zfs_vnops.c revision 260773
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. On subsequent 123 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT, 124 * to indicate that this operation has already called dmu_tx_wait(). 125 * This will ensure that we don't retry forever, waiting a short bit 126 * each time. 127 * 128 * (5) If the operation succeeded, generate the intent log entry for it 129 * before dropping locks. This ensures that the ordering of events 130 * in the intent log matches the order in which they actually occurred. 131 * During ZIL replay the zfs_log_* functions will update the sequence 132 * number to indicate the zil transaction has replayed. 133 * 134 * (6) At the end of each vnode op, the DMU tx must always commit, 135 * regardless of whether there were any errors. 136 * 137 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 138 * to ensure that synchronous semantics are provided when necessary. 139 * 140 * In general, this is how things should be ordered in each vnode op: 141 * 142 * ZFS_ENTER(zfsvfs); // exit if unmounted 143 * top: 144 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 145 * rw_enter(...); // grab any other locks you need 146 * tx = dmu_tx_create(...); // get DMU tx 147 * dmu_tx_hold_*(); // hold each object you might modify 148 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 149 * if (error) { 150 * rw_exit(...); // drop locks 151 * zfs_dirent_unlock(dl); // unlock directory entry 152 * VN_RELE(...); // release held vnodes 153 * if (error == ERESTART) { 154 * waited = B_TRUE; 155 * dmu_tx_wait(tx); 156 * dmu_tx_abort(tx); 157 * goto top; 158 * } 159 * dmu_tx_abort(tx); // abort DMU tx 160 * ZFS_EXIT(zfsvfs); // finished in zfs 161 * return (error); // really out of space 162 * } 163 * error = do_real_work(); // do whatever this VOP does 164 * if (error == 0) 165 * zfs_log_*(...); // on success, make ZIL entry 166 * dmu_tx_commit(tx); // commit DMU tx -- error or not 167 * rw_exit(...); // drop locks 168 * zfs_dirent_unlock(dl); // unlock directory entry 169 * VN_RELE(...); // release held vnodes 170 * zil_commit(zilog, foid); // synchronous when necessary 171 * ZFS_EXIT(zfsvfs); // finished in zfs 172 * return (error); // done, report error 173 */ 174 175/* ARGSUSED */ 176static int 177zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 178{ 179 znode_t *zp = VTOZ(*vpp); 180 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 181 182 ZFS_ENTER(zfsvfs); 183 ZFS_VERIFY_ZP(zp); 184 185 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 186 ((flag & FAPPEND) == 0)) { 187 ZFS_EXIT(zfsvfs); 188 return (SET_ERROR(EPERM)); 189 } 190 191 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 192 ZTOV(zp)->v_type == VREG && 193 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) { 194 if (fs_vscan(*vpp, cr, 0) != 0) { 195 ZFS_EXIT(zfsvfs); 196 return (SET_ERROR(EACCES)); 197 } 198 } 199 200 /* Keep a count of the synchronous opens in the znode */ 201 if (flag & (FSYNC | FDSYNC)) 202 atomic_inc_32(&zp->z_sync_cnt); 203 204 ZFS_EXIT(zfsvfs); 205 return (0); 206} 207 208/* ARGSUSED */ 209static int 210zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 211 caller_context_t *ct) 212{ 213 znode_t *zp = VTOZ(vp); 214 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 215 216 /* 217 * Clean up any locks held by this process on the vp. 218 */ 219 cleanlocks(vp, ddi_get_pid(), 0); 220 cleanshares(vp, ddi_get_pid()); 221 222 ZFS_ENTER(zfsvfs); 223 ZFS_VERIFY_ZP(zp); 224 225 /* Decrement the synchronous opens in the znode */ 226 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 227 atomic_dec_32(&zp->z_sync_cnt); 228 229 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 230 ZTOV(zp)->v_type == VREG && 231 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) 232 VERIFY(fs_vscan(vp, cr, 1) == 0); 233 234 ZFS_EXIT(zfsvfs); 235 return (0); 236} 237 238/* 239 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 240 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 241 */ 242static int 243zfs_holey(vnode_t *vp, u_long cmd, offset_t *off) 244{ 245 znode_t *zp = VTOZ(vp); 246 uint64_t noff = (uint64_t)*off; /* new offset */ 247 uint64_t file_sz; 248 int error; 249 boolean_t hole; 250 251 file_sz = zp->z_size; 252 if (noff >= file_sz) { 253 return (SET_ERROR(ENXIO)); 254 } 255 256 if (cmd == _FIO_SEEK_HOLE) 257 hole = B_TRUE; 258 else 259 hole = B_FALSE; 260 261 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 262 263 /* end of file? */ 264 if ((error == ESRCH) || (noff > file_sz)) { 265 /* 266 * Handle the virtual hole at the end of file. 267 */ 268 if (hole) { 269 *off = file_sz; 270 return (0); 271 } 272 return (SET_ERROR(ENXIO)); 273 } 274 275 if (noff < *off) 276 return (error); 277 *off = noff; 278 return (error); 279} 280 281/* ARGSUSED */ 282static int 283zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred, 284 int *rvalp, caller_context_t *ct) 285{ 286 offset_t off; 287 int error; 288 zfsvfs_t *zfsvfs; 289 znode_t *zp; 290 291 switch (com) { 292 case _FIOFFS: 293 return (0); 294 295 /* 296 * The following two ioctls are used by bfu. Faking out, 297 * necessary to avoid bfu errors. 298 */ 299 case _FIOGDIO: 300 case _FIOSDIO: 301 return (0); 302 303 case _FIO_SEEK_DATA: 304 case _FIO_SEEK_HOLE: 305#ifdef sun 306 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 307 return (SET_ERROR(EFAULT)); 308#else 309 off = *(offset_t *)data; 310#endif 311 zp = VTOZ(vp); 312 zfsvfs = zp->z_zfsvfs; 313 ZFS_ENTER(zfsvfs); 314 ZFS_VERIFY_ZP(zp); 315 316 /* offset parameter is in/out */ 317 error = zfs_holey(vp, com, &off); 318 ZFS_EXIT(zfsvfs); 319 if (error) 320 return (error); 321#ifdef sun 322 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 323 return (SET_ERROR(EFAULT)); 324#else 325 *(offset_t *)data = off; 326#endif 327 return (0); 328 } 329 return (SET_ERROR(ENOTTY)); 330} 331 332static vm_page_t 333page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) 334{ 335 vm_object_t obj; 336 vm_page_t pp; 337 int64_t end; 338 339 /* 340 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE 341 * aligned boundaries, if the range is not aligned. As a result a 342 * DEV_BSIZE subrange with partially dirty data may get marked as clean. 343 * It may happen that all DEV_BSIZE subranges are marked clean and thus 344 * the whole page would be considred clean despite have some dirty data. 345 * For this reason we should shrink the range to DEV_BSIZE aligned 346 * boundaries before calling vm_page_clear_dirty. 347 */ 348 end = rounddown2(off + nbytes, DEV_BSIZE); 349 off = roundup2(off, DEV_BSIZE); 350 nbytes = end - off; 351 352 obj = vp->v_object; 353 zfs_vmobject_assert_wlocked(obj); 354 355 for (;;) { 356 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 357 pp->valid) { 358 if (vm_page_xbusied(pp)) { 359 /* 360 * Reference the page before unlocking and 361 * sleeping so that the page daemon is less 362 * likely to reclaim it. 363 */ 364 vm_page_reference(pp); 365 vm_page_lock(pp); 366 zfs_vmobject_wunlock(obj); 367 vm_page_busy_sleep(pp, "zfsmwb"); 368 zfs_vmobject_wlock(obj); 369 continue; 370 } 371 vm_page_sbusy(pp); 372 } else if (pp == NULL) { 373 pp = vm_page_alloc(obj, OFF_TO_IDX(start), 374 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED | 375 VM_ALLOC_SBUSY); 376 } else { 377 ASSERT(pp != NULL && !pp->valid); 378 pp = NULL; 379 } 380 381 if (pp != NULL) { 382 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 383 vm_object_pip_add(obj, 1); 384 pmap_remove_write(pp); 385 if (nbytes != 0) 386 vm_page_clear_dirty(pp, off, nbytes); 387 } 388 break; 389 } 390 return (pp); 391} 392 393static void 394page_unbusy(vm_page_t pp) 395{ 396 397 vm_page_sunbusy(pp); 398 vm_object_pip_subtract(pp->object, 1); 399} 400 401static vm_page_t 402page_hold(vnode_t *vp, int64_t start) 403{ 404 vm_object_t obj; 405 vm_page_t pp; 406 407 obj = vp->v_object; 408 zfs_vmobject_assert_wlocked(obj); 409 410 for (;;) { 411 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 412 pp->valid) { 413 if (vm_page_xbusied(pp)) { 414 /* 415 * Reference the page before unlocking and 416 * sleeping so that the page daemon is less 417 * likely to reclaim it. 418 */ 419 vm_page_reference(pp); 420 vm_page_lock(pp); 421 zfs_vmobject_wunlock(obj); 422 vm_page_busy_sleep(pp, "zfsmwb"); 423 zfs_vmobject_wlock(obj); 424 continue; 425 } 426 427 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 428 vm_page_lock(pp); 429 vm_page_hold(pp); 430 vm_page_unlock(pp); 431 432 } else 433 pp = NULL; 434 break; 435 } 436 return (pp); 437} 438 439static void 440page_unhold(vm_page_t pp) 441{ 442 443 vm_page_lock(pp); 444 vm_page_unhold(pp); 445 vm_page_unlock(pp); 446} 447 448static caddr_t 449zfs_map_page(vm_page_t pp, struct sf_buf **sfp) 450{ 451 452 *sfp = sf_buf_alloc(pp, 0); 453 return ((caddr_t)sf_buf_kva(*sfp)); 454} 455 456static void 457zfs_unmap_page(struct sf_buf *sf) 458{ 459 460 sf_buf_free(sf); 461} 462 463/* 464 * When a file is memory mapped, we must keep the IO data synchronized 465 * between the DMU cache and the memory mapped pages. What this means: 466 * 467 * On Write: If we find a memory mapped page, we write to *both* 468 * the page and the dmu buffer. 469 */ 470static void 471update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid, 472 int segflg, dmu_tx_t *tx) 473{ 474 vm_object_t obj; 475 struct sf_buf *sf; 476 caddr_t va; 477 int off; 478 479 ASSERT(vp->v_mount != NULL); 480 obj = vp->v_object; 481 ASSERT(obj != NULL); 482 483 off = start & PAGEOFFSET; 484 zfs_vmobject_wlock(obj); 485 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 486 vm_page_t pp; 487 int nbytes = imin(PAGESIZE - off, len); 488 489 if (segflg == UIO_NOCOPY) { 490 pp = vm_page_lookup(obj, OFF_TO_IDX(start)); 491 KASSERT(pp != NULL, 492 ("zfs update_pages: NULL page in putpages case")); 493 KASSERT(off == 0, 494 ("zfs update_pages: unaligned data in putpages case")); 495 KASSERT(pp->valid == VM_PAGE_BITS_ALL, 496 ("zfs update_pages: invalid page in putpages case")); 497 KASSERT(vm_page_sbusied(pp), 498 ("zfs update_pages: unbusy page in putpages case")); 499 KASSERT(!pmap_page_is_write_mapped(pp), 500 ("zfs update_pages: writable page in putpages case")); 501 zfs_vmobject_wunlock(obj); 502 503 va = zfs_map_page(pp, &sf); 504 (void) dmu_write(os, oid, start, nbytes, va, tx); 505 zfs_unmap_page(sf); 506 507 zfs_vmobject_wlock(obj); 508 vm_page_undirty(pp); 509 } else if ((pp = page_busy(vp, start, off, nbytes)) != NULL) { 510 zfs_vmobject_wunlock(obj); 511 512 va = zfs_map_page(pp, &sf); 513 (void) dmu_read(os, oid, start+off, nbytes, 514 va+off, DMU_READ_PREFETCH);; 515 zfs_unmap_page(sf); 516 517 zfs_vmobject_wlock(obj); 518 page_unbusy(pp); 519 } 520 len -= nbytes; 521 off = 0; 522 } 523 if (segflg != UIO_NOCOPY) 524 vm_object_pip_wakeupn(obj, 0); 525 zfs_vmobject_wunlock(obj); 526} 527 528/* 529 * Read with UIO_NOCOPY flag means that sendfile(2) requests 530 * ZFS to populate a range of page cache pages with data. 531 * 532 * NOTE: this function could be optimized to pre-allocate 533 * all pages in advance, drain exclusive busy on all of them, 534 * map them into contiguous KVA region and populate them 535 * in one single dmu_read() call. 536 */ 537static int 538mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio) 539{ 540 znode_t *zp = VTOZ(vp); 541 objset_t *os = zp->z_zfsvfs->z_os; 542 struct sf_buf *sf; 543 vm_object_t obj; 544 vm_page_t pp; 545 int64_t start; 546 caddr_t va; 547 int len = nbytes; 548 int off; 549 int error = 0; 550 551 ASSERT(uio->uio_segflg == UIO_NOCOPY); 552 ASSERT(vp->v_mount != NULL); 553 obj = vp->v_object; 554 ASSERT(obj != NULL); 555 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0); 556 557 zfs_vmobject_wlock(obj); 558 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) { 559 int bytes = MIN(PAGESIZE, len); 560 561 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY | 562 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); 563 if (pp->valid == 0) { 564 zfs_vmobject_wunlock(obj); 565 va = zfs_map_page(pp, &sf); 566 error = dmu_read(os, zp->z_id, start, bytes, va, 567 DMU_READ_PREFETCH); 568 if (bytes != PAGESIZE && error == 0) 569 bzero(va + bytes, PAGESIZE - bytes); 570 zfs_unmap_page(sf); 571 zfs_vmobject_wlock(obj); 572 vm_page_sunbusy(pp); 573 vm_page_lock(pp); 574 if (error) { 575 if (pp->wire_count == 0 && pp->valid == 0 && 576 !vm_page_busied(pp)) 577 vm_page_free(pp); 578 } else { 579 pp->valid = VM_PAGE_BITS_ALL; 580 vm_page_activate(pp); 581 } 582 vm_page_unlock(pp); 583 } else { 584 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 585 vm_page_sunbusy(pp); 586 } 587 if (error) 588 break; 589 uio->uio_resid -= bytes; 590 uio->uio_offset += bytes; 591 len -= bytes; 592 } 593 zfs_vmobject_wunlock(obj); 594 return (error); 595} 596 597/* 598 * When a file is memory mapped, we must keep the IO data synchronized 599 * between the DMU cache and the memory mapped pages. What this means: 600 * 601 * On Read: We "read" preferentially from memory mapped pages, 602 * else we default from the dmu buffer. 603 * 604 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 605 * the file is memory mapped. 606 */ 607static int 608mappedread(vnode_t *vp, int nbytes, uio_t *uio) 609{ 610 znode_t *zp = VTOZ(vp); 611 objset_t *os = zp->z_zfsvfs->z_os; 612 vm_object_t obj; 613 int64_t start; 614 caddr_t va; 615 int len = nbytes; 616 int off; 617 int error = 0; 618 619 ASSERT(vp->v_mount != NULL); 620 obj = vp->v_object; 621 ASSERT(obj != NULL); 622 623 start = uio->uio_loffset; 624 off = start & PAGEOFFSET; 625 zfs_vmobject_wlock(obj); 626 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 627 vm_page_t pp; 628 uint64_t bytes = MIN(PAGESIZE - off, len); 629 630 if (pp = page_hold(vp, start)) { 631 struct sf_buf *sf; 632 caddr_t va; 633 634 zfs_vmobject_wunlock(obj); 635 va = zfs_map_page(pp, &sf); 636 error = uiomove(va + off, bytes, UIO_READ, uio); 637 zfs_unmap_page(sf); 638 zfs_vmobject_wlock(obj); 639 page_unhold(pp); 640 } else { 641 zfs_vmobject_wunlock(obj); 642 error = dmu_read_uio(os, zp->z_id, uio, bytes); 643 zfs_vmobject_wlock(obj); 644 } 645 len -= bytes; 646 off = 0; 647 if (error) 648 break; 649 } 650 zfs_vmobject_wunlock(obj); 651 return (error); 652} 653 654offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 655 656/* 657 * Read bytes from specified file into supplied buffer. 658 * 659 * IN: vp - vnode of file to be read from. 660 * uio - structure supplying read location, range info, 661 * and return buffer. 662 * ioflag - SYNC flags; used to provide FRSYNC semantics. 663 * cr - credentials of caller. 664 * ct - caller context 665 * 666 * OUT: uio - updated offset and range, buffer filled. 667 * 668 * RETURN: 0 on success, error code on failure. 669 * 670 * Side Effects: 671 * vp - atime updated if byte count > 0 672 */ 673/* ARGSUSED */ 674static int 675zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 676{ 677 znode_t *zp = VTOZ(vp); 678 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 679 objset_t *os; 680 ssize_t n, nbytes; 681 int error = 0; 682 rl_t *rl; 683 xuio_t *xuio = NULL; 684 685 ZFS_ENTER(zfsvfs); 686 ZFS_VERIFY_ZP(zp); 687 os = zfsvfs->z_os; 688 689 if (zp->z_pflags & ZFS_AV_QUARANTINED) { 690 ZFS_EXIT(zfsvfs); 691 return (SET_ERROR(EACCES)); 692 } 693 694 /* 695 * Validate file offset 696 */ 697 if (uio->uio_loffset < (offset_t)0) { 698 ZFS_EXIT(zfsvfs); 699 return (SET_ERROR(EINVAL)); 700 } 701 702 /* 703 * Fasttrack empty reads 704 */ 705 if (uio->uio_resid == 0) { 706 ZFS_EXIT(zfsvfs); 707 return (0); 708 } 709 710 /* 711 * Check for mandatory locks 712 */ 713 if (MANDMODE(zp->z_mode)) { 714 if (error = chklock(vp, FREAD, 715 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 716 ZFS_EXIT(zfsvfs); 717 return (error); 718 } 719 } 720 721 /* 722 * If we're in FRSYNC mode, sync out this znode before reading it. 723 */ 724 if (zfsvfs->z_log && 725 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)) 726 zil_commit(zfsvfs->z_log, zp->z_id); 727 728 /* 729 * Lock the range against changes. 730 */ 731 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 732 733 /* 734 * If we are reading past end-of-file we can skip 735 * to the end; but we might still need to set atime. 736 */ 737 if (uio->uio_loffset >= zp->z_size) { 738 error = 0; 739 goto out; 740 } 741 742 ASSERT(uio->uio_loffset < zp->z_size); 743 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset); 744 745#ifdef sun 746 if ((uio->uio_extflg == UIO_XUIO) && 747 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) { 748 int nblk; 749 int blksz = zp->z_blksz; 750 uint64_t offset = uio->uio_loffset; 751 752 xuio = (xuio_t *)uio; 753 if ((ISP2(blksz))) { 754 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset, 755 blksz)) / blksz; 756 } else { 757 ASSERT(offset + n <= blksz); 758 nblk = 1; 759 } 760 (void) dmu_xuio_init(xuio, nblk); 761 762 if (vn_has_cached_data(vp)) { 763 /* 764 * For simplicity, we always allocate a full buffer 765 * even if we only expect to read a portion of a block. 766 */ 767 while (--nblk >= 0) { 768 (void) dmu_xuio_add(xuio, 769 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 770 blksz), 0, blksz); 771 } 772 } 773 } 774#endif /* sun */ 775 776 while (n > 0) { 777 nbytes = MIN(n, zfs_read_chunk_size - 778 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 779 780#ifdef __FreeBSD__ 781 if (uio->uio_segflg == UIO_NOCOPY) 782 error = mappedread_sf(vp, nbytes, uio); 783 else 784#endif /* __FreeBSD__ */ 785 if (vn_has_cached_data(vp)) 786 error = mappedread(vp, nbytes, uio); 787 else 788 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 789 if (error) { 790 /* convert checksum errors into IO errors */ 791 if (error == ECKSUM) 792 error = SET_ERROR(EIO); 793 break; 794 } 795 796 n -= nbytes; 797 } 798out: 799 zfs_range_unlock(rl); 800 801 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 802 ZFS_EXIT(zfsvfs); 803 return (error); 804} 805 806/* 807 * Write the bytes to a file. 808 * 809 * IN: vp - vnode of file to be written to. 810 * uio - structure supplying write location, range info, 811 * and data buffer. 812 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is 813 * set if in append mode. 814 * cr - credentials of caller. 815 * ct - caller context (NFS/CIFS fem monitor only) 816 * 817 * OUT: uio - updated offset and range. 818 * 819 * RETURN: 0 on success, error code on failure. 820 * 821 * Timestamps: 822 * vp - ctime|mtime updated if byte count > 0 823 */ 824 825/* ARGSUSED */ 826static int 827zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 828{ 829 znode_t *zp = VTOZ(vp); 830 rlim64_t limit = MAXOFFSET_T; 831 ssize_t start_resid = uio->uio_resid; 832 ssize_t tx_bytes; 833 uint64_t end_size; 834 dmu_tx_t *tx; 835 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 836 zilog_t *zilog; 837 offset_t woff; 838 ssize_t n, nbytes; 839 rl_t *rl; 840 int max_blksz = zfsvfs->z_max_blksz; 841 int error = 0; 842 arc_buf_t *abuf; 843 iovec_t *aiov = NULL; 844 xuio_t *xuio = NULL; 845 int i_iov = 0; 846 int iovcnt = uio->uio_iovcnt; 847 iovec_t *iovp = uio->uio_iov; 848 int write_eof; 849 int count = 0; 850 sa_bulk_attr_t bulk[4]; 851 uint64_t mtime[2], ctime[2]; 852 853 /* 854 * Fasttrack empty write 855 */ 856 n = start_resid; 857 if (n == 0) 858 return (0); 859 860 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 861 limit = MAXOFFSET_T; 862 863 ZFS_ENTER(zfsvfs); 864 ZFS_VERIFY_ZP(zp); 865 866 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 867 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 868 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 869 &zp->z_size, 8); 870 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 871 &zp->z_pflags, 8); 872 873 /* 874 * If immutable or not appending then return EPERM 875 */ 876 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 877 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 878 (uio->uio_loffset < zp->z_size))) { 879 ZFS_EXIT(zfsvfs); 880 return (SET_ERROR(EPERM)); 881 } 882 883 zilog = zfsvfs->z_log; 884 885 /* 886 * Validate file offset 887 */ 888 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset; 889 if (woff < 0) { 890 ZFS_EXIT(zfsvfs); 891 return (SET_ERROR(EINVAL)); 892 } 893 894 /* 895 * Check for mandatory locks before calling zfs_range_lock() 896 * in order to prevent a deadlock with locks set via fcntl(). 897 */ 898 if (MANDMODE((mode_t)zp->z_mode) && 899 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 900 ZFS_EXIT(zfsvfs); 901 return (error); 902 } 903 904#ifdef sun 905 /* 906 * Pre-fault the pages to ensure slow (eg NFS) pages 907 * don't hold up txg. 908 * Skip this if uio contains loaned arc_buf. 909 */ 910 if ((uio->uio_extflg == UIO_XUIO) && 911 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) 912 xuio = (xuio_t *)uio; 913 else 914 uio_prefaultpages(MIN(n, max_blksz), uio); 915#endif /* sun */ 916 917 /* 918 * If in append mode, set the io offset pointer to eof. 919 */ 920 if (ioflag & FAPPEND) { 921 /* 922 * Obtain an appending range lock to guarantee file append 923 * semantics. We reset the write offset once we have the lock. 924 */ 925 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 926 woff = rl->r_off; 927 if (rl->r_len == UINT64_MAX) { 928 /* 929 * We overlocked the file because this write will cause 930 * the file block size to increase. 931 * Note that zp_size cannot change with this lock held. 932 */ 933 woff = zp->z_size; 934 } 935 uio->uio_loffset = woff; 936 } else { 937 /* 938 * Note that if the file block size will change as a result of 939 * this write, then this range lock will lock the entire file 940 * so that we can re-write the block safely. 941 */ 942 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 943 } 944 945 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) { 946 zfs_range_unlock(rl); 947 ZFS_EXIT(zfsvfs); 948 return (EFBIG); 949 } 950 951 if (woff >= limit) { 952 zfs_range_unlock(rl); 953 ZFS_EXIT(zfsvfs); 954 return (SET_ERROR(EFBIG)); 955 } 956 957 if ((woff + n) > limit || woff > (limit - n)) 958 n = limit - woff; 959 960 /* Will this write extend the file length? */ 961 write_eof = (woff + n > zp->z_size); 962 963 end_size = MAX(zp->z_size, woff + n); 964 965 /* 966 * Write the file in reasonable size chunks. Each chunk is written 967 * in a separate transaction; this keeps the intent log records small 968 * and allows us to do more fine-grained space accounting. 969 */ 970 while (n > 0) { 971 abuf = NULL; 972 woff = uio->uio_loffset; 973again: 974 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 975 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 976 if (abuf != NULL) 977 dmu_return_arcbuf(abuf); 978 error = SET_ERROR(EDQUOT); 979 break; 980 } 981 982 if (xuio && abuf == NULL) { 983 ASSERT(i_iov < iovcnt); 984 aiov = &iovp[i_iov]; 985 abuf = dmu_xuio_arcbuf(xuio, i_iov); 986 dmu_xuio_clear(xuio, i_iov); 987 DTRACE_PROBE3(zfs_cp_write, int, i_iov, 988 iovec_t *, aiov, arc_buf_t *, abuf); 989 ASSERT((aiov->iov_base == abuf->b_data) || 990 ((char *)aiov->iov_base - (char *)abuf->b_data + 991 aiov->iov_len == arc_buf_size(abuf))); 992 i_iov++; 993 } else if (abuf == NULL && n >= max_blksz && 994 woff >= zp->z_size && 995 P2PHASE(woff, max_blksz) == 0 && 996 zp->z_blksz == max_blksz) { 997 /* 998 * This write covers a full block. "Borrow" a buffer 999 * from the dmu so that we can fill it before we enter 1000 * a transaction. This avoids the possibility of 1001 * holding up the transaction if the data copy hangs 1002 * up on a pagefault (e.g., from an NFS server mapping). 