zfs_vnops.c revision 297112
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 * Copyright (c) 2014 Integros [integros.com] 26 */ 27 28/* Portions Copyright 2007 Jeremy Teo */ 29/* Portions Copyright 2010 Robert Milkowski */ 30 31#include <sys/types.h> 32#include <sys/param.h> 33#include <sys/time.h> 34#include <sys/systm.h> 35#include <sys/sysmacros.h> 36#include <sys/resource.h> 37#include <sys/vfs.h> 38#include <sys/vm.h> 39#include <sys/vnode.h> 40#include <sys/file.h> 41#include <sys/stat.h> 42#include <sys/kmem.h> 43#include <sys/taskq.h> 44#include <sys/uio.h> 45#include <sys/atomic.h> 46#include <sys/namei.h> 47#include <sys/mman.h> 48#include <sys/cmn_err.h> 49#include <sys/errno.h> 50#include <sys/unistd.h> 51#include <sys/zfs_dir.h> 52#include <sys/zfs_ioctl.h> 53#include <sys/fs/zfs.h> 54#include <sys/dmu.h> 55#include <sys/dmu_objset.h> 56#include <sys/spa.h> 57#include <sys/txg.h> 58#include <sys/dbuf.h> 59#include <sys/zap.h> 60#include <sys/sa.h> 61#include <sys/dirent.h> 62#include <sys/policy.h> 63#include <sys/sunddi.h> 64#include <sys/filio.h> 65#include <sys/sid.h> 66#include <sys/zfs_ctldir.h> 67#include <sys/zfs_fuid.h> 68#include <sys/zfs_sa.h> 69#include <sys/dnlc.h> 70#include <sys/zfs_rlock.h> 71#include <sys/extdirent.h> 72#include <sys/kidmap.h> 73#include <sys/bio.h> 74#include <sys/buf.h> 75#include <sys/sched.h> 76#include <sys/acl.h> 77#include <vm/vm_param.h> 78#include <vm/vm_pageout.h> 79 80/* 81 * Programming rules. 82 * 83 * Each vnode op performs some logical unit of work. To do this, the ZPL must 84 * properly lock its in-core state, create a DMU transaction, do the work, 85 * record this work in the intent log (ZIL), commit the DMU transaction, 86 * and wait for the intent log to commit if it is a synchronous operation. 87 * Moreover, the vnode ops must work in both normal and log replay context. 88 * The ordering of events is important to avoid deadlocks and references 89 * to freed memory. The example below illustrates the following Big Rules: 90 * 91 * (1) A check must be made in each zfs thread for a mounted file system. 92 * This is done avoiding races using ZFS_ENTER(zfsvfs). 93 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes 94 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros 95 * can return EIO from the calling function. 96 * 97 * (2) VN_RELE() should always be the last thing except for zil_commit() 98 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 99 * First, if it's the last reference, the vnode/znode 100 * can be freed, so the zp may point to freed memory. Second, the last 101 * reference will call zfs_zinactive(), which may induce a lot of work -- 102 * pushing cached pages (which acquires range locks) and syncing out 103 * cached atime changes. Third, zfs_zinactive() may require a new tx, 104 * which could deadlock the system if you were already holding one. 105 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). 106 * 107 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 108 * as they can span dmu_tx_assign() calls. 109 * 110 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to 111 * dmu_tx_assign(). This is critical because we don't want to block 112 * while holding locks. 113 * 114 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This 115 * reduces lock contention and CPU usage when we must wait (note that if 116 * throughput is constrained by the storage, nearly every transaction 117 * must wait). 118 * 119 * Note, in particular, that if a lock is sometimes acquired before 120 * the tx assigns, and sometimes after (e.g. z_lock), then failing 121 * to use a non-blocking assign can deadlock the system. The scenario: 122 * 123 * Thread A has grabbed a lock before calling dmu_tx_assign(). 124 * Thread B is in an already-assigned tx, and blocks for this lock. 125 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 126 * forever, because the previous txg can't quiesce until B's tx commits. 127 * 128 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 129 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent 130 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT, 131 * to indicate that this operation has already called dmu_tx_wait(). 132 * This will ensure that we don't retry forever, waiting a short bit 133 * each time. 134 * 135 * (5) If the operation succeeded, generate the intent log entry for it 136 * before dropping locks. This ensures that the ordering of events 137 * in the intent log matches the order in which they actually occurred. 138 * During ZIL replay the zfs_log_* functions will update the sequence 139 * number to indicate the zil transaction has replayed. 140 * 141 * (6) At the end of each vnode op, the DMU tx must always commit, 142 * regardless of whether there were any errors. 