1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or https://opensource.org/licenses/CDDL-1.0. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2012, 2015 by Delphix. All rights reserved. 25 * Copyright (c) 2014 Integros [integros.com] 26 * Copyright 2017 Nexenta Systems, Inc. 27 */ 28 29 /* Portions Copyright 2007 Jeremy Teo */ 30 /* Portions Copyright 2010 Robert Milkowski */ 31 32 #include <sys/param.h> 33 #include <sys/time.h> 34 #include <sys/systm.h> 35 #include <sys/sysmacros.h> 36 #include <sys/resource.h> 37 #include <security/mac/mac_framework.h> 38 #include <sys/vfs.h> 39 #include <sys/endian.h> 40 #include <sys/vm.h> 41 #include <sys/vnode.h> 42 #if __FreeBSD_version >= 1300102 43 #include <sys/smr.h> 44 #endif 45 #include <sys/dirent.h> 46 #include <sys/file.h> 47 #include <sys/stat.h> 48 #include <sys/kmem.h> 49 #include <sys/taskq.h> 50 #include <sys/uio.h> 51 #include <sys/atomic.h> 52 #include <sys/namei.h> 53 #include <sys/mman.h> 54 #include <sys/cmn_err.h> 55 #include <sys/kdb.h> 56 #include <sys/sysproto.h> 57 #include <sys/errno.h> 58 #include <sys/unistd.h> 59 #include <sys/zfs_dir.h> 60 #include <sys/zfs_ioctl.h> 61 #include <sys/fs/zfs.h> 62 #include <sys/dmu.h> 63 #include <sys/dmu_objset.h> 64 #include <sys/spa.h> 65 #include <sys/txg.h> 66 #include <sys/dbuf.h> 67 #include <sys/zap.h> 68 #include <sys/sa.h> 69 #include <sys/policy.h> 70 #include <sys/sunddi.h> 71 #include <sys/filio.h> 72 #include <sys/sid.h> 73 #include <sys/zfs_ctldir.h> 74 #include <sys/zfs_fuid.h> 75 #include <sys/zfs_quota.h> 76 #include <sys/zfs_sa.h> 77 #include <sys/zfs_rlock.h> 78 #include <sys/bio.h> 79 #include <sys/buf.h> 80 #include <sys/sched.h> 81 #include <sys/acl.h> 82 #include <sys/vmmeter.h> 83 #include <vm/vm_param.h> 84 #include <sys/zil.h> 85 #include <sys/zfs_vnops.h> 86 #include <sys/module.h> 87 #include <sys/sysent.h> 88 #include <sys/dmu_impl.h> 89 #include <sys/brt.h> 90 #include <sys/zfeature.h> 91 92 #include <vm/vm_object.h> 93 94 #include <sys/extattr.h> 95 #include <sys/priv.h> 96 97 #ifndef VN_OPEN_INVFS 98 #define VN_OPEN_INVFS 0x0 99 #endif 100 101 VFS_SMR_DECLARE; 102 103 #if __FreeBSD_version < 1300103 104 #define NDFREE_PNBUF(ndp) NDFREE((ndp), NDF_ONLY_PNBUF) 105 #endif 106 107 #if __FreeBSD_version >= 1300047 108 #define vm_page_wire_lock(pp) 109 #define vm_page_wire_unlock(pp) 110 #else 111 #define vm_page_wire_lock(pp) vm_page_lock(pp) 112 #define vm_page_wire_unlock(pp) vm_page_unlock(pp) 113 #endif 114 115 #ifdef DEBUG_VFS_LOCKS 116 #define VNCHECKREF(vp) \ 117 VNASSERT((vp)->v_holdcnt > 0 && (vp)->v_usecount > 0, vp, \ 118 ("%s: wrong ref counts", __func__)); 119 #else 120 #define VNCHECKREF(vp) 121 #endif 122 123 #if __FreeBSD_version >= 1400045 124 typedef uint64_t cookie_t; 125 #else 126 typedef ulong_t cookie_t; 127 #endif 128 129 /* 130 * Programming rules. 131 * 132 * Each vnode op performs some logical unit of work. To do this, the ZPL must 133 * properly lock its in-core state, create a DMU transaction, do the work, 134 * record this work in the intent log (ZIL), commit the DMU transaction, 135 * and wait for the intent log to commit if it is a synchronous operation. 136 * Moreover, the vnode ops must work in both normal and log replay context. 137 * The ordering of events is important to avoid deadlocks and references 138 * to freed memory. The example below illustrates the following Big Rules: 139 * 140 * (1) A check must be made in each zfs thread for a mounted file system. 141 * This is done avoiding races using zfs_enter(zfsvfs). 142 * A zfs_exit(zfsvfs) is needed before all returns. Any znodes 143 * must be checked with zfs_verify_zp(zp). Both of these macros 144 * can return EIO from the calling function. 145 * 146 * (2) VN_RELE() should always be the last thing except for zil_commit() 147 * (if necessary) and zfs_exit(). This is for 3 reasons: 148 * First, if it's the last reference, the vnode/znode 149 * can be freed, so the zp may point to freed memory. Second, the last 150 * reference will call zfs_zinactive(), which may induce a lot of work -- 151 * pushing cached pages (which acquires range locks) and syncing out 152 * cached atime changes. Third, zfs_zinactive() may require a new tx, 153 * which could deadlock the system if you were already holding one. 154 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). 155 * 156 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 157 * as they can span dmu_tx_assign() calls. 158 * 159 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to 160 * dmu_tx_assign(). This is critical because we don't want to block 161 * while holding locks. 162 * 163 * If no ZPL locks are held (aside from zfs_enter()), use TXG_WAIT. This 164 * reduces lock contention and CPU usage when we must wait (note that if 165 * throughput is constrained by the storage, nearly every transaction 166 * must wait). 167 * 168 * Note, in particular, that if a lock is sometimes acquired before 169 * the tx assigns, and sometimes after (e.g. z_lock), then failing 170 * to use a non-blocking assign can deadlock the system. The scenario: 171 * 172 * Thread A has grabbed a lock before calling dmu_tx_assign(). 173 * Thread B is in an already-assigned tx, and blocks for this lock. 174 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 175 * forever, because the previous txg can't quiesce until B's tx commits. 176 * 177 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 178 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent 179 * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT, 180 * to indicate that this operation has already called dmu_tx_wait(). 181 * This will ensure that we don't retry forever, waiting a short bit 182 * each time. 183 * 184 * (5) If the operation succeeded, generate the intent log entry for it 185 * before dropping locks. This ensures that the ordering of events 186 * in the intent log matches the order in which they actually occurred. 187 * During ZIL replay the zfs_log_* functions will update the sequence 188 * number to indicate the zil transaction has replayed. 189 * 190 * (6) At the end of each vnode op, the DMU tx must always commit, 191 * regardless of whether there were any errors. 192 * 193 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 194 * to ensure that synchronous semantics are provided when necessary. 195 * 196 * In general, this is how things should be ordered in each vnode op: 197 * 198 * zfs_enter(zfsvfs); // exit if unmounted 199 * top: 200 * zfs_dirent_lookup(&dl, ...) // lock directory entry (may VN_HOLD()) 201 * rw_enter(...); // grab any other locks you need 202 * tx = dmu_tx_create(...); // get DMU tx 203 * dmu_tx_hold_*(); // hold each object you might modify 204 * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 205 * if (error) { 206 * rw_exit(...); // drop locks 207 * zfs_dirent_unlock(dl); // unlock directory entry 208 * VN_RELE(...); // release held vnodes 209 * if (error == ERESTART) { 210 * waited = B_TRUE; 211 * dmu_tx_wait(tx); 212 * dmu_tx_abort(tx); 213 * goto top; 214 * } 215 * dmu_tx_abort(tx); // abort DMU tx 216 * zfs_exit(zfsvfs); // finished in zfs 217 * return (error); // really out of space 218 * } 219 * error = do_real_work(); // do whatever this VOP does 220 * if (error == 0) 221 * zfs_log_*(...); // on success, make ZIL entry 222 * dmu_tx_commit(tx); // commit DMU tx -- error or not 223 * rw_exit(...); // drop locks 224 * zfs_dirent_unlock(dl); // unlock directory entry 225 * VN_RELE(...); // release held vnodes 226 * zil_commit(zilog, foid); // synchronous when necessary 227 * zfs_exit(zfsvfs); // finished in zfs 228 * return (error); // done, report error 229 */ 230 static int 231 zfs_open(vnode_t **vpp, int flag, cred_t *cr) 232 { 233 (void) cr; 234 znode_t *zp = VTOZ(*vpp); 235 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 236 int error; 237 238 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 239 return (error); 240 241 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 242 ((flag & FAPPEND) == 0)) { 243 zfs_exit(zfsvfs, FTAG); 244 return (SET_ERROR(EPERM)); 245 } 246 247 /* Keep a count of the synchronous opens in the znode */ 248 if (flag & O_SYNC) 249 atomic_inc_32(&zp->z_sync_cnt); 250 251 zfs_exit(zfsvfs, FTAG); 252 return (0); 253 } 254 255 static int 256 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr) 257 { 258 (void) offset, (void) cr; 259 znode_t *zp = VTOZ(vp); 260 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 261 int error; 262 263 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 264 return (error); 265 266 /* Decrement the synchronous opens in the znode */ 267 if ((flag & O_SYNC) && (count == 1)) 268 atomic_dec_32(&zp->z_sync_cnt); 269 270 zfs_exit(zfsvfs, FTAG); 271 return (0); 272 } 273 274 static int 275 zfs_ioctl(vnode_t *vp, ulong_t com, intptr_t data, int flag, cred_t *cred, 276 int *rvalp) 277 { 278 (void) flag, (void) cred, (void) rvalp; 279 loff_t off; 280 int error; 281 282 switch (com) { 283 case _FIOFFS: 284 { 285 return (0); 286 287 /* 288 * The following two ioctls are used by bfu. Faking out, 289 * necessary to avoid bfu errors. 290 */ 291 } 292 case _FIOGDIO: 293 case _FIOSDIO: 294 { 295 return (0); 296 } 297 298 case F_SEEK_DATA: 299 case F_SEEK_HOLE: 300 { 301 off = *(offset_t *)data; 302 /* offset parameter is in/out */ 303 error = zfs_holey(VTOZ(vp), com, &off); 304 if (error) 305 return (error); 306 *(offset_t *)data = off; 307 return (0); 308 } 309 } 310 return (SET_ERROR(ENOTTY)); 311 } 312 313 static vm_page_t 314 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) 315 { 316 vm_object_t obj; 317 vm_page_t pp; 318 int64_t end; 319 320 /* 321 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE 322 * aligned boundaries, if the range is not aligned. As a result a 323 * DEV_BSIZE subrange with partially dirty data may get marked as clean. 324 * It may happen that all DEV_BSIZE subranges are marked clean and thus 325 * the whole page would be considered clean despite have some 326 * dirty data. 327 * For this reason we should shrink the range to DEV_BSIZE aligned 328 * boundaries before calling vm_page_clear_dirty. 329 */ 330 end = rounddown2(off + nbytes, DEV_BSIZE); 331 off = roundup2(off, DEV_BSIZE); 332 nbytes = end - off; 333 334 obj = vp->v_object; 335 zfs_vmobject_assert_wlocked_12(obj); 336 #if __FreeBSD_version < 1300050 337 for (;;) { 338 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 339 pp->valid) { 340 if (vm_page_xbusied(pp)) { 341 /* 342 * Reference the page before unlocking and 343 * sleeping so that the page daemon is less 344 * likely to reclaim it. 345 */ 346 vm_page_reference(pp); 347 vm_page_lock(pp); 348 zfs_vmobject_wunlock(obj); 349 vm_page_busy_sleep(pp, "zfsmwb", true); 350 zfs_vmobject_wlock(obj); 351 continue; 352 } 353 vm_page_sbusy(pp); 354 } else if (pp != NULL) { 355 ASSERT(!pp->valid); 356 pp = NULL; 357 } 358 if (pp != NULL) { 359 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 360 vm_object_pip_add(obj, 1); 361 pmap_remove_write(pp); 362 if (nbytes != 0) 363 vm_page_clear_dirty(pp, off, nbytes); 364 } 365 break; 366 } 367 #else 368 vm_page_grab_valid_unlocked(&pp, obj, OFF_TO_IDX(start), 369 VM_ALLOC_NOCREAT | VM_ALLOC_SBUSY | VM_ALLOC_NORMAL | 370 VM_ALLOC_IGN_SBUSY); 371 if (pp != NULL) { 372 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 373 vm_object_pip_add(obj, 1); 374 pmap_remove_write(pp); 375 if (nbytes != 0) 376 vm_page_clear_dirty(pp, off, nbytes); 377 } 378 #endif 379 return (pp); 380 } 381 382 static void 383 page_unbusy(vm_page_t pp) 384 { 385 386 vm_page_sunbusy(pp); 387 #if __FreeBSD_version >= 1300041 388 vm_object_pip_wakeup(pp->object); 389 #else 390 vm_object_pip_subtract(pp->object, 1); 391 #endif 392 } 393 394 #if __FreeBSD_version > 1300051 395 static vm_page_t 396 page_hold(vnode_t *vp, int64_t start) 397 { 398 vm_object_t obj; 399 vm_page_t m; 400 401 obj = vp->v_object; 402 vm_page_grab_valid_unlocked(&m, obj, OFF_TO_IDX(start), 403 VM_ALLOC_NOCREAT | VM_ALLOC_WIRED | VM_ALLOC_IGN_SBUSY | 404 VM_ALLOC_NOBUSY); 405 return (m); 406 } 407 #else 408 static vm_page_t 409 page_hold(vnode_t *vp, int64_t start) 410 { 411 vm_object_t obj; 412 vm_page_t pp; 413 414 obj = vp->v_object; 415 zfs_vmobject_assert_wlocked(obj); 416 417 for (;;) { 418 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 419 pp->valid) { 420 if (vm_page_xbusied(pp)) { 421 /* 422 * Reference the page before unlocking and 423 * sleeping so that the page daemon is less 424 * likely to reclaim it. 425 */ 426 vm_page_reference(pp); 427 vm_page_lock(pp); 428 zfs_vmobject_wunlock(obj); 429 vm_page_busy_sleep(pp, "zfsmwb", true); 430 zfs_vmobject_wlock(obj); 431 continue; 432 } 433 434 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 435 vm_page_wire_lock(pp); 436 vm_page_hold(pp); 437 vm_page_wire_unlock(pp); 438 439 } else 440 pp = NULL; 441 break; 442 } 443 return (pp); 444 } 445 #endif 446 447 static void 448 page_unhold(vm_page_t pp) 449 { 450 451 vm_page_wire_lock(pp); 452 #if __FreeBSD_version >= 1300035 453 vm_page_unwire(pp, PQ_ACTIVE); 454 #else 455 vm_page_unhold(pp); 456 #endif 457 vm_page_wire_unlock(pp); 458 } 459 460 /* 461 * When a file is memory mapped, we must keep the IO data synchronized 462 * between the DMU cache and the memory mapped pages. What this means: 463 * 464 * On Write: If we find a memory mapped page, we write to *both* 465 * the page and the dmu buffer. 466 */ 467 void 468 update_pages(znode_t *zp, int64_t start, int len, objset_t *os) 469 { 470 vm_object_t obj; 471 struct sf_buf *sf; 472 vnode_t *vp = ZTOV(zp); 473 caddr_t va; 474 int off; 475 476 ASSERT3P(vp->v_mount, !=, NULL); 477 obj = vp->v_object; 478 ASSERT3P(obj, !=, NULL); 479 480 off = start & PAGEOFFSET; 481 zfs_vmobject_wlock_12(obj); 482 #if __FreeBSD_version >= 1300041 483 vm_object_pip_add(obj, 1); 484 #endif 485 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 486 vm_page_t pp; 487 int nbytes = imin(PAGESIZE - off, len); 488 489 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) { 490 zfs_vmobject_wunlock_12(obj); 491 492 va = zfs_map_page(pp, &sf); 493 (void) dmu_read(os, zp->z_id, start + off, nbytes, 494 va + off, DMU_READ_PREFETCH); 495 zfs_unmap_page(sf); 496 497 zfs_vmobject_wlock_12(obj); 498 page_unbusy(pp); 499 } 500 len -= nbytes; 501 off = 0; 502 } 503 #if __FreeBSD_version >= 1300041 504 vm_object_pip_wakeup(obj); 505 #else 506 vm_object_pip_wakeupn(obj, 0); 507 #endif 508 zfs_vmobject_wunlock_12(obj); 509 } 510 511 /* 512 * Read with UIO_NOCOPY flag means that sendfile(2) requests 513 * ZFS to populate a range of page cache pages with data. 514 * 515 * NOTE: this function could be optimized to pre-allocate 516 * all pages in advance, drain exclusive busy on all of them, 517 * map them into contiguous KVA region and populate them 518 * in one single dmu_read() call. 519 */ 520 int 521 mappedread_sf(znode_t *zp, int nbytes, zfs_uio_t *uio) 522 { 523 vnode_t *vp = ZTOV(zp); 524 objset_t *os = zp->z_zfsvfs->z_os; 525 struct sf_buf *sf; 526 vm_object_t obj; 527 vm_page_t pp; 528 int64_t start; 529 caddr_t va; 530 int len = nbytes; 531 int error = 0; 532 533 ASSERT3U(zfs_uio_segflg(uio), ==, UIO_NOCOPY); 534 ASSERT3P(vp->v_mount, !=, NULL); 535 obj = vp->v_object; 536 ASSERT3P(obj, !=, NULL); 537 ASSERT0(zfs_uio_offset(uio) & PAGEOFFSET); 538 539 zfs_vmobject_wlock_12(obj); 540 for (start = zfs_uio_offset(uio); len > 0; start += PAGESIZE) { 541 int bytes = MIN(PAGESIZE, len); 542 543 pp = vm_page_grab_unlocked(obj, OFF_TO_IDX(start), 544 VM_ALLOC_SBUSY | VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); 545 if (vm_page_none_valid(pp)) { 546 zfs_vmobject_wunlock_12(obj); 547 va = zfs_map_page(pp, &sf); 548 error = dmu_read(os, zp->z_id, start, bytes, va, 549 DMU_READ_PREFETCH); 550 if (bytes != PAGESIZE && error == 0) 551 memset(va + bytes, 0, PAGESIZE - bytes); 552 zfs_unmap_page(sf); 553 zfs_vmobject_wlock_12(obj); 554 #if __FreeBSD_version >= 1300081 555 if (error == 0) { 556 vm_page_valid(pp); 557 vm_page_activate(pp); 558 vm_page_do_sunbusy(pp); 559 } else { 560 zfs_vmobject_wlock(obj); 561 if (!vm_page_wired(pp) && pp->valid == 0 && 562 vm_page_busy_tryupgrade(pp)) 563 vm_page_free(pp); 564 else 565 vm_page_sunbusy(pp); 566 zfs_vmobject_wunlock(obj); 567 } 568 #else 569 vm_page_do_sunbusy(pp); 570 vm_page_lock(pp); 571 if (error) { 572 if (pp->wire_count == 0 && pp->valid == 0 && 573 !vm_page_busied(pp)) 574 vm_page_free(pp); 575 } else { 576 pp->valid = VM_PAGE_BITS_ALL; 577 vm_page_activate(pp); 578 } 579 vm_page_unlock(pp); 580 #endif 581 } else { 582 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 583 vm_page_do_sunbusy(pp); 584 } 585 if (error) 586 break; 587 zfs_uio_advance(uio, bytes); 588 len -= bytes; 589 } 590 zfs_vmobject_wunlock_12(obj); 591 return (error); 592 } 593 594 /* 595 * When a file is memory mapped, we must keep the IO data synchronized 596 * between the DMU cache and the memory mapped pages. What this means: 597 * 598 * On Read: We "read" preferentially from memory mapped pages, 599 * else we default from the dmu buffer. 600 * 601 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 602 * the file is memory mapped. 603 */ 604 int 605 mappedread(znode_t *zp, int nbytes, zfs_uio_t *uio) 606 { 607 vnode_t *vp = ZTOV(zp); 608 vm_object_t obj; 609 int64_t start; 610 int len = nbytes; 611 int off; 612 int error = 0; 613 614 ASSERT3P(vp->v_mount, !=, NULL); 615 obj = vp->v_object; 616 ASSERT3P(obj, !=, NULL); 617 618 start = zfs_uio_offset(uio); 619 off = start & PAGEOFFSET; 620 zfs_vmobject_wlock_12(obj); 621 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 622 vm_page_t pp; 623 uint64_t bytes = MIN(PAGESIZE - off, len); 624 625 if ((pp = page_hold(vp, start))) { 626 struct sf_buf *sf; 627 caddr_t va; 628 629 zfs_vmobject_wunlock_12(obj); 630 va = zfs_map_page(pp, &sf); 631 error = vn_io_fault_uiomove(va + off, bytes, 632 GET_UIO_STRUCT(uio)); 633 zfs_unmap_page(sf); 634 zfs_vmobject_wlock_12(obj); 635 page_unhold(pp); 636 } else { 637 zfs_vmobject_wunlock_12(obj); 638 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 639 uio, bytes); 640 zfs_vmobject_wlock_12(obj); 641 } 642 len -= bytes; 643 off = 0; 644 if (error) 645 break; 646 } 647 zfs_vmobject_wunlock_12(obj); 648 return (error); 649 } 650 651 int 652 zfs_write_simple(znode_t *zp, const void *data, size_t len, 653 loff_t pos, size_t *presid) 654 { 655 int error = 0; 656 ssize_t resid; 657 658 error = vn_rdwr(UIO_WRITE, ZTOV(zp), __DECONST(void *, data), len, pos, 659 UIO_SYSSPACE, IO_SYNC, kcred, NOCRED, &resid, curthread); 660 661 if (error) { 662 return (SET_ERROR(error)); 663 } else if (presid == NULL) { 664 if (resid != 0) { 665 error = SET_ERROR(EIO); 666 } 667 } else { 668 *presid = resid; 669 } 670 return (error); 671 } 672 673 void 674 zfs_zrele_async(znode_t *zp) 675 { 676 vnode_t *vp = ZTOV(zp); 677 objset_t *os = ITOZSB(vp)->z_os; 678 679 VN_RELE_ASYNC(vp, dsl_pool_zrele_taskq(dmu_objset_pool(os))); 680 } 681 682 static int 683 zfs_dd_callback(struct mount *mp, void *arg, int lkflags, struct vnode **vpp) 684 { 685 int error; 686 687 *vpp = arg; 688 error = vn_lock(*vpp, lkflags); 689 if (error != 0) 690 vrele(*vpp); 691 return (error); 692 } 693 694 static int 695 zfs_lookup_lock(vnode_t *dvp, vnode_t *vp, const char *name, int lkflags) 696 { 697 znode_t *zdp = VTOZ(dvp); 698 zfsvfs_t *zfsvfs __unused = zdp->z_zfsvfs; 699 int error; 700 int ltype; 701 702 if (zfsvfs->z_replay == B_FALSE) 703 ASSERT_VOP_LOCKED(dvp, __func__); 704 705 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) { 706 ASSERT3P(dvp, ==, vp); 707 vref(dvp); 708 ltype = lkflags & LK_TYPE_MASK; 709 if (ltype != VOP_ISLOCKED(dvp)) { 710 if (ltype == LK_EXCLUSIVE) 711 vn_lock(dvp, LK_UPGRADE | LK_RETRY); 712 else /* if (ltype == LK_SHARED) */ 713 vn_lock(dvp, LK_DOWNGRADE | LK_RETRY); 714 715 /* 716 * Relock for the "." case could leave us with 717 * reclaimed vnode. 718 */ 719 if (VN_IS_DOOMED(dvp)) { 720 vrele(dvp); 721 return (SET_ERROR(ENOENT)); 722 } 723 } 724 return (0); 725 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) { 726 /* 727 * Note that in this case, dvp is the child vnode, and we 728 * are looking up the parent vnode - exactly reverse from 729 * normal operation. Unlocking dvp requires some rather 730 * tricky unlock/relock dance to prevent mp from being freed; 731 * use vn_vget_ino_gen() which takes care of all that. 732 * 733 * XXX Note that there is a time window when both vnodes are 734 * unlocked. It is possible, although highly unlikely, that 735 * during that window the parent-child relationship between 736 * the vnodes may change, for example, get reversed. 737 * In that case we would have a wrong lock order for the vnodes. 738 * All other filesystems seem to ignore this problem, so we 739 * do the same here. 740 * A potential solution could be implemented as follows: 741 * - using LK_NOWAIT when locking the second vnode and retrying 742 * if necessary 743 * - checking that the parent-child relationship still holds 744 * after locking both vnodes and retrying if it doesn't 745 */ 746 error = vn_vget_ino_gen(dvp, zfs_dd_callback, vp, lkflags, &vp); 747 return (error); 748 } else { 749 error = vn_lock(vp, lkflags); 750 if (error != 0) 751 vrele(vp); 752 return (error); 753 } 754 } 755 756 /* 757 * Lookup an entry in a directory, or an extended attribute directory. 758 * If it exists, return a held vnode reference for it. 759 * 760 * IN: dvp - vnode of directory to search. 761 * nm - name of entry to lookup. 762 * pnp - full pathname to lookup [UNUSED]. 763 * flags - LOOKUP_XATTR set if looking for an attribute. 764 * rdir - root directory vnode [UNUSED]. 765 * cr - credentials of caller. 766 * ct - caller context 767 * 768 * OUT: vpp - vnode of located entry, NULL if not found. 769 * 770 * RETURN: 0 on success, error code on failure. 771 * 772 * Timestamps: 773 * NA 774 */ 775 static int 776 zfs_lookup(vnode_t *dvp, const char *nm, vnode_t **vpp, 777 struct componentname *cnp, int nameiop, cred_t *cr, int flags, 778 boolean_t cached) 779 { 780 znode_t *zdp = VTOZ(dvp); 781 znode_t *zp; 782 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 783 #if __FreeBSD_version > 1300124 784 seqc_t dvp_seqc; 785 #endif 786 int error = 0; 787 788 /* 789 * Fast path lookup, however we must skip DNLC lookup 790 * for case folding or normalizing lookups because the 791 * DNLC code only stores the passed in name. This means 792 * creating 'a' and removing 'A' on a case insensitive 793 * file system would work, but DNLC still thinks 'a' 794 * exists and won't let you create it again on the next 795 * pass through fast path. 796 */ 797 if (!