zfs_vnops.c revision 306819
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2015 by Delphix. All rights reserved. 24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved. 25 * Copyright (c) 2014 Integros [integros.com] 26 */ 27 28/* Portions Copyright 2007 Jeremy Teo */ 29/* Portions Copyright 2010 Robert Milkowski */ 30 31#include <sys/types.h> 32#include <sys/param.h> 33#include <sys/time.h> 34#include <sys/systm.h> 35#include <sys/sysmacros.h> 36#include <sys/resource.h> 37#include <sys/vfs.h> 38#include <sys/vm.h> 39#include <sys/vnode.h> 40#include <sys/file.h> 41#include <sys/stat.h> 42#include <sys/kmem.h> 43#include <sys/taskq.h> 44#include <sys/uio.h> 45#include <sys/atomic.h> 46#include <sys/namei.h> 47#include <sys/mman.h> 48#include <sys/cmn_err.h> 49#include <sys/errno.h> 50#include <sys/unistd.h> 51#include <sys/zfs_dir.h> 52#include <sys/zfs_ioctl.h> 53#include <sys/fs/zfs.h> 54#include <sys/dmu.h> 55#include <sys/dmu_objset.h> 56#include <sys/spa.h> 57#include <sys/txg.h> 58#include <sys/dbuf.h> 59#include <sys/zap.h> 60#include <sys/sa.h> 61#include <sys/dirent.h> 62#include <sys/policy.h> 63#include <sys/sunddi.h> 64#include <sys/filio.h> 65#include <sys/sid.h> 66#include <sys/zfs_ctldir.h> 67#include <sys/zfs_fuid.h> 68#include <sys/zfs_sa.h> 69#include <sys/zfs_rlock.h> 70#include <sys/extdirent.h> 71#include <sys/kidmap.h> 72#include <sys/bio.h> 73#include <sys/buf.h> 74#include <sys/sched.h> 75#include <sys/acl.h> 76#include <vm/vm_param.h> 77 78/* 79 * Programming rules. 80 * 81 * Each vnode op performs some logical unit of work. To do this, the ZPL must 82 * properly lock its in-core state, create a DMU transaction, do the work, 83 * record this work in the intent log (ZIL), commit the DMU transaction, 84 * and wait for the intent log to commit if it is a synchronous operation. 85 * Moreover, the vnode ops must work in both normal and log replay context. 86 * The ordering of events is important to avoid deadlocks and references 87 * to freed memory. The example below illustrates the following Big Rules: 88 * 89 * (1) A check must be made in each zfs thread for a mounted file system. 90 * This is done avoiding races using ZFS_ENTER(zfsvfs). 91 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes 92 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros 93 * can return EIO from the calling function. 94 * 95 * (2) VN_RELE() should always be the last thing except for zil_commit() 96 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 97 * First, if it's the last reference, the vnode/znode 98 * can be freed, so the zp may point to freed memory. Second, the last 99 * reference will call zfs_zinactive(), which may induce a lot of work -- 100 * pushing cached pages (which acquires range locks) and syncing out 101 * cached atime changes. Third, zfs_zinactive() may require a new tx, 102 * which could deadlock the system if you were already holding one. 103 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). 104 * 105 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 106 * as they can span dmu_tx_assign() calls. 107 * 108 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to 109 * dmu_tx_assign(). This is critical because we don't want to block 110 * while holding locks. 111 * 112 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This 113 * reduces lock contention and CPU usage when we must wait (note that if 114 * throughput is constrained by the storage, nearly every transaction 115 * must wait). 116 * 117 * Note, in particular, that if a lock is sometimes acquired before 118 * the tx assigns, and sometimes after (e.g. z_lock), then failing 119 * to use a non-blocking assign can deadlock the system. The scenario: 120 * 121 * Thread A has grabbed a lock before calling dmu_tx_assign(). 122 * Thread B is in an already-assigned tx, and blocks for this lock. 123 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 124 * forever, because the previous txg can't quiesce until B's tx commits. 125 * 126 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 127 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent 128 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT, 129 * to indicate that this operation has already called dmu_tx_wait(). 130 * This will ensure that we don't retry forever, waiting a short bit 131 * each time. 132 * 133 * (5) If the operation succeeded, generate the intent log entry for it 134 * before dropping locks. This ensures that the ordering of events 135 * in the intent log matches the order in which they actually occurred. 136 * During ZIL replay the zfs_log_* functions will update the sequence 137 * number to indicate the zil transaction has replayed. 138 * 139 * (6) At the end of each vnode op, the DMU tx must always commit, 140 * regardless of whether there were any errors. 141 * 142 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 143 * to ensure that synchronous semantics are provided when necessary. 144 * 145 * In general, this is how things should be ordered in each vnode op: 146 * 147 * ZFS_ENTER(zfsvfs); // exit if unmounted 148 * top: 149 * zfs_dirent_lookup(&dl, ...) // lock directory entry (may VN_HOLD()) 150 * rw_enter(...); // grab any other locks you need 151 * tx = dmu_tx_create(...); // get DMU tx 152 * dmu_tx_hold_*(); // hold each object you might modify 153 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 154 * if (error) { 155 * rw_exit(...); // drop locks 156 * zfs_dirent_unlock(dl); // unlock directory entry 157 * VN_RELE(...); // release held vnodes 158 * if (error == ERESTART) { 159 * waited = B_TRUE; 160 * dmu_tx_wait(tx); 161 * dmu_tx_abort(tx); 162 * goto top; 163 * } 164 * dmu_tx_abort(tx); // abort DMU tx 165 * ZFS_EXIT(zfsvfs); // finished in zfs 166 * return (error); // really out of space 167 * } 168 * error = do_real_work(); // do whatever this VOP does 169 * if (error == 0) 170 * zfs_log_*(...); // on success, make ZIL entry 171 * dmu_tx_commit(tx); // commit DMU tx -- error or not 172 * rw_exit(...); // drop locks 173 * zfs_dirent_unlock(dl); // unlock directory entry 174 * VN_RELE(...); // release held vnodes 175 * zil_commit(zilog, foid); // synchronous when necessary 176 * ZFS_EXIT(zfsvfs); // finished in zfs 177 * return (error); // done, report error 178 */ 179 180/* ARGSUSED */ 181static int 182zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 183{ 184 znode_t *zp = VTOZ(*vpp); 185 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 186 187 ZFS_ENTER(zfsvfs); 188 ZFS_VERIFY_ZP(zp); 189 190 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 191 ((flag & FAPPEND) == 0)) { 192 ZFS_EXIT(zfsvfs); 193 return (SET_ERROR(EPERM)); 194 } 195 196 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 197 ZTOV(zp)->v_type == VREG && 198 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) { 199 if (fs_vscan(*vpp, cr, 0) != 0) { 200 ZFS_EXIT(zfsvfs); 201 return (SET_ERROR(EACCES)); 202 } 203 } 204 205 /* Keep a count of the synchronous opens in the znode */ 206 if (flag & (FSYNC | FDSYNC)) 207 atomic_inc_32(&zp->z_sync_cnt); 208 209 ZFS_EXIT(zfsvfs); 210 return (0); 211} 212 213/* ARGSUSED */ 214static int 215zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 216 caller_context_t *ct) 217{ 218 znode_t *zp = VTOZ(vp); 219 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 220 221 /* 222 * Clean up any locks held by this process on the vp. 223 */ 224 cleanlocks(vp, ddi_get_pid(), 0); 225 cleanshares(vp, ddi_get_pid()); 226 227 ZFS_ENTER(zfsvfs); 228 ZFS_VERIFY_ZP(zp); 229 230 /* Decrement the synchronous opens in the znode */ 231 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 232 atomic_dec_32(&zp->z_sync_cnt); 233 234 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 235 ZTOV(zp)->v_type == VREG && 236 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) 237 VERIFY(fs_vscan(vp, cr, 1) == 0); 238 239 ZFS_EXIT(zfsvfs); 240 return (0); 241} 242 243/* 244 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 245 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 246 */ 247static int 248zfs_holey(vnode_t *vp, u_long cmd, offset_t *off) 249{ 250 znode_t *zp = VTOZ(vp); 251 uint64_t noff = (uint64_t)*off; /* new offset */ 252 uint64_t file_sz; 253 int error; 254 boolean_t hole; 255 256 file_sz = zp->z_size; 257 if (noff >= file_sz) { 258 return (SET_ERROR(ENXIO)); 259 } 260 261 if (cmd == _FIO_SEEK_HOLE) 262 hole = B_TRUE; 263 else 264 hole = B_FALSE; 265 266 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 267 268 if (error == ESRCH) 269 return (SET_ERROR(ENXIO)); 270 271 /* 272 * We could find a hole that begins after the logical end-of-file, 273 * because dmu_offset_next() only works on whole blocks. If the 274 * EOF falls mid-block, then indicate that the "virtual hole" 275 * at the end of the file begins at the logical EOF, rather than 276 * at the end of the last block. 277 */ 278 if (noff > file_sz) { 279 ASSERT(hole); 280 noff = file_sz; 281 } 282 283 if (noff < *off) 284 return (error); 285 *off = noff; 286 return (error); 287} 288 289/* ARGSUSED */ 290static int 291zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred, 292 int *rvalp, caller_context_t *ct) 293{ 294 offset_t off; 295 offset_t ndata; 296 dmu_object_info_t doi; 297 int error; 298 zfsvfs_t *zfsvfs; 299 znode_t *zp; 300 301 switch (com) { 302 case _FIOFFS: 303 { 304 return (0); 305 306 /* 307 * The following two ioctls are used by bfu. Faking out, 308 * necessary to avoid bfu errors. 309 */ 310 } 311 case _FIOGDIO: 312 case _FIOSDIO: 313 { 314 return (0); 315 } 316 317 case _FIO_SEEK_DATA: 318 case _FIO_SEEK_HOLE: 319 { 320#ifdef illumos 321 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 322 return (SET_ERROR(EFAULT)); 323#else 324 off = *(offset_t *)data; 325#endif 326 zp = VTOZ(vp); 327 zfsvfs = zp->z_zfsvfs; 328 ZFS_ENTER(zfsvfs); 329 ZFS_VERIFY_ZP(zp); 330 331 /* offset parameter is in/out */ 332 error = zfs_holey(vp, com, &off); 333 ZFS_EXIT(zfsvfs); 334 if (error) 335 return (error); 336#ifdef illumos 337 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 338 return (SET_ERROR(EFAULT)); 339#else 340 *(offset_t *)data = off; 341#endif 342 return (0); 343 } 344#ifdef illumos 345 case _FIO_COUNT_FILLED: 346 { 347 /* 348 * _FIO_COUNT_FILLED adds a new ioctl command which 349 * exposes the number of filled blocks in a 350 * ZFS object. 351 */ 352 zp = VTOZ(vp); 353 zfsvfs = zp->z_zfsvfs; 354 ZFS_ENTER(zfsvfs); 355 ZFS_VERIFY_ZP(zp); 356 357 /* 358 * Wait for all dirty blocks for this object 359 * to get synced out to disk, and the DMU info 360 * updated. 361 */ 362 error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id); 363 if (error) { 364 ZFS_EXIT(zfsvfs); 365 return (error); 366 } 367 368 /* 369 * Retrieve fill count from DMU object. 370 */ 371 error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi); 372 if (error) { 373 ZFS_EXIT(zfsvfs); 374 return (error); 375 } 376 377 ndata = doi.doi_fill_count; 378 379 ZFS_EXIT(zfsvfs); 380 if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag)) 381 return (SET_ERROR(EFAULT)); 382 return (0); 383 } 384#endif 385 } 386 return (SET_ERROR(ENOTTY)); 387} 388 389static vm_page_t 390page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) 391{ 392 vm_object_t obj; 393 vm_page_t pp; 394 int64_t end; 395 396 /* 397 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE 398 * aligned boundaries, if the range is not aligned. As a result a 399 * DEV_BSIZE subrange with partially dirty data may get marked as clean. 400 * It may happen that all DEV_BSIZE subranges are marked clean and thus 401 * the whole page would be considred clean despite have some dirty data. 402 * For this reason we should shrink the range to DEV_BSIZE aligned 403 * boundaries before calling vm_page_clear_dirty. 404 */ 405 end = rounddown2(off + nbytes, DEV_BSIZE); 406 off = roundup2(off, DEV_BSIZE); 407 nbytes = end - off; 408 409 obj = vp->v_object; 410 zfs_vmobject_assert_wlocked(obj); 411 412 for (;;) { 413 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 414 pp->valid) { 415 if (vm_page_xbusied(pp)) { 416 /* 417 * Reference the page before unlocking and 418 * sleeping so that the page daemon is less 419 * likely to reclaim it. 420 */ 421 vm_page_reference(pp); 422 vm_page_lock(pp); 423 zfs_vmobject_wunlock(obj); 424 vm_page_busy_sleep(pp, "zfsmwb"); 425 zfs_vmobject_wlock(obj); 426 continue; 427 } 428 vm_page_sbusy(pp); 429 } else if (pp == NULL) { 430 pp = vm_page_alloc(obj, OFF_TO_IDX(start), 431 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED | 432 VM_ALLOC_SBUSY); 433 } else { 434 ASSERT(pp != NULL && !pp->valid); 435 pp = NULL; 436 } 437 438 if (pp != NULL) { 439 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 440 vm_object_pip_add(obj, 1); 441 pmap_remove_write(pp); 442 if (nbytes != 0) 443 vm_page_clear_dirty(pp, off, nbytes); 444 } 445 break; 446 } 447 return (pp); 448} 449 450static void 451page_unbusy(vm_page_t pp) 452{ 453 454 vm_page_sunbusy(pp); 455 vm_object_pip_subtract(pp->object, 1); 456} 457 458static vm_page_t 459page_hold(vnode_t *vp, int64_t start) 460{ 461 vm_object_t obj; 462 vm_page_t pp; 463 464 obj = vp->v_object; 465 zfs_vmobject_assert_wlocked(obj); 466 467 for (;;) { 468 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 469 pp->valid) { 470 if (vm_page_xbusied(pp)) { 471 /* 472 * Reference the page before unlocking and 473 * sleeping so that the page daemon is less 474 * likely to reclaim it. 475 */ 476 vm_page_reference(pp); 477 vm_page_lock(pp); 478 zfs_vmobject_wunlock(obj); 479 vm_page_busy_sleep(pp, "zfsmwb"); 480 zfs_vmobject_wlock(obj); 481 continue; 482 } 483 484 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 485 vm_page_lock(pp); 486 vm_page_hold(pp); 487 vm_page_unlock(pp); 488 489 } else 490 pp = NULL; 491 break; 492 } 493 return (pp); 494} 495 496static void 497page_unhold(vm_page_t pp) 498{ 499 500 vm_page_lock(pp); 501 vm_page_unhold(pp); 502 vm_page_unlock(pp); 503} 504 505/* 506 * When a file is memory mapped, we must keep the IO data synchronized 507 * between the DMU cache and the memory mapped pages. What this means: 508 * 509 * On Write: If we find a memory mapped page, we write to *both* 510 * the page and the dmu buffer. 511 */ 512static void 513update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid, 514 int segflg, dmu_tx_t *tx) 515{ 516 vm_object_t obj; 517 struct sf_buf *sf; 518 caddr_t va; 519 int off; 520 521 ASSERT(segflg != UIO_NOCOPY); 522 ASSERT(vp->v_mount != NULL); 523 obj = vp->v_object; 524 ASSERT(obj != NULL); 525 526 off = start & PAGEOFFSET; 527 zfs_vmobject_wlock(obj); 528 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 529 vm_page_t pp; 530 int nbytes = imin(PAGESIZE - off, len); 531 532 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) { 533 zfs_vmobject_wunlock(obj); 534 535 va = zfs_map_page(pp, &sf); 536 (void) dmu_read(os, oid, start+off, nbytes, 537 va+off, DMU_READ_PREFETCH);; 538 zfs_unmap_page(sf); 539 540 zfs_vmobject_wlock(obj); 541 page_unbusy(pp); 542 } 543 len -= nbytes; 544 off = 0; 545 } 546 vm_object_pip_wakeupn(obj, 0); 547 zfs_vmobject_wunlock(obj); 548} 549 550/* 551 * Read with UIO_NOCOPY flag means that sendfile(2) requests 552 * ZFS to populate a range of page cache pages with data. 553 * 554 * NOTE: this function could be optimized to pre-allocate 555 * all pages in advance, drain exclusive busy on all of them, 556 * map them into contiguous KVA region and populate them 557 * in one single dmu_read() call. 558 */ 559static int 560mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio) 561{ 562 znode_t *zp = VTOZ(vp); 563 objset_t *os = zp->z_zfsvfs->z_os; 564 struct sf_buf *sf; 565 vm_object_t obj; 566 vm_page_t pp; 567 int64_t start; 568 caddr_t va; 569 int len = nbytes; 570 int off; 571 int error = 0; 572 573 ASSERT(uio->uio_segflg == UIO_NOCOPY); 574 ASSERT(vp->v_mount != NULL); 575 obj = vp->v_object; 576 ASSERT(obj != NULL); 577 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0); 578 579 zfs_vmobject_wlock(obj); 580 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) { 581 int bytes = MIN(PAGESIZE, len); 582 583 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY | 584 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); 585 if (pp->valid == 0) { 586 zfs_vmobject_wunlock(obj); 587 va = zfs_map_page(pp, &sf); 588 error = dmu_read(os, zp->z_id, start, bytes, va, 589 DMU_READ_PREFETCH); 590 if (bytes != PAGESIZE && error == 0) 591 bzero(va + bytes, PAGESIZE - bytes); 592 zfs_unmap_page(sf); 593 zfs_vmobject_wlock(obj); 594 vm_page_sunbusy(pp); 595 vm_page_lock(pp); 596 if (error) { 597 if (pp->wire_count == 0 && pp->valid == 0 && 598 !vm_page_busied(pp)) 599 vm_page_free(pp); 600 } else { 601 pp->valid = VM_PAGE_BITS_ALL; 602 vm_page_activate(pp); 603 } 604 vm_page_unlock(pp); 605 } else { 606 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 607 vm_page_sunbusy(pp); 608 } 609 if (error) 610 break; 611 uio->uio_resid -= bytes; 612 uio->uio_offset += bytes; 613 len -= bytes; 614 } 615 zfs_vmobject_wunlock(obj); 616 return (error); 617} 618 619/* 620 * When a file is memory mapped, we must keep the IO data synchronized 621 * between the DMU cache and the memory mapped pages. What this means: 622 * 623 * On Read: We "read" preferentially from memory mapped pages, 624 * else we default from the dmu buffer. 625 * 626 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 627 * the file is memory mapped. 628 */ 629static int 630mappedread(vnode_t *vp, int nbytes, uio_t *uio) 631{ 632 znode_t *zp = VTOZ(vp); 633 vm_object_t obj; 634 int64_t start; 635 caddr_t va; 636 int len = nbytes; 637 int off; 638 int error = 0; 639 640 ASSERT(vp->v_mount != NULL); 641 obj = vp->v_object; 642 ASSERT(obj != NULL); 643 644 start = uio->uio_loffset; 645 off = start & PAGEOFFSET; 646 zfs_vmobject_wlock(obj); 647 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 648 vm_page_t pp; 649 uint64_t bytes = MIN(PAGESIZE - off, len); 650 651 if (pp = page_hold(vp, start)) { 652 struct sf_buf *sf; 653 caddr_t va; 654 655 zfs_vmobject_wunlock(obj); 656 va = zfs_map_page(pp, &sf); 657#ifdef illumos 658 error = uiomove(va + off, bytes, UIO_READ, uio); 659#else 660 error = vn_io_fault_uiomove(va + off, bytes, uio); 661#endif 662 zfs_unmap_page(sf); 663 zfs_vmobject_wlock(obj); 664 page_unhold(pp); 665 } else { 666 zfs_vmobject_wunlock(obj); 667 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 668 uio, bytes); 669 zfs_vmobject_wlock(obj); 670 } 671 len -= bytes; 672 off = 0; 673 if (error) 674 break; 675 } 676 zfs_vmobject_wunlock(obj); 677 return (error); 678} 679 680offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 681 682/* 683 * Read bytes from specified file into supplied buffer. 684 * 685 * IN: vp - vnode of file to be read from. 686 * uio - structure supplying read location, range info, 687 * and return buffer. 688 * ioflag - SYNC flags; used to provide FRSYNC semantics. 689 * cr - credentials of caller. 690 * ct - caller context 691 * 692 * OUT: uio - updated offset and range, buffer filled. 693 * 694 * RETURN: 0 on success, error code on failure. 695 * 696 * Side Effects: 697 * vp - atime updated if byte count > 0 698 */ 699/* ARGSUSED */ 700static int 701zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 702{ 703 znode_t *zp = VTOZ(vp); 704 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 705 ssize_t n, nbytes; 706 int error = 0; 707 rl_t *rl; 708 xuio_t *xuio = NULL; 709 710 ZFS_ENTER(zfsvfs); 711 ZFS_VERIFY_ZP(zp); 712 713 if (zp->z_pflags & ZFS_AV_QUARANTINED) { 714 ZFS_EXIT(zfsvfs); 715 return (SET_ERROR(EACCES)); 716 } 717 718 /* 719 * Validate file offset 720 */ 721 if (uio->uio_loffset < (offset_t)0) { 722 ZFS_EXIT(zfsvfs); 723 return (SET_ERROR(EINVAL)); 724 } 725 726 /* 727 * Fasttrack empty reads 728 */ 729 if (uio->uio_resid == 0) { 730 ZFS_EXIT(zfsvfs); 731 return (0); 732 } 733 734 /* 735 * Check for mandatory locks 736 */ 737 if (MANDMODE(zp->z_mode)) { 738 if (error = chklock(vp, FREAD, 739 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 740 ZFS_EXIT(zfsvfs); 741 return (error); 742 } 743 } 744 745 /* 746 * If we're in FRSYNC mode, sync out this znode before reading it. 747 */ 748 if (zfsvfs->z_log && 749 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)) 750 zil_commit(zfsvfs->z_log, zp->z_id); 751 752 /* 753 * Lock the range against changes. 754 */ 755 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 756 757 /* 758 * If we are reading past end-of-file we can skip 759 * to the end; but we might still need to set atime. 