1 // SPDX-License-Identifier: CDDL-1.0 2 /* 3 * CDDL HEADER START 4 * 5 * The contents of this file are subject to the terms of the 6 * Common Development and Distribution License (the "License"). 7 * You may not use this file except in compliance with the License. 8 * 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10 * or https://opensource.org/licenses/CDDL-1.0. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When distributing Covered Code, include this CDDL HEADER in each 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16 * If applicable, add the following below this CDDL HEADER, with the 17 * fields enclosed by brackets "[]" replaced with your own identifying 18 * information: Portions Copyright [yyyy] [name of copyright owner] 19 * 20 * CDDL HEADER END 21 */ 22 23 /* 24 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 25 * Copyright (c) 2012, 2015 by Delphix. All rights reserved. 26 * Copyright (c) 2014 Integros [integros.com] 27 * Copyright 2017 Nexenta Systems, Inc. 28 * Copyright (c) 2025, Klara, Inc. 29 */ 30 31 /* Portions Copyright 2007 Jeremy Teo */ 32 /* Portions Copyright 2010 Robert Milkowski */ 33 34 #include <sys/param.h> 35 #include <sys/time.h> 36 #include <sys/systm.h> 37 #include <sys/sysmacros.h> 38 #include <sys/resource.h> 39 #include <security/mac/mac_framework.h> 40 #include <sys/vfs.h> 41 #include <sys/endian.h> 42 #include <sys/vm.h> 43 #include <sys/vnode.h> 44 #include <sys/smr.h> 45 #include <sys/dirent.h> 46 #include <sys/file.h> 47 #include <sys/stat.h> 48 #include <sys/kmem.h> 49 #include <sys/taskq.h> 50 #include <sys/uio.h> 51 #include <sys/atomic.h> 52 #include <sys/namei.h> 53 #include <sys/mman.h> 54 #include <sys/cmn_err.h> 55 #include <sys/kdb.h> 56 #include <sys/sysproto.h> 57 #include <sys/errno.h> 58 #include <sys/unistd.h> 59 #include <sys/zfs_dir.h> 60 #include <sys/zfs_ioctl.h> 61 #include <sys/fs/zfs.h> 62 #include <sys/dmu.h> 63 #include <sys/dmu_objset.h> 64 #include <sys/dsl_dataset.h> 65 #include <sys/spa.h> 66 #include <sys/txg.h> 67 #include <sys/dbuf.h> 68 #include <sys/zap.h> 69 #include <sys/sa.h> 70 #include <sys/policy.h> 71 #include <sys/sunddi.h> 72 #include <sys/filio.h> 73 #include <sys/sid.h> 74 #include <sys/zfs_ctldir.h> 75 #include <sys/zfs_fuid.h> 76 #include <sys/zfs_quota.h> 77 #include <sys/zfs_sa.h> 78 #include <sys/zfs_rlock.h> 79 #include <sys/zfs_project.h> 80 #include <sys/bio.h> 81 #include <sys/buf.h> 82 #include <sys/sched.h> 83 #include <sys/acl.h> 84 #include <sys/vmmeter.h> 85 #include <vm/vm_param.h> 86 #include <sys/zil.h> 87 #include <sys/zfs_vnops.h> 88 #include <sys/module.h> 89 #include <sys/sysent.h> 90 #include <sys/dmu_impl.h> 91 #include <sys/brt.h> 92 #include <sys/zfeature.h> 93 94 #include <vm/vm_object.h> 95 96 #include <sys/extattr.h> 97 #include <sys/priv.h> 98 99 #ifndef VN_OPEN_INVFS 100 #define VN_OPEN_INVFS 0x0 101 #endif 102 103 VFS_SMR_DECLARE; 104 105 #ifdef DEBUG_VFS_LOCKS 106 #define VNCHECKREF(vp) \ 107 VNASSERT((vp)->v_holdcnt > 0 && (vp)->v_usecount > 0, vp, \ 108 ("%s: wrong ref counts", __func__)); 109 #else 110 #define VNCHECKREF(vp) 111 #endif 112 113 #if __FreeBSD_version >= 1400045 114 typedef uint64_t cookie_t; 115 #else 116 typedef ulong_t cookie_t; 117 #endif 118 119 static int zfs_check_attrname(const char *name); 120 121 /* 122 * Programming rules. 123 * 124 * Each vnode op performs some logical unit of work. To do this, the ZPL must 125 * properly lock its in-core state, create a DMU transaction, do the work, 126 * record this work in the intent log (ZIL), commit the DMU transaction, 127 * and wait for the intent log to commit if it is a synchronous operation. 128 * Moreover, the vnode ops must work in both normal and log replay context. 129 * The ordering of events is important to avoid deadlocks and references 130 * to freed memory. The example below illustrates the following Big Rules: 131 * 132 * (1) A check must be made in each zfs thread for a mounted file system. 133 * This is done avoiding races using zfs_enter(zfsvfs). 134 * A zfs_exit(zfsvfs) is needed before all returns. Any znodes 135 * must be checked with zfs_verify_zp(zp). Both of these macros 136 * can return EIO from the calling function. 137 * 138 * (2) VN_RELE() should always be the last thing except for zil_commit() 139 * (if necessary) and zfs_exit(). This is for 3 reasons: 140 * First, if it's the last reference, the vnode/znode 141 * can be freed, so the zp may point to freed memory. Second, the last 142 * reference will call zfs_zinactive(), which may induce a lot of work -- 143 * pushing cached pages (which acquires range locks) and syncing out 144 * cached atime changes. Third, zfs_zinactive() may require a new tx, 145 * which could deadlock the system if you were already holding one. 146 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). 147 * 148 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 149 * as they can span dmu_tx_assign() calls. 150 * 151 * (4) If ZPL locks are held, pass DMU_TX_NOWAIT as the second argument to 152 * dmu_tx_assign(). This is critical because we don't want to block 153 * while holding locks. 154 * 155 * If no ZPL locks are held (aside from zfs_enter()), use DMU_TX_WAIT. 156 * This reduces lock contention and CPU usage when we must wait (note 157 * that if throughput is constrained by the storage, nearly every 158 * transaction must wait). 159 * 160 * Note, in particular, that if a lock is sometimes acquired before 161 * the tx assigns, and sometimes after (e.g. z_lock), then failing 162 * to use a non-blocking assign can deadlock the system. The scenario: 163 * 164 * Thread A has grabbed a lock before calling dmu_tx_assign(). 165 * Thread B is in an already-assigned tx, and blocks for this lock. 166 * Thread A calls dmu_tx_assign(DMU_TX_WAIT) and blocks in 167 * txg_wait_open() forever, because the previous txg can't quiesce 168 * until B's tx commits. 169 * 170 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is 171 * DMU_TX_NOWAIT, then drop all locks, call dmu_tx_wait(), and try 172 * again. On subsequent calls to dmu_tx_assign(), pass 173 * DMU_TX_NOTHROTTLE in addition to DMU_TX_NOWAIT, to indicate that 174 * this operation has already called dmu_tx_wait(). This will ensure 175 * that we don't retry forever, waiting a short bit each time. 176 * 177 * (5) If the operation succeeded, generate the intent log entry for it 178 * before dropping locks. This ensures that the ordering of events 179 * in the intent log matches the order in which they actually occurred. 180 * During ZIL replay the zfs_log_* functions will update the sequence 181 * number to indicate the zil transaction has replayed. 182 * 183 * (6) At the end of each vnode op, the DMU tx must always commit, 184 * regardless of whether there were any errors. 185 * 186 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 187 * to ensure that synchronous semantics are provided when necessary. 188 * 189 * In general, this is how things should be ordered in each vnode op: 190 * 191 * zfs_enter(zfsvfs); // exit if unmounted 192 * top: 193 * zfs_dirent_lookup(&dl, ...) // lock directory entry (may VN_HOLD()) 194 * rw_enter(...); // grab any other locks you need 195 * tx = dmu_tx_create(...); // get DMU tx 196 * dmu_tx_hold_*(); // hold each object you might modify 197 * error = dmu_tx_assign(tx, 198 * (waited ? DMU_TX_NOTHROTTLE : 0) | DMU_TX_NOWAIT); 199 * if (error) { 200 * rw_exit(...); // drop locks 201 * zfs_dirent_unlock(dl); // unlock directory entry 202 * VN_RELE(...); // release held vnodes 203 * if (error == ERESTART) { 204 * waited = B_TRUE; 205 * dmu_tx_wait(tx); 206 * dmu_tx_abort(tx); 207 * goto top; 208 * } 209 * dmu_tx_abort(tx); // abort DMU tx 210 * zfs_exit(zfsvfs); // finished in zfs 211 * return (error); // really out of space 212 * } 213 * error = do_real_work(); // do whatever this VOP does 214 * if (error == 0) 215 * zfs_log_*(...); // on success, make ZIL entry 216 * dmu_tx_commit(tx); // commit DMU tx -- error or not 217 * rw_exit(...); // drop locks 218 * zfs_dirent_unlock(dl); // unlock directory entry 219 * VN_RELE(...); // release held vnodes 220 * zil_commit(zilog, foid); // synchronous when necessary 221 * zfs_exit(zfsvfs); // finished in zfs 222 * return (error); // done, report error 223 */ 224 static int 225 zfs_open(vnode_t **vpp, int flag, cred_t *cr) 226 { 227 (void) cr; 228 znode_t *zp = VTOZ(*vpp); 229 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 230 int error; 231 232 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 233 return (error); 234 235 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 236 ((flag & FAPPEND) == 0)) { 237 zfs_exit(zfsvfs, FTAG); 238 return (SET_ERROR(EPERM)); 239 } 240 241 /* 242 * Keep a count of the synchronous opens in the znode. On first 243 * synchronous open we must convert all previous async transactions 244 * into sync to keep correct ordering. 245 */ 246 if (flag & O_SYNC) { 247 if (atomic_inc_32_nv(&zp->z_sync_cnt) == 1) 248 zil_async_to_sync(zfsvfs->z_log, zp->z_id); 249 } 250 251 zfs_exit(zfsvfs, FTAG); 252 return (0); 253 } 254 255 static int 256 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr) 257 { 258 (void) offset, (void) cr; 259 znode_t *zp = VTOZ(vp); 260 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 261 int error; 262 263 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 264 return (error); 265 266 /* Decrement the synchronous opens in the znode */ 267 if ((flag & O_SYNC) && (count == 1)) 268 atomic_dec_32(&zp->z_sync_cnt); 269 270 zfs_exit(zfsvfs, FTAG); 271 return (0); 272 } 273 274 static int 275 zfs_ioctl_getxattr(vnode_t *vp, zfsxattr_t *fsx) 276 { 277 znode_t *zp = VTOZ(vp); 278 279 memset(fsx, 0, sizeof (*fsx)); 280 fsx->fsx_xflags = (zp->z_pflags & ZFS_PROJINHERIT) ? 281 FS_PROJINHERIT_FL : 0; 282 fsx->fsx_projid = zp->z_projid; 283 284 return (0); 285 } 286 287 static int 288 zfs_ioctl_setflags(vnode_t *vp, uint32_t ioctl_flags, xvattr_t *xva) 289 { 290 uint64_t zfs_flags = VTOZ(vp)->z_pflags; 291 xoptattr_t *xoap; 292 293 if (ioctl_flags & ~(FS_PROJINHERIT_FL)) 294 return (SET_ERROR(EOPNOTSUPP)); 295 296 xva_init(xva); 297 xoap = xva_getxoptattr(xva); 298 299 #define FLAG_CHANGE(iflag, zflag, xflag, xfield) do { \ 300 if (((ioctl_flags & (iflag)) && !(zfs_flags & (zflag))) || \ 301 ((zfs_flags & (zflag)) && !(ioctl_flags & (iflag)))) { \ 302 XVA_SET_REQ(xva, (xflag)); \ 303 (xfield) = ((ioctl_flags & (iflag)) != 0); \ 304 } \ 305 } while (0) 306 307 FLAG_CHANGE(FS_PROJINHERIT_FL, ZFS_PROJINHERIT, XAT_PROJINHERIT, 308 xoap->xoa_projinherit); 309 310 #undef FLAG_CHANGE 311 312 return (0); 313 } 314 315 static int 316 zfs_ioctl_setxattr(vnode_t *vp, zfsxattr_t *fsx, cred_t *cr) 317 { 318 znode_t *zp = VTOZ(vp); 319 xvattr_t xva; 320 xoptattr_t *xoap; 321 int err; 322 323 if (!zpl_is_valid_projid(fsx->fsx_projid)) 324 return (SET_ERROR(EINVAL)); 325 326 err = zfs_ioctl_setflags(vp, fsx->fsx_xflags, &xva); 327 if (err) 328 return (err); 329 330 xoap = xva_getxoptattr(&xva); 331 XVA_SET_REQ(&xva, XAT_PROJID); 332 xoap->xoa_projid = fsx->fsx_projid; 333 334 err = zfs_setattr(zp, (vattr_t *)&xva, 0, cr, NULL); 335 336 return (err); 337 } 338 339 static int 340 zfs_ioctl(vnode_t *vp, ulong_t com, intptr_t data, int flag, cred_t *cred, 341 int *rvalp) 342 { 343 (void) flag, (void) cred, (void) rvalp; 344 loff_t off; 345 int error; 346 347 switch (com) { 348 case _FIOFFS: 349 { 350 return (0); 351 352 /* 353 * The following two ioctls are used by bfu. Faking out, 354 * necessary to avoid bfu errors. 355 */ 356 } 357 case _FIOGDIO: 358 case _FIOSDIO: 359 { 360 return (0); 361 } 362 363 case F_SEEK_DATA: 364 case F_SEEK_HOLE: 365 { 366 off = *(offset_t *)data; 367 error = vn_lock(vp, LK_SHARED); 368 if (error) 369 return (error); 370 /* offset parameter is in/out */ 371 error = zfs_holey(VTOZ(vp), com, &off); 372 VOP_UNLOCK(vp); 373 if (error) 374 return (error); 375 *(offset_t *)data = off; 376 return (0); 377 } 378 case ZFS_IOC_FSGETXATTR: { 379 zfsxattr_t *fsx = (zfsxattr_t *)data; 380 error = vn_lock(vp, LK_SHARED); 381 if (error) 382 return (error); 383 error = zfs_ioctl_getxattr(vp, fsx); 384 VOP_UNLOCK(vp); 385 return (error); 386 } 387 case ZFS_IOC_FSSETXATTR: { 388 zfsxattr_t *fsx = (zfsxattr_t *)data; 389 error = vn_lock(vp, LK_EXCLUSIVE); 390 if (error) 391 return (error); 392 vn_seqc_write_begin(vp); 393 error = zfs_ioctl_setxattr(vp, fsx, cred); 394 vn_seqc_write_end(vp); 395 VOP_UNLOCK(vp); 396 return (error); 397 } 398 case ZFS_IOC_REWRITE: { 399 zfs_rewrite_args_t *args = (zfs_rewrite_args_t *)data; 400 if ((flag & FWRITE) == 0) 401 return (SET_ERROR(EBADF)); 402 error = vn_lock(vp, LK_SHARED); 403 if (error) 404 return (error); 405 error = zfs_rewrite(VTOZ(vp), args->off, args->len, 406 args->flags, args->arg); 407 VOP_UNLOCK(vp); 408 return (error); 409 } 410 } 411 return (SET_ERROR(ENOTTY)); 412 } 413 414 static vm_page_t 415 page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) 416 { 417 vm_object_t obj; 418 vm_page_t pp; 419 int64_t end; 420 421 /* 422 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE 423 * aligned boundaries, if the range is not aligned. As a result a 424 * DEV_BSIZE subrange with partially dirty data may get marked as clean. 425 * It may happen that all DEV_BSIZE subranges are marked clean and thus 426 * the whole page would be considered clean despite have some 427 * dirty data. 428 * For this reason we should shrink the range to DEV_BSIZE aligned 429 * boundaries before calling vm_page_clear_dirty. 430 */ 431 end = rounddown2(off + nbytes, DEV_BSIZE); 432 off = roundup2(off, DEV_BSIZE); 433 nbytes = end - off; 434 435 obj = vp->v_object; 436 vm_page_grab_valid_unlocked(&pp, obj, OFF_TO_IDX(start), 437 VM_ALLOC_NOCREAT | VM_ALLOC_SBUSY | VM_ALLOC_NORMAL | 438 VM_ALLOC_IGN_SBUSY); 439 if (pp != NULL) { 440 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 441 vm_object_pip_add(obj, 1); 442 pmap_remove_write(pp); 443 if (nbytes != 0) 444 vm_page_clear_dirty(pp, off, nbytes); 445 } 446 return (pp); 447 } 448 449 static void 450 page_unbusy(vm_page_t pp) 451 { 452 453 vm_page_sunbusy(pp); 454 vm_object_pip_wakeup(pp->object); 455 } 456 457 static vm_page_t 458 page_hold(vnode_t *vp, int64_t start) 459 { 460 vm_object_t obj; 461 vm_page_t m; 462 463 obj = vp->v_object; 464 vm_page_grab_valid_unlocked(&m, obj, OFF_TO_IDX(start), 465 VM_ALLOC_NOCREAT | VM_ALLOC_WIRED | VM_ALLOC_IGN_SBUSY | 466 VM_ALLOC_NOBUSY); 467 return (m); 468 } 469 470 static void 471 page_unhold(vm_page_t pp) 472 { 473 vm_page_unwire(pp, PQ_ACTIVE); 474 } 475 476 /* 477 * When a file is memory mapped, we must keep the IO data synchronized 478 * between the DMU cache and the memory mapped pages. What this means: 479 * 480 * On Write: If we find a memory mapped page, we write to *both* 481 * the page and the dmu buffer. 482 */ 483 void 484 update_pages(znode_t *zp, int64_t start, int len, objset_t *os) 485 { 486 vm_object_t obj; 487 struct sf_buf *sf; 488 vnode_t *vp = ZTOV(zp); 489 caddr_t va; 490 int off; 491 492 ASSERT3P(vp->v_mount, !=, NULL); 493 obj = vp->v_object; 494 ASSERT3P(obj, !=, NULL); 495 496 off = start & PAGEOFFSET; 497 vm_object_pip_add(obj, 1); 498 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 499 vm_page_t pp; 500 int nbytes = imin(PAGESIZE - off, len); 501 502 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) { 503 va = zfs_map_page(pp, &sf); 504 (void) dmu_read(os, zp->z_id, start + off, nbytes, 505 va + off, DMU_READ_PREFETCH); 506 zfs_unmap_page(sf); 507 page_unbusy(pp); 508 } 509 len -= nbytes; 510 off = 0; 511 } 512 vm_object_pip_wakeup(obj); 513 } 514 515 /* 516 * Read with UIO_NOCOPY flag means that sendfile(2) requests 517 * ZFS to populate a range of page cache pages with data. 518 * 519 * NOTE: this function could be optimized to pre-allocate 520 * all pages in advance, drain exclusive busy on all of them, 521 * map them into contiguous KVA region and populate them 522 * in one single dmu_read() call. 523 */ 524 int 525 mappedread_sf(znode_t *zp, int nbytes, zfs_uio_t *uio) 526 { 527 vnode_t *vp = ZTOV(zp); 528 objset_t *os = zp->z_zfsvfs->z_os; 529 struct sf_buf *sf; 530 vm_object_t obj; 531 vm_page_t pp; 532 int64_t start; 533 caddr_t va; 534 int len = nbytes; 535 int error = 0; 536 537 ASSERT3U(zfs_uio_segflg(uio), ==, UIO_NOCOPY); 538 ASSERT3P(vp->v_mount, !=, NULL); 539 obj = vp->v_object; 540 ASSERT3P(obj, !=, NULL); 541 ASSERT0(zfs_uio_offset(uio) & PAGEOFFSET); 542 543 for (start = zfs_uio_offset(uio); len > 0; start += PAGESIZE) { 544 int bytes = MIN(PAGESIZE, len); 545 546 pp = vm_page_grab_unlocked(obj, OFF_TO_IDX(start), 547 VM_ALLOC_SBUSY | VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); 548 if (vm_page_none_valid(pp)) { 549 va = zfs_map_page(pp, &sf); 550 error = dmu_read(os, zp->z_id, start, bytes, va, 551 DMU_READ_PREFETCH); 552 if (bytes != PAGESIZE && error == 0) 553 memset(va + bytes, 0, PAGESIZE - bytes); 554 zfs_unmap_page(sf); 555 if (error == 0) { 556 vm_page_valid(pp); 557 vm_page_activate(pp); 558 vm_page_sunbusy(pp); 559 } else { 560 zfs_vmobject_wlock(obj); 561 if (!vm_page_wired(pp) && pp->valid == 0 && 562 vm_page_busy_tryupgrade(pp)) 563 vm_page_free(pp); 564 else { 565 vm_page_deactivate_noreuse(pp); 566 vm_page_sunbusy(pp); 567 } 568 zfs_vmobject_wunlock(obj); 569 } 570 } else { 571 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 572 vm_page_sunbusy(pp); 573 } 574 if (error) 575 break; 576 zfs_uio_advance(uio, bytes); 577 len -= bytes; 578 } 579 return (error); 580 } 581 582 /* 583 * When a file is memory mapped, we must keep the IO data synchronized 584 * between the DMU cache and the memory mapped pages. What this means: 585 * 586 * On Read: We "read" preferentially from memory mapped pages, 587 * else we default from the dmu buffer. 588 * 589 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 590 * the file is memory mapped. 591 */ 592 int 593 mappedread(znode_t *zp, int nbytes, zfs_uio_t *uio) 594 { 595 vnode_t *vp = ZTOV(zp); 596 vm_object_t obj; 597 int64_t start; 598 int len = nbytes; 599 int off; 600 int error = 0; 601 602 ASSERT3P(vp->v_mount, !=, NULL); 603 obj = vp->v_object; 604 ASSERT3P(obj, !=, NULL); 605 606 start = zfs_uio_offset(uio); 607 off = start & PAGEOFFSET; 608 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 609 vm_page_t pp; 610 uint64_t bytes = MIN(PAGESIZE - off, len); 611 612 if ((pp = page_hold(vp, start))) { 613 struct sf_buf *sf; 614 caddr_t va; 615 616 va = zfs_map_page(pp, &sf); 617 error = vn_io_fault_uiomove(va + off, bytes, 618 GET_UIO_STRUCT(uio)); 619 zfs_unmap_page(sf); 620 page_unhold(pp); 621 } else { 622 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 623 uio, bytes, DMU_READ_PREFETCH); 624 } 625 len -= bytes; 626 off = 0; 627 if (error) 628 break; 629 } 630 return (error); 631 } 632 633 int 634 zfs_write_simple(znode_t *zp, const void *data, size_t len, 635 loff_t pos, size_t *presid) 636 { 637 int error = 0; 638 ssize_t resid; 639 640 error = vn_rdwr(UIO_WRITE, ZTOV(zp), __DECONST(void *, data), len, pos, 641 UIO_SYSSPACE, IO_SYNC, kcred, NOCRED, &resid, curthread); 642 643 if (error) { 644 return (SET_ERROR(error)); 645 } else if (presid == NULL) { 646 if (resid != 0) { 647 error = SET_ERROR(EIO); 648 } 649 } else { 650 *presid = resid; 651 } 652 return (error); 653 } 654 655 void 656 zfs_zrele_async(znode_t *zp) 657 { 658 vnode_t *vp = ZTOV(zp); 659 objset_t *os = ITOZSB(vp)->z_os; 660 661 VN_RELE_ASYNC(vp, dsl_pool_zrele_taskq(dmu_objset_pool(os))); 662 } 663 664 static int 665 zfs_dd_callback(struct mount *mp, void *arg, int lkflags, struct vnode **vpp) 666 { 667 int error; 668 669 *vpp = arg; 670 error = vn_lock(*vpp, lkflags); 671 if (error != 0) 672 vrele(*vpp); 673 return (error); 674 } 675 676 static int 677 zfs_lookup_lock(vnode_t *dvp, vnode_t *vp, const char *name, int lkflags) 678 { 679 znode_t *zdp = VTOZ(dvp); 680 zfsvfs_t *zfsvfs __unused = zdp->z_zfsvfs; 681 int error; 682 int ltype; 683 684 if (zfsvfs->z_replay == B_FALSE) 685 ASSERT_VOP_LOCKED(dvp, __func__); 686 687 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) { 688 ASSERT3P(dvp, ==, vp); 689 vref(dvp); 690 ltype = lkflags & LK_TYPE_MASK; 691 if (ltype != VOP_ISLOCKED(dvp)) { 692 if (ltype == LK_EXCLUSIVE) 693 vn_lock(dvp, LK_UPGRADE | LK_RETRY); 694 else /* if (ltype == LK_SHARED) */ 695 vn_lock(dvp, LK_DOWNGRADE | LK_RETRY); 696 697 /* 698 * Relock for the "." case could leave us with 699 * reclaimed vnode. 700 */ 701 if (VN_IS_DOOMED(dvp)) { 702 vrele(dvp); 703 return (SET_ERROR(ENOENT)); 704 } 705 } 706 return (0); 707 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) { 708 /* 709 * Note that in this case, dvp is the child vnode, and we 710 * are looking up the parent vnode - exactly reverse from 711 * normal operation. Unlocking dvp requires some rather 712 * tricky unlock/relock dance to prevent mp from being freed; 713 * use vn_vget_ino_gen() which takes care of all that. 714 * 715 * XXX Note that there is a time window when both vnodes are 716 * unlocked. It is possible, although highly unlikely, that 717 * during that window the parent-child relationship between 718 * the vnodes may change, for example, get reversed. 719 * In that case we would have a wrong lock order for the vnodes. 720 * All other filesystems seem to ignore this problem, so we 721 * do the same here. 722 * A potential solution could be implemented as follows: 723 * - using LK_NOWAIT when locking the second vnode and retrying 724 * if necessary 725 * - checking that the parent-child relationship still holds 726 * after locking both vnodes and retrying if it doesn't 727 */ 728 error = vn_vget_ino_gen(dvp, zfs_dd_callback, vp, lkflags, &vp); 729 return (error); 730 } else { 731 error = vn_lock(vp, lkflags); 732 if (error != 0) 733 vrele(vp); 734 return (error); 735 } 736 } 737 738 /* 739 * Lookup an entry in a directory, or an extended attribute directory. 740 * If it exists, return a held vnode reference for it. 741 * 742 * IN: dvp - vnode of directory to search. 743 * nm - name of entry to lookup. 744 * pnp - full pathname to lookup [UNUSED]. 745 * flags - LOOKUP_XATTR set if looking for an attribute. 746 * rdir - root directory vnode [UNUSED]. 747 * cr - credentials of caller. 748 * ct - caller context 749 * 750 * OUT: vpp - vnode of located entry, NULL if not found. 751 * 752 * RETURN: 0 on success, error code on failure. 753 * 754 * Timestamps: 755 * NA 756 */ 757 static int 758 zfs_lookup(vnode_t *dvp, const char *nm, vnode_t **vpp, 759 struct componentname *cnp, int nameiop, cred_t *cr, int flags, 760 boolean_t cached) 761 { 762 znode_t *zdp = VTOZ(dvp); 763 znode_t *zp; 764 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 765 seqc_t dvp_seqc; 766 int error = 0; 767 768 /* 769 * Fast path lookup, however we must skip DNLC lookup 770 * for case folding or normalizing lookups because the 771 * DNLC code only stores the passed in name. This means 772 * creating 'a' and removing 'A' on a case insensitive 773 * file system would work, but DNLC still thinks 'a' 774 * exists and won't let you create it again on the next 775 * pass through fast path. 776 */ 777 if (!(flags & LOOKUP_XATTR)) { 778 if (dvp->v_type != VDIR) { 779 return (SET_ERROR(ENOTDIR)); 780 } else if (zdp->z_sa_hdl == NULL) { 781 return (SET_ERROR(EIO)); 782 } 783 } 784 785 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, 786 const char *, nm); 787 788 if ((error = zfs_enter_verify_zp(zfsvfs, zdp, FTAG)) != 0) 789 return (error); 790 791 dvp_seqc = vn_seqc_read_notmodify(dvp); 792 793 *vpp = NULL; 794 795 if (flags & LOOKUP_XATTR) { 796 /* 797 * If the xattr property is off, refuse the lookup request. 798 */ 799 if (!(zfsvfs->z_flags & ZSB_XATTR)) { 800 zfs_exit(zfsvfs, FTAG); 801 return (SET_ERROR(EOPNOTSUPP)); 802 } 803 804 /* 805 * We don't allow recursive attributes.. 806 * Maybe someday we will. 807 */ 808 if (zdp->z_pflags & ZFS_XATTR) { 809 zfs_exit(zfsvfs, FTAG); 810 return (SET_ERROR(EINVAL)); 811 } 812 813 if ((error = zfs_get_xattrdir(VTOZ(dvp), &zp, cr, flags))) { 814 zfs_exit(zfsvfs, FTAG); 815 return (error); 816 } 817 *vpp = ZTOV(zp); 818 819 /* 820 * Do we have permission to get into attribute directory? 821 */ 822 if (flags & LOOKUP_NAMED_ATTR) 823 error = zfs_zaccess(zp, ACE_EXECUTE, V_NAMEDATTR, 824 B_FALSE, cr, NULL); 825 else 826 error = zfs_zaccess(zp, ACE_EXECUTE, 0, B_FALSE, cr, 827 NULL); 828 if (error) { 829 vrele(ZTOV(zp)); 830 } 831 832 zfs_exit(zfsvfs, FTAG); 833 return (error); 834 } 835 836 /* 837 * Check accessibility of directory if we're not coming in via 838 * VOP_CACHEDLOOKUP. 