1003 */ 1004 size_t cbytes; 1005 1006 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 1007 max_blksz); 1008 ASSERT(abuf != NULL); 1009 ASSERT(arc_buf_size(abuf) == max_blksz); 1010 if (error = uiocopy(abuf->b_data, max_blksz, 1011 UIO_WRITE, uio, &cbytes)) { 1012 dmu_return_arcbuf(abuf); 1013 break; 1014 } 1015 ASSERT(cbytes == max_blksz); 1016 } 1017 1018 /* 1019 * Start a transaction. 1020 */ 1021 tx = dmu_tx_create(zfsvfs->z_os); 1022 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1023 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 1024 zfs_sa_upgrade_txholds(tx, zp); 1025 error = dmu_tx_assign(tx, TXG_NOWAIT); 1026 if (error) { 1027 if (error == ERESTART) { 1028 dmu_tx_wait(tx); 1029 dmu_tx_abort(tx); 1030 goto again; 1031 } 1032 dmu_tx_abort(tx); 1033 if (abuf != NULL) 1034 dmu_return_arcbuf(abuf); 1035 break; 1036 } 1037 1038 /* 1039 * If zfs_range_lock() over-locked we grow the blocksize 1040 * and then reduce the lock range. This will only happen 1041 * on the first iteration since zfs_range_reduce() will 1042 * shrink down r_len to the appropriate size. 1043 */ 1044 if (rl->r_len == UINT64_MAX) { 1045 uint64_t new_blksz; 1046 1047 if (zp->z_blksz > max_blksz) { 1048 ASSERT(!ISP2(zp->z_blksz)); 1049 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 1050 } else { 1051 new_blksz = MIN(end_size, max_blksz); 1052 } 1053 zfs_grow_blocksize(zp, new_blksz, tx); 1054 zfs_range_reduce(rl, woff, n); 1055 } 1056 1057 /* 1058 * XXX - should we really limit each write to z_max_blksz? 1059 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 1060 */ 1061 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 1062 1063 if (woff + nbytes > zp->z_size) 1064 vnode_pager_setsize(vp, woff + nbytes); 1065 1066 if (abuf == NULL) { 1067 tx_bytes = uio->uio_resid; 1068 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), 1069 uio, nbytes, tx); 1070 tx_bytes -= uio->uio_resid; 1071 } else { 1072 tx_bytes = nbytes; 1073 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len); 1074 /* 1075 * If this is not a full block write, but we are 1076 * extending the file past EOF and this data starts 1077 * block-aligned, use assign_arcbuf(). Otherwise, 1078 * write via dmu_write(). 1079 */ 1080 if (tx_bytes < max_blksz && (!write_eof || 1081 aiov->iov_base != abuf->b_data)) { 1082 ASSERT(xuio); 1083 dmu_write(zfsvfs->z_os, zp->z_id, woff, 1084 aiov->iov_len, aiov->iov_base, tx); 1085 dmu_return_arcbuf(abuf); 1086 xuio_stat_wbuf_copied(); 1087 } else { 1088 ASSERT(xuio || tx_bytes == max_blksz); 1089 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), 1090 woff, abuf, tx); 1091 } 1092 ASSERT(tx_bytes <= uio->uio_resid); 1093 uioskip(uio, tx_bytes); 1094 } 1095 if (tx_bytes && vn_has_cached_data(vp)) { 1096 update_pages(vp, woff, tx_bytes, zfsvfs->z_os, 1097 zp->z_id, uio->uio_segflg, tx); 1098 } 1099 1100 /* 1101 * If we made no progress, we're done. If we made even 1102 * partial progress, update the znode and ZIL accordingly. 1103 */ 1104 if (tx_bytes == 0) { 1105 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 1106 (void *)&zp->z_size, sizeof (uint64_t), tx); 1107 dmu_tx_commit(tx); 1108 ASSERT(error != 0); 1109 break; 1110 } 1111 1112 /* 1113 * Clear Set-UID/Set-GID bits on successful write if not 1114 * privileged and at least one of the excute bits is set. 1115 * 1116 * It would be nice to to this after all writes have 1117 * been done, but that would still expose the ISUID/ISGID 1118 * to another app after the partial write is committed. 1119 * 1120 * Note: we don't call zfs_fuid_map_id() here because 1121 * user 0 is not an ephemeral uid. 1122 */ 1123 mutex_enter(&zp->z_acl_lock); 1124 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | 1125 (S_IXUSR >> 6))) != 0 && 1126 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && 1127 secpolicy_vnode_setid_retain(vp, cr, 1128 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) { 1129 uint64_t newmode; 1130 zp->z_mode &= ~(S_ISUID | S_ISGID); 1131 newmode = zp->z_mode; 1132 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), 1133 (void *)&newmode, sizeof (uint64_t), tx); 1134 } 1135 mutex_exit(&zp->z_acl_lock); 1136 1137 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 1138 B_TRUE); 1139 1140 /* 1141 * Update the file size (zp_size) if it has changed; 1142 * account for possible concurrent updates. 1143 */ 1144 while ((end_size = zp->z_size) < uio->uio_loffset) { 1145 (void) atomic_cas_64(&zp->z_size, end_size, 1146 uio->uio_loffset); 1147 ASSERT(error == 0); 1148 } 1149 /* 1150 * If we are replaying and eof is non zero then force 1151 * the file size to the specified eof. Note, there's no 1152 * concurrency during replay. 1153 */ 1154 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) 1155 zp->z_size = zfsvfs->z_replay_eof; 1156 1157 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1158 1159 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 1160 dmu_tx_commit(tx); 1161 1162 if (error != 0) 1163 break; 1164 ASSERT(tx_bytes == nbytes); 1165 n -= nbytes; 1166 1167#ifdef sun 1168 if (!xuio && n > 0) 1169 uio_prefaultpages(MIN(n, max_blksz), uio); 1170#endif /* sun */ 1171 } 1172 1173 zfs_range_unlock(rl); 1174 1175 /* 1176 * If we're in replay mode, or we made no progress, return error. 1177 * Otherwise, it's at least a partial write, so it's successful. 1178 */ 1179 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 1180 ZFS_EXIT(zfsvfs); 1181 return (error); 1182 } 1183 1184 if (ioflag & (FSYNC | FDSYNC) || 1185 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1186 zil_commit(zilog, zp->z_id); 1187 1188 ZFS_EXIT(zfsvfs); 1189 return (0); 1190} 1191 1192void 1193zfs_get_done(zgd_t *zgd, int error) 1194{ 1195 znode_t *zp = zgd->zgd_private; 1196 objset_t *os = zp->z_zfsvfs->z_os; 1197 1198 if (zgd->zgd_db) 1199 dmu_buf_rele(zgd->zgd_db, zgd); 1200 1201 zfs_range_unlock(zgd->zgd_rl); 1202 1203 /* 1204 * Release the vnode asynchronously as we currently have the 1205 * txg stopped from syncing. 1206 */ 1207 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1208 1209 if (error == 0 && zgd->zgd_bp) 1210 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1211 1212 kmem_free(zgd, sizeof (zgd_t)); 1213} 1214 1215#ifdef DEBUG 1216static int zil_fault_io = 0; 1217#endif 1218 1219/* 1220 * Get data to generate a TX_WRITE intent log record. 1221 */ 1222int 1223zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1224{ 1225 zfsvfs_t *zfsvfs = arg; 1226 objset_t *os = zfsvfs->z_os; 1227 znode_t *zp; 1228 uint64_t object = lr->lr_foid; 1229 uint64_t offset = lr->lr_offset; 1230 uint64_t size = lr->lr_length; 1231 blkptr_t *bp = &lr->lr_blkptr; 1232 dmu_buf_t *db; 1233 zgd_t *zgd; 1234 int error = 0; 1235 1236 ASSERT(zio != NULL); 1237 ASSERT(size != 0); 1238 1239 /* 1240 * Nothing to do if the file has been removed 1241 */ 1242 if (zfs_zget(zfsvfs, object, &zp) != 0) 1243 return (SET_ERROR(ENOENT)); 1244 if (zp->z_unlinked) { 1245 /* 1246 * Release the vnode asynchronously as we currently have the 1247 * txg stopped from syncing. 1248 */ 1249 VN_RELE_ASYNC(ZTOV(zp), 1250 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1251 return (SET_ERROR(ENOENT)); 1252 } 1253 1254 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1255 zgd->zgd_zilog = zfsvfs->z_log; 1256 zgd->zgd_private = zp; 1257 1258 /* 1259 * Write records come in two flavors: immediate and indirect. 1260 * For small writes it's cheaper to store the data with the 1261 * log record (immediate); for large writes it's cheaper to 1262 * sync the data and get a pointer to it (indirect) so that 1263 * we don't have to write the data twice. 1264 */ 1265 if (buf != NULL) { /* immediate write */ 1266 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER); 1267 /* test for truncation needs to be done while range locked */ 1268 if (offset >= zp->z_size) { 1269 error = SET_ERROR(ENOENT); 1270 } else { 1271 error = dmu_read(os, object, offset, size, buf, 1272 DMU_READ_NO_PREFETCH); 1273 } 1274 ASSERT(error == 0 || error == ENOENT); 1275 } else { /* indirect write */ 1276 /* 1277 * Have to lock the whole block to ensure when it's 1278 * written out and it's checksum is being calculated 1279 * that no one can change the data. We need to re-check 1280 * blocksize after we get the lock in case it's changed! 1281 */ 1282 for (;;) { 1283 uint64_t blkoff; 1284 size = zp->z_blksz; 1285 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; 1286 offset -= blkoff; 1287 zgd->zgd_rl = zfs_range_lock(zp, offset, size, 1288 RL_READER); 1289 if (zp->z_blksz == size) 1290 break; 1291 offset += blkoff; 1292 zfs_range_unlock(zgd->zgd_rl); 1293 } 1294 /* test for truncation needs to be done while range locked */ 1295 if (lr->lr_offset >= zp->z_size) 1296 error = SET_ERROR(ENOENT); 1297#ifdef DEBUG 1298 if (zil_fault_io) { 1299 error = SET_ERROR(EIO); 1300 zil_fault_io = 0; 1301 } 1302#endif 1303 if (error == 0) 1304 error = dmu_buf_hold(os, object, offset, zgd, &db, 1305 DMU_READ_NO_PREFETCH); 1306 1307 if (error == 0) { 1308 blkptr_t *obp = dmu_buf_get_blkptr(db); 1309 if (obp) { 1310 ASSERT(BP_IS_HOLE(bp)); 1311 *bp = *obp; 1312 } 1313 1314 zgd->zgd_db = db; 1315 zgd->zgd_bp = bp; 1316 1317 ASSERT(db->db_offset == offset); 1318 ASSERT(db->db_size == size); 1319 1320 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1321 zfs_get_done, zgd); 1322 ASSERT(error || lr->lr_length <= zp->z_blksz); 1323 1324 /* 1325 * On success, we need to wait for the write I/O 1326 * initiated by dmu_sync() to complete before we can 1327 * release this dbuf. We will finish everything up 1328 * in the zfs_get_done() callback. 1329 */ 1330 if (error == 0) 1331 return (0); 1332 1333 if (error == EALREADY) { 1334 lr->lr_common.lrc_txtype = TX_WRITE2; 1335 error = 0; 1336 } 1337 } 1338 } 1339 1340 zfs_get_done(zgd, error); 1341 1342 return (error); 1343} 1344 1345/*ARGSUSED*/ 1346static int 1347zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 1348 caller_context_t *ct) 1349{ 1350 znode_t *zp = VTOZ(vp); 1351 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1352 int error; 1353 1354 ZFS_ENTER(zfsvfs); 1355 ZFS_VERIFY_ZP(zp); 1356 1357 if (flag & V_ACE_MASK) 1358 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 1359 else 1360 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1361 1362 ZFS_EXIT(zfsvfs); 1363 return (error); 1364} 1365 1366/* 1367 * If vnode is for a device return a specfs vnode instead. 1368 */ 1369static int 1370specvp_check(vnode_t **vpp, cred_t *cr) 1371{ 1372 int error = 0; 1373 1374 if (IS_DEVVP(*vpp)) { 1375 struct vnode *svp; 1376 1377 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1378 VN_RELE(*vpp); 1379 if (svp == NULL) 1380 error = SET_ERROR(ENOSYS); 1381 *vpp = svp; 1382 } 1383 return (error); 1384} 1385 1386 1387/* 1388 * Lookup an entry in a directory, or an extended attribute directory. 1389 * If it exists, return a held vnode reference for it. 1390 * 1391 * IN: dvp - vnode of directory to search. 1392 * nm - name of entry to lookup. 1393 * pnp - full pathname to lookup [UNUSED]. 1394 * flags - LOOKUP_XATTR set if looking for an attribute. 1395 * rdir - root directory vnode [UNUSED]. 1396 * cr - credentials of caller. 1397 * ct - caller context 1398 * direntflags - directory lookup flags 1399 * realpnp - returned pathname. 1400 * 1401 * OUT: vpp - vnode of located entry, NULL if not found. 1402 * 1403 * RETURN: 0 on success, error code on failure. 1404 * 1405 * Timestamps: 1406 * NA 1407 */ 1408/* ARGSUSED */ 1409static int 1410zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 1411 int nameiop, cred_t *cr, kthread_t *td, int flags) 1412{ 1413 znode_t *zdp = VTOZ(dvp); 1414 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1415 int error = 0; 1416 int *direntflags = NULL; 1417 void *realpnp = NULL; 1418 1419 /* fast path */ 1420 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 1421 1422 if (dvp->v_type != VDIR) { 1423 return (SET_ERROR(ENOTDIR)); 1424 } else if (zdp->z_sa_hdl == NULL) { 1425 return (SET_ERROR(EIO)); 1426 } 1427 1428 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 1429 error = zfs_fastaccesschk_execute(zdp, cr); 1430 if (!error) { 1431 *vpp = dvp; 1432 VN_HOLD(*vpp); 1433 return (0); 1434 } 1435 return (error); 1436 } else { 1437 vnode_t *tvp = dnlc_lookup(dvp, nm); 1438 1439 if (tvp) { 1440 error = zfs_fastaccesschk_execute(zdp, cr); 1441 if (error) { 1442 VN_RELE(tvp); 1443 return (error); 1444 } 1445 if (tvp == DNLC_NO_VNODE) { 1446 VN_RELE(tvp); 1447 return (SET_ERROR(ENOENT)); 1448 } else { 1449 *vpp = tvp; 1450 return (specvp_check(vpp, cr)); 1451 } 1452 } 1453 } 1454 } 1455 1456 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1457 1458 ZFS_ENTER(zfsvfs); 1459 ZFS_VERIFY_ZP(zdp); 1460 1461 *vpp = NULL; 1462 1463 if (flags & LOOKUP_XATTR) { 1464#ifdef TODO 1465 /* 1466 * If the xattr property is off, refuse the lookup request. 1467 */ 1468 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1469 ZFS_EXIT(zfsvfs); 1470 return (SET_ERROR(EINVAL)); 1471 } 1472#endif 1473 1474 /* 1475 * We don't allow recursive attributes.. 1476 * Maybe someday we will. 1477 */ 1478 if (zdp->z_pflags & ZFS_XATTR) { 1479 ZFS_EXIT(zfsvfs); 1480 return (SET_ERROR(EINVAL)); 1481 } 1482 1483 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1484 ZFS_EXIT(zfsvfs); 1485 return (error); 1486 } 1487 1488 /* 1489 * Do we have permission to get into attribute directory? 1490 */ 1491 1492 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1493 B_FALSE, cr)) { 1494 VN_RELE(*vpp); 1495 *vpp = NULL; 1496 } 1497 1498 ZFS_EXIT(zfsvfs); 1499 return (error); 1500 } 1501 1502 if (dvp->v_type != VDIR) { 1503 ZFS_EXIT(zfsvfs); 1504 return (SET_ERROR(ENOTDIR)); 1505 } 1506 1507 /* 1508 * Check accessibility of directory. 1509 */ 1510 1511 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1512 ZFS_EXIT(zfsvfs); 1513 return (error); 1514 } 1515 1516 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1517 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1518 ZFS_EXIT(zfsvfs); 1519 return (SET_ERROR(EILSEQ)); 1520 } 1521 1522 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1523 if (error == 0) 1524 error = specvp_check(vpp, cr); 1525 1526 /* Translate errors and add SAVENAME when needed. */ 1527 if (cnp->cn_flags & ISLASTCN) { 1528 switch (nameiop) { 1529 case CREATE: 1530 case RENAME: 1531 if (error == ENOENT) { 1532 error = EJUSTRETURN; 1533 cnp->cn_flags |= SAVENAME; 1534 break; 1535 } 1536 /* FALLTHROUGH */ 1537 case DELETE: 1538 if (error == 0) 1539 cnp->cn_flags |= SAVENAME; 1540 break; 1541 } 1542 } 1543 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) { 1544 int ltype = 0; 1545 1546 if (cnp->cn_flags & ISDOTDOT) { 1547 ltype = VOP_ISLOCKED(dvp); 1548 VOP_UNLOCK(dvp, 0); 1549 } 1550 ZFS_EXIT(zfsvfs); 1551 error = vn_lock(*vpp, cnp->cn_lkflags); 1552 if (cnp->cn_flags & ISDOTDOT) 1553 vn_lock(dvp, ltype | LK_RETRY); 1554 if (error != 0) { 1555 VN_RELE(*vpp); 1556 *vpp = NULL; 1557 return (error); 1558 } 1559 } else { 1560 ZFS_EXIT(zfsvfs); 1561 } 1562 1563#ifdef FREEBSD_NAMECACHE 1564 /* 1565 * Insert name into cache (as non-existent) if appropriate. 1566 */ 1567 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE) 1568 cache_enter(dvp, *vpp, cnp); 1569 /* 1570 * Insert name into cache if appropriate. 1571 */ 1572 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1573 if (!(cnp->cn_flags & ISLASTCN) || 1574 (nameiop != DELETE && nameiop != RENAME)) { 1575 cache_enter(dvp, *vpp, cnp); 1576 } 1577 } 1578#endif 1579 1580 return (error); 1581} 1582 1583/* 1584 * Attempt to create a new entry in a directory. If the entry 1585 * already exists, truncate the file if permissible, else return 1586 * an error. Return the vp of the created or trunc'd file. 1587 * 1588 * IN: dvp - vnode of directory to put new file entry in. 1589 * name - name of new file entry. 1590 * vap - attributes of new file. 1591 * excl - flag indicating exclusive or non-exclusive mode. 1592 * mode - mode to open file with. 1593 * cr - credentials of caller. 1594 * flag - large file flag [UNUSED]. 1595 * ct - caller context 1596 * vsecp - ACL to be set 1597 * 1598 * OUT: vpp - vnode of created or trunc'd entry. 1599 * 1600 * RETURN: 0 on success, error code on failure. 1601 * 1602 * Timestamps: 1603 * dvp - ctime|mtime updated if new entry created 1604 * vp - ctime|mtime always, atime if new 1605 */ 1606 1607/* ARGSUSED */ 1608static int 1609zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1610 vnode_t **vpp, cred_t *cr, kthread_t *td) 1611{ 1612 znode_t *zp, *dzp = VTOZ(dvp); 1613 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1614 zilog_t *zilog; 1615 objset_t *os; 1616 zfs_dirlock_t *dl; 1617 dmu_tx_t *tx; 1618 int error; 1619 ksid_t *ksid; 1620 uid_t uid; 1621 gid_t gid = crgetgid(cr); 1622 zfs_acl_ids_t acl_ids; 1623 boolean_t fuid_dirtied; 1624 boolean_t have_acl = B_FALSE; 1625 boolean_t waited = B_FALSE; 1626 void *vsecp = NULL; 1627 int flag = 0; 1628 1629 /* 1630 * If we have an ephemeral id, ACL, or XVATTR then 1631 * make sure file system is at proper version 1632 */ 1633 1634 ksid = crgetsid(cr, KSID_OWNER); 1635 if (ksid) 1636 uid = ksid_getid(ksid); 1637 else 1638 uid = crgetuid(cr); 1639 1640 if (zfsvfs->z_use_fuids == B_FALSE && 1641 (vsecp || (vap->va_mask & AT_XVATTR) || 1642 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1643 return (SET_ERROR(EINVAL)); 1644 1645 ZFS_ENTER(zfsvfs); 1646 ZFS_VERIFY_ZP(dzp); 1647 os = zfsvfs->z_os; 1648 zilog = zfsvfs->z_log; 1649 1650 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1651 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1652 ZFS_EXIT(zfsvfs); 1653 return (SET_ERROR(EILSEQ)); 1654 } 1655 1656 if (vap->va_mask & AT_XVATTR) { 1657 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 1658 crgetuid(cr), cr, vap->va_type)) != 0) { 1659 ZFS_EXIT(zfsvfs); 1660 return (error); 1661 } 1662 } 1663top: 1664 *vpp = NULL; 1665 1666 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) 1667 vap->va_mode &= ~S_ISVTX; 1668 1669 if (*name == '\0') { 1670 /* 1671 * Null component name refers to the directory itself. 1672 */ 1673 VN_HOLD(dvp); 1674 zp = dzp; 1675 dl = NULL; 1676 error = 0; 1677 } else { 1678 /* possible VN_HOLD(zp) */ 1679 int zflg = 0; 1680 1681 if (flag & FIGNORECASE) 1682 zflg |= ZCILOOK; 1683 1684 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1685 NULL, NULL); 1686 if (error) { 1687 if (have_acl) 1688 zfs_acl_ids_free(&acl_ids); 1689 if (strcmp(name, "..") == 0) 1690 error = SET_ERROR(EISDIR); 1691 ZFS_EXIT(zfsvfs); 1692 return (error); 1693 } 1694 } 1695 1696 if (zp == NULL) { 1697 uint64_t txtype; 1698 1699 /* 1700 * Create a new file object and update the directory 1701 * to reference it. 1702 */ 1703 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1704 if (have_acl) 1705 zfs_acl_ids_free(&acl_ids); 1706 goto out; 1707 } 1708 1709 /* 1710 * We only support the creation of regular files in 1711 * extended attribute directories. 1712 */ 1713 1714 if ((dzp->z_pflags & ZFS_XATTR) && 1715 (vap->va_type != VREG)) { 1716 if (have_acl) 1717 zfs_acl_ids_free(&acl_ids); 1718 error = SET_ERROR(EINVAL); 1719 goto out; 1720 } 1721 1722 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 1723 cr, vsecp, &acl_ids)) != 0) 1724 goto out; 1725 have_acl = B_TRUE; 1726 1727 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1728 zfs_acl_ids_free(&acl_ids); 1729 error = SET_ERROR(EDQUOT); 1730 goto out; 1731 } 1732 1733 tx = dmu_tx_create(os); 1734 1735 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1736 ZFS_SA_BASE_ATTR_SIZE); 1737 1738 fuid_dirtied = zfsvfs->z_fuid_dirty; 1739 if (fuid_dirtied) 1740 zfs_fuid_txhold(zfsvfs, tx); 1741 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1742 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1743 if (!zfsvfs->z_use_sa && 1744 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1745 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1746 0, acl_ids.z_aclp->z_acl_bytes); 1747 } 1748 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 1749 if (error) { 1750 zfs_dirent_unlock(dl); 1751 if (error == ERESTART) { 1752 waited = B_TRUE; 1753 dmu_tx_wait(tx); 1754 dmu_tx_abort(tx); 1755 goto top; 1756 } 1757 zfs_acl_ids_free(&acl_ids); 1758 dmu_tx_abort(tx); 1759 ZFS_EXIT(zfsvfs); 1760 return (error); 1761 } 1762 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1763 1764 if (fuid_dirtied) 1765 zfs_fuid_sync(zfsvfs, tx); 1766 1767 (void) zfs_link_create(dl, zp, tx, ZNEW); 1768 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1769 if (flag & FIGNORECASE) 1770 txtype |= TX_CI; 1771 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1772 vsecp, acl_ids.z_fuidp, vap); 1773 zfs_acl_ids_free(&acl_ids); 1774 dmu_tx_commit(tx); 1775 } else { 1776 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1777 1778 if (have_acl) 1779 zfs_acl_ids_free(&acl_ids); 1780 have_acl = B_FALSE; 1781 1782 /* 1783 * A directory entry already exists for this name. 1784 */ 1785 /* 1786 * Can't truncate an existing file if in exclusive mode. 1787 */ 1788 if (excl == EXCL) { 1789 error = SET_ERROR(EEXIST); 1790 goto out; 1791 } 1792 /* 1793 * Can't open a directory for writing. 1794 */ 1795 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1796 error = SET_ERROR(EISDIR); 1797 goto out; 1798 } 1799 /* 1800 * Verify requested access to file. 1801 */ 1802 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1803 goto out; 1804 } 1805 1806 mutex_enter(&dzp->z_lock); 1807 dzp->z_seq++; 1808 mutex_exit(&dzp->z_lock); 1809 1810 /* 1811 * Truncate regular files if requested. 1812 */ 1813 if ((ZTOV(zp)->v_type == VREG) && 1814 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1815 /* we can't hold any locks when calling zfs_freesp() */ 1816 zfs_dirent_unlock(dl); 1817 dl = NULL; 1818 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1819 if (error == 0) { 1820 vnevent_create(ZTOV(zp), ct); 1821 } 1822 } 1823 } 1824out: 1825 if (dl) 1826 zfs_dirent_unlock(dl); 1827 1828 if (error) { 1829 if (zp) 1830 VN_RELE(ZTOV(zp)); 1831 } else { 1832 *vpp = ZTOV(zp); 1833 error = specvp_check(vpp, cr); 1834 } 1835 1836 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1837 zil_commit(zilog, 0); 1838 1839 ZFS_EXIT(zfsvfs); 1840 return (error); 1841} 1842 1843/* 1844 * Remove an entry from a directory. 1845 * 1846 * IN: dvp - vnode of directory to remove entry from. 1847 * name - name of entry to remove. 1848 * cr - credentials of caller. 1849 * ct - caller context 1850 * flags - case flags 1851 * 1852 * RETURN: 0 on success, error code on failure. 1853 * 1854 * Timestamps: 1855 * dvp - ctime|mtime 1856 * vp - ctime (if nlink > 0) 1857 */ 1858 1859uint64_t null_xattr = 0; 1860 1861/*ARGSUSED*/ 1862static int 1863zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1864 int flags) 1865{ 1866 znode_t *zp, *dzp = VTOZ(dvp); 1867 znode_t *xzp; 1868 vnode_t *vp; 1869 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1870 zilog_t *zilog; 1871 uint64_t acl_obj, xattr_obj; 1872 uint64_t xattr_obj_unlinked = 0; 1873 uint64_t obj = 0; 1874 zfs_dirlock_t *dl; 1875 dmu_tx_t *tx; 1876 boolean_t may_delete_now, delete_now = FALSE; 1877 boolean_t unlinked, toobig = FALSE; 1878 uint64_t txtype; 1879 pathname_t *realnmp = NULL; 1880 pathname_t realnm; 1881 int error; 1882 int zflg = ZEXISTS; 1883 boolean_t waited = B_FALSE; 1884 1885 ZFS_ENTER(zfsvfs); 1886 ZFS_VERIFY_ZP(dzp); 1887 zilog = zfsvfs->z_log; 1888 1889 if (flags & FIGNORECASE) { 1890 zflg |= ZCILOOK; 1891 pn_alloc(&realnm); 1892 realnmp = &realnm; 1893 } 1894 1895top: 1896 xattr_obj = 0; 1897 xzp = NULL; 1898 /* 1899 * Attempt to lock directory; fail if entry doesn't exist. 