1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
25 */
26
27 /* Portions Copyright 2007 Jeremy Teo */
28 /* Portions Copyright 2010 Robert Milkowski */
29
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/time.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
36 #include <sys/vfs.h>
37 #include <sys/vfs_opreg.h>
38 #include <sys/vnode.h>
39 #include <sys/file.h>
40 #include <sys/stat.h>
41 #include <sys/kmem.h>
42 #include <sys/taskq.h>
43 #include <sys/uio.h>
44 #include <sys/vmsystm.h>
45 #include <sys/atomic.h>
46 #include <sys/vm.h>
47 #include <vm/seg_vn.h>
48 #include <vm/pvn.h>
49 #include <vm/as.h>
50 #include <vm/kpm.h>
51 #include <vm/seg_kpm.h>
52 #include <sys/mman.h>
53 #include <sys/pathname.h>
54 #include <sys/cmn_err.h>
55 #include <sys/errno.h>
56 #include <sys/unistd.h>
57 #include <sys/zfs_dir.h>
58 #include <sys/zfs_acl.h>
59 #include <sys/zfs_ioctl.h>
60 #include <sys/fs/zfs.h>
61 #include <sys/dmu.h>
62 #include <sys/dmu_objset.h>
63 #include <sys/spa.h>
64 #include <sys/txg.h>
65 #include <sys/dbuf.h>
66 #include <sys/zap.h>
67 #include <sys/sa.h>
68 #include <sys/dirent.h>
69 #include <sys/policy.h>
70 #include <sys/sunddi.h>
71 #include <sys/filio.h>
72 #include <sys/sid.h>
73 #include "fs/fs_subr.h"
74 #include <sys/zfs_ctldir.h>
75 #include <sys/zfs_fuid.h>
76 #include <sys/zfs_sa.h>
77 #include <sys/dnlc.h>
78 #include <sys/zfs_rlock.h>
79 #include <sys/extdirent.h>
80 #include <sys/kidmap.h>
81 #include <sys/cred.h>
82 #include <sys/attr.h>
83
84 /*
85 * Programming rules.
86 *
87 * Each vnode op performs some logical unit of work. To do this, the ZPL must
88 * properly lock its in-core state, create a DMU transaction, do the work,
89 * record this work in the intent log (ZIL), commit the DMU transaction,
90 * and wait for the intent log to commit if it is a synchronous operation.
91 * Moreover, the vnode ops must work in both normal and log replay context.
92 * The ordering of events is important to avoid deadlocks and references
93 * to freed memory. The example below illustrates the following Big Rules:
94 *
95 * (1) A check must be made in each zfs thread for a mounted file system.
96 * This is done avoiding races using ZFS_ENTER(zfsvfs).
97 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
98 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
99 * can return EIO from the calling function.
100 *
101 * (2) VN_RELE() should always be the last thing except for zil_commit()
102 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
103 * First, if it's the last reference, the vnode/znode
104 * can be freed, so the zp may point to freed memory. Second, the last
105 * reference will call zfs_zinactive(), which may induce a lot of work --
106 * pushing cached pages (which acquires range locks) and syncing out
107 * cached atime changes. Third, zfs_zinactive() may require a new tx,
108 * which could deadlock the system if you were already holding one.
109 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
110 *
111 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
112 * as they can span dmu_tx_assign() calls.
113 *
114 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
115 * dmu_tx_assign(). This is critical because we don't want to block
116 * while holding locks.
117 *
118 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
119 * reduces lock contention and CPU usage when we must wait (note that if
120 * throughput is constrained by the storage, nearly every transaction
121 * must wait).
122 *
123 * Note, in particular, that if a lock is sometimes acquired before
124 * the tx assigns, and sometimes after (e.g. z_lock), then failing
125 * to use a non-blocking assign can deadlock the system. The scenario:
126 *
127 * Thread A has grabbed a lock before calling dmu_tx_assign().
128 * Thread B is in an already-assigned tx, and blocks for this lock.
129 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
130 * forever, because the previous txg can't quiesce until B's tx commits.
131 *
132 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
133 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
134 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
135 * to indicate that this operation has already called dmu_tx_wait().
136 * This will ensure that we don't retry forever, waiting a short bit
137 * each time.
138 *
139 * (5) If the operation succeeded, generate the intent log entry for it
140 * before dropping locks. This ensures that the ordering of events
141 * in the intent log matches the order in which they actually occurred.
142 * During ZIL replay the zfs_log_* functions will update the sequence
143 * number to indicate the zil transaction has replayed.
144 *
145 * (6) At the end of each vnode op, the DMU tx must always commit,
146 * regardless of whether there were any errors.
147 *
148 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
149 * to ensure that synchronous semantics are provided when necessary.
150 *
151 * In general, this is how things should be ordered in each vnode op:
152 *
153 * ZFS_ENTER(zfsvfs); // exit if unmounted
154 * top:
155 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
156 * rw_enter(...); // grab any other locks you need
157 * tx = dmu_tx_create(...); // get DMU tx
158 * dmu_tx_hold_*(); // hold each object you might modify
159 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
160 * if (error) {
161 * rw_exit(...); // drop locks
162 * zfs_dirent_unlock(dl); // unlock directory entry
163 * VN_RELE(...); // release held vnodes
164 * if (error == ERESTART) {
165 * waited = B_TRUE;
166 * dmu_tx_wait(tx);
167 * dmu_tx_abort(tx);
168 * goto top;
169 * }
170 * dmu_tx_abort(tx); // abort DMU tx
171 * ZFS_EXIT(zfsvfs); // finished in zfs
172 * return (error); // really out of space
173 * }
174 * error = do_real_work(); // do whatever this VOP does
175 * if (error == 0)
176 * zfs_log_*(...); // on success, make ZIL entry
177 * dmu_tx_commit(tx); // commit DMU tx -- error or not
178 * rw_exit(...); // drop locks
179 * zfs_dirent_unlock(dl); // unlock directory entry
180 * VN_RELE(...); // release held vnodes
181 * zil_commit(zilog, foid); // synchronous when necessary
182 * ZFS_EXIT(zfsvfs); // finished in zfs
183 * return (error); // done, report error
184 */
185
186 /* ARGSUSED */
187 static int
188 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
189 {
190 znode_t *zp = VTOZ(*vpp);
191 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
192
193 ZFS_ENTER(zfsvfs);
194 ZFS_VERIFY_ZP(zp);
195
196 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
197 ((flag & FAPPEND) == 0)) {
198 ZFS_EXIT(zfsvfs);
199 return (SET_ERROR(EPERM));
200 }
201
202 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
203 ZTOV(zp)->v_type == VREG &&
204 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
205 if (fs_vscan(*vpp, cr, 0) != 0) {
206 ZFS_EXIT(zfsvfs);
207 return (SET_ERROR(EACCES));
208 }
209 }
210
211 /* Keep a count of the synchronous opens in the znode */
212 if (flag & (FSYNC | FDSYNC))
213 atomic_inc_32(&zp->z_sync_cnt);
214
215 ZFS_EXIT(zfsvfs);
216 return (0);
217 }
218
219 /* ARGSUSED */
220 static int
221 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
222 caller_context_t *ct)
223 {
224 znode_t *zp = VTOZ(vp);
225 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
226
227 /*
228 * Clean up any locks held by this process on the vp.
229 */
230 cleanlocks(vp, ddi_get_pid(), 0);
231 cleanshares(vp, ddi_get_pid());
232
233 ZFS_ENTER(zfsvfs);
234 ZFS_VERIFY_ZP(zp);
235
236 /* Decrement the synchronous opens in the znode */
237 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
238 atomic_dec_32(&zp->z_sync_cnt);
239
240 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
241 ZTOV(zp)->v_type == VREG &&
242 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
243 VERIFY(fs_vscan(vp, cr, 1) == 0);
244
245 ZFS_EXIT(zfsvfs);
246 return (0);
247 }
248
249 /*
250 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
251 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
252 */
253 static int
254 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
255 {
256 znode_t *zp = VTOZ(vp);
257 uint64_t noff = (uint64_t)*off; /* new offset */
258 uint64_t file_sz;
259 int error;
260 boolean_t hole;
261
262 file_sz = zp->z_size;
263 if (noff >= file_sz) {
264 return (SET_ERROR(ENXIO));
265 }
266
267 if (cmd == _FIO_SEEK_HOLE)
268 hole = B_TRUE;
269 else
270 hole = B_FALSE;
271
272 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
273
274 /* end of file? */
275 if ((error == ESRCH) || (noff > file_sz)) {
276 /*
277 * Handle the virtual hole at the end of file.
278 */
279 if (hole) {
280 *off = file_sz;
281 return (0);
282 }
283 return (SET_ERROR(ENXIO));
284 }
285
286 if (noff < *off)
287 return (error);
288 *off = noff;
289 return (error);
290 }
291
292 /* ARGSUSED */
293 static int
294 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
295 int *rvalp, caller_context_t *ct)
296 {
297 offset_t off;
298 int error;
299 zfsvfs_t *zfsvfs;
300 znode_t *zp;
301
302 switch (com) {
303 case _FIOFFS:
304 return (zfs_sync(vp->v_vfsp, 0, cred));
305
306 /*
307 * The following two ioctls are used by bfu. Faking out,
308 * necessary to avoid bfu errors.
309 */
310 case _FIOGDIO:
311 case _FIOSDIO:
312 return (0);
313
314 case _FIO_SEEK_DATA:
315 case _FIO_SEEK_HOLE:
316 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
317 return (SET_ERROR(EFAULT));
318
319 zp = VTOZ(vp);
320 zfsvfs = zp->z_zfsvfs;
321 ZFS_ENTER(zfsvfs);
322 ZFS_VERIFY_ZP(zp);
323
324 /* offset parameter is in/out */
325 error = zfs_holey(vp, com, &off);
326 ZFS_EXIT(zfsvfs);
327 if (error)
328 return (error);
329 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
330 return (SET_ERROR(EFAULT));
331 return (0);
332 }
333 return (SET_ERROR(ENOTTY));
334 }
335
336 /*
337 * Utility functions to map and unmap a single physical page. These
338 * are used to manage the mappable copies of ZFS file data, and therefore
339 * do not update ref/mod bits.
340 */
341 caddr_t
342 zfs_map_page(page_t *pp, enum seg_rw rw)
343 {
344 if (kpm_enable)
345 return (hat_kpm_mapin(pp, 0));
346 ASSERT(rw == S_READ || rw == S_WRITE);
347 return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
348 (caddr_t)-1));
349 }
350
351 void
352 zfs_unmap_page(page_t *pp, caddr_t addr)
353 {
354 if (kpm_enable) {
355 hat_kpm_mapout(pp, 0, addr);
356 } else {
357 ppmapout(addr);
358 }
359 }
360
361 /*
362 * When a file is memory mapped, we must keep the IO data synchronized
363 * between the DMU cache and the memory mapped pages. What this means:
364 *
365 * On Write: If we find a memory mapped page, we write to *both*
366 * the page and the dmu buffer.
367 */
368 static void
369 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
370 {
371 int64_t off;
372
373 off = start & PAGEOFFSET;
374 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
375 page_t *pp;
376 uint64_t nbytes = MIN(PAGESIZE - off, len);
377
378 if (pp = page_lookup(vp, start, SE_SHARED)) {
379 caddr_t va;
380
381 va = zfs_map_page(pp, S_WRITE);
382 (void) dmu_read(os, oid, start+off, nbytes, va+off,
383 DMU_READ_PREFETCH);
384 zfs_unmap_page(pp, va);
385 page_unlock(pp);
386 }
387 len -= nbytes;
388 off = 0;
389 }
390 }
391
392 /*
393 * When a file is memory mapped, we must keep the IO data synchronized
394 * between the DMU cache and the memory mapped pages. What this means:
395 *
396 * On Read: We "read" preferentially from memory mapped pages,
397 * else we default from the dmu buffer.
398 *
399 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
400 * the file is memory mapped.
401 */
402 static int
403 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
404 {
405 znode_t *zp = VTOZ(vp);
406 objset_t *os = zp->z_zfsvfs->z_os;
407 int64_t start, off;
408 int len = nbytes;
409 int error = 0;
410
411 start = uio->uio_loffset;
412 off = start & PAGEOFFSET;
413 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
414 page_t *pp;
415 uint64_t bytes = MIN(PAGESIZE - off, len);
416
417 if (pp = page_lookup(vp, start, SE_SHARED)) {
418 caddr_t va;
419
420 va = zfs_map_page(pp, S_READ);
421 error = uiomove(va + off, bytes, UIO_READ, uio);
422 zfs_unmap_page(pp, va);
423 page_unlock(pp);
424 } else {
425 error = dmu_read_uio(os, zp->z_id, uio, bytes);
426 }
427 len -= bytes;
428 off = 0;
429 if (error)
430 break;
431 }
432 return (error);
433 }
434
435 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
436
437 /*
438 * Read bytes from specified file into supplied buffer.
439 *
440 * IN: vp - vnode of file to be read from.
441 * uio - structure supplying read location, range info,
442 * and return buffer.
443 * ioflag - SYNC flags; used to provide FRSYNC semantics.
444 * cr - credentials of caller.
445 * ct - caller context
446 *
447 * OUT: uio - updated offset and range, buffer filled.
448 *
449 * RETURN: 0 on success, error code on failure.
450 *
451 * Side Effects:
452 * vp - atime updated if byte count > 0
453 */
454 /* ARGSUSED */
455 static int
456 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
457 {
458 znode_t *zp = VTOZ(vp);
459 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
460 objset_t *os;
461 ssize_t n, nbytes;
462 int error = 0;
463 rl_t *rl;
464 xuio_t *xuio = NULL;
465
466 ZFS_ENTER(zfsvfs);
467 ZFS_VERIFY_ZP(zp);
468 os = zfsvfs->z_os;
469
470 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
471 ZFS_EXIT(zfsvfs);
472 return (SET_ERROR(EACCES));
473 }
474
475 /*
476 * Validate file offset
477 */
478 if (uio->uio_loffset < (offset_t)0) {
479 ZFS_EXIT(zfsvfs);
480 return (SET_ERROR(EINVAL));
481 }
482
483 /*
484 * Fasttrack empty reads
485 */
486 if (uio->uio_resid == 0) {
487 ZFS_EXIT(zfsvfs);
488 return (0);
489 }
490
491 /*
492 * Check for mandatory locks
493 */
494 if (MANDMODE(zp->z_mode)) {
495 if (error = chklock(vp, FREAD,
496 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
497 ZFS_EXIT(zfsvfs);
498 return (error);
499 }
500 }
501
502 /*
503 * If we're in FRSYNC mode, sync out this znode before reading it.
504 */
505 if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
506 zil_commit(zfsvfs->z_log, zp->z_id);
507
508 /*
509 * Lock the range against changes.
510 */
511 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
512
513 /*
514 * If we are reading past end-of-file we can skip
515 * to the end; but we might still need to set atime.
516 */
517 if (uio->uio_loffset >= zp->z_size) {
518 error = 0;
519 goto out;
520 }
521
522 ASSERT(uio->uio_loffset < zp->z_size);
523 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
524
525 if ((uio->uio_extflg == UIO_XUIO) &&
526 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
527 int nblk;
528 int blksz = zp->z_blksz;
529 uint64_t offset = uio->uio_loffset;
530
531 xuio = (xuio_t *)uio;
532 if ((ISP2(blksz))) {
533 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
534 blksz)) / blksz;
535 } else {
536 ASSERT(offset + n <= blksz);
537 nblk = 1;
538 }
539 (void) dmu_xuio_init(xuio, nblk);
540
541 if (vn_has_cached_data(vp)) {
542 /*
543 * For simplicity, we always allocate a full buffer
544 * even if we only expect to read a portion of a block.
545 */
546 while (--nblk >= 0) {
547 (void) dmu_xuio_add(xuio,
548 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
549 blksz), 0, blksz);
550 }
551 }
552 }
553
554 while (n > 0) {
555 nbytes = MIN(n, zfs_read_chunk_size -
556 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
557
558 if (vn_has_cached_data(vp))
559 error = mappedread(vp, nbytes, uio);
560 else
561 error = dmu_read_uio(os, zp->z_id, uio, nbytes);
562 if (error) {
563 /* convert checksum errors into IO errors */
564 if (error == ECKSUM)
565 error = SET_ERROR(EIO);
566 break;
567 }
568
569 n -= nbytes;
570 }
571 out:
572 zfs_range_unlock(rl);
573
574 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
575 ZFS_EXIT(zfsvfs);
576 return (error);
577 }
578
579 /*
580 * Write the bytes to a file.
581 *
582 * IN: vp - vnode of file to be written to.
583 * uio - structure supplying write location, range info,
584 * and data buffer.
585 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
586 * set if in append mode.
587 * cr - credentials of caller.
588 * ct - caller context (NFS/CIFS fem monitor only)
589 *
590 * OUT: uio - updated offset and range.
591 *
592 * RETURN: 0 on success, error code on failure.
593 *
594 * Timestamps:
595 * vp - ctime|mtime updated if byte count > 0
596 */
597
598 /* ARGSUSED */
599 static int
600 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
601 {
602 znode_t *zp = VTOZ(vp);
603 rlim64_t limit = uio->uio_llimit;
604 ssize_t start_resid = uio->uio_resid;
605 ssize_t tx_bytes;
606 uint64_t end_size;
607 dmu_tx_t *tx;
608 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
609 zilog_t *zilog;
610 offset_t woff;
611 ssize_t n, nbytes;
612 rl_t *rl;
613 int max_blksz = zfsvfs->z_max_blksz;
614 int error = 0;
615 arc_buf_t *abuf;
616 iovec_t *aiov = NULL;
617 xuio_t *xuio = NULL;
618 int i_iov = 0;
619 int iovcnt = uio->uio_iovcnt;
620 iovec_t *iovp = uio->uio_iov;
621 int write_eof;
622 int count = 0;
623 sa_bulk_attr_t bulk[4];
624 uint64_t mtime[2], ctime[2];
625
626 /*
627 * Fasttrack empty write
628 */
629 n = start_resid;
630 if (n == 0)
631 return (0);
632
633 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
634 limit = MAXOFFSET_T;
635
636 ZFS_ENTER(zfsvfs);
637 ZFS_VERIFY_ZP(zp);
638
639 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
640 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
641 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
642 &zp->z_size, 8);
643 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
644 &zp->z_pflags, 8);
645
646 /*
647 * If immutable or not appending then return EPERM
648 */
649 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
650 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
651 (uio->uio_loffset < zp->z_size))) {
652 ZFS_EXIT(zfsvfs);
653 return (SET_ERROR(EPERM));
654 }
655
656 zilog = zfsvfs->z_log;
657
658 /*
659 * Validate file offset
660 */
661 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
662 if (woff < 0) {
663 ZFS_EXIT(zfsvfs);
664 return (SET_ERROR(EINVAL));
665 }
666
667 /*
668 * Check for mandatory locks before calling zfs_range_lock()
669 * in order to prevent a deadlock with locks set via fcntl().
670 */
671 if (MANDMODE((mode_t)zp->z_mode) &&
672 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
673 ZFS_EXIT(zfsvfs);
674 return (error);
675 }
676
677 /*
678 * Pre-fault the pages to ensure slow (eg NFS) pages
679 * don't hold up txg.
