zfs_vnops.c revision 297096
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2015 by Delphix. All rights reserved. 24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved. 25 */ 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/sched.h> 75#include <sys/acl.h> 76#include <vm/vm_param.h> 77#include <vm/vm_pageout.h> 78 79/* 80 * Programming rules. 81 * 82 * Each vnode op performs some logical unit of work. To do this, the ZPL must 83 * properly lock its in-core state, create a DMU transaction, do the work, 84 * record this work in the intent log (ZIL), commit the DMU transaction, 85 * and wait for the intent log to commit if it is a synchronous operation. 86 * Moreover, the vnode ops must work in both normal and log replay context. 87 * The ordering of events is important to avoid deadlocks and references 88 * to freed memory. The example below illustrates the following Big Rules: 89 * 90 * (1) A check must be made in each zfs thread for a mounted file system. 91 * This is done avoiding races using ZFS_ENTER(zfsvfs). 92 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes 93 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros 94 * can return EIO from the calling function. 95 * 96 * (2) VN_RELE() should always be the last thing except for zil_commit() 97 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 98 * First, if it's the last reference, the vnode/znode 99 * can be freed, so the zp may point to freed memory. Second, the last 100 * reference will call zfs_zinactive(), which may induce a lot of work -- 101 * pushing cached pages (which acquires range locks) and syncing out 102 * cached atime changes. Third, zfs_zinactive() may require a new tx, 103 * which could deadlock the system if you were already holding one. 104 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). 105 * 106 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 107 * as they can span dmu_tx_assign() calls. 108 * 109 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to 110 * dmu_tx_assign(). This is critical because we don't want to block 111 * while holding locks. 112 * 113 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This 114 * reduces lock contention and CPU usage when we must wait (note that if 115 * throughput is constrained by the storage, nearly every transaction 116 * must wait). 117 * 118 * Note, in particular, that if a lock is sometimes acquired before 119 * the tx assigns, and sometimes after (e.g. z_lock), then failing 120 * to use a non-blocking assign can deadlock the system. The scenario: 121 * 122 * Thread A has grabbed a lock before calling dmu_tx_assign(). 123 * Thread B is in an already-assigned tx, and blocks for this lock. 124 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 125 * forever, because the previous txg can't quiesce until B's tx commits. 126 * 127 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 128 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent 129 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT, 130 * to indicate that this operation has already called dmu_tx_wait(). 131 * This will ensure that we don't retry forever, waiting a short bit 132 * each time. 133 * 134 * (5) If the operation succeeded, generate the intent log entry for it 135 * before dropping locks. This ensures that the ordering of events 136 * in the intent log matches the order in which they actually occurred. 137 * During ZIL replay the zfs_log_* functions will update the sequence 138 * number to indicate the zil transaction has replayed. 139 * 140 * (6) At the end of each vnode op, the DMU tx must always commit, 141 * regardless of whether there were any errors. 142 * 143 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 144 * to ensure that synchronous semantics are provided when necessary. 145 * 146 * In general, this is how things should be ordered in each vnode op: 147 * 148 * ZFS_ENTER(zfsvfs); // exit if unmounted 149 * top: 150 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 151 * rw_enter(...); // grab any other locks you need 152 * tx = dmu_tx_create(...); // get DMU tx 153 * dmu_tx_hold_*(); // hold each object you might modify 154 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 155 * if (error) { 156 * rw_exit(...); // drop locks 157 * zfs_dirent_unlock(dl); // unlock directory entry 158 * VN_RELE(...); // release held vnodes 159 * if (error == ERESTART) { 160 * waited = B_TRUE; 161 * dmu_tx_wait(tx); 162 * dmu_tx_abort(tx); 163 * goto top; 164 * } 165 * dmu_tx_abort(tx); // abort DMU tx 166 * ZFS_EXIT(zfsvfs); // finished in zfs 167 * return (error); // really out of space 168 * } 169 * error = do_real_work(); // do whatever this VOP does 170 * if (error == 0) 171 * zfs_log_*(...); // on success, make ZIL entry 172 * dmu_tx_commit(tx); // commit DMU tx -- error or not 173 * rw_exit(...); // drop locks 174 * zfs_dirent_unlock(dl); // unlock directory entry 175 * VN_RELE(...); // release held vnodes 176 * zil_commit(zilog, foid); // synchronous when necessary 177 * ZFS_EXIT(zfsvfs); // finished in zfs 178 * return (error); // done, report error 179 */ 180 181/* ARGSUSED */ 182static int 183zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 184{ 185 znode_t *zp = VTOZ(*vpp); 186 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 187 188 ZFS_ENTER(zfsvfs); 189 ZFS_VERIFY_ZP(zp); 190 191 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 192 ((flag & FAPPEND) == 0)) { 193 ZFS_EXIT(zfsvfs); 194 return (SET_ERROR(EPERM)); 195 } 196 197 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 198 ZTOV(zp)->v_type == VREG && 199 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) { 200 if (fs_vscan(*vpp, cr, 0) != 0) { 201 ZFS_EXIT(zfsvfs); 202 return (SET_ERROR(EACCES)); 203 } 204 } 205 206 /* Keep a count of the synchronous opens in the znode */ 207 if (flag & (FSYNC | FDSYNC)) 208 atomic_inc_32(&zp->z_sync_cnt); 209 210 ZFS_EXIT(zfsvfs); 211 return (0); 212} 213 214/* ARGSUSED */ 215static int 216zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 217 caller_context_t *ct) 218{ 219 znode_t *zp = VTOZ(vp); 220 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 221 222 /* 223 * Clean up any locks held by this process on the vp. 224 */ 225 cleanlocks(vp, ddi_get_pid(), 0); 226 cleanshares(vp, ddi_get_pid()); 227 228 ZFS_ENTER(zfsvfs); 229 ZFS_VERIFY_ZP(zp); 230 231 /* Decrement the synchronous opens in the znode */ 232 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 233 atomic_dec_32(&zp->z_sync_cnt); 234 235 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 236 ZTOV(zp)->v_type == VREG && 237 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) 238 VERIFY(fs_vscan(vp, cr, 1) == 0); 239 240 ZFS_EXIT(zfsvfs); 241 return (0); 242} 243 244/* 245 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 246 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 247 */ 248static int 249zfs_holey(vnode_t *vp, u_long cmd, offset_t *off) 250{ 251 znode_t *zp = VTOZ(vp); 252 uint64_t noff = (uint64_t)*off; /* new offset */ 253 uint64_t file_sz; 254 int error; 255 boolean_t hole; 256 257 file_sz = zp->z_size; 258 if (noff >= file_sz) { 259 return (SET_ERROR(ENXIO)); 260 } 261 262 if (cmd == _FIO_SEEK_HOLE) 263 hole = B_TRUE; 264 else 265 hole = B_FALSE; 266 267 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 268 269 if (error == ESRCH) 270 return (SET_ERROR(ENXIO)); 271 272 /* 273 * We could find a hole that begins after the logical end-of-file, 274 * because dmu_offset_next() only works on whole blocks. If the 275 * EOF falls mid-block, then indicate that the "virtual hole" 276 * at the end of the file begins at the logical EOF, rather than 277 * at the end of the last block. 278 */ 279 if (noff > file_sz) { 280 ASSERT(hole); 281 noff = file_sz; 282 } 283 284 if (noff < *off) 285 return (error); 286 *off = noff; 287 return (error); 288} 289 290/* ARGSUSED */ 291static int 292zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred, 293 int *rvalp, caller_context_t *ct) 294{ 295 offset_t off; 296 offset_t ndata; 297 dmu_object_info_t doi; 298 int error; 299 zfsvfs_t *zfsvfs; 300 znode_t *zp; 301 302 switch (com) { 303 case _FIOFFS: 304 { 305 return (0); 306 307 /* 308 * The following two ioctls are used by bfu. Faking out, 309 * necessary to avoid bfu errors. 310 */ 311 } 312 case _FIOGDIO: 313 case _FIOSDIO: 314 { 315 return (0); 316 } 317 318 case _FIO_SEEK_DATA: 319 case _FIO_SEEK_HOLE: 320 { 321#ifdef illumos 322 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 323 return (SET_ERROR(EFAULT)); 324#else 325 off = *(offset_t *)data; 326#endif 327 zp = VTOZ(vp); 328 zfsvfs = zp->z_zfsvfs; 329 ZFS_ENTER(zfsvfs); 330 ZFS_VERIFY_ZP(zp); 331 332 /* offset parameter is in/out */ 333 error = zfs_holey(vp, com, &off); 334 ZFS_EXIT(zfsvfs); 335 if (error) 336 return (error); 337#ifdef illumos 338 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 339 return (SET_ERROR(EFAULT)); 340#else 341 *(offset_t *)data = off; 342#endif 343 return (0); 344 } 345#ifdef illumos 346 case _FIO_COUNT_FILLED: 347 { 348 /* 349 * _FIO_COUNT_FILLED adds a new ioctl command which 350 * exposes the number of filled blocks in a 351 * ZFS object. 352 */ 353 zp = VTOZ(vp); 354 zfsvfs = zp->z_zfsvfs; 355 ZFS_ENTER(zfsvfs); 356 ZFS_VERIFY_ZP(zp); 357 358 /* 359 * Wait for all dirty blocks for this object 360 * to get synced out to disk, and the DMU info 361 * updated. 362 */ 363 error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id); 364 if (error) { 365 ZFS_EXIT(zfsvfs); 366 return (error); 367 } 368 369 /* 370 * Retrieve fill count from DMU object. 371 */ 372 error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi); 373 if (error) { 374 ZFS_EXIT(zfsvfs); 375 return (error); 376 } 377 378 ndata = doi.doi_fill_count; 379 380 ZFS_EXIT(zfsvfs); 381 if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag)) 382 return (SET_ERROR(EFAULT)); 383 return (0); 384 } 385#endif 386 } 387 return (SET_ERROR(ENOTTY)); 388} 389 390static vm_page_t 391page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) 392{ 393 vm_object_t obj; 394 vm_page_t pp; 395 int64_t end; 396 397 /* 398 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE 399 * aligned boundaries, if the range is not aligned. As a result a 400 * DEV_BSIZE subrange with partially dirty data may get marked as clean. 401 * It may happen that all DEV_BSIZE subranges are marked clean and thus 402 * the whole page would be considred clean despite have some dirty data. 403 * For this reason we should shrink the range to DEV_BSIZE aligned 404 * boundaries before calling vm_page_clear_dirty. 405 */ 406 end = rounddown2(off + nbytes, DEV_BSIZE); 407 off = roundup2(off, DEV_BSIZE); 408 nbytes = end - off; 409 410 obj = vp->v_object; 411 zfs_vmobject_assert_wlocked(obj); 412 413 for (;;) { 414 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 415 pp->valid) { 416 if (vm_page_xbusied(pp)) { 417 /* 418 * Reference the page before unlocking and 419 * sleeping so that the page daemon is less 420 * likely to reclaim it. 421 */ 422 vm_page_reference(pp); 423 vm_page_lock(pp); 424 zfs_vmobject_wunlock(obj); 425 vm_page_busy_sleep(pp, "zfsmwb"); 426 zfs_vmobject_wlock(obj); 427 continue; 428 } 429 vm_page_sbusy(pp); 430 } else if (pp == NULL) { 431 pp = vm_page_alloc(obj, OFF_TO_IDX(start), 432 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED | 433 VM_ALLOC_SBUSY); 434 } else { 435 ASSERT(pp != NULL && !pp->valid); 436 pp = NULL; 437 } 438 439 if (pp != NULL) { 440 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 441 vm_object_pip_add(obj, 1); 442 pmap_remove_write(pp); 443 if (nbytes != 0) 444 vm_page_clear_dirty(pp, off, nbytes); 445 } 446 break; 447 } 448 return (pp); 449} 450 451static void 452page_unbusy(vm_page_t pp) 453{ 454 455 vm_page_sunbusy(pp); 456 vm_object_pip_subtract(pp->object, 1); 457} 458 459static vm_page_t 460page_hold(vnode_t *vp, int64_t start) 461{ 462 vm_object_t obj; 463 vm_page_t pp; 464 465 obj = vp->v_object; 466 zfs_vmobject_assert_wlocked(obj); 467 468 for (;;) { 469 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 470 pp->valid) { 471 if (vm_page_xbusied(pp)) { 472 /* 473 * Reference the page before unlocking and 474 * sleeping so that the page daemon is less 475 * likely to reclaim it. 476 */ 477 vm_page_reference(pp); 478 vm_page_lock(pp); 479 zfs_vmobject_wunlock(obj); 480 vm_page_busy_sleep(pp, "zfsmwb"); 481 zfs_vmobject_wlock(obj); 482 continue; 483 } 484 485 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 486 vm_page_lock(pp); 487 vm_page_hold(pp); 488 vm_page_unlock(pp); 489 490 } else 491 pp = NULL; 492 break; 493 } 494 return (pp); 495} 496 497static void 498page_unhold(vm_page_t pp) 499{ 500 501 vm_page_lock(pp); 502 vm_page_unhold(pp); 503 vm_page_unlock(pp); 504} 505 506/* 507 * When a file is memory mapped, we must keep the IO data synchronized 508 * between the DMU cache and the memory mapped pages. What this means: 509 * 510 * On Write: If we find a memory mapped page, we write to *both* 511 * the page and the dmu buffer. 512 */ 513static void 514update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid, 515 int segflg, dmu_tx_t *tx) 516{ 517 vm_object_t obj; 518 struct sf_buf *sf; 519 caddr_t va; 520 int off; 521 522 ASSERT(segflg != UIO_NOCOPY); 523 ASSERT(vp->v_mount != NULL); 524 obj = vp->v_object; 525 ASSERT(obj != NULL); 526 527 off = start & PAGEOFFSET; 528 zfs_vmobject_wlock(obj); 529 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 530 vm_page_t pp; 531 int nbytes = imin(PAGESIZE - off, len); 532 533 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) { 534 zfs_vmobject_wunlock(obj); 535 536 va = zfs_map_page(pp, &sf); 537 (void) dmu_read(os, oid, start+off, nbytes, 538 va+off, DMU_READ_PREFETCH);; 539 zfs_unmap_page(sf); 540 541 zfs_vmobject_wlock(obj); 542 page_unbusy(pp); 543 } 544 len -= nbytes; 545 off = 0; 546 } 547 vm_object_pip_wakeupn(obj, 0); 548 zfs_vmobject_wunlock(obj); 549} 550 551/* 552 * Read with UIO_NOCOPY flag means that sendfile(2) requests 553 * ZFS to populate a range of page cache pages with data. 554 * 555 * NOTE: this function could be optimized to pre-allocate 556 * all pages in advance, drain exclusive busy on all of them, 557 * map them into contiguous KVA region and populate them 558 * in one single dmu_read() call. 559 */ 560static int 561mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio) 562{ 563 znode_t *zp = VTOZ(vp); 564 objset_t *os = zp->z_zfsvfs->z_os; 565 struct sf_buf *sf; 566 vm_object_t obj; 567 vm_page_t pp; 568 int64_t start; 569 caddr_t va; 570 int len = nbytes; 571 int off; 572 int error = 0; 573 574 ASSERT(uio->uio_segflg == UIO_NOCOPY); 575 ASSERT(vp->v_mount != NULL); 576 obj = vp->v_object; 577 ASSERT(obj != NULL); 578 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0); 579 580 zfs_vmobject_wlock(obj); 581 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) { 582 int bytes = MIN(PAGESIZE, len); 583 584 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY | 585 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); 586 if (pp->valid == 0) { 587 zfs_vmobject_wunlock(obj); 588 va = zfs_map_page(pp, &sf); 589 error = dmu_read(os, zp->z_id, start, bytes, va, 590 DMU_READ_PREFETCH); 591 if (bytes != PAGESIZE && error == 0) 592 bzero(va + bytes, PAGESIZE - bytes); 593 zfs_unmap_page(sf); 594 zfs_vmobject_wlock(obj); 595 vm_page_sunbusy(pp); 596 vm_page_lock(pp); 597 if (error) { 598 if (pp->wire_count == 0 && pp->valid == 0 && 599 !vm_page_busied(pp)) 600 vm_page_free(pp); 601 } else { 602 pp->valid = VM_PAGE_BITS_ALL; 603 vm_page_activate(pp); 604 } 605 vm_page_unlock(pp); 606 } else { 607 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 608 vm_page_sunbusy(pp); 609 } 610 if (error) 611 break; 612 uio->uio_resid -= bytes; 613 uio->uio_offset += bytes; 614 len -= bytes; 615 } 616 zfs_vmobject_wunlock(obj); 617 return (error); 618} 619 620/* 621 * When a file is memory mapped, we must keep the IO data synchronized 622 * between the DMU cache and the memory mapped pages. What this means: 623 * 624 * On Read: We "read" preferentially from memory mapped pages, 625 * else we default from the dmu buffer. 626 * 627 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 628 * the file is memory mapped. 629 */ 630static int 631mappedread(vnode_t *vp, int nbytes, uio_t *uio) 632{ 633 znode_t *zp = VTOZ(vp); 634 vm_object_t obj; 635 int64_t start; 636 caddr_t va; 637 int len = nbytes; 638 int off; 639 int error = 0; 640 641 ASSERT(vp->v_mount != NULL); 642 obj = vp->v_object; 643 ASSERT(obj != NULL); 644 645 start = uio->uio_loffset; 646 off = start & PAGEOFFSET; 647 zfs_vmobject_wlock(obj); 648 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 649 vm_page_t pp; 650 uint64_t bytes = MIN(PAGESIZE - off, len); 651 652 if (pp = page_hold(vp, start)) { 653 struct sf_buf *sf; 654 caddr_t va; 655 656 zfs_vmobject_wunlock(obj); 657 va = zfs_map_page(pp, &sf); 658 error = uiomove(va + off, bytes, UIO_READ, uio); 659 zfs_unmap_page(sf); 660 zfs_vmobject_wlock(obj); 661 page_unhold(pp); 662 } else { 663 zfs_vmobject_wunlock(obj); 664 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 665 uio, bytes); 666 zfs_vmobject_wlock(obj); 667 } 668 len -= bytes; 669 off = 0; 670 if (error) 671 break; 672 } 673 zfs_vmobject_wunlock(obj); 674 return (error); 675} 676 677offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 678 679/* 680 * Read bytes from specified file into supplied buffer. 681 * 682 * IN: vp - vnode of file to be read from. 683 * uio - structure supplying read location, range info, 684 * and return buffer. 685 * ioflag - SYNC flags; used to provide FRSYNC semantics. 686 * cr - credentials of caller. 687 * ct - caller context 688 * 689 * OUT: uio - updated offset and range, buffer filled. 690 * 691 * RETURN: 0 on success, error code on failure. 692 * 693 * Side Effects: 694 * vp - atime updated if byte count > 0 695 */ 696/* ARGSUSED */ 697static int 698zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 699{ 700 znode_t *zp = VTOZ(vp); 701 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 702 ssize_t n, nbytes; 703 int error = 0; 704 rl_t *rl; 705 xuio_t *xuio = NULL; 706 707 ZFS_ENTER(zfsvfs); 708 ZFS_VERIFY_ZP(zp); 709 710 if (zp->z_pflags & ZFS_AV_QUARANTINED) { 711 ZFS_EXIT(zfsvfs); 712 return (SET_ERROR(EACCES)); 713 } 714 715 /* 716 * Validate file offset 717 */ 718 if (uio->uio_loffset < (offset_t)0) { 719 ZFS_EXIT(zfsvfs); 720 return (SET_ERROR(EINVAL)); 721 } 722 723 /* 724 * Fasttrack empty reads 725 */ 726 if (uio->uio_resid == 0) { 727 ZFS_EXIT(zfsvfs); 728 return (0); 729 } 730 731 /* 732 * Check for mandatory locks 733 */ 734 if (MANDMODE(zp->z_mode)) { 735 if (error = chklock(vp, FREAD, 736 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 737 ZFS_EXIT(zfsvfs); 738 return (error); 739 } 740 } 741 742 /* 743 * If we're in FRSYNC mode, sync out this znode before reading it. 744 */ 745 if (zfsvfs->z_log && 746 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)) 747 zil_commit(zfsvfs->z_log, zp->z_id); 748 749 /* 750 * Lock the range against changes. 751 */ 752 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 753 754 /* 755 * If we are reading past end-of-file we can skip 756 * to the end; but we might still need to set atime. 757 */ 758 if (uio->uio_loffset >= zp->z_size) { 759 error = 0; 760 goto out; 761 } 762 763 ASSERT(uio->uio_loffset < zp->z_size); 764 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset); 765 766#ifdef illumos 767 if ((uio->uio_extflg == UIO_XUIO) && 768 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) { 769 int nblk; 770 int blksz = zp->z_blksz; 771 uint64_t offset = uio->uio_loffset; 772 773 xuio = (xuio_t *)uio; 774 if ((ISP2(blksz))) { 775 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset, 776 blksz)) / blksz; 777 } else { 778 ASSERT(offset + n <= blksz); 779 nblk = 1; 780 } 781 (void) dmu_xuio_init(xuio, nblk); 782 783 if (vn_has_cached_data(vp)) { 784 /* 785 * For simplicity, we always allocate a full buffer 786 * even if we only expect to read a portion of a block. 787 */ 788 while (--nblk >= 0) { 789 (void) dmu_xuio_add(xuio, 790 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 791 blksz), 0, blksz); 792 } 793 } 794 } 795#endif /* illumos */ 796 797 while (n > 0) { 798 nbytes = MIN(n, zfs_read_chunk_size - 799 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 800 801#ifdef __FreeBSD__ 802 if (uio->uio_segflg == UIO_NOCOPY) 803 error = mappedread_sf(vp, nbytes, uio); 804 else 805#endif /* __FreeBSD__ */ 806 if (vn_has_cached_data(vp)) { 807 error = mappedread(vp, nbytes, uio); 808 } else { 809 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 810 uio, nbytes); 811 } 812 if (error) { 813 /* convert checksum errors into IO errors */ 814 if (error == ECKSUM) 815 error = SET_ERROR(EIO); 816 break; 817 } 818 819 n -= nbytes; 820 } 821out: 822 zfs_range_unlock(rl); 823 824 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 825 ZFS_EXIT(zfsvfs); 826 return (error); 827} 828 829/* 830 * Write the bytes to a file. 831 * 832 * IN: vp - vnode of file to be written to. 833 * uio - structure supplying write location, range info, 834 * and data buffer. 835 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is 836 * set if in append mode. 837 * cr - credentials of caller. 838 * ct - caller context (NFS/CIFS fem monitor only) 839 * 840 * OUT: uio - updated offset and range. 841 * 842 * RETURN: 0 on success, error code on failure. 843 * 844 * Timestamps: 845 * vp - ctime|mtime updated if byte count > 0 846 */ 847 848/* ARGSUSED */ 849static int 850zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 851{ 852 znode_t *zp = VTOZ(vp); 853 rlim64_t limit = MAXOFFSET_T; 854 ssize_t start_resid = uio->uio_resid; 855 ssize_t tx_bytes; 856 uint64_t end_size; 857 dmu_tx_t *tx; 858 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 859 zilog_t *zilog; 860 offset_t woff; 861 ssize_t n, nbytes; 862 rl_t *rl; 863 int max_blksz = zfsvfs->z_max_blksz; 864 int error = 0; 865 arc_buf_t *abuf; 866 iovec_t *aiov = NULL; 867 xuio_t *xuio = NULL; 868 int i_iov = 0; 869 int iovcnt = uio->uio_iovcnt; 870 iovec_t *iovp = uio->uio_iov; 871 int write_eof; 872 int count = 0; 873 sa_bulk_attr_t bulk[4]; 874 uint64_t mtime[2], ctime[2]; 875 876 /* 877 * Fasttrack empty write 878 */ 879 n = start_resid; 880 if (n == 0) 881 return (0); 882 883 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 884 limit = MAXOFFSET_T; 885 886 ZFS_ENTER(zfsvfs); 887 ZFS_VERIFY_ZP(zp); 888 889 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 890 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 891 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 892 &zp->z_size, 8); 893 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 894 &zp->z_pflags, 8); 895 896 /* 897 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our 898 * callers might not be able to detect properly that we are read-only, 899 * so check it explicitly here. 900 */ 901 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 902 ZFS_EXIT(zfsvfs); 903 return (SET_ERROR(EROFS)); 904 } 905 906 /* 907 * If immutable or not appending then return EPERM 908 */ 909 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 910 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 911 (uio->uio_loffset < zp->z_size))) { 912 ZFS_EXIT(zfsvfs); 913 return (SET_ERROR(EPERM)); 914 } 915 916 zilog = zfsvfs->z_log; 917 918 /* 919 * Validate file offset 920 */ 921 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset; 922 if (woff < 0) { 923 ZFS_EXIT(zfsvfs); 924 return (SET_ERROR(EINVAL)); 925 } 926 927 /* 928 * Check for mandatory locks before calling zfs_range_lock() 929 * in order to prevent a deadlock with locks set via fcntl(). 930 */ 931 if (MANDMODE((mode_t)zp->z_mode) && 932 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 933 ZFS_EXIT(zfsvfs); 934 return (error); 935 } 936 937#ifdef illumos 938 /* 939 * Pre-fault the pages to ensure slow (eg NFS) pages 940 * don't hold up txg. 941 * Skip this if uio contains loaned arc_buf. 942 */ 943 if ((uio->uio_extflg == UIO_XUIO) && 944 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) 945 xuio = (xuio_t *)uio; 946 else 947 uio_prefaultpages(MIN(n, max_blksz), uio); 948#endif 949 950 /* 951 * If in append mode, set the io offset pointer to eof. 952 */ 953 if (ioflag & FAPPEND) { 954 /* 955 * Obtain an appending range lock to guarantee file append 956 * semantics. We reset the write offset once we have the lock. 957 */ 958 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 959 woff = rl->r_off; 960 if (rl->r_len == UINT64_MAX) { 961 /* 962 * We overlocked the file because this write will cause 963 * the file block size to increase. 