1900 */ 1901 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1902 NULL, realnmp)) { 1903 if (realnmp) 1904 pn_free(realnmp); 1905 ZFS_EXIT(zfsvfs); 1906 return (error); 1907 } 1908 1909 vp = ZTOV(zp); 1910 1911 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1912 goto out; 1913 } 1914 1915 /* 1916 * Need to use rmdir for removing directories. 1917 */ 1918 if (vp->v_type == VDIR) { 1919 error = SET_ERROR(EPERM); 1920 goto out; 1921 } 1922 1923 vnevent_remove(vp, dvp, name, ct); 1924 1925 if (realnmp) 1926 dnlc_remove(dvp, realnmp->pn_buf); 1927 else 1928 dnlc_remove(dvp, name); 1929 1930 VI_LOCK(vp); 1931 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1932 VI_UNLOCK(vp); 1933 1934 /* 1935 * We may delete the znode now, or we may put it in the unlinked set; 1936 * it depends on whether we're the last link, and on whether there are 1937 * other holds on the vnode. So we dmu_tx_hold() the right things to 1938 * allow for either case. 1939 */ 1940 obj = zp->z_id; 1941 tx = dmu_tx_create(zfsvfs->z_os); 1942 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1943 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1944 zfs_sa_upgrade_txholds(tx, zp); 1945 zfs_sa_upgrade_txholds(tx, dzp); 1946 if (may_delete_now) { 1947 toobig = 1948 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 1949 /* if the file is too big, only hold_free a token amount */ 1950 dmu_tx_hold_free(tx, zp->z_id, 0, 1951 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1952 } 1953 1954 /* are there any extended attributes? */ 1955 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1956 &xattr_obj, sizeof (xattr_obj)); 1957 if (error == 0 && xattr_obj) { 1958 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1959 ASSERT0(error); 1960 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1961 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1962 } 1963 1964 mutex_enter(&zp->z_lock); 1965 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now) 1966 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1967 mutex_exit(&zp->z_lock); 1968 1969 /* charge as an update -- would be nice not to charge at all */ 1970 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1971 1972 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 1973 if (error) { 1974 zfs_dirent_unlock(dl); 1975 VN_RELE(vp); 1976 if (xzp) 1977 VN_RELE(ZTOV(xzp)); 1978 if (error == ERESTART) { 1979 waited = B_TRUE; 1980 dmu_tx_wait(tx); 1981 dmu_tx_abort(tx); 1982 goto top; 1983 } 1984 if (realnmp) 1985 pn_free(realnmp); 1986 dmu_tx_abort(tx); 1987 ZFS_EXIT(zfsvfs); 1988 return (error); 1989 } 1990 1991 /* 1992 * Remove the directory entry. 1993 */ 1994 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1995 1996 if (error) { 1997 dmu_tx_commit(tx); 1998 goto out; 1999 } 2000 2001 if (unlinked) { 2002 2003 /* 2004 * Hold z_lock so that we can make sure that the ACL obj 2005 * hasn't changed. Could have been deleted due to 2006 * zfs_sa_upgrade(). 2007 */ 2008 mutex_enter(&zp->z_lock); 2009 VI_LOCK(vp); 2010 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 2011 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked)); 2012 delete_now = may_delete_now && !toobig && 2013 vp->v_count == 1 && !vn_has_cached_data(vp) && 2014 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) == 2015 acl_obj; 2016 VI_UNLOCK(vp); 2017 } 2018 2019 if (delete_now) { 2020#ifdef __FreeBSD__ 2021 panic("zfs_remove: delete_now branch taken"); 2022#endif 2023 if (xattr_obj_unlinked) { 2024 ASSERT3U(xzp->z_links, ==, 2); 2025 mutex_enter(&xzp->z_lock); 2026 xzp->z_unlinked = 1; 2027 xzp->z_links = 0; 2028 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 2029 &xzp->z_links, sizeof (xzp->z_links), tx); 2030 ASSERT3U(error, ==, 0); 2031 mutex_exit(&xzp->z_lock); 2032 zfs_unlinked_add(xzp, tx); 2033 2034 if (zp->z_is_sa) 2035 error = sa_remove(zp->z_sa_hdl, 2036 SA_ZPL_XATTR(zfsvfs), tx); 2037 else 2038 error = sa_update(zp->z_sa_hdl, 2039 SA_ZPL_XATTR(zfsvfs), &null_xattr, 2040 sizeof (uint64_t), tx); 2041 ASSERT0(error); 2042 } 2043 VI_LOCK(vp); 2044 vp->v_count--; 2045 ASSERT0(vp->v_count); 2046 VI_UNLOCK(vp); 2047 mutex_exit(&zp->z_lock); 2048 zfs_znode_delete(zp, tx); 2049 } else if (unlinked) { 2050 mutex_exit(&zp->z_lock); 2051 zfs_unlinked_add(zp, tx); 2052#ifdef __FreeBSD__ 2053 vp->v_vflag |= VV_NOSYNC; 2054#endif 2055 } 2056 2057 txtype = TX_REMOVE; 2058 if (flags & FIGNORECASE) 2059 txtype |= TX_CI; 2060 zfs_log_remove(zilog, tx, txtype, dzp, name, obj); 2061 2062 dmu_tx_commit(tx); 2063out: 2064 if (realnmp) 2065 pn_free(realnmp); 2066 2067 zfs_dirent_unlock(dl); 2068 2069 if (!delete_now) 2070 VN_RELE(vp); 2071 if (xzp) 2072 VN_RELE(ZTOV(xzp)); 2073 2074 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2075 zil_commit(zilog, 0); 2076 2077 ZFS_EXIT(zfsvfs); 2078 return (error); 2079} 2080 2081/* 2082 * Create a new directory and insert it into dvp using the name 2083 * provided. Return a pointer to the inserted directory. 2084 * 2085 * IN: dvp - vnode of directory to add subdir to. 2086 * dirname - name of new directory. 2087 * vap - attributes of new directory. 2088 * cr - credentials of caller. 2089 * ct - caller context 2090 * flags - case flags 2091 * vsecp - ACL to be set 2092 * 2093 * OUT: vpp - vnode of created directory. 2094 * 2095 * RETURN: 0 on success, error code on failure. 2096 * 2097 * Timestamps: 2098 * dvp - ctime|mtime updated 2099 * vp - ctime|mtime|atime updated 2100 */ 2101/*ARGSUSED*/ 2102static int 2103zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 2104 caller_context_t *ct, int flags, vsecattr_t *vsecp) 2105{ 2106 znode_t *zp, *dzp = VTOZ(dvp); 2107 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2108 zilog_t *zilog; 2109 zfs_dirlock_t *dl; 2110 uint64_t txtype; 2111 dmu_tx_t *tx; 2112 int error; 2113 int zf = ZNEW; 2114 ksid_t *ksid; 2115 uid_t uid; 2116 gid_t gid = crgetgid(cr); 2117 zfs_acl_ids_t acl_ids; 2118 boolean_t fuid_dirtied; 2119 boolean_t waited = B_FALSE; 2120 2121 ASSERT(vap->va_type == VDIR); 2122 2123 /* 2124 * If we have an ephemeral id, ACL, or XVATTR then 2125 * make sure file system is at proper version 2126 */ 2127 2128 ksid = crgetsid(cr, KSID_OWNER); 2129 if (ksid) 2130 uid = ksid_getid(ksid); 2131 else 2132 uid = crgetuid(cr); 2133 if (zfsvfs->z_use_fuids == B_FALSE && 2134 (vsecp || (vap->va_mask & AT_XVATTR) || 2135 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 2136 return (SET_ERROR(EINVAL)); 2137 2138 ZFS_ENTER(zfsvfs); 2139 ZFS_VERIFY_ZP(dzp); 2140 zilog = zfsvfs->z_log; 2141 2142 if (dzp->z_pflags & ZFS_XATTR) { 2143 ZFS_EXIT(zfsvfs); 2144 return (SET_ERROR(EINVAL)); 2145 } 2146 2147 if (zfsvfs->z_utf8 && u8_validate(dirname, 2148 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2149 ZFS_EXIT(zfsvfs); 2150 return (SET_ERROR(EILSEQ)); 2151 } 2152 if (flags & FIGNORECASE) 2153 zf |= ZCILOOK; 2154 2155 if (vap->va_mask & AT_XVATTR) { 2156 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 2157 crgetuid(cr), cr, vap->va_type)) != 0) { 2158 ZFS_EXIT(zfsvfs); 2159 return (error); 2160 } 2161 } 2162 2163 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 2164 vsecp, &acl_ids)) != 0) { 2165 ZFS_EXIT(zfsvfs); 2166 return (error); 2167 } 2168 /* 2169 * First make sure the new directory doesn't exist. 2170 * 2171 * Existence is checked first to make sure we don't return 2172 * EACCES instead of EEXIST which can cause some applications 2173 * to fail. 2174 */ 2175top: 2176 *vpp = NULL; 2177 2178 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 2179 NULL, NULL)) { 2180 zfs_acl_ids_free(&acl_ids); 2181 ZFS_EXIT(zfsvfs); 2182 return (error); 2183 } 2184 2185 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 2186 zfs_acl_ids_free(&acl_ids); 2187 zfs_dirent_unlock(dl); 2188 ZFS_EXIT(zfsvfs); 2189 return (error); 2190 } 2191 2192 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 2193 zfs_acl_ids_free(&acl_ids); 2194 zfs_dirent_unlock(dl); 2195 ZFS_EXIT(zfsvfs); 2196 return (SET_ERROR(EDQUOT)); 2197 } 2198 2199 /* 2200 * Add a new entry to the directory. 2201 */ 2202 tx = dmu_tx_create(zfsvfs->z_os); 2203 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 2204 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2205 fuid_dirtied = zfsvfs->z_fuid_dirty; 2206 if (fuid_dirtied) 2207 zfs_fuid_txhold(zfsvfs, tx); 2208 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2209 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2210 acl_ids.z_aclp->z_acl_bytes); 2211 } 2212 2213 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 2214 ZFS_SA_BASE_ATTR_SIZE); 2215 2216 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2217 if (error) { 2218 zfs_dirent_unlock(dl); 2219 if (error == ERESTART) { 2220 waited = B_TRUE; 2221 dmu_tx_wait(tx); 2222 dmu_tx_abort(tx); 2223 goto top; 2224 } 2225 zfs_acl_ids_free(&acl_ids); 2226 dmu_tx_abort(tx); 2227 ZFS_EXIT(zfsvfs); 2228 return (error); 2229 } 2230 2231 /* 2232 * Create new node. 2233 */ 2234 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 2235 2236 if (fuid_dirtied) 2237 zfs_fuid_sync(zfsvfs, tx); 2238 2239 /* 2240 * Now put new name in parent dir. 2241 */ 2242 (void) zfs_link_create(dl, zp, tx, ZNEW); 2243 2244 *vpp = ZTOV(zp); 2245 2246 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 2247 if (flags & FIGNORECASE) 2248 txtype |= TX_CI; 2249 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 2250 acl_ids.z_fuidp, vap); 2251 2252 zfs_acl_ids_free(&acl_ids); 2253 2254 dmu_tx_commit(tx); 2255 2256 zfs_dirent_unlock(dl); 2257 2258 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2259 zil_commit(zilog, 0); 2260 2261 ZFS_EXIT(zfsvfs); 2262 return (0); 2263} 2264 2265/* 2266 * Remove a directory subdir entry. If the current working 2267 * directory is the same as the subdir to be removed, the 2268 * remove will fail. 2269 * 2270 * IN: dvp - vnode of directory to remove from. 2271 * name - name of directory to be removed. 2272 * cwd - vnode of current working directory. 2273 * cr - credentials of caller. 2274 * ct - caller context 2275 * flags - case flags 2276 * 2277 * RETURN: 0 on success, error code on failure. 2278 * 2279 * Timestamps: 2280 * dvp - ctime|mtime updated 2281 */ 2282/*ARGSUSED*/ 2283static int 2284zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 2285 caller_context_t *ct, int flags) 2286{ 2287 znode_t *dzp = VTOZ(dvp); 2288 znode_t *zp; 2289 vnode_t *vp; 2290 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2291 zilog_t *zilog; 2292 zfs_dirlock_t *dl; 2293 dmu_tx_t *tx; 2294 int error; 2295 int zflg = ZEXISTS; 2296 boolean_t waited = B_FALSE; 2297 2298 ZFS_ENTER(zfsvfs); 2299 ZFS_VERIFY_ZP(dzp); 2300 zilog = zfsvfs->z_log; 2301 2302 if (flags & FIGNORECASE) 2303 zflg |= ZCILOOK; 2304top: 2305 zp = NULL; 2306 2307 /* 2308 * Attempt to lock directory; fail if entry doesn't exist. 2309 */ 2310 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 2311 NULL, NULL)) { 2312 ZFS_EXIT(zfsvfs); 2313 return (error); 2314 } 2315 2316 vp = ZTOV(zp); 2317 2318 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 2319 goto out; 2320 } 2321 2322 if (vp->v_type != VDIR) { 2323 error = SET_ERROR(ENOTDIR); 2324 goto out; 2325 } 2326 2327 if (vp == cwd) { 2328 error = SET_ERROR(EINVAL); 2329 goto out; 2330 } 2331 2332 vnevent_rmdir(vp, dvp, name, ct); 2333 2334 /* 2335 * Grab a lock on the directory to make sure that noone is 2336 * trying to add (or lookup) entries while we are removing it. 2337 */ 2338 rw_enter(&zp->z_name_lock, RW_WRITER); 2339 2340 /* 2341 * Grab a lock on the parent pointer to make sure we play well 2342 * with the treewalk and directory rename code. 2343 */ 2344 rw_enter(&zp->z_parent_lock, RW_WRITER); 2345 2346 tx = dmu_tx_create(zfsvfs->z_os); 2347 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2348 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2349 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2350 zfs_sa_upgrade_txholds(tx, zp); 2351 zfs_sa_upgrade_txholds(tx, dzp); 2352 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2353 if (error) { 2354 rw_exit(&zp->z_parent_lock); 2355 rw_exit(&zp->z_name_lock); 2356 zfs_dirent_unlock(dl); 2357 VN_RELE(vp); 2358 if (error == ERESTART) { 2359 waited = B_TRUE; 2360 dmu_tx_wait(tx); 2361 dmu_tx_abort(tx); 2362 goto top; 2363 } 2364 dmu_tx_abort(tx); 2365 ZFS_EXIT(zfsvfs); 2366 return (error); 2367 } 2368 2369#ifdef FREEBSD_NAMECACHE 2370 cache_purge(dvp); 2371#endif 2372 2373 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 2374 2375 if (error == 0) { 2376 uint64_t txtype = TX_RMDIR; 2377 if (flags & FIGNORECASE) 2378 txtype |= TX_CI; 2379 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT); 2380 } 2381 2382 dmu_tx_commit(tx); 2383 2384 rw_exit(&zp->z_parent_lock); 2385 rw_exit(&zp->z_name_lock); 2386#ifdef FREEBSD_NAMECACHE 2387 cache_purge(vp); 2388#endif 2389out: 2390 zfs_dirent_unlock(dl); 2391 2392 VN_RELE(vp); 2393 2394 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2395 zil_commit(zilog, 0); 2396 2397 ZFS_EXIT(zfsvfs); 2398 return (error); 2399} 2400 2401/* 2402 * Read as many directory entries as will fit into the provided 2403 * buffer from the given directory cursor position (specified in 2404 * the uio structure). 2405 * 2406 * IN: vp - vnode of directory to read. 2407 * uio - structure supplying read location, range info, 2408 * and return buffer. 2409 * cr - credentials of caller. 2410 * ct - caller context 2411 * flags - case flags 2412 * 2413 * OUT: uio - updated offset and range, buffer filled. 2414 * eofp - set to true if end-of-file detected. 2415 * 2416 * RETURN: 0 on success, error code on failure. 2417 * 2418 * Timestamps: 2419 * vp - atime updated 2420 * 2421 * Note that the low 4 bits of the cookie returned by zap is always zero. 2422 * This allows us to use the low range for "special" directory entries: 2423 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 2424 * we use the offset 2 for the '.zfs' directory. 2425 */ 2426/* ARGSUSED */ 2427static int 2428zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 2429{ 2430 znode_t *zp = VTOZ(vp); 2431 iovec_t *iovp; 2432 edirent_t *eodp; 2433 dirent64_t *odp; 2434 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2435 objset_t *os; 2436 caddr_t outbuf; 2437 size_t bufsize; 2438 zap_cursor_t zc; 2439 zap_attribute_t zap; 2440 uint_t bytes_wanted; 2441 uint64_t offset; /* must be unsigned; checks for < 1 */ 2442 uint64_t parent; 2443 int local_eof; 2444 int outcount; 2445 int error; 2446 uint8_t prefetch; 2447 boolean_t check_sysattrs; 2448 uint8_t type; 2449 int ncooks; 2450 u_long *cooks = NULL; 2451 int flags = 0; 2452 2453 ZFS_ENTER(zfsvfs); 2454 ZFS_VERIFY_ZP(zp); 2455 2456 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 2457 &parent, sizeof (parent))) != 0) { 2458 ZFS_EXIT(zfsvfs); 2459 return (error); 2460 } 2461 2462 /* 2463 * If we are not given an eof variable, 2464 * use a local one. 2465 */ 2466 if (eofp == NULL) 2467 eofp = &local_eof; 2468 2469 /* 2470 * Check for valid iov_len. 2471 */ 2472 if (uio->uio_iov->iov_len <= 0) { 2473 ZFS_EXIT(zfsvfs); 2474 return (SET_ERROR(EINVAL)); 2475 } 2476 2477 /* 2478 * Quit if directory has been removed (posix) 2479 */ 2480 if ((*eofp = zp->z_unlinked) != 0) { 2481 ZFS_EXIT(zfsvfs); 2482 return (0); 2483 } 2484 2485 error = 0; 2486 os = zfsvfs->z_os; 2487 offset = uio->uio_loffset; 2488 prefetch = zp->z_zn_prefetch; 2489 2490 /* 2491 * Initialize the iterator cursor. 2492 */ 2493 if (offset <= 3) { 2494 /* 2495 * Start iteration from the beginning of the directory. 2496 */ 2497 zap_cursor_init(&zc, os, zp->z_id); 2498 } else { 2499 /* 2500 * The offset is a serialized cursor. 2501 */ 2502 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 2503 } 2504 2505 /* 2506 * Get space to change directory entries into fs independent format. 2507 */ 2508 iovp = uio->uio_iov; 2509 bytes_wanted = iovp->iov_len; 2510 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 2511 bufsize = bytes_wanted; 2512 outbuf = kmem_alloc(bufsize, KM_SLEEP); 2513 odp = (struct dirent64 *)outbuf; 2514 } else { 2515 bufsize = bytes_wanted; 2516 outbuf = NULL; 2517 odp = (struct dirent64 *)iovp->iov_base; 2518 } 2519 eodp = (struct edirent *)odp; 2520 2521 if (ncookies != NULL) { 2522 /* 2523 * Minimum entry size is dirent size and 1 byte for a file name. 2524 */ 2525 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 2526 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 2527 *cookies = cooks; 2528 *ncookies = ncooks; 2529 } 2530 /* 2531 * If this VFS supports the system attribute view interface; and 2532 * we're looking at an extended attribute directory; and we care 2533 * about normalization conflicts on this vfs; then we must check 2534 * for normalization conflicts with the sysattr name space. 2535 */ 2536#ifdef TODO 2537 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 2538 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 2539 (flags & V_RDDIR_ENTFLAGS); 2540#else 2541 check_sysattrs = 0; 2542#endif 2543 2544 /* 2545 * Transform to file-system independent format 2546 */ 2547 outcount = 0; 2548 while (outcount < bytes_wanted) { 2549 ino64_t objnum; 2550 ushort_t reclen; 2551 off64_t *next = NULL; 2552 2553 /* 2554 * Special case `.', `..', and `.zfs'. 2555 */ 2556 if (offset == 0) { 2557 (void) strcpy(zap.za_name, "."); 2558 zap.za_normalization_conflict = 0; 2559 objnum = zp->z_id; 2560 type = DT_DIR; 2561 } else if (offset == 1) { 2562 (void) strcpy(zap.za_name, ".."); 2563 zap.za_normalization_conflict = 0; 2564 objnum = parent; 2565 type = DT_DIR; 2566 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2567 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2568 zap.za_normalization_conflict = 0; 2569 objnum = ZFSCTL_INO_ROOT; 2570 type = DT_DIR; 2571 } else { 2572 /* 2573 * Grab next entry. 2574 */ 2575 if (error = zap_cursor_retrieve(&zc, &zap)) { 2576 if ((*eofp = (error == ENOENT)) != 0) 2577 break; 2578 else 2579 goto update; 2580 } 2581 2582 if (zap.za_integer_length != 8 || 2583 zap.za_num_integers != 1) { 2584 cmn_err(CE_WARN, "zap_readdir: bad directory " 2585 "entry, obj = %lld, offset = %lld\n", 2586 (u_longlong_t)zp->z_id, 2587 (u_longlong_t)offset); 2588 error = SET_ERROR(ENXIO); 2589 goto update; 2590 } 2591 2592 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2593 /* 2594 * MacOS X can extract the object type here such as: 2595 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2596 */ 2597 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2598 2599 if (check_sysattrs && !zap.za_normalization_conflict) { 2600#ifdef TODO 2601 zap.za_normalization_conflict = 2602 xattr_sysattr_casechk(zap.za_name); 2603#else 2604 panic("%s:%u: TODO", __func__, __LINE__); 2605#endif 2606 } 2607 } 2608 2609 if (flags & V_RDDIR_ACCFILTER) { 2610 /* 2611 * If we have no access at all, don't include 2612 * this entry in the returned information 2613 */ 2614 znode_t *ezp; 2615 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2616 goto skip_entry; 2617 if (!zfs_has_access(ezp, cr)) { 2618 VN_RELE(ZTOV(ezp)); 2619 goto skip_entry; 2620 } 2621 VN_RELE(ZTOV(ezp)); 2622 } 2623 2624 if (flags & V_RDDIR_ENTFLAGS) 2625 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2626 else 2627 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2628 2629 /* 2630 * Will this entry fit in the buffer? 2631 */ 2632 if (outcount + reclen > bufsize) { 2633 /* 2634 * Did we manage to fit anything in the buffer? 2635 */ 2636 if (!outcount) { 2637 error = SET_ERROR(EINVAL); 2638 goto update; 2639 } 2640 break; 2641 } 2642 if (flags & V_RDDIR_ENTFLAGS) { 2643 /* 2644 * Add extended flag entry: 2645 */ 2646 eodp->ed_ino = objnum; 2647 eodp->ed_reclen = reclen; 2648 /* NOTE: ed_off is the offset for the *next* entry */ 2649 next = &(eodp->ed_off); 2650 eodp->ed_eflags = zap.za_normalization_conflict ? 2651 ED_CASE_CONFLICT : 0; 2652 (void) strncpy(eodp->ed_name, zap.za_name, 2653 EDIRENT_NAMELEN(reclen)); 2654 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2655 } else { 2656 /* 2657 * Add normal entry: 2658 */ 2659 odp->d_ino = objnum; 2660 odp->d_reclen = reclen; 2661 odp->d_namlen = strlen(zap.za_name); 2662 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 2663 odp->d_type = type; 2664 odp = (dirent64_t *)((intptr_t)odp + reclen); 2665 } 2666 outcount += reclen; 2667 2668 ASSERT(outcount <= bufsize); 2669 2670 /* Prefetch znode */ 2671 if (prefetch) 2672 dmu_prefetch(os, objnum, 0, 0); 2673 2674 skip_entry: 2675 /* 2676 * Move to the next entry, fill in the previous offset. 2677 */ 2678 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2679 zap_cursor_advance(&zc); 2680 offset = zap_cursor_serialize(&zc); 2681 } else { 2682 offset += 1; 2683 } 2684 2685 if (cooks != NULL) { 2686 *cooks++ = offset; 2687 ncooks--; 2688 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 2689 } 2690 } 2691 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2692 2693 /* Subtract unused cookies */ 2694 if (ncookies != NULL) 2695 *ncookies -= ncooks; 2696 2697 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2698 iovp->iov_base += outcount; 2699 iovp->iov_len -= outcount; 2700 uio->uio_resid -= outcount; 2701 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2702 /* 2703 * Reset the pointer. 2704 */ 2705 offset = uio->uio_loffset; 2706 } 2707 2708update: 2709 zap_cursor_fini(&zc); 2710 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2711 kmem_free(outbuf, bufsize); 2712 2713 if (error == ENOENT) 2714 error = 0; 2715 2716 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2717 2718 uio->uio_loffset = offset; 2719 ZFS_EXIT(zfsvfs); 2720 if (error != 0 && cookies != NULL) { 2721 free(*cookies, M_TEMP); 2722 *cookies = NULL; 2723 *ncookies = 0; 2724 } 2725 return (error); 2726} 2727 2728ulong_t zfs_fsync_sync_cnt = 4; 2729 2730static int 2731zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2732{ 2733 znode_t *zp = VTOZ(vp); 2734 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2735 2736 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2737 2738 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { 2739 ZFS_ENTER(zfsvfs); 2740 ZFS_VERIFY_ZP(zp); 2741 zil_commit(zfsvfs->z_log, zp->z_id); 2742 ZFS_EXIT(zfsvfs); 2743 } 2744 return (0); 2745} 2746 2747 2748/* 2749 * Get the requested file attributes and place them in the provided 2750 * vattr structure. 2751 * 2752 * IN: vp - vnode of file. 2753 * vap - va_mask identifies requested attributes. 2754 * If AT_XVATTR set, then optional attrs are requested 2755 * flags - ATTR_NOACLCHECK (CIFS server context) 2756 * cr - credentials of caller. 2757 * ct - caller context 2758 * 2759 * OUT: vap - attribute values. 2760 * 2761 * RETURN: 0 (always succeeds). 2762 */ 2763/* ARGSUSED */ 2764static int 2765zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2766 caller_context_t *ct) 2767{ 2768 znode_t *zp = VTOZ(vp); 2769 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2770 int error = 0; 2771 uint32_t blksize; 2772 u_longlong_t nblocks; 2773 uint64_t links; 2774 uint64_t mtime[2], ctime[2], crtime[2], rdev; 2775 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2776 xoptattr_t *xoap = NULL; 2777 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2778 sa_bulk_attr_t bulk[4]; 2779 int count = 0; 2780 2781 ZFS_ENTER(zfsvfs); 2782 ZFS_VERIFY_ZP(zp); 2783 2784 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2785 2786 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 2787 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 2788 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16); 2789 if (vp->v_type == VBLK || vp->v_type == VCHR) 2790 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, 2791 &rdev, 8); 2792 2793 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 2794 ZFS_EXIT(zfsvfs); 2795 return (error); 2796 } 2797 2798 /* 2799 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2800 * Also, if we are the owner don't bother, since owner should 2801 * always be allowed to read basic attributes of file. 2802 */ 2803 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 2804 (vap->va_uid != crgetuid(cr))) { 2805 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2806 skipaclchk, cr)) { 2807 ZFS_EXIT(zfsvfs); 2808 return (error); 2809 } 2810 } 2811 2812 /* 2813 * Return all attributes. It's cheaper to provide the answer 2814 * than to determine whether we were asked the question. 