(flags & LOOKUP_XATTR)) { 798 if (dvp->v_type != VDIR) { 799 return (SET_ERROR(ENOTDIR)); 800 } else if (zdp->z_sa_hdl == NULL) { 801 return (SET_ERROR(EIO)); 802 } 803 } 804 805 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, 806 const char *, nm); 807 808 if ((error = zfs_enter_verify_zp(zfsvfs, zdp, FTAG)) != 0) 809 return (error); 810 811 #if __FreeBSD_version > 1300124 812 dvp_seqc = vn_seqc_read_notmodify(dvp); 813 #endif 814 815 *vpp = NULL; 816 817 if (flags & LOOKUP_XATTR) { 818 /* 819 * If the xattr property is off, refuse the lookup request. 820 */ 821 if (!(zfsvfs->z_flags & ZSB_XATTR)) { 822 zfs_exit(zfsvfs, FTAG); 823 return (SET_ERROR(EOPNOTSUPP)); 824 } 825 826 /* 827 * We don't allow recursive attributes.. 828 * Maybe someday we will. 829 */ 830 if (zdp->z_pflags & ZFS_XATTR) { 831 zfs_exit(zfsvfs, FTAG); 832 return (SET_ERROR(EINVAL)); 833 } 834 835 if ((error = zfs_get_xattrdir(VTOZ(dvp), &zp, cr, flags))) { 836 zfs_exit(zfsvfs, FTAG); 837 return (error); 838 } 839 *vpp = ZTOV(zp); 840 841 /* 842 * Do we have permission to get into attribute directory? 843 */ 844 error = zfs_zaccess(zp, ACE_EXECUTE, 0, B_FALSE, cr, NULL); 845 if (error) { 846 vrele(ZTOV(zp)); 847 } 848 849 zfs_exit(zfsvfs, FTAG); 850 return (error); 851 } 852 853 /* 854 * Check accessibility of directory if we're not coming in via 855 * VOP_CACHEDLOOKUP. 856 */ 857 if (!cached) { 858 #ifdef NOEXECCHECK 859 if ((cnp->cn_flags & NOEXECCHECK) != 0) { 860 cnp->cn_flags &= ~NOEXECCHECK; 861 } else 862 #endif 863 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr, 864 NULL))) { 865 zfs_exit(zfsvfs, FTAG); 866 return (error); 867 } 868 } 869 870 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 871 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 872 zfs_exit(zfsvfs, FTAG); 873 return (SET_ERROR(EILSEQ)); 874 } 875 876 877 /* 878 * First handle the special cases. 879 */ 880 if ((cnp->cn_flags & ISDOTDOT) != 0) { 881 /* 882 * If we are a snapshot mounted under .zfs, return 883 * the vp for the snapshot directory. 884 */ 885 if (zdp->z_id == zfsvfs->z_root && zfsvfs->z_parent != zfsvfs) { 886 struct componentname cn; 887 vnode_t *zfsctl_vp; 888 int ltype; 889 890 zfs_exit(zfsvfs, FTAG); 891 ltype = VOP_ISLOCKED(dvp); 892 VOP_UNLOCK1(dvp); 893 error = zfsctl_root(zfsvfs->z_parent, LK_SHARED, 894 &zfsctl_vp); 895 if (error == 0) { 896 cn.cn_nameptr = "snapshot"; 897 cn.cn_namelen = strlen(cn.cn_nameptr); 898 cn.cn_nameiop = cnp->cn_nameiop; 899 cn.cn_flags = cnp->cn_flags & ~ISDOTDOT; 900 cn.cn_lkflags = cnp->cn_lkflags; 901 error = VOP_LOOKUP(zfsctl_vp, vpp, &cn); 902 vput(zfsctl_vp); 903 } 904 vn_lock(dvp, ltype | LK_RETRY); 905 return (error); 906 } 907 } 908 if (zfs_has_ctldir(zdp) && strcmp(nm, ZFS_CTLDIR_NAME) == 0) { 909 zfs_exit(zfsvfs, FTAG); 910 if ((cnp->cn_flags & ISLASTCN) != 0 && nameiop != LOOKUP) 911 return (SET_ERROR(ENOTSUP)); 912 error = zfsctl_root(zfsvfs, cnp->cn_lkflags, vpp); 913 return (error); 914 } 915 916 /* 917 * The loop is retry the lookup if the parent-child relationship 918 * changes during the dot-dot locking complexities. 919 */ 920 for (;;) { 921 uint64_t parent; 922 923 error = zfs_dirlook(zdp, nm, &zp); 924 if (error == 0) 925 *vpp = ZTOV(zp); 926 927 zfs_exit(zfsvfs, FTAG); 928 if (error != 0) 929 break; 930 931 error = zfs_lookup_lock(dvp, *vpp, nm, cnp->cn_lkflags); 932 if (error != 0) { 933 /* 934 * If we've got a locking error, then the vnode 935 * got reclaimed because of a force unmount. 936 * We never enter doomed vnodes into the name cache. 937 */ 938 *vpp = NULL; 939 return (error); 940 } 941 942 if ((cnp->cn_flags & ISDOTDOT) == 0) 943 break; 944 945 if ((error = zfs_enter(zfsvfs, FTAG)) != 0) { 946 vput(ZTOV(zp)); 947 *vpp = NULL; 948 return (error); 949 } 950 if (zdp->z_sa_hdl == NULL) { 951 error = SET_ERROR(EIO); 952 } else { 953 error = sa_lookup(zdp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 954 &parent, sizeof (parent)); 955 } 956 if (error != 0) { 957 zfs_exit(zfsvfs, FTAG); 958 vput(ZTOV(zp)); 959 break; 960 } 961 if (zp->z_id == parent) { 962 zfs_exit(zfsvfs, FTAG); 963 break; 964 } 965 vput(ZTOV(zp)); 966 } 967 968 if (error != 0) 969 *vpp = NULL; 970 971 /* Translate errors and add SAVENAME when needed. */ 972 if (cnp->cn_flags & ISLASTCN) { 973 switch (nameiop) { 974 case CREATE: 975 case RENAME: 976 if (error == ENOENT) { 977 error = EJUSTRETURN; 978 #if __FreeBSD_version < 1400068 979 cnp->cn_flags |= SAVENAME; 980 #endif 981 break; 982 } 983 zfs_fallthrough; 984 case DELETE: 985 #if __FreeBSD_version < 1400068 986 if (error == 0) 987 cnp->cn_flags |= SAVENAME; 988 #endif 989 break; 990 } 991 } 992 993 #if __FreeBSD_version > 1300124 994 if ((cnp->cn_flags & ISDOTDOT) != 0) { 995 /* 996 * FIXME: zfs_lookup_lock relocks vnodes and does nothing to 997 * handle races. In particular different callers may end up 998 * with different vnodes and will try to add conflicting 999 * entries to the namecache. 1000 * 1001 * While finding different result may be acceptable in face 1002 * of concurrent modification, adding conflicting entries 1003 * trips over an assert in the namecache. 1004 * 1005 * Ultimately let an entry through once everything settles. 1006 */ 1007 if (!vn_seqc_consistent(dvp, dvp_seqc)) { 1008 cnp->cn_flags &= ~MAKEENTRY; 1009 } 1010 } 1011 #endif 1012 1013 /* Insert name into cache (as non-existent) if appropriate. */ 1014 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay && 1015 error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0) 1016 cache_enter(dvp, NULL, cnp); 1017 1018 /* Insert name into cache if appropriate. */ 1019 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay && 1020 error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1021 if (!(cnp->cn_flags & ISLASTCN) || 1022 (nameiop != DELETE && nameiop != RENAME)) { 1023 cache_enter(dvp, *vpp, cnp); 1024 } 1025 } 1026 1027 return (error); 1028 } 1029 1030 /* 1031 * Attempt to create a new entry in a directory. If the entry 1032 * already exists, truncate the file if permissible, else return 1033 * an error. Return the vp of the created or trunc'd file. 1034 * 1035 * IN: dvp - vnode of directory to put new file entry in. 1036 * name - name of new file entry. 1037 * vap - attributes of new file. 1038 * excl - flag indicating exclusive or non-exclusive mode. 1039 * mode - mode to open file with. 1040 * cr - credentials of caller. 1041 * flag - large file flag [UNUSED]. 1042 * ct - caller context 1043 * vsecp - ACL to be set 1044 * mnt_ns - Unused on FreeBSD 1045 * 1046 * OUT: vpp - vnode of created or trunc'd entry. 1047 * 1048 * RETURN: 0 on success, error code on failure. 1049 * 1050 * Timestamps: 1051 * dvp - ctime|mtime updated if new entry created 1052 * vp - ctime|mtime always, atime if new 1053 */ 1054 int 1055 zfs_create(znode_t *dzp, const char *name, vattr_t *vap, int excl, int mode, 1056 znode_t **zpp, cred_t *cr, int flag, vsecattr_t *vsecp, zidmap_t *mnt_ns) 1057 { 1058 (void) excl, (void) mode, (void) flag; 1059 znode_t *zp; 1060 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1061 zilog_t *zilog; 1062 objset_t *os; 1063 dmu_tx_t *tx; 1064 int error; 1065 uid_t uid = crgetuid(cr); 1066 gid_t gid = crgetgid(cr); 1067 uint64_t projid = ZFS_DEFAULT_PROJID; 1068 zfs_acl_ids_t acl_ids; 1069 boolean_t fuid_dirtied; 1070 uint64_t txtype; 1071 #ifdef DEBUG_VFS_LOCKS 1072 vnode_t *dvp = ZTOV(dzp); 1073 #endif 1074 1075 /* 1076 * If we have an ephemeral id, ACL, or XVATTR then 1077 * make sure file system is at proper version 1078 */ 1079 if (zfsvfs->z_use_fuids == B_FALSE && 1080 (vsecp || (vap->va_mask & AT_XVATTR) || 1081 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1082 return (SET_ERROR(EINVAL)); 1083 1084 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 1085 return (error); 1086 os = zfsvfs->z_os; 1087 zilog = zfsvfs->z_log; 1088 1089 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1090 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1091 zfs_exit(zfsvfs, FTAG); 1092 return (SET_ERROR(EILSEQ)); 1093 } 1094 1095 if (vap->va_mask & AT_XVATTR) { 1096 if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap, 1097 crgetuid(cr), cr, vap->va_type)) != 0) { 1098 zfs_exit(zfsvfs, FTAG); 1099 return (error); 1100 } 1101 } 1102 1103 *zpp = NULL; 1104 1105 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) 1106 vap->va_mode &= ~S_ISVTX; 1107 1108 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW); 1109 if (error) { 1110 zfs_exit(zfsvfs, FTAG); 1111 return (error); 1112 } 1113 ASSERT3P(zp, ==, NULL); 1114 1115 /* 1116 * Create a new file object and update the directory 1117 * to reference it. 1118 */ 1119 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) { 1120 goto out; 1121 } 1122 1123 /* 1124 * We only support the creation of regular files in 1125 * extended attribute directories. 1126 */ 1127 1128 if ((dzp->z_pflags & ZFS_XATTR) && 1129 (vap->va_type != VREG)) { 1130 error = SET_ERROR(EINVAL); 1131 goto out; 1132 } 1133 1134 if ((error = zfs_acl_ids_create(dzp, 0, vap, 1135 cr, vsecp, &acl_ids, NULL)) != 0) 1136 goto out; 1137 1138 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode)) 1139 projid = zfs_inherit_projid(dzp); 1140 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) { 1141 zfs_acl_ids_free(&acl_ids); 1142 error = SET_ERROR(EDQUOT); 1143 goto out; 1144 } 1145 1146 getnewvnode_reserve_(); 1147 1148 tx = dmu_tx_create(os); 1149 1150 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1151 ZFS_SA_BASE_ATTR_SIZE); 1152 1153 fuid_dirtied = zfsvfs->z_fuid_dirty; 1154 if (fuid_dirtied) 1155 zfs_fuid_txhold(zfsvfs, tx); 1156 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1157 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1158 if (!zfsvfs->z_use_sa && 1159 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1160 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1161 0, acl_ids.z_aclp->z_acl_bytes); 1162 } 1163 error = dmu_tx_assign(tx, TXG_WAIT); 1164 if (error) { 1165 zfs_acl_ids_free(&acl_ids); 1166 dmu_tx_abort(tx); 1167 getnewvnode_drop_reserve(); 1168 zfs_exit(zfsvfs, FTAG); 1169 return (error); 1170 } 1171 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1172 if (fuid_dirtied) 1173 zfs_fuid_sync(zfsvfs, tx); 1174 1175 (void) zfs_link_create(dzp, name, zp, tx, ZNEW); 1176 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1177 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1178 vsecp, acl_ids.z_fuidp, vap); 1179 zfs_acl_ids_free(&acl_ids); 1180 dmu_tx_commit(tx); 1181 1182 getnewvnode_drop_reserve(); 1183 1184 out: 1185 VNCHECKREF(dvp); 1186 if (error == 0) { 1187 *zpp = zp; 1188 } 1189 1190 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1191 zil_commit(zilog, 0); 1192 1193 zfs_exit(zfsvfs, FTAG); 1194 return (error); 1195 } 1196 1197 /* 1198 * Remove an entry from a directory. 1199 * 1200 * IN: dvp - vnode of directory to remove entry from. 1201 * name - name of entry to remove. 1202 * cr - credentials of caller. 1203 * ct - caller context 1204 * flags - case flags 1205 * 1206 * RETURN: 0 on success, error code on failure. 1207 * 1208 * Timestamps: 1209 * dvp - ctime|mtime 1210 * vp - ctime (if nlink > 0) 1211 */ 1212 static int 1213 zfs_remove_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr) 1214 { 1215 znode_t *dzp = VTOZ(dvp); 1216 znode_t *zp; 1217 znode_t *xzp; 1218 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1219 zilog_t *zilog; 1220 uint64_t xattr_obj; 1221 uint64_t obj = 0; 1222 dmu_tx_t *tx; 1223 boolean_t unlinked; 1224 uint64_t txtype; 1225 int error; 1226 1227 1228 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 1229 return (error); 1230 zp = VTOZ(vp); 1231 if ((error = zfs_verify_zp(zp)) != 0) { 1232 zfs_exit(zfsvfs, FTAG); 1233 return (error); 1234 } 1235 zilog = zfsvfs->z_log; 1236 1237 xattr_obj = 0; 1238 xzp = NULL; 1239 1240 if ((error = zfs_zaccess_delete(dzp, zp, cr, NULL))) { 1241 goto out; 1242 } 1243 1244 /* 1245 * Need to use rmdir for removing directories. 1246 */ 1247 if (vp->v_type == VDIR) { 1248 error = SET_ERROR(EPERM); 1249 goto out; 1250 } 1251 1252 vnevent_remove(vp, dvp, name, ct); 1253 1254 obj = zp->z_id; 1255 1256 /* are there any extended attributes? */ 1257 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1258 &xattr_obj, sizeof (xattr_obj)); 1259 if (error == 0 && xattr_obj) { 1260 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1261 ASSERT0(error); 1262 } 1263 1264 /* 1265 * We may delete the znode now, or we may put it in the unlinked set; 1266 * it depends on whether we're the last link, and on whether there are 1267 * other holds on the vnode. So we dmu_tx_hold() the right things to 1268 * allow for either case. 1269 */ 1270 tx = dmu_tx_create(zfsvfs->z_os); 1271 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1272 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1273 zfs_sa_upgrade_txholds(tx, zp); 1274 zfs_sa_upgrade_txholds(tx, dzp); 1275 1276 if (xzp) { 1277 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1278 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1279 } 1280 1281 /* charge as an update -- would be nice not to charge at all */ 1282 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1283 1284 /* 1285 * Mark this transaction as typically resulting in a net free of space 1286 */ 1287 dmu_tx_mark_netfree(tx); 1288 1289 error = dmu_tx_assign(tx, TXG_WAIT); 1290 if (error) { 1291 dmu_tx_abort(tx); 1292 zfs_exit(zfsvfs, FTAG); 1293 return (error); 1294 } 1295 1296 /* 1297 * Remove the directory entry. 1298 */ 1299 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, &unlinked); 1300 1301 if (error) { 1302 dmu_tx_commit(tx); 1303 goto out; 1304 } 1305 1306 if (unlinked) { 1307 zfs_unlinked_add(zp, tx); 1308 vp->v_vflag |= VV_NOSYNC; 1309 } 1310 /* XXX check changes to linux vnops */ 1311 txtype = TX_REMOVE; 1312 zfs_log_remove(zilog, tx, txtype, dzp, name, obj, unlinked); 1313 1314 dmu_tx_commit(tx); 1315 out: 1316 1317 if (xzp) 1318 vrele(ZTOV(xzp)); 1319 1320 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1321 zil_commit(zilog, 0); 1322 1323 1324 zfs_exit(zfsvfs, FTAG); 1325 return (error); 1326 } 1327 1328 1329 static int 1330 zfs_lookup_internal(znode_t *dzp, const char *name, vnode_t **vpp, 1331 struct componentname *cnp, int nameiop) 1332 { 1333 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1334 int error; 1335 1336 cnp->cn_nameptr = __DECONST(char *, name); 1337 cnp->cn_namelen = strlen(name); 1338 cnp->cn_nameiop = nameiop; 1339 cnp->cn_flags = ISLASTCN; 1340 #if __FreeBSD_version < 1400068 1341 cnp->cn_flags |= SAVENAME; 1342 #endif 1343 cnp->cn_lkflags = LK_EXCLUSIVE | LK_RETRY; 1344 cnp->cn_cred = kcred; 1345 #if __FreeBSD_version < 1400037 1346 cnp->cn_thread = curthread; 1347 #endif 1348 1349 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay) { 1350 struct vop_lookup_args a; 1351 1352 a.a_gen.a_desc = &vop_lookup_desc; 1353 a.a_dvp = ZTOV(dzp); 1354 a.a_vpp = vpp; 1355 a.a_cnp = cnp; 1356 error = vfs_cache_lookup(&a); 1357 } else { 1358 error = zfs_lookup(ZTOV(dzp), name, vpp, cnp, nameiop, kcred, 0, 1359 B_FALSE); 1360 } 1361 #ifdef ZFS_DEBUG 1362 if (error) { 1363 printf("got error %d on name %s on op %d\n", error, name, 1364 nameiop); 1365 kdb_backtrace(); 1366 } 1367 #endif 1368 return (error); 1369 } 1370 1371 int 1372 zfs_remove(znode_t *dzp, const char *name, cred_t *cr, int flags) 1373 { 1374 vnode_t *vp; 1375 int error; 1376 struct componentname cn; 1377 1378 if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE))) 1379 return (error); 1380 1381 error = zfs_remove_(ZTOV(dzp), vp, name, cr); 1382 vput(vp); 1383 return (error); 1384 } 1385 /* 1386 * Create a new directory and insert it into dvp using the name 1387 * provided. Return a pointer to the inserted directory. 1388 * 1389 * IN: dvp - vnode of directory to add subdir to. 1390 * dirname - name of new directory. 1391 * vap - attributes of new directory. 1392 * cr - credentials of caller. 1393 * ct - caller context 1394 * flags - case flags 1395 * vsecp - ACL to be set 1396 * mnt_ns - Unused on FreeBSD 1397 * 1398 * OUT: vpp - vnode of created directory. 1399 * 1400 * RETURN: 0 on success, error code on failure. 1401 * 1402 * Timestamps: 1403 * dvp - ctime|mtime updated 1404 * vp - ctime|mtime|atime updated 1405 */ 1406 int 1407 zfs_mkdir(znode_t *dzp, const char *dirname, vattr_t *vap, znode_t **zpp, 1408 cred_t *cr, int flags, vsecattr_t *vsecp, zidmap_t *mnt_ns) 1409 { 1410 (void) flags, (void) vsecp; 1411 znode_t *zp; 1412 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1413 zilog_t *zilog; 1414 uint64_t txtype; 1415 dmu_tx_t *tx; 1416 int error; 1417 uid_t uid = crgetuid(cr); 1418 gid_t gid = crgetgid(cr); 1419 zfs_acl_ids_t acl_ids; 1420 boolean_t fuid_dirtied; 1421 1422 ASSERT3U(vap->va_type, ==, VDIR); 1423 1424 /* 1425 * If we have an ephemeral id, ACL, or XVATTR then 1426 * make sure file system is at proper version 1427 */ 1428 if (zfsvfs->z_use_fuids == B_FALSE && 1429 ((vap->va_mask & AT_XVATTR) || 1430 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1431 return (SET_ERROR(EINVAL)); 1432 1433 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 1434 return (error); 1435 zilog = zfsvfs->z_log; 1436 1437 if (dzp->z_pflags & ZFS_XATTR) { 1438 zfs_exit(zfsvfs, FTAG); 1439 return (SET_ERROR(EINVAL)); 1440 } 1441 1442 if (zfsvfs->z_utf8 && u8_validate(dirname, 1443 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1444 zfs_exit(zfsvfs, FTAG); 1445 return (SET_ERROR(EILSEQ)); 1446 } 1447 1448 if (vap->va_mask & AT_XVATTR) { 1449 if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap, 1450 crgetuid(cr), cr, vap->va_type)) != 0) { 1451 zfs_exit(zfsvfs, FTAG); 1452 return (error); 1453 } 1454 } 1455 1456 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 1457 NULL, &acl_ids, NULL)) != 0) { 1458 zfs_exit(zfsvfs, FTAG); 1459 return (error); 1460 } 1461 1462 /* 1463 * First make sure the new directory doesn't exist. 1464 * 1465 * Existence is checked first to make sure we don't return 1466 * EACCES instead of EEXIST which can cause some applications 1467 * to fail. 1468 */ 1469 *zpp = NULL; 1470 1471 if ((error = zfs_dirent_lookup(dzp, dirname, &zp, ZNEW))) { 1472 zfs_acl_ids_free(&acl_ids); 1473 zfs_exit(zfsvfs, FTAG); 1474 return (error); 1475 } 1476 ASSERT3P(zp, ==, NULL); 1477 1478 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr, 1479 mnt_ns))) { 1480 zfs_acl_ids_free(&acl_ids); 1481 zfs_exit(zfsvfs, FTAG); 1482 return (error); 1483 } 1484 1485 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zfs_inherit_projid(dzp))) { 1486 zfs_acl_ids_free(&acl_ids); 1487 zfs_exit(zfsvfs, FTAG); 1488 return (SET_ERROR(EDQUOT)); 1489 } 1490 1491 /* 1492 * Add a new entry to the directory. 1493 */ 1494 getnewvnode_reserve_(); 1495 tx = dmu_tx_create(zfsvfs->z_os); 1496 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1497 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1498 fuid_dirtied = zfsvfs->z_fuid_dirty; 1499 if (fuid_dirtied) 1500 zfs_fuid_txhold(zfsvfs, tx); 1501 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1502 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1503 acl_ids.z_aclp->z_acl_bytes); 1504 } 1505 1506 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1507 ZFS_SA_BASE_ATTR_SIZE); 1508 1509 error = dmu_tx_assign(tx, TXG_WAIT); 1510 if (error) { 1511 zfs_acl_ids_free(&acl_ids); 1512 dmu_tx_abort(tx); 1513 getnewvnode_drop_reserve(); 1514 zfs_exit(zfsvfs, FTAG); 1515 return (error); 1516 } 1517 1518 /* 1519 * Create new node. 1520 */ 1521 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1522 1523 if (fuid_dirtied) 1524 zfs_fuid_sync(zfsvfs, tx); 1525 1526 /* 1527 * Now put new name in parent dir. 1528 */ 1529 (void) zfs_link_create(dzp, dirname, zp, tx, ZNEW); 1530 1531 *zpp = zp; 1532 1533 txtype = zfs_log_create_txtype(Z_DIR, NULL, vap); 1534 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, NULL, 1535 acl_ids.z_fuidp, vap); 1536 1537 zfs_acl_ids_free(&acl_ids); 1538 1539 dmu_tx_commit(tx); 1540 1541 getnewvnode_drop_reserve(); 1542 1543 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1544 zil_commit(zilog, 0); 1545 1546 zfs_exit(zfsvfs, FTAG); 1547 return (0); 1548 } 1549 1550 #if __FreeBSD_version < 1300124 1551 static void 1552 cache_vop_rmdir(struct vnode *dvp, struct vnode *vp) 1553 { 1554 1555 cache_purge(dvp); 1556 cache_purge(vp); 1557 } 1558 #endif 1559 1560 /* 1561 * Remove a directory subdir entry. If the current working 1562 * directory is the same as the subdir to be removed, the 1563 * remove will fail. 1564 * 1565 * IN: dvp - vnode of directory to remove from. 1566 * name - name of directory to be removed. 1567 * cwd - vnode of current working directory. 1568 * cr - credentials of caller. 1569 * ct - caller context 1570 * flags - case flags 1571 * 1572 * RETURN: 0 on success, error code on failure. 1573 * 1574 * Timestamps: 1575 * dvp - ctime|mtime updated 1576 */ 1577 static int 1578 zfs_rmdir_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr) 1579 { 1580 znode_t *dzp = VTOZ(dvp); 1581 znode_t *zp = VTOZ(vp); 1582 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1583 zilog_t *zilog; 1584 dmu_tx_t *tx; 1585 int error; 1586 1587 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 1588 return (error); 1589 if ((error = zfs_verify_zp(zp)) != 0) { 1590 zfs_exit(zfsvfs, FTAG); 1591 return (error); 1592 } 1593 zilog = zfsvfs->z_log; 1594 1595 1596 if ((error = zfs_zaccess_delete(dzp, zp, cr, NULL))) { 1597 goto out; 1598 } 1599 1600 if (vp->v_type != VDIR) { 1601 error = SET_ERROR(ENOTDIR); 1602 goto out; 1603 } 1604 1605 vnevent_rmdir(vp, dvp, name, ct); 1606 1607 tx = dmu_tx_create(zfsvfs->z_os); 1608 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1609 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1610 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1611 zfs_sa_upgrade_txholds(tx, zp); 1612 zfs_sa_upgrade_txholds(tx, dzp); 1613 dmu_tx_mark_netfree(tx); 1614 error = dmu_tx_assign(tx, TXG_WAIT); 1615 if (error) { 1616 dmu_tx_abort(tx); 1617 zfs_exit(zfsvfs, FTAG); 1618 return (error); 1619 } 1620 1621 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, NULL); 1622 1623 if (error == 0) { 1624 uint64_t txtype = TX_RMDIR; 1625 zfs_log_remove(zilog, tx, txtype, dzp, name, 1626 ZFS_NO_OBJECT, B_FALSE); 1627 } 1628 1629 dmu_tx_commit(tx); 1630 1631 if (zfsvfs->z_use_namecache) 1632 cache_vop_rmdir(dvp, vp); 1633 out: 1634 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1635 zil_commit(zilog, 0); 1636 1637 zfs_exit(zfsvfs, FTAG); 1638 return (error); 1639 } 1640 1641 int 1642 zfs_rmdir(znode_t *dzp, const char *name, znode_t *cwd, cred_t *cr, int flags) 1643 { 1644 struct componentname cn; 1645 vnode_t *vp; 1646 int error; 1647 1648 if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE))) 1649 return (error); 1650 1651 error = zfs_rmdir_(ZTOV(dzp), vp, name, cr); 1652 vput(vp); 1653 return (error); 1654 } 1655 1656 /* 1657 * Read as many directory entries as will fit into the provided 1658 * buffer from the given directory cursor position (specified in 1659 * the uio structure). 