839 */ 840 if (!cached) { 841 #ifdef NOEXECCHECK 842 if ((cnp->cn_flags & NOEXECCHECK) != 0) { 843 cnp->cn_flags &= ~NOEXECCHECK; 844 } else 845 #endif 846 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr, 847 NULL))) { 848 zfs_exit(zfsvfs, FTAG); 849 return (error); 850 } 851 } 852 853 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 854 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 855 zfs_exit(zfsvfs, FTAG); 856 return (SET_ERROR(EILSEQ)); 857 } 858 859 860 /* 861 * First handle the special cases. 862 */ 863 if ((cnp->cn_flags & ISDOTDOT) != 0) { 864 /* 865 * If we are a snapshot mounted under .zfs, return 866 * the vp for the snapshot directory. 867 */ 868 if (zdp->z_id == zfsvfs->z_root && zfsvfs->z_parent != zfsvfs) { 869 struct componentname cn; 870 vnode_t *zfsctl_vp; 871 int ltype; 872 873 zfs_exit(zfsvfs, FTAG); 874 ltype = VOP_ISLOCKED(dvp); 875 VOP_UNLOCK(dvp); 876 error = zfsctl_root(zfsvfs->z_parent, LK_SHARED, 877 &zfsctl_vp); 878 if (error == 0) { 879 cn.cn_nameptr = "snapshot"; 880 cn.cn_namelen = strlen(cn.cn_nameptr); 881 cn.cn_nameiop = cnp->cn_nameiop; 882 cn.cn_flags = cnp->cn_flags & ~ISDOTDOT; 883 cn.cn_lkflags = cnp->cn_lkflags; 884 error = VOP_LOOKUP(zfsctl_vp, vpp, &cn); 885 vput(zfsctl_vp); 886 } 887 vn_lock(dvp, ltype | LK_RETRY); 888 return (error); 889 } 890 } 891 if (zfs_has_ctldir(zdp) && strcmp(nm, ZFS_CTLDIR_NAME) == 0) { 892 zfs_exit(zfsvfs, FTAG); 893 if (zfsvfs->z_show_ctldir == ZFS_SNAPDIR_DISABLED) 894 return (SET_ERROR(ENOENT)); 895 if ((cnp->cn_flags & ISLASTCN) != 0 && nameiop != LOOKUP) 896 return (SET_ERROR(ENOTSUP)); 897 error = zfsctl_root(zfsvfs, cnp->cn_lkflags, vpp); 898 return (error); 899 } 900 901 /* 902 * The loop is retry the lookup if the parent-child relationship 903 * changes during the dot-dot locking complexities. 904 */ 905 for (;;) { 906 uint64_t parent; 907 908 error = zfs_dirlook(zdp, nm, &zp); 909 if (error == 0) 910 *vpp = ZTOV(zp); 911 912 zfs_exit(zfsvfs, FTAG); 913 if (error != 0) 914 break; 915 916 error = zfs_lookup_lock(dvp, *vpp, nm, cnp->cn_lkflags); 917 if (error != 0) { 918 /* 919 * If we've got a locking error, then the vnode 920 * got reclaimed because of a force unmount. 921 * We never enter doomed vnodes into the name cache. 922 */ 923 *vpp = NULL; 924 return (error); 925 } 926 927 if ((cnp->cn_flags & ISDOTDOT) == 0) 928 break; 929 930 if ((error = zfs_enter(zfsvfs, FTAG)) != 0) { 931 vput(ZTOV(zp)); 932 *vpp = NULL; 933 return (error); 934 } 935 if (zdp->z_sa_hdl == NULL) { 936 error = SET_ERROR(EIO); 937 } else { 938 error = sa_lookup(zdp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 939 &parent, sizeof (parent)); 940 } 941 if (error != 0) { 942 zfs_exit(zfsvfs, FTAG); 943 vput(ZTOV(zp)); 944 break; 945 } 946 if (zp->z_id == parent) { 947 zfs_exit(zfsvfs, FTAG); 948 break; 949 } 950 vput(ZTOV(zp)); 951 } 952 953 if (error != 0) 954 *vpp = NULL; 955 956 /* Translate errors and add SAVENAME when needed. */ 957 if (cnp->cn_flags & ISLASTCN) { 958 switch (nameiop) { 959 case CREATE: 960 case RENAME: 961 if (error == ENOENT) { 962 error = EJUSTRETURN; 963 #if __FreeBSD_version < 1400068 964 cnp->cn_flags |= SAVENAME; 965 #endif 966 break; 967 } 968 zfs_fallthrough; 969 case DELETE: 970 #if __FreeBSD_version < 1400068 971 if (error == 0) 972 cnp->cn_flags |= SAVENAME; 973 #endif 974 break; 975 } 976 } 977 978 if ((cnp->cn_flags & ISDOTDOT) != 0) { 979 /* 980 * FIXME: zfs_lookup_lock relocks vnodes and does nothing to 981 * handle races. In particular different callers may end up 982 * with different vnodes and will try to add conflicting 983 * entries to the namecache. 984 * 985 * While finding different result may be acceptable in face 986 * of concurrent modification, adding conflicting entries 987 * trips over an assert in the namecache. 988 * 989 * Ultimately let an entry through once everything settles. 990 */ 991 if (!vn_seqc_consistent(dvp, dvp_seqc)) { 992 cnp->cn_flags &= ~MAKEENTRY; 993 } 994 } 995 996 /* Insert name into cache (as non-existent) if appropriate. */ 997 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay && 998 error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0) 999 cache_enter(dvp, NULL, cnp); 1000 1001 /* Insert name into cache if appropriate. */ 1002 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay && 1003 error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1004 if (!(cnp->cn_flags & ISLASTCN) || 1005 (nameiop != DELETE && nameiop != RENAME)) { 1006 cache_enter(dvp, *vpp, cnp); 1007 } 1008 } 1009 1010 return (error); 1011 } 1012 1013 static inline bool 1014 is_nametoolong(zfsvfs_t *zfsvfs, const char *name) 1015 { 1016 size_t dlen = strlen(name); 1017 return ((!zfsvfs->z_longname && dlen >= ZAP_MAXNAMELEN) || 1018 dlen >= ZAP_MAXNAMELEN_NEW); 1019 } 1020 1021 /* 1022 * Attempt to create a new entry in a directory. If the entry 1023 * already exists, truncate the file if permissible, else return 1024 * an error. Return the vp of the created or trunc'd file. 1025 * 1026 * IN: dvp - vnode of directory to put new file entry in. 1027 * name - name of new file entry. 1028 * vap - attributes of new file. 1029 * excl - flag indicating exclusive or non-exclusive mode. 1030 * mode - mode to open file with. 1031 * cr - credentials of caller. 1032 * flag - large file flag [UNUSED]. 1033 * ct - caller context 1034 * vsecp - ACL to be set 1035 * mnt_ns - Unused on FreeBSD 1036 * 1037 * OUT: vpp - vnode of created or trunc'd entry. 1038 * 1039 * RETURN: 0 on success, error code on failure. 1040 * 1041 * Timestamps: 1042 * dvp - ctime|mtime updated if new entry created 1043 * vp - ctime|mtime always, atime if new 1044 */ 1045 int 1046 zfs_create(znode_t *dzp, const char *name, vattr_t *vap, int excl, int mode, 1047 znode_t **zpp, cred_t *cr, int flag, vsecattr_t *vsecp, zidmap_t *mnt_ns) 1048 { 1049 (void) excl, (void) mode, (void) flag; 1050 znode_t *zp; 1051 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1052 zilog_t *zilog; 1053 objset_t *os; 1054 dmu_tx_t *tx; 1055 int error; 1056 uid_t uid = crgetuid(cr); 1057 gid_t gid = crgetgid(cr); 1058 uint64_t projid = ZFS_DEFAULT_PROJID; 1059 zfs_acl_ids_t acl_ids; 1060 boolean_t fuid_dirtied; 1061 uint64_t txtype; 1062 #ifdef DEBUG_VFS_LOCKS 1063 vnode_t *dvp = ZTOV(dzp); 1064 #endif 1065 1066 if (is_nametoolong(zfsvfs, name)) 1067 return (SET_ERROR(ENAMETOOLONG)); 1068 1069 /* 1070 * If we have an ephemeral id, ACL, or XVATTR then 1071 * make sure file system is at proper version 1072 */ 1073 if (zfsvfs->z_use_fuids == B_FALSE && 1074 (vsecp || (vap->va_mask & AT_XVATTR) || 1075 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1076 return (SET_ERROR(EINVAL)); 1077 1078 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 1079 return (error); 1080 os = zfsvfs->z_os; 1081 zilog = zfsvfs->z_log; 1082 1083 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1084 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1085 zfs_exit(zfsvfs, FTAG); 1086 return (SET_ERROR(EILSEQ)); 1087 } 1088 1089 if (vap->va_mask & AT_XVATTR) { 1090 if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap, 1091 crgetuid(cr), cr, vap->va_type)) != 0) { 1092 zfs_exit(zfsvfs, FTAG); 1093 return (error); 1094 } 1095 } 1096 1097 *zpp = NULL; 1098 1099 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) 1100 vap->va_mode &= ~S_ISVTX; 1101 1102 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW); 1103 if (error) { 1104 zfs_exit(zfsvfs, FTAG); 1105 return (error); 1106 } 1107 ASSERT0P(zp); 1108 1109 /* 1110 * Create a new file object and update the directory 1111 * to reference it. 1112 */ 1113 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) { 1114 goto out; 1115 } 1116 1117 /* 1118 * We only support the creation of regular files in 1119 * extended attribute directories. 1120 */ 1121 1122 if ((dzp->z_pflags & ZFS_XATTR) && 1123 (vap->va_type != VREG)) { 1124 error = SET_ERROR(EINVAL); 1125 goto out; 1126 } 1127 1128 if ((error = zfs_acl_ids_create(dzp, 0, vap, 1129 cr, vsecp, &acl_ids, NULL)) != 0) 1130 goto out; 1131 1132 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode)) 1133 projid = zfs_inherit_projid(dzp); 1134 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) { 1135 zfs_acl_ids_free(&acl_ids); 1136 error = SET_ERROR(EDQUOT); 1137 goto out; 1138 } 1139 1140 getnewvnode_reserve(); 1141 1142 tx = dmu_tx_create(os); 1143 1144 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1145 ZFS_SA_BASE_ATTR_SIZE); 1146 1147 fuid_dirtied = zfsvfs->z_fuid_dirty; 1148 if (fuid_dirtied) 1149 zfs_fuid_txhold(zfsvfs, tx); 1150 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1151 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1152 if (!zfsvfs->z_use_sa && 1153 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1154 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1155 0, acl_ids.z_aclp->z_acl_bytes); 1156 } 1157 error = dmu_tx_assign(tx, DMU_TX_WAIT); 1158 if (error) { 1159 zfs_acl_ids_free(&acl_ids); 1160 dmu_tx_abort(tx); 1161 getnewvnode_drop_reserve(); 1162 zfs_exit(zfsvfs, FTAG); 1163 return (error); 1164 } 1165 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1166 1167 error = zfs_link_create(dzp, name, zp, tx, ZNEW); 1168 if (error != 0) { 1169 /* 1170 * Since, we failed to add the directory entry for it, 1171 * delete the newly created dnode. 1172 */ 1173 zfs_znode_delete(zp, tx); 1174 VOP_UNLOCK(ZTOV(zp)); 1175 zrele(zp); 1176 zfs_acl_ids_free(&acl_ids); 1177 dmu_tx_commit(tx); 1178 getnewvnode_drop_reserve(); 1179 goto out; 1180 } 1181 1182 if (fuid_dirtied) 1183 zfs_fuid_sync(zfsvfs, tx); 1184 1185 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1186 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1187 vsecp, acl_ids.z_fuidp, vap); 1188 zfs_acl_ids_free(&acl_ids); 1189 dmu_tx_commit(tx); 1190 1191 getnewvnode_drop_reserve(); 1192 1193 out: 1194 VNCHECKREF(dvp); 1195 if (error == 0) { 1196 *zpp = zp; 1197 } 1198 1199 if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1200 error = zil_commit(zilog, 0); 1201 1202 zfs_exit(zfsvfs, FTAG); 1203 return (error); 1204 } 1205 1206 /* 1207 * Remove an entry from a directory. 1208 * 1209 * IN: dvp - vnode of directory to remove entry from. 1210 * name - name of entry to remove. 1211 * cr - credentials of caller. 1212 * ct - caller context 1213 * flags - case flags 1214 * 1215 * RETURN: 0 on success, error code on failure. 1216 * 1217 * Timestamps: 1218 * dvp - ctime|mtime 1219 * vp - ctime (if nlink > 0) 1220 */ 1221 static int 1222 zfs_remove_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr) 1223 { 1224 znode_t *dzp = VTOZ(dvp); 1225 znode_t *zp; 1226 znode_t *xzp; 1227 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1228 zilog_t *zilog; 1229 uint64_t xattr_obj; 1230 uint64_t obj = 0; 1231 dmu_tx_t *tx; 1232 boolean_t unlinked; 1233 uint64_t txtype; 1234 int error; 1235 1236 1237 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 1238 return (error); 1239 zp = VTOZ(vp); 1240 if ((error = zfs_verify_zp(zp)) != 0) { 1241 zfs_exit(zfsvfs, FTAG); 1242 return (error); 1243 } 1244 zilog = zfsvfs->z_log; 1245 1246 xattr_obj = 0; 1247 xzp = NULL; 1248 1249 if ((error = zfs_zaccess_delete(dzp, zp, cr, NULL))) { 1250 goto out; 1251 } 1252 1253 /* 1254 * Need to use rmdir for removing directories. 1255 */ 1256 if (vp->v_type == VDIR) { 1257 error = SET_ERROR(EPERM); 1258 goto out; 1259 } 1260 1261 vnevent_remove(vp, dvp, name, ct); 1262 1263 obj = zp->z_id; 1264 1265 /* are there any extended attributes? */ 1266 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1267 &xattr_obj, sizeof (xattr_obj)); 1268 if (error == 0 && xattr_obj) { 1269 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1270 ASSERT0(error); 1271 } 1272 1273 /* 1274 * We may delete the znode now, or we may put it in the unlinked set; 1275 * it depends on whether we're the last link, and on whether there are 1276 * other holds on the vnode. So we dmu_tx_hold() the right things to 1277 * allow for either case. 1278 */ 1279 tx = dmu_tx_create(zfsvfs->z_os); 1280 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1281 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1282 zfs_sa_upgrade_txholds(tx, zp); 1283 zfs_sa_upgrade_txholds(tx, dzp); 1284 1285 if (xzp) { 1286 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1287 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1288 } 1289 1290 /* charge as an update -- would be nice not to charge at all */ 1291 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1292 1293 /* 1294 * Mark this transaction as typically resulting in a net free of space 1295 */ 1296 dmu_tx_mark_netfree(tx); 1297 1298 error = dmu_tx_assign(tx, DMU_TX_WAIT); 1299 if (error) { 1300 dmu_tx_abort(tx); 1301 zfs_exit(zfsvfs, FTAG); 1302 return (error); 1303 } 1304 1305 /* 1306 * Remove the directory entry. 1307 */ 1308 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, &unlinked); 1309 1310 if (error) { 1311 dmu_tx_commit(tx); 1312 goto out; 1313 } 1314 1315 if (unlinked) { 1316 zfs_unlinked_add(zp, tx); 1317 vp->v_vflag |= VV_NOSYNC; 1318 } 1319 /* XXX check changes to linux vnops */ 1320 txtype = TX_REMOVE; 1321 zfs_log_remove(zilog, tx, txtype, dzp, name, obj, unlinked); 1322 1323 dmu_tx_commit(tx); 1324 out: 1325 1326 if (xzp) 1327 vrele(ZTOV(xzp)); 1328 1329 if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1330 error = zil_commit(zilog, 0); 1331 1332 zfs_exit(zfsvfs, FTAG); 1333 return (error); 1334 } 1335 1336 1337 static int 1338 zfs_lookup_internal(znode_t *dzp, const char *name, vnode_t **vpp, 1339 struct componentname *cnp, int nameiop) 1340 { 1341 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1342 int error; 1343 1344 cnp->cn_nameptr = __DECONST(char *, name); 1345 cnp->cn_namelen = strlen(name); 1346 cnp->cn_nameiop = nameiop; 1347 cnp->cn_flags = ISLASTCN; 1348 #if __FreeBSD_version < 1400068 1349 cnp->cn_flags |= SAVENAME; 1350 #endif 1351 cnp->cn_lkflags = LK_EXCLUSIVE | LK_RETRY; 1352 cnp->cn_cred = kcred; 1353 #if __FreeBSD_version < 1400037 1354 cnp->cn_thread = curthread; 1355 #endif 1356 1357 if (zfsvfs->z_use_namecache && !zfsvfs->z_replay) { 1358 struct vop_lookup_args a; 1359 1360 a.a_gen.a_desc = &vop_lookup_desc; 1361 a.a_dvp = ZTOV(dzp); 1362 a.a_vpp = vpp; 1363 a.a_cnp = cnp; 1364 error = vfs_cache_lookup(&a); 1365 } else { 1366 error = zfs_lookup(ZTOV(dzp), name, vpp, cnp, nameiop, kcred, 0, 1367 B_FALSE); 1368 } 1369 #ifdef ZFS_DEBUG 1370 if (error) { 1371 printf("got error %d on name %s on op %d\n", error, name, 1372 nameiop); 1373 kdb_backtrace(); 1374 } 1375 #endif 1376 return (error); 1377 } 1378 1379 int 1380 zfs_remove(znode_t *dzp, const char *name, cred_t *cr, int flags) 1381 { 1382 vnode_t *vp; 1383 int error; 1384 struct componentname cn; 1385 1386 if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE))) 1387 return (error); 1388 1389 error = zfs_remove_(ZTOV(dzp), vp, name, cr); 1390 vput(vp); 1391 return (error); 1392 } 1393 /* 1394 * Create a new directory and insert it into dvp using the name 1395 * provided. Return a pointer to the inserted directory. 1396 * 1397 * IN: dvp - vnode of directory to add subdir to. 1398 * dirname - name of new directory. 1399 * vap - attributes of new directory. 1400 * cr - credentials of caller. 1401 * ct - caller context 1402 * flags - case flags 1403 * vsecp - ACL to be set 1404 * mnt_ns - Unused on FreeBSD 1405 * 1406 * OUT: vpp - vnode of created directory. 1407 * 1408 * RETURN: 0 on success, error code on failure. 1409 * 1410 * Timestamps: 1411 * dvp - ctime|mtime updated 1412 * vp - ctime|mtime|atime updated 1413 */ 1414 int 1415 zfs_mkdir(znode_t *dzp, const char *dirname, vattr_t *vap, znode_t **zpp, 1416 cred_t *cr, int flags, vsecattr_t *vsecp, zidmap_t *mnt_ns) 1417 { 1418 (void) flags, (void) vsecp; 1419 znode_t *zp; 1420 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1421 zilog_t *zilog; 1422 uint64_t txtype; 1423 dmu_tx_t *tx; 1424 int error; 1425 uid_t uid = crgetuid(cr); 1426 gid_t gid = crgetgid(cr); 1427 zfs_acl_ids_t acl_ids; 1428 boolean_t fuid_dirtied; 1429 1430 ASSERT3U(vap->va_type, ==, VDIR); 1431 1432 if (is_nametoolong(zfsvfs, dirname)) 1433 return (SET_ERROR(ENAMETOOLONG)); 1434 1435 /* 1436 * If we have an ephemeral id, ACL, or XVATTR then 1437 * make sure file system is at proper version 1438 */ 1439 if (zfsvfs->z_use_fuids == B_FALSE && 1440 ((vap->va_mask & AT_XVATTR) || 1441 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1442 return (SET_ERROR(EINVAL)); 1443 1444 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 1445 return (error); 1446 zilog = zfsvfs->z_log; 1447 1448 if (dzp->z_pflags & ZFS_XATTR) { 1449 zfs_exit(zfsvfs, FTAG); 1450 return (SET_ERROR(EINVAL)); 1451 } 1452 1453 if (zfsvfs->z_utf8 && u8_validate(dirname, 1454 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1455 zfs_exit(zfsvfs, FTAG); 1456 return (SET_ERROR(EILSEQ)); 1457 } 1458 1459 if (vap->va_mask & AT_XVATTR) { 1460 if ((error = secpolicy_xvattr(ZTOV(dzp), (xvattr_t *)vap, 1461 crgetuid(cr), cr, vap->va_type)) != 0) { 1462 zfs_exit(zfsvfs, FTAG); 1463 return (error); 1464 } 1465 } 1466 1467 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 1468 NULL, &acl_ids, NULL)) != 0) { 1469 zfs_exit(zfsvfs, FTAG); 1470 return (error); 1471 } 1472 1473 /* 1474 * First make sure the new directory doesn't exist. 1475 * 1476 * Existence is checked first to make sure we don't return 1477 * EACCES instead of EEXIST which can cause some applications 1478 * to fail. 1479 */ 1480 *zpp = NULL; 1481 1482 if ((error = zfs_dirent_lookup(dzp, dirname, &zp, ZNEW))) { 1483 zfs_acl_ids_free(&acl_ids); 1484 zfs_exit(zfsvfs, FTAG); 1485 return (error); 1486 } 1487 ASSERT0P(zp); 1488 1489 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr, 1490 mnt_ns))) { 1491 zfs_acl_ids_free(&acl_ids); 1492 zfs_exit(zfsvfs, FTAG); 1493 return (error); 1494 } 1495 1496 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zfs_inherit_projid(dzp))) { 1497 zfs_acl_ids_free(&acl_ids); 1498 zfs_exit(zfsvfs, FTAG); 1499 return (SET_ERROR(EDQUOT)); 1500 } 1501 1502 /* 1503 * Add a new entry to the directory. 1504 */ 1505 getnewvnode_reserve(); 1506 tx = dmu_tx_create(zfsvfs->z_os); 1507 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1508 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1509 fuid_dirtied = zfsvfs->z_fuid_dirty; 1510 if (fuid_dirtied) 1511 zfs_fuid_txhold(zfsvfs, tx); 1512 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1513 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1514 acl_ids.z_aclp->z_acl_bytes); 1515 } 1516 1517 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1518 ZFS_SA_BASE_ATTR_SIZE); 1519 1520 error = dmu_tx_assign(tx, DMU_TX_WAIT); 1521 if (error) { 1522 zfs_acl_ids_free(&acl_ids); 1523 dmu_tx_abort(tx); 1524 getnewvnode_drop_reserve(); 1525 zfs_exit(zfsvfs, FTAG); 1526 return (error); 1527 } 1528 1529 /* 1530 * Create new node. 1531 */ 1532 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1533 1534 /* 1535 * Now put new name in parent dir. 1536 */ 1537 error = zfs_link_create(dzp, dirname, zp, tx, ZNEW); 1538 if (error != 0) { 1539 zfs_znode_delete(zp, tx); 1540 VOP_UNLOCK(ZTOV(zp)); 1541 zrele(zp); 1542 goto out; 1543 } 1544 1545 if (fuid_dirtied) 1546 zfs_fuid_sync(zfsvfs, tx); 1547 1548 *zpp = zp; 1549 1550 txtype = zfs_log_create_txtype(Z_DIR, NULL, vap); 1551 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, NULL, 1552 acl_ids.z_fuidp, vap); 1553 1554 out: 1555 zfs_acl_ids_free(&acl_ids); 1556 1557 dmu_tx_commit(tx); 1558 1559 getnewvnode_drop_reserve(); 1560 1561 if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1562 error = zil_commit(zilog, 0); 1563 1564 zfs_exit(zfsvfs, FTAG); 1565 return (error); 1566 } 1567 1568 /* 1569 * Remove a directory subdir entry. If the current working 1570 * directory is the same as the subdir to be removed, the 1571 * remove will fail. 1572 * 1573 * IN: dvp - vnode of directory to remove from. 1574 * name - name of directory to be removed. 1575 * cwd - vnode of current working directory. 1576 * cr - credentials of caller. 1577 * ct - caller context 1578 * flags - case flags 1579 * 1580 * RETURN: 0 on success, error code on failure. 1581 * 1582 * Timestamps: 1583 * dvp - ctime|mtime updated 1584 */ 1585 static int 1586 zfs_rmdir_(vnode_t *dvp, vnode_t *vp, const char *name, cred_t *cr) 1587 { 1588 znode_t *dzp = VTOZ(dvp); 1589 znode_t *zp = VTOZ(vp); 1590 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1591 zilog_t *zilog; 1592 dmu_tx_t *tx; 1593 int error; 1594 1595 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 1596 return (error); 1597 if ((error = zfs_verify_zp(zp)) != 0) { 1598 zfs_exit(zfsvfs, FTAG); 1599 return (error); 1600 } 1601 zilog = zfsvfs->z_log; 1602 1603 1604 if ((error = zfs_zaccess_delete(dzp, zp, cr, NULL))) { 1605 goto out; 1606 } 1607 1608 if (vp->v_type != VDIR) { 1609 error = SET_ERROR(ENOTDIR); 1610 goto out; 1611 } 1612 1613 vnevent_rmdir(vp, dvp, name, ct); 1614 1615 tx = dmu_tx_create(zfsvfs->z_os); 1616 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1617 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1618 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1619 zfs_sa_upgrade_txholds(tx, zp); 1620 zfs_sa_upgrade_txholds(tx, dzp); 1621 dmu_tx_mark_netfree(tx); 1622 error = dmu_tx_assign(tx, DMU_TX_WAIT); 1623 if (error) { 1624 dmu_tx_abort(tx); 1625 zfs_exit(zfsvfs, FTAG); 1626 return (error); 1627 } 1628 1629 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, NULL); 1630 1631 if (error == 0) { 1632 uint64_t txtype = TX_RMDIR; 1633 zfs_log_remove(zilog, tx, txtype, dzp, name, 1634 ZFS_NO_OBJECT, B_FALSE); 1635 } 1636 1637 dmu_tx_commit(tx); 1638 1639 if (zfsvfs->z_use_namecache) 1640 cache_vop_rmdir(dvp, vp); 1641 out: 1642 if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1643 error = zil_commit(zilog, 0); 1644 1645 zfs_exit(zfsvfs, FTAG); 1646 return (error); 1647 } 1648 1649 int 1650 zfs_rmdir(znode_t *dzp, const char *name, znode_t *cwd, cred_t *cr, int flags) 1651 { 1652 struct componentname cn; 1653 vnode_t *vp; 1654 int error; 1655 1656 if ((error = zfs_lookup_internal(dzp, name, &vp, &cn, DELETE))) 1657 return (error); 1658 1659 error = zfs_rmdir_(ZTOV(dzp), vp, name, cr); 1660 vput(vp); 1661 return (error); 1662 } 1663 1664 /* 1665 * Read as many directory entries as will fit into the provided 1666 * buffer from the given directory cursor position (specified in 1667 * the uio structure). 1668 * 1669 * IN: vp - vnode of directory to read. 1670 * uio - structure supplying read location, range info, 1671 * and return buffer. 1672 * cr - credentials of caller. 1673 * ct - caller context 1674 * 1675 * OUT: uio - updated offset and range, buffer filled. 1676 * eofp - set to true if end-of-file detected. 1677 * ncookies- number of entries in cookies 1678 * cookies - offsets to directory entries 1679 * 1680 * RETURN: 0 on success, error code on failure. 1681 * 1682 * Timestamps: 1683 * vp - atime updated 1684 * 1685 * Note that the low 4 bits of the cookie returned by zap is always zero. 1686 * This allows us to use the low range for "special" directory entries: 1687 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1688 * we use the offset 2 for the '.zfs' directory. 1689 */ 1690 static int 1691 zfs_readdir(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, int *eofp, 1692 int *ncookies, cookie_t **cookies) 1693 { 1694 znode_t *zp = VTOZ(vp); 1695 iovec_t *iovp; 1696 dirent64_t *odp; 1697 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1698 objset_t *os; 1699 caddr_t outbuf; 1700 size_t bufsize; 1701 zap_cursor_t zc; 1702 zap_attribute_t *zap; 1703 uint_t bytes_wanted; 1704 uint64_t offset; /* must be unsigned; checks for < 1 */ 1705 uint64_t parent; 1706 int local_eof; 1707 int outcount; 1708 int error; 1709 uint8_t prefetch; 1710 uint8_t type; 1711 int ncooks; 1712 cookie_t *cooks = NULL; 1713 1714 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 1715 return (error); 1716 1717 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 1718 &parent, sizeof (parent))) != 0) { 1719 zfs_exit(zfsvfs, FTAG); 1720 return (error); 1721 } 1722 1723 /* 1724 * If we are not given an eof variable, 1725 * use a local one. 1726 */ 1727 if (eofp == NULL) 1728 eofp = &local_eof; 1729 1730 /* 1731 * Check for valid iov_len. 1732 */ 1733 if (GET_UIO_STRUCT(uio)->uio_iov->iov_len <= 0) { 1734 zfs_exit(zfsvfs, FTAG); 1735 return (SET_ERROR(EINVAL)); 1736 } 1737 1738 /* 1739 * Quit if directory has been removed (posix) 1740 */ 1741 if ((*eofp = (zp->z_unlinked != 0)) != 0) { 1742 zfs_exit(zfsvfs, FTAG); 1743 return (0); 1744 } 1745 1746 error = 0; 1747 os = zfsvfs->z_os; 1748 offset = zfs_uio_offset(uio); 1749 prefetch = zp->z_zn_prefetch; 1750 zap = zap_attribute_long_alloc(); 1751 1752 /* 1753 * Initialize the iterator cursor. 1754 */ 1755 if (offset <= 3) { 1756 /* 1757 * Start iteration from the beginning of the directory. 1758 */ 1759 zap_cursor_init(&zc, os, zp->z_id); 1760 } else { 1761 /* 1762 * The offset is a serialized cursor. 