Print this page
Possibility to physically reserve space without writing leaf blocks
| 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, 2014 by Delphix. All rights reserved.
24 24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
25 25 */
26 26
27 27 /* Portions Copyright 2007 Jeremy Teo */
28 28 /* Portions Copyright 2010 Robert Milkowski */
29 29
30 30 #include <sys/types.h>
31 31 #include <sys/param.h>
32 32 #include <sys/time.h>
33 33 #include <sys/systm.h>
34 34 #include <sys/sysmacros.h>
35 35 #include <sys/resource.h>
36 36 #include <sys/vfs.h>
37 37 #include <sys/vfs_opreg.h>
38 38 #include <sys/vnode.h>
39 39 #include <sys/file.h>
40 40 #include <sys/stat.h>
41 41 #include <sys/kmem.h>
42 42 #include <sys/taskq.h>
43 43 #include <sys/uio.h>
44 44 #include <sys/vmsystm.h>
45 45 #include <sys/atomic.h>
46 46 #include <sys/vm.h>
47 47 #include <vm/seg_vn.h>
48 48 #include <vm/pvn.h>
49 49 #include <vm/as.h>
50 50 #include <vm/kpm.h>
51 51 #include <vm/seg_kpm.h>
52 52 #include <sys/mman.h>
53 53 #include <sys/pathname.h>
54 54 #include <sys/cmn_err.h>
55 55 #include <sys/errno.h>
56 56 #include <sys/unistd.h>
57 57 #include <sys/zfs_dir.h>
58 58 #include <sys/zfs_acl.h>
59 59 #include <sys/zfs_ioctl.h>
60 60 #include <sys/fs/zfs.h>
61 61 #include <sys/dmu.h>
62 62 #include <sys/dmu_objset.h>
63 63 #include <sys/spa.h>
64 64 #include <sys/txg.h>
65 65 #include <sys/dbuf.h>
66 66 #include <sys/zap.h>
|
↓ open down ↓ |
66 lines elided |
↑ open up ↑ |
67 67 #include <sys/sa.h>
68 68 #include <sys/dirent.h>
69 69 #include <sys/policy.h>
70 70 #include <sys/sunddi.h>
71 71 #include <sys/filio.h>
72 72 #include <sys/sid.h>
73 73 #include "fs/fs_subr.h"
74 74 #include <sys/zfs_ctldir.h>
75 75 #include <sys/zfs_fuid.h>
76 76 #include <sys/zfs_sa.h>
77 +#include <sys/zfeature.h>
77 78 #include <sys/dnlc.h>
78 79 #include <sys/zfs_rlock.h>
79 80 #include <sys/extdirent.h>
80 81 #include <sys/kidmap.h>
81 82 #include <sys/cred.h>
82 83 #include <sys/attr.h>
83 84
84 85 /*
85 86 * Programming rules.
86 87 *
87 88 * Each vnode op performs some logical unit of work. To do this, the ZPL must
88 89 * properly lock its in-core state, create a DMU transaction, do the work,
89 90 * record this work in the intent log (ZIL), commit the DMU transaction,
90 91 * and wait for the intent log to commit if it is a synchronous operation.
91 92 * Moreover, the vnode ops must work in both normal and log replay context.
92 93 * The ordering of events is important to avoid deadlocks and references
93 94 * to freed memory. The example below illustrates the following Big Rules:
94 95 *
95 96 * (1) A check must be made in each zfs thread for a mounted file system.
96 97 * This is done avoiding races using ZFS_ENTER(zfsvfs).
97 98 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
98 99 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
99 100 * can return EIO from the calling function.
100 101 *
101 102 * (2) VN_RELE() should always be the last thing except for zil_commit()
102 103 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
103 104 * First, if it's the last reference, the vnode/znode
104 105 * can be freed, so the zp may point to freed memory. Second, the last
105 106 * reference will call zfs_zinactive(), which may induce a lot of work --
106 107 * pushing cached pages (which acquires range locks) and syncing out
107 108 * cached atime changes. Third, zfs_zinactive() may require a new tx,
108 109 * which could deadlock the system if you were already holding one.
109 110 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
110 111 *
111 112 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
112 113 * as they can span dmu_tx_assign() calls.
113 114 *
114 115 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
115 116 * dmu_tx_assign(). This is critical because we don't want to block
116 117 * while holding locks.
117 118 *
118 119 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
119 120 * reduces lock contention and CPU usage when we must wait (note that if
120 121 * throughput is constrained by the storage, nearly every transaction
121 122 * must wait).
122 123 *
123 124 * Note, in particular, that if a lock is sometimes acquired before
124 125 * the tx assigns, and sometimes after (e.g. z_lock), then failing
125 126 * to use a non-blocking assign can deadlock the system. The scenario:
126 127 *
127 128 * Thread A has grabbed a lock before calling dmu_tx_assign().
128 129 * Thread B is in an already-assigned tx, and blocks for this lock.
129 130 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
130 131 * forever, because the previous txg can't quiesce until B's tx commits.
131 132 *
132 133 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
133 134 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
134 135 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
135 136 * to indicate that this operation has already called dmu_tx_wait().
136 137 * This will ensure that we don't retry forever, waiting a short bit
137 138 * each time.
138 139 *
139 140 * (5) If the operation succeeded, generate the intent log entry for it
140 141 * before dropping locks. This ensures that the ordering of events
141 142 * in the intent log matches the order in which they actually occurred.
142 143 * During ZIL replay the zfs_log_* functions will update the sequence
143 144 * number to indicate the zil transaction has replayed.
144 145 *
145 146 * (6) At the end of each vnode op, the DMU tx must always commit,
146 147 * regardless of whether there were any errors.
147 148 *
148 149 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
149 150 * to ensure that synchronous semantics are provided when necessary.
150 151 *
151 152 * In general, this is how things should be ordered in each vnode op:
152 153 *
153 154 * ZFS_ENTER(zfsvfs); // exit if unmounted
154 155 * top:
155 156 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
156 157 * rw_enter(...); // grab any other locks you need
157 158 * tx = dmu_tx_create(...); // get DMU tx
158 159 * dmu_tx_hold_*(); // hold each object you might modify
159 160 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
160 161 * if (error) {
161 162 * rw_exit(...); // drop locks
162 163 * zfs_dirent_unlock(dl); // unlock directory entry
163 164 * VN_RELE(...); // release held vnodes
164 165 * if (error == ERESTART) {
165 166 * waited = B_TRUE;
166 167 * dmu_tx_wait(tx);
167 168 * dmu_tx_abort(tx);
168 169 * goto top;
169 170 * }
170 171 * dmu_tx_abort(tx); // abort DMU tx
171 172 * ZFS_EXIT(zfsvfs); // finished in zfs
172 173 * return (error); // really out of space
173 174 * }
174 175 * error = do_real_work(); // do whatever this VOP does
175 176 * if (error == 0)
176 177 * zfs_log_*(...); // on success, make ZIL entry
177 178 * dmu_tx_commit(tx); // commit DMU tx -- error or not
178 179 * rw_exit(...); // drop locks
179 180 * zfs_dirent_unlock(dl); // unlock directory entry
180 181 * VN_RELE(...); // release held vnodes
181 182 * zil_commit(zilog, foid); // synchronous when necessary
182 183 * ZFS_EXIT(zfsvfs); // finished in zfs
183 184 * return (error); // done, report error
184 185 */
185 186
186 187 /* ARGSUSED */
187 188 static int
188 189 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
189 190 {
190 191 znode_t *zp = VTOZ(*vpp);
191 192 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
192 193
193 194 ZFS_ENTER(zfsvfs);
194 195 ZFS_VERIFY_ZP(zp);
195 196
196 197 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
197 198 ((flag & FAPPEND) == 0)) {
198 199 ZFS_EXIT(zfsvfs);
199 200 return (SET_ERROR(EPERM));
200 201 }
201 202
202 203 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
203 204 ZTOV(zp)->v_type == VREG &&
204 205 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
205 206 if (fs_vscan(*vpp, cr, 0) != 0) {
206 207 ZFS_EXIT(zfsvfs);
207 208 return (SET_ERROR(EACCES));
208 209 }
209 210 }
210 211
211 212 /* Keep a count of the synchronous opens in the znode */
212 213 if (flag & (FSYNC | FDSYNC))
213 214 atomic_inc_32(&zp->z_sync_cnt);
214 215
215 216 ZFS_EXIT(zfsvfs);
216 217 return (0);
217 218 }
218 219
219 220 /* ARGSUSED */
220 221 static int
221 222 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
222 223 caller_context_t *ct)
223 224 {
224 225 znode_t *zp = VTOZ(vp);
225 226 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
226 227
227 228 /*
228 229 * Clean up any locks held by this process on the vp.
229 230 */
230 231 cleanlocks(vp, ddi_get_pid(), 0);
231 232 cleanshares(vp, ddi_get_pid());
232 233
233 234 ZFS_ENTER(zfsvfs);
234 235 ZFS_VERIFY_ZP(zp);
235 236
236 237 /* Decrement the synchronous opens in the znode */
237 238 if ((flag & (FSYNC | FDSYNC)) && (count == 1))
238 239 atomic_dec_32(&zp->z_sync_cnt);
239 240
240 241 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
241 242 ZTOV(zp)->v_type == VREG &&
242 243 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
243 244 VERIFY(fs_vscan(vp, cr, 1) == 0);
244 245
245 246 ZFS_EXIT(zfsvfs);
246 247 return (0);
247 248 }
248 249
249 250 /*
250 251 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
251 252 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
252 253 */
253 254 static int
254 255 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
255 256 {
256 257 znode_t *zp = VTOZ(vp);
257 258 uint64_t noff = (uint64_t)*off; /* new offset */
258 259 uint64_t file_sz;
259 260 int error;
260 261 boolean_t hole;
261 262
262 263 file_sz = zp->z_size;
263 264 if (noff >= file_sz) {
264 265 return (SET_ERROR(ENXIO));
265 266 }
266 267
267 268 if (cmd == _FIO_SEEK_HOLE)
268 269 hole = B_TRUE;
269 270 else
270 271 hole = B_FALSE;
271 272
272 273 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
273 274
274 275 if (error == ESRCH)
275 276 return (SET_ERROR(ENXIO));
276 277
277 278 /*
278 279 * We could find a hole that begins after the logical end-of-file,
279 280 * because dmu_offset_next() only works on whole blocks. If the
280 281 * EOF falls mid-block, then indicate that the "virtual hole"
281 282 * at the end of the file begins at the logical EOF, rather than
282 283 * at the end of the last block.
283 284 */
284 285 if (noff > file_sz) {
|
↓ open down ↓ |
198 lines elided |
↑ open up ↑ |
285 286 ASSERT(hole);
286 287 noff = file_sz;
287 288 }
288 289
289 290 if (noff < *off)
290 291 return (error);
291 292 *off = noff;
292 293 return (error);
293 294 }
294 295
296 +
297 +static int zfs_zero_write(vnode_t *vp, uint64_t size, cred_t *cr,
298 + caller_context_t *ct);
299 +
295 300 /* ARGSUSED */
296 301 static int
297 302 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
298 303 int *rvalp, caller_context_t *ct)
299 304 {
300 305 offset_t off;
301 306 int error;
302 307 zfsvfs_t *zfsvfs;
303 308 znode_t *zp;
309 + uint64_t size;
304 310
305 311 switch (com) {
306 312 case _FIOFFS:
307 313 return (zfs_sync(vp->v_vfsp, 0, cred));
308 314
309 315 /*
310 316 * The following two ioctls are used by bfu. Faking out,
311 317 * necessary to avoid bfu errors.
312 318 */
313 319 case _FIOGDIO:
314 320 case _FIOSDIO:
315 321 return (0);
316 322
317 323 case _FIO_SEEK_DATA:
318 324 case _FIO_SEEK_HOLE:
319 325 if (ddi_copyin((void *)data, &off, sizeof (off), flag))
320 326 return (SET_ERROR(EFAULT));
321 327
322 328 zp = VTOZ(vp);
323 329 zfsvfs = zp->z_zfsvfs;
324 330 ZFS_ENTER(zfsvfs);
|
↓ open down ↓ |
11 lines elided |
↑ open up ↑ |
325 331 ZFS_VERIFY_ZP(zp);
326 332
327 333 /* offset parameter is in/out */
328 334 error = zfs_holey(vp, com, &off);
329 335 ZFS_EXIT(zfsvfs);
330 336 if (error)
331 337 return (error);
332 338 if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
333 339 return (SET_ERROR(EFAULT));
334 340 return (0);
341 + case _FIO_RESERVE_SPACE:
342 + if (ddi_copyin((void *)data, &size, sizeof (size), flag))
343 + return (EFAULT);
344 + error = zfs_zero_write(vp, size, cred, ct);
345 + return (error);
335 346 }
336 347 return (SET_ERROR(ENOTTY));
337 348 }
338 349
339 350 /*
340 351 * Utility functions to map and unmap a single physical page. These
341 352 * are used to manage the mappable copies of ZFS file data, and therefore
342 353 * do not update ref/mod bits.
343 354 */
344 355 caddr_t
345 356 zfs_map_page(page_t *pp, enum seg_rw rw)
346 357 {
347 358 if (kpm_enable)
348 359 return (hat_kpm_mapin(pp, 0));
349 360 ASSERT(rw == S_READ || rw == S_WRITE);
350 361 return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
351 362 (caddr_t)-1));
352 363 }
353 364
354 365 void
355 366 zfs_unmap_page(page_t *pp, caddr_t addr)
356 367 {
357 368 if (kpm_enable) {
358 369 hat_kpm_mapout(pp, 0, addr);
359 370 } else {
360 371 ppmapout(addr);
361 372 }
362 373 }
363 374
364 375 /*
365 376 * When a file is memory mapped, we must keep the IO data synchronized
366 377 * between the DMU cache and the memory mapped pages. What this means:
367 378 *
368 379 * On Write: If we find a memory mapped page, we write to *both*
369 380 * the page and the dmu buffer.
370 381 */
371 382 static void
372 383 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
373 384 {
374 385 int64_t off;
375 386
376 387 off = start & PAGEOFFSET;
377 388 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
378 389 page_t *pp;
379 390 uint64_t nbytes = MIN(PAGESIZE - off, len);
380 391
381 392 if (pp = page_lookup(vp, start, SE_SHARED)) {
382 393 caddr_t va;
383 394
384 395 va = zfs_map_page(pp, S_WRITE);
385 396 (void) dmu_read(os, oid, start+off, nbytes, va+off,
386 397 DMU_READ_PREFETCH);
387 398 zfs_unmap_page(pp, va);
388 399 page_unlock(pp);
389 400 }
390 401 len -= nbytes;
391 402 off = 0;
392 403 }
393 404 }
394 405
395 406 /*
396 407 * When a file is memory mapped, we must keep the IO data synchronized
397 408 * between the DMU cache and the memory mapped pages. What this means:
398 409 *
399 410 * On Read: We "read" preferentially from memory mapped pages,
400 411 * else we default from the dmu buffer.
401 412 *
402 413 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
403 414 * the file is memory mapped.
404 415 */
405 416 static int
406 417 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
407 418 {
408 419 znode_t *zp = VTOZ(vp);
409 420 int64_t start, off;
410 421 int len = nbytes;
411 422 int error = 0;
412 423
413 424 start = uio->uio_loffset;
414 425 off = start & PAGEOFFSET;
415 426 for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
416 427 page_t *pp;
417 428 uint64_t bytes = MIN(PAGESIZE - off, len);
418 429
419 430 if (pp = page_lookup(vp, start, SE_SHARED)) {
420 431 caddr_t va;
421 432
422 433 va = zfs_map_page(pp, S_READ);
423 434 error = uiomove(va + off, bytes, UIO_READ, uio);
424 435 zfs_unmap_page(pp, va);
425 436 page_unlock(pp);
426 437 } else {
427 438 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
428 439 uio, bytes);
429 440 }
430 441 len -= bytes;
431 442 off = 0;
432 443 if (error)
433 444 break;
434 445 }
435 446 return (error);
436 447 }
437 448
438 449 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
439 450
440 451 /*
441 452 * Read bytes from specified file into supplied buffer.
442 453 *
443 454 * IN: vp - vnode of file to be read from.
444 455 * uio - structure supplying read location, range info,
445 456 * and return buffer.
446 457 * ioflag - SYNC flags; used to provide FRSYNC semantics.
447 458 * cr - credentials of caller.
448 459 * ct - caller context
449 460 *
450 461 * OUT: uio - updated offset and range, buffer filled.
451 462 *
452 463 * RETURN: 0 on success, error code on failure.
453 464 *
454 465 * Side Effects:
455 466 * vp - atime updated if byte count > 0
456 467 */
457 468 /* ARGSUSED */
458 469 static int
459 470 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
460 471 {
461 472 znode_t *zp = VTOZ(vp);
462 473 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
463 474 ssize_t n, nbytes;
464 475 int error = 0;
465 476 rl_t *rl;
466 477 xuio_t *xuio = NULL;
467 478
468 479 ZFS_ENTER(zfsvfs);
469 480 ZFS_VERIFY_ZP(zp);
470 481
471 482 if (zp->z_pflags & ZFS_AV_QUARANTINED) {
472 483 ZFS_EXIT(zfsvfs);
473 484 return (SET_ERROR(EACCES));
474 485 }
475 486
476 487 /*
477 488 * Validate file offset
478 489 */
479 490 if (uio->uio_loffset < (offset_t)0) {
480 491 ZFS_EXIT(zfsvfs);
481 492 return (SET_ERROR(EINVAL));
482 493 }
483 494
484 495 /*
485 496 * Fasttrack empty reads
486 497 */
487 498 if (uio->uio_resid == 0) {
488 499 ZFS_EXIT(zfsvfs);
489 500 return (0);
490 501 }
491 502
492 503 /*
493 504 * Check for mandatory locks
494 505 */
495 506 if (MANDMODE(zp->z_mode)) {
496 507 if (error = chklock(vp, FREAD,
497 508 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
498 509 ZFS_EXIT(zfsvfs);
499 510 return (error);
500 511 }
501 512 }
502 513
503 514 /*
504 515 * If we're in FRSYNC mode, sync out this znode before reading it.
505 516 */
506 517 if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
507 518 zil_commit(zfsvfs->z_log, zp->z_id);
508 519
509 520 /*
510 521 * Lock the range against changes.
511 522 */
512 523 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
513 524
514 525 /*
515 526 * If we are reading past end-of-file we can skip
516 527 * to the end; but we might still need to set atime.
517 528 */
518 529 if (uio->uio_loffset >= zp->z_size) {
519 530 error = 0;
520 531 goto out;
521 532 }
522 533
523 534 ASSERT(uio->uio_loffset < zp->z_size);
524 535 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
525 536
526 537 if ((uio->uio_extflg == UIO_XUIO) &&
527 538 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
528 539 int nblk;
529 540 int blksz = zp->z_blksz;
530 541 uint64_t offset = uio->uio_loffset;
531 542
532 543 xuio = (xuio_t *)uio;
533 544 if ((ISP2(blksz))) {
534 545 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
535 546 blksz)) / blksz;
536 547 } else {
537 548 ASSERT(offset + n <= blksz);
538 549 nblk = 1;
539 550 }
540 551 (void) dmu_xuio_init(xuio, nblk);
541 552
542 553 if (vn_has_cached_data(vp)) {
543 554 /*
544 555 * For simplicity, we always allocate a full buffer
545 556 * even if we only expect to read a portion of a block.
546 557 */
547 558 while (--nblk >= 0) {
548 559 (void) dmu_xuio_add(xuio,
549 560 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
550 561 blksz), 0, blksz);
551 562 }
552 563 }
553 564 }
554 565
555 566 while (n > 0) {
556 567 nbytes = MIN(n, zfs_read_chunk_size -
557 568 P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
558 569
559 570 if (vn_has_cached_data(vp)) {
560 571 error = mappedread(vp, nbytes, uio);
561 572 } else {
562 573 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
563 574 uio, nbytes);
564 575 }
565 576 if (error) {
566 577 /* convert checksum errors into IO errors */
567 578 if (error == ECKSUM)
568 579 error = SET_ERROR(EIO);
569 580 break;
570 581 }
571 582
572 583 n -= nbytes;
573 584 }
574 585 out:
575 586 zfs_range_unlock(rl);
576 587
577 588 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
578 589 ZFS_EXIT(zfsvfs);
579 590 return (error);
580 591 }
581 592
582 593 /*
583 594 * Write the bytes to a file.
584 595 *
585 596 * IN: vp - vnode of file to be written to.
586 597 * uio - structure supplying write location, range info,
587 598 * and data buffer.
588 599 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
589 600 * set if in append mode.
590 601 * cr - credentials of caller.
591 602 * ct - caller context (NFS/CIFS fem monitor only)
592 603 *
593 604 * OUT: uio - updated offset and range.
594 605 *
595 606 * RETURN: 0 on success, error code on failure.
596 607 *
597 608 * Timestamps:
598 609 * vp - ctime|mtime updated if byte count > 0
599 610 */
600 611
601 612 /* ARGSUSED */
602 613 static int
603 614 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
604 615 {
605 616 znode_t *zp = VTOZ(vp);
606 617 rlim64_t limit = uio->uio_llimit;
607 618 ssize_t start_resid = uio->uio_resid;
608 619 ssize_t tx_bytes;
609 620 uint64_t end_size;
610 621 dmu_tx_t *tx;
611 622 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
612 623 zilog_t *zilog;
613 624 offset_t woff;
614 625 ssize_t n, nbytes;
615 626 rl_t *rl;
616 627 int max_blksz = zfsvfs->z_max_blksz;
617 628 int error = 0;
618 629 arc_buf_t *abuf;
619 630 iovec_t *aiov = NULL;
620 631 xuio_t *xuio = NULL;
621 632 int i_iov = 0;
622 633 int iovcnt = uio->uio_iovcnt;
623 634 iovec_t *iovp = uio->uio_iov;
624 635 int write_eof;
625 636 int count = 0;
626 637 sa_bulk_attr_t bulk[4];
627 638 uint64_t mtime[2], ctime[2];
628 639
629 640 /*
630 641 * Fasttrack empty write
631 642 */
632 643 n = start_resid;
633 644 if (n == 0)
634 645 return (0);
635 646
636 647 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
637 648 limit = MAXOFFSET_T;
638 649
639 650 ZFS_ENTER(zfsvfs);
640 651 ZFS_VERIFY_ZP(zp);
641 652
642 653 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
643 654 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
644 655 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
645 656 &zp->z_size, 8);
646 657 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
647 658 &zp->z_pflags, 8);
648 659
649 660 /*
650 661 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
651 662 * callers might not be able to detect properly that we are read-only,
652 663 * so check it explicitly here.
653 664 */
654 665 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
655 666 ZFS_EXIT(zfsvfs);
656 667 return (SET_ERROR(EROFS));
657 668 }
658 669
659 670 /*
660 671 * If immutable or not appending then return EPERM
661 672 */
662 673 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
663 674 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
664 675 (uio->uio_loffset < zp->z_size))) {
665 676 ZFS_EXIT(zfsvfs);
666 677 return (SET_ERROR(EPERM));
667 678 }
668 679
669 680 zilog = zfsvfs->z_log;
670 681
671 682 /*
672 683 * Validate file offset
673 684 */
674 685 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
675 686 if (woff < 0) {
676 687 ZFS_EXIT(zfsvfs);
677 688 return (SET_ERROR(EINVAL));
678 689 }
679 690
680 691 /*
681 692 * Check for mandatory locks before calling zfs_range_lock()
682 693 * in order to prevent a deadlock with locks set via fcntl().
683 694 */
684 695 if (MANDMODE((mode_t)zp->z_mode) &&
685 696 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
686 697 ZFS_EXIT(zfsvfs);
687 698 return (error);
688 699 }
689 700
690 701 /*
691 702 * Pre-fault the pages to ensure slow (eg NFS) pages
692 703 * don't hold up txg.
693 704 * Skip this if uio contains loaned arc_buf.
694 705 */
695 706 if ((uio->uio_extflg == UIO_XUIO) &&
696 707 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
697 708 xuio = (xuio_t *)uio;
698 709 else
699 710 uio_prefaultpages(MIN(n, max_blksz), uio);
700 711
701 712 /*
702 713 * If in append mode, set the io offset pointer to eof.
703 714 */
704 715 if (ioflag & FAPPEND) {
705 716 /*
706 717 * Obtain an appending range lock to guarantee file append
707 718 * semantics. We reset the write offset once we have the lock.