1660 * 1661 * IN: vp - vnode of directory to read. 1662 * uio - structure supplying read location, range info, 1663 * and return buffer. 1664 * cr - credentials of caller. 1665 * ct - caller context 1666 * 1667 * OUT: uio - updated offset and range, buffer filled. 1668 * eofp - set to true if end-of-file detected. 1669 * ncookies- number of entries in cookies 1670 * cookies - offsets to directory entries 1671 * 1672 * RETURN: 0 on success, error code on failure. 1673 * 1674 * Timestamps: 1675 * vp - atime updated 1676 * 1677 * Note that the low 4 bits of the cookie returned by zap is always zero. 1678 * This allows us to use the low range for "special" directory entries: 1679 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1680 * we use the offset 2 for the '.zfs' directory. 1681 */ 1682 static int 1683 zfs_readdir(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, int *eofp, 1684 int *ncookies, cookie_t **cookies) 1685 { 1686 znode_t *zp = VTOZ(vp); 1687 iovec_t *iovp; 1688 dirent64_t *odp; 1689 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1690 objset_t *os; 1691 caddr_t outbuf; 1692 size_t bufsize; 1693 zap_cursor_t zc; 1694 zap_attribute_t zap; 1695 uint_t bytes_wanted; 1696 uint64_t offset; /* must be unsigned; checks for < 1 */ 1697 uint64_t parent; 1698 int local_eof; 1699 int outcount; 1700 int error; 1701 uint8_t prefetch; 1702 uint8_t type; 1703 int ncooks; 1704 cookie_t *cooks = NULL; 1705 1706 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 1707 return (error); 1708 1709 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 1710 &parent, sizeof (parent))) != 0) { 1711 zfs_exit(zfsvfs, FTAG); 1712 return (error); 1713 } 1714 1715 /* 1716 * If we are not given an eof variable, 1717 * use a local one. 1718 */ 1719 if (eofp == NULL) 1720 eofp = &local_eof; 1721 1722 /* 1723 * Check for valid iov_len. 1724 */ 1725 if (GET_UIO_STRUCT(uio)->uio_iov->iov_len <= 0) { 1726 zfs_exit(zfsvfs, FTAG); 1727 return (SET_ERROR(EINVAL)); 1728 } 1729 1730 /* 1731 * Quit if directory has been removed (posix) 1732 */ 1733 if ((*eofp = zp->z_unlinked) != 0) { 1734 zfs_exit(zfsvfs, FTAG); 1735 return (0); 1736 } 1737 1738 error = 0; 1739 os = zfsvfs->z_os; 1740 offset = zfs_uio_offset(uio); 1741 prefetch = zp->z_zn_prefetch; 1742 1743 /* 1744 * Initialize the iterator cursor. 1745 */ 1746 if (offset <= 3) { 1747 /* 1748 * Start iteration from the beginning of the directory. 1749 */ 1750 zap_cursor_init(&zc, os, zp->z_id); 1751 } else { 1752 /* 1753 * The offset is a serialized cursor. 1754 */ 1755 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1756 } 1757 1758 /* 1759 * Get space to change directory entries into fs independent format. 1760 */ 1761 iovp = GET_UIO_STRUCT(uio)->uio_iov; 1762 bytes_wanted = iovp->iov_len; 1763 if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1) { 1764 bufsize = bytes_wanted; 1765 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1766 odp = (struct dirent64 *)outbuf; 1767 } else { 1768 bufsize = bytes_wanted; 1769 outbuf = NULL; 1770 odp = (struct dirent64 *)iovp->iov_base; 1771 } 1772 1773 if (ncookies != NULL) { 1774 /* 1775 * Minimum entry size is dirent size and 1 byte for a file name. 1776 */ 1777 ncooks = zfs_uio_resid(uio) / (sizeof (struct dirent) - 1778 sizeof (((struct dirent *)NULL)->d_name) + 1); 1779 cooks = malloc(ncooks * sizeof (*cooks), M_TEMP, M_WAITOK); 1780 *cookies = cooks; 1781 *ncookies = ncooks; 1782 } 1783 1784 /* 1785 * Transform to file-system independent format 1786 */ 1787 outcount = 0; 1788 while (outcount < bytes_wanted) { 1789 ino64_t objnum; 1790 ushort_t reclen; 1791 off64_t *next = NULL; 1792 1793 /* 1794 * Special case `.', `..', and `.zfs'. 1795 */ 1796 if (offset == 0) { 1797 (void) strcpy(zap.za_name, "."); 1798 zap.za_normalization_conflict = 0; 1799 objnum = zp->z_id; 1800 type = DT_DIR; 1801 } else if (offset == 1) { 1802 (void) strcpy(zap.za_name, ".."); 1803 zap.za_normalization_conflict = 0; 1804 objnum = parent; 1805 type = DT_DIR; 1806 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1807 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1808 zap.za_normalization_conflict = 0; 1809 objnum = ZFSCTL_INO_ROOT; 1810 type = DT_DIR; 1811 } else { 1812 /* 1813 * Grab next entry. 1814 */ 1815 if ((error = zap_cursor_retrieve(&zc, &zap))) { 1816 if ((*eofp = (error == ENOENT)) != 0) 1817 break; 1818 else 1819 goto update; 1820 } 1821 1822 if (zap.za_integer_length != 8 || 1823 zap.za_num_integers != 1) { 1824 cmn_err(CE_WARN, "zap_readdir: bad directory " 1825 "entry, obj = %lld, offset = %lld\n", 1826 (u_longlong_t)zp->z_id, 1827 (u_longlong_t)offset); 1828 error = SET_ERROR(ENXIO); 1829 goto update; 1830 } 1831 1832 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 1833 /* 1834 * MacOS X can extract the object type here such as: 1835 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1836 */ 1837 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1838 } 1839 1840 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1841 1842 /* 1843 * Will this entry fit in the buffer? 1844 */ 1845 if (outcount + reclen > bufsize) { 1846 /* 1847 * Did we manage to fit anything in the buffer? 1848 */ 1849 if (!outcount) { 1850 error = SET_ERROR(EINVAL); 1851 goto update; 1852 } 1853 break; 1854 } 1855 /* 1856 * Add normal entry: 1857 */ 1858 odp->d_ino = objnum; 1859 odp->d_reclen = reclen; 1860 odp->d_namlen = strlen(zap.za_name); 1861 /* NOTE: d_off is the offset for the *next* entry. */ 1862 next = &odp->d_off; 1863 strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 1864 odp->d_type = type; 1865 dirent_terminate(odp); 1866 odp = (dirent64_t *)((intptr_t)odp + reclen); 1867 1868 outcount += reclen; 1869 1870 ASSERT3S(outcount, <=, bufsize); 1871 1872 /* Prefetch znode */ 1873 if (prefetch) 1874 dmu_prefetch(os, objnum, 0, 0, 0, 1875 ZIO_PRIORITY_SYNC_READ); 1876 1877 /* 1878 * Move to the next entry, fill in the previous offset. 1879 */ 1880 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1881 zap_cursor_advance(&zc); 1882 offset = zap_cursor_serialize(&zc); 1883 } else { 1884 offset += 1; 1885 } 1886 1887 /* Fill the offset right after advancing the cursor. */ 1888 if (next != NULL) 1889 *next = offset; 1890 if (cooks != NULL) { 1891 *cooks++ = offset; 1892 ncooks--; 1893 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 1894 } 1895 } 1896 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1897 1898 /* Subtract unused cookies */ 1899 if (ncookies != NULL) 1900 *ncookies -= ncooks; 1901 1902 if (zfs_uio_segflg(uio) == UIO_SYSSPACE && zfs_uio_iovcnt(uio) == 1) { 1903 iovp->iov_base += outcount; 1904 iovp->iov_len -= outcount; 1905 zfs_uio_resid(uio) -= outcount; 1906 } else if ((error = 1907 zfs_uiomove(outbuf, (long)outcount, UIO_READ, uio))) { 1908 /* 1909 * Reset the pointer. 1910 */ 1911 offset = zfs_uio_offset(uio); 1912 } 1913 1914 update: 1915 zap_cursor_fini(&zc); 1916 if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1) 1917 kmem_free(outbuf, bufsize); 1918 1919 if (error == ENOENT) 1920 error = 0; 1921 1922 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1923 1924 zfs_uio_setoffset(uio, offset); 1925 zfs_exit(zfsvfs, FTAG); 1926 if (error != 0 && cookies != NULL) { 1927 free(*cookies, M_TEMP); 1928 *cookies = NULL; 1929 *ncookies = 0; 1930 } 1931 return (error); 1932 } 1933 1934 /* 1935 * Get the requested file attributes and place them in the provided 1936 * vattr structure. 1937 * 1938 * IN: vp - vnode of file. 1939 * vap - va_mask identifies requested attributes. 1940 * If AT_XVATTR set, then optional attrs are requested 1941 * flags - ATTR_NOACLCHECK (CIFS server context) 1942 * cr - credentials of caller. 1943 * 1944 * OUT: vap - attribute values. 1945 * 1946 * RETURN: 0 (always succeeds). 1947 */ 1948 static int 1949 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1950 { 1951 znode_t *zp = VTOZ(vp); 1952 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1953 int error = 0; 1954 uint32_t blksize; 1955 u_longlong_t nblocks; 1956 uint64_t mtime[2], ctime[2], crtime[2], rdev; 1957 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 1958 xoptattr_t *xoap = NULL; 1959 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 1960 sa_bulk_attr_t bulk[4]; 1961 int count = 0; 1962 1963 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 1964 return (error); 1965 1966 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 1967 1968 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 1969 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 1970 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16); 1971 if (vp->v_type == VBLK || vp->v_type == VCHR) 1972 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, 1973 &rdev, 8); 1974 1975 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 1976 zfs_exit(zfsvfs, FTAG); 1977 return (error); 1978 } 1979 1980 /* 1981 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 1982 * Also, if we are the owner don't bother, since owner should 1983 * always be allowed to read basic attributes of file. 1984 */ 1985 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 1986 (vap->va_uid != crgetuid(cr))) { 1987 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 1988 skipaclchk, cr, NULL))) { 1989 zfs_exit(zfsvfs, FTAG); 1990 return (error); 1991 } 1992 } 1993 1994 /* 1995 * Return all attributes. It's cheaper to provide the answer 1996 * than to determine whether we were asked the question. 1997 */ 1998 1999 vap->va_type = IFTOVT(zp->z_mode); 2000 vap->va_mode = zp->z_mode & ~S_IFMT; 2001 vn_fsid(vp, vap); 2002 vap->va_nodeid = zp->z_id; 2003 vap->va_nlink = zp->z_links; 2004 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp) && 2005 zp->z_links < ZFS_LINK_MAX) 2006 vap->va_nlink++; 2007 vap->va_size = zp->z_size; 2008 if (vp->v_type == VBLK || vp->v_type == VCHR) 2009 vap->va_rdev = zfs_cmpldev(rdev); 2010 vap->va_gen = zp->z_gen; 2011 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2012 vap->va_filerev = zp->z_seq; 2013 2014 /* 2015 * Add in any requested optional attributes and the create time. 2016 * Also set the corresponding bits in the returned attribute bitmap. 2017 */ 2018 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2019 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2020 xoap->xoa_archive = 2021 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2022 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2023 } 2024 2025 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2026 xoap->xoa_readonly = 2027 ((zp->z_pflags & ZFS_READONLY) != 0); 2028 XVA_SET_RTN(xvap, XAT_READONLY); 2029 } 2030 2031 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2032 xoap->xoa_system = 2033 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2034 XVA_SET_RTN(xvap, XAT_SYSTEM); 2035 } 2036 2037 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2038 xoap->xoa_hidden = 2039 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2040 XVA_SET_RTN(xvap, XAT_HIDDEN); 2041 } 2042 2043 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2044 xoap->xoa_nounlink = 2045 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2046 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2047 } 2048 2049 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2050 xoap->xoa_immutable = 2051 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2052 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2053 } 2054 2055 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2056 xoap->xoa_appendonly = 2057 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2058 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2059 } 2060 2061 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2062 xoap->xoa_nodump = 2063 ((zp->z_pflags & ZFS_NODUMP) != 0); 2064 XVA_SET_RTN(xvap, XAT_NODUMP); 2065 } 2066 2067 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2068 xoap->xoa_opaque = 2069 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2070 XVA_SET_RTN(xvap, XAT_OPAQUE); 2071 } 2072 2073 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2074 xoap->xoa_av_quarantined = 2075 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2076 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2077 } 2078 2079 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2080 xoap->xoa_av_modified = 2081 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2082 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2083 } 2084 2085 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2086 vp->v_type == VREG) { 2087 zfs_sa_get_scanstamp(zp, xvap); 2088 } 2089 2090 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2091 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2092 XVA_SET_RTN(xvap, XAT_REPARSE); 2093 } 2094 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2095 xoap->xoa_generation = zp->z_gen; 2096 XVA_SET_RTN(xvap, XAT_GEN); 2097 } 2098 2099 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2100 xoap->xoa_offline = 2101 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2102 XVA_SET_RTN(xvap, XAT_OFFLINE); 2103 } 2104 2105 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2106 xoap->xoa_sparse = 2107 ((zp->z_pflags & ZFS_SPARSE) != 0); 2108 XVA_SET_RTN(xvap, XAT_SPARSE); 2109 } 2110 2111 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) { 2112 xoap->xoa_projinherit = 2113 ((zp->z_pflags & ZFS_PROJINHERIT) != 0); 2114 XVA_SET_RTN(xvap, XAT_PROJINHERIT); 2115 } 2116 2117 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) { 2118 xoap->xoa_projid = zp->z_projid; 2119 XVA_SET_RTN(xvap, XAT_PROJID); 2120 } 2121 } 2122 2123 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2124 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2125 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2126 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 2127 2128 2129 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 2130 vap->va_blksize = blksize; 2131 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2132 2133 if (zp->z_blksz == 0) { 2134 /* 2135 * Block size hasn't been set; suggest maximal I/O transfers. 2136 */ 2137 vap->va_blksize = zfsvfs->z_max_blksz; 2138 } 2139 2140 zfs_exit(zfsvfs, FTAG); 2141 return (0); 2142 } 2143 2144 /* 2145 * Set the file attributes to the values contained in the 2146 * vattr structure. 2147 * 2148 * IN: zp - znode of file to be modified. 2149 * vap - new attribute values. 2150 * If AT_XVATTR set, then optional attrs are being set 2151 * flags - ATTR_UTIME set if non-default time values provided. 2152 * - ATTR_NOACLCHECK (CIFS context only). 2153 * cr - credentials of caller. 2154 * mnt_ns - Unused on FreeBSD 2155 * 2156 * RETURN: 0 on success, error code on failure. 2157 * 2158 * Timestamps: 2159 * vp - ctime updated, mtime updated if size changed. 2160 */ 2161 int 2162 zfs_setattr(znode_t *zp, vattr_t *vap, int flags, cred_t *cr, zidmap_t *mnt_ns) 2163 { 2164 vnode_t *vp = ZTOV(zp); 2165 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2166 objset_t *os; 2167 zilog_t *zilog; 2168 dmu_tx_t *tx; 2169 vattr_t oldva; 2170 xvattr_t tmpxvattr; 2171 uint_t mask = vap->va_mask; 2172 uint_t saved_mask = 0; 2173 uint64_t saved_mode; 2174 int trim_mask = 0; 2175 uint64_t new_mode; 2176 uint64_t new_uid, new_gid; 2177 uint64_t xattr_obj; 2178 uint64_t mtime[2], ctime[2]; 2179 uint64_t projid = ZFS_INVALID_PROJID; 2180 znode_t *attrzp; 2181 int need_policy = FALSE; 2182 int err, err2; 2183 zfs_fuid_info_t *fuidp = NULL; 2184 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2185 xoptattr_t *xoap; 2186 zfs_acl_t *aclp; 2187 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2188 boolean_t fuid_dirtied = B_FALSE; 2189 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 2190 int count = 0, xattr_count = 0; 2191 2192 if (mask == 0) 2193 return (0); 2194 2195 if (mask & AT_NOSET) 2196 return (SET_ERROR(EINVAL)); 2197 2198 if ((err = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 2199 return (err); 2200 2201 os = zfsvfs->z_os; 2202 zilog = zfsvfs->z_log; 2203 2204 /* 2205 * Make sure that if we have ephemeral uid/gid or xvattr specified 2206 * that file system is at proper version level 2207 */ 2208 2209 if (zfsvfs->z_use_fuids == B_FALSE && 2210 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2211 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2212 (mask & AT_XVATTR))) { 2213 zfs_exit(zfsvfs, FTAG); 2214 return (SET_ERROR(EINVAL)); 2215 } 2216 2217 if (mask & AT_SIZE && vp->v_type == VDIR) { 2218 zfs_exit(zfsvfs, FTAG); 2219 return (SET_ERROR(EISDIR)); 2220 } 2221 2222 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 2223 zfs_exit(zfsvfs, FTAG); 2224 return (SET_ERROR(EINVAL)); 2225 } 2226 2227 /* 2228 * If this is an xvattr_t, then get a pointer to the structure of 2229 * optional attributes. If this is NULL, then we have a vattr_t. 2230 */ 2231 xoap = xva_getxoptattr(xvap); 2232 2233 xva_init(&tmpxvattr); 2234 2235 /* 2236 * Immutable files can only alter immutable bit and atime 2237 */ 2238 if ((zp->z_pflags & ZFS_IMMUTABLE) && 2239 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2240 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 2241 zfs_exit(zfsvfs, FTAG); 2242 return (SET_ERROR(EPERM)); 2243 } 2244 2245 /* 2246 * Note: ZFS_READONLY is handled in zfs_zaccess_common. 2247 */ 2248 2249 /* 2250 * Verify timestamps doesn't overflow 32 bits. 2251 * ZFS can handle large timestamps, but 32bit syscalls can't 2252 * handle times greater than 2039. This check should be removed 2253 * once large timestamps are fully supported. 2254 */ 2255 if (mask & (AT_ATIME | AT_MTIME)) { 2256 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 2257 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 2258 zfs_exit(zfsvfs, FTAG); 2259 return (SET_ERROR(EOVERFLOW)); 2260 } 2261 } 2262 if (xoap != NULL && (mask & AT_XVATTR)) { 2263 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME) && 2264 TIMESPEC_OVERFLOW(&vap->va_birthtime)) { 2265 zfs_exit(zfsvfs, FTAG); 2266 return (SET_ERROR(EOVERFLOW)); 2267 } 2268 2269 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) { 2270 if (!dmu_objset_projectquota_enabled(os) || 2271 (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode))) { 2272 zfs_exit(zfsvfs, FTAG); 2273 return (SET_ERROR(EOPNOTSUPP)); 2274 } 2275 2276 projid = xoap->xoa_projid; 2277 if (unlikely(projid == ZFS_INVALID_PROJID)) { 2278 zfs_exit(zfsvfs, FTAG); 2279 return (SET_ERROR(EINVAL)); 2280 } 2281 2282 if (projid == zp->z_projid && zp->z_pflags & ZFS_PROJID) 2283 projid = ZFS_INVALID_PROJID; 2284 else 2285 need_policy = TRUE; 2286 } 2287 2288 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT) && 2289 (xoap->xoa_projinherit != 2290 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) && 2291 (!dmu_objset_projectquota_enabled(os) || 2292 (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode)))) { 2293 zfs_exit(zfsvfs, FTAG); 2294 return (SET_ERROR(EOPNOTSUPP)); 2295 } 2296 } 2297 2298 attrzp = NULL; 2299 aclp = NULL; 2300 2301 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2302 zfs_exit(zfsvfs, FTAG); 2303 return (SET_ERROR(EROFS)); 2304 } 2305 2306 /* 2307 * First validate permissions 2308 */ 2309 2310 if (mask & AT_SIZE) { 2311 /* 2312 * XXX - Note, we are not providing any open 2313 * mode flags here (like FNDELAY), so we may 2314 * block if there are locks present... this 2315 * should be addressed in openat(). 2316 */ 2317 /* XXX - would it be OK to generate a log record here? */ 2318 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2319 if (err) { 2320 zfs_exit(zfsvfs, FTAG); 2321 return (err); 2322 } 2323 } 2324 2325 if (mask & (AT_ATIME|AT_MTIME) || 2326 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2327 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2328 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2329 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 2330 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 2331 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2332 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 2333 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2334 skipaclchk, cr, mnt_ns); 2335 } 2336 2337 if (mask & (AT_UID|AT_GID)) { 2338 int idmask = (mask & (AT_UID|AT_GID)); 2339 int take_owner; 2340 int take_group; 2341 2342 /* 2343 * NOTE: even if a new mode is being set, 2344 * we may clear S_ISUID/S_ISGID bits. 2345 */ 2346 2347 if (!(mask & AT_MODE)) 2348 vap->va_mode = zp->z_mode; 2349 2350 /* 2351 * Take ownership or chgrp to group we are a member of 2352 */ 2353 2354 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2355 take_group = (mask & AT_GID) && 2356 zfs_groupmember(zfsvfs, vap->va_gid, cr); 2357 2358 /* 2359 * If both AT_UID and AT_GID are set then take_owner and 2360 * take_group must both be set in order to allow taking 2361 * ownership. 2362 * 2363 * Otherwise, send the check through secpolicy_vnode_setattr() 2364 * 2365 */ 2366 2367 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2368 ((idmask == AT_UID) && take_owner) || 2369 ((idmask == AT_GID) && take_group)) { 2370 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 2371 skipaclchk, cr, mnt_ns) == 0) { 2372 /* 2373 * Remove setuid/setgid for non-privileged users 2374 */ 2375 secpolicy_setid_clear(vap, vp, cr); 2376 trim_mask = (mask & (AT_UID|AT_GID)); 2377 } else { 2378 need_policy = TRUE; 2379 } 2380 } else { 2381 need_policy = TRUE; 2382 } 2383 } 2384 2385 oldva.va_mode = zp->z_mode; 2386 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 2387 if (mask & AT_XVATTR) { 2388 /* 2389 * Update xvattr mask to include only those attributes 2390 * that are actually changing. 