zfs_vnops.c revision 258563
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2013 by Delphix. All rights reserved. 24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved. 25 */ 26 27/* Portions Copyright 2007 Jeremy Teo */ 28/* Portions Copyright 2010 Robert Milkowski */ 29 30#include <sys/types.h> 31#include <sys/param.h> 32#include <sys/time.h> 33#include <sys/systm.h> 34#include <sys/sysmacros.h> 35#include <sys/resource.h> 36#include <sys/vfs.h> 37#include <sys/vm.h> 38#include <sys/vnode.h> 39#include <sys/file.h> 40#include <sys/stat.h> 41#include <sys/kmem.h> 42#include <sys/taskq.h> 43#include <sys/uio.h> 44#include <sys/atomic.h> 45#include <sys/namei.h> 46#include <sys/mman.h> 47#include <sys/cmn_err.h> 48#include <sys/errno.h> 49#include <sys/unistd.h> 50#include <sys/zfs_dir.h> 51#include <sys/zfs_ioctl.h> 52#include <sys/fs/zfs.h> 53#include <sys/dmu.h> 54#include <sys/dmu_objset.h> 55#include <sys/spa.h> 56#include <sys/txg.h> 57#include <sys/dbuf.h> 58#include <sys/zap.h> 59#include <sys/sa.h> 60#include <sys/dirent.h> 61#include <sys/policy.h> 62#include <sys/sunddi.h> 63#include <sys/filio.h> 64#include <sys/sid.h> 65#include <sys/zfs_ctldir.h> 66#include <sys/zfs_fuid.h> 67#include <sys/zfs_sa.h> 68#include <sys/dnlc.h> 69#include <sys/zfs_rlock.h> 70#include <sys/extdirent.h> 71#include <sys/kidmap.h> 72#include <sys/bio.h> 73#include <sys/buf.h> 74#include <sys/sf_buf.h> 75#include <sys/sched.h> 76#include <sys/acl.h> 77#include <vm/vm_param.h> 78#include <vm/vm_pageout.h> 79 80/* 81 * Programming rules. 82 * 83 * Each vnode op performs some logical unit of work. To do this, the ZPL must 84 * properly lock its in-core state, create a DMU transaction, do the work, 85 * record this work in the intent log (ZIL), commit the DMU transaction, 86 * and wait for the intent log to commit if it is a synchronous operation. 87 * Moreover, the vnode ops must work in both normal and log replay context. 88 * The ordering of events is important to avoid deadlocks and references 89 * to freed memory. The example below illustrates the following Big Rules: 90 * 91 * (1) A check must be made in each zfs thread for a mounted file system. 92 * This is done avoiding races using ZFS_ENTER(zfsvfs). 93 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes 94 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros 95 * can return EIO from the calling function. 96 * 97 * (2) VN_RELE() should always be the last thing except for zil_commit() 98 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 99 * First, if it's the last reference, the vnode/znode 100 * can be freed, so the zp may point to freed memory. Second, the last 101 * reference will call zfs_zinactive(), which may induce a lot of work -- 102 * pushing cached pages (which acquires range locks) and syncing out 103 * cached atime changes. Third, zfs_zinactive() may require a new tx, 104 * which could deadlock the system if you were already holding one. 105 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). 106 * 107 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 108 * as they can span dmu_tx_assign() calls. 109 * 110 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign(). 111 * This is critical because we don't want to block while holding locks. 112 * Note, in particular, that if a lock is sometimes acquired before 113 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 114 * use a non-blocking assign can deadlock the system. The scenario: 115 * 116 * Thread A has grabbed a lock before calling dmu_tx_assign(). 117 * Thread B is in an already-assigned tx, and blocks for this lock. 118 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 119 * forever, because the previous txg can't quiesce until B's tx commits. 120 * 121 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 122 * then drop all locks, call dmu_tx_wait(), and try again. 123 * 124 * (5) If the operation succeeded, generate the intent log entry for it 125 * before dropping locks. This ensures that the ordering of events 126 * in the intent log matches the order in which they actually occurred. 127 * During ZIL replay the zfs_log_* functions will update the sequence 128 * number to indicate the zil transaction has replayed. 129 * 130 * (6) At the end of each vnode op, the DMU tx must always commit, 131 * regardless of whether there were any errors. 132 * 133 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 134 * to ensure that synchronous semantics are provided when necessary. 135 * 136 * In general, this is how things should be ordered in each vnode op: 137 * 138 * ZFS_ENTER(zfsvfs); // exit if unmounted 139 * top: 140 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 141 * rw_enter(...); // grab any other locks you need 142 * tx = dmu_tx_create(...); // get DMU tx 143 * dmu_tx_hold_*(); // hold each object you might modify 144 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign 145 * if (error) { 146 * rw_exit(...); // drop locks 147 * zfs_dirent_unlock(dl); // unlock directory entry 148 * VN_RELE(...); // release held vnodes 149 * if (error == ERESTART) { 150 * dmu_tx_wait(tx); 151 * dmu_tx_abort(tx); 152 * goto top; 153 * } 154 * dmu_tx_abort(tx); // abort DMU tx 155 * ZFS_EXIT(zfsvfs); // finished in zfs 156 * return (error); // really out of space 157 * } 158 * error = do_real_work(); // do whatever this VOP does 159 * if (error == 0) 160 * zfs_log_*(...); // on success, make ZIL entry 161 * dmu_tx_commit(tx); // commit DMU tx -- error or not 162 * rw_exit(...); // drop locks 163 * zfs_dirent_unlock(dl); // unlock directory entry 164 * VN_RELE(...); // release held vnodes 165 * zil_commit(zilog, foid); // synchronous when necessary 166 * ZFS_EXIT(zfsvfs); // finished in zfs 167 * return (error); // done, report error 168 */ 169 170/* ARGSUSED */ 171static int 172zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 173{ 174 znode_t *zp = VTOZ(*vpp); 175 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 176 177 ZFS_ENTER(zfsvfs); 178 ZFS_VERIFY_ZP(zp); 179 180 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 181 ((flag & FAPPEND) == 0)) { 182 ZFS_EXIT(zfsvfs); 183 return (SET_ERROR(EPERM)); 184 } 185 186 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 187 ZTOV(zp)->v_type == VREG && 188 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) { 189 if (fs_vscan(*vpp, cr, 0) != 0) { 190 ZFS_EXIT(zfsvfs); 191 return (SET_ERROR(EACCES)); 192 } 193 } 194 195 /* Keep a count of the synchronous opens in the znode */ 196 if (flag & (FSYNC | FDSYNC)) 197 atomic_inc_32(&zp->z_sync_cnt); 198 199 ZFS_EXIT(zfsvfs); 200 return (0); 201} 202 203/* ARGSUSED */ 204static int 205zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 206 caller_context_t *ct) 207{ 208 znode_t *zp = VTOZ(vp); 209 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 210 211 /* 212 * Clean up any locks held by this process on the vp. 213 */ 214 cleanlocks(vp, ddi_get_pid(), 0); 215 cleanshares(vp, ddi_get_pid()); 216 217 ZFS_ENTER(zfsvfs); 218 ZFS_VERIFY_ZP(zp); 219 220 /* Decrement the synchronous opens in the znode */ 221 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 222 atomic_dec_32(&zp->z_sync_cnt); 223 224 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 225 ZTOV(zp)->v_type == VREG && 226 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) 227 VERIFY(fs_vscan(vp, cr, 1) == 0); 228 229 ZFS_EXIT(zfsvfs); 230 return (0); 231} 232 233/* 234 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 235 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 236 */ 237static int 238zfs_holey(vnode_t *vp, u_long cmd, offset_t *off) 239{ 240 znode_t *zp = VTOZ(vp); 241 uint64_t noff = (uint64_t)*off; /* new offset */ 242 uint64_t file_sz; 243 int error; 244 boolean_t hole; 245 246 file_sz = zp->z_size; 247 if (noff >= file_sz) { 248 return (SET_ERROR(ENXIO)); 249 } 250 251 if (cmd == _FIO_SEEK_HOLE) 252 hole = B_TRUE; 253 else 254 hole = B_FALSE; 255 256 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 257 258 /* end of file? */ 259 if ((error == ESRCH) || (noff > file_sz)) { 260 /* 261 * Handle the virtual hole at the end of file. 262 */ 263 if (hole) { 264 *off = file_sz; 265 return (0); 266 } 267 return (SET_ERROR(ENXIO)); 268 } 269 270 if (noff < *off) 271 return (error); 272 *off = noff; 273 return (error); 274} 275 276/* ARGSUSED */ 277static int 278zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred, 279 int *rvalp, caller_context_t *ct) 280{ 281 offset_t off; 282 int error; 283 zfsvfs_t *zfsvfs; 284 znode_t *zp; 285 286 switch (com) { 287 case _FIOFFS: 288 return (0); 289 290 /* 291 * The following two ioctls are used by bfu. Faking out, 292 * necessary to avoid bfu errors. 293 */ 294 case _FIOGDIO: 295 case _FIOSDIO: 296 return (0); 297 298 case _FIO_SEEK_DATA: 299 case _FIO_SEEK_HOLE: 300#ifdef sun 301 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 302 return (SET_ERROR(EFAULT)); 303#else 304 off = *(offset_t *)data; 305#endif 306 zp = VTOZ(vp); 307 zfsvfs = zp->z_zfsvfs; 308 ZFS_ENTER(zfsvfs); 309 ZFS_VERIFY_ZP(zp); 310 311 /* offset parameter is in/out */ 312 error = zfs_holey(vp, com, &off); 313 ZFS_EXIT(zfsvfs); 314 if (error) 315 return (error); 316#ifdef sun 317 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 318 return (SET_ERROR(EFAULT)); 319#else 320 *(offset_t *)data = off; 321#endif 322 return (0); 323 } 324 return (SET_ERROR(ENOTTY)); 325} 326 327static vm_page_t 328page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) 329{ 330 vm_object_t obj; 331 vm_page_t pp; 332 int64_t end; 333 334 /* 335 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE 336 * aligned boundaries, if the range is not aligned. As a result a 337 * DEV_BSIZE subrange with partially dirty data may get marked as clean. 338 * It may happen that all DEV_BSIZE subranges are marked clean and thus 339 * the whole page would be considred clean despite have some dirty data. 340 * For this reason we should shrink the range to DEV_BSIZE aligned 341 * boundaries before calling vm_page_clear_dirty. 342 */ 343 end = rounddown2(off + nbytes, DEV_BSIZE); 344 off = roundup2(off, DEV_BSIZE); 345 nbytes = end - off; 346 347 obj = vp->v_object; 348 zfs_vmobject_assert_wlocked(obj); 349 350 for (;;) { 351 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 352 pp->valid) { 353 if (vm_page_xbusied(pp)) { 354 /* 355 * Reference the page before unlocking and 356 * sleeping so that the page daemon is less 357 * likely to reclaim it. 358 */ 359 vm_page_reference(pp); 360 vm_page_lock(pp); 361 zfs_vmobject_wunlock(obj); 362 vm_page_busy_sleep(pp, "zfsmwb"); 363 zfs_vmobject_wlock(obj); 364 continue; 365 } 366 vm_page_sbusy(pp); 367 } else if (pp == NULL) { 368 pp = vm_page_alloc(obj, OFF_TO_IDX(start), 369 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED | 370 VM_ALLOC_SBUSY); 371 } else { 372 ASSERT(pp != NULL && !pp->valid); 373 pp = NULL; 374 } 375 376 if (pp != NULL) { 377 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 378 vm_object_pip_add(obj, 1); 379 pmap_remove_write(pp); 380 if (nbytes != 0) 381 vm_page_clear_dirty(pp, off, nbytes); 382 } 383 break; 384 } 385 return (pp); 386} 387 388static void 389page_unbusy(vm_page_t pp) 390{ 391 392 vm_page_sunbusy(pp); 393 vm_object_pip_subtract(pp->object, 1); 394} 395 396static vm_page_t 397page_hold(vnode_t *vp, int64_t start) 398{ 399 vm_object_t obj; 400 vm_page_t pp; 401 402 obj = vp->v_object; 403 zfs_vmobject_assert_wlocked(obj); 404 405 for (;;) { 406 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 407 pp->valid) { 408 if (vm_page_xbusied(pp)) { 409 /* 410 * Reference the page before unlocking and 411 * sleeping so that the page daemon is less 412 * likely to reclaim it. 413 */ 414 vm_page_reference(pp); 415 vm_page_lock(pp); 416 zfs_vmobject_wunlock(obj); 417 vm_page_busy_sleep(pp, "zfsmwb"); 418 zfs_vmobject_wlock(obj); 419 continue; 420 } 421 422 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 423 vm_page_lock(pp); 424 vm_page_hold(pp); 425 vm_page_unlock(pp); 426 427 } else 428 pp = NULL; 429 break; 430 } 431 return (pp); 432} 433 434static void 435page_unhold(vm_page_t pp) 436{ 437 438 vm_page_lock(pp); 439 vm_page_unhold(pp); 440 vm_page_unlock(pp); 441} 442 443static caddr_t 444zfs_map_page(vm_page_t pp, struct sf_buf **sfp) 445{ 446 447 *sfp = sf_buf_alloc(pp, 0); 448 return ((caddr_t)sf_buf_kva(*sfp)); 449} 450 451static void 452zfs_unmap_page(struct sf_buf *sf) 453{ 454 455 sf_buf_free(sf); 456} 457 458/* 459 * When a file is memory mapped, we must keep the IO data synchronized 460 * between the DMU cache and the memory mapped pages. What this means: 461 * 462 * On Write: If we find a memory mapped page, we write to *both* 463 * the page and the dmu buffer. 464 */ 465static void 466update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid, 467 int segflg, dmu_tx_t *tx) 468{ 469 vm_object_t obj; 470 struct sf_buf *sf; 471 caddr_t va; 472 int off; 473 474 ASSERT(vp->v_mount != NULL); 475 obj = vp->v_object; 476 ASSERT(obj != NULL); 477 478 off = start & PAGEOFFSET; 479 zfs_vmobject_wlock(obj); 480 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 481 vm_page_t pp; 482 int nbytes = imin(PAGESIZE - off, len); 483 484 if (segflg == UIO_NOCOPY) { 485 pp = vm_page_lookup(obj, OFF_TO_IDX(start)); 486 KASSERT(pp != NULL, 487 ("zfs update_pages: NULL page in putpages case")); 488 KASSERT(off == 0, 489 ("zfs update_pages: unaligned data in putpages case")); 490 KASSERT(pp->valid == VM_PAGE_BITS_ALL, 491 ("zfs update_pages: invalid page in putpages case")); 492 KASSERT(vm_page_sbusied(pp), 493 ("zfs update_pages: unbusy page in putpages case")); 494 KASSERT(!pmap_page_is_write_mapped(pp), 495 ("zfs update_pages: writable page in putpages case")); 496 zfs_vmobject_wunlock(obj); 497 498 va = zfs_map_page(pp, &sf); 499 (void) dmu_write(os, oid, start, nbytes, va, tx); 500 zfs_unmap_page(sf); 501 502 zfs_vmobject_wlock(obj); 503 vm_page_undirty(pp); 504 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) { 505 zfs_vmobject_wunlock(obj); 506 507 va = zfs_map_page(pp, &sf); 508 (void) dmu_read(os, oid, start+off, nbytes, 509 va+off, DMU_READ_PREFETCH);; 510 zfs_unmap_page(sf); 511 512 zfs_vmobject_wlock(obj); 513 page_unbusy(pp); 514 } 515 len -= nbytes; 516 off = 0; 517 } 518 if (segflg != UIO_NOCOPY) 519 vm_object_pip_wakeupn(obj, 0); 520 zfs_vmobject_wunlock(obj); 521} 522 523/* 524 * Read with UIO_NOCOPY flag means that sendfile(2) requests 525 * ZFS to populate a range of page cache pages with data. 526 * 527 * NOTE: this function could be optimized to pre-allocate 528 * all pages in advance, drain exclusive busy on all of them, 529 * map them into contiguous KVA region and populate them 530 * in one single dmu_read() call. 531 */ 532static int 533mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio) 534{ 535 znode_t *zp = VTOZ(vp); 536 objset_t *os = zp->z_zfsvfs->z_os; 537 struct sf_buf *sf; 538 vm_object_t obj; 539 vm_page_t pp; 540 int64_t start; 541 caddr_t va; 542 int len = nbytes; 543 int off; 544 int error = 0; 545 546 ASSERT(uio->uio_segflg == UIO_NOCOPY); 547 ASSERT(vp->v_mount != NULL); 548 obj = vp->v_object; 549 ASSERT(obj != NULL); 550 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0); 551 552 zfs_vmobject_wlock(obj); 553 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) { 554 int bytes = MIN(PAGESIZE, len); 555 556 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY | 557 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); 558 if (pp->valid == 0) { 559 zfs_vmobject_wunlock(obj); 560 va = zfs_map_page(pp, &sf); 561 error = dmu_read(os, zp->z_id, start, bytes, va, 562 DMU_READ_PREFETCH); 563 if (bytes != PAGESIZE && error == 0) 564 bzero(va + bytes, PAGESIZE - bytes); 565 zfs_unmap_page(sf); 566 zfs_vmobject_wlock(obj); 567 vm_page_sunbusy(pp); 568 vm_page_lock(pp); 569 if (error) { 570 if (pp->wire_count == 0 && pp->valid == 0 && 571 !vm_page_busied(pp)) 572 vm_page_free(pp); 573 } else { 574 pp->valid = VM_PAGE_BITS_ALL; 575 vm_page_activate(pp); 576 } 577 vm_page_unlock(pp); 578 } else 579 vm_page_sunbusy(pp); 580 if (error) 581 break; 582 uio->uio_resid -= bytes; 583 uio->uio_offset += bytes; 584 len -= bytes; 585 } 586 zfs_vmobject_wunlock(obj); 587 return (error); 588} 589 590/* 591 * When a file is memory mapped, we must keep the IO data synchronized 592 * between the DMU cache and the memory mapped pages. What this means: 593 * 594 * On Read: We "read" preferentially from memory mapped pages, 595 * else we default from the dmu buffer. 596 * 597 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 598 * the file is memory mapped. 599 */ 600static int 601mappedread(vnode_t *vp, int nbytes, uio_t *uio) 602{ 603 znode_t *zp = VTOZ(vp); 604 objset_t *os = zp->z_zfsvfs->z_os; 605 vm_object_t obj; 606 int64_t start; 607 caddr_t va; 608 int len = nbytes; 609 int off; 610 int error = 0; 611 612 ASSERT(vp->v_mount != NULL); 613 obj = vp->v_object; 614 ASSERT(obj != NULL); 615 616 start = uio->uio_loffset; 617 off = start & PAGEOFFSET; 618 zfs_vmobject_wlock(obj); 619 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 620 vm_page_t pp; 621 uint64_t bytes = MIN(PAGESIZE - off, len); 622 623 if (pp = page_hold(vp, start)) { 624 struct sf_buf *sf; 625 caddr_t va; 626 627 zfs_vmobject_wunlock(obj); 628 va = zfs_map_page(pp, &sf); 629 error = uiomove(va + off, bytes, UIO_READ, uio); 630 zfs_unmap_page(sf); 631 zfs_vmobject_wlock(obj); 632 page_unhold(pp); 633 } else { 634 zfs_vmobject_wunlock(obj); 635 error = dmu_read_uio(os, zp->z_id, uio, bytes); 636 zfs_vmobject_wlock(obj); 637 } 638 len -= bytes; 639 off = 0; 640 if (error) 641 break; 642 } 643 zfs_vmobject_wunlock(obj); 644 return (error); 645} 646 647offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 648 649/* 650 * Read bytes from specified file into supplied buffer. 651 * 652 * IN: vp - vnode of file to be read from. 653 * uio - structure supplying read location, range info, 654 * and return buffer. 655 * ioflag - SYNC flags; used to provide FRSYNC semantics. 656 * cr - credentials of caller. 657 * ct - caller context 658 * 659 * OUT: uio - updated offset and range, buffer filled. 660 * 661 * RETURN: 0 on success, error code on failure. 662 * 663 * Side Effects: 664 * vp - atime updated if byte count > 0 665 */ 666/* ARGSUSED */ 667static int 668zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 669{ 670 znode_t *zp = VTOZ(vp); 671 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 672 objset_t *os; 673 ssize_t n, nbytes; 674 int error = 0; 675 rl_t *rl; 676 xuio_t *xuio = NULL; 677 678 ZFS_ENTER(zfsvfs); 679 ZFS_VERIFY_ZP(zp); 680 os = zfsvfs->z_os; 681 682 if (zp->z_pflags & ZFS_AV_QUARANTINED) { 683 ZFS_EXIT(zfsvfs); 684 return (SET_ERROR(EACCES)); 685 } 686 687 /* 688 * Validate file offset 689 */ 690 if (uio->uio_loffset < (offset_t)0) { 691 ZFS_EXIT(zfsvfs); 692 return (SET_ERROR(EINVAL)); 693 } 694 695 /* 696 * Fasttrack empty reads 697 */ 698 if (uio->uio_resid == 0) { 699 ZFS_EXIT(zfsvfs); 700 return (0); 701 } 702 703 /* 704 * Check for mandatory locks 705 */ 706 if (MANDMODE(zp->z_mode)) { 707 if (error = chklock(vp, FREAD, 708 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 709 ZFS_EXIT(zfsvfs); 710 return (error); 711 } 712 } 713 714 /* 715 * If we're in FRSYNC mode, sync out this znode before reading it. 716 */ 717 if (zfsvfs->z_log && 718 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)) 719 zil_commit(zfsvfs->z_log, zp->z_id); 720 721 /* 722 * Lock the range against changes. 723 */ 724 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 725 726 /* 727 * If we are reading past end-of-file we can skip 728 * to the end; but we might still need to set atime. 729 */ 730 if (uio->uio_loffset >= zp->z_size) { 731 error = 0; 732 goto out; 733 } 734 735 ASSERT(uio->uio_loffset < zp->z_size); 736 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset); 737 738#ifdef sun 739 if ((uio->uio_extflg == UIO_XUIO) && 740 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) { 741 int nblk; 742 int blksz = zp->z_blksz; 743 uint64_t offset = uio->uio_loffset; 744 745 xuio = (xuio_t *)uio; 746 if ((ISP2(blksz))) { 747 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset, 748 blksz)) / blksz; 749 } else { 750 ASSERT(offset + n <= blksz); 751 nblk = 1; 752 } 753 (void) dmu_xuio_init(xuio, nblk); 754 755 if (vn_has_cached_data(vp)) { 756 /* 757 * For simplicity, we always allocate a full buffer 758 * even if we only expect to read a portion of a block. 759 */ 760 while (--nblk >= 0) { 761 (void) dmu_xuio_add(xuio, 762 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 763 blksz), 0, blksz); 764 } 765 } 766 } 767#endif /* sun */ 768 769 while (n > 0) { 770 nbytes = MIN(n, zfs_read_chunk_size - 771 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 772 773#ifdef __FreeBSD__ 774 if (uio->uio_segflg == UIO_NOCOPY) 775 error = mappedread_sf(vp, nbytes, uio); 776 else 777#endif /* __FreeBSD__ */ 778 if (vn_has_cached_data(vp)) 779 error = mappedread(vp, nbytes, uio); 780 else 781 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 782 if (error) { 783 /* convert checksum errors into IO errors */ 784 if (error == ECKSUM) 785 error = SET_ERROR(EIO); 786 break; 787 } 788 789 n -= nbytes; 790 } 791out: 792 zfs_range_unlock(rl); 793 794 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 795 ZFS_EXIT(zfsvfs); 796 return (error); 797} 798 799/* 800 * Write the bytes to a file. 801 * 802 * IN: vp - vnode of file to be written to. 803 * uio - structure supplying write location, range info, 804 * and data buffer. 805 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is 806 * set if in append mode. 807 * cr - credentials of caller. 808 * ct - caller context (NFS/CIFS fem monitor only) 809 * 810 * OUT: uio - updated offset and range. 811 * 812 * RETURN: 0 on success, error code on failure. 813 * 814 * Timestamps: 815 * vp - ctime|mtime updated if byte count > 0 816 */ 817 818/* ARGSUSED */ 819static int 820zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 821{ 822 znode_t *zp = VTOZ(vp); 823 rlim64_t limit = MAXOFFSET_T; 824 ssize_t start_resid = uio->uio_resid; 825 ssize_t tx_bytes; 826 uint64_t end_size; 827 dmu_tx_t *tx; 828 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 829 zilog_t *zilog; 830 offset_t woff; 831 ssize_t n, nbytes; 832 rl_t *rl; 833 int max_blksz = zfsvfs->z_max_blksz; 834 int error = 0; 835 arc_buf_t *abuf; 836 iovec_t *aiov = NULL; 837 xuio_t *xuio = NULL; 838 int i_iov = 0; 839 int iovcnt = uio->uio_iovcnt; 840 iovec_t *iovp = uio->uio_iov; 841 int write_eof; 842 int count = 0; 843 sa_bulk_attr_t bulk[4]; 844 uint64_t mtime[2], ctime[2]; 845 846 /* 847 * Fasttrack empty write 848 */ 849 n = start_resid; 850 if (n == 0) 851 return (0); 852 853 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 854 limit = MAXOFFSET_T; 855 856 ZFS_ENTER(zfsvfs); 857 ZFS_VERIFY_ZP(zp); 858 859 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 860 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 861 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 862 &zp->z_size, 8); 863 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 864 &zp->z_pflags, 8); 865 866 /* 867 * If immutable or not appending then return EPERM 868 */ 869 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 870 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 871 (uio->uio_loffset < zp->z_size))) { 872 ZFS_EXIT(zfsvfs); 873 return (SET_ERROR(EPERM)); 874 } 875 876 zilog = zfsvfs->z_log; 877 878 /* 879 * Validate file offset 880 */ 881 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset; 882 if (woff < 0) { 883 ZFS_EXIT(zfsvfs); 884 return (SET_ERROR(EINVAL)); 885 } 886 887 /* 888 * Check for mandatory locks before calling zfs_range_lock() 889 * in order to prevent a deadlock with locks set via fcntl(). 