708 719 */
709 720 rl = zfs_range_lock(zp, 0, n, RL_APPEND);
710 721 woff = rl->r_off;
711 722 if (rl->r_len == UINT64_MAX) {
712 723 /*
713 724 * We overlocked the file because this write will cause
714 725 * the file block size to increase.
715 726 * Note that zp_size cannot change with this lock held.
716 727 */
717 728 woff = zp->z_size;
718 729 }
719 730 uio->uio_loffset = woff;
720 731 } else {
721 732 /*
722 733 * Note that if the file block size will change as a result of
723 734 * this write, then this range lock will lock the entire file
724 735 * so that we can re-write the block safely.
725 736 */
726 737 rl = zfs_range_lock(zp, woff, n, RL_WRITER);
727 738 }
728 739
729 740 if (woff >= limit) {
730 741 zfs_range_unlock(rl);
731 742 ZFS_EXIT(zfsvfs);
732 743 return (SET_ERROR(EFBIG));
733 744 }
734 745
735 746 if ((woff + n) > limit || woff > (limit - n))
736 747 n = limit - woff;
737 748
738 749 /* Will this write extend the file length? */
739 750 write_eof = (woff + n > zp->z_size);
740 751
741 752 end_size = MAX(zp->z_size, woff + n);
742 753
743 754 /*
744 755 * Write the file in reasonable size chunks. Each chunk is written
745 756 * in a separate transaction; this keeps the intent log records small
746 757 * and allows us to do more fine-grained space accounting.
747 758 */
748 759 while (n > 0) {
749 760 abuf = NULL;
750 761 woff = uio->uio_loffset;
751 762 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
752 763 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
753 764 if (abuf != NULL)
754 765 dmu_return_arcbuf(abuf);
755 766 error = SET_ERROR(EDQUOT);
756 767 break;
757 768 }
758 769
759 770 if (xuio && abuf == NULL) {
760 771 ASSERT(i_iov < iovcnt);
761 772 aiov = &iovp[i_iov];
762 773 abuf = dmu_xuio_arcbuf(xuio, i_iov);
763 774 dmu_xuio_clear(xuio, i_iov);
764 775 DTRACE_PROBE3(zfs_cp_write, int, i_iov,
765 776 iovec_t *, aiov, arc_buf_t *, abuf);
766 777 ASSERT((aiov->iov_base == abuf->b_data) ||
767 778 ((char *)aiov->iov_base - (char *)abuf->b_data +
768 779 aiov->iov_len == arc_buf_size(abuf)));
769 780 i_iov++;
770 781 } else if (abuf == NULL && n >= max_blksz &&
771 782 woff >= zp->z_size &&
772 783 P2PHASE(woff, max_blksz) == 0 &&
773 784 zp->z_blksz == max_blksz) {
774 785 /*
775 786 * This write covers a full block. "Borrow" a buffer
776 787 * from the dmu so that we can fill it before we enter
777 788 * a transaction. This avoids the possibility of
778 789 * holding up the transaction if the data copy hangs
779 790 * up on a pagefault (e.g., from an NFS server mapping).
780 791 */
781 792 size_t cbytes;
782 793
783 794 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
784 795 max_blksz);
785 796 ASSERT(abuf != NULL);
786 797 ASSERT(arc_buf_size(abuf) == max_blksz);
787 798 if (error = uiocopy(abuf->b_data, max_blksz,
788 799 UIO_WRITE, uio, &cbytes)) {
789 800 dmu_return_arcbuf(abuf);
790 801 break;
791 802 }
792 803 ASSERT(cbytes == max_blksz);
793 804 }
794 805
795 806 /*
796 807 * Start a transaction.
797 808 */
798 809 tx = dmu_tx_create(zfsvfs->z_os);
799 810 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
800 811 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
801 812 zfs_sa_upgrade_txholds(tx, zp);
802 813 error = dmu_tx_assign(tx, TXG_WAIT);
803 814 if (error) {
804 815 dmu_tx_abort(tx);
805 816 if (abuf != NULL)
806 817 dmu_return_arcbuf(abuf);
807 818 break;
808 819 }
809 820
810 821 /*
811 822 * If zfs_range_lock() over-locked we grow the blocksize
812 823 * and then reduce the lock range. This will only happen
813 824 * on the first iteration since zfs_range_reduce() will
814 825 * shrink down r_len to the appropriate size.
815 826 */
816 827 if (rl->r_len == UINT64_MAX) {
817 828 uint64_t new_blksz;
818 829
819 830 if (zp->z_blksz > max_blksz) {
820 831 ASSERT(!ISP2(zp->z_blksz));
821 832 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
822 833 } else {
823 834 new_blksz = MIN(end_size, max_blksz);
824 835 }
825 836 zfs_grow_blocksize(zp, new_blksz, tx);
826 837 zfs_range_reduce(rl, woff, n);
827 838 }
828 839
829 840 /*
830 841 * XXX - should we really limit each write to z_max_blksz?
831 842 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
832 843 */
833 844 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
834 845
835 846 if (abuf == NULL) {
836 847 tx_bytes = uio->uio_resid;
837 848 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
838 849 uio, nbytes, tx);
839 850 tx_bytes -= uio->uio_resid;
840 851 } else {
841 852 tx_bytes = nbytes;
842 853 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
843 854 /*
844 855 * If this is not a full block write, but we are
845 856 * extending the file past EOF and this data starts
846 857 * block-aligned, use assign_arcbuf(). Otherwise,
847 858 * write via dmu_write().
848 859 */
849 860 if (tx_bytes < max_blksz && (!write_eof ||
850 861 aiov->iov_base != abuf->b_data)) {
851 862 ASSERT(xuio);
852 863 dmu_write(zfsvfs->z_os, zp->z_id, woff,
853 864 aiov->iov_len, aiov->iov_base, tx);
854 865 dmu_return_arcbuf(abuf);
855 866 xuio_stat_wbuf_copied();
856 867 } else {
857 868 ASSERT(xuio || tx_bytes == max_blksz);
858 869 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
859 870 woff, abuf, tx);
860 871 }
861 872 ASSERT(tx_bytes <= uio->uio_resid);
862 873 uioskip(uio, tx_bytes);
863 874 }
864 875 if (tx_bytes && vn_has_cached_data(vp)) {
865 876 update_pages(vp, woff,
866 877 tx_bytes, zfsvfs->z_os, zp->z_id);
867 878 }
868 879
869 880 /*
870 881 * If we made no progress, we're done. If we made even
871 882 * partial progress, update the znode and ZIL accordingly.
872 883 */
873 884 if (tx_bytes == 0) {
874 885 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
875 886 (void *)&zp->z_size, sizeof (uint64_t), tx);
876 887 dmu_tx_commit(tx);
877 888 ASSERT(error != 0);
878 889 break;
879 890 }
880 891
881 892 /*
882 893 * Clear Set-UID/Set-GID bits on successful write if not
883 894 * privileged and at least one of the excute bits is set.
884 895 *
885 896 * It would be nice to to this after all writes have
886 897 * been done, but that would still expose the ISUID/ISGID
887 898 * to another app after the partial write is committed.
888 899 *
889 900 * Note: we don't call zfs_fuid_map_id() here because
890 901 * user 0 is not an ephemeral uid.
891 902 */
892 903 mutex_enter(&zp->z_acl_lock);
893 904 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
894 905 (S_IXUSR >> 6))) != 0 &&
895 906 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
896 907 secpolicy_vnode_setid_retain(cr,
897 908 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
898 909 uint64_t newmode;
899 910 zp->z_mode &= ~(S_ISUID | S_ISGID);
900 911 newmode = zp->z_mode;
901 912 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
902 913 (void *)&newmode, sizeof (uint64_t), tx);
903 914 }
904 915 mutex_exit(&zp->z_acl_lock);
905 916
906 917 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
907 918 B_TRUE);
908 919
909 920 /*
910 921 * Update the file size (zp_size) if it has changed;
911 922 * account for possible concurrent updates.
912 923 */
913 924 while ((end_size = zp->z_size) < uio->uio_loffset) {
914 925 (void) atomic_cas_64(&zp->z_size, end_size,
915 926 uio->uio_loffset);
916 927 ASSERT(error == 0);
917 928 }
918 929 /*
919 930 * If we are replaying and eof is non zero then force
920 931 * the file size to the specified eof. Note, there's no
921 932 * concurrency during replay.
922 933 */
923 934 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
924 935 zp->z_size = zfsvfs->z_replay_eof;
925 936
926 937 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
927 938
928 939 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
929 940 dmu_tx_commit(tx);
930 941
931 942 if (error != 0)
932 943 break;
933 944 ASSERT(tx_bytes == nbytes);
934 945 n -= nbytes;
935 946
936 947 if (!xuio && n > 0)
937 948 uio_prefaultpages(MIN(n, max_blksz), uio);
938 949 }
939 950
940 951 zfs_range_unlock(rl);
941 952
942 953 /*
943 954 * If we're in replay mode, or we made no progress, return error.
944 955 * Otherwise, it's at least a partial write, so it's successful.
945 956 */
946 957 if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
947 958 ZFS_EXIT(zfsvfs);
948 959 return (error);
|
↓ open down ↓ |
604 lines elided |
↑ open up ↑ |
949 960 }
950 961
951 962 if (ioflag & (FSYNC | FDSYNC) ||
952 963 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
953 964 zil_commit(zilog, zp->z_id);
954 965
955 966 ZFS_EXIT(zfsvfs);
956 967 return (0);
957 968 }
958 969
970 +#define ZFS_RESERVE_CHUNK (2 * 1024 * 1024)
971 +/* ARGSUSED */
972 +static int
973 +zfs_zero_write(vnode_t *vp, uint64_t size, cred_t *cr, caller_context_t *ct)
974 +{
975 + znode_t *zp = VTOZ(vp);
976 + zfsvfs_t *zfsvfs = zp->z_zfsvfs;
977 + int count = 0;
978 + sa_bulk_attr_t bulk[4];
979 + uint64_t mtime[2], ctime[2];
980 + rl_t *rl;
981 + int error = 0;
982 + dmu_tx_t *tx = NULL;
983 + uint64_t end_size;
984 + uint64_t pos = 0;
985 +
986 + if (zp->z_size > 0)
987 + return (EFBIG);
988 + if (size == 0)
989 + return (0);
990 +
991 + ZFS_ENTER(zfsvfs);
992 + ZFS_VERIFY_ZP(zp);
993 +
994 + if (!spa_feature_is_enabled(zfsvfs->z_os->os_spa,
995 + SPA_FEATURE_SPACE_RESERVATION))
996 + {
997 + ZFS_EXIT(zfsvfs);
998 + return (ENOTSUP);
999 + }
1000 +
1001 + SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
1002 + SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
1003 + SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1004 + &zp->z_size, 8);
1005 + SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1006 + &zp->z_pflags, 8);
1007 +
1008 + /*
1009 + * If immutable or not appending then return EPERM
1010 + */
1011 + if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY))) {
1012 + ZFS_EXIT(zfsvfs);
1013 + return (EPERM);
1014 + }
1015 +
1016 + rl = zfs_range_lock(zp, 0, size, RL_WRITER);
1017 +
1018 + if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
1019 + zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
1020 + error = EDQUOT;
1021 + goto out;
1022 + }
1023 +
1024 + while (pos < size) {
1025 + uint64_t length = size - pos;
1026 + length = MIN(length, ZFS_RESERVE_CHUNK);
1027 +again:
1028 + tx = dmu_tx_create(zfsvfs->z_os);
1029 + dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1030 + dmu_tx_hold_write(tx, zp->z_id, pos, length);
1031 + zfs_sa_upgrade_txholds(tx, zp);
1032 + error = dmu_tx_assign(tx, TXG_NOWAIT);
1033 + if (error) {
1034 + if (error == ERESTART) {
1035 + dmu_tx_wait(tx);
1036 + dmu_tx_abort(tx);
1037 + goto again;
1038 + }
1039 + dmu_tx_abort(tx);
1040 + goto out;
1041 + }
1042 +
1043 + if (pos == 0)
1044 + zfs_grow_blocksize(zp, MIN(size, zfsvfs->z_max_blksz), tx);
1045 + dmu_write_zero(zfsvfs->z_os, zp->z_id, pos, length, tx);
1046 +
1047 + zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1048 +
1049 + pos += length;
1050 + while ((end_size = zp->z_size) < pos)
1051 + (void) atomic_cas_64(&zp->z_size, end_size, pos);
1052 +
1053 + error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1054 +
1055 + dmu_tx_commit(tx);
1056 + if (error)
1057 + goto out;
1058 + }
1059 +out:
1060 + zfs_range_unlock(rl);
1061 + ZFS_EXIT(zfsvfs);
1062 +
1063 + return (error);
1064 +}
1065 +
959 1066 void
960 1067 zfs_get_done(zgd_t *zgd, int error)
961 1068 {
962 1069 znode_t *zp = zgd->zgd_private;
963 1070 objset_t *os = zp->z_zfsvfs->z_os;
964 1071
965 1072 if (zgd->zgd_db)
966 1073 dmu_buf_rele(zgd->zgd_db, zgd);
967 1074
968 1075 zfs_range_unlock(zgd->zgd_rl);
969 1076
970 1077 /*
971 1078 * Release the vnode asynchronously as we currently have the
972 1079 * txg stopped from syncing.
973 1080 */
974 1081 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
975 1082
976 1083 if (error == 0 && zgd->zgd_bp)
977 1084 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
978 1085
979 1086 kmem_free(zgd, sizeof (zgd_t));
980 1087 }
981 1088
982 1089 #ifdef DEBUG
983 1090 static int zil_fault_io = 0;
984 1091 #endif
985 1092
986 1093 /*
987 1094 * Get data to generate a TX_WRITE intent log record.
988 1095 */
989 1096 int
990 1097 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
991 1098 {
992 1099 zfsvfs_t *zfsvfs = arg;
993 1100 objset_t *os = zfsvfs->z_os;
994 1101 znode_t *zp;
995 1102 uint64_t object = lr->lr_foid;
996 1103 uint64_t offset = lr->lr_offset;
997 1104 uint64_t size = lr->lr_length;
998 1105 blkptr_t *bp = &lr->lr_blkptr;
999 1106 dmu_buf_t *db;
1000 1107 zgd_t *zgd;
1001 1108 int error = 0;
1002 1109
1003 1110 ASSERT(zio != NULL);
1004 1111 ASSERT(size != 0);
1005 1112
1006 1113 /*
1007 1114 * Nothing to do if the file has been removed
1008 1115 */
1009 1116 if (zfs_zget(zfsvfs, object, &zp) != 0)
1010 1117 return (SET_ERROR(ENOENT));
1011 1118 if (zp->z_unlinked) {
1012 1119 /*
1013 1120 * Release the vnode asynchronously as we currently have the
1014 1121 * txg stopped from syncing.
1015 1122 */
1016 1123 VN_RELE_ASYNC(ZTOV(zp),
1017 1124 dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1018 1125 return (SET_ERROR(ENOENT));
1019 1126 }
1020 1127
1021 1128 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1022 1129 zgd->zgd_zilog = zfsvfs->z_log;
1023 1130 zgd->zgd_private = zp;
1024 1131
1025 1132 /*
1026 1133 * Write records come in two flavors: immediate and indirect.
1027 1134 * For small writes it's cheaper to store the data with the
1028 1135 * log record (immediate); for large writes it's cheaper to
1029 1136 * sync the data and get a pointer to it (indirect) so that
1030 1137 * we don't have to write the data twice.
1031 1138 */
1032 1139 if (buf != NULL) { /* immediate write */
1033 1140 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1034 1141 /* test for truncation needs to be done while range locked */
1035 1142 if (offset >= zp->z_size) {
1036 1143 error = SET_ERROR(ENOENT);
1037 1144 } else {
1038 1145 error = dmu_read(os, object, offset, size, buf,
1039 1146 DMU_READ_NO_PREFETCH);
1040 1147 }
1041 1148 ASSERT(error == 0 || error == ENOENT);
1042 1149 } else { /* indirect write */
1043 1150 /*
1044 1151 * Have to lock the whole block to ensure when it's
1045 1152 * written out and it's checksum is being calculated
1046 1153 * that no one can change the data. We need to re-check
1047 1154 * blocksize after we get the lock in case it's changed!
1048 1155 */
1049 1156 for (;;) {
1050 1157 uint64_t blkoff;
1051 1158 size = zp->z_blksz;
1052 1159 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1053 1160 offset -= blkoff;
1054 1161 zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1055 1162 RL_READER);
1056 1163 if (zp->z_blksz == size)
1057 1164 break;
1058 1165 offset += blkoff;
1059 1166 zfs_range_unlock(zgd->zgd_rl);
1060 1167 }
1061 1168 /* test for truncation needs to be done while range locked */
1062 1169 if (lr->lr_offset >= zp->z_size)
1063 1170 error = SET_ERROR(ENOENT);
1064 1171 #ifdef DEBUG
1065 1172 if (zil_fault_io) {
1066 1173 error = SET_ERROR(EIO);
1067 1174 zil_fault_io = 0;
1068 1175 }
1069 1176 #endif
1070 1177 if (error == 0)
1071 1178 error = dmu_buf_hold(os, object, offset, zgd, &db,
1072 1179 DMU_READ_NO_PREFETCH);
1073 1180
1074 1181 if (error == 0) {
1075 1182 blkptr_t *obp = dmu_buf_get_blkptr(db);
1076 1183 if (obp) {
1077 1184 ASSERT(BP_IS_HOLE(bp));
1078 1185 *bp = *obp;
1079 1186 }
1080 1187
1081 1188 zgd->zgd_db = db;
1082 1189 zgd->zgd_bp = bp;
1083 1190
1084 1191 ASSERT(db->db_offset == offset);
1085 1192 ASSERT(db->db_size == size);
1086 1193
1087 1194 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1088 1195 zfs_get_done, zgd);
1089 1196 ASSERT(error || lr->lr_length <= zp->z_blksz);
1090 1197
1091 1198 /*
1092 1199 * On success, we need to wait for the write I/O
1093 1200 * initiated by dmu_sync() to complete before we can
1094 1201 * release this dbuf. We will finish everything up
1095 1202 * in the zfs_get_done() callback.
1096 1203 */
1097 1204 if (error == 0)
1098 1205 return (0);
1099 1206
1100 1207 if (error == EALREADY) {
1101 1208 lr->lr_common.lrc_txtype = TX_WRITE2;
1102 1209 error = 0;
1103 1210 }
1104 1211 }
1105 1212 }
1106 1213
1107 1214 zfs_get_done(zgd, error);
1108 1215
1109 1216 return (error);
1110 1217 }
1111 1218
1112 1219 /*ARGSUSED*/
1113 1220 static int
1114 1221 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1115 1222 caller_context_t *ct)
1116 1223 {
1117 1224 znode_t *zp = VTOZ(vp);
1118 1225 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1119 1226 int error;
1120 1227
1121 1228 ZFS_ENTER(zfsvfs);
1122 1229 ZFS_VERIFY_ZP(zp);
1123 1230
1124 1231 if (flag & V_ACE_MASK)
1125 1232 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1126 1233 else
1127 1234 error = zfs_zaccess_rwx(zp, mode, flag, cr);
1128 1235
1129 1236 ZFS_EXIT(zfsvfs);
1130 1237 return (error);
1131 1238 }
1132 1239
1133 1240 /*
1134 1241 * If vnode is for a device return a specfs vnode instead.
1135 1242 */
1136 1243 static int
1137 1244 specvp_check(vnode_t **vpp, cred_t *cr)
1138 1245 {
1139 1246 int error = 0;
1140 1247
1141 1248 if (IS_DEVVP(*vpp)) {
1142 1249 struct vnode *svp;
1143 1250
1144 1251 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1145 1252 VN_RELE(*vpp);
1146 1253 if (svp == NULL)
1147 1254 error = SET_ERROR(ENOSYS);
1148 1255 *vpp = svp;
1149 1256 }
1150 1257 return (error);
1151 1258 }
1152 1259
1153 1260
1154 1261 /*
1155 1262 * Lookup an entry in a directory, or an extended attribute directory.
1156 1263 * If it exists, return a held vnode reference for it.
1157 1264 *
1158 1265 * IN: dvp - vnode of directory to search.
1159 1266 * nm - name of entry to lookup.
1160 1267 * pnp - full pathname to lookup [UNUSED].
1161 1268 * flags - LOOKUP_XATTR set if looking for an attribute.
1162 1269 * rdir - root directory vnode [UNUSED].
1163 1270 * cr - credentials of caller.
1164 1271 * ct - caller context
1165 1272 * direntflags - directory lookup flags
1166 1273 * realpnp - returned pathname.
1167 1274 *
1168 1275 * OUT: vpp - vnode of located entry, NULL if not found.
1169 1276 *
1170 1277 * RETURN: 0 on success, error code on failure.
1171 1278 *
1172 1279 * Timestamps:
1173 1280 * NA
1174 1281 */
1175 1282 /* ARGSUSED */
1176 1283 static int
1177 1284 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1178 1285 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
1179 1286 int *direntflags, pathname_t *realpnp)
1180 1287 {
1181 1288 znode_t *zdp = VTOZ(dvp);
1182 1289 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1183 1290 int error = 0;
1184 1291
1185 1292 /* fast path */
1186 1293 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1187 1294
1188 1295 if (dvp->v_type != VDIR) {
1189 1296 return (SET_ERROR(ENOTDIR));
1190 1297 } else if (zdp->z_sa_hdl == NULL) {
1191 1298 return (SET_ERROR(EIO));
1192 1299 }
1193 1300
1194 1301 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1195 1302 error = zfs_fastaccesschk_execute(zdp, cr);
1196 1303 if (!error) {
1197 1304 *vpp = dvp;
1198 1305 VN_HOLD(*vpp);
1199 1306 return (0);
1200 1307 }
1201 1308 return (error);
1202 1309 } else {
1203 1310 vnode_t *tvp = dnlc_lookup(dvp, nm);
1204 1311
1205 1312 if (tvp) {
1206 1313 error = zfs_fastaccesschk_execute(zdp, cr);
1207 1314 if (error) {
1208 1315 VN_RELE(tvp);
1209 1316 return (error);
1210 1317 }
1211 1318 if (tvp == DNLC_NO_VNODE) {
1212 1319 VN_RELE(tvp);
1213 1320 return (SET_ERROR(ENOENT));
1214 1321 } else {
1215 1322 *vpp = tvp;
1216 1323 return (specvp_check(vpp, cr));
1217 1324 }
1218 1325 }
1219 1326 }
1220 1327 }
1221 1328
1222 1329 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1223 1330
1224 1331 ZFS_ENTER(zfsvfs);
1225 1332 ZFS_VERIFY_ZP(zdp);
1226 1333
1227 1334 *vpp = NULL;
1228 1335
1229 1336 if (flags & LOOKUP_XATTR) {
1230 1337 /*
1231 1338 * If the xattr property is off, refuse the lookup request.
1232 1339 */
1233 1340 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1234 1341 ZFS_EXIT(zfsvfs);
1235 1342 return (SET_ERROR(EINVAL));
1236 1343 }
1237 1344
1238 1345 /*
1239 1346 * We don't allow recursive attributes..
1240 1347 * Maybe someday we will.
1241 1348 */
1242 1349 if (zdp->z_pflags & ZFS_XATTR) {
1243 1350 ZFS_EXIT(zfsvfs);
1244 1351 return (SET_ERROR(EINVAL));
1245 1352 }
1246 1353
1247 1354 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1248 1355 ZFS_EXIT(zfsvfs);
1249 1356 return (error);
1250 1357 }
1251 1358
1252 1359 /*
1253 1360 * Do we have permission to get into attribute directory?
1254 1361 */
1255 1362
1256 1363 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1257 1364 B_FALSE, cr)) {
1258 1365 VN_RELE(*vpp);
1259 1366 *vpp = NULL;
1260 1367 }
1261 1368
1262 1369 ZFS_EXIT(zfsvfs);
1263 1370 return (error);
1264 1371 }
1265 1372
1266 1373 if (dvp->v_type != VDIR) {
1267 1374 ZFS_EXIT(zfsvfs);
1268 1375 return (SET_ERROR(ENOTDIR));
1269 1376 }
1270 1377
1271 1378 /*
1272 1379 * Check accessibility of directory.
1273 1380 */
1274 1381
1275 1382 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1276 1383 ZFS_EXIT(zfsvfs);
1277 1384 return (error);
1278 1385 }
1279 1386
1280 1387 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1281 1388 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1282 1389 ZFS_EXIT(zfsvfs);
1283 1390 return (SET_ERROR(EILSEQ));
1284 1391 }
1285 1392
1286 1393 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1287 1394 if (error == 0)
1288 1395 error = specvp_check(vpp, cr);
1289 1396
1290 1397 ZFS_EXIT(zfsvfs);
1291 1398 return (error);
1292 1399 }
1293 1400
1294 1401 /*
1295 1402 * Attempt to create a new entry in a directory. If the entry
1296 1403 * already exists, truncate the file if permissible, else return
1297 1404 * an error. Return the vp of the created or trunc'd file.
