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, 2014 by Delphix. All rights reserved.
24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
25 */
26
27 /* Portions Copyright 2007 Jeremy Teo */
28 /* Portions Copyright 2010 Robert Milkowski */
29
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/time.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
36 #include <sys/vfs.h>
37 #include <sys/vfs_opreg.h>
38 #include <sys/vnode.h>
39 #include <sys/file.h>
40 #include <sys/stat.h>
41 #include <sys/kmem.h>
42 #include <sys/taskq.h>
43 #include <sys/uio.h>
44 #include <sys/vmsystm.h>
45 #include <sys/atomic.h>
46 #include <sys/vm.h>
47 #include <vm/seg_vn.h>
48 #include <vm/pvn.h>
49 #include <vm/as.h>
50 #include <vm/kpm.h>
51 #include <vm/seg_kpm.h>
52 #include <sys/mman.h>
53 #include <sys/pathname.h>
54 #include <sys/cmn_err.h>
55 #include <sys/errno.h>
56 #include <sys/unistd.h>
57 #include <sys/zfs_dir.h>
58 #include <sys/zfs_acl.h>
59 #include <sys/zfs_ioctl.h>
60 #include <sys/fs/zfs.h>
61 #include <sys/dmu.h>
62 #include <sys/dmu_objset.h>
63 #include <sys/spa.h>
64 #include <sys/txg.h>
65 #include <sys/dbuf.h>
66 #include <sys/zap.h>
67 #include <sys/sa.h>
68 #include <sys/dirent.h>
69 #include <sys/policy.h>
70 #include <sys/sunddi.h>
71 #include <sys/filio.h>
72 #include <sys/sid.h>
73 #include "fs/fs_subr.h"
74 #include <sys/zfs_ctldir.h>
75 #include <sys/zfs_fuid.h>
76 #include <sys/zfs_sa.h>
77 #include <sys/zfeature.h>
78 #include <sys/dnlc.h>
79 #include <sys/zfs_rlock.h>
80 #include <sys/extdirent.h>
81 #include <sys/kidmap.h>
82 #include <sys/cred.h>
83 #include <sys/attr.h>
84
85 /*
86 * Programming rules.
87 *
88 * Each vnode op performs some logical unit of work. To do this, the ZPL must
89 * properly lock its in-core state, create a DMU transaction, do the work,
90 * record this work in the intent log (ZIL), commit the DMU transaction,
91 * and wait for the intent log to commit if it is a synchronous operation.
92 * Moreover, the vnode ops must work in both normal and log replay context.
93 * The ordering of events is important to avoid deadlocks and references
94 * to freed memory. The example below illustrates the following Big Rules:
95 *
96 * (1) A check must be made in each zfs thread for a mounted file system.
97 * This is done avoiding races using ZFS_ENTER(zfsvfs).
98 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
99 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
100 * can return EIO from the calling function.
101 *
102 * (2) VN_RELE() should always be the last thing except for zil_commit()
103 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
104 * First, if it's the last reference, the vnode/znode
105 * can be freed, so the zp may point to freed memory. Second, the last
106 * reference will call zfs_zinactive(), which may induce a lot of work --
107 * pushing cached pages (which acquires range locks) and syncing out
108 * cached atime changes. Third, zfs_zinactive() may require a new tx,
109 * which could deadlock the system if you were already holding one.
110 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
111 *
112 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
113 * as they can span dmu_tx_assign() calls.
114 *
115 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
116 * dmu_tx_assign(). This is critical because we don't want to block
117 * while holding locks.
118 *
119 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
120 * reduces lock contention and CPU usage when we must wait (note that if
121 * throughput is constrained by the storage, nearly every transaction
122 * must wait).
123 *
124 * Note, in particular, that if a lock is sometimes acquired before
125 * the tx assigns, and sometimes after (e.g. z_lock), then failing
126 * to use a non-blocking assign can deadlock the system. The scenario:
127 *
128 * Thread A has grabbed a lock before calling dmu_tx_assign().
129 * Thread B is in an already-assigned tx, and blocks for this lock.
130 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
131 * forever, because the previous txg can't quiesce until B's tx commits.
132 *
133 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
134 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
135 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
136 * to indicate that this operation has already called dmu_tx_wait().
137 * This will ensure that we don't retry forever, waiting a short bit
138 * each time.
139 *
140 * (5) If the operation succeeded, generate the intent log entry for it
141 * before dropping locks. This ensures that the ordering of events
142 * in the intent log matches the order in which they actually occurred.
143 * During ZIL replay the zfs_log_* functions will update the sequence
144 * number to indicate the zil transaction has replayed.
145 *
146 * (6) At the end of each vnode op, the DMU tx must always commit,
147 * regardless of whether there were any errors.
148 *
149 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
150 * to ensure that synchronous semantics are provided when necessary.
151 *
152 * In general, this is how things should be ordered in each vnode op:
153 *
154 * ZFS_ENTER(zfsvfs); // exit if unmounted
155 * top:
156 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
157 * rw_enter(...); // grab any other locks you need
158 * tx = dmu_tx_create(...); // get DMU tx
159 * dmu_tx_hold_*(); // hold each object you might modify
160 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
161 * if (error) {
162 * rw_exit(...); // drop locks
163 * zfs_dirent_unlock(dl); // unlock directory entry
164 * VN_RELE(...); // release held vnodes
165 * if (error == ERESTART) {
166 * waited = B_TRUE;
167 * dmu_tx_wait(tx);
168 * dmu_tx_abort(tx);
169 * goto top;
170 * }
171 * dmu_tx_abort(tx); // abort DMU tx
172 * ZFS_EXIT(zfsvfs); // finished in zfs
173 * return (error); // really out of space
174 * }
175 * error = do_real_work(); // do whatever this VOP does
176 * if (error == 0)
177 * zfs_log_*(...); // on success, make ZIL entry
178 * dmu_tx_commit(tx); // commit DMU tx -- error or not
179 * rw_exit(...); // drop locks
180 * zfs_dirent_unlock(dl); // unlock directory entry
181 * VN_RELE(...); // release held vnodes
182 * zil_commit(zilog, foid); // synchronous when necessary
183 * ZFS_EXIT(zfsvfs); // finished in zfs
184 * return (error); // done, report error
185 */
186
187 /* ARGSUSED */
188 static int
189 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
190 {
191 znode_t *zp = VTOZ(*vpp);
192 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
193
194 ZFS_ENTER(zfsvfs);
195 ZFS_VERIFY_ZP(zp);
196
197 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
198 ((flag & FAPPEND) == 0)) {
199 ZFS_EXIT(zfsvfs);
200 return (SET_ERROR(EPERM));
201 }
202
203 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
204 ZTOV(zp)->v_type == VREG &&
205 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
206 if (fs_vscan(*vpp, cr, 0) != 0) {
207 ZFS_EXIT(zfsvfs);
208 return (SET_ERROR(EACCES));
209 }
210 }
211
212 /* Keep a count of the synchronous opens in the znode */
213 if (flag & (FSYNC | FDSYNC))
214 atomic_inc_32(&zp->z_sync_cnt);
215
216 ZFS_EXIT(zfsvfs);
217 return (0);
218 }
219
220 /* ARGSUSED */
221 static int
222 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
223 caller_context_t *ct)
224 {
225 znode_t *zp = VTOZ(vp);
226 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
227
228 /*
229 * Clean up any locks held by this process on the vp.
230 */
231 cleanlocks(vp, ddi_get_pid(), 0);
232 cleanshares(vp, ddi_get_pid());
233
234 ZFS_ENTER(zfsvfs);
235 ZFS_VERIFY_ZP(zp);
236
237 /* Decrement the synchronous opens in the znode */
238 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
239 atomic_dec_32(&zp->z_sync_cnt);
240
241 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
242 ZTOV(zp)->v_type == VREG &&
243 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
244 VERIFY(fs_vscan(vp, cr, 1) == 0);
245
246 ZFS_EXIT(zfsvfs);
247 return (0);
248 }
249
250 /*
251 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
252 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
253 */
254 static int
255 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
256 {
257 znode_t *zp = VTOZ(vp);
258 uint64_t noff = (uint64_t)*off; /* new offset */
259 uint64_t file_sz;
260 int error;
261 boolean_t hole;
262
263 file_sz = zp->z_size;
264 if (noff >= file_sz) {
265 return (SET_ERROR(ENXIO));
266 }
267
268 if (cmd == _FIO_SEEK_HOLE)
269 hole = B_TRUE;
270 else
271 hole = B_FALSE;
272
273 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
274
275 if (error == ESRCH)
276 return (SET_ERROR(ENXIO));
277
278 /*
279 * We could find a hole that begins after the logical end-of-file,
280 * because dmu_offset_next() only works on whole blocks. If the
281 * EOF falls mid-block, then indicate that the "virtual hole"
282 * at the end of the file begins at the logical EOF, rather than
283 * at the end of the last block.
284 */
285 if (noff > file_sz) {
286 ASSERT(hole);
287 noff = file_sz;
288 }
289
290 if (noff < *off)
291 return (error);
292 *off = noff;
293 return (error);
294 }
295
296
297 static int zfs_zero_write(vnode_t *vp, uint64_t size, cred_t *cr,
298 caller_context_t *ct);
299
300 /* ARGSUSED */
301 static int
302 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
303 int *rvalp, caller_context_t *ct)
304 {
305 offset_t off;
306 int error;
307 zfsvfs_t *zfsvfs;
308 znode_t *zp;
309 uint64_t size;
310
311 switch (com) {
312 case _FIOFFS:
313 return (zfs_sync(vp->v_vfsp, 0, cred));
314
315 /*
316 * The following two ioctls are used by bfu. Faking out,
317 * necessary to avoid bfu errors.
318 */
319 case _FIOGDIO:
320 case _FIOSDIO:
321 return (0);
322
323 case _FIO_SEEK_DATA:
324 case _FIO_SEEK_HOLE:
325 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
326 return (SET_ERROR(EFAULT));
327
328 zp = VTOZ(vp);
329 zfsvfs = zp->z_zfsvfs;
330 ZFS_ENTER(zfsvfs);
331 ZFS_VERIFY_ZP(zp);
332
333 /* offset parameter is in/out */
334 error = zfs_holey(vp, com, &off);
335 ZFS_EXIT(zfsvfs);
336 if (error)
337 return (error);
338 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
339 return (SET_ERROR(EFAULT));
340 return (0);
341 case _FIO_RESERVE_SPACE:
342 if (ddi_copyin((void *)data, &size, sizeof (size), flag))
343 return (EFAULT);
344 error = zfs_zero_write(vp, size, cred, ct);
345 return (error);
346 }
347 return (SET_ERROR(ENOTTY));
348 }
349
350 /*
351 * Utility functions to map and unmap a single physical page. These
352 * are used to manage the mappable copies of ZFS file data, and therefore
353 * do not update ref/mod bits.
354 */
355 caddr_t
356 zfs_map_page(page_t *pp, enum seg_rw rw)
357 {
358 if (kpm_enable)
359 return (hat_kpm_mapin(pp, 0));
360 ASSERT(rw == S_READ || rw == S_WRITE);
361 return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
362 (caddr_t)-1));
363 }
364
365 void
366 zfs_unmap_page(page_t *pp, caddr_t addr)
367 {
368 if (kpm_enable) {
369 hat_kpm_mapout(pp, 0, addr);
370 } else {
371 ppmapout(addr);
372 }
373 }
374
375 /*
376 * When a file is memory mapped, we must keep the IO data synchronized
377 * between the DMU cache and the memory mapped pages. What this means:
378 *
379 * On Write: If we find a memory mapped page, we write to *both*
380 * the page and the dmu buffer.
381 */
382 static void
383 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
384 {
385 int64_t off;
386
387 off = start & PAGEOFFSET;
388 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
389 page_t *pp;
390 uint64_t nbytes = MIN(PAGESIZE - off, len);
391
392 if (pp = page_lookup(vp, start, SE_SHARED)) {
393 caddr_t va;
394
395 va = zfs_map_page(pp, S_WRITE);
396 (void) dmu_read(os, oid, start+off, nbytes, va+off,
397 DMU_READ_PREFETCH);
398 zfs_unmap_page(pp, va);
399 page_unlock(pp);
400 }
401 len -= nbytes;
402 off = 0;
403 }
404 }
405
406 /*
407 * When a file is memory mapped, we must keep the IO data synchronized
408 * between the DMU cache and the memory mapped pages. What this means:
409 *
410 * On Read: We "read" preferentially from memory mapped pages,
411 * else we default from the dmu buffer.
412 *
413 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
414 * the file is memory mapped.
415 */
416 static int
417 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
418 {
419 znode_t *zp = VTOZ(vp);
420 int64_t start, off;
421 int len = nbytes;
422 int error = 0;
423
424 start = uio->uio_loffset;
425 off = start & PAGEOFFSET;
426 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
427 page_t *pp;
428 uint64_t bytes = 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_READ);
434 error = uiomove(va + off, bytes, UIO_READ, uio);
435 zfs_unmap_page(pp, va);
436 page_unlock(pp);
437 } else {
438 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
439 uio, bytes);
440 }
441 len -= bytes;
442 off = 0;
443 if (error)
444 break;
445 }
446 return (error);
447 }
448
449 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
450
451 /*
452 * Read bytes from specified file into supplied buffer.
453 *
454 * IN: vp - vnode of file to be read from.
455 * uio - structure supplying read location, range info,
456 * and return buffer.
457 * ioflag - SYNC flags; used to provide FRSYNC semantics.
458 * cr - credentials of caller.
459 * ct - caller context
460 *
461 * OUT: uio - updated offset and range, buffer filled.
462 *
463 * RETURN: 0 on success, error code on failure.
464 *
465 * Side Effects:
466 * vp - atime updated if byte count > 0
467 */
468 /* ARGSUSED */
469 static int
470 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
471 {
472 znode_t *zp = VTOZ(vp);
473 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
474 ssize_t n, nbytes;
475 int error = 0;
476 rl_t *rl;
477 xuio_t *xuio = NULL;
478
479 ZFS_ENTER(zfsvfs);
480 ZFS_VERIFY_ZP(zp);
481
482 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
483 ZFS_EXIT(zfsvfs);
484 return (SET_ERROR(EACCES));
485 }
486
487 /*
488 * Validate file offset
489 */
490 if (uio->uio_loffset < (offset_t)0) {
491 ZFS_EXIT(zfsvfs);
492 return (SET_ERROR(EINVAL));
493 }
494
495 /*
496 * Fasttrack empty reads
497 */
498 if (uio->uio_resid == 0) {
499 ZFS_EXIT(zfsvfs);
500 return (0);
501 }
502
503 /*
504 * Check for mandatory locks
505 */
506 if (MANDMODE(zp->z_mode)) {
507 if (error = chklock(vp, FREAD,
508 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
509 ZFS_EXIT(zfsvfs);
510 return (error);
511 }
512 }
513
514 /*
515 * If we're in FRSYNC mode, sync out this znode before reading it.
516 */
517 if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
518 zil_commit(zfsvfs->z_log, zp->z_id);
519
520 /*
521 * Lock the range against changes.
522 */
523 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
524
525 /*
526 * If we are reading past end-of-file we can skip
527 * to the end; but we might still need to set atime.
528 */
529 if (uio->uio_loffset >= zp->z_size) {
530 error = 0;
531 goto out;
532 }
533
534 ASSERT(uio->uio_loffset < zp->z_size);
535 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
536
537 if ((uio->uio_extflg == UIO_XUIO) &&
538 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
539 int nblk;
540 int blksz = zp->z_blksz;
541 uint64_t offset = uio->uio_loffset;
542
543 xuio = (xuio_t *)uio;
544 if ((ISP2(blksz))) {
545 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
546 blksz)) / blksz;
547 } else {
548 ASSERT(offset + n <= blksz);
549 nblk = 1;
550 }
551 (void) dmu_xuio_init(xuio, nblk);
552
553 if (vn_has_cached_data(vp)) {
554 /*
555 * For simplicity, we always allocate a full buffer
556 * even if we only expect to read a portion of a block.
557 */
558 while (--nblk >= 0) {
559 (void) dmu_xuio_add(xuio,
560 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
561 blksz), 0, blksz);
562 }
563 }
564 }
565
566 while (n > 0) {
567 nbytes = MIN(n, zfs_read_chunk_size -
568 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
569
570 if (vn_has_cached_data(vp)) {
571 error = mappedread(vp, nbytes, uio);
572 } else {
573 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
574 uio, nbytes);
575 }
576 if (error) {
577 /* convert checksum errors into IO errors */
578 if (error == ECKSUM)
579 error = SET_ERROR(EIO);
580 break;
581 }
582
583 n -= nbytes;
584 }
585 out:
586 zfs_range_unlock(rl);
587
588 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
589 ZFS_EXIT(zfsvfs);
590 return (error);
591 }
592
593 /*
594 * Write the bytes to a file.
595 *
596 * IN: vp - vnode of file to be written to.
597 * uio - structure supplying write location, range info,
598 * and data buffer.
599 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
600 * set if in append mode.
601 * cr - credentials of caller.
602 * ct - caller context (NFS/CIFS fem monitor only)
603 *
604 * OUT: uio - updated offset and range.
605 *
606 * RETURN: 0 on success, error code on failure.
607 *
608 * Timestamps:
609 * vp - ctime|mtime updated if byte count > 0
610 */
611
612 /* ARGSUSED */
613 static int
614 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
615 {
616 znode_t *zp = VTOZ(vp);
617 rlim64_t limit = uio->uio_llimit;
618 ssize_t start_resid = uio->uio_resid;
619 ssize_t tx_bytes;
620 uint64_t end_size;
621 dmu_tx_t *tx;
622 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
623 zilog_t *zilog;
624 offset_t woff;
625 ssize_t n, nbytes;
626 rl_t *rl;
627 int max_blksz = zfsvfs->z_max_blksz;
628 int error = 0;
629 arc_buf_t *abuf;
630 iovec_t *aiov = NULL;
631 xuio_t *xuio = NULL;
632 int i_iov = 0;
633 int iovcnt = uio->uio_iovcnt;
634 iovec_t *iovp = uio->uio_iov;
635 int write_eof;
636 int count = 0;
637 sa_bulk_attr_t bulk[4];
638 uint64_t mtime[2], ctime[2];
639
640 /*
641 * Fasttrack empty write
642 */
643 n = start_resid;
644 if (n == 0)
645 return (0);
646
647 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
648 limit = MAXOFFSET_T;
649
650 ZFS_ENTER(zfsvfs);
651 ZFS_VERIFY_ZP(zp);
652
653 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
654 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
655 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
656 &zp->z_size, 8);
657 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
658 &zp->z_pflags, 8);
659
660 /*
661 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
662 * callers might not be able to detect properly that we are read-only,
663 * so check it explicitly here.
664 */
665 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
666 ZFS_EXIT(zfsvfs);
667 return (SET_ERROR(EROFS));
668 }
669
670 /*
671 * If immutable or not appending then return EPERM
672 */
673 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
674 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
675 (uio->uio_loffset < zp->z_size))) {
676 ZFS_EXIT(zfsvfs);
677 return (SET_ERROR(EPERM));
678 }
679
680 zilog = zfsvfs->z_log;
681
682 /*
683 * Validate file offset
684 */
685 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
686 if (woff < 0) {
687 ZFS_EXIT(zfsvfs);
688 return (SET_ERROR(EINVAL));
689 }
690
691 /*
692 * Check for mandatory locks before calling zfs_range_lock()
693 * in order to prevent a deadlock with locks set via fcntl().
694 */
695 if (MANDMODE((mode_t)zp->z_mode) &&
696 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
697 ZFS_EXIT(zfsvfs);
698 return (error);
699 }
700
701 /*
702 * Pre-fault the pages to ensure slow (eg NFS) pages
703 * don't hold up txg.
