1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 * Copyright 2016 Joyent, Inc.
27 */
28
29 /* Portions Copyright 2007 Jeremy Teo */
30 /* Portions Copyright 2010 Robert Milkowski */
31
32 #include <sys/types.h>
33 #include <sys/param.h>
34 #include <sys/time.h>
35 #include <sys/systm.h>
36 #include <sys/sysmacros.h>
37 #include <sys/resource.h>
38 #include <sys/vfs.h>
39 #include <sys/vfs_opreg.h>
40 #include <sys/vnode.h>
41 #include <sys/file.h>
42 #include <sys/stat.h>
43 #include <sys/kmem.h>
44 #include <sys/taskq.h>
45 #include <sys/uio.h>
46 #include <sys/vmsystm.h>
47 #include <sys/atomic.h>
48 #include <sys/vm.h>
49 #include <vm/seg_vn.h>
50 #include <vm/pvn.h>
51 #include <vm/as.h>
52 #include <vm/kpm.h>
53 #include <vm/seg_kpm.h>
54 #include <sys/mman.h>
55 #include <sys/pathname.h>
56 #include <sys/cmn_err.h>
57 #include <sys/errno.h>
58 #include <sys/unistd.h>
59 #include <sys/zfs_dir.h>
60 #include <sys/zfs_acl.h>
61 #include <sys/zfs_ioctl.h>
62 #include <sys/fs/zfs.h>
63 #include <sys/dmu.h>
64 #include <sys/dmu_objset.h>
65 #include <sys/spa.h>
66 #include <sys/txg.h>
67 #include <sys/dbuf.h>
68 #include <sys/zap.h>
69 #include <sys/sa.h>
70 #include <sys/dirent.h>
71 #include <sys/policy.h>
72 #include <sys/sunddi.h>
73 #include <sys/filio.h>
74 #include <sys/sid.h>
75 #include "fs/fs_subr.h"
76 #include <sys/zfs_ctldir.h>
77 #include <sys/zfs_fuid.h>
78 #include <sys/zfs_sa.h>
79 #include <sys/dnlc.h>
80 #include <sys/zfs_rlock.h>
81 #include <sys/extdirent.h>
82 #include <sys/kidmap.h>
83 #include <sys/cred.h>
84 #include <sys/attr.h>
85
86 /*
87 * Programming rules.
88 *
89 * Each vnode op performs some logical unit of work. To do this, the ZPL must
90 * properly lock its in-core state, create a DMU transaction, do the work,
91 * record this work in the intent log (ZIL), commit the DMU transaction,
92 * and wait for the intent log to commit if it is a synchronous operation.
93 * Moreover, the vnode ops must work in both normal and log replay context.
94 * The ordering of events is important to avoid deadlocks and references
95 * to freed memory. The example below illustrates the following Big Rules:
96 *
97 * (1) A check must be made in each zfs thread for a mounted file system.
98 * This is done avoiding races using ZFS_ENTER(zfsvfs).
99 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
100 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
101 * can return EIO from the calling function.
102 *
103 * (2) VN_RELE() should always be the last thing except for zil_commit()
104 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
105 * First, if it's the last reference, the vnode/znode
106 * can be freed, so the zp may point to freed memory. Second, the last
107 * reference will call zfs_zinactive(), which may induce a lot of work --
108 * pushing cached pages (which acquires range locks) and syncing out
109 * cached atime changes. Third, zfs_zinactive() may require a new tx,
110 * which could deadlock the system if you were already holding one.
111 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
112 *
113 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
114 * as they can span dmu_tx_assign() calls.
115 *
116 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
117 * dmu_tx_assign(). This is critical because we don't want to block
118 * while holding locks.
119 *
120 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
121 * reduces lock contention and CPU usage when we must wait (note that if
122 * throughput is constrained by the storage, nearly every transaction
123 * must wait).
124 *
125 * Note, in particular, that if a lock is sometimes acquired before
126 * the tx assigns, and sometimes after (e.g. z_lock), then failing
127 * to use a non-blocking assign can deadlock the system. The scenario:
128 *
129 * Thread A has grabbed a lock before calling dmu_tx_assign().
130 * Thread B is in an already-assigned tx, and blocks for this lock.
131 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
132 * forever, because the previous txg can't quiesce until B's tx commits.
133 *
134 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
135 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
136 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
137 * to indicate that this operation has already called dmu_tx_wait().
138 * This will ensure that we don't retry forever, waiting a short bit
139 * each time.
140 *
141 * (5) If the operation succeeded, generate the intent log entry for it
142 * before dropping locks. This ensures that the ordering of events
143 * in the intent log matches the order in which they actually occurred.
144 * During ZIL replay the zfs_log_* functions will update the sequence
145 * number to indicate the zil transaction has replayed.
146 *
147 * (6) At the end of each vnode op, the DMU tx must always commit,
148 * regardless of whether there were any errors.
149 *
150 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
151 * to ensure that synchronous semantics are provided when necessary.
152 *
153 * In general, this is how things should be ordered in each vnode op:
154 *
155 * ZFS_ENTER(zfsvfs); // exit if unmounted
156 * top:
157 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
158 * rw_enter(...); // grab any other locks you need
159 * tx = dmu_tx_create(...); // get DMU tx
160 * dmu_tx_hold_*(); // hold each object you might modify
161 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
162 * if (error) {
163 * rw_exit(...); // drop locks
164 * zfs_dirent_unlock(dl); // unlock directory entry
165 * VN_RELE(...); // release held vnodes
166 * if (error == ERESTART) {
167 * waited = B_TRUE;
168 * dmu_tx_wait(tx);
169 * dmu_tx_abort(tx);
170 * goto top;
171 * }
172 * dmu_tx_abort(tx); // abort DMU tx
173 * ZFS_EXIT(zfsvfs); // finished in zfs
174 * return (error); // really out of space
175 * }
176 * error = do_real_work(); // do whatever this VOP does
177 * if (error == 0)
178 * zfs_log_*(...); // on success, make ZIL entry
179 * dmu_tx_commit(tx); // commit DMU tx -- error or not
180 * rw_exit(...); // drop locks
181 * zfs_dirent_unlock(dl); // unlock directory entry
182 * VN_RELE(...); // release held vnodes
183 * zil_commit(zilog, foid); // synchronous when necessary
184 * ZFS_EXIT(zfsvfs); // finished in zfs
185 * return (error); // done, report error
186 */
187
188 /* ARGSUSED */
189 static int
190 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
191 {
192 znode_t *zp = VTOZ(*vpp);
193 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
194
195 ZFS_ENTER(zfsvfs);
196 ZFS_VERIFY_ZP(zp);
197
198 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
199 ((flag & FAPPEND) == 0)) {
200 ZFS_EXIT(zfsvfs);
201 return (SET_ERROR(EPERM));
202 }
203
204 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
205 ZTOV(zp)->v_type == VREG &&
206 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
207 if (fs_vscan(*vpp, cr, 0) != 0) {
208 ZFS_EXIT(zfsvfs);
209 return (SET_ERROR(EACCES));
210 }
211 }
212
213 /* Keep a count of the synchronous opens in the znode */
214 if (flag & (FSYNC | FDSYNC))
215 atomic_inc_32(&zp->z_sync_cnt);
216
217 ZFS_EXIT(zfsvfs);
218 return (0);
219 }
220
221 /* ARGSUSED */
222 static int
223 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
224 caller_context_t *ct)
225 {
226 znode_t *zp = VTOZ(vp);
227 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
228
229 /*
230 * Clean up any locks held by this process on the vp.
231 */
232 cleanlocks(vp, ddi_get_pid(), 0);
233 cleanshares(vp, ddi_get_pid());
234
235 ZFS_ENTER(zfsvfs);
236 ZFS_VERIFY_ZP(zp);
237
238 /* Decrement the synchronous opens in the znode */
239 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
240 atomic_dec_32(&zp->z_sync_cnt);
241
242 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
243 ZTOV(zp)->v_type == VREG &&
244 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
245 VERIFY(fs_vscan(vp, cr, 1) == 0);
246
247 ZFS_EXIT(zfsvfs);
248 return (0);
249 }
250
251 /*
252 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
253 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
254 */
255 static int
256 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
257 {
258 znode_t *zp = VTOZ(vp);
259 uint64_t noff = (uint64_t)*off; /* new offset */
260 uint64_t file_sz;
261 int error;
262 boolean_t hole;
263
264 file_sz = zp->z_size;
265 if (noff >= file_sz) {
266 return (SET_ERROR(ENXIO));
267 }
268
269 if (cmd == _FIO_SEEK_HOLE)
270 hole = B_TRUE;
271 else
272 hole = B_FALSE;
273
274 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
275
276 if (error == ESRCH)
277 return (SET_ERROR(ENXIO));
278
279 /*
280 * We could find a hole that begins after the logical end-of-file,
281 * because dmu_offset_next() only works on whole blocks. If the
282 * EOF falls mid-block, then indicate that the "virtual hole"
283 * at the end of the file begins at the logical EOF, rather than
284 * at the end of the last block.
285 */
286 if (noff > file_sz) {
287 ASSERT(hole);
288 noff = file_sz;
289 }
290
291 if (noff < *off)
292 return (error);
293 *off = noff;
294 return (error);
295 }
296
297 /* ARGSUSED */
298 static int
299 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
300 int *rvalp, caller_context_t *ct)
301 {
302 offset_t off;
303 offset_t ndata;
304 dmu_object_info_t doi;
305 int error;
306 zfsvfs_t *zfsvfs;
307 znode_t *zp;
308
309 switch (com) {
310 case _FIOFFS:
311 {
312 return (zfs_sync(vp->v_vfsp, 0, cred));
313
314 /*
315 * The following two ioctls are used by bfu. Faking out,
316 * necessary to avoid bfu errors.
317 */
318 }
319 case _FIOGDIO:
320 case _FIOSDIO:
321 {
322 return (0);
323 }
324
325 case _FIO_SEEK_DATA:
326 case _FIO_SEEK_HOLE:
327 {
328 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
329 return (SET_ERROR(EFAULT));
330
331 zp = VTOZ(vp);
332 zfsvfs = zp->z_zfsvfs;
333 ZFS_ENTER(zfsvfs);
334 ZFS_VERIFY_ZP(zp);
335
336 /* offset parameter is in/out */
337 error = zfs_holey(vp, com, &off);
338 ZFS_EXIT(zfsvfs);
339 if (error)
340 return (error);
341 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
342 return (SET_ERROR(EFAULT));
343 return (0);
344 }
345 case _FIO_COUNT_FILLED:
346 {
347 /*
348 * _FIO_COUNT_FILLED adds a new ioctl command which
349 * exposes the number of filled blocks in a
350 * ZFS object.
351 */
352 zp = VTOZ(vp);
353 zfsvfs = zp->z_zfsvfs;
354 ZFS_ENTER(zfsvfs);
355 ZFS_VERIFY_ZP(zp);
356
357 /*
358 * Wait for all dirty blocks for this object
359 * to get synced out to disk, and the DMU info
360 * updated.
361 */
362 error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id);
363 if (error) {
364 ZFS_EXIT(zfsvfs);
365 return (error);
366 }
367
368 /*
369 * Retrieve fill count from DMU object.
370 */
371 error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi);
372 if (error) {
373 ZFS_EXIT(zfsvfs);
374 return (error);
375 }
376
377 ndata = doi.doi_fill_count;
378
379 ZFS_EXIT(zfsvfs);
380 if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag))
381 return (SET_ERROR(EFAULT));
382 return (0);
383 }
384 }
385 return (SET_ERROR(ENOTTY));
386 }
387
388 /*
389 * Utility functions to map and unmap a single physical page. These
390 * are used to manage the mappable copies of ZFS file data, and therefore
391 * do not update ref/mod bits.
392 */
393 caddr_t
394 zfs_map_page(page_t *pp, enum seg_rw rw)
395 {
396 if (kpm_enable)
397 return (hat_kpm_mapin(pp, 0));
398 ASSERT(rw == S_READ || rw == S_WRITE);
399 return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
400 (caddr_t)-1));
401 }
402
403 void
404 zfs_unmap_page(page_t *pp, caddr_t addr)
405 {
406 if (kpm_enable) {
407 hat_kpm_mapout(pp, 0, addr);
408 } else {
409 ppmapout(addr);
410 }
411 }
412
413 /*
414 * When a file is memory mapped, we must keep the IO data synchronized
415 * between the DMU cache and the memory mapped pages. What this means:
416 *
417 * On Write: If we find a memory mapped page, we write to *both*
418 * the page and the dmu buffer.
419 */
420 static void
421 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
422 {
423 int64_t off;
424
425 off = start & PAGEOFFSET;
426 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
427 page_t *pp;
428 uint64_t nbytes = MIN(PAGESIZE - off, len);
429
430 if (pp = page_lookup(vp, start, SE_SHARED)) {
431 caddr_t va;
432
433 va = zfs_map_page(pp, S_WRITE);
434 (void) dmu_read(os, oid, start+off, nbytes, va+off,
435 DMU_READ_PREFETCH);
436 zfs_unmap_page(pp, va);
437 page_unlock(pp);
438 }
439 len -= nbytes;
440 off = 0;
441 }
442 }
443
444 /*
445 * When a file is memory mapped, we must keep the IO data synchronized
446 * between the DMU cache and the memory mapped pages. What this means:
447 *
448 * On Read: We "read" preferentially from memory mapped pages,
449 * else we default from the dmu buffer.
450 *
451 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
452 * the file is memory mapped.
453 */
454 static int
455 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
456 {
457 znode_t *zp = VTOZ(vp);
458 int64_t start, off;
459 int len = nbytes;
460 int error = 0;
461
462 start = uio->uio_loffset;
463 off = start & PAGEOFFSET;
464 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
465 page_t *pp;
466 uint64_t bytes = MIN(PAGESIZE - off, len);
467
468 if (pp = page_lookup(vp, start, SE_SHARED)) {
469 caddr_t va;
470
471 va = zfs_map_page(pp, S_READ);
472 error = uiomove(va + off, bytes, UIO_READ, uio);
473 zfs_unmap_page(pp, va);
474 page_unlock(pp);
475 } else {
476 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
477 uio, bytes);
478 }
479 len -= bytes;
480 off = 0;
481 if (error)
482 break;
483 }
484 return (error);
485 }
486
487 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
488
489 /*
490 * Read bytes from specified file into supplied buffer.
491 *
492 * IN: vp - vnode of file to be read from.
493 * uio - structure supplying read location, range info,
494 * and return buffer.
495 * ioflag - SYNC flags; used to provide FRSYNC semantics.
496 * cr - credentials of caller.
497 * ct - caller context
498 *
499 * OUT: uio - updated offset and range, buffer filled.
500 *
501 * RETURN: 0 on success, error code on failure.
502 *
503 * Side Effects:
504 * vp - atime updated if byte count > 0
505 */
506 /* ARGSUSED */
507 static int
508 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
509 {
510 znode_t *zp = VTOZ(vp);
511 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
512 ssize_t n, nbytes;
513 int error = 0;
514 rl_t *rl;
515 xuio_t *xuio = NULL;
516
517 ZFS_ENTER(zfsvfs);
518 ZFS_VERIFY_ZP(zp);
519
520 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
521 ZFS_EXIT(zfsvfs);
522 return (SET_ERROR(EACCES));
523 }
524
525 /*
526 * Validate file offset
527 */
528 if (uio->uio_loffset < (offset_t)0) {
529 ZFS_EXIT(zfsvfs);
530 return (SET_ERROR(EINVAL));
531 }
532
533 /*
534 * Fasttrack empty reads
535 */
536 if (uio->uio_resid == 0) {
537 ZFS_EXIT(zfsvfs);
538 return (0);
539 }
540
541 /*
542 * Check for mandatory locks
543 */
544 if (MANDMODE(zp->z_mode)) {
545 if (error = chklock(vp, FREAD,
546 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
547 ZFS_EXIT(zfsvfs);
548 return (error);
549 }
550 }
551
552 /*
553 * If we're in FRSYNC mode, sync out this znode before reading it.
554 */
555 if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
556 zil_commit(zfsvfs->z_log, zp->z_id);
557
558 /*
559 * Lock the range against changes.
560 */
561 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
562
563 /*
564 * If we are reading past end-of-file we can skip
565 * to the end; but we might still need to set atime.
566 */
567 if (uio->uio_loffset >= zp->z_size) {
568 error = 0;
569 goto out;
570 }
571
572 ASSERT(uio->uio_loffset < zp->z_size);
573 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
574
575 if ((uio->uio_extflg == UIO_XUIO) &&
576 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
577 int nblk;
578 int blksz = zp->z_blksz;
579 uint64_t offset = uio->uio_loffset;
580
581 xuio = (xuio_t *)uio;
582 if ((ISP2(blksz))) {
583 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
584 blksz)) / blksz;
585 } else {
586 ASSERT(offset + n <= blksz);
587 nblk = 1;
588 }
589 (void) dmu_xuio_init(xuio, nblk);
590
591 if (vn_has_cached_data(vp)) {
592 /*
593 * For simplicity, we always allocate a full buffer
594 * even if we only expect to read a portion of a block.
595 */
596 while (--nblk >= 0) {
597 (void) dmu_xuio_add(xuio,
598 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
599 blksz), 0, blksz);
600 }
601 }
602 }
603
604 while (n > 0) {
605 nbytes = MIN(n, zfs_read_chunk_size -
606 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
607
608 if (vn_has_cached_data(vp)) {
609 error = mappedread(vp, nbytes, uio);
610 } else {
611 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
612 uio, nbytes);
613 }
614 if (error) {
615 /* convert checksum errors into IO errors */
616 if (error == ECKSUM)
617 error = SET_ERROR(EIO);
618 break;
619 }
620
621 n -= nbytes;
622 }
623 out:
624 zfs_range_unlock(rl);
625
626 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
627 ZFS_EXIT(zfsvfs);
628 return (error);
629 }
630
631 /*
632 * Write the bytes to a file.
633 *
634 * IN: vp - vnode of file to be written to.
635 * uio - structure supplying write location, range info,
636 * and data buffer.
637 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
638 * set if in append mode.
639 * cr - credentials of caller.
640 * ct - caller context (NFS/CIFS fem monitor only)
641 *
642 * OUT: uio - updated offset and range.
643 *
644 * RETURN: 0 on success, error code on failure.
645 *
646 * Timestamps:
647 * vp - ctime|mtime updated if byte count > 0
648 */
649
650 /* ARGSUSED */
651 static int
652 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
653 {
654 znode_t *zp = VTOZ(vp);
655 rlim64_t limit = uio->uio_llimit;
656 ssize_t start_resid = uio->uio_resid;
657 ssize_t tx_bytes;
658 uint64_t end_size;
659 dmu_tx_t *tx;
660 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
661 zilog_t *zilog;
662 offset_t woff;
663 ssize_t n, nbytes;
664 rl_t *rl;
665 int max_blksz = zfsvfs->z_max_blksz;
666 int error = 0;
667 int prev_error;
668 arc_buf_t *abuf;
669 iovec_t *aiov = NULL;
670 xuio_t *xuio = NULL;
671 int i_iov = 0;
672 int iovcnt = uio->uio_iovcnt;
673 iovec_t *iovp = uio->uio_iov;
674 int write_eof;
675 int count = 0;
676 sa_bulk_attr_t bulk[4];
677 uint64_t mtime[2], ctime[2];
678
679 /*
680 * Fasttrack empty write
681 */
682 n = start_resid;
683 if (n == 0)
684 return (0);
685
686 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
687 limit = MAXOFFSET_T;
688
689 ZFS_ENTER(zfsvfs);
690 ZFS_VERIFY_ZP(zp);
691
692 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
693 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
694 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
695 &zp->z_size, 8);
696 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
697 &zp->z_pflags, 8);
698
699 /*
700 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
701 * callers might not be able to detect properly that we are read-only,
702 * so check it explicitly here.