680 * Skip this if uio contains loaned arc_buf.
681 */
682 if ((uio->uio_extflg == UIO_XUIO) &&
683 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
684 xuio = (xuio_t *)uio;
685 else
686 uio_prefaultpages(MIN(n, max_blksz), uio);
687
688 /*
689 * If in append mode, set the io offset pointer to eof.
690 */
691 if (ioflag & FAPPEND) {
692 /*
693 * Obtain an appending range lock to guarantee file append
694 * semantics. We reset the write offset once we have the lock.
695 */
696 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
697 woff = rl->r_off;
698 if (rl->r_len == UINT64_MAX) {
699 /*
700 * We overlocked the file because this write will cause
701 * the file block size to increase.
702 * Note that zp_size cannot change with this lock held.
703 */
704 woff = zp->z_size;
705 }
706 uio->uio_loffset = woff;
707 } else {
708 /*
709 * Note that if the file block size will change as a result of
710 * this write, then this range lock will lock the entire file
711 * so that we can re-write the block safely.
712 */
713 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
714 }
715
716 if (woff >= limit) {
717 zfs_range_unlock(rl);
718 ZFS_EXIT(zfsvfs);
719 return (SET_ERROR(EFBIG));
720 }
721
722 if ((woff + n) > limit || woff > (limit - n))
723 n = limit - woff;
724
725 /* Will this write extend the file length? */
726 write_eof = (woff + n > zp->z_size);
727
728 end_size = MAX(zp->z_size, woff + n);
729
730 /*
731 * Write the file in reasonable size chunks. Each chunk is written
732 * in a separate transaction; this keeps the intent log records small
733 * and allows us to do more fine-grained space accounting.
734 */
735 while (n > 0) {
736 abuf = NULL;
737 woff = uio->uio_loffset;
738 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
739 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
740 if (abuf != NULL)
741 dmu_return_arcbuf(abuf);
742 error = SET_ERROR(EDQUOT);
743 break;
744 }
745
746 if (xuio && abuf == NULL) {
747 ASSERT(i_iov < iovcnt);
748 aiov = &iovp[i_iov];
749 abuf = dmu_xuio_arcbuf(xuio, i_iov);
750 dmu_xuio_clear(xuio, i_iov);
751 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
752 iovec_t *, aiov, arc_buf_t *, abuf);
753 ASSERT((aiov->iov_base == abuf->b_data) ||
754 ((char *)aiov->iov_base - (char *)abuf->b_data +
755 aiov->iov_len == arc_buf_size(abuf)));
756 i_iov++;
757 } else if (abuf == NULL && n >= max_blksz &&
758 woff >= zp->z_size &&
759 P2PHASE(woff, max_blksz) == 0 &&
760 zp->z_blksz == max_blksz) {
761 /*
762 * This write covers a full block. "Borrow" a buffer
763 * from the dmu so that we can fill it before we enter
764 * a transaction. This avoids the possibility of
765 * holding up the transaction if the data copy hangs
766 * up on a pagefault (e.g., from an NFS server mapping).
767 */
768 size_t cbytes;
769
770 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
771 max_blksz);
772 ASSERT(abuf != NULL);
773 ASSERT(arc_buf_size(abuf) == max_blksz);
774 if (error = uiocopy(abuf->b_data, max_blksz,
775 UIO_WRITE, uio, &cbytes)) {
776 dmu_return_arcbuf(abuf);
777 break;
778 }
779 ASSERT(cbytes == max_blksz);
780 }
781
782 /*
783 * Start a transaction.
784 */
785 tx = dmu_tx_create(zfsvfs->z_os);
786 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
787 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
788 zfs_sa_upgrade_txholds(tx, zp);
789 error = dmu_tx_assign(tx, TXG_WAIT);
790 if (error) {
791 dmu_tx_abort(tx);
792 if (abuf != NULL)
793 dmu_return_arcbuf(abuf);
794 break;
795 }
796
797 /*
798 * If zfs_range_lock() over-locked we grow the blocksize
799 * and then reduce the lock range. This will only happen
800 * on the first iteration since zfs_range_reduce() will
801 * shrink down r_len to the appropriate size.
802 */
803 if (rl->r_len == UINT64_MAX) {
804 uint64_t new_blksz;
805
806 if (zp->z_blksz > max_blksz) {
807 ASSERT(!ISP2(zp->z_blksz));
808 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
809 } else {
810 new_blksz = MIN(end_size, max_blksz);
811 }
812 zfs_grow_blocksize(zp, new_blksz, tx);
813 zfs_range_reduce(rl, woff, n);
814 }
815
816 /*
817 * XXX - should we really limit each write to z_max_blksz?
818 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
819 */
820 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
821
822 if (abuf == NULL) {
823 tx_bytes = uio->uio_resid;
824 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
825 uio, nbytes, tx);
826 tx_bytes -= uio->uio_resid;
827 } else {
828 tx_bytes = nbytes;
829 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
830 /*
831 * If this is not a full block write, but we are
832 * extending the file past EOF and this data starts
833 * block-aligned, use assign_arcbuf(). Otherwise,
834 * write via dmu_write().
835 */
836 if (tx_bytes < max_blksz && (!write_eof ||
837 aiov->iov_base != abuf->b_data)) {
838 ASSERT(xuio);
839 dmu_write(zfsvfs->z_os, zp->z_id, woff,
840 aiov->iov_len, aiov->iov_base, tx);
841 dmu_return_arcbuf(abuf);
842 xuio_stat_wbuf_copied();
843 } else {
844 ASSERT(xuio || tx_bytes == max_blksz);
845 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
846 woff, abuf, tx);
847 }
848 ASSERT(tx_bytes <= uio->uio_resid);
849 uioskip(uio, tx_bytes);
850 }
851 if (tx_bytes && vn_has_cached_data(vp)) {
852 update_pages(vp, woff,
853 tx_bytes, zfsvfs->z_os, zp->z_id);
854 }
855
856 /*
857 * If we made no progress, we're done. If we made even
858 * partial progress, update the znode and ZIL accordingly.
859 */
860 if (tx_bytes == 0) {
861 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
862 (void *)&zp->z_size, sizeof (uint64_t), tx);
863 dmu_tx_commit(tx);
864 ASSERT(error != 0);
865 break;
866 }
867
868 /*
869 * Clear Set-UID/Set-GID bits on successful write if not
870 * privileged and at least one of the excute bits is set.
871 *
872 * It would be nice to to this after all writes have
873 * been done, but that would still expose the ISUID/ISGID
874 * to another app after the partial write is committed.
875 *
876 * Note: we don't call zfs_fuid_map_id() here because
877 * user 0 is not an ephemeral uid.
878 */
879 mutex_enter(&zp->z_acl_lock);
880 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
881 (S_IXUSR >> 6))) != 0 &&
882 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
883 secpolicy_vnode_setid_retain(cr,
884 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
885 uint64_t newmode;
886 zp->z_mode &= ~(S_ISUID | S_ISGID);
887 newmode = zp->z_mode;
888 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
889 (void *)&newmode, sizeof (uint64_t), tx);
890 }
891 mutex_exit(&zp->z_acl_lock);
892
893 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
894 B_TRUE);
895
896 /*
897 * Update the file size (zp_size) if it has changed;
898 * account for possible concurrent updates.
899 */
900 while ((end_size = zp->z_size) < uio->uio_loffset) {
901 (void) atomic_cas_64(&zp->z_size, end_size,
902 uio->uio_loffset);
903 ASSERT(error == 0);
904 }
905 /*
906 * If we are replaying and eof is non zero then force
907 * the file size to the specified eof. Note, there's no
908 * concurrency during replay.
909 */
910 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
911 zp->z_size = zfsvfs->z_replay_eof;
912
913 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
914
915 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
916 dmu_tx_commit(tx);
917
918 if (error != 0)
919 break;
920 ASSERT(tx_bytes == nbytes);
921 n -= nbytes;
922
923 if (!xuio && n > 0)
924 uio_prefaultpages(MIN(n, max_blksz), uio);
925 }
926
927 zfs_range_unlock(rl);
928
929 /*
930 * If we're in replay mode, or we made no progress, return error.
931 * Otherwise, it's at least a partial write, so it's successful.
932 */
933 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
934 ZFS_EXIT(zfsvfs);
935 return (error);
936 }
937
938 if (ioflag & (FSYNC | FDSYNC) ||
939 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
940 zil_commit(zilog, zp->z_id);
941
942 ZFS_EXIT(zfsvfs);
943 return (0);
944 }
945
946 void
947 zfs_get_done(zgd_t *zgd, int error)
948 {
949 znode_t *zp = zgd->zgd_private;
950 objset_t *os = zp->z_zfsvfs->z_os;
951
952 if (zgd->zgd_db)
953 dmu_buf_rele(zgd->zgd_db, zgd);
954
955 zfs_range_unlock(zgd->zgd_rl);
956
957 /*
958 * Release the vnode asynchronously as we currently have the
959 * txg stopped from syncing.
960 */
961 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
962
963 if (error == 0 && zgd->zgd_bp)
964 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
965
966 kmem_free(zgd, sizeof (zgd_t));
967 }
968
969 #ifdef DEBUG
970 static int zil_fault_io = 0;
971 #endif
972
973 /*
974 * Get data to generate a TX_WRITE intent log record.
975 */
976 int
977 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
978 {
979 zfsvfs_t *zfsvfs = arg;
980 objset_t *os = zfsvfs->z_os;
981 znode_t *zp;
982 uint64_t object = lr->lr_foid;
983 uint64_t offset = lr->lr_offset;
984 uint64_t size = lr->lr_length;
985 blkptr_t *bp = &lr->lr_blkptr;
986 dmu_buf_t *db;
987 zgd_t *zgd;
988 int error = 0;
989
990 ASSERT(zio != NULL);
991 ASSERT(size != 0);
992
993 /*
994 * Nothing to do if the file has been removed
995 */
996 if (zfs_zget(zfsvfs, object, &zp) != 0)
997 return (SET_ERROR(ENOENT));
998 if (zp->z_unlinked) {
999 /*
1000 * Release the vnode asynchronously as we currently have the
1001 * txg stopped from syncing.
1002 */
1003 VN_RELE_ASYNC(ZTOV(zp),
1004 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1005 return (SET_ERROR(ENOENT));
1006 }
1007
1008 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1009 zgd->zgd_zilog = zfsvfs->z_log;
1010 zgd->zgd_private = zp;
1011
1012 /*
1013 * Write records come in two flavors: immediate and indirect.
1014 * For small writes it's cheaper to store the data with the
1015 * log record (immediate); for large writes it's cheaper to
1016 * sync the data and get a pointer to it (indirect) so that
1017 * we don't have to write the data twice.
1018 */
1019 if (buf != NULL) { /* immediate write */
1020 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1021 /* test for truncation needs to be done while range locked */
1022 if (offset >= zp->z_size) {
1023 error = SET_ERROR(ENOENT);
1024 } else {
1025 error = dmu_read(os, object, offset, size, buf,
1026 DMU_READ_NO_PREFETCH);
1027 }
1028 ASSERT(error == 0 || error == ENOENT);
1029 } else { /* indirect write */
1030 /*
1031 * Have to lock the whole block to ensure when it's
1032 * written out and it's checksum is being calculated
1033 * that no one can change the data. We need to re-check
1034 * blocksize after we get the lock in case it's changed!
1035 */
1036 for (;;) {
1037 uint64_t blkoff;
1038 size = zp->z_blksz;
1039 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1040 offset -= blkoff;
1041 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1042 RL_READER);
1043 if (zp->z_blksz == size)
1044 break;
1045 offset += blkoff;
1046 zfs_range_unlock(zgd->zgd_rl);
1047 }
1048 /* test for truncation needs to be done while range locked */
1049 if (lr->lr_offset >= zp->z_size)
1050 error = SET_ERROR(ENOENT);
1051 #ifdef DEBUG
1052 if (zil_fault_io) {
1053 error = SET_ERROR(EIO);
1054 zil_fault_io = 0;
1055 }
1056 #endif
1057 if (error == 0)
1058 error = dmu_buf_hold(os, object, offset, zgd, &db,
1059 DMU_READ_NO_PREFETCH);
1060
1061 if (error == 0) {
1062 blkptr_t *obp = dmu_buf_get_blkptr(db);
1063 if (obp) {
1064 ASSERT(BP_IS_HOLE(bp));
1065 *bp = *obp;
1066 }
1067
1068 zgd->zgd_db = db;
1069 zgd->zgd_bp = bp;
1070
1071 ASSERT(db->db_offset == offset);
1072 ASSERT(db->db_size == size);
1073
1074 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1075 zfs_get_done, zgd);
1076 ASSERT(error || lr->lr_length <= zp->z_blksz);
1077
1078 /*
1079 * On success, we need to wait for the write I/O
1080 * initiated by dmu_sync() to complete before we can
1081 * release this dbuf. We will finish everything up
1082 * in the zfs_get_done() callback.
1083 */
1084 if (error == 0)
1085 return (0);
1086
1087 if (error == EALREADY) {
1088 lr->lr_common.lrc_txtype = TX_WRITE2;
1089 error = 0;
1090 }
1091 }
1092 }
1093
1094 zfs_get_done(zgd, error);
1095
1096 return (error);
1097 }
1098
1099 /*ARGSUSED*/
1100 static int
1101 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1102 caller_context_t *ct)
1103 {
1104 znode_t *zp = VTOZ(vp);
1105 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1106 int error;
1107
1108 ZFS_ENTER(zfsvfs);
1109 ZFS_VERIFY_ZP(zp);
1110
1111 if (flag & V_ACE_MASK)
1112 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1113 else
1114 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1115
1116 ZFS_EXIT(zfsvfs);
1117 return (error);
1118 }
1119
1120 /*
1121 * If vnode is for a device return a specfs vnode instead.
1122 */
1123 static int
1124 specvp_check(vnode_t **vpp, cred_t *cr)
1125 {
1126 int error = 0;
1127
1128 if (IS_DEVVP(*vpp)) {
1129 struct vnode *svp;
1130
1131 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1132 VN_RELE(*vpp);
1133 if (svp == NULL)
1134 error = SET_ERROR(ENOSYS);
1135 *vpp = svp;
1136 }
1137 return (error);
1138 }
1139
1140
1141 /*
1142 * Lookup an entry in a directory, or an extended attribute directory.
1143 * If it exists, return a held vnode reference for it.
1144 *
1145 * IN: dvp - vnode of directory to search.
1146 * nm - name of entry to lookup.
1147 * pnp - full pathname to lookup [UNUSED].
1148 * flags - LOOKUP_XATTR set if looking for an attribute.
1149 * rdir - root directory vnode [UNUSED].
1150 * cr - credentials of caller.
1151 * ct - caller context
1152 * direntflags - directory lookup flags
1153 * realpnp - returned pathname.
1154 *
1155 * OUT: vpp - vnode of located entry, NULL if not found.
1156 *
1157 * RETURN: 0 on success, error code on failure.
1158 *
1159 * Timestamps:
1160 * NA
1161 */
1162 /* ARGSUSED */
1163 static int
1164 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1165 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
1166 int *direntflags, pathname_t *realpnp)
1167 {
1168 znode_t *zdp = VTOZ(dvp);
1169 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1170 int error = 0;
1171
1172 /* fast path */
1173 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1174
1175 if (dvp->v_type != VDIR) {
1176 return (SET_ERROR(ENOTDIR));
1177 } else if (zdp->z_sa_hdl == NULL) {
1178 return (SET_ERROR(EIO));
1179 }
1180
1181 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1182 error = zfs_fastaccesschk_execute(zdp, cr);
1183 if (!error) {
1184 *vpp = dvp;
1185 VN_HOLD(*vpp);
1186 return (0);
1187 }
1188 return (error);
1189 } else {
1190 vnode_t *tvp = dnlc_lookup(dvp, nm);
1191
1192 if (tvp) {
1193 error = zfs_fastaccesschk_execute(zdp, cr);
1194 if (error) {
1195 VN_RELE(tvp);
1196 return (error);
1197 }
1198 if (tvp == DNLC_NO_VNODE) {
1199 VN_RELE(tvp);
1200 return (SET_ERROR(ENOENT));
1201 } else {
1202 *vpp = tvp;
1203 return (specvp_check(vpp, cr));
1204 }
1205 }
1206 }
1207 }
1208
1209 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1210
1211 ZFS_ENTER(zfsvfs);
1212 ZFS_VERIFY_ZP(zdp);
1213
1214 *vpp = NULL;
1215
1216 if (flags & LOOKUP_XATTR) {
1217 /*
1218 * If the xattr property is off, refuse the lookup request.
1219 */
1220 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1221 ZFS_EXIT(zfsvfs);
1222 return (SET_ERROR(EINVAL));
1223 }
1224
1225 /*
1226 * We don't allow recursive attributes..
1227 * Maybe someday we will.
1228 */
1229 if (zdp->z_pflags & ZFS_XATTR) {
1230 ZFS_EXIT(zfsvfs);
1231 return (SET_ERROR(EINVAL));
1232 }
1233
1234 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1235 ZFS_EXIT(zfsvfs);
1236 return (error);
1237 }
1238
1239 /*
1240 * Do we have permission to get into attribute directory?
1241 */
1242
1243 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1244 B_FALSE, cr)) {
1245 VN_RELE(*vpp);
1246 *vpp = NULL;
1247 }
1248
1249 ZFS_EXIT(zfsvfs);
1250 return (error);
1251 }
1252
1253 if (dvp->v_type != VDIR) {
1254 ZFS_EXIT(zfsvfs);
1255 return (SET_ERROR(ENOTDIR));
1256 }
1257
1258 /*
1259 * Check accessibility of directory.
1260 */
1261
1262 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1263 ZFS_EXIT(zfsvfs);
1264 return (error);
1265 }
1266
1267 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1268 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1269 ZFS_EXIT(zfsvfs);
1270 return (SET_ERROR(EILSEQ));
1271 }
1272
1273 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1274 if (error == 0)
1275 error = specvp_check(vpp, cr);
1276
1277 ZFS_EXIT(zfsvfs);
1278 return (error);
1279 }
1280
1281 /*
1282 * Attempt to create a new entry in a directory. If the entry
1283 * already exists, truncate the file if permissible, else return
1284 * an error. Return the vp of the created or trunc'd file.
1285 *
1286 * IN: dvp - vnode of directory to put new file entry in.
1287 * name - name of new file entry.
1288 * vap - attributes of new file.
1289 * excl - flag indicating exclusive or non-exclusive mode.
1290 * mode - mode to open file with.
1291 * cr - credentials of caller.
1292 * flag - large file flag [UNUSED].
1293 * ct - caller context
1294 * vsecp - ACL to be set
1295 *
1296 * OUT: vpp - vnode of created or trunc'd entry.
1297 *
1298 * RETURN: 0 on success, error code on failure.
