(zfsvfs, tx);
3118
3119 zfs_sa_upgrade_txholds(tx, zp);
3120
3121 err = dmu_tx_assign(tx, TXG_WAIT);
3122 if (err)
3123 goto out;
3124
3125 count = 0;
3126 /*
3127 * Set each attribute requested.
3128 * We group settings according to the locks they need to acquire.
3129 *
3130 * Note: you cannot set ctime directly, although it will be
3131 * updated as a side-effect of calling this function.
3132 */
3133
3134
3135 if (mask & (AT_UID|AT_GID|AT_MODE))
3136 mutex_enter(&zp->z_acl_lock);
3137 mutex_enter(&zp->z_lock);
3138
3139 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3140 &zp->z_pflags, sizeof (zp->z_pflags));
3141
3142 if (attrzp) {
3143 if (mask & (AT_UID|AT_GID|AT_MODE))
3144 mutex_enter(&attrzp->z_acl_lock);
3145 mutex_enter(&attrzp->z_lock);
3146 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3147 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3148 sizeof (attrzp->z_pflags));
3149 }
3150
3151 if (mask & (AT_UID|AT_GID)) {
3152
3153 if (mask & AT_UID) {
3154 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3155 &new_uid, sizeof (new_uid));
3156 zp->z_uid = new_uid;
3157 if (attrzp) {
3158 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3159 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3160 sizeof (new_uid));
3161 attrzp->z_uid = new_uid;
3162 }
3163 }
3164
3165 if (mask & AT_GID) {
3166 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3167 NULL, &new_gid, sizeof (new_gid));
3168 zp->z_gid = new_gid;
3169 if (attrzp) {
3170 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3171 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3172 sizeof (new_gid));
3173 attrzp->z_gid = new_gid;
3174 }
3175 }
3176 if (!(mask & AT_MODE)) {
3177 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3178 NULL, &new_mode, sizeof (new_mode));
3179 new_mode = zp->z_mode;
3180 }
3181 err = zfs_acl_chown_setattr(zp);
3182 ASSERT(err == 0);
3183 if (attrzp) {
3184 err = zfs_acl_chown_setattr(attrzp);
3185 ASSERT(err == 0);
3186 }
3187 }
3188
3189 if (mask & AT_MODE) {
3190 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3191 &new_mode, sizeof (new_mode));
3192 zp->z_mode = new_mode;
3193 ASSERT3U((uintptr_t)aclp, !=, NULL);
3194 err = zfs_aclset_common(zp, aclp, cr, tx);
3195 ASSERT0(err);
3196 if (zp->z_acl_cached)
3197 zfs_acl_free(zp->z_acl_cached);
3198 zp->z_acl_cached = aclp;
3199 aclp = NULL;
3200 }
3201
3202
3203 if (mask & AT_ATIME) {
3204 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3205 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3206 &zp->z_atime, sizeof (zp->z_atime));
3207 }
3208
3209 if (mask & AT_MTIME) {
3210 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3211 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3212 mtime, sizeof (mtime));
3213 }
3214
3215 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3216 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3217 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3218 NULL, mtime, sizeof (mtime));
3219 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3220 &ctime, sizeof (ctime));
3221 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3222 B_TRUE);
3223 } else if (mask != 0) {
3224 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3225 &ctime, sizeof (ctime));
3226 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3227 B_TRUE);
3228 if (attrzp) {
3229 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3230 SA_ZPL_CTIME(zfsvfs), NULL,
3231 &ctime, sizeof (ctime));
3232 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3233 mtime, ctime, B_TRUE);
3234 }
3235 }
3236 /*
3237 * Do this after setting timestamps to prevent timestamp
3238 * update from toggling bit
3239 */
3240
3241 if (xoap && (mask & AT_XVATTR)) {
3242
3243 /*
3244 * restore trimmed off masks
3245 * so that return masks can be set for caller.
3246 */
3247
3248 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3249 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3250 }
3251 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3252 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3253 }
3254 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3255 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3256 }
3257 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3258 XVA_SET_REQ(xvap, XAT_NODUMP);
3259 }
3260 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3261 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3262 }
3263 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3264 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3265 }
3266
3267 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3268 ASSERT(vp->v_type == VREG);
3269
3270 zfs_xvattr_set(zp, xvap, tx);
3271 }
3272
3273 if (fuid_dirtied)
3274 zfs_fuid_sync(zfsvfs, tx);
3275
3276 if (mask != 0)
3277 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3278
3279 mutex_exit(&zp->z_lock);
3280 if (mask & (AT_UID|AT_GID|AT_MODE))
3281 mutex_exit(&zp->z_acl_lock);
3282
3283 if (attrzp) {
3284 if (mask & (AT_UID|AT_GID|AT_MODE))
3285 mutex_exit(&attrzp->z_acl_lock);
3286 mutex_exit(&attrzp->z_lock);
3287 }
3288 out:
3289 if (err == 0 && attrzp) {
3290 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3291 xattr_count, tx);
3292 ASSERT(err2 == 0);
3293 }
3294
3295 if (attrzp)
3296 VN_RELE(ZTOV(attrzp));
3297
3298 if (aclp)
3299 zfs_acl_free(aclp);
3300
3301 if (fuidp) {
3302 zfs_fuid_info_free(fuidp);
3303 fuidp = NULL;
3304 }
3305
3306 if (err) {
3307 dmu_tx_abort(tx);
3308 if (err == ERESTART)
3309 goto top;
3310 } else {
3311 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3312 dmu_tx_commit(tx);
3313 }
3314
3315 out2:
3316 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3317 zil_commit(zilog, 0);
3318
3319 ZFS_EXIT(zfsvfs);
3320 return (err);
3321 }
3322
3323 typedef struct zfs_zlock {
3324 krwlock_t *zl_rwlock; /* lock we acquired */
3325 znode_t *zl_znode; /* znode we held */
3326 struct zfs_zlock *zl_next; /* next in list */
3327 } zfs_zlock_t;
3328
3329 /*
3330 * Drop locks and release vnodes that were held by zfs_rename_lock().
3331 */
3332 static void
3333 zfs_rename_unlock(zfs_zlock_t **zlpp)
3334 {
3335 zfs_zlock_t *zl;
3336
3337 while ((zl = *zlpp) != NULL) {
3338 if (zl->zl_znode != NULL)
3339 VN_RELE(ZTOV(zl->zl_znode));
3340 rw_exit(zl->zl_rwlock);
3341 *zlpp = zl->zl_next;
3342 kmem_free(zl, sizeof (*zl));
3343 }
3344 }
3345
3346 /*
3347 * Search back through the directory tree, using the ".." entries.
3348 * Lock each directory in the chain to prevent concurrent renames.
3349 * Fail any attempt to move a directory into one of its own descendants.
3350 * XXX - z_parent_lock can overlap with map or grow locks
3351 */
3352 static int
3353 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3354 {
3355 zfs_zlock_t *zl;
3356 znode_t *zp = tdzp;
3357 uint64_t rootid = zp->z_zfsvfs->z_root;
3358 uint64_t oidp = zp->z_id;
3359 krwlock_t *rwlp = &szp->z_parent_lock;
3360 krw_t rw = RW_WRITER;
3361
3362 /*
3363 * First pass write-locks szp and compares to zp->z_id.
3364 * Later passes read-lock zp and compare to zp->z_parent.
3365 */
3366 do {
3367 if (!rw_tryenter(rwlp, rw)) {
3368 /*
3369 * Another thread is renaming in this path.
3370 * Note that if we are a WRITER, we don't have any
3371 * parent_locks held yet.
3372 */
3373 if (rw == RW_READER && zp->z_id > szp->z_id) {
3374 /*
3375 * Drop our locks and restart
3376 */
3377 zfs_rename_unlock(&zl);
3378 *zlpp = NULL;
3379 zp = tdzp;
3380 oidp = zp->z_id;
3381 rwlp = &szp->z_parent_lock;
3382 rw = RW_WRITER;
3383 continue;
3384 } else {
3385 /*
3386 * Wait for other thread to drop its locks
3387 */
3388 rw_enter(rwlp, rw);
3389 }
3390 }
3391
3392 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3393 zl->zl_rwlock = rwlp;
3394 zl->zl_znode = NULL;
3395 zl->zl_next = *zlpp;
3396 *zlpp = zl;
3397
3398 if (oidp == szp->z_id) /* We're a descendant of szp */
3399 return (SET_ERROR(EINVAL));
3400
3401 if (oidp == rootid) /* We've hit the top */
3402 return (0);
3403
3404 if (rw == RW_READER) { /* i.e. not the first pass */
3405 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3406 if (error)
3407 return (error);
3408 zl->zl_znode = zp;
3409 }
3410 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3411 &oidp, sizeof (oidp));
3412 rwlp = &zp->z_parent_lock;
3413 rw = RW_READER;
3414
3415 } while (zp->z_id != sdzp->z_id);
3416
3417 return (0);
3418 }
3419
3420 /*
3421 * Move an entry from the provided source directory to the target
3422 * directory. Change the entry name as indicated.