1298 1405 *
1299 1406 * IN: dvp - vnode of directory to put new file entry in.
1300 1407 * name - name of new file entry.
1301 1408 * vap - attributes of new file.
1302 1409 * excl - flag indicating exclusive or non-exclusive mode.
1303 1410 * mode - mode to open file with.
1304 1411 * cr - credentials of caller.
1305 1412 * flag - large file flag [UNUSED].
1306 1413 * ct - caller context
1307 1414 * vsecp - ACL to be set
1308 1415 *
1309 1416 * OUT: vpp - vnode of created or trunc'd entry.
1310 1417 *
1311 1418 * RETURN: 0 on success, error code on failure.
1312 1419 *
1313 1420 * Timestamps:
1314 1421 * dvp - ctime|mtime updated if new entry created
1315 1422 * vp - ctime|mtime always, atime if new
1316 1423 */
1317 1424
1318 1425 /* ARGSUSED */
1319 1426 static int
1320 1427 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1321 1428 int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1322 1429 vsecattr_t *vsecp)
1323 1430 {
1324 1431 znode_t *zp, *dzp = VTOZ(dvp);
1325 1432 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1326 1433 zilog_t *zilog;
1327 1434 objset_t *os;
1328 1435 zfs_dirlock_t *dl;
1329 1436 dmu_tx_t *tx;
1330 1437 int error;
1331 1438 ksid_t *ksid;
1332 1439 uid_t uid;
1333 1440 gid_t gid = crgetgid(cr);
1334 1441 zfs_acl_ids_t acl_ids;
1335 1442 boolean_t fuid_dirtied;
1336 1443 boolean_t have_acl = B_FALSE;
1337 1444 boolean_t waited = B_FALSE;
1338 1445
1339 1446 /*
1340 1447 * If we have an ephemeral id, ACL, or XVATTR then
1341 1448 * make sure file system is at proper version
1342 1449 */
1343 1450
1344 1451 ksid = crgetsid(cr, KSID_OWNER);
1345 1452 if (ksid)
1346 1453 uid = ksid_getid(ksid);
1347 1454 else
1348 1455 uid = crgetuid(cr);
1349 1456
1350 1457 if (zfsvfs->z_use_fuids == B_FALSE &&
1351 1458 (vsecp || (vap->va_mask & AT_XVATTR) ||
1352 1459 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1353 1460 return (SET_ERROR(EINVAL));
1354 1461
1355 1462 ZFS_ENTER(zfsvfs);
1356 1463 ZFS_VERIFY_ZP(dzp);
1357 1464 os = zfsvfs->z_os;
1358 1465 zilog = zfsvfs->z_log;
1359 1466
1360 1467 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1361 1468 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1362 1469 ZFS_EXIT(zfsvfs);
1363 1470 return (SET_ERROR(EILSEQ));
1364 1471 }
1365 1472
1366 1473 if (vap->va_mask & AT_XVATTR) {
1367 1474 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1368 1475 crgetuid(cr), cr, vap->va_type)) != 0) {
1369 1476 ZFS_EXIT(zfsvfs);
1370 1477 return (error);
1371 1478 }
1372 1479 }
1373 1480 top:
1374 1481 *vpp = NULL;
1375 1482
1376 1483 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1377 1484 vap->va_mode &= ~VSVTX;
1378 1485
1379 1486 if (*name == '\0') {
1380 1487 /*
1381 1488 * Null component name refers to the directory itself.
1382 1489 */
1383 1490 VN_HOLD(dvp);
1384 1491 zp = dzp;
1385 1492 dl = NULL;
1386 1493 error = 0;
1387 1494 } else {
1388 1495 /* possible VN_HOLD(zp) */
1389 1496 int zflg = 0;
1390 1497
1391 1498 if (flag & FIGNORECASE)
1392 1499 zflg |= ZCILOOK;
1393 1500
1394 1501 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1395 1502 NULL, NULL);
1396 1503 if (error) {
1397 1504 if (have_acl)
1398 1505 zfs_acl_ids_free(&acl_ids);
1399 1506 if (strcmp(name, "..") == 0)
1400 1507 error = SET_ERROR(EISDIR);
1401 1508 ZFS_EXIT(zfsvfs);
1402 1509 return (error);
1403 1510 }
1404 1511 }
1405 1512
1406 1513 if (zp == NULL) {
1407 1514 uint64_t txtype;
1408 1515
1409 1516 /*
1410 1517 * Create a new file object and update the directory
1411 1518 * to reference it.
1412 1519 */
1413 1520 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1414 1521 if (have_acl)
1415 1522 zfs_acl_ids_free(&acl_ids);
1416 1523 goto out;
1417 1524 }
1418 1525
1419 1526 /*
1420 1527 * We only support the creation of regular files in
1421 1528 * extended attribute directories.
1422 1529 */
1423 1530
1424 1531 if ((dzp->z_pflags & ZFS_XATTR) &&
1425 1532 (vap->va_type != VREG)) {
1426 1533 if (have_acl)
1427 1534 zfs_acl_ids_free(&acl_ids);
1428 1535 error = SET_ERROR(EINVAL);
1429 1536 goto out;
1430 1537 }
1431 1538
1432 1539 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1433 1540 cr, vsecp, &acl_ids)) != 0)
1434 1541 goto out;
1435 1542 have_acl = B_TRUE;
1436 1543
1437 1544 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1438 1545 zfs_acl_ids_free(&acl_ids);
1439 1546 error = SET_ERROR(EDQUOT);
1440 1547 goto out;
1441 1548 }
1442 1549
1443 1550 tx = dmu_tx_create(os);
1444 1551
1445 1552 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1446 1553 ZFS_SA_BASE_ATTR_SIZE);
1447 1554
1448 1555 fuid_dirtied = zfsvfs->z_fuid_dirty;
1449 1556 if (fuid_dirtied)
1450 1557 zfs_fuid_txhold(zfsvfs, tx);
1451 1558 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1452 1559 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1453 1560 if (!zfsvfs->z_use_sa &&
1454 1561 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1455 1562 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1456 1563 0, acl_ids.z_aclp->z_acl_bytes);
1457 1564 }
1458 1565 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1459 1566 if (error) {
1460 1567 zfs_dirent_unlock(dl);
1461 1568 if (error == ERESTART) {
1462 1569 waited = B_TRUE;
1463 1570 dmu_tx_wait(tx);
1464 1571 dmu_tx_abort(tx);
1465 1572 goto top;
1466 1573 }
1467 1574 zfs_acl_ids_free(&acl_ids);
1468 1575 dmu_tx_abort(tx);
1469 1576 ZFS_EXIT(zfsvfs);
1470 1577 return (error);
1471 1578 }
1472 1579 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1473 1580
1474 1581 if (fuid_dirtied)
1475 1582 zfs_fuid_sync(zfsvfs, tx);
1476 1583
1477 1584 (void) zfs_link_create(dl, zp, tx, ZNEW);
1478 1585 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1479 1586 if (flag & FIGNORECASE)
1480 1587 txtype |= TX_CI;
1481 1588 zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1482 1589 vsecp, acl_ids.z_fuidp, vap);
1483 1590 zfs_acl_ids_free(&acl_ids);
1484 1591 dmu_tx_commit(tx);
1485 1592 } else {
1486 1593 int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1487 1594
1488 1595 if (have_acl)
1489 1596 zfs_acl_ids_free(&acl_ids);
1490 1597 have_acl = B_FALSE;
1491 1598
1492 1599 /*
1493 1600 * A directory entry already exists for this name.
1494 1601 */
1495 1602 /*
1496 1603 * Can't truncate an existing file if in exclusive mode.
1497 1604 */
1498 1605 if (excl == EXCL) {
1499 1606 error = SET_ERROR(EEXIST);
1500 1607 goto out;
1501 1608 }
1502 1609 /*
1503 1610 * Can't open a directory for writing.
1504 1611 */
1505 1612 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1506 1613 error = SET_ERROR(EISDIR);
1507 1614 goto out;
1508 1615 }
1509 1616 /*
1510 1617 * Verify requested access to file.
1511 1618 */
1512 1619 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1513 1620 goto out;
1514 1621 }
1515 1622
1516 1623 mutex_enter(&dzp->z_lock);
1517 1624 dzp->z_seq++;
1518 1625 mutex_exit(&dzp->z_lock);
1519 1626
1520 1627 /*
1521 1628 * Truncate regular files if requested.
1522 1629 */
1523 1630 if ((ZTOV(zp)->v_type == VREG) &&
1524 1631 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1525 1632 /* we can't hold any locks when calling zfs_freesp() */
1526 1633 zfs_dirent_unlock(dl);
1527 1634 dl = NULL;
1528 1635 error = zfs_freesp(zp, 0, 0, mode, TRUE);
1529 1636 if (error == 0) {
1530 1637 vnevent_create(ZTOV(zp), ct);
1531 1638 }
1532 1639 }
1533 1640 }
1534 1641 out:
1535 1642
1536 1643 if (dl)
1537 1644 zfs_dirent_unlock(dl);
1538 1645
1539 1646 if (error) {
1540 1647 if (zp)
1541 1648 VN_RELE(ZTOV(zp));
1542 1649 } else {
1543 1650 *vpp = ZTOV(zp);
1544 1651 error = specvp_check(vpp, cr);
1545 1652 }
1546 1653
1547 1654 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1548 1655 zil_commit(zilog, 0);
1549 1656
1550 1657 ZFS_EXIT(zfsvfs);
1551 1658 return (error);
1552 1659 }
1553 1660
1554 1661 /*
1555 1662 * Remove an entry from a directory.
1556 1663 *
1557 1664 * IN: dvp - vnode of directory to remove entry from.
1558 1665 * name - name of entry to remove.
1559 1666 * cr - credentials of caller.
1560 1667 * ct - caller context
1561 1668 * flags - case flags
1562 1669 *
1563 1670 * RETURN: 0 on success, error code on failure.
1564 1671 *
1565 1672 * Timestamps:
1566 1673 * dvp - ctime|mtime
1567 1674 * vp - ctime (if nlink > 0)
1568 1675 */
1569 1676
1570 1677 uint64_t null_xattr = 0;
1571 1678
1572 1679 /*ARGSUSED*/
1573 1680 static int
1574 1681 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1575 1682 int flags)
1576 1683 {
1577 1684 znode_t *zp, *dzp = VTOZ(dvp);
1578 1685 znode_t *xzp;
1579 1686 vnode_t *vp;
1580 1687 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1581 1688 zilog_t *zilog;
1582 1689 uint64_t acl_obj, xattr_obj;
1583 1690 uint64_t xattr_obj_unlinked = 0;
1584 1691 uint64_t obj = 0;
1585 1692 zfs_dirlock_t *dl;
1586 1693 dmu_tx_t *tx;
1587 1694 boolean_t may_delete_now, delete_now = FALSE;
1588 1695 boolean_t unlinked, toobig = FALSE;
1589 1696 uint64_t txtype;
1590 1697 pathname_t *realnmp = NULL;
1591 1698 pathname_t realnm;
1592 1699 int error;
1593 1700 int zflg = ZEXISTS;
1594 1701 boolean_t waited = B_FALSE;
1595 1702
1596 1703 ZFS_ENTER(zfsvfs);
1597 1704 ZFS_VERIFY_ZP(dzp);
1598 1705 zilog = zfsvfs->z_log;
1599 1706
1600 1707 if (flags & FIGNORECASE) {
1601 1708 zflg |= ZCILOOK;
1602 1709 pn_alloc(&realnm);
1603 1710 realnmp = &realnm;
1604 1711 }
1605 1712
1606 1713 top:
1607 1714 xattr_obj = 0;
1608 1715 xzp = NULL;
1609 1716 /*
1610 1717 * Attempt to lock directory; fail if entry doesn't exist.
1611 1718 */
1612 1719 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1613 1720 NULL, realnmp)) {
1614 1721 if (realnmp)
1615 1722 pn_free(realnmp);
1616 1723 ZFS_EXIT(zfsvfs);
1617 1724 return (error);
1618 1725 }
1619 1726
1620 1727 vp = ZTOV(zp);
1621 1728
1622 1729 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1623 1730 goto out;
1624 1731 }
1625 1732
1626 1733 /*
1627 1734 * Need to use rmdir for removing directories.
1628 1735 */
1629 1736 if (vp->v_type == VDIR) {
1630 1737 error = SET_ERROR(EPERM);
1631 1738 goto out;
1632 1739 }
1633 1740
1634 1741 vnevent_remove(vp, dvp, name, ct);
1635 1742
1636 1743 if (realnmp)
1637 1744 dnlc_remove(dvp, realnmp->pn_buf);
1638 1745 else
1639 1746 dnlc_remove(dvp, name);
1640 1747
1641 1748 mutex_enter(&vp->v_lock);
1642 1749 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1643 1750 mutex_exit(&vp->v_lock);
1644 1751
1645 1752 /*
1646 1753 * We may delete the znode now, or we may put it in the unlinked set;
1647 1754 * it depends on whether we're the last link, and on whether there are
1648 1755 * other holds on the vnode. So we dmu_tx_hold() the right things to
1649 1756 * allow for either case.
1650 1757 */
1651 1758 obj = zp->z_id;
1652 1759 tx = dmu_tx_create(zfsvfs->z_os);
1653 1760 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1654 1761 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1655 1762 zfs_sa_upgrade_txholds(tx, zp);
1656 1763 zfs_sa_upgrade_txholds(tx, dzp);
1657 1764 if (may_delete_now) {
1658 1765 toobig =
1659 1766 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1660 1767 /* if the file is too big, only hold_free a token amount */
1661 1768 dmu_tx_hold_free(tx, zp->z_id, 0,
1662 1769 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1663 1770 }
1664 1771
1665 1772 /* are there any extended attributes? */
1666 1773 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1667 1774 &xattr_obj, sizeof (xattr_obj));
1668 1775 if (error == 0 && xattr_obj) {
1669 1776 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1670 1777 ASSERT0(error);
1671 1778 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1672 1779 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1673 1780 }
1674 1781
1675 1782 mutex_enter(&zp->z_lock);
1676 1783 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1677 1784 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1678 1785 mutex_exit(&zp->z_lock);
1679 1786
1680 1787 /* charge as an update -- would be nice not to charge at all */
1681 1788 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1682 1789
1683 1790 /*
1684 1791 * Mark this transaction as typically resulting in a net free of
1685 1792 * space, unless object removal will be delayed indefinitely
1686 1793 * (due to active holds on the vnode due to the file being open).
1687 1794 */
1688 1795 if (may_delete_now)
1689 1796 dmu_tx_mark_netfree(tx);
1690 1797
1691 1798 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1692 1799 if (error) {
1693 1800 zfs_dirent_unlock(dl);
1694 1801 VN_RELE(vp);
1695 1802 if (xzp)
1696 1803 VN_RELE(ZTOV(xzp));
1697 1804 if (error == ERESTART) {
1698 1805 waited = B_TRUE;
1699 1806 dmu_tx_wait(tx);
1700 1807 dmu_tx_abort(tx);
1701 1808 goto top;
1702 1809 }
1703 1810 if (realnmp)
1704 1811 pn_free(realnmp);
1705 1812 dmu_tx_abort(tx);
1706 1813 ZFS_EXIT(zfsvfs);
1707 1814 return (error);
1708 1815 }
1709 1816
1710 1817 /*
1711 1818 * Remove the directory entry.
1712 1819 */
1713 1820 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1714 1821
1715 1822 if (error) {
1716 1823 dmu_tx_commit(tx);
1717 1824 goto out;
1718 1825 }
1719 1826
1720 1827 if (unlinked) {
1721 1828 /*
1722 1829 * Hold z_lock so that we can make sure that the ACL obj
1723 1830 * hasn't changed. Could have been deleted due to
1724 1831 * zfs_sa_upgrade().
1725 1832 */
1726 1833 mutex_enter(&zp->z_lock);
1727 1834 mutex_enter(&vp->v_lock);
1728 1835 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1729 1836 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1730 1837 delete_now = may_delete_now && !toobig &&
1731 1838 vp->v_count == 1 && !vn_has_cached_data(vp) &&
1732 1839 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1733 1840 acl_obj;
1734 1841 mutex_exit(&vp->v_lock);
1735 1842 }
1736 1843
1737 1844 if (delete_now) {
1738 1845 if (xattr_obj_unlinked) {
1739 1846 ASSERT3U(xzp->z_links, ==, 2);
1740 1847 mutex_enter(&xzp->z_lock);
1741 1848 xzp->z_unlinked = 1;
1742 1849 xzp->z_links = 0;
1743 1850 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1744 1851 &xzp->z_links, sizeof (xzp->z_links), tx);
1745 1852 ASSERT3U(error, ==, 0);
1746 1853 mutex_exit(&xzp->z_lock);
1747 1854 zfs_unlinked_add(xzp, tx);
1748 1855
1749 1856 if (zp->z_is_sa)
1750 1857 error = sa_remove(zp->z_sa_hdl,
1751 1858 SA_ZPL_XATTR(zfsvfs), tx);
1752 1859 else
1753 1860 error = sa_update(zp->z_sa_hdl,
1754 1861 SA_ZPL_XATTR(zfsvfs), &null_xattr,
1755 1862 sizeof (uint64_t), tx);
1756 1863 ASSERT0(error);
1757 1864 }
1758 1865 mutex_enter(&vp->v_lock);
1759 1866 vp->v_count--;
1760 1867 ASSERT0(vp->v_count);
1761 1868 mutex_exit(&vp->v_lock);
1762 1869 mutex_exit(&zp->z_lock);
1763 1870 zfs_znode_delete(zp, tx);
1764 1871 } else if (unlinked) {
1765 1872 mutex_exit(&zp->z_lock);
1766 1873 zfs_unlinked_add(zp, tx);
1767 1874 }
1768 1875
1769 1876 txtype = TX_REMOVE;
1770 1877 if (flags & FIGNORECASE)
1771 1878 txtype |= TX_CI;
1772 1879 zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1773 1880
1774 1881 dmu_tx_commit(tx);
1775 1882 out:
1776 1883 if (realnmp)
1777 1884 pn_free(realnmp);
1778 1885
1779 1886 zfs_dirent_unlock(dl);
1780 1887
1781 1888 if (!delete_now)
1782 1889 VN_RELE(vp);
1783 1890 if (xzp)
1784 1891 VN_RELE(ZTOV(xzp));
1785 1892
1786 1893 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1787 1894 zil_commit(zilog, 0);
1788 1895
1789 1896 ZFS_EXIT(zfsvfs);
1790 1897 return (error);
1791 1898 }
1792 1899
1793 1900 /*
1794 1901 * Create a new directory and insert it into dvp using the name
1795 1902 * provided. Return a pointer to the inserted directory.
1796 1903 *
1797 1904 * IN: dvp - vnode of directory to add subdir to.
1798 1905 * dirname - name of new directory.
1799 1906 * vap - attributes of new directory.
1800 1907 * cr - credentials of caller.
1801 1908 * ct - caller context
1802 1909 * flags - case flags
1803 1910 * vsecp - ACL to be set
1804 1911 *
1805 1912 * OUT: vpp - vnode of created directory.
1806 1913 *
1807 1914 * RETURN: 0 on success, error code on failure.
1808 1915 *
1809 1916 * Timestamps:
1810 1917 * dvp - ctime|mtime updated
1811 1918 * vp - ctime|mtime|atime updated
1812 1919 */
1813 1920 /*ARGSUSED*/
1814 1921 static int
1815 1922 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1816 1923 caller_context_t *ct, int flags, vsecattr_t *vsecp)
1817 1924 {
1818 1925 znode_t *zp, *dzp = VTOZ(dvp);
1819 1926 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1820 1927 zilog_t *zilog;
1821 1928 zfs_dirlock_t *dl;
1822 1929 uint64_t txtype;
1823 1930 dmu_tx_t *tx;
1824 1931 int error;
1825 1932 int zf = ZNEW;
1826 1933 ksid_t *ksid;
1827 1934 uid_t uid;
1828 1935 gid_t gid = crgetgid(cr);
1829 1936 zfs_acl_ids_t acl_ids;
1830 1937 boolean_t fuid_dirtied;
1831 1938 boolean_t waited = B_FALSE;
1832 1939
1833 1940 ASSERT(vap->va_type == VDIR);
1834 1941
1835 1942 /*
1836 1943 * If we have an ephemeral id, ACL, or XVATTR then
1837 1944 * make sure file system is at proper version
1838 1945 */
1839 1946
1840 1947 ksid = crgetsid(cr, KSID_OWNER);
1841 1948 if (ksid)
1842 1949 uid = ksid_getid(ksid);
1843 1950 else
1844 1951 uid = crgetuid(cr);
1845 1952 if (zfsvfs->z_use_fuids == B_FALSE &&
1846 1953 (vsecp || (vap->va_mask & AT_XVATTR) ||
1847 1954 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1848 1955 return (SET_ERROR(EINVAL));
1849 1956
1850 1957 ZFS_ENTER(zfsvfs);
1851 1958 ZFS_VERIFY_ZP(dzp);
1852 1959 zilog = zfsvfs->z_log;
1853 1960
1854 1961 if (dzp->z_pflags & ZFS_XATTR) {
1855 1962 ZFS_EXIT(zfsvfs);
1856 1963 return (SET_ERROR(EINVAL));
1857 1964 }
1858 1965
1859 1966 if (zfsvfs->z_utf8 && u8_validate(dirname,
1860 1967 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1861 1968 ZFS_EXIT(zfsvfs);
1862 1969 return (SET_ERROR(EILSEQ));
1863 1970 }
1864 1971 if (flags & FIGNORECASE)
1865 1972 zf |= ZCILOOK;
1866 1973
1867 1974 if (vap->va_mask & AT_XVATTR) {
1868 1975 if ((error = secpolicy_xvattr((xvattr_t *)vap,
1869 1976 crgetuid(cr), cr, vap->va_type)) != 0) {
1870 1977 ZFS_EXIT(zfsvfs);
1871 1978 return (error);
1872 1979 }
1873 1980 }
1874 1981
1875 1982 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1876 1983 vsecp, &acl_ids)) != 0) {
1877 1984 ZFS_EXIT(zfsvfs);
1878 1985 return (error);
1879 1986 }
1880 1987 /*
1881 1988 * First make sure the new directory doesn't exist.
1882 1989 *
1883 1990 * Existence is checked first to make sure we don't return
1884 1991 * EACCES instead of EEXIST which can cause some applications
1885 1992 * to fail.
1886 1993 */
1887 1994 top:
1888 1995 *vpp = NULL;
1889 1996
1890 1997 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1891 1998 NULL, NULL)) {
1892 1999 zfs_acl_ids_free(&acl_ids);
1893 2000 ZFS_EXIT(zfsvfs);
1894 2001 return (error);
1895 2002 }
1896 2003
1897 2004 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1898 2005 zfs_acl_ids_free(&acl_ids);
1899 2006 zfs_dirent_unlock(dl);
1900 2007 ZFS_EXIT(zfsvfs);
1901 2008 return (error);
1902 2009 }
1903 2010
1904 2011 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1905 2012 zfs_acl_ids_free(&acl_ids);
1906 2013 zfs_dirent_unlock(dl);
1907 2014 ZFS_EXIT(zfsvfs);
1908 2015 return (SET_ERROR(EDQUOT));
1909 2016 }
1910 2017
1911 2018 /*
1912 2019 * Add a new entry to the directory.
1913 2020 */
1914 2021 tx = dmu_tx_create(zfsvfs->z_os);
1915 2022 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1916 2023 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1917 2024 fuid_dirtied = zfsvfs->z_fuid_dirty;
1918 2025 if (fuid_dirtied)
1919 2026 zfs_fuid_txhold(zfsvfs, tx);
1920 2027 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1921 2028 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1922 2029 acl_ids.z_aclp->z_acl_bytes);
1923 2030 }
1924 2031
1925 2032 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1926 2033 ZFS_SA_BASE_ATTR_SIZE);
1927 2034
1928 2035 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1929 2036 if (error) {
1930 2037 zfs_dirent_unlock(dl);
1931 2038 if (error == ERESTART) {
1932 2039 waited = B_TRUE;
1933 2040 dmu_tx_wait(tx);
1934 2041 dmu_tx_abort(tx);
1935 2042 goto top;
1936 2043 }
1937 2044 zfs_acl_ids_free(&acl_ids);
1938 2045 dmu_tx_abort(tx);
1939 2046 ZFS_EXIT(zfsvfs);
1940 2047 return (error);
1941 2048 }
1942 2049
1943 2050 /*
1944 2051 * Create new node.