964 * Note that zp_size cannot change with this lock held. 965 */ 966 woff = zp->z_size; 967 } 968 uio->uio_loffset = woff; 969 } else { 970 /* 971 * Note that if the file block size will change as a result of 972 * this write, then this range lock will lock the entire file 973 * so that we can re-write the block safely. 974 */ 975 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 976 } 977 978 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) { 979 zfs_range_unlock(rl); 980 ZFS_EXIT(zfsvfs); 981 return (EFBIG); 982 } 983 984 if (woff >= limit) { 985 zfs_range_unlock(rl); 986 ZFS_EXIT(zfsvfs); 987 return (SET_ERROR(EFBIG)); 988 } 989 990 if ((woff + n) > limit || woff > (limit - n)) 991 n = limit - woff; 992 993 /* Will this write extend the file length? */ 994 write_eof = (woff + n > zp->z_size); 995 996 end_size = MAX(zp->z_size, woff + n); 997 998 /* 999 * Write the file in reasonable size chunks. Each chunk is written 1000 * in a separate transaction; this keeps the intent log records small 1001 * and allows us to do more fine-grained space accounting. 1002 */ 1003 while (n > 0) { 1004 abuf = NULL; 1005 woff = uio->uio_loffset; 1006 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 1007 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 1008 if (abuf != NULL) 1009 dmu_return_arcbuf(abuf); 1010 error = SET_ERROR(EDQUOT); 1011 break; 1012 } 1013 1014 if (xuio && abuf == NULL) { 1015 ASSERT(i_iov < iovcnt); 1016 aiov = &iovp[i_iov]; 1017 abuf = dmu_xuio_arcbuf(xuio, i_iov); 1018 dmu_xuio_clear(xuio, i_iov); 1019 DTRACE_PROBE3(zfs_cp_write, int, i_iov, 1020 iovec_t *, aiov, arc_buf_t *, abuf); 1021 ASSERT((aiov->iov_base == abuf->b_data) || 1022 ((char *)aiov->iov_base - (char *)abuf->b_data + 1023 aiov->iov_len == arc_buf_size(abuf))); 1024 i_iov++; 1025 } else if (abuf == NULL && n >= max_blksz && 1026 woff >= zp->z_size && 1027 P2PHASE(woff, max_blksz) == 0 && 1028 zp->z_blksz == max_blksz) { 1029 /* 1030 * This write covers a full block. "Borrow" a buffer 1031 * from the dmu so that we can fill it before we enter 1032 * a transaction. This avoids the possibility of 1033 * holding up the transaction if the data copy hangs 1034 * up on a pagefault (e.g., from an NFS server mapping). 1035 */ 1036 size_t cbytes; 1037 1038 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 1039 max_blksz); 1040 ASSERT(abuf != NULL); 1041 ASSERT(arc_buf_size(abuf) == max_blksz); 1042 if (error = uiocopy(abuf->b_data, max_blksz, 1043 UIO_WRITE, uio, &cbytes)) { 1044 dmu_return_arcbuf(abuf); 1045 break; 1046 } 1047 ASSERT(cbytes == max_blksz); 1048 } 1049 1050 /* 1051 * Start a transaction. 1052 */ 1053 tx = dmu_tx_create(zfsvfs->z_os); 1054 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1055 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 1056 zfs_sa_upgrade_txholds(tx, zp); 1057 error = dmu_tx_assign(tx, TXG_WAIT); 1058 if (error) { 1059 dmu_tx_abort(tx); 1060 if (abuf != NULL) 1061 dmu_return_arcbuf(abuf); 1062 break; 1063 } 1064 1065 /* 1066 * If zfs_range_lock() over-locked we grow the blocksize 1067 * and then reduce the lock range. This will only happen 1068 * on the first iteration since zfs_range_reduce() will 1069 * shrink down r_len to the appropriate size. 1070 */ 1071 if (rl->r_len == UINT64_MAX) { 1072 uint64_t new_blksz; 1073 1074 if (zp->z_blksz > max_blksz) { 1075 /* 1076 * File's blocksize is already larger than the 1077 * "recordsize" property. Only let it grow to 1078 * the next power of 2. 1079 */ 1080 ASSERT(!ISP2(zp->z_blksz)); 1081 new_blksz = MIN(end_size, 1082 1 << highbit64(zp->z_blksz)); 1083 } else { 1084 new_blksz = MIN(end_size, max_blksz); 1085 } 1086 zfs_grow_blocksize(zp, new_blksz, tx); 1087 zfs_range_reduce(rl, woff, n); 1088 } 1089 1090 /* 1091 * XXX - should we really limit each write to z_max_blksz? 1092 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 1093 */ 1094 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 1095 1096 if (woff + nbytes > zp->z_size) 1097 vnode_pager_setsize(vp, woff + nbytes); 1098 1099 if (abuf == NULL) { 1100 tx_bytes = uio->uio_resid; 1101 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), 1102 uio, nbytes, tx); 1103 tx_bytes -= uio->uio_resid; 1104 } else { 1105 tx_bytes = nbytes; 1106 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len); 1107 /* 1108 * If this is not a full block write, but we are 1109 * extending the file past EOF and this data starts 1110 * block-aligned, use assign_arcbuf(). Otherwise, 1111 * write via dmu_write(). 1112 */ 1113 if (tx_bytes < max_blksz && (!write_eof || 1114 aiov->iov_base != abuf->b_data)) { 1115 ASSERT(xuio); 1116 dmu_write(zfsvfs->z_os, zp->z_id, woff, 1117 aiov->iov_len, aiov->iov_base, tx); 1118 dmu_return_arcbuf(abuf); 1119 xuio_stat_wbuf_copied(); 1120 } else { 1121 ASSERT(xuio || tx_bytes == max_blksz); 1122 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), 1123 woff, abuf, tx); 1124 } 1125 ASSERT(tx_bytes <= uio->uio_resid); 1126 uioskip(uio, tx_bytes); 1127 } 1128 if (tx_bytes && vn_has_cached_data(vp)) { 1129 update_pages(vp, woff, tx_bytes, zfsvfs->z_os, 1130 zp->z_id, uio->uio_segflg, tx); 1131 } 1132 1133 /* 1134 * If we made no progress, we're done. If we made even 1135 * partial progress, update the znode and ZIL accordingly. 1136 */ 1137 if (tx_bytes == 0) { 1138 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 1139 (void *)&zp->z_size, sizeof (uint64_t), tx); 1140 dmu_tx_commit(tx); 1141 ASSERT(error != 0); 1142 break; 1143 } 1144 1145 /* 1146 * Clear Set-UID/Set-GID bits on successful write if not 1147 * privileged and at least one of the excute bits is set. 1148 * 1149 * It would be nice to to this after all writes have 1150 * been done, but that would still expose the ISUID/ISGID 1151 * to another app after the partial write is committed. 1152 * 1153 * Note: we don't call zfs_fuid_map_id() here because 1154 * user 0 is not an ephemeral uid. 1155 */ 1156 mutex_enter(&zp->z_acl_lock); 1157 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | 1158 (S_IXUSR >> 6))) != 0 && 1159 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && 1160 secpolicy_vnode_setid_retain(vp, cr, 1161 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) { 1162 uint64_t newmode; 1163 zp->z_mode &= ~(S_ISUID | S_ISGID); 1164 newmode = zp->z_mode; 1165 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), 1166 (void *)&newmode, sizeof (uint64_t), tx); 1167 } 1168 mutex_exit(&zp->z_acl_lock); 1169 1170 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 1171 B_TRUE); 1172 1173 /* 1174 * Update the file size (zp_size) if it has changed; 1175 * account for possible concurrent updates. 1176 */ 1177 while ((end_size = zp->z_size) < uio->uio_loffset) { 1178 (void) atomic_cas_64(&zp->z_size, end_size, 1179 uio->uio_loffset); 1180 ASSERT(error == 0); 1181 } 1182 /* 1183 * If we are replaying and eof is non zero then force 1184 * the file size to the specified eof. Note, there's no 1185 * concurrency during replay. 1186 */ 1187 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) 1188 zp->z_size = zfsvfs->z_replay_eof; 1189 1190 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1191 1192 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 1193 dmu_tx_commit(tx); 1194 1195 if (error != 0) 1196 break; 1197 ASSERT(tx_bytes == nbytes); 1198 n -= nbytes; 1199 1200#ifdef illumos 1201 if (!xuio && n > 0) 1202 uio_prefaultpages(MIN(n, max_blksz), uio); 1203#endif 1204 } 1205 1206 zfs_range_unlock(rl); 1207 1208 /* 1209 * If we're in replay mode, or we made no progress, return error. 1210 * Otherwise, it's at least a partial write, so it's successful. 1211 */ 1212 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 1213 ZFS_EXIT(zfsvfs); 1214 return (error); 1215 } 1216 1217 if (ioflag & (FSYNC | FDSYNC) || 1218 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1219 zil_commit(zilog, zp->z_id); 1220 1221 ZFS_EXIT(zfsvfs); 1222 return (0); 1223} 1224 1225void 1226zfs_get_done(zgd_t *zgd, int error) 1227{ 1228 znode_t *zp = zgd->zgd_private; 1229 objset_t *os = zp->z_zfsvfs->z_os; 1230 1231 if (zgd->zgd_db) 1232 dmu_buf_rele(zgd->zgd_db, zgd); 1233 1234 zfs_range_unlock(zgd->zgd_rl); 1235 1236 /* 1237 * Release the vnode asynchronously as we currently have the 1238 * txg stopped from syncing. 1239 */ 1240 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1241 1242 if (error == 0 && zgd->zgd_bp) 1243 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1244 1245 kmem_free(zgd, sizeof (zgd_t)); 1246} 1247 1248#ifdef DEBUG 1249static int zil_fault_io = 0; 1250#endif 1251 1252/* 1253 * Get data to generate a TX_WRITE intent log record. 1254 */ 1255int 1256zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1257{ 1258 zfsvfs_t *zfsvfs = arg; 1259 objset_t *os = zfsvfs->z_os; 1260 znode_t *zp; 1261 uint64_t object = lr->lr_foid; 1262 uint64_t offset = lr->lr_offset; 1263 uint64_t size = lr->lr_length; 1264 blkptr_t *bp = &lr->lr_blkptr; 1265 dmu_buf_t *db; 1266 zgd_t *zgd; 1267 int error = 0; 1268 1269 ASSERT(zio != NULL); 1270 ASSERT(size != 0); 1271 1272 /* 1273 * Nothing to do if the file has been removed 1274 */ 1275 if (zfs_zget(zfsvfs, object, &zp) != 0) 1276 return (SET_ERROR(ENOENT)); 1277 if (zp->z_unlinked) { 1278 /* 1279 * Release the vnode asynchronously as we currently have the 1280 * txg stopped from syncing. 1281 */ 1282 VN_RELE_ASYNC(ZTOV(zp), 1283 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1284 return (SET_ERROR(ENOENT)); 1285 } 1286 1287 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1288 zgd->zgd_zilog = zfsvfs->z_log; 1289 zgd->zgd_private = zp; 1290 1291 /* 1292 * Write records come in two flavors: immediate and indirect. 1293 * For small writes it's cheaper to store the data with the 1294 * log record (immediate); for large writes it's cheaper to 1295 * sync the data and get a pointer to it (indirect) so that 1296 * we don't have to write the data twice. 1297 */ 1298 if (buf != NULL) { /* immediate write */ 1299 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER); 1300 /* test for truncation needs to be done while range locked */ 1301 if (offset >= zp->z_size) { 1302 error = SET_ERROR(ENOENT); 1303 } else { 1304 error = dmu_read(os, object, offset, size, buf, 1305 DMU_READ_NO_PREFETCH); 1306 } 1307 ASSERT(error == 0 || error == ENOENT); 1308 } else { /* indirect write */ 1309 /* 1310 * Have to lock the whole block to ensure when it's 1311 * written out and it's checksum is being calculated 1312 * that no one can change the data. We need to re-check 1313 * blocksize after we get the lock in case it's changed! 1314 */ 1315 for (;;) { 1316 uint64_t blkoff; 1317 size = zp->z_blksz; 1318 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; 1319 offset -= blkoff; 1320 zgd->zgd_rl = zfs_range_lock(zp, offset, size, 1321 RL_READER); 1322 if (zp->z_blksz == size) 1323 break; 1324 offset += blkoff; 1325 zfs_range_unlock(zgd->zgd_rl); 1326 } 1327 /* test for truncation needs to be done while range locked */ 1328 if (lr->lr_offset >= zp->z_size) 1329 error = SET_ERROR(ENOENT); 1330#ifdef DEBUG 1331 if (zil_fault_io) { 1332 error = SET_ERROR(EIO); 1333 zil_fault_io = 0; 1334 } 1335#endif 1336 if (error == 0) 1337 error = dmu_buf_hold(os, object, offset, zgd, &db, 1338 DMU_READ_NO_PREFETCH); 1339 1340 if (error == 0) { 1341 blkptr_t *obp = dmu_buf_get_blkptr(db); 1342 if (obp) { 1343 ASSERT(BP_IS_HOLE(bp)); 1344 *bp = *obp; 1345 } 1346 1347 zgd->zgd_db = db; 1348 zgd->zgd_bp = bp; 1349 1350 ASSERT(db->db_offset == offset); 1351 ASSERT(db->db_size == size); 1352 1353 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1354 zfs_get_done, zgd); 1355 ASSERT(error || lr->lr_length <= zp->z_blksz); 1356 1357 /* 1358 * On success, we need to wait for the write I/O 1359 * initiated by dmu_sync() to complete before we can 1360 * release this dbuf. We will finish everything up 1361 * in the zfs_get_done() callback. 1362 */ 1363 if (error == 0) 1364 return (0); 1365 1366 if (error == EALREADY) { 1367 lr->lr_common.lrc_txtype = TX_WRITE2; 1368 error = 0; 1369 } 1370 } 1371 } 1372 1373 zfs_get_done(zgd, error); 1374 1375 return (error); 1376} 1377 1378/*ARGSUSED*/ 1379static int 1380zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 1381 caller_context_t *ct) 1382{ 1383 znode_t *zp = VTOZ(vp); 1384 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1385 int error; 1386 1387 ZFS_ENTER(zfsvfs); 1388 ZFS_VERIFY_ZP(zp); 1389 1390 if (flag & V_ACE_MASK) 1391 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 1392 else 1393 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1394 1395 ZFS_EXIT(zfsvfs); 1396 return (error); 1397} 1398 1399/* 1400 * If vnode is for a device return a specfs vnode instead. 1401 */ 1402static int 1403specvp_check(vnode_t **vpp, cred_t *cr) 1404{ 1405 int error = 0; 1406 1407 if (IS_DEVVP(*vpp)) { 1408 struct vnode *svp; 1409 1410 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1411 VN_RELE(*vpp); 1412 if (svp == NULL) 1413 error = SET_ERROR(ENOSYS); 1414 *vpp = svp; 1415 } 1416 return (error); 1417} 1418 1419 1420/* 1421 * Lookup an entry in a directory, or an extended attribute directory. 1422 * If it exists, return a held vnode reference for it. 1423 * 1424 * IN: dvp - vnode of directory to search. 1425 * nm - name of entry to lookup. 1426 * pnp - full pathname to lookup [UNUSED]. 1427 * flags - LOOKUP_XATTR set if looking for an attribute. 1428 * rdir - root directory vnode [UNUSED]. 1429 * cr - credentials of caller. 1430 * ct - caller context 1431 * direntflags - directory lookup flags 1432 * realpnp - returned pathname. 1433 * 1434 * OUT: vpp - vnode of located entry, NULL if not found. 1435 * 1436 * RETURN: 0 on success, error code on failure. 1437 * 1438 * Timestamps: 1439 * NA 1440 */ 1441/* ARGSUSED */ 1442static int 1443zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 1444 int nameiop, cred_t *cr, kthread_t *td, int flags) 1445{ 1446 znode_t *zdp = VTOZ(dvp); 1447 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1448 int error = 0; 1449 int *direntflags = NULL; 1450 void *realpnp = NULL; 1451 1452 /* fast path */ 1453 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 1454 1455 if (dvp->v_type != VDIR) { 1456 return (SET_ERROR(ENOTDIR)); 1457 } else if (zdp->z_sa_hdl == NULL) { 1458 return (SET_ERROR(EIO)); 1459 } 1460 1461 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 1462 error = zfs_fastaccesschk_execute(zdp, cr); 1463 if (!error) { 1464 *vpp = dvp; 1465 VN_HOLD(*vpp); 1466 return (0); 1467 } 1468 return (error); 1469 } else { 1470 vnode_t *tvp = dnlc_lookup(dvp, nm); 1471 1472 if (tvp) { 1473 error = zfs_fastaccesschk_execute(zdp, cr); 1474 if (error) { 1475 VN_RELE(tvp); 1476 return (error); 1477 } 1478 if (tvp == DNLC_NO_VNODE) { 1479 VN_RELE(tvp); 1480 return (SET_ERROR(ENOENT)); 1481 } else { 1482 *vpp = tvp; 1483 return (specvp_check(vpp, cr)); 1484 } 1485 } 1486 } 1487 } 1488 1489 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1490 1491 ZFS_ENTER(zfsvfs); 1492 ZFS_VERIFY_ZP(zdp); 1493 1494 *vpp = NULL; 1495 1496 if (flags & LOOKUP_XATTR) { 1497#ifdef TODO 1498 /* 1499 * If the xattr property is off, refuse the lookup request. 1500 */ 1501 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1502 ZFS_EXIT(zfsvfs); 1503 return (SET_ERROR(EINVAL)); 1504 } 1505#endif 1506 1507 /* 1508 * We don't allow recursive attributes.. 1509 * Maybe someday we will. 1510 */ 1511 if (zdp->z_pflags & ZFS_XATTR) { 1512 ZFS_EXIT(zfsvfs); 1513 return (SET_ERROR(EINVAL)); 1514 } 1515 1516 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1517 ZFS_EXIT(zfsvfs); 1518 return (error); 1519 } 1520 1521 /* 1522 * Do we have permission to get into attribute directory? 1523 */ 1524 1525 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1526 B_FALSE, cr)) { 1527 VN_RELE(*vpp); 1528 *vpp = NULL; 1529 } 1530 1531 ZFS_EXIT(zfsvfs); 1532 return (error); 1533 } 1534 1535 if (dvp->v_type != VDIR) { 1536 ZFS_EXIT(zfsvfs); 1537 return (SET_ERROR(ENOTDIR)); 1538 } 1539 1540 /* 1541 * Check accessibility of directory. 1542 */ 1543 1544 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1545 ZFS_EXIT(zfsvfs); 1546 return (error); 1547 } 1548 1549 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1550 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1551 ZFS_EXIT(zfsvfs); 1552 return (SET_ERROR(EILSEQ)); 1553 } 1554 1555 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1556 if (error == 0) 1557 error = specvp_check(vpp, cr); 1558 1559 /* Translate errors and add SAVENAME when needed. */ 1560 if (cnp->cn_flags & ISLASTCN) { 1561 switch (nameiop) { 1562 case CREATE: 1563 case RENAME: 1564 if (error == ENOENT) { 1565 error = EJUSTRETURN; 1566 cnp->cn_flags |= SAVENAME; 1567 break; 1568 } 1569 /* FALLTHROUGH */ 1570 case DELETE: 1571 if (error == 0) 1572 cnp->cn_flags |= SAVENAME; 1573 break; 1574 } 1575 } 1576 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) { 1577 int ltype = 0; 1578 1579 if (cnp->cn_flags & ISDOTDOT) { 1580 ltype = VOP_ISLOCKED(dvp); 1581 VOP_UNLOCK(dvp, 0); 1582 } 1583 ZFS_EXIT(zfsvfs); 1584 error = vn_lock(*vpp, cnp->cn_lkflags); 1585 if (cnp->cn_flags & ISDOTDOT) 1586 vn_lock(dvp, ltype | LK_RETRY); 1587 if (error != 0) { 1588 VN_RELE(*vpp); 1589 *vpp = NULL; 1590 return (error); 1591 } 1592 } else { 1593 ZFS_EXIT(zfsvfs); 1594 } 1595 1596#ifdef FREEBSD_NAMECACHE 1597 /* 1598 * Insert name into cache (as non-existent) if appropriate. 1599 */ 1600 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0) 1601 cache_enter(dvp, *vpp, cnp); 1602 /* 1603 * Insert name into cache if appropriate. 1604 */ 1605 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1606 if (!(cnp->cn_flags & ISLASTCN) || 1607 (nameiop != DELETE && nameiop != RENAME)) { 1608 cache_enter(dvp, *vpp, cnp); 1609 } 1610 } 1611#endif 1612 1613 return (error); 1614} 1615 1616/* 1617 * Attempt to create a new entry in a directory. If the entry 1618 * already exists, truncate the file if permissible, else return 1619 * an error. Return the vp of the created or trunc'd file. 1620 * 1621 * IN: dvp - vnode of directory to put new file entry in. 1622 * name - name of new file entry. 1623 * vap - attributes of new file. 1624 * excl - flag indicating exclusive or non-exclusive mode. 1625 * mode - mode to open file with. 1626 * cr - credentials of caller. 1627 * flag - large file flag [UNUSED]. 1628 * ct - caller context 1629 * vsecp - ACL to be set 1630 * 1631 * OUT: vpp - vnode of created or trunc'd entry. 1632 * 1633 * RETURN: 0 on success, error code on failure. 1634 * 1635 * Timestamps: 1636 * dvp - ctime|mtime updated if new entry created 1637 * vp - ctime|mtime always, atime if new 1638 */ 1639 1640/* ARGSUSED */ 1641static int 1642zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1643 vnode_t **vpp, cred_t *cr, kthread_t *td) 1644{ 1645 znode_t *zp, *dzp = VTOZ(dvp); 1646 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1647 zilog_t *zilog; 1648 objset_t *os; 1649 zfs_dirlock_t *dl; 1650 dmu_tx_t *tx; 1651 int error; 1652 ksid_t *ksid; 1653 uid_t uid; 1654 gid_t gid = crgetgid(cr); 1655 zfs_acl_ids_t acl_ids; 1656 boolean_t fuid_dirtied; 1657 boolean_t have_acl = B_FALSE; 1658 boolean_t waited = B_FALSE; 1659 void *vsecp = NULL; 1660 int flag = 0; 1661 1662 /* 1663 * If we have an ephemeral id, ACL, or XVATTR then 1664 * make sure file system is at proper version 1665 */ 1666 1667 ksid = crgetsid(cr, KSID_OWNER); 1668 if (ksid) 1669 uid = ksid_getid(ksid); 1670 else 1671 uid = crgetuid(cr); 1672 1673 if (zfsvfs->z_use_fuids == B_FALSE && 1674 (vsecp || (vap->va_mask & AT_XVATTR) || 1675 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1676 return (SET_ERROR(EINVAL)); 1677 1678 ZFS_ENTER(zfsvfs); 1679 ZFS_VERIFY_ZP(dzp); 1680 os = zfsvfs->z_os; 1681 zilog = zfsvfs->z_log; 1682 1683 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1684 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1685 ZFS_EXIT(zfsvfs); 1686 return (SET_ERROR(EILSEQ)); 1687 } 1688 1689 if (vap->va_mask & AT_XVATTR) { 1690 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 1691 crgetuid(cr), cr, vap->va_type)) != 0) { 1692 ZFS_EXIT(zfsvfs); 1693 return (error); 1694 } 1695 } 1696 1697 getnewvnode_reserve(1); 1698 1699top: 1700 *vpp = NULL; 1701 1702 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) 1703 vap->va_mode &= ~S_ISVTX; 1704 1705 if (*name == '\0') { 1706 /* 1707 * Null component name refers to the directory itself. 1708 */ 1709 VN_HOLD(dvp); 1710 zp = dzp; 1711 dl = NULL; 1712 error = 0; 1713 } else { 1714 /* possible VN_HOLD(zp) */ 1715 int zflg = 0; 1716 1717 if (flag & FIGNORECASE) 1718 zflg |= ZCILOOK; 1719 1720 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1721 NULL, NULL); 1722 if (error) { 1723 if (have_acl) 1724 zfs_acl_ids_free(&acl_ids); 1725 if (strcmp(name, "..") == 0) 1726 error = SET_ERROR(EISDIR); 1727 getnewvnode_drop_reserve(); 1728 ZFS_EXIT(zfsvfs); 1729 return (error); 1730 } 1731 } 1732 1733 if (zp == NULL) { 1734 uint64_t txtype; 1735 1736 /* 1737 * Create a new file object and update the directory 1738 * to reference it. 1739 */ 1740 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1741 if (have_acl) 1742 zfs_acl_ids_free(&acl_ids); 1743 goto out; 1744 } 1745 1746 /* 1747 * We only support the creation of regular files in 1748 * extended attribute directories. 1749 */ 1750 1751 if ((dzp->z_pflags & ZFS_XATTR) && 1752 (vap->va_type != VREG)) { 1753 if (have_acl) 1754 zfs_acl_ids_free(&acl_ids); 1755 error = SET_ERROR(EINVAL); 1756 goto out; 1757 } 1758 1759 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 1760 cr, vsecp, &acl_ids)) != 0) 1761 goto out; 1762 have_acl = B_TRUE; 1763 1764 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1765 zfs_acl_ids_free(&acl_ids); 1766 error = SET_ERROR(EDQUOT); 1767 goto out; 1768 } 1769 1770 tx = dmu_tx_create(os); 1771 1772 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1773 ZFS_SA_BASE_ATTR_SIZE); 1774 1775 fuid_dirtied = zfsvfs->z_fuid_dirty; 1776 if (fuid_dirtied) 1777 zfs_fuid_txhold(zfsvfs, tx); 1778 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1779 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1780 if (!zfsvfs->z_use_sa && 1781 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1782 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1783 0, acl_ids.z_aclp->z_acl_bytes); 1784 } 1785 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 1786 if (error) { 1787 zfs_dirent_unlock(dl); 1788 if (error == ERESTART) { 1789 waited = B_TRUE; 1790 dmu_tx_wait(tx); 1791 dmu_tx_abort(tx); 1792 goto top; 1793 } 1794 zfs_acl_ids_free(&acl_ids); 1795 dmu_tx_abort(tx); 1796 getnewvnode_drop_reserve(); 1797 ZFS_EXIT(zfsvfs); 1798 return (error); 1799 } 1800 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1801 1802 if (fuid_dirtied) 1803 zfs_fuid_sync(zfsvfs, tx); 1804 1805 (void) zfs_link_create(dl, zp, tx, ZNEW); 1806 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1807 if (flag & FIGNORECASE) 1808 txtype |= TX_CI; 1809 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1810 vsecp, acl_ids.z_fuidp, vap); 1811 zfs_acl_ids_free(&acl_ids); 1812 dmu_tx_commit(tx); 1813 } else { 1814 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1815 1816 if (have_acl) 1817 zfs_acl_ids_free(&acl_ids); 1818 have_acl = B_FALSE; 1819 1820 /* 1821 * A directory entry already exists for this name. 1822 */ 1823 /* 1824 * Can't truncate an existing file if in exclusive mode. 1825 */ 1826 if (excl == EXCL) { 1827 error = SET_ERROR(EEXIST); 1828 goto out; 1829 } 1830 /* 1831 * Can't open a directory for writing. 1832 */ 1833 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1834 error = SET_ERROR(EISDIR); 1835 goto out; 1836 } 1837 /* 1838 * Verify requested access to file. 1839 */ 1840 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1841 goto out; 1842 } 1843 1844 mutex_enter(&dzp->z_lock); 1845 dzp->z_seq++; 1846 mutex_exit(&dzp->z_lock); 1847 1848 /* 1849 * Truncate regular files if requested. 1850 */ 1851 if ((ZTOV(zp)->v_type == VREG) && 1852 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1853 /* we can't hold any locks when calling zfs_freesp() */ 1854 zfs_dirent_unlock(dl); 1855 dl = NULL; 1856 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1857 if (error == 0) { 1858 vnevent_create(ZTOV(zp), ct); 1859 } 1860 } 1861 } 1862out: 1863 getnewvnode_drop_reserve(); 1864 if (dl) 1865 zfs_dirent_unlock(dl); 1866 1867 if (error) { 1868 if (zp) 1869 VN_RELE(ZTOV(zp)); 1870 } else { 1871 *vpp = ZTOV(zp); 1872 error = specvp_check(vpp, cr); 1873 } 1874 1875 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1876 zil_commit(zilog, 0); 1877 1878 ZFS_EXIT(zfsvfs); 1879 return (error); 1880} 1881 1882/* 1883 * Remove an entry from a directory. 