3423 *
3424 * IN: sdvp - Source directory containing the "old entry".
3425 * snm - Old entry name.
3426 * tdvp - Target directory to contain the "new entry".
3427 * tnm - New entry name.
3428 * cr - credentials of caller.
3429 * ct - caller context
3430 * flags - case flags
3431 *
3432 * RETURN: 0 on success, error code on failure.
3433 *
3434 * Timestamps:
3435 * sdvp,tdvp - ctime|mtime updated
3436 */
3437 /*ARGSUSED*/
3438 static int
3439 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3440 caller_context_t *ct, int flags)
3441 {
3442 znode_t *tdzp, *szp, *tzp;
3443 znode_t *sdzp = VTOZ(sdvp);
3444 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3445 zilog_t *zilog;
3446 vnode_t *realvp;
3447 zfs_dirlock_t *sdl, *tdl;
3448 dmu_tx_t *tx;
3449 zfs_zlock_t *zl;
3450 int cmp, serr, terr;
3451 int error = 0, rm_err = 0;
3452 int zflg = 0;
3453 boolean_t waited = B_FALSE;
3454
3455 ZFS_ENTER(zfsvfs);
3456 ZFS_VERIFY_ZP(sdzp);
3457 zilog = zfsvfs->z_log;
3458
3459 /*
3460 * Make sure we have the real vp for the target directory.
3461 */
3462 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3463 tdvp = realvp;
3464
3465 tdzp = VTOZ(tdvp);
3466 ZFS_VERIFY_ZP(tdzp);
3467
3468 /*
3469 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3470 * ctldir appear to have the same v_vfsp.
3471 */
3472 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3473 ZFS_EXIT(zfsvfs);
3474 return (SET_ERROR(EXDEV));
3475 }
3476
3477 if (zfsvfs->z_utf8 && u8_validate(tnm,
3478 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3479 ZFS_EXIT(zfsvfs);
3480 return (SET_ERROR(EILSEQ));
3481 }
3482
3483 if (flags & FIGNORECASE)
3484 zflg |= ZCILOOK;
3485
3486 top:
3487 szp = NULL;
3488 tzp = NULL;
3489 zl = NULL;
3490
3491 /*
3492 * This is to prevent the creation of links into attribute space
3493 * by renaming a linked file into/outof an attribute directory.
3494 * See the comment in zfs_link() for why this is considered bad.
3495 */
3496 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3497 ZFS_EXIT(zfsvfs);
3498 return (SET_ERROR(EINVAL));
3499 }
3500
3501 /*
3502 * Lock source and target directory entries. To prevent deadlock,
3503 * a lock ordering must be defined. We lock the directory with
3504 * the smallest object id first, or if it's a tie, the one with
3505 * the lexically first name.
3506 */
3507 if (sdzp->z_id < tdzp->z_id) {
3508 cmp = -1;
3509 } else if (sdzp->z_id > tdzp->z_id) {
3510 cmp = 1;
3511 } else {
3512 /*
3513 * First compare the two name arguments without
3514 * considering any case folding.
3515 */
3516 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3517
3518 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3519 ASSERT(error == 0 || !zfsvfs->z_utf8);
3520 if (cmp == 0) {
3521 /*
3522 * POSIX: "If the old argument and the new argument
3523 * both refer to links to the same existing file,
3524 * the rename() function shall return successfully
3525 * and perform no other action."
3526 */
3527 ZFS_EXIT(zfsvfs);
3528 return (0);
3529 }
3530 /*
3531 * If the file system is case-folding, then we may
3532 * have some more checking to do. A case-folding file
3533 * system is either supporting mixed case sensitivity
3534 * access or is completely case-insensitive. Note
3535 * that the file system is always case preserving.
3536 *
3537 * In mixed sensitivity mode case sensitive behavior
3538 * is the default. FIGNORECASE must be used to
3539 * explicitly request case insensitive behavior.
3540 *
3541 * If the source and target names provided differ only
3542 * by case (e.g., a request to rename 'tim' to 'Tim'),
3543 * we will treat this as a special case in the
3544 * case-insensitive mode: as long as the source name
3545 * is an exact match, we will allow this to proceed as
3546 * a name-change request.
3547 */
3548 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3549 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3550 flags & FIGNORECASE)) &&
3551 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3552 &error) == 0) {
3553 /*
3554 * case preserving rename request, require exact
3555 * name matches
3556 */
3557 zflg |= ZCIEXACT;
3558 zflg &= ~ZCILOOK;
3559 }
3560 }
3561
3562 /*
3563 * If the source and destination directories are the same, we should
3564 * grab the z_name_lock of that directory only once.
3565 */
3566 if (sdzp == tdzp) {
3567 zflg |= ZHAVELOCK;
3568 rw_enter(&sdzp->z_name_lock, RW_READER);
3569 }
3570
3571 if (cmp < 0) {
3572 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3573 ZEXISTS | zflg, NULL, NULL);
3574 terr = zfs_dirent_lock(&tdl,
3575 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3576 } else {
3577 terr = zfs_dirent_lock(&tdl,
3578 tdzp, tnm, &tzp, zflg, NULL, NULL);
3579 serr = zfs_dirent_lock(&sdl,
3580 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3581 NULL, NULL);
3582 }
3583
3584 if (serr) {
3585 /*
3586 * Source entry invalid or not there.
3587 */
3588 if (!terr) {
3589 zfs_dirent_unlock(tdl);
3590 if (tzp)
3591 VN_RELE(ZTOV(tzp));
3592 }
3593
3594 if (sdzp == tdzp)
3595 rw_exit(&sdzp->z_name_lock);
3596
3597 if (strcmp(snm, "..") == 0)
3598 serr = SET_ERROR(EINVAL);
3599 ZFS_EXIT(zfsvfs);
3600 return (serr);
3601 }
3602 if (terr) {
3603 zfs_dirent_unlock(sdl);
3604 VN_RELE(ZTOV(szp));
3605
3606 if (sdzp == tdzp)
3607 rw_exit(&sdzp->z_name_lock);
3608
3609 if (strcmp(tnm, "..") == 0)
3610 terr = SET_ERROR(EINVAL);
3611 ZFS_EXIT(zfsvfs);
3612 return (terr);
3613 }
3614
3615 /*
3616 * Must have write access at the source to remove the old entry
3617 * and write access at the target to create the new entry.
3618 * Note that if target and source are the same, this can be
3619 * done in a single check.
3620 */
3621
3622 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3623 goto out;
3624
3625 if (ZTOV(szp)->v_type == VDIR) {
3626 /*
3627 * Check to make sure rename is valid.
3628 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3629 */
3630 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3631 goto out;
3632 }
3633
3634 /*
3635 * Does target exist?
3636 */
3637 if (tzp) {
3638 /*
3639 * Source and target must be the same type.
3640 */
3641 if (ZTOV(szp)->v_type == VDIR) {
3642 if (ZTOV(tzp)->v_type != VDIR) {
3643 error = SET_ERROR(ENOTDIR);
3644 goto out;
3645 }
3646 } else {
3647 if (ZTOV(tzp)->v_type == VDIR) {
3648 error = SET_ERROR(EISDIR);
3649 goto out;
3650 }
3651 }
3652 /*
3653 * POSIX dictates that when the source and target
3654 * entries refer to the same file object, rename
3655 * must do nothing and exit without error.
3656 */
3657 if (szp->z_id == tzp->z_id) {
3658 error = 0;
3659 goto out;
3660 }
3661 }
3662
3663 vnevent_pre_rename_src(ZTOV(szp), sdvp, snm, ct);
3664 if (tzp)
3665 vnevent_pre_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3666
3667 /*
3668 * notify the target directory if it is not the same
3669 * as source directory.