2815 */ 2816 2817 mutex_enter(&zp->z_lock); 2818 vap->va_type = IFTOVT(zp->z_mode); 2819 vap->va_mode = zp->z_mode & ~S_IFMT; 2820#ifdef sun 2821 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2822#else 2823 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 2824#endif 2825 vap->va_nodeid = zp->z_id; 2826 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2827 links = zp->z_links + 1; 2828 else 2829 links = zp->z_links; 2830 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */ 2831 vap->va_size = zp->z_size; 2832#ifdef sun 2833 vap->va_rdev = vp->v_rdev; 2834#else 2835 if (vp->v_type == VBLK || vp->v_type == VCHR) 2836 vap->va_rdev = zfs_cmpldev(rdev); 2837#endif 2838 vap->va_seq = zp->z_seq; 2839 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2840 2841 /* 2842 * Add in any requested optional attributes and the create time. 2843 * Also set the corresponding bits in the returned attribute bitmap. 2844 */ 2845 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2846 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2847 xoap->xoa_archive = 2848 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2849 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2850 } 2851 2852 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2853 xoap->xoa_readonly = 2854 ((zp->z_pflags & ZFS_READONLY) != 0); 2855 XVA_SET_RTN(xvap, XAT_READONLY); 2856 } 2857 2858 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2859 xoap->xoa_system = 2860 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2861 XVA_SET_RTN(xvap, XAT_SYSTEM); 2862 } 2863 2864 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2865 xoap->xoa_hidden = 2866 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2867 XVA_SET_RTN(xvap, XAT_HIDDEN); 2868 } 2869 2870 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2871 xoap->xoa_nounlink = 2872 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2873 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2874 } 2875 2876 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2877 xoap->xoa_immutable = 2878 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2879 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2880 } 2881 2882 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2883 xoap->xoa_appendonly = 2884 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2885 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2886 } 2887 2888 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2889 xoap->xoa_nodump = 2890 ((zp->z_pflags & ZFS_NODUMP) != 0); 2891 XVA_SET_RTN(xvap, XAT_NODUMP); 2892 } 2893 2894 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2895 xoap->xoa_opaque = 2896 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2897 XVA_SET_RTN(xvap, XAT_OPAQUE); 2898 } 2899 2900 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2901 xoap->xoa_av_quarantined = 2902 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2903 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2904 } 2905 2906 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2907 xoap->xoa_av_modified = 2908 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2909 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2910 } 2911 2912 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2913 vp->v_type == VREG) { 2914 zfs_sa_get_scanstamp(zp, xvap); 2915 } 2916 2917 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2918 uint64_t times[2]; 2919 2920 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 2921 times, sizeof (times)); 2922 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 2923 XVA_SET_RTN(xvap, XAT_CREATETIME); 2924 } 2925 2926 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2927 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2928 XVA_SET_RTN(xvap, XAT_REPARSE); 2929 } 2930 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2931 xoap->xoa_generation = zp->z_gen; 2932 XVA_SET_RTN(xvap, XAT_GEN); 2933 } 2934 2935 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2936 xoap->xoa_offline = 2937 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2938 XVA_SET_RTN(xvap, XAT_OFFLINE); 2939 } 2940 2941 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2942 xoap->xoa_sparse = 2943 ((zp->z_pflags & ZFS_SPARSE) != 0); 2944 XVA_SET_RTN(xvap, XAT_SPARSE); 2945 } 2946 } 2947 2948 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2949 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2950 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2951 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 2952 2953 mutex_exit(&zp->z_lock); 2954 2955 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 2956 vap->va_blksize = blksize; 2957 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2958 2959 if (zp->z_blksz == 0) { 2960 /* 2961 * Block size hasn't been set; suggest maximal I/O transfers. 2962 */ 2963 vap->va_blksize = zfsvfs->z_max_blksz; 2964 } 2965 2966 ZFS_EXIT(zfsvfs); 2967 return (0); 2968} 2969 2970/* 2971 * Set the file attributes to the values contained in the 2972 * vattr structure. 2973 * 2974 * IN: vp - vnode of file to be modified. 2975 * vap - new attribute values. 2976 * If AT_XVATTR set, then optional attrs are being set 2977 * flags - ATTR_UTIME set if non-default time values provided. 2978 * - ATTR_NOACLCHECK (CIFS context only). 2979 * cr - credentials of caller. 2980 * ct - caller context 2981 * 2982 * RETURN: 0 on success, error code on failure. 2983 * 2984 * Timestamps: 2985 * vp - ctime updated, mtime updated if size changed. 2986 */ 2987/* ARGSUSED */ 2988static int 2989zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2990 caller_context_t *ct) 2991{ 2992 znode_t *zp = VTOZ(vp); 2993 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2994 zilog_t *zilog; 2995 dmu_tx_t *tx; 2996 vattr_t oldva; 2997 xvattr_t tmpxvattr; 2998 uint_t mask = vap->va_mask; 2999 uint_t saved_mask = 0; 3000 uint64_t saved_mode; 3001 int trim_mask = 0; 3002 uint64_t new_mode; 3003 uint64_t new_uid, new_gid; 3004 uint64_t xattr_obj; 3005 uint64_t mtime[2], ctime[2]; 3006 znode_t *attrzp; 3007 int need_policy = FALSE; 3008 int err, err2; 3009 zfs_fuid_info_t *fuidp = NULL; 3010 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 3011 xoptattr_t *xoap; 3012 zfs_acl_t *aclp; 3013 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 3014 boolean_t fuid_dirtied = B_FALSE; 3015 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 3016 int count = 0, xattr_count = 0; 3017 3018 if (mask == 0) 3019 return (0); 3020 3021 if (mask & AT_NOSET) 3022 return (SET_ERROR(EINVAL)); 3023 3024 ZFS_ENTER(zfsvfs); 3025 ZFS_VERIFY_ZP(zp); 3026 3027 zilog = zfsvfs->z_log; 3028 3029 /* 3030 * Make sure that if we have ephemeral uid/gid or xvattr specified 3031 * that file system is at proper version level 3032 */ 3033 3034 if (zfsvfs->z_use_fuids == B_FALSE && 3035 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 3036 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 3037 (mask & AT_XVATTR))) { 3038 ZFS_EXIT(zfsvfs); 3039 return (SET_ERROR(EINVAL)); 3040 } 3041 3042 if (mask & AT_SIZE && vp->v_type == VDIR) { 3043 ZFS_EXIT(zfsvfs); 3044 return (SET_ERROR(EISDIR)); 3045 } 3046 3047 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 3048 ZFS_EXIT(zfsvfs); 3049 return (SET_ERROR(EINVAL)); 3050 } 3051 3052 /* 3053 * If this is an xvattr_t, then get a pointer to the structure of 3054 * optional attributes. If this is NULL, then we have a vattr_t. 3055 */ 3056 xoap = xva_getxoptattr(xvap); 3057 3058 xva_init(&tmpxvattr); 3059 3060 /* 3061 * Immutable files can only alter immutable bit and atime 3062 */ 3063 if ((zp->z_pflags & ZFS_IMMUTABLE) && 3064 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 3065 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 3066 ZFS_EXIT(zfsvfs); 3067 return (SET_ERROR(EPERM)); 3068 } 3069 3070 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 3071 ZFS_EXIT(zfsvfs); 3072 return (SET_ERROR(EPERM)); 3073 } 3074 3075 /* 3076 * Verify timestamps doesn't overflow 32 bits. 3077 * ZFS can handle large timestamps, but 32bit syscalls can't 3078 * handle times greater than 2039. This check should be removed 3079 * once large timestamps are fully supported. 3080 */ 3081 if (mask & (AT_ATIME | AT_MTIME)) { 3082 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 3083 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 3084 ZFS_EXIT(zfsvfs); 3085 return (SET_ERROR(EOVERFLOW)); 3086 } 3087 } 3088 3089top: 3090 attrzp = NULL; 3091 aclp = NULL; 3092 3093 /* Can this be moved to before the top label? */ 3094 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 3095 ZFS_EXIT(zfsvfs); 3096 return (SET_ERROR(EROFS)); 3097 } 3098 3099 /* 3100 * First validate permissions 3101 */ 3102 3103 if (mask & AT_SIZE) { 3104 /* 3105 * XXX - Note, we are not providing any open 3106 * mode flags here (like FNDELAY), so we may 3107 * block if there are locks present... this 3108 * should be addressed in openat(). 3109 */ 3110 /* XXX - would it be OK to generate a log record here? */ 3111 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 3112 if (err) { 3113 ZFS_EXIT(zfsvfs); 3114 return (err); 3115 } 3116 } 3117 3118 if (mask & (AT_ATIME|AT_MTIME) || 3119 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 3120 XVA_ISSET_REQ(xvap, XAT_READONLY) || 3121 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 3122 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 3123 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 3124 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 3125 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 3126 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 3127 skipaclchk, cr); 3128 } 3129 3130 if (mask & (AT_UID|AT_GID)) { 3131 int idmask = (mask & (AT_UID|AT_GID)); 3132 int take_owner; 3133 int take_group; 3134 3135 /* 3136 * NOTE: even if a new mode is being set, 3137 * we may clear S_ISUID/S_ISGID bits. 3138 */ 3139 3140 if (!(mask & AT_MODE)) 3141 vap->va_mode = zp->z_mode; 3142 3143 /* 3144 * Take ownership or chgrp to group we are a member of 3145 */ 3146 3147 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 3148 take_group = (mask & AT_GID) && 3149 zfs_groupmember(zfsvfs, vap->va_gid, cr); 3150 3151 /* 3152 * If both AT_UID and AT_GID are set then take_owner and 3153 * take_group must both be set in order to allow taking 3154 * ownership. 3155 * 3156 * Otherwise, send the check through secpolicy_vnode_setattr() 3157 * 3158 */ 3159 3160 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 3161 ((idmask == AT_UID) && take_owner) || 3162 ((idmask == AT_GID) && take_group)) { 3163 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 3164 skipaclchk, cr) == 0) { 3165 /* 3166 * Remove setuid/setgid for non-privileged users 3167 */ 3168 secpolicy_setid_clear(vap, vp, cr); 3169 trim_mask = (mask & (AT_UID|AT_GID)); 3170 } else { 3171 need_policy = TRUE; 3172 } 3173 } else { 3174 need_policy = TRUE; 3175 } 3176 } 3177 3178 mutex_enter(&zp->z_lock); 3179 oldva.va_mode = zp->z_mode; 3180 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 3181 if (mask & AT_XVATTR) { 3182 /* 3183 * Update xvattr mask to include only those attributes 3184 * that are actually changing. 3185 * 3186 * the bits will be restored prior to actually setting 3187 * the attributes so the caller thinks they were set. 3188 */ 3189 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 3190 if (xoap->xoa_appendonly != 3191 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 3192 need_policy = TRUE; 3193 } else { 3194 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 3195 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 3196 } 3197 } 3198 3199 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 3200 if (xoap->xoa_nounlink != 3201 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 3202 need_policy = TRUE; 3203 } else { 3204 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 3205 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 3206 } 3207 } 3208 3209 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 3210 if (xoap->xoa_immutable != 3211 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 3212 need_policy = TRUE; 3213 } else { 3214 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 3215 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 3216 } 3217 } 3218 3219 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 3220 if (xoap->xoa_nodump != 3221 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 3222 need_policy = TRUE; 3223 } else { 3224 XVA_CLR_REQ(xvap, XAT_NODUMP); 3225 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 3226 } 3227 } 3228 3229 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 3230 if (xoap->xoa_av_modified != 3231 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 3232 need_policy = TRUE; 3233 } else { 3234 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 3235 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 3236 } 3237 } 3238 3239 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 3240 if ((vp->v_type != VREG && 3241 xoap->xoa_av_quarantined) || 3242 xoap->xoa_av_quarantined != 3243 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 3244 need_policy = TRUE; 3245 } else { 3246 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 3247 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 3248 } 3249 } 3250 3251 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 3252 mutex_exit(&zp->z_lock); 3253 ZFS_EXIT(zfsvfs); 3254 return (SET_ERROR(EPERM)); 3255 } 3256 3257 if (need_policy == FALSE && 3258 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 3259 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 3260 need_policy = TRUE; 3261 } 3262 } 3263 3264 mutex_exit(&zp->z_lock); 3265 3266 if (mask & AT_MODE) { 3267 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 3268 err = secpolicy_setid_setsticky_clear(vp, vap, 3269 &oldva, cr); 3270 if (err) { 3271 ZFS_EXIT(zfsvfs); 3272 return (err); 3273 } 3274 trim_mask |= AT_MODE; 3275 } else { 3276 need_policy = TRUE; 3277 } 3278 } 3279 3280 if (need_policy) { 3281 /* 3282 * If trim_mask is set then take ownership 3283 * has been granted or write_acl is present and user 3284 * has the ability to modify mode. In that case remove 3285 * UID|GID and or MODE from mask so that 3286 * secpolicy_vnode_setattr() doesn't revoke it. 3287 */ 3288 3289 if (trim_mask) { 3290 saved_mask = vap->va_mask; 3291 vap->va_mask &= ~trim_mask; 3292 if (trim_mask & AT_MODE) { 3293 /* 3294 * Save the mode, as secpolicy_vnode_setattr() 3295 * will overwrite it with ova.va_mode. 3296 */ 3297 saved_mode = vap->va_mode; 3298 } 3299 } 3300 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 3301 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 3302 if (err) { 3303 ZFS_EXIT(zfsvfs); 3304 return (err); 3305 } 3306 3307 if (trim_mask) { 3308 vap->va_mask |= saved_mask; 3309 if (trim_mask & AT_MODE) { 3310 /* 3311 * Recover the mode after 3312 * secpolicy_vnode_setattr(). 3313 */ 3314 vap->va_mode = saved_mode; 3315 } 3316 } 3317 } 3318 3319 /* 3320 * secpolicy_vnode_setattr, or take ownership may have 3321 * changed va_mask 3322 */ 3323 mask = vap->va_mask; 3324 3325 if ((mask & (AT_UID | AT_GID))) { 3326 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 3327 &xattr_obj, sizeof (xattr_obj)); 3328 3329 if (err == 0 && xattr_obj) { 3330 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 3331 if (err) 3332 goto out2; 3333 } 3334 if (mask & AT_UID) { 3335 new_uid = zfs_fuid_create(zfsvfs, 3336 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 3337 if (new_uid != zp->z_uid && 3338 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 3339 if (attrzp) 3340 VN_RELE(ZTOV(attrzp)); 3341 err = SET_ERROR(EDQUOT); 3342 goto out2; 3343 } 3344 } 3345 3346 if (mask & AT_GID) { 3347 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 3348 cr, ZFS_GROUP, &fuidp); 3349 if (new_gid != zp->z_gid && 3350 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 3351 if (attrzp) 3352 VN_RELE(ZTOV(attrzp)); 3353 err = SET_ERROR(EDQUOT); 3354 goto out2; 3355 } 3356 } 3357 } 3358 tx = dmu_tx_create(zfsvfs->z_os); 3359 3360 if (mask & AT_MODE) { 3361 uint64_t pmode = zp->z_mode; 3362 uint64_t acl_obj; 3363 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 3364 3365 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && 3366 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 3367 err = SET_ERROR(EPERM); 3368 goto out; 3369 } 3370 3371 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 3372 goto out; 3373 3374 mutex_enter(&zp->z_lock); 3375 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 3376 /* 3377 * Are we upgrading ACL from old V0 format 3378 * to V1 format? 3379 */ 3380 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 3381 zfs_znode_acl_version(zp) == 3382 ZFS_ACL_VERSION_INITIAL) { 3383 dmu_tx_hold_free(tx, acl_obj, 0, 3384 DMU_OBJECT_END); 3385 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3386 0, aclp->z_acl_bytes); 3387 } else { 3388 dmu_tx_hold_write(tx, acl_obj, 0, 3389 aclp->z_acl_bytes); 3390 } 3391 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3392 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3393 0, aclp->z_acl_bytes); 3394 } 3395 mutex_exit(&zp->z_lock); 3396 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3397 } else { 3398 if ((mask & AT_XVATTR) && 3399 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3400 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3401 else 3402 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3403 } 3404 3405 if (attrzp) { 3406 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 3407 } 3408 3409 fuid_dirtied = zfsvfs->z_fuid_dirty; 3410 if (fuid_dirtied) 3411 zfs_fuid_txhold(zfsvfs, tx); 3412 3413 zfs_sa_upgrade_txholds(tx, zp); 3414 3415 err = dmu_tx_assign(tx, TXG_NOWAIT); 3416 if (err) { 3417 if (err == ERESTART) 3418 dmu_tx_wait(tx); 3419 goto out; 3420 } 3421 3422 count = 0; 3423 /* 3424 * Set each attribute requested. 3425 * We group settings according to the locks they need to acquire. 3426 * 3427 * Note: you cannot set ctime directly, although it will be 3428 * updated as a side-effect of calling this function. 3429 */ 3430 3431 3432 if (mask & (AT_UID|AT_GID|AT_MODE)) 3433 mutex_enter(&zp->z_acl_lock); 3434 mutex_enter(&zp->z_lock); 3435 3436 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3437 &zp->z_pflags, sizeof (zp->z_pflags)); 3438 3439 if (attrzp) { 3440 if (mask & (AT_UID|AT_GID|AT_MODE)) 3441 mutex_enter(&attrzp->z_acl_lock); 3442 mutex_enter(&attrzp->z_lock); 3443 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3444 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3445 sizeof (attrzp->z_pflags)); 3446 } 3447 3448 if (mask & (AT_UID|AT_GID)) { 3449 3450 if (mask & AT_UID) { 3451 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3452 &new_uid, sizeof (new_uid)); 3453 zp->z_uid = new_uid; 3454 if (attrzp) { 3455 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3456 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3457 sizeof (new_uid)); 3458 attrzp->z_uid = new_uid; 3459 } 3460 } 3461 3462 if (mask & AT_GID) { 3463 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3464 NULL, &new_gid, sizeof (new_gid)); 3465 zp->z_gid = new_gid; 3466 if (attrzp) { 3467 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3468 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 3469 sizeof (new_gid)); 3470 attrzp->z_gid = new_gid; 3471 } 3472 } 3473 if (!(mask & AT_MODE)) { 3474 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 3475 NULL, &new_mode, sizeof (new_mode)); 3476 new_mode = zp->z_mode; 3477 } 3478 err = zfs_acl_chown_setattr(zp); 3479 ASSERT(err == 0); 3480 if (attrzp) { 3481 err = zfs_acl_chown_setattr(attrzp); 3482 ASSERT(err == 0); 3483 } 3484 } 3485 3486 if (mask & AT_MODE) { 3487 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 3488 &new_mode, sizeof (new_mode)); 3489 zp->z_mode = new_mode; 3490 ASSERT3U((uintptr_t)aclp, !=, 0); 3491 err = zfs_aclset_common(zp, aclp, cr, tx); 3492 ASSERT0(err); 3493 if (zp->z_acl_cached) 3494 zfs_acl_free(zp->z_acl_cached); 3495 zp->z_acl_cached = aclp; 3496 aclp = NULL; 3497 } 3498 3499 3500 if (mask & AT_ATIME) { 3501 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 3502 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 3503 &zp->z_atime, sizeof (zp->z_atime)); 3504 } 3505 3506 if (mask & AT_MTIME) { 3507 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 3508 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 3509 mtime, sizeof (mtime)); 3510 } 3511 3512 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 3513 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 3514 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 3515 NULL, mtime, sizeof (mtime)); 3516 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3517 &ctime, sizeof (ctime)); 3518 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 3519 B_TRUE); 3520 } else if (mask != 0) { 3521 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3522 &ctime, sizeof (ctime)); 3523 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime, 3524 B_TRUE); 3525 if (attrzp) { 3526 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3527 SA_ZPL_CTIME(zfsvfs), NULL, 3528 &ctime, sizeof (ctime)); 3529 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 3530 mtime, ctime, B_TRUE); 3531 } 3532 } 3533 /* 3534 * Do this after setting timestamps to prevent timestamp 3535 * update from toggling bit 3536 */ 3537 3538 if (xoap && (mask & AT_XVATTR)) { 3539 3540 /* 3541 * restore trimmed off masks 3542 * so that return masks can be set for caller. 3543 */ 3544 3545 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 3546 XVA_SET_REQ(xvap, XAT_APPENDONLY); 3547 } 3548 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 3549 XVA_SET_REQ(xvap, XAT_NOUNLINK); 3550 } 3551 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 3552 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 3553 } 3554 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 3555 XVA_SET_REQ(xvap, XAT_NODUMP); 3556 } 3557 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 3558 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 3559 } 3560 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 3561 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 3562 } 3563 3564 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3565 ASSERT(vp->v_type == VREG); 3566 3567 zfs_xvattr_set(zp, xvap, tx); 3568 } 3569 3570 if (fuid_dirtied) 3571 zfs_fuid_sync(zfsvfs, tx); 3572 3573 if (mask != 0) 3574 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 3575 3576 mutex_exit(&zp->z_lock); 3577 if (mask & (AT_UID|AT_GID|AT_MODE)) 3578 mutex_exit(&zp->z_acl_lock); 3579 3580 if (attrzp) { 3581 if (mask & (AT_UID|AT_GID|AT_MODE)) 3582 mutex_exit(&attrzp->z_acl_lock); 3583 mutex_exit(&attrzp->z_lock); 3584 } 3585out: 3586 if (err == 0 && attrzp) { 3587 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 3588 xattr_count, tx); 3589 ASSERT(err2 == 0); 3590 } 3591 3592 if (attrzp) 3593 VN_RELE(ZTOV(attrzp)); 3594 3595 if (aclp) 3596 zfs_acl_free(aclp); 3597 3598 if (fuidp) { 3599 zfs_fuid_info_free(fuidp); 3600 fuidp = NULL; 3601 } 3602 3603 if (err) { 3604 dmu_tx_abort(tx); 3605 if (err == ERESTART) 3606 goto top; 3607 } else { 3608 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3609 dmu_tx_commit(tx); 3610 } 3611 3612out2: 3613 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3614 zil_commit(zilog, 0); 3615 3616 ZFS_EXIT(zfsvfs); 3617 return (err); 3618} 3619 3620typedef struct zfs_zlock { 3621 krwlock_t *zl_rwlock; /* lock we acquired */ 3622 znode_t *zl_znode; /* znode we held */ 3623 struct zfs_zlock *zl_next; /* next in list */ 3624} zfs_zlock_t; 3625 3626/* 3627 * Drop locks and release vnodes that were held by zfs_rename_lock(). 