704 * Skip this if uio contains loaned arc_buf.
705 */
706 if ((uio->uio_extflg == UIO_XUIO) &&
707 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
708 xuio = (xuio_t *)uio;
709 else
710 uio_prefaultpages(MIN(n, max_blksz), uio);
711
712 /*
713 * If in append mode, set the io offset pointer to eof.
714 */
715 if (ioflag & FAPPEND) {
716 /*
717 * Obtain an appending range lock to guarantee file append
718 * semantics. We reset the write offset once we have the lock.
719 */
720 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
721 woff = rl->r_off;
722 if (rl->r_len == UINT64_MAX) {
723 /*
724 * We overlocked the file because this write will cause
725 * the file block size to increase.
726 * Note that zp_size cannot change with this lock held.
727 */
728 woff = zp->z_size;
729 }
730 uio->uio_loffset = woff;
731 } else {
732 /*
733 * Note that if the file block size will change as a result of
734 * this write, then this range lock will lock the entire file
735 * so that we can re-write the block safely.
736 */
737 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
738 }
739
740 if (woff >= limit) {
741 zfs_range_unlock(rl);
742 ZFS_EXIT(zfsvfs);
743 return (SET_ERROR(EFBIG));
744 }
745
746 if ((woff + n) > limit || woff > (limit - n))
747 n = limit - woff;
748
749 /* Will this write extend the file length? */
750 write_eof = (woff + n > zp->z_size);
751
752 end_size = MAX(zp->z_size, woff + n);
753
754 /*
755 * Write the file in reasonable size chunks. Each chunk is written
756 * in a separate transaction; this keeps the intent log records small
757 * and allows us to do more fine-grained space accounting.
758 */
759 while (n > 0) {
760 abuf = NULL;
761 woff = uio->uio_loffset;
762 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
763 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
764 if (abuf != NULL)
765 dmu_return_arcbuf(abuf);
766 error = SET_ERROR(EDQUOT);
767 break;
768 }
769
770 if (xuio && abuf == NULL) {
771 ASSERT(i_iov < iovcnt);
772 aiov = &iovp[i_iov];
773 abuf = dmu_xuio_arcbuf(xuio, i_iov);
774 dmu_xuio_clear(xuio, i_iov);
775 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
776 iovec_t *, aiov, arc_buf_t *, abuf);
777 ASSERT((aiov->iov_base == abuf->b_data) ||
778 ((char *)aiov->iov_base - (char *)abuf->b_data +
779 aiov->iov_len == arc_buf_size(abuf)));
780 i_iov++;
781 } else if (abuf == NULL && n >= max_blksz &&
782 woff >= zp->z_size &&
783 P2PHASE(woff, max_blksz) == 0 &&
784 zp->z_blksz == max_blksz) {
785 /*
786 * This write covers a full block. "Borrow" a buffer
787 * from the dmu so that we can fill it before we enter
788 * a transaction. This avoids the possibility of
789 * holding up the transaction if the data copy hangs
790 * up on a pagefault (e.g., from an NFS server mapping).
791 */
792 size_t cbytes;
793
794 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
795 max_blksz);
796 ASSERT(abuf != NULL);
797 ASSERT(arc_buf_size(abuf) == max_blksz);
798 if (error = uiocopy(abuf->b_data, max_blksz,
799 UIO_WRITE, uio, &cbytes)) {
800 dmu_return_arcbuf(abuf);
801 break;
802 }
803 ASSERT(cbytes == max_blksz);
804 }
805
806 /*
807 * Start a transaction.
808 */
809 tx = dmu_tx_create(zfsvfs->z_os);
810 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
811 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
812 zfs_sa_upgrade_txholds(tx, zp);
813 error = dmu_tx_assign(tx, TXG_WAIT);
814 if (error) {
815 dmu_tx_abort(tx);
816 if (abuf != NULL)
817 dmu_return_arcbuf(abuf);
818 break;
819 }
820
821 /*
822 * If zfs_range_lock() over-locked we grow the blocksize
823 * and then reduce the lock range. This will only happen
824 * on the first iteration since zfs_range_reduce() will
825 * shrink down r_len to the appropriate size.
826 */
827 if (rl->r_len == UINT64_MAX) {
828 uint64_t new_blksz;
829
830 if (zp->z_blksz > max_blksz) {
831 ASSERT(!ISP2(zp->z_blksz));
832 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
833 } else {
834 new_blksz = MIN(end_size, max_blksz);
835 }
836 zfs_grow_blocksize(zp, new_blksz, tx);
837 zfs_range_reduce(rl, woff, n);
838 }
839
840 /*
841 * XXX - should we really limit each write to z_max_blksz?
842 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
843 */
844 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
845
846 if (abuf == NULL) {
847 tx_bytes = uio->uio_resid;
848 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
849 uio, nbytes, tx);
850 tx_bytes -= uio->uio_resid;
851 } else {
852 tx_bytes = nbytes;
853 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
854 /*
855 * If this is not a full block write, but we are
856 * extending the file past EOF and this data starts
857 * block-aligned, use assign_arcbuf(). Otherwise,
858 * write via dmu_write().
859 */
860 if (tx_bytes < max_blksz && (!write_eof ||
861 aiov->iov_base != abuf->b_data)) {
862 ASSERT(xuio);
863 dmu_write(zfsvfs->z_os, zp->z_id, woff,
864 aiov->iov_len, aiov->iov_base, tx);
865 dmu_return_arcbuf(abuf);
866 xuio_stat_wbuf_copied();
867 } else {
868 ASSERT(xuio || tx_bytes == max_blksz);
869 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
870 woff, abuf, tx);
871 }
872 ASSERT(tx_bytes <= uio->uio_resid);
873 uioskip(uio, tx_bytes);
874 }
875 if (tx_bytes && vn_has_cached_data(vp)) {
876 update_pages(vp, woff,
877 tx_bytes, zfsvfs->z_os, zp->z_id);
878 }
879
880 /*
881 * If we made no progress, we're done. If we made even
882 * partial progress, update the znode and ZIL accordingly.
883 */
884 if (tx_bytes == 0) {
885 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
886 (void *)&zp->z_size, sizeof (uint64_t), tx);
887 dmu_tx_commit(tx);
888 ASSERT(error != 0);
889 break;
890 }
891
892 /*
893 * Clear Set-UID/Set-GID bits on successful write if not
894 * privileged and at least one of the excute bits is set.
895 *
896 * It would be nice to to this after all writes have
897 * been done, but that would still expose the ISUID/ISGID
898 * to another app after the partial write is committed.
899 *
900 * Note: we don't call zfs_fuid_map_id() here because
901 * user 0 is not an ephemeral uid.
902 */
903 mutex_enter(&zp->z_acl_lock);
904 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
905 (S_IXUSR >> 6))) != 0 &&
906 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
907 secpolicy_vnode_setid_retain(cr,
908 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
909 uint64_t newmode;
910 zp->z_mode &= ~(S_ISUID | S_ISGID);
911 newmode = zp->z_mode;
912 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
913 (void *)&newmode, sizeof (uint64_t), tx);
914 }
915 mutex_exit(&zp->z_acl_lock);
916
917 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
918 B_TRUE);
919
920 /*
921 * Update the file size (zp_size) if it has changed;
922 * account for possible concurrent updates.
923 */
924 while ((end_size = zp->z_size) < uio->uio_loffset) {
925 (void) atomic_cas_64(&zp->z_size, end_size,
926 uio->uio_loffset);
927 ASSERT(error == 0);
928 }
929 /*
930 * If we are replaying and eof is non zero then force
931 * the file size to the specified eof. Note, there's no
932 * concurrency during replay.
933 */
934 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
935 zp->z_size = zfsvfs->z_replay_eof;
936
937 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
938
939 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
940 dmu_tx_commit(tx);
941
942 if (error != 0)
943 break;
944 ASSERT(tx_bytes == nbytes);
945 n -= nbytes;
946
947 if (!xuio && n > 0)
948 uio_prefaultpages(MIN(n, max_blksz), uio);
949 }
950
951 zfs_range_unlock(rl);
952
953 /*
954 * If we're in replay mode, or we made no progress, return error.
955 * Otherwise, it's at least a partial write, so it's successful.
956 */
957 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
958 ZFS_EXIT(zfsvfs);
959 return (error);
960 }
961
962 if (ioflag & (FSYNC | FDSYNC) ||
963 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
964 zil_commit(zilog, zp->z_id);
965
966 ZFS_EXIT(zfsvfs);
967 return (0);
968 }
969
970 #define ZFS_RESERVE_CHUNK (2 * 1024 * 1024)
971 /* ARGSUSED */
972 static int
973 zfs_zero_write(vnode_t *vp, uint64_t size, cred_t *cr, caller_context_t *ct)
974 {
975 znode_t *zp = VTOZ(vp);
976 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
977 int count = 0;
978 sa_bulk_attr_t bulk[4];
979 uint64_t mtime[2], ctime[2];
980 rl_t *rl;
981 int error = 0;
982 dmu_tx_t *tx = NULL;
983 uint64_t end_size;
984 uint64_t pos = 0;
985
986 if (zp->z_size > 0)
987 return (EFBIG);
988 if (size == 0)
989 return (0);
990
991 ZFS_ENTER(zfsvfs);
992 ZFS_VERIFY_ZP(zp);
993
994 if (!spa_feature_is_enabled(zfsvfs->z_os->os_spa,
995 SPA_FEATURE_SPACE_RESERVATION))
996 {
997 ZFS_EXIT(zfsvfs);
998 return (ENOTSUP);
999 }
1000
1001 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
1002 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
1003 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1004 &zp->z_size, 8);
1005 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1006 &zp->z_pflags, 8);
1007
1008 /*
1009 * If immutable or not appending then return EPERM
1010 */
1011 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY))) {
1012 ZFS_EXIT(zfsvfs);
1013 return (EPERM);
1014 }
1015
1016 rl = zfs_range_lock(zp, 0, size, RL_WRITER);
1017
1018 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
1019 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
1020 error = EDQUOT;
1021 goto out;
1022 }
1023
1024 while (pos < size) {
1025 uint64_t length = size - pos;
1026 length = MIN(length, ZFS_RESERVE_CHUNK);
1027 again:
1028 tx = dmu_tx_create(zfsvfs->z_os);
1029 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1030 dmu_tx_hold_write(tx, zp->z_id, pos, length);
1031 zfs_sa_upgrade_txholds(tx, zp);
1032 error = dmu_tx_assign(tx, TXG_NOWAIT);
1033 if (error) {
1034 if (error == ERESTART) {
1035 dmu_tx_wait(tx);
1036 dmu_tx_abort(tx);
1037 goto again;
1038 }
1039 dmu_tx_abort(tx);
1040 goto out;
1041 }
1042
1043 if (pos == 0)
1044 zfs_grow_blocksize(zp, MIN(size, zfsvfs->z_max_blksz), tx);
1045 dmu_write_zero(zfsvfs->z_os, zp->z_id, pos, length, tx);
1046
1047 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1048
1049 pos += length;
1050 while ((end_size = zp->z_size) < pos)
1051 (void) atomic_cas_64(&zp->z_size, end_size, pos);
1052
1053 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1054
1055 dmu_tx_commit(tx);
1056 if (error)
1057 goto out;
1058 }
1059 out:
1060 zfs_range_unlock(rl);
1061 ZFS_EXIT(zfsvfs);
1062
1063 return (error);
1064 }
1065
1066 void
1067 zfs_get_done(zgd_t *zgd, int error)
1068 {
1069 znode_t *zp = zgd->zgd_private;
1070 objset_t *os = zp->z_zfsvfs->z_os;
1071
1072 if (zgd->zgd_db)
1073 dmu_buf_rele(zgd->zgd_db, zgd);
1074
1075 zfs_range_unlock(zgd->zgd_rl);
1076
1077 /*
1078 * Release the vnode asynchronously as we currently have the
1079 * txg stopped from syncing.
1080 */
1081 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1082
1083 if (error == 0 && zgd->zgd_bp)
1084 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1085
1086 kmem_free(zgd, sizeof (zgd_t));
1087 }
1088
1089 #ifdef DEBUG
1090 static int zil_fault_io = 0;
1091 #endif
1092
1093 /*
1094 * Get data to generate a TX_WRITE intent log record.
1095 */
1096 int
1097 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1098 {
1099 zfsvfs_t *zfsvfs = arg;
1100 objset_t *os = zfsvfs->z_os;
1101 znode_t *zp;
1102 uint64_t object = lr->lr_foid;
1103 uint64_t offset = lr->lr_offset;
1104 uint64_t size = lr->lr_length;
1105 blkptr_t *bp = &lr->lr_blkptr;
1106 dmu_buf_t *db;
1107 zgd_t *zgd;
1108 int error = 0;
1109
1110 ASSERT(zio != NULL);
1111 ASSERT(size != 0);
1112
1113 /*
1114 * Nothing to do if the file has been removed
1115 */
1116 if (zfs_zget(zfsvfs, object, &zp) != 0)
1117 return (SET_ERROR(ENOENT));
1118 if (zp->z_unlinked) {
1119 /*
1120 * Release the vnode asynchronously as we currently have the
1121 * txg stopped from syncing.
1122 */
1123 VN_RELE_ASYNC(ZTOV(zp),
1124 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1125 return (SET_ERROR(ENOENT));
1126 }
1127
1128 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1129 zgd->zgd_zilog = zfsvfs->z_log;
1130 zgd->zgd_private = zp;
1131
1132 /*
1133 * Write records come in two flavors: immediate and indirect.
1134 * For small writes it's cheaper to store the data with the
1135 * log record (immediate); for large writes it's cheaper to
1136 * sync the data and get a pointer to it (indirect) so that
1137 * we don't have to write the data twice.
1138 */
1139 if (buf != NULL) { /* immediate write */
1140 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1141 /* test for truncation needs to be done while range locked */
1142 if (offset >= zp->z_size) {
1143 error = SET_ERROR(ENOENT);
1144 } else {
1145 error = dmu_read(os, object, offset, size, buf,
1146 DMU_READ_NO_PREFETCH);
1147 }
1148 ASSERT(error == 0 || error == ENOENT);
1149 } else { /* indirect write */
1150 /*
1151 * Have to lock the whole block to ensure when it's
1152 * written out and it's checksum is being calculated
1153 * that no one can change the data. We need to re-check
1154 * blocksize after we get the lock in case it's changed!
1155 */
1156 for (;;) {
1157 uint64_t blkoff;
1158 size = zp->z_blksz;
1159 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1160 offset -= blkoff;
1161 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1162 RL_READER);
1163 if (zp->z_blksz == size)
1164 break;
1165 offset += blkoff;
1166 zfs_range_unlock(zgd->zgd_rl);
1167 }
1168 /* test for truncation needs to be done while range locked */
1169 if (lr->lr_offset >= zp->z_size)
1170 error = SET_ERROR(ENOENT);
1171 #ifdef DEBUG
1172 if (zil_fault_io) {
1173 error = SET_ERROR(EIO);
1174 zil_fault_io = 0;
1175 }
1176 #endif
1177 if (error == 0)
1178 error = dmu_buf_hold(os, object, offset, zgd, &db,
1179 DMU_READ_NO_PREFETCH);
1180
1181 if (error == 0) {
1182 blkptr_t *obp = dmu_buf_get_blkptr(db);
1183 if (obp) {
1184 ASSERT(BP_IS_HOLE(bp));
1185 *bp = *obp;
1186 }
1187
1188 zgd->zgd_db = db;
1189 zgd->zgd_bp = bp;
1190
1191 ASSERT(db->db_offset == offset);
1192 ASSERT(db->db_size == size);
1193
1194 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1195 zfs_get_done, zgd);
1196 ASSERT(error || lr->lr_length <= zp->z_blksz);
1197
1198 /*
1199 * On success, we need to wait for the write I/O
1200 * initiated by dmu_sync() to complete before we can
1201 * release this dbuf. We will finish everything up
1202 * in the zfs_get_done() callback.
1203 */
1204 if (error == 0)
1205 return (0);
1206
1207 if (error == EALREADY) {
1208 lr->lr_common.lrc_txtype = TX_WRITE2;
1209 error = 0;
1210 }
1211 }
1212 }
1213
1214 zfs_get_done(zgd, error);
1215
1216 return (error);
1217 }
1218
1219 /*ARGSUSED*/
1220 static int
1221 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1222 caller_context_t *ct)
1223 {
1224 znode_t *zp = VTOZ(vp);
1225 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1226 int error;
1227
1228 ZFS_ENTER(zfsvfs);
1229 ZFS_VERIFY_ZP(zp);
1230
1231 if (flag & V_ACE_MASK)
1232 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1233 else
1234 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1235
1236 ZFS_EXIT(zfsvfs);
1237 return (error);
1238 }
1239
1240 /*
1241 * If vnode is for a device return a specfs vnode instead.
1242 */
1243 static int
1244 specvp_check(vnode_t **vpp, cred_t *cr)
1245 {
1246 int error = 0;
1247
1248 if (IS_DEVVP(*vpp)) {
1249 struct vnode *svp;
1250
1251 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1252 VN_RELE(*vpp);
1253 if (svp == NULL)
1254 error = SET_ERROR(ENOSYS);
1255 *vpp = svp;
1256 }
1257 return (error);
1258 }
1259
1260
1261 /*
1262 * Lookup an entry in a directory, or an extended attribute directory.
1263 * If it exists, return a held vnode reference for it.
1264 *
1265 * IN: dvp - vnode of directory to search.
1266 * nm - name of entry to lookup.
1267 * pnp - full pathname to lookup [UNUSED].
1268 * flags - LOOKUP_XATTR set if looking for an attribute.
1269 * rdir - root directory vnode [UNUSED].
1270 * cr - credentials of caller.
1271 * ct - caller context
1272 * direntflags - directory lookup flags
1273 * realpnp - returned pathname.
1274 *
1275 * OUT: vpp - vnode of located entry, NULL if not found.
1276 *
1277 * RETURN: 0 on success, error code on failure.
1278 *
1279 * Timestamps:
1280 * NA
1281 */
1282 /* ARGSUSED */
1283 static int
1284 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1285 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
1286 int *direntflags, pathname_t *realpnp)
1287 {
1288 znode_t *zdp = VTOZ(dvp);
1289 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1290 int error = 0;
1291
1292 /* fast path */
1293 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1294
1295 if (dvp->v_type != VDIR) {
1296 return (SET_ERROR(ENOTDIR));
1297 } else if (zdp->z_sa_hdl == NULL) {
1298 return (SET_ERROR(EIO));
1299 }
1300
1301 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1302 error = zfs_fastaccesschk_execute(zdp, cr);
1303 if (!error) {
1304 *vpp = dvp;
1305 VN_HOLD(*vpp);
1306 return (0);
1307 }
1308 return (error);
1309 } else {
1310 vnode_t *tvp = dnlc_lookup(dvp, nm);
1311
1312 if (tvp) {
1313 error = zfs_fastaccesschk_execute(zdp, cr);
1314 if (error) {
1315 VN_RELE(tvp);
1316 return (error);
1317 }
1318 if (tvp == DNLC_NO_VNODE) {
1319 VN_RELE(tvp);
1320 return (SET_ERROR(ENOENT));
1321 } else {
1322 *vpp = tvp;
1323 return (specvp_check(vpp, cr));
1324 }
1325 }
1326 }
1327 }
1328
1329 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1330
1331 ZFS_ENTER(zfsvfs);
1332 ZFS_VERIFY_ZP(zdp);
1333
1334 *vpp = NULL;
1335
1336 if (flags & LOOKUP_XATTR) {
1337 /*
1338 * If the xattr property is off, refuse the lookup request.