1763 */ 1764 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1765 } 1766 1767 /* 1768 * Get space to change directory entries into fs independent format. 1769 */ 1770 iovp = GET_UIO_STRUCT(uio)->uio_iov; 1771 bytes_wanted = iovp->iov_len; 1772 if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1) { 1773 bufsize = bytes_wanted; 1774 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1775 odp = (struct dirent64 *)outbuf; 1776 } else { 1777 bufsize = bytes_wanted; 1778 outbuf = NULL; 1779 odp = (struct dirent64 *)iovp->iov_base; 1780 } 1781 1782 if (ncookies != NULL) { 1783 /* 1784 * Minimum entry size is dirent size and 1 byte for a file name. 1785 */ 1786 ncooks = zfs_uio_resid(uio) / (sizeof (struct dirent) - 1787 sizeof (((struct dirent *)NULL)->d_name) + 1); 1788 cooks = malloc(ncooks * sizeof (*cooks), M_TEMP, M_WAITOK); 1789 *cookies = cooks; 1790 *ncookies = ncooks; 1791 } 1792 1793 /* 1794 * Transform to file-system independent format 1795 */ 1796 outcount = 0; 1797 while (outcount < bytes_wanted) { 1798 ino64_t objnum; 1799 ushort_t reclen; 1800 off64_t *next = NULL; 1801 1802 /* 1803 * Special case `.', `..', and `.zfs'. 1804 */ 1805 if (offset == 0) { 1806 (void) strcpy(zap->za_name, "."); 1807 zap->za_normalization_conflict = 0; 1808 objnum = zp->z_id; 1809 type = DT_DIR; 1810 } else if (offset == 1) { 1811 (void) strcpy(zap->za_name, ".."); 1812 zap->za_normalization_conflict = 0; 1813 objnum = parent; 1814 type = DT_DIR; 1815 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1816 (void) strcpy(zap->za_name, ZFS_CTLDIR_NAME); 1817 zap->za_normalization_conflict = 0; 1818 objnum = ZFSCTL_INO_ROOT; 1819 type = DT_DIR; 1820 } else { 1821 /* 1822 * Grab next entry. 1823 */ 1824 if ((error = zap_cursor_retrieve(&zc, zap))) { 1825 if ((*eofp = (error == ENOENT)) != 0) 1826 break; 1827 else 1828 goto update; 1829 } 1830 1831 if (zap->za_integer_length != 8 || 1832 zap->za_num_integers != 1) { 1833 cmn_err(CE_WARN, "zap_readdir: bad directory " 1834 "entry, obj = %lld, offset = %lld\n", 1835 (u_longlong_t)zp->z_id, 1836 (u_longlong_t)offset); 1837 error = SET_ERROR(ENXIO); 1838 goto update; 1839 } 1840 1841 objnum = ZFS_DIRENT_OBJ(zap->za_first_integer); 1842 /* 1843 * MacOS X can extract the object type here such as: 1844 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1845 */ 1846 type = ZFS_DIRENT_TYPE(zap->za_first_integer); 1847 } 1848 1849 reclen = DIRENT64_RECLEN(strlen(zap->za_name)); 1850 1851 /* 1852 * Will this entry fit in the buffer? 1853 */ 1854 if (outcount + reclen > bufsize) { 1855 /* 1856 * Did we manage to fit anything in the buffer? 1857 */ 1858 if (!outcount) { 1859 error = SET_ERROR(EINVAL); 1860 goto update; 1861 } 1862 break; 1863 } 1864 /* 1865 * Add normal entry: 1866 */ 1867 odp->d_ino = objnum; 1868 odp->d_reclen = reclen; 1869 odp->d_namlen = strlen(zap->za_name); 1870 /* NOTE: d_off is the offset for the *next* entry. */ 1871 next = &odp->d_off; 1872 strlcpy(odp->d_name, zap->za_name, odp->d_namlen + 1); 1873 odp->d_type = type; 1874 dirent_terminate(odp); 1875 odp = (dirent64_t *)((intptr_t)odp + reclen); 1876 1877 outcount += reclen; 1878 1879 ASSERT3S(outcount, <=, bufsize); 1880 1881 if (prefetch) 1882 dmu_prefetch_dnode(os, objnum, ZIO_PRIORITY_SYNC_READ); 1883 1884 /* 1885 * Move to the next entry, fill in the previous offset. 1886 */ 1887 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1888 zap_cursor_advance(&zc); 1889 offset = zap_cursor_serialize(&zc); 1890 } else { 1891 offset += 1; 1892 } 1893 1894 /* Fill the offset right after advancing the cursor. */ 1895 if (next != NULL) 1896 *next = offset; 1897 if (cooks != NULL) { 1898 *cooks++ = offset; 1899 ncooks--; 1900 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 1901 } 1902 } 1903 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1904 1905 /* Subtract unused cookies */ 1906 if (ncookies != NULL) 1907 *ncookies -= ncooks; 1908 1909 if (zfs_uio_segflg(uio) == UIO_SYSSPACE && zfs_uio_iovcnt(uio) == 1) { 1910 iovp->iov_base += outcount; 1911 iovp->iov_len -= outcount; 1912 zfs_uio_resid(uio) -= outcount; 1913 } else if ((error = 1914 zfs_uiomove(outbuf, (long)outcount, UIO_READ, uio))) { 1915 /* 1916 * Reset the pointer. 1917 */ 1918 offset = zfs_uio_offset(uio); 1919 } 1920 1921 update: 1922 zap_cursor_fini(&zc); 1923 zap_attribute_free(zap); 1924 if (zfs_uio_segflg(uio) != UIO_SYSSPACE || zfs_uio_iovcnt(uio) != 1) 1925 kmem_free(outbuf, bufsize); 1926 1927 if (error == ENOENT) 1928 error = 0; 1929 1930 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1931 1932 zfs_uio_setoffset(uio, offset); 1933 zfs_exit(zfsvfs, FTAG); 1934 if (error != 0 && cookies != NULL) { 1935 free(*cookies, M_TEMP); 1936 *cookies = NULL; 1937 *ncookies = 0; 1938 } 1939 return (error); 1940 } 1941 1942 /* 1943 * Get the requested file attributes and place them in the provided 1944 * vattr structure. 1945 * 1946 * IN: vp - vnode of file. 1947 * vap - va_mask identifies requested attributes. 1948 * If AT_XVATTR set, then optional attrs are requested 1949 * flags - ATTR_NOACLCHECK (CIFS server context) 1950 * cr - credentials of caller. 1951 * 1952 * OUT: vap - attribute values. 1953 * 1954 * RETURN: 0 (always succeeds). 1955 */ 1956 static int 1957 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1958 { 1959 znode_t *zp = VTOZ(vp); 1960 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1961 int error = 0; 1962 uint32_t blksize; 1963 u_longlong_t nblocks; 1964 uint64_t mtime[2], ctime[2], crtime[2], rdev; 1965 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 1966 xoptattr_t *xoap = NULL; 1967 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 1968 sa_bulk_attr_t bulk[4]; 1969 int count = 0; 1970 1971 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 1972 return (error); 1973 1974 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 1975 1976 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 1977 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 1978 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16); 1979 if (vp->v_type == VBLK || vp->v_type == VCHR) 1980 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, 1981 &rdev, 8); 1982 1983 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 1984 zfs_exit(zfsvfs, FTAG); 1985 return (error); 1986 } 1987 1988 /* 1989 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 1990 * Also, if we are the owner don't bother, since owner should 1991 * always be allowed to read basic attributes of file. 1992 */ 1993 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 1994 (vap->va_uid != crgetuid(cr))) { 1995 if ((error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 1996 skipaclchk, cr, NULL))) { 1997 zfs_exit(zfsvfs, FTAG); 1998 return (error); 1999 } 2000 } 2001 2002 /* 2003 * Return all attributes. It's cheaper to provide the answer 2004 * than to determine whether we were asked the question. 2005 */ 2006 2007 vap->va_type = IFTOVT(zp->z_mode); 2008 vap->va_mode = zp->z_mode & ~S_IFMT; 2009 vn_fsid(vp, vap); 2010 vap->va_nodeid = zp->z_id; 2011 vap->va_nlink = zp->z_links; 2012 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp) && 2013 zp->z_links < ZFS_LINK_MAX) 2014 vap->va_nlink++; 2015 vap->va_size = zp->z_size; 2016 if (vp->v_type == VBLK || vp->v_type == VCHR) 2017 vap->va_rdev = zfs_cmpldev(rdev); 2018 else 2019 vap->va_rdev = NODEV; 2020 vap->va_gen = zp->z_gen; 2021 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2022 vap->va_filerev = zp->z_seq; 2023 2024 /* 2025 * Add in any requested optional attributes and the create time. 2026 * Also set the corresponding bits in the returned attribute bitmap. 2027 */ 2028 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2029 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2030 xoap->xoa_archive = 2031 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2032 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2033 } 2034 2035 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2036 xoap->xoa_readonly = 2037 ((zp->z_pflags & ZFS_READONLY) != 0); 2038 XVA_SET_RTN(xvap, XAT_READONLY); 2039 } 2040 2041 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2042 xoap->xoa_system = 2043 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2044 XVA_SET_RTN(xvap, XAT_SYSTEM); 2045 } 2046 2047 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2048 xoap->xoa_hidden = 2049 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2050 XVA_SET_RTN(xvap, XAT_HIDDEN); 2051 } 2052 2053 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2054 xoap->xoa_nounlink = 2055 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2056 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2057 } 2058 2059 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2060 xoap->xoa_immutable = 2061 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2062 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2063 } 2064 2065 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2066 xoap->xoa_appendonly = 2067 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2068 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2069 } 2070 2071 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2072 xoap->xoa_nodump = 2073 ((zp->z_pflags & ZFS_NODUMP) != 0); 2074 XVA_SET_RTN(xvap, XAT_NODUMP); 2075 } 2076 2077 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2078 xoap->xoa_opaque = 2079 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2080 XVA_SET_RTN(xvap, XAT_OPAQUE); 2081 } 2082 2083 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2084 xoap->xoa_av_quarantined = 2085 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2086 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2087 } 2088 2089 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2090 xoap->xoa_av_modified = 2091 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2092 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2093 } 2094 2095 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2096 vp->v_type == VREG) { 2097 zfs_sa_get_scanstamp(zp, xvap); 2098 } 2099 2100 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2101 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2102 XVA_SET_RTN(xvap, XAT_REPARSE); 2103 } 2104 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2105 xoap->xoa_generation = zp->z_gen; 2106 XVA_SET_RTN(xvap, XAT_GEN); 2107 } 2108 2109 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2110 xoap->xoa_offline = 2111 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2112 XVA_SET_RTN(xvap, XAT_OFFLINE); 2113 } 2114 2115 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2116 xoap->xoa_sparse = 2117 ((zp->z_pflags & ZFS_SPARSE) != 0); 2118 XVA_SET_RTN(xvap, XAT_SPARSE); 2119 } 2120 2121 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) { 2122 xoap->xoa_projinherit = 2123 ((zp->z_pflags & ZFS_PROJINHERIT) != 0); 2124 XVA_SET_RTN(xvap, XAT_PROJINHERIT); 2125 } 2126 2127 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) { 2128 xoap->xoa_projid = zp->z_projid; 2129 XVA_SET_RTN(xvap, XAT_PROJID); 2130 } 2131 } 2132 2133 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2134 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2135 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2136 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 2137 2138 2139 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 2140 vap->va_blksize = blksize; 2141 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2142 2143 if (zp->z_blksz == 0) { 2144 /* 2145 * Block size hasn't been set; suggest maximal I/O transfers. 2146 */ 2147 vap->va_blksize = zfsvfs->z_max_blksz; 2148 } 2149 2150 zfs_exit(zfsvfs, FTAG); 2151 return (0); 2152 } 2153 2154 /* 2155 * For the operation of changing file's user/group/project, we need to 2156 * handle not only the main object that is assigned to the file directly, 2157 * but also the ones that are used by the file via hidden xattr directory. 2158 * 2159 * Because the xattr directory may contains many EA entries, as to it may 2160 * be impossible to change all of them via the transaction of changing the 2161 * main object's user/group/project attributes. Then we have to change them 2162 * via other multiple independent transactions one by one. It may be not good 2163 * solution, but we have no better idea yet. 2164 */ 2165 static int 2166 zfs_setattr_dir(znode_t *dzp) 2167 { 2168 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2169 objset_t *os = zfsvfs->z_os; 2170 zap_cursor_t zc; 2171 zap_attribute_t *zap; 2172 znode_t *zp = NULL; 2173 dmu_tx_t *tx = NULL; 2174 uint64_t uid, gid; 2175 sa_bulk_attr_t bulk[4]; 2176 int count; 2177 int err; 2178 2179 zap = zap_attribute_alloc(); 2180 zap_cursor_init(&zc, os, dzp->z_id); 2181 while ((err = zap_cursor_retrieve(&zc, zap)) == 0) { 2182 count = 0; 2183 if (zap->za_integer_length != 8 || zap->za_num_integers != 1) { 2184 err = ENXIO; 2185 break; 2186 } 2187 2188 err = zfs_dirent_lookup(dzp, zap->za_name, &zp, ZEXISTS); 2189 if (err == ENOENT) 2190 goto next; 2191 if (err) 2192 break; 2193 2194 if (zp->z_uid == dzp->z_uid && 2195 zp->z_gid == dzp->z_gid && 2196 zp->z_projid == dzp->z_projid) 2197 goto next; 2198 2199 tx = dmu_tx_create(os); 2200 if (!(zp->z_pflags & ZFS_PROJID)) 2201 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2202 else 2203 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2204 2205 err = dmu_tx_assign(tx, DMU_TX_WAIT); 2206 if (err) 2207 break; 2208 2209 vn_seqc_write_begin(ZTOV(zp)); 2210 mutex_enter(&dzp->z_lock); 2211 2212 if (zp->z_uid != dzp->z_uid) { 2213 uid = dzp->z_uid; 2214 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 2215 &uid, sizeof (uid)); 2216 zp->z_uid = uid; 2217 } 2218 2219 if (zp->z_gid != dzp->z_gid) { 2220 gid = dzp->z_gid; 2221 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 2222 &gid, sizeof (gid)); 2223 zp->z_gid = gid; 2224 } 2225 2226 uint64_t projid = dzp->z_projid; 2227 if (zp->z_projid != projid) { 2228 if (!(zp->z_pflags & ZFS_PROJID)) { 2229 err = sa_add_projid(zp->z_sa_hdl, tx, projid); 2230 if (unlikely(err == EEXIST)) { 2231 err = 0; 2232 } else if (err != 0) { 2233 goto sa_add_projid_err; 2234 } else { 2235 projid = ZFS_INVALID_PROJID; 2236 } 2237 } 2238 2239 if (projid != ZFS_INVALID_PROJID) { 2240 zp->z_projid = projid; 2241 SA_ADD_BULK_ATTR(bulk, count, 2242 SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid, 2243 sizeof (zp->z_projid)); 2244 } 2245 } 2246 2247 sa_add_projid_err: 2248 mutex_exit(&dzp->z_lock); 2249 2250 if (likely(count > 0)) { 2251 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 2252 dmu_tx_commit(tx); 2253 } else if (projid == ZFS_INVALID_PROJID) { 2254 dmu_tx_commit(tx); 2255 } else { 2256 dmu_tx_abort(tx); 2257 } 2258 tx = NULL; 2259 vn_seqc_write_end(ZTOV(zp)); 2260 if (err != 0 && err != ENOENT) 2261 break; 2262 2263 next: 2264 if (zp) { 2265 zrele(zp); 2266 zp = NULL; 2267 } 2268 zap_cursor_advance(&zc); 2269 } 2270 2271 if (tx) 2272 dmu_tx_abort(tx); 2273 if (zp) { 2274 zrele(zp); 2275 } 2276 zap_cursor_fini(&zc); 2277 zap_attribute_free(zap); 2278 2279 return (err == ENOENT ? 0 : err); 2280 } 2281 2282 /* 2283 * Set the file attributes to the values contained in the 2284 * vattr structure. 2285 * 2286 * IN: zp - znode of file to be modified. 2287 * vap - new attribute values. 2288 * If AT_XVATTR set, then optional attrs are being set 2289 * flags - ATTR_UTIME set if non-default time values provided. 2290 * - ATTR_NOACLCHECK (CIFS context only). 2291 * cr - credentials of caller. 2292 * mnt_ns - Unused on FreeBSD 2293 * 2294 * RETURN: 0 on success, error code on failure. 2295 * 2296 * Timestamps: 2297 * vp - ctime updated, mtime updated if size changed. 2298 */ 2299 int 2300 zfs_setattr(znode_t *zp, vattr_t *vap, int flags, cred_t *cr, zidmap_t *mnt_ns) 2301 { 2302 vnode_t *vp = ZTOV(zp); 2303 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2304 objset_t *os; 2305 zilog_t *zilog; 2306 dmu_tx_t *tx; 2307 vattr_t oldva; 2308 xvattr_t tmpxvattr; 2309 uint_t mask = vap->va_mask; 2310 uint_t saved_mask = 0; 2311 uint64_t saved_mode; 2312 int trim_mask = 0; 2313 uint64_t new_mode; 2314 uint64_t new_uid, new_gid; 2315 uint64_t xattr_obj; 2316 uint64_t mtime[2], ctime[2]; 2317 uint64_t projid = ZFS_INVALID_PROJID; 2318 znode_t *attrzp; 2319 int need_policy = FALSE; 2320 int err, err2; 2321 zfs_fuid_info_t *fuidp = NULL; 2322 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2323 xoptattr_t *xoap; 2324 zfs_acl_t *aclp; 2325 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2326 boolean_t fuid_dirtied = B_FALSE; 2327 boolean_t handle_eadir = B_FALSE; 2328 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 2329 int count = 0, xattr_count = 0; 2330 2331 if (mask == 0) 2332 return (0); 2333 2334 if (mask & AT_NOSET) 2335 return (SET_ERROR(EINVAL)); 2336 2337 if ((err = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 2338 return (err); 2339 2340 os = zfsvfs->z_os; 2341 zilog = zfsvfs->z_log; 2342 2343 /* 2344 * Make sure that if we have ephemeral uid/gid or xvattr specified 2345 * that file system is at proper version level 2346 */ 2347 2348 if (zfsvfs->z_use_fuids == B_FALSE && 2349 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2350 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2351 (mask & AT_XVATTR))) { 2352 zfs_exit(zfsvfs, FTAG); 2353 return (SET_ERROR(EINVAL)); 2354 } 2355 2356 if (mask & AT_SIZE && vp->v_type == VDIR) { 2357 zfs_exit(zfsvfs, FTAG); 2358 return (SET_ERROR(EISDIR)); 2359 } 2360 2361 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 2362 zfs_exit(zfsvfs, FTAG); 2363 return (SET_ERROR(EINVAL)); 2364 } 2365 2366 /* 2367 * If this is an xvattr_t, then get a pointer to the structure of 2368 * optional attributes. If this is NULL, then we have a vattr_t. 2369 */ 2370 xoap = xva_getxoptattr(xvap); 2371 2372 xva_init(&tmpxvattr); 2373 2374 /* 2375 * Immutable files can only alter immutable bit and atime 2376 */ 2377 if ((zp->z_pflags & ZFS_IMMUTABLE) && 2378 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2379 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 2380 zfs_exit(zfsvfs, FTAG); 2381 return (SET_ERROR(EPERM)); 2382 } 2383 2384 /* 2385 * Note: ZFS_READONLY is handled in zfs_zaccess_common. 2386 */ 2387 2388 /* 2389 * Verify timestamps doesn't overflow 32 bits. 2390 * ZFS can handle large timestamps, but 32bit syscalls can't 2391 * handle times greater than 2039. This check should be removed 2392 * once large timestamps are fully supported. 2393 */ 2394 if (mask & (AT_ATIME | AT_MTIME)) { 2395 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 2396 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 2397 zfs_exit(zfsvfs, FTAG); 2398 return (SET_ERROR(EOVERFLOW)); 2399 } 2400 } 2401 if (xoap != NULL && (mask & AT_XVATTR)) { 2402 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME) && 2403 TIMESPEC_OVERFLOW(&vap->va_birthtime)) { 2404 zfs_exit(zfsvfs, FTAG); 2405 return (SET_ERROR(EOVERFLOW)); 2406 } 2407 2408 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) { 2409 if (!dmu_objset_projectquota_enabled(os) || 2410 (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode))) { 2411 zfs_exit(zfsvfs, FTAG); 2412 return (SET_ERROR(EOPNOTSUPP)); 2413 } 2414 2415 projid = xoap->xoa_projid; 2416 if (unlikely(projid == ZFS_INVALID_PROJID)) { 2417 zfs_exit(zfsvfs, FTAG); 2418 return (SET_ERROR(EINVAL)); 2419 } 2420 2421 if (projid == zp->z_projid && zp->z_pflags & ZFS_PROJID) 2422 projid = ZFS_INVALID_PROJID; 2423 else 2424 need_policy = TRUE; 2425 } 2426 2427 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT) && 2428 (xoap->xoa_projinherit != 2429 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) && 2430 (!dmu_objset_projectquota_enabled(os) || 2431 (!S_ISREG(zp->z_mode) && !S_ISDIR(zp->z_mode)))) { 2432 zfs_exit(zfsvfs, FTAG); 2433 return (SET_ERROR(EOPNOTSUPP)); 2434 } 2435 } 2436 2437 attrzp = NULL; 2438 aclp = NULL; 2439 2440 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2441 zfs_exit(zfsvfs, FTAG); 2442 return (SET_ERROR(EROFS)); 2443 } 2444 2445 /* 2446 * First validate permissions 2447 */ 2448 2449 if (mask & AT_SIZE) { 2450 /* 2451 * XXX - Note, we are not providing any open 2452 * mode flags here (like FNDELAY), so we may 2453 * block if there are locks present... this 2454 * should be addressed in openat(). 2455 */ 2456 /* XXX - would it be OK to generate a log record here? */ 2457 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2458 if (err) { 2459 zfs_exit(zfsvfs, FTAG); 2460 return (err); 2461 } 2462 } 2463 2464 if (mask & (AT_ATIME|AT_MTIME) || 2465 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2466 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2467 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2468 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 2469 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 2470 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2471 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 2472 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2473 skipaclchk, cr, mnt_ns); 2474 } 2475 2476 if (mask & (AT_UID|AT_GID)) { 2477 int idmask = (mask & (AT_UID|AT_GID)); 2478 int take_owner; 2479 int take_group; 2480 2481 /* 2482 * NOTE: even if a new mode is being set, 2483 * we may clear S_ISUID/S_ISGID bits. 2484 */ 2485 2486 if (!(mask & AT_MODE)) 2487 vap->va_mode = zp->z_mode; 2488 2489 /* 2490 * Take ownership or chgrp to group we are a member of 2491 */ 2492 2493 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2494 take_group = (mask & AT_GID) && 2495 zfs_groupmember(zfsvfs, vap->va_gid, cr); 2496 2497 /* 2498 * If both AT_UID and AT_GID are set then take_owner and 2499 * take_group must both be set in order to allow taking 2500 * ownership. 2501 * 2502 * Otherwise, send the check through secpolicy_vnode_setattr() 2503 * 2504 */ 2505 2506 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2507 ((idmask == AT_UID) && take_owner) || 2508 ((idmask == AT_GID) && take_group)) { 2509 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 2510 skipaclchk, cr, mnt_ns) == 0) { 2511 /* 2512 * Remove setuid/setgid for non-privileged users 2513 */ 2514 secpolicy_setid_clear(vap, vp, cr); 2515 trim_mask = (mask & (AT_UID|AT_GID)); 2516 } else { 2517 need_policy = TRUE; 2518 } 2519 } else { 2520 need_policy = TRUE; 2521 } 2522 } 2523 2524 oldva.va_mode = zp->z_mode; 2525 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 2526 if (mask & AT_XVATTR) { 2527 /* 2528 * Update xvattr mask to include only those attributes 2529 * that are actually changing. 2530 * 2531 * the bits will be restored prior to actually setting 2532 * the attributes so the caller thinks they were set. 2533 */ 2534 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2535 if (xoap->xoa_appendonly != 2536 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 2537 need_policy = TRUE; 2538 } else { 2539 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 2540 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 2541 } 2542 } 2543 2544 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) { 2545 if (xoap->xoa_projinherit != 2546 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) { 2547 need_policy = TRUE; 2548 } else { 2549 XVA_CLR_REQ(xvap, XAT_PROJINHERIT); 2550 XVA_SET_REQ(&tmpxvattr, XAT_PROJINHERIT); 2551 } 2552 } 2553 2554 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2555 if (xoap->xoa_nounlink != 2556 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 2557 need_policy = TRUE; 2558 } else { 2559 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 2560 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 2561 } 2562 } 2563 2564 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2565 if (xoap->xoa_immutable != 2566 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 2567 need_policy = TRUE; 2568 } else { 2569 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 2570 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 2571 } 2572 } 2573 2574 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2575 if (xoap->xoa_nodump != 2576 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 2577 need_policy = TRUE; 2578 } else { 2579 XVA_CLR_REQ(xvap, XAT_NODUMP); 2580 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 2581 } 2582 } 2583 2584 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2585 if (xoap->xoa_av_modified != 2586 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 2587 need_policy = TRUE; 2588 } else { 2589 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 2590 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 2591 } 2592 } 2593 2594 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2595 if ((vp->v_type != VREG && 2596 xoap->xoa_av_quarantined) || 2597 xoap->xoa_av_quarantined != 2598 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 2599 need_policy = TRUE; 2600 } else { 2601 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 2602 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 2603 } 2604 } 2605 2606 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2607 zfs_exit(zfsvfs, FTAG); 2608 return (SET_ERROR(EPERM)); 2609 } 2610 2611 if (need_policy == FALSE && 2612 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 2613 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 2614 need_policy = TRUE; 2615 } 2616 } 2617 2618 if (mask & AT_MODE) { 2619 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr, 2620 mnt_ns) == 0) { 2621 err = secpolicy_setid_setsticky_clear(vp, vap, 2622 &oldva, cr); 2623 if (err) { 2624 zfs_exit(zfsvfs, FTAG); 2625 return (err); 2626 } 2627 trim_mask |= AT_MODE; 2628 } else { 2629 need_policy = TRUE; 2630 } 2631 } 2632 2633 if (need_policy) { 2634 /* 2635 * If trim_mask is set then take ownership 2636 * has been granted or write_acl is present and user 2637 * has the ability to modify mode. In that case remove 2638 * UID|GID and or MODE from mask so that 2639 * secpolicy_vnode_setattr() doesn't revoke it. 2640 */ 2641 2642 if (trim_mask) { 2643 saved_mask = vap->va_mask; 2644 vap->va_mask &= ~trim_mask; 2645 if (trim_mask & AT_MODE) { 2646 /* 2647 * Save the mode, as secpolicy_vnode_setattr() 2648 * will overwrite it with ova.va_mode. 2649 */ 2650 saved_mode = vap->va_mode; 2651 } 2652 } 2653 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2654 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 2655 if (err) { 2656 zfs_exit(zfsvfs, FTAG); 2657 return (err); 2658 } 2659 2660 if (trim_mask) { 2661 vap->va_mask |= saved_mask; 2662 if (trim_mask & AT_MODE) { 2663 /* 2664 * Recover the mode after 2665 * secpolicy_vnode_setattr(). 