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