1299 *
1300 * Timestamps:
1301 * dvp - ctime|mtime updated if new entry created
1302 * vp - ctime|mtime always, atime if new
1303 */
1304
1305 /* ARGSUSED */
1306 static int
1307 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1308 int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1309 vsecattr_t *vsecp)
1310 {
1311 znode_t *zp, *dzp = VTOZ(dvp);
1312 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1313 zilog_t *zilog;
1314 objset_t *os;
1315 zfs_dirlock_t *dl;
1316 dmu_tx_t *tx;
1317 int error;
1318 ksid_t *ksid;
1319 uid_t uid;
1320 gid_t gid = crgetgid(cr);
1321 zfs_acl_ids_t acl_ids;
1322 boolean_t fuid_dirtied;
1323 boolean_t have_acl = B_FALSE;
1324 boolean_t waited = B_FALSE;
1325
1326 /*
1327 * If we have an ephemeral id, ACL, or XVATTR then
1328 * make sure file system is at proper version
1329 */
1330
1331 ksid = crgetsid(cr, KSID_OWNER);
1332 if (ksid)
1333 uid = ksid_getid(ksid);
1334 else
1335 uid = crgetuid(cr);
1336
1337 if (zfsvfs->z_use_fuids == B_FALSE &&
1338 (vsecp || (vap->va_mask & AT_XVATTR) ||
1339 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1340 return (SET_ERROR(EINVAL));
1341
1342 ZFS_ENTER(zfsvfs);
1343 ZFS_VERIFY_ZP(dzp);
1344 os = zfsvfs->z_os;
1345 zilog = zfsvfs->z_log;
1346
1347 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1348 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1349 ZFS_EXIT(zfsvfs);
1350 return (SET_ERROR(EILSEQ));
1351 }
1352
1353 if (vap->va_mask & AT_XVATTR) {
1354 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1355 crgetuid(cr), cr, vap->va_type)) != 0) {
1356 ZFS_EXIT(zfsvfs);
1357 return (error);
1358 }
1359 }
1360 top:
1361 *vpp = NULL;
1362
1363 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1364 vap->va_mode &= ~VSVTX;
1365
1366 if (*name == '\0') {
1367 /*
1368 * Null component name refers to the directory itself.
1369 */
1370 VN_HOLD(dvp);
1371 zp = dzp;
1372 dl = NULL;
1373 error = 0;
1374 } else {
1375 /* possible VN_HOLD(zp) */
1376 int zflg = 0;
1377
1378 if (flag & FIGNORECASE)
1379 zflg |= ZCILOOK;
1380
1381 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1382 NULL, NULL);
1383 if (error) {
1384 if (have_acl)
1385 zfs_acl_ids_free(&acl_ids);
1386 if (strcmp(name, "..") == 0)
1387 error = SET_ERROR(EISDIR);
1388 ZFS_EXIT(zfsvfs);
1389 return (error);
1390 }
1391 }
1392
1393 if (zp == NULL) {
1394 uint64_t txtype;
1395
1396 /*
1397 * Create a new file object and update the directory
1398 * to reference it.
1399 */
1400 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1401 if (have_acl)
1402 zfs_acl_ids_free(&acl_ids);
1403 goto out;
1404 }
1405
1406 /*
1407 * We only support the creation of regular files in
1408 * extended attribute directories.
1409 */
1410
1411 if ((dzp->z_pflags & ZFS_XATTR) &&
1412 (vap->va_type != VREG)) {
1413 if (have_acl)
1414 zfs_acl_ids_free(&acl_ids);
1415 error = SET_ERROR(EINVAL);
1416 goto out;
1417 }
1418
1419 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1420 cr, vsecp, &acl_ids)) != 0)
1421 goto out;
1422 have_acl = B_TRUE;
1423
1424 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1425 zfs_acl_ids_free(&acl_ids);
1426 error = SET_ERROR(EDQUOT);
1427 goto out;
1428 }
1429
1430 tx = dmu_tx_create(os);
1431
1432 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1433 ZFS_SA_BASE_ATTR_SIZE);
1434
1435 fuid_dirtied = zfsvfs->z_fuid_dirty;
1436 if (fuid_dirtied)
1437 zfs_fuid_txhold(zfsvfs, tx);
1438 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1439 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1440 if (!zfsvfs->z_use_sa &&
1441 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1442 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1443 0, acl_ids.z_aclp->z_acl_bytes);
1444 }
1445 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1446 if (error) {
1447 zfs_dirent_unlock(dl);
1448 if (error == ERESTART) {
1449 waited = B_TRUE;
1450 dmu_tx_wait(tx);
1451 dmu_tx_abort(tx);
1452 goto top;
1453 }
1454 zfs_acl_ids_free(&acl_ids);
1455 dmu_tx_abort(tx);
1456 ZFS_EXIT(zfsvfs);
1457 return (error);
1458 }
1459 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1460
1461 if (fuid_dirtied)
1462 zfs_fuid_sync(zfsvfs, tx);
1463
1464 (void) zfs_link_create(dl, zp, tx, ZNEW);
1465 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1466 if (flag & FIGNORECASE)
1467 txtype |= TX_CI;
1468 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1469 vsecp, acl_ids.z_fuidp, vap);
1470 zfs_acl_ids_free(&acl_ids);
1471 dmu_tx_commit(tx);
1472 } else {
1473 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1474
1475 if (have_acl)
1476 zfs_acl_ids_free(&acl_ids);
1477 have_acl = B_FALSE;
1478
1479 /*
1480 * A directory entry already exists for this name.
1481 */
1482 /*
1483 * Can't truncate an existing file if in exclusive mode.
1484 */
1485 if (excl == EXCL) {
1486 error = SET_ERROR(EEXIST);
1487 goto out;
1488 }
1489 /*
1490 * Can't open a directory for writing.
1491 */
1492 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1493 error = SET_ERROR(EISDIR);
1494 goto out;
1495 }
1496 /*
1497 * Verify requested access to file.
1498 */
1499 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1500 goto out;
1501 }
1502
1503 mutex_enter(&dzp->z_lock);
1504 dzp->z_seq++;
1505 mutex_exit(&dzp->z_lock);
1506
1507 /*
1508 * Truncate regular files if requested.
1509 */
1510 if ((ZTOV(zp)->v_type == VREG) &&
1511 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1512 /* we can't hold any locks when calling zfs_freesp() */
1513 zfs_dirent_unlock(dl);
1514 dl = NULL;
1515 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1516 if (error == 0) {
1517 vnevent_create(ZTOV(zp), ct);
1518 }
1519 }
1520 }
1521 out:
1522
1523 if (dl)
1524 zfs_dirent_unlock(dl);
1525
1526 if (error) {
1527 if (zp)
1528 VN_RELE(ZTOV(zp));
1529 } else {
1530 *vpp = ZTOV(zp);
1531 error = specvp_check(vpp, cr);
1532 }
1533
1534 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1535 zil_commit(zilog, 0);
1536
1537 ZFS_EXIT(zfsvfs);
1538 return (error);
1539 }
1540
1541 /*
1542 * Remove an entry from a directory.
1543 *
1544 * IN: dvp - vnode of directory to remove entry from.
1545 * name - name of entry to remove.
1546 * cr - credentials of caller.
1547 * ct - caller context
1548 * flags - case flags
1549 *
1550 * RETURN: 0 on success, error code on failure.
1551 *
1552 * Timestamps:
1553 * dvp - ctime|mtime
1554 * vp - ctime (if nlink > 0)
1555 */
1556
1557 uint64_t null_xattr = 0;
1558
1559 /*ARGSUSED*/
1560 static int
1561 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1562 int flags)
1563 {
1564 znode_t *zp, *dzp = VTOZ(dvp);
1565 znode_t *xzp;
1566 vnode_t *vp;
1567 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1568 zilog_t *zilog;
1569 uint64_t acl_obj, xattr_obj;
1570 uint64_t xattr_obj_unlinked = 0;
1571 uint64_t obj = 0;
1572 zfs_dirlock_t *dl;
1573 dmu_tx_t *tx;
1574 boolean_t may_delete_now, delete_now = FALSE;
1575 boolean_t unlinked, toobig = FALSE;
1576 uint64_t txtype;
1577 pathname_t *realnmp = NULL;
1578 pathname_t realnm;
1579 int error;
1580 int zflg = ZEXISTS;
1581 boolean_t waited = B_FALSE;
1582
1583 ZFS_ENTER(zfsvfs);
1584 ZFS_VERIFY_ZP(dzp);
1585 zilog = zfsvfs->z_log;
1586
1587 if (flags & FIGNORECASE) {
1588 zflg |= ZCILOOK;
1589 pn_alloc(&realnm);
1590 realnmp = &realnm;
1591 }
1592
1593 top:
1594 xattr_obj = 0;
1595 xzp = NULL;
1596 /*
1597 * Attempt to lock directory; fail if entry doesn't exist.
1598 */
1599 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1600 NULL, realnmp)) {
1601 if (realnmp)
1602 pn_free(realnmp);
1603 ZFS_EXIT(zfsvfs);
1604 return (error);
1605 }
1606
1607 vp = ZTOV(zp);
1608
1609 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1610 goto out;
1611 }
1612
1613 /*
1614 * Need to use rmdir for removing directories.
1615 */
1616 if (vp->v_type == VDIR) {
1617 error = SET_ERROR(EPERM);
1618 goto out;
1619 }
1620
1621 vnevent_remove(vp, dvp, name, ct);
1622
1623 if (realnmp)
1624 dnlc_remove(dvp, realnmp->pn_buf);
1625 else
1626 dnlc_remove(dvp, name);
1627
1628 mutex_enter(&vp->v_lock);
1629 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1630 mutex_exit(&vp->v_lock);
1631
1632 /*
1633 * We may delete the znode now, or we may put it in the unlinked set;
1634 * it depends on whether we're the last link, and on whether there are
1635 * other holds on the vnode. So we dmu_tx_hold() the right things to
1636 * allow for either case.
1637 */
1638 obj = zp->z_id;
1639 tx = dmu_tx_create(zfsvfs->z_os);
1640 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1641 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1642 zfs_sa_upgrade_txholds(tx, zp);
1643 zfs_sa_upgrade_txholds(tx, dzp);
1644 if (may_delete_now) {
1645 toobig =
1646 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1647 /* if the file is too big, only hold_free a token amount */
1648 dmu_tx_hold_free(tx, zp->z_id, 0,
1649 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1650 }
1651
1652 /* are there any extended attributes? */
1653 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1654 &xattr_obj, sizeof (xattr_obj));
1655 if (error == 0 && xattr_obj) {
1656 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1657 ASSERT0(error);
1658 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1659 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1660 }
1661
1662 mutex_enter(&zp->z_lock);
1663 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1664 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1665 mutex_exit(&zp->z_lock);
1666
1667 /* charge as an update -- would be nice not to charge at all */
1668 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1669
1670 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1671 if (error) {
1672 zfs_dirent_unlock(dl);
1673 VN_RELE(vp);
1674 if (xzp)
1675 VN_RELE(ZTOV(xzp));
1676 if (error == ERESTART) {
1677 waited = B_TRUE;
1678 dmu_tx_wait(tx);
1679 dmu_tx_abort(tx);
1680 goto top;
1681 }
1682 if (realnmp)
1683 pn_free(realnmp);
1684 dmu_tx_abort(tx);
1685 ZFS_EXIT(zfsvfs);
1686 return (error);
1687 }
1688
1689 /*
1690 * Remove the directory entry.
1691 */
1692 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1693
1694 if (error) {
1695 dmu_tx_commit(tx);
1696 goto out;
1697 }
1698
1699 if (unlinked) {
1700
1701 /*
1702 * Hold z_lock so that we can make sure that the ACL obj
1703 * hasn't changed. Could have been deleted due to
1704 * zfs_sa_upgrade().
1705 */
1706 mutex_enter(&zp->z_lock);
1707 mutex_enter(&vp->v_lock);
1708 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1709 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1710 delete_now = may_delete_now && !toobig &&
1711 vp->v_count == 1 && !vn_has_cached_data(vp) &&
1712 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1713 acl_obj;
1714 mutex_exit(&vp->v_lock);
1715 }
1716
1717 if (delete_now) {
1718 if (xattr_obj_unlinked) {
1719 ASSERT3U(xzp->z_links, ==, 2);
1720 mutex_enter(&xzp->z_lock);
1721 xzp->z_unlinked = 1;
1722 xzp->z_links = 0;
1723 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1724 &xzp->z_links, sizeof (xzp->z_links), tx);
1725 ASSERT3U(error, ==, 0);
1726 mutex_exit(&xzp->z_lock);
1727 zfs_unlinked_add(xzp, tx);
1728
1729 if (zp->z_is_sa)
1730 error = sa_remove(zp->z_sa_hdl,
1731 SA_ZPL_XATTR(zfsvfs), tx);
1732 else
1733 error = sa_update(zp->z_sa_hdl,
1734 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1735 sizeof (uint64_t), tx);
1736 ASSERT0(error);
1737 }
1738 mutex_enter(&vp->v_lock);
1739 vp->v_count--;
1740 ASSERT0(vp->v_count);
1741 mutex_exit(&vp->v_lock);
1742 mutex_exit(&zp->z_lock);
1743 zfs_znode_delete(zp, tx);
1744 } else if (unlinked) {
1745 mutex_exit(&zp->z_lock);
1746 zfs_unlinked_add(zp, tx);
1747 }
1748
1749 txtype = TX_REMOVE;
1750 if (flags & FIGNORECASE)
1751 txtype |= TX_CI;
1752 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1753
1754 dmu_tx_commit(tx);
1755 out:
1756 if (realnmp)
1757 pn_free(realnmp);
1758
1759 zfs_dirent_unlock(dl);
1760
1761 if (!delete_now)
1762 VN_RELE(vp);
1763 if (xzp)
1764 VN_RELE(ZTOV(xzp));
1765
1766 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1767 zil_commit(zilog, 0);
1768
1769 ZFS_EXIT(zfsvfs);
1770 return (error);
1771 }
1772
1773 /*
1774 * Create a new directory and insert it into dvp using the name
1775 * provided. Return a pointer to the inserted directory.
1776 *
1777 * IN: dvp - vnode of directory to add subdir to.
1778 * dirname - name of new directory.
1779 * vap - attributes of new directory.
1780 * cr - credentials of caller.
1781 * ct - caller context
1782 * flags - case flags
1783 * vsecp - ACL to be set
1784 *
1785 * OUT: vpp - vnode of created directory.
1786 *
1787 * RETURN: 0 on success, error code on failure.
1788 *
1789 * Timestamps:
1790 * dvp - ctime|mtime updated
1791 * vp - ctime|mtime|atime updated
1792 */
1793 /*ARGSUSED*/
1794 static int
1795 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1796 caller_context_t *ct, int flags, vsecattr_t *vsecp)
1797 {
1798 znode_t *zp, *dzp = VTOZ(dvp);
1799 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1800 zilog_t *zilog;
1801 zfs_dirlock_t *dl;
1802 uint64_t txtype;
1803 dmu_tx_t *tx;
1804 int error;
1805 int zf = ZNEW;
1806 ksid_t *ksid;
1807 uid_t uid;
1808 gid_t gid = crgetgid(cr);
1809 zfs_acl_ids_t acl_ids;
1810 boolean_t fuid_dirtied;
1811 boolean_t waited = B_FALSE;
1812
1813 ASSERT(vap->va_type == VDIR);
1814
1815 /*
1816 * If we have an ephemeral id, ACL, or XVATTR then
1817 * make sure file system is at proper version
1818 */
1819
1820 ksid = crgetsid(cr, KSID_OWNER);
1821 if (ksid)
1822 uid = ksid_getid(ksid);
1823 else
1824 uid = crgetuid(cr);
1825 if (zfsvfs->z_use_fuids == B_FALSE &&
1826 (vsecp || (vap->va_mask & AT_XVATTR) ||
1827 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1828 return (SET_ERROR(EINVAL));
1829
1830 ZFS_ENTER(zfsvfs);
1831 ZFS_VERIFY_ZP(dzp);
1832 zilog = zfsvfs->z_log;
1833
1834 if (dzp->z_pflags & ZFS_XATTR) {
1835 ZFS_EXIT(zfsvfs);
1836 return (SET_ERROR(EINVAL));
1837 }
1838
1839 if (zfsvfs->z_utf8 && u8_validate(dirname,
1840 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1841 ZFS_EXIT(zfsvfs);
1842 return (SET_ERROR(EILSEQ));
1843 }
1844 if (flags & FIGNORECASE)
1845 zf |= ZCILOOK;
1846
1847 if (vap->va_mask & AT_XVATTR) {
1848 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1849 crgetuid(cr), cr, vap->va_type)) != 0) {
1850 ZFS_EXIT(zfsvfs);
1851 return (error);
1852 }
1853 }
1854
1855 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1856 vsecp, &acl_ids)) != 0) {
1857 ZFS_EXIT(zfsvfs);
1858 return (error);
1859 }
1860 /*
1861 * First make sure the new directory doesn't exist.
1862 *
1863 * Existence is checked first to make sure we don't return
1864 * EACCES instead of EEXIST which can cause some applications
1865 * to fail.
1866 */
1867 top:
1868 *vpp = NULL;
1869
1870 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1871 NULL, NULL)) {
1872 zfs_acl_ids_free(&acl_ids);
1873 ZFS_EXIT(zfsvfs);
1874 return (error);
1875 }
1876
1877 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1878 zfs_acl_ids_free(&acl_ids);
1879 zfs_dirent_unlock(dl);
1880 ZFS_EXIT(zfsvfs);
1881 return (error);
1882 }
1883
1884 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1885 zfs_acl_ids_free(&acl_ids);
1886 zfs_dirent_unlock(dl);
1887 ZFS_EXIT(zfsvfs);
1888 return (SET_ERROR(EDQUOT));
1889 }
1890
1891 /*
1892 * Add a new entry to the directory.
1893 */
1894 tx = dmu_tx_create(zfsvfs->z_os);
1895 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1896 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1897 fuid_dirtied = zfsvfs->z_fuid_dirty;
1898 if (fuid_dirtied)
1899 zfs_fuid_txhold(zfsvfs, tx);
1900 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1901 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1902 acl_ids.z_aclp->z_acl_bytes);
1903 }
1904
1905 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1906 ZFS_SA_BASE_ATTR_SIZE);
1907
1908 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1909 if (error) {
1910 zfs_dirent_unlock(dl);
1911 if (error == ERESTART) {
1912 waited = B_TRUE;
1913 dmu_tx_wait(tx);
1914 dmu_tx_abort(tx);
1915 goto top;
1916 }
1917 zfs_acl_ids_free(&acl_ids);
1918 dmu_tx_abort(tx);
1919 ZFS_EXIT(zfsvfs);
1920 return (error);
1921 }
1922
1923 /*
1924 * Create new node.
1925 */
1926 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1927
1928 if (fuid_dirtied)
1929 zfs_fuid_sync(zfsvfs, tx);
1930
1931 /*
1932 * Now put new name in parent dir.
1933 */
1934 (void) zfs_link_create(dl, zp, tx, ZNEW);
1935
1936 *vpp = ZTOV(zp);
1937
1938 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1939 if (flags & FIGNORECASE)
1940 txtype |= TX_CI;
1941 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1942 acl_ids.z_fuidp, vap);
1943
1944 zfs_acl_ids_free(&acl_ids);
1945
1946 dmu_tx_commit(tx);
1947
1948 zfs_dirent_unlock(dl);
1949
1950 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1951 zil_commit(zilog, 0);
1952
1953 ZFS_EXIT(zfsvfs);
1954 return (0);
1955 }
1956
1957 /*
1958 * Remove a directory subdir entry. If the current working
1959 * directory is the same as the subdir to be removed, the
1960 * remove will fail.