3628 */ 3629static void 3630zfs_rename_unlock(zfs_zlock_t **zlpp) 3631{ 3632 zfs_zlock_t *zl; 3633 3634 while ((zl = *zlpp) != NULL) { 3635 if (zl->zl_znode != NULL) 3636 VN_RELE(ZTOV(zl->zl_znode)); 3637 rw_exit(zl->zl_rwlock); 3638 *zlpp = zl->zl_next; 3639 kmem_free(zl, sizeof (*zl)); 3640 } 3641} 3642 3643/* 3644 * Search back through the directory tree, using the ".." entries. 3645 * Lock each directory in the chain to prevent concurrent renames. 3646 * Fail any attempt to move a directory into one of its own descendants. 3647 * XXX - z_parent_lock can overlap with map or grow locks 3648 */ 3649static int 3650zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 3651{ 3652 zfs_zlock_t *zl; 3653 znode_t *zp = tdzp; 3654 uint64_t rootid = zp->z_zfsvfs->z_root; 3655 uint64_t oidp = zp->z_id; 3656 krwlock_t *rwlp = &szp->z_parent_lock; 3657 krw_t rw = RW_WRITER; 3658 3659 /* 3660 * First pass write-locks szp and compares to zp->z_id. 3661 * Later passes read-lock zp and compare to zp->z_parent. 3662 */ 3663 do { 3664 if (!rw_tryenter(rwlp, rw)) { 3665 /* 3666 * Another thread is renaming in this path. 3667 * Note that if we are a WRITER, we don't have any 3668 * parent_locks held yet. 3669 */ 3670 if (rw == RW_READER && zp->z_id > szp->z_id) { 3671 /* 3672 * Drop our locks and restart 3673 */ 3674 zfs_rename_unlock(&zl); 3675 *zlpp = NULL; 3676 zp = tdzp; 3677 oidp = zp->z_id; 3678 rwlp = &szp->z_parent_lock; 3679 rw = RW_WRITER; 3680 continue; 3681 } else { 3682 /* 3683 * Wait for other thread to drop its locks 3684 */ 3685 rw_enter(rwlp, rw); 3686 } 3687 } 3688 3689 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 3690 zl->zl_rwlock = rwlp; 3691 zl->zl_znode = NULL; 3692 zl->zl_next = *zlpp; 3693 *zlpp = zl; 3694 3695 if (oidp == szp->z_id) /* We're a descendant of szp */ 3696 return (SET_ERROR(EINVAL)); 3697 3698 if (oidp == rootid) /* We've hit the top */ 3699 return (0); 3700 3701 if (rw == RW_READER) { /* i.e. not the first pass */ 3702 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp); 3703 if (error) 3704 return (error); 3705 zl->zl_znode = zp; 3706 } 3707 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs), 3708 &oidp, sizeof (oidp)); 3709 rwlp = &zp->z_parent_lock; 3710 rw = RW_READER; 3711 3712 } while (zp->z_id != sdzp->z_id); 3713 3714 return (0); 3715} 3716 3717/* 3718 * Move an entry from the provided source directory to the target 3719 * directory. Change the entry name as indicated. 3720 * 3721 * IN: sdvp - Source directory containing the "old entry". 3722 * snm - Old entry name. 3723 * tdvp - Target directory to contain the "new entry". 3724 * tnm - New entry name. 3725 * cr - credentials of caller. 3726 * ct - caller context 3727 * flags - case flags 3728 * 3729 * RETURN: 0 on success, error code on failure. 3730 * 3731 * Timestamps: 3732 * sdvp,tdvp - ctime|mtime updated 3733 */ 3734/*ARGSUSED*/ 3735static int 3736zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 3737 caller_context_t *ct, int flags) 3738{ 3739 znode_t *tdzp, *szp, *tzp; 3740 znode_t *sdzp = VTOZ(sdvp); 3741 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 3742 zilog_t *zilog; 3743 vnode_t *realvp; 3744 zfs_dirlock_t *sdl, *tdl; 3745 dmu_tx_t *tx; 3746 zfs_zlock_t *zl; 3747 int cmp, serr, terr; 3748 int error = 0; 3749 int zflg = 0; 3750 boolean_t waited = B_FALSE; 3751 3752 ZFS_ENTER(zfsvfs); 3753 ZFS_VERIFY_ZP(sdzp); 3754 zilog = zfsvfs->z_log; 3755 3756 /* 3757 * Make sure we have the real vp for the target directory. 3758 */ 3759 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 3760 tdvp = realvp; 3761 3762 tdzp = VTOZ(tdvp); 3763 ZFS_VERIFY_ZP(tdzp); 3764 3765 /* 3766 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 3767 * ctldir appear to have the same v_vfsp. 3768 */ 3769 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) { 3770 ZFS_EXIT(zfsvfs); 3771 return (SET_ERROR(EXDEV)); 3772 } 3773 3774 if (zfsvfs->z_utf8 && u8_validate(tnm, 3775 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3776 ZFS_EXIT(zfsvfs); 3777 return (SET_ERROR(EILSEQ)); 3778 } 3779 3780 if (flags & FIGNORECASE) 3781 zflg |= ZCILOOK; 3782 3783top: 3784 szp = NULL; 3785 tzp = NULL; 3786 zl = NULL; 3787 3788 /* 3789 * This is to prevent the creation of links into attribute space 3790 * by renaming a linked file into/outof an attribute directory. 3791 * See the comment in zfs_link() for why this is considered bad. 3792 */ 3793 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3794 ZFS_EXIT(zfsvfs); 3795 return (SET_ERROR(EINVAL)); 3796 } 3797 3798 /* 3799 * Lock source and target directory entries. To prevent deadlock, 3800 * a lock ordering must be defined. We lock the directory with 3801 * the smallest object id first, or if it's a tie, the one with 3802 * the lexically first name. 3803 */ 3804 if (sdzp->z_id < tdzp->z_id) { 3805 cmp = -1; 3806 } else if (sdzp->z_id > tdzp->z_id) { 3807 cmp = 1; 3808 } else { 3809 /* 3810 * First compare the two name arguments without 3811 * considering any case folding. 3812 */ 3813 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 3814 3815 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 3816 ASSERT(error == 0 || !zfsvfs->z_utf8); 3817 if (cmp == 0) { 3818 /* 3819 * POSIX: "If the old argument and the new argument 3820 * both refer to links to the same existing file, 3821 * the rename() function shall return successfully 3822 * and perform no other action." 3823 */ 3824 ZFS_EXIT(zfsvfs); 3825 return (0); 3826 } 3827 /* 3828 * If the file system is case-folding, then we may 3829 * have some more checking to do. A case-folding file 3830 * system is either supporting mixed case sensitivity 3831 * access or is completely case-insensitive. Note 3832 * that the file system is always case preserving. 3833 * 3834 * In mixed sensitivity mode case sensitive behavior 3835 * is the default. FIGNORECASE must be used to 3836 * explicitly request case insensitive behavior. 3837 * 3838 * If the source and target names provided differ only 3839 * by case (e.g., a request to rename 'tim' to 'Tim'), 3840 * we will treat this as a special case in the 3841 * case-insensitive mode: as long as the source name 3842 * is an exact match, we will allow this to proceed as 3843 * a name-change request. 3844 */ 3845 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3846 (zfsvfs->z_case == ZFS_CASE_MIXED && 3847 flags & FIGNORECASE)) && 3848 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3849 &error) == 0) { 3850 /* 3851 * case preserving rename request, require exact 3852 * name matches 3853 */ 3854 zflg |= ZCIEXACT; 3855 zflg &= ~ZCILOOK; 3856 } 3857 } 3858 3859 /* 3860 * If the source and destination directories are the same, we should 3861 * grab the z_name_lock of that directory only once. 3862 */ 3863 if (sdzp == tdzp) { 3864 zflg |= ZHAVELOCK; 3865 rw_enter(&sdzp->z_name_lock, RW_READER); 3866 } 3867 3868 if (cmp < 0) { 3869 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3870 ZEXISTS | zflg, NULL, NULL); 3871 terr = zfs_dirent_lock(&tdl, 3872 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3873 } else { 3874 terr = zfs_dirent_lock(&tdl, 3875 tdzp, tnm, &tzp, zflg, NULL, NULL); 3876 serr = zfs_dirent_lock(&sdl, 3877 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3878 NULL, NULL); 3879 } 3880 3881 if (serr) { 3882 /* 3883 * Source entry invalid or not there. 3884 */ 3885 if (!terr) { 3886 zfs_dirent_unlock(tdl); 3887 if (tzp) 3888 VN_RELE(ZTOV(tzp)); 3889 } 3890 3891 if (sdzp == tdzp) 3892 rw_exit(&sdzp->z_name_lock); 3893 3894 /* 3895 * FreeBSD: In OpenSolaris they only check if rename source is 3896 * ".." here, because "." is handled in their lookup. This is 3897 * not the case for FreeBSD, so we check for "." explicitly. 3898 */ 3899 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 3900 serr = SET_ERROR(EINVAL); 3901 ZFS_EXIT(zfsvfs); 3902 return (serr); 3903 } 3904 if (terr) { 3905 zfs_dirent_unlock(sdl); 3906 VN_RELE(ZTOV(szp)); 3907 3908 if (sdzp == tdzp) 3909 rw_exit(&sdzp->z_name_lock); 3910 3911 if (strcmp(tnm, "..") == 0) 3912 terr = SET_ERROR(EINVAL); 3913 ZFS_EXIT(zfsvfs); 3914 return (terr); 3915 } 3916 3917 /* 3918 * Must have write access at the source to remove the old entry 3919 * and write access at the target to create the new entry. 3920 * Note that if target and source are the same, this can be 3921 * done in a single check. 3922 */ 3923 3924 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3925 goto out; 3926 3927 if (ZTOV(szp)->v_type == VDIR) { 3928 /* 3929 * Check to make sure rename is valid. 3930 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3931 */ 3932 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3933 goto out; 3934 } 3935 3936 /* 3937 * Does target exist? 3938 */ 3939 if (tzp) { 3940 /* 3941 * Source and target must be the same type. 3942 */ 3943 if (ZTOV(szp)->v_type == VDIR) { 3944 if (ZTOV(tzp)->v_type != VDIR) { 3945 error = SET_ERROR(ENOTDIR); 3946 goto out; 3947 } 3948 } else { 3949 if (ZTOV(tzp)->v_type == VDIR) { 3950 error = SET_ERROR(EISDIR); 3951 goto out; 3952 } 3953 } 3954 /* 3955 * POSIX dictates that when the source and target 3956 * entries refer to the same file object, rename 3957 * must do nothing and exit without error. 3958 */ 3959 if (szp->z_id == tzp->z_id) { 3960 error = 0; 3961 goto out; 3962 } 3963 } 3964 3965 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3966 if (tzp) 3967 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3968 3969 /* 3970 * notify the target directory if it is not the same 3971 * as source directory. 3972 */ 3973 if (tdvp != sdvp) { 3974 vnevent_rename_dest_dir(tdvp, ct); 3975 } 3976 3977 tx = dmu_tx_create(zfsvfs->z_os); 3978 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3979 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 3980 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3981 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3982 if (sdzp != tdzp) { 3983 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 3984 zfs_sa_upgrade_txholds(tx, tdzp); 3985 } 3986 if (tzp) { 3987 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 3988 zfs_sa_upgrade_txholds(tx, tzp); 3989 } 3990 3991 zfs_sa_upgrade_txholds(tx, szp); 3992 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3993 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 3994 if (error) { 3995 if (zl != NULL) 3996 zfs_rename_unlock(&zl); 3997 zfs_dirent_unlock(sdl); 3998 zfs_dirent_unlock(tdl); 3999 4000 if (sdzp == tdzp) 4001 rw_exit(&sdzp->z_name_lock); 4002 4003 VN_RELE(ZTOV(szp)); 4004 if (tzp) 4005 VN_RELE(ZTOV(tzp)); 4006 if (error == ERESTART) { 4007 waited = B_TRUE; 4008 dmu_tx_wait(tx); 4009 dmu_tx_abort(tx); 4010 goto top; 4011 } 4012 dmu_tx_abort(tx); 4013 ZFS_EXIT(zfsvfs); 4014 return (error); 4015 } 4016 4017 if (tzp) /* Attempt to remove the existing target */ 4018 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 4019 4020 if (error == 0) { 4021 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 4022 if (error == 0) { 4023 szp->z_pflags |= ZFS_AV_MODIFIED; 4024 4025 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 4026 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 4027 ASSERT0(error); 4028 4029 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 4030 if (error == 0) { 4031 zfs_log_rename(zilog, tx, TX_RENAME | 4032 (flags & FIGNORECASE ? TX_CI : 0), sdzp, 4033 sdl->dl_name, tdzp, tdl->dl_name, szp); 4034 4035 /* 4036 * Update path information for the target vnode 4037 */ 4038 vn_renamepath(tdvp, ZTOV(szp), tnm, 4039 strlen(tnm)); 4040 } else { 4041 /* 4042 * At this point, we have successfully created 4043 * the target name, but have failed to remove 4044 * the source name. Since the create was done 4045 * with the ZRENAMING flag, there are 4046 * complications; for one, the link count is 4047 * wrong. The easiest way to deal with this 4048 * is to remove the newly created target, and 4049 * return the original error. This must 4050 * succeed; fortunately, it is very unlikely to 4051 * fail, since we just created it. 4052 */ 4053 VERIFY3U(zfs_link_destroy(tdl, szp, tx, 4054 ZRENAMING, NULL), ==, 0); 4055 } 4056 } 4057#ifdef FREEBSD_NAMECACHE 4058 if (error == 0) { 4059 cache_purge(sdvp); 4060 cache_purge(tdvp); 4061 cache_purge(ZTOV(szp)); 4062 if (tzp) 4063 cache_purge(ZTOV(tzp)); 4064 } 4065#endif 4066 } 4067 4068 dmu_tx_commit(tx); 4069out: 4070 if (zl != NULL) 4071 zfs_rename_unlock(&zl); 4072 4073 zfs_dirent_unlock(sdl); 4074 zfs_dirent_unlock(tdl); 4075 4076 if (sdzp == tdzp) 4077 rw_exit(&sdzp->z_name_lock); 4078 4079 4080 VN_RELE(ZTOV(szp)); 4081 if (tzp) 4082 VN_RELE(ZTOV(tzp)); 4083 4084 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4085 zil_commit(zilog, 0); 4086 4087 ZFS_EXIT(zfsvfs); 4088 4089 return (error); 4090} 4091 4092/* 4093 * Insert the indicated symbolic reference entry into the directory. 4094 * 4095 * IN: dvp - Directory to contain new symbolic link. 4096 * link - Name for new symlink entry. 4097 * vap - Attributes of new entry. 4098 * cr - credentials of caller. 4099 * ct - caller context 4100 * flags - case flags 4101 * 4102 * RETURN: 0 on success, error code on failure. 4103 * 4104 * Timestamps: 4105 * dvp - ctime|mtime updated 4106 */ 4107/*ARGSUSED*/ 4108static int 4109zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, 4110 cred_t *cr, kthread_t *td) 4111{ 4112 znode_t *zp, *dzp = VTOZ(dvp); 4113 zfs_dirlock_t *dl; 4114 dmu_tx_t *tx; 4115 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4116 zilog_t *zilog; 4117 uint64_t len = strlen(link); 4118 int error; 4119 int zflg = ZNEW; 4120 zfs_acl_ids_t acl_ids; 4121 boolean_t fuid_dirtied; 4122 uint64_t txtype = TX_SYMLINK; 4123 boolean_t waited = B_FALSE; 4124 int flags = 0; 4125 4126 ASSERT(vap->va_type == VLNK); 4127 4128 ZFS_ENTER(zfsvfs); 4129 ZFS_VERIFY_ZP(dzp); 4130 zilog = zfsvfs->z_log; 4131 4132 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 4133 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4134 ZFS_EXIT(zfsvfs); 4135 return (SET_ERROR(EILSEQ)); 4136 } 4137 if (flags & FIGNORECASE) 4138 zflg |= ZCILOOK; 4139 4140 if (len > MAXPATHLEN) { 4141 ZFS_EXIT(zfsvfs); 4142 return (SET_ERROR(ENAMETOOLONG)); 4143 } 4144 4145 if ((error = zfs_acl_ids_create(dzp, 0, 4146 vap, cr, NULL, &acl_ids)) != 0) { 4147 ZFS_EXIT(zfsvfs); 4148 return (error); 4149 } 4150top: 4151 /* 4152 * Attempt to lock directory; fail if entry already exists. 4153 */ 4154 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 4155 if (error) { 4156 zfs_acl_ids_free(&acl_ids); 4157 ZFS_EXIT(zfsvfs); 4158 return (error); 4159 } 4160 4161 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4162 zfs_acl_ids_free(&acl_ids); 4163 zfs_dirent_unlock(dl); 4164 ZFS_EXIT(zfsvfs); 4165 return (error); 4166 } 4167 4168 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 4169 zfs_acl_ids_free(&acl_ids); 4170 zfs_dirent_unlock(dl); 4171 ZFS_EXIT(zfsvfs); 4172 return (SET_ERROR(EDQUOT)); 4173 } 4174 tx = dmu_tx_create(zfsvfs->z_os); 4175 fuid_dirtied = zfsvfs->z_fuid_dirty; 4176 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 4177 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4178 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 4179 ZFS_SA_BASE_ATTR_SIZE + len); 4180 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 4181 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 4182 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 4183 acl_ids.z_aclp->z_acl_bytes); 4184 } 4185 if (fuid_dirtied) 4186 zfs_fuid_txhold(zfsvfs, tx); 4187 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4188 if (error) { 4189 zfs_dirent_unlock(dl); 4190 if (error == ERESTART) { 4191 waited = B_TRUE; 4192 dmu_tx_wait(tx); 4193 dmu_tx_abort(tx); 4194 goto top; 4195 } 4196 zfs_acl_ids_free(&acl_ids); 4197 dmu_tx_abort(tx); 4198 ZFS_EXIT(zfsvfs); 4199 return (error); 4200 } 4201 4202 /* 4203 * Create a new object for the symlink. 4204 * for version 4 ZPL datsets the symlink will be an SA attribute 4205 */ 4206 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 4207 4208 if (fuid_dirtied) 4209 zfs_fuid_sync(zfsvfs, tx); 4210 4211 mutex_enter(&zp->z_lock); 4212 if (zp->z_is_sa) 4213 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 4214 link, len, tx); 4215 else 4216 zfs_sa_symlink(zp, link, len, tx); 4217 mutex_exit(&zp->z_lock); 4218 4219 zp->z_size = len; 4220 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 4221 &zp->z_size, sizeof (zp->z_size), tx); 4222 /* 4223 * Insert the new object into the directory. 4224 */ 4225 (void) zfs_link_create(dl, zp, tx, ZNEW); 4226 4227 if (flags & FIGNORECASE) 4228 txtype |= TX_CI; 4229 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 4230 *vpp = ZTOV(zp); 4231 4232 zfs_acl_ids_free(&acl_ids); 4233 4234 dmu_tx_commit(tx); 4235 4236 zfs_dirent_unlock(dl); 4237 4238 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4239 zil_commit(zilog, 0); 4240 4241 ZFS_EXIT(zfsvfs); 4242 return (error); 4243} 4244 4245/* 4246 * Return, in the buffer contained in the provided uio structure, 4247 * the symbolic path referred to by vp. 4248 * 4249 * IN: vp - vnode of symbolic link. 4250 * uio - structure to contain the link path. 4251 * cr - credentials of caller. 4252 * ct - caller context 4253 * 4254 * OUT: uio - structure containing the link path. 4255 * 4256 * RETURN: 0 on success, error code on failure. 4257 * 4258 * Timestamps: 4259 * vp - atime updated 4260 */ 4261/* ARGSUSED */ 4262static int 4263zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 4264{ 4265 znode_t *zp = VTOZ(vp); 4266 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4267 int error; 4268 4269 ZFS_ENTER(zfsvfs); 4270 ZFS_VERIFY_ZP(zp); 4271 4272 mutex_enter(&zp->z_lock); 4273 if (zp->z_is_sa) 4274 error = sa_lookup_uio(zp->z_sa_hdl, 4275 SA_ZPL_SYMLINK(zfsvfs), uio); 4276 else 4277 error = zfs_sa_readlink(zp, uio); 4278 mutex_exit(&zp->z_lock); 4279 4280 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4281 4282 ZFS_EXIT(zfsvfs); 4283 return (error); 4284} 4285 4286/* 4287 * Insert a new entry into directory tdvp referencing svp. 4288 * 4289 * IN: tdvp - Directory to contain new entry. 4290 * svp - vnode of new entry. 4291 * name - name of new entry. 4292 * cr - credentials of caller. 4293 * ct - caller context 4294 * 4295 * RETURN: 0 on success, error code on failure. 4296 * 4297 * Timestamps: 4298 * tdvp - ctime|mtime updated 4299 * svp - ctime updated 4300 */ 4301/* ARGSUSED */ 4302static int 4303zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 4304 caller_context_t *ct, int flags) 4305{ 4306 znode_t *dzp = VTOZ(tdvp); 4307 znode_t *tzp, *szp; 4308 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4309 zilog_t *zilog; 4310 zfs_dirlock_t *dl; 4311 dmu_tx_t *tx; 4312 vnode_t *realvp; 4313 int error; 4314 int zf = ZNEW; 4315 uint64_t parent; 4316 uid_t owner; 4317 boolean_t waited = B_FALSE; 4318 4319 ASSERT(tdvp->v_type == VDIR); 4320 4321 ZFS_ENTER(zfsvfs); 4322 ZFS_VERIFY_ZP(dzp); 4323 zilog = zfsvfs->z_log; 4324 4325 if (VOP_REALVP(svp, &realvp, ct) == 0) 4326 svp = realvp; 4327 4328 /* 4329 * POSIX dictates that we return EPERM here. 4330 * Better choices include ENOTSUP or EISDIR. 4331 */ 4332 if (svp->v_type == VDIR) { 4333 ZFS_EXIT(zfsvfs); 4334 return (SET_ERROR(EPERM)); 4335 } 4336 4337 szp = VTOZ(svp); 4338 ZFS_VERIFY_ZP(szp); 4339 4340 /* 4341 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 4342 * ctldir appear to have the same v_vfsp. 4343 */ 4344 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) { 4345 ZFS_EXIT(zfsvfs); 4346 return (SET_ERROR(EXDEV)); 4347 } 4348 4349 /* Prevent links to .zfs/shares files */ 4350 4351 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 4352 &parent, sizeof (uint64_t))) != 0) { 4353 ZFS_EXIT(zfsvfs); 4354 return (error); 4355 } 4356 if (parent == zfsvfs->z_shares_dir) { 4357 ZFS_EXIT(zfsvfs); 4358 return (SET_ERROR(EPERM)); 4359 } 4360 4361 if (zfsvfs->z_utf8 && u8_validate(name, 4362 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4363 ZFS_EXIT(zfsvfs); 4364 return (SET_ERROR(EILSEQ)); 4365 } 4366 if (flags & FIGNORECASE) 4367 zf |= ZCILOOK; 4368 4369 /* 4370 * We do not support links between attributes and non-attributes 4371 * because of the potential security risk of creating links 4372 * into "normal" file space in order to circumvent restrictions 4373 * imposed in attribute space. 4374 */ 4375 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 4376 ZFS_EXIT(zfsvfs); 4377 return (SET_ERROR(EINVAL)); 4378 } 4379 4380 4381 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 4382 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) { 4383 ZFS_EXIT(zfsvfs); 4384 return (SET_ERROR(EPERM)); 4385 } 4386 4387 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4388 ZFS_EXIT(zfsvfs); 4389 return (error); 4390 } 4391 4392top: 4393 /* 4394 * Attempt to lock directory; fail if entry already exists. 4395 */ 4396 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 4397 if (error) { 4398 ZFS_EXIT(zfsvfs); 4399 return (error); 4400 } 4401 4402 tx = dmu_tx_create(zfsvfs->z_os); 4403 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4404 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4405 zfs_sa_upgrade_txholds(tx, szp); 4406 zfs_sa_upgrade_txholds(tx, dzp); 4407 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4408 if (error) { 4409 zfs_dirent_unlock(dl); 4410 if (error == ERESTART) { 4411 waited = B_TRUE; 4412 dmu_tx_wait(tx); 4413 dmu_tx_abort(tx); 4414 goto top; 4415 } 4416 dmu_tx_abort(tx); 4417 ZFS_EXIT(zfsvfs); 4418 return (error); 4419 } 4420 4421 error = zfs_link_create(dl, szp, tx, 0); 4422 4423 if (error == 0) { 4424 uint64_t txtype = TX_LINK; 4425 if (flags & FIGNORECASE) 4426 txtype |= TX_CI; 4427 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 4428 } 4429 4430 dmu_tx_commit(tx); 4431 4432 zfs_dirent_unlock(dl); 4433 4434 if (error == 0) { 4435 vnevent_link(svp, ct); 4436 } 4437 4438 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4439 zil_commit(zilog, 0); 4440 4441 ZFS_EXIT(zfsvfs); 4442 return (error); 4443} 4444 4445#ifdef sun 4446/* 4447 * zfs_null_putapage() is used when the file system has been force 4448 * unmounted. It just drops the pages. 4449 */ 4450/* ARGSUSED */ 4451static int 4452zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4453 size_t *lenp, int flags, cred_t *cr) 4454{ 4455 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 4456 return (0); 4457} 4458 4459/* 4460 * Push a page out to disk, klustering if possible. 4461 * 4462 * IN: vp - file to push page to. 4463 * pp - page to push. 4464 * flags - additional flags. 4465 * cr - credentials of caller. 4466 * 4467 * OUT: offp - start of range pushed. 4468 * lenp - len of range pushed. 