760 */ 761 if (uio->uio_loffset >= zp->z_size) { 762 error = 0; 763 goto out; 764 } 765 766 ASSERT(uio->uio_loffset < zp->z_size); 767 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset); 768 769#ifdef illumos 770 if ((uio->uio_extflg == UIO_XUIO) && 771 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) { 772 int nblk; 773 int blksz = zp->z_blksz; 774 uint64_t offset = uio->uio_loffset; 775 776 xuio = (xuio_t *)uio; 777 if ((ISP2(blksz))) { 778 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset, 779 blksz)) / blksz; 780 } else { 781 ASSERT(offset + n <= blksz); 782 nblk = 1; 783 } 784 (void) dmu_xuio_init(xuio, nblk); 785 786 if (vn_has_cached_data(vp)) { 787 /* 788 * For simplicity, we always allocate a full buffer 789 * even if we only expect to read a portion of a block. 790 */ 791 while (--nblk >= 0) { 792 (void) dmu_xuio_add(xuio, 793 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 794 blksz), 0, blksz); 795 } 796 } 797 } 798#endif /* illumos */ 799 800 while (n > 0) { 801 nbytes = MIN(n, zfs_read_chunk_size - 802 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 803 804#ifdef __FreeBSD__ 805 if (uio->uio_segflg == UIO_NOCOPY) 806 error = mappedread_sf(vp, nbytes, uio); 807 else 808#endif /* __FreeBSD__ */ 809 if (vn_has_cached_data(vp)) { 810 error = mappedread(vp, nbytes, uio); 811 } else { 812 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 813 uio, nbytes); 814 } 815 if (error) { 816 /* convert checksum errors into IO errors */ 817 if (error == ECKSUM) 818 error = SET_ERROR(EIO); 819 break; 820 } 821 822 n -= nbytes; 823 } 824out: 825 zfs_range_unlock(rl); 826 827 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 828 ZFS_EXIT(zfsvfs); 829 return (error); 830} 831 832/* 833 * Write the bytes to a file. 834 * 835 * IN: vp - vnode of file to be written to. 836 * uio - structure supplying write location, range info, 837 * and data buffer. 838 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is 839 * set if in append mode. 840 * cr - credentials of caller. 841 * ct - caller context (NFS/CIFS fem monitor only) 842 * 843 * OUT: uio - updated offset and range. 844 * 845 * RETURN: 0 on success, error code on failure. 846 * 847 * Timestamps: 848 * vp - ctime|mtime updated if byte count > 0 849 */ 850 851/* ARGSUSED */ 852static int 853zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 854{ 855 znode_t *zp = VTOZ(vp); 856 rlim64_t limit = MAXOFFSET_T; 857 ssize_t start_resid = uio->uio_resid; 858 ssize_t tx_bytes; 859 uint64_t end_size; 860 dmu_tx_t *tx; 861 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 862 zilog_t *zilog; 863 offset_t woff; 864 ssize_t n, nbytes; 865 rl_t *rl; 866 int max_blksz = zfsvfs->z_max_blksz; 867 int error = 0; 868 arc_buf_t *abuf; 869 iovec_t *aiov = NULL; 870 xuio_t *xuio = NULL; 871 int i_iov = 0; 872 int iovcnt = uio->uio_iovcnt; 873 iovec_t *iovp = uio->uio_iov; 874 int write_eof; 875 int count = 0; 876 sa_bulk_attr_t bulk[4]; 877 uint64_t mtime[2], ctime[2]; 878 879 /* 880 * Fasttrack empty write 881 */ 882 n = start_resid; 883 if (n == 0) 884 return (0); 885 886 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 887 limit = MAXOFFSET_T; 888 889 ZFS_ENTER(zfsvfs); 890 ZFS_VERIFY_ZP(zp); 891 892 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 893 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 894 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 895 &zp->z_size, 8); 896 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 897 &zp->z_pflags, 8); 898 899 /* 900 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our 901 * callers might not be able to detect properly that we are read-only, 902 * so check it explicitly here. 903 */ 904 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 905 ZFS_EXIT(zfsvfs); 906 return (SET_ERROR(EROFS)); 907 } 908 909 /* 910 * If immutable or not appending then return EPERM 911 */ 912 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 913 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 914 (uio->uio_loffset < zp->z_size))) { 915 ZFS_EXIT(zfsvfs); 916 return (SET_ERROR(EPERM)); 917 } 918 919 zilog = zfsvfs->z_log; 920 921 /* 922 * Validate file offset 923 */ 924 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset; 925 if (woff < 0) { 926 ZFS_EXIT(zfsvfs); 927 return (SET_ERROR(EINVAL)); 928 } 929 930 /* 931 * Check for mandatory locks before calling zfs_range_lock() 932 * in order to prevent a deadlock with locks set via fcntl(). 933 */ 934 if (MANDMODE((mode_t)zp->z_mode) && 935 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 936 ZFS_EXIT(zfsvfs); 937 return (error); 938 } 939 940#ifdef illumos 941 /* 942 * Pre-fault the pages to ensure slow (eg NFS) pages 943 * don't hold up txg. 944 * Skip this if uio contains loaned arc_buf. 945 */ 946 if ((uio->uio_extflg == UIO_XUIO) && 947 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) 948 xuio = (xuio_t *)uio; 949 else 950 uio_prefaultpages(MIN(n, max_blksz), uio); 951#endif 952 953 /* 954 * If in append mode, set the io offset pointer to eof. 955 */ 956 if (ioflag & FAPPEND) { 957 /* 958 * Obtain an appending range lock to guarantee file append 959 * semantics. We reset the write offset once we have the lock. 960 */ 961 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 962 woff = rl->r_off; 963 if (rl->r_len == UINT64_MAX) { 964 /* 965 * We overlocked the file because this write will cause 966 * the file block size to increase. 967 * Note that zp_size cannot change with this lock held. 968 */ 969 woff = zp->z_size; 970 } 971 uio->uio_loffset = woff; 972 } else { 973 /* 974 * Note that if the file block size will change as a result of 975 * this write, then this range lock will lock the entire file 976 * so that we can re-write the block safely. 977 */ 978 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 979 } 980 981 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) { 982 zfs_range_unlock(rl); 983 ZFS_EXIT(zfsvfs); 984 return (EFBIG); 985 } 986 987 if (woff >= limit) { 988 zfs_range_unlock(rl); 989 ZFS_EXIT(zfsvfs); 990 return (SET_ERROR(EFBIG)); 991 } 992 993 if ((woff + n) > limit || woff > (limit - n)) 994 n = limit - woff; 995 996 /* Will this write extend the file length? */ 997 write_eof = (woff + n > zp->z_size); 998 999 end_size = MAX(zp->z_size, woff + n); 1000 1001 /* 1002 * Write the file in reasonable size chunks. Each chunk is written 1003 * in a separate transaction; this keeps the intent log records small 1004 * and allows us to do more fine-grained space accounting. 1005 */ 1006 while (n > 0) { 1007 abuf = NULL; 1008 woff = uio->uio_loffset; 1009 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 1010 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 1011 if (abuf != NULL) 1012 dmu_return_arcbuf(abuf); 1013 error = SET_ERROR(EDQUOT); 1014 break; 1015 } 1016 1017 if (xuio && abuf == NULL) { 1018 ASSERT(i_iov < iovcnt); 1019 aiov = &iovp[i_iov]; 1020 abuf = dmu_xuio_arcbuf(xuio, i_iov); 1021 dmu_xuio_clear(xuio, i_iov); 1022 DTRACE_PROBE3(zfs_cp_write, int, i_iov, 1023 iovec_t *, aiov, arc_buf_t *, abuf); 1024 ASSERT((aiov->iov_base == abuf->b_data) || 1025 ((char *)aiov->iov_base - (char *)abuf->b_data + 1026 aiov->iov_len == arc_buf_size(abuf))); 1027 i_iov++; 1028 } else if (abuf == NULL && n >= max_blksz && 1029 woff >= zp->z_size && 1030 P2PHASE(woff, max_blksz) == 0 && 1031 zp->z_blksz == max_blksz) { 1032 /* 1033 * This write covers a full block. "Borrow" a buffer 1034 * from the dmu so that we can fill it before we enter 1035 * a transaction. This avoids the possibility of 1036 * holding up the transaction if the data copy hangs 1037 * up on a pagefault (e.g., from an NFS server mapping). 1038 */ 1039#ifdef illumos 1040 size_t cbytes; 1041#endif 1042 1043 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 1044 max_blksz); 1045 ASSERT(abuf != NULL); 1046 ASSERT(arc_buf_size(abuf) == max_blksz); 1047#ifdef illumos 1048 if (error = uiocopy(abuf->b_data, max_blksz, 1049 UIO_WRITE, uio, &cbytes)) { 1050 dmu_return_arcbuf(abuf); 1051 break; 1052 } 1053 ASSERT(cbytes == max_blksz); 1054#else 1055 ssize_t resid = uio->uio_resid; 1056 error = vn_io_fault_uiomove(abuf->b_data, max_blksz, uio); 1057 if (error != 0) { 1058 uio->uio_offset -= resid - uio->uio_resid; 1059 uio->uio_resid = resid; 1060 dmu_return_arcbuf(abuf); 1061 break; 1062 } 1063#endif 1064 } 1065 1066 /* 1067 * Start a transaction. 1068 */ 1069 tx = dmu_tx_create(zfsvfs->z_os); 1070 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1071 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 1072 zfs_sa_upgrade_txholds(tx, zp); 1073 error = dmu_tx_assign(tx, TXG_WAIT); 1074 if (error) { 1075 dmu_tx_abort(tx); 1076 if (abuf != NULL) 1077 dmu_return_arcbuf(abuf); 1078 break; 1079 } 1080 1081 /* 1082 * If zfs_range_lock() over-locked we grow the blocksize 1083 * and then reduce the lock range. This will only happen 1084 * on the first iteration since zfs_range_reduce() will 1085 * shrink down r_len to the appropriate size. 1086 */ 1087 if (rl->r_len == UINT64_MAX) { 1088 uint64_t new_blksz; 1089 1090 if (zp->z_blksz > max_blksz) { 1091 /* 1092 * File's blocksize is already larger than the 1093 * "recordsize" property. Only let it grow to 1094 * the next power of 2. 1095 */ 1096 ASSERT(!ISP2(zp->z_blksz)); 1097 new_blksz = MIN(end_size, 1098 1 << highbit64(zp->z_blksz)); 1099 } else { 1100 new_blksz = MIN(end_size, max_blksz); 1101 } 1102 zfs_grow_blocksize(zp, new_blksz, tx); 1103 zfs_range_reduce(rl, woff, n); 1104 } 1105 1106 /* 1107 * XXX - should we really limit each write to z_max_blksz? 1108 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 1109 */ 1110 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 1111 1112 if (woff + nbytes > zp->z_size) 1113 vnode_pager_setsize(vp, woff + nbytes); 1114 1115 if (abuf == NULL) { 1116 tx_bytes = uio->uio_resid; 1117 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), 1118 uio, nbytes, tx); 1119 tx_bytes -= uio->uio_resid; 1120 } else { 1121 tx_bytes = nbytes; 1122 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len); 1123 /* 1124 * If this is not a full block write, but we are 1125 * extending the file past EOF and this data starts 1126 * block-aligned, use assign_arcbuf(). Otherwise, 1127 * write via dmu_write(). 1128 */ 1129 if (tx_bytes < max_blksz && (!write_eof || 1130 aiov->iov_base != abuf->b_data)) { 1131 ASSERT(xuio); 1132 dmu_write(zfsvfs->z_os, zp->z_id, woff, 1133 aiov->iov_len, aiov->iov_base, tx); 1134 dmu_return_arcbuf(abuf); 1135 xuio_stat_wbuf_copied(); 1136 } else { 1137 ASSERT(xuio || tx_bytes == max_blksz); 1138 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), 1139 woff, abuf, tx); 1140 } 1141#ifdef illumos 1142 ASSERT(tx_bytes <= uio->uio_resid); 1143 uioskip(uio, tx_bytes); 1144#endif 1145 } 1146 if (tx_bytes && vn_has_cached_data(vp)) { 1147 update_pages(vp, woff, tx_bytes, zfsvfs->z_os, 1148 zp->z_id, uio->uio_segflg, tx); 1149 } 1150 1151 /* 1152 * If we made no progress, we're done. If we made even 1153 * partial progress, update the znode and ZIL accordingly. 1154 */ 1155 if (tx_bytes == 0) { 1156 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 1157 (void *)&zp->z_size, sizeof (uint64_t), tx); 1158 dmu_tx_commit(tx); 1159 ASSERT(error != 0); 1160 break; 1161 } 1162 1163 /* 1164 * Clear Set-UID/Set-GID bits on successful write if not 1165 * privileged and at least one of the excute bits is set. 1166 * 1167 * It would be nice to to this after all writes have 1168 * been done, but that would still expose the ISUID/ISGID 1169 * to another app after the partial write is committed. 1170 * 1171 * Note: we don't call zfs_fuid_map_id() here because 1172 * user 0 is not an ephemeral uid. 1173 */ 1174 mutex_enter(&zp->z_acl_lock); 1175 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | 1176 (S_IXUSR >> 6))) != 0 && 1177 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && 1178 secpolicy_vnode_setid_retain(vp, cr, 1179 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) { 1180 uint64_t newmode; 1181 zp->z_mode &= ~(S_ISUID | S_ISGID); 1182 newmode = zp->z_mode; 1183 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), 1184 (void *)&newmode, sizeof (uint64_t), tx); 1185 } 1186 mutex_exit(&zp->z_acl_lock); 1187 1188 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 1189 B_TRUE); 1190 1191 /* 1192 * Update the file size (zp_size) if it has changed; 1193 * account for possible concurrent updates. 1194 */ 1195 while ((end_size = zp->z_size) < uio->uio_loffset) { 1196 (void) atomic_cas_64(&zp->z_size, end_size, 1197 uio->uio_loffset); 1198#ifdef illumos 1199 ASSERT(error == 0); 1200#else 1201 ASSERT(error == 0 || error == EFAULT); 1202#endif 1203 } 1204 /* 1205 * If we are replaying and eof is non zero then force 1206 * the file size to the specified eof. Note, there's no 1207 * concurrency during replay. 1208 */ 1209 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) 1210 zp->z_size = zfsvfs->z_replay_eof; 1211 1212 if (error == 0) 1213 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1214 else 1215 (void) sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1216 1217 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 1218 dmu_tx_commit(tx); 1219 1220 if (error != 0) 1221 break; 1222 ASSERT(tx_bytes == nbytes); 1223 n -= nbytes; 1224 1225#ifdef illumos 1226 if (!xuio && n > 0) 1227 uio_prefaultpages(MIN(n, max_blksz), uio); 1228#endif 1229 } 1230 1231 zfs_range_unlock(rl); 1232 1233 /* 1234 * If we're in replay mode, or we made no progress, return error. 1235 * Otherwise, it's at least a partial write, so it's successful. 1236 */ 1237 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 1238 ZFS_EXIT(zfsvfs); 1239 return (error); 1240 } 1241 1242#ifdef __FreeBSD__ 1243 /* 1244 * EFAULT means that at least one page of the source buffer was not 1245 * available. VFS will re-try remaining I/O upon this error. 1246 */ 1247 if (error == EFAULT) { 1248 ZFS_EXIT(zfsvfs); 1249 return (error); 1250 } 1251#endif 1252 1253 if (ioflag & (FSYNC | FDSYNC) || 1254 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1255 zil_commit(zilog, zp->z_id); 1256 1257 ZFS_EXIT(zfsvfs); 1258 return (0); 1259} 1260 1261void 1262zfs_get_done(zgd_t *zgd, int error) 1263{ 1264 znode_t *zp = zgd->zgd_private; 1265 objset_t *os = zp->z_zfsvfs->z_os; 1266 1267 if (zgd->zgd_db) 1268 dmu_buf_rele(zgd->zgd_db, zgd); 1269 1270 zfs_range_unlock(zgd->zgd_rl); 1271 1272 /* 1273 * Release the vnode asynchronously as we currently have the 1274 * txg stopped from syncing. 1275 */ 1276 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1277 1278 if (error == 0 && zgd->zgd_bp) 1279 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1280 1281 kmem_free(zgd, sizeof (zgd_t)); 1282} 1283 1284#ifdef DEBUG 1285static int zil_fault_io = 0; 1286#endif 1287 1288/* 1289 * Get data to generate a TX_WRITE intent log record. 1290 */ 1291int 1292zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1293{ 1294 zfsvfs_t *zfsvfs = arg; 1295 objset_t *os = zfsvfs->z_os; 1296 znode_t *zp; 1297 uint64_t object = lr->lr_foid; 1298 uint64_t offset = lr->lr_offset; 1299 uint64_t size = lr->lr_length; 1300 blkptr_t *bp = &lr->lr_blkptr; 1301 dmu_buf_t *db; 1302 zgd_t *zgd; 1303 int error = 0; 1304 1305 ASSERT(zio != NULL); 1306 ASSERT(size != 0); 1307 1308 /* 1309 * Nothing to do if the file has been removed 1310 */ 1311 if (zfs_zget(zfsvfs, object, &zp) != 0) 1312 return (SET_ERROR(ENOENT)); 1313 if (zp->z_unlinked) { 1314 /* 1315 * Release the vnode asynchronously as we currently have the 1316 * txg stopped from syncing. 1317 */ 1318 VN_RELE_ASYNC(ZTOV(zp), 1319 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1320 return (SET_ERROR(ENOENT)); 1321 } 1322 1323 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1324 zgd->zgd_zilog = zfsvfs->z_log; 1325 zgd->zgd_private = zp; 1326 1327 /* 1328 * Write records come in two flavors: immediate and indirect. 1329 * For small writes it's cheaper to store the data with the 1330 * log record (immediate); for large writes it's cheaper to 1331 * sync the data and get a pointer to it (indirect) so that 1332 * we don't have to write the data twice. 1333 */ 1334 if (buf != NULL) { /* immediate write */ 1335 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER); 1336 /* test for truncation needs to be done while range locked */ 1337 if (offset >= zp->z_size) { 1338 error = SET_ERROR(ENOENT); 1339 } else { 1340 error = dmu_read(os, object, offset, size, buf, 1341 DMU_READ_NO_PREFETCH); 1342 } 1343 ASSERT(error == 0 || error == ENOENT); 1344 } else { /* indirect write */ 1345 /* 1346 * Have to lock the whole block to ensure when it's 1347 * written out and it's checksum is being calculated 1348 * that no one can change the data. We need to re-check 1349 * blocksize after we get the lock in case it's changed! 1350 */ 1351 for (;;) { 1352 uint64_t blkoff; 1353 size = zp->z_blksz; 1354 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; 1355 offset -= blkoff; 1356 zgd->zgd_rl = zfs_range_lock(zp, offset, size, 1357 RL_READER); 1358 if (zp->z_blksz == size) 1359 break; 1360 offset += blkoff; 1361 zfs_range_unlock(zgd->zgd_rl); 1362 } 1363 /* test for truncation needs to be done while range locked */ 1364 if (lr->lr_offset >= zp->z_size) 1365 error = SET_ERROR(ENOENT); 1366#ifdef DEBUG 1367 if (zil_fault_io) { 1368 error = SET_ERROR(EIO); 1369 zil_fault_io = 0; 1370 } 1371#endif 1372 if (error == 0) 1373 error = dmu_buf_hold(os, object, offset, zgd, &db, 1374 DMU_READ_NO_PREFETCH); 1375 1376 if (error == 0) { 1377 blkptr_t *obp = dmu_buf_get_blkptr(db); 1378 if (obp) { 1379 ASSERT(BP_IS_HOLE(bp)); 1380 *bp = *obp; 1381 } 1382 1383 zgd->zgd_db = db; 1384 zgd->zgd_bp = bp; 1385 1386 ASSERT(db->db_offset == offset); 1387 ASSERT(db->db_size == size); 1388 1389 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1390 zfs_get_done, zgd); 1391 ASSERT(error || lr->lr_length <= zp->z_blksz); 1392 1393 /* 1394 * On success, we need to wait for the write I/O 1395 * initiated by dmu_sync() to complete before we can 1396 * release this dbuf. We will finish everything up 1397 * in the zfs_get_done() callback. 1398 */ 1399 if (error == 0) 1400 return (0); 1401 1402 if (error == EALREADY) { 1403 lr->lr_common.lrc_txtype = TX_WRITE2; 1404 error = 0; 1405 } 1406 } 1407 } 1408 1409 zfs_get_done(zgd, error); 1410 1411 return (error); 1412} 1413 1414/*ARGSUSED*/ 1415static int 1416zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 1417 caller_context_t *ct) 1418{ 1419 znode_t *zp = VTOZ(vp); 1420 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1421 int error; 1422 1423 ZFS_ENTER(zfsvfs); 1424 ZFS_VERIFY_ZP(zp); 1425 1426 if (flag & V_ACE_MASK) 1427 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 1428 else 1429 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1430 1431 ZFS_EXIT(zfsvfs); 1432 return (error); 1433} 1434 1435static int 1436zfs_dd_callback(struct mount *mp, void *arg, int lkflags, struct vnode **vpp) 1437{ 1438 int error; 1439 1440 *vpp = arg; 1441 error = vn_lock(*vpp, lkflags); 1442 if (error != 0) 1443 vrele(*vpp); 1444 return (error); 1445} 1446 1447static int 1448zfs_lookup_lock(vnode_t *dvp, vnode_t *vp, const char *name, int lkflags) 1449{ 1450 znode_t *zdp = VTOZ(dvp); 1451 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1452 int error; 1453 int ltype; 1454 1455 ASSERT_VOP_LOCKED(dvp, __func__); 1456#ifdef DIAGNOSTIC 1457 ASSERT(!RRM_LOCK_HELD(&zfsvfs->z_teardown_lock)); 1458#endif 1459 1460 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) { 1461 ASSERT3P(dvp, ==, vp); 1462 vref(dvp); 1463 ltype = lkflags & LK_TYPE_MASK; 1464 if (ltype != VOP_ISLOCKED(dvp)) { 1465 if (ltype == LK_EXCLUSIVE) 1466 vn_lock(dvp, LK_UPGRADE | LK_RETRY); 1467 else /* if (ltype == LK_SHARED) */ 1468 vn_lock(dvp, LK_DOWNGRADE | LK_RETRY); 1469 1470 /* 1471 * Relock for the "." case could leave us with 1472 * reclaimed vnode. 1473 */ 1474 if (dvp->v_iflag & VI_DOOMED) { 1475 vrele(dvp); 1476 return (SET_ERROR(ENOENT)); 1477 } 1478 } 1479 return (0); 1480 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) { 1481 /* 1482 * Note that in this case, dvp is the child vnode, and we 1483 * are looking up the parent vnode - exactly reverse from 1484 * normal operation. Unlocking dvp requires some rather 1485 * tricky unlock/relock dance to prevent mp from being freed; 1486 * use vn_vget_ino_gen() which takes care of all that. 1487 * 1488 * XXX Note that there is a time window when both vnodes are 1489 * unlocked. It is possible, although highly unlikely, that 1490 * during that window the parent-child relationship between 1491 * the vnodes may change, for example, get reversed. 1492 * In that case we would have a wrong lock order for the vnodes. 