1961 *
1962 * IN: dvp - vnode of directory to remove from.
1963 * name - name of directory to be removed.
1964 * cwd - vnode of current working directory.
1965 * cr - credentials of caller.
1966 * ct - caller context
1967 * flags - case flags
1968 *
1969 * RETURN: 0 on success, error code on failure.
1970 *
1971 * Timestamps:
1972 * dvp - ctime|mtime updated
1973 */
1974 /*ARGSUSED*/
1975 static int
1976 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1977 caller_context_t *ct, int flags)
1978 {
1979 znode_t *dzp = VTOZ(dvp);
1980 znode_t *zp;
1981 vnode_t *vp;
1982 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1983 zilog_t *zilog;
1984 zfs_dirlock_t *dl;
1985 dmu_tx_t *tx;
1986 int error;
1987 int zflg = ZEXISTS;
1988 boolean_t waited = B_FALSE;
1989
1990 ZFS_ENTER(zfsvfs);
1991 ZFS_VERIFY_ZP(dzp);
1992 zilog = zfsvfs->z_log;
1993
1994 if (flags & FIGNORECASE)
1995 zflg |= ZCILOOK;
1996 top:
1997 zp = NULL;
1998
1999 /*
2000 * Attempt to lock directory; fail if entry doesn't exist.
2001 */
2002 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2003 NULL, NULL)) {
2004 ZFS_EXIT(zfsvfs);
2005 return (error);
2006 }
2007
2008 vp = ZTOV(zp);
2009
2010 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2011 goto out;
2012 }
2013
2014 if (vp->v_type != VDIR) {
2015 error = SET_ERROR(ENOTDIR);
2016 goto out;
2017 }
2018
2019 if (vp == cwd) {
2020 error = SET_ERROR(EINVAL);
2021 goto out;
2022 }
2023
2024 vnevent_rmdir(vp, dvp, name, ct);
2025
2026 /*
2027 * Grab a lock on the directory to make sure that noone is
2028 * trying to add (or lookup) entries while we are removing it.
2029 */
2030 rw_enter(&zp->z_name_lock, RW_WRITER);
2031
2032 /*
2033 * Grab a lock on the parent pointer to make sure we play well
2034 * with the treewalk and directory rename code.
2035 */
2036 rw_enter(&zp->z_parent_lock, RW_WRITER);
2037
2038 tx = dmu_tx_create(zfsvfs->z_os);
2039 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2040 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2041 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2042 zfs_sa_upgrade_txholds(tx, zp);
2043 zfs_sa_upgrade_txholds(tx, dzp);
2044 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2045 if (error) {
2046 rw_exit(&zp->z_parent_lock);
2047 rw_exit(&zp->z_name_lock);
2048 zfs_dirent_unlock(dl);
2049 VN_RELE(vp);
2050 if (error == ERESTART) {
2051 waited = B_TRUE;
2052 dmu_tx_wait(tx);
2053 dmu_tx_abort(tx);
2054 goto top;
2055 }
2056 dmu_tx_abort(tx);
2057 ZFS_EXIT(zfsvfs);
2058 return (error);
2059 }
2060
2061 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2062
2063 if (error == 0) {
2064 uint64_t txtype = TX_RMDIR;
2065 if (flags & FIGNORECASE)
2066 txtype |= TX_CI;
2067 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2068 }
2069
2070 dmu_tx_commit(tx);
2071
2072 rw_exit(&zp->z_parent_lock);
2073 rw_exit(&zp->z_name_lock);
2074 out:
2075 zfs_dirent_unlock(dl);
2076
2077 VN_RELE(vp);
2078
2079 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2080 zil_commit(zilog, 0);
2081
2082 ZFS_EXIT(zfsvfs);
2083 return (error);
2084 }
2085
2086 /*
2087 * Read as many directory entries as will fit into the provided
2088 * buffer from the given directory cursor position (specified in
2089 * the uio structure).
2090 *
2091 * IN: vp - vnode of directory to read.
2092 * uio - structure supplying read location, range info,
2093 * and return buffer.
2094 * cr - credentials of caller.
2095 * ct - caller context
2096 * flags - case flags
2097 *
2098 * OUT: uio - updated offset and range, buffer filled.
2099 * eofp - set to true if end-of-file detected.
2100 *
2101 * RETURN: 0 on success, error code on failure.
2102 *
2103 * Timestamps:
2104 * vp - atime updated
2105 *
2106 * Note that the low 4 bits of the cookie returned by zap is always zero.
2107 * This allows us to use the low range for "special" directory entries:
2108 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2109 * we use the offset 2 for the '.zfs' directory.
2110 */
2111 /* ARGSUSED */
2112 static int
2113 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2114 caller_context_t *ct, int flags)
2115 {
2116 znode_t *zp = VTOZ(vp);
2117 iovec_t *iovp;
2118 edirent_t *eodp;
2119 dirent64_t *odp;
2120 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2121 objset_t *os;
2122 caddr_t outbuf;
2123 size_t bufsize;
2124 zap_cursor_t zc;
2125 zap_attribute_t zap;
2126 uint_t bytes_wanted;
2127 uint64_t offset; /* must be unsigned; checks for < 1 */
2128 uint64_t parent;
2129 int local_eof;
2130 int outcount;
2131 int error;
2132 uint8_t prefetch;
2133 boolean_t check_sysattrs;
2134
2135 ZFS_ENTER(zfsvfs);
2136 ZFS_VERIFY_ZP(zp);
2137
2138 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2139 &parent, sizeof (parent))) != 0) {
2140 ZFS_EXIT(zfsvfs);
2141 return (error);
2142 }
2143
2144 /*
2145 * If we are not given an eof variable,
2146 * use a local one.
2147 */
2148 if (eofp == NULL)
2149 eofp = &local_eof;
2150
2151 /*
2152 * Check for valid iov_len.
2153 */
2154 if (uio->uio_iov->iov_len <= 0) {
2155 ZFS_EXIT(zfsvfs);
2156 return (SET_ERROR(EINVAL));
2157 }
2158
2159 /*
2160 * Quit if directory has been removed (posix)
2161 */
2162 if ((*eofp = zp->z_unlinked) != 0) {
2163 ZFS_EXIT(zfsvfs);
2164 return (0);
2165 }
2166
2167 error = 0;
2168 os = zfsvfs->z_os;
2169 offset = uio->uio_loffset;
2170 prefetch = zp->z_zn_prefetch;
2171
2172 /*
2173 * Initialize the iterator cursor.
2174 */
2175 if (offset <= 3) {
2176 /*
2177 * Start iteration from the beginning of the directory.
2178 */
2179 zap_cursor_init(&zc, os, zp->z_id);
2180 } else {
2181 /*
2182 * The offset is a serialized cursor.
2183 */
2184 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2185 }
2186
2187 /*
2188 * Get space to change directory entries into fs independent format.
2189 */
2190 iovp = uio->uio_iov;
2191 bytes_wanted = iovp->iov_len;
2192 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2193 bufsize = bytes_wanted;
2194 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2195 odp = (struct dirent64 *)outbuf;
2196 } else {
2197 bufsize = bytes_wanted;
2198 outbuf = NULL;
2199 odp = (struct dirent64 *)iovp->iov_base;
2200 }
2201 eodp = (struct edirent *)odp;
2202
2203 /*
2204 * If this VFS supports the system attribute view interface; and
2205 * we're looking at an extended attribute directory; and we care
2206 * about normalization conflicts on this vfs; then we must check
2207 * for normalization conflicts with the sysattr name space.
2208 */
2209 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2210 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2211 (flags & V_RDDIR_ENTFLAGS);
2212
2213 /*
2214 * Transform to file-system independent format
2215 */
2216 outcount = 0;
2217 while (outcount < bytes_wanted) {
2218 ino64_t objnum;
2219 ushort_t reclen;
2220 off64_t *next = NULL;
2221
2222 /*
2223 * Special case `.', `..', and `.zfs'.
2224 */
2225 if (offset == 0) {
2226 (void) strcpy(zap.za_name, ".");
2227 zap.za_normalization_conflict = 0;
2228 objnum = zp->z_id;
2229 } else if (offset == 1) {
2230 (void) strcpy(zap.za_name, "..");
2231 zap.za_normalization_conflict = 0;
2232 objnum = parent;
2233 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2234 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2235 zap.za_normalization_conflict = 0;
2236 objnum = ZFSCTL_INO_ROOT;
2237 } else {
2238 /*
2239 * Grab next entry.
2240 */
2241 if (error = zap_cursor_retrieve(&zc, &zap)) {
2242 if ((*eofp = (error == ENOENT)) != 0)
2243 break;
2244 else
2245 goto update;
2246 }
2247
2248 if (zap.za_integer_length != 8 ||
2249 zap.za_num_integers != 1) {
2250 cmn_err(CE_WARN, "zap_readdir: bad directory "
2251 "entry, obj = %lld, offset = %lld\n",
2252 (u_longlong_t)zp->z_id,
2253 (u_longlong_t)offset);
2254 error = SET_ERROR(ENXIO);
2255 goto update;
2256 }
2257
2258 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2259 /*
2260 * MacOS X can extract the object type here such as:
2261 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2262 */
2263
2264 if (check_sysattrs && !zap.za_normalization_conflict) {
2265 zap.za_normalization_conflict =
2266 xattr_sysattr_casechk(zap.za_name);
2267 }
2268 }
2269
2270 if (flags & V_RDDIR_ACCFILTER) {
2271 /*
2272 * If we have no access at all, don't include
2273 * this entry in the returned information
2274 */
2275 znode_t *ezp;
2276 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2277 goto skip_entry;
2278 if (!zfs_has_access(ezp, cr)) {
2279 VN_RELE(ZTOV(ezp));
2280 goto skip_entry;
2281 }
2282 VN_RELE(ZTOV(ezp));
2283 }
2284
2285 if (flags & V_RDDIR_ENTFLAGS)
2286 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2287 else
2288 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2289
2290 /*
2291 * Will this entry fit in the buffer?
2292 */
2293 if (outcount + reclen > bufsize) {
2294 /*
2295 * Did we manage to fit anything in the buffer?
2296 */
2297 if (!outcount) {
2298 error = SET_ERROR(EINVAL);
2299 goto update;
2300 }
2301 break;
2302 }
2303 if (flags & V_RDDIR_ENTFLAGS) {
2304 /*
2305 * Add extended flag entry:
2306 */
2307 eodp->ed_ino = objnum;
2308 eodp->ed_reclen = reclen;
2309 /* NOTE: ed_off is the offset for the *next* entry */
2310 next = &(eodp->ed_off);
2311 eodp->ed_eflags = zap.za_normalization_conflict ?
2312 ED_CASE_CONFLICT : 0;
2313 (void) strncpy(eodp->ed_name, zap.za_name,
2314 EDIRENT_NAMELEN(reclen));
2315 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2316 } else {
2317 /*
2318 * Add normal entry:
2319 */
2320 odp->d_ino = objnum;
2321 odp->d_reclen = reclen;
2322 /* NOTE: d_off is the offset for the *next* entry */
2323 next = &(odp->d_off);
2324 (void) strncpy(odp->d_name, zap.za_name,
2325 DIRENT64_NAMELEN(reclen));
2326 odp = (dirent64_t *)((intptr_t)odp + reclen);
2327 }
2328 outcount += reclen;
2329
2330 ASSERT(outcount <= bufsize);
2331
2332 /* Prefetch znode */
2333 if (prefetch)
2334 dmu_prefetch(os, objnum, 0, 0);
2335
2336 skip_entry:
2337 /*
2338 * Move to the next entry, fill in the previous offset.
2339 */
2340 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2341 zap_cursor_advance(&zc);
2342 offset = zap_cursor_serialize(&zc);
2343 } else {
2344 offset += 1;
2345 }
2346 if (next)
2347 *next = offset;
2348 }
2349 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2350
2351 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2352 iovp->iov_base += outcount;
2353 iovp->iov_len -= outcount;
2354 uio->uio_resid -= outcount;
2355 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2356 /*
2357 * Reset the pointer.
2358 */
2359 offset = uio->uio_loffset;
2360 }
2361
2362 update:
2363 zap_cursor_fini(&zc);
2364 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2365 kmem_free(outbuf, bufsize);
2366
2367 if (error == ENOENT)
2368 error = 0;
2369
2370 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2371
2372 uio->uio_loffset = offset;
2373 ZFS_EXIT(zfsvfs);
2374 return (error);
2375 }
2376
2377 ulong_t zfs_fsync_sync_cnt = 4;
2378
2379 static int
2380 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2381 {
2382 znode_t *zp = VTOZ(vp);
2383 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2384
2385 /*
2386 * Regardless of whether this is required for standards conformance,
2387 * this is the logical behavior when fsync() is called on a file with
2388 * dirty pages. We use B_ASYNC since the ZIL transactions are already
2389 * going to be pushed out as part of the zil_commit().
2390 */
2391 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2392 (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2393 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2394
2395 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2396
2397 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2398 ZFS_ENTER(zfsvfs);
2399 ZFS_VERIFY_ZP(zp);
2400 zil_commit(zfsvfs->z_log, zp->z_id);
2401 ZFS_EXIT(zfsvfs);
2402 }
2403 return (0);
2404 }
2405
2406
2407 /*
2408 * Get the requested file attributes and place them in the provided
2409 * vattr structure.
2410 *
2411 * IN: vp - vnode of file.
2412 * vap - va_mask identifies requested attributes.
2413 * If AT_XVATTR set, then optional attrs are requested
2414 * flags - ATTR_NOACLCHECK (CIFS server context)
2415 * cr - credentials of caller.
2416 * ct - caller context
2417 *
2418 * OUT: vap - attribute values.
2419 *
2420 * RETURN: 0 (always succeeds).
2421 */
2422 /* ARGSUSED */
2423 static int
2424 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2425 caller_context_t *ct)
2426 {
2427 znode_t *zp = VTOZ(vp);
2428 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2429 int error = 0;
2430 uint64_t links;
2431 uint64_t mtime[2], ctime[2];
2432 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2433 xoptattr_t *xoap = NULL;
2434 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2435 sa_bulk_attr_t bulk[2];
2436 int count = 0;
2437
2438 ZFS_ENTER(zfsvfs);
2439 ZFS_VERIFY_ZP(zp);
2440
2441 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2442
2443 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2444 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2445
2446 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2447 ZFS_EXIT(zfsvfs);
2448 return (error);
2449 }
2450
2451 /*
2452 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2453 * Also, if we are the owner don't bother, since owner should
2454 * always be allowed to read basic attributes of file.
2455 */
2456 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2457 (vap->va_uid != crgetuid(cr))) {
2458 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2459 skipaclchk, cr)) {
2460 ZFS_EXIT(zfsvfs);
2461 return (error);
2462 }
2463 }
2464
2465 /*
2466 * Return all attributes. It's cheaper to provide the answer
2467 * than to determine whether we were asked the question.
2468 */
2469
2470 mutex_enter(&zp->z_lock);
2471 vap->va_type = vp->v_type;
2472 vap->va_mode = zp->z_mode & MODEMASK;
2473 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2474 vap->va_nodeid = zp->z_id;
2475 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2476 links = zp->z_links + 1;
2477 else
2478 links = zp->z_links;
2479 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */
2480 vap->va_size = zp->z_size;
2481 vap->va_rdev = vp->v_rdev;
2482 vap->va_seq = zp->z_seq;
2483
2484 /*
2485 * Add in any requested optional attributes and the create time.
2486 * Also set the corresponding bits in the returned attribute bitmap.
2487 */
2488 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2489 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2490 xoap->xoa_archive =
2491 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2492 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2493 }
2494
2495 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2496 xoap->xoa_readonly =
2497 ((zp->z_pflags & ZFS_READONLY) != 0);
2498 XVA_SET_RTN(xvap, XAT_READONLY);
2499 }
2500
2501 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2502 xoap->xoa_system =
2503 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2504 XVA_SET_RTN(xvap, XAT_SYSTEM);
2505 }
2506
2507 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2508 xoap->xoa_hidden =
2509 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2510 XVA_SET_RTN(xvap, XAT_HIDDEN);
2511 }
2512
2513 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2514 xoap->xoa_nounlink =
2515 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2516 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2517 }
2518
2519 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2520 xoap->xoa_immutable =
2521 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2522 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2523 }
2524
2525 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2526 xoap->xoa_appendonly =
2527 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2528 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2529 }
2530
2531 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2532 xoap->xoa_nodump =
2533 ((zp->z_pflags & ZFS_NODUMP) != 0);
2534 XVA_SET_RTN(xvap, XAT_NODUMP);
2535 }
2536
2537 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2538 xoap->xoa_opaque =
2539 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2540 XVA_SET_RTN(xvap, XAT_OPAQUE);
2541 }
2542
2543 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2544 xoap->xoa_av_quarantined =
2545 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2546 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2547 }
2548
2549 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2550 xoap->xoa_av_modified =
2551 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2552 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2553 }
2554
2555 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2556 vp->v_type == VREG) {
2557 zfs_sa_get_scanstamp(zp, xvap);
2558 }
2559
2560 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2561 uint64_t times[2];
2562
2563 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2564 times, sizeof (times));
2565 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2566 XVA_SET_RTN(xvap, XAT_CREATETIME);
2567 }
2568
2569 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2570 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2571 XVA_SET_RTN(xvap, XAT_REPARSE);
2572 }
2573 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2574 xoap->xoa_generation = zp->z_gen;
2575 XVA_SET_RTN(xvap, XAT_GEN);
2576 }
2577
2578 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2579 xoap->xoa_offline =
2580 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2581 XVA_SET_RTN(xvap, XAT_OFFLINE);
2582 }
2583
2584 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2585 xoap->xoa_sparse =
2586 ((zp->z_pflags & ZFS_SPARSE) != 0);
2587 XVA_SET_RTN(xvap, XAT_SPARSE);
2588 }
2589 }
2590
2591 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2592 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2593 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2594
2595 mutex_exit(&zp->z_lock);
2596
2597 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2598
2599 if (zp->z_blksz == 0) {
2600 /*
2601 * Block size hasn't been set; suggest maximal I/O transfers.
2602 */
2603 vap->va_blksize = zfsvfs->z_max_blksz;
2604 }
2605
2606 ZFS_EXIT(zfsvfs);
2607 return (0);
2608 }
2609
2610 /*
2611 * Set the file attributes to the values contained in the
2612 * vattr structure.
2613 *
2614 * IN: vp - vnode of file to be modified.
2615 * vap - new attribute values.
2616 * If AT_XVATTR set, then optional attrs are being set
2617 * flags - ATTR_UTIME set if non-default time values provided.
2618 * - ATTR_NOACLCHECK (CIFS context only).
2619 * cr - credentials of caller.
2620 * ct - caller context
2621 *
2622 * RETURN: 0 on success, error code on failure.
2623 *
2624 * Timestamps:
2625 * vp - ctime updated, mtime updated if size changed.