703 */
704 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
705 ZFS_EXIT(zfsvfs);
706 return (SET_ERROR(EROFS));
707 }
708
709 /*
710 * If immutable or not appending then return EPERM
711 */
712 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
713 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
714 (uio->uio_loffset < zp->z_size))) {
715 ZFS_EXIT(zfsvfs);
716 return (SET_ERROR(EPERM));
717 }
718
719 zilog = zfsvfs->z_log;
720
721 /*
722 * Validate file offset
723 */
724 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
725 if (woff < 0) {
726 ZFS_EXIT(zfsvfs);
727 return (SET_ERROR(EINVAL));
728 }
729
730 /*
731 * Check for mandatory locks before calling zfs_range_lock()
732 * in order to prevent a deadlock with locks set via fcntl().
733 */
734 if (MANDMODE((mode_t)zp->z_mode) &&
735 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
736 ZFS_EXIT(zfsvfs);
737 return (error);
738 }
739
740 /*
741 * Pre-fault the pages to ensure slow (eg NFS) pages
742 * don't hold up txg.
743 * Skip this if uio contains loaned arc_buf.
744 */
745 if ((uio->uio_extflg == UIO_XUIO) &&
746 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
747 xuio = (xuio_t *)uio;
748 else
749 uio_prefaultpages(MIN(n, max_blksz), uio);
750
751 /*
752 * If in append mode, set the io offset pointer to eof.
753 */
754 if (ioflag & FAPPEND) {
755 /*
756 * Obtain an appending range lock to guarantee file append
757 * semantics. We reset the write offset once we have the lock.
758 */
759 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
760 woff = rl->r_off;
761 if (rl->r_len == UINT64_MAX) {
762 /*
763 * We overlocked the file because this write will cause
764 * the file block size to increase.
765 * Note that zp_size cannot change with this lock held.
766 */
767 woff = zp->z_size;
768 }
769 uio->uio_loffset = woff;
770 } else {
771 /*
772 * Note that if the file block size will change as a result of
773 * this write, then this range lock will lock the entire file
774 * so that we can re-write the block safely.
775 */
776 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
777 }
778
779 if (woff >= limit) {
780 zfs_range_unlock(rl);
781 ZFS_EXIT(zfsvfs);
782 return (SET_ERROR(EFBIG));
783 }
784
785 if ((woff + n) > limit || woff > (limit - n))
786 n = limit - woff;
787
788 /* Will this write extend the file length? */
789 write_eof = (woff + n > zp->z_size);
790
791 end_size = MAX(zp->z_size, woff + n);
792
793 /*
794 * Write the file in reasonable size chunks. Each chunk is written
795 * in a separate transaction; this keeps the intent log records small
796 * and allows us to do more fine-grained space accounting.
797 */
798 while (n > 0) {
799 abuf = NULL;
800 woff = uio->uio_loffset;
801 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
802 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
803 if (abuf != NULL)
804 dmu_return_arcbuf(abuf);
805 error = SET_ERROR(EDQUOT);
806 break;
807 }
808
809 if (xuio && abuf == NULL) {
810 ASSERT(i_iov < iovcnt);
811 aiov = &iovp[i_iov];
812 abuf = dmu_xuio_arcbuf(xuio, i_iov);
813 dmu_xuio_clear(xuio, i_iov);
814 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
815 iovec_t *, aiov, arc_buf_t *, abuf);
816 ASSERT((aiov->iov_base == abuf->b_data) ||
817 ((char *)aiov->iov_base - (char *)abuf->b_data +
818 aiov->iov_len == arc_buf_size(abuf)));
819 i_iov++;
820 } else if (abuf == NULL && n >= max_blksz &&
821 woff >= zp->z_size &&
822 P2PHASE(woff, max_blksz) == 0 &&
823 zp->z_blksz == max_blksz) {
824 /*
825 * This write covers a full block. "Borrow" a buffer
826 * from the dmu so that we can fill it before we enter
827 * a transaction. This avoids the possibility of
828 * holding up the transaction if the data copy hangs
829 * up on a pagefault (e.g., from an NFS server mapping).
830 */
831 size_t cbytes;
832
833 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
834 max_blksz);
835 ASSERT(abuf != NULL);
836 ASSERT(arc_buf_size(abuf) == max_blksz);
837 if (error = uiocopy(abuf->b_data, max_blksz,
838 UIO_WRITE, uio, &cbytes)) {
839 dmu_return_arcbuf(abuf);
840 break;
841 }
842 ASSERT(cbytes == max_blksz);
843 }
844
845 /*
846 * Start a transaction.
847 */
848 tx = dmu_tx_create(zfsvfs->z_os);
849 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
850 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
851 zfs_sa_upgrade_txholds(tx, zp);
852 error = dmu_tx_assign(tx, TXG_WAIT);
853 if (error) {
854 dmu_tx_abort(tx);
855 if (abuf != NULL)
856 dmu_return_arcbuf(abuf);
857 break;
858 }
859
860 /*
861 * If zfs_range_lock() over-locked we grow the blocksize
862 * and then reduce the lock range. This will only happen
863 * on the first iteration since zfs_range_reduce() will
864 * shrink down r_len to the appropriate size.
865 */
866 if (rl->r_len == UINT64_MAX) {
867 uint64_t new_blksz;
868
869 if (zp->z_blksz > max_blksz) {
870 /*
871 * File's blocksize is already larger than the
872 * "recordsize" property. Only let it grow to
873 * the next power of 2.
874 */
875 ASSERT(!ISP2(zp->z_blksz));
876 new_blksz = MIN(end_size,
877 1 << highbit64(zp->z_blksz));
878 } else {
879 new_blksz = MIN(end_size, max_blksz);
880 }
881 zfs_grow_blocksize(zp, new_blksz, tx);
882 zfs_range_reduce(rl, woff, n);
883 }
884
885 /*
886 * XXX - should we really limit each write to z_max_blksz?
887 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
888 */
889 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
890
891 if (abuf == NULL) {
892 tx_bytes = uio->uio_resid;
893 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
894 uio, nbytes, tx);
895 tx_bytes -= uio->uio_resid;
896 } else {
897 tx_bytes = nbytes;
898 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
899 /*
900 * If this is not a full block write, but we are
901 * extending the file past EOF and this data starts
902 * block-aligned, use assign_arcbuf(). Otherwise,
903 * write via dmu_write().
904 */
905 if (tx_bytes < max_blksz && (!write_eof ||
906 aiov->iov_base != abuf->b_data)) {
907 ASSERT(xuio);
908 dmu_write(zfsvfs->z_os, zp->z_id, woff,
909 aiov->iov_len, aiov->iov_base, tx);
910 dmu_return_arcbuf(abuf);
911 xuio_stat_wbuf_copied();
912 } else {
913 ASSERT(xuio || tx_bytes == max_blksz);
914 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
915 woff, abuf, tx);
916 }
917 ASSERT(tx_bytes <= uio->uio_resid);
918 uioskip(uio, tx_bytes);
919 }
920 if (tx_bytes && vn_has_cached_data(vp)) {
921 update_pages(vp, woff,
922 tx_bytes, zfsvfs->z_os, zp->z_id);
923 }
924
925 /*
926 * If we made no progress, we're done. If we made even
927 * partial progress, update the znode and ZIL accordingly.
928 */
929 if (tx_bytes == 0) {
930 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
931 (void *)&zp->z_size, sizeof (uint64_t), tx);
932 dmu_tx_commit(tx);
933 ASSERT(error != 0);
934 break;
935 }
936
937 /*
938 * Clear Set-UID/Set-GID bits on successful write if not
939 * privileged and at least one of the excute bits is set.
940 *
941 * It would be nice to to this after all writes have
942 * been done, but that would still expose the ISUID/ISGID
943 * to another app after the partial write is committed.
944 *
945 * Note: we don't call zfs_fuid_map_id() here because
946 * user 0 is not an ephemeral uid.
947 */
948 mutex_enter(&zp->z_acl_lock);
949 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
950 (S_IXUSR >> 6))) != 0 &&
951 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
952 secpolicy_vnode_setid_retain(cr,
953 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
954 uint64_t newmode;
955 zp->z_mode &= ~(S_ISUID | S_ISGID);
956 newmode = zp->z_mode;
957 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
958 (void *)&newmode, sizeof (uint64_t), tx);
959 }
960 mutex_exit(&zp->z_acl_lock);
961
962 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
963 B_TRUE);
964
965 /*
966 * Update the file size (zp_size) if it has changed;
967 * account for possible concurrent updates.
968 */
969 while ((end_size = zp->z_size) < uio->uio_loffset) {
970 (void) atomic_cas_64(&zp->z_size, end_size,
971 uio->uio_loffset);
972 }
973 /*
974 * If we are replaying and eof is non zero then force
975 * the file size to the specified eof. Note, there's no
976 * concurrency during replay.
977 */
978 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
979 zp->z_size = zfsvfs->z_replay_eof;
980
981 /*
982 * Keep track of a possible pre-existing error from a partial
983 * write via dmu_write_uio_dbuf above.
984 */
985 prev_error = error;
986 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
987
988 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
989 dmu_tx_commit(tx);
990
991 if (prev_error != 0 || error != 0)
992 break;
993 ASSERT(tx_bytes == nbytes);
994 n -= nbytes;
995
996 if (!xuio && n > 0)
997 uio_prefaultpages(MIN(n, max_blksz), uio);
998 }
999
1000 zfs_range_unlock(rl);
1001
1002 /*
1003 * If we're in replay mode, or we made no progress, return error.
1004 * Otherwise, it's at least a partial write, so it's successful.
1005 */
1006 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1007 ZFS_EXIT(zfsvfs);
1008 return (error);
1009 }
1010
1011 if (ioflag & (FSYNC | FDSYNC) ||
1012 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1013 zil_commit(zilog, zp->z_id);
1014
1015 ZFS_EXIT(zfsvfs);
1016 return (0);
1017 }
1018
1019 void
1020 zfs_get_done(zgd_t *zgd, int error)
1021 {
1022 znode_t *zp = zgd->zgd_private;
1023 objset_t *os = zp->z_zfsvfs->z_os;
1024
1025 if (zgd->zgd_db)
1026 dmu_buf_rele(zgd->zgd_db, zgd);
1027
1028 zfs_range_unlock(zgd->zgd_rl);
1029
1030 /*
1031 * Release the vnode asynchronously as we currently have the
1032 * txg stopped from syncing.
1033 */
1034 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1035
1036 if (error == 0 && zgd->zgd_bp)
1037 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1038
1039 kmem_free(zgd, sizeof (zgd_t));
1040 }
1041
1042 #ifdef DEBUG
1043 static int zil_fault_io = 0;
1044 #endif
1045
1046 /*
1047 * Get data to generate a TX_WRITE intent log record.
1048 */
1049 int
1050 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1051 {
1052 zfsvfs_t *zfsvfs = arg;
1053 objset_t *os = zfsvfs->z_os;
1054 znode_t *zp;
1055 uint64_t object = lr->lr_foid;
1056 uint64_t offset = lr->lr_offset;
1057 uint64_t size = lr->lr_length;
1058 blkptr_t *bp = &lr->lr_blkptr;
1059 dmu_buf_t *db;
1060 zgd_t *zgd;
1061 int error = 0;
1062
1063 ASSERT(zio != NULL);
1064 ASSERT(size != 0);
1065
1066 /*
1067 * Nothing to do if the file has been removed
1068 */
1069 if (zfs_zget(zfsvfs, object, &zp) != 0)
1070 return (SET_ERROR(ENOENT));
1071 if (zp->z_unlinked) {
1072 /*
1073 * Release the vnode asynchronously as we currently have the
1074 * txg stopped from syncing.
1075 */
1076 VN_RELE_ASYNC(ZTOV(zp),
1077 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1078 return (SET_ERROR(ENOENT));
1079 }
1080
1081 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1082 zgd->zgd_zilog = zfsvfs->z_log;
1083 zgd->zgd_private = zp;
1084
1085 /*
1086 * Write records come in two flavors: immediate and indirect.
1087 * For small writes it's cheaper to store the data with the
1088 * log record (immediate); for large writes it's cheaper to
1089 * sync the data and get a pointer to it (indirect) so that
1090 * we don't have to write the data twice.
1091 */
1092 if (buf != NULL) { /* immediate write */
1093 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1094 /* test for truncation needs to be done while range locked */
1095 if (offset >= zp->z_size) {
1096 error = SET_ERROR(ENOENT);
1097 } else {
1098 error = dmu_read(os, object, offset, size, buf,
1099 DMU_READ_NO_PREFETCH);
1100 }
1101 ASSERT(error == 0 || error == ENOENT);
1102 } else { /* indirect write */
1103 /*
1104 * Have to lock the whole block to ensure when it's
1105 * written out and it's checksum is being calculated
1106 * that no one can change the data. We need to re-check
1107 * blocksize after we get the lock in case it's changed!
1108 */
1109 for (;;) {
1110 uint64_t blkoff;
1111 size = zp->z_blksz;
1112 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1113 offset -= blkoff;
1114 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1115 RL_READER);
1116 if (zp->z_blksz == size)
1117 break;
1118 offset += blkoff;
1119 zfs_range_unlock(zgd->zgd_rl);
1120 }
1121 /* test for truncation needs to be done while range locked */
1122 if (lr->lr_offset >= zp->z_size)
1123 error = SET_ERROR(ENOENT);
1124 #ifdef DEBUG
1125 if (zil_fault_io) {
1126 error = SET_ERROR(EIO);
1127 zil_fault_io = 0;
1128 }
1129 #endif
1130 if (error == 0)
1131 error = dmu_buf_hold(os, object, offset, zgd, &db,
1132 DMU_READ_NO_PREFETCH);
1133
1134 if (error == 0) {
1135 blkptr_t *obp = dmu_buf_get_blkptr(db);
1136 if (obp) {
1137 ASSERT(BP_IS_HOLE(bp));
1138 *bp = *obp;
1139 }
1140
1141 zgd->zgd_db = db;
1142 zgd->zgd_bp = bp;
1143
1144 ASSERT(db->db_offset == offset);
1145 ASSERT(db->db_size == size);
1146
1147 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1148 zfs_get_done, zgd);
1149 ASSERT(error || lr->lr_length <= zp->z_blksz);
1150
1151 /*
1152 * On success, we need to wait for the write I/O
1153 * initiated by dmu_sync() to complete before we can
1154 * release this dbuf. We will finish everything up
1155 * in the zfs_get_done() callback.
1156 */
1157 if (error == 0)
1158 return (0);
1159
1160 if (error == EALREADY) {
1161 lr->lr_common.lrc_txtype = TX_WRITE2;
1162 error = 0;
1163 }
1164 }
1165 }
1166
1167 zfs_get_done(zgd, error);
1168
1169 return (error);
1170 }
1171
1172 /*ARGSUSED*/
1173 static int
1174 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1175 caller_context_t *ct)
1176 {
1177 znode_t *zp = VTOZ(vp);
1178 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1179 int error;
1180
1181 ZFS_ENTER(zfsvfs);
1182 ZFS_VERIFY_ZP(zp);
1183
1184 if (flag & V_ACE_MASK)
1185 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1186 else
1187 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1188
1189 ZFS_EXIT(zfsvfs);
1190 return (error);
1191 }
1192
1193 /*
1194 * If vnode is for a device return a specfs vnode instead.
1195 */
1196 static int
1197 specvp_check(vnode_t **vpp, cred_t *cr)
1198 {
1199 int error = 0;
1200
1201 if (IS_DEVVP(*vpp)) {
1202 struct vnode *svp;
1203
1204 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1205 VN_RELE(*vpp);
1206 if (svp == NULL)
1207 error = SET_ERROR(ENOSYS);
1208 *vpp = svp;
1209 }
1210 return (error);
1211 }
1212
1213
1214 /*
1215 * Lookup an entry in a directory, or an extended attribute directory.
1216 * If it exists, return a held vnode reference for it.
1217 *
1218 * IN: dvp - vnode of directory to search.
1219 * nm - name of entry to lookup.
1220 * pnp - full pathname to lookup [UNUSED].
1221 * flags - LOOKUP_XATTR set if looking for an attribute.
1222 * rdir - root directory vnode [UNUSED].
1223 * cr - credentials of caller.
1224 * ct - caller context
1225 * direntflags - directory lookup flags
1226 * realpnp - returned pathname.
1227 *
1228 * OUT: vpp - vnode of located entry, NULL if not found.
1229 *
1230 * RETURN: 0 on success, error code on failure.
1231 *
1232 * Timestamps:
1233 * NA
1234 */
1235 /* ARGSUSED */
1236 static int
1237 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1238 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
1239 int *direntflags, pathname_t *realpnp)
1240 {
1241 znode_t *zdp = VTOZ(dvp);
1242 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1243 int error = 0;
1244
1245 /* fast path */
1246 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1247
1248 if (dvp->v_type != VDIR) {
1249 return (SET_ERROR(ENOTDIR));
1250 } else if (zdp->z_sa_hdl == NULL) {
1251 return (SET_ERROR(EIO));
1252 }
1253
1254 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1255 error = zfs_fastaccesschk_execute(zdp, cr);
1256 if (!error) {
1257 *vpp = dvp;
1258 VN_HOLD(*vpp);
1259 return (0);
1260 }
1261 return (error);
1262 } else {
1263 vnode_t *tvp = dnlc_lookup(dvp, nm);
1264
1265 if (tvp) {
1266 error = zfs_fastaccesschk_execute(zdp, cr);
1267 if (error) {
1268 VN_RELE(tvp);
1269 return (error);
1270 }
1271 if (tvp == DNLC_NO_VNODE) {
1272 VN_RELE(tvp);
1273 return (SET_ERROR(ENOENT));
1274 } else {
1275 *vpp = tvp;
1276 return (specvp_check(vpp, cr));
1277 }
1278 }
1279 }
1280 }
1281
1282 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1283
1284 ZFS_ENTER(zfsvfs);
1285 ZFS_VERIFY_ZP(zdp);
1286
1287 *vpp = NULL;
1288
1289 if (flags & LOOKUP_XATTR) {
1290 /*
1291 * If the xattr property is off, refuse the lookup request.
1292 */
1293 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1294 ZFS_EXIT(zfsvfs);
1295 return (SET_ERROR(EINVAL));
1296 }
1297
1298 /*
1299 * We don't allow recursive attributes..
1300 * Maybe someday we will.
1301 */
1302 if (zdp->z_pflags & ZFS_XATTR) {
1303 ZFS_EXIT(zfsvfs);
1304 return (SET_ERROR(EINVAL));
1305 }
1306
1307 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1308 ZFS_EXIT(zfsvfs);
1309 return (error);
1310 }
1311
1312 /*
1313 * Do we have permission to get into attribute directory?
1314 */
1315
1316 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1317 B_FALSE, cr)) {
1318 VN_RELE(*vpp);
1319 *vpp = NULL;
1320 }
1321
1322 ZFS_EXIT(zfsvfs);
1323 return (error);
1324 }
1325
1326 if (dvp->v_type != VDIR) {
1327 ZFS_EXIT(zfsvfs);
1328 return (SET_ERROR(ENOTDIR));
1329 }
1330
1331 /*
1332 * Check accessibility of directory.