2666 */ 2667 vap->va_mode = saved_mode; 2668 } 2669 } 2670 } 2671 2672 /* 2673 * secpolicy_vnode_setattr, or take ownership may have 2674 * changed va_mask 2675 */ 2676 mask = vap->va_mask; 2677 2678 if ((mask & (AT_UID | AT_GID)) || projid != ZFS_INVALID_PROJID) { 2679 handle_eadir = B_TRUE; 2680 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 2681 &xattr_obj, sizeof (xattr_obj)); 2682 2683 if (err == 0 && xattr_obj) { 2684 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 2685 if (err == 0) { 2686 err = vn_lock(ZTOV(attrzp), LK_EXCLUSIVE); 2687 if (err != 0) 2688 vrele(ZTOV(attrzp)); 2689 } 2690 if (err) 2691 goto out2; 2692 } 2693 if (mask & AT_UID) { 2694 new_uid = zfs_fuid_create(zfsvfs, 2695 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 2696 if (new_uid != zp->z_uid && 2697 zfs_id_overquota(zfsvfs, DMU_USERUSED_OBJECT, 2698 new_uid)) { 2699 if (attrzp) 2700 vput(ZTOV(attrzp)); 2701 err = SET_ERROR(EDQUOT); 2702 goto out2; 2703 } 2704 } 2705 2706 if (mask & AT_GID) { 2707 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 2708 cr, ZFS_GROUP, &fuidp); 2709 if (new_gid != zp->z_gid && 2710 zfs_id_overquota(zfsvfs, DMU_GROUPUSED_OBJECT, 2711 new_gid)) { 2712 if (attrzp) 2713 vput(ZTOV(attrzp)); 2714 err = SET_ERROR(EDQUOT); 2715 goto out2; 2716 } 2717 } 2718 2719 if (projid != ZFS_INVALID_PROJID && 2720 zfs_id_overquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid)) { 2721 if (attrzp) 2722 vput(ZTOV(attrzp)); 2723 err = SET_ERROR(EDQUOT); 2724 goto out2; 2725 } 2726 } 2727 tx = dmu_tx_create(os); 2728 2729 if (mask & AT_MODE) { 2730 uint64_t pmode = zp->z_mode; 2731 uint64_t acl_obj; 2732 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2733 2734 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && 2735 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 2736 err = SET_ERROR(EPERM); 2737 goto out; 2738 } 2739 2740 if ((err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))) 2741 goto out; 2742 2743 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 2744 /* 2745 * Are we upgrading ACL from old V0 format 2746 * to V1 format? 2747 */ 2748 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 2749 zfs_znode_acl_version(zp) == 2750 ZFS_ACL_VERSION_INITIAL) { 2751 dmu_tx_hold_free(tx, acl_obj, 0, 2752 DMU_OBJECT_END); 2753 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2754 0, aclp->z_acl_bytes); 2755 } else { 2756 dmu_tx_hold_write(tx, acl_obj, 0, 2757 aclp->z_acl_bytes); 2758 } 2759 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2760 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2761 0, aclp->z_acl_bytes); 2762 } 2763 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2764 } else { 2765 if (((mask & AT_XVATTR) && 2766 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) || 2767 (projid != ZFS_INVALID_PROJID && 2768 !(zp->z_pflags & ZFS_PROJID))) 2769 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2770 else 2771 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2772 } 2773 2774 if (attrzp) { 2775 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 2776 } 2777 2778 fuid_dirtied = zfsvfs->z_fuid_dirty; 2779 if (fuid_dirtied) 2780 zfs_fuid_txhold(zfsvfs, tx); 2781 2782 zfs_sa_upgrade_txholds(tx, zp); 2783 2784 err = dmu_tx_assign(tx, DMU_TX_WAIT); 2785 if (err) 2786 goto out; 2787 2788 count = 0; 2789 /* 2790 * Set each attribute requested. 2791 * We group settings according to the locks they need to acquire. 2792 * 2793 * Note: you cannot set ctime directly, although it will be 2794 * updated as a side-effect of calling this function. 2795 */ 2796 2797 if (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID)) { 2798 /* 2799 * For the existed object that is upgraded from old system, 2800 * its on-disk layout has no slot for the project ID attribute. 2801 * But quota accounting logic needs to access related slots by 2802 * offset directly. So we need to adjust old objects' layout 2803 * to make the project ID to some unified and fixed offset. 2804 */ 2805 if (attrzp) 2806 err = sa_add_projid(attrzp->z_sa_hdl, tx, projid); 2807 if (err == 0) 2808 err = sa_add_projid(zp->z_sa_hdl, tx, projid); 2809 2810 if (unlikely(err == EEXIST)) 2811 err = 0; 2812 else if (err != 0) 2813 goto out; 2814 else 2815 projid = ZFS_INVALID_PROJID; 2816 } 2817 2818 if (mask & (AT_UID|AT_GID|AT_MODE)) 2819 mutex_enter(&zp->z_acl_lock); 2820 2821 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 2822 &zp->z_pflags, sizeof (zp->z_pflags)); 2823 2824 if (attrzp) { 2825 if (mask & (AT_UID|AT_GID|AT_MODE)) 2826 mutex_enter(&attrzp->z_acl_lock); 2827 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2828 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 2829 sizeof (attrzp->z_pflags)); 2830 if (projid != ZFS_INVALID_PROJID) { 2831 attrzp->z_projid = projid; 2832 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2833 SA_ZPL_PROJID(zfsvfs), NULL, &attrzp->z_projid, 2834 sizeof (attrzp->z_projid)); 2835 } 2836 } 2837 2838 if (mask & (AT_UID|AT_GID)) { 2839 2840 if (mask & AT_UID) { 2841 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 2842 &new_uid, sizeof (new_uid)); 2843 zp->z_uid = new_uid; 2844 if (attrzp) { 2845 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2846 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 2847 sizeof (new_uid)); 2848 attrzp->z_uid = new_uid; 2849 } 2850 } 2851 2852 if (mask & AT_GID) { 2853 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 2854 NULL, &new_gid, sizeof (new_gid)); 2855 zp->z_gid = new_gid; 2856 if (attrzp) { 2857 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2858 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 2859 sizeof (new_gid)); 2860 attrzp->z_gid = new_gid; 2861 } 2862 } 2863 if (!(mask & AT_MODE)) { 2864 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 2865 NULL, &new_mode, sizeof (new_mode)); 2866 new_mode = zp->z_mode; 2867 } 2868 err = zfs_acl_chown_setattr(zp); 2869 ASSERT0(err); 2870 if (attrzp) { 2871 vn_seqc_write_begin(ZTOV(attrzp)); 2872 err = zfs_acl_chown_setattr(attrzp); 2873 vn_seqc_write_end(ZTOV(attrzp)); 2874 ASSERT0(err); 2875 } 2876 } 2877 2878 if (mask & AT_MODE) { 2879 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 2880 &new_mode, sizeof (new_mode)); 2881 zp->z_mode = new_mode; 2882 ASSERT3P(aclp, !=, NULL); 2883 err = zfs_aclset_common(zp, aclp, cr, tx); 2884 ASSERT0(err); 2885 if (zp->z_acl_cached) 2886 zfs_acl_free(zp->z_acl_cached); 2887 zp->z_acl_cached = aclp; 2888 aclp = NULL; 2889 } 2890 2891 2892 if (mask & AT_ATIME) { 2893 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 2894 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 2895 &zp->z_atime, sizeof (zp->z_atime)); 2896 } 2897 2898 if (mask & AT_MTIME) { 2899 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 2900 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 2901 mtime, sizeof (mtime)); 2902 } 2903 2904 if (projid != ZFS_INVALID_PROJID) { 2905 zp->z_projid = projid; 2906 SA_ADD_BULK_ATTR(bulk, count, 2907 SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid, 2908 sizeof (zp->z_projid)); 2909 } 2910 2911 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 2912 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 2913 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 2914 NULL, mtime, sizeof (mtime)); 2915 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 2916 &ctime, sizeof (ctime)); 2917 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime); 2918 } else if (mask != 0) { 2919 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 2920 &ctime, sizeof (ctime)); 2921 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime); 2922 if (attrzp) { 2923 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2924 SA_ZPL_CTIME(zfsvfs), NULL, 2925 &ctime, sizeof (ctime)); 2926 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 2927 mtime, ctime); 2928 } 2929 } 2930 2931 /* 2932 * Do this after setting timestamps to prevent timestamp 2933 * update from toggling bit 2934 */ 2935 2936 if (xoap && (mask & AT_XVATTR)) { 2937 2938 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) 2939 xoap->xoa_createtime = vap->va_birthtime; 2940 /* 2941 * restore trimmed off masks 2942 * so that return masks can be set for caller. 2943 */ 2944 2945 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 2946 XVA_SET_REQ(xvap, XAT_APPENDONLY); 2947 } 2948 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 2949 XVA_SET_REQ(xvap, XAT_NOUNLINK); 2950 } 2951 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 2952 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 2953 } 2954 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 2955 XVA_SET_REQ(xvap, XAT_NODUMP); 2956 } 2957 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 2958 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 2959 } 2960 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 2961 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 2962 } 2963 if (XVA_ISSET_REQ(&tmpxvattr, XAT_PROJINHERIT)) { 2964 XVA_SET_REQ(xvap, XAT_PROJINHERIT); 2965 } 2966 2967 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 2968 ASSERT3S(vp->v_type, ==, VREG); 2969 2970 zfs_xvattr_set(zp, xvap, tx); 2971 } 2972 2973 if (fuid_dirtied) 2974 zfs_fuid_sync(zfsvfs, tx); 2975 2976 if (mask != 0) 2977 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 2978 2979 if (mask & (AT_UID|AT_GID|AT_MODE)) 2980 mutex_exit(&zp->z_acl_lock); 2981 2982 if (attrzp) { 2983 if (mask & (AT_UID|AT_GID|AT_MODE)) 2984 mutex_exit(&attrzp->z_acl_lock); 2985 } 2986 out: 2987 if (err == 0 && attrzp) { 2988 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 2989 xattr_count, tx); 2990 ASSERT0(err2); 2991 } 2992 2993 if (attrzp) 2994 vput(ZTOV(attrzp)); 2995 2996 if (aclp) 2997 zfs_acl_free(aclp); 2998 2999 if (fuidp) { 3000 zfs_fuid_info_free(fuidp); 3001 fuidp = NULL; 3002 } 3003 3004 if (err) { 3005 dmu_tx_abort(tx); 3006 } else { 3007 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3008 dmu_tx_commit(tx); 3009 if (attrzp) { 3010 if (err2 == 0 && handle_eadir) 3011 err = zfs_setattr_dir(attrzp); 3012 } 3013 } 3014 3015 out2: 3016 if (err == 0 && os->os_sync == ZFS_SYNC_ALWAYS) 3017 err = zil_commit(zilog, 0); 3018 3019 zfs_exit(zfsvfs, FTAG); 3020 return (err); 3021 } 3022 3023 /* 3024 * Look up the directory entries corresponding to the source and target 3025 * directory/name pairs. 3026 */ 3027 static int 3028 zfs_rename_relock_lookup(znode_t *sdzp, const struct componentname *scnp, 3029 znode_t **szpp, znode_t *tdzp, const struct componentname *tcnp, 3030 znode_t **tzpp) 3031 { 3032 zfsvfs_t *zfsvfs; 3033 znode_t *szp, *tzp; 3034 int error; 3035 3036 /* 3037 * Before using sdzp and tdzp we must ensure that they are live. 3038 * As a porting legacy from illumos we have two things to worry 3039 * about. One is typical for FreeBSD and it is that the vnode is 3040 * not reclaimed (doomed). The other is that the znode is live. 3041 * The current code can invalidate the znode without acquiring the 3042 * corresponding vnode lock if the object represented by the znode 3043 * and vnode is no longer valid after a rollback or receive operation. 3044 * z_teardown_lock hidden behind zfs_enter and zfs_exit is the lock 3045 * that protects the znodes from the invalidation. 3046 */ 3047 zfsvfs = sdzp->z_zfsvfs; 3048 ASSERT3P(zfsvfs, ==, tdzp->z_zfsvfs); 3049 if ((error = zfs_enter_verify_zp(zfsvfs, sdzp, FTAG)) != 0) 3050 return (error); 3051 if ((error = zfs_verify_zp(tdzp)) != 0) { 3052 zfs_exit(zfsvfs, FTAG); 3053 return (error); 3054 } 3055 3056 /* 3057 * Re-resolve svp to be certain it still exists and fetch the 3058 * correct vnode. 3059 */ 3060 error = zfs_dirent_lookup(sdzp, scnp->cn_nameptr, &szp, ZEXISTS); 3061 if (error != 0) { 3062 /* Source entry invalid or not there. */ 3063 if ((scnp->cn_flags & ISDOTDOT) != 0 || 3064 (scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.')) 3065 error = SET_ERROR(EINVAL); 3066 goto out; 3067 } 3068 *szpp = szp; 3069 3070 /* 3071 * Re-resolve tvp, if it disappeared we just carry on. 3072 */ 3073 error = zfs_dirent_lookup(tdzp, tcnp->cn_nameptr, &tzp, 0); 3074 if (error != 0) { 3075 vrele(ZTOV(szp)); 3076 if ((tcnp->cn_flags & ISDOTDOT) != 0) 3077 error = SET_ERROR(EINVAL); 3078 goto out; 3079 } 3080 *tzpp = tzp; 3081 out: 3082 zfs_exit(zfsvfs, FTAG); 3083 return (error); 3084 } 3085 3086 /* 3087 * We acquire all but fdvp locks using non-blocking acquisitions. If we 3088 * fail to acquire any lock in the path we will drop all held locks, 3089 * acquire the new lock in a blocking fashion, and then release it and 3090 * restart the rename. This acquire/release step ensures that we do not 3091 * spin on a lock waiting for release. On error release all vnode locks 3092 * and decrement references the way tmpfs_rename() would do. 3093 */ 3094 static int 3095 zfs_rename_relock(struct vnode *sdvp, struct vnode **svpp, 3096 struct vnode *tdvp, struct vnode **tvpp, 3097 const struct componentname *scnp, const struct componentname *tcnp) 3098 { 3099 struct vnode *nvp, *svp, *tvp; 3100 znode_t *sdzp, *tdzp, *szp, *tzp; 3101 int error; 3102 3103 VOP_UNLOCK(tdvp); 3104 if (*tvpp != NULL && *tvpp != tdvp) 3105 VOP_UNLOCK(*tvpp); 3106 3107 relock: 3108 error = vn_lock(sdvp, LK_EXCLUSIVE); 3109 if (error) 3110 goto out; 3111 error = vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT); 3112 if (error != 0) { 3113 VOP_UNLOCK(sdvp); 3114 if (error != EBUSY) 3115 goto out; 3116 error = vn_lock(tdvp, LK_EXCLUSIVE); 3117 if (error) 3118 goto out; 3119 VOP_UNLOCK(tdvp); 3120 goto relock; 3121 } 3122 tdzp = VTOZ(tdvp); 3123 sdzp = VTOZ(sdvp); 3124 3125 error = zfs_rename_relock_lookup(sdzp, scnp, &szp, tdzp, tcnp, &tzp); 3126 if (error != 0) { 3127 VOP_UNLOCK(sdvp); 3128 VOP_UNLOCK(tdvp); 3129 goto out; 3130 } 3131 svp = ZTOV(szp); 3132 tvp = tzp != NULL ? ZTOV(tzp) : NULL; 3133 3134 /* 3135 * Now try acquire locks on svp and tvp. 3136 */ 3137 nvp = svp; 3138 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT); 3139 if (error != 0) { 3140 VOP_UNLOCK(sdvp); 3141 VOP_UNLOCK(tdvp); 3142 if (tvp != NULL) 3143 vrele(tvp); 3144 if (error != EBUSY) { 3145 vrele(nvp); 3146 goto out; 3147 } 3148 error = vn_lock(nvp, LK_EXCLUSIVE); 3149 if (error != 0) { 3150 vrele(nvp); 3151 goto out; 3152 } 3153 VOP_UNLOCK(nvp); 3154 /* 3155 * Concurrent rename race. 3156 * XXX ? 3157 */ 3158 if (nvp == tdvp) { 3159 vrele(nvp); 3160 error = SET_ERROR(EINVAL); 3161 goto out; 3162 } 3163 vrele(*svpp); 3164 *svpp = nvp; 3165 goto relock; 3166 } 3167 vrele(*svpp); 3168 *svpp = nvp; 3169 3170 if (*tvpp != NULL) 3171 vrele(*tvpp); 3172 *tvpp = NULL; 3173 if (tvp != NULL) { 3174 nvp = tvp; 3175 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT); 3176 if (error != 0) { 3177 VOP_UNLOCK(sdvp); 3178 VOP_UNLOCK(tdvp); 3179 VOP_UNLOCK(*svpp); 3180 if (error != EBUSY) { 3181 vrele(nvp); 3182 goto out; 3183 } 3184 error = vn_lock(nvp, LK_EXCLUSIVE); 3185 if (error != 0) { 3186 vrele(nvp); 3187 goto out; 3188 } 3189 vput(nvp); 3190 goto relock; 3191 } 3192 *tvpp = nvp; 3193 } 3194 3195 return (0); 3196 3197 out: 3198 return (error); 3199 } 3200 3201 /* 3202 * Note that we must use VRELE_ASYNC in this function as it walks 3203 * up the directory tree and vrele may need to acquire an exclusive 3204 * lock if a last reference to a vnode is dropped. 3205 */ 3206 static int 3207 zfs_rename_check(znode_t *szp, znode_t *sdzp, znode_t *tdzp) 3208 { 3209 zfsvfs_t *zfsvfs; 3210 znode_t *zp, *zp1; 3211 uint64_t parent; 3212 int error; 3213 3214 zfsvfs = tdzp->z_zfsvfs; 3215 if (tdzp == szp) 3216 return (SET_ERROR(EINVAL)); 3217 if (tdzp == sdzp) 3218 return (0); 3219 if (tdzp->z_id == zfsvfs->z_root) 3220 return (0); 3221 zp = tdzp; 3222 for (;;) { 3223 ASSERT(!zp->z_unlinked); 3224 if ((error = sa_lookup(zp->z_sa_hdl, 3225 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) 3226 break; 3227 3228 if (parent == szp->z_id) { 3229 error = SET_ERROR(EINVAL); 3230 break; 3231 } 3232 if (parent == zfsvfs->z_root) 3233 break; 3234 if (parent == sdzp->z_id) 3235 break; 3236 3237 error = zfs_zget(zfsvfs, parent, &zp1); 3238 if (error != 0) 3239 break; 3240 3241 if (zp != tdzp) 3242 VN_RELE_ASYNC(ZTOV(zp), 3243 dsl_pool_zrele_taskq( 3244 dmu_objset_pool(zfsvfs->z_os))); 3245 zp = zp1; 3246 } 3247 3248 if (error == ENOTDIR) 3249 panic("checkpath: .. not a directory\n"); 3250 if (zp != tdzp) 3251 VN_RELE_ASYNC(ZTOV(zp), 3252 dsl_pool_zrele_taskq(dmu_objset_pool(zfsvfs->z_os))); 3253 return (error); 3254 } 3255 3256 static int 3257 zfs_do_rename_impl(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, 3258 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, 3259 cred_t *cr); 3260 3261 /* 3262 * Move an entry from the provided source directory to the target 3263 * directory. Change the entry name as indicated. 3264 * 3265 * IN: sdvp - Source directory containing the "old entry". 3266 * scnp - Old entry name. 3267 * tdvp - Target directory to contain the "new entry". 3268 * tcnp - New entry name. 3269 * cr - credentials of caller. 3270 * INOUT: svpp - Source file 3271 * tvpp - Target file, may point to NULL initially 3272 * 3273 * RETURN: 0 on success, error code on failure. 3274 * 3275 * Timestamps: 3276 * sdvp,tdvp - ctime|mtime updated 3277 */ 3278 static int 3279 zfs_do_rename(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, 3280 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, 3281 cred_t *cr) 3282 { 3283 int error; 3284 3285 ASSERT_VOP_ELOCKED(tdvp, __func__); 3286 if (*tvpp != NULL) 3287 ASSERT_VOP_ELOCKED(*tvpp, __func__); 3288 3289 /* Reject renames across filesystems. */ 3290 if ((*svpp)->v_mount != tdvp->v_mount || 3291 ((*tvpp) != NULL && (*svpp)->v_mount != (*tvpp)->v_mount)) { 3292 error = SET_ERROR(EXDEV); 3293 goto out; 3294 } 3295 3296 if (zfsctl_is_node(tdvp)) { 3297 error = SET_ERROR(EXDEV); 3298 goto out; 3299 } 3300 3301 /* 3302 * Lock all four vnodes to ensure safety and semantics of renaming. 3303 */ 3304 error = zfs_rename_relock(sdvp, svpp, tdvp, tvpp, scnp, tcnp); 3305 if (error != 0) { 3306 /* no vnodes are locked in the case of error here */ 3307 return (error); 3308 } 3309 3310 error = zfs_do_rename_impl(sdvp, svpp, scnp, tdvp, tvpp, tcnp, cr); 3311 VOP_UNLOCK(sdvp); 3312 VOP_UNLOCK(*svpp); 3313 out: 3314 if (*tvpp != NULL) 3315 VOP_UNLOCK(*tvpp); 3316 if (tdvp != *tvpp) 3317 VOP_UNLOCK(tdvp); 3318 3319 return (error); 3320 } 3321 3322 static int 3323 zfs_do_rename_impl(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, 3324 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, 3325 cred_t *cr) 3326 { 3327 dmu_tx_t *tx; 3328 zfsvfs_t *zfsvfs; 3329 zilog_t *zilog; 3330 znode_t *tdzp, *sdzp, *tzp, *szp; 3331 const char *snm = scnp->cn_nameptr; 3332 const char *tnm = tcnp->cn_nameptr; 3333 int error; 3334 3335 tdzp = VTOZ(tdvp); 3336 sdzp = VTOZ(sdvp); 3337 zfsvfs = tdzp->z_zfsvfs; 3338 3339 if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0) 3340 return (error); 3341 if ((error = zfs_verify_zp(sdzp)) != 0) { 3342 zfs_exit(zfsvfs, FTAG); 3343 return (error); 3344 } 3345 zilog = zfsvfs->z_log; 3346 3347 if (zfsvfs->z_utf8 && u8_validate(tnm, 3348 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3349 error = SET_ERROR(EILSEQ); 3350 goto out; 3351 } 3352 3353 /* If source and target are the same file, there is nothing to do. */ 3354 if ((*svpp) == (*tvpp)) { 3355 error = 0; 3356 goto out; 3357 } 3358 3359 if (((*svpp)->v_type == VDIR && (*svpp)->v_mountedhere != NULL) || 3360 ((*tvpp) != NULL && (*tvpp)->v_type == VDIR && 3361 (*tvpp)->v_mountedhere != NULL)) { 3362 error = SET_ERROR(EXDEV); 3363 goto out; 3364 } 3365 3366 szp = VTOZ(*svpp); 3367 if ((error = zfs_verify_zp(szp)) != 0) { 3368 zfs_exit(zfsvfs, FTAG); 3369 return (error); 3370 } 3371 tzp = *tvpp == NULL ? NULL : VTOZ(*tvpp); 3372 if (tzp != NULL) { 3373 if ((error = zfs_verify_zp(tzp)) != 0) { 3374 zfs_exit(zfsvfs, FTAG); 3375 return (error); 3376 } 3377 } 3378 3379 /* 3380 * This is to prevent the creation of links into attribute space 3381 * by renaming a linked file into/outof an attribute directory. 3382 * See the comment in zfs_link() for why this is considered bad. 3383 */ 3384 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3385 error = SET_ERROR(EINVAL); 3386 goto out; 3387 } 3388 3389 /* 3390 * If we are using project inheritance, means if the directory has 3391 * ZFS_PROJINHERIT set, then its descendant directories will inherit 3392 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under 3393 * such case, we only allow renames into our tree when the project 3394 * IDs are the same. 3395 */ 3396 if (tdzp->z_pflags & ZFS_PROJINHERIT && 3397 tdzp->z_projid != szp->z_projid) { 3398 error = SET_ERROR(EXDEV); 3399 goto out; 3400 } 3401 3402 /* 3403 * Must have write access at the source to remove the old entry 3404 * and write access at the target to create the new entry. 3405 * Note that if target and source are the same, this can be 3406 * done in a single check. 3407 */ 3408 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr, NULL))) 3409 goto out; 3410 3411 if ((*svpp)->v_type == VDIR) { 3412 /* 3413 * Avoid ".", "..", and aliases of "." for obvious reasons. 3414 */ 3415 if ((scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.') || 3416 sdzp == szp || 3417 (scnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) { 3418 error = EINVAL; 3419 goto out; 3420 } 3421 3422 /* 3423 * Check to make sure rename is valid. 3424 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3425 */ 3426 if ((error = zfs_rename_check(szp, sdzp, tdzp))) 3427 goto out; 3428 } 3429 3430 /* 3431 * Does target exist? 3432 */ 3433 if (tzp) { 3434 /* 3435 * Source and target must be the same type. 3436 */ 3437 if ((*svpp)->v_type == VDIR) { 3438 if ((*tvpp)->v_type != VDIR) { 3439 error = SET_ERROR(ENOTDIR); 3440 goto out; 3441 } else { 3442 cache_purge(tdvp); 3443 if (sdvp != tdvp) 3444 cache_purge(sdvp); 3445 } 3446 } else { 3447 if ((*tvpp)->v_type == VDIR) { 3448 error = SET_ERROR(EISDIR); 3449 goto out; 3450 } 3451 } 3452 } 3453 3454 vn_seqc_write_begin(*svpp); 3455 vn_seqc_write_begin(sdvp); 3456 if (*tvpp != NULL) 3457 vn_seqc_write_begin(*tvpp); 3458 if (tdvp != *tvpp) 3459 vn_seqc_write_begin(tdvp); 3460 3461 vnevent_rename_src(*svpp, sdvp, scnp->cn_nameptr, ct); 3462 if (tzp) 3463 vnevent_rename_dest(*tvpp, tdvp, tnm, ct); 3464 3465 /* 3466 * notify the target directory if it is not the same 3467 * as source directory. 