2626 */
2627 /* ARGSUSED */
2628 static int
2629 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2630 caller_context_t *ct)
2631 {
2632 znode_t *zp = VTOZ(vp);
2633 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2634 zilog_t *zilog;
2635 dmu_tx_t *tx;
2636 vattr_t oldva;
2637 xvattr_t tmpxvattr;
2638 uint_t mask = vap->va_mask;
2639 uint_t saved_mask = 0;
2640 int trim_mask = 0;
2641 uint64_t new_mode;
2642 uint64_t new_uid, new_gid;
2643 uint64_t xattr_obj;
2644 uint64_t mtime[2], ctime[2];
2645 znode_t *attrzp;
2646 int need_policy = FALSE;
2647 int err, err2;
2648 zfs_fuid_info_t *fuidp = NULL;
2649 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2650 xoptattr_t *xoap;
2651 zfs_acl_t *aclp;
2652 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2653 boolean_t fuid_dirtied = B_FALSE;
2654 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2655 int count = 0, xattr_count = 0;
2656
2657 if (mask == 0)
2658 return (0);
2659
2660 if (mask & AT_NOSET)
2661 return (SET_ERROR(EINVAL));
2662
2663 ZFS_ENTER(zfsvfs);
2664 ZFS_VERIFY_ZP(zp);
2665
2666 zilog = zfsvfs->z_log;
2667
2668 /*
2669 * Make sure that if we have ephemeral uid/gid or xvattr specified
2670 * that file system is at proper version level
2671 */
2672
2673 if (zfsvfs->z_use_fuids == B_FALSE &&
2674 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2675 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2676 (mask & AT_XVATTR))) {
2677 ZFS_EXIT(zfsvfs);
2678 return (SET_ERROR(EINVAL));
2679 }
2680
2681 if (mask & AT_SIZE && vp->v_type == VDIR) {
2682 ZFS_EXIT(zfsvfs);
2683 return (SET_ERROR(EISDIR));
2684 }
2685
2686 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2687 ZFS_EXIT(zfsvfs);
2688 return (SET_ERROR(EINVAL));
2689 }
2690
2691 /*
2692 * If this is an xvattr_t, then get a pointer to the structure of
2693 * optional attributes. If this is NULL, then we have a vattr_t.
2694 */
2695 xoap = xva_getxoptattr(xvap);
2696
2697 xva_init(&tmpxvattr);
2698
2699 /*
2700 * Immutable files can only alter immutable bit and atime
2701 */
2702 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2703 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2704 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2705 ZFS_EXIT(zfsvfs);
2706 return (SET_ERROR(EPERM));
2707 }
2708
2709 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2710 ZFS_EXIT(zfsvfs);
2711 return (SET_ERROR(EPERM));
2712 }
2713
2714 /*
2715 * Verify timestamps doesn't overflow 32 bits.
2716 * ZFS can handle large timestamps, but 32bit syscalls can't
2717 * handle times greater than 2039. This check should be removed
2718 * once large timestamps are fully supported.
2719 */
2720 if (mask & (AT_ATIME | AT_MTIME)) {
2721 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2722 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2723 ZFS_EXIT(zfsvfs);
2724 return (SET_ERROR(EOVERFLOW));
2725 }
2726 }
2727
2728 top:
2729 attrzp = NULL;
2730 aclp = NULL;
2731
2732 /* Can this be moved to before the top label? */
2733 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2734 ZFS_EXIT(zfsvfs);
2735 return (SET_ERROR(EROFS));
2736 }
2737
2738 /*
2739 * First validate permissions
2740 */
2741
2742 if (mask & AT_SIZE) {
2743 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2744 if (err) {
2745 ZFS_EXIT(zfsvfs);
2746 return (err);
2747 }
2748 /*
2749 * XXX - Note, we are not providing any open
2750 * mode flags here (like FNDELAY), so we may
2751 * block if there are locks present... this
2752 * should be addressed in openat().
2753 */
2754 /* XXX - would it be OK to generate a log record here? */
2755 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2756 if (err) {
2757 ZFS_EXIT(zfsvfs);
2758 return (err);
2759 }
2760
2761 if (vap->va_size == 0)
2762 vnevent_truncate(ZTOV(zp), ct);
2763 }
2764
2765 if (mask & (AT_ATIME|AT_MTIME) ||
2766 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2767 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2768 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2769 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2770 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2771 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2772 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2773 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2774 skipaclchk, cr);
2775 }
2776
2777 if (mask & (AT_UID|AT_GID)) {
2778 int idmask = (mask & (AT_UID|AT_GID));
2779 int take_owner;
2780 int take_group;
2781
2782 /*
2783 * NOTE: even if a new mode is being set,
2784 * we may clear S_ISUID/S_ISGID bits.
2785 */
2786
2787 if (!(mask & AT_MODE))
2788 vap->va_mode = zp->z_mode;
2789
2790 /*
2791 * Take ownership or chgrp to group we are a member of
2792 */
2793
2794 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2795 take_group = (mask & AT_GID) &&
2796 zfs_groupmember(zfsvfs, vap->va_gid, cr);
2797
2798 /*
2799 * If both AT_UID and AT_GID are set then take_owner and
2800 * take_group must both be set in order to allow taking
2801 * ownership.
2802 *
2803 * Otherwise, send the check through secpolicy_vnode_setattr()
2804 *
2805 */
2806
2807 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2808 ((idmask == AT_UID) && take_owner) ||
2809 ((idmask == AT_GID) && take_group)) {
2810 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2811 skipaclchk, cr) == 0) {
2812 /*
2813 * Remove setuid/setgid for non-privileged users
2814 */
2815 secpolicy_setid_clear(vap, cr);
2816 trim_mask = (mask & (AT_UID|AT_GID));
2817 } else {
2818 need_policy = TRUE;
2819 }
2820 } else {
2821 need_policy = TRUE;
2822 }
2823 }
2824
2825 mutex_enter(&zp->z_lock);
2826 oldva.va_mode = zp->z_mode;
2827 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2828 if (mask & AT_XVATTR) {
2829 /*
2830 * Update xvattr mask to include only those attributes
2831 * that are actually changing.
2832 *
2833 * the bits will be restored prior to actually setting
2834 * the attributes so the caller thinks they were set.
2835 */
2836 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2837 if (xoap->xoa_appendonly !=
2838 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2839 need_policy = TRUE;
2840 } else {
2841 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2842 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2843 }
2844 }
2845
2846 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2847 if (xoap->xoa_nounlink !=
2848 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2849 need_policy = TRUE;
2850 } else {
2851 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2852 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2853 }
2854 }
2855
2856 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2857 if (xoap->xoa_immutable !=
2858 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2859 need_policy = TRUE;
2860 } else {
2861 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2862 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2863 }
2864 }
2865
2866 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2867 if (xoap->xoa_nodump !=
2868 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2869 need_policy = TRUE;
2870 } else {
2871 XVA_CLR_REQ(xvap, XAT_NODUMP);
2872 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2873 }
2874 }
2875
2876 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2877 if (xoap->xoa_av_modified !=
2878 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2879 need_policy = TRUE;
2880 } else {
2881 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2882 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2883 }
2884 }
2885
2886 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2887 if ((vp->v_type != VREG &&
2888 xoap->xoa_av_quarantined) ||
2889 xoap->xoa_av_quarantined !=
2890 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2891 need_policy = TRUE;
2892 } else {
2893 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2894 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2895 }
2896 }
2897
2898 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2899 mutex_exit(&zp->z_lock);
2900 ZFS_EXIT(zfsvfs);
2901 return (SET_ERROR(EPERM));
2902 }
2903
2904 if (need_policy == FALSE &&
2905 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2906 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2907 need_policy = TRUE;
2908 }
2909 }
2910
2911 mutex_exit(&zp->z_lock);
2912
2913 if (mask & AT_MODE) {
2914 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2915 err = secpolicy_setid_setsticky_clear(vp, vap,
2916 &oldva, cr);
2917 if (err) {
2918 ZFS_EXIT(zfsvfs);
2919 return (err);
2920 }
2921 trim_mask |= AT_MODE;
2922 } else {
2923 need_policy = TRUE;
2924 }
2925 }
2926
2927 if (need_policy) {
2928 /*
2929 * If trim_mask is set then take ownership
2930 * has been granted or write_acl is present and user
2931 * has the ability to modify mode. In that case remove
2932 * UID|GID and or MODE from mask so that
2933 * secpolicy_vnode_setattr() doesn't revoke it.
2934 */
2935
2936 if (trim_mask) {
2937 saved_mask = vap->va_mask;
2938 vap->va_mask &= ~trim_mask;
2939 }
2940 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2941 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2942 if (err) {
2943 ZFS_EXIT(zfsvfs);
2944 return (err);
2945 }
2946
2947 if (trim_mask)
2948 vap->va_mask |= saved_mask;
2949 }
2950
2951 /*
2952 * secpolicy_vnode_setattr, or take ownership may have
2953 * changed va_mask
2954 */
2955 mask = vap->va_mask;
2956
2957 if ((mask & (AT_UID | AT_GID))) {
2958 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2959 &xattr_obj, sizeof (xattr_obj));
2960
2961 if (err == 0 && xattr_obj) {
2962 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
2963 if (err)
2964 goto out2;
2965 }
2966 if (mask & AT_UID) {
2967 new_uid = zfs_fuid_create(zfsvfs,
2968 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2969 if (new_uid != zp->z_uid &&
2970 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
2971 if (attrzp)
2972 VN_RELE(ZTOV(attrzp));
2973 err = SET_ERROR(EDQUOT);
2974 goto out2;
2975 }
2976 }
2977
2978 if (mask & AT_GID) {
2979 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
2980 cr, ZFS_GROUP, &fuidp);
2981 if (new_gid != zp->z_gid &&
2982 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
2983 if (attrzp)
2984 VN_RELE(ZTOV(attrzp));
2985 err = SET_ERROR(EDQUOT);
2986 goto out2;
2987 }
2988 }
2989 }
2990 tx = dmu_tx_create(zfsvfs->z_os);
2991
2992 if (mask & AT_MODE) {
2993 uint64_t pmode = zp->z_mode;
2994 uint64_t acl_obj;
2995 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2996
2997 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
2998 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
2999 err = SET_ERROR(EPERM);
3000 goto out;
3001 }
3002
3003 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3004 goto out;
3005
3006 mutex_enter(&zp->z_lock);
3007 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3008 /*
3009 * Are we upgrading ACL from old V0 format
3010 * to V1 format?
3011 */
3012 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3013 zfs_znode_acl_version(zp) ==
3014 ZFS_ACL_VERSION_INITIAL) {
3015 dmu_tx_hold_free(tx, acl_obj, 0,
3016 DMU_OBJECT_END);
3017 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3018 0, aclp->z_acl_bytes);
3019 } else {
3020 dmu_tx_hold_write(tx, acl_obj, 0,
3021 aclp->z_acl_bytes);
3022 }
3023 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3024 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3025 0, aclp->z_acl_bytes);
3026 }
3027 mutex_exit(&zp->z_lock);
3028 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3029 } else {
3030 if ((mask & AT_XVATTR) &&
3031 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3032 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3033 else
3034 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3035 }
3036
3037 if (attrzp) {
3038 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3039 }
3040
3041 fuid_dirtied = zfsvfs->z_fuid_dirty;
3042 if (fuid_dirtied)
3043 zfs_fuid_txhold(zfsvfs, tx);
3044
3045 zfs_sa_upgrade_txholds(tx, zp);
3046
3047 err = dmu_tx_assign(tx, TXG_WAIT);
3048 if (err)
3049 goto out;
3050
3051 count = 0;
3052 /*
3053 * Set each attribute requested.
3054 * We group settings according to the locks they need to acquire.
3055 *
3056 * Note: you cannot set ctime directly, although it will be
3057 * updated as a side-effect of calling this function.
3058 */
3059
3060
3061 if (mask & (AT_UID|AT_GID|AT_MODE))
3062 mutex_enter(&zp->z_acl_lock);
3063 mutex_enter(&zp->z_lock);
3064
3065 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3066 &zp->z_pflags, sizeof (zp->z_pflags));
3067
3068 if (attrzp) {
3069 if (mask & (AT_UID|AT_GID|AT_MODE))
3070 mutex_enter(&attrzp->z_acl_lock);
3071 mutex_enter(&attrzp->z_lock);
3072 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3073 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3074 sizeof (attrzp->z_pflags));
3075 }
3076
3077 if (mask & (AT_UID|AT_GID)) {
3078
3079 if (mask & AT_UID) {
3080 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3081 &new_uid, sizeof (new_uid));
3082 zp->z_uid = new_uid;
3083 if (attrzp) {
3084 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3085 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3086 sizeof (new_uid));
3087 attrzp->z_uid = new_uid;
3088 }
3089 }
3090
3091 if (mask & AT_GID) {
3092 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3093 NULL, &new_gid, sizeof (new_gid));
3094 zp->z_gid = new_gid;
3095 if (attrzp) {
3096 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3097 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3098 sizeof (new_gid));
3099 attrzp->z_gid = new_gid;
3100 }
3101 }
3102 if (!(mask & AT_MODE)) {
3103 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3104 NULL, &new_mode, sizeof (new_mode));
3105 new_mode = zp->z_mode;
3106 }
3107 err = zfs_acl_chown_setattr(zp);
3108 ASSERT(err == 0);
3109 if (attrzp) {
3110 err = zfs_acl_chown_setattr(attrzp);
3111 ASSERT(err == 0);
3112 }
3113 }
3114
3115 if (mask & AT_MODE) {
3116 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3117 &new_mode, sizeof (new_mode));
3118 zp->z_mode = new_mode;
3119 ASSERT3U((uintptr_t)aclp, !=, NULL);
3120 err = zfs_aclset_common(zp, aclp, cr, tx);
3121 ASSERT0(err);
3122 if (zp->z_acl_cached)
3123 zfs_acl_free(zp->z_acl_cached);
3124 zp->z_acl_cached = aclp;
3125 aclp = NULL;
3126 }
3127
3128
3129 if (mask & AT_ATIME) {
3130 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3131 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3132 &zp->z_atime, sizeof (zp->z_atime));
3133 }
3134
3135 if (mask & AT_MTIME) {
3136 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3137 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3138 mtime, sizeof (mtime));
3139 }
3140
3141 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3142 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3143 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3144 NULL, mtime, sizeof (mtime));
3145 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3146 &ctime, sizeof (ctime));
3147 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3148 B_TRUE);
3149 } else if (mask != 0) {
3150 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3151 &ctime, sizeof (ctime));
3152 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3153 B_TRUE);
3154 if (attrzp) {
3155 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3156 SA_ZPL_CTIME(zfsvfs), NULL,
3157 &ctime, sizeof (ctime));
3158 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3159 mtime, ctime, B_TRUE);
3160 }
3161 }
3162 /*
3163 * Do this after setting timestamps to prevent timestamp
3164 * update from toggling bit
3165 */
3166
3167 if (xoap && (mask & AT_XVATTR)) {
3168
3169 /*
3170 * restore trimmed off masks
3171 * so that return masks can be set for caller.
3172 */
3173
3174 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3175 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3176 }
3177 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3178 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3179 }
3180 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3181 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3182 }
3183 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3184 XVA_SET_REQ(xvap, XAT_NODUMP);
3185 }
3186 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3187 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3188 }
3189 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3190 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3191 }
3192
3193 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3194 ASSERT(vp->v_type == VREG);
3195
3196 zfs_xvattr_set(zp, xvap, tx);
3197 }
3198
3199 if (fuid_dirtied)
3200 zfs_fuid_sync(zfsvfs, tx);
3201
3202 if (mask != 0)
3203 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3204
3205 mutex_exit(&zp->z_lock);
3206 if (mask & (AT_UID|AT_GID|AT_MODE))
3207 mutex_exit(&zp->z_acl_lock);
3208
3209 if (attrzp) {
3210 if (mask & (AT_UID|AT_GID|AT_MODE))
3211 mutex_exit(&attrzp->z_acl_lock);
3212 mutex_exit(&attrzp->z_lock);
3213 }
3214 out:
3215 if (err == 0 && attrzp) {
3216 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3217 xattr_count, tx);
3218 ASSERT(err2 == 0);
3219 }
3220
3221 if (attrzp)
3222 VN_RELE(ZTOV(attrzp));
3223
3224 if (aclp)
3225 zfs_acl_free(aclp);
3226
3227 if (fuidp) {
3228 zfs_fuid_info_free(fuidp);
3229 fuidp = NULL;
3230 }
3231
3232 if (err) {
3233 dmu_tx_abort(tx);
3234 if (err == ERESTART)
3235 goto top;
3236 } else {
3237 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3238 dmu_tx_commit(tx);
3239 }
3240
3241 out2:
3242 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3243 zil_commit(zilog, 0);
3244
3245 ZFS_EXIT(zfsvfs);
3246 return (err);
3247 }
3248
3249 typedef struct zfs_zlock {
3250 krwlock_t *zl_rwlock; /* lock we acquired */
3251 znode_t *zl_znode; /* znode we held */
3252 struct zfs_zlock *zl_next; /* next in list */
3253 } zfs_zlock_t;
3254
3255 /*
3256 * Drop locks and release vnodes that were held by zfs_rename_lock().
3257 */
3258 static void
3259 zfs_rename_unlock(zfs_zlock_t **zlpp)
3260 {
3261 zfs_zlock_t *zl;
3262
3263 while ((zl = *zlpp) != NULL) {
3264 if (zl->zl_znode != NULL)
3265 VN_RELE(ZTOV(zl->zl_znode));
3266 rw_exit(zl->zl_rwlock);
3267 *zlpp = zl->zl_next;
3268 kmem_free(zl, sizeof (*zl));
3269 }
3270 }
3271
3272 /*
3273 * Search back through the directory tree, using the ".." entries.
3274 * Lock each directory in the chain to prevent concurrent renames.
3275 * Fail any attempt to move a directory into one of its own descendants.
3276 * XXX - z_parent_lock can overlap with map or grow locks
3277 */
3278 static int
3279 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3280 {
3281 zfs_zlock_t *zl;
3282 znode_t *zp = tdzp;
3283 uint64_t rootid = zp->z_zfsvfs->z_root;
3284 uint64_t oidp = zp->z_id;
3285 krwlock_t *rwlp = &szp->z_parent_lock;
3286 krw_t rw = RW_WRITER;
3287
3288 /*
3289 * First pass write-locks szp and compares to zp->z_id.
3290 * Later passes read-lock zp and compare to zp->z_parent.
3291 */
3292 do {
3293 if (!rw_tryenter(rwlp, rw)) {
3294 /*
3295 * Another thread is renaming in this path.
3296 * Note that if we are a WRITER, we don't have any
3297 * parent_locks held yet.