1884 * 1885 * IN: dvp - vnode of directory to remove entry from. 1886 * name - name of entry to remove. 1887 * cr - credentials of caller. 1888 * ct - caller context 1889 * flags - case flags 1890 * 1891 * RETURN: 0 on success, error code on failure. 1892 * 1893 * Timestamps: 1894 * dvp - ctime|mtime 1895 * vp - ctime (if nlink > 0) 1896 */ 1897 1898uint64_t null_xattr = 0; 1899 1900/*ARGSUSED*/ 1901static int 1902zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1903 int flags) 1904{ 1905 znode_t *zp, *dzp = VTOZ(dvp); 1906 znode_t *xzp; 1907 vnode_t *vp; 1908 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1909 zilog_t *zilog; 1910 uint64_t acl_obj, xattr_obj; 1911 uint64_t xattr_obj_unlinked = 0; 1912 uint64_t obj = 0; 1913 zfs_dirlock_t *dl; 1914 dmu_tx_t *tx; 1915 boolean_t may_delete_now, delete_now = FALSE; 1916 boolean_t unlinked, toobig = FALSE; 1917 uint64_t txtype; 1918 pathname_t *realnmp = NULL; 1919 pathname_t realnm; 1920 int error; 1921 int zflg = ZEXISTS; 1922 boolean_t waited = B_FALSE; 1923 1924 ZFS_ENTER(zfsvfs); 1925 ZFS_VERIFY_ZP(dzp); 1926 zilog = zfsvfs->z_log; 1927 1928 if (flags & FIGNORECASE) { 1929 zflg |= ZCILOOK; 1930 pn_alloc(&realnm); 1931 realnmp = &realnm; 1932 } 1933 1934top: 1935 xattr_obj = 0; 1936 xzp = NULL; 1937 /* 1938 * Attempt to lock directory; fail if entry doesn't exist. 1939 */ 1940 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1941 NULL, realnmp)) { 1942 if (realnmp) 1943 pn_free(realnmp); 1944 ZFS_EXIT(zfsvfs); 1945 return (error); 1946 } 1947 1948 vp = ZTOV(zp); 1949 1950 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1951 goto out; 1952 } 1953 1954 /* 1955 * Need to use rmdir for removing directories. 1956 */ 1957 if (vp->v_type == VDIR) { 1958 error = SET_ERROR(EPERM); 1959 goto out; 1960 } 1961 1962 vnevent_remove(vp, dvp, name, ct); 1963 1964 if (realnmp) 1965 dnlc_remove(dvp, realnmp->pn_buf); 1966 else 1967 dnlc_remove(dvp, name); 1968 1969 VI_LOCK(vp); 1970 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1971 VI_UNLOCK(vp); 1972 1973 /* 1974 * We may delete the znode now, or we may put it in the unlinked set; 1975 * it depends on whether we're the last link, and on whether there are 1976 * other holds on the vnode. So we dmu_tx_hold() the right things to 1977 * allow for either case. 1978 */ 1979 obj = zp->z_id; 1980 tx = dmu_tx_create(zfsvfs->z_os); 1981 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1982 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1983 zfs_sa_upgrade_txholds(tx, zp); 1984 zfs_sa_upgrade_txholds(tx, dzp); 1985 if (may_delete_now) { 1986 toobig = 1987 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 1988 /* if the file is too big, only hold_free a token amount */ 1989 dmu_tx_hold_free(tx, zp->z_id, 0, 1990 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1991 } 1992 1993 /* are there any extended attributes? */ 1994 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1995 &xattr_obj, sizeof (xattr_obj)); 1996 if (error == 0 && xattr_obj) { 1997 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1998 ASSERT0(error); 1999 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2000 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 2001 } 2002 2003 mutex_enter(&zp->z_lock); 2004 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now) 2005 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 2006 mutex_exit(&zp->z_lock); 2007 2008 /* charge as an update -- would be nice not to charge at all */ 2009 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2010 2011 /* 2012 * Mark this transaction as typically resulting in a net free of space 2013 */ 2014 dmu_tx_mark_netfree(tx); 2015 2016 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2017 if (error) { 2018 zfs_dirent_unlock(dl); 2019 VN_RELE(vp); 2020 if (xzp) 2021 VN_RELE(ZTOV(xzp)); 2022 if (error == ERESTART) { 2023 waited = B_TRUE; 2024 dmu_tx_wait(tx); 2025 dmu_tx_abort(tx); 2026 goto top; 2027 } 2028 if (realnmp) 2029 pn_free(realnmp); 2030 dmu_tx_abort(tx); 2031 ZFS_EXIT(zfsvfs); 2032 return (error); 2033 } 2034 2035 /* 2036 * Remove the directory entry. 2037 */ 2038 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 2039 2040 if (error) { 2041 dmu_tx_commit(tx); 2042 goto out; 2043 } 2044 2045 if (unlinked) { 2046 /* 2047 * Hold z_lock so that we can make sure that the ACL obj 2048 * hasn't changed. Could have been deleted due to 2049 * zfs_sa_upgrade(). 2050 */ 2051 mutex_enter(&zp->z_lock); 2052 VI_LOCK(vp); 2053 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 2054 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked)); 2055 delete_now = may_delete_now && !toobig && 2056 vp->v_count == 1 && !vn_has_cached_data(vp) && 2057 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) == 2058 acl_obj; 2059 VI_UNLOCK(vp); 2060 } 2061 2062 if (delete_now) { 2063#ifdef __FreeBSD__ 2064 panic("zfs_remove: delete_now branch taken"); 2065#endif 2066 if (xattr_obj_unlinked) { 2067 ASSERT3U(xzp->z_links, ==, 2); 2068 mutex_enter(&xzp->z_lock); 2069 xzp->z_unlinked = 1; 2070 xzp->z_links = 0; 2071 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 2072 &xzp->z_links, sizeof (xzp->z_links), tx); 2073 ASSERT3U(error, ==, 0); 2074 mutex_exit(&xzp->z_lock); 2075 zfs_unlinked_add(xzp, tx); 2076 2077 if (zp->z_is_sa) 2078 error = sa_remove(zp->z_sa_hdl, 2079 SA_ZPL_XATTR(zfsvfs), tx); 2080 else 2081 error = sa_update(zp->z_sa_hdl, 2082 SA_ZPL_XATTR(zfsvfs), &null_xattr, 2083 sizeof (uint64_t), tx); 2084 ASSERT0(error); 2085 } 2086 VI_LOCK(vp); 2087 vp->v_count--; 2088 ASSERT0(vp->v_count); 2089 VI_UNLOCK(vp); 2090 mutex_exit(&zp->z_lock); 2091 zfs_znode_delete(zp, tx); 2092 } else if (unlinked) { 2093 mutex_exit(&zp->z_lock); 2094 zfs_unlinked_add(zp, tx); 2095#ifdef __FreeBSD__ 2096 vp->v_vflag |= VV_NOSYNC; 2097#endif 2098 } 2099 2100 txtype = TX_REMOVE; 2101 if (flags & FIGNORECASE) 2102 txtype |= TX_CI; 2103 zfs_log_remove(zilog, tx, txtype, dzp, name, obj); 2104 2105 dmu_tx_commit(tx); 2106out: 2107 if (realnmp) 2108 pn_free(realnmp); 2109 2110 zfs_dirent_unlock(dl); 2111 2112 if (!delete_now) 2113 VN_RELE(vp); 2114 if (xzp) 2115 VN_RELE(ZTOV(xzp)); 2116 2117 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2118 zil_commit(zilog, 0); 2119 2120 ZFS_EXIT(zfsvfs); 2121 return (error); 2122} 2123 2124/* 2125 * Create a new directory and insert it into dvp using the name 2126 * provided. Return a pointer to the inserted directory. 2127 * 2128 * IN: dvp - vnode of directory to add subdir to. 2129 * dirname - name of new directory. 2130 * vap - attributes of new directory. 2131 * cr - credentials of caller. 2132 * ct - caller context 2133 * flags - case flags 2134 * vsecp - ACL to be set 2135 * 2136 * OUT: vpp - vnode of created directory. 2137 * 2138 * RETURN: 0 on success, error code on failure. 2139 * 2140 * Timestamps: 2141 * dvp - ctime|mtime updated 2142 * vp - ctime|mtime|atime updated 2143 */ 2144/*ARGSUSED*/ 2145static int 2146zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 2147 caller_context_t *ct, int flags, vsecattr_t *vsecp) 2148{ 2149 znode_t *zp, *dzp = VTOZ(dvp); 2150 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2151 zilog_t *zilog; 2152 zfs_dirlock_t *dl; 2153 uint64_t txtype; 2154 dmu_tx_t *tx; 2155 int error; 2156 int zf = ZNEW; 2157 ksid_t *ksid; 2158 uid_t uid; 2159 gid_t gid = crgetgid(cr); 2160 zfs_acl_ids_t acl_ids; 2161 boolean_t fuid_dirtied; 2162 boolean_t waited = B_FALSE; 2163 2164 ASSERT(vap->va_type == VDIR); 2165 2166 /* 2167 * If we have an ephemeral id, ACL, or XVATTR then 2168 * make sure file system is at proper version 2169 */ 2170 2171 ksid = crgetsid(cr, KSID_OWNER); 2172 if (ksid) 2173 uid = ksid_getid(ksid); 2174 else 2175 uid = crgetuid(cr); 2176 if (zfsvfs->z_use_fuids == B_FALSE && 2177 (vsecp || (vap->va_mask & AT_XVATTR) || 2178 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 2179 return (SET_ERROR(EINVAL)); 2180 2181 ZFS_ENTER(zfsvfs); 2182 ZFS_VERIFY_ZP(dzp); 2183 zilog = zfsvfs->z_log; 2184 2185 if (dzp->z_pflags & ZFS_XATTR) { 2186 ZFS_EXIT(zfsvfs); 2187 return (SET_ERROR(EINVAL)); 2188 } 2189 2190 if (zfsvfs->z_utf8 && u8_validate(dirname, 2191 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2192 ZFS_EXIT(zfsvfs); 2193 return (SET_ERROR(EILSEQ)); 2194 } 2195 if (flags & FIGNORECASE) 2196 zf |= ZCILOOK; 2197 2198 if (vap->va_mask & AT_XVATTR) { 2199 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 2200 crgetuid(cr), cr, vap->va_type)) != 0) { 2201 ZFS_EXIT(zfsvfs); 2202 return (error); 2203 } 2204 } 2205 2206 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 2207 vsecp, &acl_ids)) != 0) { 2208 ZFS_EXIT(zfsvfs); 2209 return (error); 2210 } 2211 2212 getnewvnode_reserve(1); 2213 2214 /* 2215 * First make sure the new directory doesn't exist. 2216 * 2217 * Existence is checked first to make sure we don't return 2218 * EACCES instead of EEXIST which can cause some applications 2219 * to fail. 2220 */ 2221top: 2222 *vpp = NULL; 2223 2224 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 2225 NULL, NULL)) { 2226 zfs_acl_ids_free(&acl_ids); 2227 getnewvnode_drop_reserve(); 2228 ZFS_EXIT(zfsvfs); 2229 return (error); 2230 } 2231 2232 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 2233 zfs_acl_ids_free(&acl_ids); 2234 zfs_dirent_unlock(dl); 2235 getnewvnode_drop_reserve(); 2236 ZFS_EXIT(zfsvfs); 2237 return (error); 2238 } 2239 2240 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 2241 zfs_acl_ids_free(&acl_ids); 2242 zfs_dirent_unlock(dl); 2243 getnewvnode_drop_reserve(); 2244 ZFS_EXIT(zfsvfs); 2245 return (SET_ERROR(EDQUOT)); 2246 } 2247 2248 /* 2249 * Add a new entry to the directory. 2250 */ 2251 tx = dmu_tx_create(zfsvfs->z_os); 2252 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 2253 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2254 fuid_dirtied = zfsvfs->z_fuid_dirty; 2255 if (fuid_dirtied) 2256 zfs_fuid_txhold(zfsvfs, tx); 2257 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2258 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2259 acl_ids.z_aclp->z_acl_bytes); 2260 } 2261 2262 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 2263 ZFS_SA_BASE_ATTR_SIZE); 2264 2265 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2266 if (error) { 2267 zfs_dirent_unlock(dl); 2268 if (error == ERESTART) { 2269 waited = B_TRUE; 2270 dmu_tx_wait(tx); 2271 dmu_tx_abort(tx); 2272 goto top; 2273 } 2274 zfs_acl_ids_free(&acl_ids); 2275 dmu_tx_abort(tx); 2276 getnewvnode_drop_reserve(); 2277 ZFS_EXIT(zfsvfs); 2278 return (error); 2279 } 2280 2281 /* 2282 * Create new node. 2283 */ 2284 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 2285 2286 if (fuid_dirtied) 2287 zfs_fuid_sync(zfsvfs, tx); 2288 2289 /* 2290 * Now put new name in parent dir. 2291 */ 2292 (void) zfs_link_create(dl, zp, tx, ZNEW); 2293 2294 *vpp = ZTOV(zp); 2295 2296 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 2297 if (flags & FIGNORECASE) 2298 txtype |= TX_CI; 2299 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 2300 acl_ids.z_fuidp, vap); 2301 2302 zfs_acl_ids_free(&acl_ids); 2303 2304 dmu_tx_commit(tx); 2305 2306 getnewvnode_drop_reserve(); 2307 2308 zfs_dirent_unlock(dl); 2309 2310 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2311 zil_commit(zilog, 0); 2312 2313 ZFS_EXIT(zfsvfs); 2314 return (0); 2315} 2316 2317/* 2318 * Remove a directory subdir entry. If the current working 2319 * directory is the same as the subdir to be removed, the 2320 * remove will fail. 2321 * 2322 * IN: dvp - vnode of directory to remove from. 2323 * name - name of directory to be removed. 2324 * cwd - vnode of current working directory. 2325 * cr - credentials of caller. 2326 * ct - caller context 2327 * flags - case flags 2328 * 2329 * RETURN: 0 on success, error code on failure. 2330 * 2331 * Timestamps: 2332 * dvp - ctime|mtime updated 2333 */ 2334/*ARGSUSED*/ 2335static int 2336zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 2337 caller_context_t *ct, int flags) 2338{ 2339 znode_t *dzp = VTOZ(dvp); 2340 znode_t *zp; 2341 vnode_t *vp; 2342 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2343 zilog_t *zilog; 2344 zfs_dirlock_t *dl; 2345 dmu_tx_t *tx; 2346 int error; 2347 int zflg = ZEXISTS; 2348 boolean_t waited = B_FALSE; 2349 2350 ZFS_ENTER(zfsvfs); 2351 ZFS_VERIFY_ZP(dzp); 2352 zilog = zfsvfs->z_log; 2353 2354 if (flags & FIGNORECASE) 2355 zflg |= ZCILOOK; 2356top: 2357 zp = NULL; 2358 2359 /* 2360 * Attempt to lock directory; fail if entry doesn't exist. 2361 */ 2362 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 2363 NULL, NULL)) { 2364 ZFS_EXIT(zfsvfs); 2365 return (error); 2366 } 2367 2368 vp = ZTOV(zp); 2369 2370 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 2371 goto out; 2372 } 2373 2374 if (vp->v_type != VDIR) { 2375 error = SET_ERROR(ENOTDIR); 2376 goto out; 2377 } 2378 2379 if (vp == cwd) { 2380 error = SET_ERROR(EINVAL); 2381 goto out; 2382 } 2383 2384 vnevent_rmdir(vp, dvp, name, ct); 2385 2386 /* 2387 * Grab a lock on the directory to make sure that noone is 2388 * trying to add (or lookup) entries while we are removing it. 2389 */ 2390 rw_enter(&zp->z_name_lock, RW_WRITER); 2391 2392 /* 2393 * Grab a lock on the parent pointer to make sure we play well 2394 * with the treewalk and directory rename code. 2395 */ 2396 rw_enter(&zp->z_parent_lock, RW_WRITER); 2397 2398 tx = dmu_tx_create(zfsvfs->z_os); 2399 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2400 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2401 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2402 zfs_sa_upgrade_txholds(tx, zp); 2403 zfs_sa_upgrade_txholds(tx, dzp); 2404 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2405 if (error) { 2406 rw_exit(&zp->z_parent_lock); 2407 rw_exit(&zp->z_name_lock); 2408 zfs_dirent_unlock(dl); 2409 VN_RELE(vp); 2410 if (error == ERESTART) { 2411 waited = B_TRUE; 2412 dmu_tx_wait(tx); 2413 dmu_tx_abort(tx); 2414 goto top; 2415 } 2416 dmu_tx_abort(tx); 2417 ZFS_EXIT(zfsvfs); 2418 return (error); 2419 } 2420 2421#ifdef FREEBSD_NAMECACHE 2422 cache_purge(dvp); 2423#endif 2424 2425 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 2426 2427 if (error == 0) { 2428 uint64_t txtype = TX_RMDIR; 2429 if (flags & FIGNORECASE) 2430 txtype |= TX_CI; 2431 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT); 2432 } 2433 2434 dmu_tx_commit(tx); 2435 2436 rw_exit(&zp->z_parent_lock); 2437 rw_exit(&zp->z_name_lock); 2438#ifdef FREEBSD_NAMECACHE 2439 cache_purge(vp); 2440#endif 2441out: 2442 zfs_dirent_unlock(dl); 2443 2444 VN_RELE(vp); 2445 2446 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2447 zil_commit(zilog, 0); 2448 2449 ZFS_EXIT(zfsvfs); 2450 return (error); 2451} 2452 2453/* 2454 * Read as many directory entries as will fit into the provided 2455 * buffer from the given directory cursor position (specified in 2456 * the uio structure). 2457 * 2458 * IN: vp - vnode of directory to read. 2459 * uio - structure supplying read location, range info, 2460 * and return buffer. 2461 * cr - credentials of caller. 2462 * ct - caller context 2463 * flags - case flags 2464 * 2465 * OUT: uio - updated offset and range, buffer filled. 2466 * eofp - set to true if end-of-file detected. 2467 * 2468 * RETURN: 0 on success, error code on failure. 2469 * 2470 * Timestamps: 2471 * vp - atime updated 2472 * 2473 * Note that the low 4 bits of the cookie returned by zap is always zero. 2474 * This allows us to use the low range for "special" directory entries: 2475 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 2476 * we use the offset 2 for the '.zfs' directory. 2477 */ 2478/* ARGSUSED */ 2479static int 2480zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 2481{ 2482 znode_t *zp = VTOZ(vp); 2483 iovec_t *iovp; 2484 edirent_t *eodp; 2485 dirent64_t *odp; 2486 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2487 objset_t *os; 2488 caddr_t outbuf; 2489 size_t bufsize; 2490 zap_cursor_t zc; 2491 zap_attribute_t zap; 2492 uint_t bytes_wanted; 2493 uint64_t offset; /* must be unsigned; checks for < 1 */ 2494 uint64_t parent; 2495 int local_eof; 2496 int outcount; 2497 int error; 2498 uint8_t prefetch; 2499 boolean_t check_sysattrs; 2500 uint8_t type; 2501 int ncooks; 2502 u_long *cooks = NULL; 2503 int flags = 0; 2504 2505 ZFS_ENTER(zfsvfs); 2506 ZFS_VERIFY_ZP(zp); 2507 2508 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 2509 &parent, sizeof (parent))) != 0) { 2510 ZFS_EXIT(zfsvfs); 2511 return (error); 2512 } 2513 2514 /* 2515 * If we are not given an eof variable, 2516 * use a local one. 2517 */ 2518 if (eofp == NULL) 2519 eofp = &local_eof; 2520 2521 /* 2522 * Check for valid iov_len. 2523 */ 2524 if (uio->uio_iov->iov_len <= 0) { 2525 ZFS_EXIT(zfsvfs); 2526 return (SET_ERROR(EINVAL)); 2527 } 2528 2529 /* 2530 * Quit if directory has been removed (posix) 2531 */ 2532 if ((*eofp = zp->z_unlinked) != 0) { 2533 ZFS_EXIT(zfsvfs); 2534 return (0); 2535 } 2536 2537 error = 0; 2538 os = zfsvfs->z_os; 2539 offset = uio->uio_loffset; 2540 prefetch = zp->z_zn_prefetch; 2541 2542 /* 2543 * Initialize the iterator cursor. 2544 */ 2545 if (offset <= 3) { 2546 /* 2547 * Start iteration from the beginning of the directory. 2548 */ 2549 zap_cursor_init(&zc, os, zp->z_id); 2550 } else { 2551 /* 2552 * The offset is a serialized cursor. 2553 */ 2554 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 2555 } 2556 2557 /* 2558 * Get space to change directory entries into fs independent format. 2559 */ 2560 iovp = uio->uio_iov; 2561 bytes_wanted = iovp->iov_len; 2562 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 2563 bufsize = bytes_wanted; 2564 outbuf = kmem_alloc(bufsize, KM_SLEEP); 2565 odp = (struct dirent64 *)outbuf; 2566 } else { 2567 bufsize = bytes_wanted; 2568 outbuf = NULL; 2569 odp = (struct dirent64 *)iovp->iov_base; 2570 } 2571 eodp = (struct edirent *)odp; 2572 2573 if (ncookies != NULL) { 2574 /* 2575 * Minimum entry size is dirent size and 1 byte for a file name. 2576 */ 2577 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 2578 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 2579 *cookies = cooks; 2580 *ncookies = ncooks; 2581 } 2582 /* 2583 * If this VFS supports the system attribute view interface; and 2584 * we're looking at an extended attribute directory; and we care 2585 * about normalization conflicts on this vfs; then we must check 2586 * for normalization conflicts with the sysattr name space. 2587 */ 2588#ifdef TODO 2589 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 2590 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 2591 (flags & V_RDDIR_ENTFLAGS); 2592#else 2593 check_sysattrs = 0; 2594#endif 2595 2596 /* 2597 * Transform to file-system independent format 2598 */ 2599 outcount = 0; 2600 while (outcount < bytes_wanted) { 2601 ino64_t objnum; 2602 ushort_t reclen; 2603 off64_t *next = NULL; 2604 2605 /* 2606 * Special case `.', `..', and `.zfs'. 2607 */ 2608 if (offset == 0) { 2609 (void) strcpy(zap.za_name, "."); 2610 zap.za_normalization_conflict = 0; 2611 objnum = zp->z_id; 2612 type = DT_DIR; 2613 } else if (offset == 1) { 2614 (void) strcpy(zap.za_name, ".."); 2615 zap.za_normalization_conflict = 0; 2616 objnum = parent; 2617 type = DT_DIR; 2618 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2619 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2620 zap.za_normalization_conflict = 0; 2621 objnum = ZFSCTL_INO_ROOT; 2622 type = DT_DIR; 2623 } else { 2624 /* 2625 * Grab next entry. 2626 */ 2627 if (error = zap_cursor_retrieve(&zc, &zap)) { 2628 if ((*eofp = (error == ENOENT)) != 0) 2629 break; 2630 else 2631 goto update; 2632 } 2633 2634 if (zap.za_integer_length != 8 || 2635 zap.za_num_integers != 1) { 2636 cmn_err(CE_WARN, "zap_readdir: bad directory " 2637 "entry, obj = %lld, offset = %lld\n", 2638 (u_longlong_t)zp->z_id, 2639 (u_longlong_t)offset); 2640 error = SET_ERROR(ENXIO); 2641 goto update; 2642 } 2643 2644 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2645 /* 2646 * MacOS X can extract the object type here such as: 2647 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2648 */ 2649 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2650 2651 if (check_sysattrs && !zap.za_normalization_conflict) { 2652#ifdef TODO 2653 zap.za_normalization_conflict = 2654 xattr_sysattr_casechk(zap.za_name); 2655#else 2656 panic("%s:%u: TODO", __func__, __LINE__); 2657#endif 2658 } 2659 } 2660 2661 if (flags & V_RDDIR_ACCFILTER) { 2662 /* 2663 * If we have no access at all, don't include 2664 * this entry in the returned information 2665 */ 2666 znode_t *ezp; 2667 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2668 goto skip_entry; 2669 if (!zfs_has_access(ezp, cr)) { 2670 VN_RELE(ZTOV(ezp)); 2671 goto skip_entry; 2672 } 2673 VN_RELE(ZTOV(ezp)); 2674 } 2675 2676 if (flags & V_RDDIR_ENTFLAGS) 2677 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2678 else 2679 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2680 2681 /* 2682 * Will this entry fit in the buffer? 2683 */ 2684 if (outcount + reclen > bufsize) { 2685 /* 2686 * Did we manage to fit anything in the buffer? 2687 */ 2688 if (!outcount) { 2689 error = SET_ERROR(EINVAL); 2690 goto update; 2691 } 2692 break; 2693 } 2694 if (flags & V_RDDIR_ENTFLAGS) { 2695 /* 2696 * Add extended flag entry: 2697 */ 2698 eodp->ed_ino = objnum; 2699 eodp->ed_reclen = reclen; 2700 /* NOTE: ed_off is the offset for the *next* entry */ 2701 next = &(eodp->ed_off); 2702 eodp->ed_eflags = zap.za_normalization_conflict ? 2703 ED_CASE_CONFLICT : 0; 2704 (void) strncpy(eodp->ed_name, zap.za_name, 2705 EDIRENT_NAMELEN(reclen)); 2706 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2707 } else { 2708 /* 2709 * Add normal entry: 2710 */ 2711 odp->d_ino = objnum; 2712 odp->d_reclen = reclen; 2713 odp->d_namlen = strlen(zap.za_name); 2714 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 2715 odp->d_type = type; 2716 odp = (dirent64_t *)((intptr_t)odp + reclen); 2717 } 2718 outcount += reclen; 2719 2720 ASSERT(outcount <= bufsize); 2721 2722 /* Prefetch znode */ 2723 if (prefetch) 2724 dmu_prefetch(os, objnum, 0, 0, 0, 2725 ZIO_PRIORITY_SYNC_READ); 2726 2727 skip_entry: 2728 /* 2729 * Move to the next entry, fill in the previous offset. 2730 */ 2731 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2732 zap_cursor_advance(&zc); 2733 offset = zap_cursor_serialize(&zc); 2734 } else { 2735 offset += 1; 2736 } 2737 2738 if (cooks != NULL) { 2739 *cooks++ = offset; 2740 ncooks--; 2741 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 2742 } 2743 } 2744 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2745 2746 /* Subtract unused cookies */ 2747 if (ncookies != NULL) 2748 *ncookies -= ncooks; 2749 2750 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2751 iovp->iov_base += outcount; 2752 iovp->iov_len -= outcount; 2753 uio->uio_resid -= outcount; 2754 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2755 /* 2756 * Reset the pointer. 2757 */ 2758 offset = uio->uio_loffset; 2759 } 2760 2761update: 2762 zap_cursor_fini(&zc); 2763 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2764 kmem_free(outbuf, bufsize); 2765 2766 if (error == ENOENT) 2767 error = 0; 2768 2769 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2770 2771 uio->uio_loffset = offset; 2772 ZFS_EXIT(zfsvfs); 2773 if (error != 0 && cookies != NULL) { 2774 free(*cookies, M_TEMP); 2775 *cookies = NULL; 2776 *ncookies = 0; 2777 } 2778 return (error); 2779} 2780 2781ulong_t zfs_fsync_sync_cnt = 4; 2782 2783static int 2784zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2785{ 2786 znode_t *zp = VTOZ(vp); 2787 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2788 2789 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2790 2791 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { 2792 ZFS_ENTER(zfsvfs); 2793 ZFS_VERIFY_ZP(zp); 2794 zil_commit(zfsvfs->z_log, zp->z_id); 2795 ZFS_EXIT(zfsvfs); 2796 } 2797 return (0); 2798} 2799 2800 2801/* 2802 * Get the requested file attributes and place them in the provided 2803 * vattr structure. 2804 * 2805 * IN: vp - vnode of file. 2806 * vap - va_mask identifies requested attributes. 2807 * If AT_XVATTR set, then optional attrs are requested 2808 * flags - ATTR_NOACLCHECK (CIFS server context) 2809 * cr - credentials of caller. 2810 * ct - caller context 2811 * 2812 * OUT: vap - attribute values. 2813 * 2814 * RETURN: 0 (always succeeds). 