1945 2052 */
1946 2053 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1947 2054
1948 2055 if (fuid_dirtied)
1949 2056 zfs_fuid_sync(zfsvfs, tx);
1950 2057
1951 2058 /*
1952 2059 * Now put new name in parent dir.
1953 2060 */
1954 2061 (void) zfs_link_create(dl, zp, tx, ZNEW);
1955 2062
1956 2063 *vpp = ZTOV(zp);
1957 2064
1958 2065 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1959 2066 if (flags & FIGNORECASE)
1960 2067 txtype |= TX_CI;
1961 2068 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1962 2069 acl_ids.z_fuidp, vap);
1963 2070
1964 2071 zfs_acl_ids_free(&acl_ids);
1965 2072
1966 2073 dmu_tx_commit(tx);
1967 2074
1968 2075 zfs_dirent_unlock(dl);
1969 2076
1970 2077 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1971 2078 zil_commit(zilog, 0);
1972 2079
1973 2080 ZFS_EXIT(zfsvfs);
1974 2081 return (0);
1975 2082 }
1976 2083
1977 2084 /*
1978 2085 * Remove a directory subdir entry. If the current working
1979 2086 * directory is the same as the subdir to be removed, the
1980 2087 * remove will fail.
1981 2088 *
1982 2089 * IN: dvp - vnode of directory to remove from.
1983 2090 * name - name of directory to be removed.
1984 2091 * cwd - vnode of current working directory.
1985 2092 * cr - credentials of caller.
1986 2093 * ct - caller context
1987 2094 * flags - case flags
1988 2095 *
1989 2096 * RETURN: 0 on success, error code on failure.
1990 2097 *
1991 2098 * Timestamps:
1992 2099 * dvp - ctime|mtime updated
1993 2100 */
1994 2101 /*ARGSUSED*/
1995 2102 static int
1996 2103 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1997 2104 caller_context_t *ct, int flags)
1998 2105 {
1999 2106 znode_t *dzp = VTOZ(dvp);
2000 2107 znode_t *zp;
2001 2108 vnode_t *vp;
2002 2109 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
2003 2110 zilog_t *zilog;
2004 2111 zfs_dirlock_t *dl;
2005 2112 dmu_tx_t *tx;
2006 2113 int error;
2007 2114 int zflg = ZEXISTS;
2008 2115 boolean_t waited = B_FALSE;
2009 2116
2010 2117 ZFS_ENTER(zfsvfs);
2011 2118 ZFS_VERIFY_ZP(dzp);
2012 2119 zilog = zfsvfs->z_log;
2013 2120
2014 2121 if (flags & FIGNORECASE)
2015 2122 zflg |= ZCILOOK;
2016 2123 top:
2017 2124 zp = NULL;
2018 2125
2019 2126 /*
2020 2127 * Attempt to lock directory; fail if entry doesn't exist.
2021 2128 */
2022 2129 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2023 2130 NULL, NULL)) {
2024 2131 ZFS_EXIT(zfsvfs);
2025 2132 return (error);
2026 2133 }
2027 2134
2028 2135 vp = ZTOV(zp);
2029 2136
2030 2137 if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2031 2138 goto out;
2032 2139 }
2033 2140
2034 2141 if (vp->v_type != VDIR) {
2035 2142 error = SET_ERROR(ENOTDIR);
2036 2143 goto out;
2037 2144 }
2038 2145
2039 2146 if (vp == cwd) {
2040 2147 error = SET_ERROR(EINVAL);
2041 2148 goto out;
2042 2149 }
2043 2150
2044 2151 vnevent_rmdir(vp, dvp, name, ct);
2045 2152
2046 2153 /*
2047 2154 * Grab a lock on the directory to make sure that noone is
2048 2155 * trying to add (or lookup) entries while we are removing it.
2049 2156 */
2050 2157 rw_enter(&zp->z_name_lock, RW_WRITER);
2051 2158
2052 2159 /*
2053 2160 * Grab a lock on the parent pointer to make sure we play well
2054 2161 * with the treewalk and directory rename code.
2055 2162 */
2056 2163 rw_enter(&zp->z_parent_lock, RW_WRITER);
2057 2164
2058 2165 tx = dmu_tx_create(zfsvfs->z_os);
2059 2166 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2060 2167 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2061 2168 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2062 2169 zfs_sa_upgrade_txholds(tx, zp);
2063 2170 zfs_sa_upgrade_txholds(tx, dzp);
2064 2171 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2065 2172 if (error) {
2066 2173 rw_exit(&zp->z_parent_lock);
2067 2174 rw_exit(&zp->z_name_lock);
2068 2175 zfs_dirent_unlock(dl);
2069 2176 VN_RELE(vp);
2070 2177 if (error == ERESTART) {
2071 2178 waited = B_TRUE;
2072 2179 dmu_tx_wait(tx);
2073 2180 dmu_tx_abort(tx);
2074 2181 goto top;
2075 2182 }
2076 2183 dmu_tx_abort(tx);
2077 2184 ZFS_EXIT(zfsvfs);
2078 2185 return (error);
2079 2186 }
2080 2187
2081 2188 error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2082 2189
2083 2190 if (error == 0) {
2084 2191 uint64_t txtype = TX_RMDIR;
2085 2192 if (flags & FIGNORECASE)
2086 2193 txtype |= TX_CI;
2087 2194 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2088 2195 }
2089 2196
2090 2197 dmu_tx_commit(tx);
2091 2198
2092 2199 rw_exit(&zp->z_parent_lock);
2093 2200 rw_exit(&zp->z_name_lock);
2094 2201 out:
2095 2202 zfs_dirent_unlock(dl);
2096 2203
2097 2204 VN_RELE(vp);
2098 2205
2099 2206 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2100 2207 zil_commit(zilog, 0);
2101 2208
2102 2209 ZFS_EXIT(zfsvfs);
2103 2210 return (error);
2104 2211 }
2105 2212
2106 2213 /*
2107 2214 * Read as many directory entries as will fit into the provided
2108 2215 * buffer from the given directory cursor position (specified in
2109 2216 * the uio structure).
2110 2217 *
2111 2218 * IN: vp - vnode of directory to read.
2112 2219 * uio - structure supplying read location, range info,
2113 2220 * and return buffer.
2114 2221 * cr - credentials of caller.
2115 2222 * ct - caller context
2116 2223 * flags - case flags
2117 2224 *
2118 2225 * OUT: uio - updated offset and range, buffer filled.
2119 2226 * eofp - set to true if end-of-file detected.
2120 2227 *
2121 2228 * RETURN: 0 on success, error code on failure.
2122 2229 *
2123 2230 * Timestamps:
2124 2231 * vp - atime updated
2125 2232 *
2126 2233 * Note that the low 4 bits of the cookie returned by zap is always zero.
2127 2234 * This allows us to use the low range for "special" directory entries:
2128 2235 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2129 2236 * we use the offset 2 for the '.zfs' directory.
2130 2237 */
2131 2238 /* ARGSUSED */
2132 2239 static int
2133 2240 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2134 2241 caller_context_t *ct, int flags)
2135 2242 {
2136 2243 znode_t *zp = VTOZ(vp);
2137 2244 iovec_t *iovp;
2138 2245 edirent_t *eodp;
2139 2246 dirent64_t *odp;
2140 2247 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2141 2248 objset_t *os;
2142 2249 caddr_t outbuf;
2143 2250 size_t bufsize;
2144 2251 zap_cursor_t zc;
2145 2252 zap_attribute_t zap;
2146 2253 uint_t bytes_wanted;
2147 2254 uint64_t offset; /* must be unsigned; checks for < 1 */
2148 2255 uint64_t parent;
2149 2256 int local_eof;
2150 2257 int outcount;
2151 2258 int error;
2152 2259 uint8_t prefetch;
2153 2260 boolean_t check_sysattrs;
2154 2261
2155 2262 ZFS_ENTER(zfsvfs);
2156 2263 ZFS_VERIFY_ZP(zp);
2157 2264
2158 2265 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2159 2266 &parent, sizeof (parent))) != 0) {
2160 2267 ZFS_EXIT(zfsvfs);
2161 2268 return (error);
2162 2269 }
2163 2270
2164 2271 /*
2165 2272 * If we are not given an eof variable,
2166 2273 * use a local one.
2167 2274 */
2168 2275 if (eofp == NULL)
2169 2276 eofp = &local_eof;
2170 2277
2171 2278 /*
2172 2279 * Check for valid iov_len.
2173 2280 */
2174 2281 if (uio->uio_iov->iov_len <= 0) {
2175 2282 ZFS_EXIT(zfsvfs);
2176 2283 return (SET_ERROR(EINVAL));
2177 2284 }
2178 2285
2179 2286 /*
2180 2287 * Quit if directory has been removed (posix)
2181 2288 */
2182 2289 if ((*eofp = zp->z_unlinked) != 0) {
2183 2290 ZFS_EXIT(zfsvfs);
2184 2291 return (0);
2185 2292 }
2186 2293
2187 2294 error = 0;
2188 2295 os = zfsvfs->z_os;
2189 2296 offset = uio->uio_loffset;
2190 2297 prefetch = zp->z_zn_prefetch;
2191 2298
2192 2299 /*
2193 2300 * Initialize the iterator cursor.
2194 2301 */
2195 2302 if (offset <= 3) {
2196 2303 /*
2197 2304 * Start iteration from the beginning of the directory.
2198 2305 */
2199 2306 zap_cursor_init(&zc, os, zp->z_id);
2200 2307 } else {
2201 2308 /*
2202 2309 * The offset is a serialized cursor.
2203 2310 */
2204 2311 zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2205 2312 }
2206 2313
2207 2314 /*
2208 2315 * Get space to change directory entries into fs independent format.
2209 2316 */
2210 2317 iovp = uio->uio_iov;
2211 2318 bytes_wanted = iovp->iov_len;
2212 2319 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2213 2320 bufsize = bytes_wanted;
2214 2321 outbuf = kmem_alloc(bufsize, KM_SLEEP);
2215 2322 odp = (struct dirent64 *)outbuf;
2216 2323 } else {
2217 2324 bufsize = bytes_wanted;
2218 2325 outbuf = NULL;
2219 2326 odp = (struct dirent64 *)iovp->iov_base;
2220 2327 }
2221 2328 eodp = (struct edirent *)odp;
2222 2329
2223 2330 /*
2224 2331 * If this VFS supports the system attribute view interface; and
2225 2332 * we're looking at an extended attribute directory; and we care
2226 2333 * about normalization conflicts on this vfs; then we must check
2227 2334 * for normalization conflicts with the sysattr name space.
2228 2335 */
2229 2336 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2230 2337 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2231 2338 (flags & V_RDDIR_ENTFLAGS);
2232 2339
2233 2340 /*
2234 2341 * Transform to file-system independent format
2235 2342 */
2236 2343 outcount = 0;
2237 2344 while (outcount < bytes_wanted) {
2238 2345 ino64_t objnum;
2239 2346 ushort_t reclen;
2240 2347 off64_t *next = NULL;
2241 2348
2242 2349 /*
2243 2350 * Special case `.', `..', and `.zfs'.
2244 2351 */
2245 2352 if (offset == 0) {
2246 2353 (void) strcpy(zap.za_name, ".");
2247 2354 zap.za_normalization_conflict = 0;
2248 2355 objnum = zp->z_id;
2249 2356 } else if (offset == 1) {
2250 2357 (void) strcpy(zap.za_name, "..");
2251 2358 zap.za_normalization_conflict = 0;
2252 2359 objnum = parent;
2253 2360 } else if (offset == 2 && zfs_show_ctldir(zp)) {
2254 2361 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2255 2362 zap.za_normalization_conflict = 0;
2256 2363 objnum = ZFSCTL_INO_ROOT;
2257 2364 } else {
2258 2365 /*
2259 2366 * Grab next entry.
2260 2367 */
2261 2368 if (error = zap_cursor_retrieve(&zc, &zap)) {
2262 2369 if ((*eofp = (error == ENOENT)) != 0)
2263 2370 break;
2264 2371 else
2265 2372 goto update;
2266 2373 }
2267 2374
2268 2375 if (zap.za_integer_length != 8 ||
2269 2376 zap.za_num_integers != 1) {
2270 2377 cmn_err(CE_WARN, "zap_readdir: bad directory "
2271 2378 "entry, obj = %lld, offset = %lld\n",
2272 2379 (u_longlong_t)zp->z_id,
2273 2380 (u_longlong_t)offset);
2274 2381 error = SET_ERROR(ENXIO);
2275 2382 goto update;
2276 2383 }
2277 2384
2278 2385 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2279 2386 /*
2280 2387 * MacOS X can extract the object type here such as:
2281 2388 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2282 2389 */
2283 2390
2284 2391 if (check_sysattrs && !zap.za_normalization_conflict) {
2285 2392 zap.za_normalization_conflict =
2286 2393 xattr_sysattr_casechk(zap.za_name);
2287 2394 }
2288 2395 }
2289 2396
2290 2397 if (flags & V_RDDIR_ACCFILTER) {
2291 2398 /*
2292 2399 * If we have no access at all, don't include
2293 2400 * this entry in the returned information
2294 2401 */
2295 2402 znode_t *ezp;
2296 2403 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2297 2404 goto skip_entry;
2298 2405 if (!zfs_has_access(ezp, cr)) {
2299 2406 VN_RELE(ZTOV(ezp));
2300 2407 goto skip_entry;
2301 2408 }
2302 2409 VN_RELE(ZTOV(ezp));
2303 2410 }
2304 2411
2305 2412 if (flags & V_RDDIR_ENTFLAGS)
2306 2413 reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2307 2414 else
2308 2415 reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2309 2416
2310 2417 /*
2311 2418 * Will this entry fit in the buffer?
2312 2419 */
2313 2420 if (outcount + reclen > bufsize) {
2314 2421 /*
2315 2422 * Did we manage to fit anything in the buffer?
2316 2423 */
2317 2424 if (!outcount) {
2318 2425 error = SET_ERROR(EINVAL);
2319 2426 goto update;
2320 2427 }
2321 2428 break;
2322 2429 }
2323 2430 if (flags & V_RDDIR_ENTFLAGS) {
2324 2431 /*
2325 2432 * Add extended flag entry:
2326 2433 */
2327 2434 eodp->ed_ino = objnum;
2328 2435 eodp->ed_reclen = reclen;
2329 2436 /* NOTE: ed_off is the offset for the *next* entry */
2330 2437 next = &(eodp->ed_off);
2331 2438 eodp->ed_eflags = zap.za_normalization_conflict ?
2332 2439 ED_CASE_CONFLICT : 0;
2333 2440 (void) strncpy(eodp->ed_name, zap.za_name,
2334 2441 EDIRENT_NAMELEN(reclen));
2335 2442 eodp = (edirent_t *)((intptr_t)eodp + reclen);
2336 2443 } else {
2337 2444 /*
2338 2445 * Add normal entry:
2339 2446 */
2340 2447 odp->d_ino = objnum;
2341 2448 odp->d_reclen = reclen;
2342 2449 /* NOTE: d_off is the offset for the *next* entry */
2343 2450 next = &(odp->d_off);
2344 2451 (void) strncpy(odp->d_name, zap.za_name,
2345 2452 DIRENT64_NAMELEN(reclen));
2346 2453 odp = (dirent64_t *)((intptr_t)odp + reclen);
2347 2454 }
2348 2455 outcount += reclen;
2349 2456
2350 2457 ASSERT(outcount <= bufsize);
2351 2458
2352 2459 /* Prefetch znode */
2353 2460 if (prefetch)
2354 2461 dmu_prefetch(os, objnum, 0, 0);
2355 2462
2356 2463 skip_entry:
2357 2464 /*
2358 2465 * Move to the next entry, fill in the previous offset.
2359 2466 */
2360 2467 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2361 2468 zap_cursor_advance(&zc);
2362 2469 offset = zap_cursor_serialize(&zc);
2363 2470 } else {
2364 2471 offset += 1;
2365 2472 }
2366 2473 if (next)
2367 2474 *next = offset;
2368 2475 }
2369 2476 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2370 2477
2371 2478 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2372 2479 iovp->iov_base += outcount;
2373 2480 iovp->iov_len -= outcount;
2374 2481 uio->uio_resid -= outcount;
2375 2482 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2376 2483 /*
2377 2484 * Reset the pointer.
2378 2485 */
2379 2486 offset = uio->uio_loffset;
2380 2487 }
2381 2488
2382 2489 update:
2383 2490 zap_cursor_fini(&zc);
2384 2491 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2385 2492 kmem_free(outbuf, bufsize);
2386 2493
2387 2494 if (error == ENOENT)
2388 2495 error = 0;
2389 2496
2390 2497 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2391 2498
2392 2499 uio->uio_loffset = offset;
2393 2500 ZFS_EXIT(zfsvfs);
2394 2501 return (error);
2395 2502 }
2396 2503
2397 2504 ulong_t zfs_fsync_sync_cnt = 4;
2398 2505
2399 2506 static int
2400 2507 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2401 2508 {
2402 2509 znode_t *zp = VTOZ(vp);
2403 2510 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2404 2511
2405 2512 /*
2406 2513 * Regardless of whether this is required for standards conformance,
2407 2514 * this is the logical behavior when fsync() is called on a file with
2408 2515 * dirty pages. We use B_ASYNC since the ZIL transactions are already
2409 2516 * going to be pushed out as part of the zil_commit().
2410 2517 */
2411 2518 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2412 2519 (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2413 2520 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2414 2521
2415 2522 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2416 2523
2417 2524 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2418 2525 ZFS_ENTER(zfsvfs);
2419 2526 ZFS_VERIFY_ZP(zp);
2420 2527 zil_commit(zfsvfs->z_log, zp->z_id);
2421 2528 ZFS_EXIT(zfsvfs);
2422 2529 }
2423 2530 return (0);
2424 2531 }
2425 2532
2426 2533
2427 2534 /*
2428 2535 * Get the requested file attributes and place them in the provided
2429 2536 * vattr structure.
2430 2537 *
2431 2538 * IN: vp - vnode of file.
2432 2539 * vap - va_mask identifies requested attributes.
2433 2540 * If AT_XVATTR set, then optional attrs are requested
2434 2541 * flags - ATTR_NOACLCHECK (CIFS server context)
2435 2542 * cr - credentials of caller.
2436 2543 * ct - caller context
2437 2544 *
2438 2545 * OUT: vap - attribute values.
2439 2546 *
2440 2547 * RETURN: 0 (always succeeds).
2441 2548 */
2442 2549 /* ARGSUSED */
2443 2550 static int
2444 2551 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2445 2552 caller_context_t *ct)
2446 2553 {
2447 2554 znode_t *zp = VTOZ(vp);
2448 2555 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2449 2556 int error = 0;
2450 2557 uint64_t links;
2451 2558 uint64_t mtime[2], ctime[2];
2452 2559 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2453 2560 xoptattr_t *xoap = NULL;
2454 2561 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2455 2562 sa_bulk_attr_t bulk[2];
2456 2563 int count = 0;
2457 2564
2458 2565 ZFS_ENTER(zfsvfs);
2459 2566 ZFS_VERIFY_ZP(zp);
2460 2567
2461 2568 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2462 2569
2463 2570 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2464 2571 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2465 2572
2466 2573 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2467 2574 ZFS_EXIT(zfsvfs);
2468 2575 return (error);
2469 2576 }
2470 2577
2471 2578 /*
2472 2579 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2473 2580 * Also, if we are the owner don't bother, since owner should
2474 2581 * always be allowed to read basic attributes of file.
2475 2582 */
2476 2583 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2477 2584 (vap->va_uid != crgetuid(cr))) {
2478 2585 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2479 2586 skipaclchk, cr)) {
2480 2587 ZFS_EXIT(zfsvfs);
2481 2588 return (error);
2482 2589 }
2483 2590 }
2484 2591
2485 2592 /*
2486 2593 * Return all attributes. It's cheaper to provide the answer
2487 2594 * than to determine whether we were asked the question.
2488 2595 */
2489 2596
2490 2597 mutex_enter(&zp->z_lock);
2491 2598 vap->va_type = vp->v_type;
2492 2599 vap->va_mode = zp->z_mode & MODEMASK;
2493 2600 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2494 2601 vap->va_nodeid = zp->z_id;
2495 2602 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2496 2603 links = zp->z_links + 1;
2497 2604 else
2498 2605 links = zp->z_links;
2499 2606 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */
2500 2607 vap->va_size = zp->z_size;
2501 2608 vap->va_rdev = vp->v_rdev;
2502 2609 vap->va_seq = zp->z_seq;
2503 2610
2504 2611 /*
2505 2612 * Add in any requested optional attributes and the create time.
2506 2613 * Also set the corresponding bits in the returned attribute bitmap.
2507 2614 */
2508 2615 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2509 2616 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2510 2617 xoap->xoa_archive =
2511 2618 ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2512 2619 XVA_SET_RTN(xvap, XAT_ARCHIVE);
2513 2620 }
2514 2621
2515 2622 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2516 2623 xoap->xoa_readonly =
2517 2624 ((zp->z_pflags & ZFS_READONLY) != 0);
2518 2625 XVA_SET_RTN(xvap, XAT_READONLY);
2519 2626 }
2520 2627
2521 2628 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2522 2629 xoap->xoa_system =
2523 2630 ((zp->z_pflags & ZFS_SYSTEM) != 0);
2524 2631 XVA_SET_RTN(xvap, XAT_SYSTEM);
2525 2632 }
2526 2633
2527 2634 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2528 2635 xoap->xoa_hidden =
2529 2636 ((zp->z_pflags & ZFS_HIDDEN) != 0);
2530 2637 XVA_SET_RTN(xvap, XAT_HIDDEN);
2531 2638 }
2532 2639
2533 2640 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2534 2641 xoap->xoa_nounlink =
2535 2642 ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2536 2643 XVA_SET_RTN(xvap, XAT_NOUNLINK);
2537 2644 }
2538 2645
2539 2646 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2540 2647 xoap->xoa_immutable =
2541 2648 ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2542 2649 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2543 2650 }
2544 2651
2545 2652 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2546 2653 xoap->xoa_appendonly =
2547 2654 ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2548 2655 XVA_SET_RTN(xvap, XAT_APPENDONLY);
2549 2656 }
2550 2657
2551 2658 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2552 2659 xoap->xoa_nodump =
2553 2660 ((zp->z_pflags & ZFS_NODUMP) != 0);
2554 2661 XVA_SET_RTN(xvap, XAT_NODUMP);
2555 2662 }
2556 2663
2557 2664 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2558 2665 xoap->xoa_opaque =
2559 2666 ((zp->z_pflags & ZFS_OPAQUE) != 0);
2560 2667 XVA_SET_RTN(xvap, XAT_OPAQUE);
2561 2668 }
2562 2669
2563 2670 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2564 2671 xoap->xoa_av_quarantined =
2565 2672 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2566 2673 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2567 2674 }
2568 2675
2569 2676 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2570 2677 xoap->xoa_av_modified =
2571 2678 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2572 2679 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2573 2680 }
2574 2681
2575 2682 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2576 2683 vp->v_type == VREG) {
2577 2684 zfs_sa_get_scanstamp(zp, xvap);
2578 2685 }
2579 2686
2580 2687 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2581 2688 uint64_t times[2];
2582 2689
2583 2690 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2584 2691 times, sizeof (times));
2585 2692 ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2586 2693 XVA_SET_RTN(xvap, XAT_CREATETIME);
2587 2694 }
2588 2695
2589 2696 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2590 2697 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2591 2698 XVA_SET_RTN(xvap, XAT_REPARSE);
2592 2699 }
2593 2700 if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2594 2701 xoap->xoa_generation = zp->z_gen;
2595 2702 XVA_SET_RTN(xvap, XAT_GEN);
2596 2703 }
2597 2704
2598 2705 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2599 2706 xoap->xoa_offline =
2600 2707 ((zp->z_pflags & ZFS_OFFLINE) != 0);
2601 2708 XVA_SET_RTN(xvap, XAT_OFFLINE);
2602 2709 }
2603 2710
2604 2711 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2605 2712 xoap->xoa_sparse =
2606 2713 ((zp->z_pflags & ZFS_SPARSE) != 0);
2607 2714 XVA_SET_RTN(xvap, XAT_SPARSE);
2608 2715 }
2609 2716 }
2610 2717
2611 2718 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2612 2719 ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2613 2720 ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2614 2721
2615 2722 mutex_exit(&zp->z_lock);
2616 2723
2617 2724 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2618 2725
2619 2726 if (zp->z_blksz == 0) {
2620 2727 /*
2621 2728 * Block size hasn't been set; suggest maximal I/O transfers.