3298 */
3299 if (rw == RW_READER && zp->z_id > szp->z_id) {
3300 /*
3301 * Drop our locks and restart
3302 */
3303 zfs_rename_unlock(&zl);
3304 *zlpp = NULL;
3305 zp = tdzp;
3306 oidp = zp->z_id;
3307 rwlp = &szp->z_parent_lock;
3308 rw = RW_WRITER;
3309 continue;
3310 } else {
3311 /*
3312 * Wait for other thread to drop its locks
3313 */
3314 rw_enter(rwlp, rw);
3315 }
3316 }
3317
3318 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3319 zl->zl_rwlock = rwlp;
3320 zl->zl_znode = NULL;
3321 zl->zl_next = *zlpp;
3322 *zlpp = zl;
3323
3324 if (oidp == szp->z_id) /* We're a descendant of szp */
3325 return (SET_ERROR(EINVAL));
3326
3327 if (oidp == rootid) /* We've hit the top */
3328 return (0);
3329
3330 if (rw == RW_READER) { /* i.e. not the first pass */
3331 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3332 if (error)
3333 return (error);
3334 zl->zl_znode = zp;
3335 }
3336 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3337 &oidp, sizeof (oidp));
3338 rwlp = &zp->z_parent_lock;
3339 rw = RW_READER;
3340
3341 } while (zp->z_id != sdzp->z_id);
3342
3343 return (0);
3344 }
3345
3346 /*
3347 * Move an entry from the provided source directory to the target
3348 * directory. Change the entry name as indicated.
3349 *
3350 * IN: sdvp - Source directory containing the "old entry".
3351 * snm - Old entry name.
3352 * tdvp - Target directory to contain the "new entry".
3353 * tnm - New entry name.
3354 * cr - credentials of caller.
3355 * ct - caller context
3356 * flags - case flags
3357 *
3358 * RETURN: 0 on success, error code on failure.
3359 *
3360 * Timestamps:
3361 * sdvp,tdvp - ctime|mtime updated
3362 */
3363 /*ARGSUSED*/
3364 static int
3365 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3366 caller_context_t *ct, int flags)
3367 {
3368 znode_t *tdzp, *szp, *tzp;
3369 znode_t *sdzp = VTOZ(sdvp);
3370 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3371 zilog_t *zilog;
3372 vnode_t *realvp;
3373 zfs_dirlock_t *sdl, *tdl;
3374 dmu_tx_t *tx;
3375 zfs_zlock_t *zl;
3376 int cmp, serr, terr;
3377 int error = 0;
3378 int zflg = 0;
3379 boolean_t waited = B_FALSE;
3380
3381 ZFS_ENTER(zfsvfs);
3382 ZFS_VERIFY_ZP(sdzp);
3383 zilog = zfsvfs->z_log;
3384
3385 /*
3386 * Make sure we have the real vp for the target directory.
3387 */
3388 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3389 tdvp = realvp;
3390
3391 tdzp = VTOZ(tdvp);
3392 ZFS_VERIFY_ZP(tdzp);
3393
3394 /*
3395 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3396 * ctldir appear to have the same v_vfsp.
3397 */
3398 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3399 ZFS_EXIT(zfsvfs);
3400 return (SET_ERROR(EXDEV));
3401 }
3402
3403 if (zfsvfs->z_utf8 && u8_validate(tnm,
3404 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3405 ZFS_EXIT(zfsvfs);
3406 return (SET_ERROR(EILSEQ));
3407 }
3408
3409 if (flags & FIGNORECASE)
3410 zflg |= ZCILOOK;
3411
3412 top:
3413 szp = NULL;
3414 tzp = NULL;
3415 zl = NULL;
3416
3417 /*
3418 * This is to prevent the creation of links into attribute space
3419 * by renaming a linked file into/outof an attribute directory.
3420 * See the comment in zfs_link() for why this is considered bad.
3421 */
3422 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3423 ZFS_EXIT(zfsvfs);
3424 return (SET_ERROR(EINVAL));
3425 }
3426
3427 /*
3428 * Lock source and target directory entries. To prevent deadlock,
3429 * a lock ordering must be defined. We lock the directory with
3430 * the smallest object id first, or if it's a tie, the one with
3431 * the lexically first name.
3432 */
3433 if (sdzp->z_id < tdzp->z_id) {
3434 cmp = -1;
3435 } else if (sdzp->z_id > tdzp->z_id) {
3436 cmp = 1;
3437 } else {
3438 /*
3439 * First compare the two name arguments without
3440 * considering any case folding.
3441 */
3442 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3443
3444 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3445 ASSERT(error == 0 || !zfsvfs->z_utf8);
3446 if (cmp == 0) {
3447 /*
3448 * POSIX: "If the old argument and the new argument
3449 * both refer to links to the same existing file,
3450 * the rename() function shall return successfully
3451 * and perform no other action."
3452 */
3453 ZFS_EXIT(zfsvfs);
3454 return (0);
3455 }
3456 /*
3457 * If the file system is case-folding, then we may
3458 * have some more checking to do. A case-folding file
3459 * system is either supporting mixed case sensitivity
3460 * access or is completely case-insensitive. Note
3461 * that the file system is always case preserving.
3462 *
3463 * In mixed sensitivity mode case sensitive behavior
3464 * is the default. FIGNORECASE must be used to
3465 * explicitly request case insensitive behavior.
3466 *
3467 * If the source and target names provided differ only
3468 * by case (e.g., a request to rename 'tim' to 'Tim'),
3469 * we will treat this as a special case in the
3470 * case-insensitive mode: as long as the source name
3471 * is an exact match, we will allow this to proceed as
3472 * a name-change request.
3473 */
3474 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3475 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3476 flags & FIGNORECASE)) &&
3477 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3478 &error) == 0) {
3479 /*
3480 * case preserving rename request, require exact
3481 * name matches
3482 */
3483 zflg |= ZCIEXACT;
3484 zflg &= ~ZCILOOK;
3485 }
3486 }
3487
3488 /*
3489 * If the source and destination directories are the same, we should
3490 * grab the z_name_lock of that directory only once.
3491 */
3492 if (sdzp == tdzp) {
3493 zflg |= ZHAVELOCK;
3494 rw_enter(&sdzp->z_name_lock, RW_READER);
3495 }
3496
3497 if (cmp < 0) {
3498 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3499 ZEXISTS | zflg, NULL, NULL);
3500 terr = zfs_dirent_lock(&tdl,
3501 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3502 } else {
3503 terr = zfs_dirent_lock(&tdl,
3504 tdzp, tnm, &tzp, zflg, NULL, NULL);
3505 serr = zfs_dirent_lock(&sdl,
3506 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3507 NULL, NULL);
3508 }
3509
3510 if (serr) {
3511 /*
3512 * Source entry invalid or not there.
3513 */
3514 if (!terr) {
3515 zfs_dirent_unlock(tdl);
3516 if (tzp)
3517 VN_RELE(ZTOV(tzp));
3518 }
3519
3520 if (sdzp == tdzp)
3521 rw_exit(&sdzp->z_name_lock);
3522
3523 if (strcmp(snm, "..") == 0)
3524 serr = SET_ERROR(EINVAL);
3525 ZFS_EXIT(zfsvfs);
3526 return (serr);
3527 }
3528 if (terr) {
3529 zfs_dirent_unlock(sdl);
3530 VN_RELE(ZTOV(szp));
3531
3532 if (sdzp == tdzp)
3533 rw_exit(&sdzp->z_name_lock);
3534
3535 if (strcmp(tnm, "..") == 0)
3536 terr = SET_ERROR(EINVAL);
3537 ZFS_EXIT(zfsvfs);
3538 return (terr);
3539 }
3540
3541 /*
3542 * Must have write access at the source to remove the old entry
3543 * and write access at the target to create the new entry.
3544 * Note that if target and source are the same, this can be
3545 * done in a single check.
3546 */
3547
3548 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3549 goto out;
3550
3551 if (ZTOV(szp)->v_type == VDIR) {
3552 /*
3553 * Check to make sure rename is valid.
3554 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3555 */
3556 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3557 goto out;
3558 }
3559
3560 /*
3561 * Does target exist?
3562 */
3563 if (tzp) {
3564 /*
3565 * Source and target must be the same type.
3566 */
3567 if (ZTOV(szp)->v_type == VDIR) {
3568 if (ZTOV(tzp)->v_type != VDIR) {
3569 error = SET_ERROR(ENOTDIR);
3570 goto out;
3571 }
3572 } else {
3573 if (ZTOV(tzp)->v_type == VDIR) {
3574 error = SET_ERROR(EISDIR);
3575 goto out;
3576 }
3577 }
3578 /*
3579 * POSIX dictates that when the source and target
3580 * entries refer to the same file object, rename
3581 * must do nothing and exit without error.
3582 */
3583 if (szp->z_id == tzp->z_id) {
3584 error = 0;
3585 goto out;
3586 }
3587 }
3588
3589 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3590 if (tzp)
3591 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3592
3593 /*
3594 * notify the target directory if it is not the same
3595 * as source directory.
3596 */
3597 if (tdvp != sdvp) {
3598 vnevent_rename_dest_dir(tdvp, ct);
3599 }
3600
3601 tx = dmu_tx_create(zfsvfs->z_os);
3602 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3603 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3604 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3605 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3606 if (sdzp != tdzp) {
3607 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3608 zfs_sa_upgrade_txholds(tx, tdzp);
3609 }
3610 if (tzp) {
3611 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3612 zfs_sa_upgrade_txholds(tx, tzp);
3613 }
3614
3615 zfs_sa_upgrade_txholds(tx, szp);
3616 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3617 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3618 if (error) {
3619 if (zl != NULL)
3620 zfs_rename_unlock(&zl);
3621 zfs_dirent_unlock(sdl);
3622 zfs_dirent_unlock(tdl);
3623
3624 if (sdzp == tdzp)
3625 rw_exit(&sdzp->z_name_lock);
3626
3627 VN_RELE(ZTOV(szp));
3628 if (tzp)
3629 VN_RELE(ZTOV(tzp));
3630 if (error == ERESTART) {
3631 waited = B_TRUE;
3632 dmu_tx_wait(tx);
3633 dmu_tx_abort(tx);
3634 goto top;
3635 }
3636 dmu_tx_abort(tx);
3637 ZFS_EXIT(zfsvfs);
3638 return (error);
3639 }
3640
3641 if (tzp) /* Attempt to remove the existing target */
3642 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3643
3644 if (error == 0) {
3645 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3646 if (error == 0) {
3647 szp->z_pflags |= ZFS_AV_MODIFIED;
3648
3649 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3650 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3651 ASSERT0(error);
3652
3653 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3654 if (error == 0) {
3655 zfs_log_rename(zilog, tx, TX_RENAME |
3656 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3657 sdl->dl_name, tdzp, tdl->dl_name, szp);
3658
3659 /*
3660 * Update path information for the target vnode
3661 */
3662 vn_renamepath(tdvp, ZTOV(szp), tnm,
3663 strlen(tnm));
3664 } else {
3665 /*
3666 * At this point, we have successfully created
3667 * the target name, but have failed to remove
3668 * the source name. Since the create was done
3669 * with the ZRENAMING flag, there are
3670 * complications; for one, the link count is
3671 * wrong. The easiest way to deal with this
3672 * is to remove the newly created target, and
3673 * return the original error. This must
3674 * succeed; fortunately, it is very unlikely to
3675 * fail, since we just created it.
3676 */
3677 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3678 ZRENAMING, NULL), ==, 0);
3679 }
3680 }
3681 }
3682
3683 dmu_tx_commit(tx);
3684 out:
3685 if (zl != NULL)
3686 zfs_rename_unlock(&zl);
3687
3688 zfs_dirent_unlock(sdl);
3689 zfs_dirent_unlock(tdl);
3690
3691 if (sdzp == tdzp)
3692 rw_exit(&sdzp->z_name_lock);
3693
3694
3695 VN_RELE(ZTOV(szp));
3696 if (tzp)
3697 VN_RELE(ZTOV(tzp));
3698
3699 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3700 zil_commit(zilog, 0);
3701
3702 ZFS_EXIT(zfsvfs);
3703 return (error);
3704 }
3705
3706 /*
3707 * Insert the indicated symbolic reference entry into the directory.
3708 *
3709 * IN: dvp - Directory to contain new symbolic link.
3710 * link - Name for new symlink entry.
3711 * vap - Attributes of new entry.
3712 * cr - credentials of caller.
3713 * ct - caller context
3714 * flags - case flags
3715 *
3716 * RETURN: 0 on success, error code on failure.
3717 *
3718 * Timestamps:
3719 * dvp - ctime|mtime updated
3720 */
3721 /*ARGSUSED*/
3722 static int
3723 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3724 caller_context_t *ct, int flags)
3725 {
3726 znode_t *zp, *dzp = VTOZ(dvp);
3727 zfs_dirlock_t *dl;
3728 dmu_tx_t *tx;
3729 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
3730 zilog_t *zilog;
3731 uint64_t len = strlen(link);
3732 int error;
3733 int zflg = ZNEW;
3734 zfs_acl_ids_t acl_ids;
3735 boolean_t fuid_dirtied;
3736 uint64_t txtype = TX_SYMLINK;
3737 boolean_t waited = B_FALSE;
3738
3739 ASSERT(vap->va_type == VLNK);
3740
3741 ZFS_ENTER(zfsvfs);
3742 ZFS_VERIFY_ZP(dzp);
3743 zilog = zfsvfs->z_log;
3744
3745 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3746 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3747 ZFS_EXIT(zfsvfs);
3748 return (SET_ERROR(EILSEQ));
3749 }
3750 if (flags & FIGNORECASE)
3751 zflg |= ZCILOOK;
3752
3753 if (len > MAXPATHLEN) {
3754 ZFS_EXIT(zfsvfs);
3755 return (SET_ERROR(ENAMETOOLONG));
3756 }
3757
3758 if ((error = zfs_acl_ids_create(dzp, 0,
3759 vap, cr, NULL, &acl_ids)) != 0) {
3760 ZFS_EXIT(zfsvfs);
3761 return (error);
3762 }
3763 top:
3764 /*
3765 * Attempt to lock directory; fail if entry already exists.
3766 */
3767 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3768 if (error) {
3769 zfs_acl_ids_free(&acl_ids);
3770 ZFS_EXIT(zfsvfs);
3771 return (error);
3772 }
3773
3774 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3775 zfs_acl_ids_free(&acl_ids);
3776 zfs_dirent_unlock(dl);
3777 ZFS_EXIT(zfsvfs);
3778 return (error);
3779 }
3780
3781 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3782 zfs_acl_ids_free(&acl_ids);
3783 zfs_dirent_unlock(dl);
3784 ZFS_EXIT(zfsvfs);
3785 return (SET_ERROR(EDQUOT));
3786 }
3787 tx = dmu_tx_create(zfsvfs->z_os);
3788 fuid_dirtied = zfsvfs->z_fuid_dirty;
3789 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3790 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3791 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3792 ZFS_SA_BASE_ATTR_SIZE + len);
3793 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3794 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3795 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3796 acl_ids.z_aclp->z_acl_bytes);
3797 }
3798 if (fuid_dirtied)
3799 zfs_fuid_txhold(zfsvfs, tx);
3800 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3801 if (error) {
3802 zfs_dirent_unlock(dl);
3803 if (error == ERESTART) {
3804 waited = B_TRUE;
3805 dmu_tx_wait(tx);
3806 dmu_tx_abort(tx);
3807 goto top;
3808 }
3809 zfs_acl_ids_free(&acl_ids);
3810 dmu_tx_abort(tx);
3811 ZFS_EXIT(zfsvfs);
3812 return (error);
3813 }
3814
3815 /*
3816 * Create a new object for the symlink.
3817 * for version 4 ZPL datsets the symlink will be an SA attribute
3818 */
3819 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3820
3821 if (fuid_dirtied)
3822 zfs_fuid_sync(zfsvfs, tx);
3823
3824 mutex_enter(&zp->z_lock);
3825 if (zp->z_is_sa)
3826 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3827 link, len, tx);
3828 else
3829 zfs_sa_symlink(zp, link, len, tx);
3830 mutex_exit(&zp->z_lock);
3831
3832 zp->z_size = len;
3833 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3834 &zp->z_size, sizeof (zp->z_size), tx);
3835 /*
3836 * Insert the new object into the directory.
3837 */
3838 (void) zfs_link_create(dl, zp, tx, ZNEW);
3839
3840 if (flags & FIGNORECASE)
3841 txtype |= TX_CI;
3842 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3843
3844 zfs_acl_ids_free(&acl_ids);
3845
3846 dmu_tx_commit(tx);
3847
3848 zfs_dirent_unlock(dl);
3849
3850 VN_RELE(ZTOV(zp));
3851
3852 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3853 zil_commit(zilog, 0);
3854
3855 ZFS_EXIT(zfsvfs);
3856 return (error);
3857 }
3858
3859 /*
3860 * Return, in the buffer contained in the provided uio structure,
3861 * the symbolic path referred to by vp.
3862 *
3863 * IN: vp - vnode of symbolic link.
3864 * uio - structure to contain the link path.
3865 * cr - credentials of caller.
3866 * ct - caller context
3867 *
3868 * OUT: uio - structure containing the link path.
3869 *
3870 * RETURN: 0 on success, error code on failure.
3871 *
3872 * Timestamps:
3873 * vp - atime updated
3874 */
3875 /* ARGSUSED */
3876 static int
3877 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3878 {
3879 znode_t *zp = VTOZ(vp);
3880 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3881 int error;
3882
3883 ZFS_ENTER(zfsvfs);
3884 ZFS_VERIFY_ZP(zp);
3885
3886 mutex_enter(&zp->z_lock);
3887 if (zp->z_is_sa)
3888 error = sa_lookup_uio(zp->z_sa_hdl,
3889 SA_ZPL_SYMLINK(zfsvfs), uio);
3890 else
3891 error = zfs_sa_readlink(zp, uio);
3892 mutex_exit(&zp->z_lock);
3893
3894 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3895
3896 ZFS_EXIT(zfsvfs);
3897 return (error);
3898 }
3899
3900 /*
3901 * Insert a new entry into directory tdvp referencing svp.
3902 *
3903 * IN: tdvp - Directory to contain new entry.
3904 * svp - vnode of new entry.
3905 * name - name of new entry.
3906 * cr - credentials of caller.
3907 * ct - caller context
3908 *
3909 * RETURN: 0 on success, error code on failure.
3910 *
3911 * Timestamps:
3912 * tdvp - ctime|mtime updated
3913 * svp - ctime updated
3914 */
3915 /* ARGSUSED */
3916 static int
3917 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3918 caller_context_t *ct, int flags)
3919 {
3920 znode_t *dzp = VTOZ(tdvp);
3921 znode_t *tzp, *szp;
3922 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
3923 zilog_t *zilog;
3924 zfs_dirlock_t *dl;
3925 dmu_tx_t *tx;
3926 vnode_t *realvp;
3927 int error;
3928 int zf = ZNEW;
3929 uint64_t parent;
3930 uid_t owner;
3931 boolean_t waited = B_FALSE;
3932
3933 ASSERT(tdvp->v_type == VDIR);
3934
3935 ZFS_ENTER(zfsvfs);
3936 ZFS_VERIFY_ZP(dzp);
3937 zilog = zfsvfs->z_log;
3938
3939 if (VOP_REALVP(svp, &realvp, ct) == 0)
3940 svp = realvp;
3941
3942 /*
3943 * POSIX dictates that we return EPERM here.
3944 * Better choices include ENOTSUP or EISDIR.