2391 * 2392 * the bits will be restored prior to actually setting 2393 * the attributes so the caller thinks they were set. 2394 */ 2395 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2396 if (xoap->xoa_appendonly != 2397 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 2398 need_policy = TRUE; 2399 } else { 2400 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 2401 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 2402 } 2403 } 2404 2405 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) { 2406 if (xoap->xoa_projinherit != 2407 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) { 2408 need_policy = TRUE; 2409 } else { 2410 XVA_CLR_REQ(xvap, XAT_PROJINHERIT); 2411 XVA_SET_REQ(&tmpxvattr, XAT_PROJINHERIT); 2412 } 2413 } 2414 2415 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2416 if (xoap->xoa_nounlink != 2417 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 2418 need_policy = TRUE; 2419 } else { 2420 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 2421 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 2422 } 2423 } 2424 2425 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2426 if (xoap->xoa_immutable != 2427 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 2428 need_policy = TRUE; 2429 } else { 2430 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 2431 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 2432 } 2433 } 2434 2435 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2436 if (xoap->xoa_nodump != 2437 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 2438 need_policy = TRUE; 2439 } else { 2440 XVA_CLR_REQ(xvap, XAT_NODUMP); 2441 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 2442 } 2443 } 2444 2445 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2446 if (xoap->xoa_av_modified != 2447 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 2448 need_policy = TRUE; 2449 } else { 2450 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 2451 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 2452 } 2453 } 2454 2455 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2456 if ((vp->v_type != VREG && 2457 xoap->xoa_av_quarantined) || 2458 xoap->xoa_av_quarantined != 2459 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 2460 need_policy = TRUE; 2461 } else { 2462 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 2463 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 2464 } 2465 } 2466 2467 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2468 zfs_exit(zfsvfs, FTAG); 2469 return (SET_ERROR(EPERM)); 2470 } 2471 2472 if (need_policy == FALSE && 2473 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 2474 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 2475 need_policy = TRUE; 2476 } 2477 } 2478 2479 if (mask & AT_MODE) { 2480 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr, 2481 mnt_ns) == 0) { 2482 err = secpolicy_setid_setsticky_clear(vp, vap, 2483 &oldva, cr); 2484 if (err) { 2485 zfs_exit(zfsvfs, FTAG); 2486 return (err); 2487 } 2488 trim_mask |= AT_MODE; 2489 } else { 2490 need_policy = TRUE; 2491 } 2492 } 2493 2494 if (need_policy) { 2495 /* 2496 * If trim_mask is set then take ownership 2497 * has been granted or write_acl is present and user 2498 * has the ability to modify mode. In that case remove 2499 * UID|GID and or MODE from mask so that 2500 * secpolicy_vnode_setattr() doesn't revoke it. 2501 */ 2502 2503 if (trim_mask) { 2504 saved_mask = vap->va_mask; 2505 vap->va_mask &= ~trim_mask; 2506 if (trim_mask & AT_MODE) { 2507 /* 2508 * Save the mode, as secpolicy_vnode_setattr() 2509 * will overwrite it with ova.va_mode. 2510 */ 2511 saved_mode = vap->va_mode; 2512 } 2513 } 2514 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2515 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 2516 if (err) { 2517 zfs_exit(zfsvfs, FTAG); 2518 return (err); 2519 } 2520 2521 if (trim_mask) { 2522 vap->va_mask |= saved_mask; 2523 if (trim_mask & AT_MODE) { 2524 /* 2525 * Recover the mode after 2526 * secpolicy_vnode_setattr(). 2527 */ 2528 vap->va_mode = saved_mode; 2529 } 2530 } 2531 } 2532 2533 /* 2534 * secpolicy_vnode_setattr, or take ownership may have 2535 * changed va_mask 2536 */ 2537 mask = vap->va_mask; 2538 2539 if ((mask & (AT_UID | AT_GID)) || projid != ZFS_INVALID_PROJID) { 2540 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 2541 &xattr_obj, sizeof (xattr_obj)); 2542 2543 if (err == 0 && xattr_obj) { 2544 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 2545 if (err == 0) { 2546 err = vn_lock(ZTOV(attrzp), LK_EXCLUSIVE); 2547 if (err != 0) 2548 vrele(ZTOV(attrzp)); 2549 } 2550 if (err) 2551 goto out2; 2552 } 2553 if (mask & AT_UID) { 2554 new_uid = zfs_fuid_create(zfsvfs, 2555 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 2556 if (new_uid != zp->z_uid && 2557 zfs_id_overquota(zfsvfs, DMU_USERUSED_OBJECT, 2558 new_uid)) { 2559 if (attrzp) 2560 vput(ZTOV(attrzp)); 2561 err = SET_ERROR(EDQUOT); 2562 goto out2; 2563 } 2564 } 2565 2566 if (mask & AT_GID) { 2567 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 2568 cr, ZFS_GROUP, &fuidp); 2569 if (new_gid != zp->z_gid && 2570 zfs_id_overquota(zfsvfs, DMU_GROUPUSED_OBJECT, 2571 new_gid)) { 2572 if (attrzp) 2573 vput(ZTOV(attrzp)); 2574 err = SET_ERROR(EDQUOT); 2575 goto out2; 2576 } 2577 } 2578 2579 if (projid != ZFS_INVALID_PROJID && 2580 zfs_id_overquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid)) { 2581 if (attrzp) 2582 vput(ZTOV(attrzp)); 2583 err = SET_ERROR(EDQUOT); 2584 goto out2; 2585 } 2586 } 2587 tx = dmu_tx_create(os); 2588 2589 if (mask & AT_MODE) { 2590 uint64_t pmode = zp->z_mode; 2591 uint64_t acl_obj; 2592 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2593 2594 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && 2595 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 2596 err = SET_ERROR(EPERM); 2597 goto out; 2598 } 2599 2600 if ((err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))) 2601 goto out; 2602 2603 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 2604 /* 2605 * Are we upgrading ACL from old V0 format 2606 * to V1 format? 2607 */ 2608 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 2609 zfs_znode_acl_version(zp) == 2610 ZFS_ACL_VERSION_INITIAL) { 2611 dmu_tx_hold_free(tx, acl_obj, 0, 2612 DMU_OBJECT_END); 2613 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2614 0, aclp->z_acl_bytes); 2615 } else { 2616 dmu_tx_hold_write(tx, acl_obj, 0, 2617 aclp->z_acl_bytes); 2618 } 2619 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2620 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2621 0, aclp->z_acl_bytes); 2622 } 2623 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2624 } else { 2625 if (((mask & AT_XVATTR) && 2626 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) || 2627 (projid != ZFS_INVALID_PROJID && 2628 !(zp->z_pflags & ZFS_PROJID))) 2629 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2630 else 2631 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2632 } 2633 2634 if (attrzp) { 2635 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 2636 } 2637 2638 fuid_dirtied = zfsvfs->z_fuid_dirty; 2639 if (fuid_dirtied) 2640 zfs_fuid_txhold(zfsvfs, tx); 2641 2642 zfs_sa_upgrade_txholds(tx, zp); 2643 2644 err = dmu_tx_assign(tx, TXG_WAIT); 2645 if (err) 2646 goto out; 2647 2648 count = 0; 2649 /* 2650 * Set each attribute requested. 2651 * We group settings according to the locks they need to acquire. 2652 * 2653 * Note: you cannot set ctime directly, although it will be 2654 * updated as a side-effect of calling this function. 2655 */ 2656 2657 if (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID)) { 2658 /* 2659 * For the existed object that is upgraded from old system, 2660 * its on-disk layout has no slot for the project ID attribute. 2661 * But quota accounting logic needs to access related slots by 2662 * offset directly. So we need to adjust old objects' layout 2663 * to make the project ID to some unified and fixed offset. 2664 */ 2665 if (attrzp) 2666 err = sa_add_projid(attrzp->z_sa_hdl, tx, projid); 2667 if (err == 0) 2668 err = sa_add_projid(zp->z_sa_hdl, tx, projid); 2669 2670 if (unlikely(err == EEXIST)) 2671 err = 0; 2672 else if (err != 0) 2673 goto out; 2674 else 2675 projid = ZFS_INVALID_PROJID; 2676 } 2677 2678 if (mask & (AT_UID|AT_GID|AT_MODE)) 2679 mutex_enter(&zp->z_acl_lock); 2680 2681 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 2682 &zp->z_pflags, sizeof (zp->z_pflags)); 2683 2684 if (attrzp) { 2685 if (mask & (AT_UID|AT_GID|AT_MODE)) 2686 mutex_enter(&attrzp->z_acl_lock); 2687 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2688 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 2689 sizeof (attrzp->z_pflags)); 2690 if (projid != ZFS_INVALID_PROJID) { 2691 attrzp->z_projid = projid; 2692 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2693 SA_ZPL_PROJID(zfsvfs), NULL, &attrzp->z_projid, 2694 sizeof (attrzp->z_projid)); 2695 } 2696 } 2697 2698 if (mask & (AT_UID|AT_GID)) { 2699 2700 if (mask & AT_UID) { 2701 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 2702 &new_uid, sizeof (new_uid)); 2703 zp->z_uid = new_uid; 2704 if (attrzp) { 2705 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2706 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 2707 sizeof (new_uid)); 2708 attrzp->z_uid = new_uid; 2709 } 2710 } 2711 2712 if (mask & AT_GID) { 2713 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 2714 NULL, &new_gid, sizeof (new_gid)); 2715 zp->z_gid = new_gid; 2716 if (attrzp) { 2717 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2718 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 2719 sizeof (new_gid)); 2720 attrzp->z_gid = new_gid; 2721 } 2722 } 2723 if (!(mask & AT_MODE)) { 2724 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 2725 NULL, &new_mode, sizeof (new_mode)); 2726 new_mode = zp->z_mode; 2727 } 2728 err = zfs_acl_chown_setattr(zp); 2729 ASSERT0(err); 2730 if (attrzp) { 2731 vn_seqc_write_begin(ZTOV(attrzp)); 2732 err = zfs_acl_chown_setattr(attrzp); 2733 vn_seqc_write_end(ZTOV(attrzp)); 2734 ASSERT0(err); 2735 } 2736 } 2737 2738 if (mask & AT_MODE) { 2739 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 2740 &new_mode, sizeof (new_mode)); 2741 zp->z_mode = new_mode; 2742 ASSERT3P(aclp, !=, NULL); 2743 err = zfs_aclset_common(zp, aclp, cr, tx); 2744 ASSERT0(err); 2745 if (zp->z_acl_cached) 2746 zfs_acl_free(zp->z_acl_cached); 2747 zp->z_acl_cached = aclp; 2748 aclp = NULL; 2749 } 2750 2751 2752 if (mask & AT_ATIME) { 2753 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 2754 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 2755 &zp->z_atime, sizeof (zp->z_atime)); 2756 } 2757 2758 if (mask & AT_MTIME) { 2759 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 2760 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 2761 mtime, sizeof (mtime)); 2762 } 2763 2764 if (projid != ZFS_INVALID_PROJID) { 2765 zp->z_projid = projid; 2766 SA_ADD_BULK_ATTR(bulk, count, 2767 SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid, 2768 sizeof (zp->z_projid)); 2769 } 2770 2771 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 2772 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 2773 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 2774 NULL, mtime, sizeof (mtime)); 2775 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 2776 &ctime, sizeof (ctime)); 2777 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime); 2778 } else if (mask != 0) { 2779 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 2780 &ctime, sizeof (ctime)); 2781 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime); 2782 if (attrzp) { 2783 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2784 SA_ZPL_CTIME(zfsvfs), NULL, 2785 &ctime, sizeof (ctime)); 2786 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 2787 mtime, ctime); 2788 } 2789 } 2790 2791 /* 2792 * Do this after setting timestamps to prevent timestamp 2793 * update from toggling bit 2794 */ 2795 2796 if (xoap && (mask & AT_XVATTR)) { 2797 2798 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) 2799 xoap->xoa_createtime = vap->va_birthtime; 2800 /* 2801 * restore trimmed off masks 2802 * so that return masks can be set for caller. 2803 */ 2804 2805 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 2806 XVA_SET_REQ(xvap, XAT_APPENDONLY); 2807 } 2808 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 2809 XVA_SET_REQ(xvap, XAT_NOUNLINK); 2810 } 2811 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 2812 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 2813 } 2814 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 2815 XVA_SET_REQ(xvap, XAT_NODUMP); 2816 } 2817 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 2818 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 2819 } 2820 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 2821 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 2822 } 2823 if (XVA_ISSET_REQ(&tmpxvattr, XAT_PROJINHERIT)) { 2824 XVA_SET_REQ(xvap, XAT_PROJINHERIT); 2825 } 2826 2827 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 2828 ASSERT3S(vp->v_type, ==, VREG); 2829 2830 zfs_xvattr_set(zp, xvap, tx); 2831 } 2832 2833 if (fuid_dirtied) 2834 zfs_fuid_sync(zfsvfs, tx); 2835 2836 if (mask != 0) 2837 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 2838 2839 if (mask & (AT_UID|AT_GID|AT_MODE)) 2840 mutex_exit(&zp->z_acl_lock); 2841 2842 if (attrzp) { 2843 if (mask & (AT_UID|AT_GID|AT_MODE)) 2844 mutex_exit(&attrzp->z_acl_lock); 2845 } 2846 out: 2847 if (err == 0 && attrzp) { 2848 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 2849 xattr_count, tx); 2850 ASSERT0(err2); 2851 } 2852 2853 if (attrzp) 2854 vput(ZTOV(attrzp)); 2855 2856 if (aclp) 2857 zfs_acl_free(aclp); 2858 2859 if (fuidp) { 2860 zfs_fuid_info_free(fuidp); 2861 fuidp = NULL; 2862 } 2863 2864 if (err) { 2865 dmu_tx_abort(tx); 2866 } else { 2867 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 2868 dmu_tx_commit(tx); 2869 } 2870 2871 out2: 2872 if (os->os_sync == ZFS_SYNC_ALWAYS) 2873 zil_commit(zilog, 0); 2874 2875 zfs_exit(zfsvfs, FTAG); 2876 return (err); 2877 } 2878 2879 /* 2880 * Look up the directory entries corresponding to the source and target 2881 * directory/name pairs. 2882 */ 2883 static int 2884 zfs_rename_relock_lookup(znode_t *sdzp, const struct componentname *scnp, 2885 znode_t **szpp, znode_t *tdzp, const struct componentname *tcnp, 2886 znode_t **tzpp) 2887 { 2888 zfsvfs_t *zfsvfs; 2889 znode_t *szp, *tzp; 2890 int error; 2891 2892 /* 2893 * Before using sdzp and tdzp we must ensure that they are live. 2894 * As a porting legacy from illumos we have two things to worry 2895 * about. One is typical for FreeBSD and it is that the vnode is 2896 * not reclaimed (doomed). The other is that the znode is live. 2897 * The current code can invalidate the znode without acquiring the 2898 * corresponding vnode lock if the object represented by the znode 2899 * and vnode is no longer valid after a rollback or receive operation. 2900 * z_teardown_lock hidden behind zfs_enter and zfs_exit is the lock 2901 * that protects the znodes from the invalidation. 2902 */ 2903 zfsvfs = sdzp->z_zfsvfs; 2904 ASSERT3P(zfsvfs, ==, tdzp->z_zfsvfs); 2905 if ((error = zfs_enter_verify_zp(zfsvfs, sdzp, FTAG)) != 0) 2906 return (error); 2907 if ((error = zfs_verify_zp(tdzp)) != 0) { 2908 zfs_exit(zfsvfs, FTAG); 2909 return (error); 2910 } 2911 2912 /* 2913 * Re-resolve svp to be certain it still exists and fetch the 2914 * correct vnode. 2915 */ 2916 error = zfs_dirent_lookup(sdzp, scnp->cn_nameptr, &szp, ZEXISTS); 2917 if (error != 0) { 2918 /* Source entry invalid or not there. */ 2919 if ((scnp->cn_flags & ISDOTDOT) != 0 || 2920 (scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.')) 2921 error = SET_ERROR(EINVAL); 2922 goto out; 2923 } 2924 *szpp = szp; 2925 2926 /* 2927 * Re-resolve tvp, if it disappeared we just carry on. 2928 */ 2929 error = zfs_dirent_lookup(tdzp, tcnp->cn_nameptr, &tzp, 0); 2930 if (error != 0) { 2931 vrele(ZTOV(szp)); 2932 if ((tcnp->cn_flags & ISDOTDOT) != 0) 2933 error = SET_ERROR(EINVAL); 2934 goto out; 2935 } 2936 *tzpp = tzp; 2937 out: 2938 zfs_exit(zfsvfs, FTAG); 2939 return (error); 2940 } 2941 2942 /* 2943 * We acquire all but fdvp locks using non-blocking acquisitions. If we 2944 * fail to acquire any lock in the path we will drop all held locks, 2945 * acquire the new lock in a blocking fashion, and then release it and 2946 * restart the rename. This acquire/release step ensures that we do not 2947 * spin on a lock waiting for release. On error release all vnode locks 2948 * and decrement references the way tmpfs_rename() would do. 2949 */ 2950 static int 2951 zfs_rename_relock(struct vnode *sdvp, struct vnode **svpp, 2952 struct vnode *tdvp, struct vnode **tvpp, 2953 const struct componentname *scnp, const struct componentname *tcnp) 2954 { 2955 struct vnode *nvp, *svp, *tvp; 2956 znode_t *sdzp, *tdzp, *szp, *tzp; 2957 int error; 2958 2959 VOP_UNLOCK1(tdvp); 2960 if (*tvpp != NULL && *tvpp != tdvp) 2961 VOP_UNLOCK1(*tvpp); 2962 2963 relock: 2964 error = vn_lock(sdvp, LK_EXCLUSIVE); 2965 if (error) 2966 goto out; 2967 error = vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT); 2968 if (error != 0) { 2969 VOP_UNLOCK1(sdvp); 2970 if (error != EBUSY) 2971 goto out; 2972 error = vn_lock(tdvp, LK_EXCLUSIVE); 2973 if (error) 2974 goto out; 2975 VOP_UNLOCK1(tdvp); 2976 goto relock; 2977 } 2978 tdzp = VTOZ(tdvp); 2979 sdzp = VTOZ(sdvp); 2980 2981 error = zfs_rename_relock_lookup(sdzp, scnp, &szp, tdzp, tcnp, &tzp); 2982 if (error != 0) { 2983 VOP_UNLOCK1(sdvp); 2984 VOP_UNLOCK1(tdvp); 2985 goto out; 2986 } 2987 svp = ZTOV(szp); 2988 tvp = tzp != NULL ? ZTOV(tzp) : NULL; 2989 2990 /* 2991 * Now try acquire locks on svp and tvp. 2992 */ 2993 nvp = svp; 2994 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT); 2995 if (error != 0) { 2996 VOP_UNLOCK1(sdvp); 2997 VOP_UNLOCK1(tdvp); 2998 if (tvp != NULL) 2999 vrele(tvp); 3000 if (error != EBUSY) { 3001 vrele(nvp); 3002 goto out; 3003 } 3004 error = vn_lock(nvp, LK_EXCLUSIVE); 3005 if (error != 0) { 3006 vrele(nvp); 3007 goto out; 3008 } 3009 VOP_UNLOCK1(nvp); 3010 /* 3011 * Concurrent rename race. 3012 * XXX ? 3013 */ 3014 if (nvp == tdvp) { 3015 vrele(nvp); 3016 error = SET_ERROR(EINVAL); 3017 goto out; 3018 } 3019 vrele(*svpp); 3020 *svpp = nvp; 3021 goto relock; 3022 } 3023 vrele(*svpp); 3024 *svpp = nvp; 3025 3026 if (*tvpp != NULL) 3027 vrele(*tvpp); 3028 *tvpp = NULL; 3029 if (tvp != NULL) { 3030 nvp = tvp; 3031 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT); 3032 if (error != 0) { 3033 VOP_UNLOCK1(sdvp); 3034 VOP_UNLOCK1(tdvp); 3035 VOP_UNLOCK1(*svpp); 3036 if (error != EBUSY) { 3037 vrele(nvp); 3038 goto out; 3039 } 3040 error = vn_lock(nvp, LK_EXCLUSIVE); 3041 if (error != 0) { 3042 vrele(nvp); 3043 goto out; 3044 } 3045 vput(nvp); 3046 goto relock; 3047 } 3048 *tvpp = nvp; 3049 } 3050 3051 return (0); 3052 3053 out: 3054 return (error); 3055 } 3056 3057 /* 3058 * Note that we must use VRELE_ASYNC in this function as it walks 3059 * up the directory tree and vrele may need to acquire an exclusive 3060 * lock if a last reference to a vnode is dropped. 3061 */ 3062 static int 3063 zfs_rename_check(znode_t *szp, znode_t *sdzp, znode_t *tdzp) 3064 { 3065 zfsvfs_t *zfsvfs; 3066 znode_t *zp, *zp1; 3067 uint64_t parent; 3068 int error; 3069 3070 zfsvfs = tdzp->z_zfsvfs; 3071 if (tdzp == szp) 3072 return (SET_ERROR(EINVAL)); 3073 if (tdzp == sdzp) 3074 return (0); 3075 if (tdzp->z_id == zfsvfs->z_root) 3076 return (0); 3077 zp = tdzp; 3078 for (;;) { 3079 ASSERT(!zp->z_unlinked); 3080 if ((error = sa_lookup(zp->z_sa_hdl, 3081 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) 3082 break; 3083 3084 if (parent == szp->z_id) { 3085 error = SET_ERROR(EINVAL); 3086 break; 3087 } 3088 if (parent == zfsvfs->z_root) 3089 break; 3090 if (parent == sdzp->z_id) 3091 break; 3092 3093 error = zfs_zget(zfsvfs, parent, &zp1); 3094 if (error != 0) 3095 break; 3096 3097 if (zp != tdzp) 3098 VN_RELE_ASYNC(ZTOV(zp), 3099 dsl_pool_zrele_taskq( 3100 dmu_objset_pool(zfsvfs->z_os))); 3101 zp = zp1; 3102 } 3103 3104 if (error == ENOTDIR) 3105 panic("checkpath: .. not a directory\n"); 3106 if (zp != tdzp) 3107 VN_RELE_ASYNC(ZTOV(zp), 3108 dsl_pool_zrele_taskq(dmu_objset_pool(zfsvfs->z_os))); 3109 return (error); 3110 } 3111 3112 #if __FreeBSD_version < 1300124 3113 static void 3114 cache_vop_rename(struct vnode *fdvp, struct vnode *fvp, struct vnode *tdvp, 3115 struct vnode *tvp, struct componentname *fcnp, struct componentname *tcnp) 3116 { 3117 3118 cache_purge(fvp); 3119 if (tvp != NULL) 3120 cache_purge(tvp); 3121 cache_purge_negative(tdvp); 3122 } 3123 #endif 3124 3125 static int 3126 zfs_do_rename_impl(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, 3127 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, 3128 cred_t *cr); 3129 3130 /* 3131 * Move an entry from the provided source directory to the target 3132 * directory. Change the entry name as indicated. 3133 * 3134 * IN: sdvp - Source directory containing the "old entry". 3135 * scnp - Old entry name. 3136 * tdvp - Target directory to contain the "new entry". 3137 * tcnp - New entry name. 3138 * cr - credentials of caller. 3139 * INOUT: svpp - Source file 3140 * tvpp - Target file, may point to NULL initially 3141 * 3142 * RETURN: 0 on success, error code on failure. 3143 * 3144 * Timestamps: 3145 * sdvp,tdvp - ctime|mtime updated 3146 */ 3147 static int 3148 zfs_do_rename(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, 3149 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, 3150 cred_t *cr) 3151 { 3152 int error; 3153 3154 ASSERT_VOP_ELOCKED(tdvp, __func__); 3155 if (*tvpp != NULL) 3156 ASSERT_VOP_ELOCKED(*tvpp, __func__); 3157 3158 /* Reject renames across filesystems. */ 3159 if ((*svpp)->v_mount != tdvp->v_mount || 3160 ((*tvpp) != NULL && (*svpp)->v_mount != (*tvpp)->v_mount)) { 3161 error = SET_ERROR(EXDEV); 3162 goto out; 3163 } 3164 3165 if (zfsctl_is_node(tdvp)) { 3166 error = SET_ERROR(EXDEV); 3167 goto out; 3168 } 3169 3170 /* 3171 * Lock all four vnodes to ensure safety and semantics of renaming. 