1339 */
1340 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1341 ZFS_EXIT(zfsvfs);
1342 return (SET_ERROR(EINVAL));
1343 }
1344
1345 /*
1346 * We don't allow recursive attributes..
1347 * Maybe someday we will.
1348 */
1349 if (zdp->z_pflags & ZFS_XATTR) {
1350 ZFS_EXIT(zfsvfs);
1351 return (SET_ERROR(EINVAL));
1352 }
1353
1354 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1355 ZFS_EXIT(zfsvfs);
1356 return (error);
1357 }
1358
1359 /*
1360 * Do we have permission to get into attribute directory?
1361 */
1362
1363 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1364 B_FALSE, cr)) {
1365 VN_RELE(*vpp);
1366 *vpp = NULL;
1367 }
1368
1369 ZFS_EXIT(zfsvfs);
1370 return (error);
1371 }
1372
1373 if (dvp->v_type != VDIR) {
1374 ZFS_EXIT(zfsvfs);
1375 return (SET_ERROR(ENOTDIR));
1376 }
1377
1378 /*
1379 * Check accessibility of directory.
1380 */
1381
1382 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1383 ZFS_EXIT(zfsvfs);
1384 return (error);
1385 }
1386
1387 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1388 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1389 ZFS_EXIT(zfsvfs);
1390 return (SET_ERROR(EILSEQ));
1391 }
1392
1393 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1394 if (error == 0)
1395 error = specvp_check(vpp, cr);
1396
1397 ZFS_EXIT(zfsvfs);
1398 return (error);
1399 }
1400
1401 /*
1402 * Attempt to create a new entry in a directory. If the entry
1403 * already exists, truncate the file if permissible, else return
1404 * an error. Return the vp of the created or trunc'd file.
1405 *
1406 * IN: dvp - vnode of directory to put new file entry in.
1407 * name - name of new file entry.
1408 * vap - attributes of new file.
1409 * excl - flag indicating exclusive or non-exclusive mode.
1410 * mode - mode to open file with.
1411 * cr - credentials of caller.
1412 * flag - large file flag [UNUSED].
1413 * ct - caller context
1414 * vsecp - ACL to be set
1415 *
1416 * OUT: vpp - vnode of created or trunc'd entry.
1417 *
1418 * RETURN: 0 on success, error code on failure.
1419 *
1420 * Timestamps:
1421 * dvp - ctime|mtime updated if new entry created
1422 * vp - ctime|mtime always, atime if new
1423 */
1424
1425 /* ARGSUSED */
1426 static int
1427 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1428 int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1429 vsecattr_t *vsecp)
1430 {
1431 znode_t *zp, *dzp = VTOZ(dvp);
1432 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1433 zilog_t *zilog;
1434 objset_t *os;
1435 zfs_dirlock_t *dl;
1436 dmu_tx_t *tx;
1437 int error;
1438 ksid_t *ksid;
1439 uid_t uid;
1440 gid_t gid = crgetgid(cr);
1441 zfs_acl_ids_t acl_ids;
1442 boolean_t fuid_dirtied;
1443 boolean_t have_acl = B_FALSE;
1444 boolean_t waited = B_FALSE;
1445
1446 /*
1447 * If we have an ephemeral id, ACL, or XVATTR then
1448 * make sure file system is at proper version
1449 */
1450
1451 ksid = crgetsid(cr, KSID_OWNER);
1452 if (ksid)
1453 uid = ksid_getid(ksid);
1454 else
1455 uid = crgetuid(cr);
1456
1457 if (zfsvfs->z_use_fuids == B_FALSE &&
1458 (vsecp || (vap->va_mask & AT_XVATTR) ||
1459 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1460 return (SET_ERROR(EINVAL));
1461
1462 ZFS_ENTER(zfsvfs);
1463 ZFS_VERIFY_ZP(dzp);
1464 os = zfsvfs->z_os;
1465 zilog = zfsvfs->z_log;
1466
1467 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1468 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1469 ZFS_EXIT(zfsvfs);
1470 return (SET_ERROR(EILSEQ));
1471 }
1472
1473 if (vap->va_mask & AT_XVATTR) {
1474 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1475 crgetuid(cr), cr, vap->va_type)) != 0) {
1476 ZFS_EXIT(zfsvfs);
1477 return (error);
1478 }
1479 }
1480 top:
1481 *vpp = NULL;
1482
1483 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1484 vap->va_mode &= ~VSVTX;
1485
1486 if (*name == '\0') {
1487 /*
1488 * Null component name refers to the directory itself.
1489 */
1490 VN_HOLD(dvp);
1491 zp = dzp;
1492 dl = NULL;
1493 error = 0;
1494 } else {
1495 /* possible VN_HOLD(zp) */
1496 int zflg = 0;
1497
1498 if (flag & FIGNORECASE)
1499 zflg |= ZCILOOK;
1500
1501 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1502 NULL, NULL);
1503 if (error) {
1504 if (have_acl)
1505 zfs_acl_ids_free(&acl_ids);
1506 if (strcmp(name, "..") == 0)
1507 error = SET_ERROR(EISDIR);
1508 ZFS_EXIT(zfsvfs);
1509 return (error);
1510 }
1511 }
1512
1513 if (zp == NULL) {
1514 uint64_t txtype;
1515
1516 /*
1517 * Create a new file object and update the directory
1518 * to reference it.
1519 */
1520 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1521 if (have_acl)
1522 zfs_acl_ids_free(&acl_ids);
1523 goto out;
1524 }
1525
1526 /*
1527 * We only support the creation of regular files in
1528 * extended attribute directories.
1529 */
1530
1531 if ((dzp->z_pflags & ZFS_XATTR) &&
1532 (vap->va_type != VREG)) {
1533 if (have_acl)
1534 zfs_acl_ids_free(&acl_ids);
1535 error = SET_ERROR(EINVAL);
1536 goto out;
1537 }
1538
1539 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1540 cr, vsecp, &acl_ids)) != 0)
1541 goto out;
1542 have_acl = B_TRUE;
1543
1544 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1545 zfs_acl_ids_free(&acl_ids);
1546 error = SET_ERROR(EDQUOT);
1547 goto out;
1548 }
1549
1550 tx = dmu_tx_create(os);
1551
1552 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1553 ZFS_SA_BASE_ATTR_SIZE);
1554
1555 fuid_dirtied = zfsvfs->z_fuid_dirty;
1556 if (fuid_dirtied)
1557 zfs_fuid_txhold(zfsvfs, tx);
1558 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1559 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1560 if (!zfsvfs->z_use_sa &&
1561 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1562 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1563 0, acl_ids.z_aclp->z_acl_bytes);
1564 }
1565 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1566 if (error) {
1567 zfs_dirent_unlock(dl);
1568 if (error == ERESTART) {
1569 waited = B_TRUE;
1570 dmu_tx_wait(tx);
1571 dmu_tx_abort(tx);
1572 goto top;
1573 }
1574 zfs_acl_ids_free(&acl_ids);
1575 dmu_tx_abort(tx);
1576 ZFS_EXIT(zfsvfs);
1577 return (error);
1578 }
1579 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1580
1581 if (fuid_dirtied)
1582 zfs_fuid_sync(zfsvfs, tx);
1583
1584 (void) zfs_link_create(dl, zp, tx, ZNEW);
1585 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1586 if (flag & FIGNORECASE)
1587 txtype |= TX_CI;
1588 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1589 vsecp, acl_ids.z_fuidp, vap);
1590 zfs_acl_ids_free(&acl_ids);
1591 dmu_tx_commit(tx);
1592 } else {
1593 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1594
1595 if (have_acl)
1596 zfs_acl_ids_free(&acl_ids);
1597 have_acl = B_FALSE;
1598
1599 /*
1600 * A directory entry already exists for this name.
1601 */
1602 /*
1603 * Can't truncate an existing file if in exclusive mode.
1604 */
1605 if (excl == EXCL) {
1606 error = SET_ERROR(EEXIST);
1607 goto out;
1608 }
1609 /*
1610 * Can't open a directory for writing.
1611 */
1612 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1613 error = SET_ERROR(EISDIR);
1614 goto out;
1615 }
1616 /*
1617 * Verify requested access to file.
1618 */
1619 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1620 goto out;
1621 }
1622
1623 mutex_enter(&dzp->z_lock);
1624 dzp->z_seq++;
1625 mutex_exit(&dzp->z_lock);
1626
1627 /*
1628 * Truncate regular files if requested.
1629 */
1630 if ((ZTOV(zp)->v_type == VREG) &&
1631 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1632 /* we can't hold any locks when calling zfs_freesp() */
1633 zfs_dirent_unlock(dl);
1634 dl = NULL;
1635 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1636 if (error == 0) {
1637 vnevent_create(ZTOV(zp), ct);
1638 }
1639 }
1640 }
1641 out:
1642
1643 if (dl)
1644 zfs_dirent_unlock(dl);
1645
1646 if (error) {
1647 if (zp)
1648 VN_RELE(ZTOV(zp));
1649 } else {
1650 *vpp = ZTOV(zp);
1651 error = specvp_check(vpp, cr);
1652 }
1653
1654 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1655 zil_commit(zilog, 0);
1656
1657 ZFS_EXIT(zfsvfs);
1658 return (error);
1659 }
1660
1661 /*
1662 * Remove an entry from a directory.
1663 *
1664 * IN: dvp - vnode of directory to remove entry from.
1665 * name - name of entry to remove.
1666 * cr - credentials of caller.
1667 * ct - caller context
1668 * flags - case flags
1669 *
1670 * RETURN: 0 on success, error code on failure.
1671 *
1672 * Timestamps:
1673 * dvp - ctime|mtime
1674 * vp - ctime (if nlink > 0)
1675 */
1676
1677 uint64_t null_xattr = 0;
1678
1679 /*ARGSUSED*/
1680 static int
1681 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1682 int flags)
1683 {
1684 znode_t *zp, *dzp = VTOZ(dvp);
1685 znode_t *xzp;
1686 vnode_t *vp;
1687 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1688 zilog_t *zilog;
1689 uint64_t acl_obj, xattr_obj;
1690 uint64_t xattr_obj_unlinked = 0;
1691 uint64_t obj = 0;
1692 zfs_dirlock_t *dl;
1693 dmu_tx_t *tx;
1694 boolean_t may_delete_now, delete_now = FALSE;
1695 boolean_t unlinked, toobig = FALSE;
1696 uint64_t txtype;
1697 pathname_t *realnmp = NULL;
1698 pathname_t realnm;
1699 int error;
1700 int zflg = ZEXISTS;
1701 boolean_t waited = B_FALSE;
1702
1703 ZFS_ENTER(zfsvfs);
1704 ZFS_VERIFY_ZP(dzp);
1705 zilog = zfsvfs->z_log;
1706
1707 if (flags & FIGNORECASE) {
1708 zflg |= ZCILOOK;
1709 pn_alloc(&realnm);
1710 realnmp = &realnm;
1711 }
1712
1713 top:
1714 xattr_obj = 0;
1715 xzp = NULL;
1716 /*
1717 * Attempt to lock directory; fail if entry doesn't exist.
1718 */
1719 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1720 NULL, realnmp)) {
1721 if (realnmp)
1722 pn_free(realnmp);
1723 ZFS_EXIT(zfsvfs);
1724 return (error);
1725 }
1726
1727 vp = ZTOV(zp);
1728
1729 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1730 goto out;
1731 }
1732
1733 /*
1734 * Need to use rmdir for removing directories.
1735 */
1736 if (vp->v_type == VDIR) {
1737 error = SET_ERROR(EPERM);
1738 goto out;
1739 }
1740
1741 vnevent_remove(vp, dvp, name, ct);
1742
1743 if (realnmp)
1744 dnlc_remove(dvp, realnmp->pn_buf);
1745 else
1746 dnlc_remove(dvp, name);
1747
1748 mutex_enter(&vp->v_lock);
1749 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1750 mutex_exit(&vp->v_lock);
1751
1752 /*
1753 * We may delete the znode now, or we may put it in the unlinked set;
1754 * it depends on whether we're the last link, and on whether there are
1755 * other holds on the vnode. So we dmu_tx_hold() the right things to
1756 * allow for either case.
1757 */
1758 obj = zp->z_id;
1759 tx = dmu_tx_create(zfsvfs->z_os);
1760 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1761 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1762 zfs_sa_upgrade_txholds(tx, zp);
1763 zfs_sa_upgrade_txholds(tx, dzp);
1764 if (may_delete_now) {
1765 toobig =
1766 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1767 /* if the file is too big, only hold_free a token amount */
1768 dmu_tx_hold_free(tx, zp->z_id, 0,
1769 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1770 }
1771
1772 /* are there any extended attributes? */
1773 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1774 &xattr_obj, sizeof (xattr_obj));
1775 if (error == 0 && xattr_obj) {
1776 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1777 ASSERT0(error);
1778 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1779 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1780 }
1781
1782 mutex_enter(&zp->z_lock);
1783 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1784 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1785 mutex_exit(&zp->z_lock);
1786
1787 /* charge as an update -- would be nice not to charge at all */
1788 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1789
1790 /*
1791 * Mark this transaction as typically resulting in a net free of
1792 * space, unless object removal will be delayed indefinitely
1793 * (due to active holds on the vnode due to the file being open).
1794 */
1795 if (may_delete_now)
1796 dmu_tx_mark_netfree(tx);
1797
1798 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1799 if (error) {
1800 zfs_dirent_unlock(dl);
1801 VN_RELE(vp);
1802 if (xzp)
1803 VN_RELE(ZTOV(xzp));
1804 if (error == ERESTART) {
1805 waited = B_TRUE;
1806 dmu_tx_wait(tx);
1807 dmu_tx_abort(tx);
1808 goto top;
1809 }
1810 if (realnmp)
1811 pn_free(realnmp);
1812 dmu_tx_abort(tx);
1813 ZFS_EXIT(zfsvfs);
1814 return (error);
1815 }
1816
1817 /*
1818 * Remove the directory entry.
1819 */
1820 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1821
1822 if (error) {
1823 dmu_tx_commit(tx);
1824 goto out;
1825 }
1826
1827 if (unlinked) {
1828 /*
1829 * Hold z_lock so that we can make sure that the ACL obj
1830 * hasn't changed. Could have been deleted due to
1831 * zfs_sa_upgrade().
1832 */
1833 mutex_enter(&zp->z_lock);
1834 mutex_enter(&vp->v_lock);
1835 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1836 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1837 delete_now = may_delete_now && !toobig &&
1838 vp->v_count == 1 && !vn_has_cached_data(vp) &&
1839 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1840 acl_obj;
1841 mutex_exit(&vp->v_lock);
1842 }
1843
1844 if (delete_now) {
1845 if (xattr_obj_unlinked) {
1846 ASSERT3U(xzp->z_links, ==, 2);
1847 mutex_enter(&xzp->z_lock);
1848 xzp->z_unlinked = 1;
1849 xzp->z_links = 0;
1850 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1851 &xzp->z_links, sizeof (xzp->z_links), tx);
1852 ASSERT3U(error, ==, 0);
1853 mutex_exit(&xzp->z_lock);
1854 zfs_unlinked_add(xzp, tx);
1855
1856 if (zp->z_is_sa)
1857 error = sa_remove(zp->z_sa_hdl,
1858 SA_ZPL_XATTR(zfsvfs), tx);
1859 else
1860 error = sa_update(zp->z_sa_hdl,
1861 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1862 sizeof (uint64_t), tx);
1863 ASSERT0(error);
1864 }
1865 mutex_enter(&vp->v_lock);
1866 vp->v_count--;
1867 ASSERT0(vp->v_count);
1868 mutex_exit(&vp->v_lock);
1869 mutex_exit(&zp->z_lock);
1870 zfs_znode_delete(zp, tx);
1871 } else if (unlinked) {
1872 mutex_exit(&zp->z_lock);
1873 zfs_unlinked_add(zp, tx);
1874 }
1875
1876 txtype = TX_REMOVE;
1877 if (flags & FIGNORECASE)
1878 txtype |= TX_CI;
1879 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1880
1881 dmu_tx_commit(tx);
1882 out:
1883 if (realnmp)
1884 pn_free(realnmp);
1885
1886 zfs_dirent_unlock(dl);
1887
1888 if (!delete_now)
1889 VN_RELE(vp);
1890 if (xzp)
1891 VN_RELE(ZTOV(xzp));
1892
1893 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1894 zil_commit(zilog, 0);
1895
1896 ZFS_EXIT(zfsvfs);
1897 return (error);
1898 }
1899
1900 /*
1901 * Create a new directory and insert it into dvp using the name
1902 * provided. Return a pointer to the inserted directory.
1903 *
1904 * IN: dvp - vnode of directory to add subdir to.
1905 * dirname - name of new directory.
1906 * vap - attributes of new directory.
1907 * cr - credentials of caller.
1908 * ct - caller context
1909 * flags - case flags
1910 * vsecp - ACL to be set
1911 *
1912 * OUT: vpp - vnode of created directory.
1913 *
1914 * RETURN: 0 on success, error code on failure.
1915 *
1916 * Timestamps:
1917 * dvp - ctime|mtime updated
1918 * vp - ctime|mtime|atime updated
1919 */
1920 /*ARGSUSED*/
1921 static int
1922 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1923 caller_context_t *ct, int flags, vsecattr_t *vsecp)
1924 {
1925 znode_t *zp, *dzp = VTOZ(dvp);
1926 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1927 zilog_t *zilog;
1928 zfs_dirlock_t *dl;
1929 uint64_t txtype;
1930 dmu_tx_t *tx;
1931 int error;
1932 int zf = ZNEW;
1933 ksid_t *ksid;
1934 uid_t uid;
1935 gid_t gid = crgetgid(cr);
1936 zfs_acl_ids_t acl_ids;
1937 boolean_t fuid_dirtied;
1938 boolean_t waited = B_FALSE;
1939
1940 ASSERT(vap->va_type == VDIR);
1941
1942 /*
1943 * If we have an ephemeral id, ACL, or XVATTR then
1944 * make sure file system is at proper version
1945 */
1946
1947 ksid = crgetsid(cr, KSID_OWNER);
1948 if (ksid)
1949 uid = ksid_getid(ksid);
1950 else
1951 uid = crgetuid(cr);
1952 if (zfsvfs->z_use_fuids == B_FALSE &&
1953 (vsecp || (vap->va_mask & AT_XVATTR) ||
1954 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1955 return (SET_ERROR(EINVAL));
1956
1957 ZFS_ENTER(zfsvfs);
1958 ZFS_VERIFY_ZP(dzp);
1959 zilog = zfsvfs->z_log;
1960
1961 if (dzp->z_pflags & ZFS_XATTR) {
1962 ZFS_EXIT(zfsvfs);
1963 return (SET_ERROR(EINVAL));
1964 }
1965
1966 if (zfsvfs->z_utf8 && u8_validate(dirname,
1967 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1968 ZFS_EXIT(zfsvfs);
1969 return (SET_ERROR(EILSEQ));
1970 }
1971 if (flags & FIGNORECASE)
1972 zf |= ZCILOOK;
1973
1974 if (vap->va_mask & AT_XVATTR) {
1975 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1976 crgetuid(cr), cr, vap->va_type)) != 0) {
1977 ZFS_EXIT(zfsvfs);
1978 return (error);
1979 }
1980 }
1981
1982 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1983 vsecp, &acl_ids)) != 0) {
1984 ZFS_EXIT(zfsvfs);
1985 return (error);
1986 }
1987 /*
1988 * First make sure the new directory doesn't exist.
1989 *
1990 * Existence is checked first to make sure we don't return
1991 * EACCES instead of EEXIST which can cause some applications
1992 * to fail.
