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 #include <sys/smr.h> 43 #include <sys/dirent.h> 44 #include <sys/file.h> 45 #include <sys/stat.h> 46 #include <sys/kmem.h> 47 #include <sys/taskq.h> 48 #include <sys/uio.h> 49 #include <sys/atomic.h> 50 #include <sys/namei.h> 51 #include <sys/mman.h> 52 #include <sys/cmn_err.h> 53 #include <sys/kdb.h> 54 #include <sys/sysproto.h> 55 #include <sys/errno.h> 56 #include <sys/unistd.h> 57 #include <sys/zfs_dir.h> 58 #include <sys/zfs_ioctl.h> 59 #include <sys/fs/zfs.h> 60 #include <sys/dmu.h> 61 #include <sys/dmu_objset.h> 62 #include <sys/spa.h> 63 #include <sys/txg.h> 64 #include <sys/dbuf.h> 65 #include <sys/zap.h> 66 #include <sys/sa.h> 67 #include <sys/policy.h> 68 #include <sys/sunddi.h> 69 #include <sys/filio.h> 70 #include <sys/sid.h> 71 #include <sys/zfs_ctldir.h> 72 #include <sys/zfs_fuid.h> 73 #include <sys/zfs_quota.h> 74 #include <sys/zfs_sa.h> 75 #include <sys/zfs_rlock.h> 76 #include <sys/bio.h> 77 #include <sys/buf.h> 78 #include <sys/sched.h> 79 #include <sys/acl.h> 80 #include <sys/vmmeter.h> 81 #include <vm/vm_param.h> 82 #include <sys/zil.h> 83 #include <sys/zfs_vnops.h> 84 #include <sys/module.h> 85 #include <sys/sysent.h> 86 #include <sys/dmu_impl.h> 87 #include <sys/brt.h> 88 #include <sys/zfeature.h> 89 90 #include <vm/vm_object.h> 91 92 #include <sys/extattr.h> 93 #include <sys/priv.h> 94 95 #ifndef VN_OPEN_INVFS 96 #define VN_OPEN_INVFS 0x0 97 #endif 98 99 VFS_SMR_DECLARE; 100 101 #ifdef DEBUG_VFS_LOCKS 102 #define VNCHECKREF(vp) \ 103 VNASSERT((vp)->v_holdcnt > 0 && (vp)->v_usecount > 0, vp, \ 104 ("%s: wrong ref counts", __func__)); 105 #else 106 #define VNCHECKREF(vp) 107 #endif 108 109 #if __FreeBSD_version >= 1400045 110 typedef uint64_t cookie_t; 111 #else 112 typedef ulong_t cookie_t; 113 #endif 114 115 /* 116 * Programming rules. 117 * 118 * Each vnode op performs some logical unit of work. To do this, the ZPL must 119 * properly lock its in-core state, create a DMU transaction, do the work, 120 * record this work in the intent log (ZIL), commit the DMU transaction, 121 * and wait for the intent log to commit if it is a synchronous operation. 122 * Moreover, the vnode ops must work in both normal and log replay context. 123 * The ordering of events is important to avoid deadlocks and references 124 * to freed memory. The example below illustrates the following Big Rules: 125 * 126 * (1) A check must be made in each zfs thread for a mounted file system. 127 * This is done avoiding races using zfs_enter(zfsvfs). 128 * A zfs_exit(zfsvfs) is needed before all returns. Any znodes 129 * must be checked with zfs_verify_zp(zp). Both of these macros 130 * can return EIO from the calling function. 131 * 132 * (2) VN_RELE() should always be the last thing except for zil_commit() 133 * (if necessary) and zfs_exit(). This is for 3 reasons: 134 * First, if it's the last reference, the vnode/znode 135 * can be freed, so the zp may point to freed memory. Second, the last 136 * reference will call zfs_zinactive(), which may induce a lot of work -- 137 * pushing cached pages (which acquires range locks) and syncing out 138 * cached atime changes. Third, zfs_zinactive() may require a new tx, 139 * which could deadlock the system if you were already holding one. 140 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). 141 * 142 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 143 * as they can span dmu_tx_assign() calls. 144 * 145 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to 146 * dmu_tx_assign(). This is critical because we don't want to block 147 * while holding locks. 148 * 149 * If no ZPL locks are held (aside from zfs_enter()), use TXG_WAIT. This 150 * reduces lock contention and CPU usage when we must wait (note that if 151 * throughput is constrained by the storage, nearly every transaction 152 * must wait). 153 * 154 * Note, in particular, that if a lock is sometimes acquired before 155 * the tx assigns, and sometimes after (e.g. z_lock), then failing 156 * to use a non-blocking assign can deadlock the system. The scenario: 157 * 158 * Thread A has grabbed a lock before calling dmu_tx_assign(). 159 * Thread B is in an already-assigned tx, and blocks for this lock. 160 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 161 * forever, because the previous txg can't quiesce until B's tx commits. 162 * 163 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 164 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent 165 * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT, 166 * to indicate that this operation has already called dmu_tx_wait(). 167 * This will ensure that we don't retry forever, waiting a short bit 168 * each time. 169 * 170 * (5) If the operation succeeded, generate the intent log entry for it 171 * before dropping locks. This ensures that the ordering of events 172 * in the intent log matches the order in which they actually occurred. 173 * During ZIL replay the zfs_log_* functions will update the sequence 174 * number to indicate the zil transaction has replayed. 175 * 176 * (6) At the end of each vnode op, the DMU tx must always commit, 177 * regardless of whether there were any errors. 178 * 179 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 180 * to ensure that synchronous semantics are provided when necessary. 181 * 182 * In general, this is how things should be ordered in each vnode op: 183 * 184 * zfs_enter(zfsvfs); // exit if unmounted 185 * top: 186 * zfs_dirent_lookup(&dl, ...) // lock directory entry (may VN_HOLD()) 187 * rw_enter(...); // grab any other locks you need 188 * tx = dmu_tx_create(...); // get DMU tx 189 * dmu_tx_hold_*(); // hold each object you might modify 190 * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 191 * if (error) { 192 * rw_exit(...); // drop locks 193 * zfs_dirent_unlock(dl); // unlock directory entry 194 * VN_RELE(...); // release held vnodes 195 * if (error == ERESTART) { 196 * waited = B_TRUE; 197 * dmu_tx_wait(tx); 198 * dmu_tx_abort(tx); 199 * goto top; 200 * } 201 * dmu_tx_abort(tx); // abort DMU tx 202 * zfs_exit(zfsvfs); // finished in zfs 203 * return (error); // really out of space 204 * } 205 * error = do_real_work(); // do whatever this VOP does 206 * if (error == 0) 207 * zfs_log_*(...); // on success, make ZIL entry 208 * dmu_tx_commit(tx); // commit DMU tx -- error or not 209 * rw_exit(...); // drop locks 210 * zfs_dirent_unlock(dl); // unlock directory entry 211 * VN_RELE(...); // release held vnodes 212 * zil_commit(zilog, foid); // synchronous when necessary 213 * zfs_exit(zfsvfs); // finished in zfs 214 * return (error); // done, report error 215 */ 216 static int 217 zfs_open(vnode_t **vpp, int flag, cred_t *cr) 218 { 219 (void) cr; 220 znode_t *zp = VTOZ(*vpp); 221 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 222 int error; 223 224 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 225 return (error); 226 227 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 228 ((flag & FAPPEND) == 0)) { 229 zfs_exit(zfsvfs, FTAG); 230 return (SET_ERROR(EPERM)); 231 } 232 233 /* 234 * Keep a count of the synchronous opens in the znode. On first 235 * synchronous open we must convert all previous async transactions 236 * into sync to keep correct ordering. 237 */ 238 if (flag & O_SYNC) { 239 if (atomic_inc_32_nv(&zp->z_sync_cnt) == 1) 240 zil_async_to_sync(zfsvfs->z_log, zp->z_id); 241 } 242 243 zfs_exit(zfsvfs, FTAG); 244 return (0); 245 } 246 247 static int 248 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr) 249 { 250 (void) offset, (void) cr; 251 znode_t *zp = VTOZ(vp); 252 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 253 int error; 254 255 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 256 return (error); 257 258 /* Decrement the synchronous opens in the znode */ 259 if ((flag & O_SYNC) && (count == 1)) 260 atomic_dec_32(&zp->z_sync_cnt); 261 262 zfs_exit(zfsvfs, FTAG); 263 return (0); 264 } 265 266 static int 267 zfs_ioctl(vnode_t *vp, ulong_t com, intptr_t data, int flag, cred_t *cred, 268 int *rvalp) 269 { 270 (void) flag, (void) cred, (void) rvalp; 271 loff_t off; 272 int error; 273 274 switch (com) { 275 case _FIOFFS: 276 { 277 return (0); 278 279 /* 280 * The following two ioctls are used by bfu. Faking out, 281 * necessary to avoid bfu errors. 282 */ 283 } 284 case _FIOGDIO: 285 case _FIOSDIO: 286 { 287 return (0); 288 } 289 290 case F_SEEK_DATA: 291 case F_SEEK_HOLE: 292 { 293 off = *(offset_t *)data; 294 error = vn_lock(vp, LK_SHARED); 295 if (error) 296 return (error); 297 /* offset parameter is in/out */ 298 error = zfs_holey(VTOZ(vp), com, &off); 299 VOP_UNLOCK(vp); 300 if (error) 301 return (error); 302 *(offset_t *)data = off; 303 return (0); 304 } 305 } 306 return (SET_ERROR(ENOTTY)); 307 } 308 309 static vm_page_t 310 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) 311 { 312 vm_object_t obj; 313 vm_page_t pp; 314 int64_t end; 315 316 /* 317 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE 318 * aligned boundaries, if the range is not aligned. As a result a 319 * DEV_BSIZE subrange with partially dirty data may get marked as clean. 320 * It may happen that all DEV_BSIZE subranges are marked clean and thus 321 * the whole page would be considered clean despite have some 322 * dirty data. 323 * For this reason we should shrink the range to DEV_BSIZE aligned 324 * boundaries before calling vm_page_clear_dirty. 325 */ 326 end = rounddown2(off + nbytes, DEV_BSIZE); 327 off = roundup2(off, DEV_BSIZE); 328 nbytes = end - off; 329 330 obj = vp->v_object; 331 vm_page_grab_valid_unlocked(&pp, obj, OFF_TO_IDX(start), 332 VM_ALLOC_NOCREAT | VM_ALLOC_SBUSY | VM_ALLOC_NORMAL | 333 VM_ALLOC_IGN_SBUSY); 334 if (pp != NULL) { 335 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 336 vm_object_pip_add(obj, 1); 337 pmap_remove_write(pp); 338 if (nbytes != 0) 339 vm_page_clear_dirty(pp, off, nbytes); 340 } 341 return (pp); 342 } 343 344 static void 345 page_unbusy(vm_page_t pp) 346 { 347 348 vm_page_sunbusy(pp); 349 vm_object_pip_wakeup(pp->object); 350 } 351 352 static vm_page_t 353 page_hold(vnode_t *vp, int64_t start) 354 { 355 vm_object_t obj; 356 vm_page_t m; 357 358 obj = vp->v_object; 359 vm_page_grab_valid_unlocked(&m, obj, OFF_TO_IDX(start), 360 VM_ALLOC_NOCREAT | VM_ALLOC_WIRED | VM_ALLOC_IGN_SBUSY | 361 VM_ALLOC_NOBUSY); 362 return (m); 363 } 364 365 static void 366 page_unhold(vm_page_t pp) 367 { 368 vm_page_unwire(pp, PQ_ACTIVE); 369 } 370 371 /* 372 * When a file is memory mapped, we must keep the IO data synchronized 373 * between the DMU cache and the memory mapped pages. What this means: 374 * 375 * On Write: If we find a memory mapped page, we write to *both* 376 * the page and the dmu buffer. 377 */ 378 void 379 update_pages(znode_t *zp, int64_t start, int len, objset_t *os) 380 { 381 vm_object_t obj; 382 struct sf_buf *sf; 383 vnode_t *vp = ZTOV(zp); 384 caddr_t va; 385 int off; 386 387 ASSERT3P(vp->v_mount, !=, NULL); 388 obj = vp->v_object; 389 ASSERT3P(obj, !=, NULL); 390 391 off = start & PAGEOFFSET; 392 vm_object_pip_add(obj, 1); 393 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 394 vm_page_t pp; 395 int nbytes = imin(PAGESIZE - off, len); 396 397 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) { 398 va = zfs_map_page(pp, &sf); 399 (void) dmu_read(os, zp->z_id, start + off, nbytes, 400 va + off, DMU_READ_PREFETCH); 401 zfs_unmap_page(sf); 402 page_unbusy(pp); 403 } 404 len -= nbytes; 405 off = 0; 406 } 407 vm_object_pip_wakeup(obj); 408 } 409 410 /* 411 * Read with UIO_NOCOPY flag means that sendfile(2) requests 412 * ZFS to populate a range of page cache pages with data. 413 * 414 * NOTE: this function could be optimized to pre-allocate 415 * all pages in advance, drain exclusive busy on all of them, 416 * map them into contiguous KVA region and populate them 417 * in one single dmu_read() call. 418 */ 419 int 420 mappedread_sf(znode_t *zp, int nbytes, zfs_uio_t *uio) 421 { 422 vnode_t *vp = ZTOV(zp); 423 objset_t *os = zp->z_zfsvfs->z_os; 424 struct sf_buf *sf; 425 vm_object_t obj; 426 vm_page_t pp; 427 int64_t start; 428 caddr_t va; 429 int len = nbytes; 430 int error = 0; 431 432 ASSERT3U(zfs_uio_segflg(uio), ==, UIO_NOCOPY); 433 ASSERT3P(vp->v_mount, !=, NULL); 434 obj = vp->v_object; 435 ASSERT3P(obj, !=, NULL); 436 ASSERT0(zfs_uio_offset(uio) & PAGEOFFSET); 437 438 for (start = zfs_uio_offset(uio); len > 0; start += PAGESIZE) { 439 int bytes = MIN(PAGESIZE, len); 440 441 pp = vm_page_grab_unlocked(obj, OFF_TO_IDX(start), 442 VM_ALLOC_SBUSY | VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); 443 if (vm_page_none_valid(pp)) { 444 va = zfs_map_page(pp, &sf); 445 error = dmu_read(os, zp->z_id, start, bytes, va, 446 DMU_READ_PREFETCH); 447 if (bytes != PAGESIZE && error == 0) 448 memset(va + bytes, 0, PAGESIZE - bytes); 449 zfs_unmap_page(sf); 450 if (error == 0) { 451 vm_page_valid(pp); 452 vm_page_activate(pp); 453 vm_page_sunbusy(pp); 454 } else { 455 zfs_vmobject_wlock(obj); 456 if (!vm_page_wired(pp) && pp->valid == 0 && 457 vm_page_busy_tryupgrade(pp)) 458 vm_page_free(pp); 459 else { 460 vm_page_deactivate_noreuse(pp); 461 vm_page_sunbusy(pp); 462 } 463 zfs_vmobject_wunlock(obj); 464 } 465 } else { 466 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 467 vm_page_sunbusy(pp); 468 } 469 if (error) 470 break; 471 zfs_uio_advance(uio, bytes); 472 len -= bytes; 473 } 474 return (error); 475 } 476 477 /* 478 * When a file is memory mapped, we must keep the IO data synchronized 479 * between the DMU cache and the memory mapped pages. What this means: 480 * 481 * On Read: We "read" preferentially from memory mapped pages, 482 * else we default from the dmu buffer. 483 * 484 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 485 * the file is memory mapped. 486 */ 487 int 488 mappedread(znode_t *zp, int nbytes, zfs_uio_t *uio) 489 { 490 vnode_t *vp = ZTOV(zp); 491 vm_object_t obj; 492 int64_t start; 493 int len = nbytes; 494 int off; 495 int error = 0; 496 497 ASSERT3P(vp->v_mount, !=, NULL); 498 obj = vp->v_object; 499 ASSERT3P(obj, !=, NULL); 500 501 start = zfs_uio_offset(uio); 502 off = start & PAGEOFFSET; 503 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 504 vm_page_t pp; 505 uint64_t bytes = MIN(PAGESIZE - off, len); 506 507 if ((pp = page_hold(vp, start))) { 508 struct sf_buf *sf; 509 caddr_t va; 510 511 va = zfs_map_page(pp, &sf); 512 error = vn_io_fault_uiomove(va + off, bytes, 513 GET_UIO_STRUCT(uio)); 514 zfs_unmap_page(sf); 515 page_unhold(pp); 516 } else { 517 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 518 uio, bytes); 519 } 520 len -= bytes; 521 off = 0; 522 if (error) 523 break; 524 } 525 return (error); 526 } 527 528 int 529 zfs_write_simple(znode_t *zp, const void *data, size_t len, 530 loff_t pos, size_t *presid) 531 { 532 int error = 0; 533 ssize_t resid; 534 535 error = vn_rdwr(UIO_WRITE, ZTOV(zp), __DECONST(void *, data), len, pos, 536 UIO_SYSSPACE, IO_SYNC, kcred, NOCRED, &resid, curthread); 537 538 if (error) { 539 return (SET_ERROR(error)); 540 } else if (presid == NULL) { 541 if (resid != 0) { 542 error = SET_ERROR(EIO); 543 } 544 } else { 545 *presid = resid; 546 } 547 return (error); 548 } 549 550 void 551 zfs_zrele_async(znode_t *zp) 552 { 553 vnode_t *vp = ZTOV(zp); 554 objset_t *os = ITOZSB(vp)->z_os; 555 556 VN_RELE_ASYNC(vp, dsl_pool_zrele_taskq(dmu_objset_pool(os))); 557 } 558 559 static int 560 zfs_dd_callback(struct mount *mp, void *arg, int lkflags, struct vnode **vpp) 561 { 562 int error; 563 564 *vpp = arg; 565 error = vn_lock(*vpp, lkflags); 566 if (error != 0) 567 vrele(*vpp); 568 return (error); 569 } 570 571 static int 572 zfs_lookup_lock(vnode_t *dvp, vnode_t *vp, const char *name, int lkflags) 573 { 574 znode_t *zdp = VTOZ(dvp); 575 zfsvfs_t *zfsvfs __unused = zdp->z_zfsvfs; 576 int error; 577 int ltype; 578 579 if (zfsvfs->z_replay == B_FALSE) 580 ASSERT_VOP_LOCKED(dvp, __func__); 581 582 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) { 583 ASSERT3P(dvp, ==, vp); 584 vref(dvp); 585 ltype = lkflags & LK_TYPE_MASK; 586 if (ltype != VOP_ISLOCKED(dvp)) { 587 if (ltype == LK_EXCLUSIVE) 588 vn_lock(dvp, LK_UPGRADE | LK_RETRY); 589 else /* if (ltype == LK_SHARED) */ 590 vn_lock(dvp, LK_DOWNGRADE | LK_RETRY); 591 592 /* 593 * Relock for the "." case could leave us with 594 * reclaimed vnode. 595 */ 596 if (VN_IS_DOOMED(dvp)) { 597 vrele(dvp); 598 return (SET_ERROR(ENOENT)); 599 } 600 } 601 return (0); 602 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) { 603 /* 604 * Note that in this case, dvp is the child vnode, and we 605 * are looking up the parent vnode - exactly reverse from 606 * normal operation. Unlocking dvp requires some rather 607 * tricky unlock/relock dance to prevent mp from being freed; 608 * use vn_vget_ino_gen() which takes care of all that. 609 * 610 * XXX Note that there is a time window when both vnodes are 611 * unlocked. It is possible, although highly unlikely, that 612 * during that window the parent-child relationship between 613 * the vnodes may change, for example, get reversed. 614 * In that case we would have a wrong lock order for the vnodes. 615 * All other filesystems seem to ignore this problem, so we 616 * do the same here. 617 * A potential solution could be implemented as follows: 618 * - using LK_NOWAIT when locking the second vnode and retrying 619 * if necessary 620 * - checking that the parent-child relationship still holds 621 * after locking both vnodes and retrying if it doesn't 622 */ 623 error = vn_vget_ino_gen(dvp, zfs_dd_callback, vp, lkflags, &vp); 624 return (error); 625 } else { 626 error = vn_lock(vp, lkflags); 627 if (error != 0) 628 vrele(vp); 629 return (error); 630 } 631 } 632 633 /* 634 * Lookup an entry in a directory, or an extended attribute directory. 635 * If it exists, return a held vnode reference for it. 636 * 637 * IN: dvp - vnode of directory to search. 638 * nm - name of entry to lookup. 639 * pnp - full pathname to lookup [UNUSED]. 640 * flags - LOOKUP_XATTR set if looking for an attribute. 641 * rdir - root directory vnode [UNUSED]. 642 * cr - credentials of caller. 643 * ct - caller context 644 * 645 * OUT: vpp - vnode of located entry, NULL if not found. 646 * 647 * RETURN: 0 on success, error code on failure. 648 * 649 * Timestamps: 650 * NA 651 */ 652 static int 653 zfs_lookup(vnode_t *dvp, const char *nm, vnode_t **vpp, 654 struct componentname *cnp, int nameiop, cred_t *cr, int flags, 655 boolean_t cached) 656 { 657 znode_t *zdp = VTOZ(dvp); 658 znode_t *zp; 659 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 660 seqc_t dvp_seqc; 661 int error = 0; 662 663 /* 664 * Fast path lookup, however we must skip DNLC lookup 665 * for case folding or normalizing lookups because the 666 * DNLC code only stores the passed in name. This means 667 * creating 'a' and removing 'A' on a case insensitive 668 * file system would work, but DNLC still thinks 'a' 669 * exists and won't let you create it again on the next 670 * pass through fast path. 671 */ 672 if (!(flags & LOOKUP_XATTR)) { 673 if (dvp->v_type != VDIR) { 674 return (SET_ERROR(ENOTDIR)); 675 } else if (zdp->z_sa_hdl == NULL) { 676 return (SET_ERROR(EIO)); 677 } 678 } 679 680 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, 681 const char *, nm); 682 683 if ((error = zfs_enter_verify_zp(zfsvfs, zdp, FTAG)) != 0) 684 return (error); 685 686 dvp_seqc = vn_seqc_read_notmodify(dvp); 687 688 *vpp = NULL; 689 690 if (flags & LOOKUP_XATTR) { 691 /* 692 * If the xattr property is off, refuse the lookup request. 693 */ 694 if (!(zfsvfs->z_flags & ZSB_XATTR)) { 695 zfs_exit(zfsvfs, FTAG); 696 return (SET_ERROR(EOPNOTSUPP)); 697 } 698 699 /* 700 * We don't allow recursive attributes.. 701 * Maybe someday we will. 702 */ 703 if (zdp->z_pflags & ZFS_XATTR) { 704 zfs_exit(zfsvfs, FTAG); 705 return (SET_ERROR(EINVAL)); 706 } 707 708 if ((error = zfs_get_xattrdir(VTOZ(dvp), &zp, cr, flags))) { 709 zfs_exit(zfsvfs, FTAG); 710 return (error); 711 } 712 *vpp = ZTOV(zp); 713 714 /* 715 * Do we have permission to get into attribute directory? 716 */ 717 error = zfs_zaccess(zp, ACE_EXECUTE, 0, B_FALSE, cr, NULL); 718 if (error) { 719 vrele(ZTOV(zp)); 720 } 721 722 zfs_exit(zfsvfs, FTAG); 723 return (error); 724 } 725 726 /* 727 * Check accessibility of directory if we're not coming in via 728 * VOP_CACHEDLOOKUP. 729 */ 730 if (!cached) { 731 #ifdef NOEXECCHECK 732 if ((cnp->cn_flags & NOEXECCHECK) != 0) { 733 cnp->cn_flags &= ~NOEXECCHECK; 734 } else 735 #endif 736 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr, 737 NULL))) { 738 zfs_exit(zfsvfs, FTAG); 739 return (error); 740 } 741 } 742 743 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 744 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 745 zfs_exit(zfsvfs, FTAG); 746 return (SET_ERROR(EILSEQ)); 747 } 748 749 750 /* 751 * First handle the special cases. 752 */ 753 if ((cnp->cn_flags & ISDOTDOT) != 0) { 754 /* 755 * If we are a snapshot mounted under .zfs, return 756 * the vp for the snapshot directory. 757 */ 758 if (zdp->z_id == zfsvfs->z_root && zfsvfs->z_parent != zfsvfs) { 759 struct componentname cn; 760 vnode_t *zfsctl_vp; 761 int ltype; 762 763 zfs_exit(zfsvfs, FTAG); 764 ltype = VOP_ISLOCKED(dvp); 765 VOP_UNLOCK(dvp); 766 error = zfsctl_root(zfsvfs->z_parent, LK_SHARED, 767 &zfsctl_vp); 768 if (error == 0) { 769 cn.cn_nameptr = "snapshot"; 770 cn.cn_namelen = strlen(cn.cn_nameptr); 771 cn.cn_nameiop = cnp->cn_nameiop; 772 cn.cn_flags = cnp->cn_flags & ~ISDOTDOT; 773 cn.cn_lkflags = cnp->cn_lkflags; 774 error = VOP_LOOKUP(zfsctl_vp, vpp, &cn); 775 vput(zfsctl_vp); 776 } 777 vn_lock(dvp, ltype | LK_RETRY); 778 return (error); 779 } 780 } 781 if (zfs_has_ctldir(zdp) && strcmp(nm, ZFS_CTLDIR_NAME) == 0) { 782 zfs_exit(zfsvfs, FTAG); 783 if (zfsvfs->z_show_ctldir == ZFS_SNAPDIR_DISABLED) 784 return (SET_ERROR(ENOENT)); 785 if ((cnp->cn_flags & ISLASTCN) != 0 && nameiop != LOOKUP) 786 return (SET_ERROR(ENOTSUP)); 787 error = zfsctl_root(zfsvfs, cnp->cn_lkflags, vpp); 788 return (error); 789 } 790 791 /* 792 * The loop is retry the lookup if the parent-child relationship 793 * changes during the dot-dot locking complexities. 