1493 * All other filesystems seem to ignore this problem, so we 1494 * do the same here. 1495 * A potential solution could be implemented as follows: 1496 * - using LK_NOWAIT when locking the second vnode and retrying 1497 * if necessary 1498 * - checking that the parent-child relationship still holds 1499 * after locking both vnodes and retrying if it doesn't 1500 */ 1501 error = vn_vget_ino_gen(dvp, zfs_dd_callback, vp, lkflags, &vp); 1502 return (error); 1503 } else { 1504 error = vn_lock(vp, lkflags); 1505 if (error != 0) 1506 vrele(vp); 1507 return (error); 1508 } 1509} 1510 1511/* 1512 * Lookup an entry in a directory, or an extended attribute directory. 1513 * If it exists, return a held vnode reference for it. 1514 * 1515 * IN: dvp - vnode of directory to search. 1516 * nm - name of entry to lookup. 1517 * pnp - full pathname to lookup [UNUSED]. 1518 * flags - LOOKUP_XATTR set if looking for an attribute. 1519 * rdir - root directory vnode [UNUSED]. 1520 * cr - credentials of caller. 1521 * ct - caller context 1522 * 1523 * OUT: vpp - vnode of located entry, NULL if not found. 1524 * 1525 * RETURN: 0 on success, error code on failure. 1526 * 1527 * Timestamps: 1528 * NA 1529 */ 1530/* ARGSUSED */ 1531static int 1532zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 1533 int nameiop, cred_t *cr, kthread_t *td, int flags) 1534{ 1535 znode_t *zdp = VTOZ(dvp); 1536 znode_t *zp; 1537 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1538 int error = 0; 1539 1540 /* fast path (should be redundant with vfs namecache) */ 1541 if (!(flags & LOOKUP_XATTR)) { 1542 if (dvp->v_type != VDIR) { 1543 return (SET_ERROR(ENOTDIR)); 1544 } else if (zdp->z_sa_hdl == NULL) { 1545 return (SET_ERROR(EIO)); 1546 } 1547 } 1548 1549 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1550 1551 ZFS_ENTER(zfsvfs); 1552 ZFS_VERIFY_ZP(zdp); 1553 1554 *vpp = NULL; 1555 1556 if (flags & LOOKUP_XATTR) { 1557#ifdef TODO 1558 /* 1559 * If the xattr property is off, refuse the lookup request. 1560 */ 1561 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1562 ZFS_EXIT(zfsvfs); 1563 return (SET_ERROR(EINVAL)); 1564 } 1565#endif 1566 1567 /* 1568 * We don't allow recursive attributes.. 1569 * Maybe someday we will. 1570 */ 1571 if (zdp->z_pflags & ZFS_XATTR) { 1572 ZFS_EXIT(zfsvfs); 1573 return (SET_ERROR(EINVAL)); 1574 } 1575 1576 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1577 ZFS_EXIT(zfsvfs); 1578 return (error); 1579 } 1580 1581 /* 1582 * Do we have permission to get into attribute directory? 1583 */ 1584 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1585 B_FALSE, cr)) { 1586 vrele(*vpp); 1587 *vpp = NULL; 1588 } 1589 1590 ZFS_EXIT(zfsvfs); 1591 return (error); 1592 } 1593 1594 /* 1595 * Check accessibility of directory. 1596 */ 1597 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1598 ZFS_EXIT(zfsvfs); 1599 return (error); 1600 } 1601 1602 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1603 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1604 ZFS_EXIT(zfsvfs); 1605 return (SET_ERROR(EILSEQ)); 1606 } 1607 1608 1609 /* 1610 * First handle the special cases. 1611 */ 1612 if ((cnp->cn_flags & ISDOTDOT) != 0) { 1613 /* 1614 * If we are a snapshot mounted under .zfs, return 1615 * the vp for the snapshot directory. 1616 */ 1617 if (zdp->z_id == zfsvfs->z_root && zfsvfs->z_parent != zfsvfs) { 1618 error = zfsctl_root_lookup(zfsvfs->z_parent->z_ctldir, 1619 "snapshot", vpp, NULL, 0, NULL, kcred, 1620 NULL, NULL, NULL); 1621 ZFS_EXIT(zfsvfs); 1622 if (error == 0) { 1623 error = zfs_lookup_lock(dvp, *vpp, nm, 1624 cnp->cn_lkflags); 1625 } 1626 goto out; 1627 } 1628 } 1629 if (zfs_has_ctldir(zdp) && strcmp(nm, ZFS_CTLDIR_NAME) == 0) { 1630 error = 0; 1631 if ((cnp->cn_flags & ISLASTCN) != 0 && nameiop != LOOKUP) 1632 error = SET_ERROR(ENOTSUP); 1633 else 1634 *vpp = zfsctl_root(zdp); 1635 ZFS_EXIT(zfsvfs); 1636 if (error == 0) 1637 error = zfs_lookup_lock(dvp, *vpp, nm, cnp->cn_lkflags); 1638 goto out; 1639 } 1640 1641 /* 1642 * The loop is retry the lookup if the parent-child relationship 1643 * changes during the dot-dot locking complexities. 1644 */ 1645 for (;;) { 1646 uint64_t parent; 1647 1648 error = zfs_dirlook(zdp, nm, &zp); 1649 if (error == 0) 1650 *vpp = ZTOV(zp); 1651 1652 ZFS_EXIT(zfsvfs); 1653 if (error != 0) 1654 break; 1655 1656 error = zfs_lookup_lock(dvp, *vpp, nm, cnp->cn_lkflags); 1657 if (error != 0) { 1658 /* 1659 * If we've got a locking error, then the vnode 1660 * got reclaimed because of a force unmount. 1661 * We never enter doomed vnodes into the name cache. 1662 */ 1663 *vpp = NULL; 1664 return (error); 1665 } 1666 1667 if ((cnp->cn_flags & ISDOTDOT) == 0) 1668 break; 1669 1670 ZFS_ENTER(zfsvfs); 1671 if (zdp->z_sa_hdl == NULL) { 1672 error = SET_ERROR(EIO); 1673 } else { 1674 error = sa_lookup(zdp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 1675 &parent, sizeof (parent)); 1676 } 1677 if (error != 0) { 1678 ZFS_EXIT(zfsvfs); 1679 vput(ZTOV(zp)); 1680 break; 1681 } 1682 if (zp->z_id == parent) { 1683 ZFS_EXIT(zfsvfs); 1684 break; 1685 } 1686 vput(ZTOV(zp)); 1687 } 1688 1689out: 1690 if (error != 0) 1691 *vpp = NULL; 1692 1693 /* Translate errors and add SAVENAME when needed. */ 1694 if (cnp->cn_flags & ISLASTCN) { 1695 switch (nameiop) { 1696 case CREATE: 1697 case RENAME: 1698 if (error == ENOENT) { 1699 error = EJUSTRETURN; 1700 cnp->cn_flags |= SAVENAME; 1701 break; 1702 } 1703 /* FALLTHROUGH */ 1704 case DELETE: 1705 if (error == 0) 1706 cnp->cn_flags |= SAVENAME; 1707 break; 1708 } 1709 } 1710 1711 /* Insert name into cache (as non-existent) if appropriate. */ 1712 if (zfsvfs->z_use_namecache && 1713 error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0) 1714 cache_enter(dvp, NULL, cnp); 1715 1716 /* Insert name into cache if appropriate. */ 1717 if (zfsvfs->z_use_namecache && 1718 error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1719 if (!(cnp->cn_flags & ISLASTCN) || 1720 (nameiop != DELETE && nameiop != RENAME)) { 1721 cache_enter(dvp, *vpp, cnp); 1722 } 1723 } 1724 1725 return (error); 1726} 1727 1728/* 1729 * Attempt to create a new entry in a directory. If the entry 1730 * already exists, truncate the file if permissible, else return 1731 * an error. Return the vp of the created or trunc'd file. 1732 * 1733 * IN: dvp - vnode of directory to put new file entry in. 1734 * name - name of new file entry. 1735 * vap - attributes of new file. 1736 * excl - flag indicating exclusive or non-exclusive mode. 1737 * mode - mode to open file with. 1738 * cr - credentials of caller. 1739 * flag - large file flag [UNUSED]. 1740 * ct - caller context 1741 * vsecp - ACL to be set 1742 * 1743 * OUT: vpp - vnode of created or trunc'd entry. 1744 * 1745 * RETURN: 0 on success, error code on failure. 1746 * 1747 * Timestamps: 1748 * dvp - ctime|mtime updated if new entry created 1749 * vp - ctime|mtime always, atime if new 1750 */ 1751 1752/* ARGSUSED */ 1753static int 1754zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1755 vnode_t **vpp, cred_t *cr, kthread_t *td) 1756{ 1757 znode_t *zp, *dzp = VTOZ(dvp); 1758 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1759 zilog_t *zilog; 1760 objset_t *os; 1761 dmu_tx_t *tx; 1762 int error; 1763 ksid_t *ksid; 1764 uid_t uid; 1765 gid_t gid = crgetgid(cr); 1766 zfs_acl_ids_t acl_ids; 1767 boolean_t fuid_dirtied; 1768 void *vsecp = NULL; 1769 int flag = 0; 1770 uint64_t txtype; 1771 1772 /* 1773 * If we have an ephemeral id, ACL, or XVATTR then 1774 * make sure file system is at proper version 1775 */ 1776 1777 ksid = crgetsid(cr, KSID_OWNER); 1778 if (ksid) 1779 uid = ksid_getid(ksid); 1780 else 1781 uid = crgetuid(cr); 1782 1783 if (zfsvfs->z_use_fuids == B_FALSE && 1784 (vsecp || (vap->va_mask & AT_XVATTR) || 1785 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1786 return (SET_ERROR(EINVAL)); 1787 1788 ZFS_ENTER(zfsvfs); 1789 ZFS_VERIFY_ZP(dzp); 1790 os = zfsvfs->z_os; 1791 zilog = zfsvfs->z_log; 1792 1793 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1794 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1795 ZFS_EXIT(zfsvfs); 1796 return (SET_ERROR(EILSEQ)); 1797 } 1798 1799 if (vap->va_mask & AT_XVATTR) { 1800 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 1801 crgetuid(cr), cr, vap->va_type)) != 0) { 1802 ZFS_EXIT(zfsvfs); 1803 return (error); 1804 } 1805 } 1806 1807 *vpp = NULL; 1808 1809 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) 1810 vap->va_mode &= ~S_ISVTX; 1811 1812 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW); 1813 if (error) { 1814 ZFS_EXIT(zfsvfs); 1815 return (error); 1816 } 1817 ASSERT3P(zp, ==, NULL); 1818 1819 /* 1820 * Create a new file object and update the directory 1821 * to reference it. 1822 */ 1823 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1824 goto out; 1825 } 1826 1827 /* 1828 * We only support the creation of regular files in 1829 * extended attribute directories. 1830 */ 1831 1832 if ((dzp->z_pflags & ZFS_XATTR) && 1833 (vap->va_type != VREG)) { 1834 error = SET_ERROR(EINVAL); 1835 goto out; 1836 } 1837 1838 if ((error = zfs_acl_ids_create(dzp, 0, vap, 1839 cr, vsecp, &acl_ids)) != 0) 1840 goto out; 1841 1842 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1843 zfs_acl_ids_free(&acl_ids); 1844 error = SET_ERROR(EDQUOT); 1845 goto out; 1846 } 1847 1848 getnewvnode_reserve(1); 1849 1850 tx = dmu_tx_create(os); 1851 1852 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1853 ZFS_SA_BASE_ATTR_SIZE); 1854 1855 fuid_dirtied = zfsvfs->z_fuid_dirty; 1856 if (fuid_dirtied) 1857 zfs_fuid_txhold(zfsvfs, tx); 1858 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1859 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1860 if (!zfsvfs->z_use_sa && 1861 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1862 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1863 0, acl_ids.z_aclp->z_acl_bytes); 1864 } 1865 error = dmu_tx_assign(tx, TXG_WAIT); 1866 if (error) { 1867 zfs_acl_ids_free(&acl_ids); 1868 dmu_tx_abort(tx); 1869 getnewvnode_drop_reserve(); 1870 ZFS_EXIT(zfsvfs); 1871 return (error); 1872 } 1873 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1874 1875 if (fuid_dirtied) 1876 zfs_fuid_sync(zfsvfs, tx); 1877 1878 (void) zfs_link_create(dzp, name, zp, tx, ZNEW); 1879 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1880 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1881 vsecp, acl_ids.z_fuidp, vap); 1882 zfs_acl_ids_free(&acl_ids); 1883 dmu_tx_commit(tx); 1884 1885 getnewvnode_drop_reserve(); 1886 1887out: 1888 if (error == 0) { 1889 *vpp = ZTOV(zp); 1890 } 1891 1892 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1893 zil_commit(zilog, 0); 1894 1895 ZFS_EXIT(zfsvfs); 1896 return (error); 1897} 1898 1899/* 1900 * Remove an entry from a directory. 1901 * 1902 * IN: dvp - vnode of directory to remove entry from. 1903 * name - name of entry to remove. 1904 * cr - credentials of caller. 1905 * ct - caller context 1906 * flags - case flags 1907 * 1908 * RETURN: 0 on success, error code on failure. 1909 * 1910 * Timestamps: 1911 * dvp - ctime|mtime 1912 * vp - ctime (if nlink > 0) 1913 */ 1914 1915/*ARGSUSED*/ 1916static int 1917zfs_remove(vnode_t *dvp, vnode_t *vp, char *name, cred_t *cr) 1918{ 1919 znode_t *dzp = VTOZ(dvp); 1920 znode_t *zp = VTOZ(vp); 1921 znode_t *xzp; 1922 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1923 zilog_t *zilog; 1924 uint64_t acl_obj, xattr_obj; 1925 uint64_t obj = 0; 1926 dmu_tx_t *tx; 1927 boolean_t unlinked, toobig = FALSE; 1928 uint64_t txtype; 1929 int error; 1930 1931 ZFS_ENTER(zfsvfs); 1932 ZFS_VERIFY_ZP(dzp); 1933 ZFS_VERIFY_ZP(zp); 1934 zilog = zfsvfs->z_log; 1935 zp = VTOZ(vp); 1936 1937 xattr_obj = 0; 1938 xzp = NULL; 1939 1940 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1941 goto out; 1942 } 1943 1944 /* 1945 * Need to use rmdir for removing directories. 1946 */ 1947 if (vp->v_type == VDIR) { 1948 error = SET_ERROR(EPERM); 1949 goto out; 1950 } 1951 1952 vnevent_remove(vp, dvp, name, ct); 1953 1954 obj = zp->z_id; 1955 1956 /* are there any extended attributes? */ 1957 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1958 &xattr_obj, sizeof (xattr_obj)); 1959 if (error == 0 && xattr_obj) { 1960 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1961 ASSERT0(error); 1962 } 1963 1964 /* 1965 * We may delete the znode now, or we may put it in the unlinked set; 1966 * it depends on whether we're the last link, and on whether there are 1967 * other holds on the vnode. So we dmu_tx_hold() the right things to 1968 * allow for either case. 1969 */ 1970 tx = dmu_tx_create(zfsvfs->z_os); 1971 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1972 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1973 zfs_sa_upgrade_txholds(tx, zp); 1974 zfs_sa_upgrade_txholds(tx, dzp); 1975 1976 if (xzp) { 1977 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1978 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1979 } 1980 1981 /* charge as an update -- would be nice not to charge at all */ 1982 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1983 1984 /* 1985 * Mark this transaction as typically resulting in a net free of space 1986 */ 1987 dmu_tx_mark_netfree(tx); 1988 1989 error = dmu_tx_assign(tx, TXG_WAIT); 1990 if (error) { 1991 dmu_tx_abort(tx); 1992 ZFS_EXIT(zfsvfs); 1993 return (error); 1994 } 1995 1996 /* 1997 * Remove the directory entry. 1998 */ 1999 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, &unlinked); 2000 2001 if (error) { 2002 dmu_tx_commit(tx); 2003 goto out; 2004 } 2005 2006 if (unlinked) { 2007 zfs_unlinked_add(zp, tx); 2008 vp->v_vflag |= VV_NOSYNC; 2009 } 2010 2011 txtype = TX_REMOVE; 2012 zfs_log_remove(zilog, tx, txtype, dzp, name, obj); 2013 2014 dmu_tx_commit(tx); 2015out: 2016 2017 if (xzp) 2018 vrele(ZTOV(xzp)); 2019 2020 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2021 zil_commit(zilog, 0); 2022 2023 ZFS_EXIT(zfsvfs); 2024 return (error); 2025} 2026 2027/* 2028 * Create a new directory and insert it into dvp using the name 2029 * provided. Return a pointer to the inserted directory. 2030 * 2031 * IN: dvp - vnode of directory to add subdir to. 2032 * dirname - name of new directory. 2033 * vap - attributes of new directory. 2034 * cr - credentials of caller. 2035 * ct - caller context 2036 * flags - case flags 2037 * vsecp - ACL to be set 2038 * 2039 * OUT: vpp - vnode of created directory. 2040 * 2041 * RETURN: 0 on success, error code on failure. 2042 * 2043 * Timestamps: 2044 * dvp - ctime|mtime updated 2045 * vp - ctime|mtime|atime updated 2046 */ 2047/*ARGSUSED*/ 2048static int 2049zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr) 2050{ 2051 znode_t *zp, *dzp = VTOZ(dvp); 2052 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2053 zilog_t *zilog; 2054 uint64_t txtype; 2055 dmu_tx_t *tx; 2056 int error; 2057 ksid_t *ksid; 2058 uid_t uid; 2059 gid_t gid = crgetgid(cr); 2060 zfs_acl_ids_t acl_ids; 2061 boolean_t fuid_dirtied; 2062 2063 ASSERT(vap->va_type == VDIR); 2064 2065 /* 2066 * If we have an ephemeral id, ACL, or XVATTR then 2067 * make sure file system is at proper version 2068 */ 2069 2070 ksid = crgetsid(cr, KSID_OWNER); 2071 if (ksid) 2072 uid = ksid_getid(ksid); 2073 else 2074 uid = crgetuid(cr); 2075 if (zfsvfs->z_use_fuids == B_FALSE && 2076 ((vap->va_mask & AT_XVATTR) || 2077 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 2078 return (SET_ERROR(EINVAL)); 2079 2080 ZFS_ENTER(zfsvfs); 2081 ZFS_VERIFY_ZP(dzp); 2082 zilog = zfsvfs->z_log; 2083 2084 if (dzp->z_pflags & ZFS_XATTR) { 2085 ZFS_EXIT(zfsvfs); 2086 return (SET_ERROR(EINVAL)); 2087 } 2088 2089 if (zfsvfs->z_utf8 && u8_validate(dirname, 2090 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2091 ZFS_EXIT(zfsvfs); 2092 return (SET_ERROR(EILSEQ)); 2093 } 2094 2095 if (vap->va_mask & AT_XVATTR) { 2096 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 2097 crgetuid(cr), cr, vap->va_type)) != 0) { 2098 ZFS_EXIT(zfsvfs); 2099 return (error); 2100 } 2101 } 2102 2103 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 2104 NULL, &acl_ids)) != 0) { 2105 ZFS_EXIT(zfsvfs); 2106 return (error); 2107 } 2108 2109 /* 2110 * First make sure the new directory doesn't exist. 2111 * 2112 * Existence is checked first to make sure we don't return 2113 * EACCES instead of EEXIST which can cause some applications 2114 * to fail. 2115 */ 2116 *vpp = NULL; 2117 2118 if (error = zfs_dirent_lookup(dzp, dirname, &zp, ZNEW)) { 2119 zfs_acl_ids_free(&acl_ids); 2120 ZFS_EXIT(zfsvfs); 2121 return (error); 2122 } 2123 ASSERT3P(zp, ==, NULL); 2124 2125 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 2126 zfs_acl_ids_free(&acl_ids); 2127 ZFS_EXIT(zfsvfs); 2128 return (error); 2129 } 2130 2131 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 2132 zfs_acl_ids_free(&acl_ids); 2133 ZFS_EXIT(zfsvfs); 2134 return (SET_ERROR(EDQUOT)); 2135 } 2136 2137 /* 2138 * Add a new entry to the directory. 2139 */ 2140 getnewvnode_reserve(1); 2141 tx = dmu_tx_create(zfsvfs->z_os); 2142 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 2143 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2144 fuid_dirtied = zfsvfs->z_fuid_dirty; 2145 if (fuid_dirtied) 2146 zfs_fuid_txhold(zfsvfs, tx); 2147 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2148 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2149 acl_ids.z_aclp->z_acl_bytes); 2150 } 2151 2152 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 2153 ZFS_SA_BASE_ATTR_SIZE); 2154 2155 error = dmu_tx_assign(tx, TXG_WAIT); 2156 if (error) { 2157 zfs_acl_ids_free(&acl_ids); 2158 dmu_tx_abort(tx); 2159 getnewvnode_drop_reserve(); 2160 ZFS_EXIT(zfsvfs); 2161 return (error); 2162 } 2163 2164 /* 2165 * Create new node. 2166 */ 2167 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 2168 2169 if (fuid_dirtied) 2170 zfs_fuid_sync(zfsvfs, tx); 2171 2172 /* 2173 * Now put new name in parent dir. 2174 */ 2175 (void) zfs_link_create(dzp, dirname, zp, tx, ZNEW); 2176 2177 *vpp = ZTOV(zp); 2178 2179 txtype = zfs_log_create_txtype(Z_DIR, NULL, vap); 2180 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, NULL, 2181 acl_ids.z_fuidp, vap); 2182 2183 zfs_acl_ids_free(&acl_ids); 2184 2185 dmu_tx_commit(tx); 2186 2187 getnewvnode_drop_reserve(); 2188 2189 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2190 zil_commit(zilog, 0); 2191 2192 ZFS_EXIT(zfsvfs); 2193 return (0); 2194} 2195 2196/* 2197 * Remove a directory subdir entry. If the current working 2198 * directory is the same as the subdir to be removed, the 2199 * remove will fail. 2200 * 2201 * IN: dvp - vnode of directory to remove from. 2202 * name - name of directory to be removed. 2203 * cwd - vnode of current working directory. 2204 * cr - credentials of caller. 2205 * ct - caller context 2206 * flags - case flags 2207 * 2208 * RETURN: 0 on success, error code on failure. 2209 * 2210 * Timestamps: 2211 * dvp - ctime|mtime updated 2212 */ 2213/*ARGSUSED*/ 2214static int 2215zfs_rmdir(vnode_t *dvp, vnode_t *vp, char *name, cred_t *cr) 2216{ 2217 znode_t *dzp = VTOZ(dvp); 2218 znode_t *zp = VTOZ(vp); 2219 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2220 zilog_t *zilog; 2221 dmu_tx_t *tx; 2222 int error; 2223 2224 ZFS_ENTER(zfsvfs); 2225 ZFS_VERIFY_ZP(dzp); 2226 ZFS_VERIFY_ZP(zp); 2227 zilog = zfsvfs->z_log; 2228 2229 2230 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 2231 goto out; 2232 } 2233 2234 if (vp->v_type != VDIR) { 2235 error = SET_ERROR(ENOTDIR); 2236 goto out; 2237 } 2238 2239 vnevent_rmdir(vp, dvp, name, ct); 2240 2241 tx = dmu_tx_create(zfsvfs->z_os); 2242 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2243 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2244 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2245 zfs_sa_upgrade_txholds(tx, zp); 2246 zfs_sa_upgrade_txholds(tx, dzp); 2247 dmu_tx_mark_netfree(tx); 2248 error = dmu_tx_assign(tx, TXG_WAIT); 2249 if (error) { 2250 dmu_tx_abort(tx); 2251 ZFS_EXIT(zfsvfs); 2252 return (error); 2253 } 2254 2255 cache_purge(dvp); 2256 2257 error = zfs_link_destroy(dzp, name, zp, tx, ZEXISTS, NULL); 2258 2259 if (error == 0) { 2260 uint64_t txtype = TX_RMDIR; 2261 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT); 2262 } 2263 2264 dmu_tx_commit(tx); 2265 2266 cache_purge(vp); 2267out: 2268 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2269 zil_commit(zilog, 0); 2270 2271 ZFS_EXIT(zfsvfs); 2272 return (error); 2273} 2274 2275/* 2276 * Read as many directory entries as will fit into the provided 2277 * buffer from the given directory cursor position (specified in 2278 * the uio structure). 2279 * 2280 * IN: vp - vnode of directory to read. 2281 * uio - structure supplying read location, range info, 2282 * and return buffer. 2283 * cr - credentials of caller. 2284 * ct - caller context 2285 * flags - case flags 2286 * 2287 * OUT: uio - updated offset and range, buffer filled. 2288 * eofp - set to true if end-of-file detected. 2289 * 2290 * RETURN: 0 on success, error code on failure. 