3468 */ 3469 if (tdvp != sdvp) { 3470 vnevent_rename_dest_dir(tdvp, ct); 3471 } 3472 3473 tx = dmu_tx_create(zfsvfs->z_os); 3474 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3475 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 3476 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3477 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3478 if (sdzp != tdzp) { 3479 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 3480 zfs_sa_upgrade_txholds(tx, tdzp); 3481 } 3482 if (tzp) { 3483 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 3484 zfs_sa_upgrade_txholds(tx, tzp); 3485 } 3486 3487 zfs_sa_upgrade_txholds(tx, szp); 3488 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3489 error = dmu_tx_assign(tx, DMU_TX_WAIT); 3490 if (error) { 3491 dmu_tx_abort(tx); 3492 goto out_seq; 3493 } 3494 3495 if (tzp) /* Attempt to remove the existing target */ 3496 error = zfs_link_destroy(tdzp, tnm, tzp, tx, 0, NULL); 3497 3498 if (error == 0) { 3499 error = zfs_link_create(tdzp, tnm, szp, tx, ZRENAMING); 3500 if (error == 0) { 3501 szp->z_pflags |= ZFS_AV_MODIFIED; 3502 3503 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 3504 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 3505 ASSERT0(error); 3506 3507 error = zfs_link_destroy(sdzp, snm, szp, tx, ZRENAMING, 3508 NULL); 3509 if (error == 0) { 3510 zfs_log_rename(zilog, tx, TX_RENAME, sdzp, 3511 snm, tdzp, tnm, szp); 3512 } else { 3513 /* 3514 * At this point, we have successfully created 3515 * the target name, but have failed to remove 3516 * the source name. Since the create was done 3517 * with the ZRENAMING flag, there are 3518 * complications; for one, the link count is 3519 * wrong. The easiest way to deal with this 3520 * is to remove the newly created target, and 3521 * return the original error. This must 3522 * succeed; fortunately, it is very unlikely to 3523 * fail, since we just created it. 3524 */ 3525 VERIFY0(zfs_link_destroy(tdzp, tnm, szp, tx, 3526 ZRENAMING, NULL)); 3527 } 3528 } 3529 if (error == 0) { 3530 cache_vop_rename(sdvp, *svpp, tdvp, *tvpp, scnp, tcnp); 3531 } 3532 } 3533 3534 dmu_tx_commit(tx); 3535 3536 out_seq: 3537 vn_seqc_write_end(*svpp); 3538 vn_seqc_write_end(sdvp); 3539 if (*tvpp != NULL) 3540 vn_seqc_write_end(*tvpp); 3541 if (tdvp != *tvpp) 3542 vn_seqc_write_end(tdvp); 3543 3544 out: 3545 if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3546 error = zil_commit(zilog, 0); 3547 zfs_exit(zfsvfs, FTAG); 3548 3549 return (error); 3550 } 3551 3552 int 3553 zfs_rename(znode_t *sdzp, const char *sname, znode_t *tdzp, const char *tname, 3554 cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap, zidmap_t *mnt_ns) 3555 { 3556 struct componentname scn, tcn; 3557 vnode_t *sdvp, *tdvp; 3558 vnode_t *svp, *tvp; 3559 int error; 3560 svp = tvp = NULL; 3561 3562 if (is_nametoolong(tdzp->z_zfsvfs, tname)) 3563 return (SET_ERROR(ENAMETOOLONG)); 3564 3565 if (rflags != 0 || wo_vap != NULL) 3566 return (SET_ERROR(EINVAL)); 3567 3568 sdvp = ZTOV(sdzp); 3569 tdvp = ZTOV(tdzp); 3570 error = zfs_lookup_internal(sdzp, sname, &svp, &scn, DELETE); 3571 if (sdzp->z_zfsvfs->z_replay == B_FALSE) 3572 VOP_UNLOCK(sdvp); 3573 if (error != 0) 3574 goto fail; 3575 VOP_UNLOCK(svp); 3576 3577 vn_lock(tdvp, LK_EXCLUSIVE | LK_RETRY); 3578 error = zfs_lookup_internal(tdzp, tname, &tvp, &tcn, RENAME); 3579 if (error == EJUSTRETURN) 3580 tvp = NULL; 3581 else if (error != 0) { 3582 VOP_UNLOCK(tdvp); 3583 goto fail; 3584 } 3585 3586 error = zfs_do_rename(sdvp, &svp, &scn, tdvp, &tvp, &tcn, cr); 3587 fail: 3588 if (svp != NULL) 3589 vrele(svp); 3590 if (tvp != NULL) 3591 vrele(tvp); 3592 3593 return (error); 3594 } 3595 3596 /* 3597 * Insert the indicated symbolic reference entry into the directory. 3598 * 3599 * IN: dvp - Directory to contain new symbolic link. 3600 * link - Name for new symlink entry. 3601 * vap - Attributes of new entry. 3602 * cr - credentials of caller. 3603 * ct - caller context 3604 * flags - case flags 3605 * mnt_ns - Unused on FreeBSD 3606 * 3607 * RETURN: 0 on success, error code on failure. 3608 * 3609 * Timestamps: 3610 * dvp - ctime|mtime updated 3611 */ 3612 int 3613 zfs_symlink(znode_t *dzp, const char *name, vattr_t *vap, 3614 const char *link, znode_t **zpp, cred_t *cr, int flags, zidmap_t *mnt_ns) 3615 { 3616 (void) flags; 3617 znode_t *zp; 3618 dmu_tx_t *tx; 3619 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3620 zilog_t *zilog; 3621 uint64_t len = strlen(link); 3622 int error; 3623 zfs_acl_ids_t acl_ids; 3624 boolean_t fuid_dirtied; 3625 uint64_t txtype = TX_SYMLINK; 3626 3627 ASSERT3S(vap->va_type, ==, VLNK); 3628 3629 if (is_nametoolong(zfsvfs, name)) 3630 return (SET_ERROR(ENAMETOOLONG)); 3631 3632 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 3633 return (error); 3634 zilog = zfsvfs->z_log; 3635 3636 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 3637 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3638 zfs_exit(zfsvfs, FTAG); 3639 return (SET_ERROR(EILSEQ)); 3640 } 3641 3642 if (len > MAXPATHLEN) { 3643 zfs_exit(zfsvfs, FTAG); 3644 return (SET_ERROR(ENAMETOOLONG)); 3645 } 3646 3647 if ((error = zfs_acl_ids_create(dzp, 0, 3648 vap, cr, NULL, &acl_ids, NULL)) != 0) { 3649 zfs_exit(zfsvfs, FTAG); 3650 return (error); 3651 } 3652 3653 /* 3654 * Attempt to lock directory; fail if entry already exists. 3655 */ 3656 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW); 3657 if (error) { 3658 zfs_acl_ids_free(&acl_ids); 3659 zfs_exit(zfsvfs, FTAG); 3660 return (error); 3661 } 3662 3663 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) { 3664 zfs_acl_ids_free(&acl_ids); 3665 zfs_exit(zfsvfs, FTAG); 3666 return (error); 3667 } 3668 3669 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, ZFS_DEFAULT_PROJID)) { 3670 zfs_acl_ids_free(&acl_ids); 3671 zfs_exit(zfsvfs, FTAG); 3672 return (SET_ERROR(EDQUOT)); 3673 } 3674 3675 getnewvnode_reserve(); 3676 tx = dmu_tx_create(zfsvfs->z_os); 3677 fuid_dirtied = zfsvfs->z_fuid_dirty; 3678 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 3679 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3680 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 3681 ZFS_SA_BASE_ATTR_SIZE + len); 3682 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 3683 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3684 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 3685 acl_ids.z_aclp->z_acl_bytes); 3686 } 3687 if (fuid_dirtied) 3688 zfs_fuid_txhold(zfsvfs, tx); 3689 error = dmu_tx_assign(tx, DMU_TX_WAIT); 3690 if (error) { 3691 zfs_acl_ids_free(&acl_ids); 3692 dmu_tx_abort(tx); 3693 getnewvnode_drop_reserve(); 3694 zfs_exit(zfsvfs, FTAG); 3695 return (error); 3696 } 3697 3698 /* 3699 * Create a new object for the symlink. 3700 * for version 4 ZPL datasets the symlink will be an SA attribute 3701 */ 3702 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 3703 3704 if (fuid_dirtied) 3705 zfs_fuid_sync(zfsvfs, tx); 3706 3707 if (zp->z_is_sa) 3708 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 3709 __DECONST(void *, link), len, tx); 3710 else 3711 zfs_sa_symlink(zp, __DECONST(char *, link), len, tx); 3712 3713 zp->z_size = len; 3714 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 3715 &zp->z_size, sizeof (zp->z_size), tx); 3716 /* 3717 * Insert the new object into the directory. 3718 */ 3719 error = zfs_link_create(dzp, name, zp, tx, ZNEW); 3720 if (error != 0) { 3721 zfs_znode_delete(zp, tx); 3722 VOP_UNLOCK(ZTOV(zp)); 3723 zrele(zp); 3724 } else { 3725 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 3726 } 3727 3728 zfs_acl_ids_free(&acl_ids); 3729 3730 dmu_tx_commit(tx); 3731 3732 getnewvnode_drop_reserve(); 3733 3734 if (error == 0) { 3735 *zpp = zp; 3736 3737 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3738 error = zil_commit(zilog, 0); 3739 } 3740 3741 zfs_exit(zfsvfs, FTAG); 3742 return (error); 3743 } 3744 3745 /* 3746 * Return, in the buffer contained in the provided uio structure, 3747 * the symbolic path referred to by vp. 3748 * 3749 * IN: vp - vnode of symbolic link. 3750 * uio - structure to contain the link path. 3751 * cr - credentials of caller. 3752 * ct - caller context 3753 * 3754 * OUT: uio - structure containing the link path. 3755 * 3756 * RETURN: 0 on success, error code on failure. 3757 * 3758 * Timestamps: 3759 * vp - atime updated 3760 */ 3761 static int 3762 zfs_readlink(vnode_t *vp, zfs_uio_t *uio, cred_t *cr, caller_context_t *ct) 3763 { 3764 (void) cr, (void) ct; 3765 znode_t *zp = VTOZ(vp); 3766 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3767 int error; 3768 3769 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3770 return (error); 3771 3772 if (zp->z_is_sa) 3773 error = sa_lookup_uio(zp->z_sa_hdl, 3774 SA_ZPL_SYMLINK(zfsvfs), uio); 3775 else 3776 error = zfs_sa_readlink(zp, uio); 3777 3778 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3779 3780 zfs_exit(zfsvfs, FTAG); 3781 return (error); 3782 } 3783 3784 /* 3785 * Insert a new entry into directory tdvp referencing svp. 3786 * 3787 * IN: tdvp - Directory to contain new entry. 3788 * svp - vnode of new entry. 3789 * name - name of new entry. 3790 * cr - credentials of caller. 3791 * 3792 * RETURN: 0 on success, error code on failure. 3793 * 3794 * Timestamps: 3795 * tdvp - ctime|mtime updated 3796 * svp - ctime updated 3797 */ 3798 int 3799 zfs_link(znode_t *tdzp, znode_t *szp, const char *name, cred_t *cr, 3800 int flags) 3801 { 3802 (void) flags; 3803 znode_t *tzp; 3804 zfsvfs_t *zfsvfs = tdzp->z_zfsvfs; 3805 zilog_t *zilog; 3806 dmu_tx_t *tx; 3807 int error; 3808 uint64_t parent; 3809 uid_t owner; 3810 3811 ASSERT3S(ZTOV(tdzp)->v_type, ==, VDIR); 3812 3813 if (is_nametoolong(zfsvfs, name)) 3814 return (SET_ERROR(ENAMETOOLONG)); 3815 3816 if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0) 3817 return (error); 3818 zilog = zfsvfs->z_log; 3819 3820 /* 3821 * POSIX dictates that we return EPERM here. 3822 * Better choices include ENOTSUP or EISDIR. 3823 */ 3824 if (ZTOV(szp)->v_type == VDIR) { 3825 zfs_exit(zfsvfs, FTAG); 3826 return (SET_ERROR(EPERM)); 3827 } 3828 3829 if ((error = zfs_verify_zp(szp)) != 0) { 3830 zfs_exit(zfsvfs, FTAG); 3831 return (error); 3832 } 3833 3834 /* 3835 * If we are using project inheritance, means if the directory has 3836 * ZFS_PROJINHERIT set, then its descendant directories will inherit 3837 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under 3838 * such case, we only allow hard link creation in our tree when the 3839 * project IDs are the same. 3840 */ 3841 if (tdzp->z_pflags & ZFS_PROJINHERIT && 3842 tdzp->z_projid != szp->z_projid) { 3843 zfs_exit(zfsvfs, FTAG); 3844 return (SET_ERROR(EXDEV)); 3845 } 3846 3847 if (szp->z_pflags & (ZFS_APPENDONLY | 3848 ZFS_IMMUTABLE | ZFS_READONLY)) { 3849 zfs_exit(zfsvfs, FTAG); 3850 return (SET_ERROR(EPERM)); 3851 } 3852 3853 /* Prevent links to .zfs/shares files */ 3854 3855 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 3856 &parent, sizeof (uint64_t))) != 0) { 3857 zfs_exit(zfsvfs, FTAG); 3858 return (error); 3859 } 3860 if (parent == zfsvfs->z_shares_dir) { 3861 zfs_exit(zfsvfs, FTAG); 3862 return (SET_ERROR(EPERM)); 3863 } 3864 3865 if (zfsvfs->z_utf8 && u8_validate(name, 3866 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3867 zfs_exit(zfsvfs, FTAG); 3868 return (SET_ERROR(EILSEQ)); 3869 } 3870 3871 /* 3872 * We do not support links between attributes and non-attributes 3873 * because of the potential security risk of creating links 3874 * into "normal" file space in order to circumvent restrictions 3875 * imposed in attribute space. 3876 */ 3877 if ((szp->z_pflags & ZFS_XATTR) != (tdzp->z_pflags & ZFS_XATTR)) { 3878 zfs_exit(zfsvfs, FTAG); 3879 return (SET_ERROR(EINVAL)); 3880 } 3881 3882 3883 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 3884 if (owner != crgetuid(cr) && secpolicy_basic_link(ZTOV(szp), cr) != 0) { 3885 zfs_exit(zfsvfs, FTAG); 3886 return (SET_ERROR(EPERM)); 3887 } 3888 3889 if ((error = zfs_zaccess(tdzp, ACE_ADD_FILE, 0, B_FALSE, cr, NULL))) { 3890 zfs_exit(zfsvfs, FTAG); 3891 return (error); 3892 } 3893 3894 /* 3895 * Attempt to lock directory; fail if entry already exists. 3896 */ 3897 error = zfs_dirent_lookup(tdzp, name, &tzp, ZNEW); 3898 if (error) { 3899 zfs_exit(zfsvfs, FTAG); 3900 return (error); 3901 } 3902 3903 tx = dmu_tx_create(zfsvfs->z_os); 3904 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3905 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, name); 3906 zfs_sa_upgrade_txholds(tx, szp); 3907 zfs_sa_upgrade_txholds(tx, tdzp); 3908 error = dmu_tx_assign(tx, DMU_TX_WAIT); 3909 if (error) { 3910 dmu_tx_abort(tx); 3911 zfs_exit(zfsvfs, FTAG); 3912 return (error); 3913 } 3914 3915 error = zfs_link_create(tdzp, name, szp, tx, 0); 3916 3917 if (error == 0) { 3918 uint64_t txtype = TX_LINK; 3919 zfs_log_link(zilog, tx, txtype, tdzp, szp, name); 3920 } 3921 3922 dmu_tx_commit(tx); 3923 3924 if (error == 0) { 3925 vnevent_link(ZTOV(szp), ct); 3926 } 3927 3928 if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3929 error = zil_commit(zilog, 0); 3930 3931 zfs_exit(zfsvfs, FTAG); 3932 return (error); 3933 } 3934 3935 /* 3936 * Free or allocate space in a file. Currently, this function only 3937 * supports the `F_FREESP' command. However, this command is somewhat 3938 * misnamed, as its functionality includes the ability to allocate as 3939 * well as free space. 3940 * 3941 * IN: ip - inode of file to free data in. 3942 * cmd - action to take (only F_FREESP supported). 3943 * bfp - section of file to free/alloc. 3944 * flag - current file open mode flags. 3945 * offset - current file offset. 3946 * cr - credentials of caller. 3947 * 3948 * RETURN: 0 on success, error code on failure. 3949 * 3950 * Timestamps: 3951 * ip - ctime|mtime updated 3952 */ 3953 int 3954 zfs_space(znode_t *zp, int cmd, flock64_t *bfp, int flag, 3955 offset_t offset, cred_t *cr) 3956 { 3957 (void) offset; 3958 zfsvfs_t *zfsvfs = ZTOZSB(zp); 3959 uint64_t off, len; 3960 int error; 3961 3962 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3963 return (error); 3964 3965 if (cmd != F_FREESP) { 3966 zfs_exit(zfsvfs, FTAG); 3967 return (SET_ERROR(EINVAL)); 3968 } 3969 3970 /* 3971 * Callers might not be able to detect properly that we are read-only, 3972 * so check it explicitly here. 3973 */ 3974 if (zfs_is_readonly(zfsvfs)) { 3975 zfs_exit(zfsvfs, FTAG); 3976 return (SET_ERROR(EROFS)); 3977 } 3978 3979 if (bfp->l_len < 0) { 3980 zfs_exit(zfsvfs, FTAG); 3981 return (SET_ERROR(EINVAL)); 3982 } 3983 3984 /* 3985 * Permissions aren't checked on Solaris because on this OS 3986 * zfs_space() can only be called with an opened file handle. 3987 * On Linux we can get here through truncate_range() which 3988 * operates directly on inodes, so we need to check access rights. 3989 */ 3990 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr, NULL))) { 3991 zfs_exit(zfsvfs, FTAG); 3992 return (error); 3993 } 3994 3995 off = bfp->l_start; 3996 len = bfp->l_len; /* 0 means from off to end of file */ 3997 3998 error = zfs_freesp(zp, off, len, flag, TRUE); 3999 4000 zfs_exit(zfsvfs, FTAG); 4001 return (error); 4002 } 4003 4004 static void 4005 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4006 { 4007 (void) cr, (void) ct; 4008 znode_t *zp = VTOZ(vp); 4009 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4010 int error; 4011 4012 ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs); 4013 if (zp->z_sa_hdl == NULL) { 4014 /* 4015 * The fs has been unmounted, or we did a 4016 * suspend/resume and this file no longer exists. 4017 */ 4018 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 4019 vrecycle(vp); 4020 return; 4021 } 4022 4023 if (zp->z_unlinked) { 4024 /* 4025 * Fast path to recycle a vnode of a removed file. 4026 */ 4027 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 4028 vrecycle(vp); 4029 return; 4030 } 4031 4032 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4033 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4034 4035 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4036 zfs_sa_upgrade_txholds(tx, zp); 4037 error = dmu_tx_assign(tx, DMU_TX_WAIT); 4038 if (error) { 4039 dmu_tx_abort(tx); 4040 } else { 4041 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4042 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4043 zp->z_atime_dirty = 0; 4044 dmu_tx_commit(tx); 4045 } 4046 } 4047 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 4048 } 4049 4050 4051 _Static_assert(sizeof (struct zfid_short) <= sizeof (struct fid), 4052 "struct zfid_short bigger than struct fid"); 4053 _Static_assert(sizeof (struct zfid_long) <= sizeof (struct fid), 4054 "struct zfid_long bigger than struct fid"); 4055 4056 static int 4057 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 4058 { 4059 (void) ct; 4060 znode_t *zp = VTOZ(vp); 4061 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4062 uint32_t gen; 4063 uint64_t gen64; 4064 uint64_t object = zp->z_id; 4065 zfid_short_t *zfid; 4066 int size, i, error; 4067 4068 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 4069 return (error); 4070 4071 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 4072 &gen64, sizeof (uint64_t))) != 0) { 4073 zfs_exit(zfsvfs, FTAG); 4074 return (error); 4075 } 4076 4077 gen = (uint32_t)gen64; 4078 4079 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 4080 fidp->fid_len = size; 4081 4082 zfid = (zfid_short_t *)fidp; 4083 4084 zfid->zf_len = size; 4085 4086 for (i = 0; i < sizeof (zfid->zf_object); i++) 4087 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 4088 4089 /* Must have a non-zero generation number to distinguish from .zfs */ 4090 if (gen == 0) 4091 gen = 1; 4092 for (i = 0; i < sizeof (zfid->zf_gen); i++) 4093 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 4094 4095 if (size == LONG_FID_LEN) { 4096 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 4097 zfid_long_t *zlfid; 4098 4099 zlfid = (zfid_long_t *)fidp; 4100 4101 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 4102 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 4103 4104 /* XXX - this should be the generation number for the objset */ 4105 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 4106 zlfid->zf_setgen[i] = 0; 4107 } 4108 4109 zfs_exit(zfsvfs, FTAG); 4110 return (0); 4111 } 4112 4113 static int 4114 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 4115 caller_context_t *ct) 4116 { 4117 znode_t *zp; 4118 zfsvfs_t *zfsvfs; 4119 uint_t blksize, iosize; 4120 int error; 4121 4122 switch (cmd) { 4123 case _PC_LINK_MAX: 4124 *valp = MIN(LONG_MAX, ZFS_LINK_MAX); 4125 return (0); 4126 4127 case _PC_FILESIZEBITS: 4128 *valp = 64; 4129 return (0); 4130 case _PC_MIN_HOLE_SIZE: 4131 iosize = vp->v_mount->mnt_stat.f_iosize; 4132 if (vp->v_type == VREG) { 4133 zp = VTOZ(vp); 4134 blksize = zp->z_blksz; 4135 if (zp->z_size <= blksize) 4136 blksize = MAX(blksize, iosize); 4137 *valp = (int)blksize; 4138 return (0); 4139 } 4140 if (vp->v_type == VDIR) { 4141 *valp = (int)iosize; 4142 return (0); 4143 } 4144 return (EINVAL); 4145 case _PC_ACL_EXTENDED: 4146 #if 0 /* POSIX ACLs are not implemented for ZFS on FreeBSD yet. */ 4147 zp = VTOZ(vp); 4148 zfsvfs = zp->z_zfsvfs; 4149 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 4150 return (error); 4151 *valp = zfsvfs->z_acl_type == ZFSACLTYPE_POSIX ? 1 : 0; 4152 zfs_exit(zfsvfs, FTAG); 4153 #else 4154 *valp = 0; 4155 #endif 4156 return (0); 4157 4158 case _PC_ACL_NFS4: 4159 zp = VTOZ(vp); 4160 zfsvfs = zp->z_zfsvfs; 4161 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 4162 return (error); 4163 *valp = zfsvfs->z_acl_type == ZFS_ACLTYPE_NFSV4 ? 1 : 0; 4164 zfs_exit(zfsvfs, FTAG); 4165 return (0); 4166 4167 case _PC_ACL_PATH_MAX: 4168 *valp = ACL_MAX_ENTRIES; 4169 return (0); 4170 4171 default: 4172 return (EOPNOTSUPP); 4173 } 4174 } 4175 4176 static int 4177 zfs_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind, 4178 int *rahead) 4179 { 4180 znode_t *zp = VTOZ(vp); 4181 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4182 zfs_locked_range_t *lr; 4183 vm_object_t object; 4184 off_t start, end, obj_size; 4185 uint_t blksz; 4186 int pgsin_b, pgsin_a; 4187 int error; 4188 4189 if (zfs_enter_verify_zp(zfsvfs, zp, FTAG) != 0) 4190 return (zfs_vm_pagerret_error); 4191 4192 object = ma[0]->object; 4193 start = IDX_TO_OFF(ma[0]->pindex); 4194 end = IDX_TO_OFF(ma[count - 1]->pindex + 1); 4195 4196 /* 4197 * Lock a range covering all required and optional pages. 4198 * Note that we need to handle the case of the block size growing. 4199 */ 4200 for (;;) { 4201 uint64_t len; 4202 4203 blksz = zp->z_blksz; 4204 len = roundup(end, blksz) - rounddown(start, blksz); 4205 4206 lr = zfs_rangelock_tryenter(&zp->z_rangelock, 4207 rounddown(start, blksz), len, RL_READER); 4208 if (lr == NULL) { 4209 /* 4210 * Avoid a deadlock with update_pages(). We need to 4211 * hold the range lock when copying from the DMU, so 4212 * give up the busy lock to allow update_pages() to 4213 * proceed. We might need to allocate new pages, which 4214 * isn't quite right since this allocation isn't subject 4215 * to the page fault handler's OOM logic, but this is 4216 * the best we can do for now. 4217 */ 4218 for (int i = 0; i < count; i++) 4219 vm_page_xunbusy(ma[i]); 4220 4221 lr = zfs_rangelock_enter(&zp->z_rangelock, 4222 rounddown(start, blksz), len, RL_READER); 4223 4224 zfs_vmobject_wlock(object); 4225 (void) vm_page_grab_pages(object, OFF_TO_IDX(start), 4226 VM_ALLOC_NORMAL | VM_ALLOC_WAITOK, 4227 ma, count); 4228 if (!vm_page_all_valid(ma[count - 1])) { 4229 /* 4230 * Later in this function, we copy DMU data to 4231 * invalid pages only. The last page may not be 4232 * entirely filled though, if the file does not 4233 * end on a page boundary. Therefore, we zero 4234 * that last page here to make sure it does not 4235 * contain garbage after the end of file. 4236 */ 4237 ASSERT(vm_page_none_valid(ma[count - 1])); 4238 vm_page_zero_invalid(ma[count - 1], FALSE); 4239 } 4240 zfs_vmobject_wunlock(object); 4241 } 4242 if (blksz == zp->z_blksz) 4243 break; 4244 zfs_rangelock_exit(lr); 4245 } 4246 4247 zfs_vmobject_wlock(object); 4248 obj_size = object->un_pager.vnp.vnp_size; 4249 zfs_vmobject_wunlock(object); 4250 if (IDX_TO_OFF(ma[count - 1]->pindex) >= obj_size) { 4251 zfs_rangelock_exit(lr); 4252 zfs_exit(zfsvfs, FTAG); 4253 return (zfs_vm_pagerret_bad); 4254 } 4255 4256 pgsin_b = 0; 4257 if (rbehind != NULL) { 4258 pgsin_b = OFF_TO_IDX(start - rounddown(start, blksz)); 4259 pgsin_b = MIN(*rbehind, pgsin_b); 4260 } 4261 4262 pgsin_a = 0; 4263 if (rahead != NULL) { 4264 pgsin_a = OFF_TO_IDX(roundup(end, blksz) - end); 4265 if (end + IDX_TO_OFF(pgsin_a) >= obj_size) 4266 pgsin_a = OFF_TO_IDX(round_page(obj_size) - end); 4267 pgsin_a = MIN(*rahead, pgsin_a); 4268 } 4269 4270 /* 4271 * NB: we need to pass the exact byte size of the data that we expect 4272 * to read after accounting for the file size. This is required because 4273 * ZFS will panic if we request DMU to read beyond the end of the last 4274 * allocated block. 4275 */ 4276 for (int i = 0; i < count; i++) { 4277 int dummypgsin, count1, j, last_size; 4278 4279 if (vm_page_any_valid(ma[i])) { 4280 ASSERT(vm_page_all_valid(ma[i])); 4281 continue; 4282 } 4283 for (j = i + 1; j < count; j++) { 4284 if (vm_page_any_valid(ma[j])) { 4285 ASSERT(vm_page_all_valid(ma[j])); 4286 break; 4287 } 4288 } 4289 count1 = j - i; 4290 dummypgsin = 0; 4291 last_size = j == count ? 4292 MIN(end, obj_size) - (end - PAGE_SIZE) : PAGE_SIZE; 4293 error = dmu_read_pages(zfsvfs->z_os, zp->z_id, &ma[i], count1, 4294 i == 0 ? &pgsin_b : &dummypgsin, 4295 j == count ? &pgsin_a : &dummypgsin, 4296 last_size); 4297 if (error != 0) 4298 break; 4299 i += count1 - 1; 4300 } 4301 4302 zfs_rangelock_exit(lr); 4303 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4304 4305 dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, count*PAGE_SIZE); 4306 4307 zfs_exit(zfsvfs, FTAG); 4308 4309 if (error != 0) 4310 return (zfs_vm_pagerret_error); 4311 4312 VM_CNT_INC(v_vnodein); 4313 VM_CNT_ADD(v_vnodepgsin, count + pgsin_b + pgsin_a); 4314 if (rbehind != NULL) 4315 *rbehind = pgsin_b; 4316 if (rahead != NULL) 4317 *rahead = pgsin_a; 4318 return (zfs_vm_pagerret_ok); 4319 } 4320 4321 #ifndef _SYS_SYSPROTO_H_ 4322 struct vop_getpages_args { 4323 struct vnode *a_vp; 4324 vm_page_t *a_m; 4325 int a_count; 4326 int *a_rbehind; 4327 int *a_rahead; 4328 }; 4329 #endif 4330 4331 static int 4332 zfs_freebsd_getpages(struct vop_getpages_args *ap) 4333 { 4334 4335 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind, 4336 ap->a_rahead)); 4337 } 4338 4339 typedef struct { 4340 uint_t pca_npages; 4341 vm_page_t pca_pages[]; 4342 } putpage_commit_arg_t; 4343 4344 static void 4345 zfs_putpage_commit_cb(void *arg, int err) 4346 { 4347 putpage_commit_arg_t *pca = arg; 4348 vm_object_t object = pca->pca_pages[0]->object; 4349 4350 zfs_vmobject_wlock(object); 4351 4352 for (uint_t i = 0; i < pca->pca_npages; i++) { 4353 vm_page_t pp = pca->pca_pages[i]; 4354 4355 if (err == 0) { 4356 /* 4357 * Writeback succeeded, so undirty the page. If it 4358 * fails, we leave it in the same state it was. That's 4359 * most likely dirty, so it will get tried again some 4360 * other time. 4361 */ 4362 vm_page_undirty(pp); 4363 } 4364 4365 vm_page_sunbusy(pp); 4366 } 4367 4368 vm_object_pip_wakeupn(object, pca->pca_npages); 4369 4370 zfs_vmobject_wunlock(object); 4371 4372 kmem_free(pca, 4373 offsetof(putpage_commit_arg_t, pca_pages[pca->pca_npages])); 4374 } 4375 4376 static int 4377 zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags, 4378 int *rtvals) 4379 { 4380 znode_t *zp = VTOZ(vp); 4381 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4382 zfs_locked_range_t *lr; 4383 dmu_tx_t *tx; 4384 struct sf_buf *sf; 4385 vm_object_t object; 4386 vm_page_t m; 4387 caddr_t va; 4388 size_t tocopy; 4389 size_t lo_len; 4390 vm_ooffset_t lo_off; 4391 vm_ooffset_t off; 4392 uint_t blksz; 4393 int ncount; 4394 int pcount; 4395 int err; 4396 int i; 4397 4398 object = vp->v_object; 4399 KASSERT(ma[0]->object == object, ("mismatching object")); 4400 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length")); 4401 4402 pcount = btoc(len); 4403 ncount = pcount; 4404 for (i = 0; i < pcount; i++) 4405 rtvals[i] = zfs_vm_pagerret_error; 4406 4407 if (zfs_enter_verify_zp(zfsvfs, zp, FTAG) != 0) 4408 return (zfs_vm_pagerret_error); 4409 4410 off = IDX_TO_OFF(ma[0]->pindex); 4411 blksz = zp->z_blksz; 4412 lo_off = rounddown(off, blksz); 4413 lo_len = roundup(len + (off - lo_off), blksz); 4414 lr = zfs_rangelock_enter(&zp->z_rangelock, lo_off, lo_len, RL_WRITER); 4415 4416 zfs_vmobject_wlock(object); 4417 if (len + off > object->un_pager.vnp.vnp_size) { 4418 if (object->un_pager.vnp.vnp_size > off) { 4419 int pgoff; 4420 4421 len = object->un_pager.vnp.vnp_size - off; 4422 ncount = btoc(len); 4423 if ((pgoff = (int)len & PAGE_MASK) != 0) { 4424 /* 4425 * If the object is locked and the following 4426 * conditions hold, then the page's dirty 4427 * field cannot be concurrently changed by a 4428 * pmap operation. 