2622 2729 */
2623 2730 vap->va_blksize = zfsvfs->z_max_blksz;
2624 2731 }
2625 2732
2626 2733 ZFS_EXIT(zfsvfs);
2627 2734 return (0);
2628 2735 }
2629 2736
2630 2737 /*
2631 2738 * Set the file attributes to the values contained in the
2632 2739 * vattr structure.
2633 2740 *
2634 2741 * IN: vp - vnode of file to be modified.
2635 2742 * vap - new attribute values.
2636 2743 * If AT_XVATTR set, then optional attrs are being set
2637 2744 * flags - ATTR_UTIME set if non-default time values provided.
2638 2745 * - ATTR_NOACLCHECK (CIFS context only).
2639 2746 * cr - credentials of caller.
2640 2747 * ct - caller context
2641 2748 *
2642 2749 * RETURN: 0 on success, error code on failure.
2643 2750 *
2644 2751 * Timestamps:
2645 2752 * vp - ctime updated, mtime updated if size changed.
2646 2753 */
2647 2754 /* ARGSUSED */
2648 2755 static int
2649 2756 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2650 2757 caller_context_t *ct)
2651 2758 {
2652 2759 znode_t *zp = VTOZ(vp);
2653 2760 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2654 2761 zilog_t *zilog;
2655 2762 dmu_tx_t *tx;
2656 2763 vattr_t oldva;
2657 2764 xvattr_t tmpxvattr;
2658 2765 uint_t mask = vap->va_mask;
2659 2766 uint_t saved_mask = 0;
2660 2767 int trim_mask = 0;
2661 2768 uint64_t new_mode;
2662 2769 uint64_t new_uid, new_gid;
2663 2770 uint64_t xattr_obj;
2664 2771 uint64_t mtime[2], ctime[2];
2665 2772 znode_t *attrzp;
2666 2773 int need_policy = FALSE;
2667 2774 int err, err2;
2668 2775 zfs_fuid_info_t *fuidp = NULL;
2669 2776 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */
2670 2777 xoptattr_t *xoap;
2671 2778 zfs_acl_t *aclp;
2672 2779 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2673 2780 boolean_t fuid_dirtied = B_FALSE;
2674 2781 sa_bulk_attr_t bulk[7], xattr_bulk[7];
2675 2782 int count = 0, xattr_count = 0;
2676 2783
2677 2784 if (mask == 0)
2678 2785 return (0);
2679 2786
2680 2787 if (mask & AT_NOSET)
2681 2788 return (SET_ERROR(EINVAL));
2682 2789
2683 2790 ZFS_ENTER(zfsvfs);
2684 2791 ZFS_VERIFY_ZP(zp);
2685 2792
2686 2793 zilog = zfsvfs->z_log;
2687 2794
2688 2795 /*
2689 2796 * Make sure that if we have ephemeral uid/gid or xvattr specified
2690 2797 * that file system is at proper version level
2691 2798 */
2692 2799
2693 2800 if (zfsvfs->z_use_fuids == B_FALSE &&
2694 2801 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2695 2802 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2696 2803 (mask & AT_XVATTR))) {
2697 2804 ZFS_EXIT(zfsvfs);
2698 2805 return (SET_ERROR(EINVAL));
2699 2806 }
2700 2807
2701 2808 if (mask & AT_SIZE && vp->v_type == VDIR) {
2702 2809 ZFS_EXIT(zfsvfs);
2703 2810 return (SET_ERROR(EISDIR));
2704 2811 }
2705 2812
2706 2813 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2707 2814 ZFS_EXIT(zfsvfs);
2708 2815 return (SET_ERROR(EINVAL));
2709 2816 }
2710 2817
2711 2818 /*
2712 2819 * If this is an xvattr_t, then get a pointer to the structure of
2713 2820 * optional attributes. If this is NULL, then we have a vattr_t.
2714 2821 */
2715 2822 xoap = xva_getxoptattr(xvap);
2716 2823
2717 2824 xva_init(&tmpxvattr);
2718 2825
2719 2826 /*
2720 2827 * Immutable files can only alter immutable bit and atime
2721 2828 */
2722 2829 if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2723 2830 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2724 2831 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2725 2832 ZFS_EXIT(zfsvfs);
2726 2833 return (SET_ERROR(EPERM));
2727 2834 }
2728 2835
2729 2836 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2730 2837 ZFS_EXIT(zfsvfs);
2731 2838 return (SET_ERROR(EPERM));
2732 2839 }
2733 2840
2734 2841 /*
2735 2842 * Verify timestamps doesn't overflow 32 bits.
2736 2843 * ZFS can handle large timestamps, but 32bit syscalls can't
2737 2844 * handle times greater than 2039. This check should be removed
2738 2845 * once large timestamps are fully supported.
2739 2846 */
2740 2847 if (mask & (AT_ATIME | AT_MTIME)) {
2741 2848 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2742 2849 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2743 2850 ZFS_EXIT(zfsvfs);
2744 2851 return (SET_ERROR(EOVERFLOW));
2745 2852 }
2746 2853 }
2747 2854
2748 2855 top:
2749 2856 attrzp = NULL;
2750 2857 aclp = NULL;
2751 2858
2752 2859 /* Can this be moved to before the top label? */
2753 2860 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2754 2861 ZFS_EXIT(zfsvfs);
2755 2862 return (SET_ERROR(EROFS));
2756 2863 }
2757 2864
2758 2865 /*
2759 2866 * First validate permissions
2760 2867 */
2761 2868
2762 2869 if (mask & AT_SIZE) {
2763 2870 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2764 2871 if (err) {
2765 2872 ZFS_EXIT(zfsvfs);
2766 2873 return (err);
2767 2874 }
2768 2875 /*
2769 2876 * XXX - Note, we are not providing any open
2770 2877 * mode flags here (like FNDELAY), so we may
2771 2878 * block if there are locks present... this
2772 2879 * should be addressed in openat().
2773 2880 */
2774 2881 /* XXX - would it be OK to generate a log record here? */
2775 2882 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2776 2883 if (err) {
2777 2884 ZFS_EXIT(zfsvfs);
2778 2885 return (err);
2779 2886 }
2780 2887
2781 2888 if (vap->va_size == 0)
2782 2889 vnevent_truncate(ZTOV(zp), ct);
2783 2890 }
2784 2891
2785 2892 if (mask & (AT_ATIME|AT_MTIME) ||
2786 2893 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2787 2894 XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2788 2895 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2789 2896 XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2790 2897 XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2791 2898 XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2792 2899 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2793 2900 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2794 2901 skipaclchk, cr);
2795 2902 }
2796 2903
2797 2904 if (mask & (AT_UID|AT_GID)) {
2798 2905 int idmask = (mask & (AT_UID|AT_GID));
2799 2906 int take_owner;
2800 2907 int take_group;
2801 2908
2802 2909 /*
2803 2910 * NOTE: even if a new mode is being set,
2804 2911 * we may clear S_ISUID/S_ISGID bits.
2805 2912 */
2806 2913
2807 2914 if (!(mask & AT_MODE))
2808 2915 vap->va_mode = zp->z_mode;
2809 2916
2810 2917 /*
2811 2918 * Take ownership or chgrp to group we are a member of
2812 2919 */
2813 2920
2814 2921 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2815 2922 take_group = (mask & AT_GID) &&
2816 2923 zfs_groupmember(zfsvfs, vap->va_gid, cr);
2817 2924
2818 2925 /*
2819 2926 * If both AT_UID and AT_GID are set then take_owner and
2820 2927 * take_group must both be set in order to allow taking
2821 2928 * ownership.
2822 2929 *
2823 2930 * Otherwise, send the check through secpolicy_vnode_setattr()
2824 2931 *
2825 2932 */
2826 2933
2827 2934 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2828 2935 ((idmask == AT_UID) && take_owner) ||
2829 2936 ((idmask == AT_GID) && take_group)) {
2830 2937 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2831 2938 skipaclchk, cr) == 0) {
2832 2939 /*
2833 2940 * Remove setuid/setgid for non-privileged users
2834 2941 */
2835 2942 secpolicy_setid_clear(vap, cr);
2836 2943 trim_mask = (mask & (AT_UID|AT_GID));
2837 2944 } else {
2838 2945 need_policy = TRUE;
2839 2946 }
2840 2947 } else {
2841 2948 need_policy = TRUE;
2842 2949 }
2843 2950 }
2844 2951
2845 2952 mutex_enter(&zp->z_lock);
2846 2953 oldva.va_mode = zp->z_mode;
2847 2954 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2848 2955 if (mask & AT_XVATTR) {
2849 2956 /*
2850 2957 * Update xvattr mask to include only those attributes
2851 2958 * that are actually changing.
2852 2959 *
2853 2960 * the bits will be restored prior to actually setting
2854 2961 * the attributes so the caller thinks they were set.
2855 2962 */
2856 2963 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2857 2964 if (xoap->xoa_appendonly !=
2858 2965 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2859 2966 need_policy = TRUE;
2860 2967 } else {
2861 2968 XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2862 2969 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2863 2970 }
2864 2971 }
2865 2972
2866 2973 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2867 2974 if (xoap->xoa_nounlink !=
2868 2975 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2869 2976 need_policy = TRUE;
2870 2977 } else {
2871 2978 XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2872 2979 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2873 2980 }
2874 2981 }
2875 2982
2876 2983 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2877 2984 if (xoap->xoa_immutable !=
2878 2985 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2879 2986 need_policy = TRUE;
2880 2987 } else {
2881 2988 XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2882 2989 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2883 2990 }
2884 2991 }
2885 2992
2886 2993 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2887 2994 if (xoap->xoa_nodump !=
2888 2995 ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2889 2996 need_policy = TRUE;
2890 2997 } else {
2891 2998 XVA_CLR_REQ(xvap, XAT_NODUMP);
2892 2999 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2893 3000 }
2894 3001 }
2895 3002
2896 3003 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2897 3004 if (xoap->xoa_av_modified !=
2898 3005 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2899 3006 need_policy = TRUE;
2900 3007 } else {
2901 3008 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2902 3009 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2903 3010 }
2904 3011 }
2905 3012
2906 3013 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2907 3014 if ((vp->v_type != VREG &&
2908 3015 xoap->xoa_av_quarantined) ||
2909 3016 xoap->xoa_av_quarantined !=
2910 3017 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2911 3018 need_policy = TRUE;
2912 3019 } else {
2913 3020 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2914 3021 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2915 3022 }
2916 3023 }
2917 3024
2918 3025 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2919 3026 mutex_exit(&zp->z_lock);
2920 3027 ZFS_EXIT(zfsvfs);
2921 3028 return (SET_ERROR(EPERM));
2922 3029 }
2923 3030
2924 3031 if (need_policy == FALSE &&
2925 3032 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2926 3033 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2927 3034 need_policy = TRUE;
2928 3035 }
2929 3036 }
2930 3037
2931 3038 mutex_exit(&zp->z_lock);
2932 3039
2933 3040 if (mask & AT_MODE) {
2934 3041 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2935 3042 err = secpolicy_setid_setsticky_clear(vp, vap,
2936 3043 &oldva, cr);
2937 3044 if (err) {
2938 3045 ZFS_EXIT(zfsvfs);
2939 3046 return (err);
2940 3047 }
2941 3048 trim_mask |= AT_MODE;
2942 3049 } else {
2943 3050 need_policy = TRUE;
2944 3051 }
2945 3052 }
2946 3053
2947 3054 if (need_policy) {
2948 3055 /*
2949 3056 * If trim_mask is set then take ownership
2950 3057 * has been granted or write_acl is present and user
2951 3058 * has the ability to modify mode. In that case remove
2952 3059 * UID|GID and or MODE from mask so that
2953 3060 * secpolicy_vnode_setattr() doesn't revoke it.
2954 3061 */
2955 3062
2956 3063 if (trim_mask) {
2957 3064 saved_mask = vap->va_mask;
2958 3065 vap->va_mask &= ~trim_mask;
2959 3066 }
2960 3067 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2961 3068 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2962 3069 if (err) {
2963 3070 ZFS_EXIT(zfsvfs);
2964 3071 return (err);
2965 3072 }
2966 3073
2967 3074 if (trim_mask)
2968 3075 vap->va_mask |= saved_mask;
2969 3076 }
2970 3077
2971 3078 /*
2972 3079 * secpolicy_vnode_setattr, or take ownership may have
2973 3080 * changed va_mask
2974 3081 */
2975 3082 mask = vap->va_mask;
2976 3083
2977 3084 if ((mask & (AT_UID | AT_GID))) {
2978 3085 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2979 3086 &xattr_obj, sizeof (xattr_obj));
2980 3087
2981 3088 if (err == 0 && xattr_obj) {
2982 3089 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
2983 3090 if (err)
2984 3091 goto out2;
2985 3092 }
2986 3093 if (mask & AT_UID) {
2987 3094 new_uid = zfs_fuid_create(zfsvfs,
2988 3095 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2989 3096 if (new_uid != zp->z_uid &&
2990 3097 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
2991 3098 if (attrzp)
2992 3099 VN_RELE(ZTOV(attrzp));
2993 3100 err = SET_ERROR(EDQUOT);
2994 3101 goto out2;
2995 3102 }
2996 3103 }
2997 3104
2998 3105 if (mask & AT_GID) {
2999 3106 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3000 3107 cr, ZFS_GROUP, &fuidp);
3001 3108 if (new_gid != zp->z_gid &&
3002 3109 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3003 3110 if (attrzp)
3004 3111 VN_RELE(ZTOV(attrzp));
3005 3112 err = SET_ERROR(EDQUOT);
3006 3113 goto out2;
3007 3114 }
3008 3115 }
3009 3116 }
3010 3117 tx = dmu_tx_create(zfsvfs->z_os);
3011 3118
3012 3119 if (mask & AT_MODE) {
3013 3120 uint64_t pmode = zp->z_mode;
3014 3121 uint64_t acl_obj;
3015 3122 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3016 3123
3017 3124 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3018 3125 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3019 3126 err = SET_ERROR(EPERM);
3020 3127 goto out;
3021 3128 }
3022 3129
3023 3130 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3024 3131 goto out;
3025 3132
3026 3133 mutex_enter(&zp->z_lock);
3027 3134 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3028 3135 /*
3029 3136 * Are we upgrading ACL from old V0 format
3030 3137 * to V1 format?
3031 3138 */
3032 3139 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3033 3140 zfs_znode_acl_version(zp) ==
3034 3141 ZFS_ACL_VERSION_INITIAL) {
3035 3142 dmu_tx_hold_free(tx, acl_obj, 0,
3036 3143 DMU_OBJECT_END);
3037 3144 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3038 3145 0, aclp->z_acl_bytes);
3039 3146 } else {
3040 3147 dmu_tx_hold_write(tx, acl_obj, 0,
3041 3148 aclp->z_acl_bytes);
3042 3149 }
3043 3150 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3044 3151 dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3045 3152 0, aclp->z_acl_bytes);
3046 3153 }
3047 3154 mutex_exit(&zp->z_lock);
3048 3155 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3049 3156 } else {
3050 3157 if ((mask & AT_XVATTR) &&
3051 3158 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3052 3159 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3053 3160 else
3054 3161 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3055 3162 }
3056 3163
3057 3164 if (attrzp) {
3058 3165 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3059 3166 }
3060 3167
3061 3168 fuid_dirtied = zfsvfs->z_fuid_dirty;
3062 3169 if (fuid_dirtied)
3063 3170 zfs_fuid_txhold(zfsvfs, tx);
3064 3171
3065 3172 zfs_sa_upgrade_txholds(tx, zp);
3066 3173
3067 3174 err = dmu_tx_assign(tx, TXG_WAIT);
3068 3175 if (err)
3069 3176 goto out;
3070 3177
3071 3178 count = 0;
3072 3179 /*
3073 3180 * Set each attribute requested.
3074 3181 * We group settings according to the locks they need to acquire.
3075 3182 *
3076 3183 * Note: you cannot set ctime directly, although it will be
3077 3184 * updated as a side-effect of calling this function.
3078 3185 */
3079 3186
3080 3187
3081 3188 if (mask & (AT_UID|AT_GID|AT_MODE))
3082 3189 mutex_enter(&zp->z_acl_lock);
3083 3190 mutex_enter(&zp->z_lock);
3084 3191
3085 3192 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3086 3193 &zp->z_pflags, sizeof (zp->z_pflags));
3087 3194
3088 3195 if (attrzp) {
3089 3196 if (mask & (AT_UID|AT_GID|AT_MODE))
3090 3197 mutex_enter(&attrzp->z_acl_lock);
3091 3198 mutex_enter(&attrzp->z_lock);
3092 3199 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3093 3200 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3094 3201 sizeof (attrzp->z_pflags));
3095 3202 }
3096 3203
3097 3204 if (mask & (AT_UID|AT_GID)) {
3098 3205
3099 3206 if (mask & AT_UID) {
3100 3207 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3101 3208 &new_uid, sizeof (new_uid));
3102 3209 zp->z_uid = new_uid;
3103 3210 if (attrzp) {
3104 3211 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3105 3212 SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3106 3213 sizeof (new_uid));
3107 3214 attrzp->z_uid = new_uid;
3108 3215 }
3109 3216 }
3110 3217
3111 3218 if (mask & AT_GID) {
3112 3219 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3113 3220 NULL, &new_gid, sizeof (new_gid));
3114 3221 zp->z_gid = new_gid;
3115 3222 if (attrzp) {
3116 3223 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3117 3224 SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3118 3225 sizeof (new_gid));
3119 3226 attrzp->z_gid = new_gid;
3120 3227 }
3121 3228 }
3122 3229 if (!(mask & AT_MODE)) {
3123 3230 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3124 3231 NULL, &new_mode, sizeof (new_mode));
3125 3232 new_mode = zp->z_mode;
3126 3233 }
3127 3234 err = zfs_acl_chown_setattr(zp);
3128 3235 ASSERT(err == 0);
3129 3236 if (attrzp) {
3130 3237 err = zfs_acl_chown_setattr(attrzp);
3131 3238 ASSERT(err == 0);
3132 3239 }
3133 3240 }
3134 3241
3135 3242 if (mask & AT_MODE) {
3136 3243 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3137 3244 &new_mode, sizeof (new_mode));
3138 3245 zp->z_mode = new_mode;
3139 3246 ASSERT3U((uintptr_t)aclp, !=, NULL);
3140 3247 err = zfs_aclset_common(zp, aclp, cr, tx);
3141 3248 ASSERT0(err);
3142 3249 if (zp->z_acl_cached)
3143 3250 zfs_acl_free(zp->z_acl_cached);
3144 3251 zp->z_acl_cached = aclp;
3145 3252 aclp = NULL;
3146 3253 }
3147 3254
3148 3255
3149 3256 if (mask & AT_ATIME) {
3150 3257 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3151 3258 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3152 3259 &zp->z_atime, sizeof (zp->z_atime));
3153 3260 }
3154 3261
3155 3262 if (mask & AT_MTIME) {
3156 3263 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3157 3264 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3158 3265 mtime, sizeof (mtime));
3159 3266 }
3160 3267
3161 3268 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3162 3269 if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3163 3270 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3164 3271 NULL, mtime, sizeof (mtime));
3165 3272 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3166 3273 &ctime, sizeof (ctime));
3167 3274 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3168 3275 B_TRUE);
3169 3276 } else if (mask != 0) {
3170 3277 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3171 3278 &ctime, sizeof (ctime));
3172 3279 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3173 3280 B_TRUE);
3174 3281 if (attrzp) {
3175 3282 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3176 3283 SA_ZPL_CTIME(zfsvfs), NULL,
3177 3284 &ctime, sizeof (ctime));
3178 3285 zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3179 3286 mtime, ctime, B_TRUE);
3180 3287 }
3181 3288 }
3182 3289 /*
3183 3290 * Do this after setting timestamps to prevent timestamp
3184 3291 * update from toggling bit
3185 3292 */
3186 3293
3187 3294 if (xoap && (mask & AT_XVATTR)) {
3188 3295
3189 3296 /*
3190 3297 * restore trimmed off masks
3191 3298 * so that return masks can be set for caller.
3192 3299 */
3193 3300
3194 3301 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3195 3302 XVA_SET_REQ(xvap, XAT_APPENDONLY);
3196 3303 }
3197 3304 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3198 3305 XVA_SET_REQ(xvap, XAT_NOUNLINK);
3199 3306 }
3200 3307 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3201 3308 XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3202 3309 }
3203 3310 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3204 3311 XVA_SET_REQ(xvap, XAT_NODUMP);
3205 3312 }
3206 3313 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3207 3314 XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3208 3315 }
3209 3316 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3210 3317 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3211 3318 }
3212 3319
3213 3320 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3214 3321 ASSERT(vp->v_type == VREG);
3215 3322
3216 3323 zfs_xvattr_set(zp, xvap, tx);
3217 3324 }
3218 3325
3219 3326 if (fuid_dirtied)
3220 3327 zfs_fuid_sync(zfsvfs, tx);
3221 3328
3222 3329 if (mask != 0)
3223 3330 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3224 3331
3225 3332 mutex_exit(&zp->z_lock);
3226 3333 if (mask & (AT_UID|AT_GID|AT_MODE))
3227 3334 mutex_exit(&zp->z_acl_lock);
3228 3335
3229 3336 if (attrzp) {
3230 3337 if (mask & (AT_UID|AT_GID|AT_MODE))
3231 3338 mutex_exit(&attrzp->z_acl_lock);
3232 3339 mutex_exit(&attrzp->z_lock);
3233 3340 }
3234 3341 out:
3235 3342 if (err == 0 && attrzp) {
3236 3343 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3237 3344 xattr_count, tx);
3238 3345 ASSERT(err2 == 0);
3239 3346 }
3240 3347
3241 3348 if (attrzp)
3242 3349 VN_RELE(ZTOV(attrzp));
3243 3350
3244 3351 if (aclp)
3245 3352 zfs_acl_free(aclp);
3246 3353
3247 3354 if (fuidp) {
3248 3355 zfs_fuid_info_free(fuidp);
3249 3356 fuidp = NULL;
3250 3357 }
3251 3358
3252 3359 if (err) {
3253 3360 dmu_tx_abort(tx);
3254 3361 if (err == ERESTART)
3255 3362 goto top;
3256 3363 } else {
3257 3364 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3258 3365 dmu_tx_commit(tx);
3259 3366 }
3260 3367
3261 3368 out2:
3262 3369 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3263 3370 zil_commit(zilog, 0);
3264 3371
3265 3372 ZFS_EXIT(zfsvfs);
3266 3373 return (err);
3267 3374 }
3268 3375
3269 3376 typedef struct zfs_zlock {
3270 3377 krwlock_t *zl_rwlock; /* lock we acquired */
3271 3378 znode_t *zl_znode; /* znode we held */
3272 3379 struct zfs_zlock *zl_next; /* next in list */
3273 3380 } zfs_zlock_t;
3274 3381
3275 3382 /*
3276 3383 * Drop locks and release vnodes that were held by zfs_rename_lock().
3277 3384 */
3278 3385 static void
3279 3386 zfs_rename_unlock(zfs_zlock_t **zlpp)
3280 3387 {
3281 3388 zfs_zlock_t *zl;
3282 3389
3283 3390 while ((zl = *zlpp) != NULL) {
3284 3391 if (zl->zl_znode != NULL)
3285 3392 VN_RELE(ZTOV(zl->zl_znode));
3286 3393 rw_exit(zl->zl_rwlock);
3287 3394 *zlpp = zl->zl_next;
3288 3395 kmem_free(zl, sizeof (*zl));
3289 3396 }
3290 3397 }
3291 3398
3292 3399 /*
3293 3400 * Search back through the directory tree, using the ".." entries.
3294 3401 * Lock each directory in the chain to prevent concurrent renames.
3295 3402 * Fail any attempt to move a directory into one of its own descendants.
3296 3403 * XXX - z_parent_lock can overlap with map or grow locks
3297 3404 */
3298 3405 static int
3299 3406 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3300 3407 {
3301 3408 zfs_zlock_t *zl;
3302 3409 znode_t *zp = tdzp;
3303 3410 uint64_t rootid = zp->z_zfsvfs->z_root;
3304 3411 uint64_t oidp = zp->z_id;
3305 3412 krwlock_t *rwlp = &szp->z_parent_lock;
3306 3413 krw_t rw = RW_WRITER;
3307 3414
3308 3415 /*
3309 3416 * First pass write-locks szp and compares to zp->z_id.