794 */ 795 for (;;) { 796 uint64_t parent; 797 798 error = zfs_dirlook(zdp, nm, &zp); 799 if (error == 0) 800 *vpp = ZTOV(zp); 801 802 zfs_exit(zfsvfs, FTAG); 803 if (error != 0) 804 break; 805 806 error = zfs_lookup_lock(dvp, *vpp, nm, cnp->cn_lkflags); 807 if (error != 0) { 808 /* 809 * If we've got a locking error, then the vnode 810 * got reclaimed because of a force unmount. 811 * We never enter doomed vnodes into the name cache. 812 */ 813 *vpp = NULL; 814 return (error); 815 } 816 817 if ((cnp->cn_flags & ISDOTDOT) == 0) 818 break; 819 820 if ((error = zfs_enter(zfsvfs, FTAG)) != 0) { 821 vput(ZTOV(zp)); 822 *vpp = NULL; 823 return (error); 824 } 825 if (zdp->z_sa_hdl == NULL) { 826 error = SET_ERROR(EIO); 827 } else { 828 error = sa_lookup(zdp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 829 &parent, sizeof (parent)); 830 } 831 if (error != 0) { 832 zfs_exit(zfsvfs, FTAG); 833 vput(ZTOV(zp)); 834 break; 835 } 836 if (zp->z_id == parent) { 837 zfs_exit(zfsvfs, FTAG); 838 break; 839 } 840 vput(ZTOV(zp)); 841 } 842 843 if (error != 0) 844 *vpp = NULL; 845 846 /* Translate errors and add SAVENAME when needed. */ 847 if (cnp->cn_flags & ISLASTCN) { 848 switch (nameiop) { 849 case CREATE: 850 case RENAME: 851 if (error == ENOENT) { 852 error = EJUSTRETURN; 853 #if __FreeBSD_version < 1400068 854 cnp->cn_flags |= SAVENAME; 855 #endif 856 break; 857 } 858 zfs_fallthrough; 859 case DELETE: 860 #if __FreeBSD_version < 1400068 861 if (error == 0) 862 cnp->cn_flags |= SAVENAME; 863 #endif 864 break; 865 } 866 } 867 868 if ((cnp->cn_flags & ISDOTDOT) != 0) { 869 /* 870 * FIXME: zfs_lookup_lock relocks vnodes and does nothing to 871 * handle races. In particular different callers may end up 872 * with different vnodes and will try to add conflicting 873 * entries to the namecache. 874 * 875 * While finding different result may be acceptable in face 876 * of concurrent modification, adding conflicting entries 877 * trips over an assert in the namecache. 878 * 879 * Ultimately let an entry through once everything settles. 880 */ 881 if (!vn_seqc_consistent(dvp, dvp_seqc)) { 882 cnp->cn_flags &= ~MAKEENTRY; 883 } 884 } 885 886 /* Insert name into cache (as non-existent) if appropriate. */ 887 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay && 888 error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0) 889 cache_enter(dvp, NULL, cnp); 890 891 /* Insert name into cache if appropriate. */ 892 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay && 893 error == 0 && (cnp->cn_flags & MAKEENTRY)) { 894 if (!(cnp->cn_flags & ISLASTCN) || 895 (nameiop != DELETE && nameiop != RENAME)) { 896 cache_enter(dvp, *vpp, cnp); 897 } 898 } 899 900 return (error); 901 } 902 903 static inline bool 904 is_nametoolong(zfsvfs_t *zfsvfs, const char *name) 905 { 906 size_t dlen = strlen(name); 907 return ((!zfsvfs->z_longname && dlen >= ZAP_MAXNAMELEN) || 908 dlen >= ZAP_MAXNAMELEN_NEW); 909 } 910 911 /* 912 * Attempt to create a new entry in a directory. If the entry 913 * already exists, truncate the file if permissible, else return 914 * an error. Return the vp of the created or trunc'd file. 915 * 916 * IN: dvp - vnode of directory to put new file entry in. 917 * name - name of new file entry. 918 * vap - attributes of new file. 919 * excl - flag indicating exclusive or non-exclusive mode. 920 * mode - mode to open file with. 921 * cr - credentials of caller. 922 * flag - large file flag [UNUSED]. 923 * ct - caller context 924 * vsecp - ACL to be set 925 * mnt_ns - Unused on FreeBSD 926 * 927 * OUT: vpp - vnode of created or trunc'd entry. 928 * 929 * RETURN: 0 on success, error code on failure. 930 * 931 * Timestamps: 932 * dvp - ctime|mtime updated if new entry created 933 * vp - ctime|mtime always, atime if new 934 */ 935 int 936 zfs_create(znode_t *dzp, const char *name, vattr_t *vap, int excl, int mode, 937 znode_t **zpp, cred_t *cr, int flag, vsecattr_t *vsecp, zidmap_t *mnt_ns) 938 { 939 (void) excl, (void) mode, (void) flag; 940 znode_t *zp; 941 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 942 zilog_t *zilog; 943 objset_t *os; 944 dmu_tx_t *tx; 945 int error; 946 uid_t uid = crgetuid(cr); 947 gid_t gid = crgetgid(cr); 948 uint64_t projid = ZFS_DEFAULT_PROJID; 949 zfs_acl_ids_t acl_ids; 950 boolean_t fuid_dirtied; 951 uint64_t txtype; 952 #ifdef DEBUG_VFS_LOCKS 953 vnode_t *dvp = ZTOV(dzp); 954 #endif 955 956 if (is_nametoolong(zfsvfs, name)) 957 return (SET_ERROR(ENAMETOOLONG)); 958 959 /* 960 * If we have an ephemeral id, ACL, or XVATTR then 961 * make sure file system is at proper version 962 */ 963 if (zfsvfs->z_use_fuids == B_FALSE && 964 (vsecp || (vap->va_mask & AT_XVATTR) || 965 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 966 return (SET_ERROR(EINVAL)); 967 968 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 969 return (error); 970 os = zfsvfs->z_os; 971 zilog = zfsvfs->z_log; 972 973 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 974 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 975 zfs_exit(zfsvfs, FTAG); 976 return (SET_ERROR(EILSEQ)); 977 } 978 979 if (vap->va_mask & AT_XVATTR) { 980 if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap, 981 crgetuid(cr), cr, vap->va_type)) != 0) { 982 zfs_exit(zfsvfs, FTAG); 983 return (error); 984 } 985 } 986 987 *zpp = NULL; 988 989 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) 990 vap->va_mode &= ~S_ISVTX; 991 992 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW); 993 if (error) { 994 zfs_exit(zfsvfs, FTAG); 995 return (error); 996 } 997 ASSERT3P(zp, ==, NULL); 998 999 /* 1000 * Create a new file object and update the directory 1001 * to reference it. 1002 */ 1003 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) { 1004 goto out; 1005 } 1006 1007 /* 1008 * We only support the creation of regular files in 1009 * extended attribute directories. 1010 */ 1011 1012 if ((dzp->z_pflags & ZFS_XATTR) && 1013 (vap->va_type != VREG)) { 1014 error = SET_ERROR(EINVAL); 1015 goto out; 1016 } 1017 1018 if ((error = zfs_acl_ids_create(dzp, 0, vap, 1019 cr, vsecp, &acl_ids, NULL)) != 0) 1020 goto out; 1021 1022 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode)) 1023 projid = zfs_inherit_projid(dzp); 1024 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) { 1025 zfs_acl_ids_free(&acl_ids); 1026 error = SET_ERROR(EDQUOT); 1027 goto out; 1028 } 1029 1030 getnewvnode_reserve(); 1031 1032 tx = dmu_tx_create(os); 1033 1034 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1035 ZFS_SA_BASE_ATTR_SIZE); 1036 1037 fuid_dirtied = zfsvfs->z_fuid_dirty; 1038 if (fuid_dirtied) 1039 zfs_fuid_txhold(zfsvfs, tx); 1040 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1041 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1042 if (!zfsvfs->z_use_sa && 1043 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1044 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1045 0, acl_ids.z_aclp->z_acl_bytes); 1046 } 1047 error = dmu_tx_assign(tx, TXG_WAIT); 1048 if (error) { 1049 zfs_acl_ids_free(&acl_ids); 1050 dmu_tx_abort(tx); 1051 getnewvnode_drop_reserve(); 1052 zfs_exit(zfsvfs, FTAG); 1053 return (error); 1054 } 1055 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1056 1057 error = zfs_link_create(dzp, name, zp, tx, ZNEW); 1058 if (error != 0) { 1059 /* 1060 * Since, we failed to add the directory entry for it, 1061 * delete the newly created dnode. 1062 */ 1063 zfs_znode_delete(zp, tx); 1064 VOP_UNLOCK(ZTOV(zp)); 1065 zrele(zp); 1066 zfs_acl_ids_free(&acl_ids); 1067 dmu_tx_commit(tx); 1068 getnewvnode_drop_reserve(); 1069 goto out; 1070 } 1071 1072 if (fuid_dirtied) 1073 zfs_fuid_sync(zfsvfs, tx); 1074 1075 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1076 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1077 vsecp, acl_ids.z_fuidp, vap); 1078 zfs_acl_ids_free(&acl_ids); 1079 dmu_tx_commit(tx); 1080 1081 getnewvnode_drop_reserve(); 1082 1083 out: 1084 VNCHECKREF(dvp); 1085 if (error == 0) { 1086 *zpp = zp; 1087 } 1088 1089 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1090 zil_commit(zilog, 0); 1091 1092 zfs_exit(zfsvfs, FTAG); 1093 return (error); 1094 } 1095 1096 /* 1097 * Remove an entry from a directory. 1098 * 1099 * IN: dvp - vnode of directory to remove entry from. 1100 * name - name of entry to remove. 1101 * cr - credentials of caller. 1102 * ct - caller context 1103 * flags - case flags 1104 * 1105 * RETURN: 0 on success, error code on failure. 1106 * 1107 * Timestamps: 1108 * dvp - ctime|mtime 1109 * vp - ctime (if nlink > 0) 1110 */ 1111 static int 1112 zfs_remove_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr) 1113 { 1114 znode_t *dzp = VTOZ(dvp); 1115 znode_t *zp; 1116 znode_t *xzp; 1117 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1118 zilog_t *zilog; 1119 uint64_t xattr_obj; 1120 uint64_t obj = 0; 1121 dmu_tx_t *tx; 1122 boolean_t unlinked; 1123 uint64_t txtype; 1124 int error; 1125 1126 1127 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 1128 return (error); 1129 zp = VTOZ(vp); 1130 if ((error = zfs_verify_zp(zp)) != 0) { 1131 zfs_exit(zfsvfs, FTAG); 1132 return (error); 1133 } 1134 zilog = zfsvfs->z_log; 1135 1136 xattr_obj = 0; 1137 xzp = NULL; 1138 1139 if ((error = zfs_zaccess_delete(dzp, zp, cr, NULL))) { 1140 goto out; 1141 } 1142 1143 /* 1144 * Need to use rmdir for removing directories. 1145 */ 1146 if (vp->v_type == VDIR) { 1147 error = SET_ERROR(EPERM); 1148 goto out; 1149 } 1150 1151 vnevent_remove(vp, dvp, name, ct); 1152 1153 obj = zp->z_id; 1154 1155 /* are there any extended attributes? */ 1156 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1157 &xattr_obj, sizeof (xattr_obj)); 1158 if (error == 0 && xattr_obj) { 1159 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1160 ASSERT0(error); 1161 } 1162 1163 /* 1164 * We may delete the znode now, or we may put it in the unlinked set; 1165 * it depends on whether we're the last link, and on whether there are 1166 * other holds on the vnode. So we dmu_tx_hold() the right things to 1167 * allow for either case. 1168 */ 1169 tx = dmu_tx_create(zfsvfs->z_os); 1170 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1171 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1172 zfs_sa_upgrade_txholds(tx, zp); 1173 zfs_sa_upgrade_txholds(tx, dzp); 1174 1175 if (xzp) { 1176 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1177 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1178 } 1179 1180 /* charge as an update -- would be nice not to charge at all */ 1181 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1182 1183 /* 1184 * Mark this transaction as typically resulting in a net free of space 1185 */ 1186 dmu_tx_mark_netfree(tx); 1187 1188 error = dmu_tx_assign(tx, TXG_WAIT); 1189 if (error) { 1190 dmu_tx_abort(tx); 1191 zfs_exit(zfsvfs, FTAG); 1192 return (error); 1193 } 1194 1195 /* 1196 * Remove the directory entry. 1197 */ 1198 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, &unlinked); 1199 1200 if (error) { 1201 dmu_tx_commit(tx); 1202 goto out; 1203 } 1204 1205 if (unlinked) { 1206 zfs_unlinked_add(zp, tx); 1207 vp->v_vflag |= VV_NOSYNC; 1208 } 1209 /* XXX check changes to linux vnops */ 1210 txtype = TX_REMOVE; 1211 zfs_log_remove(zilog, tx, txtype, dzp, name, obj, unlinked); 1212 1213 dmu_tx_commit(tx); 1214 out: 1215 1216 if (xzp) 1217 vrele(ZTOV(xzp)); 1218 1219 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1220 zil_commit(zilog, 0); 1221 1222 1223 zfs_exit(zfsvfs, FTAG); 1224 return (error); 1225 } 1226 1227 1228 static int 1229 zfs_lookup_internal(znode_t *dzp, const char *name, vnode_t **vpp, 1230 struct componentname *cnp, int nameiop) 1231 { 1232 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1233 int error; 1234 1235 cnp->cn_nameptr = __DECONST(char *, name); 1236 cnp->cn_namelen = strlen(name); 1237 cnp->cn_nameiop = nameiop; 1238 cnp->cn_flags = ISLASTCN; 1239 #if __FreeBSD_version < 1400068 1240 cnp->cn_flags |= SAVENAME; 1241 #endif 1242 cnp->cn_lkflags = LK_EXCLUSIVE | LK_RETRY; 1243 cnp->cn_cred = kcred; 1244 #if __FreeBSD_version < 1400037 1245 cnp->cn_thread = curthread; 1246 #endif 1247 1248 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay) { 1249 struct vop_lookup_args a; 1250 1251 a.a_gen.a_desc = &vop_lookup_desc; 1252 a.a_dvp = ZTOV(dzp); 1253 a.a_vpp = vpp; 1254 a.a_cnp = cnp; 1255 error = vfs_cache_lookup(&a); 1256 } else { 1257 error = zfs_lookup(ZTOV(dzp), name, vpp, cnp, nameiop, kcred, 0, 1258 B_FALSE); 1259 } 1260 #ifdef ZFS_DEBUG 1261 if (error) { 1262 printf("got error %d on name %s on op %d\n", error, name, 1263 nameiop); 1264 kdb_backtrace(); 1265 } 1266 #endif 1267 return (error); 1268 } 1269 1270 int 1271 zfs_remove(znode_t *dzp, const char *name, cred_t *cr, int flags) 1272 { 1273 vnode_t *vp; 1274 int error; 1275 struct componentname cn; 1276 1277 if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE))) 1278 return (error); 1279 1280 error = zfs_remove_(ZTOV(dzp), vp, name, cr); 1281 vput(vp); 1282 return (error); 1283 } 1284 /* 1285 * Create a new directory and insert it into dvp using the name 1286 * provided. Return a pointer to the inserted directory. 1287 * 1288 * IN: dvp - vnode of directory to add subdir to. 1289 * dirname - name of new directory. 1290 * vap - attributes of new directory. 1291 * cr - credentials of caller. 1292 * ct - caller context 1293 * flags - case flags 1294 * vsecp - ACL to be set 1295 * mnt_ns - Unused on FreeBSD 1296 * 1297 * OUT: vpp - vnode of created directory. 1298 * 1299 * RETURN: 0 on success, error code on failure. 1300 * 1301 * Timestamps: 1302 * dvp - ctime|mtime updated 1303 * vp - ctime|mtime|atime updated 1304 */ 1305 int 1306 zfs_mkdir(znode_t *dzp, const char *dirname, vattr_t *vap, znode_t **zpp, 1307 cred_t *cr, int flags, vsecattr_t *vsecp, zidmap_t *mnt_ns) 1308 { 1309 (void) flags, (void) vsecp; 1310 znode_t *zp; 1311 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1312 zilog_t *zilog; 1313 uint64_t txtype; 1314 dmu_tx_t *tx; 1315 int error; 1316 uid_t uid = crgetuid(cr); 1317 gid_t gid = crgetgid(cr); 1318 zfs_acl_ids_t acl_ids; 1319 boolean_t fuid_dirtied; 1320 1321 ASSERT3U(vap->va_type, ==, VDIR); 1322 1323 if (is_nametoolong(zfsvfs, dirname)) 1324 return (SET_ERROR(ENAMETOOLONG)); 1325 1326 /* 1327 * If we have an ephemeral id, ACL, or XVATTR then 1328 * make sure file system is at proper version 1329 */ 1330 if (zfsvfs->z_use_fuids == B_FALSE && 1331 ((vap->va_mask & AT_XVATTR) || 1332 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1333 return (SET_ERROR(EINVAL)); 1334 1335 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 1336 return (error); 1337 zilog = zfsvfs->z_log; 1338 1339 if (dzp->z_pflags & ZFS_XATTR) { 1340 zfs_exit(zfsvfs, FTAG); 1341 return (SET_ERROR(EINVAL)); 1342 } 1343 1344 if (zfsvfs->z_utf8 && u8_validate(dirname, 1345 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1346 zfs_exit(zfsvfs, FTAG); 1347 return (SET_ERROR(EILSEQ)); 1348 } 1349 1350 if (vap->va_mask & AT_XVATTR) { 1351 if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap, 1352 crgetuid(cr), cr, vap->va_type)) != 0) { 1353 zfs_exit(zfsvfs, FTAG); 1354 return (error); 1355 } 1356 } 1357 1358 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 1359 NULL, &acl_ids, NULL)) != 0) { 1360 zfs_exit(zfsvfs, FTAG); 1361 return (error); 1362 } 1363 1364 /* 1365 * First make sure the new directory doesn't exist. 1366 * 1367 * Existence is checked first to make sure we don't return 1368 * EACCES instead of EEXIST which can cause some applications 1369 * to fail. 1370 */ 1371 *zpp = NULL; 1372 1373 if ((error = zfs_dirent_lookup(dzp, dirname, &zp, ZNEW))) { 1374 zfs_acl_ids_free(&acl_ids); 1375 zfs_exit(zfsvfs, FTAG); 1376 return (error); 1377 } 1378 ASSERT3P(zp, ==, NULL); 1379 1380 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr, 1381 mnt_ns))) { 1382 zfs_acl_ids_free(&acl_ids); 1383 zfs_exit(zfsvfs, FTAG); 1384 return (error); 1385 } 1386 1387 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zfs_inherit_projid(dzp))) { 1388 zfs_acl_ids_free(&acl_ids); 1389 zfs_exit(zfsvfs, FTAG); 1390 return (SET_ERROR(EDQUOT)); 1391 } 1392 1393 /* 1394 * Add a new entry to the directory. 1395 */ 1396 getnewvnode_reserve(); 1397 tx = dmu_tx_create(zfsvfs->z_os); 1398 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1399 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1400 fuid_dirtied = zfsvfs->z_fuid_dirty; 1401 if (fuid_dirtied) 1402 zfs_fuid_txhold(zfsvfs, tx); 1403 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1404 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1405 acl_ids.z_aclp->z_acl_bytes); 1406 } 1407 1408 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1409 ZFS_SA_BASE_ATTR_SIZE); 1410 1411 error = dmu_tx_assign(tx, TXG_WAIT); 1412 if (error) { 1413 zfs_acl_ids_free(&acl_ids); 1414 dmu_tx_abort(tx); 1415 getnewvnode_drop_reserve(); 1416 zfs_exit(zfsvfs, FTAG); 1417 return (error); 1418 } 1419 1420 /* 1421 * Create new node. 1422 */ 1423 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1424 1425 /* 1426 * Now put new name in parent dir. 1427 */ 1428 error = zfs_link_create(dzp, dirname, zp, tx, ZNEW); 1429 if (error != 0) { 1430 zfs_znode_delete(zp, tx); 1431 VOP_UNLOCK(ZTOV(zp)); 1432 zrele(zp); 1433 goto out; 1434 } 1435 1436 if (fuid_dirtied) 1437 zfs_fuid_sync(zfsvfs, tx); 1438 1439 *zpp = zp; 1440 1441 txtype = zfs_log_create_txtype(Z_DIR, NULL, vap); 1442 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, NULL, 1443 acl_ids.z_fuidp, vap); 1444 1445 out: 1446 zfs_acl_ids_free(&acl_ids); 1447 1448 dmu_tx_commit(tx); 1449 1450 getnewvnode_drop_reserve(); 1451 1452 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1453 zil_commit(zilog, 0); 1454 1455 zfs_exit(zfsvfs, FTAG); 1456 return (error); 1457 } 1458 1459 /* 1460 * Remove a directory subdir entry. If the current working 1461 * directory is the same as the subdir to be removed, the 1462 * remove will fail. 1463 * 1464 * IN: dvp - vnode of directory to remove from. 1465 * name - name of directory to be removed. 1466 * cwd - vnode of current working directory. 1467 * cr - credentials of caller. 1468 * ct - caller context 1469 * flags - case flags 1470 * 1471 * RETURN: 0 on success, error code on failure. 1472 * 1473 * Timestamps: 1474 * dvp - ctime|mtime updated 1475 */ 1476 static int 1477 zfs_rmdir_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr) 1478 { 1479 znode_t *dzp = VTOZ(dvp); 1480 znode_t *zp = VTOZ(vp); 1481 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1482 zilog_t *zilog; 1483 dmu_tx_t *tx; 1484 int error; 1485 1486 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 1487 return (error); 1488 if ((error = zfs_verify_zp(zp)) != 0) { 1489 zfs_exit(zfsvfs, FTAG); 1490 return (error); 1491 } 1492 zilog = zfsvfs->z_log; 1493 1494 1495 if ((error = zfs_zaccess_delete(dzp, zp, cr, NULL))) { 1496 goto out; 1497 } 1498 1499 if (vp->v_type != VDIR) { 1500 error = SET_ERROR(ENOTDIR); 1501 goto out; 1502 } 1503 1504 vnevent_rmdir(vp, dvp, name, ct); 1505 1506 tx = dmu_tx_create(zfsvfs->z_os); 1507 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1508 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1509 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1510 zfs_sa_upgrade_txholds(tx, zp); 1511 zfs_sa_upgrade_txholds(tx, dzp); 1512 dmu_tx_mark_netfree(tx); 1513 error = dmu_tx_assign(tx, TXG_WAIT); 1514 if (error) { 1515 dmu_tx_abort(tx); 1516 zfs_exit(zfsvfs, FTAG); 1517 return (error); 1518 } 1519 1520 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, NULL); 1521 1522 if (error == 0) { 1523 uint64_t txtype = TX_RMDIR; 1524 zfs_log_remove(zilog, tx, txtype, dzp, name, 1525 ZFS_NO_OBJECT, B_FALSE); 1526 } 1527 1528 dmu_tx_commit(tx); 1529 1530 if (zfsvfs->z_use_namecache) 1531 cache_vop_rmdir(dvp, vp); 1532 out: 1533 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1534 zil_commit(zilog, 0); 1535 1536 zfs_exit(zfsvfs, FTAG); 1537 return (error); 1538 } 1539 1540 int 1541 zfs_rmdir(znode_t *dzp, const char *name, znode_t *cwd, cred_t *cr, int flags) 1542 { 1543 struct componentname cn; 1544 vnode_t *vp; 1545 int error; 1546 1547 if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE))) 1548 return (error); 1549 1550 error = zfs_rmdir_(ZTOV(dzp), vp, name, cr); 1551 vput(vp); 1552 return (error); 1553 } 1554 1555 /* 1556 * Read as many directory entries as will fit into the provided 1557 * buffer from the given directory cursor position (specified in 1558 * the uio structure). 1559 * 1560 * IN: vp - vnode of directory to read. 1561 * uio - structure supplying read location, range info, 1562 * and return buffer. 1563 * cr - credentials of caller. 1564 * ct - caller context 1565 * 1566 * OUT: uio - updated offset and range, buffer filled. 1567 * eofp - set to true if end-of-file detected. 1568 * ncookies- number of entries in cookies 1569 * cookies - offsets to directory entries 1570 * 1571 * RETURN: 0 on success, error code on failure. 1572 * 1573 * Timestamps: 1574 * vp - atime updated 1575 * 1576 * Note that the low 4 bits of the cookie returned by zap is always zero. 1577 * This allows us to use the low range for "special" directory entries: 1578 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1579 * we use the offset 2 for the '.zfs' directory. 