2815 */ 2816/* ARGSUSED */ 2817static int 2818zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2819 caller_context_t *ct) 2820{ 2821 znode_t *zp = VTOZ(vp); 2822 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2823 int error = 0; 2824 uint32_t blksize; 2825 u_longlong_t nblocks; 2826 uint64_t links; 2827 uint64_t mtime[2], ctime[2], crtime[2], rdev; 2828 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2829 xoptattr_t *xoap = NULL; 2830 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2831 sa_bulk_attr_t bulk[4]; 2832 int count = 0; 2833 2834 ZFS_ENTER(zfsvfs); 2835 ZFS_VERIFY_ZP(zp); 2836 2837 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2838 2839 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 2840 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 2841 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16); 2842 if (vp->v_type == VBLK || vp->v_type == VCHR) 2843 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, 2844 &rdev, 8); 2845 2846 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 2847 ZFS_EXIT(zfsvfs); 2848 return (error); 2849 } 2850 2851 /* 2852 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2853 * Also, if we are the owner don't bother, since owner should 2854 * always be allowed to read basic attributes of file. 2855 */ 2856 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 2857 (vap->va_uid != crgetuid(cr))) { 2858 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2859 skipaclchk, cr)) { 2860 ZFS_EXIT(zfsvfs); 2861 return (error); 2862 } 2863 } 2864 2865 /* 2866 * Return all attributes. It's cheaper to provide the answer 2867 * than to determine whether we were asked the question. 2868 */ 2869 2870 mutex_enter(&zp->z_lock); 2871 vap->va_type = IFTOVT(zp->z_mode); 2872 vap->va_mode = zp->z_mode & ~S_IFMT; 2873#ifdef illumos 2874 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2875#else 2876 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 2877#endif 2878 vap->va_nodeid = zp->z_id; 2879 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2880 links = zp->z_links + 1; 2881 else 2882 links = zp->z_links; 2883 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */ 2884 vap->va_size = zp->z_size; 2885#ifdef illumos 2886 vap->va_rdev = vp->v_rdev; 2887#else 2888 if (vp->v_type == VBLK || vp->v_type == VCHR) 2889 vap->va_rdev = zfs_cmpldev(rdev); 2890#endif 2891 vap->va_seq = zp->z_seq; 2892 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2893 vap->va_filerev = zp->z_seq; 2894 2895 /* 2896 * Add in any requested optional attributes and the create time. 2897 * Also set the corresponding bits in the returned attribute bitmap. 2898 */ 2899 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2900 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2901 xoap->xoa_archive = 2902 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2903 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2904 } 2905 2906 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2907 xoap->xoa_readonly = 2908 ((zp->z_pflags & ZFS_READONLY) != 0); 2909 XVA_SET_RTN(xvap, XAT_READONLY); 2910 } 2911 2912 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2913 xoap->xoa_system = 2914 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2915 XVA_SET_RTN(xvap, XAT_SYSTEM); 2916 } 2917 2918 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2919 xoap->xoa_hidden = 2920 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2921 XVA_SET_RTN(xvap, XAT_HIDDEN); 2922 } 2923 2924 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2925 xoap->xoa_nounlink = 2926 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2927 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2928 } 2929 2930 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2931 xoap->xoa_immutable = 2932 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2933 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2934 } 2935 2936 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2937 xoap->xoa_appendonly = 2938 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2939 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2940 } 2941 2942 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2943 xoap->xoa_nodump = 2944 ((zp->z_pflags & ZFS_NODUMP) != 0); 2945 XVA_SET_RTN(xvap, XAT_NODUMP); 2946 } 2947 2948 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2949 xoap->xoa_opaque = 2950 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2951 XVA_SET_RTN(xvap, XAT_OPAQUE); 2952 } 2953 2954 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2955 xoap->xoa_av_quarantined = 2956 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2957 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2958 } 2959 2960 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2961 xoap->xoa_av_modified = 2962 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2963 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2964 } 2965 2966 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2967 vp->v_type == VREG) { 2968 zfs_sa_get_scanstamp(zp, xvap); 2969 } 2970 2971 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2972 uint64_t times[2]; 2973 2974 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 2975 times, sizeof (times)); 2976 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 2977 XVA_SET_RTN(xvap, XAT_CREATETIME); 2978 } 2979 2980 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2981 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2982 XVA_SET_RTN(xvap, XAT_REPARSE); 2983 } 2984 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2985 xoap->xoa_generation = zp->z_gen; 2986 XVA_SET_RTN(xvap, XAT_GEN); 2987 } 2988 2989 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2990 xoap->xoa_offline = 2991 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2992 XVA_SET_RTN(xvap, XAT_OFFLINE); 2993 } 2994 2995 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2996 xoap->xoa_sparse = 2997 ((zp->z_pflags & ZFS_SPARSE) != 0); 2998 XVA_SET_RTN(xvap, XAT_SPARSE); 2999 } 3000 } 3001 3002 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 3003 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 3004 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 3005 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 3006 3007 mutex_exit(&zp->z_lock); 3008 3009 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 3010 vap->va_blksize = blksize; 3011 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 3012 3013 if (zp->z_blksz == 0) { 3014 /* 3015 * Block size hasn't been set; suggest maximal I/O transfers. 3016 */ 3017 vap->va_blksize = zfsvfs->z_max_blksz; 3018 } 3019 3020 ZFS_EXIT(zfsvfs); 3021 return (0); 3022} 3023 3024/* 3025 * Set the file attributes to the values contained in the 3026 * vattr structure. 3027 * 3028 * IN: vp - vnode of file to be modified. 3029 * vap - new attribute values. 3030 * If AT_XVATTR set, then optional attrs are being set 3031 * flags - ATTR_UTIME set if non-default time values provided. 3032 * - ATTR_NOACLCHECK (CIFS context only). 3033 * cr - credentials of caller. 3034 * ct - caller context 3035 * 3036 * RETURN: 0 on success, error code on failure. 3037 * 3038 * Timestamps: 3039 * vp - ctime updated, mtime updated if size changed. 3040 */ 3041/* ARGSUSED */ 3042static int 3043zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 3044 caller_context_t *ct) 3045{ 3046 znode_t *zp = VTOZ(vp); 3047 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3048 zilog_t *zilog; 3049 dmu_tx_t *tx; 3050 vattr_t oldva; 3051 xvattr_t tmpxvattr; 3052 uint_t mask = vap->va_mask; 3053 uint_t saved_mask = 0; 3054 uint64_t saved_mode; 3055 int trim_mask = 0; 3056 uint64_t new_mode; 3057 uint64_t new_uid, new_gid; 3058 uint64_t xattr_obj; 3059 uint64_t mtime[2], ctime[2]; 3060 znode_t *attrzp; 3061 int need_policy = FALSE; 3062 int err, err2; 3063 zfs_fuid_info_t *fuidp = NULL; 3064 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 3065 xoptattr_t *xoap; 3066 zfs_acl_t *aclp; 3067 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 3068 boolean_t fuid_dirtied = B_FALSE; 3069 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 3070 int count = 0, xattr_count = 0; 3071 3072 if (mask == 0) 3073 return (0); 3074 3075 if (mask & AT_NOSET) 3076 return (SET_ERROR(EINVAL)); 3077 3078 ZFS_ENTER(zfsvfs); 3079 ZFS_VERIFY_ZP(zp); 3080 3081 zilog = zfsvfs->z_log; 3082 3083 /* 3084 * Make sure that if we have ephemeral uid/gid or xvattr specified 3085 * that file system is at proper version level 3086 */ 3087 3088 if (zfsvfs->z_use_fuids == B_FALSE && 3089 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 3090 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 3091 (mask & AT_XVATTR))) { 3092 ZFS_EXIT(zfsvfs); 3093 return (SET_ERROR(EINVAL)); 3094 } 3095 3096 if (mask & AT_SIZE && vp->v_type == VDIR) { 3097 ZFS_EXIT(zfsvfs); 3098 return (SET_ERROR(EISDIR)); 3099 } 3100 3101 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 3102 ZFS_EXIT(zfsvfs); 3103 return (SET_ERROR(EINVAL)); 3104 } 3105 3106 /* 3107 * If this is an xvattr_t, then get a pointer to the structure of 3108 * optional attributes. If this is NULL, then we have a vattr_t. 3109 */ 3110 xoap = xva_getxoptattr(xvap); 3111 3112 xva_init(&tmpxvattr); 3113 3114 /* 3115 * Immutable files can only alter immutable bit and atime 3116 */ 3117 if ((zp->z_pflags & ZFS_IMMUTABLE) && 3118 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 3119 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 3120 ZFS_EXIT(zfsvfs); 3121 return (SET_ERROR(EPERM)); 3122 } 3123 3124 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 3125 ZFS_EXIT(zfsvfs); 3126 return (SET_ERROR(EPERM)); 3127 } 3128 3129 /* 3130 * Verify timestamps doesn't overflow 32 bits. 3131 * ZFS can handle large timestamps, but 32bit syscalls can't 3132 * handle times greater than 2039. This check should be removed 3133 * once large timestamps are fully supported. 3134 */ 3135 if (mask & (AT_ATIME | AT_MTIME)) { 3136 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 3137 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 3138 ZFS_EXIT(zfsvfs); 3139 return (SET_ERROR(EOVERFLOW)); 3140 } 3141 } 3142 3143top: 3144 attrzp = NULL; 3145 aclp = NULL; 3146 3147 /* Can this be moved to before the top label? */ 3148 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 3149 ZFS_EXIT(zfsvfs); 3150 return (SET_ERROR(EROFS)); 3151 } 3152 3153 /* 3154 * First validate permissions 3155 */ 3156 3157 if (mask & AT_SIZE) { 3158 /* 3159 * XXX - Note, we are not providing any open 3160 * mode flags here (like FNDELAY), so we may 3161 * block if there are locks present... this 3162 * should be addressed in openat(). 3163 */ 3164 /* XXX - would it be OK to generate a log record here? */ 3165 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 3166 if (err) { 3167 ZFS_EXIT(zfsvfs); 3168 return (err); 3169 } 3170 } 3171 3172 if (mask & (AT_ATIME|AT_MTIME) || 3173 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 3174 XVA_ISSET_REQ(xvap, XAT_READONLY) || 3175 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 3176 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 3177 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 3178 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 3179 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 3180 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 3181 skipaclchk, cr); 3182 } 3183 3184 if (mask & (AT_UID|AT_GID)) { 3185 int idmask = (mask & (AT_UID|AT_GID)); 3186 int take_owner; 3187 int take_group; 3188 3189 /* 3190 * NOTE: even if a new mode is being set, 3191 * we may clear S_ISUID/S_ISGID bits. 3192 */ 3193 3194 if (!(mask & AT_MODE)) 3195 vap->va_mode = zp->z_mode; 3196 3197 /* 3198 * Take ownership or chgrp to group we are a member of 3199 */ 3200 3201 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 3202 take_group = (mask & AT_GID) && 3203 zfs_groupmember(zfsvfs, vap->va_gid, cr); 3204 3205 /* 3206 * If both AT_UID and AT_GID are set then take_owner and 3207 * take_group must both be set in order to allow taking 3208 * ownership. 3209 * 3210 * Otherwise, send the check through secpolicy_vnode_setattr() 3211 * 3212 */ 3213 3214 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 3215 ((idmask == AT_UID) && take_owner) || 3216 ((idmask == AT_GID) && take_group)) { 3217 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 3218 skipaclchk, cr) == 0) { 3219 /* 3220 * Remove setuid/setgid for non-privileged users 3221 */ 3222 secpolicy_setid_clear(vap, vp, cr); 3223 trim_mask = (mask & (AT_UID|AT_GID)); 3224 } else { 3225 need_policy = TRUE; 3226 } 3227 } else { 3228 need_policy = TRUE; 3229 } 3230 } 3231 3232 mutex_enter(&zp->z_lock); 3233 oldva.va_mode = zp->z_mode; 3234 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 3235 if (mask & AT_XVATTR) { 3236 /* 3237 * Update xvattr mask to include only those attributes 3238 * that are actually changing. 3239 * 3240 * the bits will be restored prior to actually setting 3241 * the attributes so the caller thinks they were set. 3242 */ 3243 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 3244 if (xoap->xoa_appendonly != 3245 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 3246 need_policy = TRUE; 3247 } else { 3248 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 3249 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 3250 } 3251 } 3252 3253 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 3254 if (xoap->xoa_nounlink != 3255 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 3256 need_policy = TRUE; 3257 } else { 3258 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 3259 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 3260 } 3261 } 3262 3263 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 3264 if (xoap->xoa_immutable != 3265 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 3266 need_policy = TRUE; 3267 } else { 3268 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 3269 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 3270 } 3271 } 3272 3273 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 3274 if (xoap->xoa_nodump != 3275 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 3276 need_policy = TRUE; 3277 } else { 3278 XVA_CLR_REQ(xvap, XAT_NODUMP); 3279 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 3280 } 3281 } 3282 3283 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 3284 if (xoap->xoa_av_modified != 3285 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 3286 need_policy = TRUE; 3287 } else { 3288 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 3289 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 3290 } 3291 } 3292 3293 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 3294 if ((vp->v_type != VREG && 3295 xoap->xoa_av_quarantined) || 3296 xoap->xoa_av_quarantined != 3297 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 3298 need_policy = TRUE; 3299 } else { 3300 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 3301 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 3302 } 3303 } 3304 3305 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 3306 mutex_exit(&zp->z_lock); 3307 ZFS_EXIT(zfsvfs); 3308 return (SET_ERROR(EPERM)); 3309 } 3310 3311 if (need_policy == FALSE && 3312 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 3313 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 3314 need_policy = TRUE; 3315 } 3316 } 3317 3318 mutex_exit(&zp->z_lock); 3319 3320 if (mask & AT_MODE) { 3321 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 3322 err = secpolicy_setid_setsticky_clear(vp, vap, 3323 &oldva, cr); 3324 if (err) { 3325 ZFS_EXIT(zfsvfs); 3326 return (err); 3327 } 3328 trim_mask |= AT_MODE; 3329 } else { 3330 need_policy = TRUE; 3331 } 3332 } 3333 3334 if (need_policy) { 3335 /* 3336 * If trim_mask is set then take ownership 3337 * has been granted or write_acl is present and user 3338 * has the ability to modify mode. In that case remove 3339 * UID|GID and or MODE from mask so that 3340 * secpolicy_vnode_setattr() doesn't revoke it. 3341 */ 3342 3343 if (trim_mask) { 3344 saved_mask = vap->va_mask; 3345 vap->va_mask &= ~trim_mask; 3346 if (trim_mask & AT_MODE) { 3347 /* 3348 * Save the mode, as secpolicy_vnode_setattr() 3349 * will overwrite it with ova.va_mode. 3350 */ 3351 saved_mode = vap->va_mode; 3352 } 3353 } 3354 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 3355 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 3356 if (err) { 3357 ZFS_EXIT(zfsvfs); 3358 return (err); 3359 } 3360 3361 if (trim_mask) { 3362 vap->va_mask |= saved_mask; 3363 if (trim_mask & AT_MODE) { 3364 /* 3365 * Recover the mode after 3366 * secpolicy_vnode_setattr(). 3367 */ 3368 vap->va_mode = saved_mode; 3369 } 3370 } 3371 } 3372 3373 /* 3374 * secpolicy_vnode_setattr, or take ownership may have 3375 * changed va_mask 3376 */ 3377 mask = vap->va_mask; 3378 3379 if ((mask & (AT_UID | AT_GID))) { 3380 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 3381 &xattr_obj, sizeof (xattr_obj)); 3382 3383 if (err == 0 && xattr_obj) { 3384 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 3385 if (err) 3386 goto out2; 3387 } 3388 if (mask & AT_UID) { 3389 new_uid = zfs_fuid_create(zfsvfs, 3390 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 3391 if (new_uid != zp->z_uid && 3392 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 3393 if (attrzp) 3394 VN_RELE(ZTOV(attrzp)); 3395 err = SET_ERROR(EDQUOT); 3396 goto out2; 3397 } 3398 } 3399 3400 if (mask & AT_GID) { 3401 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 3402 cr, ZFS_GROUP, &fuidp); 3403 if (new_gid != zp->z_gid && 3404 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 3405 if (attrzp) 3406 VN_RELE(ZTOV(attrzp)); 3407 err = SET_ERROR(EDQUOT); 3408 goto out2; 3409 } 3410 } 3411 } 3412 tx = dmu_tx_create(zfsvfs->z_os); 3413 3414 if (mask & AT_MODE) { 3415 uint64_t pmode = zp->z_mode; 3416 uint64_t acl_obj; 3417 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 3418 3419 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && 3420 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 3421 err = SET_ERROR(EPERM); 3422 goto out; 3423 } 3424 3425 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 3426 goto out; 3427 3428 mutex_enter(&zp->z_lock); 3429 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 3430 /* 3431 * Are we upgrading ACL from old V0 format 3432 * to V1 format? 3433 */ 3434 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 3435 zfs_znode_acl_version(zp) == 3436 ZFS_ACL_VERSION_INITIAL) { 3437 dmu_tx_hold_free(tx, acl_obj, 0, 3438 DMU_OBJECT_END); 3439 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3440 0, aclp->z_acl_bytes); 3441 } else { 3442 dmu_tx_hold_write(tx, acl_obj, 0, 3443 aclp->z_acl_bytes); 3444 } 3445 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3446 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3447 0, aclp->z_acl_bytes); 3448 } 3449 mutex_exit(&zp->z_lock); 3450 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3451 } else { 3452 if ((mask & AT_XVATTR) && 3453 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3454 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3455 else 3456 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3457 } 3458 3459 if (attrzp) { 3460 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 3461 } 3462 3463 fuid_dirtied = zfsvfs->z_fuid_dirty; 3464 if (fuid_dirtied) 3465 zfs_fuid_txhold(zfsvfs, tx); 3466 3467 zfs_sa_upgrade_txholds(tx, zp); 3468 3469 err = dmu_tx_assign(tx, TXG_WAIT); 3470 if (err) 3471 goto out; 3472 3473 count = 0; 3474 /* 3475 * Set each attribute requested. 3476 * We group settings according to the locks they need to acquire. 3477 * 3478 * Note: you cannot set ctime directly, although it will be 3479 * updated as a side-effect of calling this function. 3480 */ 3481 3482 3483 if (mask & (AT_UID|AT_GID|AT_MODE)) 3484 mutex_enter(&zp->z_acl_lock); 3485 mutex_enter(&zp->z_lock); 3486 3487 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3488 &zp->z_pflags, sizeof (zp->z_pflags)); 3489 3490 if (attrzp) { 3491 if (mask & (AT_UID|AT_GID|AT_MODE)) 3492 mutex_enter(&attrzp->z_acl_lock); 3493 mutex_enter(&attrzp->z_lock); 3494 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3495 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3496 sizeof (attrzp->z_pflags)); 3497 } 3498 3499 if (mask & (AT_UID|AT_GID)) { 3500 3501 if (mask & AT_UID) { 3502 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3503 &new_uid, sizeof (new_uid)); 3504 zp->z_uid = new_uid; 3505 if (attrzp) { 3506 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3507 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3508 sizeof (new_uid)); 3509 attrzp->z_uid = new_uid; 3510 } 3511 } 3512 3513 if (mask & AT_GID) { 3514 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3515 NULL, &new_gid, sizeof (new_gid)); 3516 zp->z_gid = new_gid; 3517 if (attrzp) { 3518 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3519 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 3520 sizeof (new_gid)); 3521 attrzp->z_gid = new_gid; 3522 } 3523 } 3524 if (!(mask & AT_MODE)) { 3525 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 3526 NULL, &new_mode, sizeof (new_mode)); 3527 new_mode = zp->z_mode; 3528 } 3529 err = zfs_acl_chown_setattr(zp); 3530 ASSERT(err == 0); 3531 if (attrzp) { 3532 err = zfs_acl_chown_setattr(attrzp); 3533 ASSERT(err == 0); 3534 } 3535 } 3536 3537 if (mask & AT_MODE) { 3538 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 3539 &new_mode, sizeof (new_mode)); 3540 zp->z_mode = new_mode; 3541 ASSERT3U((uintptr_t)aclp, !=, 0); 3542 err = zfs_aclset_common(zp, aclp, cr, tx); 3543 ASSERT0(err); 3544 if (zp->z_acl_cached) 3545 zfs_acl_free(zp->z_acl_cached); 3546 zp->z_acl_cached = aclp; 3547 aclp = NULL; 3548 } 3549 3550 3551 if (mask & AT_ATIME) { 3552 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 3553 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 3554 &zp->z_atime, sizeof (zp->z_atime)); 3555 } 3556 3557 if (mask & AT_MTIME) { 3558 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 3559 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 3560 mtime, sizeof (mtime)); 3561 } 3562 3563 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 3564 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 3565 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 3566 NULL, mtime, sizeof (mtime)); 3567 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3568 &ctime, sizeof (ctime)); 3569 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 3570 B_TRUE); 3571 } else if (mask != 0) { 3572 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3573 &ctime, sizeof (ctime)); 3574 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime, 3575 B_TRUE); 3576 if (attrzp) { 3577 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3578 SA_ZPL_CTIME(zfsvfs), NULL, 3579 &ctime, sizeof (ctime)); 3580 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 3581 mtime, ctime, B_TRUE); 3582 } 3583 } 3584 /* 3585 * Do this after setting timestamps to prevent timestamp 3586 * update from toggling bit 3587 */ 3588 3589 if (xoap && (mask & AT_XVATTR)) { 3590 3591 /* 3592 * restore trimmed off masks 3593 * so that return masks can be set for caller. 3594 */ 3595 3596 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 3597 XVA_SET_REQ(xvap, XAT_APPENDONLY); 3598 } 3599 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 3600 XVA_SET_REQ(xvap, XAT_NOUNLINK); 3601 } 3602 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 3603 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 3604 } 3605 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 3606 XVA_SET_REQ(xvap, XAT_NODUMP); 3607 } 3608 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 3609 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 3610 } 3611 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 3612 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 3613 } 3614 3615 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3616 ASSERT(vp->v_type == VREG); 3617 3618 zfs_xvattr_set(zp, xvap, tx); 3619 } 3620 3621 if (fuid_dirtied) 3622 zfs_fuid_sync(zfsvfs, tx); 3623 3624 if (mask != 0) 3625 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 3626 3627 mutex_exit(&zp->z_lock); 3628 if (mask & (AT_UID|AT_GID|AT_MODE)) 3629 mutex_exit(&zp->z_acl_lock); 3630 3631 if (attrzp) { 3632 if (mask & (AT_UID|AT_GID|AT_MODE)) 3633 mutex_exit(&attrzp->z_acl_lock); 3634 mutex_exit(&attrzp->z_lock); 3635 } 3636out: 3637 if (err == 0 && attrzp) { 3638 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 3639 xattr_count, tx); 3640 ASSERT(err2 == 0); 3641 } 3642 3643 if (attrzp) 3644 VN_RELE(ZTOV(attrzp)); 3645 3646 if (aclp) 3647 zfs_acl_free(aclp); 3648 3649 if (fuidp) { 3650 zfs_fuid_info_free(fuidp); 3651 fuidp = NULL; 3652 } 3653 3654 if (err) { 3655 dmu_tx_abort(tx); 3656 if (err == ERESTART) 3657 goto top; 3658 } else { 3659 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3660 dmu_tx_commit(tx); 3661 } 3662 3663out2: 3664 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3665 zil_commit(zilog, 0); 3666 3667 ZFS_EXIT(zfsvfs); 3668 return (err); 3669} 3670 3671typedef struct zfs_zlock { 3672 krwlock_t *zl_rwlock; /* lock we acquired */ 3673 znode_t *zl_znode; /* znode we held */ 3674 struct zfs_zlock *zl_next; /* next in list */ 3675} zfs_zlock_t; 3676 3677/* 3678 * Drop locks and release vnodes that were held by zfs_rename_lock(). 