3670 */
3671 if (tdvp != sdvp) {
3672 vnevent_pre_rename_dest_dir(tdvp, ZTOV(szp), tnm, ct);
3673 }
3674
3675 tx = dmu_tx_create(zfsvfs->z_os);
3676 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3677 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3678 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3679 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3680 if (sdzp != tdzp) {
3681 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3682 zfs_sa_upgrade_txholds(tx, tdzp);
3683 }
3684 if (tzp) {
3685 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3686 zfs_sa_upgrade_txholds(tx, tzp);
3687 }
3688
3689 zfs_sa_upgrade_txholds(tx, szp);
3690 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3691 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3692 if (error) {
3693 if (zl != NULL)
3694 zfs_rename_unlock(&zl);
3695 zfs_dirent_unlock(sdl);
3696 zfs_dirent_unlock(tdl);
3697
3698 if (sdzp == tdzp)
3699 rw_exit(&sdzp->z_name_lock);
3700
3701 VN_RELE(ZTOV(szp));
3702 if (tzp)
3703 VN_RELE(ZTOV(tzp));
3704 if (error == ERESTART) {
3705 waited = B_TRUE;
3706 dmu_tx_wait(tx);
3707 dmu_tx_abort(tx);
3708 goto top;
3709 }
3710 dmu_tx_abort(tx);
3711 ZFS_EXIT(zfsvfs);
3712 return (error);
3713 }
3714
3715 if (tzp) /* Attempt to remove the existing target */
3716 error = rm_err = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3717
3718 if (error == 0) {
3719 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3720 if (error == 0) {
3721 szp->z_pflags |= ZFS_AV_MODIFIED;
3722
3723 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3724 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3725 ASSERT0(error);
3726
3727 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3728 if (error == 0) {
3729 zfs_log_rename(zilog, tx, TX_RENAME |
3730 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3731 sdl->dl_name, tdzp, tdl->dl_name, szp);
3732
3733 /*
3734 * Update path information for the target vnode
3735 */
3736 vn_renamepath(tdvp, ZTOV(szp), tnm,
3737 strlen(tnm));
3738 } else {
3739 /*
3740 * At this point, we have successfully created
3741 * the target name, but have failed to remove
3742 * the source name. Since the create was done
3743 * with the ZRENAMING flag, there are
3744 * complications; for one, the link count is
3745 * wrong. The easiest way to deal with this
3746 * is to remove the newly created target, and
3747 * return the original error. This must
3748 * succeed; fortunately, it is very unlikely to
3749 * fail, since we just created it.
3750 */
3751 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3752 ZRENAMING, NULL), ==, 0);
3753 }
3754 }
3755 }
3756
3757 dmu_tx_commit(tx);
3758
3759 if (tzp && rm_err == 0)
3760 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3761
3762 if (error == 0) {
3763 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3764 /* notify the target dir if it is not the same as source dir */
3765 if (tdvp != sdvp)
3766 vnevent_rename_dest_dir(tdvp, ct);
3767 }
3768 out:
3769 if (zl != NULL)
3770 zfs_rename_unlock(&zl);
3771
3772 zfs_dirent_unlock(sdl);
3773 zfs_dirent_unlock(tdl);
3774
3775 if (sdzp == tdzp)
3776 rw_exit(&sdzp->z_name_lock);
3777
3778
3779 VN_RELE(ZTOV(szp));
3780 if (tzp)
3781 VN_RELE(ZTOV(tzp));
3782
3783 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3784 zil_commit(zilog, 0);
3785
3786 ZFS_EXIT(zfsvfs);
3787 return (error);
3788 }
3789
3790 /*
3791 * Insert the indicated symbolic reference entry into the directory.
3792 *
3793 * IN: dvp - Directory to contain new symbolic link.
3794 * link - Name for new symlink entry.
3795 * vap - Attributes of new entry.
3796 * cr - credentials of caller.
3797 * ct - caller context
3798 * flags - case flags
3799 *
3800 * RETURN: 0 on success, error code on failure.
3801 *
3802 * Timestamps:
3803 * dvp - ctime|mtime updated
3804 */
3805 /*ARGSUSED*/
3806 static int
3807 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3808 caller_context_t *ct, int flags)
3809 {
3810 znode_t *zp, *dzp = VTOZ(dvp);
3811 zfs_dirlock_t *dl;
3812 dmu_tx_t *tx;
3813 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
3814 zilog_t *zilog;
3815 uint64_t len = strlen(link);
3816 int error;
3817 int zflg = ZNEW;
3818 zfs_acl_ids_t acl_ids;
3819 boolean_t fuid_dirtied;
3820 uint64_t txtype = TX_SYMLINK;
3821 boolean_t waited = B_FALSE;
3822
3823 ASSERT(vap->va_type == VLNK);
3824
3825 ZFS_ENTER(zfsvfs);
3826 ZFS_VERIFY_ZP(dzp);
3827 zilog = zfsvfs->z_log;
3828
3829 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3830 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3831 ZFS_EXIT(zfsvfs);
3832 return (SET_ERROR(EILSEQ));
3833 }
3834 if (flags & FIGNORECASE)
3835 zflg |= ZCILOOK;
3836
3837 if (len > MAXPATHLEN) {
3838 ZFS_EXIT(zfsvfs);
3839 return (SET_ERROR(ENAMETOOLONG));
3840 }
3841
3842 if ((error = zfs_acl_ids_create(dzp, 0,
3843 vap, cr, NULL, &acl_ids)) != 0) {
3844 ZFS_EXIT(zfsvfs);
3845 return (error);
3846 }
3847 top:
3848 /*
3849 * Attempt to lock directory; fail if entry already exists.
3850 */
3851 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3852 if (error) {
3853 zfs_acl_ids_free(&acl_ids);
3854 ZFS_EXIT(zfsvfs);
3855 return (error);
3856 }
3857
3858 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3859 zfs_acl_ids_free(&acl_ids);
3860 zfs_dirent_unlock(dl);
3861 ZFS_EXIT(zfsvfs);
3862 return (error);
3863 }
3864
3865 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3866 zfs_acl_ids_free(&acl_ids);
3867 zfs_dirent_unlock(dl);
3868 ZFS_EXIT(zfsvfs);
3869 return (SET_ERROR(EDQUOT));
3870 }
3871 tx = dmu_tx_create(zfsvfs->z_os);
3872 fuid_dirtied = zfsvfs->z_fuid_dirty;
3873 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3874 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3875 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3876 ZFS_SA_BASE_ATTR_SIZE + len);
3877 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3878 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3879 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3880 acl_ids.z_aclp->z_acl_bytes);
3881 }
3882 if (fuid_dirtied)
3883 zfs_fuid_txhold(zfsvfs, tx);
3884 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3885 if (error) {
3886 zfs_dirent_unlock(dl);
3887 if (error == ERESTART) {
3888 waited = B_TRUE;
3889 dmu_tx_wait(tx);
3890 dmu_tx_abort(tx);
3891 goto top;
3892 }
3893 zfs_acl_ids_free(&acl_ids);
3894 dmu_tx_abort(tx);
3895 ZFS_EXIT(zfsvfs);
3896 return (error);
3897 }
3898
3899 /*
3900 * Create a new object for the symlink.
3901 * for version 4 ZPL datsets the symlink will be an SA attribute
3902 */
3903 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3904
3905 if (fuid_dirtied)
3906 zfs_fuid_sync(zfsvfs, tx);
3907
3908 mutex_enter(&zp->z_lock);
3909 if (zp->z_is_sa)
3910 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3911 link, len, tx);
3912 else
3913 zfs_sa_symlink(zp, link, len, tx);
3914 mutex_exit(&zp->z_lock);
3915
3916 zp->z_size = len;
3917 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3918 &zp->z_size, sizeof (zp->z_size), tx);
3919 /*
3920 * Insert the new object into the directory.
3921 */
3922 (void) zfs_link_create(dl, zp, tx, ZNEW);
3923
3924 if (flags & FIGNORECASE)
3925 txtype |= TX_CI;
3926 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3927
3928 zfs_acl_ids_free(&acl_ids);
3929
3930 dmu_tx_commit(tx);
3931
3932 zfs_dirent_unlock(dl);
3933
3934 VN_RELE(ZTOV(zp));
3935
3936 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3937 zil_commit(zilog, 0);
3938
3939 ZFS_EXIT(zfsvfs);
3940 return (error);
3941 }
3942
3943 /*
3944 * Return, in the buffer contained in the provided uio structure,
3945 * the symbolic path referred to by vp.
3946 *
3947 * IN: vp - vnode of symbolic link.
3948 * uio - structure to contain the link path.
3949 * cr - credentials of caller.
3950 * ct - caller context
3951 *
3952 * OUT: uio - structure containing the link path.
3953 *
3954 * RETURN: 0 on success, error code on failure.
3955 *
3956 * Timestamps:
3957 * vp - atime updated
3958 */
3959 /* ARGSUSED */
3960 static int
3961 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3962 {
3963 znode_t *zp = VTOZ(vp);
3964 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3965 int error;
3966
3967 ZFS_ENTER(zfsvfs);
3968 ZFS_VERIFY_ZP(zp);
3969
3970 mutex_enter(&zp->z_lock);
3971 if (zp->z_is_sa)
3972 error = sa_lookup_uio(zp->z_sa_hdl,
3973 SA_ZPL_SYMLINK(zfsvfs), uio);
3974 else
3975 error = zfs_sa_readlink(zp, uio);
3976 mutex_exit(&zp->z_lock);
3977
3978 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3979
3980 ZFS_EXIT(zfsvfs);
3981 return (error);
3982 }
3983
3984 /*
3985 * Insert a new entry into directory tdvp referencing svp.