1993 */
1994 top:
1995 *vpp = NULL;
1996
1997 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1998 NULL, NULL)) {
1999 zfs_acl_ids_free(&acl_ids);
2000 ZFS_EXIT(zfsvfs);
2001 return (error);
2002 }
2003
2004 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
2005 zfs_acl_ids_free(&acl_ids);
2006 zfs_dirent_unlock(dl);
2007 ZFS_EXIT(zfsvfs);
2008 return (error);
2009 }
2010
2011 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
2012 zfs_acl_ids_free(&acl_ids);
2013 zfs_dirent_unlock(dl);
2014 ZFS_EXIT(zfsvfs);
2015 return (SET_ERROR(EDQUOT));
2016 }
2017
2018 /*
2019 * Add a new entry to the directory.
2020 */
2021 tx = dmu_tx_create(zfsvfs->z_os);
2022 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
2023 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2024 fuid_dirtied = zfsvfs->z_fuid_dirty;
2025 if (fuid_dirtied)
2026 zfs_fuid_txhold(zfsvfs, tx);
2027 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2028 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2029 acl_ids.z_aclp->z_acl_bytes);
2030 }
2031
2032 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
2033 ZFS_SA_BASE_ATTR_SIZE);
2034
2035 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2036 if (error) {
2037 zfs_dirent_unlock(dl);
2038 if (error == ERESTART) {
2039 waited = B_TRUE;
2040 dmu_tx_wait(tx);
2041 dmu_tx_abort(tx);
2042 goto top;
2043 }
2044 zfs_acl_ids_free(&acl_ids);
2045 dmu_tx_abort(tx);
2046 ZFS_EXIT(zfsvfs);
2047 return (error);
2048 }
2049
2050 /*
2051 * Create new node.
2052 */
2053 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2054
2055 if (fuid_dirtied)
2056 zfs_fuid_sync(zfsvfs, tx);
2057
2058 /*
2059 * Now put new name in parent dir.
2060 */
2061 (void) zfs_link_create(dl, zp, tx, ZNEW);
2062
2063 *vpp = ZTOV(zp);
2064
2065 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2066 if (flags & FIGNORECASE)
2067 txtype |= TX_CI;
2068 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2069 acl_ids.z_fuidp, vap);
2070
2071 zfs_acl_ids_free(&acl_ids);
2072
2073 dmu_tx_commit(tx);
2074
2075 zfs_dirent_unlock(dl);
2076
2077 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2078 zil_commit(zilog, 0);
2079
2080 ZFS_EXIT(zfsvfs);
2081 return (0);
2082 }
2083
2084 /*
2085 * Remove a directory subdir entry. If the current working
2086 * directory is the same as the subdir to be removed, the
2087 * remove will fail.
2088 *
2089 * IN: dvp - vnode of directory to remove from.
2090 * name - name of directory to be removed.
2091 * cwd - vnode of current working directory.
2092 * cr - credentials of caller.
2093 * ct - caller context
2094 * flags - case flags
2095 *
2096 * RETURN: 0 on success, error code on failure.
2097 *
2098 * Timestamps:
2099 * dvp - ctime|mtime updated
2100 */
2101 /*ARGSUSED*/
2102 static int
2103 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2104 caller_context_t *ct, int flags)
2105 {
2106 znode_t *dzp = VTOZ(dvp);
2107 znode_t *zp;
2108 vnode_t *vp;
2109 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2110 zilog_t *zilog;
2111 zfs_dirlock_t *dl;
2112 dmu_tx_t *tx;
2113 int error;
2114 int zflg = ZEXISTS;
2115 boolean_t waited = B_FALSE;
2116
2117 ZFS_ENTER(zfsvfs);
2118 ZFS_VERIFY_ZP(dzp);
2119 zilog = zfsvfs->z_log;
2120
2121 if (flags & FIGNORECASE)
2122 zflg |= ZCILOOK;
2123 top:
2124 zp = NULL;
2125
2126 /*
2127 * Attempt to lock directory; fail if entry doesn't exist.
2128 */
2129 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2130 NULL, NULL)) {
2131 ZFS_EXIT(zfsvfs);
2132 return (error);
2133 }
2134
2135 vp = ZTOV(zp);
2136
2137 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2138 goto out;
2139 }
2140
2141 if (vp->v_type != VDIR) {
2142 error = SET_ERROR(ENOTDIR);
2143 goto out;
2144 }
2145
2146 if (vp == cwd) {
2147 error = SET_ERROR(EINVAL);
2148 goto out;
2149 }
2150
2151 vnevent_rmdir(vp, dvp, name, ct);
2152
2153 /*
2154 * Grab a lock on the directory to make sure that noone is
2155 * trying to add (or lookup) entries while we are removing it.
2156 */
2157 rw_enter(&zp->z_name_lock, RW_WRITER);
2158
2159 /*
2160 * Grab a lock on the parent pointer to make sure we play well
2161 * with the treewalk and directory rename code.
2162 */
2163 rw_enter(&zp->z_parent_lock, RW_WRITER);
2164
2165 tx = dmu_tx_create(zfsvfs->z_os);
2166 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2167 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2168 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2169 zfs_sa_upgrade_txholds(tx, zp);
2170 zfs_sa_upgrade_txholds(tx, dzp);
2171 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2172 if (error) {
2173 rw_exit(&zp->z_parent_lock);
2174 rw_exit(&zp->z_name_lock);
2175 zfs_dirent_unlock(dl);
2176 VN_RELE(vp);
2177 if (error == ERESTART) {
2178 waited = B_TRUE;
2179 dmu_tx_wait(tx);
2180 dmu_tx_abort(tx);
2181 goto top;
2182 }
2183 dmu_tx_abort(tx);
2184 ZFS_EXIT(zfsvfs);
2185 return (error);
2186 }
2187
2188 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2189
2190 if (error == 0) {
2191 uint64_t txtype = TX_RMDIR;
2192 if (flags & FIGNORECASE)
2193 txtype |= TX_CI;
2194 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2195 }
2196
2197 dmu_tx_commit(tx);
2198
2199 rw_exit(&zp->z_parent_lock);
2200 rw_exit(&zp->z_name_lock);
2201 out:
2202 zfs_dirent_unlock(dl);
2203
2204 VN_RELE(vp);
2205
2206 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2207 zil_commit(zilog, 0);
2208
2209 ZFS_EXIT(zfsvfs);
2210 return (error);
2211 }
2212
2213 /*
2214 * Read as many directory entries as will fit into the provided
2215 * buffer from the given directory cursor position (specified in
2216 * the uio structure).
2217 *
2218 * IN: vp - vnode of directory to read.
2219 * uio - structure supplying read location, range info,
2220 * and return buffer.
2221 * cr - credentials of caller.
2222 * ct - caller context
2223 * flags - case flags
2224 *
2225 * OUT: uio - updated offset and range, buffer filled.
2226 * eofp - set to true if end-of-file detected.
2227 *
2228 * RETURN: 0 on success, error code on failure.
2229 *
2230 * Timestamps:
2231 * vp - atime updated
2232 *
2233 * Note that the low 4 bits of the cookie returned by zap is always zero.
2234 * This allows us to use the low range for "special" directory entries:
2235 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2236 * we use the offset 2 for the '.zfs' directory.
2237 */
2238 /* ARGSUSED */
2239 static int
2240 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2241 caller_context_t *ct, int flags)
2242 {
2243 znode_t *zp = VTOZ(vp);
2244 iovec_t *iovp;
2245 edirent_t *eodp;
2246 dirent64_t *odp;
2247 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2248 objset_t *os;
2249 caddr_t outbuf;
2250 size_t bufsize;
2251 zap_cursor_t zc;
2252 zap_attribute_t zap;
2253 uint_t bytes_wanted;
2254 uint64_t offset; /* must be unsigned; checks for < 1 */
2255 uint64_t parent;
2256 int local_eof;
2257 int outcount;
2258 int error;
2259 uint8_t prefetch;
2260 boolean_t check_sysattrs;
2261
2262 ZFS_ENTER(zfsvfs);
2263 ZFS_VERIFY_ZP(zp);
2264
2265 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2266 &parent, sizeof (parent))) != 0) {
2267 ZFS_EXIT(zfsvfs);
2268 return (error);
2269 }
2270
2271 /*
2272 * If we are not given an eof variable,
2273 * use a local one.
2274 */
2275 if (eofp == NULL)
2276 eofp = &local_eof;
2277
2278 /*
2279 * Check for valid iov_len.
2280 */
2281 if (uio->uio_iov->iov_len <= 0) {
2282 ZFS_EXIT(zfsvfs);
2283 return (SET_ERROR(EINVAL));
2284 }
2285
2286 /*
2287 * Quit if directory has been removed (posix)
2288 */
2289 if ((*eofp = zp->z_unlinked) != 0) {
2290 ZFS_EXIT(zfsvfs);
2291 return (0);
2292 }
2293
2294 error = 0;
2295 os = zfsvfs->z_os;
2296 offset = uio->uio_loffset;
2297 prefetch = zp->z_zn_prefetch;
2298
2299 /*
2300 * Initialize the iterator cursor.
2301 */
2302 if (offset <= 3) {
2303 /*
2304 * Start iteration from the beginning of the directory.
2305 */
2306 zap_cursor_init(&zc, os, zp->z_id);
2307 } else {
2308 /*
2309 * The offset is a serialized cursor.
2310 */
2311 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2312 }
2313
2314 /*
2315 * Get space to change directory entries into fs independent format.
2316 */
2317 iovp = uio->uio_iov;
2318 bytes_wanted = iovp->iov_len;
2319 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2320 bufsize = bytes_wanted;
2321 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2322 odp = (struct dirent64 *)outbuf;
2323 } else {
2324 bufsize = bytes_wanted;
2325 outbuf = NULL;
2326 odp = (struct dirent64 *)iovp->iov_base;
2327 }
2328 eodp = (struct edirent *)odp;
2329
2330 /*
2331 * If this VFS supports the system attribute view interface; and
2332 * we're looking at an extended attribute directory; and we care
2333 * about normalization conflicts on this vfs; then we must check
2334 * for normalization conflicts with the sysattr name space.
2335 */
2336 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2337 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2338 (flags & V_RDDIR_ENTFLAGS);
2339
2340 /*
2341 * Transform to file-system independent format
2342 */
2343 outcount = 0;
2344 while (outcount < bytes_wanted) {
2345 ino64_t objnum;
2346 ushort_t reclen;
2347 off64_t *next = NULL;
2348
2349 /*
2350 * Special case `.', `..', and `.zfs'.
2351 */
2352 if (offset == 0) {
2353 (void) strcpy(zap.za_name, ".");
2354 zap.za_normalization_conflict = 0;
2355 objnum = zp->z_id;
2356 } else if (offset == 1) {
2357 (void) strcpy(zap.za_name, "..");
2358 zap.za_normalization_conflict = 0;
2359 objnum = parent;
2360 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2361 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2362 zap.za_normalization_conflict = 0;
2363 objnum = ZFSCTL_INO_ROOT;
2364 } else {
2365 /*
2366 * Grab next entry.
2367 */
2368 if (error = zap_cursor_retrieve(&zc, &zap)) {
2369 if ((*eofp = (error == ENOENT)) != 0)
2370 break;
2371 else
2372 goto update;
2373 }
2374
2375 if (zap.za_integer_length != 8 ||
2376 zap.za_num_integers != 1) {
2377 cmn_err(CE_WARN, "zap_readdir: bad directory "
2378 "entry, obj = %lld, offset = %lld\n",
2379 (u_longlong_t)zp->z_id,
2380 (u_longlong_t)offset);
2381 error = SET_ERROR(ENXIO);
2382 goto update;
2383 }
2384
2385 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2386 /*
2387 * MacOS X can extract the object type here such as:
2388 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2389 */
2390
2391 if (check_sysattrs && !zap.za_normalization_conflict) {
2392 zap.za_normalization_conflict =
2393 xattr_sysattr_casechk(zap.za_name);
2394 }
2395 }
2396
2397 if (flags & V_RDDIR_ACCFILTER) {
2398 /*
2399 * If we have no access at all, don't include
2400 * this entry in the returned information
2401 */
2402 znode_t *ezp;
2403 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2404 goto skip_entry;
2405 if (!zfs_has_access(ezp, cr)) {
2406 VN_RELE(ZTOV(ezp));
2407 goto skip_entry;
2408 }
2409 VN_RELE(ZTOV(ezp));
2410 }
2411
2412 if (flags & V_RDDIR_ENTFLAGS)
2413 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2414 else
2415 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2416
2417 /*
2418 * Will this entry fit in the buffer?
2419 */
2420 if (outcount + reclen > bufsize) {
2421 /*
2422 * Did we manage to fit anything in the buffer?
2423 */
2424 if (!outcount) {
2425 error = SET_ERROR(EINVAL);
2426 goto update;
2427 }
2428 break;
2429 }
2430 if (flags & V_RDDIR_ENTFLAGS) {
2431 /*
2432 * Add extended flag entry:
2433 */
2434 eodp->ed_ino = objnum;
2435 eodp->ed_reclen = reclen;
2436 /* NOTE: ed_off is the offset for the *next* entry */
2437 next = &(eodp->ed_off);
2438 eodp->ed_eflags = zap.za_normalization_conflict ?
2439 ED_CASE_CONFLICT : 0;
2440 (void) strncpy(eodp->ed_name, zap.za_name,
2441 EDIRENT_NAMELEN(reclen));
2442 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2443 } else {
2444 /*
2445 * Add normal entry:
2446 */
2447 odp->d_ino = objnum;
2448 odp->d_reclen = reclen;
2449 /* NOTE: d_off is the offset for the *next* entry */
2450 next = &(odp->d_off);
2451 (void) strncpy(odp->d_name, zap.za_name,
2452 DIRENT64_NAMELEN(reclen));
2453 odp = (dirent64_t *)((intptr_t)odp + reclen);
2454 }
2455 outcount += reclen;
2456
2457 ASSERT(outcount <= bufsize);
2458
2459 /* Prefetch znode */
2460 if (prefetch)
2461 dmu_prefetch(os, objnum, 0, 0);
2462
2463 skip_entry:
2464 /*
2465 * Move to the next entry, fill in the previous offset.
2466 */
2467 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2468 zap_cursor_advance(&zc);
2469 offset = zap_cursor_serialize(&zc);
2470 } else {
2471 offset += 1;
2472 }
2473 if (next)
2474 *next = offset;
2475 }
2476 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2477
2478 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2479 iovp->iov_base += outcount;
2480 iovp->iov_len -= outcount;
2481 uio->uio_resid -= outcount;
2482 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2483 /*
2484 * Reset the pointer.
2485 */
2486 offset = uio->uio_loffset;
2487 }
2488
2489 update:
2490 zap_cursor_fini(&zc);
2491 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2492 kmem_free(outbuf, bufsize);
2493
2494 if (error == ENOENT)
2495 error = 0;
2496
2497 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2498
2499 uio->uio_loffset = offset;
2500 ZFS_EXIT(zfsvfs);
2501 return (error);
2502 }
2503
2504 ulong_t zfs_fsync_sync_cnt = 4;
2505
2506 static int
2507 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2508 {
2509 znode_t *zp = VTOZ(vp);
2510 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2511
2512 /*
2513 * Regardless of whether this is required for standards conformance,
2514 * this is the logical behavior when fsync() is called on a file with
2515 * dirty pages. We use B_ASYNC since the ZIL transactions are already
2516 * going to be pushed out as part of the zil_commit().
2517 */
2518 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2519 (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2520 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2521
2522 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2523
2524 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2525 ZFS_ENTER(zfsvfs);
2526 ZFS_VERIFY_ZP(zp);
2527 zil_commit(zfsvfs->z_log, zp->z_id);
2528 ZFS_EXIT(zfsvfs);
2529 }
2530 return (0);
2531 }
2532
2533
2534 /*
2535 * Get the requested file attributes and place them in the provided
2536 * vattr structure.
2537 *
2538 * IN: vp - vnode of file.
2539 * vap - va_mask identifies requested attributes.
2540 * If AT_XVATTR set, then optional attrs are requested
2541 * flags - ATTR_NOACLCHECK (CIFS server context)
2542 * cr - credentials of caller.
2543 * ct - caller context
2544 *
2545 * OUT: vap - attribute values.
2546 *
2547 * RETURN: 0 (always succeeds).
2548 */
2549 /* ARGSUSED */
2550 static int
2551 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2552 caller_context_t *ct)
2553 {
2554 znode_t *zp = VTOZ(vp);
2555 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2556 int error = 0;
2557 uint64_t links;
2558 uint64_t mtime[2], ctime[2];
2559 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2560 xoptattr_t *xoap = NULL;
2561 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2562 sa_bulk_attr_t bulk[2];
2563 int count = 0;
2564
2565 ZFS_ENTER(zfsvfs);
2566 ZFS_VERIFY_ZP(zp);
2567
2568 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2569
2570 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2571 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2572
2573 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2574 ZFS_EXIT(zfsvfs);
2575 return (error);
2576 }
2577
2578 /*
2579 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2580 * Also, if we are the owner don't bother, since owner should
2581 * always be allowed to read basic attributes of file.
2582 */
2583 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2584 (vap->va_uid != crgetuid(cr))) {
2585 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2586 skipaclchk, cr)) {
2587 ZFS_EXIT(zfsvfs);
2588 return (error);
2589 }
2590 }
2591
2592 /*
2593 * Return all attributes. It's cheaper to provide the answer
2594 * than to determine whether we were asked the question.
2595 */
2596
2597 mutex_enter(&zp->z_lock);
2598 vap->va_type = vp->v_type;
2599 vap->va_mode = zp->z_mode & MODEMASK;
2600 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2601 vap->va_nodeid = zp->z_id;
2602 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2603 links = zp->z_links + 1;
2604 else
2605 links = zp->z_links;
2606 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */
2607 vap->va_size = zp->z_size;
2608 vap->va_rdev = vp->v_rdev;
2609 vap->va_seq = zp->z_seq;
2610
2611 /*
2612 * Add in any requested optional attributes and the create time.
2613 * Also set the corresponding bits in the returned attribute bitmap.
2614 */
2615 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2616 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2617 xoap->xoa_archive =
2618 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2619 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2620 }
2621
2622 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2623 xoap->xoa_readonly =
2624 ((zp->z_pflags & ZFS_READONLY) != 0);
2625 XVA_SET_RTN(xvap, XAT_READONLY);
2626 }
2627
2628 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2629 xoap->xoa_system =
2630 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2631 XVA_SET_RTN(xvap, XAT_SYSTEM);
2632 }
2633
2634 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2635 xoap->xoa_hidden =
2636 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2637 XVA_SET_RTN(xvap, XAT_HIDDEN);
2638 }
2639
2640 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2641 xoap->xoa_nounlink =
2642 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2643 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2644 }
2645
2646 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2647 xoap->xoa_immutable =
2648 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2649 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2650 }
2651
2652 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2653 xoap->xoa_appendonly =
2654 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2655 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2656 }
2657
2658 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2659 xoap->xoa_nodump =
2660 ((zp->z_pflags & ZFS_NODUMP) != 0);
2661 XVA_SET_RTN(xvap, XAT_NODUMP);
2662 }
2663
2664 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2665 xoap->xoa_opaque =
2666 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2667 XVA_SET_RTN(xvap, XAT_OPAQUE);
2668 }
2669
2670 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2671 xoap->xoa_av_quarantined =
2672 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2673 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2674 }
2675
2676 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2677 xoap->xoa_av_modified =
2678 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2679 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2680 }
2681
2682 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2683 vp->v_type == VREG) {
2684 zfs_sa_get_scanstamp(zp, xvap);
2685 }
2686
2687 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2688 uint64_t times[2];
2689
2690 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2691 times, sizeof (times));
2692 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2693 XVA_SET_RTN(xvap, XAT_CREATETIME);
2694 }
2695
2696 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2697 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2698 XVA_SET_RTN(xvap, XAT_REPARSE);
2699 }
2700 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2701 xoap->xoa_generation = zp->z_gen;
2702 XVA_SET_RTN(xvap, XAT_GEN);
2703 }
2704
2705 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2706 xoap->xoa_offline =
2707 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2708 XVA_SET_RTN(xvap, XAT_OFFLINE);
2709 }
2710
2711 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2712 xoap->xoa_sparse =
2713 ((zp->z_pflags & ZFS_SPARSE) != 0);
2714 XVA_SET_RTN(xvap, XAT_SPARSE);
2715 }
2716 }
2717
2718 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2719 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2720 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2721
2722 mutex_exit(&zp->z_lock);
2723
2724 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2725
2726 if (zp->z_blksz == 0) {
2727 /*
2728 * Block size hasn't been set; suggest maximal I/O transfers.