3172 */ 3173 error = zfs_rename_relock(sdvp, svpp, tdvp, tvpp, scnp, tcnp); 3174 if (error != 0) { 3175 /* no vnodes are locked in the case of error here */ 3176 return (error); 3177 } 3178 3179 error = zfs_do_rename_impl(sdvp, svpp, scnp, tdvp, tvpp, tcnp, cr); 3180 VOP_UNLOCK1(sdvp); 3181 VOP_UNLOCK1(*svpp); 3182 out: 3183 if (*tvpp != NULL) 3184 VOP_UNLOCK1(*tvpp); 3185 if (tdvp != *tvpp) 3186 VOP_UNLOCK1(tdvp); 3187 3188 return (error); 3189 } 3190 3191 static int 3192 zfs_do_rename_impl(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, 3193 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, 3194 cred_t *cr) 3195 { 3196 dmu_tx_t *tx; 3197 zfsvfs_t *zfsvfs; 3198 zilog_t *zilog; 3199 znode_t *tdzp, *sdzp, *tzp, *szp; 3200 const char *snm = scnp->cn_nameptr; 3201 const char *tnm = tcnp->cn_nameptr; 3202 int error; 3203 3204 tdzp = VTOZ(tdvp); 3205 sdzp = VTOZ(sdvp); 3206 zfsvfs = tdzp->z_zfsvfs; 3207 3208 if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0) 3209 return (error); 3210 if ((error = zfs_verify_zp(sdzp)) != 0) { 3211 zfs_exit(zfsvfs, FTAG); 3212 return (error); 3213 } 3214 zilog = zfsvfs->z_log; 3215 3216 if (zfsvfs->z_utf8 && u8_validate(tnm, 3217 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3218 error = SET_ERROR(EILSEQ); 3219 goto out; 3220 } 3221 3222 /* If source and target are the same file, there is nothing to do. */ 3223 if ((*svpp) == (*tvpp)) { 3224 error = 0; 3225 goto out; 3226 } 3227 3228 if (((*svpp)->v_type == VDIR && (*svpp)->v_mountedhere != NULL) || 3229 ((*tvpp) != NULL && (*tvpp)->v_type == VDIR && 3230 (*tvpp)->v_mountedhere != NULL)) { 3231 error = SET_ERROR(EXDEV); 3232 goto out; 3233 } 3234 3235 szp = VTOZ(*svpp); 3236 if ((error = zfs_verify_zp(szp)) != 0) { 3237 zfs_exit(zfsvfs, FTAG); 3238 return (error); 3239 } 3240 tzp = *tvpp == NULL ? NULL : VTOZ(*tvpp); 3241 if (tzp != NULL) { 3242 if ((error = zfs_verify_zp(tzp)) != 0) { 3243 zfs_exit(zfsvfs, FTAG); 3244 return (error); 3245 } 3246 } 3247 3248 /* 3249 * This is to prevent the creation of links into attribute space 3250 * by renaming a linked file into/outof an attribute directory. 3251 * See the comment in zfs_link() for why this is considered bad. 3252 */ 3253 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3254 error = SET_ERROR(EINVAL); 3255 goto out; 3256 } 3257 3258 /* 3259 * If we are using project inheritance, means if the directory has 3260 * ZFS_PROJINHERIT set, then its descendant directories will inherit 3261 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under 3262 * such case, we only allow renames into our tree when the project 3263 * IDs are the same. 3264 */ 3265 if (tdzp->z_pflags & ZFS_PROJINHERIT && 3266 tdzp->z_projid != szp->z_projid) { 3267 error = SET_ERROR(EXDEV); 3268 goto out; 3269 } 3270 3271 /* 3272 * Must have write access at the source to remove the old entry 3273 * and write access at the target to create the new entry. 3274 * Note that if target and source are the same, this can be 3275 * done in a single check. 3276 */ 3277 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr, NULL))) 3278 goto out; 3279 3280 if ((*svpp)->v_type == VDIR) { 3281 /* 3282 * Avoid ".", "..", and aliases of "." for obvious reasons. 3283 */ 3284 if ((scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.') || 3285 sdzp == szp || 3286 (scnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) { 3287 error = EINVAL; 3288 goto out; 3289 } 3290 3291 /* 3292 * Check to make sure rename is valid. 3293 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3294 */ 3295 if ((error = zfs_rename_check(szp, sdzp, tdzp))) 3296 goto out; 3297 } 3298 3299 /* 3300 * Does target exist? 3301 */ 3302 if (tzp) { 3303 /* 3304 * Source and target must be the same type. 3305 */ 3306 if ((*svpp)->v_type == VDIR) { 3307 if ((*tvpp)->v_type != VDIR) { 3308 error = SET_ERROR(ENOTDIR); 3309 goto out; 3310 } else { 3311 cache_purge(tdvp); 3312 if (sdvp != tdvp) 3313 cache_purge(sdvp); 3314 } 3315 } else { 3316 if ((*tvpp)->v_type == VDIR) { 3317 error = SET_ERROR(EISDIR); 3318 goto out; 3319 } 3320 } 3321 } 3322 3323 vn_seqc_write_begin(*svpp); 3324 vn_seqc_write_begin(sdvp); 3325 if (*tvpp != NULL) 3326 vn_seqc_write_begin(*tvpp); 3327 if (tdvp != *tvpp) 3328 vn_seqc_write_begin(tdvp); 3329 3330 vnevent_rename_src(*svpp, sdvp, scnp->cn_nameptr, ct); 3331 if (tzp) 3332 vnevent_rename_dest(*tvpp, tdvp, tnm, ct); 3333 3334 /* 3335 * notify the target directory if it is not the same 3336 * as source directory. 3337 */ 3338 if (tdvp != sdvp) { 3339 vnevent_rename_dest_dir(tdvp, ct); 3340 } 3341 3342 tx = dmu_tx_create(zfsvfs->z_os); 3343 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3344 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 3345 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3346 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3347 if (sdzp != tdzp) { 3348 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 3349 zfs_sa_upgrade_txholds(tx, tdzp); 3350 } 3351 if (tzp) { 3352 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 3353 zfs_sa_upgrade_txholds(tx, tzp); 3354 } 3355 3356 zfs_sa_upgrade_txholds(tx, szp); 3357 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3358 error = dmu_tx_assign(tx, TXG_WAIT); 3359 if (error) { 3360 dmu_tx_abort(tx); 3361 goto out_seq; 3362 } 3363 3364 if (tzp) /* Attempt to remove the existing target */ 3365 error = zfs_link_destroy(tdzp, tnm, tzp, tx, 0, NULL); 3366 3367 if (error == 0) { 3368 error = zfs_link_create(tdzp, tnm, szp, tx, ZRENAMING); 3369 if (error == 0) { 3370 szp->z_pflags |= ZFS_AV_MODIFIED; 3371 3372 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 3373 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 3374 ASSERT0(error); 3375 3376 error = zfs_link_destroy(sdzp, snm, szp, tx, ZRENAMING, 3377 NULL); 3378 if (error == 0) { 3379 zfs_log_rename(zilog, tx, TX_RENAME, sdzp, 3380 snm, tdzp, tnm, szp); 3381 } else { 3382 /* 3383 * At this point, we have successfully created 3384 * the target name, but have failed to remove 3385 * the source name. Since the create was done 3386 * with the ZRENAMING flag, there are 3387 * complications; for one, the link count is 3388 * wrong. The easiest way to deal with this 3389 * is to remove the newly created target, and 3390 * return the original error. This must 3391 * succeed; fortunately, it is very unlikely to 3392 * fail, since we just created it. 3393 */ 3394 VERIFY0(zfs_link_destroy(tdzp, tnm, szp, tx, 3395 ZRENAMING, NULL)); 3396 } 3397 } 3398 if (error == 0) { 3399 cache_vop_rename(sdvp, *svpp, tdvp, *tvpp, scnp, tcnp); 3400 } 3401 } 3402 3403 dmu_tx_commit(tx); 3404 3405 out_seq: 3406 vn_seqc_write_end(*svpp); 3407 vn_seqc_write_end(sdvp); 3408 if (*tvpp != NULL) 3409 vn_seqc_write_end(*tvpp); 3410 if (tdvp != *tvpp) 3411 vn_seqc_write_end(tdvp); 3412 3413 out: 3414 if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3415 zil_commit(zilog, 0); 3416 zfs_exit(zfsvfs, FTAG); 3417 3418 return (error); 3419 } 3420 3421 int 3422 zfs_rename(znode_t *sdzp, const char *sname, znode_t *tdzp, const char *tname, 3423 cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap, zidmap_t *mnt_ns) 3424 { 3425 struct componentname scn, tcn; 3426 vnode_t *sdvp, *tdvp; 3427 vnode_t *svp, *tvp; 3428 int error; 3429 svp = tvp = NULL; 3430 3431 if (rflags != 0 || wo_vap != NULL) 3432 return (SET_ERROR(EINVAL)); 3433 3434 sdvp = ZTOV(sdzp); 3435 tdvp = ZTOV(tdzp); 3436 error = zfs_lookup_internal(sdzp, sname, &svp, &scn, DELETE); 3437 if (sdzp->z_zfsvfs->z_replay == B_FALSE) 3438 VOP_UNLOCK1(sdvp); 3439 if (error != 0) 3440 goto fail; 3441 VOP_UNLOCK1(svp); 3442 3443 vn_lock(tdvp, LK_EXCLUSIVE | LK_RETRY); 3444 error = zfs_lookup_internal(tdzp, tname, &tvp, &tcn, RENAME); 3445 if (error == EJUSTRETURN) 3446 tvp = NULL; 3447 else if (error != 0) { 3448 VOP_UNLOCK1(tdvp); 3449 goto fail; 3450 } 3451 3452 error = zfs_do_rename(sdvp, &svp, &scn, tdvp, &tvp, &tcn, cr); 3453 fail: 3454 if (svp != NULL) 3455 vrele(svp); 3456 if (tvp != NULL) 3457 vrele(tvp); 3458 3459 return (error); 3460 } 3461 3462 /* 3463 * Insert the indicated symbolic reference entry into the directory. 3464 * 3465 * IN: dvp - Directory to contain new symbolic link. 3466 * link - Name for new symlink entry. 3467 * vap - Attributes of new entry. 3468 * cr - credentials of caller. 3469 * ct - caller context 3470 * flags - case flags 3471 * mnt_ns - Unused on FreeBSD 3472 * 3473 * RETURN: 0 on success, error code on failure. 3474 * 3475 * Timestamps: 3476 * dvp - ctime|mtime updated 3477 */ 3478 int 3479 zfs_symlink(znode_t *dzp, const char *name, vattr_t *vap, 3480 const char *link, znode_t **zpp, cred_t *cr, int flags, zidmap_t *mnt_ns) 3481 { 3482 (void) flags; 3483 znode_t *zp; 3484 dmu_tx_t *tx; 3485 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3486 zilog_t *zilog; 3487 uint64_t len = strlen(link); 3488 int error; 3489 zfs_acl_ids_t acl_ids; 3490 boolean_t fuid_dirtied; 3491 uint64_t txtype = TX_SYMLINK; 3492 3493 ASSERT3S(vap->va_type, ==, VLNK); 3494 3495 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 3496 return (error); 3497 zilog = zfsvfs->z_log; 3498 3499 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 3500 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3501 zfs_exit(zfsvfs, FTAG); 3502 return (SET_ERROR(EILSEQ)); 3503 } 3504 3505 if (len > MAXPATHLEN) { 3506 zfs_exit(zfsvfs, FTAG); 3507 return (SET_ERROR(ENAMETOOLONG)); 3508 } 3509 3510 if ((error = zfs_acl_ids_create(dzp, 0, 3511 vap, cr, NULL, &acl_ids, NULL)) != 0) { 3512 zfs_exit(zfsvfs, FTAG); 3513 return (error); 3514 } 3515 3516 /* 3517 * Attempt to lock directory; fail if entry already exists. 3518 */ 3519 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW); 3520 if (error) { 3521 zfs_acl_ids_free(&acl_ids); 3522 zfs_exit(zfsvfs, FTAG); 3523 return (error); 3524 } 3525 3526 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) { 3527 zfs_acl_ids_free(&acl_ids); 3528 zfs_exit(zfsvfs, FTAG); 3529 return (error); 3530 } 3531 3532 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, 3533 0 /* projid */)) { 3534 zfs_acl_ids_free(&acl_ids); 3535 zfs_exit(zfsvfs, FTAG); 3536 return (SET_ERROR(EDQUOT)); 3537 } 3538 3539 getnewvnode_reserve_(); 3540 tx = dmu_tx_create(zfsvfs->z_os); 3541 fuid_dirtied = zfsvfs->z_fuid_dirty; 3542 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 3543 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3544 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 3545 ZFS_SA_BASE_ATTR_SIZE + len); 3546 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 3547 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3548 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 3549 acl_ids.z_aclp->z_acl_bytes); 3550 } 3551 if (fuid_dirtied) 3552 zfs_fuid_txhold(zfsvfs, tx); 3553 error = dmu_tx_assign(tx, TXG_WAIT); 3554 if (error) { 3555 zfs_acl_ids_free(&acl_ids); 3556 dmu_tx_abort(tx); 3557 getnewvnode_drop_reserve(); 3558 zfs_exit(zfsvfs, FTAG); 3559 return (error); 3560 } 3561 3562 /* 3563 * Create a new object for the symlink. 3564 * for version 4 ZPL datasets the symlink will be an SA attribute 3565 */ 3566 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 3567 3568 if (fuid_dirtied) 3569 zfs_fuid_sync(zfsvfs, tx); 3570 3571 if (zp->z_is_sa) 3572 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 3573 __DECONST(void *, link), len, tx); 3574 else 3575 zfs_sa_symlink(zp, __DECONST(char *, link), len, tx); 3576 3577 zp->z_size = len; 3578 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 3579 &zp->z_size, sizeof (zp->z_size), tx); 3580 /* 3581 * Insert the new object into the directory. 3582 */ 3583 (void) zfs_link_create(dzp, name, zp, tx, ZNEW); 3584 3585 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 3586 *zpp = zp; 3587 3588 zfs_acl_ids_free(&acl_ids); 3589 3590 dmu_tx_commit(tx); 3591 3592 getnewvnode_drop_reserve(); 3593 3594 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3595 zil_commit(zilog, 0); 3596 3597 zfs_exit(zfsvfs, FTAG); 3598 return (error); 3599 } 3600 3601 /* 3602 * Return, in the buffer contained in the provided uio structure, 3603 * the symbolic path referred to by vp. 3604 * 3605 * IN: vp - vnode of symbolic link. 3606 * uio - structure to contain the link path. 3607 * cr - credentials of caller. 3608 * ct - caller context 3609 * 3610 * OUT: uio - structure containing the link path. 3611 * 3612 * RETURN: 0 on success, error code on failure. 3613 * 3614 * Timestamps: 3615 * vp - atime updated 3616 */ 3617 static int 3618 zfs_readlink(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, caller_context_t *ct) 3619 { 3620 (void) cr, (void) ct; 3621 znode_t *zp = VTOZ(vp); 3622 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3623 int error; 3624 3625 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3626 return (error); 3627 3628 if (zp->z_is_sa) 3629 error = sa_lookup_uio(zp->z_sa_hdl, 3630 SA_ZPL_SYMLINK(zfsvfs), uio); 3631 else 3632 error = zfs_sa_readlink(zp, uio); 3633 3634 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3635 3636 zfs_exit(zfsvfs, FTAG); 3637 return (error); 3638 } 3639 3640 /* 3641 * Insert a new entry into directory tdvp referencing svp. 3642 * 3643 * IN: tdvp - Directory to contain new entry. 3644 * svp - vnode of new entry. 3645 * name - name of new entry. 3646 * cr - credentials of caller. 3647 * 3648 * RETURN: 0 on success, error code on failure. 3649 * 3650 * Timestamps: 3651 * tdvp - ctime|mtime updated 3652 * svp - ctime updated 3653 */ 3654 int 3655 zfs_link(znode_t *tdzp, znode_t *szp, const char *name, cred_t *cr, 3656 int flags) 3657 { 3658 (void) flags; 3659 znode_t *tzp; 3660 zfsvfs_t *zfsvfs = tdzp->z_zfsvfs; 3661 zilog_t *zilog; 3662 dmu_tx_t *tx; 3663 int error; 3664 uint64_t parent; 3665 uid_t owner; 3666 3667 ASSERT3S(ZTOV(tdzp)->v_type, ==, VDIR); 3668 3669 if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0) 3670 return (error); 3671 zilog = zfsvfs->z_log; 3672 3673 /* 3674 * POSIX dictates that we return EPERM here. 3675 * Better choices include ENOTSUP or EISDIR. 3676 */ 3677 if (ZTOV(szp)->v_type == VDIR) { 3678 zfs_exit(zfsvfs, FTAG); 3679 return (SET_ERROR(EPERM)); 3680 } 3681 3682 if ((error = zfs_verify_zp(szp)) != 0) { 3683 zfs_exit(zfsvfs, FTAG); 3684 return (error); 3685 } 3686 3687 /* 3688 * If we are using project inheritance, means if the directory has 3689 * ZFS_PROJINHERIT set, then its descendant directories will inherit 3690 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under 3691 * such case, we only allow hard link creation in our tree when the 3692 * project IDs are the same. 3693 */ 3694 if (tdzp->z_pflags & ZFS_PROJINHERIT && 3695 tdzp->z_projid != szp->z_projid) { 3696 zfs_exit(zfsvfs, FTAG); 3697 return (SET_ERROR(EXDEV)); 3698 } 3699 3700 if (szp->z_pflags & (ZFS_APPENDONLY | 3701 ZFS_IMMUTABLE | ZFS_READONLY)) { 3702 zfs_exit(zfsvfs, FTAG); 3703 return (SET_ERROR(EPERM)); 3704 } 3705 3706 /* Prevent links to .zfs/shares files */ 3707 3708 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 3709 &parent, sizeof (uint64_t))) != 0) { 3710 zfs_exit(zfsvfs, FTAG); 3711 return (error); 3712 } 3713 if (parent == zfsvfs->z_shares_dir) { 3714 zfs_exit(zfsvfs, FTAG); 3715 return (SET_ERROR(EPERM)); 3716 } 3717 3718 if (zfsvfs->z_utf8 && u8_validate(name, 3719 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3720 zfs_exit(zfsvfs, FTAG); 3721 return (SET_ERROR(EILSEQ)); 3722 } 3723 3724 /* 3725 * We do not support links between attributes and non-attributes 3726 * because of the potential security risk of creating links 3727 * into "normal" file space in order to circumvent restrictions 3728 * imposed in attribute space. 3729 */ 3730 if ((szp->z_pflags & ZFS_XATTR) != (tdzp->z_pflags & ZFS_XATTR)) { 3731 zfs_exit(zfsvfs, FTAG); 3732 return (SET_ERROR(EINVAL)); 3733 } 3734 3735 3736 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 3737 if (owner != crgetuid(cr) && secpolicy_basic_link(ZTOV(szp), cr) != 0) { 3738 zfs_exit(zfsvfs, FTAG); 3739 return (SET_ERROR(EPERM)); 3740 } 3741 3742 if ((error = zfs_zaccess(tdzp, ACE_ADD_FILE, 0, B_FALSE, cr, NULL))) { 3743 zfs_exit(zfsvfs, FTAG); 3744 return (error); 3745 } 3746 3747 /* 3748 * Attempt to lock directory; fail if entry already exists. 3749 */ 3750 error = zfs_dirent_lookup(tdzp, name, &tzp, ZNEW); 3751 if (error) { 3752 zfs_exit(zfsvfs, FTAG); 3753 return (error); 3754 } 3755 3756 tx = dmu_tx_create(zfsvfs->z_os); 3757 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3758 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, name); 3759 zfs_sa_upgrade_txholds(tx, szp); 3760 zfs_sa_upgrade_txholds(tx, tdzp); 3761 error = dmu_tx_assign(tx, TXG_WAIT); 3762 if (error) { 3763 dmu_tx_abort(tx); 3764 zfs_exit(zfsvfs, FTAG); 3765 return (error); 3766 } 3767 3768 error = zfs_link_create(tdzp, name, szp, tx, 0); 3769 3770 if (error == 0) { 3771 uint64_t txtype = TX_LINK; 3772 zfs_log_link(zilog, tx, txtype, tdzp, szp, name); 3773 } 3774 3775 dmu_tx_commit(tx); 3776 3777 if (error == 0) { 3778 vnevent_link(ZTOV(szp), ct); 3779 } 3780 3781 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3782 zil_commit(zilog, 0); 3783 3784 zfs_exit(zfsvfs, FTAG); 3785 return (error); 3786 } 3787 3788 /* 3789 * Free or allocate space in a file. Currently, this function only 3790 * supports the `F_FREESP' command. However, this command is somewhat 3791 * misnamed, as its functionality includes the ability to allocate as 3792 * well as free space. 3793 * 3794 * IN: ip - inode of file to free data in. 3795 * cmd - action to take (only F_FREESP supported). 3796 * bfp - section of file to free/alloc. 3797 * flag - current file open mode flags. 3798 * offset - current file offset. 3799 * cr - credentials of caller. 3800 * 3801 * RETURN: 0 on success, error code on failure. 3802 * 3803 * Timestamps: 3804 * ip - ctime|mtime updated 3805 */ 3806 int 3807 zfs_space(znode_t *zp, int cmd, flock64_t *bfp, int flag, 3808 offset_t offset, cred_t *cr) 3809 { 3810 (void) offset; 3811 zfsvfs_t *zfsvfs = ZTOZSB(zp); 3812 uint64_t off, len; 3813 int error; 3814 3815 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3816 return (error); 3817 3818 if (cmd != F_FREESP) { 3819 zfs_exit(zfsvfs, FTAG); 3820 return (SET_ERROR(EINVAL)); 3821 } 3822 3823 /* 3824 * Callers might not be able to detect properly that we are read-only, 3825 * so check it explicitly here. 3826 */ 3827 if (zfs_is_readonly(zfsvfs)) { 3828 zfs_exit(zfsvfs, FTAG); 3829 return (SET_ERROR(EROFS)); 3830 } 3831 3832 if (bfp->l_len < 0) { 3833 zfs_exit(zfsvfs, FTAG); 3834 return (SET_ERROR(EINVAL)); 3835 } 3836 3837 /* 3838 * Permissions aren't checked on Solaris because on this OS 3839 * zfs_space() can only be called with an opened file handle. 3840 * On Linux we can get here through truncate_range() which 3841 * operates directly on inodes, so we need to check access rights. 3842 */ 3843 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr, NULL))) { 3844 zfs_exit(zfsvfs, FTAG); 3845 return (error); 3846 } 3847 3848 off = bfp->l_start; 3849 len = bfp->l_len; /* 0 means from off to end of file */ 3850 3851 error = zfs_freesp(zp, off, len, flag, TRUE); 3852 3853 zfs_exit(zfsvfs, FTAG); 3854 return (error); 3855 } 3856 3857 static void 3858 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 3859 { 3860 (void) cr, (void) ct; 3861 znode_t *zp = VTOZ(vp); 3862 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3863 int error; 3864 3865 ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs); 3866 if (zp->z_sa_hdl == NULL) { 3867 /* 3868 * The fs has been unmounted, or we did a 3869 * suspend/resume and this file no longer exists. 3870 */ 3871 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 3872 vrecycle(vp); 3873 return; 3874 } 3875 3876 if (zp->z_unlinked) { 3877 /* 3878 * Fast path to recycle a vnode of a removed file. 3879 */ 3880 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 3881 vrecycle(vp); 3882 return; 3883 } 3884 3885 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 3886 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 3887 3888 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3889 zfs_sa_upgrade_txholds(tx, zp); 3890 error = dmu_tx_assign(tx, TXG_WAIT); 3891 if (error) { 3892 dmu_tx_abort(tx); 3893 } else { 3894 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 3895 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 3896 zp->z_atime_dirty = 0; 3897 dmu_tx_commit(tx); 3898 } 3899 } 3900 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 3901 } 3902 3903 3904 _Static_assert(sizeof (struct zfid_short) <= sizeof (struct fid), 3905 "struct zfid_short bigger than struct fid"); 3906 _Static_assert(sizeof (struct zfid_long) <= sizeof (struct fid), 3907 "struct zfid_long bigger than struct fid"); 3908 3909 static int 3910 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 3911 { 3912 (void) ct; 3913 znode_t *zp = VTOZ(vp); 3914 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3915 uint32_t gen; 3916 uint64_t gen64; 3917 uint64_t object = zp->z_id; 3918 zfid_short_t *zfid; 3919 int size, i, error; 3920 3921 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3922 return (error); 3923 3924 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 3925 &gen64, sizeof (uint64_t))) != 0) { 3926 zfs_exit(zfsvfs, FTAG); 3927 return (error); 3928 } 3929 3930 gen = (uint32_t)gen64; 3931 3932 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 3933 fidp->fid_len = size; 3934 3935 zfid = (zfid_short_t *)fidp; 3936 3937 zfid->zf_len = size; 3938 3939 for (i = 0; i < sizeof (zfid->zf_object); i++) 3940 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 3941 3942 /* Must have a non-zero generation number to distinguish from .zfs */ 3943 if (gen == 0) 3944 gen = 1; 3945 for (i = 0; i < sizeof (zfid->zf_gen); i++) 3946 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 3947 3948 if (size == LONG_FID_LEN) { 3949 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 3950 zfid_long_t *zlfid; 3951 3952 zlfid = (zfid_long_t *)fidp; 3953 3954 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 3955 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 3956 3957 /* XXX - this should be the generation number for the objset */ 3958 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 3959 zlfid->zf_setgen[i] = 0; 3960 } 3961 3962 zfs_exit(zfsvfs, FTAG); 3963 return (0); 3964 } 3965 3966 static int 3967 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 3968 caller_context_t *ct) 3969 { 3970 znode_t *zp; 3971 zfsvfs_t *zfsvfs; 3972 int error; 3973 3974 switch (cmd) { 3975 case _PC_LINK_MAX: 3976 *valp = MIN(LONG_MAX, ZFS_LINK_MAX); 3977 return (0); 3978 3979 case _PC_FILESIZEBITS: 3980 *valp = 64; 3981 return (0); 3982 case _PC_MIN_HOLE_SIZE: 3983 *valp = (int)SPA_MINBLOCKSIZE; 3984 return (0); 3985 case _PC_ACL_EXTENDED: 3986 #if 0 /* POSIX ACLs are not implemented for ZFS on FreeBSD yet. */ 3987 zp = VTOZ(vp); 3988 zfsvfs = zp->z_zfsvfs; 3989 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3990 return (error); 3991 *valp = zfsvfs->z_acl_type == ZFSACLTYPE_POSIX ? 