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