2291 * 2292 * Timestamps: 2293 * vp - atime updated 2294 * 2295 * Note that the low 4 bits of the cookie returned by zap is always zero. 2296 * This allows us to use the low range for "special" directory entries: 2297 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 2298 * we use the offset 2 for the '.zfs' directory. 2299 */ 2300/* ARGSUSED */ 2301static int 2302zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 2303{ 2304 znode_t *zp = VTOZ(vp); 2305 iovec_t *iovp; 2306 edirent_t *eodp; 2307 dirent64_t *odp; 2308 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2309 objset_t *os; 2310 caddr_t outbuf; 2311 size_t bufsize; 2312 zap_cursor_t zc; 2313 zap_attribute_t zap; 2314 uint_t bytes_wanted; 2315 uint64_t offset; /* must be unsigned; checks for < 1 */ 2316 uint64_t parent; 2317 int local_eof; 2318 int outcount; 2319 int error; 2320 uint8_t prefetch; 2321 boolean_t check_sysattrs; 2322 uint8_t type; 2323 int ncooks; 2324 u_long *cooks = NULL; 2325 int flags = 0; 2326 2327 ZFS_ENTER(zfsvfs); 2328 ZFS_VERIFY_ZP(zp); 2329 2330 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 2331 &parent, sizeof (parent))) != 0) { 2332 ZFS_EXIT(zfsvfs); 2333 return (error); 2334 } 2335 2336 /* 2337 * If we are not given an eof variable, 2338 * use a local one. 2339 */ 2340 if (eofp == NULL) 2341 eofp = &local_eof; 2342 2343 /* 2344 * Check for valid iov_len. 2345 */ 2346 if (uio->uio_iov->iov_len <= 0) { 2347 ZFS_EXIT(zfsvfs); 2348 return (SET_ERROR(EINVAL)); 2349 } 2350 2351 /* 2352 * Quit if directory has been removed (posix) 2353 */ 2354 if ((*eofp = zp->z_unlinked) != 0) { 2355 ZFS_EXIT(zfsvfs); 2356 return (0); 2357 } 2358 2359 error = 0; 2360 os = zfsvfs->z_os; 2361 offset = uio->uio_loffset; 2362 prefetch = zp->z_zn_prefetch; 2363 2364 /* 2365 * Initialize the iterator cursor. 2366 */ 2367 if (offset <= 3) { 2368 /* 2369 * Start iteration from the beginning of the directory. 2370 */ 2371 zap_cursor_init(&zc, os, zp->z_id); 2372 } else { 2373 /* 2374 * The offset is a serialized cursor. 2375 */ 2376 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 2377 } 2378 2379 /* 2380 * Get space to change directory entries into fs independent format. 2381 */ 2382 iovp = uio->uio_iov; 2383 bytes_wanted = iovp->iov_len; 2384 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 2385 bufsize = bytes_wanted; 2386 outbuf = kmem_alloc(bufsize, KM_SLEEP); 2387 odp = (struct dirent64 *)outbuf; 2388 } else { 2389 bufsize = bytes_wanted; 2390 outbuf = NULL; 2391 odp = (struct dirent64 *)iovp->iov_base; 2392 } 2393 eodp = (struct edirent *)odp; 2394 2395 if (ncookies != NULL) { 2396 /* 2397 * Minimum entry size is dirent size and 1 byte for a file name. 2398 */ 2399 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 2400 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 2401 *cookies = cooks; 2402 *ncookies = ncooks; 2403 } 2404 /* 2405 * If this VFS supports the system attribute view interface; and 2406 * we're looking at an extended attribute directory; and we care 2407 * about normalization conflicts on this vfs; then we must check 2408 * for normalization conflicts with the sysattr name space. 2409 */ 2410#ifdef TODO 2411 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 2412 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 2413 (flags & V_RDDIR_ENTFLAGS); 2414#else 2415 check_sysattrs = 0; 2416#endif 2417 2418 /* 2419 * Transform to file-system independent format 2420 */ 2421 outcount = 0; 2422 while (outcount < bytes_wanted) { 2423 ino64_t objnum; 2424 ushort_t reclen; 2425 off64_t *next = NULL; 2426 2427 /* 2428 * Special case `.', `..', and `.zfs'. 2429 */ 2430 if (offset == 0) { 2431 (void) strcpy(zap.za_name, "."); 2432 zap.za_normalization_conflict = 0; 2433 objnum = zp->z_id; 2434 type = DT_DIR; 2435 } else if (offset == 1) { 2436 (void) strcpy(zap.za_name, ".."); 2437 zap.za_normalization_conflict = 0; 2438 objnum = parent; 2439 type = DT_DIR; 2440 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2441 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2442 zap.za_normalization_conflict = 0; 2443 objnum = ZFSCTL_INO_ROOT; 2444 type = DT_DIR; 2445 } else { 2446 /* 2447 * Grab next entry. 2448 */ 2449 if (error = zap_cursor_retrieve(&zc, &zap)) { 2450 if ((*eofp = (error == ENOENT)) != 0) 2451 break; 2452 else 2453 goto update; 2454 } 2455 2456 if (zap.za_integer_length != 8 || 2457 zap.za_num_integers != 1) { 2458 cmn_err(CE_WARN, "zap_readdir: bad directory " 2459 "entry, obj = %lld, offset = %lld\n", 2460 (u_longlong_t)zp->z_id, 2461 (u_longlong_t)offset); 2462 error = SET_ERROR(ENXIO); 2463 goto update; 2464 } 2465 2466 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2467 /* 2468 * MacOS X can extract the object type here such as: 2469 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2470 */ 2471 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2472 2473 if (check_sysattrs && !zap.za_normalization_conflict) { 2474#ifdef TODO 2475 zap.za_normalization_conflict = 2476 xattr_sysattr_casechk(zap.za_name); 2477#else 2478 panic("%s:%u: TODO", __func__, __LINE__); 2479#endif 2480 } 2481 } 2482 2483 if (flags & V_RDDIR_ACCFILTER) { 2484 /* 2485 * If we have no access at all, don't include 2486 * this entry in the returned information 2487 */ 2488 znode_t *ezp; 2489 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2490 goto skip_entry; 2491 if (!zfs_has_access(ezp, cr)) { 2492 vrele(ZTOV(ezp)); 2493 goto skip_entry; 2494 } 2495 vrele(ZTOV(ezp)); 2496 } 2497 2498 if (flags & V_RDDIR_ENTFLAGS) 2499 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2500 else 2501 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2502 2503 /* 2504 * Will this entry fit in the buffer? 2505 */ 2506 if (outcount + reclen > bufsize) { 2507 /* 2508 * Did we manage to fit anything in the buffer? 2509 */ 2510 if (!outcount) { 2511 error = SET_ERROR(EINVAL); 2512 goto update; 2513 } 2514 break; 2515 } 2516 if (flags & V_RDDIR_ENTFLAGS) { 2517 /* 2518 * Add extended flag entry: 2519 */ 2520 eodp->ed_ino = objnum; 2521 eodp->ed_reclen = reclen; 2522 /* NOTE: ed_off is the offset for the *next* entry */ 2523 next = &(eodp->ed_off); 2524 eodp->ed_eflags = zap.za_normalization_conflict ? 2525 ED_CASE_CONFLICT : 0; 2526 (void) strncpy(eodp->ed_name, zap.za_name, 2527 EDIRENT_NAMELEN(reclen)); 2528 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2529 } else { 2530 /* 2531 * Add normal entry: 2532 */ 2533 odp->d_ino = objnum; 2534 odp->d_reclen = reclen; 2535 odp->d_namlen = strlen(zap.za_name); 2536 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 2537 odp->d_type = type; 2538 odp = (dirent64_t *)((intptr_t)odp + reclen); 2539 } 2540 outcount += reclen; 2541 2542 ASSERT(outcount <= bufsize); 2543 2544 /* Prefetch znode */ 2545 if (prefetch) 2546 dmu_prefetch(os, objnum, 0, 0, 0, 2547 ZIO_PRIORITY_SYNC_READ); 2548 2549 skip_entry: 2550 /* 2551 * Move to the next entry, fill in the previous offset. 2552 */ 2553 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2554 zap_cursor_advance(&zc); 2555 offset = zap_cursor_serialize(&zc); 2556 } else { 2557 offset += 1; 2558 } 2559 2560 if (cooks != NULL) { 2561 *cooks++ = offset; 2562 ncooks--; 2563 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 2564 } 2565 } 2566 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2567 2568 /* Subtract unused cookies */ 2569 if (ncookies != NULL) 2570 *ncookies -= ncooks; 2571 2572 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2573 iovp->iov_base += outcount; 2574 iovp->iov_len -= outcount; 2575 uio->uio_resid -= outcount; 2576 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2577 /* 2578 * Reset the pointer. 2579 */ 2580 offset = uio->uio_loffset; 2581 } 2582 2583update: 2584 zap_cursor_fini(&zc); 2585 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2586 kmem_free(outbuf, bufsize); 2587 2588 if (error == ENOENT) 2589 error = 0; 2590 2591 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2592 2593 uio->uio_loffset = offset; 2594 ZFS_EXIT(zfsvfs); 2595 if (error != 0 && cookies != NULL) { 2596 free(*cookies, M_TEMP); 2597 *cookies = NULL; 2598 *ncookies = 0; 2599 } 2600 return (error); 2601} 2602 2603ulong_t zfs_fsync_sync_cnt = 4; 2604 2605static int 2606zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2607{ 2608 znode_t *zp = VTOZ(vp); 2609 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2610 2611 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2612 2613 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { 2614 ZFS_ENTER(zfsvfs); 2615 ZFS_VERIFY_ZP(zp); 2616 zil_commit(zfsvfs->z_log, zp->z_id); 2617 ZFS_EXIT(zfsvfs); 2618 } 2619 return (0); 2620} 2621 2622 2623/* 2624 * Get the requested file attributes and place them in the provided 2625 * vattr structure. 2626 * 2627 * IN: vp - vnode of file. 2628 * vap - va_mask identifies requested attributes. 2629 * If AT_XVATTR set, then optional attrs are requested 2630 * flags - ATTR_NOACLCHECK (CIFS server context) 2631 * cr - credentials of caller. 2632 * ct - caller context 2633 * 2634 * OUT: vap - attribute values. 2635 * 2636 * RETURN: 0 (always succeeds). 2637 */ 2638/* ARGSUSED */ 2639static int 2640zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2641 caller_context_t *ct) 2642{ 2643 znode_t *zp = VTOZ(vp); 2644 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2645 int error = 0; 2646 uint32_t blksize; 2647 u_longlong_t nblocks; 2648 uint64_t links; 2649 uint64_t mtime[2], ctime[2], crtime[2], rdev; 2650 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2651 xoptattr_t *xoap = NULL; 2652 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2653 sa_bulk_attr_t bulk[4]; 2654 int count = 0; 2655 2656 ZFS_ENTER(zfsvfs); 2657 ZFS_VERIFY_ZP(zp); 2658 2659 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2660 2661 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 2662 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 2663 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16); 2664 if (vp->v_type == VBLK || vp->v_type == VCHR) 2665 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, 2666 &rdev, 8); 2667 2668 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 2669 ZFS_EXIT(zfsvfs); 2670 return (error); 2671 } 2672 2673 /* 2674 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2675 * Also, if we are the owner don't bother, since owner should 2676 * always be allowed to read basic attributes of file. 2677 */ 2678 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 2679 (vap->va_uid != crgetuid(cr))) { 2680 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2681 skipaclchk, cr)) { 2682 ZFS_EXIT(zfsvfs); 2683 return (error); 2684 } 2685 } 2686 2687 /* 2688 * Return all attributes. It's cheaper to provide the answer 2689 * than to determine whether we were asked the question. 2690 */ 2691 2692 vap->va_type = IFTOVT(zp->z_mode); 2693 vap->va_mode = zp->z_mode & ~S_IFMT; 2694#ifdef illumos 2695 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2696#else 2697 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 2698#endif 2699 vap->va_nodeid = zp->z_id; 2700 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2701 links = zp->z_links + 1; 2702 else 2703 links = zp->z_links; 2704 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */ 2705 vap->va_size = zp->z_size; 2706#ifdef illumos 2707 vap->va_rdev = vp->v_rdev; 2708#else 2709 if (vp->v_type == VBLK || vp->v_type == VCHR) 2710 vap->va_rdev = zfs_cmpldev(rdev); 2711#endif 2712 vap->va_seq = zp->z_seq; 2713 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2714 vap->va_filerev = zp->z_seq; 2715 2716 /* 2717 * Add in any requested optional attributes and the create time. 2718 * Also set the corresponding bits in the returned attribute bitmap. 2719 */ 2720 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2721 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2722 xoap->xoa_archive = 2723 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2724 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2725 } 2726 2727 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2728 xoap->xoa_readonly = 2729 ((zp->z_pflags & ZFS_READONLY) != 0); 2730 XVA_SET_RTN(xvap, XAT_READONLY); 2731 } 2732 2733 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2734 xoap->xoa_system = 2735 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2736 XVA_SET_RTN(xvap, XAT_SYSTEM); 2737 } 2738 2739 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2740 xoap->xoa_hidden = 2741 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2742 XVA_SET_RTN(xvap, XAT_HIDDEN); 2743 } 2744 2745 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2746 xoap->xoa_nounlink = 2747 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2748 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2749 } 2750 2751 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2752 xoap->xoa_immutable = 2753 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2754 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2755 } 2756 2757 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2758 xoap->xoa_appendonly = 2759 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2760 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2761 } 2762 2763 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2764 xoap->xoa_nodump = 2765 ((zp->z_pflags & ZFS_NODUMP) != 0); 2766 XVA_SET_RTN(xvap, XAT_NODUMP); 2767 } 2768 2769 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2770 xoap->xoa_opaque = 2771 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2772 XVA_SET_RTN(xvap, XAT_OPAQUE); 2773 } 2774 2775 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2776 xoap->xoa_av_quarantined = 2777 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2778 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2779 } 2780 2781 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2782 xoap->xoa_av_modified = 2783 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2784 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2785 } 2786 2787 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2788 vp->v_type == VREG) { 2789 zfs_sa_get_scanstamp(zp, xvap); 2790 } 2791 2792 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2793 uint64_t times[2]; 2794 2795 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 2796 times, sizeof (times)); 2797 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 2798 XVA_SET_RTN(xvap, XAT_CREATETIME); 2799 } 2800 2801 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2802 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2803 XVA_SET_RTN(xvap, XAT_REPARSE); 2804 } 2805 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2806 xoap->xoa_generation = zp->z_gen; 2807 XVA_SET_RTN(xvap, XAT_GEN); 2808 } 2809 2810 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2811 xoap->xoa_offline = 2812 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2813 XVA_SET_RTN(xvap, XAT_OFFLINE); 2814 } 2815 2816 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2817 xoap->xoa_sparse = 2818 ((zp->z_pflags & ZFS_SPARSE) != 0); 2819 XVA_SET_RTN(xvap, XAT_SPARSE); 2820 } 2821 } 2822 2823 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2824 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2825 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2826 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 2827 2828 2829 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 2830 vap->va_blksize = blksize; 2831 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2832 2833 if (zp->z_blksz == 0) { 2834 /* 2835 * Block size hasn't been set; suggest maximal I/O transfers. 2836 */ 2837 vap->va_blksize = zfsvfs->z_max_blksz; 2838 } 2839 2840 ZFS_EXIT(zfsvfs); 2841 return (0); 2842} 2843 2844/* 2845 * Set the file attributes to the values contained in the 2846 * vattr structure. 2847 * 2848 * IN: vp - vnode of file to be modified. 2849 * vap - new attribute values. 2850 * If AT_XVATTR set, then optional attrs are being set 2851 * flags - ATTR_UTIME set if non-default time values provided. 2852 * - ATTR_NOACLCHECK (CIFS context only). 2853 * cr - credentials of caller. 2854 * ct - caller context 2855 * 2856 * RETURN: 0 on success, error code on failure. 2857 * 2858 * Timestamps: 2859 * vp - ctime updated, mtime updated if size changed. 2860 */ 2861/* ARGSUSED */ 2862static int 2863zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2864 caller_context_t *ct) 2865{ 2866 znode_t *zp = VTOZ(vp); 2867 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2868 zilog_t *zilog; 2869 dmu_tx_t *tx; 2870 vattr_t oldva; 2871 xvattr_t tmpxvattr; 2872 uint_t mask = vap->va_mask; 2873 uint_t saved_mask = 0; 2874 uint64_t saved_mode; 2875 int trim_mask = 0; 2876 uint64_t new_mode; 2877 uint64_t new_uid, new_gid; 2878 uint64_t xattr_obj; 2879 uint64_t mtime[2], ctime[2]; 2880 znode_t *attrzp; 2881 int need_policy = FALSE; 2882 int err, err2; 2883 zfs_fuid_info_t *fuidp = NULL; 2884 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2885 xoptattr_t *xoap; 2886 zfs_acl_t *aclp; 2887 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2888 boolean_t fuid_dirtied = B_FALSE; 2889 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 2890 int count = 0, xattr_count = 0; 2891 2892 if (mask == 0) 2893 return (0); 2894 2895 if (mask & AT_NOSET) 2896 return (SET_ERROR(EINVAL)); 2897 2898 ZFS_ENTER(zfsvfs); 2899 ZFS_VERIFY_ZP(zp); 2900 2901 zilog = zfsvfs->z_log; 2902 2903 /* 2904 * Make sure that if we have ephemeral uid/gid or xvattr specified 2905 * that file system is at proper version level 2906 */ 2907 2908 if (zfsvfs->z_use_fuids == B_FALSE && 2909 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2910 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2911 (mask & AT_XVATTR))) { 2912 ZFS_EXIT(zfsvfs); 2913 return (SET_ERROR(EINVAL)); 2914 } 2915 2916 if (mask & AT_SIZE && vp->v_type == VDIR) { 2917 ZFS_EXIT(zfsvfs); 2918 return (SET_ERROR(EISDIR)); 2919 } 2920 2921 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 2922 ZFS_EXIT(zfsvfs); 2923 return (SET_ERROR(EINVAL)); 2924 } 2925 2926 /* 2927 * If this is an xvattr_t, then get a pointer to the structure of 2928 * optional attributes. If this is NULL, then we have a vattr_t. 2929 */ 2930 xoap = xva_getxoptattr(xvap); 2931 2932 xva_init(&tmpxvattr); 2933 2934 /* 2935 * Immutable files can only alter immutable bit and atime 2936 */ 2937 if ((zp->z_pflags & ZFS_IMMUTABLE) && 2938 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2939 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 2940 ZFS_EXIT(zfsvfs); 2941 return (SET_ERROR(EPERM)); 2942 } 2943 2944 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 2945 ZFS_EXIT(zfsvfs); 2946 return (SET_ERROR(EPERM)); 2947 } 2948 2949 /* 2950 * Verify timestamps doesn't overflow 32 bits. 2951 * ZFS can handle large timestamps, but 32bit syscalls can't 2952 * handle times greater than 2039. This check should be removed 2953 * once large timestamps are fully supported. 2954 */ 2955 if (mask & (AT_ATIME | AT_MTIME)) { 2956 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 2957 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 2958 ZFS_EXIT(zfsvfs); 2959 return (SET_ERROR(EOVERFLOW)); 2960 } 2961 } 2962 2963 attrzp = NULL; 2964 aclp = NULL; 2965 2966 /* Can this be moved to before the top label? */ 2967 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2968 ZFS_EXIT(zfsvfs); 2969 return (SET_ERROR(EROFS)); 2970 } 2971 2972 /* 2973 * First validate permissions 2974 */ 2975 2976 if (mask & AT_SIZE) { 2977 /* 2978 * XXX - Note, we are not providing any open 2979 * mode flags here (like FNDELAY), so we may 2980 * block if there are locks present... this 2981 * should be addressed in openat(). 2982 */ 2983 /* XXX - would it be OK to generate a log record here? */ 2984 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2985 if (err) { 2986 ZFS_EXIT(zfsvfs); 2987 return (err); 2988 } 2989 } 2990 2991 if (mask & (AT_ATIME|AT_MTIME) || 2992 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2993 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2994 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2995 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 2996 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 2997 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2998 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 2999 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 3000 skipaclchk, cr); 3001 } 3002 3003 if (mask & (AT_UID|AT_GID)) { 3004 int idmask = (mask & (AT_UID|AT_GID)); 3005 int take_owner; 3006 int take_group; 3007 3008 /* 3009 * NOTE: even if a new mode is being set, 3010 * we may clear S_ISUID/S_ISGID bits. 3011 */ 3012 3013 if (!