2729 */
2730 vap->va_blksize = zfsvfs->z_max_blksz;
2731 }
2732
2733 ZFS_EXIT(zfsvfs);
2734 return (0);
2735 }
2736
2737 /*
2738 * Set the file attributes to the values contained in the
2739 * vattr structure.
2740 *
2741 * IN: vp - vnode of file to be modified.
2742 * vap - new attribute values.
2743 * If AT_XVATTR set, then optional attrs are being set
2744 * flags - ATTR_UTIME set if non-default time values provided.
2745 * - ATTR_NOACLCHECK (CIFS context only).
2746 * cr - credentials of caller.
2747 * ct - caller context
2748 *
2749 * RETURN: 0 on success, error code on failure.
2750 *
2751 * Timestamps:
2752 * vp - ctime updated, mtime updated if size changed.
2753 */
2754 /* ARGSUSED */
2755 static int
2756 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2757 caller_context_t *ct)
2758 {
2759 znode_t *zp = VTOZ(vp);
2760 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2761 zilog_t *zilog;
2762 dmu_tx_t *tx;
2763 vattr_t oldva;
2764 xvattr_t tmpxvattr;
2765 uint_t mask = vap->va_mask;
2766 uint_t saved_mask = 0;
2767 int trim_mask = 0;
2768 uint64_t new_mode;
2769 uint64_t new_uid, new_gid;
2770 uint64_t xattr_obj;
2771 uint64_t mtime[2], ctime[2];
2772 znode_t *attrzp;
2773 int need_policy = FALSE;
2774 int err, err2;
2775 zfs_fuid_info_t *fuidp = NULL;
2776 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2777 xoptattr_t *xoap;
2778 zfs_acl_t *aclp;
2779 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2780 boolean_t fuid_dirtied = B_FALSE;
2781 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2782 int count = 0, xattr_count = 0;
2783
2784 if (mask == 0)
2785 return (0);
2786
2787 if (mask & AT_NOSET)
2788 return (SET_ERROR(EINVAL));
2789
2790 ZFS_ENTER(zfsvfs);
2791 ZFS_VERIFY_ZP(zp);
2792
2793 zilog = zfsvfs->z_log;
2794
2795 /*
2796 * Make sure that if we have ephemeral uid/gid or xvattr specified
2797 * that file system is at proper version level
2798 */
2799
2800 if (zfsvfs->z_use_fuids == B_FALSE &&
2801 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2802 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2803 (mask & AT_XVATTR))) {
2804 ZFS_EXIT(zfsvfs);
2805 return (SET_ERROR(EINVAL));
2806 }
2807
2808 if (mask & AT_SIZE && vp->v_type == VDIR) {
2809 ZFS_EXIT(zfsvfs);
2810 return (SET_ERROR(EISDIR));
2811 }
2812
2813 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2814 ZFS_EXIT(zfsvfs);
2815 return (SET_ERROR(EINVAL));
2816 }
2817
2818 /*
2819 * If this is an xvattr_t, then get a pointer to the structure of
2820 * optional attributes. If this is NULL, then we have a vattr_t.
2821 */
2822 xoap = xva_getxoptattr(xvap);
2823
2824 xva_init(&tmpxvattr);
2825
2826 /*
2827 * Immutable files can only alter immutable bit and atime
2828 */
2829 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2830 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2831 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2832 ZFS_EXIT(zfsvfs);
2833 return (SET_ERROR(EPERM));
2834 }
2835
2836 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2837 ZFS_EXIT(zfsvfs);
2838 return (SET_ERROR(EPERM));
2839 }
2840
2841 /*
2842 * Verify timestamps doesn't overflow 32 bits.
2843 * ZFS can handle large timestamps, but 32bit syscalls can't
2844 * handle times greater than 2039. This check should be removed
2845 * once large timestamps are fully supported.
2846 */
2847 if (mask & (AT_ATIME | AT_MTIME)) {
2848 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2849 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2850 ZFS_EXIT(zfsvfs);
2851 return (SET_ERROR(EOVERFLOW));
2852 }
2853 }
2854
2855 top:
2856 attrzp = NULL;
2857 aclp = NULL;
2858
2859 /* Can this be moved to before the top label? */
2860 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2861 ZFS_EXIT(zfsvfs);
2862 return (SET_ERROR(EROFS));
2863 }
2864
2865 /*
2866 * First validate permissions
2867 */
2868
2869 if (mask & AT_SIZE) {
2870 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2871 if (err) {
2872 ZFS_EXIT(zfsvfs);
2873 return (err);
2874 }
2875 /*
2876 * XXX - Note, we are not providing any open
2877 * mode flags here (like FNDELAY), so we may
2878 * block if there are locks present... this
2879 * should be addressed in openat().
2880 */
2881 /* XXX - would it be OK to generate a log record here? */
2882 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2883 if (err) {
2884 ZFS_EXIT(zfsvfs);
2885 return (err);
2886 }
2887
2888 if (vap->va_size == 0)
2889 vnevent_truncate(ZTOV(zp), ct);
2890 }
2891
2892 if (mask & (AT_ATIME|AT_MTIME) ||
2893 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2894 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2895 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2896 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2897 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2898 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2899 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2900 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2901 skipaclchk, cr);
2902 }
2903
2904 if (mask & (AT_UID|AT_GID)) {
2905 int idmask = (mask & (AT_UID|AT_GID));
2906 int take_owner;
2907 int take_group;
2908
2909 /*
2910 * NOTE: even if a new mode is being set,
2911 * we may clear S_ISUID/S_ISGID bits.
2912 */
2913
2914 if (!(mask & AT_MODE))
2915 vap->va_mode = zp->z_mode;
2916
2917 /*
2918 * Take ownership or chgrp to group we are a member of
2919 */
2920
2921 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2922 take_group = (mask & AT_GID) &&
2923 zfs_groupmember(zfsvfs, vap->va_gid, cr);
2924
2925 /*
2926 * If both AT_UID and AT_GID are set then take_owner and
2927 * take_group must both be set in order to allow taking
2928 * ownership.
2929 *
2930 * Otherwise, send the check through secpolicy_vnode_setattr()
2931 *
2932 */
2933
2934 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2935 ((idmask == AT_UID) && take_owner) ||
2936 ((idmask == AT_GID) && take_group)) {
2937 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2938 skipaclchk, cr) == 0) {
2939 /*
2940 * Remove setuid/setgid for non-privileged users
2941 */
2942 secpolicy_setid_clear(vap, cr);
2943 trim_mask = (mask & (AT_UID|AT_GID));
2944 } else {
2945 need_policy = TRUE;
2946 }
2947 } else {
2948 need_policy = TRUE;
2949 }
2950 }
2951
2952 mutex_enter(&zp->z_lock);
2953 oldva.va_mode = zp->z_mode;
2954 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2955 if (mask & AT_XVATTR) {
2956 /*
2957 * Update xvattr mask to include only those attributes
2958 * that are actually changing.
2959 *
2960 * the bits will be restored prior to actually setting
2961 * the attributes so the caller thinks they were set.
2962 */
2963 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2964 if (xoap->xoa_appendonly !=
2965 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2966 need_policy = TRUE;
2967 } else {
2968 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2969 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2970 }
2971 }
2972
2973 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2974 if (xoap->xoa_nounlink !=
2975 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2976 need_policy = TRUE;
2977 } else {
2978 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2979 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2980 }
2981 }
2982
2983 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2984 if (xoap->xoa_immutable !=
2985 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2986 need_policy = TRUE;
2987 } else {
2988 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2989 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2990 }
2991 }
2992
2993 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2994 if (xoap->xoa_nodump !=
2995 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2996 need_policy = TRUE;
2997 } else {
2998 XVA_CLR_REQ(xvap, XAT_NODUMP);
2999 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
3000 }
3001 }
3002
3003 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
3004 if (xoap->xoa_av_modified !=
3005 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
3006 need_policy = TRUE;
3007 } else {
3008 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
3009 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
3010 }
3011 }
3012
3013 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
3014 if ((vp->v_type != VREG &&
3015 xoap->xoa_av_quarantined) ||
3016 xoap->xoa_av_quarantined !=
3017 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
3018 need_policy = TRUE;
3019 } else {
3020 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
3021 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
3022 }
3023 }
3024
3025 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
3026 mutex_exit(&zp->z_lock);
3027 ZFS_EXIT(zfsvfs);
3028 return (SET_ERROR(EPERM));
3029 }
3030
3031 if (need_policy == FALSE &&
3032 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
3033 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
3034 need_policy = TRUE;
3035 }
3036 }
3037
3038 mutex_exit(&zp->z_lock);
3039
3040 if (mask & AT_MODE) {
3041 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3042 err = secpolicy_setid_setsticky_clear(vp, vap,
3043 &oldva, cr);
3044 if (err) {
3045 ZFS_EXIT(zfsvfs);
3046 return (err);
3047 }
3048 trim_mask |= AT_MODE;
3049 } else {
3050 need_policy = TRUE;
3051 }
3052 }
3053
3054 if (need_policy) {
3055 /*
3056 * If trim_mask is set then take ownership
3057 * has been granted or write_acl is present and user
3058 * has the ability to modify mode. In that case remove
3059 * UID|GID and or MODE from mask so that
3060 * secpolicy_vnode_setattr() doesn't revoke it.
3061 */
3062
3063 if (trim_mask) {
3064 saved_mask = vap->va_mask;
3065 vap->va_mask &= ~trim_mask;
3066 }
3067 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3068 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3069 if (err) {
3070 ZFS_EXIT(zfsvfs);
3071 return (err);
3072 }
3073
3074 if (trim_mask)
3075 vap->va_mask |= saved_mask;
3076 }
3077
3078 /*
3079 * secpolicy_vnode_setattr, or take ownership may have
3080 * changed va_mask
3081 */
3082 mask = vap->va_mask;
3083
3084 if ((mask & (AT_UID | AT_GID))) {
3085 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3086 &xattr_obj, sizeof (xattr_obj));
3087
3088 if (err == 0 && xattr_obj) {
3089 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3090 if (err)
3091 goto out2;
3092 }
3093 if (mask & AT_UID) {
3094 new_uid = zfs_fuid_create(zfsvfs,
3095 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3096 if (new_uid != zp->z_uid &&
3097 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3098 if (attrzp)
3099 VN_RELE(ZTOV(attrzp));
3100 err = SET_ERROR(EDQUOT);
3101 goto out2;
3102 }
3103 }
3104
3105 if (mask & AT_GID) {
3106 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3107 cr, ZFS_GROUP, &fuidp);
3108 if (new_gid != zp->z_gid &&
3109 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3110 if (attrzp)
3111 VN_RELE(ZTOV(attrzp));
3112 err = SET_ERROR(EDQUOT);
3113 goto out2;
3114 }
3115 }
3116 }
3117 tx = dmu_tx_create(zfsvfs->z_os);
3118
3119 if (mask & AT_MODE) {
3120 uint64_t pmode = zp->z_mode;
3121 uint64_t acl_obj;
3122 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3123
3124 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3125 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3126 err = SET_ERROR(EPERM);
3127 goto out;
3128 }
3129
3130 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3131 goto out;
3132
3133 mutex_enter(&zp->z_lock);
3134 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3135 /*
3136 * Are we upgrading ACL from old V0 format
3137 * to V1 format?
3138 */
3139 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3140 zfs_znode_acl_version(zp) ==
3141 ZFS_ACL_VERSION_INITIAL) {
3142 dmu_tx_hold_free(tx, acl_obj, 0,
3143 DMU_OBJECT_END);
3144 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3145 0, aclp->z_acl_bytes);
3146 } else {
3147 dmu_tx_hold_write(tx, acl_obj, 0,
3148 aclp->z_acl_bytes);
3149 }
3150 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3151 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3152 0, aclp->z_acl_bytes);
3153 }
3154 mutex_exit(&zp->z_lock);
3155 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3156 } else {
3157 if ((mask & AT_XVATTR) &&
3158 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3159 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3160 else
3161 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3162 }
3163
3164 if (attrzp) {
3165 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3166 }
3167
3168 fuid_dirtied = zfsvfs->z_fuid_dirty;
3169 if (fuid_dirtied)
3170 zfs_fuid_txhold(zfsvfs, tx);
3171
3172 zfs_sa_upgrade_txholds(tx, zp);
3173
3174 err = dmu_tx_assign(tx, TXG_WAIT);
3175 if (err)
3176 goto out;
3177
3178 count = 0;
3179 /*
3180 * Set each attribute requested.
3181 * We group settings according to the locks they need to acquire.
3182 *
3183 * Note: you cannot set ctime directly, although it will be
3184 * updated as a side-effect of calling this function.
3185 */
3186
3187
3188 if (mask & (AT_UID|AT_GID|AT_MODE))
3189 mutex_enter(&zp->z_acl_lock);
3190 mutex_enter(&zp->z_lock);
3191
3192 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3193 &zp->z_pflags, sizeof (zp->z_pflags));
3194
3195 if (attrzp) {
3196 if (mask & (AT_UID|AT_GID|AT_MODE))
3197 mutex_enter(&attrzp->z_acl_lock);
3198 mutex_enter(&attrzp->z_lock);
3199 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3200 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3201 sizeof (attrzp->z_pflags));
3202 }
3203
3204 if (mask & (AT_UID|AT_GID)) {
3205
3206 if (mask & AT_UID) {
3207 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3208 &new_uid, sizeof (new_uid));
3209 zp->z_uid = new_uid;
3210 if (attrzp) {
3211 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3212 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3213 sizeof (new_uid));
3214 attrzp->z_uid = new_uid;
3215 }
3216 }
3217
3218 if (mask & AT_GID) {
3219 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3220 NULL, &new_gid, sizeof (new_gid));
3221 zp->z_gid = new_gid;
3222 if (attrzp) {
3223 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3224 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3225 sizeof (new_gid));
3226 attrzp->z_gid = new_gid;
3227 }
3228 }
3229 if (!(mask & AT_MODE)) {
3230 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3231 NULL, &new_mode, sizeof (new_mode));
3232 new_mode = zp->z_mode;
3233 }
3234 err = zfs_acl_chown_setattr(zp);
3235 ASSERT(err == 0);
3236 if (attrzp) {
3237 err = zfs_acl_chown_setattr(attrzp);
3238 ASSERT(err == 0);
3239 }
3240 }
3241
3242 if (mask & AT_MODE) {
3243 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3244 &new_mode, sizeof (new_mode));
3245 zp->z_mode = new_mode;
3246 ASSERT3U((uintptr_t)aclp, !=, NULL);
3247 err = zfs_aclset_common(zp, aclp, cr, tx);
3248 ASSERT0(err);
3249 if (zp->z_acl_cached)
3250 zfs_acl_free(zp->z_acl_cached);
3251 zp->z_acl_cached = aclp;
3252 aclp = NULL;
3253 }
3254
3255
3256 if (mask & AT_ATIME) {
3257 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3258 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3259 &zp->z_atime, sizeof (zp->z_atime));
3260 }
3261
3262 if (mask & AT_MTIME) {
3263 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3264 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3265 mtime, sizeof (mtime));
3266 }
3267
3268 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3269 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3270 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3271 NULL, mtime, sizeof (mtime));
3272 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3273 &ctime, sizeof (ctime));
3274 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3275 B_TRUE);
3276 } else if (mask != 0) {
3277 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3278 &ctime, sizeof (ctime));
3279 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3280 B_TRUE);
3281 if (attrzp) {
3282 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3283 SA_ZPL_CTIME(zfsvfs), NULL,
3284 &ctime, sizeof (ctime));
3285 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3286 mtime, ctime, B_TRUE);
3287 }
3288 }
3289 /*
3290 * Do this after setting timestamps to prevent timestamp
3291 * update from toggling bit
3292 */
3293
3294 if (xoap && (mask & AT_XVATTR)) {
3295
3296 /*
3297 * restore trimmed off masks
3298 * so that return masks can be set for caller.
3299 */
3300
3301 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3302 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3303 }
3304 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3305 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3306 }
3307 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3308 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3309 }
3310 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3311 XVA_SET_REQ(xvap, XAT_NODUMP);
3312 }
3313 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3314 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3315 }
3316 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3317 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3318 }
3319
3320 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3321 ASSERT(vp->v_type == VREG);
3322
3323 zfs_xvattr_set(zp, xvap, tx);
3324 }
3325
3326 if (fuid_dirtied)
3327 zfs_fuid_sync(zfsvfs, tx);
3328
3329 if (mask != 0)
3330 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3331
3332 mutex_exit(&zp->z_lock);
3333 if (mask & (AT_UID|AT_GID|AT_MODE))
3334 mutex_exit(&zp->z_acl_lock);
3335
3336 if (attrzp) {
3337 if (mask & (AT_UID|AT_GID|AT_MODE))
3338 mutex_exit(&attrzp->z_acl_lock);
3339 mutex_exit(&attrzp->z_lock);
3340 }
3341 out:
3342 if (err == 0 && attrzp) {
3343 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3344 xattr_count, tx);
3345 ASSERT(err2 == 0);
3346 }
3347
3348 if (attrzp)
3349 VN_RELE(ZTOV(attrzp));
3350
3351 if (aclp)
3352 zfs_acl_free(aclp);
3353
3354 if (fuidp) {
3355 zfs_fuid_info_free(fuidp);
3356 fuidp = NULL;
3357 }
3358
3359 if (err) {
3360 dmu_tx_abort(tx);
3361 if (err == ERESTART)
3362 goto top;
3363 } else {
3364 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3365 dmu_tx_commit(tx);
3366 }
3367
3368 out2:
3369 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3370 zil_commit(zilog, 0);
3371
3372 ZFS_EXIT(zfsvfs);
3373 return (err);
3374 }
3375
3376 typedef struct zfs_zlock {
3377 krwlock_t *zl_rwlock; /* lock we acquired */
3378 znode_t *zl_znode; /* znode we held */
3379 struct zfs_zlock *zl_next; /* next in list */
3380 } zfs_zlock_t;
3381
3382 /*
3383 * Drop locks and release vnodes that were held by zfs_rename_lock().
3384 */
3385 static void
3386 zfs_rename_unlock(zfs_zlock_t **zlpp)
3387 {
3388 zfs_zlock_t *zl;
3389
3390 while ((zl = *zlpp) != NULL) {
3391 if (zl->zl_znode != NULL)
3392 VN_RELE(ZTOV(zl->zl_znode));
3393 rw_exit(zl->zl_rwlock);
3394 *zlpp = zl->zl_next;
3395 kmem_free(zl, sizeof (*zl));
3396 }
3397 }
3398
3399 /*
3400 * Search back through the directory tree, using the ".." entries.
3401 * Lock each directory in the chain to prevent concurrent renames.