890 */ 891 if (MANDMODE((mode_t)zp->z_mode) && 892 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 893 ZFS_EXIT(zfsvfs); 894 return (error); 895 } 896 897#ifdef sun 898 /* 899 * Pre-fault the pages to ensure slow (eg NFS) pages 900 * don't hold up txg. 901 * Skip this if uio contains loaned arc_buf. 902 */ 903 if ((uio->uio_extflg == UIO_XUIO) && 904 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) 905 xuio = (xuio_t *)uio; 906 else 907 uio_prefaultpages(MIN(n, max_blksz), uio); 908#endif /* sun */ 909 910 /* 911 * If in append mode, set the io offset pointer to eof. 912 */ 913 if (ioflag & FAPPEND) { 914 /* 915 * Obtain an appending range lock to guarantee file append 916 * semantics. We reset the write offset once we have the lock. 917 */ 918 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 919 woff = rl->r_off; 920 if (rl->r_len == UINT64_MAX) { 921 /* 922 * We overlocked the file because this write will cause 923 * the file block size to increase. 924 * Note that zp_size cannot change with this lock held. 925 */ 926 woff = zp->z_size; 927 } 928 uio->uio_loffset = woff; 929 } else { 930 /* 931 * Note that if the file block size will change as a result of 932 * this write, then this range lock will lock the entire file 933 * so that we can re-write the block safely. 934 */ 935 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 936 } 937 938 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) { 939 zfs_range_unlock(rl); 940 ZFS_EXIT(zfsvfs); 941 return (EFBIG); 942 } 943 944 if (woff >= limit) { 945 zfs_range_unlock(rl); 946 ZFS_EXIT(zfsvfs); 947 return (SET_ERROR(EFBIG)); 948 } 949 950 if ((woff + n) > limit || woff > (limit - n)) 951 n = limit - woff; 952 953 /* Will this write extend the file length? */ 954 write_eof = (woff + n > zp->z_size); 955 956 end_size = MAX(zp->z_size, woff + n); 957 958 /* 959 * Write the file in reasonable size chunks. Each chunk is written 960 * in a separate transaction; this keeps the intent log records small 961 * and allows us to do more fine-grained space accounting. 962 */ 963 while (n > 0) { 964 abuf = NULL; 965 woff = uio->uio_loffset; 966again: 967 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 968 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 969 if (abuf != NULL) 970 dmu_return_arcbuf(abuf); 971 error = SET_ERROR(EDQUOT); 972 break; 973 } 974 975 if (xuio && abuf == NULL) { 976 ASSERT(i_iov < iovcnt); 977 aiov = &iovp[i_iov]; 978 abuf = dmu_xuio_arcbuf(xuio, i_iov); 979 dmu_xuio_clear(xuio, i_iov); 980 DTRACE_PROBE3(zfs_cp_write, int, i_iov, 981 iovec_t *, aiov, arc_buf_t *, abuf); 982 ASSERT((aiov->iov_base == abuf->b_data) || 983 ((char *)aiov->iov_base - (char *)abuf->b_data + 984 aiov->iov_len == arc_buf_size(abuf))); 985 i_iov++; 986 } else if (abuf == NULL && n >= max_blksz && 987 woff >= zp->z_size && 988 P2PHASE(woff, max_blksz) == 0 && 989 zp->z_blksz == max_blksz) { 990 /* 991 * This write covers a full block. "Borrow" a buffer 992 * from the dmu so that we can fill it before we enter 993 * a transaction. This avoids the possibility of 994 * holding up the transaction if the data copy hangs 995 * up on a pagefault (e.g., from an NFS server mapping). 996 */ 997 size_t cbytes; 998 999 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 1000 max_blksz); 1001 ASSERT(abuf != NULL); 1002 ASSERT(arc_buf_size(abuf) == max_blksz); 1003 if (error = uiocopy(abuf->b_data, max_blksz, 1004 UIO_WRITE, uio, &cbytes)) { 1005 dmu_return_arcbuf(abuf); 1006 break; 1007 } 1008 ASSERT(cbytes == max_blksz); 1009 } 1010 1011 /* 1012 * Start a transaction. 1013 */ 1014 tx = dmu_tx_create(zfsvfs->z_os); 1015 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1016 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 1017 zfs_sa_upgrade_txholds(tx, zp); 1018 error = dmu_tx_assign(tx, TXG_NOWAIT); 1019 if (error) { 1020 if (error == ERESTART) { 1021 dmu_tx_wait(tx); 1022 dmu_tx_abort(tx); 1023 goto again; 1024 } 1025 dmu_tx_abort(tx); 1026 if (abuf != NULL) 1027 dmu_return_arcbuf(abuf); 1028 break; 1029 } 1030 1031 /* 1032 * If zfs_range_lock() over-locked we grow the blocksize 1033 * and then reduce the lock range. This will only happen 1034 * on the first iteration since zfs_range_reduce() will 1035 * shrink down r_len to the appropriate size. 1036 */ 1037 if (rl->r_len == UINT64_MAX) { 1038 uint64_t new_blksz; 1039 1040 if (zp->z_blksz > max_blksz) { 1041 ASSERT(!ISP2(zp->z_blksz)); 1042 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 1043 } else { 1044 new_blksz = MIN(end_size, max_blksz); 1045 } 1046 zfs_grow_blocksize(zp, new_blksz, tx); 1047 zfs_range_reduce(rl, woff, n); 1048 } 1049 1050 /* 1051 * XXX - should we really limit each write to z_max_blksz? 1052 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 1053 */ 1054 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 1055 1056 if (woff + nbytes > zp->z_size) 1057 vnode_pager_setsize(vp, woff + nbytes); 1058 1059 if (abuf == NULL) { 1060 tx_bytes = uio->uio_resid; 1061 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), 1062 uio, nbytes, tx); 1063 tx_bytes -= uio->uio_resid; 1064 } else { 1065 tx_bytes = nbytes; 1066 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len); 1067 /* 1068 * If this is not a full block write, but we are 1069 * extending the file past EOF and this data starts 1070 * block-aligned, use assign_arcbuf(). Otherwise, 1071 * write via dmu_write(). 1072 */ 1073 if (tx_bytes < max_blksz && (!write_eof || 1074 aiov->iov_base != abuf->b_data)) { 1075 ASSERT(xuio); 1076 dmu_write(zfsvfs->z_os, zp->z_id, woff, 1077 aiov->iov_len, aiov->iov_base, tx); 1078 dmu_return_arcbuf(abuf); 1079 xuio_stat_wbuf_copied(); 1080 } else { 1081 ASSERT(xuio || tx_bytes == max_blksz); 1082 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), 1083 woff, abuf, tx); 1084 } 1085 ASSERT(tx_bytes <= uio->uio_resid); 1086 uioskip(uio, tx_bytes); 1087 } 1088 if (tx_bytes && vn_has_cached_data(vp)) { 1089 update_pages(vp, woff, tx_bytes, zfsvfs->z_os, 1090 zp->z_id, uio->uio_segflg, tx); 1091 } 1092 1093 /* 1094 * If we made no progress, we're done. If we made even 1095 * partial progress, update the znode and ZIL accordingly. 1096 */ 1097 if (tx_bytes == 0) { 1098 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 1099 (void *)&zp->z_size, sizeof (uint64_t), tx); 1100 dmu_tx_commit(tx); 1101 ASSERT(error != 0); 1102 break; 1103 } 1104 1105 /* 1106 * Clear Set-UID/Set-GID bits on successful write if not 1107 * privileged and at least one of the excute bits is set. 1108 * 1109 * It would be nice to to this after all writes have 1110 * been done, but that would still expose the ISUID/ISGID 1111 * to another app after the partial write is committed. 1112 * 1113 * Note: we don't call zfs_fuid_map_id() here because 1114 * user 0 is not an ephemeral uid. 1115 */ 1116 mutex_enter(&zp->z_acl_lock); 1117 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | 1118 (S_IXUSR >> 6))) != 0 && 1119 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && 1120 secpolicy_vnode_setid_retain(vp, cr, 1121 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) { 1122 uint64_t newmode; 1123 zp->z_mode &= ~(S_ISUID | S_ISGID); 1124 newmode = zp->z_mode; 1125 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), 1126 (void *)&newmode, sizeof (uint64_t), tx); 1127 } 1128 mutex_exit(&zp->z_acl_lock); 1129 1130 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 1131 B_TRUE); 1132 1133 /* 1134 * Update the file size (zp_size) if it has changed; 1135 * account for possible concurrent updates. 1136 */ 1137 while ((end_size = zp->z_size) < uio->uio_loffset) { 1138 (void) atomic_cas_64(&zp->z_size, end_size, 1139 uio->uio_loffset); 1140 ASSERT(error == 0); 1141 } 1142 /* 1143 * If we are replaying and eof is non zero then force 1144 * the file size to the specified eof. Note, there's no 1145 * concurrency during replay. 1146 */ 1147 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) 1148 zp->z_size = zfsvfs->z_replay_eof; 1149 1150 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1151 1152 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 1153 dmu_tx_commit(tx); 1154 1155 if (error != 0) 1156 break; 1157 ASSERT(tx_bytes == nbytes); 1158 n -= nbytes; 1159 1160#ifdef sun 1161 if (!xuio && n > 0) 1162 uio_prefaultpages(MIN(n, max_blksz), uio); 1163#endif /* sun */ 1164 } 1165 1166 zfs_range_unlock(rl); 1167 1168 /* 1169 * If we're in replay mode, or we made no progress, return error. 1170 * Otherwise, it's at least a partial write, so it's successful. 1171 */ 1172 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 1173 ZFS_EXIT(zfsvfs); 1174 return (error); 1175 } 1176 1177 if (ioflag & (FSYNC | FDSYNC) || 1178 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1179 zil_commit(zilog, zp->z_id); 1180 1181 ZFS_EXIT(zfsvfs); 1182 return (0); 1183} 1184 1185void 1186zfs_get_done(zgd_t *zgd, int error) 1187{ 1188 znode_t *zp = zgd->zgd_private; 1189 objset_t *os = zp->z_zfsvfs->z_os; 1190 1191 if (zgd->zgd_db) 1192 dmu_buf_rele(zgd->zgd_db, zgd); 1193 1194 zfs_range_unlock(zgd->zgd_rl); 1195 1196 /* 1197 * Release the vnode asynchronously as we currently have the 1198 * txg stopped from syncing. 1199 */ 1200 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1201 1202 if (error == 0 && zgd->zgd_bp) 1203 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1204 1205 kmem_free(zgd, sizeof (zgd_t)); 1206} 1207 1208#ifdef DEBUG 1209static int zil_fault_io = 0; 1210#endif 1211 1212/* 1213 * Get data to generate a TX_WRITE intent log record. 1214 */ 1215int 1216zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1217{ 1218 zfsvfs_t *zfsvfs = arg; 1219 objset_t *os = zfsvfs->z_os; 1220 znode_t *zp; 1221 uint64_t object = lr->lr_foid; 1222 uint64_t offset = lr->lr_offset; 1223 uint64_t size = lr->lr_length; 1224 blkptr_t *bp = &lr->lr_blkptr; 1225 dmu_buf_t *db; 1226 zgd_t *zgd; 1227 int error = 0; 1228 1229 ASSERT(zio != NULL); 1230 ASSERT(size != 0); 1231 1232 /* 1233 * Nothing to do if the file has been removed 1234 */ 1235 if (zfs_zget(zfsvfs, object, &zp) != 0) 1236 return (SET_ERROR(ENOENT)); 1237 if (zp->z_unlinked) { 1238 /* 1239 * Release the vnode asynchronously as we currently have the 1240 * txg stopped from syncing. 1241 */ 1242 VN_RELE_ASYNC(ZTOV(zp), 1243 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1244 return (SET_ERROR(ENOENT)); 1245 } 1246 1247 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1248 zgd->zgd_zilog = zfsvfs->z_log; 1249 zgd->zgd_private = zp; 1250 1251 /* 1252 * Write records come in two flavors: immediate and indirect. 1253 * For small writes it's cheaper to store the data with the 1254 * log record (immediate); for large writes it's cheaper to 1255 * sync the data and get a pointer to it (indirect) so that 1256 * we don't have to write the data twice. 1257 */ 1258 if (buf != NULL) { /* immediate write */ 1259 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER); 1260 /* test for truncation needs to be done while range locked */ 1261 if (offset >= zp->z_size) { 1262 error = SET_ERROR(ENOENT); 1263 } else { 1264 error = dmu_read(os, object, offset, size, buf, 1265 DMU_READ_NO_PREFETCH); 1266 } 1267 ASSERT(error == 0 || error == ENOENT); 1268 } else { /* indirect write */ 1269 /* 1270 * Have to lock the whole block to ensure when it's 1271 * written out and it's checksum is being calculated 1272 * that no one can change the data. We need to re-check 1273 * blocksize after we get the lock in case it's changed! 1274 */ 1275 for (;;) { 1276 uint64_t blkoff; 1277 size = zp->z_blksz; 1278 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; 1279 offset -= blkoff; 1280 zgd->zgd_rl = zfs_range_lock(zp, offset, size, 1281 RL_READER); 1282 if (zp->z_blksz == size) 1283 break; 1284 offset += blkoff; 1285 zfs_range_unlock(zgd->zgd_rl); 1286 } 1287 /* test for truncation needs to be done while range locked */ 1288 if (lr->lr_offset >= zp->z_size) 1289 error = SET_ERROR(ENOENT); 1290#ifdef DEBUG 1291 if (zil_fault_io) { 1292 error = SET_ERROR(EIO); 1293 zil_fault_io = 0; 1294 } 1295#endif 1296 if (error == 0) 1297 error = dmu_buf_hold(os, object, offset, zgd, &db, 1298 DMU_READ_NO_PREFETCH); 1299 1300 if (error == 0) { 1301 blkptr_t *obp = dmu_buf_get_blkptr(db); 1302 if (obp) { 1303 ASSERT(BP_IS_HOLE(bp)); 1304 *bp = *obp; 1305 } 1306 1307 zgd->zgd_db = db; 1308 zgd->zgd_bp = bp; 1309 1310 ASSERT(db->db_offset == offset); 1311 ASSERT(db->db_size == size); 1312 1313 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1314 zfs_get_done, zgd); 1315 ASSERT(error || lr->lr_length <= zp->z_blksz); 1316 1317 /* 1318 * On success, we need to wait for the write I/O 1319 * initiated by dmu_sync() to complete before we can 1320 * release this dbuf. We will finish everything up 1321 * in the zfs_get_done() callback. 1322 */ 1323 if (error == 0) 1324 return (0); 1325 1326 if (error == EALREADY) { 1327 lr->lr_common.lrc_txtype = TX_WRITE2; 1328 error = 0; 1329 } 1330 } 1331 } 1332 1333 zfs_get_done(zgd, error); 1334 1335 return (error); 1336} 1337 1338/*ARGSUSED*/ 1339static int 1340zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 1341 caller_context_t *ct) 1342{ 1343 znode_t *zp = VTOZ(vp); 1344 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1345 int error; 1346 1347 ZFS_ENTER(zfsvfs); 1348 ZFS_VERIFY_ZP(zp); 1349 1350 if (flag & V_ACE_MASK) 1351 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 1352 else 1353 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1354 1355 ZFS_EXIT(zfsvfs); 1356 return (error); 1357} 1358 1359/* 1360 * If vnode is for a device return a specfs vnode instead. 1361 */ 1362static int 1363specvp_check(vnode_t **vpp, cred_t *cr) 1364{ 1365 int error = 0; 1366 1367 if (IS_DEVVP(*vpp)) { 1368 struct vnode *svp; 1369 1370 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1371 VN_RELE(*vpp); 1372 if (svp == NULL) 1373 error = SET_ERROR(ENOSYS); 1374 *vpp = svp; 1375 } 1376 return (error); 1377} 1378 1379 1380/* 1381 * Lookup an entry in a directory, or an extended attribute directory. 1382 * If it exists, return a held vnode reference for it. 1383 * 1384 * IN: dvp - vnode of directory to search. 1385 * nm - name of entry to lookup. 1386 * pnp - full pathname to lookup [UNUSED]. 1387 * flags - LOOKUP_XATTR set if looking for an attribute. 1388 * rdir - root directory vnode [UNUSED]. 1389 * cr - credentials of caller. 1390 * ct - caller context 1391 * direntflags - directory lookup flags 1392 * realpnp - returned pathname. 1393 * 1394 * OUT: vpp - vnode of located entry, NULL if not found. 1395 * 1396 * RETURN: 0 on success, error code on failure. 1397 * 1398 * Timestamps: 1399 * NA 1400 */ 1401/* ARGSUSED */ 1402static int 1403zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 1404 int nameiop, cred_t *cr, kthread_t *td, int flags) 1405{ 1406 znode_t *zdp = VTOZ(dvp); 1407 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1408 int error = 0; 1409 int *direntflags = NULL; 1410 void *realpnp = NULL; 1411 1412 /* fast path */ 1413 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 1414 1415 if (dvp->v_type != VDIR) { 1416 return (SET_ERROR(ENOTDIR)); 1417 } else if (zdp->z_sa_hdl == NULL) { 1418 return (SET_ERROR(EIO)); 1419 } 1420 1421 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 1422 error = zfs_fastaccesschk_execute(zdp, cr); 1423 if (!error) { 1424 *vpp = dvp; 1425 VN_HOLD(*vpp); 1426 return (0); 1427 } 1428 return (error); 1429 } else { 1430 vnode_t *tvp = dnlc_lookup(dvp, nm); 1431 1432 if (tvp) { 1433 error = zfs_fastaccesschk_execute(zdp, cr); 1434 if (error) { 1435 VN_RELE(tvp); 1436 return (error); 1437 } 1438 if (tvp == DNLC_NO_VNODE) { 1439 VN_RELE(tvp); 1440 return (SET_ERROR(ENOENT)); 1441 } else { 1442 *vpp = tvp; 1443 return (specvp_check(vpp, cr)); 1444 } 1445 } 1446 } 1447 } 1448 1449 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1450 1451 ZFS_ENTER(zfsvfs); 1452 ZFS_VERIFY_ZP(zdp); 1453 1454 *vpp = NULL; 1455 1456 if (flags & LOOKUP_XATTR) { 1457#ifdef TODO 1458 /* 1459 * If the xattr property is off, refuse the lookup request. 1460 */ 1461 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1462 ZFS_EXIT(zfsvfs); 1463 return (SET_ERROR(EINVAL)); 1464 } 1465#endif 1466 1467 /* 1468 * We don't allow recursive attributes.. 1469 * Maybe someday we will. 1470 */ 1471 if (zdp->z_pflags & ZFS_XATTR) { 1472 ZFS_EXIT(zfsvfs); 1473 return (SET_ERROR(EINVAL)); 1474 } 1475 1476 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1477 ZFS_EXIT(zfsvfs); 1478 return (error); 1479 } 1480 1481 /* 1482 * Do we have permission to get into attribute directory? 1483 */ 1484 1485 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1486 B_FALSE, cr)) { 1487 VN_RELE(*vpp); 1488 *vpp = NULL; 1489 } 1490 1491 ZFS_EXIT(zfsvfs); 1492 return (error); 1493 } 1494 1495 if (dvp->v_type != VDIR) { 1496 ZFS_EXIT(zfsvfs); 1497 return (SET_ERROR(ENOTDIR)); 1498 } 1499 1500 /* 1501 * Check accessibility of directory. 1502 */ 1503 1504 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1505 ZFS_EXIT(zfsvfs); 1506 return (error); 1507 } 1508 1509 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1510 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1511 ZFS_EXIT(zfsvfs); 1512 return (SET_ERROR(EILSEQ)); 1513 } 1514 1515 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1516 if (error == 0) 1517 error = specvp_check(vpp, cr); 1518 1519 /* Translate errors and add SAVENAME when needed. */ 1520 if (cnp->cn_flags & ISLASTCN) { 1521 switch (nameiop) { 1522 case CREATE: 1523 case RENAME: 1524 if (error == ENOENT) { 1525 error = EJUSTRETURN; 1526 cnp->cn_flags |= SAVENAME; 1527 break; 1528 } 1529 /* FALLTHROUGH */ 1530 case DELETE: 1531 if (error == 0) 1532 cnp->cn_flags |= SAVENAME; 1533 break; 1534 } 1535 } 1536 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) { 1537 int ltype = 0; 1538 1539 if (cnp->cn_flags & ISDOTDOT) { 1540 ltype = VOP_ISLOCKED(dvp); 1541 VOP_UNLOCK(dvp, 0); 1542 } 1543 ZFS_EXIT(zfsvfs); 1544 error = vn_lock(*vpp, cnp->cn_lkflags); 1545 if (cnp->cn_flags & ISDOTDOT) 1546 vn_lock(dvp, ltype | LK_RETRY); 1547 if (error != 0) { 1548 VN_RELE(*vpp); 1549 *vpp = NULL; 1550 return (error); 1551 } 1552 } else { 1553 ZFS_EXIT(zfsvfs); 1554 } 1555 1556#ifdef FREEBSD_NAMECACHE 1557 /* 1558 * Insert name into cache (as non-existent) if appropriate. 1559 */ 1560 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE) 1561 cache_enter(dvp, *vpp, cnp); 1562 /* 1563 * Insert name into cache if appropriate. 1564 */ 1565 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1566 if (!(cnp->cn_flags & ISLASTCN) || 1567 (nameiop != DELETE && nameiop != RENAME)) { 1568 cache_enter(dvp, *vpp, cnp); 1569 } 1570 } 1571#endif 1572 1573 return (error); 1574} 1575 1576/* 1577 * Attempt to create a new entry in a directory. If the entry 1578 * already exists, truncate the file if permissible, else return 1579 * an error. Return the vp of the created or trunc'd file. 1580 * 1581 * IN: dvp - vnode of directory to put new file entry in. 1582 * name - name of new file entry. 1583 * vap - attributes of new file. 1584 * excl - flag indicating exclusive or non-exclusive mode. 1585 * mode - mode to open file with. 1586 * cr - credentials of caller. 1587 * flag - large file flag [UNUSED]. 1588 * ct - caller context 1589 * vsecp - ACL to be set 1590 * 1591 * OUT: vpp - vnode of created or trunc'd entry. 1592 * 1593 * RETURN: 0 on success, error code on failure. 1594 * 1595 * Timestamps: 1596 * dvp - ctime|mtime updated if new entry created 1597 * vp - ctime|mtime always, atime if new 1598 */ 1599 1600/* ARGSUSED */ 1601static int 1602zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1603 vnode_t **vpp, cred_t *cr, kthread_t *td) 1604{ 1605 znode_t *zp, *dzp = VTOZ(dvp); 1606 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1607 zilog_t *zilog; 1608 objset_t *os; 1609 zfs_dirlock_t *dl; 1610 dmu_tx_t *tx; 1611 int error; 1612 ksid_t *ksid; 1613 uid_t uid; 1614 gid_t gid = crgetgid(cr); 1615 zfs_acl_ids_t acl_ids; 1616 boolean_t fuid_dirtied; 1617 boolean_t have_acl = B_FALSE; 1618 void *vsecp = NULL; 1619 int flag = 0; 1620 1621 /* 1622 * If we have an ephemeral id, ACL, or XVATTR then 1623 * make sure file system is at proper version 1624 */ 1625 1626 ksid = crgetsid(cr, KSID_OWNER); 1627 if (ksid) 1628 uid = ksid_getid(ksid); 1629 else 1630 uid = crgetuid(cr); 1631 1632 if (zfsvfs->z_use_fuids == B_FALSE && 1633 (vsecp || (vap->va_mask & AT_XVATTR) || 1634 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1635 return (SET_ERROR(EINVAL)); 1636 1637 ZFS_ENTER(zfsvfs); 1638 ZFS_VERIFY_ZP(dzp); 1639 os = zfsvfs->z_os; 1640 zilog = zfsvfs->z_log; 1641 1642 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1643 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1644 ZFS_EXIT(zfsvfs); 1645 return (SET_ERROR(EILSEQ)); 1646 } 1647 1648 if (vap->va_mask & AT_XVATTR) { 1649 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 1650 crgetuid(cr), cr, vap->va_type)) != 0) { 1651 ZFS_EXIT(zfsvfs); 1652 return (error); 1653 } 1654 } 1655top: 1656 *vpp = NULL; 1657 1658 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) 1659 vap->va_mode &= ~S_ISVTX; 1660 1661 if (*name == '\0') { 1662 /* 1663 * Null component name refers to the directory itself. 1664 */ 1665 VN_HOLD(dvp); 1666 zp = dzp; 1667 dl = NULL; 1668 error = 0; 1669 } else { 1670 /* possible VN_HOLD(zp) */ 1671 int zflg = 0; 1672 1673 if (flag & FIGNORECASE) 1674 zflg |= ZCILOOK; 1675 1676 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1677 NULL, NULL); 1678 if (error) { 1679 if (have_acl) 1680 zfs_acl_ids_free(&acl_ids); 1681 if (strcmp(name, "..") == 0) 1682 error = SET_ERROR(EISDIR); 1683 ZFS_EXIT(zfsvfs); 1684 return (error); 1685 } 1686 } 1687 1688 if (zp == NULL) { 1689 uint64_t txtype; 1690 1691 /* 1692 * Create a new file object and update the directory 1693 * to reference it. 1694 */ 1695 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1696 if (have_acl) 1697 zfs_acl_ids_free(&acl_ids); 1698 goto out; 1699 } 1700 1701 /* 1702 * We only support the creation of regular files in 1703 * extended attribute directories. 1704 */ 1705 1706 if ((dzp->z_pflags & ZFS_XATTR) && 1707 (vap->va_type != VREG)) { 1708 if (have_acl) 1709 zfs_acl_ids_free(&acl_ids); 1710 error = SET_ERROR(EINVAL); 1711 goto out; 1712 } 1713 1714 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 1715 cr, vsecp, &acl_ids)) != 0) 1716 goto out; 1717 have_acl = B_TRUE; 1718 1719 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1720 zfs_acl_ids_free(&acl_ids); 1721 error = SET_ERROR(EDQUOT); 1722 goto out; 1723 } 1724 1725 tx = dmu_tx_create(os); 1726 1727 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1728 ZFS_SA_BASE_ATTR_SIZE); 1729 1730 fuid_dirtied = zfsvfs->z_fuid_dirty; 1731 if (fuid_dirtied) 1732 zfs_fuid_txhold(zfsvfs, tx); 1733 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1734 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1735 if (!zfsvfs->z_use_sa && 1736 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1737 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1738 0, acl_ids.z_aclp->z_acl_bytes); 1739 } 1740 error = dmu_tx_assign(tx, TXG_NOWAIT); 1741 if (error) { 1742 zfs_dirent_unlock(dl); 1743 if (error == ERESTART) { 1744 dmu_tx_wait(tx); 1745 dmu_tx_abort(tx); 1746 goto top; 1747 } 1748 zfs_acl_ids_free(&acl_ids); 1749 dmu_tx_abort(tx); 1750 ZFS_EXIT(zfsvfs); 1751 return (error); 1752 } 1753 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1754 1755 if (fuid_dirtied) 1756 zfs_fuid_sync(zfsvfs, tx); 1757 1758 (void) zfs_link_create(dl, zp, tx, ZNEW); 1759 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1760 if (flag & FIGNORECASE) 1761 txtype |= TX_CI; 1762 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1763 vsecp, acl_ids.z_fuidp, vap); 1764 zfs_acl_ids_free(&acl_ids); 1765 dmu_tx_commit(tx); 1766 } else { 1767 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1768 1769 if (have_acl) 1770 zfs_acl_ids_free(&acl_ids); 1771 have_acl = B_FALSE; 1772 1773 /* 1774 * A directory entry already exists for this name. 1775 */ 1776 /* 1777 * Can't truncate an existing file if in exclusive mode. 1778 */ 1779 if (excl == EXCL) { 1780 error = SET_ERROR(EEXIST); 1781 goto out; 1782 } 1783 /* 1784 * Can't open a directory for writing. 