(mask & AT_MODE)) 3014 vap->va_mode = zp->z_mode; 3015 3016 /* 3017 * Take ownership or chgrp to group we are a member of 3018 */ 3019 3020 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 3021 take_group = (mask & AT_GID) && 3022 zfs_groupmember(zfsvfs, vap->va_gid, cr); 3023 3024 /* 3025 * If both AT_UID and AT_GID are set then take_owner and 3026 * take_group must both be set in order to allow taking 3027 * ownership. 3028 * 3029 * Otherwise, send the check through secpolicy_vnode_setattr() 3030 * 3031 */ 3032 3033 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 3034 ((idmask == AT_UID) && take_owner) || 3035 ((idmask == AT_GID) && take_group)) { 3036 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 3037 skipaclchk, cr) == 0) { 3038 /* 3039 * Remove setuid/setgid for non-privileged users 3040 */ 3041 secpolicy_setid_clear(vap, vp, cr); 3042 trim_mask = (mask & (AT_UID|AT_GID)); 3043 } else { 3044 need_policy = TRUE; 3045 } 3046 } else { 3047 need_policy = TRUE; 3048 } 3049 } 3050 3051 oldva.va_mode = zp->z_mode; 3052 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 3053 if (mask & AT_XVATTR) { 3054 /* 3055 * Update xvattr mask to include only those attributes 3056 * that are actually changing. 3057 * 3058 * the bits will be restored prior to actually setting 3059 * the attributes so the caller thinks they were set. 3060 */ 3061 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 3062 if (xoap->xoa_appendonly != 3063 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 3064 need_policy = TRUE; 3065 } else { 3066 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 3067 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 3068 } 3069 } 3070 3071 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 3072 if (xoap->xoa_nounlink != 3073 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 3074 need_policy = TRUE; 3075 } else { 3076 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 3077 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 3078 } 3079 } 3080 3081 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 3082 if (xoap->xoa_immutable != 3083 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 3084 need_policy = TRUE; 3085 } else { 3086 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 3087 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 3088 } 3089 } 3090 3091 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 3092 if (xoap->xoa_nodump != 3093 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 3094 need_policy = TRUE; 3095 } else { 3096 XVA_CLR_REQ(xvap, XAT_NODUMP); 3097 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 3098 } 3099 } 3100 3101 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 3102 if (xoap->xoa_av_modified != 3103 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 3104 need_policy = TRUE; 3105 } else { 3106 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 3107 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 3108 } 3109 } 3110 3111 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 3112 if ((vp->v_type != VREG && 3113 xoap->xoa_av_quarantined) || 3114 xoap->xoa_av_quarantined != 3115 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 3116 need_policy = TRUE; 3117 } else { 3118 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 3119 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 3120 } 3121 } 3122 3123 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 3124 ZFS_EXIT(zfsvfs); 3125 return (SET_ERROR(EPERM)); 3126 } 3127 3128 if (need_policy == FALSE && 3129 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 3130 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 3131 need_policy = TRUE; 3132 } 3133 } 3134 3135 if (mask & AT_MODE) { 3136 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 3137 err = secpolicy_setid_setsticky_clear(vp, vap, 3138 &oldva, cr); 3139 if (err) { 3140 ZFS_EXIT(zfsvfs); 3141 return (err); 3142 } 3143 trim_mask |= AT_MODE; 3144 } else { 3145 need_policy = TRUE; 3146 } 3147 } 3148 3149 if (need_policy) { 3150 /* 3151 * If trim_mask is set then take ownership 3152 * has been granted or write_acl is present and user 3153 * has the ability to modify mode. In that case remove 3154 * UID|GID and or MODE from mask so that 3155 * secpolicy_vnode_setattr() doesn't revoke it. 3156 */ 3157 3158 if (trim_mask) { 3159 saved_mask = vap->va_mask; 3160 vap->va_mask &= ~trim_mask; 3161 if (trim_mask & AT_MODE) { 3162 /* 3163 * Save the mode, as secpolicy_vnode_setattr() 3164 * will overwrite it with ova.va_mode. 3165 */ 3166 saved_mode = vap->va_mode; 3167 } 3168 } 3169 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 3170 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 3171 if (err) { 3172 ZFS_EXIT(zfsvfs); 3173 return (err); 3174 } 3175 3176 if (trim_mask) { 3177 vap->va_mask |= saved_mask; 3178 if (trim_mask & AT_MODE) { 3179 /* 3180 * Recover the mode after 3181 * secpolicy_vnode_setattr(). 3182 */ 3183 vap->va_mode = saved_mode; 3184 } 3185 } 3186 } 3187 3188 /* 3189 * secpolicy_vnode_setattr, or take ownership may have 3190 * changed va_mask 3191 */ 3192 mask = vap->va_mask; 3193 3194 if ((mask & (AT_UID | AT_GID))) { 3195 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 3196 &xattr_obj, sizeof (xattr_obj)); 3197 3198 if (err == 0 && xattr_obj) { 3199 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 3200 if (err == 0) { 3201 err = vn_lock(ZTOV(attrzp), LK_EXCLUSIVE); 3202 if (err != 0) 3203 vrele(ZTOV(attrzp)); 3204 } 3205 if (err) 3206 goto out2; 3207 } 3208 if (mask & AT_UID) { 3209 new_uid = zfs_fuid_create(zfsvfs, 3210 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 3211 if (new_uid != zp->z_uid && 3212 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 3213 if (attrzp) 3214 vput(ZTOV(attrzp)); 3215 err = SET_ERROR(EDQUOT); 3216 goto out2; 3217 } 3218 } 3219 3220 if (mask & AT_GID) { 3221 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 3222 cr, ZFS_GROUP, &fuidp); 3223 if (new_gid != zp->z_gid && 3224 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 3225 if (attrzp) 3226 vput(ZTOV(attrzp)); 3227 err = SET_ERROR(EDQUOT); 3228 goto out2; 3229 } 3230 } 3231 } 3232 tx = dmu_tx_create(zfsvfs->z_os); 3233 3234 if (mask & AT_MODE) { 3235 uint64_t pmode = zp->z_mode; 3236 uint64_t acl_obj; 3237 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 3238 3239 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && 3240 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 3241 err = SET_ERROR(EPERM); 3242 goto out; 3243 } 3244 3245 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 3246 goto out; 3247 3248 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 3249 /* 3250 * Are we upgrading ACL from old V0 format 3251 * to V1 format? 3252 */ 3253 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 3254 zfs_znode_acl_version(zp) == 3255 ZFS_ACL_VERSION_INITIAL) { 3256 dmu_tx_hold_free(tx, acl_obj, 0, 3257 DMU_OBJECT_END); 3258 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3259 0, aclp->z_acl_bytes); 3260 } else { 3261 dmu_tx_hold_write(tx, acl_obj, 0, 3262 aclp->z_acl_bytes); 3263 } 3264 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3265 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3266 0, aclp->z_acl_bytes); 3267 } 3268 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3269 } else { 3270 if ((mask & AT_XVATTR) && 3271 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3272 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3273 else 3274 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3275 } 3276 3277 if (attrzp) { 3278 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 3279 } 3280 3281 fuid_dirtied = zfsvfs->z_fuid_dirty; 3282 if (fuid_dirtied) 3283 zfs_fuid_txhold(zfsvfs, tx); 3284 3285 zfs_sa_upgrade_txholds(tx, zp); 3286 3287 err = dmu_tx_assign(tx, TXG_WAIT); 3288 if (err) 3289 goto out; 3290 3291 count = 0; 3292 /* 3293 * Set each attribute requested. 3294 * We group settings according to the locks they need to acquire. 3295 * 3296 * Note: you cannot set ctime directly, although it will be 3297 * updated as a side-effect of calling this function. 3298 */ 3299 3300 if (mask & (AT_UID|AT_GID|AT_MODE)) 3301 mutex_enter(&zp->z_acl_lock); 3302 3303 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3304 &zp->z_pflags, sizeof (zp->z_pflags)); 3305 3306 if (attrzp) { 3307 if (mask & (AT_UID|AT_GID|AT_MODE)) 3308 mutex_enter(&attrzp->z_acl_lock); 3309 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3310 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3311 sizeof (attrzp->z_pflags)); 3312 } 3313 3314 if (mask & (AT_UID|AT_GID)) { 3315 3316 if (mask & AT_UID) { 3317 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3318 &new_uid, sizeof (new_uid)); 3319 zp->z_uid = new_uid; 3320 if (attrzp) { 3321 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3322 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3323 sizeof (new_uid)); 3324 attrzp->z_uid = new_uid; 3325 } 3326 } 3327 3328 if (mask & AT_GID) { 3329 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3330 NULL, &new_gid, sizeof (new_gid)); 3331 zp->z_gid = new_gid; 3332 if (attrzp) { 3333 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3334 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 3335 sizeof (new_gid)); 3336 attrzp->z_gid = new_gid; 3337 } 3338 } 3339 if (!(mask & AT_MODE)) { 3340 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 3341 NULL, &new_mode, sizeof (new_mode)); 3342 new_mode = zp->z_mode; 3343 } 3344 err = zfs_acl_chown_setattr(zp); 3345 ASSERT(err == 0); 3346 if (attrzp) { 3347 err = zfs_acl_chown_setattr(attrzp); 3348 ASSERT(err == 0); 3349 } 3350 } 3351 3352 if (mask & AT_MODE) { 3353 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 3354 &new_mode, sizeof (new_mode)); 3355 zp->z_mode = new_mode; 3356 ASSERT3U((uintptr_t)aclp, !=, 0); 3357 err = zfs_aclset_common(zp, aclp, cr, tx); 3358 ASSERT0(err); 3359 if (zp->z_acl_cached) 3360 zfs_acl_free(zp->z_acl_cached); 3361 zp->z_acl_cached = aclp; 3362 aclp = NULL; 3363 } 3364 3365 3366 if (mask & AT_ATIME) { 3367 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 3368 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 3369 &zp->z_atime, sizeof (zp->z_atime)); 3370 } 3371 3372 if (mask & AT_MTIME) { 3373 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 3374 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 3375 mtime, sizeof (mtime)); 3376 } 3377 3378 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 3379 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 3380 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 3381 NULL, mtime, sizeof (mtime)); 3382 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3383 &ctime, sizeof (ctime)); 3384 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 3385 B_TRUE); 3386 } else if (mask != 0) { 3387 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3388 &ctime, sizeof (ctime)); 3389 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime, 3390 B_TRUE); 3391 if (attrzp) { 3392 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3393 SA_ZPL_CTIME(zfsvfs), NULL, 3394 &ctime, sizeof (ctime)); 3395 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 3396 mtime, ctime, B_TRUE); 3397 } 3398 } 3399 /* 3400 * Do this after setting timestamps to prevent timestamp 3401 * update from toggling bit 3402 */ 3403 3404 if (xoap && (mask & AT_XVATTR)) { 3405 3406 /* 3407 * restore trimmed off masks 3408 * so that return masks can be set for caller. 3409 */ 3410 3411 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 3412 XVA_SET_REQ(xvap, XAT_APPENDONLY); 3413 } 3414 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 3415 XVA_SET_REQ(xvap, XAT_NOUNLINK); 3416 } 3417 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 3418 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 3419 } 3420 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 3421 XVA_SET_REQ(xvap, XAT_NODUMP); 3422 } 3423 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 3424 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 3425 } 3426 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 3427 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 3428 } 3429 3430 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3431 ASSERT(vp->v_type == VREG); 3432 3433 zfs_xvattr_set(zp, xvap, tx); 3434 } 3435 3436 if (fuid_dirtied) 3437 zfs_fuid_sync(zfsvfs, tx); 3438 3439 if (mask != 0) 3440 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 3441 3442 if (mask & (AT_UID|AT_GID|AT_MODE)) 3443 mutex_exit(&zp->z_acl_lock); 3444 3445 if (attrzp) { 3446 if (mask & (AT_UID|AT_GID|AT_MODE)) 3447 mutex_exit(&attrzp->z_acl_lock); 3448 } 3449out: 3450 if (err == 0 && attrzp) { 3451 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 3452 xattr_count, tx); 3453 ASSERT(err2 == 0); 3454 } 3455 3456 if (attrzp) 3457 vput(ZTOV(attrzp)); 3458 3459 if (aclp) 3460 zfs_acl_free(aclp); 3461 3462 if (fuidp) { 3463 zfs_fuid_info_free(fuidp); 3464 fuidp = NULL; 3465 } 3466 3467 if (err) { 3468 dmu_tx_abort(tx); 3469 } else { 3470 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3471 dmu_tx_commit(tx); 3472 } 3473 3474out2: 3475 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3476 zil_commit(zilog, 0); 3477 3478 ZFS_EXIT(zfsvfs); 3479 return (err); 3480} 3481 3482/* 3483 * We acquire all but fdvp locks using non-blocking acquisitions. If we 3484 * fail to acquire any lock in the path we will drop all held locks, 3485 * acquire the new lock in a blocking fashion, and then release it and 3486 * restart the rename. This acquire/release step ensures that we do not 3487 * spin on a lock waiting for release. On error release all vnode locks 3488 * and decrement references the way tmpfs_rename() would do. 3489 */ 3490static int 3491zfs_rename_relock(struct vnode *sdvp, struct vnode **svpp, 3492 struct vnode *tdvp, struct vnode **tvpp, 3493 const struct componentname *scnp, const struct componentname *tcnp) 3494{ 3495 zfsvfs_t *zfsvfs; 3496 struct vnode *nvp, *svp, *tvp; 3497 znode_t *sdzp, *tdzp, *szp, *tzp; 3498 const char *snm = scnp->cn_nameptr; 3499 const char *tnm = tcnp->cn_nameptr; 3500 int error; 3501 3502 VOP_UNLOCK(tdvp, 0); 3503 if (*tvpp != NULL && *tvpp != tdvp) 3504 VOP_UNLOCK(*tvpp, 0); 3505 3506relock: 3507 error = vn_lock(sdvp, LK_EXCLUSIVE); 3508 if (error) 3509 goto out; 3510 sdzp = VTOZ(sdvp); 3511 3512 error = vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT); 3513 if (error != 0) { 3514 VOP_UNLOCK(sdvp, 0); 3515 if (error != EBUSY) 3516 goto out; 3517 error = vn_lock(tdvp, LK_EXCLUSIVE); 3518 if (error) 3519 goto out; 3520 VOP_UNLOCK(tdvp, 0); 3521 goto relock; 3522 } 3523 tdzp = VTOZ(tdvp); 3524 3525 /* 3526 * Before using sdzp and tdzp we must ensure that they are live. 3527 * As a porting legacy from illumos we have two things to worry 3528 * about. One is typical for FreeBSD and it is that the vnode is 3529 * not reclaimed (doomed). The other is that the znode is live. 3530 * The current code can invalidate the znode without acquiring the 3531 * corresponding vnode lock if the object represented by the znode 3532 * and vnode is no longer valid after a rollback or receive operation. 3533 * z_teardown_lock hidden behind ZFS_ENTER and ZFS_EXIT is the lock 3534 * that protects the znodes from the invalidation. 3535 */ 3536 zfsvfs = sdzp->z_zfsvfs; 3537 ASSERT3P(zfsvfs, ==, tdzp->z_zfsvfs); 3538 ZFS_ENTER(zfsvfs); 3539 3540 /* 3541 * We can not use ZFS_VERIFY_ZP() here because it could directly return 3542 * bypassing the cleanup code in the case of an error. 3543 */ 3544 if (tdzp->z_sa_hdl == NULL || sdzp->z_sa_hdl == NULL) { 3545 ZFS_EXIT(zfsvfs); 3546 VOP_UNLOCK(sdvp, 0); 3547 VOP_UNLOCK(tdvp, 0); 3548 error = SET_ERROR(EIO); 3549 goto out; 3550 } 3551 3552 /* 3553 * Re-resolve svp to be certain it still exists and fetch the 3554 * correct vnode. 3555 */ 3556 error = zfs_dirent_lookup(sdzp, snm, &szp, ZEXISTS); 3557 if (error != 0) { 3558 /* Source entry invalid or not there. */ 3559 ZFS_EXIT(zfsvfs); 3560 VOP_UNLOCK(sdvp, 0); 3561 VOP_UNLOCK(tdvp, 0); 3562 if ((scnp->cn_flags & ISDOTDOT) != 0 || 3563 (scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.')) 3564 error = SET_ERROR(EINVAL); 3565 goto out; 3566 } 3567 svp = ZTOV(szp); 3568 3569 /* 3570 * Re-resolve tvp, if it disappeared we just carry on. 3571 */ 3572 error = zfs_dirent_lookup(tdzp, tnm, &tzp, 0); 3573 if (error != 0) { 3574 ZFS_EXIT(zfsvfs); 3575 VOP_UNLOCK(sdvp, 0); 3576 VOP_UNLOCK(tdvp, 0); 3577 vrele(svp); 3578 if ((tcnp->cn_flags & ISDOTDOT) != 0) 3579 error = SET_ERROR(EINVAL); 3580 goto out; 3581 } 3582 if (tzp != NULL) 3583 tvp = ZTOV(tzp); 3584 else 3585 tvp = NULL; 3586 3587 /* 3588 * At present the vnode locks must be acquired before z_teardown_lock, 3589 * although it would be more logical to use the opposite order. 3590 */ 3591 ZFS_EXIT(zfsvfs); 3592 3593 /* 3594 * Now try acquire locks on svp and tvp. 3595 */ 3596 nvp = svp; 3597 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT); 3598 if (error != 0) { 3599 VOP_UNLOCK(sdvp, 0); 3600 VOP_UNLOCK(tdvp, 0); 3601 if (tvp != NULL) 3602 vrele(tvp); 3603 if (error != EBUSY) { 3604 vrele(nvp); 3605 goto out; 3606 } 3607 error = vn_lock(nvp, LK_EXCLUSIVE); 3608 if (error != 0) { 3609 vrele(nvp); 3610 goto out; 3611 } 3612 VOP_UNLOCK(nvp, 0); 3613 /* 3614 * Concurrent rename race. 3615 * XXX ? 3616 */ 3617 if (nvp == tdvp) { 3618 vrele(nvp); 3619 error = SET_ERROR(EINVAL); 3620 goto out; 3621 } 3622 vrele(*svpp); 3623 *svpp = nvp; 3624 goto relock; 3625 } 3626 vrele(*svpp); 3627 *svpp = nvp; 3628 3629 if (*tvpp != NULL) 3630 vrele(*tvpp); 3631 *tvpp = NULL; 3632 if (tvp != NULL) { 3633 nvp = tvp; 3634 error = vn_lock(nvp, LK_EXCLUSIVE | LK_NOWAIT); 3635 if (error != 0) { 3636 VOP_UNLOCK(sdvp, 0); 3637 VOP_UNLOCK(tdvp, 0); 3638 VOP_UNLOCK(*svpp, 0); 3639 if (error != EBUSY) { 3640 vrele(nvp); 3641 goto out; 3642 } 3643 error = vn_lock(nvp, LK_EXCLUSIVE); 3644 if (error != 0) { 3645 vrele(nvp); 3646 goto out; 3647 } 3648 vput(nvp); 3649 goto relock; 3650 } 3651 *tvpp = nvp; 3652 } 3653 3654 return (0); 3655 3656out: 3657 return (error); 3658} 3659 3660/* 3661 * Note that we must use VRELE_ASYNC in this function as it walks 3662 * up the directory tree and vrele may need to acquire an exclusive 3663 * lock if a last reference to a vnode is dropped. 3664 */ 3665static int 3666zfs_rename_check(znode_t *szp, znode_t *sdzp, znode_t *tdzp) 3667{ 3668 zfsvfs_t *zfsvfs; 3669 znode_t *zp, *zp1; 3670 uint64_t parent; 3671 int error; 3672 3673 zfsvfs = tdzp->z_zfsvfs; 3674 if (tdzp == szp) 3675 return (SET_ERROR(EINVAL)); 3676 if (tdzp == sdzp) 3677 return (0); 3678 if (tdzp->z_id == zfsvfs->z_root) 3679 return (0); 3680 zp = tdzp; 3681 for (;;) { 3682 ASSERT(!zp->z_unlinked); 3683 if ((error = sa_lookup(zp->z_sa_hdl, 3684 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) 3685 break; 3686 3687 if (parent == szp->z_id) { 3688 error = SET_ERROR(EINVAL); 3689 break; 3690 } 3691 if (parent == zfsvfs->z_root) 3692 break; 3693 if (parent == sdzp->z_id) 3694 break; 3695 3696 error = zfs_zget(zfsvfs, parent, &zp1); 3697 if (error != 0) 3698 break; 3699 3700 if (zp != tdzp) 3701 VN_RELE_ASYNC(ZTOV(zp), 3702 dsl_pool_vnrele_taskq(dmu_objset_pool(zfsvfs->z_os))); 3703 zp = zp1; 3704 } 3705 3706 if (error == ENOTDIR) 3707 panic("checkpath: .. not a directory\n"); 3708 if (zp != tdzp) 3709 VN_RELE_ASYNC(ZTOV(zp), 3710 dsl_pool_vnrele_taskq(dmu_objset_pool(zfsvfs->z_os))); 3711 return (error); 3712} 3713 3714/* 3715 * Move an entry from the provided source directory to the target 3716 * directory. Change the entry name as indicated. 3717 * 3718 * IN: sdvp - Source directory containing the "old entry". 3719 * snm - Old entry name. 3720 * tdvp - Target directory to contain the "new entry". 3721 * tnm - New entry name. 3722 * cr - credentials of caller. 3723 * ct - caller context 3724 * flags - case flags 3725 * 3726 * RETURN: 0 on success, error code on failure. 3727 * 3728 * Timestamps: 3729 * sdvp,tdvp - ctime|mtime updated 3730 */ 3731/*ARGSUSED*/ 3732static int 3733zfs_rename(vnode_t *sdvp, vnode_t **svpp, struct componentname *scnp, 3734 vnode_t *tdvp, vnode_t **tvpp, struct componentname *tcnp, 3735 cred_t *cr) 3736{ 3737 zfsvfs_t *zfsvfs; 3738 znode_t *sdzp, *tdzp, *szp, *tzp; 3739 zilog_t *zilog = NULL; 3740 dmu_tx_t *tx; 3741 char *snm = scnp->cn_nameptr; 3742 char *tnm = tcnp->cn_nameptr; 3743 int error = 0; 3744 3745 /* Reject renames across filesystems. */ 3746 if ((*svpp)->v_mount != tdvp->v_mount || 3747 ((*tvpp) != NULL && (*svpp)->v_mount != (*tvpp)->v_mount)) { 3748 error = SET_ERROR(EXDEV); 3749 goto out; 3750 } 3751 3752 if (zfsctl_is_node(tdvp)) { 3753 error = SET_ERROR(EXDEV); 3754 goto out; 3755 } 3756 3757 /* 3758 * Lock all four vnodes to ensure safety and semantics of renaming. 