1 : 0; 3992 zfs_exit(zfsvfs, FTAG); 3993 #else 3994 *valp = 0; 3995 #endif 3996 return (0); 3997 3998 case _PC_ACL_NFS4: 3999 zp = VTOZ(vp); 4000 zfsvfs = zp->z_zfsvfs; 4001 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 4002 return (error); 4003 *valp = zfsvfs->z_acl_type == ZFS_ACLTYPE_NFSV4 ? 1 : 0; 4004 zfs_exit(zfsvfs, FTAG); 4005 return (0); 4006 4007 case _PC_ACL_PATH_MAX: 4008 *valp = ACL_MAX_ENTRIES; 4009 return (0); 4010 4011 default: 4012 return (EOPNOTSUPP); 4013 } 4014 } 4015 4016 static int 4017 zfs_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind, 4018 int *rahead) 4019 { 4020 znode_t *zp = VTOZ(vp); 4021 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4022 zfs_locked_range_t *lr; 4023 vm_object_t object; 4024 off_t start, end, obj_size; 4025 uint_t blksz; 4026 int pgsin_b, pgsin_a; 4027 int error; 4028 4029 if (zfs_enter_verify_zp(zfsvfs, zp, FTAG) != 0) 4030 return (zfs_vm_pagerret_error); 4031 4032 start = IDX_TO_OFF(ma[0]->pindex); 4033 end = IDX_TO_OFF(ma[count - 1]->pindex + 1); 4034 4035 /* 4036 * Lock a range covering all required and optional pages. 4037 * Note that we need to handle the case of the block size growing. 4038 */ 4039 for (;;) { 4040 blksz = zp->z_blksz; 4041 lr = zfs_rangelock_tryenter(&zp->z_rangelock, 4042 rounddown(start, blksz), 4043 roundup(end, blksz) - rounddown(start, blksz), RL_READER); 4044 if (lr == NULL) { 4045 if (rahead != NULL) { 4046 *rahead = 0; 4047 rahead = NULL; 4048 } 4049 if (rbehind != NULL) { 4050 *rbehind = 0; 4051 rbehind = NULL; 4052 } 4053 break; 4054 } 4055 if (blksz == zp->z_blksz) 4056 break; 4057 zfs_rangelock_exit(lr); 4058 } 4059 4060 object = ma[0]->object; 4061 zfs_vmobject_wlock(object); 4062 obj_size = object->un_pager.vnp.vnp_size; 4063 zfs_vmobject_wunlock(object); 4064 if (IDX_TO_OFF(ma[count - 1]->pindex) >= obj_size) { 4065 if (lr != NULL) 4066 zfs_rangelock_exit(lr); 4067 zfs_exit(zfsvfs, FTAG); 4068 return (zfs_vm_pagerret_bad); 4069 } 4070 4071 pgsin_b = 0; 4072 if (rbehind != NULL) { 4073 pgsin_b = OFF_TO_IDX(start - rounddown(start, blksz)); 4074 pgsin_b = MIN(*rbehind, pgsin_b); 4075 } 4076 4077 pgsin_a = 0; 4078 if (rahead != NULL) { 4079 pgsin_a = OFF_TO_IDX(roundup(end, blksz) - end); 4080 if (end + IDX_TO_OFF(pgsin_a) >= obj_size) 4081 pgsin_a = OFF_TO_IDX(round_page(obj_size) - end); 4082 pgsin_a = MIN(*rahead, pgsin_a); 4083 } 4084 4085 /* 4086 * NB: we need to pass the exact byte size of the data that we expect 4087 * to read after accounting for the file size. This is required because 4088 * ZFS will panic if we request DMU to read beyond the end of the last 4089 * allocated block. 4090 */ 4091 error = dmu_read_pages(zfsvfs->z_os, zp->z_id, ma, count, &pgsin_b, 4092 &pgsin_a, MIN(end, obj_size) - (end - PAGE_SIZE)); 4093 4094 if (lr != NULL) 4095 zfs_rangelock_exit(lr); 4096 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4097 4098 dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, count*PAGE_SIZE); 4099 4100 zfs_exit(zfsvfs, FTAG); 4101 4102 if (error != 0) 4103 return (zfs_vm_pagerret_error); 4104 4105 VM_CNT_INC(v_vnodein); 4106 VM_CNT_ADD(v_vnodepgsin, count + pgsin_b + pgsin_a); 4107 if (rbehind != NULL) 4108 *rbehind = pgsin_b; 4109 if (rahead != NULL) 4110 *rahead = pgsin_a; 4111 return (zfs_vm_pagerret_ok); 4112 } 4113 4114 #ifndef _SYS_SYSPROTO_H_ 4115 struct vop_getpages_args { 4116 struct vnode *a_vp; 4117 vm_page_t *a_m; 4118 int a_count; 4119 int *a_rbehind; 4120 int *a_rahead; 4121 }; 4122 #endif 4123 4124 static int 4125 zfs_freebsd_getpages(struct vop_getpages_args *ap) 4126 { 4127 4128 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind, 4129 ap->a_rahead)); 4130 } 4131 4132 static int 4133 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags, 4134 int *rtvals) 4135 { 4136 znode_t *zp = VTOZ(vp); 4137 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4138 zfs_locked_range_t *lr; 4139 dmu_tx_t *tx; 4140 struct sf_buf *sf; 4141 vm_object_t object; 4142 vm_page_t m; 4143 caddr_t va; 4144 size_t tocopy; 4145 size_t lo_len; 4146 vm_ooffset_t lo_off; 4147 vm_ooffset_t off; 4148 uint_t blksz; 4149 int ncount; 4150 int pcount; 4151 int err; 4152 int i; 4153 4154 object = vp->v_object; 4155 KASSERT(ma[0]->object == object, ("mismatching object")); 4156 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length")); 4157 4158 pcount = btoc(len); 4159 ncount = pcount; 4160 for (i = 0; i < pcount; i++) 4161 rtvals[i] = zfs_vm_pagerret_error; 4162 4163 if (zfs_enter_verify_zp(zfsvfs, zp, FTAG) != 0) 4164 return (zfs_vm_pagerret_error); 4165 4166 off = IDX_TO_OFF(ma[0]->pindex); 4167 blksz = zp->z_blksz; 4168 lo_off = rounddown(off, blksz); 4169 lo_len = roundup(len + (off - lo_off), blksz); 4170 lr = zfs_rangelock_enter(&zp->z_rangelock, lo_off, lo_len, RL_WRITER); 4171 4172 zfs_vmobject_wlock(object); 4173 if (len + off > object->un_pager.vnp.vnp_size) { 4174 if (object->un_pager.vnp.vnp_size > off) { 4175 int pgoff; 4176 4177 len = object->un_pager.vnp.vnp_size - off; 4178 ncount = btoc(len); 4179 if ((pgoff = (int)len & PAGE_MASK) != 0) { 4180 /* 4181 * If the object is locked and the following 4182 * conditions hold, then the page's dirty 4183 * field cannot be concurrently changed by a 4184 * pmap operation. 4185 */ 4186 m = ma[ncount - 1]; 4187 vm_page_assert_sbusied(m); 4188 KASSERT(!pmap_page_is_write_mapped(m), 4189 ("zfs_putpages: page %p is not read-only", 4190 m)); 4191 vm_page_clear_dirty(m, pgoff, PAGE_SIZE - 4192 pgoff); 4193 } 4194 } else { 4195 len = 0; 4196 ncount = 0; 4197 } 4198 if (ncount < pcount) { 4199 for (i = ncount; i < pcount; i++) { 4200 rtvals[i] = zfs_vm_pagerret_bad; 4201 } 4202 } 4203 } 4204 zfs_vmobject_wunlock(object); 4205 4206 if (ncount == 0) 4207 goto out; 4208 4209 if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT, zp->z_uid) || 4210 zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT, zp->z_gid) || 4211 (zp->z_projid != ZFS_DEFAULT_PROJID && 4212 zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT, 4213 zp->z_projid))) { 4214 goto out; 4215 } 4216 4217 tx = dmu_tx_create(zfsvfs->z_os); 4218 dmu_tx_hold_write(tx, zp->z_id, off, len); 4219 4220 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4221 zfs_sa_upgrade_txholds(tx, zp); 4222 err = dmu_tx_assign(tx, TXG_WAIT); 4223 if (err != 0) { 4224 dmu_tx_abort(tx); 4225 goto out; 4226 } 4227 4228 if (zp->z_blksz < PAGE_SIZE) { 4229 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) { 4230 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len; 4231 va = zfs_map_page(ma[i], &sf); 4232 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx); 4233 zfs_unmap_page(sf); 4234 } 4235 } else { 4236 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx); 4237 } 4238 4239 if (err == 0) { 4240 uint64_t mtime[2], ctime[2]; 4241 sa_bulk_attr_t bulk[3]; 4242 int count = 0; 4243 4244 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4245 &mtime, 16); 4246 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4247 &ctime, 16); 4248 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4249 &zp->z_pflags, 8); 4250 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime); 4251 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 4252 ASSERT0(err); 4253 /* 4254 * XXX we should be passing a callback to undirty 4255 * but that would make the locking messier 4256 */ 4257 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, 4258 len, 0, NULL, NULL); 4259 4260 zfs_vmobject_wlock(object); 4261 for (i = 0; i < ncount; i++) { 4262 rtvals[i] = zfs_vm_pagerret_ok; 4263 vm_page_undirty(ma[i]); 4264 } 4265 zfs_vmobject_wunlock(object); 4266 VM_CNT_INC(v_vnodeout); 4267 VM_CNT_ADD(v_vnodepgsout, ncount); 4268 } 4269 dmu_tx_commit(tx); 4270 4271 out: 4272 zfs_rangelock_exit(lr); 4273 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 || 4274 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4275 zil_commit(zfsvfs->z_log, zp->z_id); 4276 4277 dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, len); 4278 4279 zfs_exit(zfsvfs, FTAG); 4280 return (rtvals[0]); 4281 } 4282 4283 #ifndef _SYS_SYSPROTO_H_ 4284 struct vop_putpages_args { 4285 struct vnode *a_vp; 4286 vm_page_t *a_m; 4287 int a_count; 4288 int a_sync; 4289 int *a_rtvals; 4290 }; 4291 #endif 4292 4293 static int 4294 zfs_freebsd_putpages(struct vop_putpages_args *ap) 4295 { 4296 4297 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync, 4298 ap->a_rtvals)); 4299 } 4300 4301 #ifndef _SYS_SYSPROTO_H_ 4302 struct vop_bmap_args { 4303 struct vnode *a_vp; 4304 daddr_t a_bn; 4305 struct bufobj **a_bop; 4306 daddr_t *a_bnp; 4307 int *a_runp; 4308 int *a_runb; 4309 }; 4310 #endif 4311 4312 static int 4313 zfs_freebsd_bmap(struct vop_bmap_args *ap) 4314 { 4315 4316 if (ap->a_bop != NULL) 4317 *ap->a_bop = &ap->a_vp->v_bufobj; 4318 if (ap->a_bnp != NULL) 4319 *ap->a_bnp = ap->a_bn; 4320 if (ap->a_runp != NULL) 4321 *ap->a_runp = 0; 4322 if (ap->a_runb != NULL) 4323 *ap->a_runb = 0; 4324 4325 return (0); 4326 } 4327 4328 #ifndef _SYS_SYSPROTO_H_ 4329 struct vop_open_args { 4330 struct vnode *a_vp; 4331 int a_mode; 4332 struct ucred *a_cred; 4333 struct thread *a_td; 4334 }; 4335 #endif 4336 4337 static int 4338 zfs_freebsd_open(struct vop_open_args *ap) 4339 { 4340 vnode_t *vp = ap->a_vp; 4341 znode_t *zp = VTOZ(vp); 4342 int error; 4343 4344 error = zfs_open(&vp, ap->a_mode, ap->a_cred); 4345 if (error == 0) 4346 vnode_create_vobject(vp, zp->z_size, ap->a_td); 4347 return (error); 4348 } 4349 4350 #ifndef _SYS_SYSPROTO_H_ 4351 struct vop_close_args { 4352 struct vnode *a_vp; 4353 int a_fflag; 4354 struct ucred *a_cred; 4355 struct thread *a_td; 4356 }; 4357 #endif 4358 4359 static int 4360 zfs_freebsd_close(struct vop_close_args *ap) 4361 { 4362 4363 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred)); 4364 } 4365 4366 #ifndef _SYS_SYSPROTO_H_ 4367 struct vop_ioctl_args { 4368 struct vnode *a_vp; 4369 ulong_t a_command; 4370 caddr_t a_data; 4371 int a_fflag; 4372 struct ucred *cred; 4373 struct thread *td; 4374 }; 4375 #endif 4376 4377 static int 4378 zfs_freebsd_ioctl(struct vop_ioctl_args *ap) 4379 { 4380 4381 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 4382 ap->a_fflag, ap->a_cred, NULL)); 4383 } 4384 4385 static int 4386 ioflags(int ioflags) 4387 { 4388 int flags = 0; 4389 4390 if (ioflags & IO_APPEND) 4391 flags |= O_APPEND; 4392 if (ioflags & IO_NDELAY) 4393 flags |= O_NONBLOCK; 4394 if (ioflags & IO_SYNC) 4395 flags |= O_SYNC; 4396 4397 return (flags); 4398 } 4399 4400 #ifndef _SYS_SYSPROTO_H_ 4401 struct vop_read_args { 4402 struct vnode *a_vp; 4403 struct uio *a_uio; 4404 int a_ioflag; 4405 struct ucred *a_cred; 4406 }; 4407 #endif 4408 4409 static int 4410 zfs_freebsd_read(struct vop_read_args *ap) 4411 { 4412 zfs_uio_t uio; 4413 zfs_uio_init(&uio, ap->a_uio); 4414 return (zfs_read(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag), 4415 ap->a_cred)); 4416 } 4417 4418 #ifndef _SYS_SYSPROTO_H_ 4419 struct vop_write_args { 4420 struct vnode *a_vp; 4421 struct uio *a_uio; 4422 int a_ioflag; 4423 struct ucred *a_cred; 4424 }; 4425 #endif 4426 4427 static int 4428 zfs_freebsd_write(struct vop_write_args *ap) 4429 { 4430 zfs_uio_t uio; 4431 zfs_uio_init(&uio, ap->a_uio); 4432 return (zfs_write(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag), 4433 ap->a_cred)); 4434 } 4435 4436 #if __FreeBSD_version >= 1300102 4437 /* 4438 * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see 4439 * the comment above cache_fplookup for details. 4440 */ 4441 static int 4442 zfs_freebsd_fplookup_vexec(struct vop_fplookup_vexec_args *v) 4443 { 4444 vnode_t *vp; 4445 znode_t *zp; 4446 uint64_t pflags; 4447 4448 vp = v->a_vp; 4449 zp = VTOZ_SMR(vp); 4450 if (__predict_false(zp == NULL)) 4451 return (EAGAIN); 4452 pflags = atomic_load_64(&zp->z_pflags); 4453 if (pflags & ZFS_AV_QUARANTINED) 4454 return (EAGAIN); 4455 if (pflags & ZFS_XATTR) 4456 return (EAGAIN); 4457 if ((pflags & ZFS_NO_EXECS_DENIED) == 0) 4458 return (EAGAIN); 4459 return (0); 4460 } 4461 #endif 4462 4463 #if __FreeBSD_version >= 1300139 4464 static int 4465 zfs_freebsd_fplookup_symlink(struct vop_fplookup_symlink_args *v) 4466 { 4467 vnode_t *vp; 4468 znode_t *zp; 4469 char *target; 4470 4471 vp = v->a_vp; 4472 zp = VTOZ_SMR(vp); 4473 if (__predict_false(zp == NULL)) { 4474 return (EAGAIN); 4475 } 4476 4477 target = atomic_load_consume_ptr(&zp->z_cached_symlink); 4478 if (target == NULL) { 4479 return (EAGAIN); 4480 } 4481 return (cache_symlink_resolve(v->a_fpl, target, strlen(target))); 4482 } 4483 #endif 4484 4485 #ifndef _SYS_SYSPROTO_H_ 4486 struct vop_access_args { 4487 struct vnode *a_vp; 4488 accmode_t a_accmode; 4489 struct ucred *a_cred; 4490 struct thread *a_td; 4491 }; 4492 #endif 4493 4494 static int 4495 zfs_freebsd_access(struct vop_access_args *ap) 4496 { 4497 vnode_t *vp = ap->a_vp; 4498 znode_t *zp = VTOZ(vp); 4499 accmode_t accmode; 4500 int error = 0; 4501 4502 4503 if (ap->a_accmode == VEXEC) { 4504 if (zfs_fastaccesschk_execute(zp, ap->a_cred) == 0) 4505 return (0); 4506 } 4507 4508 /* 4509 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 4510 */ 4511 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 4512 if (accmode != 0) 4513 error = zfs_access(zp, accmode, 0, ap->a_cred); 4514 4515 /* 4516 * VADMIN has to be handled by vaccess(). 4517 */ 4518 if (error == 0) { 4519 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 4520 if (accmode != 0) { 4521 #if __FreeBSD_version >= 1300105 4522 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 4523 zp->z_gid, accmode, ap->a_cred); 4524 #else 4525 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 4526 zp->z_gid, accmode, ap->a_cred, NULL); 4527 #endif 4528 } 4529 } 4530 4531 /* 4532 * For VEXEC, ensure that at least one execute bit is set for 4533 * non-directories. 4534 */ 4535 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 4536 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 4537 error = EACCES; 4538 } 4539 4540 return (error); 4541 } 4542 4543 #ifndef _SYS_SYSPROTO_H_ 4544 struct vop_lookup_args { 4545 struct vnode *a_dvp; 4546 struct vnode **a_vpp; 4547 struct componentname *a_cnp; 4548 }; 4549 #endif 4550 4551 static int 4552 zfs_freebsd_lookup(struct vop_lookup_args *ap, boolean_t cached) 4553 { 4554 struct componentname *cnp = ap->a_cnp; 4555 char nm[NAME_MAX + 1]; 4556 4557 ASSERT3U(cnp->cn_namelen, <, sizeof (nm)); 4558 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof (nm))); 4559 4560 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 4561 cnp->cn_cred, 0, cached)); 4562 } 4563 4564 static int 4565 zfs_freebsd_cachedlookup(struct vop_cachedlookup_args *ap) 4566 { 4567 4568 return (zfs_freebsd_lookup((struct vop_lookup_args *)ap, B_TRUE)); 4569 } 4570 4571 #ifndef _SYS_SYSPROTO_H_ 4572 struct vop_lookup_args { 4573 struct vnode *a_dvp; 4574 struct vnode **a_vpp; 4575 struct componentname *a_cnp; 4576 }; 4577 #endif 4578 4579 static int 4580 zfs_cache_lookup(struct vop_lookup_args *ap) 4581 { 4582 zfsvfs_t *zfsvfs; 4583 4584 zfsvfs = ap->a_dvp->v_mount->mnt_data; 4585 if (zfsvfs->z_use_namecache) 4586 return (vfs_cache_lookup(ap)); 4587 else 4588 return (zfs_freebsd_lookup(ap, B_FALSE)); 4589 } 4590 4591 #ifndef _SYS_SYSPROTO_H_ 4592 struct vop_create_args { 4593 struct vnode *a_dvp; 4594 struct vnode **a_vpp; 4595 struct componentname *a_cnp; 4596 struct vattr *a_vap; 4597 }; 4598 #endif 4599 4600 static int 4601 zfs_freebsd_create(struct vop_create_args *ap) 4602 { 4603 zfsvfs_t *zfsvfs; 4604 struct componentname *cnp = ap->a_cnp; 4605 vattr_t *vap = ap->a_vap; 4606 znode_t *zp = NULL; 4607 int rc, mode; 4608 4609 #if __FreeBSD_version < 1400068 4610 ASSERT(cnp->cn_flags & SAVENAME); 4611 #endif 4612 4613 vattr_init_mask(vap); 4614 mode = vap->va_mode & ALLPERMS; 4615 zfsvfs = ap->a_dvp->v_mount->mnt_data; 4616 *ap->a_vpp = NULL; 4617 4618 rc = zfs_create(VTOZ(ap->a_dvp), cnp->cn_nameptr, vap, 0, mode, 4619 &zp, cnp->cn_cred, 0 /* flag */, NULL /* vsecattr */, NULL); 4620 if (rc == 0) 4621 *ap->a_vpp = ZTOV(zp); 4622 if (zfsvfs->z_use_namecache && 4623 rc == 0 && (cnp->cn_flags & MAKEENTRY) != 0) 4624 cache_enter(ap->a_dvp, *ap->a_vpp, cnp); 4625 4626 return (rc); 4627 } 4628 4629 #ifndef _SYS_SYSPROTO_H_ 4630 struct vop_remove_args { 4631 struct vnode *a_dvp; 4632 struct vnode *a_vp; 4633 struct componentname *a_cnp; 4634 }; 4635 #endif 4636 4637 static int 4638 zfs_freebsd_remove(struct vop_remove_args *ap) 4639 { 4640 4641 #if __FreeBSD_version < 1400068 4642 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 4643 #endif 4644 4645 return (zfs_remove_(ap->a_dvp, ap->a_vp, ap->a_cnp->cn_nameptr, 4646 ap->a_cnp->cn_cred)); 4647 } 4648 4649 #ifndef _SYS_SYSPROTO_H_ 4650 struct vop_mkdir_args { 4651 struct vnode *a_dvp; 4652 struct vnode **a_vpp; 4653 struct componentname *a_cnp; 4654 struct vattr *a_vap; 4655 }; 4656 #endif 4657 4658 static int 4659 zfs_freebsd_mkdir(struct vop_mkdir_args *ap) 4660 { 4661 vattr_t *vap = ap->a_vap; 4662 znode_t *zp = NULL; 4663 int rc; 4664 4665 #if __FreeBSD_version < 1400068 4666 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 4667 #endif 4668 4669 vattr_init_mask(vap); 4670 *ap->a_vpp = NULL; 4671 4672 rc = zfs_mkdir(VTOZ(ap->a_dvp), ap->a_cnp->cn_nameptr, vap, &zp, 4673 ap->a_cnp->cn_cred, 0, NULL, NULL); 4674 4675 if (rc == 0) 4676 *ap->a_vpp = ZTOV(zp); 4677 return (rc); 4678 } 4679 4680 #ifndef _SYS_SYSPROTO_H_ 4681 struct vop_rmdir_args { 4682 struct vnode *a_dvp; 4683 struct vnode *a_vp; 4684 struct componentname *a_cnp; 4685 }; 4686 #endif 4687 4688 static int 4689 zfs_freebsd_rmdir(struct vop_rmdir_args *ap) 4690 { 4691 struct componentname *cnp = ap->a_cnp; 4692 4693 #if __FreeBSD_version < 1400068 4694 ASSERT(cnp->cn_flags & SAVENAME); 4695 #endif 4696 4697 return (zfs_rmdir_(ap->a_dvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred)); 4698 } 4699 4700 #ifndef _SYS_SYSPROTO_H_ 4701 struct vop_readdir_args { 4702 struct vnode *a_vp; 4703 struct uio *a_uio; 4704 struct ucred *a_cred; 4705 int *a_eofflag; 4706 int *a_ncookies; 4707 cookie_t **a_cookies; 4708 }; 4709 #endif 4710 4711 static int 4712 zfs_freebsd_readdir(struct vop_readdir_args *ap) 4713 { 4714 zfs_uio_t uio; 4715 zfs_uio_init(&uio, ap->a_uio); 4716 return (zfs_readdir(ap->a_vp, &uio, ap->a_cred, ap->a_eofflag, 4717 ap->a_ncookies, ap->a_cookies)); 4718 } 4719 4720 #ifndef _SYS_SYSPROTO_H_ 4721 struct vop_fsync_args { 4722 struct vnode *a_vp; 4723 int a_waitfor; 4724 struct thread *a_td; 4725 }; 4726 #endif 4727 4728 static int 4729 zfs_freebsd_fsync(struct vop_fsync_args *ap) 4730 { 4731 4732 return (zfs_fsync(VTOZ(ap->a_vp), 0, ap->a_td->td_ucred)); 4733 } 4734 4735 #ifndef _SYS_SYSPROTO_H_ 4736 struct vop_getattr_args { 4737 struct vnode *a_vp; 4738 struct vattr *a_vap; 4739 struct ucred *a_cred; 4740 }; 4741 #endif 4742 4743 static int 4744 zfs_freebsd_getattr(struct vop_getattr_args *ap) 4745 { 4746 vattr_t *vap = ap->a_vap; 4747 xvattr_t xvap; 4748 ulong_t fflags = 0; 4749 int error; 4750 4751 xva_init(&xvap); 4752 xvap.xva_vattr = *vap; 4753 xvap.xva_vattr.va_mask |= AT_XVATTR; 4754 4755 /* Convert chflags into ZFS-type flags. */ 4756 /* XXX: what about SF_SETTABLE?. */ 4757 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 4758 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 4759 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 4760 XVA_SET_REQ(&xvap, XAT_NODUMP); 4761 XVA_SET_REQ(&xvap, XAT_READONLY); 4762 XVA_SET_REQ(&xvap, XAT_ARCHIVE); 4763 XVA_SET_REQ(&xvap, XAT_SYSTEM); 4764 XVA_SET_REQ(&xvap, XAT_HIDDEN); 4765 XVA_SET_REQ(&xvap, XAT_REPARSE); 4766 XVA_SET_REQ(&xvap, XAT_OFFLINE); 4767 XVA_SET_REQ(&xvap, XAT_SPARSE); 4768 4769 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred); 4770 if (error != 0) 4771 return (error); 4772 4773 /* Convert ZFS xattr into chflags. */ 4774 #define FLAG_CHECK(fflag, xflag, xfield) do { \ 4775 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 4776 fflags |= (fflag); \ 4777 } while (0) 4778 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 4779 xvap.xva_xoptattrs.xoa_immutable); 4780 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 4781 xvap.xva_xoptattrs.xoa_appendonly); 4782 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 4783 xvap.xva_xoptattrs.xoa_nounlink); 4784 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE, 4785 xvap.xva_xoptattrs.xoa_archive); 4786 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 4787 xvap.xva_xoptattrs.xoa_nodump); 4788 FLAG_CHECK(UF_READONLY, XAT_READONLY, 4789 xvap.xva_xoptattrs.xoa_readonly); 4790 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM, 4791 xvap.xva_xoptattrs.xoa_system); 4792 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN, 4793 xvap.xva_xoptattrs.xoa_hidden); 4794 FLAG_CHECK(UF_REPARSE, XAT_REPARSE, 4795 xvap.xva_xoptattrs.xoa_reparse); 4796 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE, 4797 xvap.xva_xoptattrs.xoa_offline); 4798 FLAG_CHECK(UF_SPARSE, XAT_SPARSE, 4799 xvap.xva_xoptattrs.xoa_sparse); 4800 4801 #undef FLAG_CHECK 4802 *vap = xvap.xva_vattr; 4803 vap->va_flags = fflags; 4804 return (0); 4805 } 4806 4807 #ifndef _SYS_SYSPROTO_H_ 4808 struct vop_setattr_args { 4809 struct vnode *a_vp; 4810 struct vattr *a_vap; 4811 struct ucred *a_cred; 4812 }; 4813 #endif 4814 4815 static int 4816 zfs_freebsd_setattr(struct vop_setattr_args *ap) 4817 { 4818 vnode_t *vp = ap->a_vp; 4819 vattr_t *vap = ap->a_vap; 4820 cred_t *cred = ap->a_cred; 4821 xvattr_t xvap; 4822 ulong_t fflags; 4823 uint64_t zflags; 4824 4825 vattr_init_mask(vap); 4826 vap->va_mask &= ~AT_NOSET; 4827 4828 xva_init(&xvap); 4829 xvap.xva_vattr = *vap; 4830 4831 zflags = VTOZ(vp)->z_pflags; 4832 4833 if (vap->va_flags != VNOVAL) { 4834 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 4835 int error; 4836 4837 if (zfsvfs->z_use_fuids == B_FALSE) 4838 return (EOPNOTSUPP); 4839 4840 fflags = vap->va_flags; 4841 /* 4842 * XXX KDM 4843 * We need to figure out whether it makes sense to allow 4844 * UF_REPARSE through, since we don't really have other 4845 * facilities to handle reparse points and zfs_setattr() 4846 * doesn't currently allow setting that attribute anyway. 