3986 *
3987 * IN: tdvp - Directory to contain new entry.
3988 * svp - vnode of new entry.
3989 * name - name of new entry.
3990 * cr - credentials of caller.
3991 * ct - caller context
3992 *
3993 * RETURN: 0 on success, error code on failure.
3994 *
3995 * Timestamps:
3996 * tdvp - ctime|mtime updated
3997 * svp - ctime updated
3998 */
3999 /* ARGSUSED */
4000 static int
4001 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4002 caller_context_t *ct, int flags)
4003 {
4004 znode_t *dzp = VTOZ(tdvp);
4005 znode_t *tzp, *szp;
4006 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4007 zilog_t *zilog;
4008 zfs_dirlock_t *dl;
4009 dmu_tx_t *tx;
4010 vnode_t *realvp;
4011 int error;
4012 int zf = ZNEW;
4013 uint64_t parent;
4014 uid_t owner;
4015 boolean_t waited = B_FALSE;
4016
4017 ASSERT(tdvp->v_type == VDIR);
4018
4019 ZFS_ENTER(zfsvfs);
4020 ZFS_VERIFY_ZP(dzp);
4021 zilog = zfsvfs->z_log;
4022
4023 if (VOP_REALVP(svp, &realvp, ct) == 0)
4024 svp = realvp;
4025
4026 /*
4027 * POSIX dictates that we return EPERM here.
4028 * Better choices include ENOTSUP or EISDIR.
4029 */
4030 if (svp->v_type == VDIR) {
4031 ZFS_EXIT(zfsvfs);
4032 return (SET_ERROR(EPERM));
4033 }
4034
4035 szp = VTOZ(svp);
4036 ZFS_VERIFY_ZP(szp);
4037
4038 /*
4039 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4040 * ctldir appear to have the same v_vfsp.
4041 */
4042 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4043 ZFS_EXIT(zfsvfs);
4044 return (SET_ERROR(EXDEV));
4045 }
4046
4047 /* Prevent links to .zfs/shares files */
4048
4049 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4050 &parent, sizeof (uint64_t))) != 0) {
4051 ZFS_EXIT(zfsvfs);
4052 return (error);
4053 }
4054 if (parent == zfsvfs->z_shares_dir) {
4055 ZFS_EXIT(zfsvfs);
4056 return (SET_ERROR(EPERM));
4057 }
4058
4059 if (zfsvfs->z_utf8 && u8_validate(name,
4060 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4061 ZFS_EXIT(zfsvfs);
4062 return (SET_ERROR(EILSEQ));
4063 }
4064 if (flags & FIGNORECASE)
4065 zf |= ZCILOOK;
4066
4067 /*
4068 * We do not support links between attributes and non-attributes
4069 * because of the potential security risk of creating links
4070 * into "normal" file space in order to circumvent restrictions
4071 * imposed in attribute space.
4072 */
4073 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4074 ZFS_EXIT(zfsvfs);
4075 return (SET_ERROR(EINVAL));
4076 }
4077
4078
4079 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4080 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4081 ZFS_EXIT(zfsvfs);
4082 return (SET_ERROR(EPERM));
4083 }
4084
4085 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4086 ZFS_EXIT(zfsvfs);
4087 return (error);
4088 }
4089
4090 top:
4091 /*
4092 * Attempt to lock directory; fail if entry already exists.
4093 */
4094 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4095 if (error) {
4096 ZFS_EXIT(zfsvfs);
4097 return (error);
4098 }
4099
4100 tx = dmu_tx_create(zfsvfs->z_os);
4101 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4102 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4103 zfs_sa_upgrade_txholds(tx, szp);
4104 zfs_sa_upgrade_txholds(tx, dzp);
4105 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4106 if (error) {
4107 zfs_dirent_unlock(dl);
4108 if (error == ERESTART) {
4109 waited = B_TRUE;
4110 dmu_tx_wait(tx);
4111 dmu_tx_abort(tx);
4112 goto top;
4113 }
4114 dmu_tx_abort(tx);
4115 ZFS_EXIT(zfsvfs);
4116 return (error);
4117 }
4118
4119 error = zfs_link_create(dl, szp, tx, 0);
4120
4121 if (error == 0) {
4122 uint64_t txtype = TX_LINK;
4123 if (flags & FIGNORECASE)
4124 txtype |= TX_CI;
4125 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4126 }
4127
4128 dmu_tx_commit(tx);
4129
4130 zfs_dirent_unlock(dl);
4131
4132 if (error == 0) {
4133 vnevent_link(svp, ct);
4134 }
4135
4136 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4137 zil_commit(zilog, 0);
4138
4139 ZFS_EXIT(zfsvfs);
4140 return (error);
4141 }
4142
4143 /*
4144 * zfs_null_putapage() is used when the file system has been force
4145 * unmounted. It just drops the pages.
4146 */
4147 /* ARGSUSED */
4148 static int
4149 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4150 size_t *lenp, int flags, cred_t *cr)
4151 {
4152 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4153 return (0);
4154 }
4155
4156 /*
4157 * Push a page out to disk, klustering if possible.
4158 *
4159 * IN: vp - file to push page to.
4160 * pp - page to push.
4161 * flags - additional flags.
4162 * cr - credentials of caller.
4163 *
4164 * OUT: offp - start of range pushed.
4165 * lenp - len of range pushed.
4166 *
4167 * RETURN: 0 on success, error code on failure.
4168 *
4169 * NOTE: callers must have locked the page to be pushed. On
4170 * exit, the page (and all other pages in the kluster) must be
4171 * unlocked.
4172 */
4173 /* ARGSUSED */
4174 static int
4175 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4176 size_t *lenp, int flags, cred_t *cr)
4177 {
4178 znode_t *zp = VTOZ(vp);
4179 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4180 dmu_tx_t *tx;
4181 u_offset_t off, koff;
4182 size_t len, klen;
4183 int err;
4184
4185 off = pp->p_offset;
4186 len = PAGESIZE;
4187 /*
4188 * If our blocksize is bigger than the page size, try to kluster
4189 * multiple pages so that we write a full block (thus avoiding
4190 * a read-modify-write).
4191 */
4192 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4193 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4194 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4195 ASSERT(koff <= zp->z_size);
4196 if (koff + klen > zp->z_size)
4197 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4198 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4199 }
4200 ASSERT3U(btop(len), ==, btopr(len));
4201
4202 /*
4203 * Can't push pages past end-of-file.
4204 */
4205 if (off >= zp->z_size) {
4206 /* ignore all pages */
4207 err = 0;
4208 goto out;
4209 } else if (off + len > zp->z_size) {
4210 int npages = btopr(zp->z_size - off);
4211 page_t *trunc;
4212
4213 page_list_break(&pp, &trunc, npages);
4214 /* ignore pages past end of file */
4215 if (trunc)
4216 pvn_write_done(trunc, flags);
4217 len = zp->z_size - off;
4218 }
4219
4220 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4221 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4222 err = SET_ERROR(EDQUOT);
4223 goto out;
4224 }
4225 tx = dmu_tx_create(zfsvfs->z_os);
4226 dmu_tx_hold_write(tx, zp->z_id, off, len);
4227
4228 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4229 zfs_sa_upgrade_txholds(tx, zp);
4230 err = dmu_tx_assign(tx, TXG_WAIT);
4231 if (err != 0) {
4232 dmu_tx_abort(tx);
4233 goto out;
4234 }
4235
4236 if (zp->z_blksz <= PAGESIZE) {
4237 caddr_t va = zfs_map_page(pp, S_READ);
4238 ASSERT3U(len, <=, PAGESIZE);
4239 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4240 zfs_unmap_page(pp, va);
4241 } else {
4242 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4243 }
4244
4245 if (err == 0) {
4246 uint64_t mtime[2], ctime[2];
4247 sa_bulk_attr_t bulk[3];
4248 int count = 0;
4249
4250 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4251 &mtime, 16);
4252 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4253 &ctime, 16);
4254 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4255 &zp->z_pflags, 8);
4256 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4257 B_TRUE);
4258 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4259 }
4260 dmu_tx_commit(tx);
4261
4262 out:
4263 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4264 if (offp)
4265 *offp = off;
4266 if (lenp)
4267 *lenp = len;
4268
4269 return (err);
4270 }
4271
4272 /*
4273 * Copy the portion of the file indicated from pages into the file.
4274 * The pages are stored in a page list attached to the files vnode.
4275 *
4276 * IN: vp - vnode of file to push page data to.