3759 */ 3760 error = zfs_rename_relock(sdvp, svpp, tdvp, tvpp, scnp, tcnp); 3761 if (error != 0) { 3762 /* no vnodes are locked in the case of error here */ 3763 return (error); 3764 } 3765 3766 tdzp = VTOZ(tdvp); 3767 sdzp = VTOZ(sdvp); 3768 zfsvfs = tdzp->z_zfsvfs; 3769 zilog = zfsvfs->z_log; 3770 3771 /* 3772 * After we re-enter ZFS_ENTER() we will have to revalidate all 3773 * znodes involved. 3774 */ 3775 ZFS_ENTER(zfsvfs); 3776 3777 if (zfsvfs->z_utf8 && u8_validate(tnm, 3778 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3779 error = SET_ERROR(EILSEQ); 3780 goto unlockout; 3781 } 3782 3783 /* If source and target are the same file, there is nothing to do. */ 3784 if ((*svpp) == (*tvpp)) { 3785 error = 0; 3786 goto unlockout; 3787 } 3788 3789 if (((*svpp)->v_type == VDIR && (*svpp)->v_mountedhere != NULL) || 3790 ((*tvpp) != NULL && (*tvpp)->v_type == VDIR && 3791 (*tvpp)->v_mountedhere != NULL)) { 3792 error = SET_ERROR(EXDEV); 3793 goto unlockout; 3794 } 3795 3796 /* 3797 * We can not use ZFS_VERIFY_ZP() here because it could directly return 3798 * bypassing the cleanup code in the case of an error. 3799 */ 3800 if (tdzp->z_sa_hdl == NULL || sdzp->z_sa_hdl == NULL) { 3801 error = SET_ERROR(EIO); 3802 goto unlockout; 3803 } 3804 3805 szp = VTOZ(*svpp); 3806 tzp = *tvpp == NULL ? NULL : VTOZ(*tvpp); 3807 if (szp->z_sa_hdl == NULL || (tzp != NULL && tzp->z_sa_hdl == NULL)) { 3808 error = SET_ERROR(EIO); 3809 goto unlockout; 3810 } 3811 3812 /* 3813 * This is to prevent the creation of links into attribute space 3814 * by renaming a linked file into/outof an attribute directory. 3815 * See the comment in zfs_link() for why this is considered bad. 3816 */ 3817 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3818 error = SET_ERROR(EINVAL); 3819 goto unlockout; 3820 } 3821 3822 /* 3823 * Must have write access at the source to remove the old entry 3824 * and write access at the target to create the new entry. 3825 * Note that if target and source are the same, this can be 3826 * done in a single check. 3827 */ 3828 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3829 goto unlockout; 3830 3831 if ((*svpp)->v_type == VDIR) { 3832 /* 3833 * Avoid ".", "..", and aliases of "." for obvious reasons. 3834 */ 3835 if ((scnp->cn_namelen == 1 && scnp->cn_nameptr[0] == '.') || 3836 sdzp == szp || 3837 (scnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) { 3838 error = EINVAL; 3839 goto unlockout; 3840 } 3841 3842 /* 3843 * Check to make sure rename is valid. 3844 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3845 */ 3846 if (error = zfs_rename_check(szp, sdzp, tdzp)) 3847 goto unlockout; 3848 } 3849 3850 /* 3851 * Does target exist? 3852 */ 3853 if (tzp) { 3854 /* 3855 * Source and target must be the same type. 3856 */ 3857 if ((*svpp)->v_type == VDIR) { 3858 if ((*tvpp)->v_type != VDIR) { 3859 error = SET_ERROR(ENOTDIR); 3860 goto unlockout; 3861 } else { 3862 cache_purge(tdvp); 3863 if (sdvp != tdvp) 3864 cache_purge(sdvp); 3865 } 3866 } else { 3867 if ((*tvpp)->v_type == VDIR) { 3868 error = SET_ERROR(EISDIR); 3869 goto unlockout; 3870 } 3871 } 3872 } 3873 3874 vnevent_rename_src(*svpp, sdvp, scnp->cn_nameptr, ct); 3875 if (tzp) 3876 vnevent_rename_dest(*tvpp, tdvp, tnm, ct); 3877 3878 /* 3879 * notify the target directory if it is not the same 3880 * as source directory. 3881 */ 3882 if (tdvp != sdvp) { 3883 vnevent_rename_dest_dir(tdvp, ct); 3884 } 3885 3886 tx = dmu_tx_create(zfsvfs->z_os); 3887 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3888 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 3889 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3890 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3891 if (sdzp != tdzp) { 3892 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 3893 zfs_sa_upgrade_txholds(tx, tdzp); 3894 } 3895 if (tzp) { 3896 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 3897 zfs_sa_upgrade_txholds(tx, tzp); 3898 } 3899 3900 zfs_sa_upgrade_txholds(tx, szp); 3901 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3902 error = dmu_tx_assign(tx, TXG_WAIT); 3903 if (error) { 3904 dmu_tx_abort(tx); 3905 goto unlockout; 3906 } 3907 3908 3909 if (tzp) /* Attempt to remove the existing target */ 3910 error = zfs_link_destroy(tdzp, tnm, tzp, tx, 0, NULL); 3911 3912 if (error == 0) { 3913 error = zfs_link_create(tdzp, tnm, szp, tx, ZRENAMING); 3914 if (error == 0) { 3915 szp->z_pflags |= ZFS_AV_MODIFIED; 3916 3917 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 3918 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 3919 ASSERT0(error); 3920 3921 error = zfs_link_destroy(sdzp, snm, szp, tx, ZRENAMING, 3922 NULL); 3923 if (error == 0) { 3924 zfs_log_rename(zilog, tx, TX_RENAME, sdzp, 3925 snm, tdzp, tnm, szp); 3926 3927 /* 3928 * Update path information for the target vnode 3929 */ 3930 vn_renamepath(tdvp, *svpp, tnm, strlen(tnm)); 3931 } else { 3932 /* 3933 * At this point, we have successfully created 3934 * the target name, but have failed to remove 3935 * the source name. Since the create was done 3936 * with the ZRENAMING flag, there are 3937 * complications; for one, the link count is 3938 * wrong. The easiest way to deal with this 3939 * is to remove the newly created target, and 3940 * return the original error. This must 3941 * succeed; fortunately, it is very unlikely to 3942 * fail, since we just created it. 3943 */ 3944 VERIFY3U(zfs_link_destroy(tdzp, tnm, szp, tx, 3945 ZRENAMING, NULL), ==, 0); 3946 } 3947 } 3948 if (error == 0) { 3949 cache_purge(*svpp); 3950 if (*tvpp != NULL) 3951 cache_purge(*tvpp); 3952 cache_purge_negative(tdvp); 3953 } 3954 } 3955 3956 dmu_tx_commit(tx); 3957 3958unlockout: /* all 4 vnodes are locked, ZFS_ENTER called */ 3959 ZFS_EXIT(zfsvfs); 3960 VOP_UNLOCK(*svpp, 0); 3961 VOP_UNLOCK(sdvp, 0); 3962 3963out: /* original two vnodes are locked */ 3964 if (error == 0 && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3965 zil_commit(zilog, 0); 3966 3967 if (*tvpp != NULL) 3968 VOP_UNLOCK(*tvpp, 0); 3969 if (tdvp != *tvpp) 3970 VOP_UNLOCK(tdvp, 0); 3971 return (error); 3972} 3973 3974/* 3975 * Insert the indicated symbolic reference entry into the directory. 3976 * 3977 * IN: dvp - Directory to contain new symbolic link. 3978 * link - Name for new symlink entry. 3979 * vap - Attributes of new entry. 3980 * cr - credentials of caller. 3981 * ct - caller context 3982 * flags - case flags 3983 * 3984 * RETURN: 0 on success, error code on failure. 3985 * 3986 * Timestamps: 3987 * dvp - ctime|mtime updated 3988 */ 3989/*ARGSUSED*/ 3990static int 3991zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, 3992 cred_t *cr, kthread_t *td) 3993{ 3994 znode_t *zp, *dzp = VTOZ(dvp); 3995 dmu_tx_t *tx; 3996 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3997 zilog_t *zilog; 3998 uint64_t len = strlen(link); 3999 int error; 4000 zfs_acl_ids_t acl_ids; 4001 boolean_t fuid_dirtied; 4002 uint64_t txtype = TX_SYMLINK; 4003 int flags = 0; 4004 4005 ASSERT(vap->va_type == VLNK); 4006 4007 ZFS_ENTER(zfsvfs); 4008 ZFS_VERIFY_ZP(dzp); 4009 zilog = zfsvfs->z_log; 4010 4011 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 4012 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4013 ZFS_EXIT(zfsvfs); 4014 return (SET_ERROR(EILSEQ)); 4015 } 4016 4017 if (len > MAXPATHLEN) { 4018 ZFS_EXIT(zfsvfs); 4019 return (SET_ERROR(ENAMETOOLONG)); 4020 } 4021 4022 if ((error = zfs_acl_ids_create(dzp, 0, 4023 vap, cr, NULL, &acl_ids)) != 0) { 4024 ZFS_EXIT(zfsvfs); 4025 return (error); 4026 } 4027 4028 /* 4029 * Attempt to lock directory; fail if entry already exists. 4030 */ 4031 error = zfs_dirent_lookup(dzp, name, &zp, ZNEW); 4032 if (error) { 4033 zfs_acl_ids_free(&acl_ids); 4034 ZFS_EXIT(zfsvfs); 4035 return (error); 4036 } 4037 4038 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4039 zfs_acl_ids_free(&acl_ids); 4040 ZFS_EXIT(zfsvfs); 4041 return (error); 4042 } 4043 4044 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 4045 zfs_acl_ids_free(&acl_ids); 4046 ZFS_EXIT(zfsvfs); 4047 return (SET_ERROR(EDQUOT)); 4048 } 4049 4050 getnewvnode_reserve(1); 4051 tx = dmu_tx_create(zfsvfs->z_os); 4052 fuid_dirtied = zfsvfs->z_fuid_dirty; 4053 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 4054 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4055 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 4056 ZFS_SA_BASE_ATTR_SIZE + len); 4057 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 4058 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 4059 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 4060 acl_ids.z_aclp->z_acl_bytes); 4061 } 4062 if (fuid_dirtied) 4063 zfs_fuid_txhold(zfsvfs, tx); 4064 error = dmu_tx_assign(tx, TXG_WAIT); 4065 if (error) { 4066 zfs_acl_ids_free(&acl_ids); 4067 dmu_tx_abort(tx); 4068 getnewvnode_drop_reserve(); 4069 ZFS_EXIT(zfsvfs); 4070 return (error); 4071 } 4072 4073 /* 4074 * Create a new object for the symlink. 4075 * for version 4 ZPL datsets the symlink will be an SA attribute 4076 */ 4077 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 4078 4079 if (fuid_dirtied) 4080 zfs_fuid_sync(zfsvfs, tx); 4081 4082 if (zp->z_is_sa) 4083 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 4084 link, len, tx); 4085 else 4086 zfs_sa_symlink(zp, link, len, tx); 4087 4088 zp->z_size = len; 4089 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 4090 &zp->z_size, sizeof (zp->z_size), tx); 4091 /* 4092 * Insert the new object into the directory. 4093 */ 4094 (void) zfs_link_create(dzp, name, zp, tx, ZNEW); 4095 4096 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 4097 *vpp = ZTOV(zp); 4098 4099 zfs_acl_ids_free(&acl_ids); 4100 4101 dmu_tx_commit(tx); 4102 4103 getnewvnode_drop_reserve(); 4104 4105 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4106 zil_commit(zilog, 0); 4107 4108 ZFS_EXIT(zfsvfs); 4109 return (error); 4110} 4111 4112/* 4113 * Return, in the buffer contained in the provided uio structure, 4114 * the symbolic path referred to by vp. 4115 * 4116 * IN: vp - vnode of symbolic link. 4117 * uio - structure to contain the link path. 4118 * cr - credentials of caller. 4119 * ct - caller context 4120 * 4121 * OUT: uio - structure containing the link path. 4122 * 4123 * RETURN: 0 on success, error code on failure. 4124 * 4125 * Timestamps: 4126 * vp - atime updated 4127 */ 4128/* ARGSUSED */ 4129static int 4130zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 4131{ 4132 znode_t *zp = VTOZ(vp); 4133 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4134 int error; 4135 4136 ZFS_ENTER(zfsvfs); 4137 ZFS_VERIFY_ZP(zp); 4138 4139 if (zp->z_is_sa) 4140 error = sa_lookup_uio(zp->z_sa_hdl, 4141 SA_ZPL_SYMLINK(zfsvfs), uio); 4142 else 4143 error = zfs_sa_readlink(zp, uio); 4144 4145 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4146 4147 ZFS_EXIT(zfsvfs); 4148 return (error); 4149} 4150 4151/* 4152 * Insert a new entry into directory tdvp referencing svp. 4153 * 4154 * IN: tdvp - Directory to contain new entry. 4155 * svp - vnode of new entry. 4156 * name - name of new entry. 4157 * cr - credentials of caller. 4158 * ct - caller context 4159 * 4160 * RETURN: 0 on success, error code on failure. 4161 * 4162 * Timestamps: 4163 * tdvp - ctime|mtime updated 4164 * svp - ctime updated 4165 */ 4166/* ARGSUSED */ 4167static int 4168zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 4169 caller_context_t *ct, int flags) 4170{ 4171 znode_t *dzp = VTOZ(tdvp); 4172 znode_t *tzp, *szp; 4173 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4174 zilog_t *zilog; 4175 dmu_tx_t *tx; 4176 int error; 4177 uint64_t parent; 4178 uid_t owner; 4179 4180 ASSERT(tdvp->v_type == VDIR); 4181 4182 ZFS_ENTER(zfsvfs); 4183 ZFS_VERIFY_ZP(dzp); 4184 zilog = zfsvfs->z_log; 4185 4186 /* 4187 * POSIX dictates that we return EPERM here. 4188 * Better choices include ENOTSUP or EISDIR. 4189 */ 4190 if (svp->v_type == VDIR) { 4191 ZFS_EXIT(zfsvfs); 4192 return (SET_ERROR(EPERM)); 4193 } 4194 4195 szp = VTOZ(svp); 4196 ZFS_VERIFY_ZP(szp); 4197 4198 if (szp->z_pflags & (ZFS_APPENDONLY | ZFS_IMMUTABLE | ZFS_READONLY)) { 4199 ZFS_EXIT(zfsvfs); 4200 return (SET_ERROR(EPERM)); 4201 } 4202 4203 /* Prevent links to .zfs/shares files */ 4204 4205 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 4206 &parent, sizeof (uint64_t))) != 0) { 4207 ZFS_EXIT(zfsvfs); 4208 return (error); 4209 } 4210 if (parent == zfsvfs->z_shares_dir) { 4211 ZFS_EXIT(zfsvfs); 4212 return (SET_ERROR(EPERM)); 4213 } 4214 4215 if (zfsvfs->z_utf8 && u8_validate(name, 4216 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4217 ZFS_EXIT(zfsvfs); 4218 return (SET_ERROR(EILSEQ)); 4219 } 4220 4221 /* 4222 * We do not support links between attributes and non-attributes 4223 * because of the potential security risk of creating links 4224 * into "normal" file space in order to circumvent restrictions 4225 * imposed in attribute space. 4226 */ 4227 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 4228 ZFS_EXIT(zfsvfs); 4229 return (SET_ERROR(EINVAL)); 4230 } 4231 4232 4233 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 4234 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) { 4235 ZFS_EXIT(zfsvfs); 4236 return (SET_ERROR(EPERM)); 4237 } 4238 4239 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4240 ZFS_EXIT(zfsvfs); 4241 return (error); 4242 } 4243 4244 /* 4245 * Attempt to lock directory; fail if entry already exists. 4246 */ 4247 error = zfs_dirent_lookup(dzp, name, &tzp, ZNEW); 4248 if (error) { 4249 ZFS_EXIT(zfsvfs); 4250 return (error); 4251 } 4252 4253 tx = dmu_tx_create(zfsvfs->z_os); 4254 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4255 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4256 zfs_sa_upgrade_txholds(tx, szp); 4257 zfs_sa_upgrade_txholds(tx, dzp); 4258 error = dmu_tx_assign(tx, TXG_WAIT); 4259 if (error) { 4260 dmu_tx_abort(tx); 4261 ZFS_EXIT(zfsvfs); 4262 return (error); 4263 } 4264 4265 error = zfs_link_create(dzp, name, szp, tx, 0); 4266 4267 if (error == 0) { 4268 uint64_t txtype = TX_LINK; 4269 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 4270 } 4271 4272 dmu_tx_commit(tx); 4273 4274 if (error == 0) { 4275 vnevent_link(svp, ct); 4276 } 4277 4278 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4279 zil_commit(zilog, 0); 4280 4281 ZFS_EXIT(zfsvfs); 4282 return (error); 4283} 4284 4285 4286/*ARGSUSED*/ 4287void 4288zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4289{ 4290 znode_t *zp = VTOZ(vp); 4291 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4292 int error; 4293 4294 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4295 if (zp->z_sa_hdl == NULL) { 4296 /* 4297 * The fs has been unmounted, or we did a 4298 * suspend/resume and this file no longer exists. 4299 */ 4300 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4301 vrecycle(vp); 4302 return; 4303 } 4304 4305 if (zp->z_unlinked) { 4306 /* 4307 * Fast path to recycle a vnode of a removed file. 4308 */ 4309 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4310 vrecycle(vp); 4311 return; 4312 } 4313 4314 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4315 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4316 4317 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4318 zfs_sa_upgrade_txholds(tx, zp); 4319 error = dmu_tx_assign(tx, TXG_WAIT); 4320 if (error) { 4321 dmu_tx_abort(tx); 4322 } else { 4323 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4324 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4325 zp->z_atime_dirty = 0; 4326 dmu_tx_commit(tx); 4327 } 4328 } 4329 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4330} 4331 4332 4333CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 4334CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 4335 4336/*ARGSUSED*/ 4337static int 4338zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 4339{ 4340 znode_t *zp = VTOZ(vp); 4341 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4342 uint32_t gen; 4343 uint64_t gen64; 4344 uint64_t object = zp->z_id; 4345 zfid_short_t *zfid; 4346 int size, i, error; 4347 4348 ZFS_ENTER(zfsvfs); 4349 ZFS_VERIFY_ZP(zp); 4350 4351 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 4352 &gen64, sizeof (uint64_t))) != 0) { 4353 ZFS_EXIT(zfsvfs); 4354 return (error); 4355 } 4356 4357 gen = (uint32_t)gen64; 4358 4359 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 4360 4361#ifdef illumos 4362 if (fidp->fid_len < size) { 4363 fidp->fid_len = size; 4364 ZFS_EXIT(zfsvfs); 4365 return (SET_ERROR(ENOSPC)); 4366 } 4367#else 4368 fidp->fid_len = size; 4369#endif 4370 4371 zfid = (zfid_short_t *)fidp; 4372 4373 zfid->zf_len = size; 4374 4375 for (i = 0; i < sizeof (zfid->zf_object); i++) 4376 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 4377 4378 /* Must have a non-zero generation number to distinguish from .zfs */ 4379 if (gen == 0) 4380 gen = 1; 4381 for (i = 0; i < sizeof (zfid->zf_gen); i++) 4382 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 4383 4384 if (size == LONG_FID_LEN) { 4385 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 4386 zfid_long_t *zlfid; 4387 4388 zlfid = (zfid_long_t *)fidp; 4389 4390 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 4391 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 4392 4393 /* XXX - this should be the generation number for the objset */ 4394 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 4395 zlfid->zf_setgen[i] = 0; 4396 } 4397 4398 ZFS_EXIT(zfsvfs); 4399 return (0); 4400} 4401 4402static int 4403zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 4404 caller_context_t *ct) 4405{ 4406 znode_t *zp, *xzp; 4407 zfsvfs_t *zfsvfs; 4408 int error; 4409 4410 switch (cmd) { 4411 case _PC_LINK_MAX: 4412 *valp = INT_MAX; 4413 return (0); 4414 4415 case _PC_FILESIZEBITS: 4416 *valp = 64; 4417 return (0); 4418#ifdef illumos 4419 case _PC_XATTR_EXISTS: 4420 zp = VTOZ(vp); 4421 zfsvfs = zp->z_zfsvfs; 4422 ZFS_ENTER(zfsvfs); 4423 ZFS_VERIFY_ZP(zp); 4424 *valp = 0; 4425 error = zfs_dirent_lookup(zp, "", &xzp, 4426 ZXATTR | ZEXISTS | ZSHARED); 4427 if (error == 0) { 4428 if (!zfs_dirempty(xzp)) 4429 *valp = 1; 4430 vrele(ZTOV(xzp)); 4431 } else if (error == ENOENT) { 4432 /* 4433 * If there aren't extended attributes, it's the 4434 * same as having zero of them. 4435 */ 4436 error = 0; 4437 } 4438 ZFS_EXIT(zfsvfs); 4439 return (error); 4440 4441 case _PC_SATTR_ENABLED: 4442 case _PC_SATTR_EXISTS: 4443 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 4444 (vp->v_type == VREG || vp->v_type == VDIR); 4445 return (0); 4446 4447 case _PC_ACCESS_FILTERING: 4448 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 4449 vp->v_type == VDIR; 4450 return (0); 4451 4452 case _PC_ACL_ENABLED: 4453 *valp = _ACL_ACE_ENABLED; 4454 return (0); 4455#endif /* illumos */ 4456 case _PC_MIN_HOLE_SIZE: 4457 *valp = (int)SPA_MINBLOCKSIZE; 4458 return (0); 4459#ifdef illumos 4460 case _PC_TIMESTAMP_RESOLUTION: 4461 /* nanosecond timestamp resolution */ 4462 *valp = 1L; 4463 return (0); 4464#endif 4465 case _PC_ACL_EXTENDED: 4466 *valp = 0; 4467 return (0); 4468 4469 case _PC_ACL_NFS4: 4470 *valp = 1; 4471 return (0); 4472 4473 case _PC_ACL_PATH_MAX: 4474 *valp = ACL_MAX_ENTRIES; 4475 return (0); 4476 4477 default: 4478 return (EOPNOTSUPP); 4479 } 4480} 4481 4482/*ARGSUSED*/ 4483static int 4484zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4485 caller_context_t *ct) 4486{ 4487 znode_t *zp = VTOZ(vp); 4488 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4489 int error; 4490 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4491 4492 ZFS_ENTER(zfsvfs); 4493 ZFS_VERIFY_ZP(zp); 4494 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 4495 ZFS_EXIT(zfsvfs); 4496 4497 return (error); 4498} 4499 4500/*ARGSUSED*/ 4501int 4502zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4503 caller_context_t *ct) 4504{ 4505 znode_t *zp = VTOZ(vp); 4506 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4507 int error; 4508 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4509 zilog_t *zilog = zfsvfs->z_log; 4510 4511 ZFS_ENTER(zfsvfs); 4512 ZFS_VERIFY_ZP(zp); 4513 4514 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 4515 4516 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4517 zil_commit(zilog, 0); 4518 4519 ZFS_EXIT(zfsvfs); 4520 return (error); 4521} 4522 4523static int 4524ioflags(int ioflags) 4525{ 4526 int flags = 0; 4527 4528 if (ioflags & IO_APPEND) 4529 flags |= FAPPEND; 4530 if (ioflags & IO_NDELAY) 4531 flags |= FNONBLOCK; 4532 if (ioflags & IO_SYNC) 4533 flags |= (FSYNC | FDSYNC | FRSYNC); 4534 4535 return (flags); 4536} 4537 4538static int 4539zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage) 4540{ 4541 znode_t *zp = VTOZ(vp); 4542 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4543 objset_t *os = zp->z_zfsvfs->z_os; 4544 vm_page_t mfirst, mlast, mreq; 4545 vm_object_t object; 4546 caddr_t va; 4547 struct sf_buf *sf; 4548 off_t startoff, endoff; 