4469 * 4470 * RETURN: 0 on success, error code on failure. 4471 * 4472 * NOTE: callers must have locked the page to be pushed. On 4473 * exit, the page (and all other pages in the kluster) must be 4474 * unlocked. 4475 */ 4476/* ARGSUSED */ 4477static int 4478zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4479 size_t *lenp, int flags, cred_t *cr) 4480{ 4481 znode_t *zp = VTOZ(vp); 4482 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4483 dmu_tx_t *tx; 4484 u_offset_t off, koff; 4485 size_t len, klen; 4486 int err; 4487 4488 off = pp->p_offset; 4489 len = PAGESIZE; 4490 /* 4491 * If our blocksize is bigger than the page size, try to kluster 4492 * multiple pages so that we write a full block (thus avoiding 4493 * a read-modify-write). 4494 */ 4495 if (off < zp->z_size && zp->z_blksz > PAGESIZE) { 4496 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 4497 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 4498 ASSERT(koff <= zp->z_size); 4499 if (koff + klen > zp->z_size) 4500 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); 4501 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 4502 } 4503 ASSERT3U(btop(len), ==, btopr(len)); 4504 4505 /* 4506 * Can't push pages past end-of-file. 4507 */ 4508 if (off >= zp->z_size) { 4509 /* ignore all pages */ 4510 err = 0; 4511 goto out; 4512 } else if (off + len > zp->z_size) { 4513 int npages = btopr(zp->z_size - off); 4514 page_t *trunc; 4515 4516 page_list_break(&pp, &trunc, npages); 4517 /* ignore pages past end of file */ 4518 if (trunc) 4519 pvn_write_done(trunc, flags); 4520 len = zp->z_size - off; 4521 } 4522 4523 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4524 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4525 err = SET_ERROR(EDQUOT); 4526 goto out; 4527 } 4528top: 4529 tx = dmu_tx_create(zfsvfs->z_os); 4530 dmu_tx_hold_write(tx, zp->z_id, off, len); 4531 4532 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4533 zfs_sa_upgrade_txholds(tx, zp); 4534 err = dmu_tx_assign(tx, TXG_NOWAIT); 4535 if (err != 0) { 4536 if (err == ERESTART) { 4537 dmu_tx_wait(tx); 4538 dmu_tx_abort(tx); 4539 goto top; 4540 } 4541 dmu_tx_abort(tx); 4542 goto out; 4543 } 4544 4545 if (zp->z_blksz <= PAGESIZE) { 4546 caddr_t va = zfs_map_page(pp, S_READ); 4547 ASSERT3U(len, <=, PAGESIZE); 4548 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 4549 zfs_unmap_page(pp, va); 4550 } else { 4551 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 4552 } 4553 4554 if (err == 0) { 4555 uint64_t mtime[2], ctime[2]; 4556 sa_bulk_attr_t bulk[3]; 4557 int count = 0; 4558 4559 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4560 &mtime, 16); 4561 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4562 &ctime, 16); 4563 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4564 &zp->z_pflags, 8); 4565 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4566 B_TRUE); 4567 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4568 } 4569 dmu_tx_commit(tx); 4570 4571out: 4572 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 4573 if (offp) 4574 *offp = off; 4575 if (lenp) 4576 *lenp = len; 4577 4578 return (err); 4579} 4580 4581/* 4582 * Copy the portion of the file indicated from pages into the file. 4583 * The pages are stored in a page list attached to the files vnode. 4584 * 4585 * IN: vp - vnode of file to push page data to. 4586 * off - position in file to put data. 4587 * len - amount of data to write. 4588 * flags - flags to control the operation. 4589 * cr - credentials of caller. 4590 * ct - caller context. 4591 * 4592 * RETURN: 0 on success, error code on failure. 4593 * 4594 * Timestamps: 4595 * vp - ctime|mtime updated 4596 */ 4597/*ARGSUSED*/ 4598static int 4599zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4600 caller_context_t *ct) 4601{ 4602 znode_t *zp = VTOZ(vp); 4603 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4604 page_t *pp; 4605 size_t io_len; 4606 u_offset_t io_off; 4607 uint_t blksz; 4608 rl_t *rl; 4609 int error = 0; 4610 4611 ZFS_ENTER(zfsvfs); 4612 ZFS_VERIFY_ZP(zp); 4613 4614 /* 4615 * Align this request to the file block size in case we kluster. 4616 * XXX - this can result in pretty aggresive locking, which can 4617 * impact simultanious read/write access. One option might be 4618 * to break up long requests (len == 0) into block-by-block 4619 * operations to get narrower locking. 4620 */ 4621 blksz = zp->z_blksz; 4622 if (ISP2(blksz)) 4623 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 4624 else 4625 io_off = 0; 4626 if (len > 0 && ISP2(blksz)) 4627 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 4628 else 4629 io_len = 0; 4630 4631 if (io_len == 0) { 4632 /* 4633 * Search the entire vp list for pages >= io_off. 4634 */ 4635 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 4636 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 4637 goto out; 4638 } 4639 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 4640 4641 if (off > zp->z_size) { 4642 /* past end of file */ 4643 zfs_range_unlock(rl); 4644 ZFS_EXIT(zfsvfs); 4645 return (0); 4646 } 4647 4648 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); 4649 4650 for (off = io_off; io_off < off + len; io_off += io_len) { 4651 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 4652 pp = page_lookup(vp, io_off, 4653 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 4654 } else { 4655 pp = page_lookup_nowait(vp, io_off, 4656 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 4657 } 4658 4659 if (pp != NULL && pvn_getdirty(pp, flags)) { 4660 int err; 4661 4662 /* 4663 * Found a dirty page to push 4664 */ 4665 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 4666 if (err) 4667 error = err; 4668 } else { 4669 io_len = PAGESIZE; 4670 } 4671 } 4672out: 4673 zfs_range_unlock(rl); 4674 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4675 zil_commit(zfsvfs->z_log, zp->z_id); 4676 ZFS_EXIT(zfsvfs); 4677 return (error); 4678} 4679#endif /* sun */ 4680 4681/*ARGSUSED*/ 4682void 4683zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4684{ 4685 znode_t *zp = VTOZ(vp); 4686 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4687 int error; 4688 4689 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4690 if (zp->z_sa_hdl == NULL) { 4691 /* 4692 * The fs has been unmounted, or we did a 4693 * suspend/resume and this file no longer exists. 4694 */ 4695 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4696 vrecycle(vp); 4697 return; 4698 } 4699 4700 mutex_enter(&zp->z_lock); 4701 if (zp->z_unlinked) { 4702 /* 4703 * Fast path to recycle a vnode of a removed file. 4704 */ 4705 mutex_exit(&zp->z_lock); 4706 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4707 vrecycle(vp); 4708 return; 4709 } 4710 mutex_exit(&zp->z_lock); 4711 4712 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4713 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4714 4715 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4716 zfs_sa_upgrade_txholds(tx, zp); 4717 error = dmu_tx_assign(tx, TXG_WAIT); 4718 if (error) { 4719 dmu_tx_abort(tx); 4720 } else { 4721 mutex_enter(&zp->z_lock); 4722 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4723 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4724 zp->z_atime_dirty = 0; 4725 mutex_exit(&zp->z_lock); 4726 dmu_tx_commit(tx); 4727 } 4728 } 4729 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4730} 4731 4732#ifdef sun 4733/* 4734 * Bounds-check the seek operation. 4735 * 4736 * IN: vp - vnode seeking within 4737 * ooff - old file offset 4738 * noffp - pointer to new file offset 4739 * ct - caller context 4740 * 4741 * RETURN: 0 on success, EINVAL if new offset invalid. 4742 */ 4743/* ARGSUSED */ 4744static int 4745zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 4746 caller_context_t *ct) 4747{ 4748 if (vp->v_type == VDIR) 4749 return (0); 4750 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4751} 4752 4753/* 4754 * Pre-filter the generic locking function to trap attempts to place 4755 * a mandatory lock on a memory mapped file. 4756 */ 4757static int 4758zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 4759 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 4760{ 4761 znode_t *zp = VTOZ(vp); 4762 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4763 4764 ZFS_ENTER(zfsvfs); 4765 ZFS_VERIFY_ZP(zp); 4766 4767 /* 4768 * We are following the UFS semantics with respect to mapcnt 4769 * here: If we see that the file is mapped already, then we will 4770 * return an error, but we don't worry about races between this 4771 * function and zfs_map(). 4772 */ 4773 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { 4774 ZFS_EXIT(zfsvfs); 4775 return (SET_ERROR(EAGAIN)); 4776 } 4777 ZFS_EXIT(zfsvfs); 4778 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4779} 4780 4781/* 4782 * If we can't find a page in the cache, we will create a new page 4783 * and fill it with file data. For efficiency, we may try to fill 4784 * multiple pages at once (klustering) to fill up the supplied page 4785 * list. Note that the pages to be filled are held with an exclusive 4786 * lock to prevent access by other threads while they are being filled. 4787 */ 4788static int 4789zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 4790 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 4791{ 4792 znode_t *zp = VTOZ(vp); 4793 page_t *pp, *cur_pp; 4794 objset_t *os = zp->z_zfsvfs->z_os; 4795 u_offset_t io_off, total; 4796 size_t io_len; 4797 int err; 4798 4799 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4800 /* 4801 * We only have a single page, don't bother klustering 4802 */ 4803 io_off = off; 4804 io_len = PAGESIZE; 4805 pp = page_create_va(vp, io_off, io_len, 4806 PG_EXCL | PG_WAIT, seg, addr); 4807 } else { 4808 /* 4809 * Try to find enough pages to fill the page list 4810 */ 4811 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4812 &io_len, off, plsz, 0); 4813 } 4814 if (pp == NULL) { 4815 /* 4816 * The page already exists, nothing to do here. 4817 */ 4818 *pl = NULL; 4819 return (0); 4820 } 4821 4822 /* 4823 * Fill the pages in the kluster. 4824 */ 4825 cur_pp = pp; 4826 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4827 caddr_t va; 4828 4829 ASSERT3U(io_off, ==, cur_pp->p_offset); 4830 va = zfs_map_page(cur_pp, S_WRITE); 4831 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4832 DMU_READ_PREFETCH); 4833 zfs_unmap_page(cur_pp, va); 4834 if (err) { 4835 /* On error, toss the entire kluster */ 4836 pvn_read_done(pp, B_ERROR); 4837 /* convert checksum errors into IO errors */ 4838 if (err == ECKSUM) 4839 err = SET_ERROR(EIO); 4840 return (err); 4841 } 4842 cur_pp = cur_pp->p_next; 4843 } 4844 4845 /* 4846 * Fill in the page list array from the kluster starting 4847 * from the desired offset `off'. 4848 * NOTE: the page list will always be null terminated. 4849 */ 4850 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4851 ASSERT(pl == NULL || (*pl)->p_offset == off); 4852 4853 return (0); 4854} 4855 4856/* 4857 * Return pointers to the pages for the file region [off, off + len] 4858 * in the pl array. If plsz is greater than len, this function may 4859 * also return page pointers from after the specified region 4860 * (i.e. the region [off, off + plsz]). These additional pages are 4861 * only returned if they are already in the cache, or were created as 4862 * part of a klustered read. 4863 * 4864 * IN: vp - vnode of file to get data from. 4865 * off - position in file to get data from. 4866 * len - amount of data to retrieve. 4867 * plsz - length of provided page list. 4868 * seg - segment to obtain pages for. 4869 * addr - virtual address of fault. 4870 * rw - mode of created pages. 4871 * cr - credentials of caller. 4872 * ct - caller context. 4873 * 4874 * OUT: protp - protection mode of created pages. 4875 * pl - list of pages created. 4876 * 4877 * RETURN: 0 on success, error code on failure. 4878 * 4879 * Timestamps: 4880 * vp - atime updated 4881 */ 4882/* ARGSUSED */ 4883static int 4884zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4885 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4886 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4887{ 4888 znode_t *zp = VTOZ(vp); 4889 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4890 page_t **pl0 = pl; 4891 int err = 0; 4892 4893 /* we do our own caching, faultahead is unnecessary */ 4894 if (pl == NULL) 4895 return (0); 4896 else if (len > plsz) 4897 len = plsz; 4898 else 4899 len = P2ROUNDUP(len, PAGESIZE); 4900 ASSERT(plsz >= len); 4901 4902 ZFS_ENTER(zfsvfs); 4903 ZFS_VERIFY_ZP(zp); 4904 4905 if (protp) 4906 *protp = PROT_ALL; 4907 4908 /* 4909 * Loop through the requested range [off, off + len) looking 4910 * for pages. If we don't find a page, we will need to create 4911 * a new page and fill it with data from the file. 4912 */ 4913 while (len > 0) { 4914 if (*pl = page_lookup(vp, off, SE_SHARED)) 4915 *(pl+1) = NULL; 4916 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 4917 goto out; 4918 while (*pl) { 4919 ASSERT3U((*pl)->p_offset, ==, off); 4920 off += PAGESIZE; 4921 addr += PAGESIZE; 4922 if (len > 0) { 4923 ASSERT3U(len, >=, PAGESIZE); 4924 len -= PAGESIZE; 4925 } 4926 ASSERT3U(plsz, >=, PAGESIZE); 4927 plsz -= PAGESIZE; 4928 pl++; 4929 } 4930 } 4931 4932 /* 4933 * Fill out the page array with any pages already in the cache. 4934 */ 4935 while (plsz > 0 && 4936 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 4937 off += PAGESIZE; 4938 plsz -= PAGESIZE; 4939 } 4940out: 4941 if (err) { 4942 /* 4943 * Release any pages we have previously locked. 4944 */ 4945 while (pl > pl0) 4946 page_unlock(*--pl); 4947 } else { 4948 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4949 } 4950 4951 *pl = NULL; 4952 4953 ZFS_EXIT(zfsvfs); 4954 return (err); 4955} 4956 4957/* 4958 * Request a memory map for a section of a file. This code interacts 4959 * with common code and the VM system as follows: 4960 * 4961 * - common code calls mmap(), which ends up in smmap_common() 4962 * - this calls VOP_MAP(), which takes you into (say) zfs 4963 * - zfs_map() calls as_map(), passing segvn_create() as the callback 4964 * - segvn_create() creates the new segment and calls VOP_ADDMAP() 4965 * - zfs_addmap() updates z_mapcnt 4966 */ 4967/*ARGSUSED*/ 4968static int 4969zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4970 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4971 caller_context_t *ct) 4972{ 4973 znode_t *zp = VTOZ(vp); 4974 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4975 segvn_crargs_t vn_a; 4976 int error; 4977 4978 ZFS_ENTER(zfsvfs); 4979 ZFS_VERIFY_ZP(zp); 4980 4981 if ((prot & PROT_WRITE) && (zp->z_pflags & 4982 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 4983 ZFS_EXIT(zfsvfs); 4984 return (SET_ERROR(EPERM)); 4985 } 4986 4987 if ((prot & (PROT_READ | PROT_EXEC)) && 4988 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 4989 ZFS_EXIT(zfsvfs); 4990 return (SET_ERROR(EACCES)); 4991 } 4992 4993 if (vp->v_flag & VNOMAP) { 4994 ZFS_EXIT(zfsvfs); 4995 return (SET_ERROR(ENOSYS)); 4996 } 4997 4998 if (off < 0 || len > MAXOFFSET_T - off) { 4999 ZFS_EXIT(zfsvfs); 5000 return (SET_ERROR(ENXIO)); 5001 } 5002 5003 if (vp->v_type != VREG) { 5004 ZFS_EXIT(zfsvfs); 5005 return (SET_ERROR(ENODEV)); 5006 } 5007 5008 /* 5009 * If file is locked, disallow mapping. 5010 */ 5011 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { 5012 ZFS_EXIT(zfsvfs); 5013 return (SET_ERROR(EAGAIN)); 5014 } 5015 5016 as_rangelock(as); 5017 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 5018 if (error != 0) { 5019 as_rangeunlock(as); 5020 ZFS_EXIT(zfsvfs); 5021 return (error); 5022 } 5023 5024 vn_a.vp = vp; 5025 vn_a.offset = (u_offset_t)off; 5026 vn_a.type = flags & MAP_TYPE; 5027 vn_a.prot = prot; 5028 vn_a.maxprot = maxprot; 5029 vn_a.cred = cr; 5030 vn_a.amp = NULL; 5031 vn_a.flags = flags & ~MAP_TYPE; 5032 vn_a.szc = 0; 5033 vn_a.lgrp_mem_policy_flags = 0; 5034 5035 error = as_map(as, *addrp, len, segvn_create, &vn_a); 5036 5037 as_rangeunlock(as); 5038 ZFS_EXIT(zfsvfs); 5039 return (error); 5040} 5041 5042/* ARGSUSED */ 5043static int 5044zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5045 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 5046 caller_context_t *ct) 5047{ 5048 uint64_t pages = btopr(len); 5049 5050 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 5051 return (0); 5052} 5053 5054/* 5055 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 5056 * more accurate mtime for the associated file. Since we don't have a way of 5057 * detecting when the data was actually modified, we have to resort to 5058 * heuristics. If an explicit msync() is done, then we mark the mtime when the 5059 * last page is pushed. The problem occurs when the msync() call is omitted, 5060 * which by far the most common case: 5061 * 5062 * open() 5063 * mmap() 5064 * <modify memory> 5065 * munmap() 5066 * close() 5067 * <time lapse> 5068 * putpage() via fsflush 5069 * 5070 * If we wait until fsflush to come along, we can have a modification time that 5071 * is some arbitrary point in the future. In order to prevent this in the 5072 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 5073 * torn down. 5074 */ 5075/* ARGSUSED */ 5076static int 5077zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5078 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 5079 caller_context_t *ct) 5080{ 5081 uint64_t pages = btopr(len); 5082 5083 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 5084 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 5085 5086 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 5087 vn_has_cached_data(vp)) 5088 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 5089 5090 return (0); 5091} 5092 5093/* 5094 * Free or allocate space in a file. Currently, this function only 5095 * supports the `F_FREESP' command. However, this command is somewhat 5096 * misnamed, as its functionality includes the ability to allocate as 5097 * well as free space. 5098 * 5099 * IN: vp - vnode of file to free data in. 5100 * cmd - action to take (only F_FREESP supported). 5101 * bfp - section of file to free/alloc. 5102 * flag - current file open mode flags. 5103 * offset - current file offset. 5104 * cr - credentials of caller [UNUSED]. 5105 * ct - caller context. 5106 * 5107 * RETURN: 0 on success, error code on failure. 5108 * 5109 * Timestamps: 5110 * vp - ctime|mtime updated 5111 */ 5112/* ARGSUSED */ 5113static int 5114zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 5115 offset_t offset, cred_t *cr, caller_context_t *ct) 5116{ 5117 znode_t *zp = VTOZ(vp); 5118 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5119 uint64_t off, len; 5120 int error; 5121 5122 ZFS_ENTER(zfsvfs); 5123 ZFS_VERIFY_ZP(zp); 5124 5125 if (cmd != F_FREESP) { 5126 ZFS_EXIT(zfsvfs); 5127 return (SET_ERROR(EINVAL)); 5128 } 5129 5130 if (error = convoff(vp, bfp, 0, offset)) { 5131 ZFS_EXIT(zfsvfs); 5132 return (error); 5133 } 5134 5135 if (bfp->l_len < 0) { 5136 ZFS_EXIT(zfsvfs); 5137 return (SET_ERROR(EINVAL)); 5138 } 5139 5140 off = bfp->l_start; 5141 len = bfp->l_len; /* 0 means from off to end of file */ 5142 5143 error = zfs_freesp(zp, off, len, flag, TRUE); 5144 5145 ZFS_EXIT(zfsvfs); 5146 return (error); 5147} 5148#endif /* sun */ 5149 5150CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 5151CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 5152 5153/*ARGSUSED*/ 5154static int 5155zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 5156{ 5157 znode_t *zp = VTOZ(vp); 5158 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5159 uint32_t gen; 5160 uint64_t gen64; 5161 uint64_t object = zp->z_id; 5162 zfid_short_t *zfid; 5163 int size, i, error; 5164 5165 ZFS_ENTER(zfsvfs); 5166 ZFS_VERIFY_ZP(zp); 5167 5168 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 5169 &gen64, sizeof (uint64_t))) != 0) { 5170 ZFS_EXIT(zfsvfs); 5171 return (error); 5172 } 5173 5174 gen = (uint32_t)gen64; 5175 5176 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 5177 5178#ifdef illumos 5179 if (fidp->fid_len < size) { 5180 fidp->fid_len = size; 5181 ZFS_EXIT(zfsvfs); 5182 return (SET_ERROR(ENOSPC)); 5183 } 5184#else 5185 fidp->fid_len = size; 5186#endif 5187 5188 zfid = (zfid_short_t *)fidp; 5189 5190 zfid->zf_len = size; 5191 5192 for (i = 0; i < sizeof (zfid->zf_object); i++) 5193 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 5194 5195 /* Must have a non-zero generation number to distinguish from .zfs */ 5196 if (gen == 0) 5197 gen = 1; 5198 for (i = 0; i < sizeof (zfid->zf_gen); i++) 5199 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 5200 5201 if (size == LONG_FID_LEN) { 5202 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 5203 zfid_long_t *zlfid; 5204 5205 zlfid = (zfid_long_t *)fidp; 5206 5207 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 5208 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 5209 5210 /* XXX - this should be the generation number for the objset */ 5211 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 5212 zlfid->zf_setgen[i] = 0; 5213 } 5214 5215 ZFS_EXIT(zfsvfs); 5216 return (0); 5217} 5218 5219static int 5220zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 5221 caller_context_t *ct) 5222{ 5223 znode_t *zp, *xzp; 5224 zfsvfs_t *zfsvfs; 5225 zfs_dirlock_t *dl; 5226 int error; 5227 5228 switch (cmd) { 5229 case _PC_LINK_MAX: 5230 *valp = INT_MAX; 5231 return (0); 5232 5233 case _PC_FILESIZEBITS: 5234 *valp = 64; 5235 return (0); 5236#ifdef sun 5237 case _PC_XATTR_EXISTS: 5238 zp = VTOZ(vp); 5239 zfsvfs = zp->z_zfsvfs; 5240 ZFS_ENTER(zfsvfs); 5241 ZFS_VERIFY_ZP(zp); 5242 *valp = 0; 5243 error = zfs_dirent_lock(&dl, zp, "", &xzp, 5244 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 5245 if (error == 0) { 5246 zfs_dirent_unlock(dl); 5247 if (!zfs_dirempty(xzp)) 5248 *valp = 1; 5249 VN_RELE(ZTOV(xzp)); 5250 } else if (error == ENOENT) { 5251 /* 5252 * If there aren't extended attributes, it's the 5253 * same as having zero of them. 5254 */ 5255 error = 0; 5256 } 5257 ZFS_EXIT(zfsvfs); 5258 return (error); 5259 5260 case _PC_SATTR_ENABLED: 5261 case _PC_SATTR_EXISTS: 5262 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 5263 (vp->v_type == VREG || vp->v_type == VDIR); 5264 return (0); 5265 5266 case _PC_ACCESS_FILTERING: 5267 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 5268 vp->v_type == VDIR; 5269 return (0); 5270 5271 case _PC_ACL_ENABLED: 5272 *valp = _ACL_ACE_ENABLED; 5273 return (0); 5274#endif /* sun */ 5275 case _PC_MIN_HOLE_SIZE: 5276 *valp = (int)SPA_MINBLOCKSIZE; 5277 return (0); 5278#ifdef sun 5279 case _PC_TIMESTAMP_RESOLUTION: 5280 /* nanosecond timestamp resolution */ 5281 *valp = 1L; 5282 return (0); 5283#endif /* sun */ 5284 case _PC_ACL_EXTENDED: 5285 *valp = 0; 5286 return (0); 5287 5288 case _PC_ACL_NFS4: 5289 *valp = 1; 5290 return (0); 5291 5292 case _PC_ACL_PATH_MAX: 5293 *valp = ACL_MAX_ENTRIES; 5294 return (0); 5295 5296 default: 5297 return (EOPNOTSUPP); 5298 } 5299} 5300 5301/*ARGSUSED*/ 5302static int 5303zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5304 caller_context_t *ct) 5305{ 5306 znode_t *zp = VTOZ(vp); 5307 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5308 int error; 5309 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5310 5311 ZFS_ENTER(zfsvfs); 5312 ZFS_VERIFY_ZP(zp); 5313 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 5314 ZFS_EXIT(zfsvfs); 5315 5316 return (error); 5317} 5318 5319/*ARGSUSED*/ 5320int 5321zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5322 caller_context_t *ct) 5323{ 5324 znode_t *zp = VTOZ(vp); 5325 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5326 int error; 5327 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5328 zilog_t *zilog = zfsvfs->z_log; 5329 5330 ZFS_ENTER(zfsvfs); 5331 ZFS_VERIFY_ZP(zp); 5332 5333 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 5334 5335 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 5336 zil_commit(zilog, 0); 5337 5338 ZFS_EXIT(zfsvfs); 5339 return (error); 5340} 5341 5342#ifdef sun 5343/* 5344 * The smallest read we may consider to loan out an arcbuf. 