4277 * off - position in file to put data.
4278 * len - amount of data to write.
4279 * flags - flags to control the operation.
4280 * cr - credentials of caller.
4281 * ct - caller context.
4282 *
4283 * RETURN: 0 on success, error code on failure.
4284 *
4285 * Timestamps:
4286 * vp - ctime|mtime updated
4287 */
4288 /*ARGSUSED*/
4289 static int
4290 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4291 caller_context_t *ct)
4292 {
4293 znode_t *zp = VTOZ(vp);
4294 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4295 page_t *pp;
4296 size_t io_len;
4297 u_offset_t io_off;
4298 uint_t blksz;
4299 rl_t *rl;
4300 int error = 0;
4301
4302 ZFS_ENTER(zfsvfs);
4303 ZFS_VERIFY_ZP(zp);
4304
4305 /*
4306 * There's nothing to do if no data is cached.
4307 */
4308 if (!vn_has_cached_data(vp)) {
4309 ZFS_EXIT(zfsvfs);
4310 return (0);
4311 }
4312
4313 /*
4314 * Align this request to the file block size in case we kluster.
4315 * XXX - this can result in pretty aggresive locking, which can
4316 * impact simultanious read/write access. One option might be
4317 * to break up long requests (len == 0) into block-by-block
4318 * operations to get narrower locking.
4319 */
4320 blksz = zp->z_blksz;
4321 if (ISP2(blksz))
4322 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4323 else
4324 io_off = 0;
4325 if (len > 0 && ISP2(blksz))
4326 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4327 else
4328 io_len = 0;
4329
4330 if (io_len == 0) {
4331 /*
4332 * Search the entire vp list for pages >= io_off.
4333 */
4334 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4335 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4336 goto out;
4337 }
4338 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4339
4340 if (off > zp->z_size) {
4341 /* past end of file */
4342 zfs_range_unlock(rl);
4343 ZFS_EXIT(zfsvfs);
4344 return (0);
4345 }
4346
4347 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4348
4349 for (off = io_off; io_off < off + len; io_off += io_len) {
4350 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4351 pp = page_lookup(vp, io_off,
4352 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4353 } else {
4354 pp = page_lookup_nowait(vp, io_off,
4355 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4356 }
4357
4358 if (pp != NULL && pvn_getdirty(pp, flags)) {
4359 int err;
4360
4361 /*
4362 * Found a dirty page to push
4363 */
4364 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4365 if (err)
4366 error = err;
4367 } else {
4368 io_len = PAGESIZE;
4369 }
4370 }
4371 out:
4372 zfs_range_unlock(rl);
4373 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4374 zil_commit(zfsvfs->z_log, zp->z_id);
4375 ZFS_EXIT(zfsvfs);
4376 return (error);
4377 }
4378
4379 /*ARGSUSED*/
4380 void
4381 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4382 {
4383 znode_t *zp = VTOZ(vp);
4384 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4385 int error;
4386
4387 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4388 if (zp->z_sa_hdl == NULL) {
4389 /*
4390 * The fs has been unmounted, or we did a
4391 * suspend/resume and this file no longer exists.
4392 */
4393 if (vn_has_cached_data(vp)) {
4394 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4395 B_INVAL, cr);
4396 }
4397
4398 mutex_enter(&zp->z_lock);
4399 mutex_enter(&vp->v_lock);
4400 ASSERT(vp->v_count == 1);
4401 vp->v_count = 0;
4402 mutex_exit(&vp->v_lock);
4403 mutex_exit(&zp->z_lock);
4404 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4405 zfs_znode_free(zp);
4406 return;
4407 }
4408
4409 /*
4410 * Attempt to push any data in the page cache. If this fails
4411 * we will get kicked out later in zfs_zinactive().
4412 */
4413 if (vn_has_cached_data(vp)) {
4414 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4415 cr);
4416 }
4417
4418 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4419 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4420
4421 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4422 zfs_sa_upgrade_txholds(tx, zp);
4423 error = dmu_tx_assign(tx, TXG_WAIT);
4424 if (error) {
4425 dmu_tx_abort(tx);
4426 } else {
4427 mutex_enter(&zp->z_lock);
4428 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4429 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4430 zp->z_atime_dirty = 0;
4431 mutex_exit(&zp->z_lock);
4432 dmu_tx_commit(tx);
4433 }
4434 }
4435
4436 zfs_zinactive(zp);
4437 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4438 }
4439
4440 /*
4441 * Bounds-check the seek operation.
4442 *
4443 * IN: vp - vnode seeking within
4444 * ooff - old file offset
4445 * noffp - pointer to new file offset
4446 * ct - caller context
4447 *
4448 * RETURN: 0 on success, EINVAL if new offset invalid.
4449 */
4450 /* ARGSUSED */
4451 static int
4452 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4453 caller_context_t *ct)
4454 {
4455 if (vp->v_type == VDIR)
4456 return (0);
4457 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4458 }
4459
4460 /*
4461 * Pre-filter the generic locking function to trap attempts to place
4462 * a mandatory lock on a memory mapped file.
4463 */
4464 static int
4465 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4466 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4467 {
4468 znode_t *zp = VTOZ(vp);
4469 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4470
4471 ZFS_ENTER(zfsvfs);
4472 ZFS_VERIFY_ZP(zp);
4473
4474 /*
4475 * We are following the UFS semantics with respect to mapcnt
4476 * here: If we see that the file is mapped already, then we will
4477 * return an error, but we don't worry about races between this
4478 * function and zfs_map().
4479 */
4480 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4481 ZFS_EXIT(zfsvfs);
4482 return (SET_ERROR(EAGAIN));
4483 }
4484 ZFS_EXIT(zfsvfs);
4485 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4486 }
4487
4488 /*
4489 * If we can't find a page in the cache, we will create a new page
4490 * and fill it with file data. For efficiency, we may try to fill
4491 * multiple pages at once (klustering) to fill up the supplied page
4492 * list. Note that the pages to be filled are held with an exclusive
4493 * lock to prevent access by other threads while they are being filled.
4494 */
4495 static int
4496 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4497 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4498 {
4499 znode_t *zp = VTOZ(vp);
4500 page_t *pp, *cur_pp;
4501 objset_t *os = zp->z_zfsvfs->z_os;
4502 u_offset_t io_off, total;
4503 size_t io_len;
4504 int err;
4505
4506 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4507 /*
4508 * We only have a single page, don't bother klustering
4509 */
4510 io_off = off;
4511 io_len = PAGESIZE;
4512 pp = page_create_va(vp, io_off, io_len,
4513 PG_EXCL | PG_WAIT, seg, addr);
4514 } else {
4515 /*
4516 * Try to find enough pages to fill the page list
4517 */
4518 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4519 &io_len, off, plsz, 0);
4520 }
4521 if (pp == NULL) {
4522 /*
4523 * The page already exists, nothing to do here.
4524 */
4525 *pl = NULL;
4526 return (0);
4527 }
4528
4529 /*
4530 * Fill the pages in the kluster.
4531 */
4532 cur_pp = pp;
4533 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4534 caddr_t va;
4535
4536 ASSERT3U(io_off, ==, cur_pp->p_offset);
4537 va = zfs_map_page(cur_pp, S_WRITE);
4538 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4539 DMU_READ_PREFETCH);
4540 zfs_unmap_page(cur_pp, va);
4541 if (err) {
4542 /* On error, toss the entire kluster */
4543 pvn_read_done(pp, B_ERROR);
4544 /* convert checksum errors into IO errors */
4545 if (err == ECKSUM)
4546 err = SET_ERROR(EIO);
4547 return (err);
4548 }
4549 cur_pp = cur_pp->p_next;
4550 }
4551
4552 /*
4553 * Fill in the page list array from the kluster starting
4554 * from the desired offset `off'.
4555 * NOTE: the page list will always be null terminated.
4556 */
4557 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4558 ASSERT(pl == NULL || (*pl)->p_offset == off);
4559
4560 return (0);
4561 }
4562
4563 /*
4564 * Return pointers to the pages for the file region [off, off + len]
4565 * in the pl array. If plsz is greater than len, this function may
4566 * also return page pointers from after the specified region
4567 * (i.e. the region [off, off + plsz]). These additional pages are
4568 * only returned if they are already in the cache, or were created as
4569 * part of a klustered read.
4570 *
4571 * IN: vp - vnode of file to get data from.
4572 * off - position in file to get data from.
4573 * len - amount of data to retrieve.
4574 * plsz - length of provided page list.
4575 * seg - segment to obtain pages for.
4576 * addr - virtual address of fault.
4577 * rw - mode of created pages.
4578 * cr - credentials of caller.
4579 * ct - caller context.
4580 *
4581 * OUT: protp - protection mode of created pages.