1580 */ 1581 static int 1582 zfs_readdir(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, int *eofp, 1583 int *ncookies, cookie_t **cookies) 1584 { 1585 znode_t *zp = VTOZ(vp); 1586 iovec_t *iovp; 1587 dirent64_t *odp; 1588 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1589 objset_t *os; 1590 caddr_t outbuf; 1591 size_t bufsize; 1592 zap_cursor_t zc; 1593 zap_attribute_t *zap; 1594 uint_t bytes_wanted; 1595 uint64_t offset; /* must be unsigned; checks for < 1 */ 1596 uint64_t parent; 1597 int local_eof; 1598 int outcount; 1599 int error; 1600 uint8_t prefetch; 1601 uint8_t type; 1602 int ncooks; 1603 cookie_t *cooks = NULL; 1604 1605 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 1606 return (error); 1607 1608 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 1609 &parent, sizeof (parent))) != 0) { 1610 zfs_exit(zfsvfs, FTAG); 1611 return (error); 1612 } 1613 1614 /* 1615 * If we are not given an eof variable, 1616 * use a local one. 1617 */ 1618 if (eofp == NULL) 1619 eofp = &local_eof; 1620 1621 /* 1622 * Check for valid iov_len. 1623 */ 1624 if (GET_UIO_STRUCT(uio)->uio_iov->iov_len <= 0) { 1625 zfs_exit(zfsvfs, FTAG); 1626 return (SET_ERROR(EINVAL)); 1627 } 1628 1629 /* 1630 * Quit if directory has been removed (posix) 1631 */ 1632 if ((*eofp = zp->z_unlinked) != 0) { 1633 zfs_exit(zfsvfs, FTAG); 1634 return (0); 1635 } 1636 1637 error = 0; 1638 os = zfsvfs->z_os; 1639 offset = zfs_uio_offset(uio); 1640 prefetch = zp->z_zn_prefetch; 1641 zap = zap_attribute_long_alloc(); 1642 1643 /* 1644 * Initialize the iterator cursor. 1645 */ 1646 if (offset <= 3) { 1647 /* 1648 * Start iteration from the beginning of the directory. 1649 */ 1650 zap_cursor_init(&zc, os, zp->z_id); 1651 } else { 1652 /* 1653 * The offset is a serialized cursor. 1654 */ 1655 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1656 } 1657 1658 /* 1659 * Get space to change directory entries into fs independent format. 1660 */ 1661 iovp = GET_UIO_STRUCT(uio)->uio_iov; 1662 bytes_wanted = iovp->iov_len; 1663 if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1) { 1664 bufsize = bytes_wanted; 1665 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1666 odp = (struct dirent64 *)outbuf; 1667 } else { 1668 bufsize = bytes_wanted; 1669 outbuf = NULL; 1670 odp = (struct dirent64 *)iovp->iov_base; 1671 } 1672 1673 if (ncookies != NULL) { 1674 /* 1675 * Minimum entry size is dirent size and 1 byte for a file name. 1676 */ 1677 ncooks = zfs_uio_resid(uio) / (sizeof (struct dirent) - 1678 sizeof (((struct dirent *)NULL)->d_name) + 1); 1679 cooks = malloc(ncooks * sizeof (*cooks), M_TEMP, M_WAITOK); 1680 *cookies = cooks; 1681 *ncookies = ncooks; 1682 } 1683 1684 /* 1685 * Transform to file-system independent format 1686 */ 1687 outcount = 0; 1688 while (outcount < bytes_wanted) { 1689 ino64_t objnum; 1690 ushort_t reclen; 1691 off64_t *next = NULL; 1692 1693 /* 1694 * Special case `.', `..', and `.zfs'. 1695 */ 1696 if (offset == 0) { 1697 (void) strcpy(zap->za_name, "."); 1698 zap->za_normalization_conflict = 0; 1699 objnum = zp->z_id; 1700 type = DT_DIR; 1701 } else if (offset == 1) { 1702 (void) strcpy(zap->za_name, ".."); 1703 zap->za_normalization_conflict = 0; 1704 objnum = parent; 1705 type = DT_DIR; 1706 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1707 (void) strcpy(zap->za_name, ZFS_CTLDIR_NAME); 1708 zap->za_normalization_conflict = 0; 1709 objnum = ZFSCTL_INO_ROOT; 1710 type = DT_DIR; 1711 } else { 1712 /* 1713 * Grab next entry. 1714 */ 1715 if ((error = zap_cursor_retrieve(&zc, zap))) { 1716 if ((*eofp = (error == ENOENT)) != 0) 1717 break; 1718 else 1719 goto update; 1720 } 1721 1722 if (zap->za_integer_length != 8 || 1723 zap->za_num_integers != 1) { 1724 cmn_err(CE_WARN, "zap_readdir: bad directory " 1725 "entry, obj = %lld, offset = %lld\n", 1726 (u_longlong_t)zp->z_id, 1727 (u_longlong_t)offset); 1728 error = SET_ERROR(ENXIO); 1729 goto update; 1730 } 1731 1732 objnum = ZFS_DIRENT_OBJ(zap->za_first_integer); 1733 /* 1734 * MacOS X can extract the object type here such as: 1735 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1736 */ 1737 type = ZFS_DIRENT_TYPE(zap->za_first_integer); 1738 } 1739 1740 reclen = DIRENT64_RECLEN(strlen(zap->za_name)); 1741 1742 /* 1743 * Will this entry fit in the buffer? 1744 */ 1745 if (outcount + reclen > bufsize) { 1746 /* 1747 * Did we manage to fit anything in the buffer? 1748 */ 1749 if (!outcount) { 1750 error = SET_ERROR(EINVAL); 1751 goto update; 1752 } 1753 break; 1754 } 1755 /* 1756 * Add normal entry: 1757 */ 1758 odp->d_ino = objnum; 1759 odp->d_reclen = reclen; 1760 odp->d_namlen = strlen(zap->za_name); 1761 /* NOTE: d_off is the offset for the *next* entry. */ 1762 next = &odp->d_off; 1763 strlcpy(odp->d_name, zap->za_name, odp->d_namlen + 1); 1764 odp->d_type = type; 1765 dirent_terminate(odp); 1766 odp = (dirent64_t *)((intptr_t)odp + reclen); 1767 1768 outcount += reclen; 1769 1770 ASSERT3S(outcount, <=, bufsize); 1771 1772 if (prefetch) 1773 dmu_prefetch_dnode(os, objnum, ZIO_PRIORITY_SYNC_READ); 1774 1775 /* 1776 * Move to the next entry, fill in the previous offset. 1777 */ 1778 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1779 zap_cursor_advance(&zc); 1780 offset = zap_cursor_serialize(&zc); 1781 } else { 1782 offset += 1; 1783 } 1784 1785 /* Fill the offset right after advancing the cursor. */ 1786 if (next != NULL) 1787 *next = offset; 1788 if (cooks != NULL) { 1789 *cooks++ = offset; 1790 ncooks--; 1791 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 1792 } 1793 } 1794 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1795 1796 /* Subtract unused cookies */ 1797 if (ncookies != NULL) 1798 *ncookies -= ncooks; 1799 1800 if (zfs_uio_segflg(uio) == UIO_SYSSPACE && zfs_uio_iovcnt(uio) == 1) { 1801 iovp->iov_base += outcount; 1802 iovp->iov_len -= outcount; 1803 zfs_uio_resid(uio) -= outcount; 1804 } else if ((error = 1805 zfs_uiomove(outbuf, (long)outcount, UIO_READ, uio))) { 1806 /* 1807 * Reset the pointer. 1808 */ 1809 offset = zfs_uio_offset(uio); 1810 } 1811 1812 update: 1813 zap_cursor_fini(&zc); 1814 zap_attribute_free(zap); 1815 if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1) 1816 kmem_free(outbuf, bufsize); 1817 1818 if (error == ENOENT) 1819 error = 0; 1820 1821 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1822 1823 zfs_uio_setoffset(uio, offset); 1824 zfs_exit(zfsvfs, FTAG); 1825 if (error != 0 && cookies != NULL) { 1826 free(*cookies, M_TEMP); 1827 *cookies = NULL; 1828 *ncookies = 0; 1829 } 1830 return (error); 1831 } 1832 1833 /* 1834 * Get the requested file attributes and place them in the provided 1835 * vattr structure. 1836 * 1837 * IN: vp - vnode of file. 1838 * vap - va_mask identifies requested attributes. 1839 * If AT_XVATTR set, then optional attrs are requested 1840 * flags - ATTR_NOACLCHECK (CIFS server context) 1841 * cr - credentials of caller. 1842 * 1843 * OUT: vap - attribute values. 1844 * 1845 * RETURN: 0 (always succeeds). 1846 */ 1847 static int 1848 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1849 { 1850 znode_t *zp = VTOZ(vp); 1851 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1852 int error = 0; 1853 uint32_t blksize; 1854 u_longlong_t nblocks; 1855 uint64_t mtime[2], ctime[2], crtime[2], rdev; 1856 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 1857 xoptattr_t *xoap = NULL; 1858 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 1859 sa_bulk_attr_t bulk[4]; 1860 int count = 0; 1861 1862 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 1863 return (error); 1864 1865 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 1866 1867 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 1868 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 1869 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16); 1870 if (vp->v_type == VBLK || vp->v_type == VCHR) 1871 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, 1872 &rdev, 8); 1873 1874 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 1875 zfs_exit(zfsvfs, FTAG); 1876 return (error); 1877 } 1878 1879 /* 1880 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 1881 * Also, if we are the owner don't bother, since owner should 1882 * always be allowed to read basic attributes of file. 1883 */ 1884 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 1885 (vap->va_uid != crgetuid(cr))) { 1886 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 1887 skipaclchk, cr, NULL))) { 1888 zfs_exit(zfsvfs, FTAG); 1889 return (error); 1890 } 1891 } 1892 1893 /* 1894 * Return all attributes. It's cheaper to provide the answer 1895 * than to determine whether we were asked the question. 1896 */ 1897 1898 vap->va_type = IFTOVT(zp->z_mode); 1899 vap->va_mode = zp->z_mode & ~S_IFMT; 1900 vn_fsid(vp, vap); 1901 vap->va_nodeid = zp->z_id; 1902 vap->va_nlink = zp->z_links; 1903 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp) && 1904 zp->z_links < ZFS_LINK_MAX) 1905 vap->va_nlink++; 1906 vap->va_size = zp->z_size; 1907 if (vp->v_type == VBLK || vp->v_type == VCHR) 1908 vap->va_rdev = zfs_cmpldev(rdev); 1909 else 1910 vap->va_rdev = 0; 1911 vap->va_gen = zp->z_gen; 1912 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 1913 vap->va_filerev = zp->z_seq; 1914 1915 /* 1916 * Add in any requested optional attributes and the create time. 1917 * Also set the corresponding bits in the returned attribute bitmap. 1918 */ 1919 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 1920 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 1921 xoap->xoa_archive = 1922 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 1923 XVA_SET_RTN(xvap, XAT_ARCHIVE); 1924 } 1925 1926 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 1927 xoap->xoa_readonly = 1928 ((zp->z_pflags & ZFS_READONLY) != 0); 1929 XVA_SET_RTN(xvap, XAT_READONLY); 1930 } 1931 1932 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 1933 xoap->xoa_system = 1934 ((zp->z_pflags & ZFS_SYSTEM) != 0); 1935 XVA_SET_RTN(xvap, XAT_SYSTEM); 1936 } 1937 1938 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 1939 xoap->xoa_hidden = 1940 ((zp->z_pflags & ZFS_HIDDEN) != 0); 1941 XVA_SET_RTN(xvap, XAT_HIDDEN); 1942 } 1943 1944 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 1945 xoap->xoa_nounlink = 1946 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 1947 XVA_SET_RTN(xvap, XAT_NOUNLINK); 1948 } 1949 1950 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 1951 xoap->xoa_immutable = 1952 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 1953 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 1954 } 1955 1956 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 1957 xoap->xoa_appendonly = 1958 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 1959 XVA_SET_RTN(xvap, XAT_APPENDONLY); 1960 } 1961 1962 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 1963 xoap->xoa_nodump = 1964 ((zp->z_pflags & ZFS_NODUMP) != 0); 1965 XVA_SET_RTN(xvap, XAT_NODUMP); 1966 } 1967 1968 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 1969 xoap->xoa_opaque = 1970 ((zp->z_pflags & ZFS_OPAQUE) != 0); 1971 XVA_SET_RTN(xvap, XAT_OPAQUE); 1972 } 1973 1974 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 1975 xoap->xoa_av_quarantined = 1976 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 1977 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 1978 } 1979 1980 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 1981 xoap->xoa_av_modified = 1982 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 1983 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 1984 } 1985 1986 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 1987 vp->v_type == VREG) { 1988 zfs_sa_get_scanstamp(zp, xvap); 1989 } 1990 1991 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 1992 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 1993 XVA_SET_RTN(xvap, XAT_REPARSE); 1994 } 1995 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 1996 xoap->xoa_generation = zp->z_gen; 1997 XVA_SET_RTN(xvap, XAT_GEN); 1998 } 1999 2000 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2001 xoap->xoa_offline = 2002 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2003 XVA_SET_RTN(xvap, XAT_OFFLINE); 2004 } 2005 2006 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2007 xoap->xoa_sparse = 2008 ((zp->z_pflags & ZFS_SPARSE) != 0); 2009 XVA_SET_RTN(xvap, XAT_SPARSE); 2010 } 2011 2012 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) { 2013 xoap->xoa_projinherit = 2014 ((zp->z_pflags & ZFS_PROJINHERIT) != 0); 2015 XVA_SET_RTN(xvap, XAT_PROJINHERIT); 2016 } 2017 2018 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) { 2019 xoap->xoa_projid = zp->z_projid; 2020 XVA_SET_RTN(xvap, XAT_PROJID); 2021 } 2022 } 2023 2024 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2025 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2026 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2027 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 2028 2029 2030 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 2031 vap->va_blksize = blksize; 2032 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2033 2034 if (zp->z_blksz == 0) { 2035 /* 2036 * Block size hasn't been set; suggest maximal I/O transfers. 2037 */ 2038 vap->va_blksize = zfsvfs->z_max_blksz; 2039 } 2040 2041 zfs_exit(zfsvfs, FTAG); 2042 return (0); 2043 } 2044 2045 /* 2046 * Set the file attributes to the values contained in the 2047 * vattr structure. 2048 * 2049 * IN: zp - znode of file to be modified. 2050 * vap - new attribute values. 2051 * If AT_XVATTR set, then optional attrs are being set 2052 * flags - ATTR_UTIME set if non-default time values provided. 2053 * - ATTR_NOACLCHECK (CIFS context only). 2054 * cr - credentials of caller. 2055 * mnt_ns - Unused on FreeBSD 2056 * 2057 * RETURN: 0 on success, error code on failure. 2058 * 2059 * Timestamps: 2060 * vp - ctime updated, mtime updated if size changed. 2061 */ 2062 int 2063 zfs_setattr(znode_t *zp, vattr_t *vap, int flags, cred_t *cr, zidmap_t *mnt_ns) 2064 { 2065 vnode_t *vp = ZTOV(zp); 2066 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2067 objset_t *os; 2068 zilog_t *zilog; 2069 dmu_tx_t *tx; 2070 vattr_t oldva; 2071 xvattr_t tmpxvattr; 2072 uint_t mask = vap->va_mask; 2073 uint_t saved_mask = 0; 2074 uint64_t saved_mode; 2075 int trim_mask = 0; 2076 uint64_t new_mode; 2077 uint64_t new_uid, new_gid; 2078 uint64_t xattr_obj; 2079 uint64_t mtime[2], ctime[2]; 2080 uint64_t projid = ZFS_INVALID_PROJID; 2081 znode_t *attrzp; 2082 int need_policy = FALSE; 2083 int err, err2; 2084 zfs_fuid_info_t *fuidp = NULL; 2085 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2086 xoptattr_t *xoap; 2087 zfs_acl_t *aclp; 2088 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2089 boolean_t fuid_dirtied = B_FALSE; 2090 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 2091 int count = 0, xattr_count = 0; 2092 2093 if (mask == 0) 2094 return (0); 2095 2096 if (mask & AT_NOSET) 2097 return (SET_ERROR(EINVAL)); 2098 2099 if ((err = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 2100 return (err); 2101 2102 os = zfsvfs->z_os; 2103 zilog = zfsvfs->z_log; 2104 2105 /* 2106 * Make sure that if we have ephemeral uid/gid or xvattr specified 2107 * that file system is at proper version level 2108 */ 2109 2110 if (zfsvfs->z_use_fuids == B_FALSE && 2111 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2112 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2113 (mask & AT_XVATTR))) { 2114 zfs_exit(zfsvfs, FTAG); 2115 return (SET_ERROR(EINVAL)); 2116 } 2117 2118 if (mask & AT_SIZE && vp->v_type == VDIR) { 2119 zfs_exit(zfsvfs, FTAG); 2120 return (SET_ERROR(EISDIR)); 2121 } 2122 2123 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 2124 zfs_exit(zfsvfs, FTAG); 2125 return (SET_ERROR(EINVAL)); 2126 } 2127 2128 /* 2129 * If this is an xvattr_t, then get a pointer to the structure of 2130 * optional attributes. If this is NULL, then we have a vattr_t. 2131 */ 2132 xoap = xva_getxoptattr(xvap); 2133 2134 xva_init(&tmpxvattr); 2135 2136 /* 2137 * Immutable files can only alter immutable bit and atime 2138 */ 2139 if ((zp->z_pflags & ZFS_IMMUTABLE) && 2140 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2141 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 2142 zfs_exit(zfsvfs, FTAG); 2143 return (SET_ERROR(EPERM)); 2144 } 2145 2146 /* 2147 * Note: ZFS_READONLY is handled in zfs_zaccess_common. 2148 */ 2149 2150 /* 2151 * Verify timestamps doesn't overflow 32 bits. 2152 * ZFS can handle large timestamps, but 32bit syscalls can't 2153 * handle times greater than 2039. This check should be removed 2154 * once large timestamps are fully supported. 2155 */ 2156 if (mask & (AT_ATIME | AT_MTIME)) { 2157 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 2158 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 2159 zfs_exit(zfsvfs, FTAG); 2160 return (SET_ERROR(EOVERFLOW)); 2161 } 2162 } 2163 if (xoap != NULL && (mask & AT_XVATTR)) { 2164 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME) && 2165 TIMESPEC_OVERFLOW(&vap->va_birthtime)) { 2166 zfs_exit(zfsvfs, FTAG); 2167 return (SET_ERROR(EOVERFLOW)); 2168 } 2169 2170 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) { 2171 if (!dmu_objset_projectquota_enabled(os) || 2172 (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode))) { 2173 zfs_exit(zfsvfs, FTAG); 2174 return (SET_ERROR(EOPNOTSUPP)); 2175 } 2176 2177 projid = xoap->xoa_projid; 2178 if (unlikely(projid == ZFS_INVALID_PROJID)) { 2179 zfs_exit(zfsvfs, FTAG); 2180 return (SET_ERROR(EINVAL)); 2181 } 2182 2183 if (projid == zp->z_projid && zp->z_pflags & ZFS_PROJID) 2184 projid = ZFS_INVALID_PROJID; 2185 else 2186 need_policy = TRUE; 2187 } 2188 2189 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT) && 2190 (xoap->xoa_projinherit != 2191 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) && 2192 (!dmu_objset_projectquota_enabled(os) || 2193 (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode)))) { 2194 zfs_exit(zfsvfs, FTAG); 2195 return (SET_ERROR(EOPNOTSUPP)); 2196 } 2197 } 2198 2199 attrzp = NULL; 2200 aclp = NULL; 2201 2202 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2203 zfs_exit(zfsvfs, FTAG); 2204 return (SET_ERROR(EROFS)); 2205 } 2206 2207 /* 2208 * First validate permissions 2209 */ 2210 2211 if (mask & AT_SIZE) { 2212 /* 2213 * XXX - Note, we are not providing any open 2214 * mode flags here (like FNDELAY), so we may 2215 * block if there are locks present... this 2216 * should be addressed in openat(). 2217 */ 2218 /* XXX - would it be OK to generate a log record here? */ 2219 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2220 if (err) { 2221 zfs_exit(zfsvfs, FTAG); 2222 return (err); 2223 } 2224 } 2225 2226 if (mask & (AT_ATIME|AT_MTIME) || 2227 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2228 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2229 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2230 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 2231 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 2232 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2233 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 2234 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2235 skipaclchk, cr, mnt_ns); 2236 } 2237 2238 if (mask & (AT_UID|AT_GID)) { 2239 int idmask = (mask & (AT_UID|AT_GID)); 2240 int take_owner; 2241 int take_group; 2242 2243 /* 2244 * NOTE: even if a new mode is being set, 2245 * we may clear S_ISUID/S_ISGID bits. 2246 */ 2247 2248 if (!(mask & AT_MODE)) 2249 vap->va_mode = zp->z_mode; 2250 2251 /* 2252 * Take ownership or chgrp to group we are a member of 2253 */ 2254 2255 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2256 take_group = (mask & AT_GID) && 2257 zfs_groupmember(zfsvfs, vap->va_gid, cr); 2258 2259 /* 2260 * If both AT_UID and AT_GID are set then take_owner and 2261 * take_group must both be set in order to allow taking 2262 * ownership. 2263 * 2264 * Otherwise, send the check through secpolicy_vnode_setattr() 2265 * 2266 */ 2267 2268 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2269 ((idmask == AT_UID) && take_owner) || 2270 ((idmask == AT_GID) && take_group)) { 2271 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 2272 skipaclchk, cr, mnt_ns) == 0) { 2273 /* 2274 * Remove setuid/setgid for non-privileged users 2275 */ 2276 secpolicy_setid_clear(vap, vp, cr); 2277 trim_mask = (mask & (AT_UID|AT_GID)); 2278 } else { 2279 need_policy = TRUE; 2280 } 2281 } else { 2282 need_policy = TRUE; 2283 } 2284 } 2285 2286 oldva.va_mode = zp->z_mode; 2287 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 2288 if (mask & AT_XVATTR) { 2289 /* 2290 * Update xvattr mask to include only those attributes 2291 * that are actually changing. 2292 * 2293 * the bits will be restored prior to actually setting 2294 * the attributes so the caller thinks they were set. 2295 */ 2296 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2297 if (xoap->xoa_appendonly != 2298 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 2299 need_policy = TRUE; 2300 } else { 2301 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 2302 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 2303 } 2304 } 2305 2306 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) { 2307 if (xoap->xoa_projinherit != 2308 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) { 2309 need_policy = TRUE; 2310 } else { 2311 XVA_CLR_REQ(xvap, XAT_PROJINHERIT); 2312 XVA_SET_REQ(&tmpxvattr, XAT_PROJINHERIT); 2313 } 2314 } 2315 2316 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2317 if (xoap->xoa_nounlink != 2318 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 2319 need_policy = TRUE; 2320 } else { 2321 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 2322 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 2323 } 2324 } 2325 2326 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2327 if (xoap->xoa_immutable != 2328 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 2329 need_policy = TRUE; 2330 } else { 2331 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 2332 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 2333 } 2334 } 2335 2336 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2337 if (xoap->xoa_nodump != 2338 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 2339 need_policy = TRUE; 2340 } else { 2341 XVA_CLR_REQ(xvap, XAT_NODUMP); 2342 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 2343 } 2344 } 2345 2346 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2347 if (xoap->xoa_av_modified != 2348 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 2349 need_policy = TRUE; 2350 } else { 2351 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 2352 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 2353 } 2354 } 2355 2356 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2357 if ((vp->v_type != VREG && 2358 xoap->xoa_av_quarantined) || 2359 xoap->xoa_av_quarantined != 2360 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 2361 need_policy = TRUE; 2362 } else { 2363 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 2364 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 2365 } 2366 } 2367 2368 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2369 zfs_exit(zfsvfs, FTAG); 2370 return (SET_ERROR(EPERM)); 2371 } 2372 2373 if (need_policy == FALSE && 2374 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 2375 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 2376 need_policy = TRUE; 2377 } 2378 } 2379 2380 if (mask & AT_MODE) { 2381 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr, 2382 mnt_ns) == 0) { 2383 err = secpolicy_setid_setsticky_clear(vp, vap, 2384 &oldva, cr); 2385 if (err) { 2386 zfs_exit(zfsvfs, FTAG); 2387 return (err); 2388 } 2389 trim_mask |= AT_MODE; 2390 } else { 2391 need_policy = TRUE; 2392 } 2393 } 2394 2395 if (need_policy) { 2396 /* 2397 * If trim_mask is set then take ownership 2398 * has been granted or write_acl is present and user 2399 * has the ability to modify mode. In that case remove 2400 * UID|GID and or MODE from mask so that 2401 * secpolicy_vnode_setattr() doesn't revoke it. 2402 */ 2403 2404 if (trim_mask) { 2405 saved_mask = vap->va_mask; 2406 vap->va_mask &= ~trim_mask; 2407 if (trim_mask & AT_MODE) { 2408 /* 2409 * Save the mode, as secpolicy_vnode_setattr() 2410 * will overwrite it with ova.va_mode. 2411 */ 2412 saved_mode = vap->va_mode; 2413 } 2414 } 2415 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2416 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 2417 if (err) { 2418 zfs_exit(zfsvfs, FTAG); 2419 return (err); 2420 } 2421 2422 if (trim_mask) { 2423 vap->va_mask |= saved_mask; 2424 if (trim_mask & AT_MODE) { 2425 /* 2426 * Recover the mode after 2427 * secpolicy_vnode_setattr(). 