4429 */ 4430 m = ma[ncount - 1]; 4431 vm_page_assert_sbusied(m); 4432 KASSERT(!pmap_page_is_write_mapped(m), 4433 ("zfs_putpages: page %p is not read-only", 4434 m)); 4435 vm_page_clear_dirty(m, pgoff, PAGE_SIZE - 4436 pgoff); 4437 } 4438 } else { 4439 len = 0; 4440 ncount = 0; 4441 } 4442 if (ncount < pcount) { 4443 for (i = ncount; i < pcount; i++) { 4444 rtvals[i] = zfs_vm_pagerret_bad; 4445 } 4446 } 4447 } 4448 zfs_vmobject_wunlock(object); 4449 4450 boolean_t commit = (flags & (zfs_vm_pagerput_sync | 4451 zfs_vm_pagerput_inval)) != 0 || 4452 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS; 4453 4454 if (ncount == 0) 4455 goto out; 4456 4457 if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT, zp->z_uid) || 4458 zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT, zp->z_gid) || 4459 (zp->z_projid != ZFS_DEFAULT_PROJID && 4460 zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT, 4461 zp->z_projid))) { 4462 goto out; 4463 } 4464 4465 tx = dmu_tx_create(zfsvfs->z_os); 4466 dmu_tx_hold_write(tx, zp->z_id, off, len); 4467 4468 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4469 zfs_sa_upgrade_txholds(tx, zp); 4470 err = dmu_tx_assign(tx, DMU_TX_WAIT); 4471 if (err != 0) { 4472 dmu_tx_abort(tx); 4473 goto out; 4474 } 4475 4476 if (zp->z_blksz < PAGE_SIZE) { 4477 vm_ooffset_t woff = off; 4478 size_t wlen = len; 4479 for (i = 0; wlen > 0; woff += tocopy, wlen -= tocopy, i++) { 4480 tocopy = MIN(PAGE_SIZE, wlen); 4481 va = zfs_map_page(ma[i], &sf); 4482 dmu_write(zfsvfs->z_os, zp->z_id, woff, tocopy, va, tx, 4483 DMU_READ_PREFETCH); 4484 zfs_unmap_page(sf); 4485 } 4486 } else { 4487 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx); 4488 } 4489 4490 if (err == 0) { 4491 uint64_t mtime[2], ctime[2]; 4492 sa_bulk_attr_t bulk[3]; 4493 int count = 0; 4494 4495 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4496 &mtime, 16); 4497 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4498 &ctime, 16); 4499 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4500 &zp->z_pflags, 8); 4501 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime); 4502 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 4503 ASSERT0(err); 4504 4505 if (commit) { 4506 /* 4507 * Caller requested that we commit immediately. We set 4508 * a callback on the log entry, to be called once its 4509 * on disk after the call to zil_commit() below. The 4510 * pages will be undirtied and unbusied there. 4511 */ 4512 putpage_commit_arg_t *pca = kmem_alloc( 4513 offsetof(putpage_commit_arg_t, pca_pages[ncount]), 4514 KM_SLEEP); 4515 pca->pca_npages = ncount; 4516 memcpy(pca->pca_pages, ma, sizeof (vm_page_t) * ncount); 4517 4518 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 4519 B_TRUE, B_FALSE, zfs_putpage_commit_cb, pca); 4520 4521 for (i = 0; i < ncount; i++) 4522 rtvals[i] = zfs_vm_pagerret_pend; 4523 } else { 4524 /* 4525 * Caller just wants the page written back somewhere, 4526 * but doesn't need it committed yet. We've already 4527 * written it back to the DMU, so we just need to put 4528 * it on the async log, then undirty the page and 4529 * return. 4530 * 4531 * We cannot use a callback here, because it would keep 4532 * the page busy (locked) until it is eventually 4533 * written down at txg sync. 4534 */ 4535 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 4536 B_FALSE, B_FALSE, NULL, NULL); 4537 4538 zfs_vmobject_wlock(object); 4539 for (i = 0; i < ncount; i++) { 4540 rtvals[i] = zfs_vm_pagerret_ok; 4541 vm_page_undirty(ma[i]); 4542 } 4543 zfs_vmobject_wunlock(object); 4544 } 4545 4546 VM_CNT_INC(v_vnodeout); 4547 VM_CNT_ADD(v_vnodepgsout, ncount); 4548 } 4549 dmu_tx_commit(tx); 4550 4551 out: 4552 zfs_rangelock_exit(lr); 4553 if (commit) { 4554 err = zil_commit(zfsvfs->z_log, zp->z_id); 4555 if (err != 0) { 4556 zfs_exit(zfsvfs, FTAG); 4557 return (err); 4558 } 4559 } 4560 4561 dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, len); 4562 4563 zfs_exit(zfsvfs, FTAG); 4564 return (rtvals[0]); 4565 } 4566 4567 #ifndef _SYS_SYSPROTO_H_ 4568 struct vop_putpages_args { 4569 struct vnode *a_vp; 4570 vm_page_t *a_m; 4571 int a_count; 4572 int a_sync; 4573 int *a_rtvals; 4574 }; 4575 #endif 4576 4577 static int 4578 zfs_freebsd_putpages(struct vop_putpages_args *ap) 4579 { 4580 4581 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync, 4582 ap->a_rtvals)); 4583 } 4584 4585 #ifndef _SYS_SYSPROTO_H_ 4586 struct vop_bmap_args { 4587 struct vnode *a_vp; 4588 daddr_t a_bn; 4589 struct bufobj **a_bop; 4590 daddr_t *a_bnp; 4591 int *a_runp; 4592 int *a_runb; 4593 }; 4594 #endif 4595 4596 static int 4597 zfs_freebsd_bmap(struct vop_bmap_args *ap) 4598 { 4599 4600 if (ap->a_bop != NULL) 4601 *ap->a_bop = &ap->a_vp->v_bufobj; 4602 if (ap->a_bnp != NULL) 4603 *ap->a_bnp = ap->a_bn; 4604 if (ap->a_runp != NULL) 4605 *ap->a_runp = 0; 4606 if (ap->a_runb != NULL) 4607 *ap->a_runb = 0; 4608 4609 return (0); 4610 } 4611 4612 #ifndef _SYS_SYSPROTO_H_ 4613 struct vop_open_args { 4614 struct vnode *a_vp; 4615 int a_mode; 4616 struct ucred *a_cred; 4617 struct thread *a_td; 4618 }; 4619 #endif 4620 4621 static int 4622 zfs_freebsd_open(struct vop_open_args *ap) 4623 { 4624 vnode_t *vp = ap->a_vp; 4625 znode_t *zp = VTOZ(vp); 4626 int error; 4627 4628 error = zfs_open(&vp, ap->a_mode, ap->a_cred); 4629 if (error == 0) 4630 vnode_create_vobject(vp, zp->z_size, ap->a_td); 4631 return (error); 4632 } 4633 4634 #ifndef _SYS_SYSPROTO_H_ 4635 struct vop_close_args { 4636 struct vnode *a_vp; 4637 int a_fflag; 4638 struct ucred *a_cred; 4639 struct thread *a_td; 4640 }; 4641 #endif 4642 4643 static int 4644 zfs_freebsd_close(struct vop_close_args *ap) 4645 { 4646 4647 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred)); 4648 } 4649 4650 #ifndef _SYS_SYSPROTO_H_ 4651 struct vop_ioctl_args { 4652 struct vnode *a_vp; 4653 ulong_t a_command; 4654 caddr_t a_data; 4655 int a_fflag; 4656 struct ucred *cred; 4657 struct thread *td; 4658 }; 4659 #endif 4660 4661 static int 4662 zfs_freebsd_ioctl(struct vop_ioctl_args *ap) 4663 { 4664 4665 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 4666 ap->a_fflag, ap->a_cred, NULL)); 4667 } 4668 4669 static int 4670 ioflags(int ioflags) 4671 { 4672 int flags = 0; 4673 4674 if (ioflags & IO_APPEND) 4675 flags |= O_APPEND; 4676 if (ioflags & IO_NDELAY) 4677 flags |= O_NONBLOCK; 4678 if (ioflags & IO_DIRECT) 4679 flags |= O_DIRECT; 4680 if (ioflags & IO_SYNC) 4681 flags |= O_SYNC; 4682 4683 return (flags); 4684 } 4685 4686 #ifndef _SYS_SYSPROTO_H_ 4687 struct vop_read_args { 4688 struct vnode *a_vp; 4689 struct uio *a_uio; 4690 int a_ioflag; 4691 struct ucred *a_cred; 4692 }; 4693 #endif 4694 4695 static int 4696 zfs_freebsd_read(struct vop_read_args *ap) 4697 { 4698 zfs_uio_t uio; 4699 int error = 0; 4700 zfs_uio_init(&uio, ap->a_uio); 4701 error = zfs_read(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag), 4702 ap->a_cred); 4703 /* 4704 * XXX We occasionally get an EFAULT for Direct I/O reads on 4705 * FreeBSD 13. This still needs to be resolved. The EFAULT comes 4706 * from: 4707 * zfs_uio_get__dio_pages_alloc() -> 4708 * zfs_uio_get_dio_pages_impl() -> 4709 * zfs_uio_iov_step() -> 4710 * zfs_uio_get_user_pages(). 4711 * We return EFAULT from zfs_uio_iov_step(). When a Direct I/O 4712 * read fails to map in the user pages (returning EFAULT) the 4713 * Direct I/O request is broken up into two separate IO requests 4714 * and issued separately using Direct I/O. 4715 */ 4716 #ifdef ZFS_DEBUG 4717 if (error == EFAULT && uio.uio_extflg & UIO_DIRECT) { 4718 #if 0 4719 printf("%s(%d): Direct I/O read returning EFAULT " 4720 "uio = %p, zfs_uio_offset(uio) = %lu " 4721 "zfs_uio_resid(uio) = %lu\n", 4722 __FUNCTION__, __LINE__, &uio, zfs_uio_offset(&uio), 4723 zfs_uio_resid(&uio)); 4724 #endif 4725 } 4726 4727 #endif 4728 return (error); 4729 } 4730 4731 #ifndef _SYS_SYSPROTO_H_ 4732 struct vop_write_args { 4733 struct vnode *a_vp; 4734 struct uio *a_uio; 4735 int a_ioflag; 4736 struct ucred *a_cred; 4737 }; 4738 #endif 4739 4740 static int 4741 zfs_freebsd_write(struct vop_write_args *ap) 4742 { 4743 zfs_uio_t uio; 4744 zfs_uio_init(&uio, ap->a_uio); 4745 return (zfs_write(VTOZ(ap->a_vp), &uio, ioflags(ap->a_ioflag), 4746 ap->a_cred)); 4747 } 4748 4749 /* 4750 * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see 4751 * the comment above cache_fplookup for details. 4752 */ 4753 static int 4754 zfs_freebsd_fplookup_vexec(struct vop_fplookup_vexec_args *v) 4755 { 4756 vnode_t *vp; 4757 znode_t *zp; 4758 uint64_t pflags; 4759 4760 vp = v->a_vp; 4761 zp = VTOZ_SMR(vp); 4762 if (__predict_false(zp == NULL)) 4763 return (EAGAIN); 4764 pflags = atomic_load_64(&zp->z_pflags); 4765 if (pflags & ZFS_AV_QUARANTINED) 4766 return (EAGAIN); 4767 if (pflags & ZFS_XATTR) 4768 return (EAGAIN); 4769 if ((pflags & ZFS_NO_EXECS_DENIED) == 0) 4770 return (EAGAIN); 4771 return (0); 4772 } 4773 4774 static int 4775 zfs_freebsd_fplookup_symlink(struct vop_fplookup_symlink_args *v) 4776 { 4777 vnode_t *vp; 4778 znode_t *zp; 4779 char *target; 4780 4781 vp = v->a_vp; 4782 zp = VTOZ_SMR(vp); 4783 if (__predict_false(zp == NULL)) { 4784 return (EAGAIN); 4785 } 4786 4787 target = atomic_load_consume_ptr(&zp->z_cached_symlink); 4788 if (target == NULL) { 4789 return (EAGAIN); 4790 } 4791 return (cache_symlink_resolve(v->a_fpl, target, strlen(target))); 4792 } 4793 4794 #ifndef _SYS_SYSPROTO_H_ 4795 struct vop_access_args { 4796 struct vnode *a_vp; 4797 accmode_t a_accmode; 4798 struct ucred *a_cred; 4799 struct thread *a_td; 4800 }; 4801 #endif 4802 4803 static int 4804 zfs_freebsd_access(struct vop_access_args *ap) 4805 { 4806 vnode_t *vp = ap->a_vp; 4807 znode_t *zp = VTOZ(vp); 4808 accmode_t accmode; 4809 int error = 0; 4810 4811 4812 if (ap->a_accmode == VEXEC) { 4813 if (zfs_fastaccesschk_execute(zp, ap->a_cred) == 0) 4814 return (0); 4815 } 4816 4817 /* 4818 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 4819 */ 4820 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 4821 if (accmode != 0) { 4822 #if __FreeBSD_version >= 1500040 4823 /* For named attributes, do the checks. */ 4824 if ((vn_irflag_read(vp) & VIRF_NAMEDATTR) != 0) 4825 error = zfs_access(zp, accmode, V_NAMEDATTR, 4826 ap->a_cred); 4827 else 4828 #endif 4829 error = zfs_access(zp, accmode, 0, ap->a_cred); 4830 } 4831 4832 /* 4833 * VADMIN has to be handled by vaccess(). 4834 */ 4835 if (error == 0) { 4836 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 4837 if (accmode != 0) { 4838 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 4839 zp->z_gid, accmode, ap->a_cred); 4840 } 4841 } 4842 4843 /* 4844 * For VEXEC, ensure that at least one execute bit is set for 4845 * non-directories. 4846 */ 4847 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 4848 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 4849 error = EACCES; 4850 } 4851 4852 return (error); 4853 } 4854 4855 #ifndef _SYS_SYSPROTO_H_ 4856 struct vop_lookup_args { 4857 struct vnode *a_dvp; 4858 struct vnode **a_vpp; 4859 struct componentname *a_cnp; 4860 }; 4861 #endif 4862 4863 #if __FreeBSD_version >= 1500040 4864 static int 4865 zfs_lookup_nameddir(struct vnode *dvp, struct componentname *cnp, 4866 struct vnode **vpp) 4867 { 4868 struct vnode *xvp; 4869 int error, flags; 4870 4871 *vpp = NULL; 4872 flags = LOOKUP_XATTR | LOOKUP_NAMED_ATTR; 4873 if ((cnp->cn_flags & CREATENAMED) != 0) 4874 flags |= CREATE_XATTR_DIR; 4875 error = zfs_lookup(dvp, NULL, &xvp, NULL, 0, cnp->cn_cred, flags, 4876 B_FALSE); 4877 if (error == 0) { 4878 if ((cnp->cn_flags & LOCKLEAF) != 0) 4879 error = vn_lock(xvp, cnp->cn_lkflags); 4880 if (error == 0) { 4881 vn_irflag_set_cond(xvp, VIRF_NAMEDDIR); 4882 *vpp = xvp; 4883 } else { 4884 vrele(xvp); 4885 } 4886 } 4887 return (error); 4888 } 4889 4890 static ssize_t 4891 zfs_readdir_named(struct vnode *vp, char *buf, ssize_t blen, off_t *offp, 4892 int *eofflagp, struct ucred *cred, struct thread *td) 4893 { 4894 struct uio io; 4895 struct iovec iv; 4896 zfs_uio_t uio; 4897 int error; 4898 4899 io.uio_offset = *offp; 4900 io.uio_segflg = UIO_SYSSPACE; 4901 io.uio_rw = UIO_READ; 4902 io.uio_td = td; 4903 iv.iov_base = buf; 4904 iv.iov_len = blen; 4905 io.uio_iov = &iv; 4906 io.uio_iovcnt = 1; 4907 io.uio_resid = blen; 4908 zfs_uio_init(&uio, &io); 4909 error = zfs_readdir(vp, &uio, cred, eofflagp, NULL, NULL); 4910 if (error != 0) 4911 return (-1); 4912 *offp = io.uio_offset; 4913 return (blen - io.uio_resid); 4914 } 4915 4916 static bool 4917 zfs_has_namedattr(struct vnode *vp, struct ucred *cred) 4918 { 4919 struct componentname cn; 4920 struct vnode *xvp; 4921 struct dirent *dp; 4922 off_t offs; 4923 ssize_t rsize; 4924 char *buf, *cp, *endcp; 4925 int eofflag, error; 4926 bool ret; 4927 4928 MNT_ILOCK(vp->v_mount); 4929 if ((vp->v_mount->mnt_flag & MNT_NAMEDATTR) == 0) { 4930 MNT_IUNLOCK(vp->v_mount); 4931 return (false); 4932 } 4933 MNT_IUNLOCK(vp->v_mount); 4934 4935 /* Now see if a named attribute directory exists. */ 4936 cn.cn_flags = LOCKLEAF; 4937 cn.cn_lkflags = LK_SHARED; 4938 cn.cn_cred = cred; 4939 error = zfs_lookup_nameddir(vp, &cn, &xvp); 4940 if (error != 0) 4941 return (false); 4942 4943 /* It exists, so see if there is any entry other than "." and "..". */ 4944 buf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK); 4945 ret = false; 4946 offs = 0; 4947 do { 4948 rsize = zfs_readdir_named(xvp, buf, DEV_BSIZE, &offs, &eofflag, 4949 cred, curthread); 4950 if (rsize <= 0) 4951 break; 4952 cp = buf; 4953 endcp = &buf[rsize]; 4954 while (cp < endcp) { 4955 dp = (struct dirent *)cp; 4956 if (dp->d_fileno != 0 && (dp->d_type == DT_REG || 4957 dp->d_type == DT_UNKNOWN) && 4958 !ZFS_XA_NS_PREFIX_FORBIDDEN(dp->d_name) && 4959 ((dp->d_namlen == 1 && dp->d_name[0] != '.') || 4960 (dp->d_namlen == 2 && (dp->d_name[0] != '.' || 4961 dp->d_name[1] != '.')) || dp->d_namlen > 2)) { 4962 ret = true; 4963 break; 4964 } 4965 cp += dp->d_reclen; 4966 } 4967 } while (!ret && rsize > 0 && eofflag == 0); 4968 vput(xvp); 4969 free(buf, M_TEMP); 4970 return (ret); 4971 } 4972 4973 static int 4974 zfs_freebsd_lookup(struct vop_lookup_args *ap, boolean_t cached) 4975 { 4976 struct componentname *cnp = ap->a_cnp; 4977 char nm[NAME_MAX + 1]; 4978 int error; 4979 struct vnode **vpp = ap->a_vpp, *dvp = ap->a_dvp, *xvp; 4980 bool is_nameddir, needs_nameddir, opennamed = false; 4981 4982 /* 4983 * These variables are used to handle the named attribute cases: 4984 * opennamed - Is true when this is a call from open with O_NAMEDATTR 4985 * specified and it is the last component. 4986 * is_nameddir - Is true when the directory is a named attribute dir. 4987 * needs_nameddir - Is set when the lookup needs to look for/create 4988 * a named attribute directory. It is only set when is_nameddir 4989 * is_nameddir is false and opennamed is true. 4990 * xvp - Is the directory that the lookup needs to be done in. 4991 * Usually dvp, unless needs_nameddir is true where it is the 4992 * result of the first non-named directory lookup. 4993 * Note that name caching must be disabled for named attribute 4994 * handling. 4995 */ 4996 needs_nameddir = false; 4997 xvp = dvp; 4998 opennamed = (cnp->cn_flags & (OPENNAMED | ISLASTCN)) == 4999 (OPENNAMED | ISLASTCN); 5000 is_nameddir = (vn_irflag_read(dvp) & VIRF_NAMEDDIR) != 0; 5001 if (is_nameddir && (cnp->cn_flags & ISLASTCN) == 0) 5002 return (ENOATTR); 5003 if (opennamed && !is_nameddir && (cnp->cn_flags & ISDOTDOT) != 0) 5004 return (ENOATTR); 5005 if (opennamed || is_nameddir) 5006 cnp->cn_flags &= ~MAKEENTRY; 5007 if (opennamed && !is_nameddir) 5008 needs_nameddir = true; 5009 ASSERT3U(cnp->cn_namelen, <, sizeof (nm)); 5010 error = 0; 5011 *vpp = NULL; 5012 if (needs_nameddir) { 5013 if (VOP_ISLOCKED(dvp) != LK_EXCLUSIVE) 5014 vn_lock(dvp, LK_UPGRADE | LK_RETRY); 5015 error = zfs_lookup_nameddir(dvp, cnp, &xvp); 5016 if (error == 0) 5017 is_nameddir = true; 5018 } 5019 if (error == 0) { 5020 if (!needs_nameddir || cnp->cn_namelen != 1 || 5021 *cnp->cn_nameptr != '.') { 5022 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, 5023 sizeof (nm))); 5024 error = zfs_lookup(xvp, nm, vpp, cnp, cnp->cn_nameiop, 5025 cnp->cn_cred, 0, cached); 5026 if (is_nameddir && error == 0 && 5027 (cnp->cn_namelen != 1 || *cnp->cn_nameptr != '.') && 5028 (cnp->cn_flags & ISDOTDOT) == 0) { 5029 if ((*vpp)->v_type == VDIR) 5030 vn_irflag_set_cond(*vpp, VIRF_NAMEDDIR); 5031 else 5032 vn_irflag_set_cond(*vpp, 5033 VIRF_NAMEDATTR); 5034 } 5035 if (needs_nameddir && xvp != *vpp) 5036 vput(xvp); 5037 } else { 5038 /* 5039 * Lookup of "." when a named attribute dir is needed. 5040 */ 5041 *vpp = xvp; 5042 } 5043 } 5044 return (error); 5045 } 5046 #else 5047 static int 5048 zfs_freebsd_lookup(struct vop_lookup_args *ap, boolean_t cached) 5049 { 5050 struct componentname *cnp = ap->a_cnp; 5051 char nm[NAME_MAX + 1]; 5052 5053 ASSERT3U(cnp->cn_namelen, <, sizeof (nm)); 5054 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof (nm))); 5055 5056 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 5057 cnp->cn_cred, 0, cached)); 5058 } 5059 #endif 5060 5061 static int 5062 zfs_freebsd_cachedlookup(struct vop_cachedlookup_args *ap) 5063 { 5064 5065 return (zfs_freebsd_lookup((struct vop_lookup_args *)ap, B_TRUE)); 5066 } 5067 5068 #ifndef _SYS_SYSPROTO_H_ 5069 struct vop_lookup_args { 5070 struct vnode *a_dvp; 5071 struct vnode **a_vpp; 5072 struct componentname *a_cnp; 5073 }; 5074 #endif 5075 5076 static int 5077 zfs_cache_lookup(struct vop_lookup_args *ap) 5078 { 5079 zfsvfs_t *zfsvfs; 5080 5081 zfsvfs = ap->a_dvp->v_mount->mnt_data; 5082 #if __FreeBSD_version >= 1500040 5083 if (zfsvfs->z_use_namecache && (ap->a_cnp->cn_flags & OPENNAMED) == 0) 5084 #else 5085 if (zfsvfs->z_use_namecache) 5086 #endif 5087 return (vfs_cache_lookup(ap)); 5088 else 5089 return (zfs_freebsd_lookup(ap, B_FALSE)); 5090 } 5091 5092 #ifndef _SYS_SYSPROTO_H_ 5093 struct vop_create_args { 5094 struct vnode *a_dvp; 5095 struct vnode **a_vpp; 5096 struct componentname *a_cnp; 5097 struct vattr *a_vap; 5098 }; 5099 #endif 5100 5101 static int 5102 zfs_freebsd_create(struct vop_create_args *ap) 5103 { 5104 zfsvfs_t *zfsvfs; 5105 struct componentname *cnp = ap->a_cnp; 5106 vattr_t *vap = ap->a_vap; 5107 znode_t *zp = NULL; 5108 int rc, mode; 5109 struct vnode *dvp = ap->a_dvp; 5110 #if __FreeBSD_version >= 1500040 5111 struct vnode *xvp; 5112 bool is_nameddir; 5113 #endif 5114 5115 #if __FreeBSD_version < 1400068 5116 ASSERT(cnp->cn_flags & SAVENAME); 5117 #endif 5118 5119 vattr_init_mask(vap); 5120 mode = vap->va_mode & ALLPERMS; 5121 zfsvfs = ap->a_dvp->v_mount->mnt_data; 5122 *ap->a_vpp = NULL; 5123 5124 rc = 0; 5125 #if __FreeBSD_version >= 1500040 5126 xvp = NULL; 5127 is_nameddir = (vn_irflag_read(dvp) & VIRF_NAMEDDIR) != 0; 5128 if (!is_nameddir && (cnp->cn_flags & OPENNAMED) != 0) { 5129 /* Needs a named attribute directory. */ 5130 rc = zfs_lookup_nameddir(dvp, cnp, &xvp); 5131 if (rc == 0) { 5132 dvp = xvp; 5133 is_nameddir = true; 5134 } 5135 } 5136 if (is_nameddir && rc == 0) 5137 rc = zfs_check_attrname(cnp->cn_nameptr); 5138 #endif 5139 5140 if (rc == 0) 5141 rc = zfs_create(VTOZ(dvp), cnp->cn_nameptr, vap, 0, mode, 5142 &zp, cnp->cn_cred, 0 /* flag */, NULL /* vsecattr */, NULL); 5143 #if __FreeBSD_version >= 1500040 5144 if (xvp != NULL) 5145 vput(xvp); 5146 #endif 5147 if (rc == 0) { 5148 *ap->a_vpp = ZTOV(zp); 5149 #if __FreeBSD_version >= 1500040 5150 if (is_nameddir) 5151 vn_irflag_set_cond(*ap->a_vpp, VIRF_NAMEDATTR); 5152 #endif 5153 } 5154 if (zfsvfs->z_use_namecache && 5155 rc == 0 && (cnp->cn_flags & MAKEENTRY) != 0) 5156 cache_enter(ap->a_dvp, *ap->a_vpp, cnp); 5157 5158 return (rc); 5159 } 5160 5161 #ifndef _SYS_SYSPROTO_H_ 5162 struct vop_remove_args { 5163 struct vnode *a_dvp; 5164 struct vnode *a_vp; 5165 struct componentname *a_cnp; 5166 }; 5167 #endif 5168 5169 static int 5170 zfs_freebsd_remove(struct vop_remove_args *ap) 5171 { 5172 int error = 0; 5173 5174 #if __FreeBSD_version < 1400068 5175 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 5176 #endif 5177 5178 #if __FreeBSD_version >= 1500040 5179 if ((vn_irflag_read(ap->a_dvp) & VIRF_NAMEDDIR) != 0) 5180 error = zfs_check_attrname(ap->a_cnp->cn_nameptr); 5181 #endif 5182 5183 if (error == 0) 5184 error = zfs_remove_(ap->a_dvp, ap->a_vp, ap->a_cnp->cn_nameptr, 5185 ap->a_cnp->cn_cred); 5186 return (error); 5187 } 5188 5189 #ifndef _SYS_SYSPROTO_H_ 5190 struct vop_mkdir_args { 5191 struct vnode *a_dvp; 5192 struct vnode **a_vpp; 5193 struct componentname *a_cnp; 5194 struct vattr *a_vap; 5195 }; 5196 #endif 5197 5198 static int 5199 zfs_freebsd_mkdir(struct vop_mkdir_args *ap) 5200 { 5201 vattr_t *vap = ap->a_vap; 5202 znode_t *zp = NULL; 5203 int rc; 5204 5205 #if __FreeBSD_version < 1400068 5206 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 5207 #endif 5208 5209 vattr_init_mask(vap); 5210 *ap->a_vpp = NULL; 5211 5212 rc = zfs_mkdir(VTOZ(ap->a_dvp), ap->a_cnp->cn_nameptr, vap, &zp, 5213 ap->a_cnp->cn_cred, 0, NULL, NULL); 5214 5215 if (rc == 0) 5216 *ap->a_vpp = ZTOV(zp); 5217 return (rc); 5218 } 5219 5220 #ifndef _SYS_SYSPROTO_H_ 5221 struct vop_rmdir_args { 5222 struct vnode *a_dvp; 5223 struct vnode *a_vp; 5224 struct componentname *a_cnp; 5225 }; 5226 #endif 5227 5228 static int 5229 zfs_freebsd_rmdir(struct vop_rmdir_args *ap) 5230 { 5231 struct componentname *cnp = ap->a_cnp; 5232 5233 #if __FreeBSD_version < 1400068 5234 ASSERT(cnp->cn_flags & SAVENAME); 5235 #endif 5236 5237 return (zfs_rmdir_(ap->a_dvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred)); 5238 } 5239 5240 #ifndef _SYS_SYSPROTO_H_ 5241 struct vop_readdir_args { 5242 struct vnode *a_vp; 5243 struct uio *a_uio; 5244 struct ucred *a_cred; 5245 int *a_eofflag; 5246 int *a_ncookies; 5247 cookie_t **a_cookies; 5248 }; 5249 #endif 5250 5251 static int 5252 zfs_freebsd_readdir(struct vop_readdir_args *ap) 5253 { 5254 zfs_uio_t uio; 5255 zfs_uio_init(&uio, ap->a_uio); 5256 return (zfs_readdir(ap->a_vp, &uio, ap->a_cred, ap->a_eofflag, 5257 ap->a_ncookies, ap->a_cookies)); 5258 } 5259 5260 #ifndef _SYS_SYSPROTO_H_ 5261 struct vop_fsync_args { 5262 struct vnode *a_vp; 5263 int a_waitfor; 5264 struct thread *a_td; 5265 }; 5266 #endif 5267 5268 static int 5269 zfs_freebsd_fsync(struct vop_fsync_args *ap) 5270 { 5271 vnode_t *vp = ap->a_vp; 5272 int err = 0; 5273 5274 /* 5275 * Push any dirty mmap()'d data out to the DMU and ZIL, ready for 5276 * zil_commit() to be called in zfs_fsync(). 