4582 * pl - list of pages created.
4583 *
4584 * RETURN: 0 on success, error code on failure.
4585 *
4586 * Timestamps:
4587 * vp - atime updated
4588 */
4589 /* ARGSUSED */
4590 static int
4591 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4592 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4593 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4594 {
4595 znode_t *zp = VTOZ(vp);
4596 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4597 page_t **pl0 = pl;
4598 int err = 0;
4599
4600 /* we do our own caching, faultahead is unnecessary */
4601 if (pl == NULL)
4602 return (0);
4603 else if (len > plsz)
4604 len = plsz;
4605 else
4606 len = P2ROUNDUP(len, PAGESIZE);
4607 ASSERT(plsz >= len);
4608
4609 ZFS_ENTER(zfsvfs);
4610 ZFS_VERIFY_ZP(zp);
4611
4612 if (protp)
4613 *protp = PROT_ALL;
4614
4615 /*
4616 * Loop through the requested range [off, off + len) looking
4617 * for pages. If we don't find a page, we will need to create
4618 * a new page and fill it with data from the file.
4619 */
4620 while (len > 0) {
4621 if (*pl = page_lookup(vp, off, SE_SHARED))
4622 *(pl+1) = NULL;
4623 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4624 goto out;
4625 while (*pl) {
4626 ASSERT3U((*pl)->p_offset, ==, off);
4627 off += PAGESIZE;
4628 addr += PAGESIZE;
4629 if (len > 0) {
4630 ASSERT3U(len, >=, PAGESIZE);
4631 len -= PAGESIZE;
4632 }
4633 ASSERT3U(plsz, >=, PAGESIZE);
4634 plsz -= PAGESIZE;
4635 pl++;
4636 }
4637 }
4638
4639 /*
4640 * Fill out the page array with any pages already in the cache.
4641 */
4642 while (plsz > 0 &&
4643 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4644 off += PAGESIZE;
4645 plsz -= PAGESIZE;
4646 }
4647 out:
4648 if (err) {
4649 /*
4650 * Release any pages we have previously locked.
4651 */
4652 while (pl > pl0)
4653 page_unlock(*--pl);
4654 } else {
4655 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4656 }
4657
4658 *pl = NULL;
4659
4660 ZFS_EXIT(zfsvfs);
4661 return (err);
4662 }
4663
4664 /*
4665 * Request a memory map for a section of a file. This code interacts
4666 * with common code and the VM system as follows:
4667 *
4668 * - common code calls mmap(), which ends up in smmap_common()
4669 * - this calls VOP_MAP(), which takes you into (say) zfs
4670 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4671 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4672 * - zfs_addmap() updates z_mapcnt
4673 */
4674 /*ARGSUSED*/
4675 static int
4676 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4677 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4678 caller_context_t *ct)
4679 {
4680 znode_t *zp = VTOZ(vp);
4681 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4682 segvn_crargs_t vn_a;
4683 int error;
4684
4685 ZFS_ENTER(zfsvfs);
4686 ZFS_VERIFY_ZP(zp);
4687
4688 if ((prot & PROT_WRITE) && (zp->z_pflags &
4689 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4690 ZFS_EXIT(zfsvfs);
4691 return (SET_ERROR(EPERM));
4692 }
4693
4694 if ((prot & (PROT_READ | PROT_EXEC)) &&
4695 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4696 ZFS_EXIT(zfsvfs);
4697 return (SET_ERROR(EACCES));
4698 }
4699
4700 if (vp->v_flag & VNOMAP) {
4701 ZFS_EXIT(zfsvfs);
4702 return (SET_ERROR(ENOSYS));
4703 }
4704
4705 if (off < 0 || len > MAXOFFSET_T - off) {
4706 ZFS_EXIT(zfsvfs);
4707 return (SET_ERROR(ENXIO));
4708 }
4709
4710 if (vp->v_type != VREG) {
4711 ZFS_EXIT(zfsvfs);
4712 return (SET_ERROR(ENODEV));
4713 }
4714
4715 /*
4716 * If file is locked, disallow mapping.
4717 */
4718 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4719 ZFS_EXIT(zfsvfs);
4720 return (SET_ERROR(EAGAIN));
4721 }
4722
4723 as_rangelock(as);
4724 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4725 if (error != 0) {
4726 as_rangeunlock(as);
4727 ZFS_EXIT(zfsvfs);
4728 return (error);
4729 }
4730
4731 vn_a.vp = vp;
4732 vn_a.offset = (u_offset_t)off;
4733 vn_a.type = flags & MAP_TYPE;
4734 vn_a.prot = prot;
4735 vn_a.maxprot = maxprot;
4736 vn_a.cred = cr;
4737 vn_a.amp = NULL;
4738 vn_a.flags = flags & ~MAP_TYPE;
4739 vn_a.szc = 0;
4740 vn_a.lgrp_mem_policy_flags = 0;
4741
4742 error = as_map(as, *addrp, len, segvn_create, &vn_a);
4743
4744 as_rangeunlock(as);
4745 ZFS_EXIT(zfsvfs);
4746 return (error);
4747 }
4748
4749 /* ARGSUSED */
4750 static int
4751 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4752 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4753 caller_context_t *ct)
4754 {
4755 uint64_t pages = btopr(len);
4756
4757 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4758 return (0);
4759 }
4760
4761 /*
4762 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4763 * more accurate mtime for the associated file. Since we don't have a way of
4764 * detecting when the data was actually modified, we have to resort to
4765 * heuristics. If an explicit msync() is done, then we mark the mtime when the
4766 * last page is pushed. The problem occurs when the msync() call is omitted,
4767 * which by far the most common case:
4768 *
4769 * open()
4770 * mmap()
4771 * <modify memory>
4772 * munmap()
4773 * close()
4774 * <time lapse>
4775 * putpage() via fsflush
4776 *
4777 * If we wait until fsflush to come along, we can have a modification time that
4778 * is some arbitrary point in the future. In order to prevent this in the
4779 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4780 * torn down.
4781 */
4782 /* ARGSUSED */
4783 static int
4784 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4785 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4786 caller_context_t *ct)
4787 {
4788 uint64_t pages = btopr(len);
4789
4790 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4791 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4792
4793 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4794 vn_has_cached_data(vp))
4795 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4796
4797 return (0);
4798 }
4799
4800 /*
4801 * Free or allocate space in a file. Currently, this function only
4802 * supports the `F_FREESP' command. However, this command is somewhat
4803 * misnamed, as its functionality includes the ability to allocate as
4804 * well as free space.
4805 *
4806 * IN: vp - vnode of file to free data in.
4807 * cmd - action to take (only F_FREESP supported).
4808 * bfp - section of file to free/alloc.
4809 * flag - current file open mode flags.
4810 * offset - current file offset.
4811 * cr - credentials of caller [UNUSED].
4812 * ct - caller context.
4813 *
4814 * RETURN: 0 on success, error code on failure.
4815 *
4816 * Timestamps:
4817 * vp - ctime|mtime updated
4818 */
4819 /* ARGSUSED */
4820 static int
4821 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4822 offset_t offset, cred_t *cr, caller_context_t *ct)
4823 {
4824 znode_t *zp = VTOZ(vp);
4825 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4826 uint64_t off, len;
4827 int error;
4828
4829 ZFS_ENTER(zfsvfs);
4830 ZFS_VERIFY_ZP(zp);
4831
4832 if (cmd != F_FREESP) {
4833 ZFS_EXIT(zfsvfs);
4834 return (SET_ERROR(EINVAL));
4835 }
4836
4837 /*
4838 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
4839 * callers might not be able to detect properly that we are read-only,
4840 * so check it explicitly here.