1333 */
1334
1335 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1336 ZFS_EXIT(zfsvfs);
1337 return (error);
1338 }
1339
1340 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1341 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1342 ZFS_EXIT(zfsvfs);
1343 return (SET_ERROR(EILSEQ));
1344 }
1345
1346 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1347 if (error == 0)
1348 error = specvp_check(vpp, cr);
1349
1350 ZFS_EXIT(zfsvfs);
1351 return (error);
1352 }
1353
1354 /*
1355 * Attempt to create a new entry in a directory. If the entry
1356 * already exists, truncate the file if permissible, else return
1357 * an error. Return the vp of the created or trunc'd file.
1358 *
1359 * IN: dvp - vnode of directory to put new file entry in.
1360 * name - name of new file entry.
1361 * vap - attributes of new file.
1362 * excl - flag indicating exclusive or non-exclusive mode.
1363 * mode - mode to open file with.
1364 * cr - credentials of caller.
1365 * flag - large file flag [UNUSED].
1366 * ct - caller context
1367 * vsecp - ACL to be set
1368 *
1369 * OUT: vpp - vnode of created or trunc'd entry.
1370 *
1371 * RETURN: 0 on success, error code on failure.
1372 *
1373 * Timestamps:
1374 * dvp - ctime|mtime updated if new entry created
1375 * vp - ctime|mtime always, atime if new
1376 */
1377
1378 /* ARGSUSED */
1379 static int
1380 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1381 int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1382 vsecattr_t *vsecp)
1383 {
1384 znode_t *zp, *dzp = VTOZ(dvp);
1385 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1386 zilog_t *zilog;
1387 objset_t *os;
1388 zfs_dirlock_t *dl;
1389 dmu_tx_t *tx;
1390 int error;
1391 ksid_t *ksid;
1392 uid_t uid;
1393 gid_t gid = crgetgid(cr);
1394 zfs_acl_ids_t acl_ids;
1395 boolean_t fuid_dirtied;
1396 boolean_t have_acl = B_FALSE;
1397 boolean_t waited = B_FALSE;
1398
1399 /*
1400 * If we have an ephemeral id, ACL, or XVATTR then
1401 * make sure file system is at proper version
1402 */
1403
1404 ksid = crgetsid(cr, KSID_OWNER);
1405 if (ksid)
1406 uid = ksid_getid(ksid);
1407 else
1408 uid = crgetuid(cr);
1409
1410 if (zfsvfs->z_use_fuids == B_FALSE &&
1411 (vsecp || (vap->va_mask & AT_XVATTR) ||
1412 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1413 return (SET_ERROR(EINVAL));
1414
1415 ZFS_ENTER(zfsvfs);
1416 ZFS_VERIFY_ZP(dzp);
1417 os = zfsvfs->z_os;
1418 zilog = zfsvfs->z_log;
1419
1420 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1421 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1422 ZFS_EXIT(zfsvfs);
1423 return (SET_ERROR(EILSEQ));
1424 }
1425
1426 if (vap->va_mask & AT_XVATTR) {
1427 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1428 crgetuid(cr), cr, vap->va_type)) != 0) {
1429 ZFS_EXIT(zfsvfs);
1430 return (error);
1431 }
1432 }
1433 top:
1434 *vpp = NULL;
1435
1436 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1437 vap->va_mode &= ~VSVTX;
1438
1439 if (*name == '\0') {
1440 /*
1441 * Null component name refers to the directory itself.
1442 */
1443 VN_HOLD(dvp);
1444 zp = dzp;
1445 dl = NULL;
1446 error = 0;
1447 } else {
1448 /* possible VN_HOLD(zp) */
1449 int zflg = 0;
1450
1451 if (flag & FIGNORECASE)
1452 zflg |= ZCILOOK;
1453
1454 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1455 NULL, NULL);
1456 if (error) {
1457 if (have_acl)
1458 zfs_acl_ids_free(&acl_ids);
1459 if (strcmp(name, "..") == 0)
1460 error = SET_ERROR(EISDIR);
1461 ZFS_EXIT(zfsvfs);
1462 return (error);
1463 }
1464 }
1465
1466 if (zp == NULL) {
1467 uint64_t txtype;
1468
1469 /*
1470 * Create a new file object and update the directory
1471 * to reference it.
1472 */
1473 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1474 if (have_acl)
1475 zfs_acl_ids_free(&acl_ids);
1476 goto out;
1477 }
1478
1479 /*
1480 * We only support the creation of regular files in
1481 * extended attribute directories.
1482 */
1483
1484 if ((dzp->z_pflags & ZFS_XATTR) &&
1485 (vap->va_type != VREG)) {
1486 if (have_acl)
1487 zfs_acl_ids_free(&acl_ids);
1488 error = SET_ERROR(EINVAL);
1489 goto out;
1490 }
1491
1492 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1493 cr, vsecp, &acl_ids)) != 0)
1494 goto out;
1495 have_acl = B_TRUE;
1496
1497 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1498 zfs_acl_ids_free(&acl_ids);
1499 error = SET_ERROR(EDQUOT);
1500 goto out;
1501 }
1502
1503 tx = dmu_tx_create(os);
1504
1505 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1506 ZFS_SA_BASE_ATTR_SIZE);
1507
1508 fuid_dirtied = zfsvfs->z_fuid_dirty;
1509 if (fuid_dirtied)
1510 zfs_fuid_txhold(zfsvfs, tx);
1511 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1512 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1513 if (!zfsvfs->z_use_sa &&
1514 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1515 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1516 0, acl_ids.z_aclp->z_acl_bytes);
1517 }
1518 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1519 if (error) {
1520 zfs_dirent_unlock(dl);
1521 if (error == ERESTART) {
1522 waited = B_TRUE;
1523 dmu_tx_wait(tx);
1524 dmu_tx_abort(tx);
1525 goto top;
1526 }
1527 zfs_acl_ids_free(&acl_ids);
1528 dmu_tx_abort(tx);
1529 ZFS_EXIT(zfsvfs);
1530 return (error);
1531 }
1532 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1533
1534 if (fuid_dirtied)
1535 zfs_fuid_sync(zfsvfs, tx);
1536
1537 (void) zfs_link_create(dl, zp, tx, ZNEW);
1538 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1539 if (flag & FIGNORECASE)
1540 txtype |= TX_CI;
1541 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1542 vsecp, acl_ids.z_fuidp, vap);
1543 zfs_acl_ids_free(&acl_ids);
1544 dmu_tx_commit(tx);
1545 } else {
1546 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1547
1548 if (have_acl)
1549 zfs_acl_ids_free(&acl_ids);
1550 have_acl = B_FALSE;
1551
1552 /*
1553 * A directory entry already exists for this name.
1554 */
1555 /*
1556 * Can't truncate an existing file if in exclusive mode.
1557 */
1558 if (excl == EXCL) {
1559 error = SET_ERROR(EEXIST);
1560 goto out;
1561 }
1562 /*
1563 * Can't open a directory for writing.
1564 */
1565 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1566 error = SET_ERROR(EISDIR);
1567 goto out;
1568 }
1569 /*
1570 * Verify requested access to file.
1571 */
1572 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1573 goto out;
1574 }
1575
1576 mutex_enter(&dzp->z_lock);
1577 dzp->z_seq++;
1578 mutex_exit(&dzp->z_lock);
1579
1580 /*
1581 * Truncate regular files if requested.
1582 */
1583 if ((ZTOV(zp)->v_type == VREG) &&
1584 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1585 /* we can't hold any locks when calling zfs_freesp() */
1586 zfs_dirent_unlock(dl);
1587 dl = NULL;
1588 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1589 if (error == 0) {
1590 vnevent_create(ZTOV(zp), ct);
1591 }
1592 }
1593 }
1594 out:
1595
1596 if (dl)
1597 zfs_dirent_unlock(dl);
1598
1599 if (error) {
1600 if (zp)
1601 VN_RELE(ZTOV(zp));
1602 } else {
1603 *vpp = ZTOV(zp);
1604 error = specvp_check(vpp, cr);
1605 }
1606
1607 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1608 zil_commit(zilog, 0);
1609
1610 ZFS_EXIT(zfsvfs);
1611 return (error);
1612 }
1613
1614 /*
1615 * Remove an entry from a directory.
1616 *
1617 * IN: dvp - vnode of directory to remove entry from.
1618 * name - name of entry to remove.
1619 * cr - credentials of caller.
1620 * ct - caller context
1621 * flags - case flags
1622 *
1623 * RETURN: 0 on success, error code on failure.
1624 *
1625 * Timestamps:
1626 * dvp - ctime|mtime
1627 * vp - ctime (if nlink > 0)
1628 */
1629
1630 uint64_t null_xattr = 0;
1631
1632 /*ARGSUSED*/
1633 static int
1634 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1635 int flags)
1636 {
1637 znode_t *zp, *dzp = VTOZ(dvp);
1638 znode_t *xzp;
1639 vnode_t *vp;
1640 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1641 zilog_t *zilog;
1642 uint64_t acl_obj, xattr_obj;
1643 uint64_t xattr_obj_unlinked = 0;
1644 uint64_t obj = 0;
1645 zfs_dirlock_t *dl;
1646 dmu_tx_t *tx;
1647 boolean_t may_delete_now, delete_now = FALSE;
1648 boolean_t unlinked, toobig = FALSE;
1649 uint64_t txtype;
1650 pathname_t *realnmp = NULL;
1651 pathname_t realnm;
1652 int error;
1653 int zflg = ZEXISTS;
1654 boolean_t waited = B_FALSE;
1655
1656 ZFS_ENTER(zfsvfs);
1657 ZFS_VERIFY_ZP(dzp);
1658 zilog = zfsvfs->z_log;
1659
1660 if (flags & FIGNORECASE) {
1661 zflg |= ZCILOOK;
1662 pn_alloc(&realnm);
1663 realnmp = &realnm;
1664 }
1665
1666 top:
1667 xattr_obj = 0;
1668 xzp = NULL;
1669 /*
1670 * Attempt to lock directory; fail if entry doesn't exist.
1671 */
1672 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1673 NULL, realnmp)) {
1674 if (realnmp)
1675 pn_free(realnmp);
1676 ZFS_EXIT(zfsvfs);
1677 return (error);
1678 }
1679
1680 vp = ZTOV(zp);
1681
1682 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1683 goto out;
1684 }
1685
1686 /*
1687 * Need to use rmdir for removing directories.
1688 */
1689 if (vp->v_type == VDIR) {
1690 error = SET_ERROR(EPERM);
1691 goto out;
1692 }
1693
1694 vnevent_remove(vp, dvp, name, ct);
1695
1696 if (realnmp)
1697 dnlc_remove(dvp, realnmp->pn_buf);
1698 else
1699 dnlc_remove(dvp, name);
1700
1701 mutex_enter(&vp->v_lock);
1702 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1703 mutex_exit(&vp->v_lock);
1704
1705 /*
1706 * We may delete the znode now, or we may put it in the unlinked set;
1707 * it depends on whether we're the last link, and on whether there are
1708 * other holds on the vnode. So we dmu_tx_hold() the right things to
1709 * allow for either case.
1710 */
1711 obj = zp->z_id;
1712 tx = dmu_tx_create(zfsvfs->z_os);
1713 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1714 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1715 zfs_sa_upgrade_txholds(tx, zp);
1716 zfs_sa_upgrade_txholds(tx, dzp);
1717 if (may_delete_now) {
1718 toobig =
1719 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1720 /* if the file is too big, only hold_free a token amount */
1721 dmu_tx_hold_free(tx, zp->z_id, 0,
1722 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1723 }
1724
1725 /* are there any extended attributes? */
1726 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1727 &xattr_obj, sizeof (xattr_obj));
1728 if (error == 0 && xattr_obj) {
1729 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1730 ASSERT0(error);
1731 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1732 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1733 }
1734
1735 mutex_enter(&zp->z_lock);
1736 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1737 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1738 mutex_exit(&zp->z_lock);
1739
1740 /* charge as an update -- would be nice not to charge at all */
1741 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1742
1743 /*
1744 * Mark this transaction as typically resulting in a net free of space
1745 */
1746 dmu_tx_mark_netfree(tx);
1747
1748 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1749 if (error) {
1750 zfs_dirent_unlock(dl);
1751 VN_RELE(vp);
1752 if (xzp)
1753 VN_RELE(ZTOV(xzp));
1754 if (error == ERESTART) {
1755 waited = B_TRUE;
1756 dmu_tx_wait(tx);
1757 dmu_tx_abort(tx);
1758 goto top;
1759 }
1760 if (realnmp)
1761 pn_free(realnmp);
1762 dmu_tx_abort(tx);
1763 ZFS_EXIT(zfsvfs);
1764 return (error);
1765 }
1766
1767 /*
1768 * Remove the directory entry.
1769 */
1770 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1771
1772 if (error) {
1773 dmu_tx_commit(tx);
1774 goto out;
1775 }
1776
1777 if (unlinked) {
1778 /*
1779 * Hold z_lock so that we can make sure that the ACL obj
1780 * hasn't changed. Could have been deleted due to
1781 * zfs_sa_upgrade().
1782 */
1783 mutex_enter(&zp->z_lock);
1784 mutex_enter(&vp->v_lock);
1785 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1786 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1787 delete_now = may_delete_now && !toobig &&
1788 vp->v_count == 1 && !vn_has_cached_data(vp) &&
1789 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1790 acl_obj;
1791 mutex_exit(&vp->v_lock);
1792 }
1793
1794 if (delete_now) {
1795 if (xattr_obj_unlinked) {
1796 ASSERT3U(xzp->z_links, ==, 2);
1797 mutex_enter(&xzp->z_lock);
1798 xzp->z_unlinked = 1;
1799 xzp->z_links = 0;
1800 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1801 &xzp->z_links, sizeof (xzp->z_links), tx);
1802 ASSERT3U(error, ==, 0);
1803 mutex_exit(&xzp->z_lock);
1804 zfs_unlinked_add(xzp, tx);
1805
1806 if (zp->z_is_sa)
1807 error = sa_remove(zp->z_sa_hdl,
1808 SA_ZPL_XATTR(zfsvfs), tx);
1809 else
1810 error = sa_update(zp->z_sa_hdl,
1811 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1812 sizeof (uint64_t), tx);
1813 ASSERT0(error);
1814 }
1815 mutex_enter(&vp->v_lock);
1816 vp->v_count--;
1817 ASSERT0(vp->v_count);
1818 mutex_exit(&vp->v_lock);
1819 mutex_exit(&zp->z_lock);
1820 zfs_znode_delete(zp, tx);
1821 } else if (unlinked) {
1822 mutex_exit(&zp->z_lock);
1823 zfs_unlinked_add(zp, tx);
1824 }
1825
1826 txtype = TX_REMOVE;
1827 if (flags & FIGNORECASE)
1828 txtype |= TX_CI;
1829 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1830
1831 dmu_tx_commit(tx);
1832 out:
1833 if (realnmp)
1834 pn_free(realnmp);
1835
1836 zfs_dirent_unlock(dl);
1837
1838 if (!delete_now)
1839 VN_RELE(vp);
1840 if (xzp)
1841 VN_RELE(ZTOV(xzp));
1842
1843 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1844 zil_commit(zilog, 0);
1845
1846 ZFS_EXIT(zfsvfs);
1847 return (error);
1848 }
1849
1850 /*
1851 * Create a new directory and insert it into dvp using the name
1852 * provided. Return a pointer to the inserted directory.
1853 *
1854 * IN: dvp - vnode of directory to add subdir to.
1855 * dirname - name of new directory.
1856 * vap - attributes of new directory.
1857 * cr - credentials of caller.
1858 * ct - caller context
1859 * flags - case flags
1860 * vsecp - ACL to be set
1861 *
1862 * OUT: vpp - vnode of created directory.
1863 *
1864 * RETURN: 0 on success, error code on failure.
1865 *
1866 * Timestamps:
1867 * dvp - ctime|mtime updated
1868 * vp - ctime|mtime|atime updated
1869 */
1870 /*ARGSUSED*/
1871 static int
1872 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1873 caller_context_t *ct, int flags, vsecattr_t *vsecp)
1874 {
1875 znode_t *zp, *dzp = VTOZ(dvp);
1876 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1877 zilog_t *zilog;
1878 zfs_dirlock_t *dl;
1879 uint64_t txtype;
1880 dmu_tx_t *tx;
1881 int error;
1882 int zf = ZNEW;
1883 ksid_t *ksid;
1884 uid_t uid;
1885 gid_t gid = crgetgid(cr);
1886 zfs_acl_ids_t acl_ids;
1887 boolean_t fuid_dirtied;
1888 boolean_t waited = B_FALSE;
1889
1890 ASSERT(vap->va_type == VDIR);
1891
1892 /*
1893 * If we have an ephemeral id, ACL, or XVATTR then
1894 * make sure file system is at proper version
1895 */
1896
1897 ksid = crgetsid(cr, KSID_OWNER);
1898 if (ksid)
1899 uid = ksid_getid(ksid);
1900 else
1901 uid = crgetuid(cr);
1902 if (zfsvfs->z_use_fuids == B_FALSE &&
1903 (vsecp || (vap->va_mask & AT_XVATTR) ||
1904 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1905 return (SET_ERROR(EINVAL));
1906
1907 ZFS_ENTER(zfsvfs);
1908 ZFS_VERIFY_ZP(dzp);
1909 zilog = zfsvfs->z_log;
1910
1911 if (dzp->z_pflags & ZFS_XATTR) {
1912 ZFS_EXIT(zfsvfs);
1913 return (SET_ERROR(EINVAL));
1914 }
1915
1916 if (zfsvfs->z_utf8 && u8_validate(dirname,
1917 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1918 ZFS_EXIT(zfsvfs);
1919 return (SET_ERROR(EILSEQ));
1920 }
1921 if (flags & FIGNORECASE)
1922 zf |= ZCILOOK;
1923
1924 if (vap->va_mask & AT_XVATTR) {
1925 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1926 crgetuid(cr), cr, vap->va_type)) != 0) {
1927 ZFS_EXIT(zfsvfs);
1928 return (error);
1929 }
1930 }
1931
1932 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1933 vsecp, &acl_ids)) != 0) {
1934 ZFS_EXIT(zfsvfs);
1935 return (error);
1936 }
1937 /*
1938 * First make sure the new directory doesn't exist.
1939 *
1940 * Existence is checked first to make sure we don't return
1941 * EACCES instead of EEXIST which can cause some applications
1942 * to fail.
1943 */
1944 top:
1945 *vpp = NULL;
1946
1947 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1948 NULL, NULL)) {
1949 zfs_acl_ids_free(&acl_ids);
1950 ZFS_EXIT(zfsvfs);
1951 return (error);
1952 }
1953
1954 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1955 zfs_acl_ids_free(&acl_ids);
1956 zfs_dirent_unlock(dl);
1957 ZFS_EXIT(zfsvfs);
1958 return (error);
1959 }
1960
1961 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1962 zfs_acl_ids_free(&acl_ids);
1963 zfs_dirent_unlock(dl);
1964 ZFS_EXIT(zfsvfs);
1965 return (SET_ERROR(EDQUOT));
1966 }
1967
1968 /*
1969 * Add a new entry to the directory.
1970 */
1971 tx = dmu_tx_create(zfsvfs->z_os);
1972 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1973 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1974 fuid_dirtied = zfsvfs->z_fuid_dirty;
1975 if (fuid_dirtied)
1976 zfs_fuid_txhold(zfsvfs, tx);
1977 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1978 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1979 acl_ids.z_aclp->z_acl_bytes);
1980 }
1981
1982 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1983 ZFS_SA_BASE_ATTR_SIZE);
1984
1985 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1986 if (error) {
1987 zfs_dirent_unlock(dl);
1988 if (error == ERESTART) {
1989 waited = B_TRUE;
1990 dmu_tx_wait(tx);
1991 dmu_tx_abort(tx);
1992 goto top;
1993 }
1994 zfs_acl_ids_free(&acl_ids);
1995 dmu_tx_abort(tx);
1996 ZFS_EXIT(zfsvfs);
1997 return (error);
1998 }
1999
2000 /*
2001 * Create new node.
