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