4847 */ 4848 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE| 4849 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE| 4850 UF_OFFLINE|UF_SPARSE)) != 0) 4851 return (EOPNOTSUPP); 4852 /* 4853 * Unprivileged processes are not permitted to unset system 4854 * flags, or modify flags if any system flags are set. 4855 * Privileged non-jail processes may not modify system flags 4856 * if securelevel > 0 and any existing system flags are set. 4857 * Privileged jail processes behave like privileged non-jail 4858 * processes if the PR_ALLOW_CHFLAGS permission bit is set; 4859 * otherwise, they behave like unprivileged processes. 4860 */ 4861 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 4862 spl_priv_check_cred(cred, PRIV_VFS_SYSFLAGS) == 0) { 4863 if (zflags & 4864 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 4865 error = securelevel_gt(cred, 0); 4866 if (error != 0) 4867 return (error); 4868 } 4869 } else { 4870 /* 4871 * Callers may only modify the file flags on 4872 * objects they have VADMIN rights for. 4873 */ 4874 if ((error = VOP_ACCESS(vp, VADMIN, cred, 4875 curthread)) != 0) 4876 return (error); 4877 if (zflags & 4878 (ZFS_IMMUTABLE | ZFS_APPENDONLY | 4879 ZFS_NOUNLINK)) { 4880 return (EPERM); 4881 } 4882 if (fflags & 4883 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 4884 return (EPERM); 4885 } 4886 } 4887 4888 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 4889 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 4890 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 4891 XVA_SET_REQ(&xvap, (xflag)); \ 4892 (xfield) = ((fflags & (fflag)) != 0); \ 4893 } \ 4894 } while (0) 4895 /* Convert chflags into ZFS-type flags. */ 4896 /* XXX: what about SF_SETTABLE?. */ 4897 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 4898 xvap.xva_xoptattrs.xoa_immutable); 4899 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 4900 xvap.xva_xoptattrs.xoa_appendonly); 4901 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 4902 xvap.xva_xoptattrs.xoa_nounlink); 4903 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE, 4904 xvap.xva_xoptattrs.xoa_archive); 4905 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 4906 xvap.xva_xoptattrs.xoa_nodump); 4907 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY, 4908 xvap.xva_xoptattrs.xoa_readonly); 4909 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM, 4910 xvap.xva_xoptattrs.xoa_system); 4911 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN, 4912 xvap.xva_xoptattrs.xoa_hidden); 4913 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE, 4914 xvap.xva_xoptattrs.xoa_reparse); 4915 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE, 4916 xvap.xva_xoptattrs.xoa_offline); 4917 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE, 4918 xvap.xva_xoptattrs.xoa_sparse); 4919 #undef FLAG_CHANGE 4920 } 4921 if (vap->va_birthtime.tv_sec != VNOVAL) { 4922 xvap.xva_vattr.va_mask |= AT_XVATTR; 4923 XVA_SET_REQ(&xvap, XAT_CREATETIME); 4924 } 4925 return (zfs_setattr(VTOZ(vp), (vattr_t *)&xvap, 0, cred, NULL)); 4926 } 4927 4928 #ifndef _SYS_SYSPROTO_H_ 4929 struct vop_rename_args { 4930 struct vnode *a_fdvp; 4931 struct vnode *a_fvp; 4932 struct componentname *a_fcnp; 4933 struct vnode *a_tdvp; 4934 struct vnode *a_tvp; 4935 struct componentname *a_tcnp; 4936 }; 4937 #endif 4938 4939 static int 4940 zfs_freebsd_rename(struct vop_rename_args *ap) 4941 { 4942 vnode_t *fdvp = ap->a_fdvp; 4943 vnode_t *fvp = ap->a_fvp; 4944 vnode_t *tdvp = ap->a_tdvp; 4945 vnode_t *tvp = ap->a_tvp; 4946 int error; 4947 4948 #if __FreeBSD_version < 1400068 4949 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 4950 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 4951 #endif 4952 4953 error = zfs_do_rename(fdvp, &fvp, ap->a_fcnp, tdvp, &tvp, 4954 ap->a_tcnp, ap->a_fcnp->cn_cred); 4955 4956 vrele(fdvp); 4957 vrele(fvp); 4958 vrele(tdvp); 4959 if (tvp != NULL) 4960 vrele(tvp); 4961 4962 return (error); 4963 } 4964 4965 #ifndef _SYS_SYSPROTO_H_ 4966 struct vop_symlink_args { 4967 struct vnode *a_dvp; 4968 struct vnode **a_vpp; 4969 struct componentname *a_cnp; 4970 struct vattr *a_vap; 4971 char *a_target; 4972 }; 4973 #endif 4974 4975 static int 4976 zfs_freebsd_symlink(struct vop_symlink_args *ap) 4977 { 4978 struct componentname *cnp = ap->a_cnp; 4979 vattr_t *vap = ap->a_vap; 4980 znode_t *zp = NULL; 4981 #if __FreeBSD_version >= 1300139 4982 char *symlink; 4983 size_t symlink_len; 4984 #endif 4985 int rc; 4986 4987 #if __FreeBSD_version < 1400068 4988 ASSERT(cnp->cn_flags & SAVENAME); 4989 #endif 4990 4991 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 4992 vattr_init_mask(vap); 4993 *ap->a_vpp = NULL; 4994 4995 rc = zfs_symlink(VTOZ(ap->a_dvp), cnp->cn_nameptr, vap, 4996 ap->a_target, &zp, cnp->cn_cred, 0 /* flags */, NULL); 4997 if (rc == 0) { 4998 *ap->a_vpp = ZTOV(zp); 4999 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__); 5000 #if __FreeBSD_version >= 1300139 5001 MPASS(zp->z_cached_symlink == NULL); 5002 symlink_len = strlen(ap->a_target); 5003 symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK); 5004 if (symlink != NULL) { 5005 memcpy(symlink, ap->a_target, symlink_len); 5006 symlink[symlink_len] = '\0'; 5007 atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink, 5008 (uintptr_t)symlink); 5009 } 5010 #endif 5011 } 5012 return (rc); 5013 } 5014 5015 #ifndef _SYS_SYSPROTO_H_ 5016 struct vop_readlink_args { 5017 struct vnode *a_vp; 5018 struct uio *a_uio; 5019 struct ucred *a_cred; 5020 }; 5021 #endif 5022 5023 static int 5024 zfs_freebsd_readlink(struct vop_readlink_args *ap) 5025 { 5026 zfs_uio_t uio; 5027 int error; 5028 #if __FreeBSD_version >= 1300139 5029 znode_t *zp = VTOZ(ap->a_vp); 5030 char *symlink, *base; 5031 size_t symlink_len; 5032 bool trycache; 5033 #endif 5034 5035 zfs_uio_init(&uio, ap->a_uio); 5036 #if __FreeBSD_version >= 1300139 5037 trycache = false; 5038 if (zfs_uio_segflg(&uio) == UIO_SYSSPACE && 5039 zfs_uio_iovcnt(&uio) == 1) { 5040 base = zfs_uio_iovbase(&uio, 0); 5041 symlink_len = zfs_uio_iovlen(&uio, 0); 5042 trycache = true; 5043 } 5044 #endif 5045 error = zfs_readlink(ap->a_vp, &uio, ap->a_cred, NULL); 5046 #if __FreeBSD_version >= 1300139 5047 if (atomic_load_ptr(&zp->z_cached_symlink) != NULL || 5048 error != 0 || !trycache) { 5049 return (error); 5050 } 5051 symlink_len -= zfs_uio_resid(&uio); 5052 symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK); 5053 if (symlink != NULL) { 5054 memcpy(symlink, base, symlink_len); 5055 symlink[symlink_len] = '\0'; 5056 if (!atomic_cmpset_rel_ptr((uintptr_t *)&zp->z_cached_symlink, 5057 (uintptr_t)NULL, (uintptr_t)symlink)) { 5058 cache_symlink_free(symlink, symlink_len + 1); 5059 } 5060 } 5061 #endif 5062 return (error); 5063 } 5064 5065 #ifndef _SYS_SYSPROTO_H_ 5066 struct vop_link_args { 5067 struct vnode *a_tdvp; 5068 struct vnode *a_vp; 5069 struct componentname *a_cnp; 5070 }; 5071 #endif 5072 5073 static int 5074 zfs_freebsd_link(struct vop_link_args *ap) 5075 { 5076 struct componentname *cnp = ap->a_cnp; 5077 vnode_t *vp = ap->a_vp; 5078 vnode_t *tdvp = ap->a_tdvp; 5079 5080 if (tdvp->v_mount != vp->v_mount) 5081 return (EXDEV); 5082 5083 #if __FreeBSD_version < 1400068 5084 ASSERT(cnp->cn_flags & SAVENAME); 5085 #endif 5086 5087 return (zfs_link(VTOZ(tdvp), VTOZ(vp), 5088 cnp->cn_nameptr, cnp->cn_cred, 0)); 5089 } 5090 5091 #ifndef _SYS_SYSPROTO_H_ 5092 struct vop_inactive_args { 5093 struct vnode *a_vp; 5094 struct thread *a_td; 5095 }; 5096 #endif 5097 5098 static int 5099 zfs_freebsd_inactive(struct vop_inactive_args *ap) 5100 { 5101 vnode_t *vp = ap->a_vp; 5102 5103 #if __FreeBSD_version >= 1300123 5104 zfs_inactive(vp, curthread->td_ucred, NULL); 5105 #else 5106 zfs_inactive(vp, ap->a_td->td_ucred, NULL); 5107 #endif 5108 return (0); 5109 } 5110 5111 #if __FreeBSD_version >= 1300042 5112 #ifndef _SYS_SYSPROTO_H_ 5113 struct vop_need_inactive_args { 5114 struct vnode *a_vp; 5115 struct thread *a_td; 5116 }; 5117 #endif 5118 5119 static int 5120 zfs_freebsd_need_inactive(struct vop_need_inactive_args *ap) 5121 { 5122 vnode_t *vp = ap->a_vp; 5123 znode_t *zp = VTOZ(vp); 5124 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5125 int need; 5126 5127 if (vn_need_pageq_flush(vp)) 5128 return (1); 5129 5130 if (!ZFS_TEARDOWN_INACTIVE_TRY_ENTER_READ(zfsvfs)) 5131 return (1); 5132 need = (zp->z_sa_hdl == NULL || zp->z_unlinked || zp->z_atime_dirty); 5133 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 5134 5135 return (need); 5136 } 5137 #endif 5138 5139 #ifndef _SYS_SYSPROTO_H_ 5140 struct vop_reclaim_args { 5141 struct vnode *a_vp; 5142 struct thread *a_td; 5143 }; 5144 #endif 5145 5146 static int 5147 zfs_freebsd_reclaim(struct vop_reclaim_args *ap) 5148 { 5149 vnode_t *vp = ap->a_vp; 5150 znode_t *zp = VTOZ(vp); 5151 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5152 5153 ASSERT3P(zp, !=, NULL); 5154 5155 #if __FreeBSD_version < 1300042 5156 /* Destroy the vm object and flush associated pages. */ 5157 vnode_destroy_vobject(vp); 5158 #endif 5159 /* 5160 * z_teardown_inactive_lock protects from a race with 5161 * zfs_znode_dmu_fini in zfsvfs_teardown during 5162 * force unmount. 5163 */ 5164 ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs); 5165 if (zp->z_sa_hdl == NULL) 5166 zfs_znode_free(zp); 5167 else 5168 zfs_zinactive(zp); 5169 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 5170 5171 vp->v_data = NULL; 5172 return (0); 5173 } 5174 5175 #ifndef _SYS_SYSPROTO_H_ 5176 struct vop_fid_args { 5177 struct vnode *a_vp; 5178 struct fid *a_fid; 5179 }; 5180 #endif 5181 5182 static int 5183 zfs_freebsd_fid(struct vop_fid_args *ap) 5184 { 5185 5186 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 5187 } 5188 5189 5190 #ifndef _SYS_SYSPROTO_H_ 5191 struct vop_pathconf_args { 5192 struct vnode *a_vp; 5193 int a_name; 5194 register_t *a_retval; 5195 } *ap; 5196 #endif 5197 5198 static int 5199 zfs_freebsd_pathconf(struct vop_pathconf_args *ap) 5200 { 5201 ulong_t val; 5202 int error; 5203 5204 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, 5205 curthread->td_ucred, NULL); 5206 if (error == 0) { 5207 *ap->a_retval = val; 5208 return (error); 5209 } 5210 if (error != EOPNOTSUPP) 5211 return (error); 5212 5213 switch (ap->a_name) { 5214 case _PC_NAME_MAX: 5215 *ap->a_retval = NAME_MAX; 5216 return (0); 5217 #if __FreeBSD_version >= 1400032 5218 case _PC_DEALLOC_PRESENT: 5219 *ap->a_retval = 1; 5220 return (0); 5221 #endif 5222 case _PC_PIPE_BUF: 5223 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) { 5224 *ap->a_retval = PIPE_BUF; 5225 return (0); 5226 } 5227 return (EINVAL); 5228 default: 5229 return (vop_stdpathconf(ap)); 5230 } 5231 } 5232 5233 static int zfs_xattr_compat = 1; 5234 5235 static int 5236 zfs_check_attrname(const char *name) 5237 { 5238 /* We don't allow '/' character in attribute name. */ 5239 if (strchr(name, '/') != NULL) 5240 return (SET_ERROR(EINVAL)); 5241 /* We don't allow attribute names that start with a namespace prefix. */ 5242 if (ZFS_XA_NS_PREFIX_FORBIDDEN(name)) 5243 return (SET_ERROR(EINVAL)); 5244 return (0); 5245 } 5246 5247 /* 5248 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 5249 * extended attribute name: 5250 * 5251 * NAMESPACE XATTR_COMPAT PREFIX 5252 * system * freebsd:system: 5253 * user 1 (none, can be used to access ZFS 5254 * fsattr(5) attributes created on Solaris) 5255 * user 0 user. 5256 */ 5257 static int 5258 zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 5259 size_t size, boolean_t compat) 5260 { 5261 const char *namespace, *prefix, *suffix; 5262 5263 memset(attrname, 0, size); 5264 5265 switch (attrnamespace) { 5266 case EXTATTR_NAMESPACE_USER: 5267 if (compat) { 5268 /* 5269 * This is the default namespace by which we can access 5270 * all attributes created on Solaris. 5271 */ 5272 prefix = namespace = suffix = ""; 5273 } else { 5274 /* 5275 * This is compatible with the user namespace encoding 5276 * on Linux prior to xattr_compat, but nothing 5277 * else. 5278 */ 5279 prefix = ""; 5280 namespace = "user"; 5281 suffix = "."; 5282 } 5283 break; 5284 case EXTATTR_NAMESPACE_SYSTEM: 5285 prefix = "freebsd:"; 5286 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 5287 suffix = ":"; 5288 break; 5289 case EXTATTR_NAMESPACE_EMPTY: 5290 default: 5291 return (SET_ERROR(EINVAL)); 5292 } 5293 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 5294 name) >= size) { 5295 return (SET_ERROR(ENAMETOOLONG)); 5296 } 5297 return (0); 5298 } 5299 5300 static int 5301 zfs_ensure_xattr_cached(znode_t *zp) 5302 { 5303 int error = 0; 5304 5305 ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock)); 5306 5307 if (zp->z_xattr_cached != NULL) 5308 return (0); 5309 5310 if (rw_write_held(&zp->z_xattr_lock)) 5311 return (zfs_sa_get_xattr(zp)); 5312 5313 if (!rw_tryupgrade(&zp->z_xattr_lock)) { 5314 rw_exit(&zp->z_xattr_lock); 5315 rw_enter(&zp->z_xattr_lock, RW_WRITER); 5316 } 5317 if (zp->z_xattr_cached == NULL) 5318 error = zfs_sa_get_xattr(zp); 5319 rw_downgrade(&zp->z_xattr_lock); 5320 return (error); 5321 } 5322 5323 #ifndef _SYS_SYSPROTO_H_ 5324 struct vop_getextattr { 5325 IN struct vnode *a_vp; 5326 IN int a_attrnamespace; 5327 IN const char *a_name; 5328 INOUT struct uio *a_uio; 5329 OUT size_t *a_size; 5330 IN struct ucred *a_cred; 5331 IN struct thread *a_td; 5332 }; 5333 #endif 5334 5335 static int 5336 zfs_getextattr_dir(struct vop_getextattr_args *ap, const char *attrname) 5337 { 5338 struct thread *td = ap->a_td; 5339 struct nameidata nd; 5340 struct vattr va; 5341 vnode_t *xvp = NULL, *vp; 5342 int error, flags; 5343 5344 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, 5345 LOOKUP_XATTR, B_FALSE); 5346 if (error != 0) 5347 return (error); 5348 5349 flags = FREAD; 5350 #if __FreeBSD_version < 1400043 5351 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 5352 xvp, td); 5353 #else 5354 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp); 5355 #endif 5356 error = vn_open_cred(&nd, &flags, 0, VN_OPEN_INVFS, ap->a_cred, NULL); 5357 if (error != 0) 5358 return (SET_ERROR(error)); 5359 vp = nd.ni_vp; 5360 NDFREE_PNBUF(&nd); 5361 5362 if (ap->a_size != NULL) { 5363 error = VOP_GETATTR(vp, &va, ap->a_cred); 5364 if (error == 0) 5365 *ap->a_size = (size_t)va.va_size; 5366 } else if (ap->a_uio != NULL) 5367 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); 5368 5369 VOP_UNLOCK1(vp); 5370 vn_close(vp, flags, ap->a_cred, td); 5371 return (error); 5372 } 5373 5374 static int 5375 zfs_getextattr_sa(struct vop_getextattr_args *ap, const char *attrname) 5376 { 5377 znode_t *zp = VTOZ(ap->a_vp); 5378 uchar_t *nv_value; 5379 uint_t nv_size; 5380 int error; 5381 5382 error = zfs_ensure_xattr_cached(zp); 5383 if (error != 0) 5384 return (error); 5385 5386 ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock)); 5387 ASSERT3P(zp->z_xattr_cached, !=, NULL); 5388 5389 error = nvlist_lookup_byte_array(zp->z_xattr_cached, attrname, 5390 &nv_value, &nv_size); 5391 if (error != 0) 5392 return (SET_ERROR(error)); 5393 5394 if (ap->a_size != NULL) 5395 *ap->a_size = nv_size; 5396 else if (ap->a_uio != NULL) 5397 error = uiomove(nv_value, nv_size, ap->a_uio); 5398 if (error != 0) 5399 return (SET_ERROR(error)); 5400 5401 return (0); 5402 } 5403 5404 static int 5405 zfs_getextattr_impl(struct vop_getextattr_args *ap, boolean_t compat) 5406 { 5407 znode_t *zp = VTOZ(ap->a_vp); 5408 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5409 char attrname[EXTATTR_MAXNAMELEN+1]; 5410 int error; 5411 5412 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 5413 sizeof (attrname), compat); 5414 if (error != 0) 5415 return (error); 5416 5417 error = ENOENT; 5418 if (zfsvfs->z_use_sa && zp->z_is_sa) 5419 error = zfs_getextattr_sa(ap, attrname); 5420 if (error == ENOENT) 5421 error = zfs_getextattr_dir(ap, attrname); 5422 return (error); 5423 } 5424 5425 /* 5426 * Vnode operation to retrieve a named extended attribute. 5427 */ 5428 static int 5429 zfs_getextattr(struct vop_getextattr_args *ap) 5430 { 5431 znode_t *zp = VTOZ(ap->a_vp); 5432 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5433 int error; 5434 5435 /* 5436 * If the xattr property is off, refuse the request. 5437 */ 5438 if (!(zfsvfs->z_flags & ZSB_XATTR)) 5439 return (SET_ERROR(EOPNOTSUPP)); 5440 5441 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 5442 ap->a_cred, ap->a_td, VREAD); 5443 if (error != 0) 5444 return (SET_ERROR(error)); 5445 5446 error = zfs_check_attrname(ap->a_name); 5447 if (error != 0) 5448 return (error); 5449 5450 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 5451 return (error); 5452 error = ENOENT; 5453 rw_enter(&zp->z_xattr_lock, RW_READER); 5454 5455 error = zfs_getextattr_impl(ap, zfs_xattr_compat); 5456 if ((error == ENOENT || error == ENOATTR) && 5457 ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { 5458 /* 5459 * Fall back to the alternate namespace format if we failed to 5460 * find a user xattr. 5461 */ 5462 error = zfs_getextattr_impl(ap, !zfs_xattr_compat); 5463 } 5464 5465 rw_exit(&zp->z_xattr_lock); 5466 zfs_exit(zfsvfs, FTAG); 5467 if (error == ENOENT) 5468 error = SET_ERROR(ENOATTR); 5469 return (error); 5470 } 5471 5472 #ifndef _SYS_SYSPROTO_H_ 5473 struct vop_deleteextattr { 5474 IN struct vnode *a_vp; 5475 IN int a_attrnamespace; 5476 IN const char *a_name; 5477 IN struct ucred *a_cred; 5478 IN struct thread *a_td; 5479 }; 5480 #endif 5481 5482 static int 5483 zfs_deleteextattr_dir(struct vop_deleteextattr_args *ap, const char *attrname) 5484 { 5485 struct nameidata nd; 5486 vnode_t *xvp = NULL, *vp; 5487 int error; 5488 5489 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, 5490 LOOKUP_XATTR, B_FALSE); 5491 if (error != 0) 5492 return (error); 5493 5494 #if __FreeBSD_version < 1400043 5495 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 5496 UIO_SYSSPACE, attrname, xvp, ap->a_td); 5497 #else 5498 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 5499 UIO_SYSSPACE, attrname, xvp); 5500 #endif 5501 error = namei(&nd); 5502 if (error != 0) 5503 return (SET_ERROR(error)); 5504 5505 vp = nd.ni_vp; 5506 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 5507 NDFREE_PNBUF(&nd); 5508 5509 vput(nd.ni_dvp); 5510 if (vp == nd.ni_dvp) 5511 vrele(vp); 5512 else 5513 vput(vp); 5514 5515 return (error); 5516 } 5517 5518 static int 5519 zfs_deleteextattr_sa(struct vop_deleteextattr_args *ap, const char *attrname) 5520 { 5521 znode_t *zp = VTOZ(ap->a_vp); 5522 nvlist_t *nvl; 5523 int error; 5524 5525 error = zfs_ensure_xattr_cached(zp); 5526 if (error != 0) 5527 return (error); 5528 5529 ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock)); 5530 ASSERT3P(zp->z_xattr_cached, !=, NULL); 5531 5532 nvl = zp->z_xattr_cached; 5533 error = nvlist_remove(nvl, attrname, DATA_TYPE_BYTE_ARRAY); 5534 if (error != 0) 5535 error = SET_ERROR(error); 5536 else 5537 error = zfs_sa_set_xattr(zp, attrname, NULL, 0); 5538 if (error != 0) { 5539 zp->z_xattr_cached = NULL; 5540 nvlist_free(nvl); 5541 } 5542 return (error); 5543 } 5544 5545 static int 5546 zfs_deleteextattr_impl(struct vop_deleteextattr_args *ap, boolean_t compat) 5547 { 5548 znode_t *zp = VTOZ(ap->a_vp); 5549 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5550 char attrname[EXTATTR_MAXNAMELEN+1]; 5551 int error; 5552 5553 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 5554 sizeof (attrname), compat); 5555 if (error != 0) 5556 return (error); 5557 5558 error = ENOENT; 5559 if (zfsvfs->z_use_sa && zp->z_is_sa) 5560 error = zfs_deleteextattr_sa(ap, attrname); 5561 if (error == ENOENT) 5562 error = zfs_deleteextattr_dir(ap, attrname); 5563 return (error); 5564 } 5565 5566 /* 5567 * Vnode operation to remove a named attribute. 5568 */ 5569 static int 5570 zfs_deleteextattr(struct vop_deleteextattr_args *ap) 5571 { 5572 znode_t *zp = VTOZ(ap->a_vp); 5573 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5574 int error; 5575 5576 /* 5577 * If the xattr property is off, refuse the request. 5578 */ 5579 if (!(zfsvfs->z_flags & ZSB_XATTR)) 5580 return (SET_ERROR(EOPNOTSUPP)); 5581 5582 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 5583 ap->a_cred, ap->a_td, VWRITE); 5584 if (error != 0) 5585 return (SET_ERROR(error)); 5586 5587 error = zfs_check_attrname(ap->a_name); 5588 if (error != 0) 5589 return (error); 5590 5591 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 5592 return (error); 5593 rw_enter(&zp->z_xattr_lock, RW_WRITER); 5594 5595 error = zfs_deleteextattr_impl(ap, zfs_xattr_compat); 5596 if ((error == ENOENT || error == ENOATTR) && 5597 ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { 5598 /* 5599 * Fall back to the alternate namespace format if we failed to 5600 * find a user xattr. 5601 */ 5602 error = zfs_deleteextattr_impl(ap, !zfs_xattr_compat); 5603 } 5604 5605 rw_exit(&zp->z_xattr_lock); 5606 zfs_exit(zfsvfs, FTAG); 5607 if (error == ENOENT) 5608 error = SET_ERROR(ENOATTR); 5609 return (error); 5610 } 5611 5612 #ifndef _SYS_SYSPROTO_H_ 5613 struct vop_setextattr { 5614 IN struct vnode *a_vp; 5615 IN int a_attrnamespace; 5616 IN const char *a_name; 5617 INOUT struct uio *a_uio; 5618 IN struct ucred *a_cred; 5619 IN struct thread *a_td; 5620 }; 5621 #endif 5622 5623 static int 5624 zfs_setextattr_dir(struct vop_setextattr_args *ap, const char *attrname) 5625 { 5626 struct thread *td = ap->a_td; 5627 struct nameidata nd; 5628 struct vattr va; 5629 vnode_t *xvp = NULL, *vp; 5630 int error, flags; 5631 5632 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, 5633 LOOKUP_XATTR | CREATE_XATTR_DIR, B_FALSE); 5634 if (error != 0) 5635 return (error); 5636 5637 flags = FFLAGS(O_WRONLY | O_CREAT); 5638 #if __FreeBSD_version < 1400043 5639 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp, td); 5640 #else 5641 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp); 5642 #endif 5643 error = vn_open_cred(&nd, &flags, 0600, VN_OPEN_INVFS, ap->a_cred, 5644 NULL); 5645 if (error != 0) 5646 return (SET_ERROR(error)); 5647 vp = nd.ni_vp; 5648 NDFREE_PNBUF(&nd); 5649 5650 VATTR_NULL(&va); 5651 va.va_size = 0; 5652 error = VOP_SETATTR(vp, &va, ap->a_cred); 5653 if (error == 0) 5654 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred); 5655 5656 VOP_UNLOCK1(vp); 5657 vn_close(vp, flags, ap->a_cred, td); 5658 return (error); 5659 } 5660 5661 static int 5662 zfs_setextattr_sa(struct vop_setextattr_args *ap, const char *attrname) 5663 { 5664 znode_t *zp = VTOZ(ap->a_vp); 5665 nvlist_t *nvl; 5666 size_t sa_size; 5667 int error; 5668 5669 error = zfs_ensure_xattr_cached(zp); 5670 if (error != 0) 5671 return (error); 5672 5673 ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock)); 5674 ASSERT3P(zp->z_xattr_cached, !