3402 * Fail any attempt to move a directory into one of its own descendants.
3403 * XXX - z_parent_lock can overlap with map or grow locks
3404 */
3405 static int
3406 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3407 {
3408 zfs_zlock_t *zl;
3409 znode_t *zp = tdzp;
3410 uint64_t rootid = zp->z_zfsvfs->z_root;
3411 uint64_t oidp = zp->z_id;
3412 krwlock_t *rwlp = &szp->z_parent_lock;
3413 krw_t rw = RW_WRITER;
3414
3415 /*
3416 * First pass write-locks szp and compares to zp->z_id.
3417 * Later passes read-lock zp and compare to zp->z_parent.
3418 */
3419 do {
3420 if (!rw_tryenter(rwlp, rw)) {
3421 /*
3422 * Another thread is renaming in this path.
3423 * Note that if we are a WRITER, we don't have any
3424 * parent_locks held yet.
3425 */
3426 if (rw == RW_READER && zp->z_id > szp->z_id) {
3427 /*
3428 * Drop our locks and restart
3429 */
3430 zfs_rename_unlock(&zl);
3431 *zlpp = NULL;
3432 zp = tdzp;
3433 oidp = zp->z_id;
3434 rwlp = &szp->z_parent_lock;
3435 rw = RW_WRITER;
3436 continue;
3437 } else {
3438 /*
3439 * Wait for other thread to drop its locks
3440 */
3441 rw_enter(rwlp, rw);
3442 }
3443 }
3444
3445 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3446 zl->zl_rwlock = rwlp;
3447 zl->zl_znode = NULL;
3448 zl->zl_next = *zlpp;
3449 *zlpp = zl;
3450
3451 if (oidp == szp->z_id) /* We're a descendant of szp */
3452 return (SET_ERROR(EINVAL));
3453
3454 if (oidp == rootid) /* We've hit the top */
3455 return (0);
3456
3457 if (rw == RW_READER) { /* i.e. not the first pass */
3458 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3459 if (error)
3460 return (error);
3461 zl->zl_znode = zp;
3462 }
3463 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3464 &oidp, sizeof (oidp));
3465 rwlp = &zp->z_parent_lock;
3466 rw = RW_READER;
3467
3468 } while (zp->z_id != sdzp->z_id);
3469
3470 return (0);
3471 }
3472
3473 /*
3474 * Move an entry from the provided source directory to the target
3475 * directory. Change the entry name as indicated.
3476 *
3477 * IN: sdvp - Source directory containing the "old entry".
3478 * snm - Old entry name.
3479 * tdvp - Target directory to contain the "new entry".
3480 * tnm - New entry name.
3481 * cr - credentials of caller.
3482 * ct - caller context
3483 * flags - case flags
3484 *
3485 * RETURN: 0 on success, error code on failure.
3486 *
3487 * Timestamps:
3488 * sdvp,tdvp - ctime|mtime updated
3489 */
3490 /*ARGSUSED*/
3491 static int
3492 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3493 caller_context_t *ct, int flags)
3494 {
3495 znode_t *tdzp, *szp, *tzp;
3496 znode_t *sdzp = VTOZ(sdvp);
3497 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3498 zilog_t *zilog;
3499 vnode_t *realvp;
3500 zfs_dirlock_t *sdl, *tdl;
3501 dmu_tx_t *tx;
3502 zfs_zlock_t *zl;
3503 int cmp, serr, terr;
3504 int error = 0;
3505 int zflg = 0;
3506 boolean_t waited = B_FALSE;
3507
3508 ZFS_ENTER(zfsvfs);
3509 ZFS_VERIFY_ZP(sdzp);
3510 zilog = zfsvfs->z_log;
3511
3512 /*
3513 * Make sure we have the real vp for the target directory.
3514 */
3515 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3516 tdvp = realvp;
3517
3518 tdzp = VTOZ(tdvp);
3519 ZFS_VERIFY_ZP(tdzp);
3520
3521 /*
3522 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3523 * ctldir appear to have the same v_vfsp.
3524 */
3525 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3526 ZFS_EXIT(zfsvfs);
3527 return (SET_ERROR(EXDEV));
3528 }
3529
3530 if (zfsvfs->z_utf8 && u8_validate(tnm,
3531 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3532 ZFS_EXIT(zfsvfs);
3533 return (SET_ERROR(EILSEQ));
3534 }
3535
3536 if (flags & FIGNORECASE)
3537 zflg |= ZCILOOK;
3538
3539 top:
3540 szp = NULL;
3541 tzp = NULL;
3542 zl = NULL;
3543
3544 /*
3545 * This is to prevent the creation of links into attribute space
3546 * by renaming a linked file into/outof an attribute directory.
3547 * See the comment in zfs_link() for why this is considered bad.
3548 */
3549 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3550 ZFS_EXIT(zfsvfs);
3551 return (SET_ERROR(EINVAL));
3552 }
3553
3554 /*
3555 * Lock source and target directory entries. To prevent deadlock,
3556 * a lock ordering must be defined. We lock the directory with
3557 * the smallest object id first, or if it's a tie, the one with
3558 * the lexically first name.
3559 */
3560 if (sdzp->z_id < tdzp->z_id) {
3561 cmp = -1;
3562 } else if (sdzp->z_id > tdzp->z_id) {
3563 cmp = 1;
3564 } else {
3565 /*
3566 * First compare the two name arguments without
3567 * considering any case folding.
3568 */
3569 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3570
3571 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3572 ASSERT(error == 0 || !zfsvfs->z_utf8);
3573 if (cmp == 0) {
3574 /*
3575 * POSIX: "If the old argument and the new argument
3576 * both refer to links to the same existing file,
3577 * the rename() function shall return successfully
3578 * and perform no other action."
3579 */
3580 ZFS_EXIT(zfsvfs);
3581 return (0);
3582 }
3583 /*
3584 * If the file system is case-folding, then we may
3585 * have some more checking to do. A case-folding file
3586 * system is either supporting mixed case sensitivity
3587 * access or is completely case-insensitive. Note
3588 * that the file system is always case preserving.
3589 *
3590 * In mixed sensitivity mode case sensitive behavior
3591 * is the default. FIGNORECASE must be used to
3592 * explicitly request case insensitive behavior.
3593 *
3594 * If the source and target names provided differ only
3595 * by case (e.g., a request to rename 'tim' to 'Tim'),
3596 * we will treat this as a special case in the
3597 * case-insensitive mode: as long as the source name
3598 * is an exact match, we will allow this to proceed as
3599 * a name-change request.
3600 */
3601 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3602 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3603 flags & FIGNORECASE)) &&
3604 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3605 &error) == 0) {
3606 /*
3607 * case preserving rename request, require exact
3608 * name matches
3609 */
3610 zflg |= ZCIEXACT;
3611 zflg &= ~ZCILOOK;
3612 }
3613 }
3614
3615 /*
3616 * If the source and destination directories are the same, we should
3617 * grab the z_name_lock of that directory only once.
3618 */
3619 if (sdzp == tdzp) {
3620 zflg |= ZHAVELOCK;
3621 rw_enter(&sdzp->z_name_lock, RW_READER);
3622 }
3623
3624 if (cmp < 0) {
3625 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3626 ZEXISTS | zflg, NULL, NULL);
3627 terr = zfs_dirent_lock(&tdl,
3628 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3629 } else {
3630 terr = zfs_dirent_lock(&tdl,
3631 tdzp, tnm, &tzp, zflg, NULL, NULL);
3632 serr = zfs_dirent_lock(&sdl,
3633 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3634 NULL, NULL);
3635 }
3636
3637 if (serr) {
3638 /*
3639 * Source entry invalid or not there.
3640 */
3641 if (!terr) {
3642 zfs_dirent_unlock(tdl);
3643 if (tzp)
3644 VN_RELE(ZTOV(tzp));
3645 }
3646
3647 if (sdzp == tdzp)
3648 rw_exit(&sdzp->z_name_lock);
3649
3650 if (strcmp(snm, "..") == 0)
3651 serr = SET_ERROR(EINVAL);
3652 ZFS_EXIT(zfsvfs);
3653 return (serr);
3654 }
3655 if (terr) {
3656 zfs_dirent_unlock(sdl);
3657 VN_RELE(ZTOV(szp));
3658
3659 if (sdzp == tdzp)
3660 rw_exit(&sdzp->z_name_lock);
3661
3662 if (strcmp(tnm, "..") == 0)
3663 terr = SET_ERROR(EINVAL);
3664 ZFS_EXIT(zfsvfs);
3665 return (terr);
3666 }
3667
3668 /*
3669 * Must have write access at the source to remove the old entry
3670 * and write access at the target to create the new entry.
3671 * Note that if target and source are the same, this can be
3672 * done in a single check.
3673 */
3674
3675 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3676 goto out;
3677
3678 if (ZTOV(szp)->v_type == VDIR) {
3679 /*
3680 * Check to make sure rename is valid.
3681 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3682 */
3683 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3684 goto out;
3685 }
3686
3687 /*
3688 * Does target exist?
3689 */
3690 if (tzp) {
3691 /*
3692 * Source and target must be the same type.
3693 */
3694 if (ZTOV(szp)->v_type == VDIR) {
3695 if (ZTOV(tzp)->v_type != VDIR) {
3696 error = SET_ERROR(ENOTDIR);
3697 goto out;
3698 }
3699 } else {
3700 if (ZTOV(tzp)->v_type == VDIR) {
3701 error = SET_ERROR(EISDIR);
3702 goto out;
3703 }
3704 }
3705 /*
3706 * POSIX dictates that when the source and target
3707 * entries refer to the same file object, rename
3708 * must do nothing and exit without error.
3709 */
3710 if (szp->z_id == tzp->z_id) {
3711 error = 0;
3712 goto out;
3713 }
3714 }
3715
3716 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3717 if (tzp)
3718 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3719
3720 /*
3721 * notify the target directory if it is not the same
3722 * as source directory.
3723 */
3724 if (tdvp != sdvp) {
3725 vnevent_rename_dest_dir(tdvp, ct);
3726 }
3727
3728 tx = dmu_tx_create(zfsvfs->z_os);
3729 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3730 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3731 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3732 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3733 if (sdzp != tdzp) {
3734 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3735 zfs_sa_upgrade_txholds(tx, tdzp);
3736 }
3737 if (tzp) {
3738 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3739 zfs_sa_upgrade_txholds(tx, tzp);
3740 }
3741
3742 zfs_sa_upgrade_txholds(tx, szp);
3743 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3744 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3745 if (error) {
3746 if (zl != NULL)
3747 zfs_rename_unlock(&zl);
3748 zfs_dirent_unlock(sdl);
3749 zfs_dirent_unlock(tdl);
3750
3751 if (sdzp == tdzp)
3752 rw_exit(&sdzp->z_name_lock);
3753
3754 VN_RELE(ZTOV(szp));
3755 if (tzp)
3756 VN_RELE(ZTOV(tzp));
3757 if (error == ERESTART) {
3758 waited = B_TRUE;
3759 dmu_tx_wait(tx);
3760 dmu_tx_abort(tx);
3761 goto top;
3762 }
3763 dmu_tx_abort(tx);
3764 ZFS_EXIT(zfsvfs);
3765 return (error);
3766 }
3767
3768 if (tzp) /* Attempt to remove the existing target */
3769 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3770
3771 if (error == 0) {
3772 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3773 if (error == 0) {
3774 szp->z_pflags |= ZFS_AV_MODIFIED;
3775
3776 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3777 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3778 ASSERT0(error);
3779
3780 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3781 if (error == 0) {
3782 zfs_log_rename(zilog, tx, TX_RENAME |
3783 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3784 sdl->dl_name, tdzp, tdl->dl_name, szp);
3785
3786 /*
3787 * Update path information for the target vnode
3788 */
3789 vn_renamepath(tdvp, ZTOV(szp), tnm,
3790 strlen(tnm));
3791 } else {
3792 /*
3793 * At this point, we have successfully created
3794 * the target name, but have failed to remove
3795 * the source name. Since the create was done
3796 * with the ZRENAMING flag, there are
3797 * complications; for one, the link count is
3798 * wrong. The easiest way to deal with this
3799 * is to remove the newly created target, and
3800 * return the original error. This must
3801 * succeed; fortunately, it is very unlikely to
3802 * fail, since we just created it.
3803 */
3804 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3805 ZRENAMING, NULL), ==, 0);
3806 }
3807 }
3808 }
3809
3810 dmu_tx_commit(tx);
3811 out:
3812 if (zl != NULL)
3813 zfs_rename_unlock(&zl);
3814
3815 zfs_dirent_unlock(sdl);
3816 zfs_dirent_unlock(tdl);
3817
3818 if (sdzp == tdzp)
3819 rw_exit(&sdzp->z_name_lock);
3820
3821
3822 VN_RELE(ZTOV(szp));
3823 if (tzp)
3824 VN_RELE(ZTOV(tzp));
3825
3826 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3827 zil_commit(zilog, 0);
3828
3829 ZFS_EXIT(zfsvfs);
3830 return (error);
3831 }
3832
3833 /*
3834 * Insert the indicated symbolic reference entry into the directory.
3835 *
3836 * IN: dvp - Directory to contain new symbolic link.
3837 * link - Name for new symlink entry.
3838 * vap - Attributes of new entry.
3839 * cr - credentials of caller.
3840 * ct - caller context
3841 * flags - case flags
3842 *
3843 * RETURN: 0 on success, error code on failure.
3844 *
3845 * Timestamps:
3846 * dvp - ctime|mtime updated
3847 */
3848 /*ARGSUSED*/
3849 static int
3850 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3851 caller_context_t *ct, int flags)
3852 {
3853 znode_t *zp, *dzp = VTOZ(dvp);
3854 zfs_dirlock_t *dl;
3855 dmu_tx_t *tx;
3856 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
3857 zilog_t *zilog;
3858 uint64_t len = strlen(link);
3859 int error;
3860 int zflg = ZNEW;
3861 zfs_acl_ids_t acl_ids;
3862 boolean_t fuid_dirtied;
3863 uint64_t txtype = TX_SYMLINK;
3864 boolean_t waited = B_FALSE;
3865
3866 ASSERT(vap->va_type == VLNK);
3867
3868 ZFS_ENTER(zfsvfs);
3869 ZFS_VERIFY_ZP(dzp);
3870 zilog = zfsvfs->z_log;
3871
3872 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3873 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3874 ZFS_EXIT(zfsvfs);
3875 return (SET_ERROR(EILSEQ));
3876 }
3877 if (flags & FIGNORECASE)
3878 zflg |= ZCILOOK;
3879
3880 if (len > MAXPATHLEN) {
3881 ZFS_EXIT(zfsvfs);
3882 return (SET_ERROR(ENAMETOOLONG));
3883 }
3884
3885 if ((error = zfs_acl_ids_create(dzp, 0,
3886 vap, cr, NULL, &acl_ids)) != 0) {
3887 ZFS_EXIT(zfsvfs);
3888 return (error);
3889 }
3890 top:
3891 /*
3892 * Attempt to lock directory; fail if entry already exists.
3893 */
3894 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3895 if (error) {
3896 zfs_acl_ids_free(&acl_ids);
3897 ZFS_EXIT(zfsvfs);
3898 return (error);
3899 }
3900
3901 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3902 zfs_acl_ids_free(&acl_ids);
3903 zfs_dirent_unlock(dl);
3904 ZFS_EXIT(zfsvfs);
3905 return (error);
3906 }
3907
3908 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3909 zfs_acl_ids_free(&acl_ids);
3910 zfs_dirent_unlock(dl);
3911 ZFS_EXIT(zfsvfs);
3912 return (SET_ERROR(EDQUOT));
3913 }
3914 tx = dmu_tx_create(zfsvfs->z_os);
3915 fuid_dirtied = zfsvfs->z_fuid_dirty;
3916 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3917 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3918 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3919 ZFS_SA_BASE_ATTR_SIZE + len);
3920 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3921 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3922 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3923 acl_ids.z_aclp->z_acl_bytes);
3924 }
3925 if (fuid_dirtied)
3926 zfs_fuid_txhold(zfsvfs, tx);
3927 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3928 if (error) {
3929 zfs_dirent_unlock(dl);
3930 if (error == ERESTART) {
3931 waited = B_TRUE;
3932 dmu_tx_wait(tx);
3933 dmu_tx_abort(tx);
3934 goto top;
3935 }
3936 zfs_acl_ids_free(&acl_ids);
3937 dmu_tx_abort(tx);
3938 ZFS_EXIT(zfsvfs);
3939 return (error);
3940 }
3941
3942 /*
3943 * Create a new object for the symlink.
3944 * for version 4 ZPL datsets the symlink will be an SA attribute
3945 */
3946 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3947
3948 if (fuid_dirtied)
3949 zfs_fuid_sync(zfsvfs, tx);
3950
3951 mutex_enter(&zp->z_lock);
3952 if (zp->z_is_sa)
3953 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3954 link, len, tx);
3955 else
3956 zfs_sa_symlink(zp, link, len, tx);
3957 mutex_exit(&zp->z_lock);
3958
3959 zp->z_size = len;
3960 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3961 &zp->z_size, sizeof (zp->z_size), tx);
3962 /*
3963 * Insert the new object into the directory.
3964 */
3965 (void) zfs_link_create(dl, zp, tx, ZNEW);
3966
3967 if (flags & FIGNORECASE)
3968 txtype |= TX_CI;
3969 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3970
3971 zfs_acl_ids_free(&acl_ids);
3972
3973 dmu_tx_commit(tx);
3974
3975 zfs_dirent_unlock(dl);
3976
3977 VN_RELE(ZTOV(zp));
3978
3979 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3980 zil_commit(zilog, 0);
3981
3982 ZFS_EXIT(zfsvfs);
3983 return (error);
3984 }
3985
3986 /*
3987 * Return, in the buffer contained in the provided uio structure,
3988 * the symbolic path referred to by vp.
3989 *
3990 * IN: vp - vnode of symbolic link.
3991 * uio - structure to contain the link path.
3992 * cr - credentials of caller.
3993 * ct - caller context
3994 *
3995 * OUT: uio - structure containing the link path.
3996 *
3997 * RETURN: 0 on success, error code on failure.
3998 *
3999 * Timestamps:
4000 * vp - atime updated
4001 */
4002 /* ARGSUSED */
4003 static int
4004 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
4005 {
4006 znode_t *zp = VTOZ(vp);
4007 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4008 int error;
4009
4010 ZFS_ENTER(zfsvfs);
4011 ZFS_VERIFY_ZP(zp);
4012
4013 mutex_enter(&zp->z_lock);
4014 if (zp->z_is_sa)
4015 error = sa_lookup_uio(zp->z_sa_hdl,
4016 SA_ZPL_SYMLINK(zfsvfs), uio);
4017 else
4018 error = zfs_sa_readlink(zp, uio);
4019 mutex_exit(&zp->z_lock);
4020
4021 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4022
4023 ZFS_EXIT(zfsvfs);
4024 return (error);
4025 }
4026
4027 /*
4028 * Insert a new entry into directory tdvp referencing svp.
4029 *
4030 * IN: tdvp - Directory to contain new entry.
4031 * svp - vnode of new entry.
4032 * name - name of new entry.
4033 * cr - credentials of caller.
4034 * ct - caller context
4035 *
4036 * RETURN: 0 on success, error code on failure.