2428 */ 2429 vap->va_mode = saved_mode; 2430 } 2431 } 2432 } 2433 2434 /* 2435 * secpolicy_vnode_setattr, or take ownership may have 2436 * changed va_mask 2437 */ 2438 mask = vap->va_mask; 2439 2440 if ((mask & (AT_UID | AT_GID)) || projid != ZFS_INVALID_PROJID) { 2441 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 2442 &xattr_obj, sizeof (xattr_obj)); 2443 2444 if (err == 0 && xattr_obj) { 2445 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 2446 if (err == 0) { 2447 err = vn_lock(ZTOV(attrzp), LK_EXCLUSIVE); 2448 if (err != 0) 2449 vrele(ZTOV(attrzp)); 2450 } 2451 if (err) 2452 goto out2; 2453 } 2454 if (mask & AT_UID) { 2455 new_uid = zfs_fuid_create(zfsvfs, 2456 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 2457 if (new_uid != zp->z_uid && 2458 zfs_id_overquota(zfsvfs, DMU_USERUSED_OBJECT, 2459 new_uid)) { 2460 if (attrzp) 2461 vput(ZTOV(attrzp)); 2462 err = SET_ERROR(EDQUOT); 2463 goto out2; 2464 } 2465 } 2466 2467 if (mask & AT_GID) { 2468 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 2469 cr, ZFS_GROUP, &fuidp); 2470 if (new_gid != zp->z_gid && 2471 zfs_id_overquota(zfsvfs, DMU_GROUPUSED_OBJECT, 2472 new_gid)) { 2473 if (attrzp) 2474 vput(ZTOV(attrzp)); 2475 err = SET_ERROR(EDQUOT); 2476 goto out2; 2477 } 2478 } 2479 2480 if (projid != ZFS_INVALID_PROJID && 2481 zfs_id_overquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid)) { 2482 if (attrzp) 2483 vput(ZTOV(attrzp)); 2484 err = SET_ERROR(EDQUOT); 2485 goto out2; 2486 } 2487 } 2488 tx = dmu_tx_create(os); 2489 2490 if (mask & AT_MODE) { 2491 uint64_t pmode = zp->z_mode; 2492 uint64_t acl_obj; 2493 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2494 2495 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && 2496 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 2497 err = SET_ERROR(EPERM); 2498 goto out; 2499 } 2500 2501 if ((err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))) 2502 goto out; 2503 2504 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 2505 /* 2506 * Are we upgrading ACL from old V0 format 2507 * to V1 format? 2508 */ 2509 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 2510 zfs_znode_acl_version(zp) == 2511 ZFS_ACL_VERSION_INITIAL) { 2512 dmu_tx_hold_free(tx, acl_obj, 0, 2513 DMU_OBJECT_END); 2514 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2515 0, aclp->z_acl_bytes); 2516 } else { 2517 dmu_tx_hold_write(tx, acl_obj, 0, 2518 aclp->z_acl_bytes); 2519 } 2520 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2521 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2522 0, aclp->z_acl_bytes); 2523 } 2524 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2525 } else { 2526 if (((mask & AT_XVATTR) && 2527 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) || 2528 (projid != ZFS_INVALID_PROJID && 2529 !(zp->z_pflags & ZFS_PROJID))) 2530 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2531 else 2532 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2533 } 2534 2535 if (attrzp) { 2536 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 2537 } 2538 2539 fuid_dirtied = zfsvfs->z_fuid_dirty; 2540 if (fuid_dirtied) 2541 zfs_fuid_txhold(zfsvfs, tx); 2542 2543 zfs_sa_upgrade_txholds(tx, zp); 2544 2545 err = dmu_tx_assign(tx, TXG_WAIT); 2546 if (err) 2547 goto out; 2548 2549 count = 0; 2550 /* 2551 * Set each attribute requested. 2552 * We group settings according to the locks they need to acquire. 2553 * 2554 * Note: you cannot set ctime directly, although it will be 2555 * updated as a side-effect of calling this function. 2556 */ 2557 2558 if (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID)) { 2559 /* 2560 * For the existed object that is upgraded from old system, 2561 * its on-disk layout has no slot for the project ID attribute. 2562 * But quota accounting logic needs to access related slots by 2563 * offset directly. So we need to adjust old objects' layout 2564 * to make the project ID to some unified and fixed offset. 2565 */ 2566 if (attrzp) 2567 err = sa_add_projid(attrzp->z_sa_hdl, tx, projid); 2568 if (err == 0) 2569 err = sa_add_projid(zp->z_sa_hdl, tx, projid); 2570 2571 if (unlikely(err == EEXIST)) 2572 err = 0; 2573 else if (err != 0) 2574 goto out; 2575 else 2576 projid = ZFS_INVALID_PROJID; 2577 } 2578 2579 if (mask & (AT_UID|AT_GID|AT_MODE)) 2580 mutex_enter(&zp->z_acl_lock); 2581 2582 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 2583 &zp->z_pflags, sizeof (zp->z_pflags)); 2584 2585 if (attrzp) { 2586 if (mask & (AT_UID|AT_GID|AT_MODE)) 2587 mutex_enter(&attrzp->z_acl_lock); 2588 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2589 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 2590 sizeof (attrzp->z_pflags)); 2591 if (projid != ZFS_INVALID_PROJID) { 2592 attrzp->z_projid = projid; 2593 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2594 SA_ZPL_PROJID(zfsvfs), NULL, &attrzp->z_projid, 2595 sizeof (attrzp->z_projid)); 2596 } 2597 } 2598 2599 if (mask & (AT_UID|AT_GID)) { 2600 2601 if (mask & AT_UID) { 2602 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 2603 &new_uid, sizeof (new_uid)); 2604 zp->z_uid = new_uid; 2605 if (attrzp) { 2606 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2607 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 2608 sizeof (new_uid)); 2609 attrzp->z_uid = new_uid; 2610 } 2611 } 2612 2613 if (mask & AT_GID) { 2614 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 2615 NULL, &new_gid, sizeof (new_gid)); 2616 zp->z_gid = new_gid; 2617 if (attrzp) { 2618 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2619 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 2620 sizeof (new_gid)); 2621 attrzp->z_gid = new_gid; 2622 } 2623 } 2624 if (!(mask & AT_MODE)) { 2625 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 2626 NULL, &new_mode, sizeof (new_mode)); 2627 new_mode = zp->z_mode; 2628 } 2629 err = zfs_acl_chown_setattr(zp); 2630 ASSERT0(err); 2631 if (attrzp) { 2632 vn_seqc_write_begin(ZTOV(attrzp)); 2633 err = zfs_acl_chown_setattr(attrzp); 2634 vn_seqc_write_end(ZTOV(attrzp)); 2635 ASSERT0(err); 2636 } 2637 } 2638 2639 if (mask & AT_MODE) { 2640 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 2641 &new_mode, sizeof (new_mode)); 2642 zp->z_mode = new_mode; 2643 ASSERT3P(aclp, !=, NULL); 2644 err = zfs_aclset_common(zp, aclp, cr, tx); 2645 ASSERT0(err); 2646 if (zp->z_acl_cached) 2647 zfs_acl_free(zp->z_acl_cached); 2648 zp->z_acl_cached = aclp; 2649 aclp = NULL; 2650 } 2651 2652 2653 if (mask & AT_ATIME) { 2654 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 2655 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 2656 &zp->z_atime, sizeof (zp->z_atime)); 2657 } 2658 2659 if (mask & AT_MTIME) { 2660 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 2661 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 2662 mtime, sizeof (mtime)); 2663 } 2664 2665 if (projid != ZFS_INVALID_PROJID) { 2666 zp->z_projid = projid; 2667 SA_ADD_BULK_ATTR(bulk, count, 2668 SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid, 2669 sizeof (zp->z_projid)); 2670 } 2671 2672 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 2673 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 2674 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 2675 NULL, mtime, sizeof (mtime)); 2676 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 2677 &ctime, sizeof (ctime)); 2678 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime); 2679 } else if (mask != 0) { 2680 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 2681 &ctime, sizeof (ctime)); 2682 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime); 2683 if (attrzp) { 2684 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2685 SA_ZPL_CTIME(zfsvfs), NULL, 2686 &ctime, sizeof (ctime)); 2687 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 2688 mtime, ctime); 2689 } 2690 } 2691 2692 /* 2693 * Do this after setting timestamps to prevent timestamp 2694 * update from toggling bit 2695 */ 2696 2697 if (xoap && (mask & AT_XVATTR)) { 2698 2699 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) 2700 xoap->xoa_createtime = vap->va_birthtime; 2701 /* 2702 * restore trimmed off masks 2703 * so that return masks can be set for caller. 2704 */ 2705 2706 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 2707 XVA_SET_REQ(xvap, XAT_APPENDONLY); 2708 } 2709 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 2710 XVA_SET_REQ(xvap, XAT_NOUNLINK); 2711 } 2712 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 2713 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 2714 } 2715 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 2716 XVA_SET_REQ(xvap, XAT_NODUMP); 2717 } 2718 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 2719 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 2720 } 2721 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 2722 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 2723 } 2724 if (XVA_ISSET_REQ(&tmpxvattr, XAT_PROJINHERIT)) { 2725 XVA_SET_REQ(xvap, XAT_PROJINHERIT); 2726 } 2727 2728 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 2729 ASSERT3S(vp->v_type, ==, VREG); 2730 2731 zfs_xvattr_set(zp, xvap, tx); 2732 } 2733 2734 if (fuid_dirtied) 2735 zfs_fuid_sync(zfsvfs, tx); 2736 2737 if (mask != 0) 2738 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 2739 2740 if (mask & (AT_UID|AT_GID|AT_MODE)) 2741 mutex_exit(&zp->z_acl_lock); 2742 2743 if (attrzp) { 2744 if (mask & (AT_UID|AT_GID|AT_MODE)) 2745 mutex_exit(&attrzp->z_acl_lock); 2746 } 2747 out: 2748 if (err == 0 && attrzp) { 2749 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 2750 xattr_count, tx); 2751 ASSERT0(err2); 2752 } 2753 2754 if (attrzp) 2755 vput(ZTOV(attrzp)); 2756 2757 if (aclp) 2758 zfs_acl_free(aclp); 2759 2760 if (fuidp) { 2761 zfs_fuid_info_free(fuidp); 2762 fuidp = NULL; 2763 } 2764 2765 if (err) { 2766 dmu_tx_abort(tx); 2767 } else { 2768 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 2769 dmu_tx_commit(tx); 2770 } 2771 2772 out2: 2773 if (os->os_sync == ZFS_SYNC_ALWAYS) 2774 zil_commit(zilog, 0); 2775 2776 zfs_exit(zfsvfs, FTAG); 2777 return (err); 2778 } 2779 2780 /* 2781 * Look up the directory entries corresponding to the source and target 2782 * directory/name pairs. 2783 */ 2784 static int 2785 zfs_rename_relock_lookup(znode_t *sdzp, const struct componentname *scnp, 2786 znode_t **szpp, znode_t *tdzp, const struct componentname *tcnp, 2787 znode_t **tzpp) 2788 { 2789 zfsvfs_t *zfsvfs; 2790 znode_t *szp, *tzp; 2791 int error; 2792 2793 /* 2794 * Before using sdzp and tdzp we must ensure that they are live. 2795 * As a porting legacy from illumos we have two things to worry 2796 * about. One is typical for FreeBSD and it is that the vnode is 2797 * not reclaimed (doomed). The other is that the znode is live. 2798 * The current code can invalidate the znode without acquiring the 2799 * corresponding vnode lock if the object represented by the znode 2800 * and vnode is no longer valid after a rollback or receive operation. 2801 * z_teardown_lock hidden behind zfs_enter and zfs_exit is the lock 2802 * that protects the znodes from the invalidation. 2803 */ 2804 zfsvfs = sdzp->z_zfsvfs; 2805 ASSERT3P(zfsvfs, ==, tdzp->z_zfsvfs); 2806 if ((error = zfs_enter_verify_zp(zfsvfs, sdzp, FTAG)) != 0) 2807 return (error); 2808 if ((error = zfs_verify_zp(tdzp)) != 0) { 2809 zfs_exit(zfsvfs, FTAG); 2810 return (error); 2811 } 2812 2813 /* 2814 * Re-resolve svp to be certain it still exists and fetch the 2815 * correct vnode. 2816 */ 2817 error = zfs_dirent_lookup(sdzp, scnp->cn_nameptr, &szp, ZEXISTS); 2818 if (error != 0) { 2819 /* Source entry invalid or not there. */ 2820 if ((scnp->cn_flags & ISDOTDOT) != 0 || 2821 (scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.')) 2822 error = SET_ERROR(EINVAL); 2823 goto out; 2824 } 2825 *szpp = szp; 2826 2827 /* 2828 * Re-resolve tvp, if it disappeared we just carry on. 2829 */ 2830 error = zfs_dirent_lookup(tdzp, tcnp->cn_nameptr, &tzp, 0); 2831 if (error != 0) { 2832 vrele(ZTOV(szp)); 2833 if ((tcnp->cn_flags & ISDOTDOT) != 0) 2834 error = SET_ERROR(EINVAL); 2835 goto out; 2836 } 2837 *tzpp = tzp; 2838 out: 2839 zfs_exit(zfsvfs, FTAG); 2840 return (error); 2841 } 2842 2843 /* 2844 * We acquire all but fdvp locks using non-blocking acquisitions. If we 2845 * fail to acquire any lock in the path we will drop all held locks, 2846 * acquire the new lock in a blocking fashion, and then release it and 2847 * restart the rename. This acquire/release step ensures that we do not 2848 * spin on a lock waiting for release. On error release all vnode locks 2849 * and decrement references the way tmpfs_rename() would do. 2850 */ 2851 static int 2852 zfs_rename_relock(struct vnode *sdvp, struct vnode **svpp, 2853 struct vnode *tdvp, struct vnode **tvpp, 2854 const struct componentname *scnp, const struct componentname *tcnp) 2855 { 2856 struct vnode *nvp, *svp, *tvp; 2857 znode_t *sdzp, *tdzp, *szp, *tzp; 2858 int error; 2859 2860 VOP_UNLOCK(tdvp); 2861 if (*tvpp != NULL && *tvpp != tdvp) 2862 VOP_UNLOCK(*tvpp); 2863 2864 relock: 2865 error = vn_lock(sdvp, LK_EXCLUSIVE); 2866 if (error) 2867 goto out; 2868 error = vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT); 2869 if (error != 0) { 2870 VOP_UNLOCK(sdvp); 2871 if (error != EBUSY) 2872 goto out; 2873 error = vn_lock(tdvp, LK_EXCLUSIVE); 2874 if (error) 2875 goto out; 2876 VOP_UNLOCK(tdvp); 2877 goto relock; 2878 } 2879 tdzp = VTOZ(tdvp); 2880 sdzp = VTOZ(sdvp); 2881 2882 error = zfs_rename_relock_lookup(sdzp, scnp, &szp, tdzp, tcnp, &tzp); 2883 if (error != 0) { 2884 VOP_UNLOCK(sdvp); 2885 VOP_UNLOCK(tdvp); 2886 goto out; 2887 } 2888 svp = ZTOV(szp); 2889 tvp = tzp != NULL ? ZTOV(tzp) : NULL; 2890 2891 /* 2892 * Now try acquire locks on svp and tvp. 2893 */ 2894 nvp = svp; 2895 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT); 2896 if (error != 0) { 2897 VOP_UNLOCK(sdvp); 2898 VOP_UNLOCK(tdvp); 2899 if (tvp != NULL) 2900 vrele(tvp); 2901 if (error != EBUSY) { 2902 vrele(nvp); 2903 goto out; 2904 } 2905 error = vn_lock(nvp, LK_EXCLUSIVE); 2906 if (error != 0) { 2907 vrele(nvp); 2908 goto out; 2909 } 2910 VOP_UNLOCK(nvp); 2911 /* 2912 * Concurrent rename race. 2913 * XXX ? 2914 */ 2915 if (nvp == tdvp) { 2916 vrele(nvp); 2917 error = SET_ERROR(EINVAL); 2918 goto out; 2919 } 2920 vrele(*svpp); 2921 *svpp = nvp; 2922 goto relock; 2923 } 2924 vrele(*svpp); 2925 *svpp = nvp; 2926 2927 if (*tvpp != NULL) 2928 vrele(*tvpp); 2929 *tvpp = NULL; 2930 if (tvp != NULL) { 2931 nvp = tvp; 2932 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT); 2933 if (error != 0) { 2934 VOP_UNLOCK(sdvp); 2935 VOP_UNLOCK(tdvp); 2936 VOP_UNLOCK(*svpp); 2937 if (error != EBUSY) { 2938 vrele(nvp); 2939 goto out; 2940 } 2941 error = vn_lock(nvp, LK_EXCLUSIVE); 2942 if (error != 0) { 2943 vrele(nvp); 2944 goto out; 2945 } 2946 vput(nvp); 2947 goto relock; 2948 } 2949 *tvpp = nvp; 2950 } 2951 2952 return (0); 2953 2954 out: 2955 return (error); 2956 } 2957 2958 /* 2959 * Note that we must use VRELE_ASYNC in this function as it walks 2960 * up the directory tree and vrele may need to acquire an exclusive 2961 * lock if a last reference to a vnode is dropped. 2962 */ 2963 static int 2964 zfs_rename_check(znode_t *szp, znode_t *sdzp, znode_t *tdzp) 2965 { 2966 zfsvfs_t *zfsvfs; 2967 znode_t *zp, *zp1; 2968 uint64_t parent; 2969 int error; 2970 2971 zfsvfs = tdzp->z_zfsvfs; 2972 if (tdzp == szp) 2973 return (SET_ERROR(EINVAL)); 2974 if (tdzp == sdzp) 2975 return (0); 2976 if (tdzp->z_id == zfsvfs->z_root) 2977 return (0); 2978 zp = tdzp; 2979 for (;;) { 2980 ASSERT(!zp->z_unlinked); 2981 if ((error = sa_lookup(zp->z_sa_hdl, 2982 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) 2983 break; 2984 2985 if (parent == szp->z_id) { 2986 error = SET_ERROR(EINVAL); 2987 break; 2988 } 2989 if (parent == zfsvfs->z_root) 2990 break; 2991 if (parent == sdzp->z_id) 2992 break; 2993 2994 error = zfs_zget(zfsvfs, parent, &zp1); 2995 if (error != 0) 2996 break; 2997 2998 if (zp != tdzp) 2999 VN_RELE_ASYNC(ZTOV(zp), 3000 dsl_pool_zrele_taskq( 3001 dmu_objset_pool(zfsvfs->z_os))); 3002 zp = zp1; 3003 } 3004 3005 if (error == ENOTDIR) 3006 panic("checkpath: .. not a directory\n"); 3007 if (zp != tdzp) 3008 VN_RELE_ASYNC(ZTOV(zp), 3009 dsl_pool_zrele_taskq(dmu_objset_pool(zfsvfs->z_os))); 3010 return (error); 3011 } 3012 3013 static int 3014 zfs_do_rename_impl(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, 3015 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, 3016 cred_t *cr); 3017 3018 /* 3019 * Move an entry from the provided source directory to the target 3020 * directory. Change the entry name as indicated. 3021 * 3022 * IN: sdvp - Source directory containing the "old entry". 3023 * scnp - Old entry name. 3024 * tdvp - Target directory to contain the "new entry". 3025 * tcnp - New entry name. 3026 * cr - credentials of caller. 3027 * INOUT: svpp - Source file 3028 * tvpp - Target file, may point to NULL initially 3029 * 3030 * RETURN: 0 on success, error code on failure. 3031 * 3032 * Timestamps: 3033 * sdvp,tdvp - ctime|mtime updated 3034 */ 3035 static int 3036 zfs_do_rename(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, 3037 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, 3038 cred_t *cr) 3039 { 3040 int error; 3041 3042 ASSERT_VOP_ELOCKED(tdvp, __func__); 3043 if (*tvpp != NULL) 3044 ASSERT_VOP_ELOCKED(*tvpp, __func__); 3045 3046 /* Reject renames across filesystems. */ 3047 if ((*svpp)->v_mount != tdvp->v_mount || 3048 ((*tvpp) != NULL && (*svpp)->v_mount != (*tvpp)->v_mount)) { 3049 error = SET_ERROR(EXDEV); 3050 goto out; 3051 } 3052 3053 if (zfsctl_is_node(tdvp)) { 3054 error = SET_ERROR(EXDEV); 3055 goto out; 3056 } 3057 3058 /* 3059 * Lock all four vnodes to ensure safety and semantics of renaming. 3060 */ 3061 error = zfs_rename_relock(sdvp, svpp, tdvp, tvpp, scnp, tcnp); 3062 if (error != 0) { 3063 /* no vnodes are locked in the case of error here */ 3064 return (error); 3065 } 3066 3067 error = zfs_do_rename_impl(sdvp, svpp, scnp, tdvp, tvpp, tcnp, cr); 3068 VOP_UNLOCK(sdvp); 3069 VOP_UNLOCK(*svpp); 3070 out: 3071 if (*tvpp != NULL) 3072 VOP_UNLOCK(*tvpp); 3073 if (tdvp != *tvpp) 3074 VOP_UNLOCK(tdvp); 3075 3076 return (error); 3077 } 3078 3079 static int 3080 zfs_do_rename_impl(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, 3081 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, 3082 cred_t *cr) 3083 { 3084 dmu_tx_t *tx; 3085 zfsvfs_t *zfsvfs; 3086 zilog_t *zilog; 3087 znode_t *tdzp, *sdzp, *tzp, *szp; 3088 const char *snm = scnp->cn_nameptr; 3089 const char *tnm = tcnp->cn_nameptr; 3090 int error; 3091 3092 tdzp = VTOZ(tdvp); 3093 sdzp = VTOZ(sdvp); 3094 zfsvfs = tdzp->z_zfsvfs; 3095 3096 if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0) 3097 return (error); 3098 if ((error = zfs_verify_zp(sdzp)) != 0) { 3099 zfs_exit(zfsvfs, FTAG); 3100 return (error); 3101 } 3102 zilog = zfsvfs->z_log; 3103 3104 if (zfsvfs->z_utf8 && u8_validate(tnm, 3105 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3106 error = SET_ERROR(EILSEQ); 3107 goto out; 3108 } 3109 3110 /* If source and target are the same file, there is nothing to do. */ 3111 if ((*svpp) == (*tvpp)) { 3112 error = 0; 3113 goto out; 3114 } 3115 3116 if (((*svpp)->v_type == VDIR && (*svpp)->v_mountedhere != NULL) || 3117 ((*tvpp) != NULL && (*tvpp)->v_type == VDIR && 3118 (*tvpp)->v_mountedhere != NULL)) { 3119 error = SET_ERROR(EXDEV); 3120 goto out; 3121 } 3122 3123 szp = VTOZ(*svpp); 3124 if ((error = zfs_verify_zp(szp)) != 0) { 3125 zfs_exit(zfsvfs, FTAG); 3126 return (error); 3127 } 3128 tzp = *tvpp == NULL ? NULL : VTOZ(*tvpp); 3129 if (tzp != NULL) { 3130 if ((error = zfs_verify_zp(tzp)) != 0) { 3131 zfs_exit(zfsvfs, FTAG); 3132 return (error); 3133 } 3134 } 3135 3136 /* 3137 * This is to prevent the creation of links into attribute space 3138 * by renaming a linked file into/outof an attribute directory. 