3679 */ 3680static void 3681zfs_rename_unlock(zfs_zlock_t **zlpp) 3682{ 3683 zfs_zlock_t *zl; 3684 3685 while ((zl = *zlpp) != NULL) { 3686 if (zl->zl_znode != NULL) 3687 VN_RELE(ZTOV(zl->zl_znode)); 3688 rw_exit(zl->zl_rwlock); 3689 *zlpp = zl->zl_next; 3690 kmem_free(zl, sizeof (*zl)); 3691 } 3692} 3693 3694/* 3695 * Search back through the directory tree, using the ".." entries. 3696 * Lock each directory in the chain to prevent concurrent renames. 3697 * Fail any attempt to move a directory into one of its own descendants. 3698 * XXX - z_parent_lock can overlap with map or grow locks 3699 */ 3700static int 3701zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 3702{ 3703 zfs_zlock_t *zl; 3704 znode_t *zp = tdzp; 3705 uint64_t rootid = zp->z_zfsvfs->z_root; 3706 uint64_t oidp = zp->z_id; 3707 krwlock_t *rwlp = &szp->z_parent_lock; 3708 krw_t rw = RW_WRITER; 3709 3710 /* 3711 * First pass write-locks szp and compares to zp->z_id. 3712 * Later passes read-lock zp and compare to zp->z_parent. 3713 */ 3714 do { 3715 if (!rw_tryenter(rwlp, rw)) { 3716 /* 3717 * Another thread is renaming in this path. 3718 * Note that if we are a WRITER, we don't have any 3719 * parent_locks held yet. 3720 */ 3721 if (rw == RW_READER && zp->z_id > szp->z_id) { 3722 /* 3723 * Drop our locks and restart 3724 */ 3725 zfs_rename_unlock(&zl); 3726 *zlpp = NULL; 3727 zp = tdzp; 3728 oidp = zp->z_id; 3729 rwlp = &szp->z_parent_lock; 3730 rw = RW_WRITER; 3731 continue; 3732 } else { 3733 /* 3734 * Wait for other thread to drop its locks 3735 */ 3736 rw_enter(rwlp, rw); 3737 } 3738 } 3739 3740 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 3741 zl->zl_rwlock = rwlp; 3742 zl->zl_znode = NULL; 3743 zl->zl_next = *zlpp; 3744 *zlpp = zl; 3745 3746 if (oidp == szp->z_id) /* We're a descendant of szp */ 3747 return (SET_ERROR(EINVAL)); 3748 3749 if (oidp == rootid) /* We've hit the top */ 3750 return (0); 3751 3752 if (rw == RW_READER) { /* i.e. not the first pass */ 3753 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp); 3754 if (error) 3755 return (error); 3756 zl->zl_znode = zp; 3757 } 3758 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs), 3759 &oidp, sizeof (oidp)); 3760 rwlp = &zp->z_parent_lock; 3761 rw = RW_READER; 3762 3763 } while (zp->z_id != sdzp->z_id); 3764 3765 return (0); 3766} 3767 3768/* 3769 * Move an entry from the provided source directory to the target 3770 * directory. Change the entry name as indicated. 3771 * 3772 * IN: sdvp - Source directory containing the "old entry". 3773 * snm - Old entry name. 3774 * tdvp - Target directory to contain the "new entry". 3775 * tnm - New entry name. 3776 * cr - credentials of caller. 3777 * ct - caller context 3778 * flags - case flags 3779 * 3780 * RETURN: 0 on success, error code on failure. 3781 * 3782 * Timestamps: 3783 * sdvp,tdvp - ctime|mtime updated 3784 */ 3785/*ARGSUSED*/ 3786static int 3787zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 3788 caller_context_t *ct, int flags) 3789{ 3790 znode_t *tdzp, *sdzp, *szp, *tzp; 3791 zfsvfs_t *zfsvfs; 3792 zilog_t *zilog; 3793 vnode_t *realvp; 3794 zfs_dirlock_t *sdl, *tdl; 3795 dmu_tx_t *tx; 3796 zfs_zlock_t *zl; 3797 int cmp, serr, terr; 3798 int error = 0; 3799 int zflg = 0; 3800 boolean_t waited = B_FALSE; 3801 3802 tdzp = VTOZ(tdvp); 3803 ZFS_VERIFY_ZP(tdzp); 3804 zfsvfs = tdzp->z_zfsvfs; 3805 ZFS_ENTER(zfsvfs); 3806 zilog = zfsvfs->z_log; 3807 sdzp = VTOZ(sdvp); 3808 3809 /* 3810 * In case sdzp is not valid, let's be sure to exit from the right 3811 * zfsvfs_t. 3812 */ 3813 if (sdzp->z_sa_hdl == NULL) { 3814 ZFS_EXIT(zfsvfs); 3815 return (SET_ERROR(EIO)); 3816 } 3817 3818 /* 3819 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 3820 * ctldir appear to have the same v_vfsp. 3821 */ 3822 if (sdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) { 3823 ZFS_EXIT(zfsvfs); 3824 return (SET_ERROR(EXDEV)); 3825 } 3826 3827 if (zfsvfs->z_utf8 && u8_validate(tnm, 3828 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3829 ZFS_EXIT(zfsvfs); 3830 return (SET_ERROR(EILSEQ)); 3831 } 3832 3833 if (flags & FIGNORECASE) 3834 zflg |= ZCILOOK; 3835 3836top: 3837 szp = NULL; 3838 tzp = NULL; 3839 zl = NULL; 3840 3841 /* 3842 * This is to prevent the creation of links into attribute space 3843 * by renaming a linked file into/outof an attribute directory. 3844 * See the comment in zfs_link() for why this is considered bad. 3845 */ 3846 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3847 ZFS_EXIT(zfsvfs); 3848 return (SET_ERROR(EINVAL)); 3849 } 3850 3851 /* 3852 * Lock source and target directory entries. To prevent deadlock, 3853 * a lock ordering must be defined. We lock the directory with 3854 * the smallest object id first, or if it's a tie, the one with 3855 * the lexically first name. 3856 */ 3857 if (sdzp->z_id < tdzp->z_id) { 3858 cmp = -1; 3859 } else if (sdzp->z_id > tdzp->z_id) { 3860 cmp = 1; 3861 } else { 3862 /* 3863 * First compare the two name arguments without 3864 * considering any case folding. 3865 */ 3866 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 3867 3868 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 3869 ASSERT(error == 0 || !zfsvfs->z_utf8); 3870 if (cmp == 0) { 3871 /* 3872 * POSIX: "If the old argument and the new argument 3873 * both refer to links to the same existing file, 3874 * the rename() function shall return successfully 3875 * and perform no other action." 3876 */ 3877 ZFS_EXIT(zfsvfs); 3878 return (0); 3879 } 3880 /* 3881 * If the file system is case-folding, then we may 3882 * have some more checking to do. A case-folding file 3883 * system is either supporting mixed case sensitivity 3884 * access or is completely case-insensitive. Note 3885 * that the file system is always case preserving. 3886 * 3887 * In mixed sensitivity mode case sensitive behavior 3888 * is the default. FIGNORECASE must be used to 3889 * explicitly request case insensitive behavior. 3890 * 3891 * If the source and target names provided differ only 3892 * by case (e.g., a request to rename 'tim' to 'Tim'), 3893 * we will treat this as a special case in the 3894 * case-insensitive mode: as long as the source name 3895 * is an exact match, we will allow this to proceed as 3896 * a name-change request. 3897 */ 3898 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3899 (zfsvfs->z_case == ZFS_CASE_MIXED && 3900 flags & FIGNORECASE)) && 3901 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3902 &error) == 0) { 3903 /* 3904 * case preserving rename request, require exact 3905 * name matches 3906 */ 3907 zflg |= ZCIEXACT; 3908 zflg &= ~ZCILOOK; 3909 } 3910 } 3911 3912 /* 3913 * If the source and destination directories are the same, we should 3914 * grab the z_name_lock of that directory only once. 3915 */ 3916 if (sdzp == tdzp) { 3917 zflg |= ZHAVELOCK; 3918 rw_enter(&sdzp->z_name_lock, RW_READER); 3919 } 3920 3921 if (cmp < 0) { 3922 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3923 ZEXISTS | zflg, NULL, NULL); 3924 terr = zfs_dirent_lock(&tdl, 3925 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3926 } else { 3927 terr = zfs_dirent_lock(&tdl, 3928 tdzp, tnm, &tzp, zflg, NULL, NULL); 3929 serr = zfs_dirent_lock(&sdl, 3930 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3931 NULL, NULL); 3932 } 3933 3934 if (serr) { 3935 /* 3936 * Source entry invalid or not there. 3937 */ 3938 if (!terr) { 3939 zfs_dirent_unlock(tdl); 3940 if (tzp) 3941 VN_RELE(ZTOV(tzp)); 3942 } 3943 3944 if (sdzp == tdzp) 3945 rw_exit(&sdzp->z_name_lock); 3946 3947 /* 3948 * FreeBSD: In OpenSolaris they only check if rename source is 3949 * ".." here, because "." is handled in their lookup. This is 3950 * not the case for FreeBSD, so we check for "." explicitly. 3951 */ 3952 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 3953 serr = SET_ERROR(EINVAL); 3954 ZFS_EXIT(zfsvfs); 3955 return (serr); 3956 } 3957 if (terr) { 3958 zfs_dirent_unlock(sdl); 3959 VN_RELE(ZTOV(szp)); 3960 3961 if (sdzp == tdzp) 3962 rw_exit(&sdzp->z_name_lock); 3963 3964 if (strcmp(tnm, "..") == 0) 3965 terr = SET_ERROR(EINVAL); 3966 ZFS_EXIT(zfsvfs); 3967 return (terr); 3968 } 3969 3970 /* 3971 * Must have write access at the source to remove the old entry 3972 * and write access at the target to create the new entry. 3973 * Note that if target and source are the same, this can be 3974 * done in a single check. 3975 */ 3976 3977 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3978 goto out; 3979 3980 if (ZTOV(szp)->v_type == VDIR) { 3981 /* 3982 * Check to make sure rename is valid. 3983 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3984 */ 3985 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3986 goto out; 3987 } 3988 3989 /* 3990 * Does target exist? 3991 */ 3992 if (tzp) { 3993 /* 3994 * Source and target must be the same type. 3995 */ 3996 if (ZTOV(szp)->v_type == VDIR) { 3997 if (ZTOV(tzp)->v_type != VDIR) { 3998 error = SET_ERROR(ENOTDIR); 3999 goto out; 4000 } 4001 } else { 4002 if (ZTOV(tzp)->v_type == VDIR) { 4003 error = SET_ERROR(EISDIR); 4004 goto out; 4005 } 4006 } 4007 /* 4008 * POSIX dictates that when the source and target 4009 * entries refer to the same file object, rename 4010 * must do nothing and exit without error. 4011 */ 4012 if (szp->z_id == tzp->z_id) { 4013 error = 0; 4014 goto out; 4015 } 4016 } 4017 4018 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 4019 if (tzp) 4020 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 4021 4022 /* 4023 * notify the target directory if it is not the same 4024 * as source directory. 4025 */ 4026 if (tdvp != sdvp) { 4027 vnevent_rename_dest_dir(tdvp, ct); 4028 } 4029 4030 tx = dmu_tx_create(zfsvfs->z_os); 4031 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4032 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 4033 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 4034 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 4035 if (sdzp != tdzp) { 4036 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 4037 zfs_sa_upgrade_txholds(tx, tdzp); 4038 } 4039 if (tzp) { 4040 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 4041 zfs_sa_upgrade_txholds(tx, tzp); 4042 } 4043 4044 zfs_sa_upgrade_txholds(tx, szp); 4045 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 4046 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4047 if (error) { 4048 if (zl != NULL) 4049 zfs_rename_unlock(&zl); 4050 zfs_dirent_unlock(sdl); 4051 zfs_dirent_unlock(tdl); 4052 4053 if (sdzp == tdzp) 4054 rw_exit(&sdzp->z_name_lock); 4055 4056 VN_RELE(ZTOV(szp)); 4057 if (tzp) 4058 VN_RELE(ZTOV(tzp)); 4059 if (error == ERESTART) { 4060 waited = B_TRUE; 4061 dmu_tx_wait(tx); 4062 dmu_tx_abort(tx); 4063 goto top; 4064 } 4065 dmu_tx_abort(tx); 4066 ZFS_EXIT(zfsvfs); 4067 return (error); 4068 } 4069 4070 if (tzp) /* Attempt to remove the existing target */ 4071 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 4072 4073 if (error == 0) { 4074 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 4075 if (error == 0) { 4076 szp->z_pflags |= ZFS_AV_MODIFIED; 4077 4078 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 4079 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 4080 ASSERT0(error); 4081 4082 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 4083 if (error == 0) { 4084 zfs_log_rename(zilog, tx, TX_RENAME | 4085 (flags & FIGNORECASE ? TX_CI : 0), sdzp, 4086 sdl->dl_name, tdzp, tdl->dl_name, szp); 4087 4088 /* 4089 * Update path information for the target vnode 4090 */ 4091 vn_renamepath(tdvp, ZTOV(szp), tnm, 4092 strlen(tnm)); 4093 } else { 4094 /* 4095 * At this point, we have successfully created 4096 * the target name, but have failed to remove 4097 * the source name. Since the create was done 4098 * with the ZRENAMING flag, there are 4099 * complications; for one, the link count is 4100 * wrong. The easiest way to deal with this 4101 * is to remove the newly created target, and 4102 * return the original error. This must 4103 * succeed; fortunately, it is very unlikely to 4104 * fail, since we just created it. 4105 */ 4106 VERIFY3U(zfs_link_destroy(tdl, szp, tx, 4107 ZRENAMING, NULL), ==, 0); 4108 } 4109 } 4110#ifdef FREEBSD_NAMECACHE 4111 if (error == 0) { 4112 cache_purge(sdvp); 4113 cache_purge(tdvp); 4114 cache_purge(ZTOV(szp)); 4115 if (tzp) 4116 cache_purge(ZTOV(tzp)); 4117 } 4118#endif 4119 } 4120 4121 dmu_tx_commit(tx); 4122out: 4123 if (zl != NULL) 4124 zfs_rename_unlock(&zl); 4125 4126 zfs_dirent_unlock(sdl); 4127 zfs_dirent_unlock(tdl); 4128 4129 if (sdzp == tdzp) 4130 rw_exit(&sdzp->z_name_lock); 4131 4132 4133 VN_RELE(ZTOV(szp)); 4134 if (tzp) 4135 VN_RELE(ZTOV(tzp)); 4136 4137 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4138 zil_commit(zilog, 0); 4139 4140 ZFS_EXIT(zfsvfs); 4141 4142 return (error); 4143} 4144 4145/* 4146 * Insert the indicated symbolic reference entry into the directory. 4147 * 4148 * IN: dvp - Directory to contain new symbolic link. 4149 * link - Name for new symlink entry. 4150 * vap - Attributes of new entry. 4151 * cr - credentials of caller. 4152 * ct - caller context 4153 * flags - case flags 4154 * 4155 * RETURN: 0 on success, error code on failure. 4156 * 4157 * Timestamps: 4158 * dvp - ctime|mtime updated 4159 */ 4160/*ARGSUSED*/ 4161static int 4162zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, 4163 cred_t *cr, kthread_t *td) 4164{ 4165 znode_t *zp, *dzp = VTOZ(dvp); 4166 zfs_dirlock_t *dl; 4167 dmu_tx_t *tx; 4168 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4169 zilog_t *zilog; 4170 uint64_t len = strlen(link); 4171 int error; 4172 int zflg = ZNEW; 4173 zfs_acl_ids_t acl_ids; 4174 boolean_t fuid_dirtied; 4175 uint64_t txtype = TX_SYMLINK; 4176 boolean_t waited = B_FALSE; 4177 int flags = 0; 4178 4179 ASSERT(vap->va_type == VLNK); 4180 4181 ZFS_ENTER(zfsvfs); 4182 ZFS_VERIFY_ZP(dzp); 4183 zilog = zfsvfs->z_log; 4184 4185 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 4186 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4187 ZFS_EXIT(zfsvfs); 4188 return (SET_ERROR(EILSEQ)); 4189 } 4190 if (flags & FIGNORECASE) 4191 zflg |= ZCILOOK; 4192 4193 if (len > MAXPATHLEN) { 4194 ZFS_EXIT(zfsvfs); 4195 return (SET_ERROR(ENAMETOOLONG)); 4196 } 4197 4198 if ((error = zfs_acl_ids_create(dzp, 0, 4199 vap, cr, NULL, &acl_ids)) != 0) { 4200 ZFS_EXIT(zfsvfs); 4201 return (error); 4202 } 4203 4204 getnewvnode_reserve(1); 4205 4206top: 4207 /* 4208 * Attempt to lock directory; fail if entry already exists. 4209 */ 4210 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 4211 if (error) { 4212 zfs_acl_ids_free(&acl_ids); 4213 getnewvnode_drop_reserve(); 4214 ZFS_EXIT(zfsvfs); 4215 return (error); 4216 } 4217 4218 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4219 zfs_acl_ids_free(&acl_ids); 4220 zfs_dirent_unlock(dl); 4221 getnewvnode_drop_reserve(); 4222 ZFS_EXIT(zfsvfs); 4223 return (error); 4224 } 4225 4226 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 4227 zfs_acl_ids_free(&acl_ids); 4228 zfs_dirent_unlock(dl); 4229 getnewvnode_drop_reserve(); 4230 ZFS_EXIT(zfsvfs); 4231 return (SET_ERROR(EDQUOT)); 4232 } 4233 tx = dmu_tx_create(zfsvfs->z_os); 4234 fuid_dirtied = zfsvfs->z_fuid_dirty; 4235 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 4236 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4237 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 4238 ZFS_SA_BASE_ATTR_SIZE + len); 4239 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 4240 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 4241 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 4242 acl_ids.z_aclp->z_acl_bytes); 4243 } 4244 if (fuid_dirtied) 4245 zfs_fuid_txhold(zfsvfs, tx); 4246 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4247 if (error) { 4248 zfs_dirent_unlock(dl); 4249 if (error == ERESTART) { 4250 waited = B_TRUE; 4251 dmu_tx_wait(tx); 4252 dmu_tx_abort(tx); 4253 goto top; 4254 } 4255 zfs_acl_ids_free(&acl_ids); 4256 dmu_tx_abort(tx); 4257 getnewvnode_drop_reserve(); 4258 ZFS_EXIT(zfsvfs); 4259 return (error); 4260 } 4261 4262 /* 4263 * Create a new object for the symlink. 4264 * for version 4 ZPL datsets the symlink will be an SA attribute 4265 */ 4266 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 4267 4268 if (fuid_dirtied) 4269 zfs_fuid_sync(zfsvfs, tx); 4270 4271 mutex_enter(&zp->z_lock); 4272 if (zp->z_is_sa) 4273 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 4274 link, len, tx); 4275 else 4276 zfs_sa_symlink(zp, link, len, tx); 4277 mutex_exit(&zp->z_lock); 4278 4279 zp->z_size = len; 4280 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 4281 &zp->z_size, sizeof (zp->z_size), tx); 4282 /* 4283 * Insert the new object into the directory. 4284 */ 4285 (void) zfs_link_create(dl, zp, tx, ZNEW); 4286 4287 if (flags & FIGNORECASE) 4288 txtype |= TX_CI; 4289 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 4290 *vpp = ZTOV(zp); 4291 4292 zfs_acl_ids_free(&acl_ids); 4293 4294 dmu_tx_commit(tx); 4295 4296 getnewvnode_drop_reserve(); 4297 4298 zfs_dirent_unlock(dl); 4299 4300 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4301 zil_commit(zilog, 0); 4302 4303 ZFS_EXIT(zfsvfs); 4304 return (error); 4305} 4306 4307/* 4308 * Return, in the buffer contained in the provided uio structure, 4309 * the symbolic path referred to by vp. 4310 * 4311 * IN: vp - vnode of symbolic link. 4312 * uio - structure to contain the link path. 4313 * cr - credentials of caller. 4314 * ct - caller context 4315 * 4316 * OUT: uio - structure containing the link path. 4317 * 4318 * RETURN: 0 on success, error code on failure. 4319 * 4320 * Timestamps: 4321 * vp - atime updated 4322 */ 4323/* ARGSUSED */ 4324static int 4325zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 4326{ 4327 znode_t *zp = VTOZ(vp); 4328 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4329 int error; 4330 4331 ZFS_ENTER(zfsvfs); 4332 ZFS_VERIFY_ZP(zp); 4333 4334 mutex_enter(&zp->z_lock); 4335 if (zp->z_is_sa) 4336 error = sa_lookup_uio(zp->z_sa_hdl, 4337 SA_ZPL_SYMLINK(zfsvfs), uio); 4338 else 4339 error = zfs_sa_readlink(zp, uio); 4340 mutex_exit(&zp->z_lock); 4341 4342 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4343 4344 ZFS_EXIT(zfsvfs); 4345 return (error); 4346} 4347 4348/* 4349 * Insert a new entry into directory tdvp referencing svp. 4350 * 4351 * IN: tdvp - Directory to contain new entry. 4352 * svp - vnode of new entry. 4353 * name - name of new entry. 4354 * cr - credentials of caller. 4355 * ct - caller context 4356 * 4357 * RETURN: 0 on success, error code on failure. 4358 * 4359 * Timestamps: 4360 * tdvp - ctime|mtime updated 4361 * svp - ctime updated 4362 */ 4363/* ARGSUSED */ 4364static int 4365zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 4366 caller_context_t *ct, int flags) 4367{ 4368 znode_t *dzp = VTOZ(tdvp); 4369 znode_t *tzp, *szp; 4370 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4371 zilog_t *zilog; 4372 zfs_dirlock_t *dl; 4373 dmu_tx_t *tx; 4374 vnode_t *realvp; 4375 int error; 4376 int zf = ZNEW; 4377 uint64_t parent; 4378 uid_t owner; 4379 boolean_t waited = B_FALSE; 4380 4381 ASSERT(tdvp->v_type == VDIR); 4382 4383 ZFS_ENTER(zfsvfs); 4384 ZFS_VERIFY_ZP(dzp); 4385 zilog = zfsvfs->z_log; 4386 4387 if (VOP_REALVP(svp, &realvp, ct) == 0) 4388 svp = realvp; 4389 4390 /* 4391 * POSIX dictates that we return EPERM here. 4392 * Better choices include ENOTSUP or EISDIR. 4393 */ 4394 if (svp->v_type == VDIR) { 4395 ZFS_EXIT(zfsvfs); 4396 return (SET_ERROR(EPERM)); 4397 } 4398 4399 szp = VTOZ(svp); 4400 ZFS_VERIFY_ZP(szp); 4401 4402 if (szp->z_pflags & (ZFS_APPENDONLY | ZFS_IMMUTABLE | ZFS_READONLY)) { 4403 ZFS_EXIT(zfsvfs); 4404 return (SET_ERROR(EPERM)); 4405 } 4406 4407 /* 4408 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 4409 * ctldir appear to have the same v_vfsp. 4410 */ 4411 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) { 4412 ZFS_EXIT(zfsvfs); 4413 return (SET_ERROR(EXDEV)); 4414 } 4415 4416 /* Prevent links to .zfs/shares files */ 4417 4418 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 4419 &parent, sizeof (uint64_t))) != 0) { 4420 ZFS_EXIT(zfsvfs); 4421 return (error); 4422 } 4423 if (parent == zfsvfs->z_shares_dir) { 4424 ZFS_EXIT(zfsvfs); 4425 return (SET_ERROR(EPERM)); 4426 } 4427 4428 if (zfsvfs->z_utf8 && u8_validate(name, 4429 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4430 ZFS_EXIT(zfsvfs); 4431 return (SET_ERROR(EILSEQ)); 4432 } 4433 if (flags & FIGNORECASE) 4434 zf |= ZCILOOK; 4435 4436 /* 4437 * We do not support links between attributes and non-attributes 4438 * because of the potential security risk of creating links 4439 * into "normal" file space in order to circumvent restrictions 4440 * imposed in attribute space. 4441 */ 4442 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 4443 ZFS_EXIT(zfsvfs); 4444 return (SET_ERROR(EINVAL)); 4445 } 4446 4447 4448 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 4449 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) { 4450 ZFS_EXIT(zfsvfs); 4451 return (SET_ERROR(EPERM)); 4452 } 4453 4454 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4455 ZFS_EXIT(zfsvfs); 4456 return (error); 4457 } 4458 4459top: 4460 /* 4461 * Attempt to lock directory; fail if entry already exists. 4462 */ 4463 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 4464 if (error) { 4465 ZFS_EXIT(zfsvfs); 4466 return (error); 4467 } 4468 4469 tx = dmu_tx_create(zfsvfs->z_os); 4470 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4471 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4472 zfs_sa_upgrade_txholds(tx, szp); 4473 zfs_sa_upgrade_txholds(tx, dzp); 4474 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4475 if (error) { 4476 zfs_dirent_unlock(dl); 4477 if (error == ERESTART) { 4478 waited = B_TRUE; 4479 dmu_tx_wait(tx); 4480 dmu_tx_abort(tx); 4481 goto top; 4482 } 4483 dmu_tx_abort(tx); 4484 ZFS_EXIT(zfsvfs); 4485 return (error); 4486 } 4487 4488 error = zfs_link_create(dl, szp, tx, 0); 4489 4490 if (error == 0) { 4491 uint64_t txtype = TX_LINK; 4492 if (flags & FIGNORECASE) 4493 txtype |= TX_CI; 4494 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 4495 } 4496 4497 dmu_tx_commit(tx); 4498 4499 zfs_dirent_unlock(dl); 4500 4501 if (error == 0) { 4502 vnevent_link(svp, ct); 4503 } 4504 4505 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4506 zil_commit(zilog, 0); 4507 4508 ZFS_EXIT(zfsvfs); 4509 return (error); 4510} 4511 4512#ifdef illumos 4513/* 4514 * zfs_null_putapage() is used when the file system has been force 4515 * unmounted. It just drops the pages. 4516 */ 4517/* ARGSUSED */ 4518static int 4519zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4520 size_t *lenp, int flags, cred_t *cr) 4521{ 4522 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 4523 return (0); 4524} 4525 4526/* 4527 * Push a page out to disk, klustering if possible. 4528 * 4529 * IN: vp - file to push page to. 4530 * pp - page to push. 4531 * flags - additional flags. 4532 * cr - credentials of caller. 4533 * 4534 * OUT: offp - start of range pushed. 4535 * lenp - len of range pushed. 4536 * 4537 * RETURN: 0 on success, error code on failure. 4538 * 4539 * NOTE: callers must have locked the page to be pushed. On 4540 * exit, the page (and all other pages in the kluster) must be 4541 * unlocked. 4542 */ 4543/* ARGSUSED */ 4544static int 4545zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4546 size_t *lenp, int flags, cred_t *cr) 4547{ 4548 znode_t *zp = VTOZ(vp); 4549 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4550 dmu_tx_t *tx; 4551 u_offset_t off, koff; 4552 size_t len, klen; 4553 int err; 4554 4555 off = pp->p_offset; 4556 len = PAGESIZE; 4557 /* 4558 * If our blocksize is bigger than the page size, try to kluster 4559 * multiple pages so that we write a full block (thus avoiding 4560 * a read-modify-write). 