5345 * This must be a power of 2. 5346 */ 5347int zcr_blksz_min = (1 << 10); /* 1K */ 5348/* 5349 * If set to less than the file block size, allow loaning out of an 5350 * arcbuf for a partial block read. This must be a power of 2. 5351 */ 5352int zcr_blksz_max = (1 << 17); /* 128K */ 5353 5354/*ARGSUSED*/ 5355static int 5356zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr, 5357 caller_context_t *ct) 5358{ 5359 znode_t *zp = VTOZ(vp); 5360 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5361 int max_blksz = zfsvfs->z_max_blksz; 5362 uio_t *uio = &xuio->xu_uio; 5363 ssize_t size = uio->uio_resid; 5364 offset_t offset = uio->uio_loffset; 5365 int blksz; 5366 int fullblk, i; 5367 arc_buf_t *abuf; 5368 ssize_t maxsize; 5369 int preamble, postamble; 5370 5371 if (xuio->xu_type != UIOTYPE_ZEROCOPY) 5372 return (SET_ERROR(EINVAL)); 5373 5374 ZFS_ENTER(zfsvfs); 5375 ZFS_VERIFY_ZP(zp); 5376 switch (ioflag) { 5377 case UIO_WRITE: 5378 /* 5379 * Loan out an arc_buf for write if write size is bigger than 5380 * max_blksz, and the file's block size is also max_blksz. 5381 */ 5382 blksz = max_blksz; 5383 if (size < blksz || zp->z_blksz != blksz) { 5384 ZFS_EXIT(zfsvfs); 5385 return (SET_ERROR(EINVAL)); 5386 } 5387 /* 5388 * Caller requests buffers for write before knowing where the 5389 * write offset might be (e.g. NFS TCP write). 5390 */ 5391 if (offset == -1) { 5392 preamble = 0; 5393 } else { 5394 preamble = P2PHASE(offset, blksz); 5395 if (preamble) { 5396 preamble = blksz - preamble; 5397 size -= preamble; 5398 } 5399 } 5400 5401 postamble = P2PHASE(size, blksz); 5402 size -= postamble; 5403 5404 fullblk = size / blksz; 5405 (void) dmu_xuio_init(xuio, 5406 (preamble != 0) + fullblk + (postamble != 0)); 5407 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble, 5408 int, postamble, int, 5409 (preamble != 0) + fullblk + (postamble != 0)); 5410 5411 /* 5412 * Have to fix iov base/len for partial buffers. They 5413 * currently represent full arc_buf's. 5414 */ 5415 if (preamble) { 5416 /* data begins in the middle of the arc_buf */ 5417 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5418 blksz); 5419 ASSERT(abuf); 5420 (void) dmu_xuio_add(xuio, abuf, 5421 blksz - preamble, preamble); 5422 } 5423 5424 for (i = 0; i < fullblk; i++) { 5425 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5426 blksz); 5427 ASSERT(abuf); 5428 (void) dmu_xuio_add(xuio, abuf, 0, blksz); 5429 } 5430 5431 if (postamble) { 5432 /* data ends in the middle of the arc_buf */ 5433 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5434 blksz); 5435 ASSERT(abuf); 5436 (void) dmu_xuio_add(xuio, abuf, 0, postamble); 5437 } 5438 break; 5439 case UIO_READ: 5440 /* 5441 * Loan out an arc_buf for read if the read size is larger than 5442 * the current file block size. Block alignment is not 5443 * considered. Partial arc_buf will be loaned out for read. 5444 */ 5445 blksz = zp->z_blksz; 5446 if (blksz < zcr_blksz_min) 5447 blksz = zcr_blksz_min; 5448 if (blksz > zcr_blksz_max) 5449 blksz = zcr_blksz_max; 5450 /* avoid potential complexity of dealing with it */ 5451 if (blksz > max_blksz) { 5452 ZFS_EXIT(zfsvfs); 5453 return (SET_ERROR(EINVAL)); 5454 } 5455 5456 maxsize = zp->z_size - uio->uio_loffset; 5457 if (size > maxsize) 5458 size = maxsize; 5459 5460 if (size < blksz || vn_has_cached_data(vp)) { 5461 ZFS_EXIT(zfsvfs); 5462 return (SET_ERROR(EINVAL)); 5463 } 5464 break; 5465 default: 5466 ZFS_EXIT(zfsvfs); 5467 return (SET_ERROR(EINVAL)); 5468 } 5469 5470 uio->uio_extflg = UIO_XUIO; 5471 XUIO_XUZC_RW(xuio) = ioflag; 5472 ZFS_EXIT(zfsvfs); 5473 return (0); 5474} 5475 5476/*ARGSUSED*/ 5477static int 5478zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct) 5479{ 5480 int i; 5481 arc_buf_t *abuf; 5482 int ioflag = XUIO_XUZC_RW(xuio); 5483 5484 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY); 5485 5486 i = dmu_xuio_cnt(xuio); 5487 while (i-- > 0) { 5488 abuf = dmu_xuio_arcbuf(xuio, i); 5489 /* 5490 * if abuf == NULL, it must be a write buffer 5491 * that has been returned in zfs_write(). 5492 */ 5493 if (abuf) 5494 dmu_return_arcbuf(abuf); 5495 ASSERT(abuf || ioflag == UIO_WRITE); 5496 } 5497 5498 dmu_xuio_fini(xuio); 5499 return (0); 5500} 5501 5502/* 5503 * Predeclare these here so that the compiler assumes that 5504 * this is an "old style" function declaration that does 5505 * not include arguments => we won't get type mismatch errors 5506 * in the initializations that follow. 5507 */ 5508static int zfs_inval(); 5509static int zfs_isdir(); 5510 5511static int 5512zfs_inval() 5513{ 5514 return (SET_ERROR(EINVAL)); 5515} 5516 5517static int 5518zfs_isdir() 5519{ 5520 return (SET_ERROR(EISDIR)); 5521} 5522/* 5523 * Directory vnode operations template 5524 */ 5525vnodeops_t *zfs_dvnodeops; 5526const fs_operation_def_t zfs_dvnodeops_template[] = { 5527 VOPNAME_OPEN, { .vop_open = zfs_open }, 5528 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5529 VOPNAME_READ, { .error = zfs_isdir }, 5530 VOPNAME_WRITE, { .error = zfs_isdir }, 5531 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5532 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5533 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5534 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5535 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5536 VOPNAME_CREATE, { .vop_create = zfs_create }, 5537 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5538 VOPNAME_LINK, { .vop_link = zfs_link }, 5539 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5540 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 5541 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5542 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5543 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 5544 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5545 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5546 VOPNAME_FID, { .vop_fid = zfs_fid }, 5547 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5548 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5549 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5550 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5551 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5552 NULL, NULL 5553}; 5554 5555/* 5556 * Regular file vnode operations template 5557 */ 5558vnodeops_t *zfs_fvnodeops; 5559const fs_operation_def_t zfs_fvnodeops_template[] = { 5560 VOPNAME_OPEN, { .vop_open = zfs_open }, 5561 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5562 VOPNAME_READ, { .vop_read = zfs_read }, 5563 VOPNAME_WRITE, { .vop_write = zfs_write }, 5564 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5565 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5566 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5567 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5568 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5569 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5570 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5571 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5572 VOPNAME_FID, { .vop_fid = zfs_fid }, 5573 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5574 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 5575 VOPNAME_SPACE, { .vop_space = zfs_space }, 5576 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 5577 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 5578 VOPNAME_MAP, { .vop_map = zfs_map }, 5579 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 5580 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 5581 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5582 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5583 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5584 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5585 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf }, 5586 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf }, 5587 NULL, NULL 5588}; 5589 5590/* 5591 * Symbolic link vnode operations template 5592 */ 5593vnodeops_t *zfs_symvnodeops; 5594const fs_operation_def_t zfs_symvnodeops_template[] = { 5595 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5596 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5597 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5598 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5599 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 5600 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5601 VOPNAME_FID, { .vop_fid = zfs_fid }, 5602 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5603 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5604 NULL, NULL 5605}; 5606 5607/* 5608 * special share hidden files vnode operations template 5609 */ 5610vnodeops_t *zfs_sharevnodeops; 5611const fs_operation_def_t zfs_sharevnodeops_template[] = { 5612 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5613 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5614 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5615 VOPNAME_FID, { .vop_fid = zfs_fid }, 5616 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5617 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5618 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5619 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5620 NULL, NULL 5621}; 5622 5623/* 5624 * Extended attribute directory vnode operations template 5625 * 5626 * This template is identical to the directory vnodes 5627 * operation template except for restricted operations: 5628 * VOP_MKDIR() 5629 * VOP_SYMLINK() 5630 * 5631 * Note that there are other restrictions embedded in: 5632 * zfs_create() - restrict type to VREG 5633 * zfs_link() - no links into/out of attribute space 5634 * zfs_rename() - no moves into/out of attribute space 5635 */ 5636vnodeops_t *zfs_xdvnodeops; 5637const fs_operation_def_t zfs_xdvnodeops_template[] = { 5638 VOPNAME_OPEN, { .vop_open = zfs_open }, 5639 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5640 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5641 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5642 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5643 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5644 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5645 VOPNAME_CREATE, { .vop_create = zfs_create }, 5646 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5647 VOPNAME_LINK, { .vop_link = zfs_link }, 5648 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5649 VOPNAME_MKDIR, { .error = zfs_inval }, 5650 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5651 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5652 VOPNAME_SYMLINK, { .error = zfs_inval }, 5653 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5654 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5655 VOPNAME_FID, { .vop_fid = zfs_fid }, 5656 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5657 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5658 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5659 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5660 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5661 NULL, NULL 5662}; 5663 5664/* 5665 * Error vnode operations template 5666 */ 5667vnodeops_t *zfs_evnodeops; 5668const fs_operation_def_t zfs_evnodeops_template[] = { 5669 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5670 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5671 NULL, NULL 5672}; 5673#endif /* sun */ 5674 5675static int 5676ioflags(int ioflags) 5677{ 5678 int flags = 0; 5679 5680 if (ioflags & IO_APPEND) 5681 flags |= FAPPEND; 5682 if (ioflags & IO_NDELAY) 5683 flags |= FNONBLOCK; 5684 if (ioflags & IO_SYNC) 5685 flags |= (FSYNC | FDSYNC | FRSYNC); 5686 5687 return (flags); 5688} 5689 5690static int 5691zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage) 5692{ 5693 znode_t *zp = VTOZ(vp); 5694 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5695 objset_t *os = zp->z_zfsvfs->z_os; 5696 vm_page_t mfirst, mlast, mreq; 5697 vm_object_t object; 5698 caddr_t va; 5699 struct sf_buf *sf; 5700 off_t startoff, endoff; 5701 int i, error; 5702 vm_pindex_t reqstart, reqend; 5703 int pcount, lsize, reqsize, size; 5704 5705 ZFS_ENTER(zfsvfs); 5706 ZFS_VERIFY_ZP(zp); 5707 5708 pcount = OFF_TO_IDX(round_page(count)); 5709 mreq = m[reqpage]; 5710 object = mreq->object; 5711 error = 0; 5712 5713 KASSERT(vp->v_object == object, ("mismatching object")); 5714 5715 if (pcount > 1 && zp->z_blksz > PAGESIZE) { 5716 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz); 5717 reqstart = OFF_TO_IDX(round_page(startoff)); 5718 if (reqstart < m[0]->pindex) 5719 reqstart = 0; 5720 else 5721 reqstart = reqstart - m[0]->pindex; 5722 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE, 5723 zp->z_blksz); 5724 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1; 5725 if (reqend > m[pcount - 1]->pindex) 5726 reqend = m[pcount - 1]->pindex; 5727 reqsize = reqend - m[reqstart]->pindex + 1; 5728 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize, 5729 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds")); 5730 } else { 5731 reqstart = reqpage; 5732 reqsize = 1; 5733 } 5734 mfirst = m[reqstart]; 5735 mlast = m[reqstart + reqsize - 1]; 5736 5737 zfs_vmobject_wlock(object); 5738 5739 for (i = 0; i < reqstart; i++) { 5740 vm_page_lock(m[i]); 5741 vm_page_free(m[i]); 5742 vm_page_unlock(m[i]); 5743 } 5744 for (i = reqstart + reqsize; i < pcount; i++) { 5745 vm_page_lock(m[i]); 5746 vm_page_free(m[i]); 5747 vm_page_unlock(m[i]); 5748 } 5749 5750 if (mreq->valid && reqsize == 1) { 5751 if (mreq->valid != VM_PAGE_BITS_ALL) 5752 vm_page_zero_invalid(mreq, TRUE); 5753 zfs_vmobject_wunlock(object); 5754 ZFS_EXIT(zfsvfs); 5755 return (zfs_vm_pagerret_ok); 5756 } 5757 5758 PCPU_INC(cnt.v_vnodein); 5759 PCPU_ADD(cnt.v_vnodepgsin, reqsize); 5760 5761 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) { 5762 for (i = reqstart; i < reqstart + reqsize; i++) { 5763 if (i != reqpage) { 5764 vm_page_lock(m[i]); 5765 vm_page_free(m[i]); 5766 vm_page_unlock(m[i]); 5767 } 5768 } 5769 zfs_vmobject_wunlock(object); 5770 ZFS_EXIT(zfsvfs); 5771 return (zfs_vm_pagerret_bad); 5772 } 5773 5774 lsize = PAGE_SIZE; 5775 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size) 5776 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex); 5777 5778 zfs_vmobject_wunlock(object); 5779 5780 for (i = reqstart; i < reqstart + reqsize; i++) { 5781 size = PAGE_SIZE; 5782 if (i == (reqstart + reqsize - 1)) 5783 size = lsize; 5784 va = zfs_map_page(m[i], &sf); 5785 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex), 5786 size, va, DMU_READ_PREFETCH); 5787 if (size != PAGE_SIZE) 5788 bzero(va + size, PAGE_SIZE - size); 5789 zfs_unmap_page(sf); 5790 if (error != 0) 5791 break; 5792 } 5793 5794 zfs_vmobject_wlock(object); 5795 5796 for (i = reqstart; i < reqstart + reqsize; i++) { 5797 if (!error) 5798 m[i]->valid = VM_PAGE_BITS_ALL; 5799 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i])); 5800 if (i != reqpage) 5801 vm_page_readahead_finish(m[i]); 5802 } 5803 5804 zfs_vmobject_wunlock(object); 5805 5806 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 5807 ZFS_EXIT(zfsvfs); 5808 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok); 5809} 5810 5811static int 5812zfs_freebsd_getpages(ap) 5813 struct vop_getpages_args /* { 5814 struct vnode *a_vp; 5815 vm_page_t *a_m; 5816 int a_count; 5817 int a_reqpage; 5818 vm_ooffset_t a_offset; 5819 } */ *ap; 5820{ 5821 5822 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage)); 5823} 5824 5825static int 5826zfs_freebsd_bmap(ap) 5827 struct vop_bmap_args /* { 5828 struct vnode *a_vp; 5829 daddr_t a_bn; 5830 struct bufobj **a_bop; 5831 daddr_t *a_bnp; 5832 int *a_runp; 5833 int *a_runb; 5834 } */ *ap; 5835{ 5836 5837 if (ap->a_bop != NULL) 5838 *ap->a_bop = &ap->a_vp->v_bufobj; 5839 if (ap->a_bnp != NULL) 5840 *ap->a_bnp = ap->a_bn; 5841 if (ap->a_runp != NULL) 5842 *ap->a_runp = 0; 5843 if (ap->a_runb != NULL) 5844 *ap->a_runb = 0; 5845 5846 return (0); 5847} 5848 5849static int 5850zfs_freebsd_open(ap) 5851 struct vop_open_args /* { 5852 struct vnode *a_vp; 5853 int a_mode; 5854 struct ucred *a_cred; 5855 struct thread *a_td; 5856 } */ *ap; 5857{ 5858 vnode_t *vp = ap->a_vp; 5859 znode_t *zp = VTOZ(vp); 5860 int error; 5861 5862 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL); 5863 if (error == 0) 5864 vnode_create_vobject(vp, zp->z_size, ap->a_td); 5865 return (error); 5866} 5867 5868static int 5869zfs_freebsd_close(ap) 5870 struct vop_close_args /* { 5871 struct vnode *a_vp; 5872 int a_fflag; 5873 struct ucred *a_cred; 5874 struct thread *a_td; 5875 } */ *ap; 5876{ 5877 5878 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL)); 5879} 5880 5881static int 5882zfs_freebsd_ioctl(ap) 5883 struct vop_ioctl_args /* { 5884 struct vnode *a_vp; 5885 u_long a_command; 5886 caddr_t a_data; 5887 int a_fflag; 5888 struct ucred *cred; 5889 struct thread *td; 5890 } */ *ap; 5891{ 5892 5893 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 5894 ap->a_fflag, ap->a_cred, NULL, NULL)); 5895} 5896 5897static int 5898zfs_freebsd_read(ap) 5899 struct vop_read_args /* { 5900 struct vnode *a_vp; 5901 struct uio *a_uio; 5902 int a_ioflag; 5903 struct ucred *a_cred; 5904 } */ *ap; 5905{ 5906 5907 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 5908 ap->a_cred, NULL)); 5909} 5910 5911static int 5912zfs_freebsd_write(ap) 5913 struct vop_write_args /* { 5914 struct vnode *a_vp; 5915 struct uio *a_uio; 5916 int a_ioflag; 5917 struct ucred *a_cred; 5918 } */ *ap; 5919{ 5920 5921 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 5922 ap->a_cred, NULL)); 5923} 5924 5925static int 5926zfs_freebsd_access(ap) 5927 struct vop_access_args /* { 5928 struct vnode *a_vp; 5929 accmode_t a_accmode; 5930 struct ucred *a_cred; 5931 struct thread *a_td; 5932 } */ *ap; 5933{ 5934 vnode_t *vp = ap->a_vp; 5935 znode_t *zp = VTOZ(vp); 5936 accmode_t accmode; 5937 int error = 0; 5938 5939 /* 5940 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 5941 */ 5942 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 5943 if (accmode != 0) 5944 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL); 5945 5946 /* 5947 * VADMIN has to be handled by vaccess(). 5948 */ 5949 if (error == 0) { 5950 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 5951 if (accmode != 0) { 5952 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 5953 zp->z_gid, accmode, ap->a_cred, NULL); 5954 } 5955 } 5956 5957 /* 5958 * For VEXEC, ensure that at least one execute bit is set for 5959 * non-directories. 5960 */ 5961 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 5962 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 5963 error = EACCES; 5964 } 5965 5966 return (error); 5967} 5968 5969static int 5970zfs_freebsd_lookup(ap) 5971 struct vop_lookup_args /* { 5972 struct vnode *a_dvp; 5973 struct vnode **a_vpp; 5974 struct componentname *a_cnp; 5975 } */ *ap; 5976{ 5977 struct componentname *cnp = ap->a_cnp; 5978 char nm[NAME_MAX + 1]; 5979 5980 ASSERT(cnp->cn_namelen < sizeof(nm)); 5981 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 5982 5983 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 5984 cnp->cn_cred, cnp->cn_thread, 0)); 5985} 5986 5987static int 5988zfs_freebsd_create(ap) 5989 struct vop_create_args /* { 5990 struct vnode *a_dvp; 5991 struct vnode **a_vpp; 5992 struct componentname *a_cnp; 5993 struct vattr *a_vap; 5994 } */ *ap; 5995{ 5996 struct componentname *cnp = ap->a_cnp; 5997 vattr_t *vap = ap->a_vap; 5998 int mode; 5999 6000 ASSERT(cnp->cn_flags & SAVENAME); 6001 6002 vattr_init_mask(vap); 6003 mode = vap->va_mode & ALLPERMS; 6004 6005 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 6006 ap->a_vpp, cnp->cn_cred, cnp->cn_thread)); 6007} 6008 6009static int 6010zfs_freebsd_remove(ap) 6011 struct vop_remove_args /* { 6012 struct vnode *a_dvp; 6013 struct vnode *a_vp; 6014 struct componentname *a_cnp; 6015 } */ *ap; 6016{ 6017 6018 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6019 6020 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 6021 ap->a_cnp->cn_cred, NULL, 0)); 6022} 6023 6024static int 6025zfs_freebsd_mkdir(ap) 6026 struct vop_mkdir_args /* { 6027 struct vnode *a_dvp; 6028 struct vnode **a_vpp; 6029 struct componentname *a_cnp; 6030 struct vattr *a_vap; 6031 } */ *ap; 6032{ 6033 vattr_t *vap = ap->a_vap; 6034 6035 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6036 6037 vattr_init_mask(vap); 6038 6039 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 6040 ap->a_cnp->cn_cred, NULL, 0, NULL)); 6041} 6042 6043static int 6044zfs_freebsd_rmdir(ap) 6045 struct vop_rmdir_args /* { 6046 struct vnode *a_dvp; 6047 struct vnode *a_vp; 6048 struct componentname *a_cnp; 6049 } */ *ap; 6050{ 6051 struct componentname *cnp = ap->a_cnp; 6052 6053 ASSERT(cnp->cn_flags & SAVENAME); 6054 6055 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0)); 6056} 6057 6058static int 6059zfs_freebsd_readdir(ap) 6060 struct vop_readdir_args /* { 6061 struct vnode *a_vp; 6062 struct uio *a_uio; 6063 struct ucred *a_cred; 6064 int *a_eofflag; 6065 int *a_ncookies; 6066 u_long **a_cookies; 6067 } */ *ap; 6068{ 6069 6070 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 6071 ap->a_ncookies, ap->a_cookies)); 6072} 6073 6074static int 6075zfs_freebsd_fsync(ap) 6076 struct vop_fsync_args /* { 6077 struct vnode *a_vp; 6078 int a_waitfor; 6079 struct thread *a_td; 6080 } */ *ap; 6081{ 6082 6083 vop_stdfsync(ap); 6084 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL)); 6085} 6086 6087static int 6088zfs_freebsd_getattr(ap) 6089 struct vop_getattr_args /* { 6090 struct vnode *a_vp; 6091 struct vattr *a_vap; 6092 struct ucred *a_cred; 6093 } */ *ap; 6094{ 6095 vattr_t *vap = ap->a_vap; 6096 xvattr_t xvap; 6097 u_long fflags = 0; 6098 int error; 6099 6100 xva_init(&xvap); 6101 xvap.xva_vattr = *vap; 6102 xvap.xva_vattr.va_mask |= AT_XVATTR; 6103 6104 /* Convert chflags into ZFS-type flags. */ 6105 /* XXX: what about SF_SETTABLE?. */ 6106 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 6107 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 6108 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 6109 XVA_SET_REQ(&xvap, XAT_NODUMP); 6110 XVA_SET_REQ(&xvap, XAT_READONLY); 6111 XVA_SET_REQ(&xvap, XAT_ARCHIVE); 6112 XVA_SET_REQ(&xvap, XAT_SYSTEM); 6113 XVA_SET_REQ(&xvap, XAT_HIDDEN); 6114 XVA_SET_REQ(&xvap, XAT_REPARSE); 6115 XVA_SET_REQ(&xvap, XAT_OFFLINE); 6116 XVA_SET_REQ(&xvap, XAT_SPARSE); 6117 6118 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL); 6119 if (error != 0) 6120 return (error); 6121 6122 /* Convert ZFS xattr into chflags. */ 6123#define FLAG_CHECK(fflag, xflag, xfield) do { \ 6124 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 6125 fflags |= (fflag); \ 6126} while (0) 6127 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 6128 xvap.xva_xoptattrs.xoa_immutable); 6129 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 6130 xvap.xva_xoptattrs.xoa_appendonly); 6131 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 6132 xvap.xva_xoptattrs.xoa_nounlink); 6133 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE, 6134 xvap.xva_xoptattrs.xoa_archive); 6135 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 6136 xvap.xva_xoptattrs.xoa_nodump); 6137 FLAG_CHECK(UF_READONLY, XAT_READONLY, 6138 xvap.xva_xoptattrs.xoa_readonly); 6139 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM, 6140 xvap.xva_xoptattrs.xoa_system); 6141 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN, 6142 xvap.xva_xoptattrs.xoa_hidden); 6143 FLAG_CHECK(UF_REPARSE, XAT_REPARSE, 6144 xvap.xva_xoptattrs.xoa_reparse); 6145 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE, 6146 xvap.xva_xoptattrs.xoa_offline); 6147 FLAG_CHECK(UF_SPARSE, XAT_SPARSE, 6148 xvap.xva_xoptattrs.xoa_sparse); 6149 6150#undef FLAG_CHECK 6151 *vap = xvap.xva_vattr; 6152 vap->va_flags = fflags; 6153 return (0); 6154} 6155 6156static int 6157zfs_freebsd_setattr(ap) 6158 struct vop_setattr_args /* { 6159 struct vnode *a_vp; 6160 struct vattr *a_vap; 6161 struct ucred *a_cred; 6162 } */ *ap; 6163{ 6164 vnode_t *vp = ap->a_vp; 6165 vattr_t *vap = ap->a_vap; 6166 cred_t *cred = ap->a_cred; 6167 xvattr_t xvap; 6168 u_long fflags; 6169 uint64_t zflags; 6170 6171 vattr_init_mask(vap); 6172 vap->va_mask &= ~AT_NOSET; 6173 6174 xva_init(&xvap); 6175 xvap.xva_vattr = *vap; 6176 6177 zflags = VTOZ(vp)->z_pflags; 6178 6179 if (vap->va_flags != VNOVAL) { 6180 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 6181 int error; 6182 6183 if (zfsvfs->z_use_fuids == B_FALSE) 6184 return (EOPNOTSUPP); 6185 6186 fflags = vap->va_flags; 6187 /* 6188 * XXX KDM 6189 * We need to figure out whether it makes sense to allow 6190 * UF_REPARSE through, since we don't really have other 6191 * facilities to handle reparse points and zfs_setattr() 6192 * doesn't currently allow setting that attribute anyway. 