=, NULL); 5675 5676 nvl = zp->z_xattr_cached; 5677 size_t entry_size = ap->a_uio->uio_resid; 5678 if (entry_size > DXATTR_MAX_ENTRY_SIZE) 5679 return (SET_ERROR(EFBIG)); 5680 error = nvlist_size(nvl, &sa_size, NV_ENCODE_XDR); 5681 if (error != 0) 5682 return (SET_ERROR(error)); 5683 if (sa_size > DXATTR_MAX_SA_SIZE) 5684 return (SET_ERROR(EFBIG)); 5685 uchar_t *buf = kmem_alloc(entry_size, KM_SLEEP); 5686 error = uiomove(buf, entry_size, ap->a_uio); 5687 if (error != 0) { 5688 error = SET_ERROR(error); 5689 } else { 5690 error = nvlist_add_byte_array(nvl, attrname, buf, entry_size); 5691 if (error != 0) 5692 error = SET_ERROR(error); 5693 } 5694 if (error == 0) 5695 error = zfs_sa_set_xattr(zp, attrname, buf, entry_size); 5696 kmem_free(buf, entry_size); 5697 if (error != 0) { 5698 zp->z_xattr_cached = NULL; 5699 nvlist_free(nvl); 5700 } 5701 return (error); 5702 } 5703 5704 static int 5705 zfs_setextattr_impl(struct vop_setextattr_args *ap, boolean_t compat) 5706 { 5707 znode_t *zp = VTOZ(ap->a_vp); 5708 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5709 char attrname[EXTATTR_MAXNAMELEN+1]; 5710 int error; 5711 5712 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 5713 sizeof (attrname), compat); 5714 if (error != 0) 5715 return (error); 5716 5717 struct vop_deleteextattr_args vda = { 5718 .a_vp = ap->a_vp, 5719 .a_attrnamespace = ap->a_attrnamespace, 5720 .a_name = ap->a_name, 5721 .a_cred = ap->a_cred, 5722 .a_td = ap->a_td, 5723 }; 5724 error = ENOENT; 5725 if (zfsvfs->z_use_sa && zp->z_is_sa && zfsvfs->z_xattr_sa) { 5726 error = zfs_setextattr_sa(ap, attrname); 5727 if (error == 0) { 5728 /* 5729 * Successfully put into SA, we need to clear the one 5730 * in dir if present. 5731 */ 5732 zfs_deleteextattr_dir(&vda, attrname); 5733 } 5734 } 5735 if (error != 0) { 5736 error = zfs_setextattr_dir(ap, attrname); 5737 if (error == 0 && zp->z_is_sa) { 5738 /* 5739 * Successfully put into dir, we need to clear the one 5740 * in SA if present. 5741 */ 5742 zfs_deleteextattr_sa(&vda, attrname); 5743 } 5744 } 5745 if (error == 0 && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { 5746 /* 5747 * Also clear all versions of the alternate compat name. 5748 */ 5749 zfs_deleteextattr_impl(&vda, !compat); 5750 } 5751 return (error); 5752 } 5753 5754 /* 5755 * Vnode operation to set a named attribute. 5756 */ 5757 static int 5758 zfs_setextattr(struct vop_setextattr_args *ap) 5759 { 5760 znode_t *zp = VTOZ(ap->a_vp); 5761 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5762 int error; 5763 5764 /* 5765 * If the xattr property is off, refuse the request. 5766 */ 5767 if (!(zfsvfs->z_flags & ZSB_XATTR)) 5768 return (SET_ERROR(EOPNOTSUPP)); 5769 5770 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 5771 ap->a_cred, ap->a_td, VWRITE); 5772 if (error != 0) 5773 return (SET_ERROR(error)); 5774 5775 error = zfs_check_attrname(ap->a_name); 5776 if (error != 0) 5777 return (error); 5778 5779 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 5780 return (error); 5781 rw_enter(&zp->z_xattr_lock, RW_WRITER); 5782 5783 error = zfs_setextattr_impl(ap, zfs_xattr_compat); 5784 5785 rw_exit(&zp->z_xattr_lock); 5786 zfs_exit(zfsvfs, FTAG); 5787 return (error); 5788 } 5789 5790 #ifndef _SYS_SYSPROTO_H_ 5791 struct vop_listextattr { 5792 IN struct vnode *a_vp; 5793 IN int a_attrnamespace; 5794 INOUT struct uio *a_uio; 5795 OUT size_t *a_size; 5796 IN struct ucred *a_cred; 5797 IN struct thread *a_td; 5798 }; 5799 #endif 5800 5801 static int 5802 zfs_listextattr_dir(struct vop_listextattr_args *ap, const char *attrprefix) 5803 { 5804 struct thread *td = ap->a_td; 5805 struct nameidata nd; 5806 uint8_t dirbuf[sizeof (struct dirent)]; 5807 struct iovec aiov; 5808 struct uio auio; 5809 vnode_t *xvp = NULL, *vp; 5810 int error, eof; 5811 5812 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, 5813 LOOKUP_XATTR, B_FALSE); 5814 if (error != 0) { 5815 /* 5816 * ENOATTR means that the EA directory does not yet exist, 5817 * i.e. there are no extended attributes there. 5818 */ 5819 if (error == ENOATTR) 5820 error = 0; 5821 return (error); 5822 } 5823 5824 #if __FreeBSD_version < 1400043 5825 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 5826 UIO_SYSSPACE, ".", xvp, td); 5827 #else 5828 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 5829 UIO_SYSSPACE, ".", xvp); 5830 #endif 5831 error = namei(&nd); 5832 if (error != 0) 5833 return (SET_ERROR(error)); 5834 vp = nd.ni_vp; 5835 NDFREE_PNBUF(&nd); 5836 5837 auio.uio_iov = &aiov; 5838 auio.uio_iovcnt = 1; 5839 auio.uio_segflg = UIO_SYSSPACE; 5840 auio.uio_td = td; 5841 auio.uio_rw = UIO_READ; 5842 auio.uio_offset = 0; 5843 5844 size_t plen = strlen(attrprefix); 5845 5846 do { 5847 aiov.iov_base = (void *)dirbuf; 5848 aiov.iov_len = sizeof (dirbuf); 5849 auio.uio_resid = sizeof (dirbuf); 5850 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 5851 if (error != 0) 5852 break; 5853 int done = sizeof (dirbuf) - auio.uio_resid; 5854 for (int pos = 0; pos < done; ) { 5855 struct dirent *dp = (struct dirent *)(dirbuf + pos); 5856 pos += dp->d_reclen; 5857 /* 5858 * XXX: Temporarily we also accept DT_UNKNOWN, as this 5859 * is what we get when attribute was created on Solaris. 5860 */ 5861 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 5862 continue; 5863 else if (plen == 0 && 5864 ZFS_XA_NS_PREFIX_FORBIDDEN(dp->d_name)) 5865 continue; 5866 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 5867 continue; 5868 uint8_t nlen = dp->d_namlen - plen; 5869 if (ap->a_size != NULL) { 5870 *ap->a_size += 1 + nlen; 5871 } else if (ap->a_uio != NULL) { 5872 /* 5873 * Format of extattr name entry is one byte for 5874 * length and the rest for name. 5875 */ 5876 error = uiomove(&nlen, 1, ap->a_uio); 5877 if (error == 0) { 5878 char *namep = dp->d_name + plen; 5879 error = uiomove(namep, nlen, ap->a_uio); 5880 } 5881 if (error != 0) { 5882 error = SET_ERROR(error); 5883 break; 5884 } 5885 } 5886 } 5887 } while (!eof && error == 0); 5888 5889 vput(vp); 5890 return (error); 5891 } 5892 5893 static int 5894 zfs_listextattr_sa(struct vop_listextattr_args *ap, const char *attrprefix) 5895 { 5896 znode_t *zp = VTOZ(ap->a_vp); 5897 int error; 5898 5899 error = zfs_ensure_xattr_cached(zp); 5900 if (error != 0) 5901 return (error); 5902 5903 ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock)); 5904 ASSERT3P(zp->z_xattr_cached, !=, NULL); 5905 5906 size_t plen = strlen(attrprefix); 5907 nvpair_t *nvp = NULL; 5908 while ((nvp = nvlist_next_nvpair(zp->z_xattr_cached, nvp)) != NULL) { 5909 ASSERT3U(nvpair_type(nvp), ==, DATA_TYPE_BYTE_ARRAY); 5910 5911 const char *name = nvpair_name(nvp); 5912 if (plen == 0 && ZFS_XA_NS_PREFIX_FORBIDDEN(name)) 5913 continue; 5914 else if (strncmp(name, attrprefix, plen) != 0) 5915 continue; 5916 uint8_t nlen = strlen(name) - plen; 5917 if (ap->a_size != NULL) { 5918 *ap->a_size += 1 + nlen; 5919 } else if (ap->a_uio != NULL) { 5920 /* 5921 * Format of extattr name entry is one byte for 5922 * length and the rest for name. 5923 */ 5924 error = uiomove(&nlen, 1, ap->a_uio); 5925 if (error == 0) { 5926 char *namep = __DECONST(char *, name) + plen; 5927 error = uiomove(namep, nlen, ap->a_uio); 5928 } 5929 if (error != 0) { 5930 error = SET_ERROR(error); 5931 break; 5932 } 5933 } 5934 } 5935 5936 return (error); 5937 } 5938 5939 static int 5940 zfs_listextattr_impl(struct vop_listextattr_args *ap, boolean_t compat) 5941 { 5942 znode_t *zp = VTOZ(ap->a_vp); 5943 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5944 char attrprefix[16]; 5945 int error; 5946 5947 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 5948 sizeof (attrprefix), compat); 5949 if (error != 0) 5950 return (error); 5951 5952 if (zfsvfs->z_use_sa && zp->z_is_sa) 5953 error = zfs_listextattr_sa(ap, attrprefix); 5954 if (error == 0) 5955 error = zfs_listextattr_dir(ap, attrprefix); 5956 return (error); 5957 } 5958 5959 /* 5960 * Vnode operation to retrieve extended attributes on a vnode. 5961 */ 5962 static int 5963 zfs_listextattr(struct vop_listextattr_args *ap) 5964 { 5965 znode_t *zp = VTOZ(ap->a_vp); 5966 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5967 int error; 5968 5969 if (ap->a_size != NULL) 5970 *ap->a_size = 0; 5971 5972 /* 5973 * If the xattr property is off, refuse the request. 5974 */ 5975 if (!(zfsvfs->z_flags & ZSB_XATTR)) 5976 return (SET_ERROR(EOPNOTSUPP)); 5977 5978 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 5979 ap->a_cred, ap->a_td, VREAD); 5980 if (error != 0) 5981 return (SET_ERROR(error)); 5982 5983 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 5984 return (error); 5985 rw_enter(&zp->z_xattr_lock, RW_READER); 5986 5987 error = zfs_listextattr_impl(ap, zfs_xattr_compat); 5988 if (error == 0 && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { 5989 /* Also list user xattrs with the alternate format. */ 5990 error = zfs_listextattr_impl(ap, !zfs_xattr_compat); 5991 } 5992 5993 rw_exit(&zp->z_xattr_lock); 5994 zfs_exit(zfsvfs, FTAG); 5995 return (error); 5996 } 5997 5998 #ifndef _SYS_SYSPROTO_H_ 5999 struct vop_getacl_args { 6000 struct vnode *vp; 6001 acl_type_t type; 6002 struct acl *aclp; 6003 struct ucred *cred; 6004 struct thread *td; 6005 }; 6006 #endif 6007 6008 static int 6009 zfs_freebsd_getacl(struct vop_getacl_args *ap) 6010 { 6011 int error; 6012 vsecattr_t vsecattr; 6013 6014 if (ap->a_type != ACL_TYPE_NFS4) 6015 return (EINVAL); 6016 6017 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 6018 if ((error = zfs_getsecattr(VTOZ(ap->a_vp), 6019 &vsecattr, 0, ap->a_cred))) 6020 return (error); 6021 6022 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, 6023 vsecattr.vsa_aclcnt); 6024 if (vsecattr.vsa_aclentp != NULL) 6025 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 6026 6027 return (error); 6028 } 6029 6030 #ifndef _SYS_SYSPROTO_H_ 6031 struct vop_setacl_args { 6032 struct vnode *vp; 6033 acl_type_t type; 6034 struct acl *aclp; 6035 struct ucred *cred; 6036 struct thread *td; 6037 }; 6038 #endif 6039 6040 static int 6041 zfs_freebsd_setacl(struct vop_setacl_args *ap) 6042 { 6043 int error; 6044 vsecattr_t vsecattr; 6045 int aclbsize; /* size of acl list in bytes */ 6046 aclent_t *aaclp; 6047 6048 if (ap->a_type != ACL_TYPE_NFS4) 6049 return (EINVAL); 6050 6051 if (ap->a_aclp == NULL) 6052 return (EINVAL); 6053 6054 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 6055 return (EINVAL); 6056 6057 /* 6058 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 6059 * splitting every entry into two and appending "canonical six" 6060 * entries at the end. Don't allow for setting an ACL that would 6061 * cause chmod(2) to run out of ACL entries. 6062 */ 6063 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 6064 return (ENOSPC); 6065 6066 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 6067 if (error != 0) 6068 return (error); 6069 6070 vsecattr.vsa_mask = VSA_ACE; 6071 aclbsize = ap->a_aclp->acl_cnt * sizeof (ace_t); 6072 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 6073 aaclp = vsecattr.vsa_aclentp; 6074 vsecattr.vsa_aclentsz = aclbsize; 6075 6076 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 6077 error = zfs_setsecattr(VTOZ(ap->a_vp), &vsecattr, 0, ap->a_cred); 6078 kmem_free(aaclp, aclbsize); 6079 6080 return (error); 6081 } 6082 6083 #ifndef _SYS_SYSPROTO_H_ 6084 struct vop_aclcheck_args { 6085 struct vnode *vp; 6086 acl_type_t type; 6087 struct acl *aclp; 6088 struct ucred *cred; 6089 struct thread *td; 6090 }; 6091 #endif 6092 6093 static int 6094 zfs_freebsd_aclcheck(struct vop_aclcheck_args *ap) 6095 { 6096 6097 return (EOPNOTSUPP); 6098 } 6099 6100 static int 6101 zfs_vptocnp(struct vop_vptocnp_args *ap) 6102 { 6103 vnode_t *covered_vp; 6104 vnode_t *vp = ap->a_vp; 6105 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 6106 znode_t *zp = VTOZ(vp); 6107 int ltype; 6108 int error; 6109 6110 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 6111 return (error); 6112 6113 /* 6114 * If we are a snapshot mounted under .zfs, run the operation 6115 * on the covered vnode. 6116 */ 6117 if (zp->z_id != zfsvfs->z_root || zfsvfs->z_parent == zfsvfs) { 6118 char name[MAXNAMLEN + 1]; 6119 znode_t *dzp; 6120 size_t len; 6121 6122 error = zfs_znode_parent_and_name(zp, &dzp, name); 6123 if (error == 0) { 6124 len = strlen(name); 6125 if (*ap->a_buflen < len) 6126 error = SET_ERROR(ENOMEM); 6127 } 6128 if (error == 0) { 6129 *ap->a_buflen -= len; 6130 memcpy(ap->a_buf + *ap->a_buflen, name, len); 6131 *ap->a_vpp = ZTOV(dzp); 6132 } 6133 zfs_exit(zfsvfs, FTAG); 6134 return (error); 6135 } 6136 zfs_exit(zfsvfs, FTAG); 6137 6138 covered_vp = vp->v_mount->mnt_vnodecovered; 6139 #if __FreeBSD_version >= 1300045 6140 enum vgetstate vs = vget_prep(covered_vp); 6141 #else 6142 vhold(covered_vp); 6143 #endif 6144 ltype = VOP_ISLOCKED(vp); 6145 VOP_UNLOCK1(vp); 6146 #if __FreeBSD_version >= 1300045 6147 error = vget_finish(covered_vp, LK_SHARED, vs); 6148 #else 6149 error = vget(covered_vp, LK_SHARED | LK_VNHELD, curthread); 6150 #endif 6151 if (error == 0) { 6152 #if __FreeBSD_version >= 1300123 6153 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_buf, 6154 ap->a_buflen); 6155 #else 6156 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_cred, 6157 ap->a_buf, ap->a_buflen); 6158 #endif 6159 vput(covered_vp); 6160 } 6161 vn_lock(vp, ltype | LK_RETRY); 6162 if (VN_IS_DOOMED(vp)) 6163 error = SET_ERROR(ENOENT); 6164 return (error); 6165 } 6166 6167 #if __FreeBSD_version >= 1400032 6168 static int 6169 zfs_deallocate(struct vop_deallocate_args *ap) 6170 { 6171 znode_t *zp = VTOZ(ap->a_vp); 6172 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6173 zilog_t *zilog; 6174 off_t off, len, file_sz; 6175 int error; 6176 6177 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 6178 return (error); 6179 6180 /* 6181 * Callers might not be able to detect properly that we are read-only, 6182 * so check it explicitly here. 6183 */ 6184 if (zfs_is_readonly(zfsvfs)) { 6185 zfs_exit(zfsvfs, FTAG); 6186 return (SET_ERROR(EROFS)); 6187 } 6188 6189 zilog = zfsvfs->z_log; 6190 off = *ap->a_offset; 6191 len = *ap->a_len; 6192 file_sz = zp->z_size; 6193 if (off + len > file_sz) 6194 len = file_sz - off; 6195 /* Fast path for out-of-range request. */ 6196 if (len <= 0) { 6197 *ap->a_len = 0; 6198 zfs_exit(zfsvfs, FTAG); 6199 return (0); 6200 } 6201 6202 error = zfs_freesp(zp, off, len, O_RDWR, TRUE); 6203 if (error == 0) { 6204 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS || 6205 (ap->a_ioflag & IO_SYNC) != 0) 6206 zil_commit(zilog, zp->z_id); 6207 *ap->a_offset = off + len; 6208 *ap->a_len = 0; 6209 } 6210 6211 zfs_exit(zfsvfs, FTAG); 6212 return (error); 6213 } 6214 #endif 6215 6216 #ifndef _SYS_SYSPROTO_H_ 6217 struct vop_copy_file_range_args { 6218 struct vnode *a_invp; 6219 off_t *a_inoffp; 6220 struct vnode *a_outvp; 6221 off_t *a_outoffp; 6222 size_t *a_lenp; 6223 unsigned int a_flags; 6224 struct ucred *a_incred; 6225 struct ucred *a_outcred; 6226 struct thread *a_fsizetd; 6227 } 6228 #endif 6229 /* 6230 * TODO: FreeBSD will only call file system-specific copy_file_range() if both 6231 * files resides under the same mountpoint. In case of ZFS we want to be called 6232 * even is files are in different datasets (but on the same pools, but we need 6233 * to check that ourselves). 6234 */ 6235 static int 6236 zfs_freebsd_copy_file_range(struct vop_copy_file_range_args *ap) 6237 { 6238 zfsvfs_t *outzfsvfs; 6239 struct vnode *invp = ap->a_invp; 6240 struct vnode *outvp = ap->a_outvp; 6241 struct mount *mp; 6242 struct uio io; 6243 int error; 6244 uint64_t len = *ap->a_lenp; 6245 6246 if (!zfs_bclone_enabled) { 6247 mp = NULL; 6248 goto bad_write_fallback; 6249 } 6250 6251 /* 6252 * TODO: If offset/length is not aligned to recordsize, use 6253 * vn_generic_copy_file_range() on this fragment. 6254 * It would be better to do this after we lock the vnodes, but then we 6255 * need something else than vn_generic_copy_file_range(). 6256 */ 6257 6258 vn_start_write(outvp, &mp, V_WAIT); 6259 if (__predict_true(mp == outvp->v_mount)) { 6260 outzfsvfs = (zfsvfs_t *)mp->mnt_data; 6261 if (!spa_feature_is_enabled(dmu_objset_spa(outzfsvfs->z_os), 6262 SPA_FEATURE_BLOCK_CLONING)) { 6263 goto bad_write_fallback; 6264 } 6265 } 6266 if (invp == outvp) { 6267 if (vn_lock(outvp, LK_EXCLUSIVE) != 0) { 6268 goto bad_write_fallback; 6269 } 6270 } else { 6271 #if __FreeBSD_version >= 1400086 6272 vn_lock_pair(invp, false, LK_EXCLUSIVE, outvp, false, 6273 LK_EXCLUSIVE); 6274 #else 6275 vn_lock_pair(invp, false, outvp, false); 6276 #endif 6277 if (VN_IS_DOOMED(invp) || VN_IS_DOOMED(outvp)) { 6278 goto bad_locked_fallback; 6279 } 6280 } 6281 6282 #ifdef MAC 6283 error = mac_vnode_check_write(curthread->td_ucred, ap->a_outcred, 6284 outvp); 6285 if (error != 0) 6286 goto out_locked; 6287 #endif 6288 6289 io.uio_offset = *ap->a_outoffp; 6290 io.uio_resid = *ap->a_lenp; 6291 error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd); 6292 if (error != 0) 6293 goto out_locked; 6294 6295 error = zfs_clone_range(VTOZ(invp), ap->a_inoffp, VTOZ(outvp), 6296 ap->a_outoffp, &len, ap->a_outcred); 6297 if (error == EXDEV) 6298 goto bad_locked_fallback; 6299 *ap->a_lenp = (size_t)len; 6300 out_locked: 6301 if (invp != outvp) 6302 VOP_UNLOCK(invp); 6303 VOP_UNLOCK(outvp); 6304 if (mp != NULL) 6305 vn_finished_write(mp); 6306 return (error); 6307 6308 bad_locked_fallback: 6309 if (invp != outvp) 6310 VOP_UNLOCK(invp); 6311 VOP_UNLOCK(outvp); 6312 bad_write_fallback: 6313 if (mp != NULL) 6314 vn_finished_write(mp); 6315 error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp, 6316 ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags, 6317 ap->a_incred, ap->a_outcred, ap->a_fsizetd); 6318 return (error); 6319 } 6320 6321 struct vop_vector zfs_vnodeops; 6322 struct vop_vector zfs_fifoops; 6323 struct vop_vector zfs_shareops; 6324 6325 struct vop_vector zfs_vnodeops = { 6326 .vop_default = &default_vnodeops, 6327 .vop_inactive = zfs_freebsd_inactive, 6328 #if __FreeBSD_version >= 1300042 6329 .vop_need_inactive = zfs_freebsd_need_inactive, 6330 #endif 6331 .vop_reclaim = zfs_freebsd_reclaim, 6332 #if __FreeBSD_version >= 1300102 6333 .vop_fplookup_vexec = zfs_freebsd_fplookup_vexec, 6334 #endif 6335 #if __FreeBSD_version >= 1300139 6336 .vop_fplookup_symlink = zfs_freebsd_fplookup_symlink, 6337 #endif 6338 .vop_access = zfs_freebsd_access, 6339 .vop_allocate = VOP_EINVAL, 6340 #if __FreeBSD_version >= 1400032 6341 .vop_deallocate = zfs_deallocate, 6342 #endif 6343 .vop_lookup = zfs_cache_lookup, 6344 .vop_cachedlookup = zfs_freebsd_cachedlookup, 6345 .vop_getattr = zfs_freebsd_getattr, 6346 .vop_setattr = zfs_freebsd_setattr, 6347 .vop_create = zfs_freebsd_create, 6348 .vop_mknod = (vop_mknod_t *)zfs_freebsd_create, 6349 .vop_mkdir = zfs_freebsd_mkdir, 6350 .vop_readdir = zfs_freebsd_readdir, 6351 .vop_fsync = zfs_freebsd_fsync, 6352 .vop_open = zfs_freebsd_open, 6353 .vop_close = zfs_freebsd_close, 6354 .vop_rmdir = zfs_freebsd_rmdir, 6355 .vop_ioctl = zfs_freebsd_ioctl, 6356 .vop_link = zfs_freebsd_link, 6357 .vop_symlink = zfs_freebsd_symlink, 6358 .vop_readlink = zfs_freebsd_readlink, 6359 .vop_read = zfs_freebsd_read, 6360 .vop_write = zfs_freebsd_write, 6361 .vop_remove = zfs_freebsd_remove, 6362 .vop_rename = zfs_freebsd_rename, 6363 .vop_pathconf = zfs_freebsd_pathconf, 6364 .vop_bmap = zfs_freebsd_bmap, 6365 .vop_fid = zfs_freebsd_fid, 6366 .vop_getextattr = zfs_getextattr, 6367 .vop_deleteextattr = zfs_deleteextattr, 6368 .vop_setextattr = zfs_setextattr, 6369 .vop_listextattr = zfs_listextattr, 6370 .vop_getacl = zfs_freebsd_getacl, 6371 .vop_setacl = zfs_freebsd_setacl, 6372 .vop_aclcheck = zfs_freebsd_aclcheck, 6373 .vop_getpages = zfs_freebsd_getpages, 6374 .vop_putpages = zfs_freebsd_putpages, 6375 .vop_vptocnp = zfs_vptocnp, 6376 #if __FreeBSD_version >= 1300064 6377 .vop_lock1 = vop_lock, 6378 .vop_unlock = vop_unlock, 6379 .vop_islocked = vop_islocked, 6380 #endif 6381 #if __FreeBSD_version >= 1400043 6382 .vop_add_writecount = vop_stdadd_writecount_nomsync, 6383 #endif 6384 .vop_copy_file_range = zfs_freebsd_copy_file_range, 6385 }; 6386 VFS_VOP_VECTOR_REGISTER(zfs_vnodeops); 6387 6388 struct vop_vector zfs_fifoops = { 6389 .vop_default = &fifo_specops, 6390 .vop_fsync = zfs_freebsd_fsync, 6391 #if __FreeBSD_version >= 1300102 6392 .vop_fplookup_vexec = zfs_freebsd_fplookup_vexec, 6393 #endif 6394 #if __FreeBSD_version >= 1300139 6395 .vop_fplookup_symlink = zfs_freebsd_fplookup_symlink, 6396 #endif 6397 .vop_access = zfs_freebsd_access, 6398 .vop_getattr = zfs_freebsd_getattr, 6399 .vop_inactive = zfs_freebsd_inactive, 6400 .vop_read = VOP_PANIC, 6401 .vop_reclaim = zfs_freebsd_reclaim, 6402 .vop_setattr = zfs_freebsd_setattr, 6403 .vop_write = VOP_PANIC, 6404 .vop_pathconf = zfs_freebsd_pathconf, 6405 .vop_fid = zfs_freebsd_fid, 6406 .vop_getacl = zfs_freebsd_getacl, 6407 .vop_setacl = zfs_freebsd_setacl, 6408 .vop_aclcheck = zfs_freebsd_aclcheck, 6409 #if __FreeBSD_version >= 1400043 6410 .vop_add_writecount = vop_stdadd_writecount_nomsync, 6411 #endif 6412 }; 6413 VFS_VOP_VECTOR_REGISTER(zfs_fifoops); 6414 6415 /* 6416 * special share hidden files vnode operations template 6417 */ 6418 struct vop_vector zfs_shareops = { 6419 .vop_default = &default_vnodeops, 6420 #if __FreeBSD_version >= 1300121 6421 .vop_fplookup_vexec = VOP_EAGAIN, 6422 #endif 6423 #if __FreeBSD_version >= 1300139 6424 .vop_fplookup_symlink = VOP_EAGAIN, 6425 #endif 6426 .vop_access = zfs_freebsd_access, 6427 .vop_inactive = zfs_freebsd_inactive, 6428 .vop_reclaim = zfs_freebsd_reclaim, 6429 .vop_fid = zfs_freebsd_fid, 6430 .vop_pathconf = zfs_freebsd_pathconf, 6431 #if __FreeBSD_version >= 1400043 6432 .vop_add_writecount = vop_stdadd_writecount_nomsync, 6433 #endif 6434 }; 6435 VFS_VOP_VECTOR_REGISTER(zfs_shareops); 6436 6437 ZFS_MODULE_PARAM(zfs, zfs_, xattr_compat, INT, ZMOD_RW, 6438 "Use legacy ZFS xattr naming for writing new user namespace xattrs"); 6439