2002 */
2003 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2004
2005 if (fuid_dirtied)
2006 zfs_fuid_sync(zfsvfs, tx);
2007
2008 /*
2009 * Now put new name in parent dir.
2010 */
2011 (void) zfs_link_create(dl, zp, tx, ZNEW);
2012
2013 *vpp = ZTOV(zp);
2014
2015 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2016 if (flags & FIGNORECASE)
2017 txtype |= TX_CI;
2018 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2019 acl_ids.z_fuidp, vap);
2020
2021 zfs_acl_ids_free(&acl_ids);
2022
2023 dmu_tx_commit(tx);
2024
2025 zfs_dirent_unlock(dl);
2026
2027 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2028 zil_commit(zilog, 0);
2029
2030 ZFS_EXIT(zfsvfs);
2031 return (0);
2032 }
2033
2034 /*
2035 * Remove a directory subdir entry. If the current working
2036 * directory is the same as the subdir to be removed, the
2037 * remove will fail.
2038 *
2039 * IN: dvp - vnode of directory to remove from.
2040 * name - name of directory to be removed.
2041 * cwd - vnode of current working directory.
2042 * cr - credentials of caller.
2043 * ct - caller context
2044 * flags - case flags
2045 *
2046 * RETURN: 0 on success, error code on failure.
2047 *
2048 * Timestamps:
2049 * dvp - ctime|mtime updated
2050 */
2051 /*ARGSUSED*/
2052 static int
2053 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2054 caller_context_t *ct, int flags)
2055 {
2056 znode_t *dzp = VTOZ(dvp);
2057 znode_t *zp;
2058 vnode_t *vp;
2059 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2060 zilog_t *zilog;
2061 zfs_dirlock_t *dl;
2062 dmu_tx_t *tx;
2063 int error;
2064 int zflg = ZEXISTS;
2065 boolean_t waited = B_FALSE;
2066
2067 ZFS_ENTER(zfsvfs);
2068 ZFS_VERIFY_ZP(dzp);
2069 zilog = zfsvfs->z_log;
2070
2071 if (flags & FIGNORECASE)
2072 zflg |= ZCILOOK;
2073 top:
2074 zp = NULL;
2075
2076 /*
2077 * Attempt to lock directory; fail if entry doesn't exist.
2078 */
2079 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2080 NULL, NULL)) {
2081 ZFS_EXIT(zfsvfs);
2082 return (error);
2083 }
2084
2085 vp = ZTOV(zp);
2086
2087 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2088 goto out;
2089 }
2090
2091 if (vp->v_type != VDIR) {
2092 error = SET_ERROR(ENOTDIR);
2093 goto out;
2094 }
2095
2096 if (vp == cwd) {
2097 error = SET_ERROR(EINVAL);
2098 goto out;
2099 }
2100
2101 vnevent_rmdir(vp, dvp, name, ct);
2102
2103 /*
2104 * Grab a lock on the directory to make sure that noone is
2105 * trying to add (or lookup) entries while we are removing it.
2106 */
2107 rw_enter(&zp->z_name_lock, RW_WRITER);
2108
2109 /*
2110 * Grab a lock on the parent pointer to make sure we play well
2111 * with the treewalk and directory rename code.
2112 */
2113 rw_enter(&zp->z_parent_lock, RW_WRITER);
2114
2115 tx = dmu_tx_create(zfsvfs->z_os);
2116 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2117 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2118 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2119 zfs_sa_upgrade_txholds(tx, zp);
2120 zfs_sa_upgrade_txholds(tx, dzp);
2121 dmu_tx_mark_netfree(tx);
2122 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2123 if (error) {
2124 rw_exit(&zp->z_parent_lock);
2125 rw_exit(&zp->z_name_lock);
2126 zfs_dirent_unlock(dl);
2127 VN_RELE(vp);
2128 if (error == ERESTART) {
2129 waited = B_TRUE;
2130 dmu_tx_wait(tx);
2131 dmu_tx_abort(tx);
2132 goto top;
2133 }
2134 dmu_tx_abort(tx);
2135 ZFS_EXIT(zfsvfs);
2136 return (error);
2137 }
2138
2139 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2140
2141 if (error == 0) {
2142 uint64_t txtype = TX_RMDIR;
2143 if (flags & FIGNORECASE)
2144 txtype |= TX_CI;
2145 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2146 }
2147
2148 dmu_tx_commit(tx);
2149
2150 rw_exit(&zp->z_parent_lock);
2151 rw_exit(&zp->z_name_lock);
2152 out:
2153 zfs_dirent_unlock(dl);
2154
2155 VN_RELE(vp);
2156
2157 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2158 zil_commit(zilog, 0);
2159
2160 ZFS_EXIT(zfsvfs);
2161 return (error);
2162 }
2163
2164 /*
2165 * Read as many directory entries as will fit into the provided
2166 * buffer from the given directory cursor position (specified in
2167 * the uio structure).
2168 *
2169 * IN: vp - vnode of directory to read.
2170 * uio - structure supplying read location, range info,
2171 * and return buffer.
2172 * cr - credentials of caller.
2173 * ct - caller context
2174 * flags - case flags
2175 *
2176 * OUT: uio - updated offset and range, buffer filled.
2177 * eofp - set to true if end-of-file detected.
2178 *
2179 * RETURN: 0 on success, error code on failure.
2180 *
2181 * Timestamps:
2182 * vp - atime updated
2183 *
2184 * Note that the low 4 bits of the cookie returned by zap is always zero.
2185 * This allows us to use the low range for "special" directory entries:
2186 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2187 * we use the offset 2 for the '.zfs' directory.
2188 */
2189 /* ARGSUSED */
2190 static int
2191 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2192 caller_context_t *ct, int flags)
2193 {
2194 znode_t *zp = VTOZ(vp);
2195 iovec_t *iovp;
2196 edirent_t *eodp;
2197 dirent64_t *odp;
2198 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2199 objset_t *os;
2200 caddr_t outbuf;
2201 size_t bufsize;
2202 zap_cursor_t zc;
2203 zap_attribute_t zap;
2204 uint_t bytes_wanted;
2205 uint64_t offset; /* must be unsigned; checks for < 1 */
2206 uint64_t parent;
2207 int local_eof;
2208 int outcount;
2209 int error;
2210 uint8_t prefetch;
2211 boolean_t check_sysattrs;
2212
2213 ZFS_ENTER(zfsvfs);
2214 ZFS_VERIFY_ZP(zp);
2215
2216 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2217 &parent, sizeof (parent))) != 0) {
2218 ZFS_EXIT(zfsvfs);
2219 return (error);
2220 }
2221
2222 /*
2223 * If we are not given an eof variable,
2224 * use a local one.
2225 */
2226 if (eofp == NULL)
2227 eofp = &local_eof;
2228
2229 /*
2230 * Check for valid iov_len.
2231 */
2232 if (uio->uio_iov->iov_len <= 0) {
2233 ZFS_EXIT(zfsvfs);
2234 return (SET_ERROR(EINVAL));
2235 }
2236
2237 /*
2238 * Quit if directory has been removed (posix)
2239 */
2240 if ((*eofp = zp->z_unlinked) != 0) {
2241 ZFS_EXIT(zfsvfs);
2242 return (0);
2243 }
2244
2245 error = 0;
2246 os = zfsvfs->z_os;
2247 offset = uio->uio_loffset;
2248 prefetch = zp->z_zn_prefetch;
2249
2250 /*
2251 * Initialize the iterator cursor.
2252 */
2253 if (offset <= 3) {
2254 /*
2255 * Start iteration from the beginning of the directory.
2256 */
2257 zap_cursor_init(&zc, os, zp->z_id);
2258 } else {
2259 /*
2260 * The offset is a serialized cursor.
2261 */
2262 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2263 }
2264
2265 /*
2266 * Get space to change directory entries into fs independent format.
2267 */
2268 iovp = uio->uio_iov;
2269 bytes_wanted = iovp->iov_len;
2270 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2271 bufsize = bytes_wanted;
2272 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2273 odp = (struct dirent64 *)outbuf;
2274 } else {
2275 bufsize = bytes_wanted;
2276 outbuf = NULL;
2277 odp = (struct dirent64 *)iovp->iov_base;
2278 }
2279 eodp = (struct edirent *)odp;
2280
2281 /*
2282 * If this VFS supports the system attribute view interface; and
2283 * we're looking at an extended attribute directory; and we care
2284 * about normalization conflicts on this vfs; then we must check
2285 * for normalization conflicts with the sysattr name space.
2286 */
2287 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2288 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2289 (flags & V_RDDIR_ENTFLAGS);
2290
2291 /*
2292 * Transform to file-system independent format
2293 */
2294 outcount = 0;
2295 while (outcount < bytes_wanted) {
2296 ino64_t objnum;
2297 ushort_t reclen;
2298 off64_t *next = NULL;
2299
2300 /*
2301 * Special case `.', `..', and `.zfs'.
2302 */
2303 if (offset == 0) {
2304 (void) strcpy(zap.za_name, ".");
2305 zap.za_normalization_conflict = 0;
2306 objnum = zp->z_id;
2307 } else if (offset == 1) {
2308 (void) strcpy(zap.za_name, "..");
2309 zap.za_normalization_conflict = 0;
2310 objnum = parent;
2311 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2312 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2313 zap.za_normalization_conflict = 0;
2314 objnum = ZFSCTL_INO_ROOT;
2315 } else {
2316 /*
2317 * Grab next entry.
2318 */
2319 if (error = zap_cursor_retrieve(&zc, &zap)) {
2320 if ((*eofp = (error == ENOENT)) != 0)
2321 break;
2322 else
2323 goto update;
2324 }
2325
2326 if (zap.za_integer_length != 8 ||
2327 zap.za_num_integers != 1) {
2328 cmn_err(CE_WARN, "zap_readdir: bad directory "
2329 "entry, obj = %lld, offset = %lld\n",
2330 (u_longlong_t)zp->z_id,
2331 (u_longlong_t)offset);
2332 error = SET_ERROR(ENXIO);
2333 goto update;
2334 }
2335
2336 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2337 /*
2338 * MacOS X can extract the object type here such as:
2339 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2340 */
2341
2342 if (check_sysattrs && !zap.za_normalization_conflict) {
2343 zap.za_normalization_conflict =
2344 xattr_sysattr_casechk(zap.za_name);
2345 }
2346 }
2347
2348 if (flags & V_RDDIR_ACCFILTER) {
2349 /*
2350 * If we have no access at all, don't include
2351 * this entry in the returned information
2352 */
2353 znode_t *ezp;
2354 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2355 goto skip_entry;
2356 if (!zfs_has_access(ezp, cr)) {
2357 VN_RELE(ZTOV(ezp));
2358 goto skip_entry;
2359 }
2360 VN_RELE(ZTOV(ezp));
2361 }
2362
2363 if (flags & V_RDDIR_ENTFLAGS)
2364 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2365 else
2366 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2367
2368 /*
2369 * Will this entry fit in the buffer?
2370 */
2371 if (outcount + reclen > bufsize) {
2372 /*
2373 * Did we manage to fit anything in the buffer?
2374 */
2375 if (!outcount) {
2376 error = SET_ERROR(EINVAL);
2377 goto update;
2378 }
2379 break;
2380 }
2381 if (flags & V_RDDIR_ENTFLAGS) {
2382 /*
2383 * Add extended flag entry:
2384 */
2385 eodp->ed_ino = objnum;
2386 eodp->ed_reclen = reclen;
2387 /* NOTE: ed_off is the offset for the *next* entry */
2388 next = &(eodp->ed_off);
2389 eodp->ed_eflags = zap.za_normalization_conflict ?
2390 ED_CASE_CONFLICT : 0;
2391 (void) strncpy(eodp->ed_name, zap.za_name,
2392 EDIRENT_NAMELEN(reclen));
2393 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2394 } else {
2395 /*
2396 * Add normal entry:
2397 */
2398 odp->d_ino = objnum;
2399 odp->d_reclen = reclen;
2400 /* NOTE: d_off is the offset for the *next* entry */
2401 next = &(odp->d_off);
2402 (void) strncpy(odp->d_name, zap.za_name,
2403 DIRENT64_NAMELEN(reclen));
2404 odp = (dirent64_t *)((intptr_t)odp + reclen);
2405 }
2406 outcount += reclen;
2407
2408 ASSERT(outcount <= bufsize);
2409
2410 /* Prefetch znode */
2411 if (prefetch)
2412 dmu_prefetch(os, objnum, 0, 0, 0,
2413 ZIO_PRIORITY_SYNC_READ);
2414
2415 skip_entry:
2416 /*
2417 * Move to the next entry, fill in the previous offset.
2418 */
2419 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2420 zap_cursor_advance(&zc);
2421 offset = zap_cursor_serialize(&zc);
2422 } else {
2423 offset += 1;
2424 }
2425 if (next)
2426 *next = offset;
2427 }
2428 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2429
2430 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2431 iovp->iov_base += outcount;
2432 iovp->iov_len -= outcount;
2433 uio->uio_resid -= outcount;
2434 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2435 /*
2436 * Reset the pointer.
2437 */
2438 offset = uio->uio_loffset;
2439 }
2440
2441 update:
2442 zap_cursor_fini(&zc);
2443 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2444 kmem_free(outbuf, bufsize);
2445
2446 if (error == ENOENT)
2447 error = 0;
2448
2449 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2450
2451 uio->uio_loffset = offset;
2452 ZFS_EXIT(zfsvfs);
2453 return (error);
2454 }
2455
2456 ulong_t zfs_fsync_sync_cnt = 4;
2457
2458 static int
2459 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2460 {
2461 znode_t *zp = VTOZ(vp);
2462 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2463
2464 /*
2465 * Regardless of whether this is required for standards conformance,
2466 * this is the logical behavior when fsync() is called on a file with
2467 * dirty pages. We use B_ASYNC since the ZIL transactions are already
2468 * going to be pushed out as part of the zil_commit().
2469 */
2470 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2471 (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2472 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2473
2474 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2475
2476 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2477 ZFS_ENTER(zfsvfs);
2478 ZFS_VERIFY_ZP(zp);
2479 zil_commit(zfsvfs->z_log, zp->z_id);
2480 ZFS_EXIT(zfsvfs);
2481 }
2482 return (0);
2483 }
2484
2485
2486 /*
2487 * Get the requested file attributes and place them in the provided
2488 * vattr structure.
2489 *
2490 * IN: vp - vnode of file.
2491 * vap - va_mask identifies requested attributes.
2492 * If AT_XVATTR set, then optional attrs are requested
2493 * flags - ATTR_NOACLCHECK (CIFS server context)
2494 * cr - credentials of caller.
2495 * ct - caller context
2496 *
2497 * OUT: vap - attribute values.
2498 *
2499 * RETURN: 0 (always succeeds).
2500 */
2501 /* ARGSUSED */
2502 static int
2503 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2504 caller_context_t *ct)
2505 {
2506 znode_t *zp = VTOZ(vp);
2507 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2508 int error = 0;
2509 uint64_t links;
2510 uint64_t mtime[2], ctime[2];
2511 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2512 xoptattr_t *xoap = NULL;
2513 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2514 sa_bulk_attr_t bulk[2];
2515 int count = 0;
2516
2517 ZFS_ENTER(zfsvfs);
2518 ZFS_VERIFY_ZP(zp);
2519
2520 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2521
2522 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2523 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2524
2525 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2526 ZFS_EXIT(zfsvfs);
2527 return (error);
2528 }
2529
2530 /*
2531 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2532 * Also, if we are the owner don't bother, since owner should
2533 * always be allowed to read basic attributes of file.
2534 */
2535 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2536 (vap->va_uid != crgetuid(cr))) {
2537 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2538 skipaclchk, cr)) {
2539 ZFS_EXIT(zfsvfs);
2540 return (error);
2541 }
2542 }
2543
2544 /*
2545 * Return all attributes. It's cheaper to provide the answer
2546 * than to determine whether we were asked the question.
2547 */
2548
2549 mutex_enter(&zp->z_lock);
2550 vap->va_type = vp->v_type;
2551 vap->va_mode = zp->z_mode & MODEMASK;
2552 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2553 vap->va_nodeid = zp->z_id;
2554 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2555 links = zp->z_links + 1;
2556 else
2557 links = zp->z_links;
2558 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */
2559 vap->va_size = zp->z_size;
2560 vap->va_rdev = vp->v_rdev;
2561 vap->va_seq = zp->z_seq;
2562
2563 /*
2564 * Add in any requested optional attributes and the create time.
2565 * Also set the corresponding bits in the returned attribute bitmap.
2566 */
2567 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2568 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2569 xoap->xoa_archive =
2570 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2571 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2572 }
2573
2574 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2575 xoap->xoa_readonly =
2576 ((zp->z_pflags & ZFS_READONLY) != 0);
2577 XVA_SET_RTN(xvap, XAT_READONLY);
2578 }
2579
2580 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2581 xoap->xoa_system =
2582 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2583 XVA_SET_RTN(xvap, XAT_SYSTEM);
2584 }
2585
2586 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2587 xoap->xoa_hidden =
2588 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2589 XVA_SET_RTN(xvap, XAT_HIDDEN);
2590 }
2591
2592 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2593 xoap->xoa_nounlink =
2594 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2595 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2596 }
2597
2598 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2599 xoap->xoa_immutable =
2600 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2601 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2602 }
2603
2604 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2605 xoap->xoa_appendonly =
2606 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2607 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2608 }
2609
2610 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2611 xoap->xoa_nodump =
2612 ((zp->z_pflags & ZFS_NODUMP) != 0);
2613 XVA_SET_RTN(xvap, XAT_NODUMP);
2614 }
2615
2616 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2617 xoap->xoa_opaque =
2618 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2619 XVA_SET_RTN(xvap, XAT_OPAQUE);
2620 }
2621
2622 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2623 xoap->xoa_av_quarantined =
2624 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2625 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2626 }
2627
2628 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2629 xoap->xoa_av_modified =
2630 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2631 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2632 }
2633
2634 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2635 vp->v_type == VREG) {
2636 zfs_sa_get_scanstamp(zp, xvap);
2637 }
2638
2639 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2640 uint64_t times[2];
2641
2642 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2643 times, sizeof (times));
2644 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2645 XVA_SET_RTN(xvap, XAT_CREATETIME);
2646 }
2647
2648 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2649 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2650 XVA_SET_RTN(xvap, XAT_REPARSE);
2651 }
2652 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2653 xoap->xoa_generation = zp->z_gen;
2654 XVA_SET_RTN(xvap, XAT_GEN);
2655 }
2656
2657 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2658 xoap->xoa_offline =
2659 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2660 XVA_SET_RTN(xvap, XAT_OFFLINE);
2661 }
2662
2663 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2664 xoap->xoa_sparse =
2665 ((zp->z_pflags & ZFS_SPARSE) != 0);
2666 XVA_SET_RTN(xvap, XAT_SPARSE);
2667 }
2668 }
2669
2670 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2671 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2672 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2673
2674 mutex_exit(&zp->z_lock);
2675
2676 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2677
2678 if (zp->z_blksz == 0) {
2679 /*
2680 * Block size hasn't been set; suggest maximal I/O transfers.