4561 */ 4562 if (off < zp->z_size && zp->z_blksz > PAGESIZE) { 4563 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 4564 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 4565 ASSERT(koff <= zp->z_size); 4566 if (koff + klen > zp->z_size) 4567 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); 4568 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 4569 } 4570 ASSERT3U(btop(len), ==, btopr(len)); 4571 4572 /* 4573 * Can't push pages past end-of-file. 4574 */ 4575 if (off >= zp->z_size) { 4576 /* ignore all pages */ 4577 err = 0; 4578 goto out; 4579 } else if (off + len > zp->z_size) { 4580 int npages = btopr(zp->z_size - off); 4581 page_t *trunc; 4582 4583 page_list_break(&pp, &trunc, npages); 4584 /* ignore pages past end of file */ 4585 if (trunc) 4586 pvn_write_done(trunc, flags); 4587 len = zp->z_size - off; 4588 } 4589 4590 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4591 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4592 err = SET_ERROR(EDQUOT); 4593 goto out; 4594 } 4595 tx = dmu_tx_create(zfsvfs->z_os); 4596 dmu_tx_hold_write(tx, zp->z_id, off, len); 4597 4598 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4599 zfs_sa_upgrade_txholds(tx, zp); 4600 err = dmu_tx_assign(tx, TXG_WAIT); 4601 if (err != 0) { 4602 dmu_tx_abort(tx); 4603 goto out; 4604 } 4605 4606 if (zp->z_blksz <= PAGESIZE) { 4607 caddr_t va = zfs_map_page(pp, S_READ); 4608 ASSERT3U(len, <=, PAGESIZE); 4609 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 4610 zfs_unmap_page(pp, va); 4611 } else { 4612 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 4613 } 4614 4615 if (err == 0) { 4616 uint64_t mtime[2], ctime[2]; 4617 sa_bulk_attr_t bulk[3]; 4618 int count = 0; 4619 4620 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4621 &mtime, 16); 4622 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4623 &ctime, 16); 4624 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4625 &zp->z_pflags, 8); 4626 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4627 B_TRUE); 4628 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4629 } 4630 dmu_tx_commit(tx); 4631 4632out: 4633 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 4634 if (offp) 4635 *offp = off; 4636 if (lenp) 4637 *lenp = len; 4638 4639 return (err); 4640} 4641 4642/* 4643 * Copy the portion of the file indicated from pages into the file. 4644 * The pages are stored in a page list attached to the files vnode. 4645 * 4646 * IN: vp - vnode of file to push page data to. 4647 * off - position in file to put data. 4648 * len - amount of data to write. 4649 * flags - flags to control the operation. 4650 * cr - credentials of caller. 4651 * ct - caller context. 4652 * 4653 * RETURN: 0 on success, error code on failure. 4654 * 4655 * Timestamps: 4656 * vp - ctime|mtime updated 4657 */ 4658/*ARGSUSED*/ 4659static int 4660zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4661 caller_context_t *ct) 4662{ 4663 znode_t *zp = VTOZ(vp); 4664 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4665 page_t *pp; 4666 size_t io_len; 4667 u_offset_t io_off; 4668 uint_t blksz; 4669 rl_t *rl; 4670 int error = 0; 4671 4672 ZFS_ENTER(zfsvfs); 4673 ZFS_VERIFY_ZP(zp); 4674 4675 /* 4676 * Align this request to the file block size in case we kluster. 4677 * XXX - this can result in pretty aggresive locking, which can 4678 * impact simultanious read/write access. One option might be 4679 * to break up long requests (len == 0) into block-by-block 4680 * operations to get narrower locking. 4681 */ 4682 blksz = zp->z_blksz; 4683 if (ISP2(blksz)) 4684 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 4685 else 4686 io_off = 0; 4687 if (len > 0 && ISP2(blksz)) 4688 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 4689 else 4690 io_len = 0; 4691 4692 if (io_len == 0) { 4693 /* 4694 * Search the entire vp list for pages >= io_off. 4695 */ 4696 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 4697 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 4698 goto out; 4699 } 4700 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 4701 4702 if (off > zp->z_size) { 4703 /* past end of file */ 4704 zfs_range_unlock(rl); 4705 ZFS_EXIT(zfsvfs); 4706 return (0); 4707 } 4708 4709 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); 4710 4711 for (off = io_off; io_off < off + len; io_off += io_len) { 4712 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 4713 pp = page_lookup(vp, io_off, 4714 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 4715 } else { 4716 pp = page_lookup_nowait(vp, io_off, 4717 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 4718 } 4719 4720 if (pp != NULL && pvn_getdirty(pp, flags)) { 4721 int err; 4722 4723 /* 4724 * Found a dirty page to push 4725 */ 4726 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 4727 if (err) 4728 error = err; 4729 } else { 4730 io_len = PAGESIZE; 4731 } 4732 } 4733out: 4734 zfs_range_unlock(rl); 4735 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4736 zil_commit(zfsvfs->z_log, zp->z_id); 4737 ZFS_EXIT(zfsvfs); 4738 return (error); 4739} 4740#endif /* illumos */ 4741 4742/*ARGSUSED*/ 4743void 4744zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4745{ 4746 znode_t *zp = VTOZ(vp); 4747 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4748 int error; 4749 4750 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4751 if (zp->z_sa_hdl == NULL) { 4752 /* 4753 * The fs has been unmounted, or we did a 4754 * suspend/resume and this file no longer exists. 4755 */ 4756 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4757 vrecycle(vp); 4758 return; 4759 } 4760 4761 mutex_enter(&zp->z_lock); 4762 if (zp->z_unlinked) { 4763 /* 4764 * Fast path to recycle a vnode of a removed file. 4765 */ 4766 mutex_exit(&zp->z_lock); 4767 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4768 vrecycle(vp); 4769 return; 4770 } 4771 mutex_exit(&zp->z_lock); 4772 4773 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4774 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4775 4776 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4777 zfs_sa_upgrade_txholds(tx, zp); 4778 error = dmu_tx_assign(tx, TXG_WAIT); 4779 if (error) { 4780 dmu_tx_abort(tx); 4781 } else { 4782 mutex_enter(&zp->z_lock); 4783 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4784 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4785 zp->z_atime_dirty = 0; 4786 mutex_exit(&zp->z_lock); 4787 dmu_tx_commit(tx); 4788 } 4789 } 4790 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4791} 4792 4793#ifdef illumos 4794/* 4795 * Bounds-check the seek operation. 4796 * 4797 * IN: vp - vnode seeking within 4798 * ooff - old file offset 4799 * noffp - pointer to new file offset 4800 * ct - caller context 4801 * 4802 * RETURN: 0 on success, EINVAL if new offset invalid. 4803 */ 4804/* ARGSUSED */ 4805static int 4806zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 4807 caller_context_t *ct) 4808{ 4809 if (vp->v_type == VDIR) 4810 return (0); 4811 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4812} 4813 4814/* 4815 * Pre-filter the generic locking function to trap attempts to place 4816 * a mandatory lock on a memory mapped file. 4817 */ 4818static int 4819zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 4820 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 4821{ 4822 znode_t *zp = VTOZ(vp); 4823 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4824 4825 ZFS_ENTER(zfsvfs); 4826 ZFS_VERIFY_ZP(zp); 4827 4828 /* 4829 * We are following the UFS semantics with respect to mapcnt 4830 * here: If we see that the file is mapped already, then we will 4831 * return an error, but we don't worry about races between this 4832 * function and zfs_map(). 4833 */ 4834 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { 4835 ZFS_EXIT(zfsvfs); 4836 return (SET_ERROR(EAGAIN)); 4837 } 4838 ZFS_EXIT(zfsvfs); 4839 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4840} 4841 4842/* 4843 * If we can't find a page in the cache, we will create a new page 4844 * and fill it with file data. For efficiency, we may try to fill 4845 * multiple pages at once (klustering) to fill up the supplied page 4846 * list. Note that the pages to be filled are held with an exclusive 4847 * lock to prevent access by other threads while they are being filled. 4848 */ 4849static int 4850zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 4851 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 4852{ 4853 znode_t *zp = VTOZ(vp); 4854 page_t *pp, *cur_pp; 4855 objset_t *os = zp->z_zfsvfs->z_os; 4856 u_offset_t io_off, total; 4857 size_t io_len; 4858 int err; 4859 4860 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4861 /* 4862 * We only have a single page, don't bother klustering 4863 */ 4864 io_off = off; 4865 io_len = PAGESIZE; 4866 pp = page_create_va(vp, io_off, io_len, 4867 PG_EXCL | PG_WAIT, seg, addr); 4868 } else { 4869 /* 4870 * Try to find enough pages to fill the page list 4871 */ 4872 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4873 &io_len, off, plsz, 0); 4874 } 4875 if (pp == NULL) { 4876 /* 4877 * The page already exists, nothing to do here. 4878 */ 4879 *pl = NULL; 4880 return (0); 4881 } 4882 4883 /* 4884 * Fill the pages in the kluster. 4885 */ 4886 cur_pp = pp; 4887 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4888 caddr_t va; 4889 4890 ASSERT3U(io_off, ==, cur_pp->p_offset); 4891 va = zfs_map_page(cur_pp, S_WRITE); 4892 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4893 DMU_READ_PREFETCH); 4894 zfs_unmap_page(cur_pp, va); 4895 if (err) { 4896 /* On error, toss the entire kluster */ 4897 pvn_read_done(pp, B_ERROR); 4898 /* convert checksum errors into IO errors */ 4899 if (err == ECKSUM) 4900 err = SET_ERROR(EIO); 4901 return (err); 4902 } 4903 cur_pp = cur_pp->p_next; 4904 } 4905 4906 /* 4907 * Fill in the page list array from the kluster starting 4908 * from the desired offset `off'. 4909 * NOTE: the page list will always be null terminated. 4910 */ 4911 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4912 ASSERT(pl == NULL || (*pl)->p_offset == off); 4913 4914 return (0); 4915} 4916 4917/* 4918 * Return pointers to the pages for the file region [off, off + len] 4919 * in the pl array. If plsz is greater than len, this function may 4920 * also return page pointers from after the specified region 4921 * (i.e. the region [off, off + plsz]). These additional pages are 4922 * only returned if they are already in the cache, or were created as 4923 * part of a klustered read. 4924 * 4925 * IN: vp - vnode of file to get data from. 4926 * off - position in file to get data from. 4927 * len - amount of data to retrieve. 4928 * plsz - length of provided page list. 4929 * seg - segment to obtain pages for. 4930 * addr - virtual address of fault. 4931 * rw - mode of created pages. 4932 * cr - credentials of caller. 4933 * ct - caller context. 4934 * 4935 * OUT: protp - protection mode of created pages. 4936 * pl - list of pages created. 4937 * 4938 * RETURN: 0 on success, error code on failure. 4939 * 4940 * Timestamps: 4941 * vp - atime updated 4942 */ 4943/* ARGSUSED */ 4944static int 4945zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4946 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4947 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4948{ 4949 znode_t *zp = VTOZ(vp); 4950 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4951 page_t **pl0 = pl; 4952 int err = 0; 4953 4954 /* we do our own caching, faultahead is unnecessary */ 4955 if (pl == NULL) 4956 return (0); 4957 else if (len > plsz) 4958 len = plsz; 4959 else 4960 len = P2ROUNDUP(len, PAGESIZE); 4961 ASSERT(plsz >= len); 4962 4963 ZFS_ENTER(zfsvfs); 4964 ZFS_VERIFY_ZP(zp); 4965 4966 if (protp) 4967 *protp = PROT_ALL; 4968 4969 /* 4970 * Loop through the requested range [off, off + len) looking 4971 * for pages. If we don't find a page, we will need to create 4972 * a new page and fill it with data from the file. 4973 */ 4974 while (len > 0) { 4975 if (*pl = page_lookup(vp, off, SE_SHARED)) 4976 *(pl+1) = NULL; 4977 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 4978 goto out; 4979 while (*pl) { 4980 ASSERT3U((*pl)->p_offset, ==, off); 4981 off += PAGESIZE; 4982 addr += PAGESIZE; 4983 if (len > 0) { 4984 ASSERT3U(len, >=, PAGESIZE); 4985 len -= PAGESIZE; 4986 } 4987 ASSERT3U(plsz, >=, PAGESIZE); 4988 plsz -= PAGESIZE; 4989 pl++; 4990 } 4991 } 4992 4993 /* 4994 * Fill out the page array with any pages already in the cache. 4995 */ 4996 while (plsz > 0 && 4997 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 4998 off += PAGESIZE; 4999 plsz -= PAGESIZE; 5000 } 5001out: 5002 if (err) { 5003 /* 5004 * Release any pages we have previously locked. 5005 */ 5006 while (pl > pl0) 5007 page_unlock(*--pl); 5008 } else { 5009 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 5010 } 5011 5012 *pl = NULL; 5013 5014 ZFS_EXIT(zfsvfs); 5015 return (err); 5016} 5017 5018/* 5019 * Request a memory map for a section of a file. This code interacts 5020 * with common code and the VM system as follows: 5021 * 5022 * - common code calls mmap(), which ends up in smmap_common() 5023 * - this calls VOP_MAP(), which takes you into (say) zfs 5024 * - zfs_map() calls as_map(), passing segvn_create() as the callback 5025 * - segvn_create() creates the new segment and calls VOP_ADDMAP() 5026 * - zfs_addmap() updates z_mapcnt 5027 */ 5028/*ARGSUSED*/ 5029static int 5030zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 5031 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 5032 caller_context_t *ct) 5033{ 5034 znode_t *zp = VTOZ(vp); 5035 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5036 segvn_crargs_t vn_a; 5037 int error; 5038 5039 ZFS_ENTER(zfsvfs); 5040 ZFS_VERIFY_ZP(zp); 5041 5042 if ((prot & PROT_WRITE) && (zp->z_pflags & 5043 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 5044 ZFS_EXIT(zfsvfs); 5045 return (SET_ERROR(EPERM)); 5046 } 5047 5048 if ((prot & (PROT_READ | PROT_EXEC)) && 5049 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 5050 ZFS_EXIT(zfsvfs); 5051 return (SET_ERROR(EACCES)); 5052 } 5053 5054 if (vp->v_flag & VNOMAP) { 5055 ZFS_EXIT(zfsvfs); 5056 return (SET_ERROR(ENOSYS)); 5057 } 5058 5059 if (off < 0 || len > MAXOFFSET_T - off) { 5060 ZFS_EXIT(zfsvfs); 5061 return (SET_ERROR(ENXIO)); 5062 } 5063 5064 if (vp->v_type != VREG) { 5065 ZFS_EXIT(zfsvfs); 5066 return (SET_ERROR(ENODEV)); 5067 } 5068 5069 /* 5070 * If file is locked, disallow mapping. 5071 */ 5072 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { 5073 ZFS_EXIT(zfsvfs); 5074 return (SET_ERROR(EAGAIN)); 5075 } 5076 5077 as_rangelock(as); 5078 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 5079 if (error != 0) { 5080 as_rangeunlock(as); 5081 ZFS_EXIT(zfsvfs); 5082 return (error); 5083 } 5084 5085 vn_a.vp = vp; 5086 vn_a.offset = (u_offset_t)off; 5087 vn_a.type = flags & MAP_TYPE; 5088 vn_a.prot = prot; 5089 vn_a.maxprot = maxprot; 5090 vn_a.cred = cr; 5091 vn_a.amp = NULL; 5092 vn_a.flags = flags & ~MAP_TYPE; 5093 vn_a.szc = 0; 5094 vn_a.lgrp_mem_policy_flags = 0; 5095 5096 error = as_map(as, *addrp, len, segvn_create, &vn_a); 5097 5098 as_rangeunlock(as); 5099 ZFS_EXIT(zfsvfs); 5100 return (error); 5101} 5102 5103/* ARGSUSED */ 5104static int 5105zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5106 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 5107 caller_context_t *ct) 5108{ 5109 uint64_t pages = btopr(len); 5110 5111 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 5112 return (0); 5113} 5114 5115/* 5116 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 5117 * more accurate mtime for the associated file. Since we don't have a way of 5118 * detecting when the data was actually modified, we have to resort to 5119 * heuristics. If an explicit msync() is done, then we mark the mtime when the 5120 * last page is pushed. The problem occurs when the msync() call is omitted, 5121 * which by far the most common case: 5122 * 5123 * open() 5124 * mmap() 5125 * <modify memory> 5126 * munmap() 5127 * close() 5128 * <time lapse> 5129 * putpage() via fsflush 5130 * 5131 * If we wait until fsflush to come along, we can have a modification time that 5132 * is some arbitrary point in the future. In order to prevent this in the 5133 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 5134 * torn down. 5135 */ 5136/* ARGSUSED */ 5137static int 5138zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5139 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 5140 caller_context_t *ct) 5141{ 5142 uint64_t pages = btopr(len); 5143 5144 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 5145 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 5146 5147 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 5148 vn_has_cached_data(vp)) 5149 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 5150 5151 return (0); 5152} 5153 5154/* 5155 * Free or allocate space in a file. Currently, this function only 5156 * supports the `F_FREESP' command. However, this command is somewhat 5157 * misnamed, as its functionality includes the ability to allocate as 5158 * well as free space. 5159 * 5160 * IN: vp - vnode of file to free data in. 5161 * cmd - action to take (only F_FREESP supported). 5162 * bfp - section of file to free/alloc. 5163 * flag - current file open mode flags. 5164 * offset - current file offset. 5165 * cr - credentials of caller [UNUSED]. 5166 * ct - caller context. 5167 * 5168 * RETURN: 0 on success, error code on failure. 5169 * 5170 * Timestamps: 5171 * vp - ctime|mtime updated 5172 */ 5173/* ARGSUSED */ 5174static int 5175zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 5176 offset_t offset, cred_t *cr, caller_context_t *ct) 5177{ 5178 znode_t *zp = VTOZ(vp); 5179 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5180 uint64_t off, len; 5181 int error; 5182 5183 ZFS_ENTER(zfsvfs); 5184 ZFS_VERIFY_ZP(zp); 5185 5186 if (cmd != F_FREESP) { 5187 ZFS_EXIT(zfsvfs); 5188 return (SET_ERROR(EINVAL)); 5189 } 5190 5191 /* 5192 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our 5193 * callers might not be able to detect properly that we are read-only, 5194 * so check it explicitly here. 5195 */ 5196 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 5197 ZFS_EXIT(zfsvfs); 5198 return (SET_ERROR(EROFS)); 5199 } 5200 5201 if (error = convoff(vp, bfp, 0, offset)) { 5202 ZFS_EXIT(zfsvfs); 5203 return (error); 5204 } 5205 5206 if (bfp->l_len < 0) { 5207 ZFS_EXIT(zfsvfs); 5208 return (SET_ERROR(EINVAL)); 5209 } 5210 5211 off = bfp->l_start; 5212 len = bfp->l_len; /* 0 means from off to end of file */ 5213 5214 error = zfs_freesp(zp, off, len, flag, TRUE); 5215 5216 ZFS_EXIT(zfsvfs); 5217 return (error); 5218} 5219#endif /* illumos */ 5220 5221CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 5222CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 5223 5224/*ARGSUSED*/ 5225static int 5226zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 5227{ 5228 znode_t *zp = VTOZ(vp); 5229 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5230 uint32_t gen; 5231 uint64_t gen64; 5232 uint64_t object = zp->z_id; 5233 zfid_short_t *zfid; 5234 int size, i, error; 5235 5236 ZFS_ENTER(zfsvfs); 5237 ZFS_VERIFY_ZP(zp); 5238 5239 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 5240 &gen64, sizeof (uint64_t))) != 0) { 5241 ZFS_EXIT(zfsvfs); 5242 return (error); 5243 } 5244 5245 gen = (uint32_t)gen64; 5246 5247 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 5248 5249#ifdef illumos 5250 if (fidp->fid_len < size) { 5251 fidp->fid_len = size; 5252 ZFS_EXIT(zfsvfs); 5253 return (SET_ERROR(ENOSPC)); 5254 } 5255#else 5256 fidp->fid_len = size; 5257#endif 5258 5259 zfid = (zfid_short_t *)fidp; 5260 5261 zfid->zf_len = size; 5262 5263 for (i = 0; i < sizeof (zfid->zf_object); i++) 5264 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 5265 5266 /* Must have a non-zero generation number to distinguish from .zfs */ 5267 if (gen == 0) 5268 gen = 1; 5269 for (i = 0; i < sizeof (zfid->zf_gen); i++) 5270 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 5271 5272 if (size == LONG_FID_LEN) { 5273 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 5274 zfid_long_t *zlfid; 5275 5276 zlfid = (zfid_long_t *)fidp; 5277 5278 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 5279 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 5280 5281 /* XXX - this should be the generation number for the objset */ 5282 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 5283 zlfid->zf_setgen[i] = 0; 5284 } 5285 5286 ZFS_EXIT(zfsvfs); 5287 return (0); 5288} 5289 5290static int 5291zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 5292 caller_context_t *ct) 5293{ 5294 znode_t *zp, *xzp; 5295 zfsvfs_t *zfsvfs; 5296 zfs_dirlock_t *dl; 5297 int error; 5298 5299 switch (cmd) { 5300 case _PC_LINK_MAX: 5301 *valp = INT_MAX; 5302 return (0); 5303 5304 case _PC_FILESIZEBITS: 5305 *valp = 64; 5306 return (0); 5307#ifdef illumos 5308 case _PC_XATTR_EXISTS: 5309 zp = VTOZ(vp); 5310 zfsvfs = zp->z_zfsvfs; 5311 ZFS_ENTER(zfsvfs); 5312 ZFS_VERIFY_ZP(zp); 5313 *valp = 0; 5314 error = zfs_dirent_lock(&dl, zp, "", &xzp, 5315 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 5316 if (error == 0) { 5317 zfs_dirent_unlock(dl); 5318 if (!zfs_dirempty(xzp)) 5319 *valp = 1; 5320 VN_RELE(ZTOV(xzp)); 5321 } else if (error == ENOENT) { 5322 /* 5323 * If there aren't extended attributes, it's the 5324 * same as having zero of them. 5325 */ 5326 error = 0; 5327 } 5328 ZFS_EXIT(zfsvfs); 5329 return (error); 5330 5331 case _PC_SATTR_ENABLED: 5332 case _PC_SATTR_EXISTS: 5333 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 5334 (vp->v_type == VREG || vp->v_type == VDIR); 5335 return (0); 5336 5337 case _PC_ACCESS_FILTERING: 5338 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 5339 vp->v_type == VDIR; 5340 return (0); 5341 5342 case _PC_ACL_ENABLED: 5343 *valp = _ACL_ACE_ENABLED; 5344 return (0); 5345#endif /* illumos */ 5346 case _PC_MIN_HOLE_SIZE: 5347 *valp = (int)SPA_MINBLOCKSIZE; 5348 return (0); 5349#ifdef illumos 5350 case _PC_TIMESTAMP_RESOLUTION: 5351 /* nanosecond timestamp resolution */ 5352 *valp = 1L; 5353 return (0); 5354#endif 5355 case _PC_ACL_EXTENDED: 5356 *valp = 0; 5357 return (0); 5358 5359 case _PC_ACL_NFS4: 5360 *valp = 1; 5361 return (0); 5362 5363 case _PC_ACL_PATH_MAX: 5364 *valp = ACL_MAX_ENTRIES; 5365 return (0); 5366 5367 default: 5368 return (EOPNOTSUPP); 5369 } 5370} 5371 5372/*ARGSUSED*/ 5373static int 5374zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5375 caller_context_t *ct) 5376{ 5377 znode_t *zp = VTOZ(vp); 5378 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5379 int error; 5380 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5381 5382 ZFS_ENTER(zfsvfs); 5383 ZFS_VERIFY_ZP(zp); 5384 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 5385 ZFS_EXIT(zfsvfs); 5386 5387 return (error); 5388} 5389 5390/*ARGSUSED*/ 5391int 5392zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5393 caller_context_t *ct) 5394{ 5395 znode_t *zp = VTOZ(vp); 5396 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5397 int error; 5398 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5399 zilog_t *zilog = zfsvfs->z_log; 5400 5401 ZFS_ENTER(zfsvfs); 5402 ZFS_VERIFY_ZP(zp); 5403 5404 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 5405 5406 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 5407 zil_commit(zilog, 0); 5408 5409 ZFS_EXIT(zfsvfs); 5410 return (error); 5411} 5412 5413#ifdef illumos 5414/* 5415 * The smallest read we may consider to loan out an arcbuf. 5416 * This must be a power of 2. 5417 */ 5418int zcr_blksz_min = (1 << 10); /* 1K */ 5419/* 5420 * If set to less than the file block size, allow loaning out of an 5421 * arcbuf for a partial block read. This must be a power of 2. 5422 */ 5423int zcr_blksz_max = (1 << 17); /* 128K */ 5424 5425/*ARGSUSED*/ 5426static int 5427zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr, 5428 caller_context_t *ct) 5429{ 5430 znode_t *zp = VTOZ(vp); 5431 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5432 int max_blksz = zfsvfs->z_max_blksz; 5433 uio_t *uio = &xuio->xu_uio; 5434 ssize_t size = uio->uio_resid; 5435 offset_t offset = uio->uio_loffset; 5436 int blksz; 5437 int fullblk, i; 5438 arc_buf_t *abuf; 5439 ssize_t maxsize; 5440 int preamble, postamble; 5441 5442 if (xuio->xu_type != UIOTYPE_ZEROCOPY) 5443 return (SET_ERROR(EINVAL)); 5444 5445 ZFS_ENTER(zfsvfs); 5446 ZFS_VERIFY_ZP(zp); 5447 switch (ioflag) { 5448 case UIO_WRITE: 5449 /* 5450 * Loan out an arc_buf for write if write size is bigger than 5451 * max_blksz, and the file's block size is also max_blksz. 5452 */ 5453 blksz = max_blksz; 5454 if (size < blksz || zp->z_blksz != blksz) { 5455 ZFS_EXIT(zfsvfs); 5456 return (SET_ERROR(EINVAL)); 5457 } 5458 /* 5459 * Caller requests buffers for write before knowing where the 5460 * write offset might be (e.g. NFS TCP write). 