4549 int i, error; 4550 vm_pindex_t reqstart, reqend; 4551 int pcount, lsize, reqsize, size; 4552 4553 ZFS_ENTER(zfsvfs); 4554 ZFS_VERIFY_ZP(zp); 4555 4556 pcount = OFF_TO_IDX(round_page(count)); 4557 mreq = m[reqpage]; 4558 object = mreq->object; 4559 error = 0; 4560 4561 KASSERT(vp->v_object == object, ("mismatching object")); 4562 4563 if (pcount > 1 && zp->z_blksz > PAGESIZE) { 4564 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz); 4565 reqstart = OFF_TO_IDX(round_page(startoff)); 4566 if (reqstart < m[0]->pindex) 4567 reqstart = 0; 4568 else 4569 reqstart = reqstart - m[0]->pindex; 4570 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE, 4571 zp->z_blksz); 4572 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1; 4573 if (reqend > m[pcount - 1]->pindex) 4574 reqend = m[pcount - 1]->pindex; 4575 reqsize = reqend - m[reqstart]->pindex + 1; 4576 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize, 4577 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds")); 4578 } else { 4579 reqstart = reqpage; 4580 reqsize = 1; 4581 } 4582 mfirst = m[reqstart]; 4583 mlast = m[reqstart + reqsize - 1]; 4584 4585 zfs_vmobject_wlock(object); 4586 4587 for (i = 0; i < reqstart; i++) { 4588 vm_page_lock(m[i]); 4589 vm_page_free(m[i]); 4590 vm_page_unlock(m[i]); 4591 } 4592 for (i = reqstart + reqsize; i < pcount; i++) { 4593 vm_page_lock(m[i]); 4594 vm_page_free(m[i]); 4595 vm_page_unlock(m[i]); 4596 } 4597 4598 if (mreq->valid && reqsize == 1) { 4599 if (mreq->valid != VM_PAGE_BITS_ALL) 4600 vm_page_zero_invalid(mreq, TRUE); 4601 zfs_vmobject_wunlock(object); 4602 ZFS_EXIT(zfsvfs); 4603 return (zfs_vm_pagerret_ok); 4604 } 4605 4606 PCPU_INC(cnt.v_vnodein); 4607 PCPU_ADD(cnt.v_vnodepgsin, reqsize); 4608 4609 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) { 4610 for (i = reqstart; i < reqstart + reqsize; i++) { 4611 if (i != reqpage) { 4612 vm_page_lock(m[i]); 4613 vm_page_free(m[i]); 4614 vm_page_unlock(m[i]); 4615 } 4616 } 4617 zfs_vmobject_wunlock(object); 4618 ZFS_EXIT(zfsvfs); 4619 return (zfs_vm_pagerret_bad); 4620 } 4621 4622 lsize = PAGE_SIZE; 4623 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size) 4624 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex); 4625 4626 zfs_vmobject_wunlock(object); 4627 4628 for (i = reqstart; i < reqstart + reqsize; i++) { 4629 size = PAGE_SIZE; 4630 if (i == (reqstart + reqsize - 1)) 4631 size = lsize; 4632 va = zfs_map_page(m[i], &sf); 4633 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex), 4634 size, va, DMU_READ_PREFETCH); 4635 if (size != PAGE_SIZE) 4636 bzero(va + size, PAGE_SIZE - size); 4637 zfs_unmap_page(sf); 4638 if (error != 0) 4639 break; 4640 } 4641 4642 zfs_vmobject_wlock(object); 4643 4644 for (i = reqstart; i < reqstart + reqsize; i++) { 4645 if (!error) 4646 m[i]->valid = VM_PAGE_BITS_ALL; 4647 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i])); 4648 if (i != reqpage) 4649 vm_page_readahead_finish(m[i]); 4650 } 4651 4652 zfs_vmobject_wunlock(object); 4653 4654 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4655 ZFS_EXIT(zfsvfs); 4656 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok); 4657} 4658 4659static int 4660zfs_freebsd_getpages(ap) 4661 struct vop_getpages_args /* { 4662 struct vnode *a_vp; 4663 vm_page_t *a_m; 4664 int a_count; 4665 int a_reqpage; 4666 vm_ooffset_t a_offset; 4667 } */ *ap; 4668{ 4669 4670 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage)); 4671} 4672 4673static int 4674zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags, 4675 int *rtvals) 4676{ 4677 znode_t *zp = VTOZ(vp); 4678 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4679 rl_t *rl; 4680 dmu_tx_t *tx; 4681 struct sf_buf *sf; 4682 vm_object_t object; 4683 vm_page_t m; 4684 caddr_t va; 4685 size_t tocopy; 4686 size_t lo_len; 4687 vm_ooffset_t lo_off; 4688 vm_ooffset_t off; 4689 uint_t blksz; 4690 int ncount; 4691 int pcount; 4692 int err; 4693 int i; 4694 4695 ZFS_ENTER(zfsvfs); 4696 ZFS_VERIFY_ZP(zp); 4697 4698 object = vp->v_object; 4699 pcount = btoc(len); 4700 ncount = pcount; 4701 4702 KASSERT(ma[0]->object == object, ("mismatching object")); 4703 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length")); 4704 4705 for (i = 0; i < pcount; i++) 4706 rtvals[i] = zfs_vm_pagerret_error; 4707 4708 off = IDX_TO_OFF(ma[0]->pindex); 4709 blksz = zp->z_blksz; 4710 lo_off = rounddown(off, blksz); 4711 lo_len = roundup(len + (off - lo_off), blksz); 4712 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER); 4713 4714 zfs_vmobject_wlock(object); 4715 if (len + off > object->un_pager.vnp.vnp_size) { 4716 if (object->un_pager.vnp.vnp_size > off) { 4717 int pgoff; 4718 4719 len = object->un_pager.vnp.vnp_size - off; 4720 ncount = btoc(len); 4721 if ((pgoff = (int)len & PAGE_MASK) != 0) { 4722 /* 4723 * If the object is locked and the following 4724 * conditions hold, then the page's dirty 4725 * field cannot be concurrently changed by a 4726 * pmap operation. 4727 */ 4728 m = ma[ncount - 1]; 4729 vm_page_assert_sbusied(m); 4730 KASSERT(!pmap_page_is_write_mapped(m), 4731 ("zfs_putpages: page %p is not read-only", m)); 4732 vm_page_clear_dirty(m, pgoff, PAGE_SIZE - 4733 pgoff); 4734 } 4735 } else { 4736 len = 0; 4737 ncount = 0; 4738 } 4739 if (ncount < pcount) { 4740 for (i = ncount; i < pcount; i++) { 4741 rtvals[i] = zfs_vm_pagerret_bad; 4742 } 4743 } 4744 } 4745 zfs_vmobject_wunlock(object); 4746 4747 if (ncount == 0) 4748 goto out; 4749 4750 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4751 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4752 goto out; 4753 } 4754 4755top: 4756 tx = dmu_tx_create(zfsvfs->z_os); 4757 dmu_tx_hold_write(tx, zp->z_id, off, len); 4758 4759 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4760 zfs_sa_upgrade_txholds(tx, zp); 4761 err = dmu_tx_assign(tx, TXG_NOWAIT); 4762 if (err != 0) { 4763 if (err == ERESTART) { 4764 dmu_tx_wait(tx); 4765 dmu_tx_abort(tx); 4766 goto top; 4767 } 4768 dmu_tx_abort(tx); 4769 goto out; 4770 } 4771 4772 if (zp->z_blksz < PAGE_SIZE) { 4773 i = 0; 4774 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) { 4775 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len; 4776 va = zfs_map_page(ma[i], &sf); 4777 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx); 4778 zfs_unmap_page(sf); 4779 } 4780 } else { 4781 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx); 4782 } 4783 4784 if (err == 0) { 4785 uint64_t mtime[2], ctime[2]; 4786 sa_bulk_attr_t bulk[3]; 4787 int count = 0; 4788 4789 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4790 &mtime, 16); 4791 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4792 &ctime, 16); 4793 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4794 &zp->z_pflags, 8); 4795 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4796 B_TRUE); 4797 (void)sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 4798 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4799 4800 zfs_vmobject_wlock(object); 4801 for (i = 0; i < ncount; i++) { 4802 rtvals[i] = zfs_vm_pagerret_ok; 4803 vm_page_undirty(ma[i]); 4804 } 4805 zfs_vmobject_wunlock(object); 4806 PCPU_INC(cnt.v_vnodeout); 4807 PCPU_ADD(cnt.v_vnodepgsout, ncount); 4808 } 4809 dmu_tx_commit(tx); 4810 4811out: 4812 zfs_range_unlock(rl); 4813 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 || 4814 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4815 zil_commit(zfsvfs->z_log, zp->z_id); 4816 ZFS_EXIT(zfsvfs); 4817 return (rtvals[0]); 4818} 4819 4820int 4821zfs_freebsd_putpages(ap) 4822 struct vop_putpages_args /* { 4823 struct vnode *a_vp; 4824 vm_page_t *a_m; 4825 int a_count; 4826 int a_sync; 4827 int *a_rtvals; 4828 vm_ooffset_t a_offset; 4829 } */ *ap; 4830{ 4831 4832 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync, 4833 ap->a_rtvals)); 4834} 4835 4836static int 4837zfs_freebsd_bmap(ap) 4838 struct vop_bmap_args /* { 4839 struct vnode *a_vp; 4840 daddr_t a_bn; 4841 struct bufobj **a_bop; 4842 daddr_t *a_bnp; 4843 int *a_runp; 4844 int *a_runb; 4845 } */ *ap; 4846{ 4847 4848 if (ap->a_bop != NULL) 4849 *ap->a_bop = &ap->a_vp->v_bufobj; 4850 if (ap->a_bnp != NULL) 4851 *ap->a_bnp = ap->a_bn; 4852 if (ap->a_runp != NULL) 4853 *ap->a_runp = 0; 4854 if (ap->a_runb != NULL) 4855 *ap->a_runb = 0; 4856 4857 return (0); 4858} 4859 4860static int 4861zfs_freebsd_open(ap) 4862 struct vop_open_args /* { 4863 struct vnode *a_vp; 4864 int a_mode; 4865 struct ucred *a_cred; 4866 struct thread *a_td; 4867 } */ *ap; 4868{ 4869 vnode_t *vp = ap->a_vp; 4870 znode_t *zp = VTOZ(vp); 4871 int error; 4872 4873 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL); 4874 if (error == 0) 4875 vnode_create_vobject(vp, zp->z_size, ap->a_td); 4876 return (error); 4877} 4878 4879static int 4880zfs_freebsd_close(ap) 4881 struct vop_close_args /* { 4882 struct vnode *a_vp; 4883 int a_fflag; 4884 struct ucred *a_cred; 4885 struct thread *a_td; 4886 } */ *ap; 4887{ 4888 4889 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL)); 4890} 4891 4892static int 4893zfs_freebsd_ioctl(ap) 4894 struct vop_ioctl_args /* { 4895 struct vnode *a_vp; 4896 u_long a_command; 4897 caddr_t a_data; 4898 int a_fflag; 4899 struct ucred *cred; 4900 struct thread *td; 4901 } */ *ap; 4902{ 4903 4904 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 4905 ap->a_fflag, ap->a_cred, NULL, NULL)); 4906} 4907 4908static int 4909zfs_freebsd_read(ap) 4910 struct vop_read_args /* { 4911 struct vnode *a_vp; 4912 struct uio *a_uio; 4913 int a_ioflag; 4914 struct ucred *a_cred; 4915 } */ *ap; 4916{ 4917 4918 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 4919 ap->a_cred, NULL)); 4920} 4921 4922static int 4923zfs_freebsd_write(ap) 4924 struct vop_write_args /* { 4925 struct vnode *a_vp; 4926 struct uio *a_uio; 4927 int a_ioflag; 4928 struct ucred *a_cred; 4929 } */ *ap; 4930{ 4931 4932 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 4933 ap->a_cred, NULL)); 4934} 4935 4936static int 4937zfs_freebsd_access(ap) 4938 struct vop_access_args /* { 4939 struct vnode *a_vp; 4940 accmode_t a_accmode; 4941 struct ucred *a_cred; 4942 struct thread *a_td; 4943 } */ *ap; 4944{ 4945 vnode_t *vp = ap->a_vp; 4946 znode_t *zp = VTOZ(vp); 4947 accmode_t accmode; 4948 int error = 0; 4949 4950 /* 4951 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 4952 */ 4953 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 4954 if (accmode != 0) 4955 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL); 4956 4957 /* 4958 * VADMIN has to be handled by vaccess(). 4959 */ 4960 if (error == 0) { 4961 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 4962 if (accmode != 0) { 4963 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 4964 zp->z_gid, accmode, ap->a_cred, NULL); 4965 } 4966 } 4967 4968 /* 4969 * For VEXEC, ensure that at least one execute bit is set for 4970 * non-directories. 4971 */ 4972 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 4973 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 4974 error = EACCES; 4975 } 4976 4977 return (error); 4978} 4979 4980static int 4981zfs_freebsd_lookup(ap) 4982 struct vop_lookup_args /* { 4983 struct vnode *a_dvp; 4984 struct vnode **a_vpp; 4985 struct componentname *a_cnp; 4986 } */ *ap; 4987{ 4988 struct componentname *cnp = ap->a_cnp; 4989 char nm[NAME_MAX + 1]; 4990 4991 ASSERT(cnp->cn_namelen < sizeof(nm)); 4992 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 4993 4994 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 4995 cnp->cn_cred, cnp->cn_thread, 0)); 4996} 4997 4998static int 4999zfs_cache_lookup(ap) 5000 struct vop_lookup_args /* { 5001 struct vnode *a_dvp; 5002 struct vnode **a_vpp; 5003 struct componentname *a_cnp; 5004 } */ *ap; 5005{ 5006 zfsvfs_t *zfsvfs; 5007 5008 zfsvfs = ap->a_dvp->v_mount->mnt_data; 5009 if (zfsvfs->z_use_namecache) 5010 return (vfs_cache_lookup(ap)); 5011 else 5012 return (zfs_freebsd_lookup(ap)); 5013} 5014 5015static int 5016zfs_freebsd_create(ap) 5017 struct vop_create_args /* { 5018 struct vnode *a_dvp; 5019 struct vnode **a_vpp; 5020 struct componentname *a_cnp; 5021 struct vattr *a_vap; 5022 } */ *ap; 5023{ 5024 zfsvfs_t *zfsvfs; 5025 struct componentname *cnp = ap->a_cnp; 5026 vattr_t *vap = ap->a_vap; 5027 int error, mode; 5028 5029 ASSERT(cnp->cn_flags & SAVENAME); 5030 5031 vattr_init_mask(vap); 5032 mode = vap->va_mode & ALLPERMS; 5033 zfsvfs = ap->a_dvp->v_mount->mnt_data; 5034 5035 error = zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 5036 ap->a_vpp, cnp->cn_cred, cnp->cn_thread); 5037 if (zfsvfs->z_use_namecache && 5038 error == 0 && (cnp->cn_flags & MAKEENTRY) != 0) 5039 cache_enter(ap->a_dvp, *ap->a_vpp, cnp); 5040 return (error); 5041} 5042 5043static int 5044zfs_freebsd_remove(ap) 5045 struct vop_remove_args /* { 5046 struct vnode *a_dvp; 5047 struct vnode *a_vp; 5048 struct componentname *a_cnp; 5049 } */ *ap; 5050{ 5051 5052 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 5053 5054 return (zfs_remove(ap->a_dvp, ap->a_vp, ap->a_cnp->cn_nameptr, 5055 ap->a_cnp->cn_cred)); 5056} 5057 5058static int 5059zfs_freebsd_mkdir(ap) 5060 struct vop_mkdir_args /* { 5061 struct vnode *a_dvp; 5062 struct vnode **a_vpp; 5063 struct componentname *a_cnp; 5064 struct vattr *a_vap; 5065 } */ *ap; 5066{ 5067 vattr_t *vap = ap->a_vap; 5068 5069 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 5070 5071 vattr_init_mask(vap); 5072 5073 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 5074 ap->a_cnp->cn_cred)); 5075} 5076 5077static int 5078zfs_freebsd_rmdir(ap) 5079 struct vop_rmdir_args /* { 5080 struct vnode *a_dvp; 5081 struct vnode *a_vp; 5082 struct componentname *a_cnp; 5083 } */ *ap; 5084{ 5085 struct componentname *cnp = ap->a_cnp; 5086 5087 ASSERT(cnp->cn_flags & SAVENAME); 5088 5089 return (zfs_rmdir(ap->a_dvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred)); 5090} 5091 5092static int 5093zfs_freebsd_readdir(ap) 5094 struct vop_readdir_args /* { 5095 struct vnode *a_vp; 5096 struct uio *a_uio; 5097 struct ucred *a_cred; 5098 int *a_eofflag; 5099 int *a_ncookies; 5100 u_long **a_cookies; 5101 } */ *ap; 5102{ 5103 5104 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 5105 ap->a_ncookies, ap->a_cookies)); 5106} 5107 5108static int 5109zfs_freebsd_fsync(ap) 5110 struct vop_fsync_args /* { 5111 struct vnode *a_vp; 5112 int a_waitfor; 5113 struct thread *a_td; 5114 } */ *ap; 5115{ 5116 5117 vop_stdfsync(ap); 5118 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL)); 5119} 5120 5121static int 5122zfs_freebsd_getattr(ap) 5123 struct vop_getattr_args /* { 5124 struct vnode *a_vp; 5125 struct vattr *a_vap; 5126 struct ucred *a_cred; 5127 } */ *ap; 5128{ 5129 vattr_t *vap = ap->a_vap; 5130 xvattr_t xvap; 5131 u_long fflags = 0; 5132 int error; 5133 5134 xva_init(&xvap); 5135 xvap.xva_vattr = *vap; 5136 xvap.xva_vattr.va_mask |= AT_XVATTR; 5137 5138 /* Convert chflags into ZFS-type flags. */ 5139 /* XXX: what about SF_SETTABLE?. */ 5140 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 5141 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 5142 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 5143 XVA_SET_REQ(&xvap, XAT_NODUMP); 5144 XVA_SET_REQ(&xvap, XAT_READONLY); 5145 XVA_SET_REQ(&xvap, XAT_ARCHIVE); 5146 XVA_SET_REQ(&xvap, XAT_SYSTEM); 5147 XVA_SET_REQ(&xvap, XAT_HIDDEN); 5148 XVA_SET_REQ(&xvap, XAT_REPARSE); 5149 XVA_SET_REQ(&xvap, XAT_OFFLINE); 5150 XVA_SET_REQ(&xvap, XAT_SPARSE); 5151 5152 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL); 5153 if (error != 0) 5154 return (error); 5155 5156 /* Convert ZFS xattr into chflags. */ 5157#define FLAG_CHECK(fflag, xflag, xfield) do { \ 5158 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 5159 fflags |= (fflag); \ 5160} while (0) 5161 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 5162 xvap.xva_xoptattrs.xoa_immutable); 5163 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 5164 xvap.xva_xoptattrs.xoa_appendonly); 5165 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 5166 xvap.xva_xoptattrs.xoa_nounlink); 5167 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE, 5168 xvap.xva_xoptattrs.xoa_archive); 5169 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 5170 xvap.xva_xoptattrs.xoa_nodump); 5171 FLAG_CHECK(UF_READONLY, XAT_READONLY, 5172 xvap.xva_xoptattrs.xoa_readonly); 5173 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM, 5174 xvap.xva_xoptattrs.xoa_system); 5175 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN, 5176 xvap.xva_xoptattrs.xoa_hidden); 5177 FLAG_CHECK(UF_REPARSE, XAT_REPARSE, 5178 xvap.xva_xoptattrs.xoa_reparse); 5179 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE, 5180 xvap.xva_xoptattrs.xoa_offline); 5181 FLAG_CHECK(UF_SPARSE, XAT_SPARSE, 5182 xvap.xva_xoptattrs.xoa_sparse); 5183 5184#undef FLAG_CHECK 5185 *vap = xvap.xva_vattr; 5186 vap->va_flags = fflags; 5187 return (0); 5188} 5189 5190static int 5191zfs_freebsd_setattr(ap) 5192 struct vop_setattr_args /* { 5193 struct vnode *a_vp; 5194 struct vattr *a_vap; 5195 struct ucred *a_cred; 5196 } */ *ap; 5197{ 5198 vnode_t *vp = ap->a_vp; 5199 vattr_t *vap = ap->a_vap; 5200 cred_t *cred = ap->a_cred; 5201 xvattr_t xvap; 5202 u_long fflags; 5203 uint64_t zflags; 5204 5205 vattr_init_mask(vap); 5206 vap->va_mask &= ~AT_NOSET; 5207 5208 xva_init(&xvap); 5209 xvap.xva_vattr = *vap; 5210 5211 zflags = VTOZ(vp)->z_pflags; 5212 5213 if (vap->va_flags != VNOVAL) { 5214 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 5215 int error; 5216 5217 if (zfsvfs->z_use_fuids == B_FALSE) 5218 return (EOPNOTSUPP); 5219 5220 fflags = vap->va_flags; 5221 /* 5222 * XXX KDM 5223 * We need to figure out whether it makes sense to allow 5224 * UF_REPARSE through, since we don't really have other 5225 * facilities to handle reparse points and zfs_setattr() 5226 * doesn't currently allow setting that attribute anyway. 5227 */ 5228 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE| 5229 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE| 5230 UF_OFFLINE|UF_SPARSE)) != 0) 5231 return (EOPNOTSUPP); 5232 /* 5233 * Unprivileged processes are not permitted to unset system 5234 * flags, or modify flags if any system flags are set. 5235 * Privileged non-jail processes may not modify system flags 5236 * if securelevel > 0 and any existing system flags are set. 5237 * Privileged jail processes behave like privileged non-jail 5238 * processes if the security.jail.chflags_allowed sysctl is 5239 * is non-zero; otherwise, they behave like unprivileged 5240 * processes. 