4037 *
4038 * Timestamps:
4039 * tdvp - ctime|mtime updated
4040 * svp - ctime updated
4041 */
4042 /* ARGSUSED */
4043 static int
4044 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
4045 caller_context_t *ct, int flags)
4046 {
4047 znode_t *dzp = VTOZ(tdvp);
4048 znode_t *tzp, *szp;
4049 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
4050 zilog_t *zilog;
4051 zfs_dirlock_t *dl;
4052 dmu_tx_t *tx;
4053 vnode_t *realvp;
4054 int error;
4055 int zf = ZNEW;
4056 uint64_t parent;
4057 uid_t owner;
4058 boolean_t waited = B_FALSE;
4059
4060 ASSERT(tdvp->v_type == VDIR);
4061
4062 ZFS_ENTER(zfsvfs);
4063 ZFS_VERIFY_ZP(dzp);
4064 zilog = zfsvfs->z_log;
4065
4066 if (VOP_REALVP(svp, &realvp, ct) == 0)
4067 svp = realvp;
4068
4069 /*
4070 * POSIX dictates that we return EPERM here.
4071 * Better choices include ENOTSUP or EISDIR.
4072 */
4073 if (svp->v_type == VDIR) {
4074 ZFS_EXIT(zfsvfs);
4075 return (SET_ERROR(EPERM));
4076 }
4077
4078 szp = VTOZ(svp);
4079 ZFS_VERIFY_ZP(szp);
4080
4081 /*
4082 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4083 * ctldir appear to have the same v_vfsp.
4084 */
4085 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4086 ZFS_EXIT(zfsvfs);
4087 return (SET_ERROR(EXDEV));
4088 }
4089
4090 /* Prevent links to .zfs/shares files */
4091
4092 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4093 &parent, sizeof (uint64_t))) != 0) {
4094 ZFS_EXIT(zfsvfs);
4095 return (error);
4096 }
4097 if (parent == zfsvfs->z_shares_dir) {
4098 ZFS_EXIT(zfsvfs);
4099 return (SET_ERROR(EPERM));
4100 }
4101
4102 if (zfsvfs->z_utf8 && u8_validate(name,
4103 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4104 ZFS_EXIT(zfsvfs);
4105 return (SET_ERROR(EILSEQ));
4106 }
4107 if (flags & FIGNORECASE)
4108 zf |= ZCILOOK;
4109
4110 /*
4111 * We do not support links between attributes and non-attributes
4112 * because of the potential security risk of creating links
4113 * into "normal" file space in order to circumvent restrictions
4114 * imposed in attribute space.
4115 */
4116 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4117 ZFS_EXIT(zfsvfs);
4118 return (SET_ERROR(EINVAL));
4119 }
4120
4121
4122 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4123 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4124 ZFS_EXIT(zfsvfs);
4125 return (SET_ERROR(EPERM));
4126 }
4127
4128 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4129 ZFS_EXIT(zfsvfs);
4130 return (error);
4131 }
4132
4133 top:
4134 /*
4135 * Attempt to lock directory; fail if entry already exists.
4136 */
4137 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4138 if (error) {
4139 ZFS_EXIT(zfsvfs);
4140 return (error);
4141 }
4142
4143 tx = dmu_tx_create(zfsvfs->z_os);
4144 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4145 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4146 zfs_sa_upgrade_txholds(tx, szp);
4147 zfs_sa_upgrade_txholds(tx, dzp);
4148 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4149 if (error) {
4150 zfs_dirent_unlock(dl);
4151 if (error == ERESTART) {
4152 waited = B_TRUE;
4153 dmu_tx_wait(tx);
4154 dmu_tx_abort(tx);
4155 goto top;
4156 }
4157 dmu_tx_abort(tx);
4158 ZFS_EXIT(zfsvfs);
4159 return (error);
4160 }
4161
4162 error = zfs_link_create(dl, szp, tx, 0);
4163
4164 if (error == 0) {
4165 uint64_t txtype = TX_LINK;
4166 if (flags & FIGNORECASE)
4167 txtype |= TX_CI;
4168 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4169 }
4170
4171 dmu_tx_commit(tx);
4172
4173 zfs_dirent_unlock(dl);
4174
4175 if (error == 0) {
4176 vnevent_link(svp, ct);
4177 }
4178
4179 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4180 zil_commit(zilog, 0);
4181
4182 ZFS_EXIT(zfsvfs);
4183 return (error);
4184 }
4185
4186 /*
4187 * zfs_null_putapage() is used when the file system has been force
4188 * unmounted. It just drops the pages.
4189 */
4190 /* ARGSUSED */
4191 static int
4192 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4193 size_t *lenp, int flags, cred_t *cr)
4194 {
4195 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4196 return (0);
4197 }
4198
4199 /*
4200 * Push a page out to disk, klustering if possible.
4201 *
4202 * IN: vp - file to push page to.
4203 * pp - page to push.
4204 * flags - additional flags.
4205 * cr - credentials of caller.
4206 *
4207 * OUT: offp - start of range pushed.
4208 * lenp - len of range pushed.
4209 *
4210 * RETURN: 0 on success, error code on failure.
4211 *
4212 * NOTE: callers must have locked the page to be pushed. On
4213 * exit, the page (and all other pages in the kluster) must be
4214 * unlocked.
4215 */
4216 /* ARGSUSED */
4217 static int
4218 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4219 size_t *lenp, int flags, cred_t *cr)
4220 {
4221 znode_t *zp = VTOZ(vp);
4222 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4223 dmu_tx_t *tx;
4224 u_offset_t off, koff;
4225 size_t len, klen;
4226 int err;
4227
4228 off = pp->p_offset;
4229 len = PAGESIZE;
4230 /*
4231 * If our blocksize is bigger than the page size, try to kluster
4232 * multiple pages so that we write a full block (thus avoiding
4233 * a read-modify-write).
4234 */
4235 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4236 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4237 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4238 ASSERT(koff <= zp->z_size);
4239 if (koff + klen > zp->z_size)
4240 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4241 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4242 }
4243 ASSERT3U(btop(len), ==, btopr(len));
4244
4245 /*
4246 * Can't push pages past end-of-file.
4247 */
4248 if (off >= zp->z_size) {
4249 /* ignore all pages */
4250 err = 0;
4251 goto out;
4252 } else if (off + len > zp->z_size) {
4253 int npages = btopr(zp->z_size - off);
4254 page_t *trunc;
4255
4256 page_list_break(&pp, &trunc, npages);
4257 /* ignore pages past end of file */
4258 if (trunc)
4259 pvn_write_done(trunc, flags);
4260 len = zp->z_size - off;
4261 }
4262
4263 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4264 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4265 err = SET_ERROR(EDQUOT);
4266 goto out;
4267 }
4268 tx = dmu_tx_create(zfsvfs->z_os);
4269 dmu_tx_hold_write(tx, zp->z_id, off, len);
4270
4271 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4272 zfs_sa_upgrade_txholds(tx, zp);
4273 err = dmu_tx_assign(tx, TXG_WAIT);
4274 if (err != 0) {
4275 dmu_tx_abort(tx);
4276 goto out;
4277 }
4278
4279 if (zp->z_blksz <= PAGESIZE) {
4280 caddr_t va = zfs_map_page(pp, S_READ);
4281 ASSERT3U(len, <=, PAGESIZE);
4282 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4283 zfs_unmap_page(pp, va);
4284 } else {
4285 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4286 }
4287
4288 if (err == 0) {
4289 uint64_t mtime[2], ctime[2];
4290 sa_bulk_attr_t bulk[3];
4291 int count = 0;
4292
4293 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4294 &mtime, 16);
4295 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4296 &ctime, 16);
4297 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4298 &zp->z_pflags, 8);
4299 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4300 B_TRUE);
4301 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4302 }
4303 dmu_tx_commit(tx);
4304
4305 out:
4306 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4307 if (offp)
4308 *offp = off;
4309 if (lenp)
4310 *lenp = len;
4311
4312 return (err);
4313 }
4314
4315 /*
4316 * Copy the portion of the file indicated from pages into the file.
4317 * The pages are stored in a page list attached to the files vnode.
4318 *
4319 * IN: vp - vnode of file to push page data to.
4320 * off - position in file to put data.
4321 * len - amount of data to write.
4322 * flags - flags to control the operation.
4323 * cr - credentials of caller.
4324 * ct - caller context.
4325 *
4326 * RETURN: 0 on success, error code on failure.
4327 *
4328 * Timestamps:
4329 * vp - ctime|mtime updated
4330 */
4331 /*ARGSUSED*/
4332 static int
4333 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4334 caller_context_t *ct)
4335 {
4336 znode_t *zp = VTOZ(vp);
4337 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4338 page_t *pp;
4339 size_t io_len;
4340 u_offset_t io_off;
4341 uint_t blksz;
4342 rl_t *rl;
4343 int error = 0;
4344
4345 ZFS_ENTER(zfsvfs);
4346 ZFS_VERIFY_ZP(zp);
4347
4348 /*
4349 * There's nothing to do if no data is cached.
4350 */
4351 if (!vn_has_cached_data(vp)) {
4352 ZFS_EXIT(zfsvfs);
4353 return (0);
4354 }
4355
4356 /*
4357 * Align this request to the file block size in case we kluster.
4358 * XXX - this can result in pretty aggresive locking, which can
4359 * impact simultanious read/write access. One option might be
4360 * to break up long requests (len == 0) into block-by-block
4361 * operations to get narrower locking.
4362 */
4363 blksz = zp->z_blksz;
4364 if (ISP2(blksz))
4365 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4366 else
4367 io_off = 0;
4368 if (len > 0 && ISP2(blksz))
4369 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4370 else
4371 io_len = 0;
4372
4373 if (io_len == 0) {
4374 /*
4375 * Search the entire vp list for pages >= io_off.
4376 */
4377 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4378 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4379 goto out;
4380 }
4381 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4382
4383 if (off > zp->z_size) {
4384 /* past end of file */
4385 zfs_range_unlock(rl);
4386 ZFS_EXIT(zfsvfs);
4387 return (0);
4388 }
4389
4390 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4391
4392 for (off = io_off; io_off < off + len; io_off += io_len) {
4393 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4394 pp = page_lookup(vp, io_off,
4395 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4396 } else {
4397 pp = page_lookup_nowait(vp, io_off,
4398 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4399 }
4400
4401 if (pp != NULL && pvn_getdirty(pp, flags)) {
4402 int err;
4403
4404 /*
4405 * Found a dirty page to push
4406 */
4407 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4408 if (err)
4409 error = err;
4410 } else {
4411 io_len = PAGESIZE;
4412 }
4413 }
4414 out:
4415 zfs_range_unlock(rl);
4416 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4417 zil_commit(zfsvfs->z_log, zp->z_id);
4418 ZFS_EXIT(zfsvfs);
4419 return (error);
4420 }
4421
4422 /*ARGSUSED*/
4423 void
4424 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4425 {
4426 znode_t *zp = VTOZ(vp);
4427 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4428 int error;
4429
4430 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4431 if (zp->z_sa_hdl == NULL) {
4432 /*
4433 * The fs has been unmounted, or we did a
4434 * suspend/resume and this file no longer exists.
4435 */
4436 if (vn_has_cached_data(vp)) {
4437 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4438 B_INVAL, cr);
4439 }
4440
4441 mutex_enter(&zp->z_lock);
4442 mutex_enter(&vp->v_lock);
4443 ASSERT(vp->v_count == 1);
4444 vp->v_count = 0;
4445 mutex_exit(&vp->v_lock);
4446 mutex_exit(&zp->z_lock);
4447 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4448 zfs_znode_free(zp);
4449 return;
4450 }
4451
4452 /*
4453 * Attempt to push any data in the page cache. If this fails
4454 * we will get kicked out later in zfs_zinactive().
4455 */
4456 if (vn_has_cached_data(vp)) {
4457 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4458 cr);
4459 }
4460
4461 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4462 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4463
4464 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4465 zfs_sa_upgrade_txholds(tx, zp);
4466 error = dmu_tx_assign(tx, TXG_WAIT);
4467 if (error) {
4468 dmu_tx_abort(tx);
4469 } else {
4470 mutex_enter(&zp->z_lock);
4471 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4472 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4473 zp->z_atime_dirty = 0;
4474 mutex_exit(&zp->z_lock);
4475 dmu_tx_commit(tx);
4476 }
4477 }
4478
4479 zfs_zinactive(zp);
4480 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4481 }
4482
4483 /*
4484 * Bounds-check the seek operation.
4485 *
4486 * IN: vp - vnode seeking within
4487 * ooff - old file offset
4488 * noffp - pointer to new file offset
4489 * ct - caller context
4490 *
4491 * RETURN: 0 on success, EINVAL if new offset invalid.
4492 */
4493 /* ARGSUSED */
4494 static int
4495 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4496 caller_context_t *ct)
4497 {
4498 if (vp->v_type == VDIR)
4499 return (0);
4500 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4501 }
4502
4503 /*
4504 * Pre-filter the generic locking function to trap attempts to place
4505 * a mandatory lock on a memory mapped file.
4506 */
4507 static int
4508 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4509 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4510 {
4511 znode_t *zp = VTOZ(vp);
4512 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4513
4514 ZFS_ENTER(zfsvfs);
4515 ZFS_VERIFY_ZP(zp);
4516
4517 /*
4518 * We are following the UFS semantics with respect to mapcnt
4519 * here: If we see that the file is mapped already, then we will
4520 * return an error, but we don't worry about races between this
4521 * function and zfs_map().
4522 */
4523 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4524 ZFS_EXIT(zfsvfs);
4525 return (SET_ERROR(EAGAIN));
4526 }
4527 ZFS_EXIT(zfsvfs);
4528 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4529 }
4530
4531 /*
4532 * If we can't find a page in the cache, we will create a new page
4533 * and fill it with file data. For efficiency, we may try to fill
4534 * multiple pages at once (klustering) to fill up the supplied page
4535 * list. Note that the pages to be filled are held with an exclusive
4536 * lock to prevent access by other threads while they are being filled.
4537 */
4538 static int
4539 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4540 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4541 {
4542 znode_t *zp = VTOZ(vp);
4543 page_t *pp, *cur_pp;
4544 objset_t *os = zp->z_zfsvfs->z_os;
4545 u_offset_t io_off, total;
4546 size_t io_len;
4547 int err;
4548
4549 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4550 /*
4551 * We only have a single page, don't bother klustering
4552 */
4553 io_off = off;
4554 io_len = PAGESIZE;
4555 pp = page_create_va(vp, io_off, io_len,
4556 PG_EXCL | PG_WAIT, seg, addr);
4557 } else {
4558 /*
4559 * Try to find enough pages to fill the page list
4560 */
4561 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4562 &io_len, off, plsz, 0);
4563 }
4564 if (pp == NULL) {
4565 /*
4566 * The page already exists, nothing to do here.
4567 */
4568 *pl = NULL;
4569 return (0);
4570 }
4571
4572 /*
4573 * Fill the pages in the kluster.
4574 */
4575 cur_pp = pp;
4576 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4577 caddr_t va;
4578
4579 ASSERT3U(io_off, ==, cur_pp->p_offset);
4580 va = zfs_map_page(cur_pp, S_WRITE);
4581 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4582 DMU_READ_PREFETCH);
4583 zfs_unmap_page(cur_pp, va);
4584 if (err) {
4585 /* On error, toss the entire kluster */
4586 pvn_read_done(pp, B_ERROR);
4587 /* convert checksum errors into IO errors */
4588 if (err == ECKSUM)
4589 err = SET_ERROR(EIO);
4590 return (err);
4591 }
4592 cur_pp = cur_pp->p_next;
4593 }
4594
4595 /*
4596 * Fill in the page list array from the kluster starting
4597 * from the desired offset `off'.
4598 * NOTE: the page list will always be null terminated.
4599 */
4600 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4601 ASSERT(pl == NULL || (*pl)->p_offset == off);
4602
4603 return (0);
4604 }
4605
4606 /*
4607 * Return pointers to the pages for the file region [off, off + len]
4608 * in the pl array. If plsz is greater than len, this function may
4609 * also return page pointers from after the specified region
4610 * (i.e. the region [off, off + plsz]). These additional pages are
4611 * only returned if they are already in the cache, or were created as
4612 * part of a klustered read.
4613 *
4614 * IN: vp - vnode of file to get data from.
4615 * off - position in file to get data from.
4616 * len - amount of data to retrieve.
4617 * plsz - length of provided page list.
4618 * seg - segment to obtain pages for.
4619 * addr - virtual address of fault.
4620 * rw - mode of created pages.
4621 * cr - credentials of caller.
4622 * ct - caller context.
4623 *
4624 * OUT: protp - protection mode of created pages.
4625 * pl - list of pages created.
4626 *
4627 * RETURN: 0 on success, error code on failure.
4628 *
4629 * Timestamps:
4630 * vp - atime updated
4631 */
4632 /* ARGSUSED */
4633 static int
4634 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4635 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4636 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4637 {
4638 znode_t *zp = VTOZ(vp);
4639 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4640 page_t **pl0 = pl;
4641 int err = 0;
4642
4643 /* we do our own caching, faultahead is unnecessary */
4644 if (pl == NULL)
4645 return (0);
4646 else if (len > plsz)
4647 len = plsz;
4648 else
4649 len = P2ROUNDUP(len, PAGESIZE);
4650 ASSERT(plsz >= len);
4651
4652 ZFS_ENTER(zfsvfs);
4653 ZFS_VERIFY_ZP(zp);
4654
4655 if (protp)
4656 *protp = PROT_ALL;
4657
4658 /*
4659 * Loop through the requested range [off, off + len) looking
4660 * for pages. If we don't find a page, we will need to create
4661 * a new page and fill it with data from the file.
4662 */
4663 while (len > 0) {
4664 if (*pl = page_lookup(vp, off, SE_SHARED))
4665 *(pl+1) = NULL;
4666 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4667 goto out;
4668 while (*pl) {
4669 ASSERT3U((*pl)->p_offset, ==, off);
4670 off += PAGESIZE;
4671 addr += PAGESIZE;
4672 if (len > 0) {
4673 ASSERT3U(len, >=, PAGESIZE);
4674 len -= PAGESIZE;
4675 }
4676 ASSERT3U(plsz, >=, PAGESIZE);
4677 plsz -= PAGESIZE;
4678 pl++;
4679 }
4680 }
4681
4682 /*
4683 * Fill out the page array with any pages already in the cache.
4684 */
4685 while (plsz > 0 &&
4686 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4687 off += PAGESIZE;
4688 plsz -= PAGESIZE;
4689 }
4690 out:
4691 if (err) {
4692 /*
4693 * Release any pages we have previously locked.
4694 */
4695 while (pl > pl0)
4696 page_unlock(*--pl);
4697 } else {
4698 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4699 }
4700
4701 *pl = NULL;
4702
4703 ZFS_EXIT(zfsvfs);
4704 return (err);
4705 }
4706
4707 /*
4708 * Request a memory map for a section of a file. This code interacts
4709 * with common code and the VM system as follows:
4710 *
4711 * - common code calls mmap(), which ends up in smmap_common()
4712 * - this calls VOP_MAP(), which takes you into (say) zfs
4713 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4714 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4715 * - zfs_addmap() updates z_mapcnt
4716 */
4717 /*ARGSUSED*/
4718 static int
4719 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4720 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4721 caller_context_t *ct)
4722 {
4723 znode_t *zp = VTOZ(vp);
4724 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4725 segvn_crargs_t vn_a;
4726 int error;
4727
4728 ZFS_ENTER(zfsvfs);
4729 ZFS_VERIFY_ZP(zp);
4730
4731 if ((prot & PROT_WRITE) && (zp->z_pflags &
4732 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4733 ZFS_EXIT(zfsvfs);
4734 return (SET_ERROR(EPERM));
4735 }
4736
4737 if ((prot & (PROT_READ | PROT_EXEC)) &&
4738 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4739 ZFS_EXIT(zfsvfs);
4740 return (SET_ERROR(EACCES));
4741 }
4742
4743 if (vp->v_flag & VNOMAP) {
4744 ZFS_EXIT(zfsvfs);
4745 return (SET_ERROR(ENOSYS));
4746 }
4747
4748 if (off < 0 || len > MAXOFFSET_T - off) {
4749 ZFS_EXIT(zfsvfs);
4750 return (SET_ERROR(ENXIO));
4751 }
4752
4753 if (vp->v_type != VREG) {
4754 ZFS_EXIT(zfsvfs);
4755 return (SET_ERROR(ENODEV));
4756 }
4757
4758 /*
4759 * If file is locked, disallow mapping.