4841 */
4842 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
4843 ZFS_EXIT(zfsvfs);
4844 return (SET_ERROR(EROFS));
4845 }
4846
4847 if (error = convoff(vp, bfp, 0, offset)) {
4848 ZFS_EXIT(zfsvfs);
4849 return (error);
4850 }
4851
4852 if (bfp->l_len < 0) {
4853 ZFS_EXIT(zfsvfs);
4854 return (SET_ERROR(EINVAL));
4855 }
4856
4857 off = bfp->l_start;
4858 len = bfp->l_len; /* 0 means from off to end of file */
4859
4860 error = zfs_freesp(zp, off, len, flag, TRUE);
4861
4862 if (error == 0 && off == 0 && len == 0)
4863 vnevent_truncate(ZTOV(zp), ct);
4864
4865 ZFS_EXIT(zfsvfs);
4866 return (error);
4867 }
4868
4869 /*ARGSUSED*/
4870 static int
4871 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4872 {
4873 znode_t *zp = VTOZ(vp);
4874 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4875 uint32_t gen;
4876 uint64_t gen64;
4877 uint64_t object = zp->z_id;
4878 zfid_short_t *zfid;
4879 int size, i, error;
4880
4881 ZFS_ENTER(zfsvfs);
4882 ZFS_VERIFY_ZP(zp);
4883
4884 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4885 &gen64, sizeof (uint64_t))) != 0) {
4886 ZFS_EXIT(zfsvfs);
4887 return (error);
4888 }
4889
4890 gen = (uint32_t)gen64;
4891
4892 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4893 if (fidp->fid_len < size) {
4894 fidp->fid_len = size;
4895 ZFS_EXIT(zfsvfs);
4896 return (SET_ERROR(ENOSPC));
4897 }
4898
4899 zfid = (zfid_short_t *)fidp;
4900
4901 zfid->zf_len = size;
4902
4903 for (i = 0; i < sizeof (zfid->zf_object); i++)
4904 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4905
4906 /* Must have a non-zero generation number to distinguish from .zfs */
4907 if (gen == 0)
4908 gen = 1;
4909 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4910 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4911
4912 if (size == LONG_FID_LEN) {
4913 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
4914 zfid_long_t *zlfid;
4915
4916 zlfid = (zfid_long_t *)fidp;
4917
4918 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4919 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4920
4921 /* XXX - this should be the generation number for the objset */
4922 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4923 zlfid->zf_setgen[i] = 0;
4924 }
4925
4926 ZFS_EXIT(zfsvfs);
4927 return (0);
4928 }
4929
4930 static int
4931 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4932 caller_context_t *ct)
4933 {
4934 znode_t *zp, *xzp;
4935 zfsvfs_t *zfsvfs;
4936 zfs_dirlock_t *dl;
4937 int error;
4938
4939 switch (cmd) {
4940 case _PC_LINK_MAX:
4941 *valp = ULONG_MAX;
4942 return (0);
4943
4944 case _PC_FILESIZEBITS:
4945 *valp = 64;
4946 return (0);
4947
4948 case _PC_XATTR_EXISTS:
4949 zp = VTOZ(vp);
4950 zfsvfs = zp->z_zfsvfs;
4951 ZFS_ENTER(zfsvfs);
4952 ZFS_VERIFY_ZP(zp);
4953 *valp = 0;
4954 error = zfs_dirent_lock(&dl, zp, "", &xzp,
4955 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4956 if (error == 0) {
4957 zfs_dirent_unlock(dl);
4958 if (!zfs_dirempty(xzp))
4959 *valp = 1;
4960 VN_RELE(ZTOV(xzp));
4961 } else if (error == ENOENT) {
4962 /*
4963 * If there aren't extended attributes, it's the
4964 * same as having zero of them.
4965 */
4966 error = 0;
4967 }
4968 ZFS_EXIT(zfsvfs);
4969 return (error);
4970
4971 case _PC_SATTR_ENABLED:
4972 case _PC_SATTR_EXISTS:
4973 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4974 (vp->v_type == VREG || vp->v_type == VDIR);
4975 return (0);
4976
4977 case _PC_ACCESS_FILTERING:
4978 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4979 vp->v_type == VDIR;
4980 return (0);
4981
4982 case _PC_ACL_ENABLED:
4983 *valp = _ACL_ACE_ENABLED;
4984 return (0);
4985
4986 case _PC_MIN_HOLE_SIZE:
4987 *valp = (ulong_t)SPA_MINBLOCKSIZE;
4988 return (0);
4989
4990 case _PC_TIMESTAMP_RESOLUTION:
4991 /* nanosecond timestamp resolution */
4992 *valp = 1L;
4993 return (0);
4994
4995 default:
4996 return (fs_pathconf(vp, cmd, valp, cr, ct));
4997 }
4998 }
4999
5000 /*ARGSUSED*/
5001 static int
5002 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5003 caller_context_t *ct)
5004 {
5005 znode_t *zp = VTOZ(vp);
5006 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5007 int error;
5008 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5009
5010 ZFS_ENTER(zfsvfs);
5011 ZFS_VERIFY_ZP(zp);
5012 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5013 ZFS_EXIT(zfsvfs);
5014
5015 return (error);
5016 }
5017
5018 /*ARGSUSED*/
5019 static int
5020 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5021 caller_context_t *ct)
5022 {
5023 znode_t *zp = VTOZ(vp);
5024 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5025 int error;
5026 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5027 zilog_t *zilog = zfsvfs->z_log;
5028
5029 ZFS_ENTER(zfsvfs);
5030 ZFS_VERIFY_ZP(zp);
5031
5032 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5033
5034 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5035 zil_commit(zilog, 0);
5036
5037 ZFS_EXIT(zfsvfs);
5038 return (error);
5039 }
5040
5041 /*
5042 * The smallest read we may consider to loan out an arcbuf.
5043 * This must be a power of 2.
5044 */
5045 int zcr_blksz_min = (1 << 10); /* 1K */
5046 /*
5047 * If set to less than the file block size, allow loaning out of an
5048 * arcbuf for a partial block read. This must be a power of 2.
5049 */
5050 int zcr_blksz_max = (1 << 17); /* 128K */
5051
5052 /*ARGSUSED*/
5053 static int
5054 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5055 caller_context_t *ct)
5056 {
5057 znode_t *zp = VTOZ(vp);
5058 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5059 int max_blksz = zfsvfs->z_max_blksz;
5060 uio_t *uio = &xuio->xu_uio;
5061 ssize_t size = uio->uio_resid;
5062 offset_t offset = uio->uio_loffset;
5063 int blksz;
5064 int fullblk, i;
5065 arc_buf_t *abuf;
5066 ssize_t maxsize;
5067 int preamble, postamble;
5068
5069 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5070 return (SET_ERROR(EINVAL));
5071
5072 ZFS_ENTER(zfsvfs);
5073 ZFS_VERIFY_ZP(zp);
5074 switch (ioflag) {
5075 case UIO_WRITE:
5076 /*
5077 * Loan out an arc_buf for write if write size is bigger than
5078 * max_blksz, and the file's block size is also max_blksz.
5079 */
5080 blksz = max_blksz;
5081 if (size < blksz || zp->z_blksz != blksz) {
5082 ZFS_EXIT(zfsvfs);
5083 return (SET_ERROR(EINVAL));
5084 }
5085 /*
5086 * Caller requests buffers for write before knowing where the
5087 * write offset might be (e.g. NFS TCP write).
5088 */
5089 if (offset == -1) {
5090 preamble = 0;
5091 } else {
5092 preamble = P2PHASE(offset, blksz);
5093 if (preamble) {
5094 preamble = blksz - preamble;
5095 size -= preamble;
5096 }
5097 }
5098
5099 postamble = P2PHASE(size, blksz);
5100 size -= postamble;
5101
5102 fullblk = size / blksz;
5103 (void) dmu_xuio_init(xuio,
5104 (preamble != 0) + fullblk + (postamble != 0));
5105 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5106 int, postamble, int,
5107 (preamble != 0) + fullblk + (postamble != 0));
5108
5109 /*
5110 * Have to fix iov base/len for partial buffers. They
5111 * currently represent full arc_buf's.
5112 */
5113 if (preamble) {
5114 /* data begins in the middle of the arc_buf */
5115 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5116 blksz);
5117 ASSERT(abuf);
5118 (void) dmu_xuio_add(xuio, abuf,
5119 blksz - preamble, preamble);
5120 }
5121
5122 for (i = 0; i < fullblk; i++) {
5123 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5124 blksz);
5125 ASSERT(abuf);
5126 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5127 }
5128
5129 if (postamble) {
5130 /* data ends in the middle of the arc_buf */
5131 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5132 blksz);
5133 ASSERT(abuf);
5134 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5135 }
5136 break;
5137 case UIO_READ:
5138 /*
5139 * Loan out an arc_buf for read if the read size is larger than
5140 * the current file block size. Block alignment is not
5141 * considered. Partial arc_buf will be loaned out for read.
5142 */
5143 blksz = zp->z_blksz;
5144 if (blksz < zcr_blksz_min)
5145 blksz = zcr_blksz_min;
5146 if (blksz > zcr_blksz_max)
5147 blksz = zcr_blksz_max;
5148 /* avoid potential complexity of dealing with it */
5149 if (blksz > max_blksz) {
5150 ZFS_EXIT(zfsvfs);
5151 return (SET_ERROR(EINVAL));
5152 }
5153
5154 maxsize = zp->z_size - uio->uio_loffset;
5155 if (size > maxsize)
5156 size = maxsize;
5157
5158 if (size < blksz || vn_has_cached_data(vp)) {
5159 ZFS_EXIT(zfsvfs);
5160 return (SET_ERROR(EINVAL));
5161 }
5162 break;
5163 default:
5164 ZFS_EXIT(zfsvfs);
5165 return (SET_ERROR(EINVAL));
5166 }
5167
5168 uio->uio_extflg = UIO_XUIO;
5169 XUIO_XUZC_RW(xuio) = ioflag;
5170 ZFS_EXIT(zfsvfs);
5171 return (0);
5172 }
5173
5174 /*ARGSUSED*/
5175 static int
5176 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5177 {
5178 int i;
5179 arc_buf_t *abuf;
5180 int ioflag = XUIO_XUZC_RW(xuio);
5181
5182 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5183
5184 i = dmu_xuio_cnt(xuio);
5185 while (i-- > 0) {
5186 abuf = dmu_xuio_arcbuf(xuio, i);
5187 /*
5188 * if abuf == NULL, it must be a write buffer
5189 * that has been returned in zfs_write().