3310 3417 * Later passes read-lock zp and compare to zp->z_parent.
3311 3418 */
3312 3419 do {
3313 3420 if (!rw_tryenter(rwlp, rw)) {
3314 3421 /*
3315 3422 * Another thread is renaming in this path.
3316 3423 * Note that if we are a WRITER, we don't have any
3317 3424 * parent_locks held yet.
3318 3425 */
3319 3426 if (rw == RW_READER && zp->z_id > szp->z_id) {
3320 3427 /*
3321 3428 * Drop our locks and restart
3322 3429 */
3323 3430 zfs_rename_unlock(&zl);
3324 3431 *zlpp = NULL;
3325 3432 zp = tdzp;
3326 3433 oidp = zp->z_id;
3327 3434 rwlp = &szp->z_parent_lock;
3328 3435 rw = RW_WRITER;
3329 3436 continue;
3330 3437 } else {
3331 3438 /*
3332 3439 * Wait for other thread to drop its locks
3333 3440 */
3334 3441 rw_enter(rwlp, rw);
3335 3442 }
3336 3443 }
3337 3444
3338 3445 zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3339 3446 zl->zl_rwlock = rwlp;
3340 3447 zl->zl_znode = NULL;
3341 3448 zl->zl_next = *zlpp;
3342 3449 *zlpp = zl;
3343 3450
3344 3451 if (oidp == szp->z_id) /* We're a descendant of szp */
3345 3452 return (SET_ERROR(EINVAL));
3346 3453
3347 3454 if (oidp == rootid) /* We've hit the top */
3348 3455 return (0);
3349 3456
3350 3457 if (rw == RW_READER) { /* i.e. not the first pass */
3351 3458 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3352 3459 if (error)
3353 3460 return (error);
3354 3461 zl->zl_znode = zp;
3355 3462 }
3356 3463 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3357 3464 &oidp, sizeof (oidp));
3358 3465 rwlp = &zp->z_parent_lock;
3359 3466 rw = RW_READER;
3360 3467
3361 3468 } while (zp->z_id != sdzp->z_id);
3362 3469
3363 3470 return (0);
3364 3471 }
3365 3472
3366 3473 /*
3367 3474 * Move an entry from the provided source directory to the target
3368 3475 * directory. Change the entry name as indicated.
3369 3476 *
3370 3477 * IN: sdvp - Source directory containing the "old entry".
3371 3478 * snm - Old entry name.
3372 3479 * tdvp - Target directory to contain the "new entry".
3373 3480 * tnm - New entry name.
3374 3481 * cr - credentials of caller.
3375 3482 * ct - caller context
3376 3483 * flags - case flags
3377 3484 *
3378 3485 * RETURN: 0 on success, error code on failure.
3379 3486 *
3380 3487 * Timestamps:
3381 3488 * sdvp,tdvp - ctime|mtime updated
3382 3489 */
3383 3490 /*ARGSUSED*/
3384 3491 static int
3385 3492 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3386 3493 caller_context_t *ct, int flags)
3387 3494 {
3388 3495 znode_t *tdzp, *szp, *tzp;
3389 3496 znode_t *sdzp = VTOZ(sdvp);
3390 3497 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs;
3391 3498 zilog_t *zilog;
3392 3499 vnode_t *realvp;
3393 3500 zfs_dirlock_t *sdl, *tdl;
3394 3501 dmu_tx_t *tx;
3395 3502 zfs_zlock_t *zl;
3396 3503 int cmp, serr, terr;
3397 3504 int error = 0;
3398 3505 int zflg = 0;
3399 3506 boolean_t waited = B_FALSE;
3400 3507
3401 3508 ZFS_ENTER(zfsvfs);
3402 3509 ZFS_VERIFY_ZP(sdzp);
3403 3510 zilog = zfsvfs->z_log;
3404 3511
3405 3512 /*
3406 3513 * Make sure we have the real vp for the target directory.
3407 3514 */
3408 3515 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3409 3516 tdvp = realvp;
3410 3517
3411 3518 tdzp = VTOZ(tdvp);
3412 3519 ZFS_VERIFY_ZP(tdzp);
3413 3520
3414 3521 /*
3415 3522 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3416 3523 * ctldir appear to have the same v_vfsp.
3417 3524 */
3418 3525 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3419 3526 ZFS_EXIT(zfsvfs);
3420 3527 return (SET_ERROR(EXDEV));
3421 3528 }
3422 3529
3423 3530 if (zfsvfs->z_utf8 && u8_validate(tnm,
3424 3531 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3425 3532 ZFS_EXIT(zfsvfs);
3426 3533 return (SET_ERROR(EILSEQ));
3427 3534 }
3428 3535
3429 3536 if (flags & FIGNORECASE)
3430 3537 zflg |= ZCILOOK;
3431 3538
3432 3539 top:
3433 3540 szp = NULL;
3434 3541 tzp = NULL;
3435 3542 zl = NULL;
3436 3543
3437 3544 /*
3438 3545 * This is to prevent the creation of links into attribute space
3439 3546 * by renaming a linked file into/outof an attribute directory.
3440 3547 * See the comment in zfs_link() for why this is considered bad.
3441 3548 */
3442 3549 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3443 3550 ZFS_EXIT(zfsvfs);
3444 3551 return (SET_ERROR(EINVAL));
3445 3552 }
3446 3553
3447 3554 /*
3448 3555 * Lock source and target directory entries. To prevent deadlock,
3449 3556 * a lock ordering must be defined. We lock the directory with
3450 3557 * the smallest object id first, or if it's a tie, the one with
3451 3558 * the lexically first name.
3452 3559 */
3453 3560 if (sdzp->z_id < tdzp->z_id) {
3454 3561 cmp = -1;
3455 3562 } else if (sdzp->z_id > tdzp->z_id) {
3456 3563 cmp = 1;
3457 3564 } else {
3458 3565 /*
3459 3566 * First compare the two name arguments without
3460 3567 * considering any case folding.
3461 3568 */
3462 3569 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3463 3570
3464 3571 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3465 3572 ASSERT(error == 0 || !zfsvfs->z_utf8);
3466 3573 if (cmp == 0) {
3467 3574 /*
3468 3575 * POSIX: "If the old argument and the new argument
3469 3576 * both refer to links to the same existing file,
3470 3577 * the rename() function shall return successfully
3471 3578 * and perform no other action."
3472 3579 */
3473 3580 ZFS_EXIT(zfsvfs);
3474 3581 return (0);
3475 3582 }
3476 3583 /*
3477 3584 * If the file system is case-folding, then we may
3478 3585 * have some more checking to do. A case-folding file
3479 3586 * system is either supporting mixed case sensitivity
3480 3587 * access or is completely case-insensitive. Note
3481 3588 * that the file system is always case preserving.
3482 3589 *
3483 3590 * In mixed sensitivity mode case sensitive behavior
3484 3591 * is the default. FIGNORECASE must be used to
3485 3592 * explicitly request case insensitive behavior.
3486 3593 *
3487 3594 * If the source and target names provided differ only
3488 3595 * by case (e.g., a request to rename 'tim' to 'Tim'),
3489 3596 * we will treat this as a special case in the
3490 3597 * case-insensitive mode: as long as the source name
3491 3598 * is an exact match, we will allow this to proceed as
3492 3599 * a name-change request.
3493 3600 */
3494 3601 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3495 3602 (zfsvfs->z_case == ZFS_CASE_MIXED &&
3496 3603 flags & FIGNORECASE)) &&
3497 3604 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3498 3605 &error) == 0) {
3499 3606 /*
3500 3607 * case preserving rename request, require exact
3501 3608 * name matches
3502 3609 */
3503 3610 zflg |= ZCIEXACT;
3504 3611 zflg &= ~ZCILOOK;
3505 3612 }
3506 3613 }
3507 3614
3508 3615 /*
3509 3616 * If the source and destination directories are the same, we should
3510 3617 * grab the z_name_lock of that directory only once.
3511 3618 */
3512 3619 if (sdzp == tdzp) {
3513 3620 zflg |= ZHAVELOCK;
3514 3621 rw_enter(&sdzp->z_name_lock, RW_READER);
3515 3622 }
3516 3623
3517 3624 if (cmp < 0) {
3518 3625 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3519 3626 ZEXISTS | zflg, NULL, NULL);
3520 3627 terr = zfs_dirent_lock(&tdl,
3521 3628 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3522 3629 } else {
3523 3630 terr = zfs_dirent_lock(&tdl,
3524 3631 tdzp, tnm, &tzp, zflg, NULL, NULL);
3525 3632 serr = zfs_dirent_lock(&sdl,
3526 3633 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3527 3634 NULL, NULL);
3528 3635 }
3529 3636
3530 3637 if (serr) {
3531 3638 /*
3532 3639 * Source entry invalid or not there.
3533 3640 */
3534 3641 if (!terr) {
3535 3642 zfs_dirent_unlock(tdl);
3536 3643 if (tzp)
3537 3644 VN_RELE(ZTOV(tzp));
3538 3645 }
3539 3646
3540 3647 if (sdzp == tdzp)
3541 3648 rw_exit(&sdzp->z_name_lock);
3542 3649
3543 3650 if (strcmp(snm, "..") == 0)
3544 3651 serr = SET_ERROR(EINVAL);
3545 3652 ZFS_EXIT(zfsvfs);
3546 3653 return (serr);
3547 3654 }
3548 3655 if (terr) {
3549 3656 zfs_dirent_unlock(sdl);
3550 3657 VN_RELE(ZTOV(szp));
3551 3658
3552 3659 if (sdzp == tdzp)
3553 3660 rw_exit(&sdzp->z_name_lock);
3554 3661
3555 3662 if (strcmp(tnm, "..") == 0)
3556 3663 terr = SET_ERROR(EINVAL);
3557 3664 ZFS_EXIT(zfsvfs);
3558 3665 return (terr);
3559 3666 }
3560 3667
3561 3668 /*
3562 3669 * Must have write access at the source to remove the old entry
3563 3670 * and write access at the target to create the new entry.
3564 3671 * Note that if target and source are the same, this can be
3565 3672 * done in a single check.
3566 3673 */
3567 3674
3568 3675 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3569 3676 goto out;
3570 3677
3571 3678 if (ZTOV(szp)->v_type == VDIR) {
3572 3679 /*
3573 3680 * Check to make sure rename is valid.
3574 3681 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3575 3682 */
3576 3683 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3577 3684 goto out;
3578 3685 }
3579 3686
3580 3687 /*
3581 3688 * Does target exist?
3582 3689 */
3583 3690 if (tzp) {
3584 3691 /*
3585 3692 * Source and target must be the same type.
3586 3693 */
3587 3694 if (ZTOV(szp)->v_type == VDIR) {
3588 3695 if (ZTOV(tzp)->v_type != VDIR) {
3589 3696 error = SET_ERROR(ENOTDIR);
3590 3697 goto out;
3591 3698 }
3592 3699 } else {
3593 3700 if (ZTOV(tzp)->v_type == VDIR) {
3594 3701 error = SET_ERROR(EISDIR);
3595 3702 goto out;
3596 3703 }
3597 3704 }
3598 3705 /*
3599 3706 * POSIX dictates that when the source and target
3600 3707 * entries refer to the same file object, rename
3601 3708 * must do nothing and exit without error.
3602 3709 */
3603 3710 if (szp->z_id == tzp->z_id) {
3604 3711 error = 0;
3605 3712 goto out;
3606 3713 }
3607 3714 }
3608 3715
3609 3716 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3610 3717 if (tzp)
3611 3718 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3612 3719
3613 3720 /*
3614 3721 * notify the target directory if it is not the same
3615 3722 * as source directory.
3616 3723 */
3617 3724 if (tdvp != sdvp) {
3618 3725 vnevent_rename_dest_dir(tdvp, ct);
3619 3726 }
3620 3727
3621 3728 tx = dmu_tx_create(zfsvfs->z_os);
3622 3729 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3623 3730 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3624 3731 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3625 3732 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3626 3733 if (sdzp != tdzp) {
3627 3734 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3628 3735 zfs_sa_upgrade_txholds(tx, tdzp);
3629 3736 }
3630 3737 if (tzp) {
3631 3738 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3632 3739 zfs_sa_upgrade_txholds(tx, tzp);
3633 3740 }
3634 3741
3635 3742 zfs_sa_upgrade_txholds(tx, szp);
3636 3743 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3637 3744 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3638 3745 if (error) {
3639 3746 if (zl != NULL)
3640 3747 zfs_rename_unlock(&zl);
3641 3748 zfs_dirent_unlock(sdl);
3642 3749 zfs_dirent_unlock(tdl);
3643 3750
3644 3751 if (sdzp == tdzp)
3645 3752 rw_exit(&sdzp->z_name_lock);
3646 3753
3647 3754 VN_RELE(ZTOV(szp));
3648 3755 if (tzp)
3649 3756 VN_RELE(ZTOV(tzp));
3650 3757 if (error == ERESTART) {
3651 3758 waited = B_TRUE;
3652 3759 dmu_tx_wait(tx);
3653 3760 dmu_tx_abort(tx);
3654 3761 goto top;
3655 3762 }
3656 3763 dmu_tx_abort(tx);
3657 3764 ZFS_EXIT(zfsvfs);
3658 3765 return (error);
3659 3766 }
3660 3767
3661 3768 if (tzp) /* Attempt to remove the existing target */
3662 3769 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3663 3770
3664 3771 if (error == 0) {
3665 3772 error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3666 3773 if (error == 0) {
3667 3774 szp->z_pflags |= ZFS_AV_MODIFIED;
3668 3775
3669 3776 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3670 3777 (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3671 3778 ASSERT0(error);
3672 3779
3673 3780 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3674 3781 if (error == 0) {
3675 3782 zfs_log_rename(zilog, tx, TX_RENAME |
3676 3783 (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3677 3784 sdl->dl_name, tdzp, tdl->dl_name, szp);
3678 3785
3679 3786 /*
3680 3787 * Update path information for the target vnode
3681 3788 */
3682 3789 vn_renamepath(tdvp, ZTOV(szp), tnm,
3683 3790 strlen(tnm));
3684 3791 } else {
3685 3792 /*
3686 3793 * At this point, we have successfully created
3687 3794 * the target name, but have failed to remove
3688 3795 * the source name. Since the create was done
3689 3796 * with the ZRENAMING flag, there are
3690 3797 * complications; for one, the link count is
3691 3798 * wrong. The easiest way to deal with this
3692 3799 * is to remove the newly created target, and
3693 3800 * return the original error. This must
3694 3801 * succeed; fortunately, it is very unlikely to
3695 3802 * fail, since we just created it.
3696 3803 */
3697 3804 VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3698 3805 ZRENAMING, NULL), ==, 0);
3699 3806 }
3700 3807 }
3701 3808 }
3702 3809
3703 3810 dmu_tx_commit(tx);
3704 3811 out:
3705 3812 if (zl != NULL)
3706 3813 zfs_rename_unlock(&zl);
3707 3814
3708 3815 zfs_dirent_unlock(sdl);
3709 3816 zfs_dirent_unlock(tdl);
3710 3817
3711 3818 if (sdzp == tdzp)
3712 3819 rw_exit(&sdzp->z_name_lock);
3713 3820
3714 3821
3715 3822 VN_RELE(ZTOV(szp));
3716 3823 if (tzp)
3717 3824 VN_RELE(ZTOV(tzp));
3718 3825
3719 3826 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3720 3827 zil_commit(zilog, 0);
3721 3828
3722 3829 ZFS_EXIT(zfsvfs);
3723 3830 return (error);
3724 3831 }
3725 3832
3726 3833 /*
3727 3834 * Insert the indicated symbolic reference entry into the directory.
3728 3835 *
3729 3836 * IN: dvp - Directory to contain new symbolic link.
3730 3837 * link - Name for new symlink entry.
3731 3838 * vap - Attributes of new entry.
3732 3839 * cr - credentials of caller.
3733 3840 * ct - caller context
3734 3841 * flags - case flags
3735 3842 *
3736 3843 * RETURN: 0 on success, error code on failure.
3737 3844 *
3738 3845 * Timestamps:
3739 3846 * dvp - ctime|mtime updated
3740 3847 */
3741 3848 /*ARGSUSED*/
3742 3849 static int
3743 3850 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3744 3851 caller_context_t *ct, int flags)
3745 3852 {
3746 3853 znode_t *zp, *dzp = VTOZ(dvp);
3747 3854 zfs_dirlock_t *dl;
3748 3855 dmu_tx_t *tx;
3749 3856 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
3750 3857 zilog_t *zilog;
3751 3858 uint64_t len = strlen(link);
3752 3859 int error;
3753 3860 int zflg = ZNEW;
3754 3861 zfs_acl_ids_t acl_ids;
3755 3862 boolean_t fuid_dirtied;
3756 3863 uint64_t txtype = TX_SYMLINK;
3757 3864 boolean_t waited = B_FALSE;
3758 3865
3759 3866 ASSERT(vap->va_type == VLNK);
3760 3867
3761 3868 ZFS_ENTER(zfsvfs);
3762 3869 ZFS_VERIFY_ZP(dzp);
3763 3870 zilog = zfsvfs->z_log;
3764 3871
3765 3872 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3766 3873 NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3767 3874 ZFS_EXIT(zfsvfs);
3768 3875 return (SET_ERROR(EILSEQ));
3769 3876 }
3770 3877 if (flags & FIGNORECASE)
3771 3878 zflg |= ZCILOOK;
3772 3879
3773 3880 if (len > MAXPATHLEN) {
3774 3881 ZFS_EXIT(zfsvfs);
3775 3882 return (SET_ERROR(ENAMETOOLONG));
3776 3883 }
3777 3884
3778 3885 if ((error = zfs_acl_ids_create(dzp, 0,
3779 3886 vap, cr, NULL, &acl_ids)) != 0) {
3780 3887 ZFS_EXIT(zfsvfs);
3781 3888 return (error);
3782 3889 }
3783 3890 top:
3784 3891 /*
3785 3892 * Attempt to lock directory; fail if entry already exists.
3786 3893 */
3787 3894 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3788 3895 if (error) {
3789 3896 zfs_acl_ids_free(&acl_ids);
3790 3897 ZFS_EXIT(zfsvfs);
3791 3898 return (error);
3792 3899 }
3793 3900
3794 3901 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3795 3902 zfs_acl_ids_free(&acl_ids);
3796 3903 zfs_dirent_unlock(dl);
3797 3904 ZFS_EXIT(zfsvfs);
3798 3905 return (error);
3799 3906 }
3800 3907
3801 3908 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3802 3909 zfs_acl_ids_free(&acl_ids);
3803 3910 zfs_dirent_unlock(dl);
3804 3911 ZFS_EXIT(zfsvfs);
3805 3912 return (SET_ERROR(EDQUOT));
3806 3913 }
3807 3914 tx = dmu_tx_create(zfsvfs->z_os);
3808 3915 fuid_dirtied = zfsvfs->z_fuid_dirty;
3809 3916 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3810 3917 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3811 3918 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3812 3919 ZFS_SA_BASE_ATTR_SIZE + len);
3813 3920 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3814 3921 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3815 3922 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3816 3923 acl_ids.z_aclp->z_acl_bytes);
3817 3924 }
3818 3925 if (fuid_dirtied)
3819 3926 zfs_fuid_txhold(zfsvfs, tx);
3820 3927 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3821 3928 if (error) {
3822 3929 zfs_dirent_unlock(dl);
3823 3930 if (error == ERESTART) {
3824 3931 waited = B_TRUE;
3825 3932 dmu_tx_wait(tx);
3826 3933 dmu_tx_abort(tx);
3827 3934 goto top;
3828 3935 }
3829 3936 zfs_acl_ids_free(&acl_ids);
3830 3937 dmu_tx_abort(tx);
3831 3938 ZFS_EXIT(zfsvfs);
3832 3939 return (error);
3833 3940 }
3834 3941
3835 3942 /*
3836 3943 * Create a new object for the symlink.
3837 3944 * for version 4 ZPL datsets the symlink will be an SA attribute
3838 3945 */
3839 3946 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3840 3947
3841 3948 if (fuid_dirtied)
3842 3949 zfs_fuid_sync(zfsvfs, tx);
3843 3950
3844 3951 mutex_enter(&zp->z_lock);
3845 3952 if (zp->z_is_sa)
3846 3953 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3847 3954 link, len, tx);
3848 3955 else
3849 3956 zfs_sa_symlink(zp, link, len, tx);
3850 3957 mutex_exit(&zp->z_lock);
3851 3958
3852 3959 zp->z_size = len;
3853 3960 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3854 3961 &zp->z_size, sizeof (zp->z_size), tx);
3855 3962 /*
3856 3963 * Insert the new object into the directory.
3857 3964 */
3858 3965 (void) zfs_link_create(dl, zp, tx, ZNEW);
3859 3966
3860 3967 if (flags & FIGNORECASE)
3861 3968 txtype |= TX_CI;
3862 3969 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3863 3970
3864 3971 zfs_acl_ids_free(&acl_ids);
3865 3972
3866 3973 dmu_tx_commit(tx);
3867 3974
3868 3975 zfs_dirent_unlock(dl);
3869 3976
3870 3977 VN_RELE(ZTOV(zp));
3871 3978
3872 3979 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3873 3980 zil_commit(zilog, 0);
3874 3981
3875 3982 ZFS_EXIT(zfsvfs);
3876 3983 return (error);
3877 3984 }
3878 3985
3879 3986 /*
3880 3987 * Return, in the buffer contained in the provided uio structure,
3881 3988 * the symbolic path referred to by vp.
3882 3989 *
3883 3990 * IN: vp - vnode of symbolic link.
3884 3991 * uio - structure to contain the link path.
3885 3992 * cr - credentials of caller.
3886 3993 * ct - caller context
3887 3994 *
3888 3995 * OUT: uio - structure containing the link path.
3889 3996 *
3890 3997 * RETURN: 0 on success, error code on failure.
3891 3998 *
3892 3999 * Timestamps:
3893 4000 * vp - atime updated
3894 4001 */
3895 4002 /* ARGSUSED */
3896 4003 static int
3897 4004 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3898 4005 {
3899 4006 znode_t *zp = VTOZ(vp);
3900 4007 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3901 4008 int error;
3902 4009
3903 4010 ZFS_ENTER(zfsvfs);
3904 4011 ZFS_VERIFY_ZP(zp);
3905 4012
3906 4013 mutex_enter(&zp->z_lock);
3907 4014 if (zp->z_is_sa)
3908 4015 error = sa_lookup_uio(zp->z_sa_hdl,
3909 4016 SA_ZPL_SYMLINK(zfsvfs), uio);
3910 4017 else
3911 4018 error = zfs_sa_readlink(zp, uio);
3912 4019 mutex_exit(&zp->z_lock);
3913 4020
3914 4021 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3915 4022
3916 4023 ZFS_EXIT(zfsvfs);
3917 4024 return (error);
3918 4025 }
3919 4026
3920 4027 /*
3921 4028 * Insert a new entry into directory tdvp referencing svp.
3922 4029 *
3923 4030 * IN: tdvp - Directory to contain new entry.
3924 4031 * svp - vnode of new entry.
3925 4032 * name - name of new entry.
3926 4033 * cr - credentials of caller.
3927 4034 * ct - caller context
3928 4035 *
3929 4036 * RETURN: 0 on success, error code on failure.
3930 4037 *
3931 4038 * Timestamps:
3932 4039 * tdvp - ctime|mtime updated
3933 4040 * svp - ctime updated
3934 4041 */
3935 4042 /* ARGSUSED */
3936 4043 static int
3937 4044 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3938 4045 caller_context_t *ct, int flags)
3939 4046 {
3940 4047 znode_t *dzp = VTOZ(tdvp);
3941 4048 znode_t *tzp, *szp;
3942 4049 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
3943 4050 zilog_t *zilog;
3944 4051 zfs_dirlock_t *dl;
3945 4052 dmu_tx_t *tx;
3946 4053 vnode_t *realvp;
3947 4054 int error;
3948 4055 int zf = ZNEW;
3949 4056 uint64_t parent;
3950 4057 uid_t owner;
3951 4058 boolean_t waited = B_FALSE;
3952 4059
3953 4060 ASSERT(tdvp->v_type == VDIR);
3954 4061
3955 4062 ZFS_ENTER(zfsvfs);
3956 4063 ZFS_VERIFY_ZP(dzp);
3957 4064 zilog = zfsvfs->z_log;
3958 4065
3959 4066 if (VOP_REALVP(svp, &realvp, ct) == 0)
3960 4067 svp = realvp;
3961 4068
3962 4069 /*
3963 4070 * POSIX dictates that we return EPERM here.