6193 */ 6194 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE| 6195 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE| 6196 UF_OFFLINE|UF_SPARSE)) != 0) 6197 return (EOPNOTSUPP); 6198 /* 6199 * Unprivileged processes are not permitted to unset system 6200 * flags, or modify flags if any system flags are set. 6201 * Privileged non-jail processes may not modify system flags 6202 * if securelevel > 0 and any existing system flags are set. 6203 * Privileged jail processes behave like privileged non-jail 6204 * processes if the security.jail.chflags_allowed sysctl is 6205 * is non-zero; otherwise, they behave like unprivileged 6206 * processes. 6207 */ 6208 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 6209 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) { 6210 if (zflags & 6211 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6212 error = securelevel_gt(cred, 0); 6213 if (error != 0) 6214 return (error); 6215 } 6216 } else { 6217 /* 6218 * Callers may only modify the file flags on objects they 6219 * have VADMIN rights for. 6220 */ 6221 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0) 6222 return (error); 6223 if (zflags & 6224 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6225 return (EPERM); 6226 } 6227 if (fflags & 6228 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 6229 return (EPERM); 6230 } 6231 } 6232 6233#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 6234 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 6235 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 6236 XVA_SET_REQ(&xvap, (xflag)); \ 6237 (xfield) = ((fflags & (fflag)) != 0); \ 6238 } \ 6239} while (0) 6240 /* Convert chflags into ZFS-type flags. */ 6241 /* XXX: what about SF_SETTABLE?. */ 6242 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 6243 xvap.xva_xoptattrs.xoa_immutable); 6244 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 6245 xvap.xva_xoptattrs.xoa_appendonly); 6246 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 6247 xvap.xva_xoptattrs.xoa_nounlink); 6248 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE, 6249 xvap.xva_xoptattrs.xoa_archive); 6250 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 6251 xvap.xva_xoptattrs.xoa_nodump); 6252 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY, 6253 xvap.xva_xoptattrs.xoa_readonly); 6254 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM, 6255 xvap.xva_xoptattrs.xoa_system); 6256 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN, 6257 xvap.xva_xoptattrs.xoa_hidden); 6258 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE, 6259 xvap.xva_xoptattrs.xoa_hidden); 6260 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE, 6261 xvap.xva_xoptattrs.xoa_offline); 6262 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE, 6263 xvap.xva_xoptattrs.xoa_sparse); 6264#undef FLAG_CHANGE 6265 } 6266 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL)); 6267} 6268 6269static int 6270zfs_freebsd_rename(ap) 6271 struct vop_rename_args /* { 6272 struct vnode *a_fdvp; 6273 struct vnode *a_fvp; 6274 struct componentname *a_fcnp; 6275 struct vnode *a_tdvp; 6276 struct vnode *a_tvp; 6277 struct componentname *a_tcnp; 6278 } */ *ap; 6279{ 6280 vnode_t *fdvp = ap->a_fdvp; 6281 vnode_t *fvp = ap->a_fvp; 6282 vnode_t *tdvp = ap->a_tdvp; 6283 vnode_t *tvp = ap->a_tvp; 6284 int error; 6285 6286 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 6287 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 6288 6289 /* 6290 * Check for cross-device rename. 6291 */ 6292 if ((fdvp->v_mount != tdvp->v_mount) || 6293 (tvp && (fdvp->v_mount != tvp->v_mount))) 6294 error = EXDEV; 6295 else 6296 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 6297 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0); 6298 if (tdvp == tvp) 6299 VN_RELE(tdvp); 6300 else 6301 VN_URELE(tdvp); 6302 if (tvp) 6303 VN_URELE(tvp); 6304 VN_RELE(fdvp); 6305 VN_RELE(fvp); 6306 6307 return (error); 6308} 6309 6310static int 6311zfs_freebsd_symlink(ap) 6312 struct vop_symlink_args /* { 6313 struct vnode *a_dvp; 6314 struct vnode **a_vpp; 6315 struct componentname *a_cnp; 6316 struct vattr *a_vap; 6317 char *a_target; 6318 } */ *ap; 6319{ 6320 struct componentname *cnp = ap->a_cnp; 6321 vattr_t *vap = ap->a_vap; 6322 6323 ASSERT(cnp->cn_flags & SAVENAME); 6324 6325 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 6326 vattr_init_mask(vap); 6327 6328 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 6329 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 6330} 6331 6332static int 6333zfs_freebsd_readlink(ap) 6334 struct vop_readlink_args /* { 6335 struct vnode *a_vp; 6336 struct uio *a_uio; 6337 struct ucred *a_cred; 6338 } */ *ap; 6339{ 6340 6341 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL)); 6342} 6343 6344static int 6345zfs_freebsd_link(ap) 6346 struct vop_link_args /* { 6347 struct vnode *a_tdvp; 6348 struct vnode *a_vp; 6349 struct componentname *a_cnp; 6350 } */ *ap; 6351{ 6352 struct componentname *cnp = ap->a_cnp; 6353 vnode_t *vp = ap->a_vp; 6354 vnode_t *tdvp = ap->a_tdvp; 6355 6356 if (tdvp->v_mount != vp->v_mount) 6357 return (EXDEV); 6358 6359 ASSERT(cnp->cn_flags & SAVENAME); 6360 6361 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0)); 6362} 6363 6364static int 6365zfs_freebsd_inactive(ap) 6366 struct vop_inactive_args /* { 6367 struct vnode *a_vp; 6368 struct thread *a_td; 6369 } */ *ap; 6370{ 6371 vnode_t *vp = ap->a_vp; 6372 6373 zfs_inactive(vp, ap->a_td->td_ucred, NULL); 6374 return (0); 6375} 6376 6377static int 6378zfs_freebsd_reclaim(ap) 6379 struct vop_reclaim_args /* { 6380 struct vnode *a_vp; 6381 struct thread *a_td; 6382 } */ *ap; 6383{ 6384 vnode_t *vp = ap->a_vp; 6385 znode_t *zp = VTOZ(vp); 6386 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6387 6388 ASSERT(zp != NULL); 6389 6390 /* Destroy the vm object and flush associated pages. */ 6391 vnode_destroy_vobject(vp); 6392 6393 /* 6394 * z_teardown_inactive_lock protects from a race with 6395 * zfs_znode_dmu_fini in zfsvfs_teardown during 6396 * force unmount. 6397 */ 6398 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 6399 if (zp->z_sa_hdl == NULL) 6400 zfs_znode_free(zp); 6401 else 6402 zfs_zinactive(zp); 6403 rw_exit(&zfsvfs->z_teardown_inactive_lock); 6404 6405 vp->v_data = NULL; 6406 return (0); 6407} 6408 6409static int 6410zfs_freebsd_fid(ap) 6411 struct vop_fid_args /* { 6412 struct vnode *a_vp; 6413 struct fid *a_fid; 6414 } */ *ap; 6415{ 6416 6417 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 6418} 6419 6420static int 6421zfs_freebsd_pathconf(ap) 6422 struct vop_pathconf_args /* { 6423 struct vnode *a_vp; 6424 int a_name; 6425 register_t *a_retval; 6426 } */ *ap; 6427{ 6428 ulong_t val; 6429 int error; 6430 6431 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); 6432 if (error == 0) 6433 *ap->a_retval = val; 6434 else if (error == EOPNOTSUPP) 6435 error = vop_stdpathconf(ap); 6436 return (error); 6437} 6438 6439static int 6440zfs_freebsd_fifo_pathconf(ap) 6441 struct vop_pathconf_args /* { 6442 struct vnode *a_vp; 6443 int a_name; 6444 register_t *a_retval; 6445 } */ *ap; 6446{ 6447 6448 switch (ap->a_name) { 6449 case _PC_ACL_EXTENDED: 6450 case _PC_ACL_NFS4: 6451 case _PC_ACL_PATH_MAX: 6452 case _PC_MAC_PRESENT: 6453 return (zfs_freebsd_pathconf(ap)); 6454 default: 6455 return (fifo_specops.vop_pathconf(ap)); 6456 } 6457} 6458 6459/* 6460 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 6461 * extended attribute name: 6462 * 6463 * NAMESPACE PREFIX 6464 * system freebsd:system: 6465 * user (none, can be used to access ZFS fsattr(5) attributes 6466 * created on Solaris) 6467 */ 6468static int 6469zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 6470 size_t size) 6471{ 6472 const char *namespace, *prefix, *suffix; 6473 6474 /* We don't allow '/' character in attribute name. */ 6475 if (strchr(name, '/') != NULL) 6476 return (EINVAL); 6477 /* We don't allow attribute names that start with "freebsd:" string. */ 6478 if (strncmp(name, "freebsd:", 8) == 0) 6479 return (EINVAL); 6480 6481 bzero(attrname, size); 6482 6483 switch (attrnamespace) { 6484 case EXTATTR_NAMESPACE_USER: 6485#if 0 6486 prefix = "freebsd:"; 6487 namespace = EXTATTR_NAMESPACE_USER_STRING; 6488 suffix = ":"; 6489#else 6490 /* 6491 * This is the default namespace by which we can access all 6492 * attributes created on Solaris. 6493 */ 6494 prefix = namespace = suffix = ""; 6495#endif 6496 break; 6497 case EXTATTR_NAMESPACE_SYSTEM: 6498 prefix = "freebsd:"; 6499 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 6500 suffix = ":"; 6501 break; 6502 case EXTATTR_NAMESPACE_EMPTY: 6503 default: 6504 return (EINVAL); 6505 } 6506 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 6507 name) >= size) { 6508 return (ENAMETOOLONG); 6509 } 6510 return (0); 6511} 6512 6513/* 6514 * Vnode operating to retrieve a named extended attribute. 6515 */ 6516static int 6517zfs_getextattr(struct vop_getextattr_args *ap) 6518/* 6519vop_getextattr { 6520 IN struct vnode *a_vp; 6521 IN int a_attrnamespace; 6522 IN const char *a_name; 6523 INOUT struct uio *a_uio; 6524 OUT size_t *a_size; 6525 IN struct ucred *a_cred; 6526 IN struct thread *a_td; 6527}; 6528*/ 6529{ 6530 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6531 struct thread *td = ap->a_td; 6532 struct nameidata nd; 6533 char attrname[255]; 6534 struct vattr va; 6535 vnode_t *xvp = NULL, *vp; 6536 int error, flags; 6537 6538 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6539 ap->a_cred, ap->a_td, VREAD); 6540 if (error != 0) 6541 return (error); 6542 6543 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6544 sizeof(attrname)); 6545 if (error != 0) 6546 return (error); 6547 6548 ZFS_ENTER(zfsvfs); 6549 6550 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6551 LOOKUP_XATTR); 6552 if (error != 0) { 6553 ZFS_EXIT(zfsvfs); 6554 return (error); 6555 } 6556 6557 flags = FREAD; 6558 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6559 xvp, td); 6560 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL); 6561 vp = nd.ni_vp; 6562 NDFREE(&nd, NDF_ONLY_PNBUF); 6563 if (error != 0) { 6564 ZFS_EXIT(zfsvfs); 6565 if (error == ENOENT) 6566 error = ENOATTR; 6567 return (error); 6568 } 6569 6570 if (ap->a_size != NULL) { 6571 error = VOP_GETATTR(vp, &va, ap->a_cred); 6572 if (error == 0) 6573 *ap->a_size = (size_t)va.va_size; 6574 } else if (ap->a_uio != NULL) 6575 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6576 6577 VOP_UNLOCK(vp, 0); 6578 vn_close(vp, flags, ap->a_cred, td); 6579 ZFS_EXIT(zfsvfs); 6580 6581 return (error); 6582} 6583 6584/* 6585 * Vnode operation to remove a named attribute. 6586 */ 6587int 6588zfs_deleteextattr(struct vop_deleteextattr_args *ap) 6589/* 6590vop_deleteextattr { 6591 IN struct vnode *a_vp; 6592 IN int a_attrnamespace; 6593 IN const char *a_name; 6594 IN struct ucred *a_cred; 6595 IN struct thread *a_td; 6596}; 6597*/ 6598{ 6599 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6600 struct thread *td = ap->a_td; 6601 struct nameidata nd; 6602 char attrname[255]; 6603 struct vattr va; 6604 vnode_t *xvp = NULL, *vp; 6605 int error, flags; 6606 6607 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6608 ap->a_cred, ap->a_td, VWRITE); 6609 if (error != 0) 6610 return (error); 6611 6612 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6613 sizeof(attrname)); 6614 if (error != 0) 6615 return (error); 6616 6617 ZFS_ENTER(zfsvfs); 6618 6619 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6620 LOOKUP_XATTR); 6621 if (error != 0) { 6622 ZFS_EXIT(zfsvfs); 6623 return (error); 6624 } 6625 6626 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 6627 UIO_SYSSPACE, attrname, xvp, td); 6628 error = namei(&nd); 6629 vp = nd.ni_vp; 6630 NDFREE(&nd, NDF_ONLY_PNBUF); 6631 if (error != 0) { 6632 ZFS_EXIT(zfsvfs); 6633 if (error == ENOENT) 6634 error = ENOATTR; 6635 return (error); 6636 } 6637 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 6638 6639 vput(nd.ni_dvp); 6640 if (vp == nd.ni_dvp) 6641 vrele(vp); 6642 else 6643 vput(vp); 6644 ZFS_EXIT(zfsvfs); 6645 6646 return (error); 6647} 6648 6649/* 6650 * Vnode operation to set a named attribute. 6651 */ 6652static int 6653zfs_setextattr(struct vop_setextattr_args *ap) 6654/* 6655vop_setextattr { 6656 IN struct vnode *a_vp; 6657 IN int a_attrnamespace; 6658 IN const char *a_name; 6659 INOUT struct uio *a_uio; 6660 IN struct ucred *a_cred; 6661 IN struct thread *a_td; 6662}; 6663*/ 6664{ 6665 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6666 struct thread *td = ap->a_td; 6667 struct nameidata nd; 6668 char attrname[255]; 6669 struct vattr va; 6670 vnode_t *xvp = NULL, *vp; 6671 int error, flags; 6672 6673 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6674 ap->a_cred, ap->a_td, VWRITE); 6675 if (error != 0) 6676 return (error); 6677 6678 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6679 sizeof(attrname)); 6680 if (error != 0) 6681 return (error); 6682 6683 ZFS_ENTER(zfsvfs); 6684 6685 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6686 LOOKUP_XATTR | CREATE_XATTR_DIR); 6687 if (error != 0) { 6688 ZFS_EXIT(zfsvfs); 6689 return (error); 6690 } 6691 6692 flags = FFLAGS(O_WRONLY | O_CREAT); 6693 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6694 xvp, td); 6695 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL); 6696 vp = nd.ni_vp; 6697 NDFREE(&nd, NDF_ONLY_PNBUF); 6698 if (error != 0) { 6699 ZFS_EXIT(zfsvfs); 6700 return (error); 6701 } 6702 6703 VATTR_NULL(&va); 6704 va.va_size = 0; 6705 error = VOP_SETATTR(vp, &va, ap->a_cred); 6706 if (error == 0) 6707 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred); 6708 6709 VOP_UNLOCK(vp, 0); 6710 vn_close(vp, flags, ap->a_cred, td); 6711 ZFS_EXIT(zfsvfs); 6712 6713 return (error); 6714} 6715 6716/* 6717 * Vnode operation to retrieve extended attributes on a vnode. 6718 */ 6719static int 6720zfs_listextattr(struct vop_listextattr_args *ap) 6721/* 6722vop_listextattr { 6723 IN struct vnode *a_vp; 6724 IN int a_attrnamespace; 6725 INOUT struct uio *a_uio; 6726 OUT size_t *a_size; 6727 IN struct ucred *a_cred; 6728 IN struct thread *a_td; 6729}; 6730*/ 6731{ 6732 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6733 struct thread *td = ap->a_td; 6734 struct nameidata nd; 6735 char attrprefix[16]; 6736 u_char dirbuf[sizeof(struct dirent)]; 6737 struct dirent *dp; 6738 struct iovec aiov; 6739 struct uio auio, *uio = ap->a_uio; 6740 size_t *sizep = ap->a_size; 6741 size_t plen; 6742 vnode_t *xvp = NULL, *vp; 6743 int done, error, eof, pos; 6744 6745 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6746 ap->a_cred, ap->a_td, VREAD); 6747 if (error != 0) 6748 return (error); 6749 6750 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 6751 sizeof(attrprefix)); 6752 if (error != 0) 6753 return (error); 6754 plen = strlen(attrprefix); 6755 6756 ZFS_ENTER(zfsvfs); 6757 6758 if (sizep != NULL) 6759 *sizep = 0; 6760 6761 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6762 LOOKUP_XATTR); 6763 if (error != 0) { 6764 ZFS_EXIT(zfsvfs); 6765 /* 6766 * ENOATTR means that the EA directory does not yet exist, 6767 * i.e. there are no extended attributes there. 6768 */ 6769 if (error == ENOATTR) 6770 error = 0; 6771 return (error); 6772 } 6773 6774 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 6775 UIO_SYSSPACE, ".", xvp, td); 6776 error = namei(&nd); 6777 vp = nd.ni_vp; 6778 NDFREE(&nd, NDF_ONLY_PNBUF); 6779 if (error != 0) { 6780 ZFS_EXIT(zfsvfs); 6781 return (error); 6782 } 6783 6784 auio.uio_iov = &aiov; 6785 auio.uio_iovcnt = 1; 6786 auio.uio_segflg = UIO_SYSSPACE; 6787 auio.uio_td = td; 6788 auio.uio_rw = UIO_READ; 6789 auio.uio_offset = 0; 6790 6791 do { 6792 u_char nlen; 6793 6794 aiov.iov_base = (void *)dirbuf; 6795 aiov.iov_len = sizeof(dirbuf); 6796 auio.uio_resid = sizeof(dirbuf); 6797 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 6798 done = sizeof(dirbuf) - auio.uio_resid; 6799 if (error != 0) 6800 break; 6801 for (pos = 0; pos < done;) { 6802 dp = (struct dirent *)(dirbuf + pos); 6803 pos += dp->d_reclen; 6804 /* 6805 * XXX: Temporarily we also accept DT_UNKNOWN, as this 6806 * is what we get when attribute was created on Solaris. 6807 */ 6808 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 6809 continue; 6810 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0) 6811 continue; 6812 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 6813 continue; 6814 nlen = dp->d_namlen - plen; 6815 if (sizep != NULL) 6816 *sizep += 1 + nlen; 6817 else if (uio != NULL) { 6818 /* 6819 * Format of extattr name entry is one byte for 6820 * length and the rest for name. 6821 */ 6822 error = uiomove(&nlen, 1, uio->uio_rw, uio); 6823 if (error == 0) { 6824 error = uiomove(dp->d_name + plen, nlen, 6825 uio->uio_rw, uio); 6826 } 6827 if (error != 0) 6828 break; 6829 } 6830 } 6831 } while (!eof && error == 0); 6832 6833 vput(vp); 6834 ZFS_EXIT(zfsvfs); 6835 6836 return (error); 6837} 6838 6839int 6840zfs_freebsd_getacl(ap) 6841 struct vop_getacl_args /* { 6842 struct vnode *vp; 6843 acl_type_t type; 6844 struct acl *aclp; 6845 struct ucred *cred; 6846 struct thread *td; 6847 } */ *ap; 6848{ 6849 int error; 6850 vsecattr_t vsecattr; 6851 6852 if (ap->a_type != ACL_TYPE_NFS4) 6853 return (EINVAL); 6854 6855 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 6856 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL)) 6857 return (error); 6858 6859 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt); 6860 if (vsecattr.vsa_aclentp != NULL) 6861 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 6862 6863 return (error); 6864} 6865 6866int 6867zfs_freebsd_setacl(ap) 6868 struct vop_setacl_args /* { 6869 struct vnode *vp; 6870 acl_type_t type; 6871 struct acl *aclp; 6872 struct ucred *cred; 6873 struct thread *td; 6874 } */ *ap; 6875{ 6876 int error; 6877 vsecattr_t vsecattr; 6878 int aclbsize; /* size of acl list in bytes */ 6879 aclent_t *aaclp; 6880 6881 if (ap->a_type != ACL_TYPE_NFS4) 6882 return (EINVAL); 6883 6884 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 6885 return (EINVAL); 6886 6887 /* 6888 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 6889 * splitting every entry into two and appending "canonical six" 6890 * entries at the end. Don't allow for setting an ACL that would 6891 * cause chmod(2) to run out of ACL entries. 6892 */ 6893 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 6894 return (ENOSPC); 6895 6896 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 6897 if (error != 0) 6898 return (error); 6899 6900 vsecattr.vsa_mask = VSA_ACE; 6901 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t); 6902 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 6903 aaclp = vsecattr.vsa_aclentp; 6904 vsecattr.vsa_aclentsz = aclbsize; 6905 6906 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 6907 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL); 6908 kmem_free(aaclp, aclbsize); 6909 6910 return (error); 6911} 6912 6913int 6914zfs_freebsd_aclcheck(ap) 6915 struct vop_aclcheck_args /* { 6916 struct vnode *vp; 6917 acl_type_t type; 6918 struct acl *aclp; 6919 struct ucred *cred; 6920 struct thread *td; 6921 } */ *ap; 6922{ 6923 6924 return (EOPNOTSUPP); 6925} 6926 6927struct vop_vector zfs_vnodeops; 6928struct vop_vector zfs_fifoops; 6929struct vop_vector zfs_shareops; 6930 6931struct vop_vector zfs_vnodeops = { 6932 .vop_default = &default_vnodeops, 6933 .vop_inactive = zfs_freebsd_inactive, 6934 .vop_reclaim = zfs_freebsd_reclaim, 6935 .vop_access = zfs_freebsd_access, 6936#ifdef FREEBSD_NAMECACHE 6937 .vop_lookup = vfs_cache_lookup, 6938 .vop_cachedlookup = zfs_freebsd_lookup, 6939#else 6940 .vop_lookup = zfs_freebsd_lookup, 6941#endif 6942 .vop_getattr = zfs_freebsd_getattr, 6943 .vop_setattr = zfs_freebsd_setattr, 6944 .vop_create = zfs_freebsd_create, 6945 .vop_mknod = zfs_freebsd_create, 6946 .vop_mkdir = zfs_freebsd_mkdir, 6947 .vop_readdir = zfs_freebsd_readdir, 6948 .vop_fsync = zfs_freebsd_fsync, 6949 .vop_open = zfs_freebsd_open, 6950 .vop_close = zfs_freebsd_close, 6951 .vop_rmdir = zfs_freebsd_rmdir, 6952 .vop_ioctl = zfs_freebsd_ioctl, 6953 .vop_link = zfs_freebsd_link, 6954 .vop_symlink = zfs_freebsd_symlink, 6955 .vop_readlink = zfs_freebsd_readlink, 6956 .vop_read = zfs_freebsd_read, 6957 .vop_write = zfs_freebsd_write, 6958 .vop_remove = zfs_freebsd_remove, 6959 .vop_rename = zfs_freebsd_rename, 6960 .vop_pathconf = zfs_freebsd_pathconf, 6961 .vop_bmap = zfs_freebsd_bmap, 6962 .vop_fid = zfs_freebsd_fid, 6963 .vop_getextattr = zfs_getextattr, 6964 .vop_deleteextattr = zfs_deleteextattr, 6965 .vop_setextattr = zfs_setextattr, 6966 .vop_listextattr = zfs_listextattr, 6967 .vop_getacl = zfs_freebsd_getacl, 6968 .vop_setacl = zfs_freebsd_setacl, 6969 .vop_aclcheck = zfs_freebsd_aclcheck, 6970 .vop_getpages = zfs_freebsd_getpages, 6971}; 6972 6973struct vop_vector zfs_fifoops = { 6974 .vop_default = &fifo_specops, 6975 .vop_fsync = zfs_freebsd_fsync, 6976 .vop_access = zfs_freebsd_access, 6977 .vop_getattr = zfs_freebsd_getattr, 6978 .vop_inactive = zfs_freebsd_inactive, 6979 .vop_read = VOP_PANIC, 6980 .vop_reclaim = zfs_freebsd_reclaim, 6981 .vop_setattr = zfs_freebsd_setattr, 6982 .vop_write = VOP_PANIC, 6983 .vop_pathconf = zfs_freebsd_fifo_pathconf, 6984 .vop_fid = zfs_freebsd_fid, 6985 .vop_getacl = zfs_freebsd_getacl, 6986 .vop_setacl = zfs_freebsd_setacl, 6987 .vop_aclcheck = zfs_freebsd_aclcheck, 6988}; 6989 6990/* 6991 * special share hidden files vnode operations template 6992 */ 6993struct vop_vector zfs_shareops = { 6994 .vop_default = &default_vnodeops, 6995 .vop_access = zfs_freebsd_access, 6996 .vop_inactive = zfs_freebsd_inactive, 6997 .vop_reclaim = zfs_freebsd_reclaim, 6998 .vop_fid = zfs_freebsd_fid, 6999 .vop_pathconf = zfs_freebsd_pathconf, 7000}; 7001