3139 * See the comment in zfs_link() for why this is considered bad. 3140 */ 3141 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3142 error = SET_ERROR(EINVAL); 3143 goto out; 3144 } 3145 3146 /* 3147 * If we are using project inheritance, means if the directory has 3148 * ZFS_PROJINHERIT set, then its descendant directories will inherit 3149 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under 3150 * such case, we only allow renames into our tree when the project 3151 * IDs are the same. 3152 */ 3153 if (tdzp->z_pflags & ZFS_PROJINHERIT && 3154 tdzp->z_projid != szp->z_projid) { 3155 error = SET_ERROR(EXDEV); 3156 goto out; 3157 } 3158 3159 /* 3160 * Must have write access at the source to remove the old entry 3161 * and write access at the target to create the new entry. 3162 * Note that if target and source are the same, this can be 3163 * done in a single check. 3164 */ 3165 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr, NULL))) 3166 goto out; 3167 3168 if ((*svpp)->v_type == VDIR) { 3169 /* 3170 * Avoid ".", "..", and aliases of "." for obvious reasons. 3171 */ 3172 if ((scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.') || 3173 sdzp == szp || 3174 (scnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) { 3175 error = EINVAL; 3176 goto out; 3177 } 3178 3179 /* 3180 * Check to make sure rename is valid. 3181 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3182 */ 3183 if ((error = zfs_rename_check(szp, sdzp, tdzp))) 3184 goto out; 3185 } 3186 3187 /* 3188 * Does target exist? 3189 */ 3190 if (tzp) { 3191 /* 3192 * Source and target must be the same type. 3193 */ 3194 if ((*svpp)->v_type == VDIR) { 3195 if ((*tvpp)->v_type != VDIR) { 3196 error = SET_ERROR(ENOTDIR); 3197 goto out; 3198 } else { 3199 cache_purge(tdvp); 3200 if (sdvp != tdvp) 3201 cache_purge(sdvp); 3202 } 3203 } else { 3204 if ((*tvpp)->v_type == VDIR) { 3205 error = SET_ERROR(EISDIR); 3206 goto out; 3207 } 3208 } 3209 } 3210 3211 vn_seqc_write_begin(*svpp); 3212 vn_seqc_write_begin(sdvp); 3213 if (*tvpp != NULL) 3214 vn_seqc_write_begin(*tvpp); 3215 if (tdvp != *tvpp) 3216 vn_seqc_write_begin(tdvp); 3217 3218 vnevent_rename_src(*svpp, sdvp, scnp->cn_nameptr, ct); 3219 if (tzp) 3220 vnevent_rename_dest(*tvpp, tdvp, tnm, ct); 3221 3222 /* 3223 * notify the target directory if it is not the same 3224 * as source directory. 3225 */ 3226 if (tdvp != sdvp) { 3227 vnevent_rename_dest_dir(tdvp, ct); 3228 } 3229 3230 tx = dmu_tx_create(zfsvfs->z_os); 3231 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3232 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 3233 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3234 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3235 if (sdzp != tdzp) { 3236 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 3237 zfs_sa_upgrade_txholds(tx, tdzp); 3238 } 3239 if (tzp) { 3240 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 3241 zfs_sa_upgrade_txholds(tx, tzp); 3242 } 3243 3244 zfs_sa_upgrade_txholds(tx, szp); 3245 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3246 error = dmu_tx_assign(tx, TXG_WAIT); 3247 if (error) { 3248 dmu_tx_abort(tx); 3249 goto out_seq; 3250 } 3251 3252 if (tzp) /* Attempt to remove the existing target */ 3253 error = zfs_link_destroy(tdzp, tnm, tzp, tx, 0, NULL); 3254 3255 if (error == 0) { 3256 error = zfs_link_create(tdzp, tnm, szp, tx, ZRENAMING); 3257 if (error == 0) { 3258 szp->z_pflags |= ZFS_AV_MODIFIED; 3259 3260 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 3261 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 3262 ASSERT0(error); 3263 3264 error = zfs_link_destroy(sdzp, snm, szp, tx, ZRENAMING, 3265 NULL); 3266 if (error == 0) { 3267 zfs_log_rename(zilog, tx, TX_RENAME, sdzp, 3268 snm, tdzp, tnm, szp); 3269 } else { 3270 /* 3271 * At this point, we have successfully created 3272 * the target name, but have failed to remove 3273 * the source name. Since the create was done 3274 * with the ZRENAMING flag, there are 3275 * complications; for one, the link count is 3276 * wrong. The easiest way to deal with this 3277 * is to remove the newly created target, and 3278 * return the original error. This must 3279 * succeed; fortunately, it is very unlikely to 3280 * fail, since we just created it. 3281 */ 3282 VERIFY0(zfs_link_destroy(tdzp, tnm, szp, tx, 3283 ZRENAMING, NULL)); 3284 } 3285 } 3286 if (error == 0) { 3287 cache_vop_rename(sdvp, *svpp, tdvp, *tvpp, scnp, tcnp); 3288 } 3289 } 3290 3291 dmu_tx_commit(tx); 3292 3293 out_seq: 3294 vn_seqc_write_end(*svpp); 3295 vn_seqc_write_end(sdvp); 3296 if (*tvpp != NULL) 3297 vn_seqc_write_end(*tvpp); 3298 if (tdvp != *tvpp) 3299 vn_seqc_write_end(tdvp); 3300 3301 out: 3302 if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3303 zil_commit(zilog, 0); 3304 zfs_exit(zfsvfs, FTAG); 3305 3306 return (error); 3307 } 3308 3309 int 3310 zfs_rename(znode_t *sdzp, const char *sname, znode_t *tdzp, const char *tname, 3311 cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap, zidmap_t *mnt_ns) 3312 { 3313 struct componentname scn, tcn; 3314 vnode_t *sdvp, *tdvp; 3315 vnode_t *svp, *tvp; 3316 int error; 3317 svp = tvp = NULL; 3318 3319 if (is_nametoolong(tdzp->z_zfsvfs, tname)) 3320 return (SET_ERROR(ENAMETOOLONG)); 3321 3322 if (rflags != 0 || wo_vap != NULL) 3323 return (SET_ERROR(EINVAL)); 3324 3325 sdvp = ZTOV(sdzp); 3326 tdvp = ZTOV(tdzp); 3327 error = zfs_lookup_internal(sdzp, sname, &svp, &scn, DELETE); 3328 if (sdzp->z_zfsvfs->z_replay == B_FALSE) 3329 VOP_UNLOCK(sdvp); 3330 if (error != 0) 3331 goto fail; 3332 VOP_UNLOCK(svp); 3333 3334 vn_lock(tdvp, LK_EXCLUSIVE | LK_RETRY); 3335 error = zfs_lookup_internal(tdzp, tname, &tvp, &tcn, RENAME); 3336 if (error == EJUSTRETURN) 3337 tvp = NULL; 3338 else if (error != 0) { 3339 VOP_UNLOCK(tdvp); 3340 goto fail; 3341 } 3342 3343 error = zfs_do_rename(sdvp, &svp, &scn, tdvp, &tvp, &tcn, cr); 3344 fail: 3345 if (svp != NULL) 3346 vrele(svp); 3347 if (tvp != NULL) 3348 vrele(tvp); 3349 3350 return (error); 3351 } 3352 3353 /* 3354 * Insert the indicated symbolic reference entry into the directory. 3355 * 3356 * IN: dvp - Directory to contain new symbolic link. 3357 * link - Name for new symlink entry. 3358 * vap - Attributes of new entry. 3359 * cr - credentials of caller. 3360 * ct - caller context 3361 * flags - case flags 3362 * mnt_ns - Unused on FreeBSD 3363 * 3364 * RETURN: 0 on success, error code on failure. 3365 * 3366 * Timestamps: 3367 * dvp - ctime|mtime updated 3368 */ 3369 int 3370 zfs_symlink(znode_t *dzp, const char *name, vattr_t *vap, 3371 const char *link, znode_t **zpp, cred_t *cr, int flags, zidmap_t *mnt_ns) 3372 { 3373 (void) flags; 3374 znode_t *zp; 3375 dmu_tx_t *tx; 3376 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3377 zilog_t *zilog; 3378 uint64_t len = strlen(link); 3379 int error; 3380 zfs_acl_ids_t acl_ids; 3381 boolean_t fuid_dirtied; 3382 uint64_t txtype = TX_SYMLINK; 3383 3384 ASSERT3S(vap->va_type, ==, VLNK); 3385 3386 if (is_nametoolong(zfsvfs, name)) 3387 return (SET_ERROR(ENAMETOOLONG)); 3388 3389 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 3390 return (error); 3391 zilog = zfsvfs->z_log; 3392 3393 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 3394 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3395 zfs_exit(zfsvfs, FTAG); 3396 return (SET_ERROR(EILSEQ)); 3397 } 3398 3399 if (len > MAXPATHLEN) { 3400 zfs_exit(zfsvfs, FTAG); 3401 return (SET_ERROR(ENAMETOOLONG)); 3402 } 3403 3404 if ((error = zfs_acl_ids_create(dzp, 0, 3405 vap, cr, NULL, &acl_ids, NULL)) != 0) { 3406 zfs_exit(zfsvfs, FTAG); 3407 return (error); 3408 } 3409 3410 /* 3411 * Attempt to lock directory; fail if entry already exists. 3412 */ 3413 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW); 3414 if (error) { 3415 zfs_acl_ids_free(&acl_ids); 3416 zfs_exit(zfsvfs, FTAG); 3417 return (error); 3418 } 3419 3420 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) { 3421 zfs_acl_ids_free(&acl_ids); 3422 zfs_exit(zfsvfs, FTAG); 3423 return (error); 3424 } 3425 3426 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, 3427 0 /* projid */)) { 3428 zfs_acl_ids_free(&acl_ids); 3429 zfs_exit(zfsvfs, FTAG); 3430 return (SET_ERROR(EDQUOT)); 3431 } 3432 3433 getnewvnode_reserve(); 3434 tx = dmu_tx_create(zfsvfs->z_os); 3435 fuid_dirtied = zfsvfs->z_fuid_dirty; 3436 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 3437 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3438 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 3439 ZFS_SA_BASE_ATTR_SIZE + len); 3440 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 3441 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3442 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 3443 acl_ids.z_aclp->z_acl_bytes); 3444 } 3445 if (fuid_dirtied) 3446 zfs_fuid_txhold(zfsvfs, tx); 3447 error = dmu_tx_assign(tx, TXG_WAIT); 3448 if (error) { 3449 zfs_acl_ids_free(&acl_ids); 3450 dmu_tx_abort(tx); 3451 getnewvnode_drop_reserve(); 3452 zfs_exit(zfsvfs, FTAG); 3453 return (error); 3454 } 3455 3456 /* 3457 * Create a new object for the symlink. 3458 * for version 4 ZPL datasets the symlink will be an SA attribute 3459 */ 3460 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 3461 3462 if (fuid_dirtied) 3463 zfs_fuid_sync(zfsvfs, tx); 3464 3465 if (zp->z_is_sa) 3466 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 3467 __DECONST(void *, link), len, tx); 3468 else 3469 zfs_sa_symlink(zp, __DECONST(char *, link), len, tx); 3470 3471 zp->z_size = len; 3472 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 3473 &zp->z_size, sizeof (zp->z_size), tx); 3474 /* 3475 * Insert the new object into the directory. 3476 */ 3477 error = zfs_link_create(dzp, name, zp, tx, ZNEW); 3478 if (error != 0) { 3479 zfs_znode_delete(zp, tx); 3480 VOP_UNLOCK(ZTOV(zp)); 3481 zrele(zp); 3482 } else { 3483 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 3484 } 3485 3486 zfs_acl_ids_free(&acl_ids); 3487 3488 dmu_tx_commit(tx); 3489 3490 getnewvnode_drop_reserve(); 3491 3492 if (error == 0) { 3493 *zpp = zp; 3494 3495 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3496 zil_commit(zilog, 0); 3497 } 3498 3499 zfs_exit(zfsvfs, FTAG); 3500 return (error); 3501 } 3502 3503 /* 3504 * Return, in the buffer contained in the provided uio structure, 3505 * the symbolic path referred to by vp. 3506 * 3507 * IN: vp - vnode of symbolic link. 3508 * uio - structure to contain the link path. 3509 * cr - credentials of caller. 3510 * ct - caller context 3511 * 3512 * OUT: uio - structure containing the link path. 3513 * 3514 * RETURN: 0 on success, error code on failure. 3515 * 3516 * Timestamps: 3517 * vp - atime updated 3518 */ 3519 static int 3520 zfs_readlink(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, caller_context_t *ct) 3521 { 3522 (void) cr, (void) ct; 3523 znode_t *zp = VTOZ(vp); 3524 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3525 int error; 3526 3527 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3528 return (error); 3529 3530 if (zp->z_is_sa) 3531 error = sa_lookup_uio(zp->z_sa_hdl, 3532 SA_ZPL_SYMLINK(zfsvfs), uio); 3533 else 3534 error = zfs_sa_readlink(zp, uio); 3535 3536 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3537 3538 zfs_exit(zfsvfs, FTAG); 3539 return (error); 3540 } 3541 3542 /* 3543 * Insert a new entry into directory tdvp referencing svp. 3544 * 3545 * IN: tdvp - Directory to contain new entry. 3546 * svp - vnode of new entry. 3547 * name - name of new entry. 3548 * cr - credentials of caller. 3549 * 3550 * RETURN: 0 on success, error code on failure. 3551 * 3552 * Timestamps: 3553 * tdvp - ctime|mtime updated 3554 * svp - ctime updated 3555 */ 3556 int 3557 zfs_link(znode_t *tdzp, znode_t *szp, const char *name, cred_t *cr, 3558 int flags) 3559 { 3560 (void) flags; 3561 znode_t *tzp; 3562 zfsvfs_t *zfsvfs = tdzp->z_zfsvfs; 3563 zilog_t *zilog; 3564 dmu_tx_t *tx; 3565 int error; 3566 uint64_t parent; 3567 uid_t owner; 3568 3569 ASSERT3S(ZTOV(tdzp)->v_type, ==, VDIR); 3570 3571 if (is_nametoolong(zfsvfs, name)) 3572 return (SET_ERROR(ENAMETOOLONG)); 3573 3574 if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0) 3575 return (error); 3576 zilog = zfsvfs->z_log; 3577 3578 /* 3579 * POSIX dictates that we return EPERM here. 3580 * Better choices include ENOTSUP or EISDIR. 3581 */ 3582 if (ZTOV(szp)->v_type == VDIR) { 3583 zfs_exit(zfsvfs, FTAG); 3584 return (SET_ERROR(EPERM)); 3585 } 3586 3587 if ((error = zfs_verify_zp(szp)) != 0) { 3588 zfs_exit(zfsvfs, FTAG); 3589 return (error); 3590 } 3591 3592 /* 3593 * If we are using project inheritance, means if the directory has 3594 * ZFS_PROJINHERIT set, then its descendant directories will inherit 3595 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under 3596 * such case, we only allow hard link creation in our tree when the 3597 * project IDs are the same. 3598 */ 3599 if (tdzp->z_pflags & ZFS_PROJINHERIT && 3600 tdzp->z_projid != szp->z_projid) { 3601 zfs_exit(zfsvfs, FTAG); 3602 return (SET_ERROR(EXDEV)); 3603 } 3604 3605 if (szp->z_pflags & (ZFS_APPENDONLY | 3606 ZFS_IMMUTABLE | ZFS_READONLY)) { 3607 zfs_exit(zfsvfs, FTAG); 3608 return (SET_ERROR(EPERM)); 3609 } 3610 3611 /* Prevent links to .zfs/shares files */ 3612 3613 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 3614 &parent, sizeof (uint64_t))) != 0) { 3615 zfs_exit(zfsvfs, FTAG); 3616 return (error); 3617 } 3618 if (parent == zfsvfs->z_shares_dir) { 3619 zfs_exit(zfsvfs, FTAG); 3620 return (SET_ERROR(EPERM)); 3621 } 3622 3623 if (zfsvfs->z_utf8 && u8_validate(name, 3624 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3625 zfs_exit(zfsvfs, FTAG); 3626 return (SET_ERROR(EILSEQ)); 3627 } 3628 3629 /* 3630 * We do not support links between attributes and non-attributes 3631 * because of the potential security risk of creating links 3632 * into "normal" file space in order to circumvent restrictions 3633 * imposed in attribute space. 3634 */ 3635 if ((szp->z_pflags & ZFS_XATTR) != (tdzp->z_pflags & ZFS_XATTR)) { 3636 zfs_exit(zfsvfs, FTAG); 3637 return (SET_ERROR(EINVAL)); 3638 } 3639 3640 3641 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 3642 if (owner != crgetuid(cr) && secpolicy_basic_link(ZTOV(szp), cr) != 0) { 3643 zfs_exit(zfsvfs, FTAG); 3644 return (SET_ERROR(EPERM)); 3645 } 3646 3647 if ((error = zfs_zaccess(tdzp, ACE_ADD_FILE, 0, B_FALSE, cr, NULL))) { 3648 zfs_exit(zfsvfs, FTAG); 3649 return (error); 3650 } 3651 3652 /* 3653 * Attempt to lock directory; fail if entry already exists. 3654 */ 3655 error = zfs_dirent_lookup(tdzp, name, &tzp, ZNEW); 3656 if (error) { 3657 zfs_exit(zfsvfs, FTAG); 3658 return (error); 3659 } 3660 3661 tx = dmu_tx_create(zfsvfs->z_os); 3662 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3663 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, name); 3664 zfs_sa_upgrade_txholds(tx, szp); 3665 zfs_sa_upgrade_txholds(tx, tdzp); 3666 error = dmu_tx_assign(tx, TXG_WAIT); 3667 if (error) { 3668 dmu_tx_abort(tx); 3669 zfs_exit(zfsvfs, FTAG); 3670 return (error); 3671 } 3672 3673 error = zfs_link_create(tdzp, name, szp, tx, 0); 3674 3675 if (error == 0) { 3676 uint64_t txtype = TX_LINK; 3677 zfs_log_link(zilog, tx, txtype, tdzp, szp, name); 3678 } 3679 3680 dmu_tx_commit(tx); 3681 3682 if (error == 0) { 3683 vnevent_link(ZTOV(szp), ct); 3684 } 3685 3686 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3687 zil_commit(zilog, 0); 3688 3689 zfs_exit(zfsvfs, FTAG); 3690 return (error); 3691 } 3692 3693 /* 3694 * Free or allocate space in a file. Currently, this function only 3695 * supports the `F_FREESP' command. However, this command is somewhat 3696 * misnamed, as its functionality includes the ability to allocate as 3697 * well as free space. 3698 * 3699 * IN: ip - inode of file to free data in. 3700 * cmd - action to take (only F_FREESP supported). 3701 * bfp - section of file to free/alloc. 3702 * flag - current file open mode flags. 3703 * offset - current file offset. 3704 * cr - credentials of caller. 3705 * 3706 * RETURN: 0 on success, error code on failure. 3707 * 3708 * Timestamps: 3709 * ip - ctime|mtime updated 3710 */ 3711 int 3712 zfs_space(znode_t *zp, int cmd, flock64_t *bfp, int flag, 3713 offset_t offset, cred_t *cr) 3714 { 3715 (void) offset; 3716 zfsvfs_t *zfsvfs = ZTOZSB(zp); 3717 uint64_t off, len; 3718 int error; 3719 3720 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3721 return (error); 3722 3723 if (cmd != F_FREESP) { 3724 zfs_exit(zfsvfs, FTAG); 3725 return (SET_ERROR(EINVAL)); 3726 } 3727 3728 /* 3729 * Callers might not be able to detect properly that we are read-only, 3730 * so check it explicitly here. 3731 */ 3732 if (zfs_is_readonly(zfsvfs)) { 3733 zfs_exit(zfsvfs, FTAG); 3734 return (SET_ERROR(EROFS)); 3735 } 3736 3737 if (bfp->l_len < 0) { 3738 zfs_exit(zfsvfs, FTAG); 3739 return (SET_ERROR(EINVAL)); 3740 } 3741 3742 /* 3743 * Permissions aren't checked on Solaris because on this OS 3744 * zfs_space() can only be called with an opened file handle. 3745 * On Linux we can get here through truncate_range() which 3746 * operates directly on inodes, so we need to check access rights. 3747 */ 3748 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr, NULL))) { 3749 zfs_exit(zfsvfs, FTAG); 3750 return (error); 3751 } 3752 3753 off = bfp->l_start; 3754 len = bfp->l_len; /* 0 means from off to end of file */ 3755 3756 error = zfs_freesp(zp, off, len, flag, TRUE); 3757 3758 zfs_exit(zfsvfs, FTAG); 3759 return (error); 3760 } 3761 3762 static void 3763 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 3764 { 3765 (void) cr, (void) ct; 3766 znode_t *zp = VTOZ(vp); 3767 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3768 int error; 3769 3770 ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs); 3771 if (zp->z_sa_hdl == NULL) { 3772 /* 3773 * The fs has been unmounted, or we did a 3774 * suspend/resume and this file no longer exists. 3775 */ 3776 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 3777 vrecycle(vp); 3778 return; 3779 } 3780 3781 if (zp->z_unlinked) { 3782 /* 3783 * Fast path to recycle a vnode of a removed file. 3784 */ 3785 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 3786 vrecycle(vp); 3787 return; 3788 } 3789 3790 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 3791 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 3792 3793 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3794 zfs_sa_upgrade_txholds(tx, zp); 3795 error = dmu_tx_assign(tx, TXG_WAIT); 3796 if (error) { 3797 dmu_tx_abort(tx); 3798 } else { 3799 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 3800 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 3801 zp->z_atime_dirty = 0; 3802 dmu_tx_commit(tx); 3803 } 3804 } 3805 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 3806 } 3807 3808 3809 _Static_assert(sizeof (struct zfid_short) <= sizeof (struct fid), 3810 "struct zfid_short bigger than struct fid"); 3811 _Static_assert(sizeof (struct zfid_long) <= sizeof (struct fid), 3812 "struct zfid_long bigger than struct fid"); 3813 3814 static int 3815 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 3816 { 3817 (void) ct; 3818 znode_t *zp = VTOZ(vp); 3819 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3820 uint32_t gen; 3821 uint64_t gen64; 3822 uint64_t object = zp->z_id; 3823 zfid_short_t *zfid; 3824 int size, i, error; 3825 3826 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3827 return (error); 3828 3829 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 3830 &gen64, sizeof (uint64_t))) != 0) { 3831 zfs_exit(zfsvfs, FTAG); 3832 return (error); 3833 } 3834 3835 gen = (uint32_t)gen64; 3836 3837 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 3838 fidp->fid_len = size; 3839 3840 zfid = (zfid_short_t *)fidp; 3841 3842 zfid->zf_len = size; 3843 3844 for (i = 0; i < sizeof (zfid->zf_object); i++) 3845 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 3846 3847 /* Must have a non-zero generation number to distinguish from .zfs */ 3848 if (gen == 0) 3849 gen = 1; 3850 for (i = 0; i < sizeof (zfid->zf_gen); i++) 3851 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 3852 3853 if (size == LONG_FID_LEN) { 3854 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 3855 zfid_long_t *zlfid; 3856 3857 zlfid = (zfid_long_t *)fidp; 3858 3859 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 3860 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 3861 3862 /* XXX - this should be the generation number for the objset */ 3863 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 3864 zlfid->zf_setgen[i] = 0; 3865 } 3866 3867 zfs_exit(zfsvfs, FTAG); 3868 return (0); 3869 } 3870 3871 static int 3872 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 3873 caller_context_t *ct) 3874 { 3875 znode_t *zp; 3876 zfsvfs_t *zfsvfs; 3877 int error; 3878 3879 switch (cmd) { 3880 case _PC_LINK_MAX: 3881 *valp = MIN(LONG_MAX, ZFS_LINK_MAX); 3882 return (0); 3883 3884 case _PC_FILESIZEBITS: 3885 *valp = 64; 3886 return (0); 3887 case _PC_MIN_HOLE_SIZE: 3888 *valp = (int)SPA_MINBLOCKSIZE; 3889 return (0); 3890 case _PC_ACL_EXTENDED: 3891 #if 0 /* POSIX ACLs are not implemented for ZFS on FreeBSD yet. */ 3892 zp = VTOZ(vp); 3893 zfsvfs = zp->z_zfsvfs; 3894 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3895 return (error); 3896 *valp = zfsvfs->z_acl_type == ZFSACLTYPE_POSIX ? 1 : 0; 3897 zfs_exit(zfsvfs, FTAG); 3898 #else 3899 *valp = 0; 3900 #endif 3901 return (0); 3902 3903 case _PC_ACL_NFS4: 3904 zp = VTOZ(vp); 3905 zfsvfs = zp->z_zfsvfs; 3906 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3907 return (error); 3908 *valp = zfsvfs->z_acl_type == ZFS_ACLTYPE_NFSV4 ? 1 : 0; 3909 zfs_exit(zfsvfs, FTAG); 3910 return (0); 3911 3912 case _PC_ACL_PATH_MAX: 3913 *valp = ACL_MAX_ENTRIES; 3914 return (0); 3915 3916 default: 3917 return (EOPNOTSUPP); 3918 } 3919 } 3920 3921 static int 3922 zfs_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind, 3923 int *rahead) 3924 { 3925 znode_t *zp = VTOZ(vp); 3926 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3927 zfs_locked_range_t *lr; 3928 vm_object_t object; 3929 off_t start, end, obj_size; 3930 uint_t blksz; 3931 int pgsin_b, pgsin_a; 3932 int error; 3933 3934 if (zfs_enter_verify_zp(zfsvfs, zp, FTAG) != 0) 3935 return (zfs_vm_pagerret_error); 3936 3937 object = ma[0]->object; 3938 start = IDX_TO_OFF(ma[0]->pindex); 3939 end = IDX_TO_OFF(ma[count - 1]->pindex + 1); 3940 3941 /* 3942 * Lock a range covering all required and optional pages. 