1785 */ 1786 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1787 error = SET_ERROR(EISDIR); 1788 goto out; 1789 } 1790 /* 1791 * Verify requested access to file. 1792 */ 1793 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1794 goto out; 1795 } 1796 1797 mutex_enter(&dzp->z_lock); 1798 dzp->z_seq++; 1799 mutex_exit(&dzp->z_lock); 1800 1801 /* 1802 * Truncate regular files if requested. 1803 */ 1804 if ((ZTOV(zp)->v_type == VREG) && 1805 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1806 /* we can't hold any locks when calling zfs_freesp() */ 1807 zfs_dirent_unlock(dl); 1808 dl = NULL; 1809 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1810 if (error == 0) { 1811 vnevent_create(ZTOV(zp), ct); 1812 } 1813 } 1814 } 1815out: 1816 if (dl) 1817 zfs_dirent_unlock(dl); 1818 1819 if (error) { 1820 if (zp) 1821 VN_RELE(ZTOV(zp)); 1822 } else { 1823 *vpp = ZTOV(zp); 1824 error = specvp_check(vpp, cr); 1825 } 1826 1827 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1828 zil_commit(zilog, 0); 1829 1830 ZFS_EXIT(zfsvfs); 1831 return (error); 1832} 1833 1834/* 1835 * Remove an entry from a directory. 1836 * 1837 * IN: dvp - vnode of directory to remove entry from. 1838 * name - name of entry to remove. 1839 * cr - credentials of caller. 1840 * ct - caller context 1841 * flags - case flags 1842 * 1843 * RETURN: 0 on success, error code on failure. 1844 * 1845 * Timestamps: 1846 * dvp - ctime|mtime 1847 * vp - ctime (if nlink > 0) 1848 */ 1849 1850uint64_t null_xattr = 0; 1851 1852/*ARGSUSED*/ 1853static int 1854zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1855 int flags) 1856{ 1857 znode_t *zp, *dzp = VTOZ(dvp); 1858 znode_t *xzp; 1859 vnode_t *vp; 1860 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1861 zilog_t *zilog; 1862 uint64_t acl_obj, xattr_obj; 1863 uint64_t xattr_obj_unlinked = 0; 1864 uint64_t obj = 0; 1865 zfs_dirlock_t *dl; 1866 dmu_tx_t *tx; 1867 boolean_t may_delete_now, delete_now = FALSE; 1868 boolean_t unlinked, toobig = FALSE; 1869 uint64_t txtype; 1870 pathname_t *realnmp = NULL; 1871 pathname_t realnm; 1872 int error; 1873 int zflg = ZEXISTS; 1874 1875 ZFS_ENTER(zfsvfs); 1876 ZFS_VERIFY_ZP(dzp); 1877 zilog = zfsvfs->z_log; 1878 1879 if (flags & FIGNORECASE) { 1880 zflg |= ZCILOOK; 1881 pn_alloc(&realnm); 1882 realnmp = &realnm; 1883 } 1884 1885top: 1886 xattr_obj = 0; 1887 xzp = NULL; 1888 /* 1889 * Attempt to lock directory; fail if entry doesn't exist. 1890 */ 1891 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1892 NULL, realnmp)) { 1893 if (realnmp) 1894 pn_free(realnmp); 1895 ZFS_EXIT(zfsvfs); 1896 return (error); 1897 } 1898 1899 vp = ZTOV(zp); 1900 1901 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1902 goto out; 1903 } 1904 1905 /* 1906 * Need to use rmdir for removing directories. 1907 */ 1908 if (vp->v_type == VDIR) { 1909 error = SET_ERROR(EPERM); 1910 goto out; 1911 } 1912 1913 vnevent_remove(vp, dvp, name, ct); 1914 1915 if (realnmp) 1916 dnlc_remove(dvp, realnmp->pn_buf); 1917 else 1918 dnlc_remove(dvp, name); 1919 1920 VI_LOCK(vp); 1921 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1922 VI_UNLOCK(vp); 1923 1924 /* 1925 * We may delete the znode now, or we may put it in the unlinked set; 1926 * it depends on whether we're the last link, and on whether there are 1927 * other holds on the vnode. So we dmu_tx_hold() the right things to 1928 * allow for either case. 1929 */ 1930 obj = zp->z_id; 1931 tx = dmu_tx_create(zfsvfs->z_os); 1932 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1933 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1934 zfs_sa_upgrade_txholds(tx, zp); 1935 zfs_sa_upgrade_txholds(tx, dzp); 1936 if (may_delete_now) { 1937 toobig = 1938 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 1939 /* if the file is too big, only hold_free a token amount */ 1940 dmu_tx_hold_free(tx, zp->z_id, 0, 1941 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1942 } 1943 1944 /* are there any extended attributes? */ 1945 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1946 &xattr_obj, sizeof (xattr_obj)); 1947 if (error == 0 && xattr_obj) { 1948 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1949 ASSERT0(error); 1950 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1951 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1952 } 1953 1954 mutex_enter(&zp->z_lock); 1955 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now) 1956 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1957 mutex_exit(&zp->z_lock); 1958 1959 /* charge as an update -- would be nice not to charge at all */ 1960 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1961 1962 error = dmu_tx_assign(tx, TXG_NOWAIT); 1963 if (error) { 1964 zfs_dirent_unlock(dl); 1965 VN_RELE(vp); 1966 if (xzp) 1967 VN_RELE(ZTOV(xzp)); 1968 if (error == ERESTART) { 1969 dmu_tx_wait(tx); 1970 dmu_tx_abort(tx); 1971 goto top; 1972 } 1973 if (realnmp) 1974 pn_free(realnmp); 1975 dmu_tx_abort(tx); 1976 ZFS_EXIT(zfsvfs); 1977 return (error); 1978 } 1979 1980 /* 1981 * Remove the directory entry. 1982 */ 1983 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1984 1985 if (error) { 1986 dmu_tx_commit(tx); 1987 goto out; 1988 } 1989 1990 if (unlinked) { 1991 1992 /* 1993 * Hold z_lock so that we can make sure that the ACL obj 1994 * hasn't changed. Could have been deleted due to 1995 * zfs_sa_upgrade(). 1996 */ 1997 mutex_enter(&zp->z_lock); 1998 VI_LOCK(vp); 1999 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 2000 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked)); 2001 delete_now = may_delete_now && !toobig && 2002 vp->v_count == 1 && !vn_has_cached_data(vp) && 2003 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) == 2004 acl_obj; 2005 VI_UNLOCK(vp); 2006 } 2007 2008 if (delete_now) { 2009#ifdef __FreeBSD__ 2010 panic("zfs_remove: delete_now branch taken"); 2011#endif 2012 if (xattr_obj_unlinked) { 2013 ASSERT3U(xzp->z_links, ==, 2); 2014 mutex_enter(&xzp->z_lock); 2015 xzp->z_unlinked = 1; 2016 xzp->z_links = 0; 2017 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 2018 &xzp->z_links, sizeof (xzp->z_links), tx); 2019 ASSERT3U(error, ==, 0); 2020 mutex_exit(&xzp->z_lock); 2021 zfs_unlinked_add(xzp, tx); 2022 2023 if (zp->z_is_sa) 2024 error = sa_remove(zp->z_sa_hdl, 2025 SA_ZPL_XATTR(zfsvfs), tx); 2026 else 2027 error = sa_update(zp->z_sa_hdl, 2028 SA_ZPL_XATTR(zfsvfs), &null_xattr, 2029 sizeof (uint64_t), tx); 2030 ASSERT0(error); 2031 } 2032 VI_LOCK(vp); 2033 vp->v_count--; 2034 ASSERT0(vp->v_count); 2035 VI_UNLOCK(vp); 2036 mutex_exit(&zp->z_lock); 2037 zfs_znode_delete(zp, tx); 2038 } else if (unlinked) { 2039 mutex_exit(&zp->z_lock); 2040 zfs_unlinked_add(zp, tx); 2041#ifdef __FreeBSD__ 2042 vp->v_vflag |= VV_NOSYNC; 2043#endif 2044 } 2045 2046 txtype = TX_REMOVE; 2047 if (flags & FIGNORECASE) 2048 txtype |= TX_CI; 2049 zfs_log_remove(zilog, tx, txtype, dzp, name, obj); 2050 2051 dmu_tx_commit(tx); 2052out: 2053 if (realnmp) 2054 pn_free(realnmp); 2055 2056 zfs_dirent_unlock(dl); 2057 2058 if (!delete_now) 2059 VN_RELE(vp); 2060 if (xzp) 2061 VN_RELE(ZTOV(xzp)); 2062 2063 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2064 zil_commit(zilog, 0); 2065 2066 ZFS_EXIT(zfsvfs); 2067 return (error); 2068} 2069 2070/* 2071 * Create a new directory and insert it into dvp using the name 2072 * provided. Return a pointer to the inserted directory. 2073 * 2074 * IN: dvp - vnode of directory to add subdir to. 2075 * dirname - name of new directory. 2076 * vap - attributes of new directory. 2077 * cr - credentials of caller. 2078 * ct - caller context 2079 * flags - case flags 2080 * vsecp - ACL to be set 2081 * 2082 * OUT: vpp - vnode of created directory. 2083 * 2084 * RETURN: 0 on success, error code on failure. 2085 * 2086 * Timestamps: 2087 * dvp - ctime|mtime updated 2088 * vp - ctime|mtime|atime updated 2089 */ 2090/*ARGSUSED*/ 2091static int 2092zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 2093 caller_context_t *ct, int flags, vsecattr_t *vsecp) 2094{ 2095 znode_t *zp, *dzp = VTOZ(dvp); 2096 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2097 zilog_t *zilog; 2098 zfs_dirlock_t *dl; 2099 uint64_t txtype; 2100 dmu_tx_t *tx; 2101 int error; 2102 int zf = ZNEW; 2103 ksid_t *ksid; 2104 uid_t uid; 2105 gid_t gid = crgetgid(cr); 2106 zfs_acl_ids_t acl_ids; 2107 boolean_t fuid_dirtied; 2108 2109 ASSERT(vap->va_type == VDIR); 2110 2111 /* 2112 * If we have an ephemeral id, ACL, or XVATTR then 2113 * make sure file system is at proper version 2114 */ 2115 2116 ksid = crgetsid(cr, KSID_OWNER); 2117 if (ksid) 2118 uid = ksid_getid(ksid); 2119 else 2120 uid = crgetuid(cr); 2121 if (zfsvfs->z_use_fuids == B_FALSE && 2122 (vsecp || (vap->va_mask & AT_XVATTR) || 2123 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 2124 return (SET_ERROR(EINVAL)); 2125 2126 ZFS_ENTER(zfsvfs); 2127 ZFS_VERIFY_ZP(dzp); 2128 zilog = zfsvfs->z_log; 2129 2130 if (dzp->z_pflags & ZFS_XATTR) { 2131 ZFS_EXIT(zfsvfs); 2132 return (SET_ERROR(EINVAL)); 2133 } 2134 2135 if (zfsvfs->z_utf8 && u8_validate(dirname, 2136 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2137 ZFS_EXIT(zfsvfs); 2138 return (SET_ERROR(EILSEQ)); 2139 } 2140 if (flags & FIGNORECASE) 2141 zf |= ZCILOOK; 2142 2143 if (vap->va_mask & AT_XVATTR) { 2144 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 2145 crgetuid(cr), cr, vap->va_type)) != 0) { 2146 ZFS_EXIT(zfsvfs); 2147 return (error); 2148 } 2149 } 2150 2151 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 2152 vsecp, &acl_ids)) != 0) { 2153 ZFS_EXIT(zfsvfs); 2154 return (error); 2155 } 2156 /* 2157 * First make sure the new directory doesn't exist. 2158 * 2159 * Existence is checked first to make sure we don't return 2160 * EACCES instead of EEXIST which can cause some applications 2161 * to fail. 2162 */ 2163top: 2164 *vpp = NULL; 2165 2166 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 2167 NULL, NULL)) { 2168 zfs_acl_ids_free(&acl_ids); 2169 ZFS_EXIT(zfsvfs); 2170 return (error); 2171 } 2172 2173 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 2174 zfs_acl_ids_free(&acl_ids); 2175 zfs_dirent_unlock(dl); 2176 ZFS_EXIT(zfsvfs); 2177 return (error); 2178 } 2179 2180 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 2181 zfs_acl_ids_free(&acl_ids); 2182 zfs_dirent_unlock(dl); 2183 ZFS_EXIT(zfsvfs); 2184 return (SET_ERROR(EDQUOT)); 2185 } 2186 2187 /* 2188 * Add a new entry to the directory. 2189 */ 2190 tx = dmu_tx_create(zfsvfs->z_os); 2191 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 2192 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2193 fuid_dirtied = zfsvfs->z_fuid_dirty; 2194 if (fuid_dirtied) 2195 zfs_fuid_txhold(zfsvfs, tx); 2196 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2197 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2198 acl_ids.z_aclp->z_acl_bytes); 2199 } 2200 2201 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 2202 ZFS_SA_BASE_ATTR_SIZE); 2203 2204 error = dmu_tx_assign(tx, TXG_NOWAIT); 2205 if (error) { 2206 zfs_dirent_unlock(dl); 2207 if (error == ERESTART) { 2208 dmu_tx_wait(tx); 2209 dmu_tx_abort(tx); 2210 goto top; 2211 } 2212 zfs_acl_ids_free(&acl_ids); 2213 dmu_tx_abort(tx); 2214 ZFS_EXIT(zfsvfs); 2215 return (error); 2216 } 2217 2218 /* 2219 * Create new node. 2220 */ 2221 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 2222 2223 if (fuid_dirtied) 2224 zfs_fuid_sync(zfsvfs, tx); 2225 2226 /* 2227 * Now put new name in parent dir. 2228 */ 2229 (void) zfs_link_create(dl, zp, tx, ZNEW); 2230 2231 *vpp = ZTOV(zp); 2232 2233 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 2234 if (flags & FIGNORECASE) 2235 txtype |= TX_CI; 2236 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 2237 acl_ids.z_fuidp, vap); 2238 2239 zfs_acl_ids_free(&acl_ids); 2240 2241 dmu_tx_commit(tx); 2242 2243 zfs_dirent_unlock(dl); 2244 2245 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2246 zil_commit(zilog, 0); 2247 2248 ZFS_EXIT(zfsvfs); 2249 return (0); 2250} 2251 2252/* 2253 * Remove a directory subdir entry. If the current working 2254 * directory is the same as the subdir to be removed, the 2255 * remove will fail. 2256 * 2257 * IN: dvp - vnode of directory to remove from. 2258 * name - name of directory to be removed. 2259 * cwd - vnode of current working directory. 2260 * cr - credentials of caller. 2261 * ct - caller context 2262 * flags - case flags 2263 * 2264 * RETURN: 0 on success, error code on failure. 2265 * 2266 * Timestamps: 2267 * dvp - ctime|mtime updated 2268 */ 2269/*ARGSUSED*/ 2270static int 2271zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 2272 caller_context_t *ct, int flags) 2273{ 2274 znode_t *dzp = VTOZ(dvp); 2275 znode_t *zp; 2276 vnode_t *vp; 2277 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2278 zilog_t *zilog; 2279 zfs_dirlock_t *dl; 2280 dmu_tx_t *tx; 2281 int error; 2282 int zflg = ZEXISTS; 2283 2284 ZFS_ENTER(zfsvfs); 2285 ZFS_VERIFY_ZP(dzp); 2286 zilog = zfsvfs->z_log; 2287 2288 if (flags & FIGNORECASE) 2289 zflg |= ZCILOOK; 2290top: 2291 zp = NULL; 2292 2293 /* 2294 * Attempt to lock directory; fail if entry doesn't exist. 2295 */ 2296 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 2297 NULL, NULL)) { 2298 ZFS_EXIT(zfsvfs); 2299 return (error); 2300 } 2301 2302 vp = ZTOV(zp); 2303 2304 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 2305 goto out; 2306 } 2307 2308 if (vp->v_type != VDIR) { 2309 error = SET_ERROR(ENOTDIR); 2310 goto out; 2311 } 2312 2313 if (vp == cwd) { 2314 error = SET_ERROR(EINVAL); 2315 goto out; 2316 } 2317 2318 vnevent_rmdir(vp, dvp, name, ct); 2319 2320 /* 2321 * Grab a lock on the directory to make sure that noone is 2322 * trying to add (or lookup) entries while we are removing it. 2323 */ 2324 rw_enter(&zp->z_name_lock, RW_WRITER); 2325 2326 /* 2327 * Grab a lock on the parent pointer to make sure we play well 2328 * with the treewalk and directory rename code. 2329 */ 2330 rw_enter(&zp->z_parent_lock, RW_WRITER); 2331 2332 tx = dmu_tx_create(zfsvfs->z_os); 2333 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2334 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2335 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2336 zfs_sa_upgrade_txholds(tx, zp); 2337 zfs_sa_upgrade_txholds(tx, dzp); 2338 error = dmu_tx_assign(tx, TXG_NOWAIT); 2339 if (error) { 2340 rw_exit(&zp->z_parent_lock); 2341 rw_exit(&zp->z_name_lock); 2342 zfs_dirent_unlock(dl); 2343 VN_RELE(vp); 2344 if (error == ERESTART) { 2345 dmu_tx_wait(tx); 2346 dmu_tx_abort(tx); 2347 goto top; 2348 } 2349 dmu_tx_abort(tx); 2350 ZFS_EXIT(zfsvfs); 2351 return (error); 2352 } 2353 2354#ifdef FREEBSD_NAMECACHE 2355 cache_purge(dvp); 2356#endif 2357 2358 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 2359 2360 if (error == 0) { 2361 uint64_t txtype = TX_RMDIR; 2362 if (flags & FIGNORECASE) 2363 txtype |= TX_CI; 2364 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT); 2365 } 2366 2367 dmu_tx_commit(tx); 2368 2369 rw_exit(&zp->z_parent_lock); 2370 rw_exit(&zp->z_name_lock); 2371#ifdef FREEBSD_NAMECACHE 2372 cache_purge(vp); 2373#endif 2374out: 2375 zfs_dirent_unlock(dl); 2376 2377 VN_RELE(vp); 2378 2379 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2380 zil_commit(zilog, 0); 2381 2382 ZFS_EXIT(zfsvfs); 2383 return (error); 2384} 2385 2386/* 2387 * Read as many directory entries as will fit into the provided 2388 * buffer from the given directory cursor position (specified in 2389 * the uio structure). 2390 * 2391 * IN: vp - vnode of directory to read. 2392 * uio - structure supplying read location, range info, 2393 * and return buffer. 2394 * cr - credentials of caller. 2395 * ct - caller context 2396 * flags - case flags 2397 * 2398 * OUT: uio - updated offset and range, buffer filled. 2399 * eofp - set to true if end-of-file detected. 2400 * 2401 * RETURN: 0 on success, error code on failure. 2402 * 2403 * Timestamps: 2404 * vp - atime updated 2405 * 2406 * Note that the low 4 bits of the cookie returned by zap is always zero. 2407 * This allows us to use the low range for "special" directory entries: 2408 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 2409 * we use the offset 2 for the '.zfs' directory. 2410 */ 2411/* ARGSUSED */ 2412static int 2413zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 2414{ 2415 znode_t *zp = VTOZ(vp); 2416 iovec_t *iovp; 2417 edirent_t *eodp; 2418 dirent64_t *odp; 2419 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2420 objset_t *os; 2421 caddr_t outbuf; 2422 size_t bufsize; 2423 zap_cursor_t zc; 2424 zap_attribute_t zap; 2425 uint_t bytes_wanted; 2426 uint64_t offset; /* must be unsigned; checks for < 1 */ 2427 uint64_t parent; 2428 int local_eof; 2429 int outcount; 2430 int error; 2431 uint8_t prefetch; 2432 boolean_t check_sysattrs; 2433 uint8_t type; 2434 int ncooks; 2435 u_long *cooks = NULL; 2436 int flags = 0; 2437 2438 ZFS_ENTER(zfsvfs); 2439 ZFS_VERIFY_ZP(zp); 2440 2441 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 2442 &parent, sizeof (parent))) != 0) { 2443 ZFS_EXIT(zfsvfs); 2444 return (error); 2445 } 2446 2447 /* 2448 * If we are not given an eof variable, 2449 * use a local one. 2450 */ 2451 if (eofp == NULL) 2452 eofp = &local_eof; 2453 2454 /* 2455 * Check for valid iov_len. 2456 */ 2457 if (uio->uio_iov->iov_len <= 0) { 2458 ZFS_EXIT(zfsvfs); 2459 return (SET_ERROR(EINVAL)); 2460 } 2461 2462 /* 2463 * Quit if directory has been removed (posix) 2464 */ 2465 if ((*eofp = zp->z_unlinked) != 0) { 2466 ZFS_EXIT(zfsvfs); 2467 return (0); 2468 } 2469 2470 error = 0; 2471 os = zfsvfs->z_os; 2472 offset = uio->uio_loffset; 2473 prefetch = zp->z_zn_prefetch; 2474 2475 /* 2476 * Initialize the iterator cursor. 2477 */ 2478 if (offset <= 3) { 2479 /* 2480 * Start iteration from the beginning of the directory. 2481 */ 2482 zap_cursor_init(&zc, os, zp->z_id); 2483 } else { 2484 /* 2485 * The offset is a serialized cursor. 2486 */ 2487 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 2488 } 2489 2490 /* 2491 * Get space to change directory entries into fs independent format. 2492 */ 2493 iovp = uio->uio_iov; 2494 bytes_wanted = iovp->iov_len; 2495 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 2496 bufsize = bytes_wanted; 2497 outbuf = kmem_alloc(bufsize, KM_SLEEP); 2498 odp = (struct dirent64 *)outbuf; 2499 } else { 2500 bufsize = bytes_wanted; 2501 outbuf = NULL; 2502 odp = (struct dirent64 *)iovp->iov_base; 2503 } 2504 eodp = (struct edirent *)odp; 2505 2506 if (ncookies != NULL) { 2507 /* 2508 * Minimum entry size is dirent size and 1 byte for a file name. 2509 */ 2510 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 2511 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 2512 *cookies = cooks; 2513 *ncookies = ncooks; 2514 } 2515 /* 2516 * If this VFS supports the system attribute view interface; and 2517 * we're looking at an extended attribute directory; and we care 2518 * about normalization conflicts on this vfs; then we must check 2519 * for normalization conflicts with the sysattr name space. 2520 */ 2521#ifdef TODO 2522 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 2523 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 2524 (flags & V_RDDIR_ENTFLAGS); 2525#else 2526 check_sysattrs = 0; 2527#endif 2528 2529 /* 2530 * Transform to file-system independent format 2531 */ 2532 outcount = 0; 2533 while (outcount < bytes_wanted) { 2534 ino64_t objnum; 2535 ushort_t reclen; 2536 off64_t *next = NULL; 2537 2538 /* 2539 * Special case `.', `..', and `.zfs'. 2540 */ 2541 if (offset == 0) { 2542 (void) strcpy(zap.za_name, "."); 2543 zap.za_normalization_conflict = 0; 2544 objnum = zp->z_id; 2545 type = DT_DIR; 2546 } else if (offset == 1) { 2547 (void) strcpy(zap.za_name, ".."); 2548 zap.za_normalization_conflict = 0; 2549 objnum = parent; 2550 type = DT_DIR; 2551 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2552 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2553 zap.za_normalization_conflict = 0; 2554 objnum = ZFSCTL_INO_ROOT; 2555 type = DT_DIR; 2556 } else { 2557 /* 2558 * Grab next entry. 2559 */ 2560 if (error = zap_cursor_retrieve(&zc, &zap)) { 2561 if ((*eofp = (error == ENOENT)) != 0) 2562 break; 2563 else 2564 goto update; 2565 } 2566 2567 if (zap.za_integer_length != 8 || 2568 zap.za_num_integers != 1) { 2569 cmn_err(CE_WARN, "zap_readdir: bad directory " 2570 "entry, obj = %lld, offset = %lld\n", 2571 (u_longlong_t)zp->z_id, 2572 (u_longlong_t)offset); 2573 error = SET_ERROR(ENXIO); 2574 goto update; 2575 } 2576 2577 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2578 /* 2579 * MacOS X can extract the object type here such as: 2580 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2581 */ 2582 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2583 2584 if (check_sysattrs && !zap.za_normalization_conflict) { 2585#ifdef TODO 2586 zap.za_normalization_conflict = 2587 xattr_sysattr_casechk(zap.za_name); 2588#else 2589 panic("%s:%u: TODO", __func__, __LINE__); 2590#endif 2591 } 2592 } 2593 2594 if (flags & V_RDDIR_ACCFILTER) { 2595 /* 2596 * If we have no access at all, don't include 2597 * this entry in the returned information 2598 */ 2599 znode_t *ezp; 2600 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2601 goto skip_entry; 2602 if (!zfs_has_access(ezp, cr)) { 2603 VN_RELE(ZTOV(ezp)); 2604 goto skip_entry; 2605 } 2606 VN_RELE(ZTOV(ezp)); 2607 } 2608 2609 if (flags & V_RDDIR_ENTFLAGS) 2610 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2611 else 2612 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2613 2614 /* 2615 * Will this entry fit in the buffer? 2616 */ 2617 if (outcount + reclen > bufsize) { 2618 /* 2619 * Did we manage to fit anything in the buffer? 2620 */ 2621 if (!outcount) { 2622 error = SET_ERROR(EINVAL); 2623 goto update; 2624 } 2625 break; 2626 } 2627 if (flags & V_RDDIR_ENTFLAGS) { 2628 /* 2629 * Add extended flag entry: 2630 */ 2631 eodp->ed_ino = objnum; 2632 eodp->ed_reclen = reclen; 2633 /* NOTE: ed_off is the offset for the *next* entry */ 2634 next = &(eodp->ed_off); 2635 eodp->ed_eflags = zap.za_normalization_conflict ? 2636 ED_CASE_CONFLICT : 0; 2637 (void) strncpy(eodp->ed_name, zap.za_name, 2638 EDIRENT_NAMELEN(reclen)); 2639 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2640 } else { 2641 /* 2642 * Add normal entry: 2643 */ 2644 odp->d_ino = objnum; 2645 odp->d_reclen = reclen; 2646 odp->d_namlen = strlen(zap.za_name); 2647 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 2648 odp->d_type = type; 2649 odp = (dirent64_t *)((intptr_t)odp + reclen); 2650 } 2651 outcount += reclen; 2652 2653 ASSERT(outcount <= bufsize); 2654 2655 /* Prefetch znode */ 2656 if (prefetch) 2657 dmu_prefetch(os, objnum, 0, 0); 2658 2659 skip_entry: 2660 /* 2661 * Move to the next entry, fill in the previous offset. 2662 */ 2663 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2664 zap_cursor_advance(&zc); 2665 offset = zap_cursor_serialize(&zc); 2666 } else { 2667 offset += 1; 2668 } 2669 2670 if (cooks != NULL) { 2671 *cooks++ = offset; 2672 ncooks--; 2673 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 2674 } 2675 } 2676 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2677 2678 /* Subtract unused cookies */ 2679 if (ncookies != NULL) 2680 *ncookies -= ncooks; 2681 2682 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2683 iovp->iov_base += outcount; 2684 iovp->iov_len -= outcount; 2685 uio->uio_resid -= outcount; 2686 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2687 /* 2688 * Reset the pointer. 2689 */ 2690 offset = uio->uio_loffset; 2691 } 2692 2693update: 2694 zap_cursor_fini(&zc); 2695 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2696 kmem_free(outbuf, bufsize); 2697 2698 if (error == ENOENT) 2699 error = 0; 2700 2701 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2702 2703 uio->uio_loffset = offset; 2704 ZFS_EXIT(zfsvfs); 2705 if (error != 0 && cookies != NULL) { 2706 free(*cookies, M_TEMP); 2707 *cookies = NULL; 2708 *ncookies = 0; 2709 } 2710 return (error); 2711} 2712 2713ulong_t zfs_fsync_sync_cnt = 4; 2714 2715static int 2716zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2717{ 2718 znode_t *zp = VTOZ(vp); 2719 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2720 2721 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2722 2723 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { 2724 ZFS_ENTER(zfsvfs); 2725 ZFS_VERIFY_ZP(zp); 2726 zil_commit(zfsvfs->z_log, zp->z_id); 2727 ZFS_EXIT(zfsvfs); 2728 } 2729 return (0); 2730} 2731 2732 2733/* 2734 * Get the requested file attributes and place them in the provided 2735 * vattr structure. 2736 * 2737 * IN: vp - vnode of file. 2738 * vap - va_mask identifies requested attributes. 2739 * If AT_XVATTR set, then optional attrs are requested 2740 * flags - ATTR_NOACLCHECK (CIFS server context) 2741 * cr - credentials of caller. 2742 * ct - caller context 2743 * 2744 * OUT: vap - attribute values. 2745 * 2746 * RETURN: 0 (always succeeds). 2747 */ 2748/* ARGSUSED */ 2749static int 2750zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2751 caller_context_t *ct) 2752{ 2753 znode_t *zp = VTOZ(vp); 2754 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2755 int error = 0; 2756 uint32_t blksize; 2757 u_longlong_t nblocks; 2758 uint64_t links; 2759 uint64_t mtime[2], ctime[2], crtime[2], rdev; 2760 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2761 xoptattr_t *xoap = NULL; 2762 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2763 sa_bulk_attr_t bulk[4]; 2764 int count = 0; 2765 2766 ZFS_ENTER(zfsvfs); 2767 ZFS_VERIFY_ZP(zp); 2768 2769 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2770 2771 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 2772 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 2773 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16); 2774 if (vp->v_type == VBLK || vp->v_type == VCHR) 2775 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, 2776 &rdev, 8); 2777 2778 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 2779 ZFS_EXIT(zfsvfs); 2780 return (error); 2781 } 2782 2783 /* 2784 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2785 * Also, if we are the owner don't bother, since owner should 2786 * always be allowed to read basic attributes of file. 2787 */ 2788 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 2789 (vap->va_uid != crgetuid(cr))) { 2790 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2791 skipaclchk, cr)) { 2792 ZFS_EXIT(zfsvfs); 2793 return (error); 2794 } 2795 } 2796 2797 /* 2798 * Return all attributes. It's cheaper to provide the answer 2799 * than to determine whether we were asked the question. 2800 */ 2801 2802 mutex_enter(&zp->z_lock); 2803 vap->va_type = IFTOVT(zp->z_mode); 2804 vap->va_mode = zp->z_mode & ~S_IFMT; 2805#ifdef sun 2806 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2807#else 2808 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 2809#endif 2810 vap->va_nodeid = zp->z_id; 2811 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2812 links = zp->z_links + 1; 2813 else 2814 links = zp->z_links; 2815 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */ 2816 vap->va_size = zp->z_size; 2817#ifdef sun 2818 vap->va_rdev = vp->v_rdev; 2819#else 2820 if (vp->v_type == VBLK || vp->v_type == VCHR) 2821 vap->va_rdev = zfs_cmpldev(rdev); 2822#endif 2823 vap->va_seq = zp->z_seq; 2824 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2825 2826 /* 2827 * Add in any requested optional attributes and the create time. 2828 * Also set the corresponding bits in the returned attribute bitmap. 2829 */ 2830 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2831 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2832 xoap->xoa_archive = 2833 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2834 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2835 } 2836 2837 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2838 xoap->xoa_readonly = 2839 ((zp->z_pflags & ZFS_READONLY) != 0); 2840 XVA_SET_RTN(xvap, XAT_READONLY); 2841 } 2842 2843 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2844 xoap->xoa_system = 2845 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2846 XVA_SET_RTN(xvap, XAT_SYSTEM); 2847 } 2848 2849 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2850 xoap->xoa_hidden = 2851 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2852 XVA_SET_RTN(xvap, XAT_HIDDEN); 2853 } 2854 2855 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2856 xoap->xoa_nounlink = 2857 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2858 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2859 } 2860 2861 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2862 xoap->xoa_immutable = 2863 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2864 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2865 } 2866 2867 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2868 xoap->xoa_appendonly = 2869 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2870 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2871 } 2872 2873 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2874 xoap->xoa_nodump = 2875 ((zp->z_pflags & ZFS_NODUMP) != 0); 2876 XVA_SET_RTN(xvap, XAT_NODUMP); 2877 } 2878 2879 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2880 xoap->xoa_opaque = 2881 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2882 XVA_SET_RTN(xvap, XAT_OPAQUE); 2883 } 2884 2885 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2886 xoap->xoa_av_quarantined = 2887 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2888 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2889 } 2890 2891 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2892 xoap->xoa_av_modified = 2893 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2894 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2895 } 2896 2897 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2898 vp->v_type == VREG) { 2899 zfs_sa_get_scanstamp(zp, xvap); 2900 } 2901 2902 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2903 uint64_t times[2]; 2904 2905 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 2906 times, sizeof (times)); 2907 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 2908 XVA_SET_RTN(xvap, XAT_CREATETIME); 2909 } 2910 2911 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2912 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2913 XVA_SET_RTN(xvap, XAT_REPARSE); 2914 } 2915 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2916 xoap->xoa_generation = zp->z_gen; 2917 XVA_SET_RTN(xvap, XAT_GEN); 2918 } 2919 2920 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2921 xoap->xoa_offline = 2922 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2923 XVA_SET_RTN(xvap, XAT_OFFLINE); 2924 } 2925 2926 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2927 xoap->xoa_sparse = 2928 ((zp->z_pflags & ZFS_SPARSE) != 0); 2929 XVA_SET_RTN(xvap, XAT_SPARSE); 2930 } 2931 } 2932 2933 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2934 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2935 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2936 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 2937 2938 mutex_exit(&zp->z_lock); 2939 2940 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 2941 vap->va_blksize = blksize; 2942 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2943 2944 if (zp->z_blksz == 0) { 2945 /* 2946 * Block size hasn't been set; suggest maximal I/O transfers. 2947 */ 2948 vap->va_blksize = zfsvfs->z_max_blksz; 2949 } 2950 2951 ZFS_EXIT(zfsvfs); 2952 return (0); 2953} 2954 2955/* 2956 * Set the file attributes to the values contained in the 2957 * vattr structure. 2958 * 2959 * IN: vp - vnode of file to be modified. 2960 * vap - new attribute values. 2961 * If AT_XVATTR set, then optional attrs are being set 2962 * flags - ATTR_UTIME set if non-default time values provided. 2963 * - ATTR_NOACLCHECK (CIFS context only). 2964 * cr - credentials of caller. 2965 * ct - caller context 2966 * 2967 * RETURN: 0 on success, error code on failure. 2968 * 2969 * Timestamps: 2970 * vp - ctime updated, mtime updated if size changed. 2971 */ 2972/* ARGSUSED */ 2973static int 2974zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2975 caller_context_t *ct) 2976{ 2977 znode_t *zp = VTOZ(vp); 2978 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2979 zilog_t *zilog; 2980 dmu_tx_t *tx; 2981 vattr_t oldva; 2982 xvattr_t tmpxvattr; 2983 uint_t mask = vap->va_mask; 2984 uint_t saved_mask = 0; 2985 uint64_t saved_mode; 2986 int trim_mask = 0; 2987 uint64_t new_mode; 2988 uint64_t new_uid, new_gid; 2989 uint64_t xattr_obj; 2990 uint64_t mtime[2], ctime[2]; 2991 znode_t *attrzp; 2992 int need_policy = FALSE; 2993 int err, err2; 2994 zfs_fuid_info_t *fuidp = NULL; 2995 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2996 xoptattr_t *xoap; 2997 zfs_acl_t *aclp; 2998 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2999 boolean_t fuid_dirtied = B_FALSE; 3000 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 3001 int count = 0, xattr_count = 0; 3002 3003 if (mask == 0) 3004 return (0); 3005 3006 if (mask & AT_NOSET) 3007 return (SET_ERROR(EINVAL)); 3008 3009 ZFS_ENTER(zfsvfs); 3010 ZFS_VERIFY_ZP(zp); 3011 3012 zilog = zfsvfs->z_log; 3013 3014 /* 3015 * Make sure that if we have ephemeral uid/gid or xvattr specified 3016 * that file system is at proper version level 3017 */ 3018 3019 if (zfsvfs->z_use_fuids == B_FALSE && 3020 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 3021 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 3022 (mask & AT_XVATTR))) { 3023 ZFS_EXIT(zfsvfs); 3024 return (SET_ERROR(EINVAL)); 3025 } 3026 3027 if (mask & AT_SIZE && vp->v_type == VDIR) { 3028 ZFS_EXIT(zfsvfs); 3029 return (SET_ERROR(EISDIR)); 3030 } 3031 3032 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 3033 ZFS_EXIT(zfsvfs); 3034 return (SET_ERROR(EINVAL)); 3035 } 3036 3037 /* 3038 * If this is an xvattr_t, then get a pointer to the structure of 3039 * optional attributes. If this is NULL, then we have a vattr_t. 3040 */ 3041 xoap = xva_getxoptattr(xvap); 3042 3043 xva_init(&tmpxvattr); 3044 3045 /* 3046 * Immutable files can only alter immutable bit and atime 3047 */ 3048 if ((zp->z_pflags & ZFS_IMMUTABLE) && 3049 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 3050 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 3051 ZFS_EXIT(zfsvfs); 3052 return (SET_ERROR(EPERM)); 3053 } 3054 3055 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 3056 ZFS_EXIT(zfsvfs); 3057 return (SET_ERROR(EPERM)); 3058 } 3059 3060 /* 3061 * Verify timestamps doesn't overflow 32 bits. 3062 * ZFS can handle large timestamps, but 32bit syscalls can't 3063 * handle times greater than 2039. This check should be removed 3064 * once large timestamps are fully supported. 3065 */ 3066 if (mask & (AT_ATIME | AT_MTIME)) { 3067 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 3068 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 3069 ZFS_EXIT(zfsvfs); 3070 return (SET_ERROR(EOVERFLOW)); 3071 } 3072 } 3073 3074top: 3075 attrzp = NULL; 3076 aclp = NULL; 3077 3078 /* Can this be moved to before the top label? */ 3079 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 3080 ZFS_EXIT(zfsvfs); 3081 return (SET_ERROR(EROFS)); 3082 } 3083 3084 /* 3085 * First validate permissions 3086 */ 3087 3088 if (mask & AT_SIZE) { 3089 /* 3090 * XXX - Note, we are not providing any open 3091 * mode flags here (like FNDELAY), so we may 3092 * block if there are locks present... this 3093 * should be addressed in openat(). 3094 */ 3095 /* XXX - would it be OK to generate a log record here? */ 3096 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 3097 if (err) { 3098 ZFS_EXIT(zfsvfs); 3099 return (err); 3100 } 3101 } 3102 3103 if (mask & (AT_ATIME|AT_MTIME) || 3104 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 3105 XVA_ISSET_REQ(xvap, XAT_READONLY) || 3106 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 3107 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 3108 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 3109 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 3110 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 3111 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 3112 skipaclchk, cr); 3113 } 3114 3115 if (mask & (AT_UID|AT_GID)) { 3116 int idmask = (mask & (AT_UID|AT_GID)); 3117 int take_owner; 3118 int take_group; 3119 3120 /* 3121 * NOTE: even if a new mode is being set, 3122 * we may clear S_ISUID/S_ISGID bits. 3123 */ 3124 3125 if (!(mask & AT_MODE)) 3126 vap->va_mode = zp->z_mode; 3127 3128 /* 3129 * Take ownership or chgrp to group we are a member of 3130 */ 3131 3132 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 3133 take_group = (mask & AT_GID) && 3134 zfs_groupmember(zfsvfs, vap->va_gid, cr); 3135 3136 /* 3137 * If both AT_UID and AT_GID are set then take_owner and 3138 * take_group must both be set in order to allow taking 3139 * ownership. 3140 * 3141 * Otherwise, send the check through secpolicy_vnode_setattr() 3142 * 3143 */ 3144 3145 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 3146 ((idmask == AT_UID) && take_owner) || 3147 ((idmask == AT_GID) && take_group)) { 3148 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 3149 skipaclchk, cr) == 0) { 3150 /* 3151 * Remove setuid/setgid for non-privileged users 3152 */ 3153 secpolicy_setid_clear(vap, vp, cr); 3154 trim_mask = (mask & (AT_UID|AT_GID)); 3155 } else { 3156 need_policy = TRUE; 3157 } 3158 } else { 3159 need_policy = TRUE; 3160 } 3161 } 3162 3163 mutex_enter(&zp->z_lock); 3164 oldva.va_mode = zp->z_mode; 3165 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 3166 if (mask & AT_XVATTR) { 3167 /* 3168 * Update xvattr mask to include only those attributes 3169 * that are actually changing. 3170 * 3171 * the bits will be restored prior to actually setting 3172 * the attributes so the caller thinks they were set. 3173 */ 3174 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 3175 if (xoap->xoa_appendonly != 3176 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 3177 need_policy = TRUE; 3178 } else { 3179 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 3180 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 3181 } 3182 } 3183 3184 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 3185 if (xoap->xoa_nounlink != 3186 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 3187 need_policy = TRUE; 3188 } else { 3189 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 3190 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 3191 } 3192 } 3193 3194 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 3195 if (xoap->xoa_immutable != 3196 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 3197 need_policy = TRUE; 3198 } else { 3199 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 3200 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 3201 } 3202 } 3203 3204 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 3205 if (xoap->xoa_nodump != 3206 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 3207 need_policy = TRUE; 3208 } else { 3209 XVA_CLR_REQ(xvap, XAT_NODUMP); 3210 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 3211 } 3212 } 3213 3214 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 3215 if (xoap->xoa_av_modified != 3216 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 3217 need_policy = TRUE; 3218 } else { 3219 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 3220 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 3221 } 3222 } 3223 3224 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 3225 if ((vp->v_type != VREG && 3226 xoap->xoa_av_quarantined) || 3227 xoap->xoa_av_quarantined != 3228 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 3229 need_policy = TRUE; 3230 } else { 3231 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 3232 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 3233 } 3234 } 3235 3236 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 3237 mutex_exit(&zp->z_lock); 3238 ZFS_EXIT(zfsvfs); 3239 return (SET_ERROR(EPERM)); 3240 } 3241 3242 if (need_policy == FALSE && 3243 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 3244 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 3245 need_policy = TRUE; 3246 } 3247 } 3248 3249 mutex_exit(&zp->z_lock); 3250 3251 if (mask & AT_MODE) { 3252 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 3253 err = secpolicy_setid_setsticky_clear(vp, vap, 3254 &oldva, cr); 3255 if (err) { 3256 ZFS_EXIT(zfsvfs); 3257 return (err); 3258 } 3259 trim_mask |= AT_MODE; 3260 } else { 3261 need_policy = TRUE; 3262 } 3263 } 3264 3265 if (need_policy) { 3266 /* 3267 * If trim_mask is set then take ownership 3268 * has been granted or write_acl is present and user 3269 * has the ability to modify mode. In that case remove 3270 * UID|GID and or MODE from mask so that 3271 * secpolicy_vnode_setattr() doesn't revoke it. 3272 */ 3273 3274 if (trim_mask) { 3275 saved_mask = vap->va_mask; 3276 vap->va_mask &= ~trim_mask; 3277 if (trim_mask & AT_MODE) { 3278 /* 3279 * Save the mode, as secpolicy_vnode_setattr() 3280 * will overwrite it with ova.va_mode. 3281 */ 3282 saved_mode = vap->va_mode; 3283 } 3284 } 3285 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 3286 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 3287 if (err) { 3288 ZFS_EXIT(zfsvfs); 3289 return (err); 3290 } 3291 3292 if (trim_mask) { 3293 vap->va_mask |= saved_mask; 3294 if (trim_mask & AT_MODE) { 3295 /* 3296 * Recover the mode after 3297 * secpolicy_vnode_setattr(). 3298 */ 3299 vap->va_mode = saved_mode; 3300 } 3301 } 3302 } 3303 3304 /* 3305 * secpolicy_vnode_setattr, or take ownership may have 3306 * changed va_mask 3307 */ 3308 mask = vap->va_mask; 3309 3310 if ((mask & (AT_UID | AT_GID))) { 3311 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 3312 &xattr_obj, sizeof (xattr_obj)); 3313 3314 if (err == 0 && xattr_obj) { 3315 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 3316 if (err) 3317 goto out2; 3318 } 3319 if (mask & AT_UID) { 3320 new_uid = zfs_fuid_create(zfsvfs, 3321 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 3322 if (new_uid != zp->z_uid && 3323 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 3324 if (attrzp) 3325 VN_RELE(ZTOV(attrzp)); 3326 err = SET_ERROR(EDQUOT); 3327 goto out2; 3328 } 3329 } 3330 3331 if (mask & AT_GID) { 3332 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 3333 cr, ZFS_GROUP, &fuidp); 3334 if (new_gid != zp->z_gid && 3335 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 3336 if (attrzp) 3337 VN_RELE(ZTOV(attrzp)); 3338 err = SET_ERROR(EDQUOT); 3339 goto out2; 3340 } 3341 } 3342 } 3343 tx = dmu_tx_create(zfsvfs->z_os); 3344 3345 if (mask & AT_MODE) { 3346 uint64_t pmode = zp->z_mode; 3347 uint64_t acl_obj; 3348 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 3349 3350 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && 3351 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 3352 err = SET_ERROR(EPERM); 3353 goto out; 3354 } 3355 3356 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 3357 goto out; 3358 3359 mutex_enter(&zp->z_lock); 3360 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 3361 /* 3362 * Are we upgrading ACL from old V0 format 3363 * to V1 format? 3364 */ 3365 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 3366 zfs_znode_acl_version(zp) == 3367 ZFS_ACL_VERSION_INITIAL) { 3368 dmu_tx_hold_free(tx, acl_obj, 0, 3369 DMU_OBJECT_END); 3370 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3371 0, aclp->z_acl_bytes); 3372 } else { 3373 dmu_tx_hold_write(tx, acl_obj, 0, 3374 aclp->z_acl_bytes); 3375 } 3376 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3377 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3378 0, aclp->z_acl_bytes); 3379 } 3380 mutex_exit(&zp->z_lock); 3381 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3382 } else { 3383 if ((mask & AT_XVATTR) && 3384 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3385 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3386 else 3387 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3388 } 3389 3390 if (attrzp) { 3391 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 3392 } 3393 3394 fuid_dirtied = zfsvfs->z_fuid_dirty; 3395 if (fuid_dirtied) 3396 zfs_fuid_txhold(zfsvfs, tx); 3397 3398 zfs_sa_upgrade_txholds(tx, zp); 3399 3400 err = dmu_tx_assign(tx, TXG_NOWAIT); 3401 if (err) { 3402 if (err == ERESTART) 3403 dmu_tx_wait(tx); 3404 goto out; 3405 } 3406 3407 count = 0; 3408 /* 3409 * Set each attribute requested. 3410 * We group settings according to the locks they need to acquire. 3411 * 3412 * Note: you cannot set ctime directly, although it will be 3413 * updated as a side-effect of calling this function. 3414 */ 3415 3416 3417 if (mask & (AT_UID|AT_GID|AT_MODE)) 3418 mutex_enter(&zp->z_acl_lock); 3419 mutex_enter(&zp->z_lock); 3420 3421 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3422 &zp->z_pflags, sizeof (zp->z_pflags)); 3423 3424 if (attrzp) { 3425 if (mask & (AT_UID|AT_GID|AT_MODE)) 3426 mutex_enter(&attrzp->z_acl_lock); 3427 mutex_enter(&attrzp->z_lock); 3428 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3429 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3430 sizeof (attrzp->z_pflags)); 3431 } 3432 3433 if (mask & (AT_UID|AT_GID)) { 3434 3435 if (mask & AT_UID) { 3436 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3437 &new_uid, sizeof (new_uid)); 3438 zp->z_uid = new_uid; 3439 if (attrzp) { 3440 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3441 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3442 sizeof (new_uid)); 3443 attrzp->z_uid = new_uid; 3444 } 3445 } 3446 3447 if (mask & AT_GID) { 3448 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3449 NULL, &new_gid, sizeof (new_gid)); 3450 zp->z_gid = new_gid; 3451 if (attrzp) { 3452 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3453 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 3454 sizeof (new_gid)); 3455 attrzp->z_gid = new_gid; 3456 } 3457 } 3458 if (!(mask & AT_MODE)) { 3459 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 3460 NULL, &new_mode, sizeof (new_mode)); 3461 new_mode = zp->z_mode; 3462 } 3463 err = zfs_acl_chown_setattr(zp); 3464 ASSERT(err == 0); 3465 if (attrzp) { 3466 err = zfs_acl_chown_setattr(attrzp); 3467 ASSERT(err == 0); 3468 } 3469 } 3470 3471 if (mask & AT_MODE) { 3472 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 3473 &new_mode, sizeof (new_mode)); 3474 zp->z_mode = new_mode; 3475 ASSERT3U((uintptr_t)aclp, !=, 0); 3476 err = zfs_aclset_common(zp, aclp, cr, tx); 3477 ASSERT0(err); 3478 if (zp->z_acl_cached) 3479 zfs_acl_free(zp->z_acl_cached); 3480 zp->z_acl_cached = aclp; 3481 aclp = NULL; 3482 } 3483 3484 3485 if (mask & AT_ATIME) { 3486 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 3487 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 3488 &zp->z_atime, sizeof (zp->z_atime)); 3489 } 3490 3491 if (mask & AT_MTIME) { 3492 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 3493 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 3494 mtime, sizeof (mtime)); 3495 } 3496 3497 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 3498 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 3499 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 3500 NULL, mtime, sizeof (mtime)); 3501 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3502 &ctime, sizeof (ctime)); 3503 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 3504 B_TRUE); 3505 } else if (mask != 0) { 3506 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3507 &ctime, sizeof (ctime)); 3508 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime, 3509 B_TRUE); 3510 if (attrzp) { 3511 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3512 SA_ZPL_CTIME(zfsvfs), NULL, 3513 &ctime, sizeof (ctime)); 3514 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 3515 mtime, ctime, B_TRUE); 3516 } 3517 } 3518 /* 3519 * Do this after setting timestamps to prevent timestamp 3520 * update from toggling bit 3521 */ 3522 3523 if (xoap && (mask & AT_XVATTR)) { 3524 3525 /* 3526 * restore trimmed off masks 3527 * so that return masks can be set for caller. 3528 */ 3529 3530 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 3531 XVA_SET_REQ(xvap, XAT_APPENDONLY); 3532 } 3533 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 3534 XVA_SET_REQ(xvap, XAT_NOUNLINK); 3535 } 3536 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 3537 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 3538 } 3539 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 3540 XVA_SET_REQ(xvap, XAT_NODUMP); 3541 } 3542 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 3543 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 3544 } 3545 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 3546 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 3547 } 3548 3549 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3550 ASSERT(vp->v_type == VREG); 3551 3552 zfs_xvattr_set(zp, xvap, tx); 3553 } 3554 3555 if (fuid_dirtied) 3556 zfs_fuid_sync(zfsvfs, tx); 3557 3558 if (mask != 0) 3559 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 3560 3561 mutex_exit(&zp->z_lock); 3562 if (mask & (AT_UID|AT_GID|AT_MODE)) 3563 mutex_exit(&zp->z_acl_lock); 3564 3565 if (attrzp) { 3566 if (mask & (AT_UID|AT_GID|AT_MODE)) 3567 mutex_exit(&attrzp->z_acl_lock); 3568 mutex_exit(&attrzp->z_lock); 3569 } 3570out: 3571 if (err == 0 && attrzp) { 3572 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 3573 xattr_count, tx); 3574 ASSERT(err2 == 0); 3575 } 3576 3577 if (attrzp) 3578 VN_RELE(ZTOV(attrzp)); 3579 3580 if (aclp) 3581 zfs_acl_free(aclp); 3582 3583 if (fuidp) { 3584 zfs_fuid_info_free(fuidp); 3585 fuidp = NULL; 3586 } 3587 3588 if (err) { 3589 dmu_tx_abort(tx); 3590 if (err == ERESTART) 3591 goto top; 3592 } else { 3593 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3594 dmu_tx_commit(tx); 3595 } 3596 3597out2: 3598 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3599 zil_commit(zilog, 0); 3600 3601 ZFS_EXIT(zfsvfs); 3602 return (err); 3603} 3604 3605typedef struct zfs_zlock { 3606 krwlock_t *zl_rwlock; /* lock we acquired */ 3607 znode_t *zl_znode; /* znode we held */ 3608 struct zfs_zlock *zl_next; /* next in list */ 3609} zfs_zlock_t; 3610 3611/* 3612 * Drop locks and release vnodes that were held by zfs_rename_lock(). 3613 */ 3614static void 3615zfs_rename_unlock(zfs_zlock_t **zlpp) 3616{ 3617 zfs_zlock_t *zl; 3618 3619 while ((zl = *zlpp) != NULL) { 3620 if (zl->zl_znode != NULL) 3621 VN_RELE(ZTOV(zl->zl_znode)); 3622 rw_exit(zl->zl_rwlock); 3623 *zlpp = zl->zl_next; 3624 kmem_free(zl, sizeof (*zl)); 3625 } 3626} 3627 3628/* 3629 * Search back through the directory tree, using the ".." entries. 3630 * Lock each directory in the chain to prevent concurrent renames. 3631 * Fail any attempt to move a directory into one of its own descendants. 3632 * XXX - z_parent_lock can overlap with map or grow locks 3633 */ 3634static int 3635zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 3636{ 3637 zfs_zlock_t *zl; 3638 znode_t *zp = tdzp; 3639 uint64_t rootid = zp->z_zfsvfs->z_root; 3640 uint64_t oidp = zp->z_id; 3641 krwlock_t *rwlp = &szp->z_parent_lock; 3642 krw_t rw = RW_WRITER; 3643 3644 /* 3645 * First pass write-locks szp and compares to zp->z_id. 3646 * Later passes read-lock zp and compare to zp->z_parent. 3647 */ 3648 do { 3649 if (!rw_tryenter(rwlp, rw)) { 3650 /* 3651 * Another thread is renaming in this path. 3652 * Note that if we are a WRITER, we don't have any 3653 * parent_locks held yet. 3654 */ 3655 if (rw == RW_READER && zp->z_id > szp->z_id) { 3656 /* 3657 * Drop our locks and restart 3658 */ 3659 zfs_rename_unlock(&zl); 3660 *zlpp = NULL; 3661 zp = tdzp; 3662 oidp = zp->z_id; 3663 rwlp = &szp->z_parent_lock; 3664 rw = RW_WRITER; 3665 continue; 3666 } else { 3667 /* 3668 * Wait for other thread to drop its locks 3669 */ 3670 rw_enter(rwlp, rw); 3671 } 3672 } 3673 3674 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 3675 zl->zl_rwlock = rwlp; 3676 zl->zl_znode = NULL; 3677 zl->zl_next = *zlpp; 3678 *zlpp = zl; 3679 3680 if (oidp == szp->z_id) /* We're a descendant of szp */ 3681 return (SET_ERROR(EINVAL)); 3682 3683 if (oidp == rootid) /* We've hit the top */ 3684 return (0); 3685 3686 if (rw == RW_READER) { /* i.e. not the first pass */ 3687 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp); 3688 if (error) 3689 return (error); 3690 zl->zl_znode = zp; 3691 } 3692 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs), 3693 &oidp, sizeof (oidp)); 3694 rwlp = &zp->z_parent_lock; 3695 rw = RW_READER; 3696 3697 } while (zp->z_id != sdzp->z_id); 3698 3699 return (0); 3700} 3701 3702/* 3703 * Move an entry from the provided source directory to the target 3704 * directory. Change the entry name as indicated. 3705 * 3706 * IN: sdvp - Source directory containing the "old entry". 3707 * snm - Old entry name. 3708 * tdvp - Target directory to contain the "new entry". 3709 * tnm - New entry name. 3710 * cr - credentials of caller. 3711 * ct - caller context 3712 * flags - case flags 3713 * 3714 * RETURN: 0 on success, error code on failure. 3715 * 3716 * Timestamps: 3717 * sdvp,tdvp - ctime|mtime updated 3718 */ 3719/*ARGSUSED*/ 3720static int 3721zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 3722 caller_context_t *ct, int flags) 3723{ 3724 znode_t *tdzp, *szp, *tzp; 3725 znode_t *sdzp = VTOZ(sdvp); 3726 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 3727 zilog_t *zilog; 3728 vnode_t *realvp; 3729 zfs_dirlock_t *sdl, *tdl; 3730 dmu_tx_t *tx; 3731 zfs_zlock_t *zl; 3732 int cmp, serr, terr; 3733 int error = 0; 3734 int zflg = 0; 3735 3736 ZFS_ENTER(zfsvfs); 3737 ZFS_VERIFY_ZP(sdzp); 3738 zilog = zfsvfs->z_log; 3739 3740 /* 3741 * Make sure we have the real vp for the target directory. 3742 */ 3743 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 3744 tdvp = realvp; 3745 3746 tdzp = VTOZ(tdvp); 3747 ZFS_VERIFY_ZP(tdzp); 3748 3749 /* 3750 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 3751 * ctldir appear to have the same v_vfsp. 3752 */ 3753 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) { 3754 ZFS_EXIT(zfsvfs); 3755 return (SET_ERROR(EXDEV)); 3756 } 3757 3758 if (zfsvfs->z_utf8 && u8_validate(tnm, 3759 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3760 ZFS_EXIT(zfsvfs); 3761 return (SET_ERROR(EILSEQ)); 3762 } 3763 3764 if (flags & FIGNORECASE) 3765 zflg |= ZCILOOK; 3766 3767top: 3768 szp = NULL; 3769 tzp = NULL; 3770 zl = NULL; 3771 3772 /* 3773 * This is to prevent the creation of links into attribute space 3774 * by renaming a linked file into/outof an attribute directory. 3775 * See the comment in zfs_link() for why this is considered bad. 3776 */ 3777 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3778 ZFS_EXIT(zfsvfs); 3779 return (SET_ERROR(EINVAL)); 3780 } 3781 3782 /* 3783 * Lock source and target directory entries. To prevent deadlock, 3784 * a lock ordering must be defined. We lock the directory with 3785 * the smallest object id first, or if it's a tie, the one with 3786 * the lexically first name. 3787 */ 3788 if (sdzp->z_id < tdzp->z_id) { 3789 cmp = -1; 3790 } else if (sdzp->z_id > tdzp->z_id) { 3791 cmp = 1; 3792 } else { 3793 /* 3794 * First compare the two name arguments without 3795 * considering any case folding. 3796 */ 3797 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 3798 3799 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 3800 ASSERT(error == 0 || !zfsvfs->z_utf8); 3801 if (cmp == 0) { 3802 /* 3803 * POSIX: "If the old argument and the new argument 3804 * both refer to links to the same existing file, 3805 * the rename() function shall return successfully 3806 * and perform no other action." 3807 */ 3808 ZFS_EXIT(zfsvfs); 3809 return (0); 3810 } 3811 /* 3812 * If the file system is case-folding, then we may 3813 * have some more checking to do. A case-folding file 3814 * system is either supporting mixed case sensitivity 3815 * access or is completely case-insensitive. Note 3816 * that the file system is always case preserving. 3817 * 3818 * In mixed sensitivity mode case sensitive behavior 3819 * is the default. FIGNORECASE must be used to 3820 * explicitly request case insensitive behavior. 3821 * 3822 * If the source and target names provided differ only 3823 * by case (e.g., a request to rename 'tim' to 'Tim'), 3824 * we will treat this as a special case in the 3825 * case-insensitive mode: as long as the source name 3826 * is an exact match, we will allow this to proceed as 3827 * a name-change request. 3828 */ 3829 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3830 (zfsvfs->z_case == ZFS_CASE_MIXED && 3831 flags & FIGNORECASE)) && 3832 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3833 &error) == 0) { 3834 /* 3835 * case preserving rename request, require exact 3836 * name matches 3837 */ 3838 zflg |= ZCIEXACT; 3839 zflg &= ~ZCILOOK; 3840 } 3841 } 3842 3843 /* 3844 * If the source and destination directories are the same, we should 3845 * grab the z_name_lock of that directory only once. 3846 */ 3847 if (sdzp == tdzp) { 3848 zflg |= ZHAVELOCK; 3849 rw_enter(&sdzp->z_name_lock, RW_READER); 3850 } 3851 3852 if (cmp < 0) { 3853 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3854 ZEXISTS | zflg, NULL, NULL); 3855 terr = zfs_dirent_lock(&tdl, 3856 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3857 } else { 3858 terr = zfs_dirent_lock(&tdl, 3859 tdzp, tnm, &tzp, zflg, NULL, NULL); 3860 serr = zfs_dirent_lock(&sdl, 3861 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3862 NULL, NULL); 3863 } 3864 3865 if (serr) { 3866 /* 3867 * Source entry invalid or not there. 3868 */ 3869 if (!terr) { 3870 zfs_dirent_unlock(tdl); 3871 if (tzp) 3872 VN_RELE(ZTOV(tzp)); 3873 } 3874 3875 if (sdzp == tdzp) 3876 rw_exit(&sdzp->z_name_lock); 3877 3878 /* 3879 * FreeBSD: In OpenSolaris they only check if rename source is 3880 * ".." here, because "." is handled in their lookup. This is 3881 * not the case for FreeBSD, so we check for "." explicitly. 3882 */ 3883 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 3884 serr = SET_ERROR(EINVAL); 3885 ZFS_EXIT(zfsvfs); 3886 return (serr); 3887 } 3888 if (terr) { 3889 zfs_dirent_unlock(sdl); 3890 VN_RELE(ZTOV(szp)); 3891 3892 if (sdzp == tdzp) 3893 rw_exit(&sdzp->z_name_lock); 3894 3895 if (strcmp(tnm, "..") == 0) 3896 terr = SET_ERROR(EINVAL); 3897 ZFS_EXIT(zfsvfs); 3898 return (terr); 3899 } 3900 3901 /* 3902 * Must have write access at the source to remove the old entry 3903 * and write access at the target to create the new entry. 3904 * Note that if target and source are the same, this can be 3905 * done in a single check. 3906 */ 3907 3908 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3909 goto out; 3910 3911 if (ZTOV(szp)->v_type == VDIR) { 3912 /* 3913 * Check to make sure rename is valid. 3914 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3915 */ 3916 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3917 goto out; 3918 } 3919 3920 /* 3921 * Does target exist? 3922 */ 3923 if (tzp) { 3924 /* 3925 * Source and target must be the same type. 3926 */ 3927 if (ZTOV(szp)->v_type == VDIR) { 3928 if (ZTOV(tzp)->v_type != VDIR) { 3929 error = SET_ERROR(ENOTDIR); 3930 goto out; 3931 } 3932 } else { 3933 if (ZTOV(tzp)->v_type == VDIR) { 3934 error = SET_ERROR(EISDIR); 3935 goto out; 3936 } 3937 } 3938 /* 3939 * POSIX dictates that when the source and target 3940 * entries refer to the same file object, rename 3941 * must do nothing and exit without error. 3942 */ 3943 if (szp->z_id == tzp->z_id) { 3944 error = 0; 3945 goto out; 3946 } 3947 } 3948 3949 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3950 if (tzp) 3951 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3952 3953 /* 3954 * notify the target directory if it is not the same 3955 * as source directory. 3956 */ 3957 if (tdvp != sdvp) { 3958 vnevent_rename_dest_dir(tdvp, ct); 3959 } 3960 3961 tx = dmu_tx_create(zfsvfs->z_os); 3962 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3963 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 3964 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3965 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3966 if (sdzp != tdzp) { 3967 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 3968 zfs_sa_upgrade_txholds(tx, tdzp); 3969 } 3970 if (tzp) { 3971 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 3972 zfs_sa_upgrade_txholds(tx, tzp); 3973 } 3974 3975 zfs_sa_upgrade_txholds(tx, szp); 3976 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3977 error = dmu_tx_assign(tx, TXG_NOWAIT); 3978 if (error) { 3979 if (zl != NULL) 3980 zfs_rename_unlock(&zl); 3981 zfs_dirent_unlock(sdl); 3982 zfs_dirent_unlock(tdl); 3983 3984 if (sdzp == tdzp) 3985 rw_exit(&sdzp->z_name_lock); 3986 3987 VN_RELE(ZTOV(szp)); 3988 if (tzp) 3989 VN_RELE(ZTOV(tzp)); 3990 if (error == ERESTART) { 3991 dmu_tx_wait(tx); 3992 dmu_tx_abort(tx); 3993 goto top; 3994 } 3995 dmu_tx_abort(tx); 3996 ZFS_EXIT(zfsvfs); 3997 return (error); 3998 } 3999 4000 if (tzp) /* Attempt to remove the existing target */ 4001 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 4002 4003 if (error == 0) { 4004 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 4005 if (error == 0) { 4006 szp->z_pflags |= ZFS_AV_MODIFIED; 4007 4008 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 4009 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 4010 ASSERT0(error); 4011 4012 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 4013 if (error == 0) { 4014 zfs_log_rename(zilog, tx, TX_RENAME | 4015 (flags & FIGNORECASE ? TX_CI : 0), sdzp, 4016 sdl->dl_name, tdzp, tdl->dl_name, szp); 4017 4018 /* 4019 * Update path information for the target vnode 4020 */ 4021 vn_renamepath(tdvp, ZTOV(szp), tnm, 4022 strlen(tnm)); 4023 } else { 4024 /* 4025 * At this point, we have successfully created 4026 * the target name, but have failed to remove 4027 * the source name. Since the create was done 4028 * with the ZRENAMING flag, there are 4029 * complications; for one, the link count is 4030 * wrong. The easiest way to deal with this 4031 * is to remove the newly created target, and 4032 * return the original error. This must 4033 * succeed; fortunately, it is very unlikely to 4034 * fail, since we just created it. 4035 */ 4036 VERIFY3U(zfs_link_destroy(tdl, szp, tx, 4037 ZRENAMING, NULL), ==, 0); 4038 } 4039 } 4040#ifdef FREEBSD_NAMECACHE 4041 if (error == 0) { 4042 cache_purge(sdvp); 4043 cache_purge(tdvp); 4044 cache_purge(ZTOV(szp)); 4045 if (tzp) 4046 cache_purge(ZTOV(tzp)); 4047 } 4048#endif 4049 } 4050 4051 dmu_tx_commit(tx); 4052out: 4053 if (zl != NULL) 4054 zfs_rename_unlock(&zl); 4055 4056 zfs_dirent_unlock(sdl); 4057 zfs_dirent_unlock(tdl); 4058 4059 if (sdzp == tdzp) 4060 rw_exit(&sdzp->z_name_lock); 4061 4062 4063 VN_RELE(ZTOV(szp)); 4064 if (tzp) 4065 VN_RELE(ZTOV(tzp)); 4066 4067 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4068 zil_commit(zilog, 0); 4069 4070 ZFS_EXIT(zfsvfs); 4071 4072 return (error); 4073} 4074 4075/* 4076 * Insert the indicated symbolic reference entry into the directory. 4077 * 4078 * IN: dvp - Directory to contain new symbolic link. 4079 * link - Name for new symlink entry. 4080 * vap - Attributes of new entry. 4081 * cr - credentials of caller. 4082 * ct - caller context 4083 * flags - case flags 4084 * 4085 * RETURN: 0 on success, error code on failure. 4086 * 4087 * Timestamps: 4088 * dvp - ctime|mtime updated 4089 */ 4090/*ARGSUSED*/ 4091static int 4092zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, 4093 cred_t *cr, kthread_t *td) 4094{ 4095 znode_t *zp, *dzp = VTOZ(dvp); 4096 zfs_dirlock_t *dl; 4097 dmu_tx_t *tx; 4098 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4099 zilog_t *zilog; 4100 uint64_t len = strlen(link); 4101 int error; 4102 int zflg = ZNEW; 4103 zfs_acl_ids_t acl_ids; 4104 boolean_t fuid_dirtied; 4105 uint64_t txtype = TX_SYMLINK; 4106 int flags = 0; 4107 4108 ASSERT(vap->va_type == VLNK); 4109 4110 ZFS_ENTER(zfsvfs); 4111 ZFS_VERIFY_ZP(dzp); 4112 zilog = zfsvfs->z_log; 4113 4114 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 4115 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4116 ZFS_EXIT(zfsvfs); 4117 return (SET_ERROR(EILSEQ)); 4118 } 4119 if (flags & FIGNORECASE) 4120 zflg |= ZCILOOK; 4121 4122 if (len > MAXPATHLEN) { 4123 ZFS_EXIT(zfsvfs); 4124 return (SET_ERROR(ENAMETOOLONG)); 4125 } 4126 4127 if ((error = zfs_acl_ids_create(dzp, 0, 4128 vap, cr, NULL, &acl_ids)) != 0) { 4129 ZFS_EXIT(zfsvfs); 4130 return (error); 4131 } 4132top: 4133 /* 4134 * Attempt to lock directory; fail if entry already exists. 4135 */ 4136 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 4137 if (error) { 4138 zfs_acl_ids_free(&acl_ids); 4139 ZFS_EXIT(zfsvfs); 4140 return (error); 4141 } 4142 4143 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4144 zfs_acl_ids_free(&acl_ids); 4145 zfs_dirent_unlock(dl); 4146 ZFS_EXIT(zfsvfs); 4147 return (error); 4148 } 4149 4150 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 4151 zfs_acl_ids_free(&acl_ids); 4152 zfs_dirent_unlock(dl); 4153 ZFS_EXIT(zfsvfs); 4154 return (SET_ERROR(EDQUOT)); 4155 } 4156 tx = dmu_tx_create(zfsvfs->z_os); 4157 fuid_dirtied = zfsvfs->z_fuid_dirty; 4158 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 4159 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4160 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 4161 ZFS_SA_BASE_ATTR_SIZE + len); 4162 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 4163 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 4164 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 4165 acl_ids.z_aclp->z_acl_bytes); 4166 } 4167 if (fuid_dirtied) 4168 zfs_fuid_txhold(zfsvfs, tx); 4169 error = dmu_tx_assign(tx, TXG_NOWAIT); 4170 if (error) { 4171 zfs_dirent_unlock(dl); 4172 if (error == ERESTART) { 4173 dmu_tx_wait(tx); 4174 dmu_tx_abort(tx); 4175 goto top; 4176 } 4177 zfs_acl_ids_free(&acl_ids); 4178 dmu_tx_abort(tx); 4179 ZFS_EXIT(zfsvfs); 4180 return (error); 4181 } 4182 4183 /* 4184 * Create a new object for the symlink. 4185 * for version 4 ZPL datsets the symlink will be an SA attribute 4186 */ 4187 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 4188 4189 if (fuid_dirtied) 4190 zfs_fuid_sync(zfsvfs, tx); 4191 4192 mutex_enter(&zp->z_lock); 4193 if (zp->z_is_sa) 4194 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 4195 link, len, tx); 4196 else 4197 zfs_sa_symlink(zp, link, len, tx); 4198 mutex_exit(&zp->z_lock); 4199 4200 zp->z_size = len; 4201 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 4202 &zp->z_size, sizeof (zp->z_size), tx); 4203 /* 4204 * Insert the new object into the directory. 4205 */ 4206 (void) zfs_link_create(dl, zp, tx, ZNEW); 4207 4208 if (flags & FIGNORECASE) 4209 txtype |= TX_CI; 4210 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 4211 *vpp = ZTOV(zp); 4212 4213 zfs_acl_ids_free(&acl_ids); 4214 4215 dmu_tx_commit(tx); 4216 4217 zfs_dirent_unlock(dl); 4218 4219 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4220 zil_commit(zilog, 0); 4221 4222 ZFS_EXIT(zfsvfs); 4223 return (error); 4224} 4225 4226/* 4227 * Return, in the buffer contained in the provided uio structure, 4228 * the symbolic path referred to by vp. 4229 * 4230 * IN: vp - vnode of symbolic link. 4231 * uio - structure to contain the link path. 4232 * cr - credentials of caller. 4233 * ct - caller context 4234 * 4235 * OUT: uio - structure containing the link path. 4236 * 4237 * RETURN: 0 on success, error code on failure. 4238 * 4239 * Timestamps: 4240 * vp - atime updated 4241 */ 4242/* ARGSUSED */ 4243static int 4244zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 4245{ 4246 znode_t *zp = VTOZ(vp); 4247 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4248 int error; 4249 4250 ZFS_ENTER(zfsvfs); 4251 ZFS_VERIFY_ZP(zp); 4252 4253 mutex_enter(&zp->z_lock); 4254 if (zp->z_is_sa) 4255 error = sa_lookup_uio(zp->z_sa_hdl, 4256 SA_ZPL_SYMLINK(zfsvfs), uio); 4257 else 4258 error = zfs_sa_readlink(zp, uio); 4259 mutex_exit(&zp->z_lock); 4260 4261 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4262 4263 ZFS_EXIT(zfsvfs); 4264 return (error); 4265} 4266 4267/* 4268 * Insert a new entry into directory tdvp referencing svp. 4269 * 4270 * IN: tdvp - Directory to contain new entry. 4271 * svp - vnode of new entry. 4272 * name - name of new entry. 4273 * cr - credentials of caller. 4274 * ct - caller context 4275 * 4276 * RETURN: 0 on success, error code on failure. 4277 * 4278 * Timestamps: 4279 * tdvp - ctime|mtime updated 4280 * svp - ctime updated 4281 */ 4282/* ARGSUSED */ 4283static int 4284zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 4285 caller_context_t *ct, int flags) 4286{ 4287 znode_t *dzp = VTOZ(tdvp); 4288 znode_t *tzp, *szp; 4289 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4290 zilog_t *zilog; 4291 zfs_dirlock_t *dl; 4292 dmu_tx_t *tx; 4293 vnode_t *realvp; 4294 int error; 4295 int zf = ZNEW; 4296 uint64_t parent; 4297 uid_t owner; 4298 4299 ASSERT(tdvp->v_type == VDIR); 4300 4301 ZFS_ENTER(zfsvfs); 4302 ZFS_VERIFY_ZP(dzp); 4303 zilog = zfsvfs->z_log; 4304 4305 if (VOP_REALVP(svp, &realvp, ct) == 0) 4306 svp = realvp; 4307 4308 /* 4309 * POSIX dictates that we return EPERM here. 4310 * Better choices include ENOTSUP or EISDIR. 4311 */ 4312 if (svp->v_type == VDIR) { 4313 ZFS_EXIT(zfsvfs); 4314 return (SET_ERROR(EPERM)); 4315 } 4316 4317 szp = VTOZ(svp); 4318 ZFS_VERIFY_ZP(szp); 4319 4320 /* 4321 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 4322 * ctldir appear to have the same v_vfsp. 4323 */ 4324 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) { 4325 ZFS_EXIT(zfsvfs); 4326 return (SET_ERROR(EXDEV)); 4327 } 4328 4329 /* Prevent links to .zfs/shares files */ 4330 4331 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 4332 &parent, sizeof (uint64_t))) != 0) { 4333 ZFS_EXIT(zfsvfs); 4334 return (error); 4335 } 4336 if (parent == zfsvfs->z_shares_dir) { 4337 ZFS_EXIT(zfsvfs); 4338 return (SET_ERROR(EPERM)); 4339 } 4340 4341 if (zfsvfs->z_utf8 && u8_validate(name, 4342 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4343 ZFS_EXIT(zfsvfs); 4344 return (SET_ERROR(EILSEQ)); 4345 } 4346 if (flags & FIGNORECASE) 4347 zf |= ZCILOOK; 4348 4349 /* 4350 * We do not support links between attributes and non-attributes 4351 * because of the potential security risk of creating links 4352 * into "normal" file space in order to circumvent restrictions 4353 * imposed in attribute space. 4354 */ 4355 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 4356 ZFS_EXIT(zfsvfs); 4357 return (SET_ERROR(EINVAL)); 4358 } 4359 4360 4361 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 4362 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) { 4363 ZFS_EXIT(zfsvfs); 4364 return (SET_ERROR(EPERM)); 4365 } 4366 4367 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4368 ZFS_EXIT(zfsvfs); 4369 return (error); 4370 } 4371 4372top: 4373 /* 4374 * Attempt to lock directory; fail if entry already exists. 4375 */ 4376 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 4377 if (error) { 4378 ZFS_EXIT(zfsvfs); 4379 return (error); 4380 } 4381 4382 tx = dmu_tx_create(zfsvfs->z_os); 4383 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4384 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4385 zfs_sa_upgrade_txholds(tx, szp); 4386 zfs_sa_upgrade_txholds(tx, dzp); 4387 error = dmu_tx_assign(tx, TXG_NOWAIT); 4388 if (error) { 4389 zfs_dirent_unlock(dl); 4390 if (error == ERESTART) { 4391 dmu_tx_wait(tx); 4392 dmu_tx_abort(tx); 4393 goto top; 4394 } 4395 dmu_tx_abort(tx); 4396 ZFS_EXIT(zfsvfs); 4397 return (error); 4398 } 4399 4400 error = zfs_link_create(dl, szp, tx, 0); 4401 4402 if (error == 0) { 4403 uint64_t txtype = TX_LINK; 4404 if (flags & FIGNORECASE) 4405 txtype |= TX_CI; 4406 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 4407 } 4408 4409 dmu_tx_commit(tx); 4410 4411 zfs_dirent_unlock(dl); 4412 4413 if (error == 0) { 4414 vnevent_link(svp, ct); 4415 } 4416 4417 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4418 zil_commit(zilog, 0); 4419 4420 ZFS_EXIT(zfsvfs); 4421 return (error); 4422} 4423 4424#ifdef sun 4425/* 4426 * zfs_null_putapage() is used when the file system has been force 4427 * unmounted. It just drops the pages. 4428 */ 4429/* ARGSUSED */ 4430static int 4431zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4432 size_t *lenp, int flags, cred_t *cr) 4433{ 4434 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 4435 return (0); 4436} 4437 4438/* 4439 * Push a page out to disk, klustering if possible. 4440 * 4441 * IN: vp - file to push page to. 4442 * pp - page to push. 4443 * flags - additional flags. 4444 * cr - credentials of caller. 4445 * 4446 * OUT: offp - start of range pushed. 4447 * lenp - len of range pushed. 4448 * 4449 * RETURN: 0 on success, error code on failure. 4450 * 4451 * NOTE: callers must have locked the page to be pushed. On 4452 * exit, the page (and all other pages in the kluster) must be 4453 * unlocked. 4454 */ 4455/* ARGSUSED */ 4456static int 4457zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4458 size_t *lenp, int flags, cred_t *cr) 4459{ 4460 znode_t *zp = VTOZ(vp); 4461 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4462 dmu_tx_t *tx; 4463 u_offset_t off, koff; 4464 size_t len, klen; 4465 int err; 4466 4467 off = pp->p_offset; 4468 len = PAGESIZE; 4469 /* 4470 * If our blocksize is bigger than the page size, try to kluster 4471 * multiple pages so that we write a full block (thus avoiding 4472 * a read-modify-write). 4473 */ 4474 if (off < zp->z_size && zp->z_blksz > PAGESIZE) { 4475 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 4476 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 4477 ASSERT(koff <= zp->z_size); 4478 if (koff + klen > zp->z_size) 4479 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); 4480 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 4481 } 4482 ASSERT3U(btop(len), ==, btopr(len)); 4483 4484 /* 4485 * Can't push pages past end-of-file. 4486 */ 4487 if (off >= zp->z_size) { 4488 /* ignore all pages */ 4489 err = 0; 4490 goto out; 4491 } else if (off + len > zp->z_size) { 4492 int npages = btopr(zp->z_size - off); 4493 page_t *trunc; 4494 4495 page_list_break(&pp, &trunc, npages); 4496 /* ignore pages past end of file */ 4497 if (trunc) 4498 pvn_write_done(trunc, flags); 4499 len = zp->z_size - off; 4500 } 4501 4502 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4503 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4504 err = SET_ERROR(EDQUOT); 4505 goto out; 4506 } 4507top: 4508 tx = dmu_tx_create(zfsvfs->z_os); 4509 dmu_tx_hold_write(tx, zp->z_id, off, len); 4510 4511 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4512 zfs_sa_upgrade_txholds(tx, zp); 4513 err = dmu_tx_assign(tx, TXG_NOWAIT); 4514 if (err != 0) { 4515 if (err == ERESTART) { 4516 dmu_tx_wait(tx); 4517 dmu_tx_abort(tx); 4518 goto top; 4519 } 4520 dmu_tx_abort(tx); 4521 goto out; 4522 } 4523 4524 if (zp->z_blksz <= PAGESIZE) { 4525 caddr_t va = zfs_map_page(pp, S_READ); 4526 ASSERT3U(len, <=, PAGESIZE); 4527 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 4528 zfs_unmap_page(pp, va); 4529 } else { 4530 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 4531 } 4532 4533 if (err == 0) { 4534 uint64_t mtime[2], ctime[2]; 4535 sa_bulk_attr_t bulk[3]; 4536 int count = 0; 4537 4538 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4539 &mtime, 16); 4540 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4541 &ctime, 16); 4542 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4543 &zp->z_pflags, 8); 4544 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4545 B_TRUE); 4546 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4547 } 4548 dmu_tx_commit(tx); 4549 4550out: 4551 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 4552 if (offp) 4553 *offp = off; 4554 if (lenp) 4555 *lenp = len; 4556 4557 return (err); 4558} 4559 4560/* 4561 * Copy the portion of the file indicated from pages into the file. 4562 * The pages are stored in a page list attached to the files vnode. 4563 * 4564 * IN: vp - vnode of file to push page data to. 4565 * off - position in file to put data. 4566 * len - amount of data to write. 4567 * flags - flags to control the operation. 4568 * cr - credentials of caller. 4569 * ct - caller context. 4570 * 4571 * RETURN: 0 on success, error code on failure. 4572 * 4573 * Timestamps: 4574 * vp - ctime|mtime updated 4575 */ 4576/*ARGSUSED*/ 4577static int 4578zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4579 caller_context_t *ct) 4580{ 4581 znode_t *zp = VTOZ(vp); 4582 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4583 page_t *pp; 4584 size_t io_len; 4585 u_offset_t io_off; 4586 uint_t blksz; 4587 rl_t *rl; 4588 int error = 0; 4589 4590 ZFS_ENTER(zfsvfs); 4591 ZFS_VERIFY_ZP(zp); 4592 4593 /* 4594 * Align this request to the file block size in case we kluster. 4595 * XXX - this can result in pretty aggresive locking, which can 4596 * impact simultanious read/write access. One option might be 4597 * to break up long requests (len == 0) into block-by-block 4598 * operations to get narrower locking. 4599 */ 4600 blksz = zp->z_blksz; 4601 if (ISP2(blksz)) 4602 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 4603 else 4604 io_off = 0; 4605 if (len > 0 && ISP2(blksz)) 4606 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 4607 else 4608 io_len = 0; 4609 4610 if (io_len == 0) { 4611 /* 4612 * Search the entire vp list for pages >= io_off. 4613 */ 4614 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 4615 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 4616 goto out; 4617 } 4618 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 4619 4620 if (off > zp->z_size) { 4621 /* past end of file */ 4622 zfs_range_unlock(rl); 4623 ZFS_EXIT(zfsvfs); 4624 return (0); 4625 } 4626 4627 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); 4628 4629 for (off = io_off; io_off < off + len; io_off += io_len) { 4630 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 4631 pp = page_lookup(vp, io_off, 4632 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 4633 } else { 4634 pp = page_lookup_nowait(vp, io_off, 4635 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 4636 } 4637 4638 if (pp != NULL && pvn_getdirty(pp, flags)) { 4639 int err; 4640 4641 /* 4642 * Found a dirty page to push 4643 */ 4644 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 4645 if (err) 4646 error = err; 4647 } else { 4648 io_len = PAGESIZE; 4649 } 4650 } 4651out: 4652 zfs_range_unlock(rl); 4653 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4654 zil_commit(zfsvfs->z_log, zp->z_id); 4655 ZFS_EXIT(zfsvfs); 4656 return (error); 4657} 4658#endif /* sun */ 4659 4660/*ARGSUSED*/ 4661void 4662zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4663{ 4664 znode_t *zp = VTOZ(vp); 4665 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4666 int error; 4667 4668 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4669 if (zp->z_sa_hdl == NULL) { 4670 /* 4671 * The fs has been unmounted, or we did a 4672 * suspend/resume and this file no longer exists. 4673 */ 4674 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4675 vrecycle(vp); 4676 return; 4677 } 4678 4679 mutex_enter(&zp->z_lock); 4680 if (zp->z_unlinked) { 4681 /* 4682 * Fast path to recycle a vnode of a removed file. 4683 */ 4684 mutex_exit(&zp->z_lock); 4685 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4686 vrecycle(vp); 4687 return; 4688 } 4689 mutex_exit(&zp->z_lock); 4690 4691 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4692 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4693 4694 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4695 zfs_sa_upgrade_txholds(tx, zp); 4696 error = dmu_tx_assign(tx, TXG_WAIT); 4697 if (error) { 4698 dmu_tx_abort(tx); 4699 } else { 4700 mutex_enter(&zp->z_lock); 4701 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4702 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4703 zp->z_atime_dirty = 0; 4704 mutex_exit(&zp->z_lock); 4705 dmu_tx_commit(tx); 4706 } 4707 } 4708 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4709} 4710 4711#ifdef sun 4712/* 4713 * Bounds-check the seek operation. 4714 * 4715 * IN: vp - vnode seeking within 4716 * ooff - old file offset 4717 * noffp - pointer to new file offset 4718 * ct - caller context 4719 * 4720 * RETURN: 0 on success, EINVAL if new offset invalid. 4721 */ 4722/* ARGSUSED */ 4723static int 4724zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 4725 caller_context_t *ct) 4726{ 4727 if (vp->v_type == VDIR) 4728 return (0); 4729 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4730} 4731 4732/* 4733 * Pre-filter the generic locking function to trap attempts to place 4734 * a mandatory lock on a memory mapped file. 4735 */ 4736static int 4737zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 4738 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 4739{ 4740 znode_t *zp = VTOZ(vp); 4741 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4742 4743 ZFS_ENTER(zfsvfs); 4744 ZFS_VERIFY_ZP(zp); 4745 4746 /* 4747 * We are following the UFS semantics with respect to mapcnt 4748 * here: If we see that the file is mapped already, then we will 4749 * return an error, but we don't worry about races between this 4750 * function and zfs_map(). 4751 */ 4752 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { 4753 ZFS_EXIT(zfsvfs); 4754 return (SET_ERROR(EAGAIN)); 4755 } 4756 ZFS_EXIT(zfsvfs); 4757 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4758} 4759 4760/* 4761 * If we can't find a page in the cache, we will create a new page 4762 * and fill it with file data. For efficiency, we may try to fill 4763 * multiple pages at once (klustering) to fill up the supplied page 4764 * list. Note that the pages to be filled are held with an exclusive 4765 * lock to prevent access by other threads while they are being filled. 4766 */ 4767static int 4768zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 4769 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 4770{ 4771 znode_t *zp = VTOZ(vp); 4772 page_t *pp, *cur_pp; 4773 objset_t *os = zp->z_zfsvfs->z_os; 4774 u_offset_t io_off, total; 4775 size_t io_len; 4776 int err; 4777 4778 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4779 /* 4780 * We only have a single page, don't bother klustering 4781 */ 4782 io_off = off; 4783 io_len = PAGESIZE; 4784 pp = page_create_va(vp, io_off, io_len, 4785 PG_EXCL | PG_WAIT, seg, addr); 4786 } else { 4787 /* 4788 * Try to find enough pages to fill the page list 4789 */ 4790 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4791 &io_len, off, plsz, 0); 4792 } 4793 if (pp == NULL) { 4794 /* 4795 * The page already exists, nothing to do here. 4796 */ 4797 *pl = NULL; 4798 return (0); 4799 } 4800 4801 /* 4802 * Fill the pages in the kluster. 4803 */ 4804 cur_pp = pp; 4805 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4806 caddr_t va; 4807 4808 ASSERT3U(io_off, ==, cur_pp->p_offset); 4809 va = zfs_map_page(cur_pp, S_WRITE); 4810 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4811 DMU_READ_PREFETCH); 4812 zfs_unmap_page(cur_pp, va); 4813 if (err) { 4814 /* On error, toss the entire kluster */ 4815 pvn_read_done(pp, B_ERROR); 4816 /* convert checksum errors into IO errors */ 4817 if (err == ECKSUM) 4818 err = SET_ERROR(EIO); 4819 return (err); 4820 } 4821 cur_pp = cur_pp->p_next; 4822 } 4823 4824 /* 4825 * Fill in the page list array from the kluster starting 4826 * from the desired offset `off'. 4827 * NOTE: the page list will always be null terminated. 4828 */ 4829 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4830 ASSERT(pl == NULL || (*pl)->p_offset == off); 4831 4832 return (0); 4833} 4834 4835/* 4836 * Return pointers to the pages for the file region [off, off + len] 4837 * in the pl array. If plsz is greater than len, this function may 4838 * also return page pointers from after the specified region 4839 * (i.e. the region [off, off + plsz]). These additional pages are 4840 * only returned if they are already in the cache, or were created as 4841 * part of a klustered read. 4842 * 4843 * IN: vp - vnode of file to get data from. 4844 * off - position in file to get data from. 4845 * len - amount of data to retrieve. 4846 * plsz - length of provided page list. 4847 * seg - segment to obtain pages for. 4848 * addr - virtual address of fault. 4849 * rw - mode of created pages. 4850 * cr - credentials of caller. 4851 * ct - caller context. 4852 * 4853 * OUT: protp - protection mode of created pages. 4854 * pl - list of pages created. 4855 * 4856 * RETURN: 0 on success, error code on failure. 4857 * 4858 * Timestamps: 4859 * vp - atime updated 4860 */ 4861/* ARGSUSED */ 4862static int 4863zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4864 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4865 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4866{ 4867 znode_t *zp = VTOZ(vp); 4868 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4869 page_t **pl0 = pl; 4870 int err = 0; 4871 4872 /* we do our own caching, faultahead is unnecessary */ 4873 if (pl == NULL) 4874 return (0); 4875 else if (len > plsz) 4876 len = plsz; 4877 else 4878 len = P2ROUNDUP(len, PAGESIZE); 4879 ASSERT(plsz >= len); 4880 4881 ZFS_ENTER(zfsvfs); 4882 ZFS_VERIFY_ZP(zp); 4883 4884 if (protp) 4885 *protp = PROT_ALL; 4886 4887 /* 4888 * Loop through the requested range [off, off + len) looking 4889 * for pages. If we don't find a page, we will need to create 4890 * a new page and fill it with data from the file. 4891 */ 4892 while (len > 0) { 4893 if (*pl = page_lookup(vp, off, SE_SHARED)) 4894 *(pl+1) = NULL; 4895 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 4896 goto out; 4897 while (*pl) { 4898 ASSERT3U((*pl)->p_offset, ==, off); 4899 off += PAGESIZE; 4900 addr += PAGESIZE; 4901 if (len > 0) { 4902 ASSERT3U(len, >=, PAGESIZE); 4903 len -= PAGESIZE; 4904 } 4905 ASSERT3U(plsz, >=, PAGESIZE); 4906 plsz -= PAGESIZE; 4907 pl++; 4908 } 4909 } 4910 4911 /* 4912 * Fill out the page array with any pages already in the cache. 4913 */ 4914 while (plsz > 0 && 4915 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 4916 off += PAGESIZE; 4917 plsz -= PAGESIZE; 4918 } 4919out: 4920 if (err) { 4921 /* 4922 * Release any pages we have previously locked. 4923 */ 4924 while (pl > pl0) 4925 page_unlock(*--pl); 4926 } else { 4927 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4928 } 4929 4930 *pl = NULL; 4931 4932 ZFS_EXIT(zfsvfs); 4933 return (err); 4934} 4935 4936/* 4937 * Request a memory map for a section of a file. This code interacts 4938 * with common code and the VM system as follows: 4939 * 4940 * - common code calls mmap(), which ends up in smmap_common() 4941 * - this calls VOP_MAP(), which takes you into (say) zfs 4942 * - zfs_map() calls as_map(), passing segvn_create() as the callback 4943 * - segvn_create() creates the new segment and calls VOP_ADDMAP() 4944 * - zfs_addmap() updates z_mapcnt 4945 */ 4946/*ARGSUSED*/ 4947static int 4948zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4949 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4950 caller_context_t *ct) 4951{ 4952 znode_t *zp = VTOZ(vp); 4953 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4954 segvn_crargs_t vn_a; 4955 int error; 4956 4957 ZFS_ENTER(zfsvfs); 4958 ZFS_VERIFY_ZP(zp); 4959 4960 if ((prot & PROT_WRITE) && (zp->z_pflags & 4961 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 4962 ZFS_EXIT(zfsvfs); 4963 return (SET_ERROR(EPERM)); 4964 } 4965 4966 if ((prot & (PROT_READ | PROT_EXEC)) && 4967 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 4968 ZFS_EXIT(zfsvfs); 4969 return (SET_ERROR(EACCES)); 4970 } 4971 4972 if (vp->v_flag & VNOMAP) { 4973 ZFS_EXIT(zfsvfs); 4974 return (SET_ERROR(ENOSYS)); 4975 } 4976 4977 if (off < 0 || len > MAXOFFSET_T - off) { 4978 ZFS_EXIT(zfsvfs); 4979 return (SET_ERROR(ENXIO)); 4980 } 4981 4982 if (vp->v_type != VREG) { 4983 ZFS_EXIT(zfsvfs); 4984 return (SET_ERROR(ENODEV)); 4985 } 4986 4987 /* 4988 * If file is locked, disallow mapping. 4989 */ 4990 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { 4991 ZFS_EXIT(zfsvfs); 4992 return (SET_ERROR(EAGAIN)); 4993 } 4994 4995 as_rangelock(as); 4996 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 4997 if (error != 0) { 4998 as_rangeunlock(as); 4999 ZFS_EXIT(zfsvfs); 5000 return (error); 5001 } 5002 5003 vn_a.vp = vp; 5004 vn_a.offset = (u_offset_t)off; 5005 vn_a.type = flags & MAP_TYPE; 5006 vn_a.prot = prot; 5007 vn_a.maxprot = maxprot; 5008 vn_a.cred = cr; 5009 vn_a.amp = NULL; 5010 vn_a.flags = flags & ~MAP_TYPE; 5011 vn_a.szc = 0; 5012 vn_a.lgrp_mem_policy_flags = 0; 5013 5014 error = as_map(as, *addrp, len, segvn_create, &vn_a); 5015 5016 as_rangeunlock(as); 5017 ZFS_EXIT(zfsvfs); 5018 return (error); 5019} 5020 5021/* ARGSUSED */ 5022static int 5023zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5024 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 5025 caller_context_t *ct) 5026{ 5027 uint64_t pages = btopr(len); 5028 5029 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 5030 return (0); 5031} 5032 5033/* 5034 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 5035 * more accurate mtime for the associated file. Since we don't have a way of 5036 * detecting when the data was actually modified, we have to resort to 5037 * heuristics. If an explicit msync() is done, then we mark the mtime when the 5038 * last page is pushed. The problem occurs when the msync() call is omitted, 5039 * which by far the most common case: 5040 * 5041 * open() 5042 * mmap() 5043 * <modify memory> 5044 * munmap() 5045 * close() 5046 * <time lapse> 5047 * putpage() via fsflush 5048 * 5049 * If we wait until fsflush to come along, we can have a modification time that 5050 * is some arbitrary point in the future. In order to prevent this in the 5051 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 5052 * torn down. 5053 */ 5054/* ARGSUSED */ 5055static int 5056zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5057 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 5058 caller_context_t *ct) 5059{ 5060 uint64_t pages = btopr(len); 5061 5062 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 5063 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 5064 5065 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 5066 vn_has_cached_data(vp)) 5067 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 5068 5069 return (0); 5070} 5071 5072/* 5073 * Free or allocate space in a file. Currently, this function only 5074 * supports the `F_FREESP' command. However, this command is somewhat 5075 * misnamed, as its functionality includes the ability to allocate as 5076 * well as free space. 5077 * 5078 * IN: vp - vnode of file to free data in. 5079 * cmd - action to take (only F_FREESP supported). 5080 * bfp - section of file to free/alloc. 5081 * flag - current file open mode flags. 5082 * offset - current file offset. 5083 * cr - credentials of caller [UNUSED]. 5084 * ct - caller context. 5085 * 5086 * RETURN: 0 on success, error code on failure. 5087 * 5088 * Timestamps: 5089 * vp - ctime|mtime updated 5090 */ 5091/* ARGSUSED */ 5092static int 5093zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 5094 offset_t offset, cred_t *cr, caller_context_t *ct) 5095{ 5096 znode_t *zp = VTOZ(vp); 5097 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5098 uint64_t off, len; 5099 int error; 5100 5101 ZFS_ENTER(zfsvfs); 5102 ZFS_VERIFY_ZP(zp); 5103 5104 if (cmd != F_FREESP) { 5105 ZFS_EXIT(zfsvfs); 5106 return (SET_ERROR(EINVAL)); 5107 } 5108 5109 if (error = convoff(vp, bfp, 0, offset)) { 5110 ZFS_EXIT(zfsvfs); 5111 return (error); 5112 } 5113 5114 if (bfp->l_len < 0) { 5115 ZFS_EXIT(zfsvfs); 5116 return (SET_ERROR(EINVAL)); 5117 } 5118 5119 off = bfp->l_start; 5120 len = bfp->l_len; /* 0 means from off to end of file */ 5121 5122 error = zfs_freesp(zp, off, len, flag, TRUE); 5123 5124 ZFS_EXIT(zfsvfs); 5125 return (error); 5126} 5127#endif /* sun */ 5128 5129CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 5130CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 5131 5132/*ARGSUSED*/ 5133static int 5134zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 5135{ 5136 znode_t *zp = VTOZ(vp); 5137 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5138 uint32_t gen; 5139 uint64_t gen64; 5140 uint64_t object = zp->z_id; 5141 zfid_short_t *zfid; 5142 int size, i, error; 5143 5144 ZFS_ENTER(zfsvfs); 5145 ZFS_VERIFY_ZP(zp); 5146 5147 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 5148 &gen64, sizeof (uint64_t))) != 0) { 5149 ZFS_EXIT(zfsvfs); 5150 return (error); 5151 } 5152 5153 gen = (uint32_t)gen64; 5154 5155 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 5156 5157#ifdef illumos 5158 if (fidp->fid_len < size) { 5159 fidp->fid_len = size; 5160 ZFS_EXIT(zfsvfs); 5161 return (SET_ERROR(ENOSPC)); 5162 } 5163#else 5164 fidp->fid_len = size; 5165#endif 5166 5167 zfid = (zfid_short_t *)fidp; 5168 5169 zfid->zf_len = size; 5170 5171 for (i = 0; i < sizeof (zfid->zf_object); i++) 5172 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 5173 5174 /* Must have a non-zero generation number to distinguish from .zfs */ 5175 if (gen == 0) 5176 gen = 1; 5177 for (i = 0; i < sizeof (zfid->zf_gen); i++) 5178 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 5179 5180 if (size == LONG_FID_LEN) { 5181 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 5182 zfid_long_t *zlfid; 5183 5184 zlfid = (zfid_long_t *)fidp; 5185 5186 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 5187 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 5188 5189 /* XXX - this should be the generation number for the objset */ 5190 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 5191 zlfid->zf_setgen[i] = 0; 5192 } 5193 5194 ZFS_EXIT(zfsvfs); 5195 return (0); 5196} 5197 5198static int 5199zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 5200 caller_context_t *ct) 5201{ 5202 znode_t *zp, *xzp; 5203 zfsvfs_t *zfsvfs; 5204 zfs_dirlock_t *dl; 5205 int error; 5206 5207 switch (cmd) { 5208 case _PC_LINK_MAX: 5209 *valp = INT_MAX; 5210 return (0); 5211 5212 case _PC_FILESIZEBITS: 5213 *valp = 64; 5214 return (0); 5215#ifdef sun 5216 case _PC_XATTR_EXISTS: 5217 zp = VTOZ(vp); 5218 zfsvfs = zp->z_zfsvfs; 5219 ZFS_ENTER(zfsvfs); 5220 ZFS_VERIFY_ZP(zp); 5221 *valp = 0; 5222 error = zfs_dirent_lock(&dl, zp, "", &xzp, 5223 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 5224 if (error == 0) { 5225 zfs_dirent_unlock(dl); 5226 if (!zfs_dirempty(xzp)) 5227 *valp = 1; 5228 VN_RELE(ZTOV(xzp)); 5229 } else if (error == ENOENT) { 5230 /* 5231 * If there aren't extended attributes, it's the 5232 * same as having zero of them. 5233 */ 5234 error = 0; 5235 } 5236 ZFS_EXIT(zfsvfs); 5237 return (error); 5238 5239 case _PC_SATTR_ENABLED: 5240 case _PC_SATTR_EXISTS: 5241 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 5242 (vp->v_type == VREG || vp->v_type == VDIR); 5243 return (0); 5244 5245 case _PC_ACCESS_FILTERING: 5246 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 5247 vp->v_type == VDIR; 5248 return (0); 5249 5250 case _PC_ACL_ENABLED: 5251 *valp = _ACL_ACE_ENABLED; 5252 return (0); 5253#endif /* sun */ 5254 case _PC_MIN_HOLE_SIZE: 5255 *valp = (int)SPA_MINBLOCKSIZE; 5256 return (0); 5257#ifdef sun 5258 case _PC_TIMESTAMP_RESOLUTION: 5259 /* nanosecond timestamp resolution */ 5260 *valp = 1L; 5261 return (0); 5262#endif /* sun */ 5263 case _PC_ACL_EXTENDED: 5264 *valp = 0; 5265 return (0); 5266 5267 case _PC_ACL_NFS4: 5268 *valp = 1; 5269 return (0); 5270 5271 case _PC_ACL_PATH_MAX: 5272 *valp = ACL_MAX_ENTRIES; 5273 return (0); 5274 5275 default: 5276 return (EOPNOTSUPP); 5277 } 5278} 5279 5280/*ARGSUSED*/ 5281static int 5282zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5283 caller_context_t *ct) 5284{ 5285 znode_t *zp = VTOZ(vp); 5286 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5287 int error; 5288 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5289 5290 ZFS_ENTER(zfsvfs); 5291 ZFS_VERIFY_ZP(zp); 5292 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 5293 ZFS_EXIT(zfsvfs); 5294 5295 return (error); 5296} 5297 5298/*ARGSUSED*/ 5299int 5300zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5301 caller_context_t *ct) 5302{ 5303 znode_t *zp = VTOZ(vp); 5304 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5305 int error; 5306 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5307 zilog_t *zilog = zfsvfs->z_log; 5308 5309 ZFS_ENTER(zfsvfs); 5310 ZFS_VERIFY_ZP(zp); 5311 5312 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 5313 5314 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 5315 zil_commit(zilog, 0); 5316 5317 ZFS_EXIT(zfsvfs); 5318 return (error); 5319} 5320 5321#ifdef sun 5322/* 5323 * The smallest read we may consider to loan out an arcbuf. 5324 * This must be a power of 2. 5325 */ 5326int zcr_blksz_min = (1 << 10); /* 1K */ 5327/* 5328 * If set to less than the file block size, allow loaning out of an 5329 * arcbuf for a partial block read. This must be a power of 2. 5330 */ 5331int zcr_blksz_max = (1 << 17); /* 128K */ 5332 5333/*ARGSUSED*/ 5334static int 5335zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr, 5336 caller_context_t *ct) 5337{ 5338 znode_t *zp = VTOZ(vp); 5339 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5340 int max_blksz = zfsvfs->z_max_blksz; 5341 uio_t *uio = &xuio->xu_uio; 5342 ssize_t size = uio->uio_resid; 5343 offset_t offset = uio->uio_loffset; 5344 int blksz; 5345 int fullblk, i; 5346 arc_buf_t *abuf; 5347 ssize_t maxsize; 5348 int preamble, postamble; 5349 5350 if (xuio->xu_type != UIOTYPE_ZEROCOPY) 5351 return (SET_ERROR(EINVAL)); 5352 5353 ZFS_ENTER(zfsvfs); 5354 ZFS_VERIFY_ZP(zp); 5355 switch (ioflag) { 5356 case UIO_WRITE: 5357 /* 5358 * Loan out an arc_buf for write if write size is bigger than 5359 * max_blksz, and the file's block size is also max_blksz. 5360 */ 5361 blksz = max_blksz; 5362 if (size < blksz || zp->z_blksz != blksz) { 5363 ZFS_EXIT(zfsvfs); 5364 return (SET_ERROR(EINVAL)); 5365 } 5366 /* 5367 * Caller requests buffers for write before knowing where the 5368 * write offset might be (e.g. NFS TCP write). 5369 */ 5370 if (offset == -1) { 5371 preamble = 0; 5372 } else { 5373 preamble = P2PHASE(offset, blksz); 5374 if (preamble) { 5375 preamble = blksz - preamble; 5376 size -= preamble; 5377 } 5378 } 5379 5380 postamble = P2PHASE(size, blksz); 5381 size -= postamble; 5382 5383 fullblk = size / blksz; 5384 (void) dmu_xuio_init(xuio, 5385 (preamble != 0) + fullblk + (postamble != 0)); 5386 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble, 5387 int, postamble, int, 5388 (preamble != 0) + fullblk + (postamble != 0)); 5389 5390 /* 5391 * Have to fix iov base/len for partial buffers. They 5392 * currently represent full arc_buf's. 5393 */ 5394 if (preamble) { 5395 /* data begins in the middle of the arc_buf */ 5396 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5397 blksz); 5398 ASSERT(abuf); 5399 (void) dmu_xuio_add(xuio, abuf, 5400 blksz - preamble, preamble); 5401 } 5402 5403 for (i = 0; i < fullblk; i++) { 5404 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5405 blksz); 5406 ASSERT(abuf); 5407 (void) dmu_xuio_add(xuio, abuf, 0, blksz); 5408 } 5409 5410 if (postamble) { 5411 /* data ends in the middle of the arc_buf */ 5412 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5413 blksz); 5414 ASSERT(abuf); 5415 (void) dmu_xuio_add(xuio, abuf, 0, postamble); 5416 } 5417 break; 5418 case UIO_READ: 5419 /* 5420 * Loan out an arc_buf for read if the read size is larger than 5421 * the current file block size. Block alignment is not 5422 * considered. Partial arc_buf will be loaned out for read. 5423 */ 5424 blksz = zp->z_blksz; 5425 if (blksz < zcr_blksz_min) 5426 blksz = zcr_blksz_min; 5427 if (blksz > zcr_blksz_max) 5428 blksz = zcr_blksz_max; 5429 /* avoid potential complexity of dealing with it */ 5430 if (blksz > max_blksz) { 5431 ZFS_EXIT(zfsvfs); 5432 return (SET_ERROR(EINVAL)); 5433 } 5434 5435 maxsize = zp->z_size - uio->uio_loffset; 5436 if (size > maxsize) 5437 size = maxsize; 5438 5439 if (size < blksz || vn_has_cached_data(vp)) { 5440 ZFS_EXIT(zfsvfs); 5441 return (SET_ERROR(EINVAL)); 5442 } 5443 break; 5444 default: 5445 ZFS_EXIT(zfsvfs); 5446 return (SET_ERROR(EINVAL)); 5447 } 5448 5449 uio->uio_extflg = UIO_XUIO; 5450 XUIO_XUZC_RW(xuio) = ioflag; 5451 ZFS_EXIT(zfsvfs); 5452 return (0); 5453} 5454 5455/*ARGSUSED*/ 5456static int 5457zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct) 5458{ 5459 int i; 5460 arc_buf_t *abuf; 5461 int ioflag = XUIO_XUZC_RW(xuio); 5462 5463 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY); 5464 5465 i = dmu_xuio_cnt(xuio); 5466 while (i-- > 0) { 5467 abuf = dmu_xuio_arcbuf(xuio, i); 5468 /* 5469 * if abuf == NULL, it must be a write buffer 5470 * that has been returned in zfs_write(). 5471 */ 5472 if (abuf) 5473 dmu_return_arcbuf(abuf); 5474 ASSERT(abuf || ioflag == UIO_WRITE); 5475 } 5476 5477 dmu_xuio_fini(xuio); 5478 return (0); 5479} 5480 5481/* 5482 * Predeclare these here so that the compiler assumes that 5483 * this is an "old style" function declaration that does 5484 * not include arguments => we won't get type mismatch errors 5485 * in the initializations that follow. 5486 */ 5487static int zfs_inval(); 5488static int zfs_isdir(); 5489 5490static int 5491zfs_inval() 5492{ 5493 return (SET_ERROR(EINVAL)); 5494} 5495 5496static int 5497zfs_isdir() 5498{ 5499 return (SET_ERROR(EISDIR)); 5500} 5501/* 5502 * Directory vnode operations template 5503 */ 5504vnodeops_t *zfs_dvnodeops; 5505const fs_operation_def_t zfs_dvnodeops_template[] = { 5506 VOPNAME_OPEN, { .vop_open = zfs_open }, 5507 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5508 VOPNAME_READ, { .error = zfs_isdir }, 5509 VOPNAME_WRITE, { .error = zfs_isdir }, 5510 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5511 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5512 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5513 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5514 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5515 VOPNAME_CREATE, { .vop_create = zfs_create }, 5516 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5517 VOPNAME_LINK, { .vop_link = zfs_link }, 5518 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5519 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 5520 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5521 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5522 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 5523 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5524 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5525 VOPNAME_FID, { .vop_fid = zfs_fid }, 5526 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5527 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5528 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5529 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5530 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5531 NULL, NULL 5532}; 5533 5534/* 5535 * Regular file vnode operations template 5536 */ 5537vnodeops_t *zfs_fvnodeops; 5538const fs_operation_def_t zfs_fvnodeops_template[] = { 5539 VOPNAME_OPEN, { .vop_open = zfs_open }, 5540 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5541 VOPNAME_READ, { .vop_read = zfs_read }, 5542 VOPNAME_WRITE, { .vop_write = zfs_write }, 5543 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5544 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5545 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5546 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5547 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5548 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5549 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5550 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5551 VOPNAME_FID, { .vop_fid = zfs_fid }, 5552 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5553 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 5554 VOPNAME_SPACE, { .vop_space = zfs_space }, 5555 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 5556 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 5557 VOPNAME_MAP, { .vop_map = zfs_map }, 5558 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 5559 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 5560 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5561 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5562 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5563 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5564 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf }, 5565 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf }, 5566 NULL, NULL 5567}; 5568 5569/* 5570 * Symbolic link vnode operations template 5571 */ 5572vnodeops_t *zfs_symvnodeops; 5573const fs_operation_def_t zfs_symvnodeops_template[] = { 5574 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5575 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5576 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5577 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5578 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 5579 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5580 VOPNAME_FID, { .vop_fid = zfs_fid }, 5581 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5582 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5583 NULL, NULL 5584}; 5585 5586/* 5587 * special share hidden files vnode operations template 5588 */ 5589vnodeops_t *zfs_sharevnodeops; 5590const fs_operation_def_t zfs_sharevnodeops_template[] = { 5591 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5592 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5593 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5594 VOPNAME_FID, { .vop_fid = zfs_fid }, 5595 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5596 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5597 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5598 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5599 NULL, NULL 5600}; 5601 5602/* 5603 * Extended attribute directory vnode operations template 5604 * 5605 * This template is identical to the directory vnodes 5606 * operation template except for restricted operations: 5607 * VOP_MKDIR() 5608 * VOP_SYMLINK() 5609 * 5610 * Note that there are other restrictions embedded in: 5611 * zfs_create() - restrict type to VREG 5612 * zfs_link() - no links into/out of attribute space 5613 * zfs_rename() - no moves into/out of attribute space 5614 */ 5615vnodeops_t *zfs_xdvnodeops; 5616const fs_operation_def_t zfs_xdvnodeops_template[] = { 5617 VOPNAME_OPEN, { .vop_open = zfs_open }, 5618 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5619 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5620 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5621 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5622 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5623 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5624 VOPNAME_CREATE, { .vop_create = zfs_create }, 5625 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5626 VOPNAME_LINK, { .vop_link = zfs_link }, 5627 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5628 VOPNAME_MKDIR, { .error = zfs_inval }, 5629 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5630 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5631 VOPNAME_SYMLINK, { .error = zfs_inval }, 5632 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5633 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5634 VOPNAME_FID, { .vop_fid = zfs_fid }, 5635 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5636 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5637 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5638 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5639 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5640 NULL, NULL 5641}; 5642 5643/* 5644 * Error vnode operations template 5645 */ 5646vnodeops_t *zfs_evnodeops; 5647const fs_operation_def_t zfs_evnodeops_template[] = { 5648 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5649 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5650 NULL, NULL 5651}; 5652#endif /* sun */ 5653 5654static int 5655ioflags(int ioflags) 5656{ 5657 int flags = 0; 5658 5659 if (ioflags & IO_APPEND) 5660 flags |= FAPPEND; 5661 if (ioflags & IO_NDELAY) 5662 flags |= FNONBLOCK; 5663 if (ioflags & IO_SYNC) 5664 flags |= (FSYNC | FDSYNC | FRSYNC); 5665 5666 return (flags); 5667} 5668 5669static int 5670zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage) 5671{ 5672 znode_t *zp = VTOZ(vp); 5673 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5674 objset_t *os = zp->z_zfsvfs->z_os; 5675 vm_page_t mfirst, mlast, mreq; 5676 vm_object_t object; 5677 caddr_t va; 5678 struct sf_buf *sf; 5679 off_t startoff, endoff; 5680 int i, error; 5681 vm_pindex_t reqstart, reqend; 5682 int pcount, lsize, reqsize, size; 5683 5684 ZFS_ENTER(zfsvfs); 5685 ZFS_VERIFY_ZP(zp); 5686 5687 pcount = OFF_TO_IDX(round_page(count)); 5688 mreq = m[reqpage]; 5689 object = mreq->object; 5690 error = 0; 5691 5692 KASSERT(vp->v_object == object, ("mismatching object")); 5693 5694 if (pcount > 1 && zp->z_blksz > PAGESIZE) { 5695 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz); 5696 reqstart = OFF_TO_IDX(round_page(startoff)); 5697 if (reqstart < m[0]->pindex) 5698 reqstart = 0; 5699 else 5700 reqstart = reqstart - m[0]->pindex; 5701 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE, 5702 zp->z_blksz); 5703 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1; 5704 if (reqend > m[pcount - 1]->pindex) 5705 reqend = m[pcount - 1]->pindex; 5706 reqsize = reqend - m[reqstart]->pindex + 1; 5707 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize, 5708 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds")); 5709 } else { 5710 reqstart = reqpage; 5711 reqsize = 1; 5712 } 5713 mfirst = m[reqstart]; 5714 mlast = m[reqstart + reqsize - 1]; 5715 5716 zfs_vmobject_wlock(object); 5717 5718 for (i = 0; i < reqstart; i++) { 5719 vm_page_lock(m[i]); 5720 vm_page_free(m[i]); 5721 vm_page_unlock(m[i]); 5722 } 5723 for (i = reqstart + reqsize; i < pcount; i++) { 5724 vm_page_lock(m[i]); 5725 vm_page_free(m[i]); 5726 vm_page_unlock(m[i]); 5727 } 5728 5729 if (mreq->valid && reqsize == 1) { 5730 if (mreq->valid != VM_PAGE_BITS_ALL) 5731 vm_page_zero_invalid(mreq, TRUE); 5732 zfs_vmobject_wunlock(object); 5733 ZFS_EXIT(zfsvfs); 5734 return (zfs_vm_pagerret_ok); 5735 } 5736 5737 PCPU_INC(cnt.v_vnodein); 5738 PCPU_ADD(cnt.v_vnodepgsin, reqsize); 5739 5740 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) { 5741 for (i = reqstart; i < reqstart + reqsize; i++) { 5742 if (i != reqpage) { 5743 vm_page_lock(m[i]); 5744 vm_page_free(m[i]); 5745 vm_page_unlock(m[i]); 5746 } 5747 } 5748 zfs_vmobject_wunlock(object); 5749 ZFS_EXIT(zfsvfs); 5750 return (zfs_vm_pagerret_bad); 5751 } 5752 5753 lsize = PAGE_SIZE; 5754 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size) 5755 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex); 5756 5757 zfs_vmobject_wunlock(object); 5758 5759 for (i = reqstart; i < reqstart + reqsize; i++) { 5760 size = PAGE_SIZE; 5761 if (i == (reqstart + reqsize - 1)) 5762 size = lsize; 5763 va = zfs_map_page(m[i], &sf); 5764 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex), 5765 size, va, DMU_READ_PREFETCH); 5766 if (size != PAGE_SIZE) 5767 bzero(va + size, PAGE_SIZE - size); 5768 zfs_unmap_page(sf); 5769 if (error != 0) 5770 break; 5771 } 5772 5773 zfs_vmobject_wlock(object); 5774 5775 for (i = reqstart; i < reqstart + reqsize; i++) { 5776 if (!error) 5777 m[i]->valid = VM_PAGE_BITS_ALL; 5778 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i])); 5779 if (i != reqpage) 5780 vm_page_readahead_finish(m[i]); 5781 } 5782 5783 zfs_vmobject_wunlock(object); 5784 5785 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 5786 ZFS_EXIT(zfsvfs); 5787 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok); 5788} 5789 5790static int 5791zfs_freebsd_getpages(ap) 5792 struct vop_getpages_args /* { 5793 struct vnode *a_vp; 5794 vm_page_t *a_m; 5795 int a_count; 5796 int a_reqpage; 5797 vm_ooffset_t a_offset; 5798 } */ *ap; 5799{ 5800 5801 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage)); 5802} 5803 5804static int 5805zfs_freebsd_bmap(ap) 5806 struct vop_bmap_args /* { 5807 struct vnode *a_vp; 5808 daddr_t a_bn; 5809 struct bufobj **a_bop; 5810 daddr_t *a_bnp; 5811 int *a_runp; 5812 int *a_runb; 5813 } */ *ap; 5814{ 5815 5816 if (ap->a_bop != NULL) 5817 *ap->a_bop = &ap->a_vp->v_bufobj; 5818 if (ap->a_bnp != NULL) 5819 *ap->a_bnp = ap->a_bn; 5820 if (ap->a_runp != NULL) 5821 *ap->a_runp = 0; 5822 if (ap->a_runb != NULL) 5823 *ap->a_runb = 0; 5824 5825 return (0); 5826} 5827 5828static int 5829zfs_freebsd_open(ap) 5830 struct vop_open_args /* { 5831 struct vnode *a_vp; 5832 int a_mode; 5833 struct ucred *a_cred; 5834 struct thread *a_td; 5835 } */ *ap; 5836{ 5837 vnode_t *vp = ap->a_vp; 5838 znode_t *zp = VTOZ(vp); 5839 int error; 5840 5841 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL); 5842 if (error == 0) 5843 vnode_create_vobject(vp, zp->z_size, ap->a_td); 5844 return (error); 5845} 5846 5847static int 5848zfs_freebsd_close(ap) 5849 struct vop_close_args /* { 5850 struct vnode *a_vp; 5851 int a_fflag; 5852 struct ucred *a_cred; 5853 struct thread *a_td; 5854 } */ *ap; 5855{ 5856 5857 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL)); 5858} 5859 5860static int 5861zfs_freebsd_ioctl(ap) 5862 struct vop_ioctl_args /* { 5863 struct vnode *a_vp; 5864 u_long a_command; 5865 caddr_t a_data; 5866 int a_fflag; 5867 struct ucred *cred; 5868 struct thread *td; 5869 } */ *ap; 5870{ 5871 5872 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 5873 ap->a_fflag, ap->a_cred, NULL, NULL)); 5874} 5875 5876static int 5877zfs_freebsd_read(ap) 5878 struct vop_read_args /* { 5879 struct vnode *a_vp; 5880 struct uio *a_uio; 5881 int a_ioflag; 5882 struct ucred *a_cred; 5883 } */ *ap; 5884{ 5885 5886 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 5887 ap->a_cred, NULL)); 5888} 5889 5890static int 5891zfs_freebsd_write(ap) 5892 struct vop_write_args /* { 5893 struct vnode *a_vp; 5894 struct uio *a_uio; 5895 int a_ioflag; 5896 struct ucred *a_cred; 5897 } */ *ap; 5898{ 5899 5900 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 5901 ap->a_cred, NULL)); 5902} 5903 5904static int 5905zfs_freebsd_access(ap) 5906 struct vop_access_args /* { 5907 struct vnode *a_vp; 5908 accmode_t a_accmode; 5909 struct ucred *a_cred; 5910 struct thread *a_td; 5911 } */ *ap; 5912{ 5913 vnode_t *vp = ap->a_vp; 5914 znode_t *zp = VTOZ(vp); 5915 accmode_t accmode; 5916 int error = 0; 5917 5918 /* 5919 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 5920 */ 5921 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 5922 if (accmode != 0) 5923 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL); 5924 5925 /* 5926 * VADMIN has to be handled by vaccess(). 5927 */ 5928 if (error == 0) { 5929 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 5930 if (accmode != 0) { 5931 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 5932 zp->z_gid, accmode, ap->a_cred, NULL); 5933 } 5934 } 5935 5936 /* 5937 * For VEXEC, ensure that at least one execute bit is set for 5938 * non-directories. 5939 */ 5940 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 5941 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 5942 error = EACCES; 5943 } 5944 5945 return (error); 5946} 5947 5948static int 5949zfs_freebsd_lookup(ap) 5950 struct vop_lookup_args /* { 5951 struct vnode *a_dvp; 5952 struct vnode **a_vpp; 5953 struct componentname *a_cnp; 5954 } */ *ap; 5955{ 5956 struct componentname *cnp = ap->a_cnp; 5957 char nm[NAME_MAX + 1]; 5958 5959 ASSERT(cnp->cn_namelen < sizeof(nm)); 5960 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 5961 5962 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 5963 cnp->cn_cred, cnp->cn_thread, 0)); 5964} 5965 5966static int 5967zfs_freebsd_create(ap) 5968 struct vop_create_args /* { 5969 struct vnode *a_dvp; 5970 struct vnode **a_vpp; 5971 struct componentname *a_cnp; 5972 struct vattr *a_vap; 5973 } */ *ap; 5974{ 5975 struct componentname *cnp = ap->a_cnp; 5976 vattr_t *vap = ap->a_vap; 5977 int mode; 5978 5979 ASSERT(cnp->cn_flags & SAVENAME); 5980 5981 vattr_init_mask(vap); 5982 mode = vap->va_mode & ALLPERMS; 5983 5984 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 5985 ap->a_vpp, cnp->cn_cred, cnp->cn_thread)); 5986} 5987 5988static int 5989zfs_freebsd_remove(ap) 5990 struct vop_remove_args /* { 5991 struct vnode *a_dvp; 5992 struct vnode *a_vp; 5993 struct componentname *a_cnp; 5994 } */ *ap; 5995{ 5996 5997 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 5998 5999 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 6000 ap->a_cnp->cn_cred, NULL, 0)); 6001} 6002 6003static int 6004zfs_freebsd_mkdir(ap) 6005 struct vop_mkdir_args /* { 6006 struct vnode *a_dvp; 6007 struct vnode **a_vpp; 6008 struct componentname *a_cnp; 6009 struct vattr *a_vap; 6010 } */ *ap; 6011{ 6012 vattr_t *vap = ap->a_vap; 6013 6014 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6015 6016 vattr_init_mask(vap); 6017 6018 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 6019 ap->a_cnp->cn_cred, NULL, 0, NULL)); 6020} 6021 6022static int 6023zfs_freebsd_rmdir(ap) 6024 struct vop_rmdir_args /* { 6025 struct vnode *a_dvp; 6026 struct vnode *a_vp; 6027 struct componentname *a_cnp; 6028 } */ *ap; 6029{ 6030 struct componentname *cnp = ap->a_cnp; 6031 6032 ASSERT(cnp->cn_flags & SAVENAME); 6033 6034 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0)); 6035} 6036 6037static int 6038zfs_freebsd_readdir(ap) 6039 struct vop_readdir_args /* { 6040 struct vnode *a_vp; 6041 struct uio *a_uio; 6042 struct ucred *a_cred; 6043 int *a_eofflag; 6044 int *a_ncookies; 6045 u_long **a_cookies; 6046 } */ *ap; 6047{ 6048 6049 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 6050 ap->a_ncookies, ap->a_cookies)); 6051} 6052 6053static int 6054zfs_freebsd_fsync(ap) 6055 struct vop_fsync_args /* { 6056 struct vnode *a_vp; 6057 int a_waitfor; 6058 struct thread *a_td; 6059 } */ *ap; 6060{ 6061 6062 vop_stdfsync(ap); 6063 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL)); 6064} 6065 6066static int 6067zfs_freebsd_getattr(ap) 6068 struct vop_getattr_args /* { 6069 struct vnode *a_vp; 6070 struct vattr *a_vap; 6071 struct ucred *a_cred; 6072 } */ *ap; 6073{ 6074 vattr_t *vap = ap->a_vap; 6075 xvattr_t xvap; 6076 u_long fflags = 0; 6077 int error; 6078 6079 xva_init(&xvap); 6080 xvap.xva_vattr = *vap; 6081 xvap.xva_vattr.va_mask |= AT_XVATTR; 6082 6083 /* Convert chflags into ZFS-type flags. */ 6084 /* XXX: what about SF_SETTABLE?. */ 6085 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 6086 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 6087 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 6088 XVA_SET_REQ(&xvap, XAT_NODUMP); 6089 XVA_SET_REQ(&xvap, XAT_READONLY); 6090 XVA_SET_REQ(&xvap, XAT_ARCHIVE); 6091 XVA_SET_REQ(&xvap, XAT_SYSTEM); 6092 XVA_SET_REQ(&xvap, XAT_HIDDEN); 6093 XVA_SET_REQ(&xvap, XAT_REPARSE); 6094 XVA_SET_REQ(&xvap, XAT_OFFLINE); 6095 XVA_SET_REQ(&xvap, XAT_SPARSE); 6096 6097 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL); 6098 if (error != 0) 6099 return (error); 6100 6101 /* Convert ZFS xattr into chflags. */ 6102#define FLAG_CHECK(fflag, xflag, xfield) do { \ 6103 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 6104 fflags |= (fflag); \ 6105} while (0) 6106 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 6107 xvap.xva_xoptattrs.xoa_immutable); 6108 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 6109 xvap.xva_xoptattrs.xoa_appendonly); 6110 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 6111 xvap.xva_xoptattrs.xoa_nounlink); 6112 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE, 6113 xvap.xva_xoptattrs.xoa_archive); 6114 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 6115 xvap.xva_xoptattrs.xoa_nodump); 6116 FLAG_CHECK(UF_READONLY, XAT_READONLY, 6117 xvap.xva_xoptattrs.xoa_readonly); 6118 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM, 6119 xvap.xva_xoptattrs.xoa_system); 6120 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN, 6121 xvap.xva_xoptattrs.xoa_hidden); 6122 FLAG_CHECK(UF_REPARSE, XAT_REPARSE, 6123 xvap.xva_xoptattrs.xoa_reparse); 6124 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE, 6125 xvap.xva_xoptattrs.xoa_offline); 6126 FLAG_CHECK(UF_SPARSE, XAT_SPARSE, 6127 xvap.xva_xoptattrs.xoa_sparse); 6128 6129#undef FLAG_CHECK 6130 *vap = xvap.xva_vattr; 6131 vap->va_flags = fflags; 6132 return (0); 6133} 6134 6135static int 6136zfs_freebsd_setattr(ap) 6137 struct vop_setattr_args /* { 6138 struct vnode *a_vp; 6139 struct vattr *a_vap; 6140 struct ucred *a_cred; 6141 } */ *ap; 6142{ 6143 vnode_t *vp = ap->a_vp; 6144 vattr_t *vap = ap->a_vap; 6145 cred_t *cred = ap->a_cred; 6146 xvattr_t xvap; 6147 u_long fflags; 6148 uint64_t zflags; 6149 6150 vattr_init_mask(vap); 6151 vap->va_mask &= ~AT_NOSET; 6152 6153 xva_init(&xvap); 6154 xvap.xva_vattr = *vap; 6155 6156 zflags = VTOZ(vp)->z_pflags; 6157 6158 if (vap->va_flags != VNOVAL) { 6159 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 6160 int error; 6161 6162 if (zfsvfs->z_use_fuids == B_FALSE) 6163 return (EOPNOTSUPP); 6164 6165 fflags = vap->va_flags; 6166 /* 6167 * XXX KDM 6168 * We need to figure out whether it makes sense to allow 6169 * UF_REPARSE through, since we don't really have other 6170 * facilities to handle reparse points and zfs_setattr() 6171 * doesn't currently allow setting that attribute anyway. 6172 */ 6173 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE| 6174 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE| 6175 UF_OFFLINE|UF_SPARSE)) != 0) 6176 return (EOPNOTSUPP); 6177 /* 6178 * Unprivileged processes are not permitted to unset system 6179 * flags, or modify flags if any system flags are set. 6180 * Privileged non-jail processes may not modify system flags 6181 * if securelevel > 0 and any existing system flags are set. 6182 * Privileged jail processes behave like privileged non-jail 6183 * processes if the security.jail.chflags_allowed sysctl is 6184 * is non-zero; otherwise, they behave like unprivileged 6185 * processes. 6186 */ 6187 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 6188 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) { 6189 if (zflags & 6190 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6191 error = securelevel_gt(cred, 0); 6192 if (error != 0) 6193 return (error); 6194 } 6195 } else { 6196 /* 6197 * Callers may only modify the file flags on objects they 6198 * have VADMIN rights for. 6199 */ 6200 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0) 6201 return (error); 6202 if (zflags & 6203 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6204 return (EPERM); 6205 } 6206 if (fflags & 6207 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 6208 return (EPERM); 6209 } 6210 } 6211 6212#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 6213 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 6214 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 6215 XVA_SET_REQ(&xvap, (xflag)); \ 6216 (xfield) = ((fflags & (fflag)) != 0); \ 6217 } \ 6218} while (0) 6219 /* Convert chflags into ZFS-type flags. */ 6220 /* XXX: what about SF_SETTABLE?. */ 6221 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 6222 xvap.xva_xoptattrs.xoa_immutable); 6223 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 6224 xvap.xva_xoptattrs.xoa_appendonly); 6225 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 6226 xvap.xva_xoptattrs.xoa_nounlink); 6227 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE, 6228 xvap.xva_xoptattrs.xoa_archive); 6229 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 6230 xvap.xva_xoptattrs.xoa_nodump); 6231 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY, 6232 xvap.xva_xoptattrs.xoa_readonly); 6233 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM, 6234 xvap.xva_xoptattrs.xoa_system); 6235 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN, 6236 xvap.xva_xoptattrs.xoa_hidden); 6237 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE, 6238 xvap.xva_xoptattrs.xoa_hidden); 6239 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE, 6240 xvap.xva_xoptattrs.xoa_offline); 6241 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE, 6242 xvap.xva_xoptattrs.xoa_sparse); 6243#undef FLAG_CHANGE 6244 } 6245 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL)); 6246} 6247 6248static int 6249zfs_freebsd_rename(ap) 6250 struct vop_rename_args /* { 6251 struct vnode *a_fdvp; 6252 struct vnode *a_fvp; 6253 struct componentname *a_fcnp; 6254 struct vnode *a_tdvp; 6255 struct vnode *a_tvp; 6256 struct componentname *a_tcnp; 6257 } */ *ap; 6258{ 6259 vnode_t *fdvp = ap->a_fdvp; 6260 vnode_t *fvp = ap->a_fvp; 6261 vnode_t *tdvp = ap->a_tdvp; 6262 vnode_t *tvp = ap->a_tvp; 6263 int error; 6264 6265 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 6266 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 6267 6268 /* 6269 * Check for cross-device rename. 6270 */ 6271 if ((fdvp->v_mount != tdvp->v_mount) || 6272 (tvp && (fdvp->v_mount != tvp->v_mount))) 6273 error = EXDEV; 6274 else 6275 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 6276 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0); 6277 if (tdvp == tvp) 6278 VN_RELE(tdvp); 6279 else 6280 VN_URELE(tdvp); 6281 if (tvp) 6282 VN_URELE(tvp); 6283 VN_RELE(fdvp); 6284 VN_RELE(fvp); 6285 6286 return (error); 6287} 6288 6289static int 6290zfs_freebsd_symlink(ap) 6291 struct vop_symlink_args /* { 6292 struct vnode *a_dvp; 6293 struct vnode **a_vpp; 6294 struct componentname *a_cnp; 6295 struct vattr *a_vap; 6296 char *a_target; 6297 } */ *ap; 6298{ 6299 struct componentname *cnp = ap->a_cnp; 6300 vattr_t *vap = ap->a_vap; 6301 6302 ASSERT(cnp->cn_flags & SAVENAME); 6303 6304 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 6305 vattr_init_mask(vap); 6306 6307 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 6308 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 6309} 6310 6311static int 6312zfs_freebsd_readlink(ap) 6313 struct vop_readlink_args /* { 6314 struct vnode *a_vp; 6315 struct uio *a_uio; 6316 struct ucred *a_cred; 6317 } */ *ap; 6318{ 6319 6320 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL)); 6321} 6322 6323static int 6324zfs_freebsd_link(ap) 6325 struct vop_link_args /* { 6326 struct vnode *a_tdvp; 6327 struct vnode *a_vp; 6328 struct componentname *a_cnp; 6329 } */ *ap; 6330{ 6331 struct componentname *cnp = ap->a_cnp; 6332 vnode_t *vp = ap->a_vp; 6333 vnode_t *tdvp = ap->a_tdvp; 6334 6335 if (tdvp->v_mount != vp->v_mount) 6336 return (EXDEV); 6337 6338 ASSERT(cnp->cn_flags & SAVENAME); 6339 6340 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0)); 6341} 6342 6343static int 6344zfs_freebsd_inactive(ap) 6345 struct vop_inactive_args /* { 6346 struct vnode *a_vp; 6347 struct thread *a_td; 6348 } */ *ap; 6349{ 6350 vnode_t *vp = ap->a_vp; 6351 6352 zfs_inactive(vp, ap->a_td->td_ucred, NULL); 6353 return (0); 6354} 6355 6356static int 6357zfs_freebsd_reclaim(ap) 6358 struct vop_reclaim_args /* { 6359 struct vnode *a_vp; 6360 struct thread *a_td; 6361 } */ *ap; 6362{ 6363 vnode_t *vp = ap->a_vp; 6364 znode_t *zp = VTOZ(vp); 6365 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6366 6367 ASSERT(zp != NULL); 6368 6369 /* Destroy the vm object and flush associated pages. */ 6370 vnode_destroy_vobject(vp); 6371 6372 /* 6373 * z_teardown_inactive_lock protects from a race with 6374 * zfs_znode_dmu_fini in zfsvfs_teardown during 6375 * force unmount. 6376 */ 6377 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 6378 if (zp->z_sa_hdl == NULL) 6379 zfs_znode_free(zp); 6380 else 6381 zfs_zinactive(zp); 6382 rw_exit(&zfsvfs->z_teardown_inactive_lock); 6383 6384 vp->v_data = NULL; 6385 return (0); 6386} 6387 6388static int 6389zfs_freebsd_fid(ap) 6390 struct vop_fid_args /* { 6391 struct vnode *a_vp; 6392 struct fid *a_fid; 6393 } */ *ap; 6394{ 6395 6396 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 6397} 6398 6399static int 6400zfs_freebsd_pathconf(ap) 6401 struct vop_pathconf_args /* { 6402 struct vnode *a_vp; 6403 int a_name; 6404 register_t *a_retval; 6405 } */ *ap; 6406{ 6407 ulong_t val; 6408 int error; 6409 6410 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); 6411 if (error == 0) 6412 *ap->a_retval = val; 6413 else if (error == EOPNOTSUPP) 6414 error = vop_stdpathconf(ap); 6415 return (error); 6416} 6417 6418static int 6419zfs_freebsd_fifo_pathconf(ap) 6420 struct vop_pathconf_args /* { 6421 struct vnode *a_vp; 6422 int a_name; 6423 register_t *a_retval; 6424 } */ *ap; 6425{ 6426 6427 switch (ap->a_name) { 6428 case _PC_ACL_EXTENDED: 6429 case _PC_ACL_NFS4: 6430 case _PC_ACL_PATH_MAX: 6431 case _PC_MAC_PRESENT: 6432 return (zfs_freebsd_pathconf(ap)); 6433 default: 6434 return (fifo_specops.vop_pathconf(ap)); 6435 } 6436} 6437 6438/* 6439 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 6440 * extended attribute name: 6441 * 6442 * NAMESPACE PREFIX 6443 * system freebsd:system: 6444 * user (none, can be used to access ZFS fsattr(5) attributes 6445 * created on Solaris) 6446 */ 6447static int 6448zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 6449 size_t size) 6450{ 6451 const char *namespace, *prefix, *suffix; 6452 6453 /* We don't allow '/' character in attribute name. */ 6454 if (strchr(name, '/') != NULL) 6455 return (EINVAL); 6456 /* We don't allow attribute names that start with "freebsd:" string. */ 6457 if (strncmp(name, "freebsd:", 8) == 0) 6458 return (EINVAL); 6459 6460 bzero(attrname, size); 6461 6462 switch (attrnamespace) { 6463 case EXTATTR_NAMESPACE_USER: 6464#if 0 6465 prefix = "freebsd:"; 6466 namespace = EXTATTR_NAMESPACE_USER_STRING; 6467 suffix = ":"; 6468#else 6469 /* 6470 * This is the default namespace by which we can access all 6471 * attributes created on Solaris. 6472 */ 6473 prefix = namespace = suffix = ""; 6474#endif 6475 break; 6476 case EXTATTR_NAMESPACE_SYSTEM: 6477 prefix = "freebsd:"; 6478 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 6479 suffix = ":"; 6480 break; 6481 case EXTATTR_NAMESPACE_EMPTY: 6482 default: 6483 return (EINVAL); 6484 } 6485 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 6486 name) >= size) { 6487 return (ENAMETOOLONG); 6488 } 6489 return (0); 6490} 6491 6492/* 6493 * Vnode operating to retrieve a named extended attribute. 6494 */ 6495static int 6496zfs_getextattr(struct vop_getextattr_args *ap) 6497/* 6498vop_getextattr { 6499 IN struct vnode *a_vp; 6500 IN int a_attrnamespace; 6501 IN const char *a_name; 6502 INOUT struct uio *a_uio; 6503 OUT size_t *a_size; 6504 IN struct ucred *a_cred; 6505 IN struct thread *a_td; 6506}; 6507*/ 6508{ 6509 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6510 struct thread *td = ap->a_td; 6511 struct nameidata nd; 6512 char attrname[255]; 6513 struct vattr va; 6514 vnode_t *xvp = NULL, *vp; 6515 int error, flags; 6516 6517 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6518 ap->a_cred, ap->a_td, VREAD); 6519 if (error != 0) 6520 return (error); 6521 6522 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6523 sizeof(attrname)); 6524 if (error != 0) 6525 return (error); 6526 6527 ZFS_ENTER(zfsvfs); 6528 6529 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6530 LOOKUP_XATTR); 6531 if (error != 0) { 6532 ZFS_EXIT(zfsvfs); 6533 return (error); 6534 } 6535 6536 flags = FREAD; 6537 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6538 xvp, td); 6539 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL); 6540 vp = nd.ni_vp; 6541 NDFREE(&nd, NDF_ONLY_PNBUF); 6542 if (error != 0) { 6543 ZFS_EXIT(zfsvfs); 6544 if (error == ENOENT) 6545 error = ENOATTR; 6546 return (error); 6547 } 6548 6549 if (ap->a_size != NULL) { 6550 error = VOP_GETATTR(vp, &va, ap->a_cred); 6551 if (error == 0) 6552 *ap->a_size = (size_t)va.va_size; 6553 } else if (ap->a_uio != NULL) 6554 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6555 6556 VOP_UNLOCK(vp, 0); 6557 vn_close(vp, flags, ap->a_cred, td); 6558 ZFS_EXIT(zfsvfs); 6559 6560 return (error); 6561} 6562 6563/* 6564 * Vnode operation to remove a named attribute. 6565 */ 6566int 6567zfs_deleteextattr(struct vop_deleteextattr_args *ap) 6568/* 6569vop_deleteextattr { 6570 IN struct vnode *a_vp; 6571 IN int a_attrnamespace; 6572 IN const char *a_name; 6573 IN struct ucred *a_cred; 6574 IN struct thread *a_td; 6575}; 6576*/ 6577{ 6578 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6579 struct thread *td = ap->a_td; 6580 struct nameidata nd; 6581 char attrname[255]; 6582 struct vattr va; 6583 vnode_t *xvp = NULL, *vp; 6584 int error, flags; 6585 6586 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6587 ap->a_cred, ap->a_td, VWRITE); 6588 if (error != 0) 6589 return (error); 6590 6591 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6592 sizeof(attrname)); 6593 if (error != 0) 6594 return (error); 6595 6596 ZFS_ENTER(zfsvfs); 6597 6598 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6599 LOOKUP_XATTR); 6600 if (error != 0) { 6601 ZFS_EXIT(zfsvfs); 6602 return (error); 6603 } 6604 6605 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 6606 UIO_SYSSPACE, attrname, xvp, td); 6607 error = namei(&nd); 6608 vp = nd.ni_vp; 6609 NDFREE(&nd, NDF_ONLY_PNBUF); 6610 if (error != 0) { 6611 ZFS_EXIT(zfsvfs); 6612 if (error == ENOENT) 6613 error = ENOATTR; 6614 return (error); 6615 } 6616 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 6617 6618 vput(nd.ni_dvp); 6619 if (vp == nd.ni_dvp) 6620 vrele(vp); 6621 else 6622 vput(vp); 6623 ZFS_EXIT(zfsvfs); 6624 6625 return (error); 6626} 6627 6628/* 6629 * Vnode operation to set a named attribute. 6630 */ 6631static int 6632zfs_setextattr(struct vop_setextattr_args *ap) 6633/* 6634vop_setextattr { 6635 IN struct vnode *a_vp; 6636 IN int a_attrnamespace; 6637 IN const char *a_name; 6638 INOUT struct uio *a_uio; 6639 IN struct ucred *a_cred; 6640 IN struct thread *a_td; 6641}; 6642*/ 6643{ 6644 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6645 struct thread *td = ap->a_td; 6646 struct nameidata nd; 6647 char attrname[255]; 6648 struct vattr va; 6649 vnode_t *xvp = NULL, *vp; 6650 int error, flags; 6651 6652 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6653 ap->a_cred, ap->a_td, VWRITE); 6654 if (error != 0) 6655 return (error); 6656 6657 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6658 sizeof(attrname)); 6659 if (error != 0) 6660 return (error); 6661 6662 ZFS_ENTER(zfsvfs); 6663 6664 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6665 LOOKUP_XATTR | CREATE_XATTR_DIR); 6666 if (error != 0) { 6667 ZFS_EXIT(zfsvfs); 6668 return (error); 6669 } 6670 6671 flags = FFLAGS(O_WRONLY | O_CREAT); 6672 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6673 xvp, td); 6674 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL); 6675 vp = nd.ni_vp; 6676 NDFREE(&nd, NDF_ONLY_PNBUF); 6677 if (error != 0) { 6678 ZFS_EXIT(zfsvfs); 6679 return (error); 6680 } 6681 6682 VATTR_NULL(&va); 6683 va.va_size = 0; 6684 error = VOP_SETATTR(vp, &va, ap->a_cred); 6685 if (error == 0) 6686 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred); 6687 6688 VOP_UNLOCK(vp, 0); 6689 vn_close(vp, flags, ap->a_cred, td); 6690 ZFS_EXIT(zfsvfs); 6691 6692 return (error); 6693} 6694 6695/* 6696 * Vnode operation to retrieve extended attributes on a vnode. 6697 */ 6698static int 6699zfs_listextattr(struct vop_listextattr_args *ap) 6700/* 6701vop_listextattr { 6702 IN struct vnode *a_vp; 6703 IN int a_attrnamespace; 6704 INOUT struct uio *a_uio; 6705 OUT size_t *a_size; 6706 IN struct ucred *a_cred; 6707 IN struct thread *a_td; 6708}; 6709*/ 6710{ 6711 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6712 struct thread *td = ap->a_td; 6713 struct nameidata nd; 6714 char attrprefix[16]; 6715 u_char dirbuf[sizeof(struct dirent)]; 6716 struct dirent *dp; 6717 struct iovec aiov; 6718 struct uio auio, *uio = ap->a_uio; 6719 size_t *sizep = ap->a_size; 6720 size_t plen; 6721 vnode_t *xvp = NULL, *vp; 6722 int done, error, eof, pos; 6723 6724 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6725 ap->a_cred, ap->a_td, VREAD); 6726 if (error != 0) 6727 return (error); 6728 6729 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 6730 sizeof(attrprefix)); 6731 if (error != 0) 6732 return (error); 6733 plen = strlen(attrprefix); 6734 6735 ZFS_ENTER(zfsvfs); 6736 6737 if (sizep != NULL) 6738 *sizep = 0; 6739 6740 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6741 LOOKUP_XATTR); 6742 if (error != 0) { 6743 ZFS_EXIT(zfsvfs); 6744 /* 6745 * ENOATTR means that the EA directory does not yet exist, 6746 * i.e. there are no extended attributes there. 6747 */ 6748 if (error == ENOATTR) 6749 error = 0; 6750 return (error); 6751 } 6752 6753 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 6754 UIO_SYSSPACE, ".", xvp, td); 6755 error = namei(&nd); 6756 vp = nd.ni_vp; 6757 NDFREE(&nd, NDF_ONLY_PNBUF); 6758 if (error != 0) { 6759 ZFS_EXIT(zfsvfs); 6760 return (error); 6761 } 6762 6763 auio.uio_iov = &aiov; 6764 auio.uio_iovcnt = 1; 6765 auio.uio_segflg = UIO_SYSSPACE; 6766 auio.uio_td = td; 6767 auio.uio_rw = UIO_READ; 6768 auio.uio_offset = 0; 6769 6770 do { 6771 u_char nlen; 6772 6773 aiov.iov_base = (void *)dirbuf; 6774 aiov.iov_len = sizeof(dirbuf); 6775 auio.uio_resid = sizeof(dirbuf); 6776 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 6777 done = sizeof(dirbuf) - auio.uio_resid; 6778 if (error != 0) 6779 break; 6780 for (pos = 0; pos < done;) { 6781 dp = (struct dirent *)(dirbuf + pos); 6782 pos += dp->d_reclen; 6783 /* 6784 * XXX: Temporarily we also accept DT_UNKNOWN, as this 6785 * is what we get when attribute was created on Solaris. 6786 */ 6787 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 6788 continue; 6789 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0) 6790 continue; 6791 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 6792 continue; 6793 nlen = dp->d_namlen - plen; 6794 if (sizep != NULL) 6795 *sizep += 1 + nlen; 6796 else if (uio != NULL) { 6797 /* 6798 * Format of extattr name entry is one byte for 6799 * length and the rest for name. 6800 */ 6801 error = uiomove(&nlen, 1, uio->uio_rw, uio); 6802 if (error == 0) { 6803 error = uiomove(dp->d_name + plen, nlen, 6804 uio->uio_rw, uio); 6805 } 6806 if (error != 0) 6807 break; 6808 } 6809 } 6810 } while (!eof && error == 0); 6811 6812 vput(vp); 6813 ZFS_EXIT(zfsvfs); 6814 6815 return (error); 6816} 6817 6818int 6819zfs_freebsd_getacl(ap) 6820 struct vop_getacl_args /* { 6821 struct vnode *vp; 6822 acl_type_t type; 6823 struct acl *aclp; 6824 struct ucred *cred; 6825 struct thread *td; 6826 } */ *ap; 6827{ 6828 int error; 6829 vsecattr_t vsecattr; 6830 6831 if (ap->a_type != ACL_TYPE_NFS4) 6832 return (EINVAL); 6833 6834 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 6835 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL)) 6836 return (error); 6837 6838 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt); 6839 if (vsecattr.vsa_aclentp != NULL) 6840 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 6841 6842 return (error); 6843} 6844 6845int 6846zfs_freebsd_setacl(ap) 6847 struct vop_setacl_args /* { 6848 struct vnode *vp; 6849 acl_type_t type; 6850 struct acl *aclp; 6851 struct ucred *cred; 6852 struct thread *td; 6853 } */ *ap; 6854{ 6855 int error; 6856 vsecattr_t vsecattr; 6857 int aclbsize; /* size of acl list in bytes */ 6858 aclent_t *aaclp; 6859 6860 if (ap->a_type != ACL_TYPE_NFS4) 6861 return (EINVAL); 6862 6863 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 6864 return (EINVAL); 6865 6866 /* 6867 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 6868 * splitting every entry into two and appending "canonical six" 6869 * entries at the end. Don't allow for setting an ACL that would 6870 * cause chmod(2) to run out of ACL entries. 6871 */ 6872 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 6873 return (ENOSPC); 6874 6875 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 6876 if (error != 0) 6877 return (error); 6878 6879 vsecattr.vsa_mask = VSA_ACE; 6880 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t); 6881 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 6882 aaclp = vsecattr.vsa_aclentp; 6883 vsecattr.vsa_aclentsz = aclbsize; 6884 6885 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 6886 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL); 6887 kmem_free(aaclp, aclbsize); 6888 6889 return (error); 6890} 6891 6892int 6893zfs_freebsd_aclcheck(ap) 6894 struct vop_aclcheck_args /* { 6895 struct vnode *vp; 6896 acl_type_t type; 6897 struct acl *aclp; 6898 struct ucred *cred; 6899 struct thread *td; 6900 } */ *ap; 6901{ 6902 6903 return (EOPNOTSUPP); 6904} 6905 6906struct vop_vector zfs_vnodeops; 6907struct vop_vector zfs_fifoops; 6908struct vop_vector zfs_shareops; 6909 6910struct vop_vector zfs_vnodeops = { 6911 .vop_default = &default_vnodeops, 6912 .vop_inactive = zfs_freebsd_inactive, 6913 .vop_reclaim = zfs_freebsd_reclaim, 6914 .vop_access = zfs_freebsd_access, 6915#ifdef FREEBSD_NAMECACHE 6916 .vop_lookup = vfs_cache_lookup, 6917 .vop_cachedlookup = zfs_freebsd_lookup, 6918#else 6919 .vop_lookup = zfs_freebsd_lookup, 6920#endif 6921 .vop_getattr = zfs_freebsd_getattr, 6922 .vop_setattr = zfs_freebsd_setattr, 6923 .vop_create = zfs_freebsd_create, 6924 .vop_mknod = zfs_freebsd_create, 6925 .vop_mkdir = zfs_freebsd_mkdir, 6926 .vop_readdir = zfs_freebsd_readdir, 6927 .vop_fsync = zfs_freebsd_fsync, 6928 .vop_open = zfs_freebsd_open, 6929 .vop_close = zfs_freebsd_close, 6930 .vop_rmdir = zfs_freebsd_rmdir, 6931 .vop_ioctl = zfs_freebsd_ioctl, 6932 .vop_link = zfs_freebsd_link, 6933 .vop_symlink = zfs_freebsd_symlink, 6934 .vop_readlink = zfs_freebsd_readlink, 6935 .vop_read = zfs_freebsd_read, 6936 .vop_write = zfs_freebsd_write, 6937 .vop_remove = zfs_freebsd_remove, 6938 .vop_rename = zfs_freebsd_rename, 6939 .vop_pathconf = zfs_freebsd_pathconf, 6940 .vop_bmap = zfs_freebsd_bmap, 6941 .vop_fid = zfs_freebsd_fid, 6942 .vop_getextattr = zfs_getextattr, 6943 .vop_deleteextattr = zfs_deleteextattr, 6944 .vop_setextattr = zfs_setextattr, 6945 .vop_listextattr = zfs_listextattr, 6946 .vop_getacl = zfs_freebsd_getacl, 6947 .vop_setacl = zfs_freebsd_setacl, 6948 .vop_aclcheck = zfs_freebsd_aclcheck, 6949 .vop_getpages = zfs_freebsd_getpages, 6950}; 6951 6952struct vop_vector zfs_fifoops = { 6953 .vop_default = &fifo_specops, 6954 .vop_fsync = zfs_freebsd_fsync, 6955 .vop_access = zfs_freebsd_access, 6956 .vop_getattr = zfs_freebsd_getattr, 6957 .vop_inactive = zfs_freebsd_inactive, 6958 .vop_read = VOP_PANIC, 6959 .vop_reclaim = zfs_freebsd_reclaim, 6960 .vop_setattr = zfs_freebsd_setattr, 6961 .vop_write = VOP_PANIC, 6962 .vop_pathconf = zfs_freebsd_fifo_pathconf, 6963 .vop_fid = zfs_freebsd_fid, 6964 .vop_getacl = zfs_freebsd_getacl, 6965 .vop_setacl = zfs_freebsd_setacl, 6966 .vop_aclcheck = zfs_freebsd_aclcheck, 6967}; 6968 6969/* 6970 * special share hidden files vnode operations template 6971 */ 6972struct vop_vector zfs_shareops = { 6973 .vop_default = &default_vnodeops, 6974 .vop_access = zfs_freebsd_access, 6975 .vop_inactive = zfs_freebsd_inactive, 6976 .vop_reclaim = zfs_freebsd_reclaim, 6977 .vop_fid = zfs_freebsd_fid, 6978 .vop_pathconf = zfs_freebsd_pathconf, 6979}; 6980