3964 4071 * Better choices include ENOTSUP or EISDIR.
3965 4072 */
3966 4073 if (svp->v_type == VDIR) {
3967 4074 ZFS_EXIT(zfsvfs);
3968 4075 return (SET_ERROR(EPERM));
3969 4076 }
3970 4077
3971 4078 szp = VTOZ(svp);
3972 4079 ZFS_VERIFY_ZP(szp);
3973 4080
3974 4081 /*
3975 4082 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3976 4083 * ctldir appear to have the same v_vfsp.
3977 4084 */
3978 4085 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
3979 4086 ZFS_EXIT(zfsvfs);
3980 4087 return (SET_ERROR(EXDEV));
3981 4088 }
3982 4089
3983 4090 /* Prevent links to .zfs/shares files */
3984 4091
3985 4092 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
3986 4093 &parent, sizeof (uint64_t))) != 0) {
3987 4094 ZFS_EXIT(zfsvfs);
3988 4095 return (error);
3989 4096 }
3990 4097 if (parent == zfsvfs->z_shares_dir) {
3991 4098 ZFS_EXIT(zfsvfs);
3992 4099 return (SET_ERROR(EPERM));
3993 4100 }
3994 4101
3995 4102 if (zfsvfs->z_utf8 && u8_validate(name,
3996 4103 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3997 4104 ZFS_EXIT(zfsvfs);
3998 4105 return (SET_ERROR(EILSEQ));
3999 4106 }
4000 4107 if (flags & FIGNORECASE)
4001 4108 zf |= ZCILOOK;
4002 4109
4003 4110 /*
4004 4111 * We do not support links between attributes and non-attributes
4005 4112 * because of the potential security risk of creating links
4006 4113 * into "normal" file space in order to circumvent restrictions
4007 4114 * imposed in attribute space.
4008 4115 */
4009 4116 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4010 4117 ZFS_EXIT(zfsvfs);
4011 4118 return (SET_ERROR(EINVAL));
4012 4119 }
4013 4120
4014 4121
4015 4122 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4016 4123 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4017 4124 ZFS_EXIT(zfsvfs);
4018 4125 return (SET_ERROR(EPERM));
4019 4126 }
4020 4127
4021 4128 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4022 4129 ZFS_EXIT(zfsvfs);
4023 4130 return (error);
4024 4131 }
4025 4132
4026 4133 top:
4027 4134 /*
4028 4135 * Attempt to lock directory; fail if entry already exists.
4029 4136 */
4030 4137 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4031 4138 if (error) {
4032 4139 ZFS_EXIT(zfsvfs);
4033 4140 return (error);
4034 4141 }
4035 4142
4036 4143 tx = dmu_tx_create(zfsvfs->z_os);
4037 4144 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4038 4145 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4039 4146 zfs_sa_upgrade_txholds(tx, szp);
4040 4147 zfs_sa_upgrade_txholds(tx, dzp);
4041 4148 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4042 4149 if (error) {
4043 4150 zfs_dirent_unlock(dl);
4044 4151 if (error == ERESTART) {
4045 4152 waited = B_TRUE;
4046 4153 dmu_tx_wait(tx);
4047 4154 dmu_tx_abort(tx);
4048 4155 goto top;
4049 4156 }
4050 4157 dmu_tx_abort(tx);
4051 4158 ZFS_EXIT(zfsvfs);
4052 4159 return (error);
4053 4160 }
4054 4161
4055 4162 error = zfs_link_create(dl, szp, tx, 0);
4056 4163
4057 4164 if (error == 0) {
4058 4165 uint64_t txtype = TX_LINK;
4059 4166 if (flags & FIGNORECASE)
4060 4167 txtype |= TX_CI;
4061 4168 zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4062 4169 }
4063 4170
4064 4171 dmu_tx_commit(tx);
4065 4172
4066 4173 zfs_dirent_unlock(dl);
4067 4174
4068 4175 if (error == 0) {
4069 4176 vnevent_link(svp, ct);
4070 4177 }
4071 4178
4072 4179 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4073 4180 zil_commit(zilog, 0);
4074 4181
4075 4182 ZFS_EXIT(zfsvfs);
4076 4183 return (error);
4077 4184 }
4078 4185
4079 4186 /*
4080 4187 * zfs_null_putapage() is used when the file system has been force
4081 4188 * unmounted. It just drops the pages.
4082 4189 */
4083 4190 /* ARGSUSED */
4084 4191 static int
4085 4192 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4086 4193 size_t *lenp, int flags, cred_t *cr)
4087 4194 {
4088 4195 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4089 4196 return (0);
4090 4197 }
4091 4198
4092 4199 /*
4093 4200 * Push a page out to disk, klustering if possible.
4094 4201 *
4095 4202 * IN: vp - file to push page to.
4096 4203 * pp - page to push.
4097 4204 * flags - additional flags.
4098 4205 * cr - credentials of caller.
4099 4206 *
4100 4207 * OUT: offp - start of range pushed.
4101 4208 * lenp - len of range pushed.
4102 4209 *
4103 4210 * RETURN: 0 on success, error code on failure.
4104 4211 *
4105 4212 * NOTE: callers must have locked the page to be pushed. On
4106 4213 * exit, the page (and all other pages in the kluster) must be
4107 4214 * unlocked.
4108 4215 */
4109 4216 /* ARGSUSED */
4110 4217 static int
4111 4218 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4112 4219 size_t *lenp, int flags, cred_t *cr)
4113 4220 {
4114 4221 znode_t *zp = VTOZ(vp);
4115 4222 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4116 4223 dmu_tx_t *tx;
4117 4224 u_offset_t off, koff;
4118 4225 size_t len, klen;
4119 4226 int err;
4120 4227
4121 4228 off = pp->p_offset;
4122 4229 len = PAGESIZE;
4123 4230 /*
4124 4231 * If our blocksize is bigger than the page size, try to kluster
4125 4232 * multiple pages so that we write a full block (thus avoiding
4126 4233 * a read-modify-write).
4127 4234 */
4128 4235 if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4129 4236 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4130 4237 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4131 4238 ASSERT(koff <= zp->z_size);
4132 4239 if (koff + klen > zp->z_size)
4133 4240 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4134 4241 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4135 4242 }
4136 4243 ASSERT3U(btop(len), ==, btopr(len));
4137 4244
4138 4245 /*
4139 4246 * Can't push pages past end-of-file.
4140 4247 */
4141 4248 if (off >= zp->z_size) {
4142 4249 /* ignore all pages */
4143 4250 err = 0;
4144 4251 goto out;
4145 4252 } else if (off + len > zp->z_size) {
4146 4253 int npages = btopr(zp->z_size - off);
4147 4254 page_t *trunc;
4148 4255
4149 4256 page_list_break(&pp, &trunc, npages);
4150 4257 /* ignore pages past end of file */
4151 4258 if (trunc)
4152 4259 pvn_write_done(trunc, flags);
4153 4260 len = zp->z_size - off;
4154 4261 }
4155 4262
4156 4263 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4157 4264 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4158 4265 err = SET_ERROR(EDQUOT);
4159 4266 goto out;
4160 4267 }
4161 4268 tx = dmu_tx_create(zfsvfs->z_os);
4162 4269 dmu_tx_hold_write(tx, zp->z_id, off, len);
4163 4270
4164 4271 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4165 4272 zfs_sa_upgrade_txholds(tx, zp);
4166 4273 err = dmu_tx_assign(tx, TXG_WAIT);
4167 4274 if (err != 0) {
4168 4275 dmu_tx_abort(tx);
4169 4276 goto out;
4170 4277 }
4171 4278
4172 4279 if (zp->z_blksz <= PAGESIZE) {
4173 4280 caddr_t va = zfs_map_page(pp, S_READ);
4174 4281 ASSERT3U(len, <=, PAGESIZE);
4175 4282 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4176 4283 zfs_unmap_page(pp, va);
4177 4284 } else {
4178 4285 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4179 4286 }
4180 4287
4181 4288 if (err == 0) {
4182 4289 uint64_t mtime[2], ctime[2];
4183 4290 sa_bulk_attr_t bulk[3];
4184 4291 int count = 0;
4185 4292
4186 4293 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4187 4294 &mtime, 16);
4188 4295 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4189 4296 &ctime, 16);
4190 4297 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4191 4298 &zp->z_pflags, 8);
4192 4299 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4193 4300 B_TRUE);
4194 4301 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4195 4302 }
4196 4303 dmu_tx_commit(tx);
4197 4304
4198 4305 out:
4199 4306 pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4200 4307 if (offp)
4201 4308 *offp = off;
4202 4309 if (lenp)
4203 4310 *lenp = len;
4204 4311
4205 4312 return (err);
4206 4313 }
4207 4314
4208 4315 /*
4209 4316 * Copy the portion of the file indicated from pages into the file.
4210 4317 * The pages are stored in a page list attached to the files vnode.
4211 4318 *
4212 4319 * IN: vp - vnode of file to push page data to.
4213 4320 * off - position in file to put data.
4214 4321 * len - amount of data to write.
4215 4322 * flags - flags to control the operation.
4216 4323 * cr - credentials of caller.
4217 4324 * ct - caller context.
4218 4325 *
4219 4326 * RETURN: 0 on success, error code on failure.
4220 4327 *
4221 4328 * Timestamps:
4222 4329 * vp - ctime|mtime updated
4223 4330 */
4224 4331 /*ARGSUSED*/
4225 4332 static int
4226 4333 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4227 4334 caller_context_t *ct)
4228 4335 {
4229 4336 znode_t *zp = VTOZ(vp);
4230 4337 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4231 4338 page_t *pp;
4232 4339 size_t io_len;
4233 4340 u_offset_t io_off;
4234 4341 uint_t blksz;
4235 4342 rl_t *rl;
4236 4343 int error = 0;
4237 4344
4238 4345 ZFS_ENTER(zfsvfs);
4239 4346 ZFS_VERIFY_ZP(zp);
4240 4347
4241 4348 /*
4242 4349 * There's nothing to do if no data is cached.
4243 4350 */
4244 4351 if (!vn_has_cached_data(vp)) {
4245 4352 ZFS_EXIT(zfsvfs);
4246 4353 return (0);
4247 4354 }
4248 4355
4249 4356 /*
4250 4357 * Align this request to the file block size in case we kluster.
4251 4358 * XXX - this can result in pretty aggresive locking, which can
4252 4359 * impact simultanious read/write access. One option might be
4253 4360 * to break up long requests (len == 0) into block-by-block
4254 4361 * operations to get narrower locking.
4255 4362 */
4256 4363 blksz = zp->z_blksz;
4257 4364 if (ISP2(blksz))
4258 4365 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4259 4366 else
4260 4367 io_off = 0;
4261 4368 if (len > 0 && ISP2(blksz))
4262 4369 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4263 4370 else
4264 4371 io_len = 0;
4265 4372
4266 4373 if (io_len == 0) {
4267 4374 /*
4268 4375 * Search the entire vp list for pages >= io_off.
4269 4376 */
4270 4377 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4271 4378 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4272 4379 goto out;
4273 4380 }
4274 4381 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4275 4382
4276 4383 if (off > zp->z_size) {
4277 4384 /* past end of file */
4278 4385 zfs_range_unlock(rl);
4279 4386 ZFS_EXIT(zfsvfs);
4280 4387 return (0);
4281 4388 }
4282 4389
4283 4390 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4284 4391
4285 4392 for (off = io_off; io_off < off + len; io_off += io_len) {
4286 4393 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4287 4394 pp = page_lookup(vp, io_off,
4288 4395 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4289 4396 } else {
4290 4397 pp = page_lookup_nowait(vp, io_off,
4291 4398 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4292 4399 }
4293 4400
4294 4401 if (pp != NULL && pvn_getdirty(pp, flags)) {
4295 4402 int err;
4296 4403
4297 4404 /*
4298 4405 * Found a dirty page to push
4299 4406 */
4300 4407 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4301 4408 if (err)
4302 4409 error = err;
4303 4410 } else {
4304 4411 io_len = PAGESIZE;
4305 4412 }
4306 4413 }
4307 4414 out:
4308 4415 zfs_range_unlock(rl);
4309 4416 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4310 4417 zil_commit(zfsvfs->z_log, zp->z_id);
4311 4418 ZFS_EXIT(zfsvfs);
4312 4419 return (error);
4313 4420 }
4314 4421
4315 4422 /*ARGSUSED*/
4316 4423 void
4317 4424 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4318 4425 {
4319 4426 znode_t *zp = VTOZ(vp);
4320 4427 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4321 4428 int error;
4322 4429
4323 4430 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4324 4431 if (zp->z_sa_hdl == NULL) {
4325 4432 /*
4326 4433 * The fs has been unmounted, or we did a
4327 4434 * suspend/resume and this file no longer exists.
4328 4435 */
4329 4436 if (vn_has_cached_data(vp)) {
4330 4437 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4331 4438 B_INVAL, cr);
4332 4439 }
4333 4440
4334 4441 mutex_enter(&zp->z_lock);
4335 4442 mutex_enter(&vp->v_lock);
4336 4443 ASSERT(vp->v_count == 1);
4337 4444 vp->v_count = 0;
4338 4445 mutex_exit(&vp->v_lock);
4339 4446 mutex_exit(&zp->z_lock);
4340 4447 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4341 4448 zfs_znode_free(zp);
4342 4449 return;
4343 4450 }
4344 4451
4345 4452 /*
4346 4453 * Attempt to push any data in the page cache. If this fails
4347 4454 * we will get kicked out later in zfs_zinactive().
4348 4455 */
4349 4456 if (vn_has_cached_data(vp)) {
4350 4457 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4351 4458 cr);
4352 4459 }
4353 4460
4354 4461 if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4355 4462 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4356 4463
4357 4464 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4358 4465 zfs_sa_upgrade_txholds(tx, zp);
4359 4466 error = dmu_tx_assign(tx, TXG_WAIT);
4360 4467 if (error) {
4361 4468 dmu_tx_abort(tx);
4362 4469 } else {
4363 4470 mutex_enter(&zp->z_lock);
4364 4471 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4365 4472 (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4366 4473 zp->z_atime_dirty = 0;
4367 4474 mutex_exit(&zp->z_lock);
4368 4475 dmu_tx_commit(tx);
4369 4476 }
4370 4477 }
4371 4478
4372 4479 zfs_zinactive(zp);
4373 4480 rw_exit(&zfsvfs->z_teardown_inactive_lock);
4374 4481 }
4375 4482
4376 4483 /*
4377 4484 * Bounds-check the seek operation.
4378 4485 *
4379 4486 * IN: vp - vnode seeking within
4380 4487 * ooff - old file offset
4381 4488 * noffp - pointer to new file offset
4382 4489 * ct - caller context
4383 4490 *
4384 4491 * RETURN: 0 on success, EINVAL if new offset invalid.
4385 4492 */
4386 4493 /* ARGSUSED */
4387 4494 static int
4388 4495 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4389 4496 caller_context_t *ct)
4390 4497 {
4391 4498 if (vp->v_type == VDIR)
4392 4499 return (0);
4393 4500 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4394 4501 }
4395 4502
4396 4503 /*
4397 4504 * Pre-filter the generic locking function to trap attempts to place
4398 4505 * a mandatory lock on a memory mapped file.
4399 4506 */
4400 4507 static int
4401 4508 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4402 4509 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4403 4510 {
4404 4511 znode_t *zp = VTOZ(vp);
4405 4512 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4406 4513
4407 4514 ZFS_ENTER(zfsvfs);
4408 4515 ZFS_VERIFY_ZP(zp);
4409 4516
4410 4517 /*
4411 4518 * We are following the UFS semantics with respect to mapcnt
4412 4519 * here: If we see that the file is mapped already, then we will
4413 4520 * return an error, but we don't worry about races between this
4414 4521 * function and zfs_map().
4415 4522 */
4416 4523 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4417 4524 ZFS_EXIT(zfsvfs);
4418 4525 return (SET_ERROR(EAGAIN));
4419 4526 }
4420 4527 ZFS_EXIT(zfsvfs);
4421 4528 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4422 4529 }
4423 4530
4424 4531 /*
4425 4532 * If we can't find a page in the cache, we will create a new page
4426 4533 * and fill it with file data. For efficiency, we may try to fill
4427 4534 * multiple pages at once (klustering) to fill up the supplied page
4428 4535 * list. Note that the pages to be filled are held with an exclusive
4429 4536 * lock to prevent access by other threads while they are being filled.
4430 4537 */
4431 4538 static int
4432 4539 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4433 4540 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4434 4541 {
4435 4542 znode_t *zp = VTOZ(vp);
4436 4543 page_t *pp, *cur_pp;
4437 4544 objset_t *os = zp->z_zfsvfs->z_os;
4438 4545 u_offset_t io_off, total;
4439 4546 size_t io_len;
4440 4547 int err;
4441 4548
4442 4549 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4443 4550 /*
4444 4551 * We only have a single page, don't bother klustering
4445 4552 */
4446 4553 io_off = off;
4447 4554 io_len = PAGESIZE;
4448 4555 pp = page_create_va(vp, io_off, io_len,
4449 4556 PG_EXCL | PG_WAIT, seg, addr);
4450 4557 } else {
4451 4558 /*
4452 4559 * Try to find enough pages to fill the page list
4453 4560 */
4454 4561 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4455 4562 &io_len, off, plsz, 0);
4456 4563 }
4457 4564 if (pp == NULL) {
4458 4565 /*
4459 4566 * The page already exists, nothing to do here.
4460 4567 */
4461 4568 *pl = NULL;
4462 4569 return (0);
4463 4570 }
4464 4571
4465 4572 /*
4466 4573 * Fill the pages in the kluster.
4467 4574 */
4468 4575 cur_pp = pp;
4469 4576 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4470 4577 caddr_t va;
4471 4578
4472 4579 ASSERT3U(io_off, ==, cur_pp->p_offset);
4473 4580 va = zfs_map_page(cur_pp, S_WRITE);
4474 4581 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4475 4582 DMU_READ_PREFETCH);
4476 4583 zfs_unmap_page(cur_pp, va);
4477 4584 if (err) {
4478 4585 /* On error, toss the entire kluster */
4479 4586 pvn_read_done(pp, B_ERROR);
4480 4587 /* convert checksum errors into IO errors */
4481 4588 if (err == ECKSUM)
4482 4589 err = SET_ERROR(EIO);
4483 4590 return (err);
4484 4591 }
4485 4592 cur_pp = cur_pp->p_next;
4486 4593 }
4487 4594
4488 4595 /*
4489 4596 * Fill in the page list array from the kluster starting
4490 4597 * from the desired offset `off'.
4491 4598 * NOTE: the page list will always be null terminated.
4492 4599 */
4493 4600 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4494 4601 ASSERT(pl == NULL || (*pl)->p_offset == off);
4495 4602
4496 4603 return (0);
4497 4604 }
4498 4605
4499 4606 /*
4500 4607 * Return pointers to the pages for the file region [off, off + len]
4501 4608 * in the pl array. If plsz is greater than len, this function may
4502 4609 * also return page pointers from after the specified region
4503 4610 * (i.e. the region [off, off + plsz]). These additional pages are
4504 4611 * only returned if they are already in the cache, or were created as
4505 4612 * part of a klustered read.
4506 4613 *
4507 4614 * IN: vp - vnode of file to get data from.
4508 4615 * off - position in file to get data from.
4509 4616 * len - amount of data to retrieve.
4510 4617 * plsz - length of provided page list.
4511 4618 * seg - segment to obtain pages for.
4512 4619 * addr - virtual address of fault.
4513 4620 * rw - mode of created pages.
4514 4621 * cr - credentials of caller.
4515 4622 * ct - caller context.
4516 4623 *
4517 4624 * OUT: protp - protection mode of created pages.
4518 4625 * pl - list of pages created.
4519 4626 *
4520 4627 * RETURN: 0 on success, error code on failure.
4521 4628 *
4522 4629 * Timestamps:
4523 4630 * vp - atime updated
4524 4631 */
4525 4632 /* ARGSUSED */
4526 4633 static int
4527 4634 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4528 4635 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4529 4636 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4530 4637 {
4531 4638 znode_t *zp = VTOZ(vp);
4532 4639 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4533 4640 page_t **pl0 = pl;
4534 4641 int err = 0;
4535 4642
4536 4643 /* we do our own caching, faultahead is unnecessary */
4537 4644 if (pl == NULL)
4538 4645 return (0);
4539 4646 else if (len > plsz)
4540 4647 len = plsz;
4541 4648 else
4542 4649 len = P2ROUNDUP(len, PAGESIZE);
4543 4650 ASSERT(plsz >= len);
4544 4651
4545 4652 ZFS_ENTER(zfsvfs);
4546 4653 ZFS_VERIFY_ZP(zp);
4547 4654
4548 4655 if (protp)
4549 4656 *protp = PROT_ALL;
4550 4657
4551 4658 /*
4552 4659 * Loop through the requested range [off, off + len) looking
4553 4660 * for pages. If we don't find a page, we will need to create
4554 4661 * a new page and fill it with data from the file.
4555 4662 */
4556 4663 while (len > 0) {
4557 4664 if (*pl = page_lookup(vp, off, SE_SHARED))
4558 4665 *(pl+1) = NULL;
4559 4666 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4560 4667 goto out;
4561 4668 while (*pl) {
4562 4669 ASSERT3U((*pl)->p_offset, ==, off);
4563 4670 off += PAGESIZE;
4564 4671 addr += PAGESIZE;
4565 4672 if (len > 0) {
4566 4673 ASSERT3U(len, >=, PAGESIZE);
4567 4674 len -= PAGESIZE;
4568 4675 }
4569 4676 ASSERT3U(plsz, >=, PAGESIZE);
4570 4677 plsz -= PAGESIZE;
4571 4678 pl++;
4572 4679 }
4573 4680 }
4574 4681
4575 4682 /*
4576 4683 * Fill out the page array with any pages already in the cache.
4577 4684 */
4578 4685 while (plsz > 0 &&
4579 4686 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4580 4687 off += PAGESIZE;
4581 4688 plsz -= PAGESIZE;
4582 4689 }
4583 4690 out:
4584 4691 if (err) {
4585 4692 /*
4586 4693 * Release any pages we have previously locked.
4587 4694 */
4588 4695 while (pl > pl0)
4589 4696 page_unlock(*--pl);
4590 4697 } else {
4591 4698 ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4592 4699 }
4593 4700
4594 4701 *pl = NULL;
4595 4702
4596 4703 ZFS_EXIT(zfsvfs);
4597 4704 return (err);
4598 4705 }
4599 4706
4600 4707 /*
4601 4708 * Request a memory map for a section of a file. This code interacts
4602 4709 * with common code and the VM system as follows:
4603 4710 *
4604 4711 * - common code calls mmap(), which ends up in smmap_common()
4605 4712 * - this calls VOP_MAP(), which takes you into (say) zfs
4606 4713 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4607 4714 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4608 4715 * - zfs_addmap() updates z_mapcnt
4609 4716 */
4610 4717 /*ARGSUSED*/
4611 4718 static int
4612 4719 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4613 4720 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4614 4721 caller_context_t *ct)
4615 4722 {
4616 4723 znode_t *zp = VTOZ(vp);
4617 4724 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4618 4725 segvn_crargs_t vn_a;
4619 4726 int error;
4620 4727
4621 4728 ZFS_ENTER(zfsvfs);
4622 4729 ZFS_VERIFY_ZP(zp);
4623 4730
4624 4731 if ((prot & PROT_WRITE) && (zp->z_pflags &
4625 4732 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4626 4733 ZFS_EXIT(zfsvfs);
4627 4734 return (SET_ERROR(EPERM));
4628 4735 }
4629 4736
4630 4737 if ((prot & (PROT_READ | PROT_EXEC)) &&
4631 4738 (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4632 4739 ZFS_EXIT(zfsvfs);
4633 4740 return (SET_ERROR(EACCES));
4634 4741 }
4635 4742
4636 4743 if (vp->v_flag & VNOMAP) {
4637 4744 ZFS_EXIT(zfsvfs);
4638 4745 return (SET_ERROR(ENOSYS));
4639 4746 }
4640 4747
4641 4748 if (off < 0 || len > MAXOFFSET_T - off) {
4642 4749 ZFS_EXIT(zfsvfs);
4643 4750 return (SET_ERROR(ENXIO));
4644 4751 }
4645 4752
4646 4753 if (vp->v_type != VREG) {
4647 4754 ZFS_EXIT(zfsvfs);
4648 4755 return (SET_ERROR(ENODEV));
4649 4756 }
4650 4757
4651 4758 /*
4652 4759 * If file is locked, disallow mapping.