5461 */ 5462 if (offset == -1) { 5463 preamble = 0; 5464 } else { 5465 preamble = P2PHASE(offset, blksz); 5466 if (preamble) { 5467 preamble = blksz - preamble; 5468 size -= preamble; 5469 } 5470 } 5471 5472 postamble = P2PHASE(size, blksz); 5473 size -= postamble; 5474 5475 fullblk = size / blksz; 5476 (void) dmu_xuio_init(xuio, 5477 (preamble != 0) + fullblk + (postamble != 0)); 5478 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble, 5479 int, postamble, int, 5480 (preamble != 0) + fullblk + (postamble != 0)); 5481 5482 /* 5483 * Have to fix iov base/len for partial buffers. They 5484 * currently represent full arc_buf's. 5485 */ 5486 if (preamble) { 5487 /* data begins in the middle of the arc_buf */ 5488 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5489 blksz); 5490 ASSERT(abuf); 5491 (void) dmu_xuio_add(xuio, abuf, 5492 blksz - preamble, preamble); 5493 } 5494 5495 for (i = 0; i < fullblk; i++) { 5496 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5497 blksz); 5498 ASSERT(abuf); 5499 (void) dmu_xuio_add(xuio, abuf, 0, blksz); 5500 } 5501 5502 if (postamble) { 5503 /* data ends in the middle of the arc_buf */ 5504 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5505 blksz); 5506 ASSERT(abuf); 5507 (void) dmu_xuio_add(xuio, abuf, 0, postamble); 5508 } 5509 break; 5510 case UIO_READ: 5511 /* 5512 * Loan out an arc_buf for read if the read size is larger than 5513 * the current file block size. Block alignment is not 5514 * considered. Partial arc_buf will be loaned out for read. 5515 */ 5516 blksz = zp->z_blksz; 5517 if (blksz < zcr_blksz_min) 5518 blksz = zcr_blksz_min; 5519 if (blksz > zcr_blksz_max) 5520 blksz = zcr_blksz_max; 5521 /* avoid potential complexity of dealing with it */ 5522 if (blksz > max_blksz) { 5523 ZFS_EXIT(zfsvfs); 5524 return (SET_ERROR(EINVAL)); 5525 } 5526 5527 maxsize = zp->z_size - uio->uio_loffset; 5528 if (size > maxsize) 5529 size = maxsize; 5530 5531 if (size < blksz || vn_has_cached_data(vp)) { 5532 ZFS_EXIT(zfsvfs); 5533 return (SET_ERROR(EINVAL)); 5534 } 5535 break; 5536 default: 5537 ZFS_EXIT(zfsvfs); 5538 return (SET_ERROR(EINVAL)); 5539 } 5540 5541 uio->uio_extflg = UIO_XUIO; 5542 XUIO_XUZC_RW(xuio) = ioflag; 5543 ZFS_EXIT(zfsvfs); 5544 return (0); 5545} 5546 5547/*ARGSUSED*/ 5548static int 5549zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct) 5550{ 5551 int i; 5552 arc_buf_t *abuf; 5553 int ioflag = XUIO_XUZC_RW(xuio); 5554 5555 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY); 5556 5557 i = dmu_xuio_cnt(xuio); 5558 while (i-- > 0) { 5559 abuf = dmu_xuio_arcbuf(xuio, i); 5560 /* 5561 * if abuf == NULL, it must be a write buffer 5562 * that has been returned in zfs_write(). 5563 */ 5564 if (abuf) 5565 dmu_return_arcbuf(abuf); 5566 ASSERT(abuf || ioflag == UIO_WRITE); 5567 } 5568 5569 dmu_xuio_fini(xuio); 5570 return (0); 5571} 5572 5573/* 5574 * Predeclare these here so that the compiler assumes that 5575 * this is an "old style" function declaration that does 5576 * not include arguments => we won't get type mismatch errors 5577 * in the initializations that follow. 5578 */ 5579static int zfs_inval(); 5580static int zfs_isdir(); 5581 5582static int 5583zfs_inval() 5584{ 5585 return (SET_ERROR(EINVAL)); 5586} 5587 5588static int 5589zfs_isdir() 5590{ 5591 return (SET_ERROR(EISDIR)); 5592} 5593/* 5594 * Directory vnode operations template 5595 */ 5596vnodeops_t *zfs_dvnodeops; 5597const fs_operation_def_t zfs_dvnodeops_template[] = { 5598 VOPNAME_OPEN, { .vop_open = zfs_open }, 5599 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5600 VOPNAME_READ, { .error = zfs_isdir }, 5601 VOPNAME_WRITE, { .error = zfs_isdir }, 5602 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5603 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5604 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5605 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5606 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5607 VOPNAME_CREATE, { .vop_create = zfs_create }, 5608 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5609 VOPNAME_LINK, { .vop_link = zfs_link }, 5610 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5611 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 5612 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5613 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5614 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 5615 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5616 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5617 VOPNAME_FID, { .vop_fid = zfs_fid }, 5618 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5619 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5620 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5621 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5622 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5623 NULL, NULL 5624}; 5625 5626/* 5627 * Regular file vnode operations template 5628 */ 5629vnodeops_t *zfs_fvnodeops; 5630const fs_operation_def_t zfs_fvnodeops_template[] = { 5631 VOPNAME_OPEN, { .vop_open = zfs_open }, 5632 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5633 VOPNAME_READ, { .vop_read = zfs_read }, 5634 VOPNAME_WRITE, { .vop_write = zfs_write }, 5635 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5636 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5637 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5638 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5639 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5640 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5641 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5642 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5643 VOPNAME_FID, { .vop_fid = zfs_fid }, 5644 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5645 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 5646 VOPNAME_SPACE, { .vop_space = zfs_space }, 5647 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 5648 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 5649 VOPNAME_MAP, { .vop_map = zfs_map }, 5650 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 5651 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 5652 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5653 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5654 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5655 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5656 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf }, 5657 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf }, 5658 NULL, NULL 5659}; 5660 5661/* 5662 * Symbolic link vnode operations template 5663 */ 5664vnodeops_t *zfs_symvnodeops; 5665const fs_operation_def_t zfs_symvnodeops_template[] = { 5666 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5667 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5668 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5669 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5670 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 5671 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5672 VOPNAME_FID, { .vop_fid = zfs_fid }, 5673 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5674 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5675 NULL, NULL 5676}; 5677 5678/* 5679 * special share hidden files vnode operations template 5680 */ 5681vnodeops_t *zfs_sharevnodeops; 5682const fs_operation_def_t zfs_sharevnodeops_template[] = { 5683 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5684 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5685 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5686 VOPNAME_FID, { .vop_fid = zfs_fid }, 5687 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5688 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5689 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5690 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5691 NULL, NULL 5692}; 5693 5694/* 5695 * Extended attribute directory vnode operations template 5696 * 5697 * This template is identical to the directory vnodes 5698 * operation template except for restricted operations: 5699 * VOP_MKDIR() 5700 * VOP_SYMLINK() 5701 * 5702 * Note that there are other restrictions embedded in: 5703 * zfs_create() - restrict type to VREG 5704 * zfs_link() - no links into/out of attribute space 5705 * zfs_rename() - no moves into/out of attribute space 5706 */ 5707vnodeops_t *zfs_xdvnodeops; 5708const fs_operation_def_t zfs_xdvnodeops_template[] = { 5709 VOPNAME_OPEN, { .vop_open = zfs_open }, 5710 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5711 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5712 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5713 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5714 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5715 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5716 VOPNAME_CREATE, { .vop_create = zfs_create }, 5717 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5718 VOPNAME_LINK, { .vop_link = zfs_link }, 5719 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5720 VOPNAME_MKDIR, { .error = zfs_inval }, 5721 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5722 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5723 VOPNAME_SYMLINK, { .error = zfs_inval }, 5724 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5725 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5726 VOPNAME_FID, { .vop_fid = zfs_fid }, 5727 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5728 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5729 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5730 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5731 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5732 NULL, NULL 5733}; 5734 5735/* 5736 * Error vnode operations template 5737 */ 5738vnodeops_t *zfs_evnodeops; 5739const fs_operation_def_t zfs_evnodeops_template[] = { 5740 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5741 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5742 NULL, NULL 5743}; 5744#endif /* illumos */ 5745 5746static int 5747ioflags(int ioflags) 5748{ 5749 int flags = 0; 5750 5751 if (ioflags & IO_APPEND) 5752 flags |= FAPPEND; 5753 if (ioflags & IO_NDELAY) 5754 flags |= FNONBLOCK; 5755 if (ioflags & IO_SYNC) 5756 flags |= (FSYNC | FDSYNC | FRSYNC); 5757 5758 return (flags); 5759} 5760 5761static int 5762zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage) 5763{ 5764 znode_t *zp = VTOZ(vp); 5765 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5766 objset_t *os = zp->z_zfsvfs->z_os; 5767 vm_page_t mfirst, mlast, mreq; 5768 vm_object_t object; 5769 caddr_t va; 5770 struct sf_buf *sf; 5771 off_t startoff, endoff; 5772 int i, error; 5773 vm_pindex_t reqstart, reqend; 5774 int pcount, lsize, reqsize, size; 5775 5776 ZFS_ENTER(zfsvfs); 5777 ZFS_VERIFY_ZP(zp); 5778 5779 pcount = OFF_TO_IDX(round_page(count)); 5780 mreq = m[reqpage]; 5781 object = mreq->object; 5782 error = 0; 5783 5784 KASSERT(vp->v_object == object, ("mismatching object")); 5785 5786 if (pcount > 1 && zp->z_blksz > PAGESIZE) { 5787 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz); 5788 reqstart = OFF_TO_IDX(round_page(startoff)); 5789 if (reqstart < m[0]->pindex) 5790 reqstart = 0; 5791 else 5792 reqstart = reqstart - m[0]->pindex; 5793 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE, 5794 zp->z_blksz); 5795 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1; 5796 if (reqend > m[pcount - 1]->pindex) 5797 reqend = m[pcount - 1]->pindex; 5798 reqsize = reqend - m[reqstart]->pindex + 1; 5799 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize, 5800 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds")); 5801 } else { 5802 reqstart = reqpage; 5803 reqsize = 1; 5804 } 5805 mfirst = m[reqstart]; 5806 mlast = m[reqstart + reqsize - 1]; 5807 5808 zfs_vmobject_wlock(object); 5809 5810 for (i = 0; i < reqstart; i++) { 5811 vm_page_lock(m[i]); 5812 vm_page_free(m[i]); 5813 vm_page_unlock(m[i]); 5814 } 5815 for (i = reqstart + reqsize; i < pcount; i++) { 5816 vm_page_lock(m[i]); 5817 vm_page_free(m[i]); 5818 vm_page_unlock(m[i]); 5819 } 5820 5821 if (mreq->valid && reqsize == 1) { 5822 if (mreq->valid != VM_PAGE_BITS_ALL) 5823 vm_page_zero_invalid(mreq, TRUE); 5824 zfs_vmobject_wunlock(object); 5825 ZFS_EXIT(zfsvfs); 5826 return (zfs_vm_pagerret_ok); 5827 } 5828 5829 PCPU_INC(cnt.v_vnodein); 5830 PCPU_ADD(cnt.v_vnodepgsin, reqsize); 5831 5832 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) { 5833 for (i = reqstart; i < reqstart + reqsize; i++) { 5834 if (i != reqpage) { 5835 vm_page_lock(m[i]); 5836 vm_page_free(m[i]); 5837 vm_page_unlock(m[i]); 5838 } 5839 } 5840 zfs_vmobject_wunlock(object); 5841 ZFS_EXIT(zfsvfs); 5842 return (zfs_vm_pagerret_bad); 5843 } 5844 5845 lsize = PAGE_SIZE; 5846 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size) 5847 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex); 5848 5849 zfs_vmobject_wunlock(object); 5850 5851 for (i = reqstart; i < reqstart + reqsize; i++) { 5852 size = PAGE_SIZE; 5853 if (i == (reqstart + reqsize - 1)) 5854 size = lsize; 5855 va = zfs_map_page(m[i], &sf); 5856 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex), 5857 size, va, DMU_READ_PREFETCH); 5858 if (size != PAGE_SIZE) 5859 bzero(va + size, PAGE_SIZE - size); 5860 zfs_unmap_page(sf); 5861 if (error != 0) 5862 break; 5863 } 5864 5865 zfs_vmobject_wlock(object); 5866 5867 for (i = reqstart; i < reqstart + reqsize; i++) { 5868 if (!error) 5869 m[i]->valid = VM_PAGE_BITS_ALL; 5870 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i])); 5871 if (i != reqpage) 5872 vm_page_readahead_finish(m[i]); 5873 } 5874 5875 zfs_vmobject_wunlock(object); 5876 5877 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 5878 ZFS_EXIT(zfsvfs); 5879 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok); 5880} 5881 5882static int 5883zfs_freebsd_getpages(ap) 5884 struct vop_getpages_args /* { 5885 struct vnode *a_vp; 5886 vm_page_t *a_m; 5887 int a_count; 5888 int a_reqpage; 5889 vm_ooffset_t a_offset; 5890 } */ *ap; 5891{ 5892 5893 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage)); 5894} 5895 5896static int 5897zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags, 5898 int *rtvals) 5899{ 5900 znode_t *zp = VTOZ(vp); 5901 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5902 rl_t *rl; 5903 dmu_tx_t *tx; 5904 struct sf_buf *sf; 5905 vm_object_t object; 5906 vm_page_t m; 5907 caddr_t va; 5908 size_t tocopy; 5909 size_t lo_len; 5910 vm_ooffset_t lo_off; 5911 vm_ooffset_t off; 5912 uint_t blksz; 5913 int ncount; 5914 int pcount; 5915 int err; 5916 int i; 5917 5918 ZFS_ENTER(zfsvfs); 5919 ZFS_VERIFY_ZP(zp); 5920 5921 object = vp->v_object; 5922 pcount = btoc(len); 5923 ncount = pcount; 5924 5925 KASSERT(ma[0]->object == object, ("mismatching object")); 5926 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length")); 5927 5928 for (i = 0; i < pcount; i++) 5929 rtvals[i] = zfs_vm_pagerret_error; 5930 5931 off = IDX_TO_OFF(ma[0]->pindex); 5932 blksz = zp->z_blksz; 5933 lo_off = rounddown(off, blksz); 5934 lo_len = roundup(len + (off - lo_off), blksz); 5935 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER); 5936 5937 zfs_vmobject_wlock(object); 5938 if (len + off > object->un_pager.vnp.vnp_size) { 5939 if (object->un_pager.vnp.vnp_size > off) { 5940 int pgoff; 5941 5942 len = object->un_pager.vnp.vnp_size - off; 5943 ncount = btoc(len); 5944 if ((pgoff = (int)len & PAGE_MASK) != 0) { 5945 /* 5946 * If the object is locked and the following 5947 * conditions hold, then the page's dirty 5948 * field cannot be concurrently changed by a 5949 * pmap operation. 5950 */ 5951 m = ma[ncount - 1]; 5952 vm_page_assert_sbusied(m); 5953 KASSERT(!pmap_page_is_write_mapped(m), 5954 ("zfs_putpages: page %p is not read-only", m)); 5955 vm_page_clear_dirty(m, pgoff, PAGE_SIZE - 5956 pgoff); 5957 } 5958 } else { 5959 len = 0; 5960 ncount = 0; 5961 } 5962 if (ncount < pcount) { 5963 for (i = ncount; i < pcount; i++) { 5964 rtvals[i] = zfs_vm_pagerret_bad; 5965 } 5966 } 5967 } 5968 zfs_vmobject_wunlock(object); 5969 5970 if (ncount == 0) 5971 goto out; 5972 5973 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 5974 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 5975 goto out; 5976 } 5977 5978top: 5979 tx = dmu_tx_create(zfsvfs->z_os); 5980 dmu_tx_hold_write(tx, zp->z_id, off, len); 5981 5982 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 5983 zfs_sa_upgrade_txholds(tx, zp); 5984 err = dmu_tx_assign(tx, TXG_NOWAIT); 5985 if (err != 0) { 5986 if (err == ERESTART) { 5987 dmu_tx_wait(tx); 5988 dmu_tx_abort(tx); 5989 goto top; 5990 } 5991 dmu_tx_abort(tx); 5992 goto out; 5993 } 5994 5995 if (zp->z_blksz < PAGE_SIZE) { 5996 i = 0; 5997 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) { 5998 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len; 5999 va = zfs_map_page(ma[i], &sf); 6000 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx); 6001 zfs_unmap_page(sf); 6002 } 6003 } else { 6004 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx); 6005 } 6006 6007 if (err == 0) { 6008 uint64_t mtime[2], ctime[2]; 6009 sa_bulk_attr_t bulk[3]; 6010 int count = 0; 6011 6012 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 6013 &mtime, 16); 6014 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 6015 &ctime, 16); 6016 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 6017 &zp->z_pflags, 8); 6018 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 6019 B_TRUE); 6020 (void)sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 6021 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 6022 6023 zfs_vmobject_wlock(object); 6024 for (i = 0; i < ncount; i++) { 6025 rtvals[i] = zfs_vm_pagerret_ok; 6026 vm_page_undirty(ma[i]); 6027 } 6028 zfs_vmobject_wunlock(object); 6029 PCPU_INC(cnt.v_vnodeout); 6030 PCPU_ADD(cnt.v_vnodepgsout, ncount); 6031 } 6032 dmu_tx_commit(tx); 6033 6034out: 6035 zfs_range_unlock(rl); 6036 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 || 6037 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 6038 zil_commit(zfsvfs->z_log, zp->z_id); 6039 ZFS_EXIT(zfsvfs); 6040 return (rtvals[0]); 6041} 6042 6043int 6044zfs_freebsd_putpages(ap) 6045 struct vop_putpages_args /* { 6046 struct vnode *a_vp; 6047 vm_page_t *a_m; 6048 int a_count; 6049 int a_sync; 6050 int *a_rtvals; 6051 vm_ooffset_t a_offset; 6052 } */ *ap; 6053{ 6054 6055 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync, 6056 ap->a_rtvals)); 6057} 6058 6059static int 6060zfs_freebsd_bmap(ap) 6061 struct vop_bmap_args /* { 6062 struct vnode *a_vp; 6063 daddr_t a_bn; 6064 struct bufobj **a_bop; 6065 daddr_t *a_bnp; 6066 int *a_runp; 6067 int *a_runb; 6068 } */ *ap; 6069{ 6070 6071 if (ap->a_bop != NULL) 6072 *ap->a_bop = &ap->a_vp->v_bufobj; 6073 if (ap->a_bnp != NULL) 6074 *ap->a_bnp = ap->a_bn; 6075 if (ap->a_runp != NULL) 6076 *ap->a_runp = 0; 6077 if (ap->a_runb != NULL) 6078 *ap->a_runb = 0; 6079 6080 return (0); 6081} 6082 6083static int 6084zfs_freebsd_open(ap) 6085 struct vop_open_args /* { 6086 struct vnode *a_vp; 6087 int a_mode; 6088 struct ucred *a_cred; 6089 struct thread *a_td; 6090 } */ *ap; 6091{ 6092 vnode_t *vp = ap->a_vp; 6093 znode_t *zp = VTOZ(vp); 6094 int error; 6095 6096 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL); 6097 if (error == 0) 6098 vnode_create_vobject(vp, zp->z_size, ap->a_td); 6099 return (error); 6100} 6101 6102static int 6103zfs_freebsd_close(ap) 6104 struct vop_close_args /* { 6105 struct vnode *a_vp; 6106 int a_fflag; 6107 struct ucred *a_cred; 6108 struct thread *a_td; 6109 } */ *ap; 6110{ 6111 6112 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL)); 6113} 6114 6115static int 6116zfs_freebsd_ioctl(ap) 6117 struct vop_ioctl_args /* { 6118 struct vnode *a_vp; 6119 u_long a_command; 6120 caddr_t a_data; 6121 int a_fflag; 6122 struct ucred *cred; 6123 struct thread *td; 6124 } */ *ap; 6125{ 6126 6127 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 6128 ap->a_fflag, ap->a_cred, NULL, NULL)); 6129} 6130 6131static int 6132zfs_freebsd_read(ap) 6133 struct vop_read_args /* { 6134 struct vnode *a_vp; 6135 struct uio *a_uio; 6136 int a_ioflag; 6137 struct ucred *a_cred; 6138 } */ *ap; 6139{ 6140 6141 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 6142 ap->a_cred, NULL)); 6143} 6144 6145static int 6146zfs_freebsd_write(ap) 6147 struct vop_write_args /* { 6148 struct vnode *a_vp; 6149 struct uio *a_uio; 6150 int a_ioflag; 6151 struct ucred *a_cred; 6152 } */ *ap; 6153{ 6154 6155 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 6156 ap->a_cred, NULL)); 6157} 6158 6159static int 6160zfs_freebsd_access(ap) 6161 struct vop_access_args /* { 6162 struct vnode *a_vp; 6163 accmode_t a_accmode; 6164 struct ucred *a_cred; 6165 struct thread *a_td; 6166 } */ *ap; 6167{ 6168 vnode_t *vp = ap->a_vp; 6169 znode_t *zp = VTOZ(vp); 6170 accmode_t accmode; 6171 int error = 0; 6172 6173 /* 6174 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 6175 */ 6176 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 6177 if (accmode != 0) 6178 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL); 6179 6180 /* 6181 * VADMIN has to be handled by vaccess(). 6182 */ 6183 if (error == 0) { 6184 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 6185 if (accmode != 0) { 6186 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 6187 zp->z_gid, accmode, ap->a_cred, NULL); 6188 } 6189 } 6190 6191 /* 6192 * For VEXEC, ensure that at least one execute bit is set for 6193 * non-directories. 6194 */ 6195 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 6196 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 6197 error = EACCES; 6198 } 6199 6200 return (error); 6201} 6202 6203static int 6204zfs_freebsd_lookup(ap) 6205 struct vop_lookup_args /* { 6206 struct vnode *a_dvp; 6207 struct vnode **a_vpp; 6208 struct componentname *a_cnp; 6209 } */ *ap; 6210{ 6211 struct componentname *cnp = ap->a_cnp; 6212 char nm[NAME_MAX + 1]; 6213 6214 ASSERT(cnp->cn_namelen < sizeof(nm)); 6215 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 6216 6217 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 6218 cnp->cn_cred, cnp->cn_thread, 0)); 6219} 6220 6221static int 6222zfs_freebsd_create(ap) 6223 struct vop_create_args /* { 6224 struct vnode *a_dvp; 6225 struct vnode **a_vpp; 6226 struct componentname *a_cnp; 6227 struct vattr *a_vap; 6228 } */ *ap; 6229{ 6230 struct componentname *cnp = ap->a_cnp; 6231 vattr_t *vap = ap->a_vap; 6232 int error, mode; 6233 6234 ASSERT(cnp->cn_flags & SAVENAME); 6235 6236 vattr_init_mask(vap); 6237 mode = vap->va_mode & ALLPERMS; 6238 6239 error = zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 6240 ap->a_vpp, cnp->cn_cred, cnp->cn_thread); 6241#ifdef FREEBSD_NAMECACHE 6242 if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0) 6243 cache_enter(ap->a_dvp, *ap->a_vpp, cnp); 6244#endif 6245 return (error); 6246} 6247 6248static int 6249zfs_freebsd_remove(ap) 6250 struct vop_remove_args /* { 6251 struct vnode *a_dvp; 6252 struct vnode *a_vp; 6253 struct componentname *a_cnp; 6254 } */ *ap; 6255{ 6256 6257 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6258 6259 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 6260 ap->a_cnp->cn_cred, NULL, 0)); 6261} 6262 6263static int 6264zfs_freebsd_mkdir(ap) 6265 struct vop_mkdir_args /* { 6266 struct vnode *a_dvp; 6267 struct vnode **a_vpp; 6268 struct componentname *a_cnp; 6269 struct vattr *a_vap; 6270 } */ *ap; 6271{ 6272 vattr_t *vap = ap->a_vap; 6273 6274 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6275 6276 vattr_init_mask(vap); 6277 6278 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 6279 ap->a_cnp->cn_cred, NULL, 0, NULL)); 6280} 6281 6282static int 6283zfs_freebsd_rmdir(ap) 6284 struct vop_rmdir_args /* { 6285 struct vnode *a_dvp; 6286 struct vnode *a_vp; 6287 struct componentname *a_cnp; 6288 } */ *ap; 6289{ 6290 struct componentname *cnp = ap->a_cnp; 6291 6292 ASSERT(cnp->cn_flags & SAVENAME); 6293 6294 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0)); 6295} 6296 6297static int 6298zfs_freebsd_readdir(ap) 6299 struct vop_readdir_args /* { 6300 struct vnode *a_vp; 6301 struct uio *a_uio; 6302 struct ucred *a_cred; 6303 int *a_eofflag; 6304 int *a_ncookies; 6305 u_long **a_cookies; 6306 } */ *ap; 6307{ 6308 6309 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 6310 ap->a_ncookies, ap->a_cookies)); 6311} 6312 6313static int 6314zfs_freebsd_fsync(ap) 6315 struct vop_fsync_args /* { 6316 struct vnode *a_vp; 6317 int a_waitfor; 6318 struct thread *a_td; 6319 } */ *ap; 6320{ 6321 6322 vop_stdfsync(ap); 6323 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL)); 6324} 6325 6326static int 6327zfs_freebsd_getattr(ap) 6328 struct vop_getattr_args /* { 6329 struct vnode *a_vp; 6330 struct vattr *a_vap; 6331 struct ucred *a_cred; 6332 } */ *ap; 6333{ 6334 vattr_t *vap = ap->a_vap; 6335 xvattr_t xvap; 6336 u_long fflags = 0; 6337 int error; 6338 6339 xva_init(&xvap); 6340 xvap.xva_vattr = *vap; 6341 xvap.xva_vattr.va_mask |= AT_XVATTR; 6342 6343 /* Convert chflags into ZFS-type flags. */ 6344 /* XXX: what about SF_SETTABLE?. */ 6345 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 6346 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 6347 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 6348 XVA_SET_REQ(&xvap, XAT_NODUMP); 6349 XVA_SET_REQ(&xvap, XAT_READONLY); 6350 XVA_SET_REQ(&xvap, XAT_ARCHIVE); 6351 XVA_SET_REQ(&xvap, XAT_SYSTEM); 6352 XVA_SET_REQ(&xvap, XAT_HIDDEN); 6353 XVA_SET_REQ(&xvap, XAT_REPARSE); 6354 XVA_SET_REQ(&xvap, XAT_OFFLINE); 6355 XVA_SET_REQ(&xvap, XAT_SPARSE); 6356 6357 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL); 6358 if (error != 0) 6359 return (error); 6360 6361 /* Convert ZFS xattr into chflags. */ 6362#define FLAG_CHECK(fflag, xflag, xfield) do { \ 6363 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 6364 fflags |= (fflag); \ 6365} while (0) 6366 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 6367 xvap.xva_xoptattrs.xoa_immutable); 6368 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 6369 xvap.xva_xoptattrs.xoa_appendonly); 6370 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 6371 xvap.xva_xoptattrs.xoa_nounlink); 6372 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE, 6373 xvap.xva_xoptattrs.xoa_archive); 6374 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 6375 xvap.xva_xoptattrs.xoa_nodump); 6376 FLAG_CHECK(UF_READONLY, XAT_READONLY, 6377 xvap.xva_xoptattrs.xoa_readonly); 6378 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM, 6379 xvap.xva_xoptattrs.xoa_system); 6380 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN, 6381 xvap.xva_xoptattrs.xoa_hidden); 6382 FLAG_CHECK(UF_REPARSE, XAT_REPARSE, 6383 xvap.xva_xoptattrs.xoa_reparse); 6384 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE, 6385 xvap.xva_xoptattrs.xoa_offline); 6386 FLAG_CHECK(UF_SPARSE, XAT_SPARSE, 6387 xvap.xva_xoptattrs.xoa_sparse); 6388 6389#undef FLAG_CHECK 6390 *vap = xvap.xva_vattr; 6391 vap->va_flags = fflags; 6392 return (0); 6393} 6394 6395static int 6396zfs_freebsd_setattr(ap) 6397 struct vop_setattr_args /* { 6398 struct vnode *a_vp; 6399 struct vattr *a_vap; 6400 struct ucred *a_cred; 6401 } */ *ap; 6402{ 6403 vnode_t *vp = ap->a_vp; 6404 vattr_t *vap = ap->a_vap; 6405 cred_t *cred = ap->a_cred; 6406 xvattr_t xvap; 6407 u_long fflags; 6408 uint64_t zflags; 6409 6410 vattr_init_mask(vap); 6411 vap->va_mask &= ~AT_NOSET; 6412 6413 xva_init(&xvap); 6414 xvap.xva_vattr = *vap; 6415 6416 zflags = VTOZ(vp)->z_pflags; 6417 6418 if (vap->va_flags != VNOVAL) { 6419 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 6420 int error; 6421 6422 if (zfsvfs->z_use_fuids == B_FALSE) 6423 return (EOPNOTSUPP); 6424 6425 fflags = vap->va_flags; 6426 /* 6427 * XXX KDM 6428 * We need to figure out whether it makes sense to allow 6429 * UF_REPARSE through, since we don't really have other 6430 * facilities to handle reparse points and zfs_setattr() 6431 * doesn't currently allow setting that attribute anyway. 