5277 */ 5278 if (vp->v_object != NULL && vm_object_mightbedirty(vp->v_object)) { 5279 zfs_vmobject_wlock(vp->v_object); 5280 if (!vm_object_page_clean(vp->v_object, 0, 0, 0)) 5281 err = SET_ERROR(EIO); 5282 zfs_vmobject_wunlock(vp->v_object); 5283 if (err) { 5284 /* 5285 * Unclear what state things are in. zfs_putpages() 5286 * will ensure the pages remain dirty if they haven't 5287 * been written down to the DMU, but because there may 5288 * be nothing logged, we can't assume that zfs_sync() 5289 * -> zil_commit() will give us a useful error. It's 5290 * safest if we just error out here. 5291 */ 5292 return (err); 5293 } 5294 } 5295 5296 return (zfs_fsync(VTOZ(vp), 0, ap->a_td->td_ucred)); 5297 } 5298 5299 #ifndef _SYS_SYSPROTO_H_ 5300 struct vop_getattr_args { 5301 struct vnode *a_vp; 5302 struct vattr *a_vap; 5303 struct ucred *a_cred; 5304 }; 5305 #endif 5306 5307 static int 5308 zfs_freebsd_getattr(struct vop_getattr_args *ap) 5309 { 5310 vattr_t *vap = ap->a_vap; 5311 xvattr_t xvap; 5312 ulong_t fflags = 0; 5313 int error; 5314 5315 xva_init(&xvap); 5316 xvap.xva_vattr = *vap; 5317 xvap.xva_vattr.va_mask |= AT_XVATTR; 5318 5319 /* Convert chflags into ZFS-type flags. */ 5320 /* XXX: what about SF_SETTABLE?. */ 5321 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 5322 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 5323 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 5324 XVA_SET_REQ(&xvap, XAT_NODUMP); 5325 XVA_SET_REQ(&xvap, XAT_READONLY); 5326 XVA_SET_REQ(&xvap, XAT_ARCHIVE); 5327 XVA_SET_REQ(&xvap, XAT_SYSTEM); 5328 XVA_SET_REQ(&xvap, XAT_HIDDEN); 5329 XVA_SET_REQ(&xvap, XAT_REPARSE); 5330 XVA_SET_REQ(&xvap, XAT_OFFLINE); 5331 XVA_SET_REQ(&xvap, XAT_SPARSE); 5332 5333 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred); 5334 if (error != 0) 5335 return (error); 5336 5337 /* Convert ZFS xattr into chflags. */ 5338 #define FLAG_CHECK(fflag, xflag, xfield) do { \ 5339 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 5340 fflags |= (fflag); \ 5341 } while (0) 5342 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 5343 xvap.xva_xoptattrs.xoa_immutable); 5344 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 5345 xvap.xva_xoptattrs.xoa_appendonly); 5346 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 5347 xvap.xva_xoptattrs.xoa_nounlink); 5348 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE, 5349 xvap.xva_xoptattrs.xoa_archive); 5350 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 5351 xvap.xva_xoptattrs.xoa_nodump); 5352 FLAG_CHECK(UF_READONLY, XAT_READONLY, 5353 xvap.xva_xoptattrs.xoa_readonly); 5354 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM, 5355 xvap.xva_xoptattrs.xoa_system); 5356 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN, 5357 xvap.xva_xoptattrs.xoa_hidden); 5358 FLAG_CHECK(UF_REPARSE, XAT_REPARSE, 5359 xvap.xva_xoptattrs.xoa_reparse); 5360 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE, 5361 xvap.xva_xoptattrs.xoa_offline); 5362 FLAG_CHECK(UF_SPARSE, XAT_SPARSE, 5363 xvap.xva_xoptattrs.xoa_sparse); 5364 5365 #undef FLAG_CHECK 5366 *vap = xvap.xva_vattr; 5367 vap->va_flags = fflags; 5368 5369 #if __FreeBSD_version >= 1500040 5370 if ((vn_irflag_read(ap->a_vp) & (VIRF_NAMEDDIR | VIRF_NAMEDATTR)) != 0) 5371 vap->va_bsdflags |= SFBSD_NAMEDATTR; 5372 #endif 5373 return (0); 5374 } 5375 5376 #ifndef _SYS_SYSPROTO_H_ 5377 struct vop_setattr_args { 5378 struct vnode *a_vp; 5379 struct vattr *a_vap; 5380 struct ucred *a_cred; 5381 }; 5382 #endif 5383 5384 static int 5385 zfs_freebsd_setattr(struct vop_setattr_args *ap) 5386 { 5387 vnode_t *vp = ap->a_vp; 5388 vattr_t *vap = ap->a_vap; 5389 cred_t *cred = ap->a_cred; 5390 xvattr_t xvap; 5391 ulong_t fflags; 5392 uint64_t zflags; 5393 5394 vattr_init_mask(vap); 5395 vap->va_mask &= ~AT_NOSET; 5396 5397 xva_init(&xvap); 5398 xvap.xva_vattr = *vap; 5399 5400 zflags = VTOZ(vp)->z_pflags; 5401 5402 if (vap->va_flags != VNOVAL) { 5403 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 5404 int error; 5405 5406 if (zfsvfs->z_use_fuids == B_FALSE) 5407 return (EOPNOTSUPP); 5408 5409 fflags = vap->va_flags; 5410 /* 5411 * XXX KDM 5412 * We need to figure out whether it makes sense to allow 5413 * UF_REPARSE through, since we don't really have other 5414 * facilities to handle reparse points and zfs_setattr() 5415 * doesn't currently allow setting that attribute anyway. 5416 */ 5417 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE| 5418 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE| 5419 UF_OFFLINE|UF_SPARSE)) != 0) 5420 return (EOPNOTSUPP); 5421 /* 5422 * Unprivileged processes are not permitted to unset system 5423 * flags, or modify flags if any system flags are set. 5424 * Privileged non-jail processes may not modify system flags 5425 * if securelevel > 0 and any existing system flags are set. 5426 * Privileged jail processes behave like privileged non-jail 5427 * processes if the PR_ALLOW_CHFLAGS permission bit is set; 5428 * otherwise, they behave like unprivileged processes. 5429 */ 5430 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 5431 priv_check_cred(cred, PRIV_VFS_SYSFLAGS) == 0) { 5432 if (zflags & 5433 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 5434 error = securelevel_gt(cred, 0); 5435 if (error != 0) 5436 return (error); 5437 } 5438 } else { 5439 /* 5440 * Callers may only modify the file flags on 5441 * objects they have VADMIN rights for. 5442 */ 5443 if ((error = VOP_ACCESS(vp, VADMIN, cred, 5444 curthread)) != 0) 5445 return (error); 5446 if (zflags & 5447 (ZFS_IMMUTABLE | ZFS_APPENDONLY | 5448 ZFS_NOUNLINK)) { 5449 return (EPERM); 5450 } 5451 if (fflags & 5452 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 5453 return (EPERM); 5454 } 5455 } 5456 5457 #define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 5458 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 5459 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 5460 XVA_SET_REQ(&xvap, (xflag)); \ 5461 (xfield) = ((fflags & (fflag)) != 0); \ 5462 } \ 5463 } while (0) 5464 /* Convert chflags into ZFS-type flags. */ 5465 /* XXX: what about SF_SETTABLE?. */ 5466 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 5467 xvap.xva_xoptattrs.xoa_immutable); 5468 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 5469 xvap.xva_xoptattrs.xoa_appendonly); 5470 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 5471 xvap.xva_xoptattrs.xoa_nounlink); 5472 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE, 5473 xvap.xva_xoptattrs.xoa_archive); 5474 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 5475 xvap.xva_xoptattrs.xoa_nodump); 5476 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY, 5477 xvap.xva_xoptattrs.xoa_readonly); 5478 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM, 5479 xvap.xva_xoptattrs.xoa_system); 5480 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN, 5481 xvap.xva_xoptattrs.xoa_hidden); 5482 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE, 5483 xvap.xva_xoptattrs.xoa_reparse); 5484 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE, 5485 xvap.xva_xoptattrs.xoa_offline); 5486 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE, 5487 xvap.xva_xoptattrs.xoa_sparse); 5488 #undef FLAG_CHANGE 5489 } 5490 if (vap->va_birthtime.tv_sec != VNOVAL) { 5491 xvap.xva_vattr.va_mask |= AT_XVATTR; 5492 XVA_SET_REQ(&xvap, XAT_CREATETIME); 5493 } 5494 return (zfs_setattr(VTOZ(vp), (vattr_t *)&xvap, 0, cred, NULL)); 5495 } 5496 5497 #ifndef _SYS_SYSPROTO_H_ 5498 struct vop_rename_args { 5499 struct vnode *a_fdvp; 5500 struct vnode *a_fvp; 5501 struct componentname *a_fcnp; 5502 struct vnode *a_tdvp; 5503 struct vnode *a_tvp; 5504 struct componentname *a_tcnp; 5505 }; 5506 #endif 5507 5508 static int 5509 zfs_freebsd_rename(struct vop_rename_args *ap) 5510 { 5511 vnode_t *fdvp = ap->a_fdvp; 5512 vnode_t *fvp = ap->a_fvp; 5513 vnode_t *tdvp = ap->a_tdvp; 5514 vnode_t *tvp = ap->a_tvp; 5515 int error = 0; 5516 5517 #if __FreeBSD_version < 1400068 5518 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 5519 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 5520 #endif 5521 5522 #if __FreeBSD_version >= 1500040 5523 if ((vn_irflag_read(fdvp) & VIRF_NAMEDDIR) != 0) { 5524 error = zfs_check_attrname(ap->a_fcnp->cn_nameptr); 5525 if (error == 0) 5526 error = zfs_check_attrname(ap->a_tcnp->cn_nameptr); 5527 } 5528 #endif 5529 5530 if (error == 0) 5531 error = zfs_do_rename(fdvp, &fvp, ap->a_fcnp, tdvp, &tvp, 5532 ap->a_tcnp, ap->a_fcnp->cn_cred); 5533 5534 vrele(fdvp); 5535 vrele(fvp); 5536 vrele(tdvp); 5537 if (tvp != NULL) 5538 vrele(tvp); 5539 5540 return (error); 5541 } 5542 5543 #ifndef _SYS_SYSPROTO_H_ 5544 struct vop_symlink_args { 5545 struct vnode *a_dvp; 5546 struct vnode **a_vpp; 5547 struct componentname *a_cnp; 5548 struct vattr *a_vap; 5549 char *a_target; 5550 }; 5551 #endif 5552 5553 static int 5554 zfs_freebsd_symlink(struct vop_symlink_args *ap) 5555 { 5556 struct componentname *cnp = ap->a_cnp; 5557 vattr_t *vap = ap->a_vap; 5558 znode_t *zp = NULL; 5559 char *symlink; 5560 size_t symlink_len; 5561 int rc; 5562 5563 #if __FreeBSD_version < 1400068 5564 ASSERT(cnp->cn_flags & SAVENAME); 5565 #endif 5566 5567 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 5568 vattr_init_mask(vap); 5569 *ap->a_vpp = NULL; 5570 5571 rc = zfs_symlink(VTOZ(ap->a_dvp), cnp->cn_nameptr, vap, 5572 ap->a_target, &zp, cnp->cn_cred, 0 /* flags */, NULL); 5573 if (rc == 0) { 5574 *ap->a_vpp = ZTOV(zp); 5575 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__); 5576 MPASS(zp->z_cached_symlink == NULL); 5577 symlink_len = strlen(ap->a_target); 5578 symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK); 5579 if (symlink != NULL) { 5580 memcpy(symlink, ap->a_target, symlink_len); 5581 symlink[symlink_len] = '\0'; 5582 atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink, 5583 (uintptr_t)symlink); 5584 } 5585 } 5586 return (rc); 5587 } 5588 5589 #ifndef _SYS_SYSPROTO_H_ 5590 struct vop_readlink_args { 5591 struct vnode *a_vp; 5592 struct uio *a_uio; 5593 struct ucred *a_cred; 5594 }; 5595 #endif 5596 5597 static int 5598 zfs_freebsd_readlink(struct vop_readlink_args *ap) 5599 { 5600 zfs_uio_t uio; 5601 int error; 5602 znode_t *zp = VTOZ(ap->a_vp); 5603 char *symlink, *base; 5604 size_t symlink_len; 5605 bool trycache; 5606 5607 zfs_uio_init(&uio, ap->a_uio); 5608 trycache = false; 5609 if (zfs_uio_segflg(&uio) == UIO_SYSSPACE && 5610 zfs_uio_iovcnt(&uio) == 1) { 5611 base = zfs_uio_iovbase(&uio, 0); 5612 symlink_len = zfs_uio_iovlen(&uio, 0); 5613 trycache = true; 5614 } 5615 error = zfs_readlink(ap->a_vp, &uio, ap->a_cred, NULL); 5616 if (atomic_load_ptr(&zp->z_cached_symlink) != NULL || 5617 error != 0 || !trycache) { 5618 return (error); 5619 } 5620 symlink_len -= zfs_uio_resid(&uio); 5621 symlink = cache_symlink_alloc(symlink_len + 1, M_WAITOK); 5622 if (symlink != NULL) { 5623 memcpy(symlink, base, symlink_len); 5624 symlink[symlink_len] = '\0'; 5625 if (!atomic_cmpset_rel_ptr((uintptr_t *)&zp->z_cached_symlink, 5626 (uintptr_t)NULL, (uintptr_t)symlink)) { 5627 cache_symlink_free(symlink, symlink_len + 1); 5628 } 5629 } 5630 return (error); 5631 } 5632 5633 #ifndef _SYS_SYSPROTO_H_ 5634 struct vop_link_args { 5635 struct vnode *a_tdvp; 5636 struct vnode *a_vp; 5637 struct componentname *a_cnp; 5638 }; 5639 #endif 5640 5641 static int 5642 zfs_freebsd_link(struct vop_link_args *ap) 5643 { 5644 struct componentname *cnp = ap->a_cnp; 5645 vnode_t *vp = ap->a_vp; 5646 vnode_t *tdvp = ap->a_tdvp; 5647 5648 if (tdvp->v_mount != vp->v_mount) 5649 return (EXDEV); 5650 5651 #if __FreeBSD_version < 1400068 5652 ASSERT(cnp->cn_flags & SAVENAME); 5653 #endif 5654 5655 return (zfs_link(VTOZ(tdvp), VTOZ(vp), 5656 cnp->cn_nameptr, cnp->cn_cred, 0)); 5657 } 5658 5659 #ifndef _SYS_SYSPROTO_H_ 5660 struct vop_inactive_args { 5661 struct vnode *a_vp; 5662 struct thread *a_td; 5663 }; 5664 #endif 5665 5666 static int 5667 zfs_freebsd_inactive(struct vop_inactive_args *ap) 5668 { 5669 vnode_t *vp = ap->a_vp; 5670 5671 zfs_inactive(vp, curthread->td_ucred, NULL); 5672 return (0); 5673 } 5674 5675 #ifndef _SYS_SYSPROTO_H_ 5676 struct vop_need_inactive_args { 5677 struct vnode *a_vp; 5678 struct thread *a_td; 5679 }; 5680 #endif 5681 5682 static int 5683 zfs_freebsd_need_inactive(struct vop_need_inactive_args *ap) 5684 { 5685 vnode_t *vp = ap->a_vp; 5686 znode_t *zp = VTOZ(vp); 5687 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5688 int need; 5689 5690 if (vn_need_pageq_flush(vp)) 5691 return (1); 5692 5693 if (!ZFS_TEARDOWN_INACTIVE_TRY_ENTER_READ(zfsvfs)) 5694 return (1); 5695 need = (zp->z_sa_hdl == NULL || zp->z_unlinked || zp->z_atime_dirty); 5696 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 5697 5698 return (need); 5699 } 5700 5701 #ifndef _SYS_SYSPROTO_H_ 5702 struct vop_reclaim_args { 5703 struct vnode *a_vp; 5704 struct thread *a_td; 5705 }; 5706 #endif 5707 5708 static int 5709 zfs_freebsd_reclaim(struct vop_reclaim_args *ap) 5710 { 5711 vnode_t *vp = ap->a_vp; 5712 znode_t *zp = VTOZ(vp); 5713 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5714 5715 ASSERT3P(zp, !=, NULL); 5716 5717 /* 5718 * z_teardown_inactive_lock protects from a race with 5719 * zfs_znode_dmu_fini in zfsvfs_teardown during 5720 * force unmount. 5721 */ 5722 ZFS_TEARDOWN_INACTIVE_ENTER_READ(zfsvfs); 5723 if (zp->z_sa_hdl == NULL) 5724 zfs_znode_free(zp); 5725 else 5726 zfs_zinactive(zp); 5727 ZFS_TEARDOWN_INACTIVE_EXIT_READ(zfsvfs); 5728 5729 vp->v_data = NULL; 5730 return (0); 5731 } 5732 5733 #ifndef _SYS_SYSPROTO_H_ 5734 struct vop_fid_args { 5735 struct vnode *a_vp; 5736 struct fid *a_fid; 5737 }; 5738 #endif 5739 5740 static int 5741 zfs_freebsd_fid(struct vop_fid_args *ap) 5742 { 5743 5744 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 5745 } 5746 5747 5748 #ifndef _SYS_SYSPROTO_H_ 5749 struct vop_pathconf_args { 5750 struct vnode *a_vp; 5751 int a_name; 5752 register_t *a_retval; 5753 } *ap; 5754 #endif 5755 5756 static int 5757 zfs_freebsd_pathconf(struct vop_pathconf_args *ap) 5758 { 5759 ulong_t val; 5760 int error; 5761 #if defined(_PC_CLONE_BLKSIZE) || defined(_PC_CASE_INSENSITIVE) 5762 zfsvfs_t *zfsvfs; 5763 #endif 5764 5765 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, 5766 curthread->td_ucred, NULL); 5767 if (error == 0) { 5768 *ap->a_retval = val; 5769 return (error); 5770 } 5771 if (error != EOPNOTSUPP) 5772 return (error); 5773 5774 switch (ap->a_name) { 5775 case _PC_NAME_MAX: 5776 *ap->a_retval = NAME_MAX; 5777 return (0); 5778 #if __FreeBSD_version >= 1400032 5779 case _PC_DEALLOC_PRESENT: 5780 *ap->a_retval = 1; 5781 return (0); 5782 #endif 5783 case _PC_PIPE_BUF: 5784 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO) { 5785 *ap->a_retval = PIPE_BUF; 5786 return (0); 5787 } 5788 return (EINVAL); 5789 #if __FreeBSD_version >= 1500040 5790 case _PC_NAMEDATTR_ENABLED: 5791 MNT_ILOCK(ap->a_vp->v_mount); 5792 if ((ap->a_vp->v_mount->mnt_flag & MNT_NAMEDATTR) != 0) 5793 *ap->a_retval = 1; 5794 else 5795 *ap->a_retval = 0; 5796 MNT_IUNLOCK(ap->a_vp->v_mount); 5797 return (0); 5798 case _PC_HAS_NAMEDATTR: 5799 if (zfs_has_namedattr(ap->a_vp, curthread->td_ucred)) 5800 *ap->a_retval = 1; 5801 else 5802 *ap->a_retval = 0; 5803 return (0); 5804 #endif 5805 #ifdef _PC_HAS_HIDDENSYSTEM 5806 case _PC_HAS_HIDDENSYSTEM: 5807 *ap->a_retval = 1; 5808 return (0); 5809 #endif 5810 #ifdef _PC_CLONE_BLKSIZE 5811 case _PC_CLONE_BLKSIZE: 5812 zfsvfs = (zfsvfs_t *)ap->a_vp->v_mount->mnt_data; 5813 if (zfs_bclone_enabled && 5814 spa_feature_is_enabled(dmu_objset_spa(zfsvfs->z_os), 5815 SPA_FEATURE_BLOCK_CLONING)) 5816 *ap->a_retval = dsl_dataset_feature_is_active( 5817 zfsvfs->z_os->os_dsl_dataset, 5818 SPA_FEATURE_LARGE_BLOCKS) ? 5819 SPA_MAXBLOCKSIZE : 5820 SPA_OLD_MAXBLOCKSIZE; 5821 else 5822 *ap->a_retval = 0; 5823 return (0); 5824 #endif 5825 #ifdef _PC_CASE_INSENSITIVE 5826 case _PC_CASE_INSENSITIVE: 5827 zfsvfs = (zfsvfs_t *)ap->a_vp->v_mount->mnt_data; 5828 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) 5829 *ap->a_retval = 1; 5830 else 5831 *ap->a_retval = 0; 5832 return (0); 5833 #endif 5834 default: 5835 return (vop_stdpathconf(ap)); 5836 } 5837 } 5838 5839 int zfs_xattr_compat = 1; 5840 5841 static int 5842 zfs_check_attrname(const char *name) 5843 { 5844 /* We don't allow '/' character in attribute name. */ 5845 if (strchr(name, '/') != NULL) 5846 return (SET_ERROR(EINVAL)); 5847 /* We don't allow attribute names that start with a namespace prefix. */ 5848 if (ZFS_XA_NS_PREFIX_FORBIDDEN(name)) 5849 return (SET_ERROR(EINVAL)); 5850 return (0); 5851 } 5852 5853 /* 5854 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 5855 * extended attribute name: 5856 * 5857 * NAMESPACE XATTR_COMPAT PREFIX 5858 * system * freebsd:system: 5859 * user 1 (none, can be used to access ZFS 5860 * fsattr(5) attributes created on Solaris) 5861 * user 0 user. 5862 */ 5863 static int 5864 zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 5865 size_t size, boolean_t compat) 5866 { 5867 const char *namespace, *prefix, *suffix; 5868 5869 memset(attrname, 0, size); 5870 5871 switch (attrnamespace) { 5872 case EXTATTR_NAMESPACE_USER: 5873 if (compat) { 5874 /* 5875 * This is the default namespace by which we can access 5876 * all attributes created on Solaris. 5877 */ 5878 prefix = namespace = suffix = ""; 5879 } else { 5880 /* 5881 * This is compatible with the user namespace encoding 5882 * on Linux prior to xattr_compat, but nothing 5883 * else. 5884 */ 5885 prefix = ""; 5886 namespace = "user"; 5887 suffix = "."; 5888 } 5889 break; 5890 case EXTATTR_NAMESPACE_SYSTEM: 5891 prefix = "freebsd:"; 5892 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 5893 suffix = ":"; 5894 break; 5895 case EXTATTR_NAMESPACE_EMPTY: 5896 default: 5897 return (SET_ERROR(EINVAL)); 5898 } 5899 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 5900 name) >= size) { 5901 return (SET_ERROR(ENAMETOOLONG)); 5902 } 5903 return (0); 5904 } 5905 5906 static int 5907 zfs_ensure_xattr_cached(znode_t *zp) 5908 { 5909 int error = 0; 5910 5911 ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock)); 5912 5913 if (zp->z_xattr_cached != NULL) 5914 return (0); 5915 5916 if (rw_write_held(&zp->z_xattr_lock)) 5917 return (zfs_sa_get_xattr(zp)); 5918 5919 if (!rw_tryupgrade(&zp->z_xattr_lock)) { 5920 rw_exit(&zp->z_xattr_lock); 5921 rw_enter(&zp->z_xattr_lock, RW_WRITER); 5922 } 5923 if (zp->z_xattr_cached == NULL) 5924 error = zfs_sa_get_xattr(zp); 5925 rw_downgrade(&zp->z_xattr_lock); 5926 return (error); 5927 } 5928 5929 #ifndef _SYS_SYSPROTO_H_ 5930 struct vop_getextattr { 5931 IN struct vnode *a_vp; 5932 IN int a_attrnamespace; 5933 IN const char *a_name; 5934 INOUT struct uio *a_uio; 5935 OUT size_t *a_size; 5936 IN struct ucred *a_cred; 5937 IN struct thread *a_td; 5938 }; 5939 #endif 5940 5941 static int 5942 zfs_getextattr_dir(struct vop_getextattr_args *ap, const char *attrname) 5943 { 5944 struct thread *td = ap->a_td; 5945 struct nameidata nd; 5946 struct vattr va; 5947 vnode_t *xvp = NULL, *vp; 5948 int error, flags; 5949 5950 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, 5951 LOOKUP_XATTR, B_FALSE); 5952 if (error != 0) 5953 return (error); 5954 5955 flags = FREAD; 5956 #if __FreeBSD_version < 1400043 5957 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 5958 xvp, td); 5959 #else 5960 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp); 5961 #endif 5962 error = vn_open_cred(&nd, &flags, 0, VN_OPEN_INVFS, ap->a_cred, NULL); 5963 if (error != 0) 5964 return (SET_ERROR(error)); 5965 vp = nd.ni_vp; 5966 NDFREE_PNBUF(&nd); 5967 5968 if (ap->a_size != NULL) { 5969 error = VOP_GETATTR(vp, &va, ap->a_cred); 5970 if (error == 0) 5971 *ap->a_size = (size_t)va.va_size; 5972 } else if (ap->a_uio != NULL) 5973 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); 5974 5975 VOP_UNLOCK(vp); 5976 vn_close(vp, flags, ap->a_cred, td); 5977 return (error); 5978 } 5979 5980 static int 5981 zfs_getextattr_sa(struct vop_getextattr_args *ap, const char *attrname) 5982 { 5983 znode_t *zp = VTOZ(ap->a_vp); 5984 uchar_t *nv_value; 5985 uint_t nv_size; 5986 int error; 5987 5988 error = zfs_ensure_xattr_cached(zp); 5989 if (error != 0) 5990 return (error); 5991 5992 ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock)); 5993 ASSERT3P(zp->z_xattr_cached, !=, NULL); 5994 5995 error = nvlist_lookup_byte_array(zp->z_xattr_cached, attrname, 5996 &nv_value, &nv_size); 5997 if (error != 0) 5998 return (SET_ERROR(error)); 5999 6000 if (ap->a_size != NULL) 6001 *ap->a_size = nv_size; 6002 else if (ap->a_uio != NULL) 6003 error = uiomove(nv_value, nv_size, ap->a_uio); 6004 if (error != 0) 6005 return (SET_ERROR(error)); 6006 6007 return (0); 6008 } 6009 6010 static int 6011 zfs_getextattr_impl(struct vop_getextattr_args *ap, boolean_t compat) 6012 { 6013 znode_t *zp = VTOZ(ap->a_vp); 6014 zfsvfs_t *zfsvfs = ZTOZSB(zp); 6015 char attrname[EXTATTR_MAXNAMELEN+1]; 6016 int error; 6017 6018 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6019 sizeof (attrname), compat); 6020 if (error != 0) 6021 return (error); 6022 6023 error = ENOENT; 6024 if (zfsvfs->z_use_sa && zp->z_is_sa) 6025 error = zfs_getextattr_sa(ap, attrname); 6026 if (error == ENOENT) 6027 error = zfs_getextattr_dir(ap, attrname); 6028 return (error); 6029 } 6030 6031 /* 6032 * Vnode operation to retrieve a named extended attribute. 6033 */ 6034 static int 6035 zfs_getextattr(struct vop_getextattr_args *ap) 6036 { 6037 znode_t *zp = VTOZ(ap->a_vp); 6038 zfsvfs_t *zfsvfs = ZTOZSB(zp); 6039 int error; 6040 6041 /* 6042 * If the xattr property is off, refuse the request. 6043 */ 6044 if (!(zfsvfs->z_flags & ZSB_XATTR)) 6045 return (SET_ERROR(EOPNOTSUPP)); 6046 6047 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6048 ap->a_cred, ap->a_td, VREAD); 6049 if (error != 0) 6050 return (SET_ERROR(error)); 6051 6052 error = zfs_check_attrname(ap->a_name); 6053 if (error != 0) 6054 return (error); 6055 6056 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 6057 return (error); 6058 error = ENOENT; 6059 rw_enter(&zp->z_xattr_lock, RW_READER); 6060 6061 error = zfs_getextattr_impl(ap, zfs_xattr_compat); 6062 if ((error == ENOENT || error == ENOATTR) && 6063 ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { 6064 /* 6065 * Fall back to the alternate namespace format if we failed to 6066 * find a user xattr. 6067 */ 6068 error = zfs_getextattr_impl(ap, !zfs_xattr_compat); 6069 } 6070 6071 rw_exit(&zp->z_xattr_lock); 6072 zfs_exit(zfsvfs, FTAG); 6073 if (error == ENOENT) 6074 error = SET_ERROR(ENOATTR); 6075 return (error); 6076 } 6077 6078 #ifndef _SYS_SYSPROTO_H_ 6079 struct vop_deleteextattr { 6080 IN struct vnode *a_vp; 6081 IN int a_attrnamespace; 6082 IN const char *a_name; 6083 IN struct ucred *a_cred; 6084 IN struct thread *a_td; 6085 }; 6086 #endif 6087 6088 static int 6089 zfs_deleteextattr_dir(struct vop_deleteextattr_args *ap, const char *attrname) 6090 { 6091 struct nameidata nd; 6092 vnode_t *xvp = NULL, *vp; 6093 int error; 6094 6095 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, 6096 LOOKUP_XATTR, B_FALSE); 6097 if (error != 0) 6098 return (error); 6099 6100 #if __FreeBSD_version < 1400043 6101 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 6102 UIO_SYSSPACE, attrname, xvp, ap->a_td); 6103 #else 6104 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 6105 UIO_SYSSPACE, attrname, xvp); 6106 #endif 6107 error = namei(&nd); 6108 if (error != 0) 6109 return (SET_ERROR(error)); 6110 6111 vp = nd.ni_vp; 6112 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 6113 NDFREE_PNBUF(&nd); 6114 6115 vput(nd.ni_dvp); 6116 if (vp == nd.ni_dvp) 6117 vrele(vp); 6118 else 6119 vput(vp); 6120 6121 return (error); 6122 } 6123 6124 static int 6125 zfs_deleteextattr_sa(struct vop_deleteextattr_args *ap, const char *attrname) 6126 { 6127 znode_t *zp = VTOZ(ap->a_vp); 6128 nvlist_t *nvl; 6129 int error; 6130 6131 error = zfs_ensure_xattr_cached(zp); 6132 if (error != 0) 6133 return (error); 6134 6135 ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock)); 6136 ASSERT3P(zp->z_xattr_cached, !=, NULL); 6137 6138 nvl = zp->z_xattr_cached; 6139 error = nvlist_remove(nvl, attrname, DATA_TYPE_BYTE_ARRAY); 6140 if (error != 0) 6141 error = SET_ERROR(error); 6142 else 6143 error = zfs_sa_set_xattr(zp, attrname, NULL, 0); 6144 if (error != 0) { 6145 zp->z_xattr_cached = NULL; 6146 nvlist_free(nvl); 6147 } 6148 return (error); 6149 } 6150 6151 static int 6152 zfs_deleteextattr_impl(struct vop_deleteextattr_args *ap, boolean_t compat) 6153 { 6154 znode_t *zp = VTOZ(ap->a_vp); 6155 zfsvfs_t *zfsvfs = ZTOZSB(zp); 6156 char attrname[EXTATTR_MAXNAMELEN+1]; 6157 int error; 6158 6159 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6160 sizeof (attrname), compat); 6161 if (error != 0) 6162 return (error); 6163 6164 error = ENOENT; 6165 if (zfsvfs->z_use_sa && zp->z_is_sa) 6166 error = zfs_deleteextattr_sa(ap, attrname); 6167 if (error == ENOENT) 6168 error = zfs_deleteextattr_dir(ap, attrname); 6169 return (error); 6170 } 6171 6172 /* 6173 * Vnode operation to remove a named attribute. 6174 */ 6175 static int 6176 zfs_deleteextattr(struct vop_deleteextattr_args *ap) 6177 { 6178 znode_t *zp = VTOZ(ap->a_vp); 6179 zfsvfs_t *zfsvfs = ZTOZSB(zp); 6180 int error; 6181 6182 /* 6183 * If the xattr property is off, refuse the request. 6184 */ 6185 if (!(zfsvfs->z_flags & ZSB_XATTR)) 6186 return (SET_ERROR(EOPNOTSUPP)); 6187 6188 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6189 ap->a_cred, ap->a_td, VWRITE); 6190 if (error != 0) 6191 return (SET_ERROR(error)); 6192 6193 error = zfs_check_attrname(ap->a_name); 6194 if (error != 0) 6195 return (error); 6196 6197 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 6198 return (error); 6199 rw_enter(&zp->z_xattr_lock, RW_WRITER); 6200 6201 error = zfs_deleteextattr_impl(ap, zfs_xattr_compat); 6202 if ((error == ENOENT || error == ENOATTR) && 6203 ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { 6204 /* 6205 * Fall back to the alternate namespace format if we failed to 6206 * find a user xattr. 