4653 4760 */
4654 4761 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4655 4762 ZFS_EXIT(zfsvfs);
4656 4763 return (SET_ERROR(EAGAIN));
4657 4764 }
4658 4765
4659 4766 as_rangelock(as);
4660 4767 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4661 4768 if (error != 0) {
4662 4769 as_rangeunlock(as);
4663 4770 ZFS_EXIT(zfsvfs);
4664 4771 return (error);
4665 4772 }
4666 4773
4667 4774 vn_a.vp = vp;
4668 4775 vn_a.offset = (u_offset_t)off;
4669 4776 vn_a.type = flags & MAP_TYPE;
4670 4777 vn_a.prot = prot;
4671 4778 vn_a.maxprot = maxprot;
4672 4779 vn_a.cred = cr;
4673 4780 vn_a.amp = NULL;
4674 4781 vn_a.flags = flags & ~MAP_TYPE;
4675 4782 vn_a.szc = 0;
4676 4783 vn_a.lgrp_mem_policy_flags = 0;
4677 4784
4678 4785 error = as_map(as, *addrp, len, segvn_create, &vn_a);
4679 4786
4680 4787 as_rangeunlock(as);
4681 4788 ZFS_EXIT(zfsvfs);
4682 4789 return (error);
4683 4790 }
4684 4791
4685 4792 /* ARGSUSED */
4686 4793 static int
4687 4794 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4688 4795 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4689 4796 caller_context_t *ct)
4690 4797 {
4691 4798 uint64_t pages = btopr(len);
4692 4799
4693 4800 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4694 4801 return (0);
4695 4802 }
4696 4803
4697 4804 /*
4698 4805 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4699 4806 * more accurate mtime for the associated file. Since we don't have a way of
4700 4807 * detecting when the data was actually modified, we have to resort to
4701 4808 * heuristics. If an explicit msync() is done, then we mark the mtime when the
4702 4809 * last page is pushed. The problem occurs when the msync() call is omitted,
4703 4810 * which by far the most common case:
4704 4811 *
4705 4812 * open()
4706 4813 * mmap()
4707 4814 * <modify memory>
4708 4815 * munmap()
4709 4816 * close()
4710 4817 * <time lapse>
4711 4818 * putpage() via fsflush
4712 4819 *
4713 4820 * If we wait until fsflush to come along, we can have a modification time that
4714 4821 * is some arbitrary point in the future. In order to prevent this in the
4715 4822 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4716 4823 * torn down.
4717 4824 */
4718 4825 /* ARGSUSED */
4719 4826 static int
4720 4827 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4721 4828 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4722 4829 caller_context_t *ct)
4723 4830 {
4724 4831 uint64_t pages = btopr(len);
4725 4832
4726 4833 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4727 4834 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4728 4835
4729 4836 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4730 4837 vn_has_cached_data(vp))
4731 4838 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4732 4839
4733 4840 return (0);
4734 4841 }
4735 4842
4736 4843 /*
4737 4844 * Free or allocate space in a file. Currently, this function only
4738 4845 * supports the `F_FREESP' command. However, this command is somewhat
4739 4846 * misnamed, as its functionality includes the ability to allocate as
4740 4847 * well as free space.
4741 4848 *
4742 4849 * IN: vp - vnode of file to free data in.
4743 4850 * cmd - action to take (only F_FREESP supported).
4744 4851 * bfp - section of file to free/alloc.
4745 4852 * flag - current file open mode flags.
4746 4853 * offset - current file offset.
4747 4854 * cr - credentials of caller [UNUSED].
4748 4855 * ct - caller context.
4749 4856 *
4750 4857 * RETURN: 0 on success, error code on failure.
4751 4858 *
4752 4859 * Timestamps:
4753 4860 * vp - ctime|mtime updated
4754 4861 */
4755 4862 /* ARGSUSED */
4756 4863 static int
4757 4864 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4758 4865 offset_t offset, cred_t *cr, caller_context_t *ct)
4759 4866 {
4760 4867 znode_t *zp = VTOZ(vp);
4761 4868 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4762 4869 uint64_t off, len;
4763 4870 int error;
4764 4871
4765 4872 ZFS_ENTER(zfsvfs);
4766 4873 ZFS_VERIFY_ZP(zp);
4767 4874
4768 4875 if (cmd != F_FREESP) {
4769 4876 ZFS_EXIT(zfsvfs);
4770 4877 return (SET_ERROR(EINVAL));
4771 4878 }
4772 4879
4773 4880 /*
4774 4881 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
4775 4882 * callers might not be able to detect properly that we are read-only,
4776 4883 * so check it explicitly here.
4777 4884 */
4778 4885 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
4779 4886 ZFS_EXIT(zfsvfs);
4780 4887 return (SET_ERROR(EROFS));
4781 4888 }
4782 4889
4783 4890 if (error = convoff(vp, bfp, 0, offset)) {
4784 4891 ZFS_EXIT(zfsvfs);
4785 4892 return (error);
4786 4893 }
4787 4894
4788 4895 if (bfp->l_len < 0) {
4789 4896 ZFS_EXIT(zfsvfs);
4790 4897 return (SET_ERROR(EINVAL));
4791 4898 }
4792 4899
4793 4900 off = bfp->l_start;
4794 4901 len = bfp->l_len; /* 0 means from off to end of file */
4795 4902
4796 4903 error = zfs_freesp(zp, off, len, flag, TRUE);
4797 4904
4798 4905 if (error == 0 && off == 0 && len == 0)
4799 4906 vnevent_truncate(ZTOV(zp), ct);
4800 4907
4801 4908 ZFS_EXIT(zfsvfs);
4802 4909 return (error);
4803 4910 }
4804 4911
4805 4912 /*ARGSUSED*/
4806 4913 static int
4807 4914 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4808 4915 {
4809 4916 znode_t *zp = VTOZ(vp);
4810 4917 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4811 4918 uint32_t gen;
4812 4919 uint64_t gen64;
4813 4920 uint64_t object = zp->z_id;
4814 4921 zfid_short_t *zfid;
4815 4922 int size, i, error;
4816 4923
4817 4924 ZFS_ENTER(zfsvfs);
4818 4925 ZFS_VERIFY_ZP(zp);
4819 4926
4820 4927 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4821 4928 &gen64, sizeof (uint64_t))) != 0) {
4822 4929 ZFS_EXIT(zfsvfs);
4823 4930 return (error);
4824 4931 }
4825 4932
4826 4933 gen = (uint32_t)gen64;
4827 4934
4828 4935 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4829 4936 if (fidp->fid_len < size) {
4830 4937 fidp->fid_len = size;
4831 4938 ZFS_EXIT(zfsvfs);
4832 4939 return (SET_ERROR(ENOSPC));
4833 4940 }
4834 4941
4835 4942 zfid = (zfid_short_t *)fidp;
4836 4943
4837 4944 zfid->zf_len = size;
4838 4945
4839 4946 for (i = 0; i < sizeof (zfid->zf_object); i++)
4840 4947 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4841 4948
4842 4949 /* Must have a non-zero generation number to distinguish from .zfs */
4843 4950 if (gen == 0)
4844 4951 gen = 1;
4845 4952 for (i = 0; i < sizeof (zfid->zf_gen); i++)
4846 4953 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4847 4954
4848 4955 if (size == LONG_FID_LEN) {
4849 4956 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
4850 4957 zfid_long_t *zlfid;
4851 4958
4852 4959 zlfid = (zfid_long_t *)fidp;
4853 4960
4854 4961 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4855 4962 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4856 4963
4857 4964 /* XXX - this should be the generation number for the objset */
4858 4965 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4859 4966 zlfid->zf_setgen[i] = 0;
4860 4967 }
4861 4968
4862 4969 ZFS_EXIT(zfsvfs);
4863 4970 return (0);
4864 4971 }
4865 4972
4866 4973 static int
4867 4974 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4868 4975 caller_context_t *ct)
4869 4976 {
4870 4977 znode_t *zp, *xzp;
4871 4978 zfsvfs_t *zfsvfs;
4872 4979 zfs_dirlock_t *dl;
4873 4980 int error;
4874 4981
4875 4982 switch (cmd) {
4876 4983 case _PC_LINK_MAX:
4877 4984 *valp = ULONG_MAX;
4878 4985 return (0);
4879 4986
4880 4987 case _PC_FILESIZEBITS:
4881 4988 *valp = 64;
4882 4989 return (0);
4883 4990
4884 4991 case _PC_XATTR_EXISTS:
4885 4992 zp = VTOZ(vp);
4886 4993 zfsvfs = zp->z_zfsvfs;
4887 4994 ZFS_ENTER(zfsvfs);
4888 4995 ZFS_VERIFY_ZP(zp);
4889 4996 *valp = 0;
4890 4997 error = zfs_dirent_lock(&dl, zp, "", &xzp,
4891 4998 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4892 4999 if (error == 0) {
4893 5000 zfs_dirent_unlock(dl);
4894 5001 if (!zfs_dirempty(xzp))
4895 5002 *valp = 1;
4896 5003 VN_RELE(ZTOV(xzp));
4897 5004 } else if (error == ENOENT) {
4898 5005 /*
4899 5006 * If there aren't extended attributes, it's the
4900 5007 * same as having zero of them.
4901 5008 */
4902 5009 error = 0;
4903 5010 }
4904 5011 ZFS_EXIT(zfsvfs);
4905 5012 return (error);
4906 5013
4907 5014 case _PC_SATTR_ENABLED:
4908 5015 case _PC_SATTR_EXISTS:
4909 5016 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4910 5017 (vp->v_type == VREG || vp->v_type == VDIR);
4911 5018 return (0);
4912 5019
4913 5020 case _PC_ACCESS_FILTERING:
4914 5021 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4915 5022 vp->v_type == VDIR;
4916 5023 return (0);
4917 5024
4918 5025 case _PC_ACL_ENABLED:
4919 5026 *valp = _ACL_ACE_ENABLED;
4920 5027 return (0);
4921 5028
4922 5029 case _PC_MIN_HOLE_SIZE:
4923 5030 *valp = (ulong_t)SPA_MINBLOCKSIZE;
4924 5031 return (0);
4925 5032
4926 5033 case _PC_TIMESTAMP_RESOLUTION:
4927 5034 /* nanosecond timestamp resolution */
4928 5035 *valp = 1L;
4929 5036 return (0);
4930 5037
4931 5038 default:
4932 5039 return (fs_pathconf(vp, cmd, valp, cr, ct));
4933 5040 }
4934 5041 }
4935 5042
4936 5043 /*ARGSUSED*/
4937 5044 static int
4938 5045 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4939 5046 caller_context_t *ct)
4940 5047 {
4941 5048 znode_t *zp = VTOZ(vp);
4942 5049 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4943 5050 int error;
4944 5051 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4945 5052
4946 5053 ZFS_ENTER(zfsvfs);
4947 5054 ZFS_VERIFY_ZP(zp);
4948 5055 error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4949 5056 ZFS_EXIT(zfsvfs);
4950 5057
4951 5058 return (error);
4952 5059 }
4953 5060
4954 5061 /*ARGSUSED*/
4955 5062 static int
4956 5063 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4957 5064 caller_context_t *ct)
4958 5065 {
4959 5066 znode_t *zp = VTOZ(vp);
4960 5067 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4961 5068 int error;
4962 5069 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4963 5070 zilog_t *zilog = zfsvfs->z_log;
4964 5071
4965 5072 ZFS_ENTER(zfsvfs);
4966 5073 ZFS_VERIFY_ZP(zp);
4967 5074
4968 5075 error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4969 5076
4970 5077 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4971 5078 zil_commit(zilog, 0);
4972 5079
4973 5080 ZFS_EXIT(zfsvfs);
4974 5081 return (error);
4975 5082 }
4976 5083
4977 5084 /*
4978 5085 * The smallest read we may consider to loan out an arcbuf.
4979 5086 * This must be a power of 2.
4980 5087 */
4981 5088 int zcr_blksz_min = (1 << 10); /* 1K */
4982 5089 /*
4983 5090 * If set to less than the file block size, allow loaning out of an
4984 5091 * arcbuf for a partial block read. This must be a power of 2.
4985 5092 */
4986 5093 int zcr_blksz_max = (1 << 17); /* 128K */
4987 5094
4988 5095 /*ARGSUSED*/
4989 5096 static int
4990 5097 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
4991 5098 caller_context_t *ct)
4992 5099 {
4993 5100 znode_t *zp = VTOZ(vp);
4994 5101 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4995 5102 int max_blksz = zfsvfs->z_max_blksz;
4996 5103 uio_t *uio = &xuio->xu_uio;
4997 5104 ssize_t size = uio->uio_resid;
4998 5105 offset_t offset = uio->uio_loffset;
4999 5106 int blksz;
5000 5107 int fullblk, i;
5001 5108 arc_buf_t *abuf;
5002 5109 ssize_t maxsize;
5003 5110 int preamble, postamble;
5004 5111
5005 5112 if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5006 5113 return (SET_ERROR(EINVAL));
5007 5114
5008 5115 ZFS_ENTER(zfsvfs);
5009 5116 ZFS_VERIFY_ZP(zp);
5010 5117 switch (ioflag) {
5011 5118 case UIO_WRITE:
5012 5119 /*
5013 5120 * Loan out an arc_buf for write if write size is bigger than
5014 5121 * max_blksz, and the file's block size is also max_blksz.
5015 5122 */
5016 5123 blksz = max_blksz;
5017 5124 if (size < blksz || zp->z_blksz != blksz) {
5018 5125 ZFS_EXIT(zfsvfs);
5019 5126 return (SET_ERROR(EINVAL));
5020 5127 }
5021 5128 /*
5022 5129 * Caller requests buffers for write before knowing where the
5023 5130 * write offset might be (e.g. NFS TCP write).
5024 5131 */
5025 5132 if (offset == -1) {
5026 5133 preamble = 0;
5027 5134 } else {
5028 5135 preamble = P2PHASE(offset, blksz);
5029 5136 if (preamble) {
5030 5137 preamble = blksz - preamble;
5031 5138 size -= preamble;
5032 5139 }
5033 5140 }
5034 5141
5035 5142 postamble = P2PHASE(size, blksz);
5036 5143 size -= postamble;
5037 5144
5038 5145 fullblk = size / blksz;
5039 5146 (void) dmu_xuio_init(xuio,
5040 5147 (preamble != 0) + fullblk + (postamble != 0));
5041 5148 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5042 5149 int, postamble, int,
5043 5150 (preamble != 0) + fullblk + (postamble != 0));
5044 5151
5045 5152 /*
5046 5153 * Have to fix iov base/len for partial buffers. They
5047 5154 * currently represent full arc_buf's.
5048 5155 */
5049 5156 if (preamble) {
5050 5157 /* data begins in the middle of the arc_buf */
5051 5158 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5052 5159 blksz);
5053 5160 ASSERT(abuf);
5054 5161 (void) dmu_xuio_add(xuio, abuf,
5055 5162 blksz - preamble, preamble);
5056 5163 }
5057 5164
5058 5165 for (i = 0; i < fullblk; i++) {
5059 5166 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5060 5167 blksz);
5061 5168 ASSERT(abuf);
5062 5169 (void) dmu_xuio_add(xuio, abuf, 0, blksz);
5063 5170 }
5064 5171
5065 5172 if (postamble) {
5066 5173 /* data ends in the middle of the arc_buf */
5067 5174 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5068 5175 blksz);
5069 5176 ASSERT(abuf);
5070 5177 (void) dmu_xuio_add(xuio, abuf, 0, postamble);
5071 5178 }
5072 5179 break;
5073 5180 case UIO_READ:
5074 5181 /*
5075 5182 * Loan out an arc_buf for read if the read size is larger than
5076 5183 * the current file block size. Block alignment is not
5077 5184 * considered. Partial arc_buf will be loaned out for read.
5078 5185 */
5079 5186 blksz = zp->z_blksz;
5080 5187 if (blksz < zcr_blksz_min)
5081 5188 blksz = zcr_blksz_min;
5082 5189 if (blksz > zcr_blksz_max)
5083 5190 blksz = zcr_blksz_max;
5084 5191 /* avoid potential complexity of dealing with it */
5085 5192 if (blksz > max_blksz) {
5086 5193 ZFS_EXIT(zfsvfs);
5087 5194 return (SET_ERROR(EINVAL));
5088 5195 }
5089 5196
5090 5197 maxsize = zp->z_size - uio->uio_loffset;
5091 5198 if (size > maxsize)
5092 5199 size = maxsize;
5093 5200
5094 5201 if (size < blksz || vn_has_cached_data(vp)) {
5095 5202 ZFS_EXIT(zfsvfs);
5096 5203 return (SET_ERROR(EINVAL));
5097 5204 }
5098 5205 break;
5099 5206 default:
5100 5207 ZFS_EXIT(zfsvfs);
5101 5208 return (SET_ERROR(EINVAL));
5102 5209 }
5103 5210
5104 5211 uio->uio_extflg = UIO_XUIO;
5105 5212 XUIO_XUZC_RW(xuio) = ioflag;
5106 5213 ZFS_EXIT(zfsvfs);
5107 5214 return (0);
5108 5215 }
5109 5216
5110 5217 /*ARGSUSED*/
5111 5218 static int
5112 5219 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5113 5220 {
5114 5221 int i;
5115 5222 arc_buf_t *abuf;
5116 5223 int ioflag = XUIO_XUZC_RW(xuio);
5117 5224
5118 5225 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5119 5226
5120 5227 i = dmu_xuio_cnt(xuio);
5121 5228 while (i-- > 0) {
5122 5229 abuf = dmu_xuio_arcbuf(xuio, i);
5123 5230 /*
5124 5231 * if abuf == NULL, it must be a write buffer
5125 5232 * that has been returned in zfs_write().
5126 5233 */
5127 5234 if (abuf)
5128 5235 dmu_return_arcbuf(abuf);
5129 5236 ASSERT(abuf || ioflag == UIO_WRITE);
5130 5237 }
5131 5238
5132 5239 dmu_xuio_fini(xuio);
5133 5240 return (0);
5134 5241 }
5135 5242
5136 5243 /*
5137 5244 * Predeclare these here so that the compiler assumes that
5138 5245 * this is an "old style" function declaration that does
5139 5246 * not include arguments => we won't get type mismatch errors
5140 5247 * in the initializations that follow.
5141 5248 */
5142 5249 static int zfs_inval();
5143 5250 static int zfs_isdir();
5144 5251
5145 5252 static int
5146 5253 zfs_inval()
5147 5254 {
5148 5255 return (SET_ERROR(EINVAL));
5149 5256 }
5150 5257
5151 5258 static int
5152 5259 zfs_isdir()
5153 5260 {
5154 5261 return (SET_ERROR(EISDIR));
5155 5262 }
5156 5263 /*
5157 5264 * Directory vnode operations template
5158 5265 */
5159 5266 vnodeops_t *zfs_dvnodeops;
5160 5267 const fs_operation_def_t zfs_dvnodeops_template[] = {
5161 5268 VOPNAME_OPEN, { .vop_open = zfs_open },
5162 5269 VOPNAME_CLOSE, { .vop_close = zfs_close },
5163 5270 VOPNAME_READ, { .error = zfs_isdir },
5164 5271 VOPNAME_WRITE, { .error = zfs_isdir },
5165 5272 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5166 5273 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5167 5274 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5168 5275 VOPNAME_ACCESS, { .vop_access = zfs_access },
5169 5276 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5170 5277 VOPNAME_CREATE, { .vop_create = zfs_create },
5171 5278 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5172 5279 VOPNAME_LINK, { .vop_link = zfs_link },
5173 5280 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5174 5281 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir },
5175 5282 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5176 5283 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5177 5284 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink },
5178 5285 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5179 5286 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5180 5287 VOPNAME_FID, { .vop_fid = zfs_fid },
5181 5288 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5182 5289 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5183 5290 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5184 5291 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5185 5292 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5186 5293 NULL, NULL
5187 5294 };
5188 5295
5189 5296 /*
5190 5297 * Regular file vnode operations template
5191 5298 */
5192 5299 vnodeops_t *zfs_fvnodeops;
5193 5300 const fs_operation_def_t zfs_fvnodeops_template[] = {
5194 5301 VOPNAME_OPEN, { .vop_open = zfs_open },
5195 5302 VOPNAME_CLOSE, { .vop_close = zfs_close },
5196 5303 VOPNAME_READ, { .vop_read = zfs_read },
5197 5304 VOPNAME_WRITE, { .vop_write = zfs_write },
5198 5305 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5199 5306 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5200 5307 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5201 5308 VOPNAME_ACCESS, { .vop_access = zfs_access },
5202 5309 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5203 5310 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5204 5311 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5205 5312 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5206 5313 VOPNAME_FID, { .vop_fid = zfs_fid },
5207 5314 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5208 5315 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock },
5209 5316 VOPNAME_SPACE, { .vop_space = zfs_space },
5210 5317 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage },
5211 5318 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage },
5212 5319 VOPNAME_MAP, { .vop_map = zfs_map },
5213 5320 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap },
5214 5321 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap },
5215 5322 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5216 5323 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5217 5324 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5218 5325 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5219 5326 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf },
5220 5327 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf },
5221 5328 NULL, NULL
5222 5329 };
5223 5330
5224 5331 /*
5225 5332 * Symbolic link vnode operations template
5226 5333 */
5227 5334 vnodeops_t *zfs_symvnodeops;
5228 5335 const fs_operation_def_t zfs_symvnodeops_template[] = {
5229 5336 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5230 5337 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5231 5338 VOPNAME_ACCESS, { .vop_access = zfs_access },
5232 5339 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5233 5340 VOPNAME_READLINK, { .vop_readlink = zfs_readlink },
5234 5341 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5235 5342 VOPNAME_FID, { .vop_fid = zfs_fid },
5236 5343 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5237 5344 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5238 5345 NULL, NULL
5239 5346 };
5240 5347
5241 5348 /*
5242 5349 * special share hidden files vnode operations template
5243 5350 */
5244 5351 vnodeops_t *zfs_sharevnodeops;
5245 5352 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5246 5353 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5247 5354 VOPNAME_ACCESS, { .vop_access = zfs_access },
5248 5355 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5249 5356 VOPNAME_FID, { .vop_fid = zfs_fid },
5250 5357 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5251 5358 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5252 5359 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5253 5360 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5254 5361 NULL, NULL
5255 5362 };
5256 5363
5257 5364 /*
5258 5365 * Extended attribute directory vnode operations template
5259 5366 *
5260 5367 * This template is identical to the directory vnodes
5261 5368 * operation template except for restricted operations:
5262 5369 * VOP_MKDIR()
5263 5370 * VOP_SYMLINK()
5264 5371 *
5265 5372 * Note that there are other restrictions embedded in:
5266 5373 * zfs_create() - restrict type to VREG
5267 5374 * zfs_link() - no links into/out of attribute space
5268 5375 * zfs_rename() - no moves into/out of attribute space
5269 5376 */
5270 5377 vnodeops_t *zfs_xdvnodeops;
5271 5378 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5272 5379 VOPNAME_OPEN, { .vop_open = zfs_open },
5273 5380 VOPNAME_CLOSE, { .vop_close = zfs_close },
5274 5381 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl },
5275 5382 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr },
5276 5383 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr },
5277 5384 VOPNAME_ACCESS, { .vop_access = zfs_access },
5278 5385 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup },
5279 5386 VOPNAME_CREATE, { .vop_create = zfs_create },
5280 5387 VOPNAME_REMOVE, { .vop_remove = zfs_remove },
5281 5388 VOPNAME_LINK, { .vop_link = zfs_link },
5282 5389 VOPNAME_RENAME, { .vop_rename = zfs_rename },
5283 5390 VOPNAME_MKDIR, { .error = zfs_inval },
5284 5391 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir },
5285 5392 VOPNAME_READDIR, { .vop_readdir = zfs_readdir },
5286 5393 VOPNAME_SYMLINK, { .error = zfs_inval },
5287 5394 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync },
5288 5395 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5289 5396 VOPNAME_FID, { .vop_fid = zfs_fid },
5290 5397 VOPNAME_SEEK, { .vop_seek = zfs_seek },
5291 5398 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5292 5399 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr },
5293 5400 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr },
5294 5401 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
5295 5402 NULL, NULL
5296 5403 };
5297 5404
5298 5405 /*
5299 5406 * Error vnode operations template
5300 5407 */
5301 5408 vnodeops_t *zfs_evnodeops;
5302 5409 const fs_operation_def_t zfs_evnodeops_template[] = {
5303 5410 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive },
5304 5411 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf },
5305 5412 NULL, NULL
5306 5413 };
|
↓ open down ↓ |
4338 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX