zfs_vnops.c revision 274110
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 vap->va_filerev = zp->z_seq; 2826 2827 /* 2828 * Add in any requested optional attributes and the create time. 2829 * Also set the corresponding bits in the returned attribute bitmap. 2830 */ 2831 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2832 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2833 xoap->xoa_archive = 2834 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2835 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2836 } 2837 2838 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2839 xoap->xoa_readonly = 2840 ((zp->z_pflags & ZFS_READONLY) != 0); 2841 XVA_SET_RTN(xvap, XAT_READONLY); 2842 } 2843 2844 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2845 xoap->xoa_system = 2846 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2847 XVA_SET_RTN(xvap, XAT_SYSTEM); 2848 } 2849 2850 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2851 xoap->xoa_hidden = 2852 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2853 XVA_SET_RTN(xvap, XAT_HIDDEN); 2854 } 2855 2856 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2857 xoap->xoa_nounlink = 2858 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2859 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2860 } 2861 2862 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2863 xoap->xoa_immutable = 2864 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2865 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2866 } 2867 2868 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2869 xoap->xoa_appendonly = 2870 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2871 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2872 } 2873 2874 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2875 xoap->xoa_nodump = 2876 ((zp->z_pflags & ZFS_NODUMP) != 0); 2877 XVA_SET_RTN(xvap, XAT_NODUMP); 2878 } 2879 2880 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2881 xoap->xoa_opaque = 2882 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2883 XVA_SET_RTN(xvap, XAT_OPAQUE); 2884 } 2885 2886 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2887 xoap->xoa_av_quarantined = 2888 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2889 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2890 } 2891 2892 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2893 xoap->xoa_av_modified = 2894 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2895 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2896 } 2897 2898 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2899 vp->v_type == VREG) { 2900 zfs_sa_get_scanstamp(zp, xvap); 2901 } 2902 2903 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2904 uint64_t times[2]; 2905 2906 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 2907 times, sizeof (times)); 2908 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 2909 XVA_SET_RTN(xvap, XAT_CREATETIME); 2910 } 2911 2912 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2913 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2914 XVA_SET_RTN(xvap, XAT_REPARSE); 2915 } 2916 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2917 xoap->xoa_generation = zp->z_gen; 2918 XVA_SET_RTN(xvap, XAT_GEN); 2919 } 2920 2921 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2922 xoap->xoa_offline = 2923 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2924 XVA_SET_RTN(xvap, XAT_OFFLINE); 2925 } 2926 2927 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2928 xoap->xoa_sparse = 2929 ((zp->z_pflags & ZFS_SPARSE) != 0); 2930 XVA_SET_RTN(xvap, XAT_SPARSE); 2931 } 2932 } 2933 2934 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2935 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2936 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2937 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 2938 2939 mutex_exit(&zp->z_lock); 2940 2941 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 2942 vap->va_blksize = blksize; 2943 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2944 2945 if (zp->z_blksz == 0) { 2946 /* 2947 * Block size hasn't been set; suggest maximal I/O transfers. 2948 */ 2949 vap->va_blksize = zfsvfs->z_max_blksz; 2950 } 2951 2952 ZFS_EXIT(zfsvfs); 2953 return (0); 2954} 2955 2956/* 2957 * Set the file attributes to the values contained in the 2958 * vattr structure. 2959 * 2960 * IN: vp - vnode of file to be modified. 2961 * vap - new attribute values. 2962 * If AT_XVATTR set, then optional attrs are being set 2963 * flags - ATTR_UTIME set if non-default time values provided. 2964 * - ATTR_NOACLCHECK (CIFS context only). 2965 * cr - credentials of caller. 2966 * ct - caller context 2967 * 2968 * RETURN: 0 on success, error code on failure. 2969 * 2970 * Timestamps: 2971 * vp - ctime updated, mtime updated if size changed. 2972 */ 2973/* ARGSUSED */ 2974static int 2975zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2976 caller_context_t *ct) 2977{ 2978 znode_t *zp = VTOZ(vp); 2979 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2980 zilog_t *zilog; 2981 dmu_tx_t *tx; 2982 vattr_t oldva; 2983 xvattr_t tmpxvattr; 2984 uint_t mask = vap->va_mask; 2985 uint_t saved_mask = 0; 2986 uint64_t saved_mode; 2987 int trim_mask = 0; 2988 uint64_t new_mode; 2989 uint64_t new_uid, new_gid; 2990 uint64_t xattr_obj; 2991 uint64_t mtime[2], ctime[2]; 2992 znode_t *attrzp; 2993 int need_policy = FALSE; 2994 int err, err2; 2995 zfs_fuid_info_t *fuidp = NULL; 2996 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2997 xoptattr_t *xoap; 2998 zfs_acl_t *aclp; 2999 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 3000 boolean_t fuid_dirtied = B_FALSE; 3001 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 3002 int count = 0, xattr_count = 0; 3003 3004 if (mask == 0) 3005 return (0); 3006 3007 if (mask & AT_NOSET) 3008 return (SET_ERROR(EINVAL)); 3009 3010 ZFS_ENTER(zfsvfs); 3011 ZFS_VERIFY_ZP(zp); 3012 3013 zilog = zfsvfs->z_log; 3014 3015 /* 3016 * Make sure that if we have ephemeral uid/gid or xvattr specified 3017 * that file system is at proper version level 3018 */ 3019 3020 if (zfsvfs->z_use_fuids == B_FALSE && 3021 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 3022 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 3023 (mask & AT_XVATTR))) { 3024 ZFS_EXIT(zfsvfs); 3025 return (SET_ERROR(EINVAL)); 3026 } 3027 3028 if (mask & AT_SIZE && vp->v_type == VDIR) { 3029 ZFS_EXIT(zfsvfs); 3030 return (SET_ERROR(EISDIR)); 3031 } 3032 3033 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 3034 ZFS_EXIT(zfsvfs); 3035 return (SET_ERROR(EINVAL)); 3036 } 3037 3038 /* 3039 * If this is an xvattr_t, then get a pointer to the structure of 3040 * optional attributes. If this is NULL, then we have a vattr_t. 3041 */ 3042 xoap = xva_getxoptattr(xvap); 3043 3044 xva_init(&tmpxvattr); 3045 3046 /* 3047 * Immutable files can only alter immutable bit and atime 3048 */ 3049 if ((zp->z_pflags & ZFS_IMMUTABLE) && 3050 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 3051 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 3052 ZFS_EXIT(zfsvfs); 3053 return (SET_ERROR(EPERM)); 3054 } 3055 3056 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 3057 ZFS_EXIT(zfsvfs); 3058 return (SET_ERROR(EPERM)); 3059 } 3060 3061 /* 3062 * Verify timestamps doesn't overflow 32 bits. 3063 * ZFS can handle large timestamps, but 32bit syscalls can't 3064 * handle times greater than 2039. This check should be removed 3065 * once large timestamps are fully supported. 3066 */ 3067 if (mask & (AT_ATIME | AT_MTIME)) { 3068 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 3069 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 3070 ZFS_EXIT(zfsvfs); 3071 return (SET_ERROR(EOVERFLOW)); 3072 } 3073 } 3074 3075top: 3076 attrzp = NULL; 3077 aclp = NULL; 3078 3079 /* Can this be moved to before the top label? */ 3080 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 3081 ZFS_EXIT(zfsvfs); 3082 return (SET_ERROR(EROFS)); 3083 } 3084 3085 /* 3086 * First validate permissions 3087 */ 3088 3089 if (mask & AT_SIZE) { 3090 /* 3091 * XXX - Note, we are not providing any open 3092 * mode flags here (like FNDELAY), so we may 3093 * block if there are locks present... this 3094 * should be addressed in openat(). 3095 */ 3096 /* XXX - would it be OK to generate a log record here? */ 3097 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 3098 if (err) { 3099 ZFS_EXIT(zfsvfs); 3100 return (err); 3101 } 3102 } 3103 3104 if (mask & (AT_ATIME|AT_MTIME) || 3105 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 3106 XVA_ISSET_REQ(xvap, XAT_READONLY) || 3107 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 3108 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 3109 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 3110 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 3111 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 3112 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 3113 skipaclchk, cr); 3114 } 3115 3116 if (mask & (AT_UID|AT_GID)) { 3117 int idmask = (mask & (AT_UID|AT_GID)); 3118 int take_owner; 3119 int take_group; 3120 3121 /* 3122 * NOTE: even if a new mode is being set, 3123 * we may clear S_ISUID/S_ISGID bits. 3124 */ 3125 3126 if (!(mask & AT_MODE)) 3127 vap->va_mode = zp->z_mode; 3128 3129 /* 3130 * Take ownership or chgrp to group we are a member of 3131 */ 3132 3133 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 3134 take_group = (mask & AT_GID) && 3135 zfs_groupmember(zfsvfs, vap->va_gid, cr); 3136 3137 /* 3138 * If both AT_UID and AT_GID are set then take_owner and 3139 * take_group must both be set in order to allow taking 3140 * ownership. 3141 * 3142 * Otherwise, send the check through secpolicy_vnode_setattr() 3143 * 3144 */ 3145 3146 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 3147 ((idmask == AT_UID) && take_owner) || 3148 ((idmask == AT_GID) && take_group)) { 3149 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 3150 skipaclchk, cr) == 0) { 3151 /* 3152 * Remove setuid/setgid for non-privileged users 3153 */ 3154 secpolicy_setid_clear(vap, vp, cr); 3155 trim_mask = (mask & (AT_UID|AT_GID)); 3156 } else { 3157 need_policy = TRUE; 3158 } 3159 } else { 3160 need_policy = TRUE; 3161 } 3162 } 3163 3164 mutex_enter(&zp->z_lock); 3165 oldva.va_mode = zp->z_mode; 3166 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 3167 if (mask & AT_XVATTR) { 3168 /* 3169 * Update xvattr mask to include only those attributes 3170 * that are actually changing. 3171 * 3172 * the bits will be restored prior to actually setting 3173 * the attributes so the caller thinks they were set. 3174 */ 3175 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 3176 if (xoap->xoa_appendonly != 3177 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 3178 need_policy = TRUE; 3179 } else { 3180 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 3181 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 3182 } 3183 } 3184 3185 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 3186 if (xoap->xoa_nounlink != 3187 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 3188 need_policy = TRUE; 3189 } else { 3190 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 3191 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 3192 } 3193 } 3194 3195 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 3196 if (xoap->xoa_immutable != 3197 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 3198 need_policy = TRUE; 3199 } else { 3200 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 3201 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 3202 } 3203 } 3204 3205 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 3206 if (xoap->xoa_nodump != 3207 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 3208 need_policy = TRUE; 3209 } else { 3210 XVA_CLR_REQ(xvap, XAT_NODUMP); 3211 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 3212 } 3213 } 3214 3215 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 3216 if (xoap->xoa_av_modified != 3217 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 3218 need_policy = TRUE; 3219 } else { 3220 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 3221 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 3222 } 3223 } 3224 3225 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 3226 if ((vp->v_type != VREG && 3227 xoap->xoa_av_quarantined) || 3228 xoap->xoa_av_quarantined != 3229 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 3230 need_policy = TRUE; 3231 } else { 3232 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 3233 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 3234 } 3235 } 3236 3237 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 3238 mutex_exit(&zp->z_lock); 3239 ZFS_EXIT(zfsvfs); 3240 return (SET_ERROR(EPERM)); 3241 } 3242 3243 if (need_policy == FALSE && 3244 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 3245 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 3246 need_policy = TRUE; 3247 } 3248 } 3249 3250 mutex_exit(&zp->z_lock); 3251 3252 if (mask & AT_MODE) { 3253 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 3254 err = secpolicy_setid_setsticky_clear(vp, vap, 3255 &oldva, cr); 3256 if (err) { 3257 ZFS_EXIT(zfsvfs); 3258 return (err); 3259 } 3260 trim_mask |= AT_MODE; 3261 } else { 3262 need_policy = TRUE; 3263 } 3264 } 3265 3266 if (need_policy) { 3267 /* 3268 * If trim_mask is set then take ownership 3269 * has been granted or write_acl is present and user 3270 * has the ability to modify mode. In that case remove 3271 * UID|GID and or MODE from mask so that 3272 * secpolicy_vnode_setattr() doesn't revoke it. 3273 */ 3274 3275 if (trim_mask) { 3276 saved_mask = vap->va_mask; 3277 vap->va_mask &= ~trim_mask; 3278 if (trim_mask & AT_MODE) { 3279 /* 3280 * Save the mode, as secpolicy_vnode_setattr() 3281 * will overwrite it with ova.va_mode. 3282 */ 3283 saved_mode = vap->va_mode; 3284 } 3285 } 3286 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 3287 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 3288 if (err) { 3289 ZFS_EXIT(zfsvfs); 3290 return (err); 3291 } 3292 3293 if (trim_mask) { 3294 vap->va_mask |= saved_mask; 3295 if (trim_mask & AT_MODE) { 3296 /* 3297 * Recover the mode after 3298 * secpolicy_vnode_setattr(). 3299 */ 3300 vap->va_mode = saved_mode; 3301 } 3302 } 3303 } 3304 3305 /* 3306 * secpolicy_vnode_setattr, or take ownership may have 3307 * changed va_mask 3308 */ 3309 mask = vap->va_mask; 3310 3311 if ((mask & (AT_UID | AT_GID))) { 3312 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 3313 &xattr_obj, sizeof (xattr_obj)); 3314 3315 if (err == 0 && xattr_obj) { 3316 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 3317 if (err) 3318 goto out2; 3319 } 3320 if (mask & AT_UID) { 3321 new_uid = zfs_fuid_create(zfsvfs, 3322 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 3323 if (new_uid != zp->z_uid && 3324 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 3325 if (attrzp) 3326 VN_RELE(ZTOV(attrzp)); 3327 err = SET_ERROR(EDQUOT); 3328 goto out2; 3329 } 3330 } 3331 3332 if (mask & AT_GID) { 3333 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 3334 cr, ZFS_GROUP, &fuidp); 3335 if (new_gid != zp->z_gid && 3336 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 3337 if (attrzp) 3338 VN_RELE(ZTOV(attrzp)); 3339 err = SET_ERROR(EDQUOT); 3340 goto out2; 3341 } 3342 } 3343 } 3344 tx = dmu_tx_create(zfsvfs->z_os); 3345 3346 if (mask & AT_MODE) { 3347 uint64_t pmode = zp->z_mode; 3348 uint64_t acl_obj; 3349 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 3350 3351 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && 3352 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 3353 err = SET_ERROR(EPERM); 3354 goto out; 3355 } 3356 3357 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 3358 goto out; 3359 3360 mutex_enter(&zp->z_lock); 3361 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 3362 /* 3363 * Are we upgrading ACL from old V0 format 3364 * to V1 format? 3365 */ 3366 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 3367 zfs_znode_acl_version(zp) == 3368 ZFS_ACL_VERSION_INITIAL) { 3369 dmu_tx_hold_free(tx, acl_obj, 0, 3370 DMU_OBJECT_END); 3371 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3372 0, aclp->z_acl_bytes); 3373 } else { 3374 dmu_tx_hold_write(tx, acl_obj, 0, 3375 aclp->z_acl_bytes); 3376 } 3377 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3378 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3379 0, aclp->z_acl_bytes); 3380 } 3381 mutex_exit(&zp->z_lock); 3382 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3383 } else { 3384 if ((mask & AT_XVATTR) && 3385 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3386 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3387 else 3388 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3389 } 3390 3391 if (attrzp) { 3392 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 3393 } 3394 3395 fuid_dirtied = zfsvfs->z_fuid_dirty; 3396 if (fuid_dirtied) 3397 zfs_fuid_txhold(zfsvfs, tx); 3398 3399 zfs_sa_upgrade_txholds(tx, zp); 3400 3401 err = dmu_tx_assign(tx, TXG_NOWAIT); 3402 if (err) { 3403 if (err == ERESTART) 3404 dmu_tx_wait(tx); 3405 goto out; 3406 } 3407 3408 count = 0; 3409 /* 3410 * Set each attribute requested. 3411 * We group settings according to the locks they need to acquire. 3412 * 3413 * Note: you cannot set ctime directly, although it will be 3414 * updated as a side-effect of calling this function. 3415 */ 3416 3417 3418 if (mask & (AT_UID|AT_GID|AT_MODE)) 3419 mutex_enter(&zp->z_acl_lock); 3420 mutex_enter(&zp->z_lock); 3421 3422 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3423 &zp->z_pflags, sizeof (zp->z_pflags)); 3424 3425 if (attrzp) { 3426 if (mask & (AT_UID|AT_GID|AT_MODE)) 3427 mutex_enter(&attrzp->z_acl_lock); 3428 mutex_enter(&attrzp->z_lock); 3429 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3430 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3431 sizeof (attrzp->z_pflags)); 3432 } 3433 3434 if (mask & (AT_UID|AT_GID)) { 3435 3436 if (mask & AT_UID) { 3437 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3438 &new_uid, sizeof (new_uid)); 3439 zp->z_uid = new_uid; 3440 if (attrzp) { 3441 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3442 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3443 sizeof (new_uid)); 3444 attrzp->z_uid = new_uid; 3445 } 3446 } 3447 3448 if (mask & AT_GID) { 3449 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3450 NULL, &new_gid, sizeof (new_gid)); 3451 zp->z_gid = new_gid; 3452 if (attrzp) { 3453 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3454 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 3455 sizeof (new_gid)); 3456 attrzp->z_gid = new_gid; 3457 } 3458 } 3459 if (!(mask & AT_MODE)) { 3460 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 3461 NULL, &new_mode, sizeof (new_mode)); 3462 new_mode = zp->z_mode; 3463 } 3464 err = zfs_acl_chown_setattr(zp); 3465 ASSERT(err == 0); 3466 if (attrzp) { 3467 err = zfs_acl_chown_setattr(attrzp); 3468 ASSERT(err == 0); 3469 } 3470 } 3471 3472 if (mask & AT_MODE) { 3473 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 3474 &new_mode, sizeof (new_mode)); 3475 zp->z_mode = new_mode; 3476 ASSERT3U((uintptr_t)aclp, !=, 0); 3477 err = zfs_aclset_common(zp, aclp, cr, tx); 3478 ASSERT0(err); 3479 if (zp->z_acl_cached) 3480 zfs_acl_free(zp->z_acl_cached); 3481 zp->z_acl_cached = aclp; 3482 aclp = NULL; 3483 } 3484 3485 3486 if (mask & AT_ATIME) { 3487 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 3488 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 3489 &zp->z_atime, sizeof (zp->z_atime)); 3490 } 3491 3492 if (mask & AT_MTIME) { 3493 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 3494 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 3495 mtime, sizeof (mtime)); 3496 } 3497 3498 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 3499 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 3500 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 3501 NULL, mtime, sizeof (mtime)); 3502 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3503 &ctime, sizeof (ctime)); 3504 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 3505 B_TRUE); 3506 } else if (mask != 0) { 3507 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3508 &ctime, sizeof (ctime)); 3509 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime, 3510 B_TRUE); 3511 if (attrzp) { 3512 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3513 SA_ZPL_CTIME(zfsvfs), NULL, 3514 &ctime, sizeof (ctime)); 3515 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 3516 mtime, ctime, B_TRUE); 3517 } 3518 } 3519 /* 3520 * Do this after setting timestamps to prevent timestamp 3521 * update from toggling bit 3522 */ 3523 3524 if (xoap && (mask & AT_XVATTR)) { 3525 3526 /* 3527 * restore trimmed off masks 3528 * so that return masks can be set for caller. 3529 */ 3530 3531 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 3532 XVA_SET_REQ(xvap, XAT_APPENDONLY); 3533 } 3534 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 3535 XVA_SET_REQ(xvap, XAT_NOUNLINK); 3536 } 3537 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 3538 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 3539 } 3540 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 3541 XVA_SET_REQ(xvap, XAT_NODUMP); 3542 } 3543 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 3544 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 3545 } 3546 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 3547 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 3548 } 3549 3550 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3551 ASSERT(vp->v_type == VREG); 3552 3553 zfs_xvattr_set(zp, xvap, tx); 3554 } 3555 3556 if (fuid_dirtied) 3557 zfs_fuid_sync(zfsvfs, tx); 3558 3559 if (mask != 0) 3560 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 3561 3562 mutex_exit(&zp->z_lock); 3563 if (mask & (AT_UID|AT_GID|AT_MODE)) 3564 mutex_exit(&zp->z_acl_lock); 3565 3566 if (attrzp) { 3567 if (mask & (AT_UID|AT_GID|AT_MODE)) 3568 mutex_exit(&attrzp->z_acl_lock); 3569 mutex_exit(&attrzp->z_lock); 3570 } 3571out: 3572 if (err == 0 && attrzp) { 3573 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 3574 xattr_count, tx); 3575 ASSERT(err2 == 0); 3576 } 3577 3578 if (attrzp) 3579 VN_RELE(ZTOV(attrzp)); 3580 3581 if (aclp) 3582 zfs_acl_free(aclp); 3583 3584 if (fuidp) { 3585 zfs_fuid_info_free(fuidp); 3586 fuidp = NULL; 3587 } 3588 3589 if (err) { 3590 dmu_tx_abort(tx); 3591 if (err == ERESTART) 3592 goto top; 3593 } else { 3594 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3595 dmu_tx_commit(tx); 3596 } 3597 3598out2: 3599 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3600 zil_commit(zilog, 0); 3601 3602 ZFS_EXIT(zfsvfs); 3603 return (err); 3604} 3605 3606typedef struct zfs_zlock { 3607 krwlock_t *zl_rwlock; /* lock we acquired */ 3608 znode_t *zl_znode; /* znode we held */ 3609 struct zfs_zlock *zl_next; /* next in list */ 3610} zfs_zlock_t; 3611 3612/* 3613 * Drop locks and release vnodes that were held by zfs_rename_lock(). 3614 */ 3615static void 3616zfs_rename_unlock(zfs_zlock_t **zlpp) 3617{ 3618 zfs_zlock_t *zl; 3619 3620 while ((zl = *zlpp) != NULL) { 3621 if (zl->zl_znode != NULL) 3622 VN_RELE(ZTOV(zl->zl_znode)); 3623 rw_exit(zl->zl_rwlock); 3624 *zlpp = zl->zl_next; 3625 kmem_free(zl, sizeof (*zl)); 3626 } 3627} 3628 3629/* 3630 * Search back through the directory tree, using the ".." entries. 3631 * Lock each directory in the chain to prevent concurrent renames. 3632 * Fail any attempt to move a directory into one of its own descendants. 3633 * XXX - z_parent_lock can overlap with map or grow locks 3634 */ 3635static int 3636zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 3637{ 3638 zfs_zlock_t *zl; 3639 znode_t *zp = tdzp; 3640 uint64_t rootid = zp->z_zfsvfs->z_root; 3641 uint64_t oidp = zp->z_id; 3642 krwlock_t *rwlp = &szp->z_parent_lock; 3643 krw_t rw = RW_WRITER; 3644 3645 /* 3646 * First pass write-locks szp and compares to zp->z_id. 3647 * Later passes read-lock zp and compare to zp->z_parent. 3648 */ 3649 do { 3650 if (!rw_tryenter(rwlp, rw)) { 3651 /* 3652 * Another thread is renaming in this path. 3653 * Note that if we are a WRITER, we don't have any 3654 * parent_locks held yet. 3655 */ 3656 if (rw == RW_READER && zp->z_id > szp->z_id) { 3657 /* 3658 * Drop our locks and restart 3659 */ 3660 zfs_rename_unlock(&zl); 3661 *zlpp = NULL; 3662 zp = tdzp; 3663 oidp = zp->z_id; 3664 rwlp = &szp->z_parent_lock; 3665 rw = RW_WRITER; 3666 continue; 3667 } else { 3668 /* 3669 * Wait for other thread to drop its locks 3670 */ 3671 rw_enter(rwlp, rw); 3672 } 3673 } 3674 3675 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 3676 zl->zl_rwlock = rwlp; 3677 zl->zl_znode = NULL; 3678 zl->zl_next = *zlpp; 3679 *zlpp = zl; 3680 3681 if (oidp == szp->z_id) /* We're a descendant of szp */ 3682 return (SET_ERROR(EINVAL)); 3683 3684 if (oidp == rootid) /* We've hit the top */ 3685 return (0); 3686 3687 if (rw == RW_READER) { /* i.e. not the first pass */ 3688 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp); 3689 if (error) 3690 return (error); 3691 zl->zl_znode = zp; 3692 } 3693 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs), 3694 &oidp, sizeof (oidp)); 3695 rwlp = &zp->z_parent_lock; 3696 rw = RW_READER; 3697 3698 } while (zp->z_id != sdzp->z_id); 3699 3700 return (0); 3701} 3702 3703/* 3704 * Move an entry from the provided source directory to the target 3705 * directory. Change the entry name as indicated. 3706 * 3707 * IN: sdvp - Source directory containing the "old entry". 3708 * snm - Old entry name. 3709 * tdvp - Target directory to contain the "new entry". 3710 * tnm - New entry name. 3711 * cr - credentials of caller. 3712 * ct - caller context 3713 * flags - case flags 3714 * 3715 * RETURN: 0 on success, error code on failure. 3716 * 3717 * Timestamps: 3718 * sdvp,tdvp - ctime|mtime updated 3719 */ 3720/*ARGSUSED*/ 3721static int 3722zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 3723 caller_context_t *ct, int flags) 3724{ 3725 znode_t *tdzp, *szp, *tzp; 3726 znode_t *sdzp = VTOZ(sdvp); 3727 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 3728 zilog_t *zilog; 3729 vnode_t *realvp; 3730 zfs_dirlock_t *sdl, *tdl; 3731 dmu_tx_t *tx; 3732 zfs_zlock_t *zl; 3733 int cmp, serr, terr; 3734 int error = 0; 3735 int zflg = 0; 3736 3737 ZFS_ENTER(zfsvfs); 3738 ZFS_VERIFY_ZP(sdzp); 3739 zilog = zfsvfs->z_log; 3740 3741 /* 3742 * Make sure we have the real vp for the target directory. 3743 */ 3744 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 3745 tdvp = realvp; 3746 3747 tdzp = VTOZ(tdvp); 3748 ZFS_VERIFY_ZP(tdzp); 3749 3750 /* 3751 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 3752 * ctldir appear to have the same v_vfsp. 3753 */ 3754 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) { 3755 ZFS_EXIT(zfsvfs); 3756 return (SET_ERROR(EXDEV)); 3757 } 3758 3759 if (zfsvfs->z_utf8 && u8_validate(tnm, 3760 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3761 ZFS_EXIT(zfsvfs); 3762 return (SET_ERROR(EILSEQ)); 3763 } 3764 3765 if (flags & FIGNORECASE) 3766 zflg |= ZCILOOK; 3767 3768top: 3769 szp = NULL; 3770 tzp = NULL; 3771 zl = NULL; 3772 3773 /* 3774 * This is to prevent the creation of links into attribute space 3775 * by renaming a linked file into/outof an attribute directory. 3776 * See the comment in zfs_link() for why this is considered bad. 3777 */ 3778 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3779 ZFS_EXIT(zfsvfs); 3780 return (SET_ERROR(EINVAL)); 3781 } 3782 3783 /* 3784 * Lock source and target directory entries. To prevent deadlock, 3785 * a lock ordering must be defined. We lock the directory with 3786 * the smallest object id first, or if it's a tie, the one with 3787 * the lexically first name. 3788 */ 3789 if (sdzp->z_id < tdzp->z_id) { 3790 cmp = -1; 3791 } else if (sdzp->z_id > tdzp->z_id) { 3792 cmp = 1; 3793 } else { 3794 /* 3795 * First compare the two name arguments without 3796 * considering any case folding. 3797 */ 3798 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 3799 3800 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 3801 ASSERT(error == 0 || !zfsvfs->z_utf8); 3802 if (cmp == 0) { 3803 /* 3804 * POSIX: "If the old argument and the new argument 3805 * both refer to links to the same existing file, 3806 * the rename() function shall return successfully 3807 * and perform no other action." 3808 */ 3809 ZFS_EXIT(zfsvfs); 3810 return (0); 3811 } 3812 /* 3813 * If the file system is case-folding, then we may 3814 * have some more checking to do. A case-folding file 3815 * system is either supporting mixed case sensitivity 3816 * access or is completely case-insensitive. Note 3817 * that the file system is always case preserving. 3818 * 3819 * In mixed sensitivity mode case sensitive behavior 3820 * is the default. FIGNORECASE must be used to 3821 * explicitly request case insensitive behavior. 3822 * 3823 * If the source and target names provided differ only 3824 * by case (e.g., a request to rename 'tim' to 'Tim'), 3825 * we will treat this as a special case in the 3826 * case-insensitive mode: as long as the source name 3827 * is an exact match, we will allow this to proceed as 3828 * a name-change request. 3829 */ 3830 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3831 (zfsvfs->z_case == ZFS_CASE_MIXED && 3832 flags & FIGNORECASE)) && 3833 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3834 &error) == 0) { 3835 /* 3836 * case preserving rename request, require exact 3837 * name matches 3838 */ 3839 zflg |= ZCIEXACT; 3840 zflg &= ~ZCILOOK; 3841 } 3842 } 3843 3844 /* 3845 * If the source and destination directories are the same, we should 3846 * grab the z_name_lock of that directory only once. 3847 */ 3848 if (sdzp == tdzp) { 3849 zflg |= ZHAVELOCK; 3850 rw_enter(&sdzp->z_name_lock, RW_READER); 3851 } 3852 3853 if (cmp < 0) { 3854 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3855 ZEXISTS | zflg, NULL, NULL); 3856 terr = zfs_dirent_lock(&tdl, 3857 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3858 } else { 3859 terr = zfs_dirent_lock(&tdl, 3860 tdzp, tnm, &tzp, zflg, NULL, NULL); 3861 serr = zfs_dirent_lock(&sdl, 3862 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3863 NULL, NULL); 3864 } 3865 3866 if (serr) { 3867 /* 3868 * Source entry invalid or not there. 3869 */ 3870 if (!terr) { 3871 zfs_dirent_unlock(tdl); 3872 if (tzp) 3873 VN_RELE(ZTOV(tzp)); 3874 } 3875 3876 if (sdzp == tdzp) 3877 rw_exit(&sdzp->z_name_lock); 3878 3879 /* 3880 * FreeBSD: In OpenSolaris they only check if rename source is 3881 * ".." here, because "." is handled in their lookup. This is 3882 * not the case for FreeBSD, so we check for "." explicitly. 3883 */ 3884 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 3885 serr = SET_ERROR(EINVAL); 3886 ZFS_EXIT(zfsvfs); 3887 return (serr); 3888 } 3889 if (terr) { 3890 zfs_dirent_unlock(sdl); 3891 VN_RELE(ZTOV(szp)); 3892 3893 if (sdzp == tdzp) 3894 rw_exit(&sdzp->z_name_lock); 3895 3896 if (strcmp(tnm, "..") == 0) 3897 terr = SET_ERROR(EINVAL); 3898 ZFS_EXIT(zfsvfs); 3899 return (terr); 3900 } 3901 3902 /* 3903 * Must have write access at the source to remove the old entry 3904 * and write access at the target to create the new entry. 3905 * Note that if target and source are the same, this can be 3906 * done in a single check. 3907 */ 3908 3909 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3910 goto out; 3911 3912 if (ZTOV(szp)->v_type == VDIR) { 3913 /* 3914 * Check to make sure rename is valid. 3915 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3916 */ 3917 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3918 goto out; 3919 } 3920 3921 /* 3922 * Does target exist? 3923 */ 3924 if (tzp) { 3925 /* 3926 * Source and target must be the same type. 3927 */ 3928 if (ZTOV(szp)->v_type == VDIR) { 3929 if (ZTOV(tzp)->v_type != VDIR) { 3930 error = SET_ERROR(ENOTDIR); 3931 goto out; 3932 } 3933 } else { 3934 if (ZTOV(tzp)->v_type == VDIR) { 3935 error = SET_ERROR(EISDIR); 3936 goto out; 3937 } 3938 } 3939 /* 3940 * POSIX dictates that when the source and target 3941 * entries refer to the same file object, rename 3942 * must do nothing and exit without error. 3943 */ 3944 if (szp->z_id == tzp->z_id) { 3945 error = 0; 3946 goto out; 3947 } 3948 } 3949 3950 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3951 if (tzp) 3952 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3953 3954 /* 3955 * notify the target directory if it is not the same 3956 * as source directory. 3957 */ 3958 if (tdvp != sdvp) { 3959 vnevent_rename_dest_dir(tdvp, ct); 3960 } 3961 3962 tx = dmu_tx_create(zfsvfs->z_os); 3963 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3964 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 3965 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3966 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3967 if (sdzp != tdzp) { 3968 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 3969 zfs_sa_upgrade_txholds(tx, tdzp); 3970 } 3971 if (tzp) { 3972 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 3973 zfs_sa_upgrade_txholds(tx, tzp); 3974 } 3975 3976 zfs_sa_upgrade_txholds(tx, szp); 3977 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3978 error = dmu_tx_assign(tx, TXG_NOWAIT); 3979 if (error) { 3980 if (zl != NULL) 3981 zfs_rename_unlock(&zl); 3982 zfs_dirent_unlock(sdl); 3983 zfs_dirent_unlock(tdl); 3984 3985 if (sdzp == tdzp) 3986 rw_exit(&sdzp->z_name_lock); 3987 3988 VN_RELE(ZTOV(szp)); 3989 if (tzp) 3990 VN_RELE(ZTOV(tzp)); 3991 if (error == ERESTART) { 3992 dmu_tx_wait(tx); 3993 dmu_tx_abort(tx); 3994 goto top; 3995 } 3996 dmu_tx_abort(tx); 3997 ZFS_EXIT(zfsvfs); 3998 return (error); 3999 } 4000 4001 if (tzp) /* Attempt to remove the existing target */ 4002 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 4003 4004 if (error == 0) { 4005 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 4006 if (error == 0) { 4007 szp->z_pflags |= ZFS_AV_MODIFIED; 4008 4009 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 4010 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 4011 ASSERT0(error); 4012 4013 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 4014 if (error == 0) { 4015 zfs_log_rename(zilog, tx, TX_RENAME | 4016 (flags & FIGNORECASE ? TX_CI : 0), sdzp, 4017 sdl->dl_name, tdzp, tdl->dl_name, szp); 4018 4019 /* 4020 * Update path information for the target vnode 4021 */ 4022 vn_renamepath(tdvp, ZTOV(szp), tnm, 4023 strlen(tnm)); 4024 } else { 4025 /* 4026 * At this point, we have successfully created 4027 * the target name, but have failed to remove 4028 * the source name. Since the create was done 4029 * with the ZRENAMING flag, there are 4030 * complications; for one, the link count is 4031 * wrong. The easiest way to deal with this 4032 * is to remove the newly created target, and 4033 * return the original error. This must 4034 * succeed; fortunately, it is very unlikely to 4035 * fail, since we just created it. 4036 */ 4037 VERIFY3U(zfs_link_destroy(tdl, szp, tx, 4038 ZRENAMING, NULL), ==, 0); 4039 } 4040 } 4041#ifdef FREEBSD_NAMECACHE 4042 if (error == 0) { 4043 cache_purge(sdvp); 4044 cache_purge(tdvp); 4045 cache_purge(ZTOV(szp)); 4046 if (tzp) 4047 cache_purge(ZTOV(tzp)); 4048 } 4049#endif 4050 } 4051 4052 dmu_tx_commit(tx); 4053out: 4054 if (zl != NULL) 4055 zfs_rename_unlock(&zl); 4056 4057 zfs_dirent_unlock(sdl); 4058 zfs_dirent_unlock(tdl); 4059 4060 if (sdzp == tdzp) 4061 rw_exit(&sdzp->z_name_lock); 4062 4063 4064 VN_RELE(ZTOV(szp)); 4065 if (tzp) 4066 VN_RELE(ZTOV(tzp)); 4067 4068 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4069 zil_commit(zilog, 0); 4070 4071 ZFS_EXIT(zfsvfs); 4072 4073 return (error); 4074} 4075 4076/* 4077 * Insert the indicated symbolic reference entry into the directory. 4078 * 4079 * IN: dvp - Directory to contain new symbolic link. 4080 * link - Name for new symlink entry. 4081 * vap - Attributes of new entry. 4082 * cr - credentials of caller. 4083 * ct - caller context 4084 * flags - case flags 4085 * 4086 * RETURN: 0 on success, error code on failure. 4087 * 4088 * Timestamps: 4089 * dvp - ctime|mtime updated 4090 */ 4091/*ARGSUSED*/ 4092static int 4093zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, 4094 cred_t *cr, kthread_t *td) 4095{ 4096 znode_t *zp, *dzp = VTOZ(dvp); 4097 zfs_dirlock_t *dl; 4098 dmu_tx_t *tx; 4099 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4100 zilog_t *zilog; 4101 uint64_t len = strlen(link); 4102 int error; 4103 int zflg = ZNEW; 4104 zfs_acl_ids_t acl_ids; 4105 boolean_t fuid_dirtied; 4106 uint64_t txtype = TX_SYMLINK; 4107 int flags = 0; 4108 4109 ASSERT(vap->va_type == VLNK); 4110 4111 ZFS_ENTER(zfsvfs); 4112 ZFS_VERIFY_ZP(dzp); 4113 zilog = zfsvfs->z_log; 4114 4115 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 4116 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4117 ZFS_EXIT(zfsvfs); 4118 return (SET_ERROR(EILSEQ)); 4119 } 4120 if (flags & FIGNORECASE) 4121 zflg |= ZCILOOK; 4122 4123 if (len > MAXPATHLEN) { 4124 ZFS_EXIT(zfsvfs); 4125 return (SET_ERROR(ENAMETOOLONG)); 4126 } 4127 4128 if ((error = zfs_acl_ids_create(dzp, 0, 4129 vap, cr, NULL, &acl_ids)) != 0) { 4130 ZFS_EXIT(zfsvfs); 4131 return (error); 4132 } 4133top: 4134 /* 4135 * Attempt to lock directory; fail if entry already exists. 4136 */ 4137 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 4138 if (error) { 4139 zfs_acl_ids_free(&acl_ids); 4140 ZFS_EXIT(zfsvfs); 4141 return (error); 4142 } 4143 4144 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4145 zfs_acl_ids_free(&acl_ids); 4146 zfs_dirent_unlock(dl); 4147 ZFS_EXIT(zfsvfs); 4148 return (error); 4149 } 4150 4151 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 4152 zfs_acl_ids_free(&acl_ids); 4153 zfs_dirent_unlock(dl); 4154 ZFS_EXIT(zfsvfs); 4155 return (SET_ERROR(EDQUOT)); 4156 } 4157 tx = dmu_tx_create(zfsvfs->z_os); 4158 fuid_dirtied = zfsvfs->z_fuid_dirty; 4159 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 4160 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4161 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 4162 ZFS_SA_BASE_ATTR_SIZE + len); 4163 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 4164 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 4165 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 4166 acl_ids.z_aclp->z_acl_bytes); 4167 } 4168 if (fuid_dirtied) 4169 zfs_fuid_txhold(zfsvfs, tx); 4170 error = dmu_tx_assign(tx, TXG_NOWAIT); 4171 if (error) { 4172 zfs_dirent_unlock(dl); 4173 if (error == ERESTART) { 4174 dmu_tx_wait(tx); 4175 dmu_tx_abort(tx); 4176 goto top; 4177 } 4178 zfs_acl_ids_free(&acl_ids); 4179 dmu_tx_abort(tx); 4180 ZFS_EXIT(zfsvfs); 4181 return (error); 4182 } 4183 4184 /* 4185 * Create a new object for the symlink. 4186 * for version 4 ZPL datsets the symlink will be an SA attribute 4187 */ 4188 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 4189 4190 if (fuid_dirtied) 4191 zfs_fuid_sync(zfsvfs, tx); 4192 4193 mutex_enter(&zp->z_lock); 4194 if (zp->z_is_sa) 4195 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 4196 link, len, tx); 4197 else 4198 zfs_sa_symlink(zp, link, len, tx); 4199 mutex_exit(&zp->z_lock); 4200 4201 zp->z_size = len; 4202 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 4203 &zp->z_size, sizeof (zp->z_size), tx); 4204 /* 4205 * Insert the new object into the directory. 4206 */ 4207 (void) zfs_link_create(dl, zp, tx, ZNEW); 4208 4209 if (flags & FIGNORECASE) 4210 txtype |= TX_CI; 4211 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 4212 *vpp = ZTOV(zp); 4213 4214 zfs_acl_ids_free(&acl_ids); 4215 4216 dmu_tx_commit(tx); 4217 4218 zfs_dirent_unlock(dl); 4219 4220 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4221 zil_commit(zilog, 0); 4222 4223 ZFS_EXIT(zfsvfs); 4224 return (error); 4225} 4226 4227/* 4228 * Return, in the buffer contained in the provided uio structure, 4229 * the symbolic path referred to by vp. 4230 * 4231 * IN: vp - vnode of symbolic link. 4232 * uio - structure to contain the link path. 4233 * cr - credentials of caller. 4234 * ct - caller context 4235 * 4236 * OUT: uio - structure containing the link path. 4237 * 4238 * RETURN: 0 on success, error code on failure. 4239 * 4240 * Timestamps: 4241 * vp - atime updated 4242 */ 4243/* ARGSUSED */ 4244static int 4245zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 4246{ 4247 znode_t *zp = VTOZ(vp); 4248 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4249 int error; 4250 4251 ZFS_ENTER(zfsvfs); 4252 ZFS_VERIFY_ZP(zp); 4253 4254 mutex_enter(&zp->z_lock); 4255 if (zp->z_is_sa) 4256 error = sa_lookup_uio(zp->z_sa_hdl, 4257 SA_ZPL_SYMLINK(zfsvfs), uio); 4258 else 4259 error = zfs_sa_readlink(zp, uio); 4260 mutex_exit(&zp->z_lock); 4261 4262 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4263 4264 ZFS_EXIT(zfsvfs); 4265 return (error); 4266} 4267 4268/* 4269 * Insert a new entry into directory tdvp referencing svp. 4270 * 4271 * IN: tdvp - Directory to contain new entry. 4272 * svp - vnode of new entry. 4273 * name - name of new entry. 4274 * cr - credentials of caller. 4275 * ct - caller context 4276 * 4277 * RETURN: 0 on success, error code on failure. 4278 * 4279 * Timestamps: 4280 * tdvp - ctime|mtime updated 4281 * svp - ctime updated 4282 */ 4283/* ARGSUSED */ 4284static int 4285zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 4286 caller_context_t *ct, int flags) 4287{ 4288 znode_t *dzp = VTOZ(tdvp); 4289 znode_t *tzp, *szp; 4290 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4291 zilog_t *zilog; 4292 zfs_dirlock_t *dl; 4293 dmu_tx_t *tx; 4294 vnode_t *realvp; 4295 int error; 4296 int zf = ZNEW; 4297 uint64_t parent; 4298 uid_t owner; 4299 4300 ASSERT(tdvp->v_type == VDIR); 4301 4302 ZFS_ENTER(zfsvfs); 4303 ZFS_VERIFY_ZP(dzp); 4304 zilog = zfsvfs->z_log; 4305 4306 if (VOP_REALVP(svp, &realvp, ct) == 0) 4307 svp = realvp; 4308 4309 /* 4310 * POSIX dictates that we return EPERM here. 4311 * Better choices include ENOTSUP or EISDIR. 4312 */ 4313 if (svp->v_type == VDIR) { 4314 ZFS_EXIT(zfsvfs); 4315 return (SET_ERROR(EPERM)); 4316 } 4317 4318 szp = VTOZ(svp); 4319 ZFS_VERIFY_ZP(szp); 4320 4321 /* 4322 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 4323 * ctldir appear to have the same v_vfsp. 4324 */ 4325 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) { 4326 ZFS_EXIT(zfsvfs); 4327 return (SET_ERROR(EXDEV)); 4328 } 4329 4330 /* Prevent links to .zfs/shares files */ 4331 4332 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 4333 &parent, sizeof (uint64_t))) != 0) { 4334 ZFS_EXIT(zfsvfs); 4335 return (error); 4336 } 4337 if (parent == zfsvfs->z_shares_dir) { 4338 ZFS_EXIT(zfsvfs); 4339 return (SET_ERROR(EPERM)); 4340 } 4341 4342 if (zfsvfs->z_utf8 && u8_validate(name, 4343 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4344 ZFS_EXIT(zfsvfs); 4345 return (SET_ERROR(EILSEQ)); 4346 } 4347 if (flags & FIGNORECASE) 4348 zf |= ZCILOOK; 4349 4350 /* 4351 * We do not support links between attributes and non-attributes 4352 * because of the potential security risk of creating links 4353 * into "normal" file space in order to circumvent restrictions 4354 * imposed in attribute space. 4355 */ 4356 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 4357 ZFS_EXIT(zfsvfs); 4358 return (SET_ERROR(EINVAL)); 4359 } 4360 4361 4362 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 4363 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) { 4364 ZFS_EXIT(zfsvfs); 4365 return (SET_ERROR(EPERM)); 4366 } 4367 4368 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4369 ZFS_EXIT(zfsvfs); 4370 return (error); 4371 } 4372 4373top: 4374 /* 4375 * Attempt to lock directory; fail if entry already exists. 4376 */ 4377 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 4378 if (error) { 4379 ZFS_EXIT(zfsvfs); 4380 return (error); 4381 } 4382 4383 tx = dmu_tx_create(zfsvfs->z_os); 4384 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4385 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4386 zfs_sa_upgrade_txholds(tx, szp); 4387 zfs_sa_upgrade_txholds(tx, dzp); 4388 error = dmu_tx_assign(tx, TXG_NOWAIT); 4389 if (error) { 4390 zfs_dirent_unlock(dl); 4391 if (error == ERESTART) { 4392 dmu_tx_wait(tx); 4393 dmu_tx_abort(tx); 4394 goto top; 4395 } 4396 dmu_tx_abort(tx); 4397 ZFS_EXIT(zfsvfs); 4398 return (error); 4399 } 4400 4401 error = zfs_link_create(dl, szp, tx, 0); 4402 4403 if (error == 0) { 4404 uint64_t txtype = TX_LINK; 4405 if (flags & FIGNORECASE) 4406 txtype |= TX_CI; 4407 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 4408 } 4409 4410 dmu_tx_commit(tx); 4411 4412 zfs_dirent_unlock(dl); 4413 4414 if (error == 0) { 4415 vnevent_link(svp, ct); 4416 } 4417 4418 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4419 zil_commit(zilog, 0); 4420 4421 ZFS_EXIT(zfsvfs); 4422 return (error); 4423} 4424 4425#ifdef sun 4426/* 4427 * zfs_null_putapage() is used when the file system has been force 4428 * unmounted. It just drops the pages. 4429 */ 4430/* ARGSUSED */ 4431static int 4432zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4433 size_t *lenp, int flags, cred_t *cr) 4434{ 4435 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 4436 return (0); 4437} 4438 4439/* 4440 * Push a page out to disk, klustering if possible. 4441 * 4442 * IN: vp - file to push page to. 4443 * pp - page to push. 4444 * flags - additional flags. 4445 * cr - credentials of caller. 4446 * 4447 * OUT: offp - start of range pushed. 4448 * lenp - len of range pushed. 4449 * 4450 * RETURN: 0 on success, error code on failure. 4451 * 4452 * NOTE: callers must have locked the page to be pushed. On 4453 * exit, the page (and all other pages in the kluster) must be 4454 * unlocked. 4455 */ 4456/* ARGSUSED */ 4457static int 4458zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4459 size_t *lenp, int flags, cred_t *cr) 4460{ 4461 znode_t *zp = VTOZ(vp); 4462 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4463 dmu_tx_t *tx; 4464 u_offset_t off, koff; 4465 size_t len, klen; 4466 int err; 4467 4468 off = pp->p_offset; 4469 len = PAGESIZE; 4470 /* 4471 * If our blocksize is bigger than the page size, try to kluster 4472 * multiple pages so that we write a full block (thus avoiding 4473 * a read-modify-write). 4474 */ 4475 if (off < zp->z_size && zp->z_blksz > PAGESIZE) { 4476 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 4477 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 4478 ASSERT(koff <= zp->z_size); 4479 if (koff + klen > zp->z_size) 4480 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); 4481 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 4482 } 4483 ASSERT3U(btop(len), ==, btopr(len)); 4484 4485 /* 4486 * Can't push pages past end-of-file. 4487 */ 4488 if (off >= zp->z_size) { 4489 /* ignore all pages */ 4490 err = 0; 4491 goto out; 4492 } else if (off + len > zp->z_size) { 4493 int npages = btopr(zp->z_size - off); 4494 page_t *trunc; 4495 4496 page_list_break(&pp, &trunc, npages); 4497 /* ignore pages past end of file */ 4498 if (trunc) 4499 pvn_write_done(trunc, flags); 4500 len = zp->z_size - off; 4501 } 4502 4503 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4504 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4505 err = SET_ERROR(EDQUOT); 4506 goto out; 4507 } 4508top: 4509 tx = dmu_tx_create(zfsvfs->z_os); 4510 dmu_tx_hold_write(tx, zp->z_id, off, len); 4511 4512 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4513 zfs_sa_upgrade_txholds(tx, zp); 4514 err = dmu_tx_assign(tx, TXG_NOWAIT); 4515 if (err != 0) { 4516 if (err == ERESTART) { 4517 dmu_tx_wait(tx); 4518 dmu_tx_abort(tx); 4519 goto top; 4520 } 4521 dmu_tx_abort(tx); 4522 goto out; 4523 } 4524 4525 if (zp->z_blksz <= PAGESIZE) { 4526 caddr_t va = zfs_map_page(pp, S_READ); 4527 ASSERT3U(len, <=, PAGESIZE); 4528 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 4529 zfs_unmap_page(pp, va); 4530 } else { 4531 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 4532 } 4533 4534 if (err == 0) { 4535 uint64_t mtime[2], ctime[2]; 4536 sa_bulk_attr_t bulk[3]; 4537 int count = 0; 4538 4539 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4540 &mtime, 16); 4541 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4542 &ctime, 16); 4543 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4544 &zp->z_pflags, 8); 4545 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4546 B_TRUE); 4547 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4548 } 4549 dmu_tx_commit(tx); 4550 4551out: 4552 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 4553 if (offp) 4554 *offp = off; 4555 if (lenp) 4556 *lenp = len; 4557 4558 return (err); 4559} 4560 4561/* 4562 * Copy the portion of the file indicated from pages into the file. 4563 * The pages are stored in a page list attached to the files vnode. 4564 * 4565 * IN: vp - vnode of file to push page data to. 4566 * off - position in file to put data. 4567 * len - amount of data to write. 4568 * flags - flags to control the operation. 4569 * cr - credentials of caller. 4570 * ct - caller context. 4571 * 4572 * RETURN: 0 on success, error code on failure. 4573 * 4574 * Timestamps: 4575 * vp - ctime|mtime updated 4576 */ 4577/*ARGSUSED*/ 4578static int 4579zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4580 caller_context_t *ct) 4581{ 4582 znode_t *zp = VTOZ(vp); 4583 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4584 page_t *pp; 4585 size_t io_len; 4586 u_offset_t io_off; 4587 uint_t blksz; 4588 rl_t *rl; 4589 int error = 0; 4590 4591 ZFS_ENTER(zfsvfs); 4592 ZFS_VERIFY_ZP(zp); 4593 4594 /* 4595 * Align this request to the file block size in case we kluster. 4596 * XXX - this can result in pretty aggresive locking, which can 4597 * impact simultanious read/write access. One option might be 4598 * to break up long requests (len == 0) into block-by-block 4599 * operations to get narrower locking. 4600 */ 4601 blksz = zp->z_blksz; 4602 if (ISP2(blksz)) 4603 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 4604 else 4605 io_off = 0; 4606 if (len > 0 && ISP2(blksz)) 4607 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 4608 else 4609 io_len = 0; 4610 4611 if (io_len == 0) { 4612 /* 4613 * Search the entire vp list for pages >= io_off. 4614 */ 4615 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 4616 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 4617 goto out; 4618 } 4619 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 4620 4621 if (off > zp->z_size) { 4622 /* past end of file */ 4623 zfs_range_unlock(rl); 4624 ZFS_EXIT(zfsvfs); 4625 return (0); 4626 } 4627 4628 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); 4629 4630 for (off = io_off; io_off < off + len; io_off += io_len) { 4631 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 4632 pp = page_lookup(vp, io_off, 4633 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 4634 } else { 4635 pp = page_lookup_nowait(vp, io_off, 4636 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 4637 } 4638 4639 if (pp != NULL && pvn_getdirty(pp, flags)) { 4640 int err; 4641 4642 /* 4643 * Found a dirty page to push 4644 */ 4645 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 4646 if (err) 4647 error = err; 4648 } else { 4649 io_len = PAGESIZE; 4650 } 4651 } 4652out: 4653 zfs_range_unlock(rl); 4654 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4655 zil_commit(zfsvfs->z_log, zp->z_id); 4656 ZFS_EXIT(zfsvfs); 4657 return (error); 4658} 4659#endif /* sun */ 4660 4661/*ARGSUSED*/ 4662void 4663zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4664{ 4665 znode_t *zp = VTOZ(vp); 4666 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4667 int error; 4668 4669 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4670 if (zp->z_sa_hdl == NULL) { 4671 /* 4672 * The fs has been unmounted, or we did a 4673 * suspend/resume and this file no longer exists. 4674 */ 4675 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4676 vrecycle(vp); 4677 return; 4678 } 4679 4680 mutex_enter(&zp->z_lock); 4681 if (zp->z_unlinked) { 4682 /* 4683 * Fast path to recycle a vnode of a removed file. 4684 */ 4685 mutex_exit(&zp->z_lock); 4686 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4687 vrecycle(vp); 4688 return; 4689 } 4690 mutex_exit(&zp->z_lock); 4691 4692 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4693 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4694 4695 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4696 zfs_sa_upgrade_txholds(tx, zp); 4697 error = dmu_tx_assign(tx, TXG_WAIT); 4698 if (error) { 4699 dmu_tx_abort(tx); 4700 } else { 4701 mutex_enter(&zp->z_lock); 4702 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4703 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4704 zp->z_atime_dirty = 0; 4705 mutex_exit(&zp->z_lock); 4706 dmu_tx_commit(tx); 4707 } 4708 } 4709 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4710} 4711 4712#ifdef sun 4713/* 4714 * Bounds-check the seek operation. 4715 * 4716 * IN: vp - vnode seeking within 4717 * ooff - old file offset 4718 * noffp - pointer to new file offset 4719 * ct - caller context 4720 * 4721 * RETURN: 0 on success, EINVAL if new offset invalid. 4722 */ 4723/* ARGSUSED */ 4724static int 4725zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 4726 caller_context_t *ct) 4727{ 4728 if (vp->v_type == VDIR) 4729 return (0); 4730 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4731} 4732 4733/* 4734 * Pre-filter the generic locking function to trap attempts to place 4735 * a mandatory lock on a memory mapped file. 4736 */ 4737static int 4738zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 4739 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 4740{ 4741 znode_t *zp = VTOZ(vp); 4742 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4743 4744 ZFS_ENTER(zfsvfs); 4745 ZFS_VERIFY_ZP(zp); 4746 4747 /* 4748 * We are following the UFS semantics with respect to mapcnt 4749 * here: If we see that the file is mapped already, then we will 4750 * return an error, but we don't worry about races between this 4751 * function and zfs_map(). 4752 */ 4753 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { 4754 ZFS_EXIT(zfsvfs); 4755 return (SET_ERROR(EAGAIN)); 4756 } 4757 ZFS_EXIT(zfsvfs); 4758 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4759} 4760 4761/* 4762 * If we can't find a page in the cache, we will create a new page 4763 * and fill it with file data. For efficiency, we may try to fill 4764 * multiple pages at once (klustering) to fill up the supplied page 4765 * list. Note that the pages to be filled are held with an exclusive 4766 * lock to prevent access by other threads while they are being filled. 4767 */ 4768static int 4769zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 4770 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 4771{ 4772 znode_t *zp = VTOZ(vp); 4773 page_t *pp, *cur_pp; 4774 objset_t *os = zp->z_zfsvfs->z_os; 4775 u_offset_t io_off, total; 4776 size_t io_len; 4777 int err; 4778 4779 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4780 /* 4781 * We only have a single page, don't bother klustering 4782 */ 4783 io_off = off; 4784 io_len = PAGESIZE; 4785 pp = page_create_va(vp, io_off, io_len, 4786 PG_EXCL | PG_WAIT, seg, addr); 4787 } else { 4788 /* 4789 * Try to find enough pages to fill the page list 4790 */ 4791 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4792 &io_len, off, plsz, 0); 4793 } 4794 if (pp == NULL) { 4795 /* 4796 * The page already exists, nothing to do here. 4797 */ 4798 *pl = NULL; 4799 return (0); 4800 } 4801 4802 /* 4803 * Fill the pages in the kluster. 4804 */ 4805 cur_pp = pp; 4806 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4807 caddr_t va; 4808 4809 ASSERT3U(io_off, ==, cur_pp->p_offset); 4810 va = zfs_map_page(cur_pp, S_WRITE); 4811 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4812 DMU_READ_PREFETCH); 4813 zfs_unmap_page(cur_pp, va); 4814 if (err) { 4815 /* On error, toss the entire kluster */ 4816 pvn_read_done(pp, B_ERROR); 4817 /* convert checksum errors into IO errors */ 4818 if (err == ECKSUM) 4819 err = SET_ERROR(EIO); 4820 return (err); 4821 } 4822 cur_pp = cur_pp->p_next; 4823 } 4824 4825 /* 4826 * Fill in the page list array from the kluster starting 4827 * from the desired offset `off'. 4828 * NOTE: the page list will always be null terminated. 4829 */ 4830 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4831 ASSERT(pl == NULL || (*pl)->p_offset == off); 4832 4833 return (0); 4834} 4835 4836/* 4837 * Return pointers to the pages for the file region [off, off + len] 4838 * in the pl array. If plsz is greater than len, this function may 4839 * also return page pointers from after the specified region 4840 * (i.e. the region [off, off + plsz]). These additional pages are 4841 * only returned if they are already in the cache, or were created as 4842 * part of a klustered read. 4843 * 4844 * IN: vp - vnode of file to get data from. 4845 * off - position in file to get data from. 4846 * len - amount of data to retrieve. 4847 * plsz - length of provided page list. 4848 * seg - segment to obtain pages for. 4849 * addr - virtual address of fault. 4850 * rw - mode of created pages. 4851 * cr - credentials of caller. 4852 * ct - caller context. 4853 * 4854 * OUT: protp - protection mode of created pages. 4855 * pl - list of pages created. 4856 * 4857 * RETURN: 0 on success, error code on failure. 4858 * 4859 * Timestamps: 4860 * vp - atime updated 4861 */ 4862/* ARGSUSED */ 4863static int 4864zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4865 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4866 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4867{ 4868 znode_t *zp = VTOZ(vp); 4869 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4870 page_t **pl0 = pl; 4871 int err = 0; 4872 4873 /* we do our own caching, faultahead is unnecessary */ 4874 if (pl == NULL) 4875 return (0); 4876 else if (len > plsz) 4877 len = plsz; 4878 else 4879 len = P2ROUNDUP(len, PAGESIZE); 4880 ASSERT(plsz >= len); 4881 4882 ZFS_ENTER(zfsvfs); 4883 ZFS_VERIFY_ZP(zp); 4884 4885 if (protp) 4886 *protp = PROT_ALL; 4887 4888 /* 4889 * Loop through the requested range [off, off + len) looking 4890 * for pages. If we don't find a page, we will need to create 4891 * a new page and fill it with data from the file. 4892 */ 4893 while (len > 0) { 4894 if (*pl = page_lookup(vp, off, SE_SHARED)) 4895 *(pl+1) = NULL; 4896 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 4897 goto out; 4898 while (*pl) { 4899 ASSERT3U((*pl)->p_offset, ==, off); 4900 off += PAGESIZE; 4901 addr += PAGESIZE; 4902 if (len > 0) { 4903 ASSERT3U(len, >=, PAGESIZE); 4904 len -= PAGESIZE; 4905 } 4906 ASSERT3U(plsz, >=, PAGESIZE); 4907 plsz -= PAGESIZE; 4908 pl++; 4909 } 4910 } 4911 4912 /* 4913 * Fill out the page array with any pages already in the cache. 4914 */ 4915 while (plsz > 0 && 4916 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 4917 off += PAGESIZE; 4918 plsz -= PAGESIZE; 4919 } 4920out: 4921 if (err) { 4922 /* 4923 * Release any pages we have previously locked. 4924 */ 4925 while (pl > pl0) 4926 page_unlock(*--pl); 4927 } else { 4928 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4929 } 4930 4931 *pl = NULL; 4932 4933 ZFS_EXIT(zfsvfs); 4934 return (err); 4935} 4936 4937/* 4938 * Request a memory map for a section of a file. This code interacts 4939 * with common code and the VM system as follows: 4940 * 4941 * - common code calls mmap(), which ends up in smmap_common() 4942 * - this calls VOP_MAP(), which takes you into (say) zfs 4943 * - zfs_map() calls as_map(), passing segvn_create() as the callback 4944 * - segvn_create() creates the new segment and calls VOP_ADDMAP() 4945 * - zfs_addmap() updates z_mapcnt 4946 */ 4947/*ARGSUSED*/ 4948static int 4949zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4950 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4951 caller_context_t *ct) 4952{ 4953 znode_t *zp = VTOZ(vp); 4954 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4955 segvn_crargs_t vn_a; 4956 int error; 4957 4958 ZFS_ENTER(zfsvfs); 4959 ZFS_VERIFY_ZP(zp); 4960 4961 if ((prot & PROT_WRITE) && (zp->z_pflags & 4962 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 4963 ZFS_EXIT(zfsvfs); 4964 return (SET_ERROR(EPERM)); 4965 } 4966 4967 if ((prot & (PROT_READ | PROT_EXEC)) && 4968 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 4969 ZFS_EXIT(zfsvfs); 4970 return (SET_ERROR(EACCES)); 4971 } 4972 4973 if (vp->v_flag & VNOMAP) { 4974 ZFS_EXIT(zfsvfs); 4975 return (SET_ERROR(ENOSYS)); 4976 } 4977 4978 if (off < 0 || len > MAXOFFSET_T - off) { 4979 ZFS_EXIT(zfsvfs); 4980 return (SET_ERROR(ENXIO)); 4981 } 4982 4983 if (vp->v_type != VREG) { 4984 ZFS_EXIT(zfsvfs); 4985 return (SET_ERROR(ENODEV)); 4986 } 4987 4988 /* 4989 * If file is locked, disallow mapping. 4990 */ 4991 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { 4992 ZFS_EXIT(zfsvfs); 4993 return (SET_ERROR(EAGAIN)); 4994 } 4995 4996 as_rangelock(as); 4997 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 4998 if (error != 0) { 4999 as_rangeunlock(as); 5000 ZFS_EXIT(zfsvfs); 5001 return (error); 5002 } 5003 5004 vn_a.vp = vp; 5005 vn_a.offset = (u_offset_t)off; 5006 vn_a.type = flags & MAP_TYPE; 5007 vn_a.prot = prot; 5008 vn_a.maxprot = maxprot; 5009 vn_a.cred = cr; 5010 vn_a.amp = NULL; 5011 vn_a.flags = flags & ~MAP_TYPE; 5012 vn_a.szc = 0; 5013 vn_a.lgrp_mem_policy_flags = 0; 5014 5015 error = as_map(as, *addrp, len, segvn_create, &vn_a); 5016 5017 as_rangeunlock(as); 5018 ZFS_EXIT(zfsvfs); 5019 return (error); 5020} 5021 5022/* ARGSUSED */ 5023static int 5024zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5025 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 5026 caller_context_t *ct) 5027{ 5028 uint64_t pages = btopr(len); 5029 5030 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 5031 return (0); 5032} 5033 5034/* 5035 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 5036 * more accurate mtime for the associated file. Since we don't have a way of 5037 * detecting when the data was actually modified, we have to resort to 5038 * heuristics. If an explicit msync() is done, then we mark the mtime when the 5039 * last page is pushed. The problem occurs when the msync() call is omitted, 5040 * which by far the most common case: 5041 * 5042 * open() 5043 * mmap() 5044 * <modify memory> 5045 * munmap() 5046 * close() 5047 * <time lapse> 5048 * putpage() via fsflush 5049 * 5050 * If we wait until fsflush to come along, we can have a modification time that 5051 * is some arbitrary point in the future. In order to prevent this in the 5052 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 5053 * torn down. 5054 */ 5055/* ARGSUSED */ 5056static int 5057zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5058 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 5059 caller_context_t *ct) 5060{ 5061 uint64_t pages = btopr(len); 5062 5063 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 5064 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 5065 5066 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 5067 vn_has_cached_data(vp)) 5068 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 5069 5070 return (0); 5071} 5072 5073/* 5074 * Free or allocate space in a file. Currently, this function only 5075 * supports the `F_FREESP' command. However, this command is somewhat 5076 * misnamed, as its functionality includes the ability to allocate as 5077 * well as free space. 5078 * 5079 * IN: vp - vnode of file to free data in. 5080 * cmd - action to take (only F_FREESP supported). 5081 * bfp - section of file to free/alloc. 5082 * flag - current file open mode flags. 5083 * offset - current file offset. 5084 * cr - credentials of caller [UNUSED]. 5085 * ct - caller context. 5086 * 5087 * RETURN: 0 on success, error code on failure. 5088 * 5089 * Timestamps: 5090 * vp - ctime|mtime updated 5091 */ 5092/* ARGSUSED */ 5093static int 5094zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 5095 offset_t offset, cred_t *cr, caller_context_t *ct) 5096{ 5097 znode_t *zp = VTOZ(vp); 5098 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5099 uint64_t off, len; 5100 int error; 5101 5102 ZFS_ENTER(zfsvfs); 5103 ZFS_VERIFY_ZP(zp); 5104 5105 if (cmd != F_FREESP) { 5106 ZFS_EXIT(zfsvfs); 5107 return (SET_ERROR(EINVAL)); 5108 } 5109 5110 if (error = convoff(vp, bfp, 0, offset)) { 5111 ZFS_EXIT(zfsvfs); 5112 return (error); 5113 } 5114 5115 if (bfp->l_len < 0) { 5116 ZFS_EXIT(zfsvfs); 5117 return (SET_ERROR(EINVAL)); 5118 } 5119 5120 off = bfp->l_start; 5121 len = bfp->l_len; /* 0 means from off to end of file */ 5122 5123 error = zfs_freesp(zp, off, len, flag, TRUE); 5124 5125 ZFS_EXIT(zfsvfs); 5126 return (error); 5127} 5128#endif /* sun */ 5129 5130CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 5131CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 5132 5133/*ARGSUSED*/ 5134static int 5135zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 5136{ 5137 znode_t *zp = VTOZ(vp); 5138 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5139 uint32_t gen; 5140 uint64_t gen64; 5141 uint64_t object = zp->z_id; 5142 zfid_short_t *zfid; 5143 int size, i, error; 5144 5145 ZFS_ENTER(zfsvfs); 5146 ZFS_VERIFY_ZP(zp); 5147 5148 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 5149 &gen64, sizeof (uint64_t))) != 0) { 5150 ZFS_EXIT(zfsvfs); 5151 return (error); 5152 } 5153 5154 gen = (uint32_t)gen64; 5155 5156 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 5157 5158#ifdef illumos 5159 if (fidp->fid_len < size) { 5160 fidp->fid_len = size; 5161 ZFS_EXIT(zfsvfs); 5162 return (SET_ERROR(ENOSPC)); 5163 } 5164#else 5165 fidp->fid_len = size; 5166#endif 5167 5168 zfid = (zfid_short_t *)fidp; 5169 5170 zfid->zf_len = size; 5171 5172 for (i = 0; i < sizeof (zfid->zf_object); i++) 5173 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 5174 5175 /* Must have a non-zero generation number to distinguish from .zfs */ 5176 if (gen == 0) 5177 gen = 1; 5178 for (i = 0; i < sizeof (zfid->zf_gen); i++) 5179 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 5180 5181 if (size == LONG_FID_LEN) { 5182 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 5183 zfid_long_t *zlfid; 5184 5185 zlfid = (zfid_long_t *)fidp; 5186 5187 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 5188 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 5189 5190 /* XXX - this should be the generation number for the objset */ 5191 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 5192 zlfid->zf_setgen[i] = 0; 5193 } 5194 5195 ZFS_EXIT(zfsvfs); 5196 return (0); 5197} 5198 5199static int 5200zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 5201 caller_context_t *ct) 5202{ 5203 znode_t *zp, *xzp; 5204 zfsvfs_t *zfsvfs; 5205 zfs_dirlock_t *dl; 5206 int error; 5207 5208 switch (cmd) { 5209 case _PC_LINK_MAX: 5210 *valp = INT_MAX; 5211 return (0); 5212 5213 case _PC_FILESIZEBITS: 5214 *valp = 64; 5215 return (0); 5216#ifdef sun 5217 case _PC_XATTR_EXISTS: 5218 zp = VTOZ(vp); 5219 zfsvfs = zp->z_zfsvfs; 5220 ZFS_ENTER(zfsvfs); 5221 ZFS_VERIFY_ZP(zp); 5222 *valp = 0; 5223 error = zfs_dirent_lock(&dl, zp, "", &xzp, 5224 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 5225 if (error == 0) { 5226 zfs_dirent_unlock(dl); 5227 if (!zfs_dirempty(xzp)) 5228 *valp = 1; 5229 VN_RELE(ZTOV(xzp)); 5230 } else if (error == ENOENT) { 5231 /* 5232 * If there aren't extended attributes, it's the 5233 * same as having zero of them. 5234 */ 5235 error = 0; 5236 } 5237 ZFS_EXIT(zfsvfs); 5238 return (error); 5239 5240 case _PC_SATTR_ENABLED: 5241 case _PC_SATTR_EXISTS: 5242 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 5243 (vp->v_type == VREG || vp->v_type == VDIR); 5244 return (0); 5245 5246 case _PC_ACCESS_FILTERING: 5247 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 5248 vp->v_type == VDIR; 5249 return (0); 5250 5251 case _PC_ACL_ENABLED: 5252 *valp = _ACL_ACE_ENABLED; 5253 return (0); 5254#endif /* sun */ 5255 case _PC_MIN_HOLE_SIZE: 5256 *valp = (int)SPA_MINBLOCKSIZE; 5257 return (0); 5258#ifdef sun 5259 case _PC_TIMESTAMP_RESOLUTION: 5260 /* nanosecond timestamp resolution */ 5261 *valp = 1L; 5262 return (0); 5263#endif /* sun */ 5264 case _PC_ACL_EXTENDED: 5265 *valp = 0; 5266 return (0); 5267 5268 case _PC_ACL_NFS4: 5269 *valp = 1; 5270 return (0); 5271 5272 case _PC_ACL_PATH_MAX: 5273 *valp = ACL_MAX_ENTRIES; 5274 return (0); 5275 5276 default: 5277 return (EOPNOTSUPP); 5278 } 5279} 5280 5281/*ARGSUSED*/ 5282static int 5283zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5284 caller_context_t *ct) 5285{ 5286 znode_t *zp = VTOZ(vp); 5287 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5288 int error; 5289 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5290 5291 ZFS_ENTER(zfsvfs); 5292 ZFS_VERIFY_ZP(zp); 5293 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 5294 ZFS_EXIT(zfsvfs); 5295 5296 return (error); 5297} 5298 5299/*ARGSUSED*/ 5300int 5301zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5302 caller_context_t *ct) 5303{ 5304 znode_t *zp = VTOZ(vp); 5305 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5306 int error; 5307 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5308 zilog_t *zilog = zfsvfs->z_log; 5309 5310 ZFS_ENTER(zfsvfs); 5311 ZFS_VERIFY_ZP(zp); 5312 5313 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 5314 5315 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 5316 zil_commit(zilog, 0); 5317 5318 ZFS_EXIT(zfsvfs); 5319 return (error); 5320} 5321 5322#ifdef sun 5323/* 5324 * The smallest read we may consider to loan out an arcbuf. 5325 * This must be a power of 2. 5326 */ 5327int zcr_blksz_min = (1 << 10); /* 1K */ 5328/* 5329 * If set to less than the file block size, allow loaning out of an 5330 * arcbuf for a partial block read. This must be a power of 2. 5331 */ 5332int zcr_blksz_max = (1 << 17); /* 128K */ 5333 5334/*ARGSUSED*/ 5335static int 5336zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr, 5337 caller_context_t *ct) 5338{ 5339 znode_t *zp = VTOZ(vp); 5340 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5341 int max_blksz = zfsvfs->z_max_blksz; 5342 uio_t *uio = &xuio->xu_uio; 5343 ssize_t size = uio->uio_resid; 5344 offset_t offset = uio->uio_loffset; 5345 int blksz; 5346 int fullblk, i; 5347 arc_buf_t *abuf; 5348 ssize_t maxsize; 5349 int preamble, postamble; 5350 5351 if (xuio->xu_type != UIOTYPE_ZEROCOPY) 5352 return (SET_ERROR(EINVAL)); 5353 5354 ZFS_ENTER(zfsvfs); 5355 ZFS_VERIFY_ZP(zp); 5356 switch (ioflag) { 5357 case UIO_WRITE: 5358 /* 5359 * Loan out an arc_buf for write if write size is bigger than 5360 * max_blksz, and the file's block size is also max_blksz. 5361 */ 5362 blksz = max_blksz; 5363 if (size < blksz || zp->z_blksz != blksz) { 5364 ZFS_EXIT(zfsvfs); 5365 return (SET_ERROR(EINVAL)); 5366 } 5367 /* 5368 * Caller requests buffers for write before knowing where the 5369 * write offset might be (e.g. NFS TCP write). 5370 */ 5371 if (offset == -1) { 5372 preamble = 0; 5373 } else { 5374 preamble = P2PHASE(offset, blksz); 5375 if (preamble) { 5376 preamble = blksz - preamble; 5377 size -= preamble; 5378 } 5379 } 5380 5381 postamble = P2PHASE(size, blksz); 5382 size -= postamble; 5383 5384 fullblk = size / blksz; 5385 (void) dmu_xuio_init(xuio, 5386 (preamble != 0) + fullblk + (postamble != 0)); 5387 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble, 5388 int, postamble, int, 5389 (preamble != 0) + fullblk + (postamble != 0)); 5390 5391 /* 5392 * Have to fix iov base/len for partial buffers. They 5393 * currently represent full arc_buf's. 5394 */ 5395 if (preamble) { 5396 /* data begins in the middle of the arc_buf */ 5397 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5398 blksz); 5399 ASSERT(abuf); 5400 (void) dmu_xuio_add(xuio, abuf, 5401 blksz - preamble, preamble); 5402 } 5403 5404 for (i = 0; i < fullblk; i++) { 5405 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5406 blksz); 5407 ASSERT(abuf); 5408 (void) dmu_xuio_add(xuio, abuf, 0, blksz); 5409 } 5410 5411 if (postamble) { 5412 /* data ends in the middle of the arc_buf */ 5413 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5414 blksz); 5415 ASSERT(abuf); 5416 (void) dmu_xuio_add(xuio, abuf, 0, postamble); 5417 } 5418 break; 5419 case UIO_READ: 5420 /* 5421 * Loan out an arc_buf for read if the read size is larger than 5422 * the current file block size. Block alignment is not 5423 * considered. Partial arc_buf will be loaned out for read. 5424 */ 5425 blksz = zp->z_blksz; 5426 if (blksz < zcr_blksz_min) 5427 blksz = zcr_blksz_min; 5428 if (blksz > zcr_blksz_max) 5429 blksz = zcr_blksz_max; 5430 /* avoid potential complexity of dealing with it */ 5431 if (blksz > max_blksz) { 5432 ZFS_EXIT(zfsvfs); 5433 return (SET_ERROR(EINVAL)); 5434 } 5435 5436 maxsize = zp->z_size - uio->uio_loffset; 5437 if (size > maxsize) 5438 size = maxsize; 5439 5440 if (size < blksz || vn_has_cached_data(vp)) { 5441 ZFS_EXIT(zfsvfs); 5442 return (SET_ERROR(EINVAL)); 5443 } 5444 break; 5445 default: 5446 ZFS_EXIT(zfsvfs); 5447 return (SET_ERROR(EINVAL)); 5448 } 5449 5450 uio->uio_extflg = UIO_XUIO; 5451 XUIO_XUZC_RW(xuio) = ioflag; 5452 ZFS_EXIT(zfsvfs); 5453 return (0); 5454} 5455 5456/*ARGSUSED*/ 5457static int 5458zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct) 5459{ 5460 int i; 5461 arc_buf_t *abuf; 5462 int ioflag = XUIO_XUZC_RW(xuio); 5463 5464 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY); 5465 5466 i = dmu_xuio_cnt(xuio); 5467 while (i-- > 0) { 5468 abuf = dmu_xuio_arcbuf(xuio, i); 5469 /* 5470 * if abuf == NULL, it must be a write buffer 5471 * that has been returned in zfs_write(). 5472 */ 5473 if (abuf) 5474 dmu_return_arcbuf(abuf); 5475 ASSERT(abuf || ioflag == UIO_WRITE); 5476 } 5477 5478 dmu_xuio_fini(xuio); 5479 return (0); 5480} 5481 5482/* 5483 * Predeclare these here so that the compiler assumes that 5484 * this is an "old style" function declaration that does 5485 * not include arguments => we won't get type mismatch errors 5486 * in the initializations that follow. 5487 */ 5488static int zfs_inval(); 5489static int zfs_isdir(); 5490 5491static int 5492zfs_inval() 5493{ 5494 return (SET_ERROR(EINVAL)); 5495} 5496 5497static int 5498zfs_isdir() 5499{ 5500 return (SET_ERROR(EISDIR)); 5501} 5502/* 5503 * Directory vnode operations template 5504 */ 5505vnodeops_t *zfs_dvnodeops; 5506const fs_operation_def_t zfs_dvnodeops_template[] = { 5507 VOPNAME_OPEN, { .vop_open = zfs_open }, 5508 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5509 VOPNAME_READ, { .error = zfs_isdir }, 5510 VOPNAME_WRITE, { .error = zfs_isdir }, 5511 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5512 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5513 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5514 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5515 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5516 VOPNAME_CREATE, { .vop_create = zfs_create }, 5517 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5518 VOPNAME_LINK, { .vop_link = zfs_link }, 5519 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5520 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 5521 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5522 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5523 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 5524 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5525 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5526 VOPNAME_FID, { .vop_fid = zfs_fid }, 5527 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5528 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5529 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5530 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5531 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5532 NULL, NULL 5533}; 5534 5535/* 5536 * Regular file vnode operations template 5537 */ 5538vnodeops_t *zfs_fvnodeops; 5539const fs_operation_def_t zfs_fvnodeops_template[] = { 5540 VOPNAME_OPEN, { .vop_open = zfs_open }, 5541 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5542 VOPNAME_READ, { .vop_read = zfs_read }, 5543 VOPNAME_WRITE, { .vop_write = zfs_write }, 5544 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5545 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5546 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5547 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5548 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5549 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5550 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5551 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5552 VOPNAME_FID, { .vop_fid = zfs_fid }, 5553 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5554 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 5555 VOPNAME_SPACE, { .vop_space = zfs_space }, 5556 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 5557 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 5558 VOPNAME_MAP, { .vop_map = zfs_map }, 5559 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 5560 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 5561 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5562 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5563 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5564 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5565 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf }, 5566 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf }, 5567 NULL, NULL 5568}; 5569 5570/* 5571 * Symbolic link vnode operations template 5572 */ 5573vnodeops_t *zfs_symvnodeops; 5574const fs_operation_def_t zfs_symvnodeops_template[] = { 5575 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5576 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5577 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5578 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5579 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 5580 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5581 VOPNAME_FID, { .vop_fid = zfs_fid }, 5582 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5583 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5584 NULL, NULL 5585}; 5586 5587/* 5588 * special share hidden files vnode operations template 5589 */ 5590vnodeops_t *zfs_sharevnodeops; 5591const fs_operation_def_t zfs_sharevnodeops_template[] = { 5592 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5593 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5594 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5595 VOPNAME_FID, { .vop_fid = zfs_fid }, 5596 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5597 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5598 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5599 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5600 NULL, NULL 5601}; 5602 5603/* 5604 * Extended attribute directory vnode operations template 5605 * 5606 * This template is identical to the directory vnodes 5607 * operation template except for restricted operations: 5608 * VOP_MKDIR() 5609 * VOP_SYMLINK() 5610 * 5611 * Note that there are other restrictions embedded in: 5612 * zfs_create() - restrict type to VREG 5613 * zfs_link() - no links into/out of attribute space 5614 * zfs_rename() - no moves into/out of attribute space 5615 */ 5616vnodeops_t *zfs_xdvnodeops; 5617const fs_operation_def_t zfs_xdvnodeops_template[] = { 5618 VOPNAME_OPEN, { .vop_open = zfs_open }, 5619 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5620 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5621 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5622 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5623 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5624 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5625 VOPNAME_CREATE, { .vop_create = zfs_create }, 5626 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5627 VOPNAME_LINK, { .vop_link = zfs_link }, 5628 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5629 VOPNAME_MKDIR, { .error = zfs_inval }, 5630 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5631 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5632 VOPNAME_SYMLINK, { .error = zfs_inval }, 5633 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5634 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5635 VOPNAME_FID, { .vop_fid = zfs_fid }, 5636 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5637 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5638 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5639 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5640 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5641 NULL, NULL 5642}; 5643 5644/* 5645 * Error vnode operations template 5646 */ 5647vnodeops_t *zfs_evnodeops; 5648const fs_operation_def_t zfs_evnodeops_template[] = { 5649 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5650 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5651 NULL, NULL 5652}; 5653#endif /* sun */ 5654 5655static int 5656ioflags(int ioflags) 5657{ 5658 int flags = 0; 5659 5660 if (ioflags & IO_APPEND) 5661 flags |= FAPPEND; 5662 if (ioflags & IO_NDELAY) 5663 flags |= FNONBLOCK; 5664 if (ioflags & IO_SYNC) 5665 flags |= (FSYNC | FDSYNC | FRSYNC); 5666 5667 return (flags); 5668} 5669 5670static int 5671zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage) 5672{ 5673 znode_t *zp = VTOZ(vp); 5674 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5675 objset_t *os = zp->z_zfsvfs->z_os; 5676 vm_page_t mfirst, mlast, mreq; 5677 vm_object_t object; 5678 caddr_t va; 5679 struct sf_buf *sf; 5680 off_t startoff, endoff; 5681 int i, error; 5682 vm_pindex_t reqstart, reqend; 5683 int pcount, lsize, reqsize, size; 5684 5685 ZFS_ENTER(zfsvfs); 5686 ZFS_VERIFY_ZP(zp); 5687 5688 pcount = OFF_TO_IDX(round_page(count)); 5689 mreq = m[reqpage]; 5690 object = mreq->object; 5691 error = 0; 5692 5693 KASSERT(vp->v_object == object, ("mismatching object")); 5694 5695 if (pcount > 1 && zp->z_blksz > PAGESIZE) { 5696 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz); 5697 reqstart = OFF_TO_IDX(round_page(startoff)); 5698 if (reqstart < m[0]->pindex) 5699 reqstart = 0; 5700 else 5701 reqstart = reqstart - m[0]->pindex; 5702 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE, 5703 zp->z_blksz); 5704 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1; 5705 if (reqend > m[pcount - 1]->pindex) 5706 reqend = m[pcount - 1]->pindex; 5707 reqsize = reqend - m[reqstart]->pindex + 1; 5708 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize, 5709 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds")); 5710 } else { 5711 reqstart = reqpage; 5712 reqsize = 1; 5713 } 5714 mfirst = m[reqstart]; 5715 mlast = m[reqstart + reqsize - 1]; 5716 5717 zfs_vmobject_wlock(object); 5718 5719 for (i = 0; i < reqstart; i++) { 5720 vm_page_lock(m[i]); 5721 vm_page_free(m[i]); 5722 vm_page_unlock(m[i]); 5723 } 5724 for (i = reqstart + reqsize; i < pcount; i++) { 5725 vm_page_lock(m[i]); 5726 vm_page_free(m[i]); 5727 vm_page_unlock(m[i]); 5728 } 5729 5730 if (mreq->valid && reqsize == 1) { 5731 if (mreq->valid != VM_PAGE_BITS_ALL) 5732 vm_page_zero_invalid(mreq, TRUE); 5733 zfs_vmobject_wunlock(object); 5734 ZFS_EXIT(zfsvfs); 5735 return (zfs_vm_pagerret_ok); 5736 } 5737 5738 PCPU_INC(cnt.v_vnodein); 5739 PCPU_ADD(cnt.v_vnodepgsin, reqsize); 5740 5741 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) { 5742 for (i = reqstart; i < reqstart + reqsize; i++) { 5743 if (i != reqpage) { 5744 vm_page_lock(m[i]); 5745 vm_page_free(m[i]); 5746 vm_page_unlock(m[i]); 5747 } 5748 } 5749 zfs_vmobject_wunlock(object); 5750 ZFS_EXIT(zfsvfs); 5751 return (zfs_vm_pagerret_bad); 5752 } 5753 5754 lsize = PAGE_SIZE; 5755 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size) 5756 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex); 5757 5758 zfs_vmobject_wunlock(object); 5759 5760 for (i = reqstart; i < reqstart + reqsize; i++) { 5761 size = PAGE_SIZE; 5762 if (i == (reqstart + reqsize - 1)) 5763 size = lsize; 5764 va = zfs_map_page(m[i], &sf); 5765 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex), 5766 size, va, DMU_READ_PREFETCH); 5767 if (size != PAGE_SIZE) 5768 bzero(va + size, PAGE_SIZE - size); 5769 zfs_unmap_page(sf); 5770 if (error != 0) 5771 break; 5772 } 5773 5774 zfs_vmobject_wlock(object); 5775 5776 for (i = reqstart; i < reqstart + reqsize; i++) { 5777 if (!error) 5778 m[i]->valid = VM_PAGE_BITS_ALL; 5779 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i])); 5780 if (i != reqpage) 5781 vm_page_readahead_finish(m[i]); 5782 } 5783 5784 zfs_vmobject_wunlock(object); 5785 5786 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 5787 ZFS_EXIT(zfsvfs); 5788 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok); 5789} 5790 5791static int 5792zfs_freebsd_getpages(ap) 5793 struct vop_getpages_args /* { 5794 struct vnode *a_vp; 5795 vm_page_t *a_m; 5796 int a_count; 5797 int a_reqpage; 5798 vm_ooffset_t a_offset; 5799 } */ *ap; 5800{ 5801 5802 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage)); 5803} 5804 5805static int 5806zfs_freebsd_bmap(ap) 5807 struct vop_bmap_args /* { 5808 struct vnode *a_vp; 5809 daddr_t a_bn; 5810 struct bufobj **a_bop; 5811 daddr_t *a_bnp; 5812 int *a_runp; 5813 int *a_runb; 5814 } */ *ap; 5815{ 5816 5817 if (ap->a_bop != NULL) 5818 *ap->a_bop = &ap->a_vp->v_bufobj; 5819 if (ap->a_bnp != NULL) 5820 *ap->a_bnp = ap->a_bn; 5821 if (ap->a_runp != NULL) 5822 *ap->a_runp = 0; 5823 if (ap->a_runb != NULL) 5824 *ap->a_runb = 0; 5825 5826 return (0); 5827} 5828 5829static int 5830zfs_freebsd_open(ap) 5831 struct vop_open_args /* { 5832 struct vnode *a_vp; 5833 int a_mode; 5834 struct ucred *a_cred; 5835 struct thread *a_td; 5836 } */ *ap; 5837{ 5838 vnode_t *vp = ap->a_vp; 5839 znode_t *zp = VTOZ(vp); 5840 int error; 5841 5842 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL); 5843 if (error == 0) 5844 vnode_create_vobject(vp, zp->z_size, ap->a_td); 5845 return (error); 5846} 5847 5848static int 5849zfs_freebsd_close(ap) 5850 struct vop_close_args /* { 5851 struct vnode *a_vp; 5852 int a_fflag; 5853 struct ucred *a_cred; 5854 struct thread *a_td; 5855 } */ *ap; 5856{ 5857 5858 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL)); 5859} 5860 5861static int 5862zfs_freebsd_ioctl(ap) 5863 struct vop_ioctl_args /* { 5864 struct vnode *a_vp; 5865 u_long a_command; 5866 caddr_t a_data; 5867 int a_fflag; 5868 struct ucred *cred; 5869 struct thread *td; 5870 } */ *ap; 5871{ 5872 5873 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 5874 ap->a_fflag, ap->a_cred, NULL, NULL)); 5875} 5876 5877static int 5878zfs_freebsd_read(ap) 5879 struct vop_read_args /* { 5880 struct vnode *a_vp; 5881 struct uio *a_uio; 5882 int a_ioflag; 5883 struct ucred *a_cred; 5884 } */ *ap; 5885{ 5886 5887 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 5888 ap->a_cred, NULL)); 5889} 5890 5891static int 5892zfs_freebsd_write(ap) 5893 struct vop_write_args /* { 5894 struct vnode *a_vp; 5895 struct uio *a_uio; 5896 int a_ioflag; 5897 struct ucred *a_cred; 5898 } */ *ap; 5899{ 5900 5901 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 5902 ap->a_cred, NULL)); 5903} 5904 5905static int 5906zfs_freebsd_access(ap) 5907 struct vop_access_args /* { 5908 struct vnode *a_vp; 5909 accmode_t a_accmode; 5910 struct ucred *a_cred; 5911 struct thread *a_td; 5912 } */ *ap; 5913{ 5914 vnode_t *vp = ap->a_vp; 5915 znode_t *zp = VTOZ(vp); 5916 accmode_t accmode; 5917 int error = 0; 5918 5919 /* 5920 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 5921 */ 5922 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 5923 if (accmode != 0) 5924 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL); 5925 5926 /* 5927 * VADMIN has to be handled by vaccess(). 5928 */ 5929 if (error == 0) { 5930 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 5931 if (accmode != 0) { 5932 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 5933 zp->z_gid, accmode, ap->a_cred, NULL); 5934 } 5935 } 5936 5937 /* 5938 * For VEXEC, ensure that at least one execute bit is set for 5939 * non-directories. 5940 */ 5941 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 5942 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 5943 error = EACCES; 5944 } 5945 5946 return (error); 5947} 5948 5949static int 5950zfs_freebsd_lookup(ap) 5951 struct vop_lookup_args /* { 5952 struct vnode *a_dvp; 5953 struct vnode **a_vpp; 5954 struct componentname *a_cnp; 5955 } */ *ap; 5956{ 5957 struct componentname *cnp = ap->a_cnp; 5958 char nm[NAME_MAX + 1]; 5959 5960 ASSERT(cnp->cn_namelen < sizeof(nm)); 5961 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 5962 5963 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 5964 cnp->cn_cred, cnp->cn_thread, 0)); 5965} 5966 5967static int 5968zfs_freebsd_create(ap) 5969 struct vop_create_args /* { 5970 struct vnode *a_dvp; 5971 struct vnode **a_vpp; 5972 struct componentname *a_cnp; 5973 struct vattr *a_vap; 5974 } */ *ap; 5975{ 5976 struct componentname *cnp = ap->a_cnp; 5977 vattr_t *vap = ap->a_vap; 5978 int mode; 5979 5980 ASSERT(cnp->cn_flags & SAVENAME); 5981 5982 vattr_init_mask(vap); 5983 mode = vap->va_mode & ALLPERMS; 5984 5985 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 5986 ap->a_vpp, cnp->cn_cred, cnp->cn_thread)); 5987} 5988 5989static int 5990zfs_freebsd_remove(ap) 5991 struct vop_remove_args /* { 5992 struct vnode *a_dvp; 5993 struct vnode *a_vp; 5994 struct componentname *a_cnp; 5995 } */ *ap; 5996{ 5997 5998 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 5999 6000 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 6001 ap->a_cnp->cn_cred, NULL, 0)); 6002} 6003 6004static int 6005zfs_freebsd_mkdir(ap) 6006 struct vop_mkdir_args /* { 6007 struct vnode *a_dvp; 6008 struct vnode **a_vpp; 6009 struct componentname *a_cnp; 6010 struct vattr *a_vap; 6011 } */ *ap; 6012{ 6013 vattr_t *vap = ap->a_vap; 6014 6015 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6016 6017 vattr_init_mask(vap); 6018 6019 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 6020 ap->a_cnp->cn_cred, NULL, 0, NULL)); 6021} 6022 6023static int 6024zfs_freebsd_rmdir(ap) 6025 struct vop_rmdir_args /* { 6026 struct vnode *a_dvp; 6027 struct vnode *a_vp; 6028 struct componentname *a_cnp; 6029 } */ *ap; 6030{ 6031 struct componentname *cnp = ap->a_cnp; 6032 6033 ASSERT(cnp->cn_flags & SAVENAME); 6034 6035 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0)); 6036} 6037 6038static int 6039zfs_freebsd_readdir(ap) 6040 struct vop_readdir_args /* { 6041 struct vnode *a_vp; 6042 struct uio *a_uio; 6043 struct ucred *a_cred; 6044 int *a_eofflag; 6045 int *a_ncookies; 6046 u_long **a_cookies; 6047 } */ *ap; 6048{ 6049 6050 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 6051 ap->a_ncookies, ap->a_cookies)); 6052} 6053 6054static int 6055zfs_freebsd_fsync(ap) 6056 struct vop_fsync_args /* { 6057 struct vnode *a_vp; 6058 int a_waitfor; 6059 struct thread *a_td; 6060 } */ *ap; 6061{ 6062 6063 vop_stdfsync(ap); 6064 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL)); 6065} 6066 6067static int 6068zfs_freebsd_getattr(ap) 6069 struct vop_getattr_args /* { 6070 struct vnode *a_vp; 6071 struct vattr *a_vap; 6072 struct ucred *a_cred; 6073 } */ *ap; 6074{ 6075 vattr_t *vap = ap->a_vap; 6076 xvattr_t xvap; 6077 u_long fflags = 0; 6078 int error; 6079 6080 xva_init(&xvap); 6081 xvap.xva_vattr = *vap; 6082 xvap.xva_vattr.va_mask |= AT_XVATTR; 6083 6084 /* Convert chflags into ZFS-type flags. */ 6085 /* XXX: what about SF_SETTABLE?. */ 6086 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 6087 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 6088 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 6089 XVA_SET_REQ(&xvap, XAT_NODUMP); 6090 XVA_SET_REQ(&xvap, XAT_READONLY); 6091 XVA_SET_REQ(&xvap, XAT_ARCHIVE); 6092 XVA_SET_REQ(&xvap, XAT_SYSTEM); 6093 XVA_SET_REQ(&xvap, XAT_HIDDEN); 6094 XVA_SET_REQ(&xvap, XAT_REPARSE); 6095 XVA_SET_REQ(&xvap, XAT_OFFLINE); 6096 XVA_SET_REQ(&xvap, XAT_SPARSE); 6097 6098 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL); 6099 if (error != 0) 6100 return (error); 6101 6102 /* Convert ZFS xattr into chflags. */ 6103#define FLAG_CHECK(fflag, xflag, xfield) do { \ 6104 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 6105 fflags |= (fflag); \ 6106} while (0) 6107 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 6108 xvap.xva_xoptattrs.xoa_immutable); 6109 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 6110 xvap.xva_xoptattrs.xoa_appendonly); 6111 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 6112 xvap.xva_xoptattrs.xoa_nounlink); 6113 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE, 6114 xvap.xva_xoptattrs.xoa_archive); 6115 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 6116 xvap.xva_xoptattrs.xoa_nodump); 6117 FLAG_CHECK(UF_READONLY, XAT_READONLY, 6118 xvap.xva_xoptattrs.xoa_readonly); 6119 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM, 6120 xvap.xva_xoptattrs.xoa_system); 6121 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN, 6122 xvap.xva_xoptattrs.xoa_hidden); 6123 FLAG_CHECK(UF_REPARSE, XAT_REPARSE, 6124 xvap.xva_xoptattrs.xoa_reparse); 6125 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE, 6126 xvap.xva_xoptattrs.xoa_offline); 6127 FLAG_CHECK(UF_SPARSE, XAT_SPARSE, 6128 xvap.xva_xoptattrs.xoa_sparse); 6129 6130#undef FLAG_CHECK 6131 *vap = xvap.xva_vattr; 6132 vap->va_flags = fflags; 6133 return (0); 6134} 6135 6136static int 6137zfs_freebsd_setattr(ap) 6138 struct vop_setattr_args /* { 6139 struct vnode *a_vp; 6140 struct vattr *a_vap; 6141 struct ucred *a_cred; 6142 } */ *ap; 6143{ 6144 vnode_t *vp = ap->a_vp; 6145 vattr_t *vap = ap->a_vap; 6146 cred_t *cred = ap->a_cred; 6147 xvattr_t xvap; 6148 u_long fflags; 6149 uint64_t zflags; 6150 6151 vattr_init_mask(vap); 6152 vap->va_mask &= ~AT_NOSET; 6153 6154 xva_init(&xvap); 6155 xvap.xva_vattr = *vap; 6156 6157 zflags = VTOZ(vp)->z_pflags; 6158 6159 if (vap->va_flags != VNOVAL) { 6160 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 6161 int error; 6162 6163 if (zfsvfs->z_use_fuids == B_FALSE) 6164 return (EOPNOTSUPP); 6165 6166 fflags = vap->va_flags; 6167 /* 6168 * XXX KDM 6169 * We need to figure out whether it makes sense to allow 6170 * UF_REPARSE through, since we don't really have other 6171 * facilities to handle reparse points and zfs_setattr() 6172 * doesn't currently allow setting that attribute anyway. 6173 */ 6174 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE| 6175 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE| 6176 UF_OFFLINE|UF_SPARSE)) != 0) 6177 return (EOPNOTSUPP); 6178 /* 6179 * Unprivileged processes are not permitted to unset system 6180 * flags, or modify flags if any system flags are set. 6181 * Privileged non-jail processes may not modify system flags 6182 * if securelevel > 0 and any existing system flags are set. 6183 * Privileged jail processes behave like privileged non-jail 6184 * processes if the security.jail.chflags_allowed sysctl is 6185 * is non-zero; otherwise, they behave like unprivileged 6186 * processes. 6187 */ 6188 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 6189 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) { 6190 if (zflags & 6191 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6192 error = securelevel_gt(cred, 0); 6193 if (error != 0) 6194 return (error); 6195 } 6196 } else { 6197 /* 6198 * Callers may only modify the file flags on objects they 6199 * have VADMIN rights for. 6200 */ 6201 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0) 6202 return (error); 6203 if (zflags & 6204 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6205 return (EPERM); 6206 } 6207 if (fflags & 6208 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 6209 return (EPERM); 6210 } 6211 } 6212 6213#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 6214 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 6215 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 6216 XVA_SET_REQ(&xvap, (xflag)); \ 6217 (xfield) = ((fflags & (fflag)) != 0); \ 6218 } \ 6219} while (0) 6220 /* Convert chflags into ZFS-type flags. */ 6221 /* XXX: what about SF_SETTABLE?. */ 6222 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 6223 xvap.xva_xoptattrs.xoa_immutable); 6224 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 6225 xvap.xva_xoptattrs.xoa_appendonly); 6226 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 6227 xvap.xva_xoptattrs.xoa_nounlink); 6228 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE, 6229 xvap.xva_xoptattrs.xoa_archive); 6230 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 6231 xvap.xva_xoptattrs.xoa_nodump); 6232 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY, 6233 xvap.xva_xoptattrs.xoa_readonly); 6234 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM, 6235 xvap.xva_xoptattrs.xoa_system); 6236 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN, 6237 xvap.xva_xoptattrs.xoa_hidden); 6238 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE, 6239 xvap.xva_xoptattrs.xoa_hidden); 6240 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE, 6241 xvap.xva_xoptattrs.xoa_offline); 6242 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE, 6243 xvap.xva_xoptattrs.xoa_sparse); 6244#undef FLAG_CHANGE 6245 } 6246 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL)); 6247} 6248 6249static int 6250zfs_freebsd_rename(ap) 6251 struct vop_rename_args /* { 6252 struct vnode *a_fdvp; 6253 struct vnode *a_fvp; 6254 struct componentname *a_fcnp; 6255 struct vnode *a_tdvp; 6256 struct vnode *a_tvp; 6257 struct componentname *a_tcnp; 6258 } */ *ap; 6259{ 6260 vnode_t *fdvp = ap->a_fdvp; 6261 vnode_t *fvp = ap->a_fvp; 6262 vnode_t *tdvp = ap->a_tdvp; 6263 vnode_t *tvp = ap->a_tvp; 6264 int error; 6265 6266 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 6267 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 6268 6269 /* 6270 * Check for cross-device rename. 6271 */ 6272 if ((fdvp->v_mount != tdvp->v_mount) || 6273 (tvp && (fdvp->v_mount != tvp->v_mount))) 6274 error = EXDEV; 6275 else 6276 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 6277 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0); 6278 if (tdvp == tvp) 6279 VN_RELE(tdvp); 6280 else 6281 VN_URELE(tdvp); 6282 if (tvp) 6283 VN_URELE(tvp); 6284 VN_RELE(fdvp); 6285 VN_RELE(fvp); 6286 6287 return (error); 6288} 6289 6290static int 6291zfs_freebsd_symlink(ap) 6292 struct vop_symlink_args /* { 6293 struct vnode *a_dvp; 6294 struct vnode **a_vpp; 6295 struct componentname *a_cnp; 6296 struct vattr *a_vap; 6297 char *a_target; 6298 } */ *ap; 6299{ 6300 struct componentname *cnp = ap->a_cnp; 6301 vattr_t *vap = ap->a_vap; 6302 6303 ASSERT(cnp->cn_flags & SAVENAME); 6304 6305 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 6306 vattr_init_mask(vap); 6307 6308 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 6309 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 6310} 6311 6312static int 6313zfs_freebsd_readlink(ap) 6314 struct vop_readlink_args /* { 6315 struct vnode *a_vp; 6316 struct uio *a_uio; 6317 struct ucred *a_cred; 6318 } */ *ap; 6319{ 6320 6321 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL)); 6322} 6323 6324static int 6325zfs_freebsd_link(ap) 6326 struct vop_link_args /* { 6327 struct vnode *a_tdvp; 6328 struct vnode *a_vp; 6329 struct componentname *a_cnp; 6330 } */ *ap; 6331{ 6332 struct componentname *cnp = ap->a_cnp; 6333 vnode_t *vp = ap->a_vp; 6334 vnode_t *tdvp = ap->a_tdvp; 6335 6336 if (tdvp->v_mount != vp->v_mount) 6337 return (EXDEV); 6338 6339 ASSERT(cnp->cn_flags & SAVENAME); 6340 6341 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0)); 6342} 6343 6344static int 6345zfs_freebsd_inactive(ap) 6346 struct vop_inactive_args /* { 6347 struct vnode *a_vp; 6348 struct thread *a_td; 6349 } */ *ap; 6350{ 6351 vnode_t *vp = ap->a_vp; 6352 6353 zfs_inactive(vp, ap->a_td->td_ucred, NULL); 6354 return (0); 6355} 6356 6357static int 6358zfs_freebsd_reclaim(ap) 6359 struct vop_reclaim_args /* { 6360 struct vnode *a_vp; 6361 struct thread *a_td; 6362 } */ *ap; 6363{ 6364 vnode_t *vp = ap->a_vp; 6365 znode_t *zp = VTOZ(vp); 6366 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6367 6368 ASSERT(zp != NULL); 6369 6370 /* Destroy the vm object and flush associated pages. */ 6371 vnode_destroy_vobject(vp); 6372 6373 /* 6374 * z_teardown_inactive_lock protects from a race with 6375 * zfs_znode_dmu_fini in zfsvfs_teardown during 6376 * force unmount. 6377 */ 6378 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 6379 if (zp->z_sa_hdl == NULL) 6380 zfs_znode_free(zp); 6381 else 6382 zfs_zinactive(zp); 6383 rw_exit(&zfsvfs->z_teardown_inactive_lock); 6384 6385 vp->v_data = NULL; 6386 return (0); 6387} 6388 6389static int 6390zfs_freebsd_fid(ap) 6391 struct vop_fid_args /* { 6392 struct vnode *a_vp; 6393 struct fid *a_fid; 6394 } */ *ap; 6395{ 6396 6397 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 6398} 6399 6400static int 6401zfs_freebsd_pathconf(ap) 6402 struct vop_pathconf_args /* { 6403 struct vnode *a_vp; 6404 int a_name; 6405 register_t *a_retval; 6406 } */ *ap; 6407{ 6408 ulong_t val; 6409 int error; 6410 6411 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); 6412 if (error == 0) 6413 *ap->a_retval = val; 6414 else if (error == EOPNOTSUPP) 6415 error = vop_stdpathconf(ap); 6416 return (error); 6417} 6418 6419static int 6420zfs_freebsd_fifo_pathconf(ap) 6421 struct vop_pathconf_args /* { 6422 struct vnode *a_vp; 6423 int a_name; 6424 register_t *a_retval; 6425 } */ *ap; 6426{ 6427 6428 switch (ap->a_name) { 6429 case _PC_ACL_EXTENDED: 6430 case _PC_ACL_NFS4: 6431 case _PC_ACL_PATH_MAX: 6432 case _PC_MAC_PRESENT: 6433 return (zfs_freebsd_pathconf(ap)); 6434 default: 6435 return (fifo_specops.vop_pathconf(ap)); 6436 } 6437} 6438 6439/* 6440 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 6441 * extended attribute name: 6442 * 6443 * NAMESPACE PREFIX 6444 * system freebsd:system: 6445 * user (none, can be used to access ZFS fsattr(5) attributes 6446 * created on Solaris) 6447 */ 6448static int 6449zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 6450 size_t size) 6451{ 6452 const char *namespace, *prefix, *suffix; 6453 6454 /* We don't allow '/' character in attribute name. */ 6455 if (strchr(name, '/') != NULL) 6456 return (EINVAL); 6457 /* We don't allow attribute names that start with "freebsd:" string. */ 6458 if (strncmp(name, "freebsd:", 8) == 0) 6459 return (EINVAL); 6460 6461 bzero(attrname, size); 6462 6463 switch (attrnamespace) { 6464 case EXTATTR_NAMESPACE_USER: 6465#if 0 6466 prefix = "freebsd:"; 6467 namespace = EXTATTR_NAMESPACE_USER_STRING; 6468 suffix = ":"; 6469#else 6470 /* 6471 * This is the default namespace by which we can access all 6472 * attributes created on Solaris. 6473 */ 6474 prefix = namespace = suffix = ""; 6475#endif 6476 break; 6477 case EXTATTR_NAMESPACE_SYSTEM: 6478 prefix = "freebsd:"; 6479 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 6480 suffix = ":"; 6481 break; 6482 case EXTATTR_NAMESPACE_EMPTY: 6483 default: 6484 return (EINVAL); 6485 } 6486 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 6487 name) >= size) { 6488 return (ENAMETOOLONG); 6489 } 6490 return (0); 6491} 6492 6493/* 6494 * Vnode operating to retrieve a named extended attribute. 6495 */ 6496static int 6497zfs_getextattr(struct vop_getextattr_args *ap) 6498/* 6499vop_getextattr { 6500 IN struct vnode *a_vp; 6501 IN int a_attrnamespace; 6502 IN const char *a_name; 6503 INOUT struct uio *a_uio; 6504 OUT size_t *a_size; 6505 IN struct ucred *a_cred; 6506 IN struct thread *a_td; 6507}; 6508*/ 6509{ 6510 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6511 struct thread *td = ap->a_td; 6512 struct nameidata nd; 6513 char attrname[255]; 6514 struct vattr va; 6515 vnode_t *xvp = NULL, *vp; 6516 int error, flags; 6517 6518 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6519 ap->a_cred, ap->a_td, VREAD); 6520 if (error != 0) 6521 return (error); 6522 6523 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6524 sizeof(attrname)); 6525 if (error != 0) 6526 return (error); 6527 6528 ZFS_ENTER(zfsvfs); 6529 6530 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6531 LOOKUP_XATTR); 6532 if (error != 0) { 6533 ZFS_EXIT(zfsvfs); 6534 return (error); 6535 } 6536 6537 flags = FREAD; 6538 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6539 xvp, td); 6540 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL); 6541 vp = nd.ni_vp; 6542 NDFREE(&nd, NDF_ONLY_PNBUF); 6543 if (error != 0) { 6544 ZFS_EXIT(zfsvfs); 6545 if (error == ENOENT) 6546 error = ENOATTR; 6547 return (error); 6548 } 6549 6550 if (ap->a_size != NULL) { 6551 error = VOP_GETATTR(vp, &va, ap->a_cred); 6552 if (error == 0) 6553 *ap->a_size = (size_t)va.va_size; 6554 } else if (ap->a_uio != NULL) 6555 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6556 6557 VOP_UNLOCK(vp, 0); 6558 vn_close(vp, flags, ap->a_cred, td); 6559 ZFS_EXIT(zfsvfs); 6560 6561 return (error); 6562} 6563 6564/* 6565 * Vnode operation to remove a named attribute. 6566 */ 6567int 6568zfs_deleteextattr(struct vop_deleteextattr_args *ap) 6569/* 6570vop_deleteextattr { 6571 IN struct vnode *a_vp; 6572 IN int a_attrnamespace; 6573 IN const char *a_name; 6574 IN struct ucred *a_cred; 6575 IN struct thread *a_td; 6576}; 6577*/ 6578{ 6579 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6580 struct thread *td = ap->a_td; 6581 struct nameidata nd; 6582 char attrname[255]; 6583 struct vattr va; 6584 vnode_t *xvp = NULL, *vp; 6585 int error, flags; 6586 6587 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6588 ap->a_cred, ap->a_td, VWRITE); 6589 if (error != 0) 6590 return (error); 6591 6592 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6593 sizeof(attrname)); 6594 if (error != 0) 6595 return (error); 6596 6597 ZFS_ENTER(zfsvfs); 6598 6599 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6600 LOOKUP_XATTR); 6601 if (error != 0) { 6602 ZFS_EXIT(zfsvfs); 6603 return (error); 6604 } 6605 6606 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 6607 UIO_SYSSPACE, attrname, xvp, td); 6608 error = namei(&nd); 6609 vp = nd.ni_vp; 6610 NDFREE(&nd, NDF_ONLY_PNBUF); 6611 if (error != 0) { 6612 ZFS_EXIT(zfsvfs); 6613 if (error == ENOENT) 6614 error = ENOATTR; 6615 return (error); 6616 } 6617 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 6618 6619 vput(nd.ni_dvp); 6620 if (vp == nd.ni_dvp) 6621 vrele(vp); 6622 else 6623 vput(vp); 6624 ZFS_EXIT(zfsvfs); 6625 6626 return (error); 6627} 6628 6629/* 6630 * Vnode operation to set a named attribute. 6631 */ 6632static int 6633zfs_setextattr(struct vop_setextattr_args *ap) 6634/* 6635vop_setextattr { 6636 IN struct vnode *a_vp; 6637 IN int a_attrnamespace; 6638 IN const char *a_name; 6639 INOUT struct uio *a_uio; 6640 IN struct ucred *a_cred; 6641 IN struct thread *a_td; 6642}; 6643*/ 6644{ 6645 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6646 struct thread *td = ap->a_td; 6647 struct nameidata nd; 6648 char attrname[255]; 6649 struct vattr va; 6650 vnode_t *xvp = NULL, *vp; 6651 int error, flags; 6652 6653 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6654 ap->a_cred, ap->a_td, VWRITE); 6655 if (error != 0) 6656 return (error); 6657 6658 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6659 sizeof(attrname)); 6660 if (error != 0) 6661 return (error); 6662 6663 ZFS_ENTER(zfsvfs); 6664 6665 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6666 LOOKUP_XATTR | CREATE_XATTR_DIR); 6667 if (error != 0) { 6668 ZFS_EXIT(zfsvfs); 6669 return (error); 6670 } 6671 6672 flags = FFLAGS(O_WRONLY | O_CREAT); 6673 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6674 xvp, td); 6675 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL); 6676 vp = nd.ni_vp; 6677 NDFREE(&nd, NDF_ONLY_PNBUF); 6678 if (error != 0) { 6679 ZFS_EXIT(zfsvfs); 6680 return (error); 6681 } 6682 6683 VATTR_NULL(&va); 6684 va.va_size = 0; 6685 error = VOP_SETATTR(vp, &va, ap->a_cred); 6686 if (error == 0) 6687 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred); 6688 6689 VOP_UNLOCK(vp, 0); 6690 vn_close(vp, flags, ap->a_cred, td); 6691 ZFS_EXIT(zfsvfs); 6692 6693 return (error); 6694} 6695 6696/* 6697 * Vnode operation to retrieve extended attributes on a vnode. 6698 */ 6699static int 6700zfs_listextattr(struct vop_listextattr_args *ap) 6701/* 6702vop_listextattr { 6703 IN struct vnode *a_vp; 6704 IN int a_attrnamespace; 6705 INOUT struct uio *a_uio; 6706 OUT size_t *a_size; 6707 IN struct ucred *a_cred; 6708 IN struct thread *a_td; 6709}; 6710*/ 6711{ 6712 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6713 struct thread *td = ap->a_td; 6714 struct nameidata nd; 6715 char attrprefix[16]; 6716 u_char dirbuf[sizeof(struct dirent)]; 6717 struct dirent *dp; 6718 struct iovec aiov; 6719 struct uio auio, *uio = ap->a_uio; 6720 size_t *sizep = ap->a_size; 6721 size_t plen; 6722 vnode_t *xvp = NULL, *vp; 6723 int done, error, eof, pos; 6724 6725 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6726 ap->a_cred, ap->a_td, VREAD); 6727 if (error != 0) 6728 return (error); 6729 6730 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 6731 sizeof(attrprefix)); 6732 if (error != 0) 6733 return (error); 6734 plen = strlen(attrprefix); 6735 6736 ZFS_ENTER(zfsvfs); 6737 6738 if (sizep != NULL) 6739 *sizep = 0; 6740 6741 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6742 LOOKUP_XATTR); 6743 if (error != 0) { 6744 ZFS_EXIT(zfsvfs); 6745 /* 6746 * ENOATTR means that the EA directory does not yet exist, 6747 * i.e. there are no extended attributes there. 6748 */ 6749 if (error == ENOATTR) 6750 error = 0; 6751 return (error); 6752 } 6753 6754 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 6755 UIO_SYSSPACE, ".", xvp, td); 6756 error = namei(&nd); 6757 vp = nd.ni_vp; 6758 NDFREE(&nd, NDF_ONLY_PNBUF); 6759 if (error != 0) { 6760 ZFS_EXIT(zfsvfs); 6761 return (error); 6762 } 6763 6764 auio.uio_iov = &aiov; 6765 auio.uio_iovcnt = 1; 6766 auio.uio_segflg = UIO_SYSSPACE; 6767 auio.uio_td = td; 6768 auio.uio_rw = UIO_READ; 6769 auio.uio_offset = 0; 6770 6771 do { 6772 u_char nlen; 6773 6774 aiov.iov_base = (void *)dirbuf; 6775 aiov.iov_len = sizeof(dirbuf); 6776 auio.uio_resid = sizeof(dirbuf); 6777 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 6778 done = sizeof(dirbuf) - auio.uio_resid; 6779 if (error != 0) 6780 break; 6781 for (pos = 0; pos < done;) { 6782 dp = (struct dirent *)(dirbuf + pos); 6783 pos += dp->d_reclen; 6784 /* 6785 * XXX: Temporarily we also accept DT_UNKNOWN, as this 6786 * is what we get when attribute was created on Solaris. 6787 */ 6788 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 6789 continue; 6790 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0) 6791 continue; 6792 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 6793 continue; 6794 nlen = dp->d_namlen - plen; 6795 if (sizep != NULL) 6796 *sizep += 1 + nlen; 6797 else if (uio != NULL) { 6798 /* 6799 * Format of extattr name entry is one byte for 6800 * length and the rest for name. 6801 */ 6802 error = uiomove(&nlen, 1, uio->uio_rw, uio); 6803 if (error == 0) { 6804 error = uiomove(dp->d_name + plen, nlen, 6805 uio->uio_rw, uio); 6806 } 6807 if (error != 0) 6808 break; 6809 } 6810 } 6811 } while (!eof && error == 0); 6812 6813 vput(vp); 6814 ZFS_EXIT(zfsvfs); 6815 6816 return (error); 6817} 6818 6819int 6820zfs_freebsd_getacl(ap) 6821 struct vop_getacl_args /* { 6822 struct vnode *vp; 6823 acl_type_t type; 6824 struct acl *aclp; 6825 struct ucred *cred; 6826 struct thread *td; 6827 } */ *ap; 6828{ 6829 int error; 6830 vsecattr_t vsecattr; 6831 6832 if (ap->a_type != ACL_TYPE_NFS4) 6833 return (EINVAL); 6834 6835 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 6836 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL)) 6837 return (error); 6838 6839 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt); 6840 if (vsecattr.vsa_aclentp != NULL) 6841 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 6842 6843 return (error); 6844} 6845 6846int 6847zfs_freebsd_setacl(ap) 6848 struct vop_setacl_args /* { 6849 struct vnode *vp; 6850 acl_type_t type; 6851 struct acl *aclp; 6852 struct ucred *cred; 6853 struct thread *td; 6854 } */ *ap; 6855{ 6856 int error; 6857 vsecattr_t vsecattr; 6858 int aclbsize; /* size of acl list in bytes */ 6859 aclent_t *aaclp; 6860 6861 if (ap->a_type != ACL_TYPE_NFS4) 6862 return (EINVAL); 6863 6864 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 6865 return (EINVAL); 6866 6867 /* 6868 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 6869 * splitting every entry into two and appending "canonical six" 6870 * entries at the end. Don't allow for setting an ACL that would 6871 * cause chmod(2) to run out of ACL entries. 6872 */ 6873 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 6874 return (ENOSPC); 6875 6876 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 6877 if (error != 0) 6878 return (error); 6879 6880 vsecattr.vsa_mask = VSA_ACE; 6881 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t); 6882 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 6883 aaclp = vsecattr.vsa_aclentp; 6884 vsecattr.vsa_aclentsz = aclbsize; 6885 6886 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 6887 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL); 6888 kmem_free(aaclp, aclbsize); 6889 6890 return (error); 6891} 6892 6893int 6894zfs_freebsd_aclcheck(ap) 6895 struct vop_aclcheck_args /* { 6896 struct vnode *vp; 6897 acl_type_t type; 6898 struct acl *aclp; 6899 struct ucred *cred; 6900 struct thread *td; 6901 } */ *ap; 6902{ 6903 6904 return (EOPNOTSUPP); 6905} 6906 6907struct vop_vector zfs_vnodeops; 6908struct vop_vector zfs_fifoops; 6909struct vop_vector zfs_shareops; 6910 6911struct vop_vector zfs_vnodeops = { 6912 .vop_default = &default_vnodeops, 6913 .vop_inactive = zfs_freebsd_inactive, 6914 .vop_reclaim = zfs_freebsd_reclaim, 6915 .vop_access = zfs_freebsd_access, 6916#ifdef FREEBSD_NAMECACHE 6917 .vop_lookup = vfs_cache_lookup, 6918 .vop_cachedlookup = zfs_freebsd_lookup, 6919#else 6920 .vop_lookup = zfs_freebsd_lookup, 6921#endif 6922 .vop_getattr = zfs_freebsd_getattr, 6923 .vop_setattr = zfs_freebsd_setattr, 6924 .vop_create = zfs_freebsd_create, 6925 .vop_mknod = zfs_freebsd_create, 6926 .vop_mkdir = zfs_freebsd_mkdir, 6927 .vop_readdir = zfs_freebsd_readdir, 6928 .vop_fsync = zfs_freebsd_fsync, 6929 .vop_open = zfs_freebsd_open, 6930 .vop_close = zfs_freebsd_close, 6931 .vop_rmdir = zfs_freebsd_rmdir, 6932 .vop_ioctl = zfs_freebsd_ioctl, 6933 .vop_link = zfs_freebsd_link, 6934 .vop_symlink = zfs_freebsd_symlink, 6935 .vop_readlink = zfs_freebsd_readlink, 6936 .vop_read = zfs_freebsd_read, 6937 .vop_write = zfs_freebsd_write, 6938 .vop_remove = zfs_freebsd_remove, 6939 .vop_rename = zfs_freebsd_rename, 6940 .vop_pathconf = zfs_freebsd_pathconf, 6941 .vop_bmap = zfs_freebsd_bmap, 6942 .vop_fid = zfs_freebsd_fid, 6943 .vop_getextattr = zfs_getextattr, 6944 .vop_deleteextattr = zfs_deleteextattr, 6945 .vop_setextattr = zfs_setextattr, 6946 .vop_listextattr = zfs_listextattr, 6947 .vop_getacl = zfs_freebsd_getacl, 6948 .vop_setacl = zfs_freebsd_setacl, 6949 .vop_aclcheck = zfs_freebsd_aclcheck, 6950 .vop_getpages = zfs_freebsd_getpages, 6951}; 6952 6953struct vop_vector zfs_fifoops = { 6954 .vop_default = &fifo_specops, 6955 .vop_fsync = zfs_freebsd_fsync, 6956 .vop_access = zfs_freebsd_access, 6957 .vop_getattr = zfs_freebsd_getattr, 6958 .vop_inactive = zfs_freebsd_inactive, 6959 .vop_read = VOP_PANIC, 6960 .vop_reclaim = zfs_freebsd_reclaim, 6961 .vop_setattr = zfs_freebsd_setattr, 6962 .vop_write = VOP_PANIC, 6963 .vop_pathconf = zfs_freebsd_fifo_pathconf, 6964 .vop_fid = zfs_freebsd_fid, 6965 .vop_getacl = zfs_freebsd_getacl, 6966 .vop_setacl = zfs_freebsd_setacl, 6967 .vop_aclcheck = zfs_freebsd_aclcheck, 6968}; 6969 6970/* 6971 * special share hidden files vnode operations template 6972 */ 6973struct vop_vector zfs_shareops = { 6974 .vop_default = &default_vnodeops, 6975 .vop_access = zfs_freebsd_access, 6976 .vop_inactive = zfs_freebsd_inactive, 6977 .vop_reclaim = zfs_freebsd_reclaim, 6978 .vop_fid = zfs_freebsd_fid, 6979 .vop_pathconf = zfs_freebsd_pathconf, 6980}; 6981