3943 * Note that we need to handle the case of the block size growing. 3944 */ 3945 for (;;) { 3946 uint64_t len; 3947 3948 blksz = zp->z_blksz; 3949 len = roundup(end, blksz) - rounddown(start, blksz); 3950 3951 lr = zfs_rangelock_tryenter(&zp->z_rangelock, 3952 rounddown(start, blksz), len, RL_READER); 3953 if (lr == NULL) { 3954 /* 3955 * Avoid a deadlock with update_pages(). We need to 3956 * hold the range lock when copying from the DMU, so 3957 * give up the busy lock to allow update_pages() to 3958 * proceed. We might need to allocate new pages, which 3959 * isn't quite right since this allocation isn't subject 3960 * to the page fault handler's OOM logic, but this is 3961 * the best we can do for now. 3962 */ 3963 for (int i = 0; i < count; i++) 3964 vm_page_xunbusy(ma[i]); 3965 3966 lr = zfs_rangelock_enter(&zp->z_rangelock, 3967 rounddown(start, blksz), len, RL_READER); 3968 3969 zfs_vmobject_wlock(object); 3970 (void) vm_page_grab_pages(object, OFF_TO_IDX(start), 3971 VM_ALLOC_NORMAL | VM_ALLOC_WAITOK | VM_ALLOC_ZERO, 3972 ma, count); 3973 zfs_vmobject_wunlock(object); 3974 } 3975 if (blksz == zp->z_blksz) 3976 break; 3977 zfs_rangelock_exit(lr); 3978 } 3979 3980 zfs_vmobject_wlock(object); 3981 obj_size = object->un_pager.vnp.vnp_size; 3982 zfs_vmobject_wunlock(object); 3983 if (IDX_TO_OFF(ma[count - 1]->pindex) >= obj_size) { 3984 zfs_rangelock_exit(lr); 3985 zfs_exit(zfsvfs, FTAG); 3986 return (zfs_vm_pagerret_bad); 3987 } 3988 3989 pgsin_b = 0; 3990 if (rbehind != NULL) { 3991 pgsin_b = OFF_TO_IDX(start - rounddown(start, blksz)); 3992 pgsin_b = MIN(*rbehind, pgsin_b); 3993 } 3994 3995 pgsin_a = 0; 3996 if (rahead != NULL) { 3997 pgsin_a = OFF_TO_IDX(roundup(end, blksz) - end); 3998 if (end + IDX_TO_OFF(pgsin_a) >= obj_size) 3999 pgsin_a = OFF_TO_IDX(round_page(obj_size) - end); 4000 pgsin_a = MIN(*rahead, pgsin_a); 4001 } 4002 4003 /* 4004 * NB: we need to pass the exact byte size of the data that we expect 4005 * to read after accounting for the file size. This is required because 4006 * ZFS will panic if we request DMU to read beyond the end of the last 4007 * allocated block. 4008 */ 4009 for (int i = 0; i < count; i++) { 4010 int dummypgsin, count1, j, last_size; 4011 4012 if (vm_page_any_valid(ma[i])) { 4013 ASSERT(vm_page_all_valid(ma[i])); 4014 continue; 4015 } 4016 for (j = i + 1; j < count; j++) { 4017 if (vm_page_any_valid(ma[j])) { 4018 ASSERT(vm_page_all_valid(ma[j])); 4019 break; 4020 } 4021 } 4022 count1 = j - i; 4023 dummypgsin = 0; 4024 last_size = j == count ? 4025 MIN(end, obj_size) - (end - PAGE_SIZE) : PAGE_SIZE; 4026 error = dmu_read_pages(zfsvfs->z_os, zp->z_id, &ma[i], count1, 4027 i == 0 ? &pgsin_b : &dummypgsin, 4028 j == count ? &pgsin_a : &dummypgsin, 4029 last_size); 4030 if (error != 0) 4031 break; 4032 i += count1 - 1; 4033 } 4034 4035 zfs_rangelock_exit(lr); 4036 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4037 4038 dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, count*PAGE_SIZE); 4039 4040 zfs_exit(zfsvfs, FTAG); 4041 4042 if (error != 0) 4043 return (zfs_vm_pagerret_error); 4044 4045 VM_CNT_INC(v_vnodein); 4046 VM_CNT_ADD(v_vnodepgsin, count + pgsin_b + pgsin_a); 4047 if (rbehind != NULL) 4048 *rbehind = pgsin_b; 4049 if (rahead != NULL) 4050 *rahead = pgsin_a; 4051 return (zfs_vm_pagerret_ok); 4052 } 4053 4054 #ifndef _SYS_SYSPROTO_H_ 4055 struct vop_getpages_args { 4056 struct vnode *a_vp; 4057 vm_page_t *a_m; 4058 int a_count; 4059 int *a_rbehind; 4060 int *a_rahead; 4061 }; 4062 #endif 4063 4064 static int 4065 zfs_freebsd_getpages(struct vop_getpages_args *ap) 4066 { 4067 4068 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind, 4069 ap->a_rahead)); 4070 } 4071 4072 static int 4073 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags, 4074 int *rtvals) 4075 { 4076 znode_t *zp = VTOZ(vp); 4077 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4078 zfs_locked_range_t *lr; 4079 dmu_tx_t *tx; 4080 struct sf_buf *sf; 4081 vm_object_t object; 4082 vm_page_t m; 4083 caddr_t va; 4084 size_t tocopy; 4085 size_t lo_len; 4086 vm_ooffset_t lo_off; 4087 vm_ooffset_t off; 4088 uint_t blksz; 4089 int ncount; 4090 int pcount; 4091 int err; 4092 int i; 4093 4094 object = vp->v_object; 4095 KASSERT(ma[0]->object == object, ("mismatching object")); 4096 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length")); 4097 4098 pcount = btoc(len); 4099 ncount = pcount; 4100 for (i = 0; i < pcount; i++) 4101 rtvals[i] = zfs_vm_pagerret_error; 4102 4103 if (zfs_enter_verify_zp(zfsvfs, zp, FTAG) != 0) 4104 return (zfs_vm_pagerret_error); 4105 4106 off = IDX_TO_OFF(ma[0]->pindex); 4107 blksz = zp->z_blksz; 4108 lo_off = rounddown(off, blksz); 4109 lo_len = roundup(len + (off - lo_off), blksz); 4110 lr = zfs_rangelock_enter(&zp->z_rangelock, lo_off, lo_len, RL_WRITER); 4111 4112 zfs_vmobject_wlock(object); 4113 if (len + off > object->un_pager.vnp.vnp_size) { 4114 if (object->un_pager.vnp.vnp_size > off) { 4115 int pgoff; 4116 4117 len = object->un_pager.vnp.vnp_size - off; 4118 ncount = btoc(len); 4119 if ((pgoff = (int)len & PAGE_MASK) != 0) { 4120 /* 4121 * If the object is locked and the following 4122 * conditions hold, then the page's dirty 4123 * field cannot be concurrently changed by a 4124 * pmap operation. 4125 */ 4126 m = ma[ncount - 1]; 4127 vm_page_assert_sbusied(m); 4128 KASSERT(!pmap_page_is_write_mapped(m), 4129 ("zfs_putpages: page %p is not read-only", 4130 m)); 4131 vm_page_clear_dirty(m, pgoff, PAGE_SIZE - 4132 pgoff); 4133 } 4134 } else { 4135 len = 0; 4136 ncount = 0; 4137 } 4138 if (ncount < pcount) { 4139 for (i = ncount; i < pcount; i++) { 4140 rtvals[i] = zfs_vm_pagerret_bad; 4141 } 4142 } 4143 } 4144 zfs_vmobject_wunlock(object); 4145 4146 boolean_t commit = (flags & (zfs_vm_pagerput_sync | 4147 zfs_vm_pagerput_inval)) != 0 || 4148 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS; 4149 4150 if (ncount == 0) 4151 goto out; 4152 4153 if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT, zp->z_uid) || 4154 zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT, zp->z_gid) || 4155 (zp->z_projid != ZFS_DEFAULT_PROJID && 4156 zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT, 4157 zp->z_projid))) { 4158 goto out; 4159 } 4160 4161 tx = dmu_tx_create(zfsvfs->z_os); 4162 dmu_tx_hold_write(tx, zp->z_id, off, len); 4163 4164 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4165 zfs_sa_upgrade_txholds(tx, zp); 4166 err = dmu_tx_assign(tx, TXG_WAIT); 4167 if (err != 0) { 4168 dmu_tx_abort(tx); 4169 goto out; 4170 } 4171 4172 if (zp->z_blksz < PAGE_SIZE) { 4173 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) { 4174 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len; 4175 va = zfs_map_page(ma[i], &sf); 4176 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx); 4177 zfs_unmap_page(sf); 4178 } 4179 } else { 4180 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx); 4181 } 4182 4183 if (err == 0) { 4184 uint64_t mtime[2], ctime[2]; 4185 sa_bulk_attr_t bulk[3]; 4186 int count = 0; 4187 4188 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4189 &mtime, 16); 4190 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4191 &ctime, 16); 4192 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4193 &zp->z_pflags, 8); 4194 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime); 4195 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 4196 ASSERT0(err); 4197 /* 4198 * XXX we should be passing a callback to undirty 4199 * but that would make the locking messier 4200 */ 4201 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, 4202 len, commit, B_FALSE, NULL, NULL); 4203 4204 zfs_vmobject_wlock(object); 4205 for (i = 0; i < ncount; i++) { 4206 rtvals[i] = zfs_vm_pagerret_ok; 4207 vm_page_undirty(ma[i]); 4208 } 4209 zfs_vmobject_wunlock(object); 4210 VM_CNT_INC(v_vnodeout); 4211 VM_CNT_ADD(v_vnodepgsout, ncount); 4212 } 4213 dmu_tx_commit(tx); 4214 4215 out: 4216 zfs_rangelock_exit(lr); 4217 if (commit) 4218 zil_commit(zfsvfs->z_log, zp->z_id); 4219 4220 dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, len); 4221 4222 zfs_exit(zfsvfs, FTAG); 4223 return (rtvals[0]); 4224 } 4225 4226 #ifndef _SYS_SYSPROTO_H_ 4227 struct vop_putpages_args { 4228 struct vnode *a_vp; 4229 vm_page_t *a_m; 4230 int a_count; 4231 int a_sync; 4232 int *a_rtvals; 4233 }; 4234 #endif 4235 4236 static int 4237 zfs_freebsd_putpages(struct vop_putpages_args *ap) 4238 { 4239 4240 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync, 4241 ap->a_rtvals)); 4242 } 4243 4244 #ifndef _SYS_SYSPROTO_H_ 4245 struct vop_bmap_args { 4246 struct vnode *a_vp; 4247 daddr_t a_bn; 4248 struct bufobj **a_bop; 4249 daddr_t *a_bnp; 4250 int *a_runp; 4251 int *a_runb; 4252 }; 4253 #endif 4254 4255 static int 4256 zfs_freebsd_bmap(struct vop_bmap_args *ap) 4257 { 4258 4259 if (ap->a_bop != NULL) 4260 *ap->a_bop = &ap->a_vp->v_bufobj; 4261 if (ap->a_bnp != NULL) 4262 *ap->a_bnp = ap->a_bn; 4263 if (ap->a_runp != NULL) 4264 *ap->a_runp = 0; 4265 if (ap->a_runb != NULL) 4266 *ap->a_runb = 0; 4267 4268 return (0); 4269 } 4270 4271 #ifndef _SYS_SYSPROTO_H_ 4272 struct vop_open_args { 4273 struct vnode *a_vp; 4274 int a_mode; 4275 struct ucred *a_cred; 4276 struct thread *a_td; 4277 }; 4278 #endif 4279 4280 static int 4281 zfs_freebsd_open(struct vop_open_args *ap) 4282 { 4283 vnode_t *vp = ap->a_vp; 4284 znode_t *zp = VTOZ(vp); 4285 int error; 4286 4287 error = zfs_open(&vp, ap->a_mode, ap->a_cred); 4288 if (error == 0) 4289 vnode_create_vobject(vp, zp->z_size, ap->a_td); 4290 return (error); 4291 } 4292 4293 #ifndef _SYS_SYSPROTO_H_ 4294 struct vop_close_args { 4295 struct vnode *a_vp; 4296 int a_fflag; 4297 struct ucred *a_cred; 4298 struct thread *a_td; 4299 }; 4300 #endif 4301 4302 static int 4303 zfs_freebsd_close(struct vop_close_args *ap) 4304 { 4305 4306 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred)); 4307 } 4308 4309 #ifndef _SYS_SYSPROTO_H_ 4310 struct vop_ioctl_args { 4311 struct vnode *a_vp; 4312 ulong_t a_command; 4313 caddr_t a_data; 4314 int a_fflag; 4315 struct ucred *cred; 4316 struct thread *td; 4317 }; 4318 #endif 4319 4320 static int 4321 zfs_freebsd_ioctl(struct vop_ioctl_args *ap) 4322 { 4323 4324 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 4325 ap->a_fflag, ap->a_cred, NULL)); 4326 } 4327 4328 static int 4329 ioflags(int ioflags) 4330 { 4331 int flags = 0; 4332 4333 if (ioflags & IO_APPEND) 4334 flags |= O_APPEND; 4335 if (ioflags & IO_NDELAY) 4336 flags |= O_NONBLOCK; 4337 if (ioflags & IO_DIRECT) 4338 flags |= O_DIRECT; 4339 if (ioflags & IO_SYNC) 4340 flags |= O_SYNC; 4341 4342 return (flags); 4343 } 4344 4345 #ifndef _SYS_SYSPROTO_H_ 4346 struct vop_read_args { 4347 struct vnode *a_vp; 4348 struct uio *a_uio; 4349 int a_ioflag; 4350 struct ucred *a_cred; 4351 }; 4352 #endif 4353 4354 static int 4355 zfs_freebsd_read(struct vop_read_args *ap) 4356 { 4357 zfs_uio_t uio; 4358 int error = 0; 4359 zfs_uio_init(&uio, ap->a_uio); 4360 error = zfs_read(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag), 4361 ap->a_cred); 4362 /* 4363 * XXX We occasionally get an EFAULT for Direct I/O reads on 4364 * FreeBSD 13. This still needs to be resolved. The EFAULT comes 4365 * from: 4366 * zfs_uio_get__dio_pages_alloc() -> 4367 * zfs_uio_get_dio_pages_impl() -> 4368 * zfs_uio_iov_step() -> 4369 * zfs_uio_get_user_pages(). 4370 * We return EFAULT from zfs_uio_iov_step(). When a Direct I/O 4371 * read fails to map in the user pages (returning EFAULT) the 4372 * Direct I/O request is broken up into two separate IO requests 4373 * and issued separately using Direct I/O. 4374 */ 4375 #ifdef ZFS_DEBUG 4376 if (error == EFAULT && uio.uio_extflg & UIO_DIRECT) { 4377 #if 0 4378 printf("%s(%d): Direct I/O read returning EFAULT " 4379 "uio = %p, zfs_uio_offset(uio) = %lu " 4380 "zfs_uio_resid(uio) = %lu\n", 4381 __FUNCTION__, __LINE__, &uio, zfs_uio_offset(&uio), 4382 zfs_uio_resid(&uio)); 4383 #endif 4384 } 4385 4386 #endif 4387 return (error); 4388 } 4389 4390 #ifndef _SYS_SYSPROTO_H_ 4391 struct vop_write_args { 4392 struct vnode *a_vp; 4393 struct uio *a_uio; 4394 int a_ioflag; 4395 struct ucred *a_cred; 4396 }; 4397 #endif 4398 4399 static int 4400 zfs_freebsd_write(struct vop_write_args *ap) 4401 { 4402 zfs_uio_t uio; 4403 zfs_uio_init(&uio, ap->a_uio); 4404 return (zfs_write(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag), 4405 ap->a_cred)); 4406 } 4407 4408 /* 4409 * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see 4410 * the comment above cache_fplookup for details. 4411 */ 4412 static int 4413 zfs_freebsd_fplookup_vexec(struct vop_fplookup_vexec_args *v) 4414 { 4415 vnode_t *vp; 4416 znode_t *zp; 4417 uint64_t pflags; 4418 4419 vp = v->a_vp; 4420 zp = VTOZ_SMR(vp); 4421 if (__predict_false(zp == NULL)) 4422 return (EAGAIN); 4423 pflags = atomic_load_64(&zp->z_pflags); 4424 if (pflags & ZFS_AV_QUARANTINED) 4425 return (EAGAIN); 4426 if (pflags & ZFS_XATTR) 4427 return (EAGAIN); 4428 if ((pflags & ZFS_NO_EXECS_DENIED) == 0) 4429 return (EAGAIN); 4430 return (0); 4431 } 4432 4433 static int 4434 zfs_freebsd_fplookup_symlink(struct vop_fplookup_symlink_args *v) 4435 { 4436 vnode_t *vp; 4437 znode_t *zp; 4438 char *target; 4439 4440 vp = v->a_vp; 4441 zp = VTOZ_SMR(vp); 4442 if (__predict_false(zp == NULL)) { 4443 return (EAGAIN); 4444 } 4445 4446 target = atomic_load_consume_ptr(&zp->z_cached_symlink); 4447 if (target == NULL) { 4448 return (EAGAIN); 4449 } 4450 return (cache_symlink_resolve(v->a_fpl, target, strlen(target))); 4451 } 4452 4453 #ifndef _SYS_SYSPROTO_H_ 4454 struct vop_access_args { 4455 struct vnode *a_vp; 4456 accmode_t a_accmode; 4457 struct ucred *a_cred; 4458 struct thread *a_td; 4459 }; 4460 #endif 4461 4462 static int 4463 zfs_freebsd_access(struct vop_access_args *ap) 4464 { 4465 vnode_t *vp = ap->a_vp; 4466 znode_t *zp = VTOZ(vp); 4467 accmode_t accmode; 4468 int error = 0; 4469 4470 4471 if (ap->a_accmode == VEXEC) { 4472 if (zfs_fastaccesschk_execute(zp, ap->a_cred) == 0) 4473 return (0); 4474 } 4475 4476 /* 4477 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 4478 */ 4479 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 4480 if (accmode != 0) 4481 error = zfs_access(zp, accmode, 0, ap->a_cred); 4482 4483 /* 4484 * VADMIN has to be handled by vaccess(). 4485 */ 4486 if (error == 0) { 4487 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 4488 if (accmode != 0) { 4489 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 4490 zp->z_gid, accmode, ap->a_cred); 4491 } 4492 } 4493 4494 /* 4495 * For VEXEC, ensure that at least one execute bit is set for 4496 * non-directories. 4497 */ 4498 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 4499 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 4500 error = EACCES; 4501 } 4502 4503 return (error); 4504 } 4505 4506 #ifndef _SYS_SYSPROTO_H_ 4507 struct vop_lookup_args { 4508 struct vnode *a_dvp; 4509 struct vnode **a_vpp; 4510 struct componentname *a_cnp; 4511 }; 4512 #endif 4513 4514 static int 4515 zfs_freebsd_lookup(struct vop_lookup_args *ap, boolean_t cached) 4516 { 4517 struct componentname *cnp = ap->a_cnp; 4518 char nm[NAME_MAX + 1]; 4519 4520 ASSERT3U(cnp->cn_namelen, <, sizeof (nm)); 4521 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof (nm))); 4522 4523 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 4524 cnp->cn_cred, 0, cached)); 4525 } 4526 4527 static int 4528 zfs_freebsd_cachedlookup(struct vop_cachedlookup_args *ap) 4529 { 4530 4531 return (zfs_freebsd_lookup((struct vop_lookup_args *)ap, B_TRUE)); 4532 } 4533 4534 #ifndef _SYS_SYSPROTO_H_ 4535 struct vop_lookup_args { 4536 struct vnode *a_dvp; 4537 struct vnode **a_vpp; 4538 struct componentname *a_cnp; 4539 }; 4540 #endif 4541 4542 static int 4543 zfs_cache_lookup(struct vop_lookup_args *ap) 4544 { 4545 zfsvfs_t *zfsvfs; 4546 4547 zfsvfs = ap->a_dvp->v_mount->mnt_data; 4548 if (zfsvfs->z_use_namecache) 4549 return (vfs_cache_lookup(ap)); 4550 else 4551 return (zfs_freebsd_lookup(ap, B_FALSE)); 4552 } 4553 4554 #ifndef _SYS_SYSPROTO_H_ 4555 struct vop_create_args { 4556 struct vnode *a_dvp; 4557 struct vnode **a_vpp; 4558 struct componentname *a_cnp; 4559 struct vattr *a_vap; 4560 }; 4561 #endif 4562 4563 static int 4564 zfs_freebsd_create(struct vop_create_args *ap) 4565 { 4566 zfsvfs_t *zfsvfs; 4567 struct componentname *cnp = ap->a_cnp; 4568 vattr_t *vap = ap->a_vap; 4569 znode_t *zp = NULL; 4570 int rc, mode; 4571 4572 #if __FreeBSD_version < 1400068 4573 ASSERT(cnp->cn_flags & SAVENAME); 4574 #endif 4575 4576 vattr_init_mask(vap); 4577 mode = vap->va_mode & ALLPERMS; 4578 zfsvfs = ap->a_dvp->v_mount->mnt_data; 4579 *ap->a_vpp = NULL; 4580 4581 rc = zfs_create(VTOZ(ap->a_dvp), cnp->cn_nameptr, vap, 0, mode, 4582 &zp, cnp->cn_cred, 0 /* flag */, NULL /* vsecattr */, NULL); 4583 if (rc == 0) 4584 *ap->a_vpp = ZTOV(zp); 4585 if (zfsvfs->z_use_namecache && 4586 rc == 0 && (cnp->cn_flags & MAKEENTRY) != 0) 4587 cache_enter(ap->a_dvp, *ap->a_vpp, cnp); 4588 4589 return (rc); 4590 } 4591 4592 #ifndef _SYS_SYSPROTO_H_ 4593 struct vop_remove_args { 4594 struct vnode *a_dvp; 4595 struct vnode *a_vp; 4596 struct componentname *a_cnp; 4597 }; 4598 #endif 4599 4600 static int 4601 zfs_freebsd_remove(struct vop_remove_args *ap) 4602 { 4603 4604 #if __FreeBSD_version < 1400068 4605 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 4606 #endif 4607 4608 return (zfs_remove_(ap->a_dvp, ap->a_vp, ap->a_cnp->cn_nameptr, 4609 ap->a_cnp->cn_cred)); 4610 } 4611 4612 #ifndef _SYS_SYSPROTO_H_ 4613 struct vop_mkdir_args { 4614 struct vnode *a_dvp; 4615 struct vnode **a_vpp; 4616 struct componentname *a_cnp; 4617 struct vattr *a_vap; 4618 }; 4619 #endif 4620 4621 static int 4622 zfs_freebsd_mkdir(struct vop_mkdir_args *ap) 4623 { 4624 vattr_t *vap = ap->a_vap; 4625 znode_t *zp = NULL; 4626 int rc; 4627 4628 #if __FreeBSD_version < 1400068 4629 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 4630 #endif 4631 4632 vattr_init_mask(vap); 4633 *ap->a_vpp = NULL; 4634 4635 rc = zfs_mkdir(VTOZ(ap->a_dvp), ap->a_cnp->cn_nameptr, vap, &zp, 4636 ap->a_cnp->cn_cred, 0, NULL, NULL); 4637 4638 if (rc == 0) 4639 *ap->a_vpp = ZTOV(zp); 4640 return (rc); 4641 } 4642 4643 #ifndef _SYS_SYSPROTO_H_ 4644 struct vop_rmdir_args { 4645 struct vnode *a_dvp; 4646 struct vnode *a_vp; 4647 struct componentname *a_cnp; 4648 }; 4649 #endif 4650 4651 static int 4652 zfs_freebsd_rmdir(struct vop_rmdir_args *ap) 4653 { 4654 struct componentname *cnp = ap->a_cnp; 4655 4656 #if __FreeBSD_version < 1400068 4657 ASSERT(cnp->cn_flags & SAVENAME); 4658 #endif 4659 4660 return (zfs_rmdir_(ap->a_dvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred)); 4661 } 4662 4663 #ifndef _SYS_SYSPROTO_H_ 4664 struct vop_readdir_args { 4665 struct vnode *a_vp; 4666 struct uio *a_uio; 4667 struct ucred *a_cred; 4668 int *a_eofflag; 4669 int *a_ncookies; 4670 cookie_t **a_cookies; 4671 }; 4672 #endif 4673 4674 static int 4675 zfs_freebsd_readdir(struct vop_readdir_args *ap) 4676 { 4677 zfs_uio_t uio; 4678 zfs_uio_init(&uio, ap->a_uio); 4679 return (zfs_readdir(ap->a_vp, &uio, ap->a_cred, ap->a_eofflag, 4680 ap->a_ncookies, ap->a_cookies)); 4681 } 4682 4683 #ifndef _SYS_SYSPROTO_H_ 4684 struct vop_fsync_args { 4685 struct vnode *a_vp; 4686 int a_waitfor; 4687 struct thread *a_td; 4688 }; 4689 #endif 4690 4691 static int 4692 zfs_freebsd_fsync(struct vop_fsync_args *ap) 4693 { 4694 4695 return (zfs_fsync(VTOZ(ap->a_vp), 0, ap->a_td->td_ucred)); 4696 } 4697 4698 #ifndef _SYS_SYSPROTO_H_ 4699 struct vop_getattr_args { 4700 struct vnode *a_vp; 4701 struct vattr *a_vap; 4702 struct ucred *a_cred; 4703 }; 4704 #endif 4705 4706 static int 4707 zfs_freebsd_getattr(struct vop_getattr_args *ap) 4708 { 4709 vattr_t *vap = ap->a_vap; 4710 xvattr_t xvap; 4711 ulong_t fflags = 0; 4712 int error; 4713 4714 xva_init(&xvap); 4715 xvap.xva_vattr = *vap; 4716 xvap.xva_vattr.va_mask |= AT_XVATTR; 4717 4718 /* Convert chflags into ZFS-type flags. */ 4719 /* XXX: what about SF_SETTABLE?. */ 4720 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 4721 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 4722 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 4723 XVA_SET_REQ(&xvap, XAT_NODUMP); 4724 XVA_SET_REQ(&xvap, XAT_READONLY); 4725 XVA_SET_REQ(&xvap, XAT_ARCHIVE); 4726 XVA_SET_REQ(&xvap, XAT_SYSTEM); 4727 XVA_SET_REQ(&xvap, XAT_HIDDEN); 4728 XVA_SET_REQ(&xvap, XAT_REPARSE); 4729 XVA_SET_REQ(&xvap, XAT_OFFLINE); 4730 XVA_SET_REQ(&xvap, XAT_SPARSE); 4731 4732 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred); 4733 if (error != 0) 4734 return (error); 4735 4736 /* Convert ZFS xattr into chflags. */ 4737 #define FLAG_CHECK(fflag, xflag, xfield) do { \ 4738 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 4739 fflags |= (fflag); \ 4740 } while (0) 4741 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 4742 xvap.xva_xoptattrs.xoa_immutable); 4743 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 4744 xvap.xva_xoptattrs.xoa_appendonly); 4745 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 4746 xvap.xva_xoptattrs.xoa_nounlink); 4747 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE, 4748 xvap.xva_xoptattrs.xoa_archive); 4749 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 4750 xvap.xva_xoptattrs.xoa_nodump); 4751 FLAG_CHECK(UF_READONLY, XAT_READONLY, 4752 xvap.xva_xoptattrs.xoa_readonly); 4753 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM, 4754 xvap.xva_xoptattrs.xoa_system); 4755 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN, 4756 xvap.xva_xoptattrs.xoa_hidden); 4757 FLAG_CHECK(UF_REPARSE, XAT_REPARSE, 4758 xvap.xva_xoptattrs.xoa_reparse); 4759 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE, 4760 xvap.xva_xoptattrs.xoa_offline); 4761 FLAG_CHECK(UF_SPARSE, XAT_SPARSE, 4762 xvap.xva_xoptattrs.xoa_sparse); 4763 4764 #undef FLAG_CHECK 4765 *vap = xvap.xva_vattr; 4766 vap->va_flags = fflags; 4767 return (0); 4768 } 4769 4770 #ifndef _SYS_SYSPROTO_H_ 4771 struct vop_setattr_args { 4772 struct vnode *a_vp; 4773 struct vattr *a_vap; 4774 struct ucred *a_cred; 4775 }; 4776 #endif 4777 4778 static int 4779 zfs_freebsd_setattr(struct vop_setattr_args *ap) 4780 { 4781 vnode_t *vp = ap->a_vp; 4782 vattr_t *vap = ap->a_vap; 4783 cred_t *cred = ap->a_cred; 4784 xvattr_t xvap; 4785 ulong_t fflags; 4786 uint64_t zflags; 4787 4788 vattr_init_mask(vap); 4789 vap->va_mask &= ~AT_NOSET; 4790 4791 xva_init(&xvap); 4792 xvap.xva_vattr = *vap; 4793 4794 zflags = VTOZ(vp)->z_pflags; 4795 4796 if (vap->va_flags != VNOVAL) { 4797 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 4798 int error; 4799 4800 if (zfsvfs->z_use_fuids == B_FALSE) 4801 return (EOPNOTSUPP); 4802 4803 fflags = vap->va_flags; 4804 /* 4805 * XXX KDM 4806 * We need to figure out whether it makes sense to allow 4807 * UF_REPARSE through, since we don't really have other 4808 * facilities to handle reparse points and zfs_setattr() 4809 * doesn't currently allow setting that attribute anyway. 