6432 */ 6433 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE| 6434 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE| 6435 UF_OFFLINE|UF_SPARSE)) != 0) 6436 return (EOPNOTSUPP); 6437 /* 6438 * Unprivileged processes are not permitted to unset system 6439 * flags, or modify flags if any system flags are set. 6440 * Privileged non-jail processes may not modify system flags 6441 * if securelevel > 0 and any existing system flags are set. 6442 * Privileged jail processes behave like privileged non-jail 6443 * processes if the security.jail.chflags_allowed sysctl is 6444 * is non-zero; otherwise, they behave like unprivileged 6445 * processes. 6446 */ 6447 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 6448 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) { 6449 if (zflags & 6450 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6451 error = securelevel_gt(cred, 0); 6452 if (error != 0) 6453 return (error); 6454 } 6455 } else { 6456 /* 6457 * Callers may only modify the file flags on objects they 6458 * have VADMIN rights for. 6459 */ 6460 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0) 6461 return (error); 6462 if (zflags & 6463 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6464 return (EPERM); 6465 } 6466 if (fflags & 6467 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 6468 return (EPERM); 6469 } 6470 } 6471 6472#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 6473 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 6474 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 6475 XVA_SET_REQ(&xvap, (xflag)); \ 6476 (xfield) = ((fflags & (fflag)) != 0); \ 6477 } \ 6478} while (0) 6479 /* Convert chflags into ZFS-type flags. */ 6480 /* XXX: what about SF_SETTABLE?. */ 6481 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 6482 xvap.xva_xoptattrs.xoa_immutable); 6483 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 6484 xvap.xva_xoptattrs.xoa_appendonly); 6485 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 6486 xvap.xva_xoptattrs.xoa_nounlink); 6487 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE, 6488 xvap.xva_xoptattrs.xoa_archive); 6489 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 6490 xvap.xva_xoptattrs.xoa_nodump); 6491 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY, 6492 xvap.xva_xoptattrs.xoa_readonly); 6493 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM, 6494 xvap.xva_xoptattrs.xoa_system); 6495 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN, 6496 xvap.xva_xoptattrs.xoa_hidden); 6497 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE, 6498 xvap.xva_xoptattrs.xoa_hidden); 6499 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE, 6500 xvap.xva_xoptattrs.xoa_offline); 6501 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE, 6502 xvap.xva_xoptattrs.xoa_sparse); 6503#undef FLAG_CHANGE 6504 } 6505 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL)); 6506} 6507 6508static int 6509zfs_freebsd_rename(ap) 6510 struct vop_rename_args /* { 6511 struct vnode *a_fdvp; 6512 struct vnode *a_fvp; 6513 struct componentname *a_fcnp; 6514 struct vnode *a_tdvp; 6515 struct vnode *a_tvp; 6516 struct componentname *a_tcnp; 6517 } */ *ap; 6518{ 6519 vnode_t *fdvp = ap->a_fdvp; 6520 vnode_t *fvp = ap->a_fvp; 6521 vnode_t *tdvp = ap->a_tdvp; 6522 vnode_t *tvp = ap->a_tvp; 6523 int error; 6524 6525 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 6526 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 6527 6528 /* 6529 * Check for cross-device rename. 6530 */ 6531 if ((fdvp->v_mount != tdvp->v_mount) || 6532 (tvp && (fdvp->v_mount != tvp->v_mount))) 6533 error = EXDEV; 6534 else 6535 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 6536 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0); 6537 if (tdvp == tvp) 6538 VN_RELE(tdvp); 6539 else 6540 VN_URELE(tdvp); 6541 if (tvp) 6542 VN_URELE(tvp); 6543 VN_RELE(fdvp); 6544 VN_RELE(fvp); 6545 6546 return (error); 6547} 6548 6549static int 6550zfs_freebsd_symlink(ap) 6551 struct vop_symlink_args /* { 6552 struct vnode *a_dvp; 6553 struct vnode **a_vpp; 6554 struct componentname *a_cnp; 6555 struct vattr *a_vap; 6556 char *a_target; 6557 } */ *ap; 6558{ 6559 struct componentname *cnp = ap->a_cnp; 6560 vattr_t *vap = ap->a_vap; 6561 6562 ASSERT(cnp->cn_flags & SAVENAME); 6563 6564 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 6565 vattr_init_mask(vap); 6566 6567 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 6568 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 6569} 6570 6571static int 6572zfs_freebsd_readlink(ap) 6573 struct vop_readlink_args /* { 6574 struct vnode *a_vp; 6575 struct uio *a_uio; 6576 struct ucred *a_cred; 6577 } */ *ap; 6578{ 6579 6580 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL)); 6581} 6582 6583static int 6584zfs_freebsd_link(ap) 6585 struct vop_link_args /* { 6586 struct vnode *a_tdvp; 6587 struct vnode *a_vp; 6588 struct componentname *a_cnp; 6589 } */ *ap; 6590{ 6591 struct componentname *cnp = ap->a_cnp; 6592 vnode_t *vp = ap->a_vp; 6593 vnode_t *tdvp = ap->a_tdvp; 6594 6595 if (tdvp->v_mount != vp->v_mount) 6596 return (EXDEV); 6597 6598 ASSERT(cnp->cn_flags & SAVENAME); 6599 6600 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0)); 6601} 6602 6603static int 6604zfs_freebsd_inactive(ap) 6605 struct vop_inactive_args /* { 6606 struct vnode *a_vp; 6607 struct thread *a_td; 6608 } */ *ap; 6609{ 6610 vnode_t *vp = ap->a_vp; 6611 6612 zfs_inactive(vp, ap->a_td->td_ucred, NULL); 6613 return (0); 6614} 6615 6616static int 6617zfs_freebsd_reclaim(ap) 6618 struct vop_reclaim_args /* { 6619 struct vnode *a_vp; 6620 struct thread *a_td; 6621 } */ *ap; 6622{ 6623 vnode_t *vp = ap->a_vp; 6624 znode_t *zp = VTOZ(vp); 6625 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6626 6627 ASSERT(zp != NULL); 6628 6629 /* Destroy the vm object and flush associated pages. */ 6630 vnode_destroy_vobject(vp); 6631 6632 /* 6633 * z_teardown_inactive_lock protects from a race with 6634 * zfs_znode_dmu_fini in zfsvfs_teardown during 6635 * force unmount. 6636 */ 6637 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 6638 if (zp->z_sa_hdl == NULL) 6639 zfs_znode_free(zp); 6640 else 6641 zfs_zinactive(zp); 6642 rw_exit(&zfsvfs->z_teardown_inactive_lock); 6643 6644 vp->v_data = NULL; 6645 return (0); 6646} 6647 6648static int 6649zfs_freebsd_fid(ap) 6650 struct vop_fid_args /* { 6651 struct vnode *a_vp; 6652 struct fid *a_fid; 6653 } */ *ap; 6654{ 6655 6656 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 6657} 6658 6659static int 6660zfs_freebsd_pathconf(ap) 6661 struct vop_pathconf_args /* { 6662 struct vnode *a_vp; 6663 int a_name; 6664 register_t *a_retval; 6665 } */ *ap; 6666{ 6667 ulong_t val; 6668 int error; 6669 6670 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); 6671 if (error == 0) 6672 *ap->a_retval = val; 6673 else if (error == EOPNOTSUPP) 6674 error = vop_stdpathconf(ap); 6675 return (error); 6676} 6677 6678static int 6679zfs_freebsd_fifo_pathconf(ap) 6680 struct vop_pathconf_args /* { 6681 struct vnode *a_vp; 6682 int a_name; 6683 register_t *a_retval; 6684 } */ *ap; 6685{ 6686 6687 switch (ap->a_name) { 6688 case _PC_ACL_EXTENDED: 6689 case _PC_ACL_NFS4: 6690 case _PC_ACL_PATH_MAX: 6691 case _PC_MAC_PRESENT: 6692 return (zfs_freebsd_pathconf(ap)); 6693 default: 6694 return (fifo_specops.vop_pathconf(ap)); 6695 } 6696} 6697 6698/* 6699 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 6700 * extended attribute name: 6701 * 6702 * NAMESPACE PREFIX 6703 * system freebsd:system: 6704 * user (none, can be used to access ZFS fsattr(5) attributes 6705 * created on Solaris) 6706 */ 6707static int 6708zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 6709 size_t size) 6710{ 6711 const char *namespace, *prefix, *suffix; 6712 6713 /* We don't allow '/' character in attribute name. */ 6714 if (strchr(name, '/') != NULL) 6715 return (EINVAL); 6716 /* We don't allow attribute names that start with "freebsd:" string. */ 6717 if (strncmp(name, "freebsd:", 8) == 0) 6718 return (EINVAL); 6719 6720 bzero(attrname, size); 6721 6722 switch (attrnamespace) { 6723 case EXTATTR_NAMESPACE_USER: 6724#if 0 6725 prefix = "freebsd:"; 6726 namespace = EXTATTR_NAMESPACE_USER_STRING; 6727 suffix = ":"; 6728#else 6729 /* 6730 * This is the default namespace by which we can access all 6731 * attributes created on Solaris. 6732 */ 6733 prefix = namespace = suffix = ""; 6734#endif 6735 break; 6736 case EXTATTR_NAMESPACE_SYSTEM: 6737 prefix = "freebsd:"; 6738 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 6739 suffix = ":"; 6740 break; 6741 case EXTATTR_NAMESPACE_EMPTY: 6742 default: 6743 return (EINVAL); 6744 } 6745 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 6746 name) >= size) { 6747 return (ENAMETOOLONG); 6748 } 6749 return (0); 6750} 6751 6752/* 6753 * Vnode operating to retrieve a named extended attribute. 6754 */ 6755static int 6756zfs_getextattr(struct vop_getextattr_args *ap) 6757/* 6758vop_getextattr { 6759 IN struct vnode *a_vp; 6760 IN int a_attrnamespace; 6761 IN const char *a_name; 6762 INOUT struct uio *a_uio; 6763 OUT size_t *a_size; 6764 IN struct ucred *a_cred; 6765 IN struct thread *a_td; 6766}; 6767*/ 6768{ 6769 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6770 struct thread *td = ap->a_td; 6771 struct nameidata nd; 6772 char attrname[255]; 6773 struct vattr va; 6774 vnode_t *xvp = NULL, *vp; 6775 int error, flags; 6776 6777 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6778 ap->a_cred, ap->a_td, VREAD); 6779 if (error != 0) 6780 return (error); 6781 6782 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6783 sizeof(attrname)); 6784 if (error != 0) 6785 return (error); 6786 6787 ZFS_ENTER(zfsvfs); 6788 6789 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6790 LOOKUP_XATTR); 6791 if (error != 0) { 6792 ZFS_EXIT(zfsvfs); 6793 return (error); 6794 } 6795 6796 flags = FREAD; 6797 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6798 xvp, td); 6799 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL); 6800 vp = nd.ni_vp; 6801 NDFREE(&nd, NDF_ONLY_PNBUF); 6802 if (error != 0) { 6803 ZFS_EXIT(zfsvfs); 6804 if (error == ENOENT) 6805 error = ENOATTR; 6806 return (error); 6807 } 6808 6809 if (ap->a_size != NULL) { 6810 error = VOP_GETATTR(vp, &va, ap->a_cred); 6811 if (error == 0) 6812 *ap->a_size = (size_t)va.va_size; 6813 } else if (ap->a_uio != NULL) 6814 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6815 6816 VOP_UNLOCK(vp, 0); 6817 vn_close(vp, flags, ap->a_cred, td); 6818 ZFS_EXIT(zfsvfs); 6819 6820 return (error); 6821} 6822 6823/* 6824 * Vnode operation to remove a named attribute. 6825 */ 6826int 6827zfs_deleteextattr(struct vop_deleteextattr_args *ap) 6828/* 6829vop_deleteextattr { 6830 IN struct vnode *a_vp; 6831 IN int a_attrnamespace; 6832 IN const char *a_name; 6833 IN struct ucred *a_cred; 6834 IN struct thread *a_td; 6835}; 6836*/ 6837{ 6838 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6839 struct thread *td = ap->a_td; 6840 struct nameidata nd; 6841 char attrname[255]; 6842 struct vattr va; 6843 vnode_t *xvp = NULL, *vp; 6844 int error, flags; 6845 6846 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6847 ap->a_cred, ap->a_td, VWRITE); 6848 if (error != 0) 6849 return (error); 6850 6851 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6852 sizeof(attrname)); 6853 if (error != 0) 6854 return (error); 6855 6856 ZFS_ENTER(zfsvfs); 6857 6858 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6859 LOOKUP_XATTR); 6860 if (error != 0) { 6861 ZFS_EXIT(zfsvfs); 6862 return (error); 6863 } 6864 6865 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 6866 UIO_SYSSPACE, attrname, xvp, td); 6867 error = namei(&nd); 6868 vp = nd.ni_vp; 6869 if (error != 0) { 6870 ZFS_EXIT(zfsvfs); 6871 NDFREE(&nd, NDF_ONLY_PNBUF); 6872 if (error == ENOENT) 6873 error = ENOATTR; 6874 return (error); 6875 } 6876 6877 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 6878 NDFREE(&nd, NDF_ONLY_PNBUF); 6879 6880 vput(nd.ni_dvp); 6881 if (vp == nd.ni_dvp) 6882 vrele(vp); 6883 else 6884 vput(vp); 6885 ZFS_EXIT(zfsvfs); 6886 6887 return (error); 6888} 6889 6890/* 6891 * Vnode operation to set a named attribute. 6892 */ 6893static int 6894zfs_setextattr(struct vop_setextattr_args *ap) 6895/* 6896vop_setextattr { 6897 IN struct vnode *a_vp; 6898 IN int a_attrnamespace; 6899 IN const char *a_name; 6900 INOUT struct uio *a_uio; 6901 IN struct ucred *a_cred; 6902 IN struct thread *a_td; 6903}; 6904*/ 6905{ 6906 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6907 struct thread *td = ap->a_td; 6908 struct nameidata nd; 6909 char attrname[255]; 6910 struct vattr va; 6911 vnode_t *xvp = NULL, *vp; 6912 int error, flags; 6913 6914 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6915 ap->a_cred, ap->a_td, VWRITE); 6916 if (error != 0) 6917 return (error); 6918 6919 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6920 sizeof(attrname)); 6921 if (error != 0) 6922 return (error); 6923 6924 ZFS_ENTER(zfsvfs); 6925 6926 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6927 LOOKUP_XATTR | CREATE_XATTR_DIR); 6928 if (error != 0) { 6929 ZFS_EXIT(zfsvfs); 6930 return (error); 6931 } 6932 6933 flags = FFLAGS(O_WRONLY | O_CREAT); 6934 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6935 xvp, td); 6936 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL); 6937 vp = nd.ni_vp; 6938 NDFREE(&nd, NDF_ONLY_PNBUF); 6939 if (error != 0) { 6940 ZFS_EXIT(zfsvfs); 6941 return (error); 6942 } 6943 6944 VATTR_NULL(&va); 6945 va.va_size = 0; 6946 error = VOP_SETATTR(vp, &va, ap->a_cred); 6947 if (error == 0) 6948 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6949 6950 VOP_UNLOCK(vp, 0); 6951 vn_close(vp, flags, ap->a_cred, td); 6952 ZFS_EXIT(zfsvfs); 6953 6954 return (error); 6955} 6956 6957/* 6958 * Vnode operation to retrieve extended attributes on a vnode. 6959 */ 6960static int 6961zfs_listextattr(struct vop_listextattr_args *ap) 6962/* 6963vop_listextattr { 6964 IN struct vnode *a_vp; 6965 IN int a_attrnamespace; 6966 INOUT struct uio *a_uio; 6967 OUT size_t *a_size; 6968 IN struct ucred *a_cred; 6969 IN struct thread *a_td; 6970}; 6971*/ 6972{ 6973 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6974 struct thread *td = ap->a_td; 6975 struct nameidata nd; 6976 char attrprefix[16]; 6977 u_char dirbuf[sizeof(struct dirent)]; 6978 struct dirent *dp; 6979 struct iovec aiov; 6980 struct uio auio, *uio = ap->a_uio; 6981 size_t *sizep = ap->a_size; 6982 size_t plen; 6983 vnode_t *xvp = NULL, *vp; 6984 int done, error, eof, pos; 6985 6986 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6987 ap->a_cred, ap->a_td, VREAD); 6988 if (error != 0) 6989 return (error); 6990 6991 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 6992 sizeof(attrprefix)); 6993 if (error != 0) 6994 return (error); 6995 plen = strlen(attrprefix); 6996 6997 ZFS_ENTER(zfsvfs); 6998 6999 if (sizep != NULL) 7000 *sizep = 0; 7001 7002 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 7003 LOOKUP_XATTR); 7004 if (error != 0) { 7005 ZFS_EXIT(zfsvfs); 7006 /* 7007 * ENOATTR means that the EA directory does not yet exist, 7008 * i.e. there are no extended attributes there. 7009 */ 7010 if (error == ENOATTR) 7011 error = 0; 7012 return (error); 7013 } 7014 7015 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 7016 UIO_SYSSPACE, ".", xvp, td); 7017 error = namei(&nd); 7018 vp = nd.ni_vp; 7019 NDFREE(&nd, NDF_ONLY_PNBUF); 7020 if (error != 0) { 7021 ZFS_EXIT(zfsvfs); 7022 return (error); 7023 } 7024 7025 auio.uio_iov = &aiov; 7026 auio.uio_iovcnt = 1; 7027 auio.uio_segflg = UIO_SYSSPACE; 7028 auio.uio_td = td; 7029 auio.uio_rw = UIO_READ; 7030 auio.uio_offset = 0; 7031 7032 do { 7033 u_char nlen; 7034 7035 aiov.iov_base = (void *)dirbuf; 7036 aiov.iov_len = sizeof(dirbuf); 7037 auio.uio_resid = sizeof(dirbuf); 7038 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 7039 done = sizeof(dirbuf) - auio.uio_resid; 7040 if (error != 0) 7041 break; 7042 for (pos = 0; pos < done;) { 7043 dp = (struct dirent *)(dirbuf + pos); 7044 pos += dp->d_reclen; 7045 /* 7046 * XXX: Temporarily we also accept DT_UNKNOWN, as this 7047 * is what we get when attribute was created on Solaris. 7048 */ 7049 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 7050 continue; 7051 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0) 7052 continue; 7053 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 7054 continue; 7055 nlen = dp->d_namlen - plen; 7056 if (sizep != NULL) 7057 *sizep += 1 + nlen; 7058 else if (uio != NULL) { 7059 /* 7060 * Format of extattr name entry is one byte for 7061 * length and the rest for name. 7062 */ 7063 error = uiomove(&nlen, 1, uio->uio_rw, uio); 7064 if (error == 0) { 7065 error = uiomove(dp->d_name + plen, nlen, 7066 uio->uio_rw, uio); 7067 } 7068 if (error != 0) 7069 break; 7070 } 7071 } 7072 } while (!eof && error == 0); 7073 7074 vput(vp); 7075 ZFS_EXIT(zfsvfs); 7076 7077 return (error); 7078} 7079 7080int 7081zfs_freebsd_getacl(ap) 7082 struct vop_getacl_args /* { 7083 struct vnode *vp; 7084 acl_type_t type; 7085 struct acl *aclp; 7086 struct ucred *cred; 7087 struct thread *td; 7088 } */ *ap; 7089{ 7090 int error; 7091 vsecattr_t vsecattr; 7092 7093 if (ap->a_type != ACL_TYPE_NFS4) 7094 return (EINVAL); 7095 7096 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 7097 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL)) 7098 return (error); 7099 7100 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt); 7101 if (vsecattr.vsa_aclentp != NULL) 7102 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 7103 7104 return (error); 7105} 7106 7107int 7108zfs_freebsd_setacl(ap) 7109 struct vop_setacl_args /* { 7110 struct vnode *vp; 7111 acl_type_t type; 7112 struct acl *aclp; 7113 struct ucred *cred; 7114 struct thread *td; 7115 } */ *ap; 7116{ 7117 int error; 7118 vsecattr_t vsecattr; 7119 int aclbsize; /* size of acl list in bytes */ 7120 aclent_t *aaclp; 7121 7122 if (ap->a_type != ACL_TYPE_NFS4) 7123 return (EINVAL); 7124 7125 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 7126 return (EINVAL); 7127 7128 /* 7129 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 7130 * splitting every entry into two and appending "canonical six" 7131 * entries at the end. Don't allow for setting an ACL that would 7132 * cause chmod(2) to run out of ACL entries. 7133 */ 7134 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 7135 return (ENOSPC); 7136 7137 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 7138 if (error != 0) 7139 return (error); 7140 7141 vsecattr.vsa_mask = VSA_ACE; 7142 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t); 7143 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 7144 aaclp = vsecattr.vsa_aclentp; 7145 vsecattr.vsa_aclentsz = aclbsize; 7146 7147 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 7148 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL); 7149 kmem_free(aaclp, aclbsize); 7150 7151 return (error); 7152} 7153 7154int 7155zfs_freebsd_aclcheck(ap) 7156 struct vop_aclcheck_args /* { 7157 struct vnode *vp; 7158 acl_type_t type; 7159 struct acl *aclp; 7160 struct ucred *cred; 7161 struct thread *td; 7162 } */ *ap; 7163{ 7164 7165 return (EOPNOTSUPP); 7166} 7167 7168struct vop_vector zfs_vnodeops; 7169struct vop_vector zfs_fifoops; 7170struct vop_vector zfs_shareops; 7171 7172struct vop_vector zfs_vnodeops = { 7173 .vop_default = &default_vnodeops, 7174 .vop_inactive = zfs_freebsd_inactive, 7175 .vop_reclaim = zfs_freebsd_reclaim, 7176 .vop_access = zfs_freebsd_access, 7177#ifdef FREEBSD_NAMECACHE 7178 .vop_lookup = vfs_cache_lookup, 7179 .vop_cachedlookup = zfs_freebsd_lookup, 7180#else 7181 .vop_lookup = zfs_freebsd_lookup, 7182#endif 7183 .vop_getattr = zfs_freebsd_getattr, 7184 .vop_setattr = zfs_freebsd_setattr, 7185 .vop_create = zfs_freebsd_create, 7186 .vop_mknod = zfs_freebsd_create, 7187 .vop_mkdir = zfs_freebsd_mkdir, 7188 .vop_readdir = zfs_freebsd_readdir, 7189 .vop_fsync = zfs_freebsd_fsync, 7190 .vop_open = zfs_freebsd_open, 7191 .vop_close = zfs_freebsd_close, 7192 .vop_rmdir = zfs_freebsd_rmdir, 7193 .vop_ioctl = zfs_freebsd_ioctl, 7194 .vop_link = zfs_freebsd_link, 7195 .vop_symlink = zfs_freebsd_symlink, 7196 .vop_readlink = zfs_freebsd_readlink, 7197 .vop_read = zfs_freebsd_read, 7198 .vop_write = zfs_freebsd_write, 7199 .vop_remove = zfs_freebsd_remove, 7200 .vop_rename = zfs_freebsd_rename, 7201 .vop_pathconf = zfs_freebsd_pathconf, 7202 .vop_bmap = zfs_freebsd_bmap, 7203 .vop_fid = zfs_freebsd_fid, 7204 .vop_getextattr = zfs_getextattr, 7205 .vop_deleteextattr = zfs_deleteextattr, 7206 .vop_setextattr = zfs_setextattr, 7207 .vop_listextattr = zfs_listextattr, 7208 .vop_getacl = zfs_freebsd_getacl, 7209 .vop_setacl = zfs_freebsd_setacl, 7210 .vop_aclcheck = zfs_freebsd_aclcheck, 7211 .vop_getpages = zfs_freebsd_getpages, 7212 .vop_putpages = zfs_freebsd_putpages, 7213}; 7214 7215struct vop_vector zfs_fifoops = { 7216 .vop_default = &fifo_specops, 7217 .vop_fsync = zfs_freebsd_fsync, 7218 .vop_access = zfs_freebsd_access, 7219 .vop_getattr = zfs_freebsd_getattr, 7220 .vop_inactive = zfs_freebsd_inactive, 7221 .vop_read = VOP_PANIC, 7222 .vop_reclaim = zfs_freebsd_reclaim, 7223 .vop_setattr = zfs_freebsd_setattr, 7224 .vop_write = VOP_PANIC, 7225 .vop_pathconf = zfs_freebsd_fifo_pathconf, 7226 .vop_fid = zfs_freebsd_fid, 7227 .vop_getacl = zfs_freebsd_getacl, 7228 .vop_setacl = zfs_freebsd_setacl, 7229 .vop_aclcheck = zfs_freebsd_aclcheck, 7230}; 7231 7232/* 7233 * special share hidden files vnode operations template 7234 */ 7235struct vop_vector zfs_shareops = { 7236 .vop_default = &default_vnodeops, 7237 .vop_access = zfs_freebsd_access, 7238 .vop_inactive = zfs_freebsd_inactive, 7239 .vop_reclaim = zfs_freebsd_reclaim, 7240 .vop_fid = zfs_freebsd_fid, 7241 .vop_pathconf = zfs_freebsd_pathconf, 7242}; 7243