5241 */ 5242 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 5243 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) { 5244 if (zflags & 5245 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 5246 error = securelevel_gt(cred, 0); 5247 if (error != 0) 5248 return (error); 5249 } 5250 } else { 5251 /* 5252 * Callers may only modify the file flags on objects they 5253 * have VADMIN rights for. 5254 */ 5255 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0) 5256 return (error); 5257 if (zflags & 5258 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 5259 return (EPERM); 5260 } 5261 if (fflags & 5262 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 5263 return (EPERM); 5264 } 5265 } 5266 5267#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 5268 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 5269 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 5270 XVA_SET_REQ(&xvap, (xflag)); \ 5271 (xfield) = ((fflags & (fflag)) != 0); \ 5272 } \ 5273} while (0) 5274 /* Convert chflags into ZFS-type flags. */ 5275 /* XXX: what about SF_SETTABLE?. */ 5276 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 5277 xvap.xva_xoptattrs.xoa_immutable); 5278 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 5279 xvap.xva_xoptattrs.xoa_appendonly); 5280 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 5281 xvap.xva_xoptattrs.xoa_nounlink); 5282 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE, 5283 xvap.xva_xoptattrs.xoa_archive); 5284 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 5285 xvap.xva_xoptattrs.xoa_nodump); 5286 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY, 5287 xvap.xva_xoptattrs.xoa_readonly); 5288 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM, 5289 xvap.xva_xoptattrs.xoa_system); 5290 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN, 5291 xvap.xva_xoptattrs.xoa_hidden); 5292 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE, 5293 xvap.xva_xoptattrs.xoa_hidden); 5294 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE, 5295 xvap.xva_xoptattrs.xoa_offline); 5296 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE, 5297 xvap.xva_xoptattrs.xoa_sparse); 5298#undef FLAG_CHANGE 5299 } 5300 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL)); 5301} 5302 5303static int 5304zfs_freebsd_rename(ap) 5305 struct vop_rename_args /* { 5306 struct vnode *a_fdvp; 5307 struct vnode *a_fvp; 5308 struct componentname *a_fcnp; 5309 struct vnode *a_tdvp; 5310 struct vnode *a_tvp; 5311 struct componentname *a_tcnp; 5312 } */ *ap; 5313{ 5314 vnode_t *fdvp = ap->a_fdvp; 5315 vnode_t *fvp = ap->a_fvp; 5316 vnode_t *tdvp = ap->a_tdvp; 5317 vnode_t *tvp = ap->a_tvp; 5318 int error; 5319 5320 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 5321 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 5322 5323 error = zfs_rename(fdvp, &fvp, ap->a_fcnp, tdvp, &tvp, 5324 ap->a_tcnp, ap->a_fcnp->cn_cred); 5325 5326 vrele(fdvp); 5327 vrele(fvp); 5328 vrele(tdvp); 5329 if (tvp != NULL) 5330 vrele(tvp); 5331 5332 return (error); 5333} 5334 5335static int 5336zfs_freebsd_symlink(ap) 5337 struct vop_symlink_args /* { 5338 struct vnode *a_dvp; 5339 struct vnode **a_vpp; 5340 struct componentname *a_cnp; 5341 struct vattr *a_vap; 5342 char *a_target; 5343 } */ *ap; 5344{ 5345 struct componentname *cnp = ap->a_cnp; 5346 vattr_t *vap = ap->a_vap; 5347 5348 ASSERT(cnp->cn_flags & SAVENAME); 5349 5350 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 5351 vattr_init_mask(vap); 5352 5353 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 5354 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 5355} 5356 5357static int 5358zfs_freebsd_readlink(ap) 5359 struct vop_readlink_args /* { 5360 struct vnode *a_vp; 5361 struct uio *a_uio; 5362 struct ucred *a_cred; 5363 } */ *ap; 5364{ 5365 5366 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL)); 5367} 5368 5369static int 5370zfs_freebsd_link(ap) 5371 struct vop_link_args /* { 5372 struct vnode *a_tdvp; 5373 struct vnode *a_vp; 5374 struct componentname *a_cnp; 5375 } */ *ap; 5376{ 5377 struct componentname *cnp = ap->a_cnp; 5378 vnode_t *vp = ap->a_vp; 5379 vnode_t *tdvp = ap->a_tdvp; 5380 5381 if (tdvp->v_mount != vp->v_mount) 5382 return (EXDEV); 5383 5384 ASSERT(cnp->cn_flags & SAVENAME); 5385 5386 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0)); 5387} 5388 5389static int 5390zfs_freebsd_inactive(ap) 5391 struct vop_inactive_args /* { 5392 struct vnode *a_vp; 5393 struct thread *a_td; 5394 } */ *ap; 5395{ 5396 vnode_t *vp = ap->a_vp; 5397 5398 zfs_inactive(vp, ap->a_td->td_ucred, NULL); 5399 return (0); 5400} 5401 5402static int 5403zfs_freebsd_reclaim(ap) 5404 struct vop_reclaim_args /* { 5405 struct vnode *a_vp; 5406 struct thread *a_td; 5407 } */ *ap; 5408{ 5409 vnode_t *vp = ap->a_vp; 5410 znode_t *zp = VTOZ(vp); 5411 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5412 5413 ASSERT(zp != NULL); 5414 5415 /* Destroy the vm object and flush associated pages. */ 5416 vnode_destroy_vobject(vp); 5417 5418 /* 5419 * z_teardown_inactive_lock protects from a race with 5420 * zfs_znode_dmu_fini in zfsvfs_teardown during 5421 * force unmount. 5422 */ 5423 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 5424 if (zp->z_sa_hdl == NULL) 5425 zfs_znode_free(zp); 5426 else 5427 zfs_zinactive(zp); 5428 rw_exit(&zfsvfs->z_teardown_inactive_lock); 5429 5430 vp->v_data = NULL; 5431 return (0); 5432} 5433 5434static int 5435zfs_freebsd_fid(ap) 5436 struct vop_fid_args /* { 5437 struct vnode *a_vp; 5438 struct fid *a_fid; 5439 } */ *ap; 5440{ 5441 5442 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 5443} 5444 5445static int 5446zfs_freebsd_pathconf(ap) 5447 struct vop_pathconf_args /* { 5448 struct vnode *a_vp; 5449 int a_name; 5450 register_t *a_retval; 5451 } */ *ap; 5452{ 5453 ulong_t val; 5454 int error; 5455 5456 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); 5457 if (error == 0) 5458 *ap->a_retval = val; 5459 else if (error == EOPNOTSUPP) 5460 error = vop_stdpathconf(ap); 5461 return (error); 5462} 5463 5464static int 5465zfs_freebsd_fifo_pathconf(ap) 5466 struct vop_pathconf_args /* { 5467 struct vnode *a_vp; 5468 int a_name; 5469 register_t *a_retval; 5470 } */ *ap; 5471{ 5472 5473 switch (ap->a_name) { 5474 case _PC_ACL_EXTENDED: 5475 case _PC_ACL_NFS4: 5476 case _PC_ACL_PATH_MAX: 5477 case _PC_MAC_PRESENT: 5478 return (zfs_freebsd_pathconf(ap)); 5479 default: 5480 return (fifo_specops.vop_pathconf(ap)); 5481 } 5482} 5483 5484/* 5485 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 5486 * extended attribute name: 5487 * 5488 * NAMESPACE PREFIX 5489 * system freebsd:system: 5490 * user (none, can be used to access ZFS fsattr(5) attributes 5491 * created on Solaris) 5492 */ 5493static int 5494zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 5495 size_t size) 5496{ 5497 const char *namespace, *prefix, *suffix; 5498 5499 /* We don't allow '/' character in attribute name. */ 5500 if (strchr(name, '/') != NULL) 5501 return (EINVAL); 5502 /* We don't allow attribute names that start with "freebsd:" string. */ 5503 if (strncmp(name, "freebsd:", 8) == 0) 5504 return (EINVAL); 5505 5506 bzero(attrname, size); 5507 5508 switch (attrnamespace) { 5509 case EXTATTR_NAMESPACE_USER: 5510#if 0 5511 prefix = "freebsd:"; 5512 namespace = EXTATTR_NAMESPACE_USER_STRING; 5513 suffix = ":"; 5514#else 5515 /* 5516 * This is the default namespace by which we can access all 5517 * attributes created on Solaris. 5518 */ 5519 prefix = namespace = suffix = ""; 5520#endif 5521 break; 5522 case EXTATTR_NAMESPACE_SYSTEM: 5523 prefix = "freebsd:"; 5524 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 5525 suffix = ":"; 5526 break; 5527 case EXTATTR_NAMESPACE_EMPTY: 5528 default: 5529 return (EINVAL); 5530 } 5531 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 5532 name) >= size) { 5533 return (ENAMETOOLONG); 5534 } 5535 return (0); 5536} 5537 5538/* 5539 * Vnode operating to retrieve a named extended attribute. 5540 */ 5541static int 5542zfs_getextattr(struct vop_getextattr_args *ap) 5543/* 5544vop_getextattr { 5545 IN struct vnode *a_vp; 5546 IN int a_attrnamespace; 5547 IN const char *a_name; 5548 INOUT struct uio *a_uio; 5549 OUT size_t *a_size; 5550 IN struct ucred *a_cred; 5551 IN struct thread *a_td; 5552}; 5553*/ 5554{ 5555 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 5556 struct thread *td = ap->a_td; 5557 struct nameidata nd; 5558 char attrname[255]; 5559 struct vattr va; 5560 vnode_t *xvp = NULL, *vp; 5561 int error, flags; 5562 5563 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 5564 ap->a_cred, ap->a_td, VREAD); 5565 if (error != 0) 5566 return (error); 5567 5568 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 5569 sizeof(attrname)); 5570 if (error != 0) 5571 return (error); 5572 5573 ZFS_ENTER(zfsvfs); 5574 5575 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 5576 LOOKUP_XATTR); 5577 if (error != 0) { 5578 ZFS_EXIT(zfsvfs); 5579 return (error); 5580 } 5581 5582 flags = FREAD; 5583 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 5584 xvp, td); 5585 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL); 5586 vp = nd.ni_vp; 5587 NDFREE(&nd, NDF_ONLY_PNBUF); 5588 if (error != 0) { 5589 ZFS_EXIT(zfsvfs); 5590 if (error == ENOENT) 5591 error = ENOATTR; 5592 return (error); 5593 } 5594 5595 if (ap->a_size != NULL) { 5596 error = VOP_GETATTR(vp, &va, ap->a_cred); 5597 if (error == 0) 5598 *ap->a_size = (size_t)va.va_size; 5599 } else if (ap->a_uio != NULL) 5600 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); 5601 5602 VOP_UNLOCK(vp, 0); 5603 vn_close(vp, flags, ap->a_cred, td); 5604 ZFS_EXIT(zfsvfs); 5605 5606 return (error); 5607} 5608 5609/* 5610 * Vnode operation to remove a named attribute. 5611 */ 5612int 5613zfs_deleteextattr(struct vop_deleteextattr_args *ap) 5614/* 5615vop_deleteextattr { 5616 IN struct vnode *a_vp; 5617 IN int a_attrnamespace; 5618 IN const char *a_name; 5619 IN struct ucred *a_cred; 5620 IN struct thread *a_td; 5621}; 5622*/ 5623{ 5624 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 5625 struct thread *td = ap->a_td; 5626 struct nameidata nd; 5627 char attrname[255]; 5628 struct vattr va; 5629 vnode_t *xvp = NULL, *vp; 5630 int error, flags; 5631 5632 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 5633 ap->a_cred, ap->a_td, VWRITE); 5634 if (error != 0) 5635 return (error); 5636 5637 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 5638 sizeof(attrname)); 5639 if (error != 0) 5640 return (error); 5641 5642 ZFS_ENTER(zfsvfs); 5643 5644 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 5645 LOOKUP_XATTR); 5646 if (error != 0) { 5647 ZFS_EXIT(zfsvfs); 5648 return (error); 5649 } 5650 5651 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 5652 UIO_SYSSPACE, attrname, xvp, td); 5653 error = namei(&nd); 5654 vp = nd.ni_vp; 5655 if (error != 0) { 5656 ZFS_EXIT(zfsvfs); 5657 NDFREE(&nd, NDF_ONLY_PNBUF); 5658 if (error == ENOENT) 5659 error = ENOATTR; 5660 return (error); 5661 } 5662 5663 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 5664 NDFREE(&nd, NDF_ONLY_PNBUF); 5665 5666 vput(nd.ni_dvp); 5667 if (vp == nd.ni_dvp) 5668 vrele(vp); 5669 else 5670 vput(vp); 5671 ZFS_EXIT(zfsvfs); 5672 5673 return (error); 5674} 5675 5676/* 5677 * Vnode operation to set a named attribute. 5678 */ 5679static int 5680zfs_setextattr(struct vop_setextattr_args *ap) 5681/* 5682vop_setextattr { 5683 IN struct vnode *a_vp; 5684 IN int a_attrnamespace; 5685 IN const char *a_name; 5686 INOUT struct uio *a_uio; 5687 IN struct ucred *a_cred; 5688 IN struct thread *a_td; 5689}; 5690*/ 5691{ 5692 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 5693 struct thread *td = ap->a_td; 5694 struct nameidata nd; 5695 char attrname[255]; 5696 struct vattr va; 5697 vnode_t *xvp = NULL, *vp; 5698 int error, flags; 5699 5700 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 5701 ap->a_cred, ap->a_td, VWRITE); 5702 if (error != 0) 5703 return (error); 5704 5705 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 5706 sizeof(attrname)); 5707 if (error != 0) 5708 return (error); 5709 5710 ZFS_ENTER(zfsvfs); 5711 5712 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 5713 LOOKUP_XATTR | CREATE_XATTR_DIR); 5714 if (error != 0) { 5715 ZFS_EXIT(zfsvfs); 5716 return (error); 5717 } 5718 5719 flags = FFLAGS(O_WRONLY | O_CREAT); 5720 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 5721 xvp, td); 5722 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL); 5723 vp = nd.ni_vp; 5724 NDFREE(&nd, NDF_ONLY_PNBUF); 5725 if (error != 0) { 5726 ZFS_EXIT(zfsvfs); 5727 return (error); 5728 } 5729 5730 VATTR_NULL(&va); 5731 va.va_size = 0; 5732 error = VOP_SETATTR(vp, &va, ap->a_cred); 5733 if (error == 0) 5734 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred); 5735 5736 VOP_UNLOCK(vp, 0); 5737 vn_close(vp, flags, ap->a_cred, td); 5738 ZFS_EXIT(zfsvfs); 5739 5740 return (error); 5741} 5742 5743/* 5744 * Vnode operation to retrieve extended attributes on a vnode. 5745 */ 5746static int 5747zfs_listextattr(struct vop_listextattr_args *ap) 5748/* 5749vop_listextattr { 5750 IN struct vnode *a_vp; 5751 IN int a_attrnamespace; 5752 INOUT struct uio *a_uio; 5753 OUT size_t *a_size; 5754 IN struct ucred *a_cred; 5755 IN struct thread *a_td; 5756}; 5757*/ 5758{ 5759 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 5760 struct thread *td = ap->a_td; 5761 struct nameidata nd; 5762 char attrprefix[16]; 5763 u_char dirbuf[sizeof(struct dirent)]; 5764 struct dirent *dp; 5765 struct iovec aiov; 5766 struct uio auio, *uio = ap->a_uio; 5767 size_t *sizep = ap->a_size; 5768 size_t plen; 5769 vnode_t *xvp = NULL, *vp; 5770 int done, error, eof, pos; 5771 5772 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 5773 ap->a_cred, ap->a_td, VREAD); 5774 if (error != 0) 5775 return (error); 5776 5777 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 5778 sizeof(attrprefix)); 5779 if (error != 0) 5780 return (error); 5781 plen = strlen(attrprefix); 5782 5783 ZFS_ENTER(zfsvfs); 5784 5785 if (sizep != NULL) 5786 *sizep = 0; 5787 5788 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 5789 LOOKUP_XATTR); 5790 if (error != 0) { 5791 ZFS_EXIT(zfsvfs); 5792 /* 5793 * ENOATTR means that the EA directory does not yet exist, 5794 * i.e. there are no extended attributes there. 5795 */ 5796 if (error == ENOATTR) 5797 error = 0; 5798 return (error); 5799 } 5800 5801 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 5802 UIO_SYSSPACE, ".", xvp, td); 5803 error = namei(&nd); 5804 vp = nd.ni_vp; 5805 NDFREE(&nd, NDF_ONLY_PNBUF); 5806 if (error != 0) { 5807 ZFS_EXIT(zfsvfs); 5808 return (error); 5809 } 5810 5811 auio.uio_iov = &aiov; 5812 auio.uio_iovcnt = 1; 5813 auio.uio_segflg = UIO_SYSSPACE; 5814 auio.uio_td = td; 5815 auio.uio_rw = UIO_READ; 5816 auio.uio_offset = 0; 5817 5818 do { 5819 u_char nlen; 5820 5821 aiov.iov_base = (void *)dirbuf; 5822 aiov.iov_len = sizeof(dirbuf); 5823 auio.uio_resid = sizeof(dirbuf); 5824 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 5825 done = sizeof(dirbuf) - auio.uio_resid; 5826 if (error != 0) 5827 break; 5828 for (pos = 0; pos < done;) { 5829 dp = (struct dirent *)(dirbuf + pos); 5830 pos += dp->d_reclen; 5831 /* 5832 * XXX: Temporarily we also accept DT_UNKNOWN, as this 5833 * is what we get when attribute was created on Solaris. 5834 */ 5835 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 5836 continue; 5837 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0) 5838 continue; 5839 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 5840 continue; 5841 nlen = dp->d_namlen - plen; 5842 if (sizep != NULL) 5843 *sizep += 1 + nlen; 5844 else if (uio != NULL) { 5845 /* 5846 * Format of extattr name entry is one byte for 5847 * length and the rest for name. 5848 */ 5849 error = uiomove(&nlen, 1, uio->uio_rw, uio); 5850 if (error == 0) { 5851 error = uiomove(dp->d_name + plen, nlen, 5852 uio->uio_rw, uio); 5853 } 5854 if (error != 0) 5855 break; 5856 } 5857 } 5858 } while (!eof && error == 0); 5859 5860 vput(vp); 5861 ZFS_EXIT(zfsvfs); 5862 5863 return (error); 5864} 5865 5866int 5867zfs_freebsd_getacl(ap) 5868 struct vop_getacl_args /* { 5869 struct vnode *vp; 5870 acl_type_t type; 5871 struct acl *aclp; 5872 struct ucred *cred; 5873 struct thread *td; 5874 } */ *ap; 5875{ 5876 int error; 5877 vsecattr_t vsecattr; 5878 5879 if (ap->a_type != ACL_TYPE_NFS4) 5880 return (EINVAL); 5881 5882 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 5883 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL)) 5884 return (error); 5885 5886 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt); 5887 if (vsecattr.vsa_aclentp != NULL) 5888 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 5889 5890 return (error); 5891} 5892 5893int 5894zfs_freebsd_setacl(ap) 5895 struct vop_setacl_args /* { 5896 struct vnode *vp; 5897 acl_type_t type; 5898 struct acl *aclp; 5899 struct ucred *cred; 5900 struct thread *td; 5901 } */ *ap; 5902{ 5903 int error; 5904 vsecattr_t vsecattr; 5905 int aclbsize; /* size of acl list in bytes */ 5906 aclent_t *aaclp; 5907 5908 if (ap->a_type != ACL_TYPE_NFS4) 5909 return (EINVAL); 5910 5911 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 5912 return (EINVAL); 5913 5914 /* 5915 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 5916 * splitting every entry into two and appending "canonical six" 5917 * entries at the end. Don't allow for setting an ACL that would 5918 * cause chmod(2) to run out of ACL entries. 5919 */ 5920 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 5921 return (ENOSPC); 5922 5923 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 5924 if (error != 0) 5925 return (error); 5926 5927 vsecattr.vsa_mask = VSA_ACE; 5928 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t); 5929 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 5930 aaclp = vsecattr.vsa_aclentp; 5931 vsecattr.vsa_aclentsz = aclbsize; 5932 5933 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 5934 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL); 5935 kmem_free(aaclp, aclbsize); 5936 5937 return (error); 5938} 5939 5940int 5941zfs_freebsd_aclcheck(ap) 5942 struct vop_aclcheck_args /* { 5943 struct vnode *vp; 5944 acl_type_t type; 5945 struct acl *aclp; 5946 struct ucred *cred; 5947 struct thread *td; 5948 } */ *ap; 5949{ 5950 5951 return (EOPNOTSUPP); 5952} 5953 5954static int 5955zfs_vptocnp(struct vop_vptocnp_args *ap) 5956{ 5957 vnode_t *covered_vp; 5958 vnode_t *vp = ap->a_vp;; 5959 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 5960 znode_t *zp = VTOZ(vp); 5961 uint64_t parent; 5962 int ltype; 5963 int error; 5964 5965 ZFS_ENTER(zfsvfs); 5966 ZFS_VERIFY_ZP(zp); 5967 5968 /* 5969 * If we are a snapshot mounted under .zfs, run the operation 5970 * on the covered vnode. 5971 */ 5972 if ((error = sa_lookup(zp->z_sa_hdl, 5973 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) { 5974 ZFS_EXIT(zfsvfs); 5975 return (error); 5976 } 5977 5978 if (zp->z_id != parent || zfsvfs->z_parent == zfsvfs) { 5979 ZFS_EXIT(zfsvfs); 5980 return (vop_stdvptocnp(ap)); 5981 } 5982 ZFS_EXIT(zfsvfs); 5983 5984 covered_vp = vp->v_mount->mnt_vnodecovered; 5985 vhold(covered_vp); 5986 ltype = VOP_ISLOCKED(vp); 5987 VOP_UNLOCK(vp, 0); 5988 error = vget(covered_vp, LK_EXCLUSIVE, curthread); 5989 vdrop(covered_vp); 5990 if (error == 0) { 5991 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_cred, 5992 ap->a_buf, ap->a_buflen); 5993 vput(covered_vp); 5994 } 5995 vn_lock(vp, ltype | LK_RETRY); 5996 if ((vp->v_iflag & VI_DOOMED) != 0) 5997 error = SET_ERROR(ENOENT); 5998 return (error); 5999} 6000 6001#ifdef DIAGNOSTIC 6002static int 6003zfs_lock(ap) 6004 struct vop_lock1_args /* { 6005 struct vnode *a_vp; 6006 int a_flags; 6007 char *file; 6008 int line; 6009 } */ *ap; 6010{ 6011 zfsvfs_t *zfsvfs; 6012 znode_t *zp; 6013 vnode_t *vp; 6014 int flags; 6015 int err; 6016 6017 vp = ap->a_vp; 6018 flags = ap->a_flags; 6019 if ((flags & LK_INTERLOCK) == 0 && (flags & LK_NOWAIT) == 0 && 6020 (vp->v_iflag & VI_DOOMED) == 0 && (zp = vp->v_data) != NULL) { 6021 zfsvfs = zp->z_zfsvfs; 6022 VERIFY(!RRM_LOCK_HELD(&zfsvfs->z_teardown_lock)); 6023 } 6024 err = vop_stdlock(ap); 6025 if ((flags & LK_INTERLOCK) != 0 && (flags & LK_NOWAIT) == 0 && 6026 (vp->v_iflag & VI_DOOMED) == 0 && (zp = vp->v_data) != NULL) { 6027 zfsvfs = zp->z_zfsvfs; 6028 VERIFY(!RRM_LOCK_HELD(&zfsvfs->z_teardown_lock)); 6029 } 6030 return (err); 6031} 6032#endif 6033 6034struct vop_vector zfs_vnodeops; 6035struct vop_vector zfs_fifoops; 6036struct vop_vector zfs_shareops; 6037 6038struct vop_vector zfs_vnodeops = { 6039 .vop_default = &default_vnodeops, 6040 .vop_inactive = zfs_freebsd_inactive, 6041 .vop_reclaim = zfs_freebsd_reclaim, 6042 .vop_access = zfs_freebsd_access, 6043 .vop_lookup = zfs_cache_lookup, 6044 .vop_cachedlookup = zfs_freebsd_lookup, 6045 .vop_getattr = zfs_freebsd_getattr, 6046 .vop_setattr = zfs_freebsd_setattr, 6047 .vop_create = zfs_freebsd_create, 6048 .vop_mknod = zfs_freebsd_create, 6049 .vop_mkdir = zfs_freebsd_mkdir, 6050 .vop_readdir = zfs_freebsd_readdir, 6051 .vop_fsync = zfs_freebsd_fsync, 6052 .vop_open = zfs_freebsd_open, 6053 .vop_close = zfs_freebsd_close, 6054 .vop_rmdir = zfs_freebsd_rmdir, 6055 .vop_ioctl = zfs_freebsd_ioctl, 6056 .vop_link = zfs_freebsd_link, 6057 .vop_symlink = zfs_freebsd_symlink, 6058 .vop_readlink = zfs_freebsd_readlink, 6059 .vop_read = zfs_freebsd_read, 6060 .vop_write = zfs_freebsd_write, 6061 .vop_remove = zfs_freebsd_remove, 6062 .vop_rename = zfs_freebsd_rename, 6063 .vop_pathconf = zfs_freebsd_pathconf, 6064 .vop_bmap = zfs_freebsd_bmap, 6065 .vop_fid = zfs_freebsd_fid, 6066 .vop_getextattr = zfs_getextattr, 6067 .vop_deleteextattr = zfs_deleteextattr, 6068 .vop_setextattr = zfs_setextattr, 6069 .vop_listextattr = zfs_listextattr, 6070 .vop_getacl = zfs_freebsd_getacl, 6071 .vop_setacl = zfs_freebsd_setacl, 6072 .vop_aclcheck = zfs_freebsd_aclcheck, 6073 .vop_getpages = zfs_freebsd_getpages, 6074 .vop_putpages = zfs_freebsd_putpages, 6075 .vop_vptocnp = zfs_vptocnp, 6076#ifdef DIAGNOSTIC 6077 .vop_lock1 = zfs_lock, 6078#endif 6079}; 6080 6081struct vop_vector zfs_fifoops = { 6082 .vop_default = &fifo_specops, 6083 .vop_fsync = zfs_freebsd_fsync, 6084 .vop_access = zfs_freebsd_access, 6085 .vop_getattr = zfs_freebsd_getattr, 6086 .vop_inactive = zfs_freebsd_inactive, 6087 .vop_read = VOP_PANIC, 6088 .vop_reclaim = zfs_freebsd_reclaim, 6089 .vop_setattr = zfs_freebsd_setattr, 6090 .vop_write = VOP_PANIC, 6091 .vop_pathconf = zfs_freebsd_fifo_pathconf, 6092 .vop_fid = zfs_freebsd_fid, 6093 .vop_getacl = zfs_freebsd_getacl, 6094 .vop_setacl = zfs_freebsd_setacl, 6095 .vop_aclcheck = zfs_freebsd_aclcheck, 6096}; 6097 6098/* 6099 * special share hidden files vnode operations template 6100 */ 6101struct vop_vector zfs_shareops = { 6102 .vop_default = &default_vnodeops, 6103 .vop_access = zfs_freebsd_access, 6104 .vop_inactive = zfs_freebsd_inactive, 6105 .vop_reclaim = zfs_freebsd_reclaim, 6106 .vop_fid = zfs_freebsd_fid, 6107 .vop_pathconf = zfs_freebsd_pathconf, 6108}; 6109