4810 */ 4811 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE| 4812 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE| 4813 UF_OFFLINE|UF_SPARSE)) != 0) 4814 return (EOPNOTSUPP); 4815 /* 4816 * Unprivileged processes are not permitted to unset system 4817 * flags, or modify flags if any system flags are set. 4818 * Privileged non-jail processes may not modify system flags 4819 * if securelevel > 0 and any existing system flags are set. 4820 * Privileged jail processes behave like privileged non-jail 4821 * processes if the PR_ALLOW_CHFLAGS permission bit is set; 4822 * otherwise, they behave like unprivileged processes. 4823 */ 4824 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 4825 priv_check_cred(cred, PRIV_VFS_SYSFLAGS) == 0) { 4826 if (zflags & 4827 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 4828 error = securelevel_gt(cred, 0); 4829 if (error != 0) 4830 return (error); 4831 } 4832 } else { 4833 /* 4834 * Callers may only modify the file flags on 4835 * objects they have VADMIN rights for. 4836 */ 4837 if ((error = VOP_ACCESS(vp, VADMIN, cred, 4838 curthread)) != 0) 4839 return (error); 4840 if (zflags & 4841 (ZFS_IMMUTABLE | ZFS_APPENDONLY | 4842 ZFS_NOUNLINK)) { 4843 return (EPERM); 4844 } 4845 if (fflags & 4846 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 4847 return (EPERM); 4848 } 4849 } 4850 4851 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 4852 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 4853 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 4854 XVA_SET_REQ(&xvap, (xflag)); \ 4855 (xfield) = ((fflags & (fflag)) != 0); \ 4856 } \ 4857 } while (0) 4858 /* Convert chflags into ZFS-type flags. */ 4859 /* XXX: what about SF_SETTABLE?. */ 4860 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 4861 xvap.xva_xoptattrs.xoa_immutable); 4862 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 4863 xvap.xva_xoptattrs.xoa_appendonly); 4864 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 4865 xvap.xva_xoptattrs.xoa_nounlink); 4866 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE, 4867 xvap.xva_xoptattrs.xoa_archive); 4868 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 4869 xvap.xva_xoptattrs.xoa_nodump); 4870 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY, 4871 xvap.xva_xoptattrs.xoa_readonly); 4872 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM, 4873 xvap.xva_xoptattrs.xoa_system); 4874 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN, 4875 xvap.xva_xoptattrs.xoa_hidden); 4876 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE, 4877 xvap.xva_xoptattrs.xoa_reparse); 4878 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE, 4879 xvap.xva_xoptattrs.xoa_offline); 4880 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE, 4881 xvap.xva_xoptattrs.xoa_sparse); 4882 #undef FLAG_CHANGE 4883 } 4884 if (vap->va_birthtime.tv_sec != VNOVAL) { 4885 xvap.xva_vattr.va_mask |= AT_XVATTR; 4886 XVA_SET_REQ(&xvap, XAT_CREATETIME); 4887 } 4888 return (zfs_setattr(VTOZ(vp), (vattr_t *)&xvap, 0, cred, NULL)); 4889 } 4890 4891 #ifndef _SYS_SYSPROTO_H_ 4892 struct vop_rename_args { 4893 struct vnode *a_fdvp; 4894 struct vnode *a_fvp; 4895 struct componentname *a_fcnp; 4896 struct vnode *a_tdvp; 4897 struct vnode *a_tvp; 4898 struct componentname *a_tcnp; 4899 }; 4900 #endif 4901 4902 static int 4903 zfs_freebsd_rename(struct vop_rename_args *ap) 4904 { 4905 vnode_t *fdvp = ap->a_fdvp; 4906 vnode_t *fvp = ap->a_fvp; 4907 vnode_t *tdvp = ap->a_tdvp; 4908 vnode_t *tvp = ap->a_tvp; 4909 int error; 4910 4911 #if __FreeBSD_version < 1400068 4912 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 4913 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 4914 #endif 4915 4916 error = zfs_do_rename(fdvp, &fvp, ap->a_fcnp, tdvp, &tvp, 4917 ap->a_tcnp, ap->a_fcnp->cn_cred); 4918 4919 vrele(fdvp); 4920 vrele(fvp); 4921 vrele(tdvp); 4922 if (tvp != NULL) 4923 vrele(tvp); 4924 4925 return (error); 4926 } 4927 4928 #ifndef _SYS_SYSPROTO_H_ 4929 struct vop_symlink_args { 4930 struct vnode *a_dvp; 4931 struct vnode **a_vpp; 4932 struct componentname *a_cnp; 4933 struct vattr *a_vap; 4934 char *a_target; 4935 }; 4936 #endif 4937 4938 static int 4939 zfs_freebsd_symlink(struct vop_symlink_args *ap) 4940 { 4941 struct componentname *cnp = ap->a_cnp; 4942 vattr_t *vap = ap->a_vap; 4943 znode_t *zp = NULL; 4944 char *symlink; 4945 size_t symlink_len; 4946 int rc; 4947 4948 #if __FreeBSD_version < 1400068 4949 ASSERT(cnp->cn_flags & SAVENAME); 4950 #endif 4951 4952 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 4953 vattr_init_mask(vap); 4954 *ap->a_vpp = NULL; 4955 4956 rc = zfs_symlink(VTOZ(ap->a_dvp), cnp->cn_nameptr, vap, 4957 ap->a_target, &zp, cnp->cn_cred, 0 /* flags */, NULL); 4958 if (rc == 0) { 4959 *ap->a_vpp = ZTOV(zp); 4960 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__); 4961 MPASS(zp->z_cached_symlink == NULL); 4962 symlink_len = strlen(ap->a_target); 4963 symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK); 4964 if (symlink != NULL) { 4965 memcpy(symlink, ap->a_target, symlink_len); 4966 symlink[symlink_len] = '\0'; 4967 atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink, 4968 (uintptr_t)symlink); 4969 } 4970 } 4971 return (rc); 4972 } 4973 4974 #ifndef _SYS_SYSPROTO_H_ 4975 struct vop_readlink_args { 4976 struct vnode *a_vp; 4977 struct uio *a_uio; 4978 struct ucred *a_cred; 4979 }; 4980 #endif 4981 4982 static int 4983 zfs_freebsd_readlink(struct vop_readlink_args *ap) 4984 { 4985 zfs_uio_t uio; 4986 int error; 4987 znode_t *zp = VTOZ(ap->a_vp); 4988 char *symlink, *base; 4989 size_t symlink_len; 4990 bool trycache; 4991 4992 zfs_uio_init(&uio, ap->a_uio); 4993 trycache = false; 4994 if (zfs_uio_segflg(&uio) == UIO_SYSSPACE && 4995 zfs_uio_iovcnt(&uio) == 1) { 4996 base = zfs_uio_iovbase(&uio, 0); 4997 symlink_len = zfs_uio_iovlen(&uio, 0); 4998 trycache = true; 4999 } 5000 error = zfs_readlink(ap->a_vp, &uio, ap->a_cred, NULL); 5001 if (atomic_load_ptr(&zp->z_cached_symlink) != NULL || 5002 error != 0 || !trycache) { 5003 return (error); 5004 } 5005 symlink_len -= zfs_uio_resid(&uio); 5006 symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK); 5007 if (symlink != NULL) { 5008 memcpy(symlink, base, symlink_len); 5009 symlink[symlink_len] = '\0'; 5010 if (!atomic_cmpset_rel_ptr((uintptr_t *)&zp->z_cached_symlink, 5011 (uintptr_t)NULL, (uintptr_t)symlink)) { 5012 cache_symlink_free(symlink, symlink_len + 1); 5013 } 5014 } 5015 return (error); 5016 } 5017 5018 #ifndef _SYS_SYSPROTO_H_ 5019 struct vop_link_args { 5020 struct vnode *a_tdvp; 5021 struct vnode *a_vp; 5022 struct componentname *a_cnp; 5023 }; 5024 #endif 5025 5026 static int 5027 zfs_freebsd_link(struct vop_link_args *ap) 5028 { 5029 struct componentname *cnp = ap->a_cnp; 5030 vnode_t *vp = ap->a_vp; 5031 vnode_t *tdvp = ap->a_tdvp; 5032 5033 if (tdvp->v_mount != vp->v_mount) 5034 return (EXDEV); 5035 5036 #if __FreeBSD_version < 1400068 5037 ASSERT(cnp->cn_flags & SAVENAME); 5038 #endif 5039 5040 return (zfs_link(VTOZ(tdvp), VTOZ(vp), 5041 cnp->cn_nameptr, cnp->cn_cred, 0)); 5042 } 5043 5044 #ifndef _SYS_SYSPROTO_H_ 5045 struct vop_inactive_args { 5046 struct vnode *a_vp; 5047 struct thread *a_td; 5048 }; 5049 #endif 5050 5051 static int 5052 zfs_freebsd_inactive(struct vop_inactive_args *ap) 5053 { 5054 vnode_t *vp = ap->a_vp; 5055 5056 zfs_inactive(vp, curthread->td_ucred, NULL); 5057 return (0); 5058 } 5059 5060 #ifndef _SYS_SYSPROTO_H_ 5061 struct vop_need_inactive_args { 5062 struct vnode *a_vp; 5063 struct thread *a_td; 5064 }; 5065 #endif 5066 5067 static int 5068 zfs_freebsd_need_inactive(struct vop_need_inactive_args *ap) 5069 { 5070 vnode_t *vp = ap->a_vp; 5071 znode_t *zp = VTOZ(vp); 5072 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5073 int need; 5074 5075 if (vn_need_pageq_flush(vp)) 5076 return (1); 5077 5078 if (!ZFS_TEARDOWN_INACTIVE_TRY_ENTER_READ(zfsvfs)) 5079 return (1); 5080 need = (zp->z_sa_hdl == NULL || zp->z_unlinked || zp->z_atime_dirty); 5081 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 5082 5083 return (need); 5084 } 5085 5086 #ifndef _SYS_SYSPROTO_H_ 5087 struct vop_reclaim_args { 5088 struct vnode *a_vp; 5089 struct thread *a_td; 5090 }; 5091 #endif 5092 5093 static int 5094 zfs_freebsd_reclaim(struct vop_reclaim_args *ap) 5095 { 5096 vnode_t *vp = ap->a_vp; 5097 znode_t *zp = VTOZ(vp); 5098 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5099 5100 ASSERT3P(zp, !=, NULL); 5101 5102 /* 5103 * z_teardown_inactive_lock protects from a race with 5104 * zfs_znode_dmu_fini in zfsvfs_teardown during 5105 * force unmount. 5106 */ 5107 ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs); 5108 if (zp->z_sa_hdl == NULL) 5109 zfs_znode_free(zp); 5110 else 5111 zfs_zinactive(zp); 5112 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 5113 5114 vp->v_data = NULL; 5115 return (0); 5116 } 5117 5118 #ifndef _SYS_SYSPROTO_H_ 5119 struct vop_fid_args { 5120 struct vnode *a_vp; 5121 struct fid *a_fid; 5122 }; 5123 #endif 5124 5125 static int 5126 zfs_freebsd_fid(struct vop_fid_args *ap) 5127 { 5128 5129 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 5130 } 5131 5132 5133 #ifndef _SYS_SYSPROTO_H_ 5134 struct vop_pathconf_args { 5135 struct vnode *a_vp; 5136 int a_name; 5137 register_t *a_retval; 5138 } *ap; 5139 #endif 5140 5141 static int 5142 zfs_freebsd_pathconf(struct vop_pathconf_args *ap) 5143 { 5144 ulong_t val; 5145 int error; 5146 5147 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, 5148 curthread->td_ucred, NULL); 5149 if (error == 0) { 5150 *ap->a_retval = val; 5151 return (error); 5152 } 5153 if (error != EOPNOTSUPP) 5154 return (error); 5155 5156 switch (ap->a_name) { 5157 case _PC_NAME_MAX: 5158 *ap->a_retval = NAME_MAX; 5159 return (0); 5160 #if __FreeBSD_version >= 1400032 5161 case _PC_DEALLOC_PRESENT: 5162 *ap->a_retval = 1; 5163 return (0); 5164 #endif 5165 case _PC_PIPE_BUF: 5166 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) { 5167 *ap->a_retval = PIPE_BUF; 5168 return (0); 5169 } 5170 return (EINVAL); 5171 default: 5172 return (vop_stdpathconf(ap)); 5173 } 5174 } 5175 5176 static int zfs_xattr_compat = 1; 5177 5178 static int 5179 zfs_check_attrname(const char *name) 5180 { 5181 /* We don't allow '/' character in attribute name. */ 5182 if (strchr(name, '/') != NULL) 5183 return (SET_ERROR(EINVAL)); 5184 /* We don't allow attribute names that start with a namespace prefix. */ 5185 if (ZFS_XA_NS_PREFIX_FORBIDDEN(name)) 5186 return (SET_ERROR(EINVAL)); 5187 return (0); 5188 } 5189 5190 /* 5191 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 5192 * extended attribute name: 5193 * 5194 * NAMESPACE XATTR_COMPAT PREFIX 5195 * system * freebsd:system: 5196 * user 1 (none, can be used to access ZFS 5197 * fsattr(5) attributes created on Solaris) 5198 * user 0 user. 5199 */ 5200 static int 5201 zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 5202 size_t size, boolean_t compat) 5203 { 5204 const char *namespace, *prefix, *suffix; 5205 5206 memset(attrname, 0, size); 5207 5208 switch (attrnamespace) { 5209 case EXTATTR_NAMESPACE_USER: 5210 if (compat) { 5211 /* 5212 * This is the default namespace by which we can access 5213 * all attributes created on Solaris. 5214 */ 5215 prefix = namespace = suffix = ""; 5216 } else { 5217 /* 5218 * This is compatible with the user namespace encoding 5219 * on Linux prior to xattr_compat, but nothing 5220 * else. 5221 */ 5222 prefix = ""; 5223 namespace = "user"; 5224 suffix = "."; 5225 } 5226 break; 5227 case EXTATTR_NAMESPACE_SYSTEM: 5228 prefix = "freebsd:"; 5229 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 5230 suffix = ":"; 5231 break; 5232 case EXTATTR_NAMESPACE_EMPTY: 5233 default: 5234 return (SET_ERROR(EINVAL)); 5235 } 5236 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 5237 name) >= size) { 5238 return (SET_ERROR(ENAMETOOLONG)); 5239 } 5240 return (0); 5241 } 5242 5243 static int 5244 zfs_ensure_xattr_cached(znode_t *zp) 5245 { 5246 int error = 0; 5247 5248 ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock)); 5249 5250 if (zp->z_xattr_cached != NULL) 5251 return (0); 5252 5253 if (rw_write_held(&zp->z_xattr_lock)) 5254 return (zfs_sa_get_xattr(zp)); 5255 5256 if (!rw_tryupgrade(&zp->z_xattr_lock)) { 5257 rw_exit(&zp->z_xattr_lock); 5258 rw_enter(&zp->z_xattr_lock, RW_WRITER); 5259 } 5260 if (zp->z_xattr_cached == NULL) 5261 error = zfs_sa_get_xattr(zp); 5262 rw_downgrade(&zp->z_xattr_lock); 5263 return (error); 5264 } 5265 5266 #ifndef _SYS_SYSPROTO_H_ 5267 struct vop_getextattr { 5268 IN struct vnode *a_vp; 5269 IN int a_attrnamespace; 5270 IN const char *a_name; 5271 INOUT struct uio *a_uio; 5272 OUT size_t *a_size; 5273 IN struct ucred *a_cred; 5274 IN struct thread *a_td; 5275 }; 5276 #endif 5277 5278 static int 5279 zfs_getextattr_dir(struct vop_getextattr_args *ap, const char *attrname) 5280 { 5281 struct thread *td = ap->a_td; 5282 struct nameidata nd; 5283 struct vattr va; 5284 vnode_t *xvp = NULL, *vp; 5285 int error, flags; 5286 5287 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, 5288 LOOKUP_XATTR, B_FALSE); 5289 if (error != 0) 5290 return (error); 5291 5292 flags = FREAD; 5293 #if __FreeBSD_version < 1400043 5294 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 5295 xvp, td); 5296 #else 5297 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp); 5298 #endif 5299 error = vn_open_cred(&nd, &flags, 0, VN_OPEN_INVFS, ap->a_cred, NULL); 5300 if (error != 0) 5301 return (SET_ERROR(error)); 5302 vp = nd.ni_vp; 5303 NDFREE_PNBUF(&nd); 5304 5305 if (ap->a_size != NULL) { 5306 error = VOP_GETATTR(vp, &va, ap->a_cred); 5307 if (error == 0) 5308 *ap->a_size = (size_t)va.va_size; 5309 } else if (ap->a_uio != NULL) 5310 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); 5311 5312 VOP_UNLOCK(vp); 5313 vn_close(vp, flags, ap->a_cred, td); 5314 return (error); 5315 } 5316 5317 static int 5318 zfs_getextattr_sa(struct vop_getextattr_args *ap, const char *attrname) 5319 { 5320 znode_t *zp = VTOZ(ap->a_vp); 5321 uchar_t *nv_value; 5322 uint_t nv_size; 5323 int error; 5324 5325 error = zfs_ensure_xattr_cached(zp); 5326 if (error != 0) 5327 return (error); 5328 5329 ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock)); 5330 ASSERT3P(zp->z_xattr_cached, !=, NULL); 5331 5332 error = nvlist_lookup_byte_array(zp->z_xattr_cached, attrname, 5333 &nv_value, &nv_size); 5334 if (error != 0) 5335 return (SET_ERROR(error)); 5336 5337 if (ap->a_size != NULL) 5338 *ap->a_size = nv_size; 5339 else if (ap->a_uio != NULL) 5340 error = uiomove(nv_value, nv_size, ap->a_uio); 5341 if (error != 0) 5342 return (SET_ERROR(error)); 5343 5344 return (0); 5345 } 5346 5347 static int 5348 zfs_getextattr_impl(struct vop_getextattr_args *ap, boolean_t compat) 5349 { 5350 znode_t *zp = VTOZ(ap->a_vp); 5351 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5352 char attrname[EXTATTR_MAXNAMELEN+1]; 5353 int error; 5354 5355 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 5356 sizeof (attrname), compat); 5357 if (error != 0) 5358 return (error); 5359 5360 error = ENOENT; 5361 if (zfsvfs->z_use_sa && zp->z_is_sa) 5362 error = zfs_getextattr_sa(ap, attrname); 5363 if (error == ENOENT) 5364 error = zfs_getextattr_dir(ap, attrname); 5365 return (error); 5366 } 5367 5368 /* 5369 * Vnode operation to retrieve a named extended attribute. 5370 */ 5371 static int 5372 zfs_getextattr(struct vop_getextattr_args *ap) 5373 { 5374 znode_t *zp = VTOZ(ap->a_vp); 5375 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5376 int error; 5377 5378 /* 5379 * If the xattr property is off, refuse the request. 5380 */ 5381 if (!(zfsvfs->z_flags & ZSB_XATTR)) 5382 return (SET_ERROR(EOPNOTSUPP)); 5383 5384 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 5385 ap->a_cred, ap->a_td, VREAD); 5386 if (error != 0) 5387 return (SET_ERROR(error)); 5388 5389 error = zfs_check_attrname(ap->a_name); 5390 if (error != 0) 5391 return (error); 5392 5393 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 5394 return (error); 5395 error = ENOENT; 5396 rw_enter(&zp->z_xattr_lock, RW_READER); 5397 5398 error = zfs_getextattr_impl(ap, zfs_xattr_compat); 5399 if ((error == ENOENT || error == ENOATTR) && 5400 ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { 5401 /* 5402 * Fall back to the alternate namespace format if we failed to 5403 * find a user xattr. 5404 */ 5405 error = zfs_getextattr_impl(ap, !zfs_xattr_compat); 5406 } 5407 5408 rw_exit(&zp->z_xattr_lock); 5409 zfs_exit(zfsvfs, FTAG); 5410 if (error == ENOENT) 5411 error = SET_ERROR(ENOATTR); 5412 return (error); 5413 } 5414 5415 #ifndef _SYS_SYSPROTO_H_ 5416 struct vop_deleteextattr { 5417 IN struct vnode *a_vp; 5418 IN int a_attrnamespace; 5419 IN const char *a_name; 5420 IN struct ucred *a_cred; 5421 IN struct thread *a_td; 5422 }; 5423 #endif 5424 5425 static int 5426 zfs_deleteextattr_dir(struct vop_deleteextattr_args *ap, const char *attrname) 5427 { 5428 struct nameidata nd; 5429 vnode_t *xvp = NULL, *vp; 5430 int error; 5431 5432 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, 5433 LOOKUP_XATTR, B_FALSE); 5434 if (error != 0) 5435 return (error); 5436 5437 #if __FreeBSD_version < 1400043 5438 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 5439 UIO_SYSSPACE, attrname, xvp, ap->a_td); 5440 #else 5441 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 5442 UIO_SYSSPACE, attrname, xvp); 5443 #endif 5444 error = namei(&nd); 5445 if (error != 0) 5446 return (SET_ERROR(error)); 5447 5448 vp = nd.ni_vp; 5449 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 5450 NDFREE_PNBUF(&nd); 5451 5452 vput(nd.ni_dvp); 5453 if (vp == nd.ni_dvp) 5454 vrele(vp); 5455 else 5456 vput(vp); 5457 5458 return (error); 5459 } 5460 5461 static int 5462 zfs_deleteextattr_sa(struct vop_deleteextattr_args *ap, const char *attrname) 5463 { 5464 znode_t *zp = VTOZ(ap->a_vp); 5465 nvlist_t *nvl; 5466 int error; 5467 5468 error = zfs_ensure_xattr_cached(zp); 5469 if (error != 0) 5470 return (error); 5471 5472 ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock)); 5473 ASSERT3P(zp->z_xattr_cached, !=, NULL); 5474 5475 nvl = zp->z_xattr_cached; 5476 error = nvlist_remove(nvl, attrname, DATA_TYPE_BYTE_ARRAY); 5477 if (error != 0) 5478 error = SET_ERROR(error); 5479 else 5480 error = zfs_sa_set_xattr(zp, attrname, NULL, 0); 5481 if (error != 0) { 5482 zp->z_xattr_cached = NULL; 5483 nvlist_free(nvl); 5484 } 5485 return (error); 5486 } 5487 5488 static int 5489 zfs_deleteextattr_impl(struct vop_deleteextattr_args *ap, boolean_t compat) 5490 { 5491 znode_t *zp = VTOZ(ap->a_vp); 5492 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5493 char attrname[EXTATTR_MAXNAMELEN+1]; 5494 int error; 5495 5496 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 5497 sizeof (attrname), compat); 5498 if (error != 0) 5499 return (error); 5500 5501 error = ENOENT; 5502 if (zfsvfs->z_use_sa && zp->z_is_sa) 5503 error = zfs_deleteextattr_sa(ap, attrname); 5504 if (error == ENOENT) 5505 error = zfs_deleteextattr_dir(ap, attrname); 5506 return (error); 5507 } 5508 5509 /* 5510 * Vnode operation to remove a named attribute. 5511 */ 5512 static int 5513 zfs_deleteextattr(struct vop_deleteextattr_args *ap) 5514 { 5515 znode_t *zp = VTOZ(ap->a_vp); 5516 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5517 int error; 5518 5519 /* 5520 * If the xattr property is off, refuse the request. 5521 */ 5522 if (!(zfsvfs->z_flags & ZSB_XATTR)) 5523 return (SET_ERROR(EOPNOTSUPP)); 5524 5525 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 5526 ap->a_cred, ap->a_td, VWRITE); 5527 if (error != 0) 5528 return (SET_ERROR(error)); 5529 5530 error = zfs_check_attrname(ap->a_name); 5531 if (error != 0) 5532 return (error); 5533 5534 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 5535 return (error); 5536 rw_enter(&zp->z_xattr_lock, RW_WRITER); 5537 5538 error = zfs_deleteextattr_impl(ap, zfs_xattr_compat); 5539 if ((error == ENOENT || error == ENOATTR) && 5540 ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { 5541 /* 5542 * Fall back to the alternate namespace format if we failed to 5543 * find a user xattr. 