2681 */
2682 vap->va_blksize = zfsvfs->z_max_blksz;
2683 }
2684
2685 ZFS_EXIT(zfsvfs);
2686 return (0);
2687 }
2688
2689 /*
2690 * Set the file attributes to the values contained in the
2691 * vattr structure.
2692 *
2693 * IN: vp - vnode of file to be modified.
2694 * vap - new attribute values.
2695 * If AT_XVATTR set, then optional attrs are being set
2696 * flags - ATTR_UTIME set if non-default time values provided.
2697 * - ATTR_NOACLCHECK (CIFS context only).
2698 * cr - credentials of caller.
2699 * ct - caller context
2700 *
2701 * RETURN: 0 on success, error code on failure.
2702 *
2703 * Timestamps:
2704 * vp - ctime updated, mtime updated if size changed.
2705 */
2706 /* ARGSUSED */
2707 static int
2708 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2709 caller_context_t *ct)
2710 {
2711 znode_t *zp = VTOZ(vp);
2712 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2713 zilog_t *zilog;
2714 dmu_tx_t *tx;
2715 vattr_t oldva;
2716 xvattr_t tmpxvattr;
2717 uint_t mask = vap->va_mask;
2718 uint_t saved_mask = 0;
2719 int trim_mask = 0;
2720 uint64_t new_mode;
2721 uint64_t new_uid, new_gid;
2722 uint64_t xattr_obj;
2723 uint64_t mtime[2], ctime[2];
2724 znode_t *attrzp;
2725 int need_policy = FALSE;
2726 int err, err2;
2727 zfs_fuid_info_t *fuidp = NULL;
2728 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2729 xoptattr_t *xoap;
2730 zfs_acl_t *aclp;
2731 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2732 boolean_t fuid_dirtied = B_FALSE;
2733 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2734 int count = 0, xattr_count = 0;
2735
2736 if (mask == 0)
2737 return (0);
2738
2739 if (mask & AT_NOSET)
2740 return (SET_ERROR(EINVAL));
2741
2742 ZFS_ENTER(zfsvfs);
2743 ZFS_VERIFY_ZP(zp);
2744
2745 zilog = zfsvfs->z_log;
2746
2747 /*
2748 * Make sure that if we have ephemeral uid/gid or xvattr specified
2749 * that file system is at proper version level
2750 */
2751
2752 if (zfsvfs->z_use_fuids == B_FALSE &&
2753 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2754 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2755 (mask & AT_XVATTR))) {
2756 ZFS_EXIT(zfsvfs);
2757 return (SET_ERROR(EINVAL));
2758 }
2759
2760 if (mask & AT_SIZE && vp->v_type == VDIR) {
2761 ZFS_EXIT(zfsvfs);
2762 return (SET_ERROR(EISDIR));
2763 }
2764
2765 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2766 ZFS_EXIT(zfsvfs);
2767 return (SET_ERROR(EINVAL));
2768 }
2769
2770 /*
2771 * If this is an xvattr_t, then get a pointer to the structure of
2772 * optional attributes. If this is NULL, then we have a vattr_t.
2773 */
2774 xoap = xva_getxoptattr(xvap);
2775
2776 xva_init(&tmpxvattr);
2777
2778 /*
2779 * Immutable files can only alter immutable bit and atime
2780 */
2781 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2782 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2783 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2784 ZFS_EXIT(zfsvfs);
2785 return (SET_ERROR(EPERM));
2786 }
2787
2788 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2789 ZFS_EXIT(zfsvfs);
2790 return (SET_ERROR(EPERM));
2791 }
2792
2793 /*
2794 * Verify timestamps doesn't overflow 32 bits.
2795 * ZFS can handle large timestamps, but 32bit syscalls can't
2796 * handle times greater than 2039. This check should be removed
2797 * once large timestamps are fully supported.
2798 */
2799 if (mask & (AT_ATIME | AT_MTIME)) {
2800 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2801 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2802 ZFS_EXIT(zfsvfs);
2803 return (SET_ERROR(EOVERFLOW));
2804 }
2805 }
2806
2807 top:
2808 attrzp = NULL;
2809 aclp = NULL;
2810
2811 /* Can this be moved to before the top label? */
2812 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2813 ZFS_EXIT(zfsvfs);
2814 return (SET_ERROR(EROFS));
2815 }
2816
2817 /*
2818 * First validate permissions
2819 */
2820
2821 if (mask & AT_SIZE) {
2822 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2823 if (err) {
2824 ZFS_EXIT(zfsvfs);
2825 return (err);
2826 }
2827 /*
2828 * XXX - Note, we are not providing any open
2829 * mode flags here (like FNDELAY), so we may
2830 * block if there are locks present... this
2831 * should be addressed in openat().
2832 */
2833 /* XXX - would it be OK to generate a log record here? */
2834 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2835 if (err) {
2836 ZFS_EXIT(zfsvfs);
2837 return (err);
2838 }
2839
2840 if (vap->va_size == 0) {
2841 vnevent_truncate(ZTOV(zp), ct);
2842 } else {
2843 vnevent_resize(ZTOV(zp), ct);
2844 }
2845 }
2846
2847 if (mask & (AT_ATIME|AT_MTIME) ||
2848 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2849 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2850 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2851 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2852 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2853 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2854 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2855 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2856 skipaclchk, cr);
2857 }
2858
2859 if (mask & (AT_UID|AT_GID)) {
2860 int idmask = (mask & (AT_UID|AT_GID));
2861 int take_owner;
2862 int take_group;
2863
2864 /*
2865 * NOTE: even if a new mode is being set,
2866 * we may clear S_ISUID/S_ISGID bits.
2867 */
2868
2869 if (!(mask & AT_MODE))
2870 vap->va_mode = zp->z_mode;
2871
2872 /*
2873 * Take ownership or chgrp to group we are a member of
2874 */
2875
2876 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2877 take_group = (mask & AT_GID) &&
2878 zfs_groupmember(zfsvfs, vap->va_gid, cr);
2879
2880 /*
2881 * If both AT_UID and AT_GID are set then take_owner and
2882 * take_group must both be set in order to allow taking
2883 * ownership.
2884 *
2885 * Otherwise, send the check through secpolicy_vnode_setattr()
2886 *
2887 */
2888
2889 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2890 ((idmask == AT_UID) && take_owner) ||
2891 ((idmask == AT_GID) && take_group)) {
2892 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2893 skipaclchk, cr) == 0) {
2894 /*
2895 * Remove setuid/setgid for non-privileged users
2896 */
2897 secpolicy_setid_clear(vap, cr);
2898 trim_mask = (mask & (AT_UID|AT_GID));
2899 } else {
2900 need_policy = TRUE;
2901 }
2902 } else {
2903 need_policy = TRUE;
2904 }
2905 }
2906
2907 mutex_enter(&zp->z_lock);
2908 oldva.va_mode = zp->z_mode;
2909 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2910 if (mask & AT_XVATTR) {
2911 /*
2912 * Update xvattr mask to include only those attributes
2913 * that are actually changing.
2914 *
2915 * the bits will be restored prior to actually setting
2916 * the attributes so the caller thinks they were set.
2917 */
2918 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2919 if (xoap->xoa_appendonly !=
2920 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2921 need_policy = TRUE;
2922 } else {
2923 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2924 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2925 }
2926 }
2927
2928 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2929 if (xoap->xoa_nounlink !=
2930 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2931 need_policy = TRUE;
2932 } else {
2933 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2934 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2935 }
2936 }
2937
2938 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2939 if (xoap->xoa_immutable !=
2940 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2941 need_policy = TRUE;
2942 } else {
2943 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2944 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2945 }
2946 }
2947
2948 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2949 if (xoap->xoa_nodump !=
2950 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2951 need_policy = TRUE;
2952 } else {
2953 XVA_CLR_REQ(xvap, XAT_NODUMP);
2954 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2955 }
2956 }
2957
2958 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2959 if (xoap->xoa_av_modified !=
2960 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2961 need_policy = TRUE;
2962 } else {
2963 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2964 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2965 }
2966 }
2967
2968 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2969 if ((vp->v_type != VREG &&
2970 xoap->xoa_av_quarantined) ||
2971 xoap->xoa_av_quarantined !=
2972 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2973 need_policy = TRUE;
2974 } else {
2975 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2976 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2977 }
2978 }
2979
2980 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2981 mutex_exit(&zp->z_lock);
2982 ZFS_EXIT(zfsvfs);
2983 return (SET_ERROR(EPERM));
2984 }
2985
2986 if (need_policy == FALSE &&
2987 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2988 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2989 need_policy = TRUE;
2990 }
2991 }
2992
2993 mutex_exit(&zp->z_lock);
2994
2995 if (mask & AT_MODE) {
2996 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2997 err = secpolicy_setid_setsticky_clear(vp, vap,
2998 &oldva, cr);
2999 if (err) {
3000 ZFS_EXIT(zfsvfs);
3001 return (err);
3002 }
3003 trim_mask |= AT_MODE;
3004 } else {
3005 need_policy = TRUE;
3006 }
3007 }
3008
3009 if (need_policy) {
3010 /*
3011 * If trim_mask is set then take ownership
3012 * has been granted or write_acl is present and user
3013 * has the ability to modify mode. In that case remove
3014 * UID|GID and or MODE from mask so that
3015 * secpolicy_vnode_setattr() doesn't revoke it.
3016 */
3017
3018 if (trim_mask) {
3019 saved_mask = vap->va_mask;
3020 vap->va_mask &= ~trim_mask;
3021 }
3022 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3023 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3024 if (err) {
3025 ZFS_EXIT(zfsvfs);
3026 return (err);
3027 }
3028
3029 if (trim_mask)
3030 vap->va_mask |= saved_mask;
3031 }
3032
3033 /*
3034 * secpolicy_vnode_setattr, or take ownership may have
3035 * changed va_mask
3036 */
3037 mask = vap->va_mask;
3038
3039 if ((mask & (AT_UID | AT_GID))) {
3040 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3041 &xattr_obj, sizeof (xattr_obj));
3042
3043 if (err == 0 && xattr_obj) {
3044 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3045 if (err)
3046 goto out2;
3047 }
3048 if (mask & AT_UID) {
3049 new_uid = zfs_fuid_create(zfsvfs,
3050 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3051 if (new_uid != zp->z_uid &&
3052 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3053 if (attrzp)
3054 VN_RELE(ZTOV(attrzp));
3055 err = SET_ERROR(EDQUOT);
3056 goto out2;
3057 }
3058 }
3059
3060 if (mask & AT_GID) {
3061 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3062 cr, ZFS_GROUP, &fuidp);
3063 if (new_gid != zp->z_gid &&
3064 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3065 if (attrzp)
3066 VN_RELE(ZTOV(attrzp));
3067 err = SET_ERROR(EDQUOT);
3068 goto out2;
3069 }
3070 }
3071 }
3072 tx = dmu_tx_create(zfsvfs->z_os);
3073
3074 if (mask & AT_MODE) {
3075 uint64_t pmode = zp->z_mode;
3076 uint64_t acl_obj;
3077 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3078
3079 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3080 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3081 err = SET_ERROR(EPERM);
3082 goto out;
3083 }
3084
3085 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3086 goto out;
3087
3088 mutex_enter(&zp->z_lock);
3089 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3090 /*
3091 * Are we upgrading ACL from old V0 format
3092 * to V1 format?
3093 */
3094 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3095 zfs_znode_acl_version(zp) ==
3096 ZFS_ACL_VERSION_INITIAL) {
3097 dmu_tx_hold_free(tx, acl_obj, 0,
3098 DMU_OBJECT_END);
3099 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3100 0, aclp->z_acl_bytes);
3101 } else {
3102 dmu_tx_hold_write(tx, acl_obj, 0,
3103 aclp->z_acl_bytes);
3104 }
3105 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3106 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3107 0, aclp->z_acl_bytes);
3108 }
3109 mutex_exit(&zp->z_lock);
3110 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3111 } else {
3112 if ((mask & AT_XVATTR) &&
3113 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3114 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3115 else
3116 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3117 }
3118
3119 if (attrzp) {
3120 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3121 }
3122
3123 fuid_dirtied = zfsvfs->z_fuid_dirty;
3124 if (fuid_dirtied)
3125 zfs_fuid_txhold(zfsvfs, tx);
3126
3127 zfs_sa_upgrade_txholds(tx, zp);
3128
3129 err = dmu_tx_assign(tx, TXG_WAIT);
3130 if (err)
3131 goto out;
3132
3133 count = 0;
3134 /*
3135 * Set each attribute requested.
3136 * We group settings according to the locks they need to acquire.
3137 *
3138 * Note: you cannot set ctime directly, although it will be
3139 * updated as a side-effect of calling this function.
3140 */
3141
3142
3143 if (mask & (AT_UID|AT_GID|AT_MODE))
3144 mutex_enter(&zp->z_acl_lock);
3145 mutex_enter(&zp->z_lock);
3146
3147 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3148 &zp->z_pflags, sizeof (zp->z_pflags));
3149
3150 if (attrzp) {
3151 if (mask & (AT_UID|AT_GID|AT_MODE))
3152 mutex_enter(&attrzp->z_acl_lock);
3153 mutex_enter(&attrzp->z_lock);
3154 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3155 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3156 sizeof (attrzp->z_pflags));
3157 }
3158
3159 if (mask & (AT_UID|AT_GID)) {
3160
3161 if (mask & AT_UID) {
3162 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3163 &new_uid, sizeof (new_uid));
3164 zp->z_uid = new_uid;
3165 if (attrzp) {
3166 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3167 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3168 sizeof (new_uid));
3169 attrzp->z_uid = new_uid;
3170 }
3171 }
3172
3173 if (mask & AT_GID) {
3174 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3175 NULL, &new_gid, sizeof (new_gid));
3176 zp->z_gid = new_gid;
3177 if (attrzp) {
3178 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3179 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3180 sizeof (new_gid));
3181 attrzp->z_gid = new_gid;
3182 }
3183 }
3184 if (!(mask & AT_MODE)) {
3185 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3186 NULL, &new_mode, sizeof (new_mode));
3187 new_mode = zp->z_mode;
3188 }
3189 err = zfs_acl_chown_setattr(zp);
3190 ASSERT(err == 0);
3191 if (attrzp) {
3192 err = zfs_acl_chown_setattr(attrzp);
3193 ASSERT(err == 0);
3194 }
3195 }
3196
3197 if (mask & AT_MODE) {
3198 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3199 &new_mode, sizeof (new_mode));
3200 zp->z_mode = new_mode;
3201 ASSERT3U((uintptr_t)aclp, !=, NULL);
3202 err = zfs_aclset_common(zp, aclp, cr, tx);
3203 ASSERT0(err);
3204 if (zp->z_acl_cached)
3205 zfs_acl_free(zp->z_acl_cached);
3206 zp->z_acl_cached = aclp;
3207 aclp = NULL;
3208 }
3209
3210
3211 if (mask & AT_ATIME) {
3212 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3213 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3214 &zp->z_atime, sizeof (zp->z_atime));
3215 }
3216
3217 if (mask & AT_MTIME) {
3218 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3219 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3220 mtime, sizeof (mtime));
3221 }
3222
3223 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3224 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3225 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3226 NULL, mtime, sizeof (mtime));
3227 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3228 &ctime, sizeof (ctime));
3229 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3230 B_TRUE);
3231 } else if (mask != 0) {
3232 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3233 &ctime, sizeof (ctime));
3234 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3235 B_TRUE);
3236 if (attrzp) {
3237 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3238 SA_ZPL_CTIME(zfsvfs), NULL,
3239 &ctime, sizeof (ctime));
3240 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3241 mtime, ctime, B_TRUE);
3242 }
3243 }
3244 /*
3245 * Do this after setting timestamps to prevent timestamp
3246 * update from toggling bit
3247 */
3248
3249 if (xoap && (mask & AT_XVATTR)) {
3250
3251 /*
3252 * restore trimmed off masks
3253 * so that return masks can be set for caller.
3254 */
3255
3256 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3257 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3258 }
3259 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3260 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3261 }
3262 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3263 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3264 }
3265 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3266 XVA_SET_REQ(xvap, XAT_NODUMP);
3267 }
3268 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3269 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3270 }
3271 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3272 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3273 }
3274
3275 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3276 ASSERT(vp->v_type == VREG);
3277
3278 zfs_xvattr_set(zp, xvap, tx);
3279 }
3280
3281 if (fuid_dirtied)
3282 zfs_fuid_sync(zfsvfs, tx);
3283
3284 if (mask != 0)
3285 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3286
3287 mutex_exit(&zp->z_lock);
3288 if (mask & (AT_UID|AT_GID|AT_MODE))
3289 mutex_exit(&zp->z_acl_lock);
3290
3291 if (attrzp) {
3292 if (mask & (AT_UID|AT_GID|AT_MODE))
3293 mutex_exit(&attrzp->z_acl_lock);
3294 mutex_exit(&attrzp->z_lock);
3295 }
3296 out:
3297 if (err == 0 && attrzp) {
3298 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3299 xattr_count, tx);
3300 ASSERT(err2 == 0);
3301 }
3302
3303 if (attrzp)
3304 VN_RELE(ZTOV(attrzp));
3305
3306 if (aclp)
3307 zfs_acl_free(aclp);
3308
3309 if (fuidp) {
3310 zfs_fuid_info_free(fuidp);
3311 fuidp = NULL;
3312 }
3313
3314 if (err) {
3315 dmu_tx_abort(tx);
3316 if (err == ERESTART)
3317 goto top;
3318 } else {
3319 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3320 dmu_tx_commit(tx);
3321 }
3322
3323 out2:
3324 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3325 zil_commit(zilog, 0);
3326
3327 ZFS_EXIT(zfsvfs);
3328 return (err);
3329 }
3330
3331 typedef struct zfs_zlock {
3332 krwlock_t *zl_rwlock; /* lock we acquired */
3333 znode_t *zl_znode; /* znode we held */
3334 struct zfs_zlock *zl_next; /* next in list */
3335 } zfs_zlock_t;
3336
3337 /*
3338 * Drop locks and release vnodes that were held by zfs_rename_lock().
3339 */
3340 static void
3341 zfs_rename_unlock(zfs_zlock_t **zlpp)
3342 {
3343 zfs_zlock_t *zl;
3344
3345 while ((zl = *zlpp) != NULL) {
3346 if (zl->zl_znode != NULL)
3347 VN_RELE(ZTOV(zl->zl_znode));
3348 rw_exit(zl->zl_rwlock);
3349 *zlpp = zl->zl_next;
3350 kmem_free(zl, sizeof (*zl));
3351 }
3352 }
3353
3354 /*
3355 * Search back through the directory tree, using the ".." entries.
3356 * Lock each directory in the chain to prevent concurrent renames.
3357 * Fail any attempt to move a directory into one of its own descendants.
3358 * XXX - z_parent_lock can overlap with map or grow locks
3359 */
3360 static int
3361 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3362 {
3363 zfs_zlock_t *zl;
3364 znode_t *zp = tdzp;
3365 uint64_t rootid = zp->z_zfsvfs->z_root;
3366 uint64_t oidp = zp->z_id;
3367 krwlock_t *rwlp = &szp->z_parent_lock;
3368 krw_t rw = RW_WRITER;
3369
3370 /*
3371 * First pass write-locks szp and compares to zp->z_id.
3372 * Later passes read-lock zp and compare to zp->z_parent.
3373 */
3374 do {
3375 if (!rw_tryenter(rwlp, rw)) {
3376 /*
3377 * Another thread is renaming in this path.