3945 */
3946 if (svp->v_type == VDIR) {
3947 ZFS_EXIT(zfsvfs);
3948 return (SET_ERROR(EPERM));
3949 }
3950
3951 szp = VTOZ(svp);
3952 ZFS_VERIFY_ZP(szp);
3953
3954 /*
3955 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3956 * ctldir appear to have the same v_vfsp.
3957 */
3958 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
3959 ZFS_EXIT(zfsvfs);
3960 return (SET_ERROR(EXDEV));
3961 }
3962
3963 /* Prevent links to .zfs/shares files */
3964
3965 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
3966 &parent, sizeof (uint64_t))) != 0) {
3967 ZFS_EXIT(zfsvfs);
3968 return (error);
3969 }
3970 if (parent == zfsvfs->z_shares_dir) {
3971 ZFS_EXIT(zfsvfs);
3972 return (SET_ERROR(EPERM));
3973 }
3974
3975 if (zfsvfs->z_utf8 && u8_validate(name,
3976 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3977 ZFS_EXIT(zfsvfs);
3978 return (SET_ERROR(EILSEQ));
3979 }
3980 if (flags & FIGNORECASE)
3981 zf |= ZCILOOK;
3982
3983 /*
3984 * We do not support links between attributes and non-attributes
3985 * because of the potential security risk of creating links
3986 * into "normal" file space in order to circumvent restrictions
3987 * imposed in attribute space.
3988 */
3989 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3990 ZFS_EXIT(zfsvfs);
3991 return (SET_ERROR(EINVAL));
3992 }
3993
3994
3995 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
3996 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3997 ZFS_EXIT(zfsvfs);
3998 return (SET_ERROR(EPERM));
3999 }
4000
4001 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4002 ZFS_EXIT(zfsvfs);
4003 return (error);
4004 }
4005
4006 top:
4007 /*
4008 * Attempt to lock directory; fail if entry already exists.
4009 */
4010 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4011 if (error) {
4012 ZFS_EXIT(zfsvfs);
4013 return (error);
4014 }
4015
4016 tx = dmu_tx_create(zfsvfs->z_os);
4017 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4018 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4019 zfs_sa_upgrade_txholds(tx, szp);
4020 zfs_sa_upgrade_txholds(tx, dzp);
4021 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4022 if (error) {
4023 zfs_dirent_unlock(dl);
4024 if (error == ERESTART) {
4025 waited = B_TRUE;
4026 dmu_tx_wait(tx);
4027 dmu_tx_abort(tx);
4028 goto top;
4029 }
4030 dmu_tx_abort(tx);
4031 ZFS_EXIT(zfsvfs);
4032 return (error);
4033 }
4034
4035 error = zfs_link_create(dl, szp, tx, 0);
4036
4037 if (error == 0) {
4038 uint64_t txtype = TX_LINK;
4039 if (flags & FIGNORECASE)
4040 txtype |= TX_CI;
4041 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4042 }
4043
4044 dmu_tx_commit(tx);
4045
4046 zfs_dirent_unlock(dl);
4047
4048 if (error == 0) {
4049 vnevent_link(svp, ct);
4050 }
4051
4052 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4053 zil_commit(zilog, 0);
4054
4055 ZFS_EXIT(zfsvfs);
4056 return (error);
4057 }
4058
4059 /*
4060 * zfs_null_putapage() is used when the file system has been force
4061 * unmounted. It just drops the pages.
4062 */
4063 /* ARGSUSED */
4064 static int
4065 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4066 size_t *lenp, int flags, cred_t *cr)
4067 {
4068 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4069 return (0);
4070 }
4071
4072 /*
4073 * Push a page out to disk, klustering if possible.
4074 *
4075 * IN: vp - file to push page to.
4076 * pp - page to push.
4077 * flags - additional flags.
4078 * cr - credentials of caller.
4079 *
4080 * OUT: offp - start of range pushed.
4081 * lenp - len of range pushed.
4082 *
4083 * RETURN: 0 on success, error code on failure.
4084 *
4085 * NOTE: callers must have locked the page to be pushed. On
4086 * exit, the page (and all other pages in the kluster) must be
4087 * unlocked.
4088 */
4089 /* ARGSUSED */
4090 static int
4091 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4092 size_t *lenp, int flags, cred_t *cr)
4093 {
4094 znode_t *zp = VTOZ(vp);
4095 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4096 dmu_tx_t *tx;
4097 u_offset_t off, koff;
4098 size_t len, klen;
4099 int err;
4100
4101 off = pp->p_offset;
4102 len = PAGESIZE;
4103 /*
4104 * If our blocksize is bigger than the page size, try to kluster
4105 * multiple pages so that we write a full block (thus avoiding
4106 * a read-modify-write).
4107 */
4108 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4109 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4110 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4111 ASSERT(koff <= zp->z_size);
4112 if (koff + klen > zp->z_size)
4113 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4114 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4115 }
4116 ASSERT3U(btop(len), ==, btopr(len));
4117
4118 /*
4119 * Can't push pages past end-of-file.
4120 */
4121 if (off >= zp->z_size) {
4122 /* ignore all pages */
4123 err = 0;
4124 goto out;
4125 } else if (off + len > zp->z_size) {
4126 int npages = btopr(zp->z_size - off);
4127 page_t *trunc;
4128
4129 page_list_break(&pp, &trunc, npages);
4130 /* ignore pages past end of file */
4131 if (trunc)
4132 pvn_write_done(trunc, flags);
4133 len = zp->z_size - off;
4134 }
4135
4136 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4137 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4138 err = SET_ERROR(EDQUOT);
4139 goto out;
4140 }
4141 tx = dmu_tx_create(zfsvfs->z_os);
4142 dmu_tx_hold_write(tx, zp->z_id, off, len);
4143
4144 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4145 zfs_sa_upgrade_txholds(tx, zp);
4146 err = dmu_tx_assign(tx, TXG_WAIT);
4147 if (err != 0) {
4148 dmu_tx_abort(tx);
4149 goto out;
4150 }
4151
4152 if (zp->z_blksz <= PAGESIZE) {
4153 caddr_t va = zfs_map_page(pp, S_READ);
4154 ASSERT3U(len, <=, PAGESIZE);
4155 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4156 zfs_unmap_page(pp, va);
4157 } else {
4158 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4159 }
4160
4161 if (err == 0) {
4162 uint64_t mtime[2], ctime[2];
4163 sa_bulk_attr_t bulk[3];
4164 int count = 0;
4165
4166 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4167 &mtime, 16);
4168 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4169 &ctime, 16);
4170 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4171 &zp->z_pflags, 8);
4172 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4173 B_TRUE);
4174 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4175 }
4176 dmu_tx_commit(tx);
4177
4178 out:
4179 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4180 if (offp)
4181 *offp = off;
4182 if (lenp)
4183 *lenp = len;
4184
4185 return (err);
4186 }
4187
4188 /*
4189 * Copy the portion of the file indicated from pages into the file.
4190 * The pages are stored in a page list attached to the files vnode.
4191 *
4192 * IN: vp - vnode of file to push page data to.
4193 * off - position in file to put data.
4194 * len - amount of data to write.
4195 * flags - flags to control the operation.
4196 * cr - credentials of caller.
4197 * ct - caller context.
4198 *
4199 * RETURN: 0 on success, error code on failure.
4200 *
4201 * Timestamps:
4202 * vp - ctime|mtime updated
4203 */
4204 /*ARGSUSED*/
4205 static int
4206 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4207 caller_context_t *ct)
4208 {
4209 znode_t *zp = VTOZ(vp);
4210 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4211 page_t *pp;
4212 size_t io_len;
4213 u_offset_t io_off;
4214 uint_t blksz;
4215 rl_t *rl;
4216 int error = 0;
4217
4218 ZFS_ENTER(zfsvfs);
4219 ZFS_VERIFY_ZP(zp);
4220
4221 /*
4222 * There's nothing to do if no data is cached.
4223 */
4224 if (!vn_has_cached_data(vp)) {
4225 ZFS_EXIT(zfsvfs);
4226 return (0);
4227 }
4228
4229 /*
4230 * Align this request to the file block size in case we kluster.
4231 * XXX - this can result in pretty aggresive locking, which can
4232 * impact simultanious read/write access. One option might be
4233 * to break up long requests (len == 0) into block-by-block
4234 * operations to get narrower locking.
4235 */
4236 blksz = zp->z_blksz;
4237 if (ISP2(blksz))
4238 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4239 else
4240 io_off = 0;
4241 if (len > 0 && ISP2(blksz))
4242 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4243 else
4244 io_len = 0;
4245
4246 if (io_len == 0) {
4247 /*
4248 * Search the entire vp list for pages >= io_off.
4249 */
4250 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4251 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4252 goto out;
4253 }
4254 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4255
4256 if (off > zp->z_size) {
4257 /* past end of file */
4258 zfs_range_unlock(rl);
4259 ZFS_EXIT(zfsvfs);
4260 return (0);
4261 }
4262
4263 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4264
4265 for (off = io_off; io_off < off + len; io_off += io_len) {
4266 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4267 pp = page_lookup(vp, io_off,
4268 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4269 } else {
4270 pp = page_lookup_nowait(vp, io_off,
4271 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4272 }
4273
4274 if (pp != NULL && pvn_getdirty(pp, flags)) {
4275 int err;
4276
4277 /*
4278 * Found a dirty page to push
4279 */
4280 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4281 if (err)
4282 error = err;
4283 } else {
4284 io_len = PAGESIZE;
4285 }
4286 }
4287 out:
4288 zfs_range_unlock(rl);
4289 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4290 zil_commit(zfsvfs->z_log, zp->z_id);
4291 ZFS_EXIT(zfsvfs);
4292 return (error);
4293 }
4294
4295 /*ARGSUSED*/
4296 void
4297 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4298 {
4299 znode_t *zp = VTOZ(vp);
4300 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4301 int error;
4302
4303 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4304 if (zp->z_sa_hdl == NULL) {
4305 /*
4306 * The fs has been unmounted, or we did a
4307 * suspend/resume and this file no longer exists.
4308 */
4309 if (vn_has_cached_data(vp)) {
4310 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4311 B_INVAL, cr);
4312 }
4313
4314 mutex_enter(&zp->z_lock);
4315 mutex_enter(&vp->v_lock);
4316 ASSERT(vp->v_count == 1);
4317 vp->v_count = 0;
4318 mutex_exit(&vp->v_lock);
4319 mutex_exit(&zp->z_lock);
4320 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4321 zfs_znode_free(zp);
4322 return;
4323 }
4324
4325 /*
4326 * Attempt to push any data in the page cache. If this fails
4327 * we will get kicked out later in zfs_zinactive().
4328 */
4329 if (vn_has_cached_data(vp)) {
4330 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4331 cr);
4332 }
4333
4334 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4335 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4336
4337 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4338 zfs_sa_upgrade_txholds(tx, zp);
4339 error = dmu_tx_assign(tx, TXG_WAIT);
4340 if (error) {
4341 dmu_tx_abort(tx);
4342 } else {
4343 mutex_enter(&zp->z_lock);
4344 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4345 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4346 zp->z_atime_dirty = 0;
4347 mutex_exit(&zp->z_lock);
4348 dmu_tx_commit(tx);
4349 }
4350 }
4351
4352 zfs_zinactive(zp);
4353 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4354 }
4355
4356 /*
4357 * Bounds-check the seek operation.
4358 *
4359 * IN: vp - vnode seeking within
4360 * ooff - old file offset
4361 * noffp - pointer to new file offset
4362 * ct - caller context
4363 *
4364 * RETURN: 0 on success, EINVAL if new offset invalid.
4365 */
4366 /* ARGSUSED */
4367 static int
4368 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4369 caller_context_t *ct)
4370 {
4371 if (vp->v_type == VDIR)
4372 return (0);
4373 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4374 }
4375
4376 /*
4377 * Pre-filter the generic locking function to trap attempts to place
4378 * a mandatory lock on a memory mapped file.
4379 */
4380 static int
4381 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4382 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4383 {
4384 znode_t *zp = VTOZ(vp);
4385 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4386
4387 ZFS_ENTER(zfsvfs);
4388 ZFS_VERIFY_ZP(zp);
4389
4390 /*
4391 * We are following the UFS semantics with respect to mapcnt
4392 * here: If we see that the file is mapped already, then we will
4393 * return an error, but we don't worry about races between this
4394 * function and zfs_map().
4395 */
4396 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4397 ZFS_EXIT(zfsvfs);
4398 return (SET_ERROR(EAGAIN));
4399 }
4400 ZFS_EXIT(zfsvfs);
4401 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4402 }
4403
4404 /*
4405 * If we can't find a page in the cache, we will create a new page
4406 * and fill it with file data. For efficiency, we may try to fill
4407 * multiple pages at once (klustering) to fill up the supplied page
4408 * list. Note that the pages to be filled are held with an exclusive
4409 * lock to prevent access by other threads while they are being filled.
4410 */
4411 static int
4412 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4413 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4414 {
4415 znode_t *zp = VTOZ(vp);
4416 page_t *pp, *cur_pp;
4417 objset_t *os = zp->z_zfsvfs->z_os;
4418 u_offset_t io_off, total;
4419 size_t io_len;
4420 int err;
4421
4422 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4423 /*
4424 * We only have a single page, don't bother klustering
4425 */
4426 io_off = off;
4427 io_len = PAGESIZE;
4428 pp = page_create_va(vp, io_off, io_len,
4429 PG_EXCL | PG_WAIT, seg, addr);
4430 } else {
4431 /*
4432 * Try to find enough pages to fill the page list
4433 */
4434 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4435 &io_len, off, plsz, 0);
4436 }
4437 if (pp == NULL) {
4438 /*
4439 * The page already exists, nothing to do here.
4440 */
4441 *pl = NULL;
4442 return (0);
4443 }
4444
4445 /*
4446 * Fill the pages in the kluster.
4447 */
4448 cur_pp = pp;
4449 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4450 caddr_t va;
4451
4452 ASSERT3U(io_off, ==, cur_pp->p_offset);
4453 va = zfs_map_page(cur_pp, S_WRITE);
4454 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4455 DMU_READ_PREFETCH);
4456 zfs_unmap_page(cur_pp, va);
4457 if (err) {
4458 /* On error, toss the entire kluster */
4459 pvn_read_done(pp, B_ERROR);
4460 /* convert checksum errors into IO errors */
4461 if (err == ECKSUM)
4462 err = SET_ERROR(EIO);
4463 return (err);
4464 }
4465 cur_pp = cur_pp->p_next;
4466 }
4467
4468 /*
4469 * Fill in the page list array from the kluster starting
4470 * from the desired offset `off'.
4471 * NOTE: the page list will always be null terminated.
4472 */
4473 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4474 ASSERT(pl == NULL || (*pl)->p_offset == off);
4475
4476 return (0);
4477 }
4478
4479 /*
4480 * Return pointers to the pages for the file region [off, off + len]
4481 * in the pl array. If plsz is greater than len, this function may
4482 * also return page pointers from after the specified region
4483 * (i.e. the region [off, off + plsz]). These additional pages are
4484 * only returned if they are already in the cache, or were created as
4485 * part of a klustered read.
4486 *
4487 * IN: vp - vnode of file to get data from.
4488 * off - position in file to get data from.
4489 * len - amount of data to retrieve.
4490 * plsz - length of provided page list.
4491 * seg - segment to obtain pages for.
4492 * addr - virtual address of fault.
4493 * rw - mode of created pages.
4494 * cr - credentials of caller.
4495 * ct - caller context.
4496 *
4497 * OUT: protp - protection mode of created pages.
4498 * pl - list of pages created.
4499 *
4500 * RETURN: 0 on success, error code on failure.
4501 *
4502 * Timestamps:
4503 * vp - atime updated
4504 */
4505 /* ARGSUSED */
4506 static int
4507 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4508 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4509 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4510 {
4511 znode_t *zp = VTOZ(vp);
4512 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4513 page_t **pl0 = pl;
4514 int err = 0;
4515
4516 /* we do our own caching, faultahead is unnecessary */
4517 if (pl == NULL)
4518 return (0);
4519 else if (len > plsz)
4520 len = plsz;
4521 else
4522 len = P2ROUNDUP(len, PAGESIZE);
4523 ASSERT(plsz >= len);
4524
4525 ZFS_ENTER(zfsvfs);
4526 ZFS_VERIFY_ZP(zp);
4527
4528 if (protp)
4529 *protp = PROT_ALL;
4530
4531 /*
4532 * Loop through the requested range [off, off + len) looking
4533 * for pages. If we don't find a page, we will need to create
4534 * a new page and fill it with data from the file.
4535 */
4536 while (len > 0) {
4537 if (*pl = page_lookup(vp, off, SE_SHARED))
4538 *(pl+1) = NULL;
4539 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4540 goto out;
4541 while (*pl) {
4542 ASSERT3U((*pl)->p_offset, ==, off);
4543 off += PAGESIZE;
4544 addr += PAGESIZE;
4545 if (len > 0) {
4546 ASSERT3U(len, >=, PAGESIZE);
4547 len -= PAGESIZE;
4548 }
4549 ASSERT3U(plsz, >=, PAGESIZE);
4550 plsz -= PAGESIZE;
4551 pl++;
4552 }
4553 }
4554
4555 /*
4556 * Fill out the page array with any pages already in the cache.
4557 */
4558 while (plsz > 0 &&
4559 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4560 off += PAGESIZE;
4561 plsz -= PAGESIZE;
4562 }
4563 out:
4564 if (err) {
4565 /*
4566 * Release any pages we have previously locked.
4567 */
4568 while (pl > pl0)
4569 page_unlock(*--pl);
4570 } else {
4571 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4572 }
4573
4574 *pl = NULL;
4575
4576 ZFS_EXIT(zfsvfs);
4577 return (err);
4578 }
4579
4580 /*
4581 * Request a memory map for a section of a file. This code interacts
4582 * with common code and the VM system as follows:
4583 *
4584 * - common code calls mmap(), which ends up in smmap_common()
4585 * - this calls VOP_MAP(), which takes you into (say) zfs
4586 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4587 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4588 * - zfs_addmap() updates z_mapcnt
4589 */
4590 /*ARGSUSED*/
4591 static int
4592 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4593 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4594 caller_context_t *ct)
4595 {
4596 znode_t *zp = VTOZ(vp);
4597 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4598 segvn_crargs_t vn_a;
4599 int error;
4600
4601 ZFS_ENTER(zfsvfs);
4602 ZFS_VERIFY_ZP(zp);
4603
4604 if ((prot & PROT_WRITE) && (zp->z_pflags &
4605 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4606 ZFS_EXIT(zfsvfs);
4607 return (SET_ERROR(EPERM));
4608 }
4609
4610 if ((prot & (PROT_READ | PROT_EXEC)) &&
4611 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4612 ZFS_EXIT(zfsvfs);
4613 return (SET_ERROR(EACCES));
4614 }
4615
4616 if (vp->v_flag & VNOMAP) {
4617 ZFS_EXIT(zfsvfs);
4618 return (SET_ERROR(ENOSYS));
4619 }
4620
4621 if (off < 0 || len > MAXOFFSET_T - off) {
4622 ZFS_EXIT(zfsvfs);
4623 return (SET_ERROR(ENXIO));
4624 }
4625
4626 if (vp->v_type != VREG) {
4627 ZFS_EXIT(zfsvfs);
4628 return (SET_ERROR(ENODEV));
4629 }
4630
4631 /*
4632 * If file is locked, disallow mapping.