5190 */
5191 if (abuf)
5192 dmu_return_arcbuf(abuf);
5193 ASSERT(abuf || ioflag == UIO_WRITE);
5194 }
5195
5196 dmu_xuio_fini(xuio);
5197 return (0);
5198 }
5199
5200 /*
5201 * Predeclare these here so that the compiler assumes that
5202 * this is an "old style" function declaration that does
5203 * not include arguments => we won't get type mismatch errors
5204 * in the initializations that follow.
5205 */
5206 static int zfs_inval();
5207 static int zfs_isdir();
5208
5209 static int
5210 zfs_inval()
5211 {
5212 return (SET_ERROR(EINVAL));
5213 }
5214
5215 static int
5216 zfs_isdir()
5217 {
5218 return (SET_ERROR(EISDIR));
5219 }
5220 /*
5221 * Directory vnode operations template
5222 */
5223 vnodeops_t *zfs_dvnodeops;
5224 const fs_operation_def_t zfs_dvnodeops_template[] = {
5225 { VOPNAME_OPEN, { .vop_open = zfs_open } },
5226 { VOPNAME_CLOSE, { .vop_close = zfs_close } },
5227 { VOPNAME_READ, { .error = zfs_isdir } },
5228 { VOPNAME_WRITE, { .error = zfs_isdir } },
5229 { VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl } },
5230 { VOPNAME_GETATTR, { .vop_getattr = zfs_getattr } },
5231 { VOPNAME_SETATTR, { .vop_setattr = zfs_setattr } },
5232 { VOPNAME_ACCESS, { .vop_access = zfs_access } },
5233 { VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup } },
5234 { VOPNAME_CREATE, { .vop_create = zfs_create } },
5235 { VOPNAME_REMOVE, { .vop_remove = zfs_remove } },
5236 { VOPNAME_LINK, { .vop_link = zfs_link } },
5237 { VOPNAME_RENAME, { .vop_rename = zfs_rename } },
5238 { VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir } },
5239 { VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir } },
5240 { VOPNAME_READDIR, { .vop_readdir = zfs_readdir } },
5241 { VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink } },
5242 { VOPNAME_FSYNC, { .vop_fsync = zfs_fsync } },
5243 { VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive } },
5244 { VOPNAME_FID, { .vop_fid = zfs_fid } },
5245 { VOPNAME_SEEK, { .vop_seek = zfs_seek } },
5246 { VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf } },
5247 { VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr } },
5248 { VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr } },
5249 { VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support } },
5250 { NULL, { NULL } }
5251 };
5252
5253 /*
5254 * Regular file vnode operations template
5255 */
5256 vnodeops_t *zfs_fvnodeops;
5257 const fs_operation_def_t zfs_fvnodeops_template[] = {
5258 { VOPNAME_OPEN, { .vop_open = zfs_open } },
5259 { VOPNAME_CLOSE, { .vop_close = zfs_close } },
5260 { VOPNAME_READ, { .vop_read = zfs_read } },
5261 { VOPNAME_WRITE, { .vop_write = zfs_write } },
5262 { VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl } },
5263 { VOPNAME_GETATTR, { .vop_getattr = zfs_getattr } },
5264 { VOPNAME_SETATTR, { .vop_setattr = zfs_setattr } },
5265 { VOPNAME_ACCESS, { .vop_access = zfs_access } },
5266 { VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup } },
5267 { VOPNAME_RENAME, { .vop_rename = zfs_rename } },
5268 { VOPNAME_FSYNC, { .vop_fsync = zfs_fsync } },
5269 { VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive } },
5270 { VOPNAME_FID, { .vop_fid = zfs_fid } },
5271 { VOPNAME_SEEK, { .vop_seek = zfs_seek } },
5272 { VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock } },
5273 { VOPNAME_SPACE, { .vop_space = zfs_space } },
5274 { VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage } },
5275 { VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage } },
5276 { VOPNAME_MAP, { .vop_map = zfs_map } },
5277 { VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap } },
5278 { VOPNAME_DELMAP, { .vop_delmap = zfs_delmap } },
5279 { VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf } },
5280 { VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr } },
5281 { VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr } },
5282 { VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support } },
5283 { VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf } },
5284 { VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf } },
5285 { NULL, { NULL } }
5286 };
5287
5288 /*
5289 * Symbolic link vnode operations template
5290 */
5291 vnodeops_t *zfs_symvnodeops;
5292 const fs_operation_def_t zfs_symvnodeops_template[] = {
5293 { VOPNAME_GETATTR, { .vop_getattr = zfs_getattr } },
5294 { VOPNAME_SETATTR, { .vop_setattr = zfs_setattr } },
5295 { VOPNAME_ACCESS, { .vop_access = zfs_access } },
5296 { VOPNAME_RENAME, { .vop_rename = zfs_rename } },
5297 { VOPNAME_READLINK, { .vop_readlink = zfs_readlink } },
5298 { VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive } },
5299 { VOPNAME_FID, { .vop_fid = zfs_fid } },
5300 { VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf } },
5301 { VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support } },
5302 { NULL, { NULL } }
5303 };
5304
5305 /*
5306 * special share hidden files vnode operations template
5307 */
5308 vnodeops_t *zfs_sharevnodeops;
5309 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5310 { VOPNAME_GETATTR, { .vop_getattr = zfs_getattr } },
5311 { VOPNAME_ACCESS, { .vop_access = zfs_access } },
5312 { VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive } },
5313 { VOPNAME_FID, { .vop_fid = zfs_fid } },
5314 { VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf } },
5315 { VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr } },
5316 { VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr } },
5317 { VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support } },
5318 { NULL, { NULL } }
5319 };
5320
5321 /*
5322 * Extended attribute directory vnode operations template
5323 *
5324 * This template is identical to the directory vnodes
5325 * operation template except for restricted operations:
5326 * VOP_MKDIR()
5327 * VOP_SYMLINK()
5328 *
5329 * Note that there are other restrictions embedded in:
5330 * zfs_create() - restrict type to VREG
5331 * zfs_link() - no links into/out of attribute space
5332 * zfs_rename() - no moves into/out of attribute space
5333 */
5334 vnodeops_t *zfs_xdvnodeops;
5335 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5336 { VOPNAME_OPEN, { .vop_open = zfs_open } },
5337 { VOPNAME_CLOSE, { .vop_close = zfs_close } },
5338 { VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl } },
5339 { VOPNAME_GETATTR, { .vop_getattr = zfs_getattr } },
5340 { VOPNAME_SETATTR, { .vop_setattr = zfs_setattr } },
5341 { VOPNAME_ACCESS, { .vop_access = zfs_access } },
5342 { VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup } },
5343 { VOPNAME_CREATE, { .vop_create = zfs_create } },
5344 { VOPNAME_REMOVE, { .vop_remove = zfs_remove } },
5345 { VOPNAME_LINK, { .vop_link = zfs_link } },
5346 { VOPNAME_RENAME, { .vop_rename = zfs_rename } },
5347 { VOPNAME_MKDIR, { .error = zfs_inval } },
5348 { VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir } },
5349 { VOPNAME_READDIR, { .vop_readdir = zfs_readdir } },
5350 { VOPNAME_SYMLINK, { .error = zfs_inval } },
5351 { VOPNAME_FSYNC, { .vop_fsync = zfs_fsync } },
5352 { VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive } },
5353 { VOPNAME_FID, { .vop_fid = zfs_fid } },
5354 { VOPNAME_SEEK, { .vop_seek = zfs_seek } },
5355 { VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf } },
5356 { VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr } },
5357 { VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr } },
5358 { VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support } },
5359 { NULL, { NULL } }
5360 };
5361
5362 /*
5363 * Error vnode operations template
5364 */
5365 vnodeops_t *zfs_evnodeops;
5366 const fs_operation_def_t zfs_evnodeops_template[] = {
5367 { VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive } },
5368 { VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf } },
5369 { NULL, { NULL } }
5370 };