3378 * Note that if we are a WRITER, we don't have any
3379 * parent_locks held yet.
3380 */
3381 if (rw == RW_READER && zp->z_id > szp->z_id) {
3382 /*
3383 * Drop our locks and restart
3384 */
3385 zfs_rename_unlock(&zl);
3386 *zlpp = NULL;
3387 zp = tdzp;
3388 oidp = zp->z_id;
3389 rwlp = &szp->z_parent_lock;
3390 rw = RW_WRITER;
3391 continue;
3392 } else {
3393 /*
3394 * Wait for other thread to drop its locks
3395 */
3396 rw_enter(rwlp, rw);
3397 }
3398 }
3399
3400 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3401 zl->zl_rwlock = rwlp;
3402 zl->zl_znode = NULL;
3403 zl->zl_next = *zlpp;
3404 *zlpp = zl;
3405
3406 if (oidp == szp->z_id) /* We're a descendant of szp */
3407 return (SET_ERROR(EINVAL));
3408
3409 if (oidp == rootid) /* We've hit the top */
3410 return (0);
3411
3412 if (rw == RW_READER) { /* i.e. not the first pass */
3413 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3414 if (error)
3415 return (error);
3416 zl->zl_znode = zp;
3417 }
3418 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3419 &oidp, sizeof (oidp));
3420 rwlp = &zp->z_parent_lock;
3421 rw = RW_READER;
3422
3423 } while (zp->z_id != sdzp->z_id);
3424
3425 return (0);
3426 }
3427
3428 /*
3429 * Move an entry from the provided source directory to the target
3430 * directory. Change the entry name as indicated.
3431 *
3432 * IN: sdvp - Source directory containing the "old entry".
3433 * snm - Old entry name.
3434 * tdvp - Target directory to contain the "new entry".
3435 * tnm - New entry name.
3436 * cr - credentials of caller.
3437 * ct - caller context
3438 * flags - case flags
3439 *
3440 * RETURN: 0 on success, error code on failure.
3441 *
3442 * Timestamps:
3443 * sdvp,tdvp - ctime|mtime updated
3444 */
3445 /*ARGSUSED*/
3446 static int
3447 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3448 caller_context_t *ct, int flags)
3449 {
3450 znode_t *tdzp, *szp, *tzp;
3451 znode_t *sdzp = VTOZ(sdvp);
3452 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3453 zilog_t *zilog;
3454 vnode_t *realvp;
3455 zfs_dirlock_t *sdl, *tdl;
3456 dmu_tx_t *tx;
3457 zfs_zlock_t *zl;
3458 int cmp, serr, terr;
3459 int error = 0, rm_err = 0;
3460 int zflg = 0;
3461 boolean_t waited = B_FALSE;
3462
3463 ZFS_ENTER(zfsvfs);
3464 ZFS_VERIFY_ZP(sdzp);
3465 zilog = zfsvfs->z_log;
3466
3467 /*
3468 * Make sure we have the real vp for the target directory.
3469 */
3470 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3471 tdvp = realvp;
3472
3473 tdzp = VTOZ(tdvp);
3474 ZFS_VERIFY_ZP(tdzp);
3475
3476 /*
3477 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3478 * ctldir appear to have the same v_vfsp.
3479 */
3480 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3481 ZFS_EXIT(zfsvfs);
3482 return (SET_ERROR(EXDEV));
3483 }
3484
3485 if (zfsvfs->z_utf8 && u8_validate(tnm,
3486 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3487 ZFS_EXIT(zfsvfs);
3488 return (SET_ERROR(EILSEQ));
3489 }
3490
3491 if (flags & FIGNORECASE)
3492 zflg |= ZCILOOK;
3493
3494 top:
3495 szp = NULL;
3496 tzp = NULL;
3497 zl = NULL;
3498
3499 /*
3500 * This is to prevent the creation of links into attribute space
3501 * by renaming a linked file into/outof an attribute directory.
3502 * See the comment in zfs_link() for why this is considered bad.
3503 */
3504 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3505 ZFS_EXIT(zfsvfs);
3506 return (SET_ERROR(EINVAL));
3507 }
3508
3509 /*
3510 * Lock source and target directory entries. To prevent deadlock,
3511 * a lock ordering must be defined. We lock the directory with
3512 * the smallest object id first, or if it's a tie, the one with
3513 * the lexically first name.
3514 */
3515 if (sdzp->z_id < tdzp->z_id) {
3516 cmp = -1;
3517 } else if (sdzp->z_id > tdzp->z_id) {
3518 cmp = 1;
3519 } else {
3520 /*
3521 * First compare the two name arguments without
3522 * considering any case folding.
3523 */
3524 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3525
3526 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3527 ASSERT(error == 0 || !zfsvfs->z_utf8);
3528 if (cmp == 0) {
3529 /*
3530 * POSIX: "If the old argument and the new argument
3531 * both refer to links to the same existing file,
3532 * the rename() function shall return successfully
3533 * and perform no other action."
3534 */
3535 ZFS_EXIT(zfsvfs);
3536 return (0);
3537 }
3538 /*
3539 * If the file system is case-folding, then we may
3540 * have some more checking to do. A case-folding file
3541 * system is either supporting mixed case sensitivity
3542 * access or is completely case-insensitive. Note
3543 * that the file system is always case preserving.
3544 *
3545 * In mixed sensitivity mode case sensitive behavior
3546 * is the default. FIGNORECASE must be used to
3547 * explicitly request case insensitive behavior.
3548 *
3549 * If the source and target names provided differ only
3550 * by case (e.g., a request to rename 'tim' to 'Tim'),
3551 * we will treat this as a special case in the
3552 * case-insensitive mode: as long as the source name
3553 * is an exact match, we will allow this to proceed as
3554 * a name-change request.
3555 */
3556 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3557 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3558 flags & FIGNORECASE)) &&
3559 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3560 &error) == 0) {
3561 /*
3562 * case preserving rename request, require exact
3563 * name matches
3564 */
3565 zflg |= ZCIEXACT;
3566 zflg &= ~ZCILOOK;
3567 }
3568 }
3569
3570 /*
3571 * If the source and destination directories are the same, we should
3572 * grab the z_name_lock of that directory only once.
3573 */
3574 if (sdzp == tdzp) {
3575 zflg |= ZHAVELOCK;
3576 rw_enter(&sdzp->z_name_lock, RW_READER);
3577 }
3578
3579 if (cmp < 0) {
3580 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3581 ZEXISTS | zflg, NULL, NULL);
3582 terr = zfs_dirent_lock(&tdl,
3583 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3584 } else {
3585 terr = zfs_dirent_lock(&tdl,
3586 tdzp, tnm, &tzp, zflg, NULL, NULL);
3587 serr = zfs_dirent_lock(&sdl,
3588 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3589 NULL, NULL);
3590 }
3591
3592 if (serr) {
3593 /*
3594 * Source entry invalid or not there.
3595 */
3596 if (!terr) {
3597 zfs_dirent_unlock(tdl);
3598 if (tzp)
3599 VN_RELE(ZTOV(tzp));
3600 }
3601
3602 if (sdzp == tdzp)
3603 rw_exit(&sdzp->z_name_lock);
3604
3605 if (strcmp(snm, "..") == 0)
3606 serr = SET_ERROR(EINVAL);
3607 ZFS_EXIT(zfsvfs);
3608 return (serr);
3609 }
3610 if (terr) {
3611 zfs_dirent_unlock(sdl);
3612 VN_RELE(ZTOV(szp));
3613
3614 if (sdzp == tdzp)
3615 rw_exit(&sdzp->z_name_lock);
3616
3617 if (strcmp(tnm, "..") == 0)
3618 terr = SET_ERROR(EINVAL);
3619 ZFS_EXIT(zfsvfs);
3620 return (terr);
3621 }
3622
3623 /*
3624 * Must have write access at the source to remove the old entry
3625 * and write access at the target to create the new entry.
3626 * Note that if target and source are the same, this can be
3627 * done in a single check.
3628 */
3629
3630 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3631 goto out;
3632
3633 if (ZTOV(szp)->v_type == VDIR) {
3634 /*
3635 * Check to make sure rename is valid.
3636 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3637 */
3638 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3639 goto out;
3640 }
3641
3642 /*
3643 * Does target exist?
3644 */
3645 if (tzp) {
3646 /*
3647 * Source and target must be the same type.
3648 */
3649 if (ZTOV(szp)->v_type == VDIR) {
3650 if (ZTOV(tzp)->v_type != VDIR) {
3651 error = SET_ERROR(ENOTDIR);
3652 goto out;
3653 }
3654 } else {
3655 if (ZTOV(tzp)->v_type == VDIR) {
3656 error = SET_ERROR(EISDIR);
3657 goto out;
3658 }
3659 }
3660 /*
3661 * POSIX dictates that when the source and target
3662 * entries refer to the same file object, rename
3663 * must do nothing and exit without error.
3664 */
3665 if (szp->z_id == tzp->z_id) {
3666 error = 0;
3667 goto out;
3668 }
3669 }
3670
3671 vnevent_pre_rename_src(ZTOV(szp), sdvp, snm, ct);
3672 if (tzp)
3673 vnevent_pre_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3674
3675 /*
3676 * notify the target directory if it is not the same
3677 * as source directory.
3678 */
3679 if (tdvp != sdvp) {
3680 vnevent_pre_rename_dest_dir(tdvp, ZTOV(szp), tnm, ct);
3681 }
3682
3683 tx = dmu_tx_create(zfsvfs->z_os);
3684 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3685 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3686 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3687 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3688 if (sdzp != tdzp) {
3689 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3690 zfs_sa_upgrade_txholds(tx, tdzp);
3691 }
3692 if (tzp) {
3693 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3694 zfs_sa_upgrade_txholds(tx, tzp);
3695 }
3696
3697 zfs_sa_upgrade_txholds(tx, szp);
3698 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3699 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3700 if (error) {
3701 if (zl != NULL)
3702 zfs_rename_unlock(&zl);
3703 zfs_dirent_unlock(sdl);
3704 zfs_dirent_unlock(tdl);
3705
3706 if (sdzp == tdzp)
3707 rw_exit(&sdzp->z_name_lock);
3708
3709 VN_RELE(ZTOV(szp));
3710 if (tzp)
3711 VN_RELE(ZTOV(tzp));
3712 if (error == ERESTART) {
3713 waited = B_TRUE;
3714 dmu_tx_wait(tx);
3715 dmu_tx_abort(tx);
3716 goto top;
3717 }
3718 dmu_tx_abort(tx);
3719 ZFS_EXIT(zfsvfs);
3720 return (error);
3721 }
3722
3723 if (tzp) /* Attempt to remove the existing target */
3724 error = rm_err = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3725
3726 if (error == 0) {
3727 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3728 if (error == 0) {
3729 szp->z_pflags |= ZFS_AV_MODIFIED;
3730
3731 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3732 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3733 ASSERT0(error);
3734
3735 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3736 if (error == 0) {
3737 zfs_log_rename(zilog, tx, TX_RENAME |
3738 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3739 sdl->dl_name, tdzp, tdl->dl_name, szp);
3740
3741 /*
3742 * Update path information for the target vnode
3743 */
3744 vn_renamepath(tdvp, ZTOV(szp), tnm,
3745 strlen(tnm));
3746 } else {
3747 /*
3748 * At this point, we have successfully created
3749 * the target name, but have failed to remove
3750 * the source name. Since the create was done
3751 * with the ZRENAMING flag, there are
3752 * complications; for one, the link count is
3753 * wrong. The easiest way to deal with this
3754 * is to remove the newly created target, and
3755 * return the original error. This must
3756 * succeed; fortunately, it is very unlikely to
3757 * fail, since we just created it.
3758 */
3759 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3760 ZRENAMING, NULL), ==, 0);
3761 }
3762 }
3763 }
3764
3765 dmu_tx_commit(tx);
3766
3767 if (tzp && rm_err == 0)
3768 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3769
3770 if (error == 0) {
3771 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3772 vnevent_rename_dest_dir(tdvp, ZTOV(szp), tnm, ct);
3773 }
3774 out:
3775 if (zl != NULL)
3776 zfs_rename_unlock(&zl);
3777
3778 zfs_dirent_unlock(sdl);
3779 zfs_dirent_unlock(tdl);
3780
3781 if (sdzp == tdzp)
3782 rw_exit(&sdzp->z_name_lock);
3783
3784
3785 VN_RELE(ZTOV(szp));
3786 if (tzp)
3787 VN_RELE(ZTOV(tzp));
3788
3789 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3790 zil_commit(zilog, 0);
3791
3792 ZFS_EXIT(zfsvfs);
3793 return (error);
3794 }
3795
3796 /*
3797 * Insert the indicated symbolic reference entry into the directory.
3798 *
3799 * IN: dvp - Directory to contain new symbolic link.
3800 * link - Name for new symlink entry.
3801 * vap - Attributes of new entry.
3802 * cr - credentials of caller.
3803 * ct - caller context
3804 * flags - case flags
3805 *
3806 * RETURN: 0 on success, error code on failure.
3807 *
3808 * Timestamps:
3809 * dvp - ctime|mtime updated
3810 */
3811 /*ARGSUSED*/
3812 static int
3813 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3814 caller_context_t *ct, int flags)
3815 {
3816 znode_t *zp, *dzp = VTOZ(dvp);
3817 zfs_dirlock_t *dl;
3818 dmu_tx_t *tx;
3819 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
3820 zilog_t *zilog;
3821 uint64_t len = strlen(link);
3822 int error;
3823 int zflg = ZNEW;
3824 zfs_acl_ids_t acl_ids;
3825 boolean_t fuid_dirtied;
3826 uint64_t txtype = TX_SYMLINK;
3827 boolean_t waited = B_FALSE;
3828
3829 ASSERT(vap->va_type == VLNK);
3830
3831 ZFS_ENTER(zfsvfs);
3832 ZFS_VERIFY_ZP(dzp);
3833 zilog = zfsvfs->z_log;
3834
3835 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3836 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3837 ZFS_EXIT(zfsvfs);
3838 return (SET_ERROR(EILSEQ));
3839 }
3840 if (flags & FIGNORECASE)
3841 zflg |= ZCILOOK;
3842
3843 if (len > MAXPATHLEN) {
3844 ZFS_EXIT(zfsvfs);
3845 return (SET_ERROR(ENAMETOOLONG));
3846 }
3847
3848 if ((error = zfs_acl_ids_create(dzp, 0,
3849 vap, cr, NULL, &acl_ids)) != 0) {
3850 ZFS_EXIT(zfsvfs);
3851 return (error);
3852 }
3853 top:
3854 /*
3855 * Attempt to lock directory; fail if entry already exists.
3856 */
3857 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3858 if (error) {
3859 zfs_acl_ids_free(&acl_ids);
3860 ZFS_EXIT(zfsvfs);
3861 return (error);
3862 }
3863
3864 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3865 zfs_acl_ids_free(&acl_ids);
3866 zfs_dirent_unlock(dl);
3867 ZFS_EXIT(zfsvfs);
3868 return (error);
3869 }
3870
3871 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3872 zfs_acl_ids_free(&acl_ids);
3873 zfs_dirent_unlock(dl);
3874 ZFS_EXIT(zfsvfs);
3875 return (SET_ERROR(EDQUOT));
3876 }
3877 tx = dmu_tx_create(zfsvfs->z_os);
3878 fuid_dirtied = zfsvfs->z_fuid_dirty;
3879 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3880 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3881 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3882 ZFS_SA_BASE_ATTR_SIZE + len);
3883 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3884 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3885 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3886 acl_ids.z_aclp->z_acl_bytes);
3887 }
3888 if (fuid_dirtied)
3889 zfs_fuid_txhold(zfsvfs, tx);
3890 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3891 if (error) {
3892 zfs_dirent_unlock(dl);
3893 if (error == ERESTART) {
3894 waited = B_TRUE;
3895 dmu_tx_wait(tx);
3896 dmu_tx_abort(tx);
3897 goto top;
3898 }
3899 zfs_acl_ids_free(&acl_ids);
3900 dmu_tx_abort(tx);
3901 ZFS_EXIT(zfsvfs);
3902 return (error);
3903 }
3904
3905 /*
3906 * Create a new object for the symlink.
3907 * for version 4 ZPL datsets the symlink will be an SA attribute
3908 */
3909 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3910
3911 if (fuid_dirtied)
3912 zfs_fuid_sync(zfsvfs, tx);
3913
3914 mutex_enter(&zp->z_lock);
3915 if (zp->z_is_sa)
3916 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3917 link, len, tx);
3918 else
3919 zfs_sa_symlink(zp, link, len, tx);
3920 mutex_exit(&zp->z_lock);
3921
3922 zp->z_size = len;
3923 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3924 &zp->z_size, sizeof (zp->z_size), tx);
3925 /*
3926 * Insert the new object into the directory.
3927 */
3928 (void) zfs_link_create(dl, zp, tx, ZNEW);
3929
3930 if (flags & FIGNORECASE)
3931 txtype |= TX_CI;
3932 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3933
3934 zfs_acl_ids_free(&acl_ids);
3935
3936 dmu_tx_commit(tx);
3937
3938 zfs_dirent_unlock(dl);
3939
3940 VN_RELE(ZTOV(zp));
3941
3942 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3943 zil_commit(zilog, 0);
3944
3945 ZFS_EXIT(zfsvfs);
3946 return (error);
3947 }
3948
3949 /*
3950 * Return, in the buffer contained in the provided uio structure,
3951 * the symbolic path referred to by vp.
3952 *
3953 * IN: vp - vnode of symbolic link.
3954 * uio - structure to contain the link path.
3955 * cr - credentials of caller.
3956 * ct - caller context
3957 *
3958 * OUT: uio - structure containing the link path.
3959 *
3960 * RETURN: 0 on success, error code on failure.
3961 *
3962 * Timestamps:
3963 * vp - atime updated
3964 */
3965 /* ARGSUSED */
3966 static int
3967 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3968 {
3969 znode_t *zp = VTOZ(vp);
3970 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3971 int error;
3972
3973 ZFS_ENTER(zfsvfs);
3974 ZFS_VERIFY_ZP(zp);
3975
3976 mutex_enter(&zp->z_lock);
3977 if (zp->z_is_sa)
3978 error = sa_lookup_uio(zp->z_sa_hdl,
3979 SA_ZPL_SYMLINK(zfsvfs), uio);
3980 else
3981 error = zfs_sa_readlink(zp, uio);
3982 mutex_exit(&zp->z_lock);
3983
3984 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3985
3986 ZFS_EXIT(zfsvfs);
3987 return (error);
3988 }
3989
3990 /*
3991 * Insert a new entry into directory tdvp referencing svp.
3992 *
3993 * IN: tdvp - Directory to contain new entry.
3994 * svp - vnode of new entry.
3995 * name - name of new entry.
3996 * cr - credentials of caller.
3997 * ct - caller context
3998 *
3999 * RETURN: 0 on success, error code on failure.
4000 *
4001 * Timestamps:
4002 * tdvp - ctime|mtime updated
4003 * svp - ctime updated
4004 */
4005 /* ARGSUSED */
4006 static int
4007 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4008 caller_context_t *ct, int flags)
4009 {
4010 znode_t *dzp = VTOZ(tdvp);
4011 znode_t *tzp, *szp;
4012 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4013 zilog_t *zilog;
4014 zfs_dirlock_t *dl;
4015 dmu_tx_t *tx;
4016 vnode_t *realvp;
4017 int error;
4018 int zf = ZNEW;
4019 uint64_t parent;
4020 uid_t owner;
4021 boolean_t waited = B_FALSE;
4022
4023 ASSERT(tdvp->v_type == VDIR);
4024
4025 ZFS_ENTER(zfsvfs);
4026 ZFS_VERIFY_ZP(dzp);
4027 zilog = zfsvfs->z_log;
4028
4029 if (VOP_REALVP(svp, &realvp, ct) == 0)
4030 svp = realvp;
4031
4032 /*
4033 * POSIX dictates that we return EPERM here.