4633 */
4634 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4635 ZFS_EXIT(zfsvfs);
4636 return (SET_ERROR(EAGAIN));
4637 }
4638
4639 as_rangelock(as);
4640 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4641 if (error != 0) {
4642 as_rangeunlock(as);
4643 ZFS_EXIT(zfsvfs);
4644 return (error);
4645 }
4646
4647 vn_a.vp = vp;
4648 vn_a.offset = (u_offset_t)off;
4649 vn_a.type = flags & MAP_TYPE;
4650 vn_a.prot = prot;
4651 vn_a.maxprot = maxprot;
4652 vn_a.cred = cr;
4653 vn_a.amp = NULL;
4654 vn_a.flags = flags & ~MAP_TYPE;
4655 vn_a.szc = 0;
4656 vn_a.lgrp_mem_policy_flags = 0;
4657
4658 error = as_map(as, *addrp, len, segvn_create, &vn_a);
4659
4660 as_rangeunlock(as);
4661 ZFS_EXIT(zfsvfs);
4662 return (error);
4663 }
4664
4665 /* ARGSUSED */
4666 static int
4667 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4668 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4669 caller_context_t *ct)
4670 {
4671 uint64_t pages = btopr(len);
4672
4673 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4674 return (0);
4675 }
4676
4677 /*
4678 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4679 * more accurate mtime for the associated file. Since we don't have a way of
4680 * detecting when the data was actually modified, we have to resort to
4681 * heuristics. If an explicit msync() is done, then we mark the mtime when the
4682 * last page is pushed. The problem occurs when the msync() call is omitted,
4683 * which by far the most common case:
4684 *
4685 * open()
4686 * mmap()
4687 * <modify memory>
4688 * munmap()
4689 * close()
4690 * <time lapse>
4691 * putpage() via fsflush
4692 *
4693 * If we wait until fsflush to come along, we can have a modification time that
4694 * is some arbitrary point in the future. In order to prevent this in the
4695 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4696 * torn down.
4697 */
4698 /* ARGSUSED */
4699 static int
4700 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4701 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4702 caller_context_t *ct)
4703 {
4704 uint64_t pages = btopr(len);
4705
4706 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4707 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4708
4709 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4710 vn_has_cached_data(vp))
4711 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4712
4713 return (0);
4714 }
4715
4716 /*
4717 * Free or allocate space in a file. Currently, this function only
4718 * supports the `F_FREESP' command. However, this command is somewhat
4719 * misnamed, as its functionality includes the ability to allocate as
4720 * well as free space.
4721 *
4722 * IN: vp - vnode of file to free data in.
4723 * cmd - action to take (only F_FREESP supported).
4724 * bfp - section of file to free/alloc.
4725 * flag - current file open mode flags.
4726 * offset - current file offset.
4727 * cr - credentials of caller [UNUSED].
4728 * ct - caller context.
4729 *
4730 * RETURN: 0 on success, error code on failure.
4731 *
4732 * Timestamps:
4733 * vp - ctime|mtime updated
4734 */
4735 /* ARGSUSED */
4736 static int
4737 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4738 offset_t offset, cred_t *cr, caller_context_t *ct)
4739 {
4740 znode_t *zp = VTOZ(vp);
4741 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4742 uint64_t off, len;
4743 int error;
4744
4745 ZFS_ENTER(zfsvfs);
4746 ZFS_VERIFY_ZP(zp);
4747
4748 if (cmd != F_FREESP) {
4749 ZFS_EXIT(zfsvfs);
4750 return (SET_ERROR(EINVAL));
4751 }
4752
4753 if (error = convoff(vp, bfp, 0, offset)) {
4754 ZFS_EXIT(zfsvfs);
4755 return (error);
4756 }
4757
4758 if (bfp->l_len < 0) {
4759 ZFS_EXIT(zfsvfs);
4760 return (SET_ERROR(EINVAL));
4761 }
4762
4763 off = bfp->l_start;
4764 len = bfp->l_len; /* 0 means from off to end of file */
4765
4766 error = zfs_freesp(zp, off, len, flag, TRUE);
4767
4768 if (error == 0 && off == 0 && len == 0)
4769 vnevent_truncate(ZTOV(zp), ct);
4770
4771 ZFS_EXIT(zfsvfs);
4772 return (error);
4773 }
4774
4775 /*ARGSUSED*/
4776 static int
4777 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4778 {
4779 znode_t *zp = VTOZ(vp);
4780 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4781 uint32_t gen;
4782 uint64_t gen64;
4783 uint64_t object = zp->z_id;
4784 zfid_short_t *zfid;
4785 int size, i, error;
4786
4787 ZFS_ENTER(zfsvfs);
4788 ZFS_VERIFY_ZP(zp);
4789
4790 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4791 &gen64, sizeof (uint64_t))) != 0) {
4792 ZFS_EXIT(zfsvfs);
4793 return (error);
4794 }
4795
4796 gen = (uint32_t)gen64;
4797
4798 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4799 if (fidp->fid_len < size) {
4800 fidp->fid_len = size;
4801 ZFS_EXIT(zfsvfs);
4802 return (SET_ERROR(ENOSPC));
4803 }
4804
4805 zfid = (zfid_short_t *)fidp;
4806
4807 zfid->zf_len = size;
4808
4809 for (i = 0; i < sizeof (zfid->zf_object); i++)
4810 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4811
4812 /* Must have a non-zero generation number to distinguish from .zfs */
4813 if (gen == 0)
4814 gen = 1;
4815 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4816 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4817
4818 if (size == LONG_FID_LEN) {
4819 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
4820 zfid_long_t *zlfid;
4821
4822 zlfid = (zfid_long_t *)fidp;
4823
4824 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4825 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4826
4827 /* XXX - this should be the generation number for the objset */
4828 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4829 zlfid->zf_setgen[i] = 0;
4830 }
4831
4832 ZFS_EXIT(zfsvfs);
4833 return (0);
4834 }
4835
4836 static int
4837 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4838 caller_context_t *ct)
4839 {
4840 znode_t *zp, *xzp;
4841 zfsvfs_t *zfsvfs;
4842 zfs_dirlock_t *dl;
4843 int error;
4844
4845 switch (cmd) {
4846 case _PC_LINK_MAX:
4847 *valp = ULONG_MAX;
4848 return (0);
4849
4850 case _PC_FILESIZEBITS:
4851 *valp = 64;
4852 return (0);
4853
4854 case _PC_XATTR_EXISTS:
4855 zp = VTOZ(vp);
4856 zfsvfs = zp->z_zfsvfs;
4857 ZFS_ENTER(zfsvfs);
4858 ZFS_VERIFY_ZP(zp);
4859 *valp = 0;
4860 error = zfs_dirent_lock(&dl, zp, "", &xzp,
4861 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4862 if (error == 0) {
4863 zfs_dirent_unlock(dl);
4864 if (!zfs_dirempty(xzp))
4865 *valp = 1;
4866 VN_RELE(ZTOV(xzp));
4867 } else if (error == ENOENT) {
4868 /*
4869 * If there aren't extended attributes, it's the
4870 * same as having zero of them.
4871 */
4872 error = 0;
4873 }
4874 ZFS_EXIT(zfsvfs);
4875 return (error);
4876
4877 case _PC_SATTR_ENABLED:
4878 case _PC_SATTR_EXISTS:
4879 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4880 (vp->v_type == VREG || vp->v_type == VDIR);
4881 return (0);
4882
4883 case _PC_ACCESS_FILTERING:
4884 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4885 vp->v_type == VDIR;
4886 return (0);
4887
4888 case _PC_ACL_ENABLED:
4889 *valp = _ACL_ACE_ENABLED;
4890 return (0);
4891
4892 case _PC_MIN_HOLE_SIZE:
4893 *valp = (ulong_t)SPA_MINBLOCKSIZE;
4894 return (0);
4895
4896 case _PC_TIMESTAMP_RESOLUTION:
4897 /* nanosecond timestamp resolution */
4898 *valp = 1L;
4899 return (0);
4900
4901 default:
4902 return (fs_pathconf(vp, cmd, valp, cr, ct));
4903 }
4904 }
4905
4906 /*ARGSUSED*/
4907 static int
4908 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4909 caller_context_t *ct)
4910 {
4911 znode_t *zp = VTOZ(vp);
4912 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4913 int error;
4914 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4915
4916 ZFS_ENTER(zfsvfs);
4917 ZFS_VERIFY_ZP(zp);
4918 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4919 ZFS_EXIT(zfsvfs);
4920
4921 return (error);
4922 }
4923
4924 /*ARGSUSED*/
4925 static int
4926 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4927 caller_context_t *ct)
4928 {
4929 znode_t *zp = VTOZ(vp);
4930 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4931 int error;
4932 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4933 zilog_t *zilog = zfsvfs->z_log;
4934
4935 ZFS_ENTER(zfsvfs);
4936 ZFS_VERIFY_ZP(zp);
4937
4938 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4939
4940 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4941 zil_commit(zilog, 0);
4942
4943 ZFS_EXIT(zfsvfs);
4944 return (error);
4945 }
4946
4947 /*
4948 * The smallest read we may consider to loan out an arcbuf.
4949 * This must be a power of 2.
4950 */
4951 int zcr_blksz_min = (1 << 10); /* 1K */
4952 /*
4953 * If set to less than the file block size, allow loaning out of an
4954 * arcbuf for a partial block read. This must be a power of 2.
4955 */
4956 int zcr_blksz_max = (1 << 17); /* 128K */
4957
4958 /*ARGSUSED*/
4959 static int
4960 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
4961 caller_context_t *ct)
4962 {
4963 znode_t *zp = VTOZ(vp);
4964 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4965 int max_blksz = zfsvfs->z_max_blksz;
4966 uio_t *uio = &xuio->xu_uio;
4967 ssize_t size = uio->uio_resid;
4968 offset_t offset = uio->uio_loffset;
4969 int blksz;
4970 int fullblk, i;
4971 arc_buf_t *abuf;
4972 ssize_t maxsize;
4973 int preamble, postamble;
4974
4975 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4976 return (SET_ERROR(EINVAL));
4977
4978 ZFS_ENTER(zfsvfs);
4979 ZFS_VERIFY_ZP(zp);
4980 switch (ioflag) {
4981 case UIO_WRITE:
4982 /*
4983 * Loan out an arc_buf for write if write size is bigger than
4984 * max_blksz, and the file's block size is also max_blksz.
4985 */
4986 blksz = max_blksz;
4987 if (size < blksz || zp->z_blksz != blksz) {
4988 ZFS_EXIT(zfsvfs);
4989 return (SET_ERROR(EINVAL));
4990 }
4991 /*
4992 * Caller requests buffers for write before knowing where the
4993 * write offset might be (e.g. NFS TCP write).
4994 */
4995 if (offset == -1) {
4996 preamble = 0;
4997 } else {
4998 preamble = P2PHASE(offset, blksz);
4999 if (preamble) {
5000 preamble = blksz - preamble;
5001 size -= preamble;
5002 }
5003 }
5004
5005 postamble = P2PHASE(size, blksz);
5006 size -= postamble;
5007
5008 fullblk = size / blksz;
5009 (void) dmu_xuio_init(xuio,
5010 (preamble != 0) + fullblk + (postamble != 0));
5011 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5012 int, postamble, int,
5013 (preamble != 0) + fullblk + (postamble != 0));
5014
5015 /*
5016 * Have to fix iov base/len for partial buffers. They
5017 * currently represent full arc_buf's.
5018 */
5019 if (preamble) {
5020 /* data begins in the middle of the arc_buf */
5021 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5022 blksz);
5023 ASSERT(abuf);
5024 (void) dmu_xuio_add(xuio, abuf,
5025 blksz - preamble, preamble);
5026 }
5027
5028 for (i = 0; i < fullblk; i++) {
5029 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5030 blksz);
5031 ASSERT(abuf);
5032 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5033 }
5034
5035 if (postamble) {
5036 /* data ends in the middle of the arc_buf */
5037 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5038 blksz);
5039 ASSERT(abuf);
5040 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5041 }
5042 break;
5043 case UIO_READ:
5044 /*
5045 * Loan out an arc_buf for read if the read size is larger than
5046 * the current file block size. Block alignment is not
5047 * considered. Partial arc_buf will be loaned out for read.
5048 */
5049 blksz = zp->z_blksz;
5050 if (blksz < zcr_blksz_min)
5051 blksz = zcr_blksz_min;
5052 if (blksz > zcr_blksz_max)
5053 blksz = zcr_blksz_max;
5054 /* avoid potential complexity of dealing with it */
5055 if (blksz > max_blksz) {
5056 ZFS_EXIT(zfsvfs);
5057 return (SET_ERROR(EINVAL));
5058 }
5059
5060 maxsize = zp->z_size - uio->uio_loffset;
5061 if (size > maxsize)
5062 size = maxsize;
5063
5064 if (size < blksz || vn_has_cached_data(vp)) {
5065 ZFS_EXIT(zfsvfs);
5066 return (SET_ERROR(EINVAL));
5067 }
5068 break;
5069 default:
5070 ZFS_EXIT(zfsvfs);
5071 return (SET_ERROR(EINVAL));
5072 }
5073
5074 uio->uio_extflg = UIO_XUIO;
5075 XUIO_XUZC_RW(xuio) = ioflag;
5076 ZFS_EXIT(zfsvfs);
5077 return (0);
5078 }
5079
5080 /*ARGSUSED*/
5081 static int
5082 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5083 {
5084 int i;
5085 arc_buf_t *abuf;
5086 int ioflag = XUIO_XUZC_RW(xuio);
5087
5088 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5089
5090 i = dmu_xuio_cnt(xuio);
5091 while (i-- > 0) {
5092 abuf = dmu_xuio_arcbuf(xuio, i);
5093 /*
5094 * if abuf == NULL, it must be a write buffer
5095 * that has been returned in zfs_write().
5096 */
5097 if (abuf)
5098 dmu_return_arcbuf(abuf);
5099 ASSERT(abuf || ioflag == UIO_WRITE);
5100 }
5101
5102 dmu_xuio_fini(xuio);
5103 return (0);
5104 }
5105
5106 /*
5107 * Predeclare these here so that the compiler assumes that
5108 * this is an "old style" function declaration that does
5109 * not include arguments => we won't get type mismatch errors
5110 * in the initializations that follow.
5111 */
5112 static int zfs_inval();
5113 static int zfs_isdir();
5114
5115 static int
5116 zfs_inval()
5117 {
5118 return (SET_ERROR(EINVAL));
5119 }
5120
5121 static int
5122 zfs_isdir()
5123 {
5124 return (SET_ERROR(EISDIR));
5125 }
5126 /*
5127 * Directory vnode operations template
5128 */
5129 vnodeops_t *zfs_dvnodeops;
5130 const fs_operation_def_t zfs_dvnodeops_template[] = {
5131 VOPNAME_OPEN, { .vop_open = zfs_open },
5132 VOPNAME_CLOSE, { .vop_close = zfs_close },
5133 VOPNAME_READ, { .error = zfs_isdir },
5134 VOPNAME_WRITE, { .error = zfs_isdir },
5135 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5136 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5137 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5138 VOPNAME_ACCESS, { .vop_access = zfs_access },
5139 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5140 VOPNAME_CREATE, { .vop_create = zfs_create },
5141 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5142 VOPNAME_LINK, { .vop_link = zfs_link },
5143 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5144 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5145 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5146 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5147 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5148 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5149 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5150 VOPNAME_FID, { .vop_fid = zfs_fid },
5151 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5152 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5153 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5154 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5155 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5156 NULL, NULL
5157 };
5158
5159 /*
5160 * Regular file vnode operations template
5161 */
5162 vnodeops_t *zfs_fvnodeops;
5163 const fs_operation_def_t zfs_fvnodeops_template[] = {
5164 VOPNAME_OPEN, { .vop_open = zfs_open },
5165 VOPNAME_CLOSE, { .vop_close = zfs_close },
5166 VOPNAME_READ, { .vop_read = zfs_read },
5167 VOPNAME_WRITE, { .vop_write = zfs_write },
5168 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5169 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5170 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5171 VOPNAME_ACCESS, { .vop_access = zfs_access },
5172 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5173 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5174 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5175 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5176 VOPNAME_FID, { .vop_fid = zfs_fid },
5177 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5178 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5179 VOPNAME_SPACE, { .vop_space = zfs_space },
5180 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5181 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5182 VOPNAME_MAP, { .vop_map = zfs_map },
5183 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5184 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5185 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5186 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5187 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5188 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5189 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5190 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5191 NULL, NULL
5192 };
5193
5194 /*
5195 * Symbolic link vnode operations template
5196 */
5197 vnodeops_t *zfs_symvnodeops;
5198 const fs_operation_def_t zfs_symvnodeops_template[] = {
5199 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5200 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5201 VOPNAME_ACCESS, { .vop_access = zfs_access },
5202 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5203 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5204 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5205 VOPNAME_FID, { .vop_fid = zfs_fid },
5206 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5207 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5208 NULL, NULL
5209 };
5210
5211 /*
5212 * special share hidden files vnode operations template
5213 */
5214 vnodeops_t *zfs_sharevnodeops;
5215 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5216 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5217 VOPNAME_ACCESS, { .vop_access = zfs_access },
5218 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5219 VOPNAME_FID, { .vop_fid = zfs_fid },
5220 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5221 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5222 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5223 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5224 NULL, NULL
5225 };
5226
5227 /*
5228 * Extended attribute directory vnode operations template
5229 *
5230 * This template is identical to the directory vnodes
5231 * operation template except for restricted operations:
5232 * VOP_MKDIR()
5233 * VOP_SYMLINK()
5234 *
5235 * Note that there are other restrictions embedded in:
5236 * zfs_create() - restrict type to VREG
5237 * zfs_link() - no links into/out of attribute space
5238 * zfs_rename() - no moves into/out of attribute space
5239 */
5240 vnodeops_t *zfs_xdvnodeops;
5241 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5242 VOPNAME_OPEN, { .vop_open = zfs_open },
5243 VOPNAME_CLOSE, { .vop_close = zfs_close },
5244 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5245 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5246 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5247 VOPNAME_ACCESS, { .vop_access = zfs_access },
5248 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5249 VOPNAME_CREATE, { .vop_create = zfs_create },
5250 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5251 VOPNAME_LINK, { .vop_link = zfs_link },
5252 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5253 VOPNAME_MKDIR, { .error = zfs_inval },
5254 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5255 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5256 VOPNAME_SYMLINK, { .error = zfs_inval },
5257 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5258 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5259 VOPNAME_FID, { .vop_fid = zfs_fid },
5260 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5261 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5262 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5263 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5264 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5265 NULL, NULL
5266 };
5267
5268 /*
5269 * Error vnode operations template
5270 */
5271 vnodeops_t *zfs_evnodeops;
5272 const fs_operation_def_t zfs_evnodeops_template[] = {
5273 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5274 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5275 NULL, NULL
5276 };