4760 */
4761 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4762 ZFS_EXIT(zfsvfs);
4763 return (SET_ERROR(EAGAIN));
4764 }
4765
4766 as_rangelock(as);
4767 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4768 if (error != 0) {
4769 as_rangeunlock(as);
4770 ZFS_EXIT(zfsvfs);
4771 return (error);
4772 }
4773
4774 vn_a.vp = vp;
4775 vn_a.offset = (u_offset_t)off;
4776 vn_a.type = flags & MAP_TYPE;
4777 vn_a.prot = prot;
4778 vn_a.maxprot = maxprot;
4779 vn_a.cred = cr;
4780 vn_a.amp = NULL;
4781 vn_a.flags = flags & ~MAP_TYPE;
4782 vn_a.szc = 0;
4783 vn_a.lgrp_mem_policy_flags = 0;
4784
4785 error = as_map(as, *addrp, len, segvn_create, &vn_a);
4786
4787 as_rangeunlock(as);
4788 ZFS_EXIT(zfsvfs);
4789 return (error);
4790 }
4791
4792 /* ARGSUSED */
4793 static int
4794 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4795 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4796 caller_context_t *ct)
4797 {
4798 uint64_t pages = btopr(len);
4799
4800 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4801 return (0);
4802 }
4803
4804 /*
4805 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4806 * more accurate mtime for the associated file. Since we don't have a way of
4807 * detecting when the data was actually modified, we have to resort to
4808 * heuristics. If an explicit msync() is done, then we mark the mtime when the
4809 * last page is pushed. The problem occurs when the msync() call is omitted,
4810 * which by far the most common case:
4811 *
4812 * open()
4813 * mmap()
4814 * <modify memory>
4815 * munmap()
4816 * close()
4817 * <time lapse>
4818 * putpage() via fsflush
4819 *
4820 * If we wait until fsflush to come along, we can have a modification time that
4821 * is some arbitrary point in the future. In order to prevent this in the
4822 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4823 * torn down.
4824 */
4825 /* ARGSUSED */
4826 static int
4827 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4828 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4829 caller_context_t *ct)
4830 {
4831 uint64_t pages = btopr(len);
4832
4833 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4834 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4835
4836 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4837 vn_has_cached_data(vp))
4838 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4839
4840 return (0);
4841 }
4842
4843 /*
4844 * Free or allocate space in a file. Currently, this function only
4845 * supports the `F_FREESP' command. However, this command is somewhat
4846 * misnamed, as its functionality includes the ability to allocate as
4847 * well as free space.
4848 *
4849 * IN: vp - vnode of file to free data in.
4850 * cmd - action to take (only F_FREESP supported).
4851 * bfp - section of file to free/alloc.
4852 * flag - current file open mode flags.
4853 * offset - current file offset.
4854 * cr - credentials of caller [UNUSED].
4855 * ct - caller context.
4856 *
4857 * RETURN: 0 on success, error code on failure.
4858 *
4859 * Timestamps:
4860 * vp - ctime|mtime updated
4861 */
4862 /* ARGSUSED */
4863 static int
4864 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4865 offset_t offset, cred_t *cr, caller_context_t *ct)
4866 {
4867 znode_t *zp = VTOZ(vp);
4868 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4869 uint64_t off, len;
4870 int error;
4871
4872 ZFS_ENTER(zfsvfs);
4873 ZFS_VERIFY_ZP(zp);
4874
4875 if (cmd != F_FREESP) {
4876 ZFS_EXIT(zfsvfs);
4877 return (SET_ERROR(EINVAL));
4878 }
4879
4880 /*
4881 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
4882 * callers might not be able to detect properly that we are read-only,
4883 * so check it explicitly here.
4884 */
4885 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
4886 ZFS_EXIT(zfsvfs);
4887 return (SET_ERROR(EROFS));
4888 }
4889
4890 if (error = convoff(vp, bfp, 0, offset)) {
4891 ZFS_EXIT(zfsvfs);
4892 return (error);
4893 }
4894
4895 if (bfp->l_len < 0) {
4896 ZFS_EXIT(zfsvfs);
4897 return (SET_ERROR(EINVAL));
4898 }
4899
4900 off = bfp->l_start;
4901 len = bfp->l_len; /* 0 means from off to end of file */
4902
4903 error = zfs_freesp(zp, off, len, flag, TRUE);
4904
4905 if (error == 0 && off == 0 && len == 0)
4906 vnevent_truncate(ZTOV(zp), ct);
4907
4908 ZFS_EXIT(zfsvfs);
4909 return (error);
4910 }
4911
4912 /*ARGSUSED*/
4913 static int
4914 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4915 {
4916 znode_t *zp = VTOZ(vp);
4917 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4918 uint32_t gen;
4919 uint64_t gen64;
4920 uint64_t object = zp->z_id;
4921 zfid_short_t *zfid;
4922 int size, i, error;
4923
4924 ZFS_ENTER(zfsvfs);
4925 ZFS_VERIFY_ZP(zp);
4926
4927 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4928 &gen64, sizeof (uint64_t))) != 0) {
4929 ZFS_EXIT(zfsvfs);
4930 return (error);
4931 }
4932
4933 gen = (uint32_t)gen64;
4934
4935 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4936 if (fidp->fid_len < size) {
4937 fidp->fid_len = size;
4938 ZFS_EXIT(zfsvfs);
4939 return (SET_ERROR(ENOSPC));
4940 }
4941
4942 zfid = (zfid_short_t *)fidp;
4943
4944 zfid->zf_len = size;
4945
4946 for (i = 0; i < sizeof (zfid->zf_object); i++)
4947 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4948
4949 /* Must have a non-zero generation number to distinguish from .zfs */
4950 if (gen == 0)
4951 gen = 1;
4952 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4953 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4954
4955 if (size == LONG_FID_LEN) {
4956 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
4957 zfid_long_t *zlfid;
4958
4959 zlfid = (zfid_long_t *)fidp;
4960
4961 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4962 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4963
4964 /* XXX - this should be the generation number for the objset */
4965 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4966 zlfid->zf_setgen[i] = 0;
4967 }
4968
4969 ZFS_EXIT(zfsvfs);
4970 return (0);
4971 }
4972
4973 static int
4974 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4975 caller_context_t *ct)
4976 {
4977 znode_t *zp, *xzp;
4978 zfsvfs_t *zfsvfs;
4979 zfs_dirlock_t *dl;
4980 int error;
4981
4982 switch (cmd) {
4983 case _PC_LINK_MAX:
4984 *valp = ULONG_MAX;
4985 return (0);
4986
4987 case _PC_FILESIZEBITS:
4988 *valp = 64;
4989 return (0);
4990
4991 case _PC_XATTR_EXISTS:
4992 zp = VTOZ(vp);
4993 zfsvfs = zp->z_zfsvfs;
4994 ZFS_ENTER(zfsvfs);
4995 ZFS_VERIFY_ZP(zp);
4996 *valp = 0;
4997 error = zfs_dirent_lock(&dl, zp, "", &xzp,
4998 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4999 if (error == 0) {
5000 zfs_dirent_unlock(dl);
5001 if (!zfs_dirempty(xzp))
5002 *valp = 1;
5003 VN_RELE(ZTOV(xzp));
5004 } else if (error == ENOENT) {
5005 /*
5006 * If there aren't extended attributes, it's the
5007 * same as having zero of them.
5008 */
5009 error = 0;
5010 }
5011 ZFS_EXIT(zfsvfs);
5012 return (error);
5013
5014 case _PC_SATTR_ENABLED:
5015 case _PC_SATTR_EXISTS:
5016 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5017 (vp->v_type == VREG || vp->v_type == VDIR);
5018 return (0);
5019
5020 case _PC_ACCESS_FILTERING:
5021 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
5022 vp->v_type == VDIR;
5023 return (0);
5024
5025 case _PC_ACL_ENABLED:
5026 *valp = _ACL_ACE_ENABLED;
5027 return (0);
5028
5029 case _PC_MIN_HOLE_SIZE:
5030 *valp = (ulong_t)SPA_MINBLOCKSIZE;
5031 return (0);
5032
5033 case _PC_TIMESTAMP_RESOLUTION:
5034 /* nanosecond timestamp resolution */
5035 *valp = 1L;
5036 return (0);
5037
5038 default:
5039 return (fs_pathconf(vp, cmd, valp, cr, ct));
5040 }
5041 }
5042
5043 /*ARGSUSED*/
5044 static int
5045 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5046 caller_context_t *ct)
5047 {
5048 znode_t *zp = VTOZ(vp);
5049 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5050 int error;
5051 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5052
5053 ZFS_ENTER(zfsvfs);
5054 ZFS_VERIFY_ZP(zp);
5055 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5056 ZFS_EXIT(zfsvfs);
5057
5058 return (error);
5059 }
5060
5061 /*ARGSUSED*/
5062 static int
5063 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5064 caller_context_t *ct)
5065 {
5066 znode_t *zp = VTOZ(vp);
5067 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5068 int error;
5069 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5070 zilog_t *zilog = zfsvfs->z_log;
5071
5072 ZFS_ENTER(zfsvfs);
5073 ZFS_VERIFY_ZP(zp);
5074
5075 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5076
5077 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5078 zil_commit(zilog, 0);
5079
5080 ZFS_EXIT(zfsvfs);
5081 return (error);
5082 }
5083
5084 /*
5085 * The smallest read we may consider to loan out an arcbuf.
5086 * This must be a power of 2.
5087 */
5088 int zcr_blksz_min = (1 << 10); /* 1K */
5089 /*
5090 * If set to less than the file block size, allow loaning out of an
5091 * arcbuf for a partial block read. This must be a power of 2.
5092 */
5093 int zcr_blksz_max = (1 << 17); /* 128K */
5094
5095 /*ARGSUSED*/
5096 static int
5097 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5098 caller_context_t *ct)
5099 {
5100 znode_t *zp = VTOZ(vp);
5101 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5102 int max_blksz = zfsvfs->z_max_blksz;
5103 uio_t *uio = &xuio->xu_uio;
5104 ssize_t size = uio->uio_resid;
5105 offset_t offset = uio->uio_loffset;
5106 int blksz;
5107 int fullblk, i;
5108 arc_buf_t *abuf;
5109 ssize_t maxsize;
5110 int preamble, postamble;
5111
5112 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5113 return (SET_ERROR(EINVAL));
5114
5115 ZFS_ENTER(zfsvfs);
5116 ZFS_VERIFY_ZP(zp);
5117 switch (ioflag) {
5118 case UIO_WRITE:
5119 /*
5120 * Loan out an arc_buf for write if write size is bigger than
5121 * max_blksz, and the file's block size is also max_blksz.
5122 */
5123 blksz = max_blksz;
5124 if (size < blksz || zp->z_blksz != blksz) {
5125 ZFS_EXIT(zfsvfs);
5126 return (SET_ERROR(EINVAL));
5127 }
5128 /*
5129 * Caller requests buffers for write before knowing where the
5130 * write offset might be (e.g. NFS TCP write).
5131 */
5132 if (offset == -1) {
5133 preamble = 0;
5134 } else {
5135 preamble = P2PHASE(offset, blksz);
5136 if (preamble) {
5137 preamble = blksz - preamble;
5138 size -= preamble;
5139 }
5140 }
5141
5142 postamble = P2PHASE(size, blksz);
5143 size -= postamble;
5144
5145 fullblk = size / blksz;
5146 (void) dmu_xuio_init(xuio,
5147 (preamble != 0) + fullblk + (postamble != 0));
5148 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5149 int, postamble, int,
5150 (preamble != 0) + fullblk + (postamble != 0));
5151
5152 /*
5153 * Have to fix iov base/len for partial buffers. They
5154 * currently represent full arc_buf's.
5155 */
5156 if (preamble) {
5157 /* data begins in the middle of the arc_buf */
5158 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5159 blksz);
5160 ASSERT(abuf);
5161 (void) dmu_xuio_add(xuio, abuf,
5162 blksz - preamble, preamble);
5163 }
5164
5165 for (i = 0; i < fullblk; i++) {
5166 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5167 blksz);
5168 ASSERT(abuf);
5169 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5170 }
5171
5172 if (postamble) {
5173 /* data ends in the middle of the arc_buf */
5174 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5175 blksz);
5176 ASSERT(abuf);
5177 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5178 }
5179 break;
5180 case UIO_READ:
5181 /*
5182 * Loan out an arc_buf for read if the read size is larger than
5183 * the current file block size. Block alignment is not
5184 * considered. Partial arc_buf will be loaned out for read.
5185 */
5186 blksz = zp->z_blksz;
5187 if (blksz < zcr_blksz_min)
5188 blksz = zcr_blksz_min;
5189 if (blksz > zcr_blksz_max)
5190 blksz = zcr_blksz_max;
5191 /* avoid potential complexity of dealing with it */
5192 if (blksz > max_blksz) {
5193 ZFS_EXIT(zfsvfs);
5194 return (SET_ERROR(EINVAL));
5195 }
5196
5197 maxsize = zp->z_size - uio->uio_loffset;
5198 if (size > maxsize)
5199 size = maxsize;
5200
5201 if (size < blksz || vn_has_cached_data(vp)) {
5202 ZFS_EXIT(zfsvfs);
5203 return (SET_ERROR(EINVAL));
5204 }
5205 break;
5206 default:
5207 ZFS_EXIT(zfsvfs);
5208 return (SET_ERROR(EINVAL));
5209 }
5210
5211 uio->uio_extflg = UIO_XUIO;
5212 XUIO_XUZC_RW(xuio) = ioflag;
5213 ZFS_EXIT(zfsvfs);
5214 return (0);
5215 }
5216
5217 /*ARGSUSED*/
5218 static int
5219 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5220 {
5221 int i;
5222 arc_buf_t *abuf;
5223 int ioflag = XUIO_XUZC_RW(xuio);
5224
5225 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5226
5227 i = dmu_xuio_cnt(xuio);
5228 while (i-- > 0) {
5229 abuf = dmu_xuio_arcbuf(xuio, i);
5230 /*
5231 * if abuf == NULL, it must be a write buffer
5232 * that has been returned in zfs_write().
5233 */
5234 if (abuf)
5235 dmu_return_arcbuf(abuf);
5236 ASSERT(abuf || ioflag == UIO_WRITE);
5237 }
5238
5239 dmu_xuio_fini(xuio);
5240 return (0);
5241 }
5242
5243 /*
5244 * Predeclare these here so that the compiler assumes that
5245 * this is an "old style" function declaration that does
5246 * not include arguments => we won't get type mismatch errors
5247 * in the initializations that follow.
5248 */
5249 static int zfs_inval();
5250 static int zfs_isdir();
5251
5252 static int
5253 zfs_inval()
5254 {
5255 return (SET_ERROR(EINVAL));
5256 }
5257
5258 static int
5259 zfs_isdir()
5260 {
5261 return (SET_ERROR(EISDIR));
5262 }
5263 /*
5264 * Directory vnode operations template
5265 */
5266 vnodeops_t *zfs_dvnodeops;
5267 const fs_operation_def_t zfs_dvnodeops_template[] = {
5268 VOPNAME_OPEN, { .vop_open = zfs_open },
5269 VOPNAME_CLOSE, { .vop_close = zfs_close },
5270 VOPNAME_READ, { .error = zfs_isdir },
5271 VOPNAME_WRITE, { .error = zfs_isdir },
5272 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5273 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5274 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5275 VOPNAME_ACCESS, { .vop_access = zfs_access },
5276 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5277 VOPNAME_CREATE, { .vop_create = zfs_create },
5278 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5279 VOPNAME_LINK, { .vop_link = zfs_link },
5280 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5281 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5282 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5283 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5284 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5285 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5286 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5287 VOPNAME_FID, { .vop_fid = zfs_fid },
5288 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5289 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5290 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5291 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5292 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5293 NULL, NULL
5294 };
5295
5296 /*
5297 * Regular file vnode operations template
5298 */
5299 vnodeops_t *zfs_fvnodeops;
5300 const fs_operation_def_t zfs_fvnodeops_template[] = {
5301 VOPNAME_OPEN, { .vop_open = zfs_open },
5302 VOPNAME_CLOSE, { .vop_close = zfs_close },
5303 VOPNAME_READ, { .vop_read = zfs_read },
5304 VOPNAME_WRITE, { .vop_write = zfs_write },
5305 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5306 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5307 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5308 VOPNAME_ACCESS, { .vop_access = zfs_access },
5309 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5310 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5311 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5312 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5313 VOPNAME_FID, { .vop_fid = zfs_fid },
5314 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5315 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5316 VOPNAME_SPACE, { .vop_space = zfs_space },
5317 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5318 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5319 VOPNAME_MAP, { .vop_map = zfs_map },
5320 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5321 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5322 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5323 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5324 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5325 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5326 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5327 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5328 NULL, NULL
5329 };
5330
5331 /*
5332 * Symbolic link vnode operations template
5333 */
5334 vnodeops_t *zfs_symvnodeops;
5335 const fs_operation_def_t zfs_symvnodeops_template[] = {
5336 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5337 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5338 VOPNAME_ACCESS, { .vop_access = zfs_access },
5339 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5340 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5341 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5342 VOPNAME_FID, { .vop_fid = zfs_fid },
5343 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5344 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5345 NULL, NULL
5346 };
5347
5348 /*
5349 * special share hidden files vnode operations template
5350 */
5351 vnodeops_t *zfs_sharevnodeops;
5352 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5353 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5354 VOPNAME_ACCESS, { .vop_access = zfs_access },
5355 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5356 VOPNAME_FID, { .vop_fid = zfs_fid },
5357 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5358 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5359 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5360 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5361 NULL, NULL
5362 };
5363
5364 /*
5365 * Extended attribute directory vnode operations template
5366 *
5367 * This template is identical to the directory vnodes
5368 * operation template except for restricted operations:
5369 * VOP_MKDIR()
5370 * VOP_SYMLINK()
5371 *
5372 * Note that there are other restrictions embedded in:
5373 * zfs_create() - restrict type to VREG
5374 * zfs_link() - no links into/out of attribute space
5375 * zfs_rename() - no moves into/out of attribute space
5376 */
5377 vnodeops_t *zfs_xdvnodeops;
5378 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5379 VOPNAME_OPEN, { .vop_open = zfs_open },
5380 VOPNAME_CLOSE, { .vop_close = zfs_close },
5381 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5382 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5383 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5384 VOPNAME_ACCESS, { .vop_access = zfs_access },
5385 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5386 VOPNAME_CREATE, { .vop_create = zfs_create },
5387 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5388 VOPNAME_LINK, { .vop_link = zfs_link },
5389 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5390 VOPNAME_MKDIR, { .error = zfs_inval },
5391 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5392 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5393 VOPNAME_SYMLINK, { .error = zfs_inval },
5394 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5395 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5396 VOPNAME_FID, { .vop_fid = zfs_fid },
5397 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5398 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5399 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5400 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5401 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5402 NULL, NULL
5403 };
5404
5405 /*
5406 * Error vnode operations template
5407 */
5408 vnodeops_t *zfs_evnodeops;
5409 const fs_operation_def_t zfs_evnodeops_template[] = {
5410 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5411 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5412 NULL, NULL
5413 };