4034 * Better choices include ENOTSUP or EISDIR.
4035 */
4036 if (svp->v_type == VDIR) {
4037 ZFS_EXIT(zfsvfs);
4038 return (SET_ERROR(EPERM));
4039 }
4040
4041 szp = VTOZ(svp);
4042 ZFS_VERIFY_ZP(szp);
4043
4044 /*
4045 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4046 * ctldir appear to have the same v_vfsp.
4047 */
4048 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4049 ZFS_EXIT(zfsvfs);
4050 return (SET_ERROR(EXDEV));
4051 }
4052
4053 /* Prevent links to .zfs/shares files */
4054
4055 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4056 &parent, sizeof (uint64_t))) != 0) {
4057 ZFS_EXIT(zfsvfs);
4058 return (error);
4059 }
4060 if (parent == zfsvfs->z_shares_dir) {
4061 ZFS_EXIT(zfsvfs);
4062 return (SET_ERROR(EPERM));
4063 }
4064
4065 if (zfsvfs->z_utf8 && u8_validate(name,
4066 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4067 ZFS_EXIT(zfsvfs);
4068 return (SET_ERROR(EILSEQ));
4069 }
4070 if (flags & FIGNORECASE)
4071 zf |= ZCILOOK;
4072
4073 /*
4074 * We do not support links between attributes and non-attributes
4075 * because of the potential security risk of creating links
4076 * into "normal" file space in order to circumvent restrictions
4077 * imposed in attribute space.
4078 */
4079 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4080 ZFS_EXIT(zfsvfs);
4081 return (SET_ERROR(EINVAL));
4082 }
4083
4084
4085 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4086 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4087 ZFS_EXIT(zfsvfs);
4088 return (SET_ERROR(EPERM));
4089 }
4090
4091 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4092 ZFS_EXIT(zfsvfs);
4093 return (error);
4094 }
4095
4096 top:
4097 /*
4098 * Attempt to lock directory; fail if entry already exists.
4099 */
4100 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4101 if (error) {
4102 ZFS_EXIT(zfsvfs);
4103 return (error);
4104 }
4105
4106 tx = dmu_tx_create(zfsvfs->z_os);
4107 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4108 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4109 zfs_sa_upgrade_txholds(tx, szp);
4110 zfs_sa_upgrade_txholds(tx, dzp);
4111 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4112 if (error) {
4113 zfs_dirent_unlock(dl);
4114 if (error == ERESTART) {
4115 waited = B_TRUE;
4116 dmu_tx_wait(tx);
4117 dmu_tx_abort(tx);
4118 goto top;
4119 }
4120 dmu_tx_abort(tx);
4121 ZFS_EXIT(zfsvfs);
4122 return (error);
4123 }
4124
4125 error = zfs_link_create(dl, szp, tx, 0);
4126
4127 if (error == 0) {
4128 uint64_t txtype = TX_LINK;
4129 if (flags & FIGNORECASE)
4130 txtype |= TX_CI;
4131 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4132 }
4133
4134 dmu_tx_commit(tx);
4135
4136 zfs_dirent_unlock(dl);
4137
4138 if (error == 0) {
4139 vnevent_link(svp, ct);
4140 }
4141
4142 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4143 zil_commit(zilog, 0);
4144
4145 ZFS_EXIT(zfsvfs);
4146 return (error);
4147 }
4148
4149 /*
4150 * zfs_null_putapage() is used when the file system has been force
4151 * unmounted. It just drops the pages.
4152 */
4153 /* ARGSUSED */
4154 static int
4155 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4156 size_t *lenp, int flags, cred_t *cr)
4157 {
4158 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4159 return (0);
4160 }
4161
4162 /*
4163 * Push a page out to disk, klustering if possible.
4164 *
4165 * IN: vp - file to push page to.
4166 * pp - page to push.
4167 * flags - additional flags.
4168 * cr - credentials of caller.
4169 *
4170 * OUT: offp - start of range pushed.
4171 * lenp - len of range pushed.
4172 *
4173 * RETURN: 0 on success, error code on failure.
4174 *
4175 * NOTE: callers must have locked the page to be pushed. On
4176 * exit, the page (and all other pages in the kluster) must be
4177 * unlocked.
4178 */
4179 /* ARGSUSED */
4180 static int
4181 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4182 size_t *lenp, int flags, cred_t *cr)
4183 {
4184 znode_t *zp = VTOZ(vp);
4185 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4186 dmu_tx_t *tx;
4187 u_offset_t off, koff;
4188 size_t len, klen;
4189 int err;
4190
4191 off = pp->p_offset;
4192 len = PAGESIZE;
4193 /*
4194 * If our blocksize is bigger than the page size, try to kluster
4195 * multiple pages so that we write a full block (thus avoiding
4196 * a read-modify-write).
4197 */
4198 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4199 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4200 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4201 ASSERT(koff <= zp->z_size);
4202 if (koff + klen > zp->z_size)
4203 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4204 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4205 }
4206 ASSERT3U(btop(len), ==, btopr(len));
4207
4208 /*
4209 * Can't push pages past end-of-file.
4210 */
4211 if (off >= zp->z_size) {
4212 /* ignore all pages */
4213 err = 0;
4214 goto out;
4215 } else if (off + len > zp->z_size) {
4216 int npages = btopr(zp->z_size - off);
4217 page_t *trunc;
4218
4219 page_list_break(&pp, &trunc, npages);
4220 /* ignore pages past end of file */
4221 if (trunc)
4222 pvn_write_done(trunc, flags);
4223 len = zp->z_size - off;
4224 }
4225
4226 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4227 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4228 err = SET_ERROR(EDQUOT);
4229 goto out;
4230 }
4231 tx = dmu_tx_create(zfsvfs->z_os);
4232 dmu_tx_hold_write(tx, zp->z_id, off, len);
4233
4234 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4235 zfs_sa_upgrade_txholds(tx, zp);
4236 err = dmu_tx_assign(tx, TXG_WAIT);
4237 if (err != 0) {
4238 dmu_tx_abort(tx);
4239 goto out;
4240 }
4241
4242 if (zp->z_blksz <= PAGESIZE) {
4243 caddr_t va = zfs_map_page(pp, S_READ);
4244 ASSERT3U(len, <=, PAGESIZE);
4245 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4246 zfs_unmap_page(pp, va);
4247 } else {
4248 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4249 }
4250
4251 if (err == 0) {
4252 uint64_t mtime[2], ctime[2];
4253 sa_bulk_attr_t bulk[3];
4254 int count = 0;
4255
4256 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4257 &mtime, 16);
4258 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4259 &ctime, 16);
4260 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4261 &zp->z_pflags, 8);
4262 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4263 B_TRUE);
4264 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4265 }
4266 dmu_tx_commit(tx);
4267
4268 out:
4269 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4270 if (offp)
4271 *offp = off;
4272 if (lenp)
4273 *lenp = len;
4274
4275 return (err);
4276 }
4277
4278 /*
4279 * Copy the portion of the file indicated from pages into the file.
4280 * The pages are stored in a page list attached to the files vnode.
4281 *
4282 * IN: vp - vnode of file to push page data to.
4283 * off - position in file to put data.
4284 * len - amount of data to write.
4285 * flags - flags to control the operation.
4286 * cr - credentials of caller.
4287 * ct - caller context.
4288 *
4289 * RETURN: 0 on success, error code on failure.
4290 *
4291 * Timestamps:
4292 * vp - ctime|mtime updated
4293 */
4294 /*ARGSUSED*/
4295 static int
4296 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4297 caller_context_t *ct)
4298 {
4299 znode_t *zp = VTOZ(vp);
4300 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4301 page_t *pp;
4302 size_t io_len;
4303 u_offset_t io_off;
4304 uint_t blksz;
4305 rl_t *rl;
4306 int error = 0;
4307
4308 ZFS_ENTER(zfsvfs);
4309 ZFS_VERIFY_ZP(zp);
4310
4311 /*
4312 * There's nothing to do if no data is cached.
4313 */
4314 if (!vn_has_cached_data(vp)) {
4315 ZFS_EXIT(zfsvfs);
4316 return (0);
4317 }
4318
4319 /*
4320 * Align this request to the file block size in case we kluster.
4321 * XXX - this can result in pretty aggresive locking, which can
4322 * impact simultanious read/write access. One option might be
4323 * to break up long requests (len == 0) into block-by-block
4324 * operations to get narrower locking.
4325 */
4326 blksz = zp->z_blksz;
4327 if (ISP2(blksz))
4328 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4329 else
4330 io_off = 0;
4331 if (len > 0 && ISP2(blksz))
4332 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4333 else
4334 io_len = 0;
4335
4336 if (io_len == 0) {
4337 /*
4338 * Search the entire vp list for pages >= io_off.
4339 */
4340 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4341 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4342 goto out;
4343 }
4344 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4345
4346 if (off > zp->z_size) {
4347 /* past end of file */
4348 zfs_range_unlock(rl);
4349 ZFS_EXIT(zfsvfs);
4350 return (0);
4351 }
4352
4353 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4354
4355 for (off = io_off; io_off < off + len; io_off += io_len) {
4356 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4357 pp = page_lookup(vp, io_off,
4358 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4359 } else {
4360 pp = page_lookup_nowait(vp, io_off,
4361 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4362 }
4363
4364 if (pp != NULL && pvn_getdirty(pp, flags)) {
4365 int err;
4366
4367 /*
4368 * Found a dirty page to push
4369 */
4370 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4371 if (err)
4372 error = err;
4373 } else {
4374 io_len = PAGESIZE;
4375 }
4376 }
4377 out:
4378 zfs_range_unlock(rl);
4379 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4380 zil_commit(zfsvfs->z_log, zp->z_id);
4381 ZFS_EXIT(zfsvfs);
4382 return (error);
4383 }
4384
4385 /*ARGSUSED*/
4386 void
4387 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4388 {
4389 znode_t *zp = VTOZ(vp);
4390 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4391 int error;
4392
4393 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4394 if (zp->z_sa_hdl == NULL) {
4395 /*
4396 * The fs has been unmounted, or we did a
4397 * suspend/resume and this file no longer exists.
4398 */
4399 if (vn_has_cached_data(vp)) {
4400 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4401 B_INVAL, cr);
4402 }
4403
4404 mutex_enter(&zp->z_lock);
4405 mutex_enter(&vp->v_lock);
4406 ASSERT(vp->v_count == 1);
4407 vp->v_count = 0;
4408 mutex_exit(&vp->v_lock);
4409 mutex_exit(&zp->z_lock);
4410 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4411 zfs_znode_free(zp);
4412 return;
4413 }
4414
4415 /*
4416 * Attempt to push any data in the page cache. If this fails
4417 * we will get kicked out later in zfs_zinactive().
4418 */
4419 if (vn_has_cached_data(vp)) {
4420 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4421 cr);
4422 }
4423
4424 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4425 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4426
4427 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4428 zfs_sa_upgrade_txholds(tx, zp);
4429 error = dmu_tx_assign(tx, TXG_WAIT);
4430 if (error) {
4431 dmu_tx_abort(tx);
4432 } else {
4433 mutex_enter(&zp->z_lock);
4434 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4435 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4436 zp->z_atime_dirty = 0;
4437 mutex_exit(&zp->z_lock);
4438 dmu_tx_commit(tx);
4439 }
4440 }
4441
4442 zfs_zinactive(zp);
4443 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4444 }
4445
4446 /*
4447 * Bounds-check the seek operation.
4448 *
4449 * IN: vp - vnode seeking within
4450 * ooff - old file offset
4451 * noffp - pointer to new file offset
4452 * ct - caller context
4453 *
4454 * RETURN: 0 on success, EINVAL if new offset invalid.
4455 */
4456 /* ARGSUSED */
4457 static int
4458 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4459 caller_context_t *ct)
4460 {
4461 if (vp->v_type == VDIR)
4462 return (0);
4463 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4464 }
4465
4466 /*
4467 * Pre-filter the generic locking function to trap attempts to place
4468 * a mandatory lock on a memory mapped file.
4469 */
4470 static int
4471 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4472 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4473 {
4474 znode_t *zp = VTOZ(vp);
4475 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4476
4477 ZFS_ENTER(zfsvfs);
4478 ZFS_VERIFY_ZP(zp);
4479
4480 /*
4481 * We are following the UFS semantics with respect to mapcnt
4482 * here: If we see that the file is mapped already, then we will
4483 * return an error, but we don't worry about races between this
4484 * function and zfs_map().
4485 */
4486 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4487 ZFS_EXIT(zfsvfs);
4488 return (SET_ERROR(EAGAIN));
4489 }
4490 ZFS_EXIT(zfsvfs);
4491 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4492 }
4493
4494 /*
4495 * If we can't find a page in the cache, we will create a new page
4496 * and fill it with file data. For efficiency, we may try to fill
4497 * multiple pages at once (klustering) to fill up the supplied page
4498 * list. Note that the pages to be filled are held with an exclusive
4499 * lock to prevent access by other threads while they are being filled.
4500 */
4501 static int
4502 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4503 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4504 {
4505 znode_t *zp = VTOZ(vp);
4506 page_t *pp, *cur_pp;
4507 objset_t *os = zp->z_zfsvfs->z_os;
4508 u_offset_t io_off, total;
4509 size_t io_len;
4510 int err;
4511
4512 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4513 /*
4514 * We only have a single page, don't bother klustering
4515 */
4516 io_off = off;
4517 io_len = PAGESIZE;
4518 pp = page_create_va(vp, io_off, io_len,
4519 PG_EXCL | PG_WAIT, seg, addr);
4520 } else {
4521 /*
4522 * Try to find enough pages to fill the page list
4523 */
4524 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4525 &io_len, off, plsz, 0);
4526 }
4527 if (pp == NULL) {
4528 /*
4529 * The page already exists, nothing to do here.
4530 */
4531 *pl = NULL;
4532 return (0);
4533 }
4534
4535 /*
4536 * Fill the pages in the kluster.
4537 */
4538 cur_pp = pp;
4539 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4540 caddr_t va;
4541
4542 ASSERT3U(io_off, ==, cur_pp->p_offset);
4543 va = zfs_map_page(cur_pp, S_WRITE);
4544 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4545 DMU_READ_PREFETCH);
4546 zfs_unmap_page(cur_pp, va);
4547 if (err) {
4548 /* On error, toss the entire kluster */
4549 pvn_read_done(pp, B_ERROR);
4550 /* convert checksum errors into IO errors */
4551 if (err == ECKSUM)
4552 err = SET_ERROR(EIO);
4553 return (err);
4554 }
4555 cur_pp = cur_pp->p_next;
4556 }
4557
4558 /*
4559 * Fill in the page list array from the kluster starting
4560 * from the desired offset `off'.
4561 * NOTE: the page list will always be null terminated.
4562 */
4563 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4564 ASSERT(pl == NULL || (*pl)->p_offset == off);
4565
4566 return (0);
4567 }
4568
4569 /*
4570 * Return pointers to the pages for the file region [off, off + len]
4571 * in the pl array. If plsz is greater than len, this function may
4572 * also return page pointers from after the specified region
4573 * (i.e. the region [off, off + plsz]). These additional pages are
4574 * only returned if they are already in the cache, or were created as
4575 * part of a klustered read.
4576 *
4577 * IN: vp - vnode of file to get data from.
4578 * off - position in file to get data from.
4579 * len - amount of data to retrieve.
4580 * plsz - length of provided page list.
4581 * seg - segment to obtain pages for.
4582 * addr - virtual address of fault.
4583 * rw - mode of created pages.
4584 * cr - credentials of caller.
4585 * ct - caller context.
4586 *
4587 * OUT: protp - protection mode of created pages.
4588 * pl - list of pages created.
4589 *
4590 * RETURN: 0 on success, error code on failure.
4591 *
4592 * Timestamps:
4593 * vp - atime updated
4594 */
4595 /* ARGSUSED */
4596 static int
4597 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4598 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4599 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4600 {
4601 znode_t *zp = VTOZ(vp);
4602 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4603 page_t **pl0 = pl;
4604 int err = 0;
4605
4606 /* we do our own caching, faultahead is unnecessary */
4607 if (pl == NULL)
4608 return (0);
4609 else if (len > plsz)
4610 len = plsz;
4611 else
4612 len = P2ROUNDUP(len, PAGESIZE);
4613 ASSERT(plsz >= len);
4614
4615 ZFS_ENTER(zfsvfs);
4616 ZFS_VERIFY_ZP(zp);
4617
4618 if (protp)
4619 *protp = PROT_ALL;
4620
4621 /*
4622 * Loop through the requested range [off, off + len) looking
4623 * for pages. If we don't find a page, we will need to create
4624 * a new page and fill it with data from the file.
4625 */
4626 while (len > 0) {
4627 if (*pl = page_lookup(vp, off, SE_SHARED))
4628 *(pl+1) = NULL;
4629 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4630 goto out;
4631 while (*pl) {
4632 ASSERT3U((*pl)->p_offset, ==, off);
4633 off += PAGESIZE;
4634 addr += PAGESIZE;
4635 if (len > 0) {
4636 ASSERT3U(len, >=, PAGESIZE);
4637 len -= PAGESIZE;
4638 }
4639 ASSERT3U(plsz, >=, PAGESIZE);
4640 plsz -= PAGESIZE;
4641 pl++;
4642 }
4643 }
4644
4645 /*
4646 * Fill out the page array with any pages already in the cache.
4647 */
4648 while (plsz > 0 &&
4649 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4650 off += PAGESIZE;
4651 plsz -= PAGESIZE;
4652 }
4653 out:
4654 if (err) {
4655 /*
4656 * Release any pages we have previously locked.
4657 */
4658 while (pl > pl0)
4659 page_unlock(*--pl);
4660 } else {
4661 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4662 }
4663
4664 *pl = NULL;
4665
4666 ZFS_EXIT(zfsvfs);
4667 return (err);
4668 }
4669
4670 /*
4671 * Request a memory map for a section of a file. This code interacts
4672 * with common code and the VM system as follows:
4673 *
4674 * - common code calls mmap(), which ends up in smmap_common()
4675 * - this calls VOP_MAP(), which takes you into (say) zfs
4676 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4677 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4678 * - zfs_addmap() updates z_mapcnt
4679 */
4680 /*ARGSUSED*/
4681 static int
4682 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4683 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4684 caller_context_t *ct)
4685 {
4686 znode_t *zp = VTOZ(vp);
4687 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4688 segvn_crargs_t vn_a;
4689 int error;
4690
4691 ZFS_ENTER(zfsvfs);
4692 ZFS_VERIFY_ZP(zp);
4693
4694 if ((prot & PROT_WRITE) && (zp->z_pflags &
4695 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4696 ZFS_EXIT(zfsvfs);
4697 return (SET_ERROR(EPERM));
4698 }
4699
4700 if ((prot & (PROT_READ | PROT_EXEC)) &&
4701 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4702 ZFS_EXIT(zfsvfs);
4703 return (SET_ERROR(EACCES));
4704 }
4705
4706 if (vp->v_flag & VNOMAP) {
4707 ZFS_EXIT(zfsvfs);
4708 return (SET_ERROR(ENOSYS));
4709 }
4710
4711 if (off < 0 || len > MAXOFFSET_T - off) {
4712 ZFS_EXIT(zfsvfs);
4713 return (SET_ERROR(ENXIO));
4714 }
4715
4716 if (vp->v_type != VREG) {
4717 ZFS_EXIT(zfsvfs);
4718 return (SET_ERROR(ENODEV));
4719 }
4720
4721 /*
4722 * If file is locked, disallow mapping.