5544 */ 5545 error = zfs_deleteextattr_impl(ap, !zfs_xattr_compat); 5546 } 5547 5548 rw_exit(&zp->z_xattr_lock); 5549 zfs_exit(zfsvfs, FTAG); 5550 if (error == ENOENT) 5551 error = SET_ERROR(ENOATTR); 5552 return (error); 5553 } 5554 5555 #ifndef _SYS_SYSPROTO_H_ 5556 struct vop_setextattr { 5557 IN struct vnode *a_vp; 5558 IN int a_attrnamespace; 5559 IN const char *a_name; 5560 INOUT struct uio *a_uio; 5561 IN struct ucred *a_cred; 5562 IN struct thread *a_td; 5563 }; 5564 #endif 5565 5566 static int 5567 zfs_setextattr_dir(struct vop_setextattr_args *ap, const char *attrname) 5568 { 5569 struct thread *td = ap->a_td; 5570 struct nameidata nd; 5571 struct vattr va; 5572 vnode_t *xvp = NULL, *vp; 5573 int error, flags; 5574 5575 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, 5576 LOOKUP_XATTR | CREATE_XATTR_DIR, B_FALSE); 5577 if (error != 0) 5578 return (error); 5579 5580 flags = FFLAGS(O_WRONLY | O_CREAT); 5581 #if __FreeBSD_version < 1400043 5582 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp, td); 5583 #else 5584 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp); 5585 #endif 5586 error = vn_open_cred(&nd, &flags, 0600, VN_OPEN_INVFS, ap->a_cred, 5587 NULL); 5588 if (error != 0) 5589 return (SET_ERROR(error)); 5590 vp = nd.ni_vp; 5591 NDFREE_PNBUF(&nd); 5592 5593 VATTR_NULL(&va); 5594 va.va_size = 0; 5595 error = VOP_SETATTR(vp, &va, ap->a_cred); 5596 if (error == 0) 5597 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred); 5598 5599 VOP_UNLOCK(vp); 5600 vn_close(vp, flags, ap->a_cred, td); 5601 return (error); 5602 } 5603 5604 static int 5605 zfs_setextattr_sa(struct vop_setextattr_args *ap, const char *attrname) 5606 { 5607 znode_t *zp = VTOZ(ap->a_vp); 5608 nvlist_t *nvl; 5609 size_t sa_size; 5610 int error; 5611 5612 error = zfs_ensure_xattr_cached(zp); 5613 if (error != 0) 5614 return (error); 5615 5616 ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock)); 5617 ASSERT3P(zp->z_xattr_cached, !=, NULL); 5618 5619 nvl = zp->z_xattr_cached; 5620 size_t entry_size = ap->a_uio->uio_resid; 5621 if (entry_size > DXATTR_MAX_ENTRY_SIZE) 5622 return (SET_ERROR(EFBIG)); 5623 error = nvlist_size(nvl, &sa_size, NV_ENCODE_XDR); 5624 if (error != 0) 5625 return (SET_ERROR(error)); 5626 if (sa_size > DXATTR_MAX_SA_SIZE) 5627 return (SET_ERROR(EFBIG)); 5628 uchar_t *buf = kmem_alloc(entry_size, KM_SLEEP); 5629 error = uiomove(buf, entry_size, ap->a_uio); 5630 if (error != 0) { 5631 error = SET_ERROR(error); 5632 } else { 5633 error = nvlist_add_byte_array(nvl, attrname, buf, entry_size); 5634 if (error != 0) 5635 error = SET_ERROR(error); 5636 } 5637 if (error == 0) 5638 error = zfs_sa_set_xattr(zp, attrname, buf, entry_size); 5639 kmem_free(buf, entry_size); 5640 if (error != 0) { 5641 zp->z_xattr_cached = NULL; 5642 nvlist_free(nvl); 5643 } 5644 return (error); 5645 } 5646 5647 static int 5648 zfs_setextattr_impl(struct vop_setextattr_args *ap, boolean_t compat) 5649 { 5650 znode_t *zp = VTOZ(ap->a_vp); 5651 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5652 char attrname[EXTATTR_MAXNAMELEN+1]; 5653 int error; 5654 5655 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 5656 sizeof (attrname), compat); 5657 if (error != 0) 5658 return (error); 5659 5660 struct vop_deleteextattr_args vda = { 5661 .a_vp = ap->a_vp, 5662 .a_attrnamespace = ap->a_attrnamespace, 5663 .a_name = ap->a_name, 5664 .a_cred = ap->a_cred, 5665 .a_td = ap->a_td, 5666 }; 5667 error = ENOENT; 5668 if (zfsvfs->z_use_sa && zp->z_is_sa && zfsvfs->z_xattr_sa) { 5669 error = zfs_setextattr_sa(ap, attrname); 5670 if (error == 0) { 5671 /* 5672 * Successfully put into SA, we need to clear the one 5673 * in dir if present. 5674 */ 5675 zfs_deleteextattr_dir(&vda, attrname); 5676 } 5677 } 5678 if (error != 0) { 5679 error = zfs_setextattr_dir(ap, attrname); 5680 if (error == 0 && zp->z_is_sa) { 5681 /* 5682 * Successfully put into dir, we need to clear the one 5683 * in SA if present. 5684 */ 5685 zfs_deleteextattr_sa(&vda, attrname); 5686 } 5687 } 5688 if (error == 0 && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { 5689 /* 5690 * Also clear all versions of the alternate compat name. 5691 */ 5692 zfs_deleteextattr_impl(&vda, !compat); 5693 } 5694 return (error); 5695 } 5696 5697 /* 5698 * Vnode operation to set a named attribute. 5699 */ 5700 static int 5701 zfs_setextattr(struct vop_setextattr_args *ap) 5702 { 5703 znode_t *zp = VTOZ(ap->a_vp); 5704 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5705 int error; 5706 5707 /* 5708 * If the xattr property is off, refuse the request. 5709 */ 5710 if (!(zfsvfs->z_flags & ZSB_XATTR)) 5711 return (SET_ERROR(EOPNOTSUPP)); 5712 5713 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 5714 ap->a_cred, ap->a_td, VWRITE); 5715 if (error != 0) 5716 return (SET_ERROR(error)); 5717 5718 error = zfs_check_attrname(ap->a_name); 5719 if (error != 0) 5720 return (error); 5721 5722 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 5723 return (error); 5724 rw_enter(&zp->z_xattr_lock, RW_WRITER); 5725 5726 error = zfs_setextattr_impl(ap, zfs_xattr_compat); 5727 5728 rw_exit(&zp->z_xattr_lock); 5729 zfs_exit(zfsvfs, FTAG); 5730 return (error); 5731 } 5732 5733 #ifndef _SYS_SYSPROTO_H_ 5734 struct vop_listextattr { 5735 IN struct vnode *a_vp; 5736 IN int a_attrnamespace; 5737 INOUT struct uio *a_uio; 5738 OUT size_t *a_size; 5739 IN struct ucred *a_cred; 5740 IN struct thread *a_td; 5741 }; 5742 #endif 5743 5744 static int 5745 zfs_listextattr_dir(struct vop_listextattr_args *ap, const char *attrprefix) 5746 { 5747 struct thread *td = ap->a_td; 5748 struct nameidata nd; 5749 uint8_t dirbuf[sizeof (struct dirent)]; 5750 struct iovec aiov; 5751 struct uio auio; 5752 vnode_t *xvp = NULL, *vp; 5753 int error, eof; 5754 5755 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, 5756 LOOKUP_XATTR, B_FALSE); 5757 if (error != 0) { 5758 /* 5759 * ENOATTR means that the EA directory does not yet exist, 5760 * i.e. there are no extended attributes there. 5761 */ 5762 if (error == ENOATTR) 5763 error = 0; 5764 return (error); 5765 } 5766 5767 #if __FreeBSD_version < 1400043 5768 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 5769 UIO_SYSSPACE, ".", xvp, td); 5770 #else 5771 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 5772 UIO_SYSSPACE, ".", xvp); 5773 #endif 5774 error = namei(&nd); 5775 if (error != 0) 5776 return (SET_ERROR(error)); 5777 vp = nd.ni_vp; 5778 NDFREE_PNBUF(&nd); 5779 5780 auio.uio_iov = &aiov; 5781 auio.uio_iovcnt = 1; 5782 auio.uio_segflg = UIO_SYSSPACE; 5783 auio.uio_td = td; 5784 auio.uio_rw = UIO_READ; 5785 auio.uio_offset = 0; 5786 5787 size_t plen = strlen(attrprefix); 5788 5789 do { 5790 aiov.iov_base = (void *)dirbuf; 5791 aiov.iov_len = sizeof (dirbuf); 5792 auio.uio_resid = sizeof (dirbuf); 5793 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 5794 if (error != 0) 5795 break; 5796 int done = sizeof (dirbuf) - auio.uio_resid; 5797 for (int pos = 0; pos < done; ) { 5798 struct dirent *dp = (struct dirent *)(dirbuf + pos); 5799 pos += dp->d_reclen; 5800 /* 5801 * XXX: Temporarily we also accept DT_UNKNOWN, as this 5802 * is what we get when attribute was created on Solaris. 5803 */ 5804 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 5805 continue; 5806 else if (plen == 0 && 5807 ZFS_XA_NS_PREFIX_FORBIDDEN(dp->d_name)) 5808 continue; 5809 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 5810 continue; 5811 uint8_t nlen = dp->d_namlen - plen; 5812 if (ap->a_size != NULL) { 5813 *ap->a_size += 1 + nlen; 5814 } else if (ap->a_uio != NULL) { 5815 /* 5816 * Format of extattr name entry is one byte for 5817 * length and the rest for name. 5818 */ 5819 error = uiomove(&nlen, 1, ap->a_uio); 5820 if (error == 0) { 5821 char *namep = dp->d_name + plen; 5822 error = uiomove(namep, nlen, ap->a_uio); 5823 } 5824 if (error != 0) { 5825 error = SET_ERROR(error); 5826 break; 5827 } 5828 } 5829 } 5830 } while (!eof && error == 0); 5831 5832 vput(vp); 5833 return (error); 5834 } 5835 5836 static int 5837 zfs_listextattr_sa(struct vop_listextattr_args *ap, const char *attrprefix) 5838 { 5839 znode_t *zp = VTOZ(ap->a_vp); 5840 int error; 5841 5842 error = zfs_ensure_xattr_cached(zp); 5843 if (error != 0) 5844 return (error); 5845 5846 ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock)); 5847 ASSERT3P(zp->z_xattr_cached, !=, NULL); 5848 5849 size_t plen = strlen(attrprefix); 5850 nvpair_t *nvp = NULL; 5851 while ((nvp = nvlist_next_nvpair(zp->z_xattr_cached, nvp)) != NULL) { 5852 ASSERT3U(nvpair_type(nvp), ==, DATA_TYPE_BYTE_ARRAY); 5853 5854 const char *name = nvpair_name(nvp); 5855 if (plen == 0 && ZFS_XA_NS_PREFIX_FORBIDDEN(name)) 5856 continue; 5857 else if (strncmp(name, attrprefix, plen) != 0) 5858 continue; 5859 uint8_t nlen = strlen(name) - plen; 5860 if (ap->a_size != NULL) { 5861 *ap->a_size += 1 + nlen; 5862 } else if (ap->a_uio != NULL) { 5863 /* 5864 * Format of extattr name entry is one byte for 5865 * length and the rest for name. 5866 */ 5867 error = uiomove(&nlen, 1, ap->a_uio); 5868 if (error == 0) { 5869 char *namep = __DECONST(char *, name) + plen; 5870 error = uiomove(namep, nlen, ap->a_uio); 5871 } 5872 if (error != 0) { 5873 error = SET_ERROR(error); 5874 break; 5875 } 5876 } 5877 } 5878 5879 return (error); 5880 } 5881 5882 static int 5883 zfs_listextattr_impl(struct vop_listextattr_args *ap, boolean_t compat) 5884 { 5885 znode_t *zp = VTOZ(ap->a_vp); 5886 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5887 char attrprefix[16]; 5888 int error; 5889 5890 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 5891 sizeof (attrprefix), compat); 5892 if (error != 0) 5893 return (error); 5894 5895 if (zfsvfs->z_use_sa && zp->z_is_sa) 5896 error = zfs_listextattr_sa(ap, attrprefix); 5897 if (error == 0) 5898 error = zfs_listextattr_dir(ap, attrprefix); 5899 return (error); 5900 } 5901 5902 /* 5903 * Vnode operation to retrieve extended attributes on a vnode. 5904 */ 5905 static int 5906 zfs_listextattr(struct vop_listextattr_args *ap) 5907 { 5908 znode_t *zp = VTOZ(ap->a_vp); 5909 zfsvfs_t *zfsvfs = ZTOZSB(zp); 5910 int error; 5911 5912 if (ap->a_size != NULL) 5913 *ap->a_size = 0; 5914 5915 /* 5916 * If the xattr property is off, refuse the request. 5917 */ 5918 if (!(zfsvfs->z_flags & ZSB_XATTR)) 5919 return (SET_ERROR(EOPNOTSUPP)); 5920 5921 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 5922 ap->a_cred, ap->a_td, VREAD); 5923 if (error != 0) 5924 return (SET_ERROR(error)); 5925 5926 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 5927 return (error); 5928 rw_enter(&zp->z_xattr_lock, RW_READER); 5929 5930 error = zfs_listextattr_impl(ap, zfs_xattr_compat); 5931 if (error == 0 && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { 5932 /* Also list user xattrs with the alternate format. */ 5933 error = zfs_listextattr_impl(ap, !zfs_xattr_compat); 5934 } 5935 5936 rw_exit(&zp->z_xattr_lock); 5937 zfs_exit(zfsvfs, FTAG); 5938 return (error); 5939 } 5940 5941 #ifndef _SYS_SYSPROTO_H_ 5942 struct vop_getacl_args { 5943 struct vnode *vp; 5944 acl_type_t type; 5945 struct acl *aclp; 5946 struct ucred *cred; 5947 struct thread *td; 5948 }; 5949 #endif 5950 5951 static int 5952 zfs_freebsd_getacl(struct vop_getacl_args *ap) 5953 { 5954 int error; 5955 vsecattr_t vsecattr; 5956 5957 if (ap->a_type != ACL_TYPE_NFS4) 5958 return (EINVAL); 5959 5960 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 5961 if ((error = zfs_getsecattr(VTOZ(ap->a_vp), 5962 &vsecattr, 0, ap->a_cred))) 5963 return (error); 5964 5965 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, 5966 vsecattr.vsa_aclcnt); 5967 if (vsecattr.vsa_aclentp != NULL) 5968 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 5969 5970 return (error); 5971 } 5972 5973 #ifndef _SYS_SYSPROTO_H_ 5974 struct vop_setacl_args { 5975 struct vnode *vp; 5976 acl_type_t type; 5977 struct acl *aclp; 5978 struct ucred *cred; 5979 struct thread *td; 5980 }; 5981 #endif 5982 5983 static int 5984 zfs_freebsd_setacl(struct vop_setacl_args *ap) 5985 { 5986 int error; 5987 vsecattr_t vsecattr; 5988 int aclbsize; /* size of acl list in bytes */ 5989 aclent_t *aaclp; 5990 5991 if (ap->a_type != ACL_TYPE_NFS4) 5992 return (EINVAL); 5993 5994 if (ap->a_aclp == NULL) 5995 return (EINVAL); 5996 5997 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 5998 return (EINVAL); 5999 6000 /* 6001 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 6002 * splitting every entry into two and appending "canonical six" 6003 * entries at the end. Don't allow for setting an ACL that would 6004 * cause chmod(2) to run out of ACL entries. 6005 */ 6006 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 6007 return (ENOSPC); 6008 6009 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 6010 if (error != 0) 6011 return (error); 6012 6013 vsecattr.vsa_mask = VSA_ACE; 6014 aclbsize = ap->a_aclp->acl_cnt * sizeof (ace_t); 6015 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 6016 aaclp = vsecattr.vsa_aclentp; 6017 vsecattr.vsa_aclentsz = aclbsize; 6018 6019 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 6020 error = zfs_setsecattr(VTOZ(ap->a_vp), &vsecattr, 0, ap->a_cred); 6021 kmem_free(aaclp, aclbsize); 6022 6023 return (error); 6024 } 6025 6026 #ifndef _SYS_SYSPROTO_H_ 6027 struct vop_aclcheck_args { 6028 struct vnode *vp; 6029 acl_type_t type; 6030 struct acl *aclp; 6031 struct ucred *cred; 6032 struct thread *td; 6033 }; 6034 #endif 6035 6036 static int 6037 zfs_freebsd_aclcheck(struct vop_aclcheck_args *ap) 6038 { 6039 6040 return (EOPNOTSUPP); 6041 } 6042 6043 static int 6044 zfs_vptocnp(struct vop_vptocnp_args *ap) 6045 { 6046 vnode_t *covered_vp; 6047 vnode_t *vp = ap->a_vp; 6048 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 6049 znode_t *zp = VTOZ(vp); 6050 int ltype; 6051 int error; 6052 6053 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 6054 return (error); 6055 6056 /* 6057 * If we are a snapshot mounted under .zfs, run the operation 6058 * on the covered vnode. 6059 */ 6060 if (zp->z_id != zfsvfs->z_root || zfsvfs->z_parent == zfsvfs) { 6061 char name[MAXNAMLEN + 1]; 6062 znode_t *dzp; 6063 size_t len; 6064 6065 error = zfs_znode_parent_and_name(zp, &dzp, name, 6066 sizeof (name)); 6067 if (error == 0) { 6068 len = strlen(name); 6069 if (*ap->a_buflen < len) 6070 error = SET_ERROR(ENOMEM); 6071 } 6072 if (error == 0) { 6073 *ap->a_buflen -= len; 6074 memcpy(ap->a_buf + *ap->a_buflen, name, len); 6075 *ap->a_vpp = ZTOV(dzp); 6076 } 6077 zfs_exit(zfsvfs, FTAG); 6078 return (error); 6079 } 6080 zfs_exit(zfsvfs, FTAG); 6081 6082 covered_vp = vp->v_mount->mnt_vnodecovered; 6083 enum vgetstate vs = vget_prep(covered_vp); 6084 ltype = VOP_ISLOCKED(vp); 6085 VOP_UNLOCK(vp); 6086 error = vget_finish(covered_vp, LK_SHARED, vs); 6087 if (error == 0) { 6088 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_buf, 6089 ap->a_buflen); 6090 vput(covered_vp); 6091 } 6092 vn_lock(vp, ltype | LK_RETRY); 6093 if (VN_IS_DOOMED(vp)) 6094 error = SET_ERROR(ENOENT); 6095 return (error); 6096 } 6097 6098 #if __FreeBSD_version >= 1400032 6099 static int 6100 zfs_deallocate(struct vop_deallocate_args *ap) 6101 { 6102 znode_t *zp = VTOZ(ap->a_vp); 6103 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6104 zilog_t *zilog; 6105 off_t off, len, file_sz; 6106 int error; 6107 6108 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 6109 return (error); 6110 6111 /* 6112 * Callers might not be able to detect properly that we are read-only, 6113 * so check it explicitly here. 6114 */ 6115 if (zfs_is_readonly(zfsvfs)) { 6116 zfs_exit(zfsvfs, FTAG); 6117 return (SET_ERROR(EROFS)); 6118 } 6119 6120 zilog = zfsvfs->z_log; 6121 off = *ap->a_offset; 6122 len = *ap->a_len; 6123 file_sz = zp->z_size; 6124 if (off + len > file_sz) 6125 len = file_sz - off; 6126 /* Fast path for out-of-range request. */ 6127 if (len <= 0) { 6128 *ap->a_len = 0; 6129 zfs_exit(zfsvfs, FTAG); 6130 return (0); 6131 } 6132 6133 error = zfs_freesp(zp, off, len, O_RDWR, TRUE); 6134 if (error == 0) { 6135 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS || 6136 (ap->a_ioflag & IO_SYNC) != 0) 6137 zil_commit(zilog, zp->z_id); 6138 *ap->a_offset = off + len; 6139 *ap->a_len = 0; 6140 } 6141 6142 zfs_exit(zfsvfs, FTAG); 6143 return (error); 6144 } 6145 #endif 6146 6147 #ifndef _SYS_SYSPROTO_H_ 6148 struct vop_copy_file_range_args { 6149 struct vnode *a_invp; 6150 off_t *a_inoffp; 6151 struct vnode *a_outvp; 6152 off_t *a_outoffp; 6153 size_t *a_lenp; 6154 unsigned int a_flags; 6155 struct ucred *a_incred; 6156 struct ucred *a_outcred; 6157 struct thread *a_fsizetd; 6158 } 6159 #endif 6160 /* 6161 * TODO: FreeBSD will only call file system-specific copy_file_range() if both 6162 * files resides under the same mountpoint. In case of ZFS we want to be called 6163 * even is files are in different datasets (but on the same pools, but we need 6164 * to check that ourselves). 6165 */ 6166 static int 6167 zfs_freebsd_copy_file_range(struct vop_copy_file_range_args *ap) 6168 { 6169 zfsvfs_t *outzfsvfs; 6170 struct vnode *invp = ap->a_invp; 6171 struct vnode *outvp = ap->a_outvp; 6172 struct mount *mp; 6173 int error; 6174 uint64_t len = *ap->a_lenp; 6175 6176 if (!zfs_bclone_enabled) { 6177 mp = NULL; 6178 goto bad_write_fallback; 6179 } 6180 6181 /* 6182 * TODO: If offset/length is not aligned to recordsize, use 6183 * vn_generic_copy_file_range() on this fragment. 6184 * It would be better to do this after we lock the vnodes, but then we 6185 * need something else than vn_generic_copy_file_range(). 6186 */ 6187 6188 vn_start_write(outvp, &mp, V_WAIT); 6189 if (__predict_true(mp == outvp->v_mount)) { 6190 outzfsvfs = (zfsvfs_t *)mp->mnt_data; 6191 if (!spa_feature_is_enabled(dmu_objset_spa(outzfsvfs->z_os), 6192 SPA_FEATURE_BLOCK_CLONING)) { 6193 goto bad_write_fallback; 6194 } 6195 } 6196 if (invp == outvp) { 6197 if (vn_lock(outvp, LK_EXCLUSIVE) != 0) { 6198 goto bad_write_fallback; 6199 } 6200 } else { 6201 #if (__FreeBSD_version >= 1302506 && __FreeBSD_version < 1400000) || \ 6202 __FreeBSD_version >= 1400086 6203 vn_lock_pair(invp, false, LK_SHARED, outvp, false, 6204 LK_EXCLUSIVE); 6205 #else 6206 vn_lock_pair(invp, false, outvp, false); 6207 #endif 6208 if (VN_IS_DOOMED(invp) || VN_IS_DOOMED(outvp)) { 6209 goto bad_locked_fallback; 6210 } 6211 } 6212 6213 #ifdef MAC 6214 error = mac_vnode_check_write(curthread->td_ucred, ap->a_outcred, 6215 outvp); 6216 if (error != 0) 6217 goto out_locked; 6218 #endif 6219 6220 error = zfs_clone_range(VTOZ(invp), ap->a_inoffp, VTOZ(outvp), 6221 ap->a_outoffp, &len, ap->a_outcred); 6222 if (error == EXDEV || error == EAGAIN || error == EINVAL || 6223 error == EOPNOTSUPP) 6224 goto bad_locked_fallback; 6225 *ap->a_lenp = (size_t)len; 6226 #ifdef MAC 6227 out_locked: 6228 #endif 6229 if (invp != outvp) 6230 VOP_UNLOCK(invp); 6231 VOP_UNLOCK(outvp); 6232 if (mp != NULL) 6233 vn_finished_write(mp); 6234 return (error); 6235 6236 bad_locked_fallback: 6237 if (invp != outvp) 6238 VOP_UNLOCK(invp); 6239 VOP_UNLOCK(outvp); 6240 bad_write_fallback: 6241 if (mp != NULL) 6242 vn_finished_write(mp); 6243 error = ENOSYS; 6244 return (error); 6245 } 6246 6247 struct vop_vector zfs_vnodeops; 6248 struct vop_vector zfs_fifoops; 6249 struct vop_vector zfs_shareops; 6250 6251 struct vop_vector zfs_vnodeops = { 6252 .vop_default = &default_vnodeops, 6253 .vop_inactive = zfs_freebsd_inactive, 6254 .vop_need_inactive = zfs_freebsd_need_inactive, 6255 .vop_reclaim = zfs_freebsd_reclaim, 6256 .vop_fplookup_vexec = zfs_freebsd_fplookup_vexec, 6257 .vop_fplookup_symlink = zfs_freebsd_fplookup_symlink, 6258 .vop_access = zfs_freebsd_access, 6259 .vop_allocate = VOP_EOPNOTSUPP, 6260 #if __FreeBSD_version >= 1400032 6261 .vop_deallocate = zfs_deallocate, 6262 #endif 6263 .vop_lookup = zfs_cache_lookup, 6264 .vop_cachedlookup = zfs_freebsd_cachedlookup, 6265 .vop_getattr = zfs_freebsd_getattr, 6266 .vop_setattr = zfs_freebsd_setattr, 6267 .vop_create = zfs_freebsd_create, 6268 .vop_mknod = (vop_mknod_t *)zfs_freebsd_create, 6269 .vop_mkdir = zfs_freebsd_mkdir, 6270 .vop_readdir = zfs_freebsd_readdir, 6271 .vop_fsync = zfs_freebsd_fsync, 6272 .vop_open = zfs_freebsd_open, 6273 .vop_close = zfs_freebsd_close, 6274 .vop_rmdir = zfs_freebsd_rmdir, 6275 .vop_ioctl = zfs_freebsd_ioctl, 6276 .vop_link = zfs_freebsd_link, 6277 .vop_symlink = zfs_freebsd_symlink, 6278 .vop_readlink = zfs_freebsd_readlink, 6279 .vop_read = zfs_freebsd_read, 6280 .vop_write = zfs_freebsd_write, 6281 .vop_remove = zfs_freebsd_remove, 6282 .vop_rename = zfs_freebsd_rename, 6283 .vop_pathconf = zfs_freebsd_pathconf, 6284 .vop_bmap = zfs_freebsd_bmap, 6285 .vop_fid = zfs_freebsd_fid, 6286 .vop_getextattr = zfs_getextattr, 6287 .vop_deleteextattr = zfs_deleteextattr, 6288 .vop_setextattr = zfs_setextattr, 6289 .vop_listextattr = zfs_listextattr, 6290 .vop_getacl = zfs_freebsd_getacl, 6291 .vop_setacl = zfs_freebsd_setacl, 6292 .vop_aclcheck = zfs_freebsd_aclcheck, 6293 .vop_getpages = zfs_freebsd_getpages, 6294 .vop_putpages = zfs_freebsd_putpages, 6295 .vop_vptocnp = zfs_vptocnp, 6296 .vop_lock1 = vop_lock, 6297 .vop_unlock = vop_unlock, 6298 .vop_islocked = vop_islocked, 6299 #if __FreeBSD_version >= 1400043 6300 .vop_add_writecount = vop_stdadd_writecount_nomsync, 6301 #endif 6302 .vop_copy_file_range = zfs_freebsd_copy_file_range, 6303 }; 6304 VFS_VOP_VECTOR_REGISTER(zfs_vnodeops); 6305 6306 struct vop_vector zfs_fifoops = { 6307 .vop_default = &fifo_specops, 6308 .vop_fsync = zfs_freebsd_fsync, 6309 .vop_fplookup_vexec = zfs_freebsd_fplookup_vexec, 6310 .vop_fplookup_symlink = zfs_freebsd_fplookup_symlink, 6311 .vop_access = zfs_freebsd_access, 6312 .vop_getattr = zfs_freebsd_getattr, 6313 .vop_inactive = zfs_freebsd_inactive, 6314 .vop_read = VOP_PANIC, 6315 .vop_reclaim = zfs_freebsd_reclaim, 6316 .vop_setattr = zfs_freebsd_setattr, 6317 .vop_write = VOP_PANIC, 6318 .vop_pathconf = zfs_freebsd_pathconf, 6319 .vop_fid = zfs_freebsd_fid, 6320 .vop_getacl = zfs_freebsd_getacl, 6321 .vop_setacl = zfs_freebsd_setacl, 6322 .vop_aclcheck = zfs_freebsd_aclcheck, 6323 #if __FreeBSD_version >= 1400043 6324 .vop_add_writecount = vop_stdadd_writecount_nomsync, 6325 #endif 6326 }; 6327 VFS_VOP_VECTOR_REGISTER(zfs_fifoops); 6328 6329 /* 6330 * special share hidden files vnode operations template 6331 */ 6332 struct vop_vector zfs_shareops = { 6333 .vop_default = &default_vnodeops, 6334 .vop_fplookup_vexec = VOP_EAGAIN, 6335 .vop_fplookup_symlink = VOP_EAGAIN, 6336 .vop_access = zfs_freebsd_access, 6337 .vop_inactive = zfs_freebsd_inactive, 6338 .vop_reclaim = zfs_freebsd_reclaim, 6339 .vop_fid = zfs_freebsd_fid, 6340 .vop_pathconf = zfs_freebsd_pathconf, 6341 #if __FreeBSD_version >= 1400043 6342 .vop_add_writecount = vop_stdadd_writecount_nomsync, 6343 #endif 6344 }; 6345 VFS_VOP_VECTOR_REGISTER(zfs_shareops); 6346 6347 ZFS_MODULE_PARAM(zfs, zfs_, xattr_compat, INT, ZMOD_RW, 6348 "Use legacy ZFS xattr naming for writing new user namespace xattrs"); 6349