6207 */ 6208 error = zfs_deleteextattr_impl(ap, !zfs_xattr_compat); 6209 } 6210 6211 rw_exit(&zp->z_xattr_lock); 6212 zfs_exit(zfsvfs, FTAG); 6213 if (error == ENOENT) 6214 error = SET_ERROR(ENOATTR); 6215 return (error); 6216 } 6217 6218 #ifndef _SYS_SYSPROTO_H_ 6219 struct vop_setextattr { 6220 IN struct vnode *a_vp; 6221 IN int a_attrnamespace; 6222 IN const char *a_name; 6223 INOUT struct uio *a_uio; 6224 IN struct ucred *a_cred; 6225 IN struct thread *a_td; 6226 }; 6227 #endif 6228 6229 static int 6230 zfs_setextattr_dir(struct vop_setextattr_args *ap, const char *attrname) 6231 { 6232 struct thread *td = ap->a_td; 6233 struct nameidata nd; 6234 struct vattr va; 6235 vnode_t *xvp = NULL, *vp; 6236 int error, flags; 6237 6238 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, 6239 LOOKUP_XATTR | CREATE_XATTR_DIR, B_FALSE); 6240 if (error != 0) 6241 return (error); 6242 6243 flags = FFLAGS(O_WRONLY | O_CREAT); 6244 #if __FreeBSD_version < 1400043 6245 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp, td); 6246 #else 6247 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, xvp); 6248 #endif 6249 error = vn_open_cred(&nd, &flags, 0600, VN_OPEN_INVFS, ap->a_cred, 6250 NULL); 6251 if (error != 0) 6252 return (SET_ERROR(error)); 6253 vp = nd.ni_vp; 6254 NDFREE_PNBUF(&nd); 6255 6256 VATTR_NULL(&va); 6257 va.va_size = 0; 6258 error = VOP_SETATTR(vp, &va, ap->a_cred); 6259 if (error == 0) 6260 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6261 6262 VOP_UNLOCK(vp); 6263 vn_close(vp, flags, ap->a_cred, td); 6264 return (error); 6265 } 6266 6267 static int 6268 zfs_setextattr_sa(struct vop_setextattr_args *ap, const char *attrname) 6269 { 6270 znode_t *zp = VTOZ(ap->a_vp); 6271 nvlist_t *nvl; 6272 size_t sa_size; 6273 int error; 6274 6275 error = zfs_ensure_xattr_cached(zp); 6276 if (error != 0) 6277 return (error); 6278 6279 ASSERT(RW_WRITE_HELD(&zp->z_xattr_lock)); 6280 ASSERT3P(zp->z_xattr_cached, !=, NULL); 6281 6282 nvl = zp->z_xattr_cached; 6283 size_t entry_size = ap->a_uio->uio_resid; 6284 if (entry_size > DXATTR_MAX_ENTRY_SIZE) 6285 return (SET_ERROR(EFBIG)); 6286 error = nvlist_size(nvl, &sa_size, NV_ENCODE_XDR); 6287 if (error != 0) 6288 return (SET_ERROR(error)); 6289 if (sa_size > DXATTR_MAX_SA_SIZE) 6290 return (SET_ERROR(EFBIG)); 6291 uchar_t *buf = kmem_alloc(entry_size, KM_SLEEP); 6292 error = uiomove(buf, entry_size, ap->a_uio); 6293 if (error != 0) { 6294 error = SET_ERROR(error); 6295 } else { 6296 error = nvlist_add_byte_array(nvl, attrname, buf, entry_size); 6297 if (error != 0) 6298 error = SET_ERROR(error); 6299 } 6300 if (error == 0) 6301 error = zfs_sa_set_xattr(zp, attrname, buf, entry_size); 6302 kmem_free(buf, entry_size); 6303 if (error != 0) { 6304 zp->z_xattr_cached = NULL; 6305 nvlist_free(nvl); 6306 } 6307 return (error); 6308 } 6309 6310 static int 6311 zfs_setextattr_impl(struct vop_setextattr_args *ap, boolean_t compat) 6312 { 6313 znode_t *zp = VTOZ(ap->a_vp); 6314 zfsvfs_t *zfsvfs = ZTOZSB(zp); 6315 char attrname[EXTATTR_MAXNAMELEN+1]; 6316 int error; 6317 6318 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6319 sizeof (attrname), compat); 6320 if (error != 0) 6321 return (error); 6322 6323 struct vop_deleteextattr_args vda = { 6324 .a_vp = ap->a_vp, 6325 .a_attrnamespace = ap->a_attrnamespace, 6326 .a_name = ap->a_name, 6327 .a_cred = ap->a_cred, 6328 .a_td = ap->a_td, 6329 }; 6330 error = ENOENT; 6331 if (zfsvfs->z_use_sa && zp->z_is_sa && zfsvfs->z_xattr_sa) { 6332 error = zfs_setextattr_sa(ap, attrname); 6333 if (error == 0) { 6334 /* 6335 * Successfully put into SA, we need to clear the one 6336 * in dir if present. 6337 */ 6338 zfs_deleteextattr_dir(&vda, attrname); 6339 } 6340 } 6341 if (error != 0) { 6342 error = zfs_setextattr_dir(ap, attrname); 6343 if (error == 0 && zp->z_is_sa) { 6344 /* 6345 * Successfully put into dir, we need to clear the one 6346 * in SA if present. 6347 */ 6348 zfs_deleteextattr_sa(&vda, attrname); 6349 } 6350 } 6351 if (error == 0 && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { 6352 /* 6353 * Also clear all versions of the alternate compat name. 6354 */ 6355 zfs_deleteextattr_impl(&vda, !compat); 6356 } 6357 return (error); 6358 } 6359 6360 /* 6361 * Vnode operation to set a named attribute. 6362 */ 6363 static int 6364 zfs_setextattr(struct vop_setextattr_args *ap) 6365 { 6366 znode_t *zp = VTOZ(ap->a_vp); 6367 zfsvfs_t *zfsvfs = ZTOZSB(zp); 6368 int error; 6369 6370 /* 6371 * If the xattr property is off, refuse the request. 6372 */ 6373 if (!(zfsvfs->z_flags & ZSB_XATTR)) 6374 return (SET_ERROR(EOPNOTSUPP)); 6375 6376 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6377 ap->a_cred, ap->a_td, VWRITE); 6378 if (error != 0) 6379 return (SET_ERROR(error)); 6380 6381 error = zfs_check_attrname(ap->a_name); 6382 if (error != 0) 6383 return (error); 6384 6385 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 6386 return (error); 6387 rw_enter(&zp->z_xattr_lock, RW_WRITER); 6388 6389 error = zfs_setextattr_impl(ap, zfs_xattr_compat); 6390 6391 rw_exit(&zp->z_xattr_lock); 6392 zfs_exit(zfsvfs, FTAG); 6393 return (error); 6394 } 6395 6396 #ifndef _SYS_SYSPROTO_H_ 6397 struct vop_listextattr { 6398 IN struct vnode *a_vp; 6399 IN int a_attrnamespace; 6400 INOUT struct uio *a_uio; 6401 OUT size_t *a_size; 6402 IN struct ucred *a_cred; 6403 IN struct thread *a_td; 6404 }; 6405 #endif 6406 6407 static int 6408 zfs_listextattr_dir(struct vop_listextattr_args *ap, const char *attrprefix) 6409 { 6410 struct thread *td = ap->a_td; 6411 struct nameidata nd; 6412 uint8_t dirbuf[sizeof (struct dirent)]; 6413 struct iovec aiov; 6414 struct uio auio; 6415 vnode_t *xvp = NULL, *vp; 6416 int error, eof; 6417 6418 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, 6419 LOOKUP_XATTR, B_FALSE); 6420 if (error != 0) { 6421 /* 6422 * ENOATTR means that the EA directory does not yet exist, 6423 * i.e. there are no extended attributes there. 6424 */ 6425 if (error == ENOATTR) 6426 error = 0; 6427 return (error); 6428 } 6429 6430 #if __FreeBSD_version < 1400043 6431 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 6432 UIO_SYSSPACE, ".", xvp, td); 6433 #else 6434 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 6435 UIO_SYSSPACE, ".", xvp); 6436 #endif 6437 error = namei(&nd); 6438 if (error != 0) 6439 return (SET_ERROR(error)); 6440 vp = nd.ni_vp; 6441 NDFREE_PNBUF(&nd); 6442 6443 auio.uio_iov = &aiov; 6444 auio.uio_iovcnt = 1; 6445 auio.uio_segflg = UIO_SYSSPACE; 6446 auio.uio_td = td; 6447 auio.uio_rw = UIO_READ; 6448 auio.uio_offset = 0; 6449 6450 size_t plen = strlen(attrprefix); 6451 6452 do { 6453 aiov.iov_base = (void *)dirbuf; 6454 aiov.iov_len = sizeof (dirbuf); 6455 auio.uio_resid = sizeof (dirbuf); 6456 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 6457 if (error != 0) 6458 break; 6459 int done = sizeof (dirbuf) - auio.uio_resid; 6460 for (int pos = 0; pos < done; ) { 6461 struct dirent *dp = (struct dirent *)(dirbuf + pos); 6462 pos += dp->d_reclen; 6463 /* 6464 * XXX: Temporarily we also accept DT_UNKNOWN, as this 6465 * is what we get when attribute was created on Solaris. 6466 */ 6467 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 6468 continue; 6469 else if (plen == 0 && 6470 ZFS_XA_NS_PREFIX_FORBIDDEN(dp->d_name)) 6471 continue; 6472 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 6473 continue; 6474 uint8_t nlen = dp->d_namlen - plen; 6475 if (ap->a_size != NULL) { 6476 *ap->a_size += 1 + nlen; 6477 } else if (ap->a_uio != NULL) { 6478 /* 6479 * Format of extattr name entry is one byte for 6480 * length and the rest for name. 6481 */ 6482 error = uiomove(&nlen, 1, ap->a_uio); 6483 if (error == 0) { 6484 char *namep = dp->d_name + plen; 6485 error = uiomove(namep, nlen, ap->a_uio); 6486 } 6487 if (error != 0) { 6488 error = SET_ERROR(error); 6489 break; 6490 } 6491 } 6492 } 6493 } while (!eof && error == 0); 6494 6495 vput(vp); 6496 return (error); 6497 } 6498 6499 static int 6500 zfs_listextattr_sa(struct vop_listextattr_args *ap, const char *attrprefix) 6501 { 6502 znode_t *zp = VTOZ(ap->a_vp); 6503 int error; 6504 6505 error = zfs_ensure_xattr_cached(zp); 6506 if (error != 0) 6507 return (error); 6508 6509 ASSERT(RW_LOCK_HELD(&zp->z_xattr_lock)); 6510 ASSERT3P(zp->z_xattr_cached, !=, NULL); 6511 6512 size_t plen = strlen(attrprefix); 6513 nvpair_t *nvp = NULL; 6514 while ((nvp = nvlist_next_nvpair(zp->z_xattr_cached, nvp)) != NULL) { 6515 ASSERT3U(nvpair_type(nvp), ==, DATA_TYPE_BYTE_ARRAY); 6516 6517 const char *name = nvpair_name(nvp); 6518 if (plen == 0 && ZFS_XA_NS_PREFIX_FORBIDDEN(name)) 6519 continue; 6520 else if (strncmp(name, attrprefix, plen) != 0) 6521 continue; 6522 uint8_t nlen = strlen(name) - plen; 6523 if (ap->a_size != NULL) { 6524 *ap->a_size += 1 + nlen; 6525 } else if (ap->a_uio != NULL) { 6526 /* 6527 * Format of extattr name entry is one byte for 6528 * length and the rest for name. 6529 */ 6530 error = uiomove(&nlen, 1, ap->a_uio); 6531 if (error == 0) { 6532 char *namep = __DECONST(char *, name) + plen; 6533 error = uiomove(namep, nlen, ap->a_uio); 6534 } 6535 if (error != 0) { 6536 error = SET_ERROR(error); 6537 break; 6538 } 6539 } 6540 } 6541 6542 return (error); 6543 } 6544 6545 static int 6546 zfs_listextattr_impl(struct vop_listextattr_args *ap, boolean_t compat) 6547 { 6548 znode_t *zp = VTOZ(ap->a_vp); 6549 zfsvfs_t *zfsvfs = ZTOZSB(zp); 6550 char attrprefix[16]; 6551 int error; 6552 6553 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 6554 sizeof (attrprefix), compat); 6555 if (error != 0) 6556 return (error); 6557 6558 if (zfsvfs->z_use_sa && zp->z_is_sa) 6559 error = zfs_listextattr_sa(ap, attrprefix); 6560 if (error == 0) 6561 error = zfs_listextattr_dir(ap, attrprefix); 6562 return (error); 6563 } 6564 6565 /* 6566 * Vnode operation to retrieve extended attributes on a vnode. 6567 */ 6568 static int 6569 zfs_listextattr(struct vop_listextattr_args *ap) 6570 { 6571 znode_t *zp = VTOZ(ap->a_vp); 6572 zfsvfs_t *zfsvfs = ZTOZSB(zp); 6573 int error; 6574 6575 if (ap->a_size != NULL) 6576 *ap->a_size = 0; 6577 6578 /* 6579 * If the xattr property is off, refuse the request. 6580 */ 6581 if (!(zfsvfs->z_flags & ZSB_XATTR)) 6582 return (SET_ERROR(EOPNOTSUPP)); 6583 6584 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6585 ap->a_cred, ap->a_td, VREAD); 6586 if (error != 0) 6587 return (SET_ERROR(error)); 6588 6589 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 6590 return (error); 6591 rw_enter(&zp->z_xattr_lock, RW_READER); 6592 6593 error = zfs_listextattr_impl(ap, zfs_xattr_compat); 6594 if (error == 0 && ap->a_attrnamespace == EXTATTR_NAMESPACE_USER) { 6595 /* Also list user xattrs with the alternate format. */ 6596 error = zfs_listextattr_impl(ap, !zfs_xattr_compat); 6597 } 6598 6599 rw_exit(&zp->z_xattr_lock); 6600 zfs_exit(zfsvfs, FTAG); 6601 return (error); 6602 } 6603 6604 #ifndef _SYS_SYSPROTO_H_ 6605 struct vop_getacl_args { 6606 struct vnode *vp; 6607 acl_type_t type; 6608 struct acl *aclp; 6609 struct ucred *cred; 6610 struct thread *td; 6611 }; 6612 #endif 6613 6614 static int 6615 zfs_freebsd_getacl(struct vop_getacl_args *ap) 6616 { 6617 int error; 6618 vsecattr_t vsecattr; 6619 6620 if (ap->a_type != ACL_TYPE_NFS4) 6621 return (EINVAL); 6622 6623 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 6624 if ((error = zfs_getsecattr(VTOZ(ap->a_vp), 6625 &vsecattr, 0, ap->a_cred))) 6626 return (error); 6627 6628 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, 6629 vsecattr.vsa_aclcnt); 6630 if (vsecattr.vsa_aclentp != NULL) 6631 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 6632 6633 return (error); 6634 } 6635 6636 #ifndef _SYS_SYSPROTO_H_ 6637 struct vop_setacl_args { 6638 struct vnode *vp; 6639 acl_type_t type; 6640 struct acl *aclp; 6641 struct ucred *cred; 6642 struct thread *td; 6643 }; 6644 #endif 6645 6646 static int 6647 zfs_freebsd_setacl(struct vop_setacl_args *ap) 6648 { 6649 int error; 6650 vsecattr_t vsecattr; 6651 int aclbsize; /* size of acl list in bytes */ 6652 aclent_t *aaclp; 6653 6654 if (ap->a_type != ACL_TYPE_NFS4) 6655 return (EINVAL); 6656 6657 if (ap->a_aclp == NULL) 6658 return (EINVAL); 6659 6660 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 6661 return (EINVAL); 6662 6663 /* 6664 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 6665 * splitting every entry into two and appending "canonical six" 6666 * entries at the end. Don't allow for setting an ACL that would 6667 * cause chmod(2) to run out of ACL entries. 6668 */ 6669 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 6670 return (ENOSPC); 6671 6672 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 6673 if (error != 0) 6674 return (error); 6675 6676 vsecattr.vsa_mask = VSA_ACE; 6677 aclbsize = ap->a_aclp->acl_cnt * sizeof (ace_t); 6678 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 6679 aaclp = vsecattr.vsa_aclentp; 6680 vsecattr.vsa_aclentsz = aclbsize; 6681 6682 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 6683 error = zfs_setsecattr(VTOZ(ap->a_vp), &vsecattr, 0, ap->a_cred); 6684 kmem_free(aaclp, aclbsize); 6685 6686 return (error); 6687 } 6688 6689 #ifndef _SYS_SYSPROTO_H_ 6690 struct vop_aclcheck_args { 6691 struct vnode *vp; 6692 acl_type_t type; 6693 struct acl *aclp; 6694 struct ucred *cred; 6695 struct thread *td; 6696 }; 6697 #endif 6698 6699 static int 6700 zfs_freebsd_aclcheck(struct vop_aclcheck_args *ap) 6701 { 6702 6703 return (EOPNOTSUPP); 6704 } 6705 6706 #ifndef _SYS_SYSPROTO_H_ 6707 struct vop_advise_args { 6708 struct vnode *a_vp; 6709 off_t a_start; 6710 off_t a_end; 6711 int a_advice; 6712 }; 6713 #endif 6714 6715 static int 6716 zfs_freebsd_advise(struct vop_advise_args *ap) 6717 { 6718 vnode_t *vp = ap->a_vp; 6719 off_t start = ap->a_start; 6720 off_t end = ap->a_end; 6721 int advice = ap->a_advice; 6722 off_t len; 6723 znode_t *zp; 6724 zfsvfs_t *zfsvfs; 6725 objset_t *os; 6726 int error = 0; 6727 6728 if (end < start) 6729 return (EINVAL); 6730 6731 error = vn_lock(vp, LK_SHARED); 6732 if (error) 6733 return (error); 6734 6735 zp = VTOZ(vp); 6736 zfsvfs = zp->z_zfsvfs; 6737 os = zp->z_zfsvfs->z_os; 6738 6739 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 6740 goto out_unlock; 6741 6742 /* kern_posix_fadvise points to the last byte, we want one past */ 6743 if (end != OFF_MAX) 6744 end += 1; 6745 len = end - start; 6746 6747 switch (advice) { 6748 case POSIX_FADV_WILLNEED: 6749 /* 6750 * Pass on the caller's size directly, but note that 6751 * dmu_prefetch_max will effectively cap it. If there really 6752 * is a larger sequential access pattern, perhaps dmu_zfetch 6753 * will detect it. 6754 */ 6755 dmu_prefetch(os, zp->z_id, 0, start, len, 6756 ZIO_PRIORITY_ASYNC_READ); 6757 break; 6758 case POSIX_FADV_NORMAL: 6759 case POSIX_FADV_RANDOM: 6760 case POSIX_FADV_SEQUENTIAL: 6761 case POSIX_FADV_DONTNEED: 6762 case POSIX_FADV_NOREUSE: 6763 /* ignored for now */ 6764 break; 6765 default: 6766 error = EINVAL; 6767 break; 6768 } 6769 6770 zfs_exit(zfsvfs, FTAG); 6771 6772 out_unlock: 6773 VOP_UNLOCK(vp); 6774 6775 return (error); 6776 } 6777 6778 static int 6779 zfs_vptocnp(struct vop_vptocnp_args *ap) 6780 { 6781 vnode_t *covered_vp; 6782 vnode_t *vp = ap->a_vp; 6783 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 6784 znode_t *zp = VTOZ(vp); 6785 int ltype; 6786 int error; 6787 6788 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 6789 return (error); 6790 6791 /* 6792 * If we are a snapshot mounted under .zfs, run the operation 6793 * on the covered vnode. 6794 */ 6795 if (zp->z_id != zfsvfs->z_root || zfsvfs->z_parent == zfsvfs) { 6796 char name[MAXNAMLEN + 1]; 6797 znode_t *dzp; 6798 size_t len; 6799 6800 error = zfs_znode_parent_and_name(zp, &dzp, name, 6801 sizeof (name)); 6802 if (error == 0) { 6803 len = strlen(name); 6804 if (*ap->a_buflen < len) 6805 error = SET_ERROR(ENOMEM); 6806 } 6807 if (error == 0) { 6808 *ap->a_buflen -= len; 6809 memcpy(ap->a_buf + *ap->a_buflen, name, len); 6810 *ap->a_vpp = ZTOV(dzp); 6811 } 6812 zfs_exit(zfsvfs, FTAG); 6813 return (error); 6814 } 6815 zfs_exit(zfsvfs, FTAG); 6816 6817 covered_vp = vp->v_mount->mnt_vnodecovered; 6818 enum vgetstate vs = vget_prep(covered_vp); 6819 ltype = VOP_ISLOCKED(vp); 6820 VOP_UNLOCK(vp); 6821 error = vget_finish(covered_vp, LK_SHARED, vs); 6822 if (error == 0) { 6823 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_buf, 6824 ap->a_buflen); 6825 vput(covered_vp); 6826 } 6827 vn_lock(vp, ltype | LK_RETRY); 6828 if (VN_IS_DOOMED(vp)) 6829 error = SET_ERROR(ENOENT); 6830 return (error); 6831 } 6832 6833 #if __FreeBSD_version >= 1400032 6834 static int 6835 zfs_deallocate(struct vop_deallocate_args *ap) 6836 { 6837 znode_t *zp = VTOZ(ap->a_vp); 6838 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6839 zilog_t *zilog; 6840 off_t off, len, file_sz; 6841 int error; 6842 6843 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 6844 return (error); 6845 6846 /* 6847 * Callers might not be able to detect properly that we are read-only, 6848 * so check it explicitly here. 6849 */ 6850 if (zfs_is_readonly(zfsvfs)) { 6851 zfs_exit(zfsvfs, FTAG); 6852 return (SET_ERROR(EROFS)); 6853 } 6854 6855 zilog = zfsvfs->z_log; 6856 off = *ap->a_offset; 6857 len = *ap->a_len; 6858 file_sz = zp->z_size; 6859 if (off + len > file_sz) 6860 len = file_sz - off; 6861 /* Fast path for out-of-range request. */ 6862 if (len <= 0) { 6863 *ap->a_len = 0; 6864 zfs_exit(zfsvfs, FTAG); 6865 return (0); 6866 } 6867 6868 error = zfs_freesp(zp, off, len, O_RDWR, TRUE); 6869 if (error == 0) { 6870 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS || 6871 (ap->a_ioflag & IO_SYNC) != 0) 6872 error = zil_commit(zilog, zp->z_id); 6873 if (error == 0) { 6874 *ap->a_offset = off + len; 6875 *ap->a_len = 0; 6876 } 6877 } 6878 6879 zfs_exit(zfsvfs, FTAG); 6880 return (error); 6881 } 6882 #endif 6883 6884 #ifndef _SYS_SYSPROTO_H_ 6885 struct vop_copy_file_range_args { 6886 struct vnode *a_invp; 6887 off_t *a_inoffp; 6888 struct vnode *a_outvp; 6889 off_t *a_outoffp; 6890 size_t *a_lenp; 6891 unsigned int a_flags; 6892 struct ucred *a_incred; 6893 struct ucred *a_outcred; 6894 struct thread *a_fsizetd; 6895 } 6896 #endif 6897 /* 6898 * TODO: FreeBSD will only call file system-specific copy_file_range() if both 6899 * files resides under the same mountpoint. In case of ZFS we want to be called 6900 * even is files are in different datasets (but on the same pools, but we need 6901 * to check that ourselves). 6902 */ 6903 static int 6904 zfs_freebsd_copy_file_range(struct vop_copy_file_range_args *ap) 6905 { 6906 zfsvfs_t *outzfsvfs; 6907 struct vnode *invp = ap->a_invp; 6908 struct vnode *outvp = ap->a_outvp; 6909 struct mount *mp; 6910 int error; 6911 uint64_t len = *ap->a_lenp; 6912 6913 if (!zfs_bclone_enabled) { 6914 mp = NULL; 6915 goto bad_write_fallback; 6916 } 6917 6918 /* 6919 * TODO: If offset/length is not aligned to recordsize, use 6920 * vn_generic_copy_file_range() on this fragment. 6921 * It would be better to do this after we lock the vnodes, but then we 6922 * need something else than vn_generic_copy_file_range(). 6923 */ 6924 6925 vn_start_write(outvp, &mp, V_WAIT); 6926 if (__predict_true(mp == outvp->v_mount)) { 6927 outzfsvfs = (zfsvfs_t *)mp->mnt_data; 6928 if (!spa_feature_is_enabled(dmu_objset_spa(outzfsvfs->z_os), 6929 SPA_FEATURE_BLOCK_CLONING)) { 6930 goto bad_write_fallback; 6931 } 6932 } 6933 if (invp == outvp) { 6934 if (vn_lock(outvp, LK_EXCLUSIVE) != 0) { 6935 goto bad_write_fallback; 6936 } 6937 } else { 6938 #if (__FreeBSD_version >= 1302506 && __FreeBSD_version < 1400000) || \ 6939 __FreeBSD_version >= 1400086 6940 vn_lock_pair(invp, false, LK_SHARED, outvp, false, 6941 LK_EXCLUSIVE); 6942 #else 6943 vn_lock_pair(invp, false, outvp, false); 6944 #endif 6945 if (VN_IS_DOOMED(invp) || VN_IS_DOOMED(outvp)) { 6946 goto bad_locked_fallback; 6947 } 6948 } 6949 6950 #ifdef MAC 6951 error = mac_vnode_check_write(curthread->td_ucred, ap->a_outcred, 6952 outvp); 6953 if (error != 0) 6954 goto out_locked; 6955 #endif 6956 6957 error = zfs_clone_range(VTOZ(invp), ap->a_inoffp, VTOZ(outvp), 6958 ap->a_outoffp, &len, ap->a_outcred); 6959 if (error == EXDEV || error == EAGAIN || error == EINVAL || 6960 error == EOPNOTSUPP) 6961 goto bad_locked_fallback; 6962 *ap->a_lenp = (size_t)len; 6963 #ifdef MAC 6964 out_locked: 6965 #endif 6966 if (invp != outvp) 6967 VOP_UNLOCK(invp); 6968 VOP_UNLOCK(outvp); 6969 if (mp != NULL) 6970 vn_finished_write(mp); 6971 return (error); 6972 6973 bad_locked_fallback: 6974 if (invp != outvp) 6975 VOP_UNLOCK(invp); 6976 VOP_UNLOCK(outvp); 6977 bad_write_fallback: 6978 if (mp != NULL) 6979 vn_finished_write(mp); 6980 error = ENOSYS; 6981 return (error); 6982 } 6983 6984 struct vop_vector zfs_vnodeops; 6985 struct vop_vector zfs_fifoops; 6986 struct vop_vector zfs_shareops; 6987 6988 struct vop_vector zfs_vnodeops = { 6989 .vop_default = &default_vnodeops, 6990 .vop_inactive = zfs_freebsd_inactive, 6991 .vop_need_inactive = zfs_freebsd_need_inactive, 6992 .vop_reclaim = zfs_freebsd_reclaim, 6993 .vop_fplookup_vexec = zfs_freebsd_fplookup_vexec, 6994 .vop_fplookup_symlink = zfs_freebsd_fplookup_symlink, 6995 .vop_access = zfs_freebsd_access, 6996 .vop_allocate = VOP_EOPNOTSUPP, 6997 #if __FreeBSD_version >= 1400032 6998 .vop_deallocate = zfs_deallocate, 6999 #endif 7000 .vop_lookup = zfs_cache_lookup, 7001 .vop_cachedlookup = zfs_freebsd_cachedlookup, 7002 .vop_getattr = zfs_freebsd_getattr, 7003 .vop_setattr = zfs_freebsd_setattr, 7004 .vop_create = zfs_freebsd_create, 7005 .vop_mknod = (vop_mknod_t *)zfs_freebsd_create, 7006 .vop_mkdir = zfs_freebsd_mkdir, 7007 .vop_readdir = zfs_freebsd_readdir, 7008 .vop_fsync = zfs_freebsd_fsync, 7009 .vop_open = zfs_freebsd_open, 7010 .vop_close = zfs_freebsd_close, 7011 .vop_rmdir = zfs_freebsd_rmdir, 7012 .vop_ioctl = zfs_freebsd_ioctl, 7013 .vop_link = zfs_freebsd_link, 7014 .vop_symlink = zfs_freebsd_symlink, 7015 .vop_readlink = zfs_freebsd_readlink, 7016 .vop_advise = zfs_freebsd_advise, 7017 .vop_read = zfs_freebsd_read, 7018 .vop_write = zfs_freebsd_write, 7019 .vop_remove = zfs_freebsd_remove, 7020 .vop_rename = zfs_freebsd_rename, 7021 .vop_pathconf = zfs_freebsd_pathconf, 7022 .vop_bmap = zfs_freebsd_bmap, 7023 .vop_fid = zfs_freebsd_fid, 7024 .vop_getextattr = zfs_getextattr, 7025 .vop_deleteextattr = zfs_deleteextattr, 7026 .vop_setextattr = zfs_setextattr, 7027 .vop_listextattr = zfs_listextattr, 7028 .vop_getacl = zfs_freebsd_getacl, 7029 .vop_setacl = zfs_freebsd_setacl, 7030 .vop_aclcheck = zfs_freebsd_aclcheck, 7031 .vop_getpages = zfs_freebsd_getpages, 7032 .vop_putpages = zfs_freebsd_putpages, 7033 .vop_vptocnp = zfs_vptocnp, 7034 .vop_lock1 = vop_lock, 7035 .vop_unlock = vop_unlock, 7036 .vop_islocked = vop_islocked, 7037 #if __FreeBSD_version >= 1400043 7038 .vop_add_writecount = vop_stdadd_writecount_nomsync, 7039 #endif 7040 .vop_copy_file_range = zfs_freebsd_copy_file_range, 7041 }; 7042 VFS_VOP_VECTOR_REGISTER(zfs_vnodeops); 7043 7044 struct vop_vector zfs_fifoops = { 7045 .vop_default = &fifo_specops, 7046 .vop_fsync = zfs_freebsd_fsync, 7047 .vop_fplookup_vexec = zfs_freebsd_fplookup_vexec, 7048 .vop_fplookup_symlink = zfs_freebsd_fplookup_symlink, 7049 .vop_access = zfs_freebsd_access, 7050 .vop_getattr = zfs_freebsd_getattr, 7051 .vop_inactive = zfs_freebsd_inactive, 7052 .vop_read = VOP_PANIC, 7053 .vop_reclaim = zfs_freebsd_reclaim, 7054 .vop_setattr = zfs_freebsd_setattr, 7055 .vop_write = VOP_PANIC, 7056 .vop_pathconf = zfs_freebsd_pathconf, 7057 .vop_fid = zfs_freebsd_fid, 7058 .vop_getacl = zfs_freebsd_getacl, 7059 .vop_setacl = zfs_freebsd_setacl, 7060 .vop_aclcheck = zfs_freebsd_aclcheck, 7061 #if __FreeBSD_version >= 1400043 7062 .vop_add_writecount = vop_stdadd_writecount_nomsync, 7063 #endif 7064 }; 7065 VFS_VOP_VECTOR_REGISTER(zfs_fifoops); 7066 7067 /* 7068 * special share hidden files vnode operations template 7069 */ 7070 struct vop_vector zfs_shareops = { 7071 .vop_default = &default_vnodeops, 7072 .vop_fplookup_vexec = VOP_EAGAIN, 7073 .vop_fplookup_symlink = VOP_EAGAIN, 7074 .vop_access = zfs_freebsd_access, 7075 .vop_inactive = zfs_freebsd_inactive, 7076 .vop_reclaim = zfs_freebsd_reclaim, 7077 .vop_fid = zfs_freebsd_fid, 7078 .vop_pathconf = zfs_freebsd_pathconf, 7079 #if __FreeBSD_version >= 1400043 7080 .vop_add_writecount = vop_stdadd_writecount_nomsync, 7081 #endif 7082 }; 7083 VFS_VOP_VECTOR_REGISTER(zfs_shareops); 7084 7085 ZFS_MODULE_PARAM(zfs, zfs_, xattr_compat, INT, ZMOD_RW, 7086 "Use legacy ZFS xattr naming for writing new user namespace xattrs"); 7087