4723 */
4724 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4725 ZFS_EXIT(zfsvfs);
4726 return (SET_ERROR(EAGAIN));
4727 }
4728
4729 as_rangelock(as);
4730 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4731 if (error != 0) {
4732 as_rangeunlock(as);
4733 ZFS_EXIT(zfsvfs);
4734 return (error);
4735 }
4736
4737 vn_a.vp = vp;
4738 vn_a.offset = (u_offset_t)off;
4739 vn_a.type = flags & MAP_TYPE;
4740 vn_a.prot = prot;
4741 vn_a.maxprot = maxprot;
4742 vn_a.cred = cr;
4743 vn_a.amp = NULL;
4744 vn_a.flags = flags & ~MAP_TYPE;
4745 vn_a.szc = 0;
4746 vn_a.lgrp_mem_policy_flags = 0;
4747
4748 error = as_map(as, *addrp, len, segvn_create, &vn_a);
4749
4750 as_rangeunlock(as);
4751 ZFS_EXIT(zfsvfs);
4752 return (error);
4753 }
4754
4755 /* ARGSUSED */
4756 static int
4757 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4758 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4759 caller_context_t *ct)
4760 {
4761 uint64_t pages = btopr(len);
4762
4763 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4764 return (0);
4765 }
4766
4767 /*
4768 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4769 * more accurate mtime for the associated file. Since we don't have a way of
4770 * detecting when the data was actually modified, we have to resort to
4771 * heuristics. If an explicit msync() is done, then we mark the mtime when the
4772 * last page is pushed. The problem occurs when the msync() call is omitted,
4773 * which by far the most common case:
4774 *
4775 * open()
4776 * mmap()
4777 * <modify memory>
4778 * munmap()
4779 * close()
4780 * <time lapse>
4781 * putpage() via fsflush
4782 *
4783 * If we wait until fsflush to come along, we can have a modification time that
4784 * is some arbitrary point in the future. In order to prevent this in the
4785 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4786 * torn down.
4787 */
4788 /* ARGSUSED */
4789 static int
4790 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4791 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4792 caller_context_t *ct)
4793 {
4794 uint64_t pages = btopr(len);
4795
4796 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4797 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4798
4799 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4800 vn_has_cached_data(vp))
4801 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4802
4803 return (0);
4804 }
4805
4806 /*
4807 * Free or allocate space in a file. Currently, this function only
4808 * supports the `F_FREESP' command. However, this command is somewhat
4809 * misnamed, as its functionality includes the ability to allocate as
4810 * well as free space.
4811 *
4812 * IN: vp - vnode of file to free data in.
4813 * cmd - action to take (only F_FREESP supported).
4814 * bfp - section of file to free/alloc.
4815 * flag - current file open mode flags.
4816 * offset - current file offset.
4817 * cr - credentials of caller [UNUSED].
4818 * ct - caller context.
4819 *
4820 * RETURN: 0 on success, error code on failure.
4821 *
4822 * Timestamps:
4823 * vp - ctime|mtime updated
4824 */
4825 /* ARGSUSED */
4826 static int
4827 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4828 offset_t offset, cred_t *cr, caller_context_t *ct)
4829 {
4830 znode_t *zp = VTOZ(vp);
4831 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4832 uint64_t off, len;
4833 int error;
4834
4835 ZFS_ENTER(zfsvfs);
4836 ZFS_VERIFY_ZP(zp);
4837
4838 if (cmd != F_FREESP) {
4839 ZFS_EXIT(zfsvfs);
4840 return (SET_ERROR(EINVAL));
4841 }
4842
4843 /*
4844 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
4845 * callers might not be able to detect properly that we are read-only,
4846 * so check it explicitly here.
4847 */
4848 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
4849 ZFS_EXIT(zfsvfs);
4850 return (SET_ERROR(EROFS));
4851 }
4852
4853 if (error = convoff(vp, bfp, 0, offset)) {
4854 ZFS_EXIT(zfsvfs);
4855 return (error);
4856 }
4857
4858 if (bfp->l_len < 0) {
4859 ZFS_EXIT(zfsvfs);
4860 return (SET_ERROR(EINVAL));
4861 }
4862
4863 off = bfp->l_start;
4864 len = bfp->l_len; /* 0 means from off to end of file */
4865
4866 error = zfs_freesp(zp, off, len, flag, TRUE);
4867
4868 if (error == 0 && len == 0) {
4869 if (off == 0) {
4870 vnevent_truncate(ZTOV(zp), ct);
4871 } else {
4872 vnevent_resize(ZTOV(zp), ct);
4873 }
4874 }
4875
4876 ZFS_EXIT(zfsvfs);
4877 return (error);
4878 }
4879
4880 /*ARGSUSED*/
4881 static int
4882 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4883 {
4884 znode_t *zp = VTOZ(vp);
4885 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4886 uint32_t gen;
4887 uint64_t gen64;
4888 uint64_t object = zp->z_id;
4889 zfid_short_t *zfid;
4890 int size, i, error;
4891
4892 ZFS_ENTER(zfsvfs);
4893 ZFS_VERIFY_ZP(zp);
4894
4895 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4896 &gen64, sizeof (uint64_t))) != 0) {
4897 ZFS_EXIT(zfsvfs);
4898 return (error);
4899 }
4900
4901 gen = (uint32_t)gen64;
4902
4903 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4904 if (fidp->fid_len < size) {
4905 fidp->fid_len = size;
4906 ZFS_EXIT(zfsvfs);
4907 return (SET_ERROR(ENOSPC));
4908 }
4909
4910 zfid = (zfid_short_t *)fidp;
4911
4912 zfid->zf_len = size;
4913
4914 for (i = 0; i < sizeof (zfid->zf_object); i++)
4915 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4916
4917 /* Must have a non-zero generation number to distinguish from .zfs */
4918 if (gen == 0)
4919 gen = 1;
4920 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4921 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4922
4923 if (size == LONG_FID_LEN) {
4924 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
4925 zfid_long_t *zlfid;
4926
4927 zlfid = (zfid_long_t *)fidp;
4928
4929 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4930 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4931
4932 /* XXX - this should be the generation number for the objset */
4933 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4934 zlfid->zf_setgen[i] = 0;
4935 }
4936
4937 ZFS_EXIT(zfsvfs);
4938 return (0);
4939 }
4940
4941 static int
4942 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4943 caller_context_t *ct)
4944 {
4945 znode_t *zp, *xzp;
4946 zfsvfs_t *zfsvfs;
4947 zfs_dirlock_t *dl;
4948 int error;
4949
4950 switch (cmd) {
4951 case _PC_LINK_MAX:
4952 *valp = ULONG_MAX;
4953 return (0);
4954
4955 case _PC_FILESIZEBITS:
4956 *valp = 64;
4957 return (0);
4958
4959 case _PC_XATTR_EXISTS:
4960 zp = VTOZ(vp);
4961 zfsvfs = zp->z_zfsvfs;
4962 ZFS_ENTER(zfsvfs);
4963 ZFS_VERIFY_ZP(zp);
4964 *valp = 0;
4965 error = zfs_dirent_lock(&dl, zp, "", &xzp,
4966 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4967 if (error == 0) {
4968 zfs_dirent_unlock(dl);
4969 if (!zfs_dirempty(xzp))
4970 *valp = 1;
4971 VN_RELE(ZTOV(xzp));
4972 } else if (error == ENOENT) {
4973 /*
4974 * If there aren't extended attributes, it's the
4975 * same as having zero of them.
4976 */
4977 error = 0;
4978 }
4979 ZFS_EXIT(zfsvfs);
4980 return (error);
4981
4982 case _PC_SATTR_ENABLED:
4983 case _PC_SATTR_EXISTS:
4984 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4985 (vp->v_type == VREG || vp->v_type == VDIR);
4986 return (0);
4987
4988 case _PC_ACCESS_FILTERING:
4989 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4990 vp->v_type == VDIR;
4991 return (0);
4992
4993 case _PC_ACL_ENABLED:
4994 *valp = _ACL_ACE_ENABLED;
4995 return (0);
4996
4997 case _PC_MIN_HOLE_SIZE:
4998 *valp = (ulong_t)SPA_MINBLOCKSIZE;
4999 return (0);
5000
5001 case _PC_TIMESTAMP_RESOLUTION:
5002 /* nanosecond timestamp resolution */
5003 *valp = 1L;
5004 return (0);
5005
5006 default:
5007 return (fs_pathconf(vp, cmd, valp, cr, ct));
5008 }
5009 }
5010
5011 /*ARGSUSED*/
5012 static int
5013 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5014 caller_context_t *ct)
5015 {
5016 znode_t *zp = VTOZ(vp);
5017 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5018 int error;
5019 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5020
5021 ZFS_ENTER(zfsvfs);
5022 ZFS_VERIFY_ZP(zp);
5023 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5024 ZFS_EXIT(zfsvfs);
5025
5026 return (error);
5027 }
5028
5029 /*ARGSUSED*/
5030 static int
5031 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5032 caller_context_t *ct)
5033 {
5034 znode_t *zp = VTOZ(vp);
5035 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5036 int error;
5037 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5038 zilog_t *zilog = zfsvfs->z_log;
5039
5040 ZFS_ENTER(zfsvfs);
5041 ZFS_VERIFY_ZP(zp);
5042
5043 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5044
5045 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5046 zil_commit(zilog, 0);
5047
5048 ZFS_EXIT(zfsvfs);
5049 return (error);
5050 }
5051
5052 /*
5053 * The smallest read we may consider to loan out an arcbuf.
5054 * This must be a power of 2.
5055 */
5056 int zcr_blksz_min = (1 << 10); /* 1K */
5057 /*
5058 * If set to less than the file block size, allow loaning out of an
5059 * arcbuf for a partial block read. This must be a power of 2.
5060 */
5061 int zcr_blksz_max = (1 << 17); /* 128K */
5062
5063 /*ARGSUSED*/
5064 static int
5065 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5066 caller_context_t *ct)
5067 {
5068 znode_t *zp = VTOZ(vp);
5069 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5070 int max_blksz = zfsvfs->z_max_blksz;
5071 uio_t *uio = &xuio->xu_uio;
5072 ssize_t size = uio->uio_resid;
5073 offset_t offset = uio->uio_loffset;
5074 int blksz;
5075 int fullblk, i;
5076 arc_buf_t *abuf;
5077 ssize_t maxsize;
5078 int preamble, postamble;
5079
5080 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5081 return (SET_ERROR(EINVAL));
5082
5083 ZFS_ENTER(zfsvfs);
5084 ZFS_VERIFY_ZP(zp);
5085 switch (ioflag) {
5086 case UIO_WRITE:
5087 /*
5088 * Loan out an arc_buf for write if write size is bigger than
5089 * max_blksz, and the file's block size is also max_blksz.
5090 */
5091 blksz = max_blksz;
5092 if (size < blksz || zp->z_blksz != blksz) {
5093 ZFS_EXIT(zfsvfs);
5094 return (SET_ERROR(EINVAL));
5095 }
5096 /*
5097 * Caller requests buffers for write before knowing where the
5098 * write offset might be (e.g. NFS TCP write).
5099 */
5100 if (offset == -1) {
5101 preamble = 0;
5102 } else {
5103 preamble = P2PHASE(offset, blksz);
5104 if (preamble) {
5105 preamble = blksz - preamble;
5106 size -= preamble;
5107 }
5108 }
5109
5110 postamble = P2PHASE(size, blksz);
5111 size -= postamble;
5112
5113 fullblk = size / blksz;
5114 (void) dmu_xuio_init(xuio,
5115 (preamble != 0) + fullblk + (postamble != 0));
5116 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5117 int, postamble, int,
5118 (preamble != 0) + fullblk + (postamble != 0));
5119
5120 /*
5121 * Have to fix iov base/len for partial buffers. They
5122 * currently represent full arc_buf's.
5123 */
5124 if (preamble) {
5125 /* data begins in the middle of the arc_buf */
5126 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5127 blksz);
5128 ASSERT(abuf);
5129 (void) dmu_xuio_add(xuio, abuf,
5130 blksz - preamble, preamble);
5131 }
5132
5133 for (i = 0; i < fullblk; i++) {
5134 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5135 blksz);
5136 ASSERT(abuf);
5137 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5138 }
5139
5140 if (postamble) {
5141 /* data ends in the middle of the arc_buf */
5142 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5143 blksz);
5144 ASSERT(abuf);
5145 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5146 }
5147 break;
5148 case UIO_READ:
5149 /*
5150 * Loan out an arc_buf for read if the read size is larger than
5151 * the current file block size. Block alignment is not
5152 * considered. Partial arc_buf will be loaned out for read.
5153 */
5154 blksz = zp->z_blksz;
5155 if (blksz < zcr_blksz_min)
5156 blksz = zcr_blksz_min;
5157 if (blksz > zcr_blksz_max)
5158 blksz = zcr_blksz_max;
5159 /* avoid potential complexity of dealing with it */
5160 if (blksz > max_blksz) {
5161 ZFS_EXIT(zfsvfs);
5162 return (SET_ERROR(EINVAL));
5163 }
5164
5165 maxsize = zp->z_size - uio->uio_loffset;
5166 if (size > maxsize)
5167 size = maxsize;
5168
5169 if (size < blksz || vn_has_cached_data(vp)) {
5170 ZFS_EXIT(zfsvfs);
5171 return (SET_ERROR(EINVAL));
5172 }
5173 break;
5174 default:
5175 ZFS_EXIT(zfsvfs);
5176 return (SET_ERROR(EINVAL));
5177 }
5178
5179 uio->uio_extflg = UIO_XUIO;
5180 XUIO_XUZC_RW(xuio) = ioflag;
5181 ZFS_EXIT(zfsvfs);
5182 return (0);
5183 }
5184
5185 /*ARGSUSED*/
5186 static int
5187 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5188 {
5189 int i;
5190 arc_buf_t *abuf;
5191 int ioflag = XUIO_XUZC_RW(xuio);
5192
5193 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5194
5195 i = dmu_xuio_cnt(xuio);
5196 while (i-- > 0) {
5197 abuf = dmu_xuio_arcbuf(xuio, i);
5198 /*
5199 * if abuf == NULL, it must be a write buffer
5200 * that has been returned in zfs_write().
5201 */
5202 if (abuf)
5203 dmu_return_arcbuf(abuf);
5204 ASSERT(abuf || ioflag == UIO_WRITE);
5205 }
5206
5207 dmu_xuio_fini(xuio);
5208 return (0);
5209 }
5210
5211 /*
5212 * Predeclare these here so that the compiler assumes that
5213 * this is an "old style" function declaration that does
5214 * not include arguments => we won't get type mismatch errors
5215 * in the initializations that follow.
5216 */
5217 static int zfs_inval();
5218 static int zfs_isdir();
5219
5220 static int
5221 zfs_inval()
5222 {
5223 return (SET_ERROR(EINVAL));
5224 }
5225
5226 static int
5227 zfs_isdir()
5228 {
5229 return (SET_ERROR(EISDIR));
5230 }
5231 /*
5232 * Directory vnode operations template
5233 */
5234 vnodeops_t *zfs_dvnodeops;
5235 const fs_operation_def_t zfs_dvnodeops_template[] = {
5236 VOPNAME_OPEN, { .vop_open = zfs_open },
5237 VOPNAME_CLOSE, { .vop_close = zfs_close },
5238 VOPNAME_READ, { .error = zfs_isdir },
5239 VOPNAME_WRITE, { .error = zfs_isdir },
5240 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5241 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5242 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5243 VOPNAME_ACCESS, { .vop_access = zfs_access },
5244 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5245 VOPNAME_CREATE, { .vop_create = zfs_create },
5246 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5247 VOPNAME_LINK, { .vop_link = zfs_link },
5248 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5249 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5250 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5251 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5252 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5253 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5254 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5255 VOPNAME_FID, { .vop_fid = zfs_fid },
5256 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5257 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5258 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5259 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5260 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5261 NULL, NULL
5262 };
5263
5264 /*
5265 * Regular file vnode operations template
5266 */
5267 vnodeops_t *zfs_fvnodeops;
5268 const fs_operation_def_t zfs_fvnodeops_template[] = {
5269 VOPNAME_OPEN, { .vop_open = zfs_open },
5270 VOPNAME_CLOSE, { .vop_close = zfs_close },
5271 VOPNAME_READ, { .vop_read = zfs_read },
5272 VOPNAME_WRITE, { .vop_write = zfs_write },
5273 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5274 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5275 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5276 VOPNAME_ACCESS, { .vop_access = zfs_access },
5277 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5278 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5279 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5280 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5281 VOPNAME_FID, { .vop_fid = zfs_fid },
5282 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5283 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5284 VOPNAME_SPACE, { .vop_space = zfs_space },
5285 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5286 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5287 VOPNAME_MAP, { .vop_map = zfs_map },
5288 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5289 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5290 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5291 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5292 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5293 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5294 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5295 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5296 NULL, NULL
5297 };
5298
5299 /*
5300 * Symbolic link vnode operations template
5301 */
5302 vnodeops_t *zfs_symvnodeops;
5303 const fs_operation_def_t zfs_symvnodeops_template[] = {
5304 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5305 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5306 VOPNAME_ACCESS, { .vop_access = zfs_access },
5307 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5308 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5309 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5310 VOPNAME_FID, { .vop_fid = zfs_fid },
5311 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5312 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5313 NULL, NULL
5314 };
5315
5316 /*
5317 * special share hidden files vnode operations template
5318 */
5319 vnodeops_t *zfs_sharevnodeops;
5320 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5321 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5322 VOPNAME_ACCESS, { .vop_access = zfs_access },
5323 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5324 VOPNAME_FID, { .vop_fid = zfs_fid },
5325 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5326 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5327 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5328 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5329 NULL, NULL
5330 };
5331
5332 /*
5333 * Extended attribute directory vnode operations template
5334 *
5335 * This template is identical to the directory vnodes
5336 * operation template except for restricted operations:
5337 * VOP_MKDIR()
5338 * VOP_SYMLINK()
5339 *
5340 * Note that there are other restrictions embedded in:
5341 * zfs_create() - restrict type to VREG
5342 * zfs_link() - no links into/out of attribute space
5343 * zfs_rename() - no moves into/out of attribute space
5344 */
5345 vnodeops_t *zfs_xdvnodeops;
5346 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5347 VOPNAME_OPEN, { .vop_open = zfs_open },
5348 VOPNAME_CLOSE, { .vop_close = zfs_close },
5349 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5350 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5351 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5352 VOPNAME_ACCESS, { .vop_access = zfs_access },
5353 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5354 VOPNAME_CREATE, { .vop_create = zfs_create },
5355 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5356 VOPNAME_LINK, { .vop_link = zfs_link },
5357 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5358 VOPNAME_MKDIR, { .error = zfs_inval },
5359 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5360 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5361 VOPNAME_SYMLINK, { .error = zfs_inval },
5362 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5363 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5364 VOPNAME_FID, { .vop_fid = zfs_fid },
5365 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5366 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5367 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5368 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5369 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5370 NULL, NULL
5371 };
5372
5373 /*
5374 * Error vnode operations template
5375 */
5376 vnodeops_t *zfs_evnodeops;
5377 const fs_operation_def_t zfs_evnodeops_template[] = {
5378 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5379 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5380 NULL, NULL
5381 };