143 * 144 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 145 * to ensure that synchronous semantics are provided when necessary. 146 * 147 * In general, this is how things should be ordered in each vnode op: 148 * 149 * ZFS_ENTER(zfsvfs); // exit if unmounted 150 * top: 151 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 152 * rw_enter(...); // grab any other locks you need 153 * tx = dmu_tx_create(...); // get DMU tx 154 * dmu_tx_hold_*(); // hold each object you might modify 155 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 156 * if (error) { 157 * rw_exit(...); // drop locks 158 * zfs_dirent_unlock(dl); // unlock directory entry 159 * VN_RELE(...); // release held vnodes 160 * if (error == ERESTART) { 161 * waited = B_TRUE; 162 * dmu_tx_wait(tx); 163 * dmu_tx_abort(tx); 164 * goto top; 165 * } 166 * dmu_tx_abort(tx); // abort DMU tx 167 * ZFS_EXIT(zfsvfs); // finished in zfs 168 * return (error); // really out of space 169 * } 170 * error = do_real_work(); // do whatever this VOP does 171 * if (error == 0) 172 * zfs_log_*(...); // on success, make ZIL entry 173 * dmu_tx_commit(tx); // commit DMU tx -- error or not 174 * rw_exit(...); // drop locks 175 * zfs_dirent_unlock(dl); // unlock directory entry 176 * VN_RELE(...); // release held vnodes 177 * zil_commit(zilog, foid); // synchronous when necessary 178 * ZFS_EXIT(zfsvfs); // finished in zfs 179 * return (error); // done, report error 180 */ 181 182/* ARGSUSED */ 183static int 184zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 185{ 186 znode_t *zp = VTOZ(*vpp); 187 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 188 189 ZFS_ENTER(zfsvfs); 190 ZFS_VERIFY_ZP(zp); 191 192 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 193 ((flag & FAPPEND) == 0)) { 194 ZFS_EXIT(zfsvfs); 195 return (SET_ERROR(EPERM)); 196 } 197 198 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 199 ZTOV(zp)->v_type == VREG && 200 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) { 201 if (fs_vscan(*vpp, cr, 0) != 0) { 202 ZFS_EXIT(zfsvfs); 203 return (SET_ERROR(EACCES)); 204 } 205 } 206 207 /* Keep a count of the synchronous opens in the znode */ 208 if (flag & (FSYNC | FDSYNC)) 209 atomic_inc_32(&zp->z_sync_cnt); 210 211 ZFS_EXIT(zfsvfs); 212 return (0); 213} 214 215/* ARGSUSED */ 216static int 217zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 218 caller_context_t *ct) 219{ 220 znode_t *zp = VTOZ(vp); 221 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 222 223 /* 224 * Clean up any locks held by this process on the vp. 225 */ 226 cleanlocks(vp, ddi_get_pid(), 0); 227 cleanshares(vp, ddi_get_pid()); 228 229 ZFS_ENTER(zfsvfs); 230 ZFS_VERIFY_ZP(zp); 231 232 /* Decrement the synchronous opens in the znode */ 233 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 234 atomic_dec_32(&zp->z_sync_cnt); 235 236 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 237 ZTOV(zp)->v_type == VREG && 238 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) 239 VERIFY(fs_vscan(vp, cr, 1) == 0); 240 241 ZFS_EXIT(zfsvfs); 242 return (0); 243} 244 245/* 246 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 247 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 248 */ 249static int 250zfs_holey(vnode_t *vp, u_long cmd, offset_t *off) 251{ 252 znode_t *zp = VTOZ(vp); 253 uint64_t noff = (uint64_t)*off; /* new offset */ 254 uint64_t file_sz; 255 int error; 256 boolean_t hole; 257 258 file_sz = zp->z_size; 259 if (noff >= file_sz) { 260 return (SET_ERROR(ENXIO)); 261 } 262 263 if (cmd == _FIO_SEEK_HOLE) 264 hole = B_TRUE; 265 else 266 hole = B_FALSE; 267 268 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 269 270 if (error == ESRCH) 271 return (SET_ERROR(ENXIO)); 272 273 /* 274 * We could find a hole that begins after the logical end-of-file, 275 * because dmu_offset_next() only works on whole blocks. If the 276 * EOF falls mid-block, then indicate that the "virtual hole" 277 * at the end of the file begins at the logical EOF, rather than 278 * at the end of the last block. 279 */ 280 if (noff > file_sz) { 281 ASSERT(hole); 282 noff = file_sz; 283 } 284 285 if (noff < *off) 286 return (error); 287 *off = noff; 288 return (error); 289} 290 291/* ARGSUSED */ 292static int 293zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred, 294 int *rvalp, caller_context_t *ct) 295{ 296 offset_t off; 297 offset_t ndata; 298 dmu_object_info_t doi; 299 int error; 300 zfsvfs_t *zfsvfs; 301 znode_t *zp; 302 303 switch (com) { 304 case _FIOFFS: 305 { 306 return (0); 307 308 /* 309 * The following two ioctls are used by bfu. Faking out, 310 * necessary to avoid bfu errors. 311 */ 312 } 313 case _FIOGDIO: 314 case _FIOSDIO: 315 { 316 return (0); 317 } 318 319 case _FIO_SEEK_DATA: 320 case _FIO_SEEK_HOLE: 321 { 322#ifdef illumos 323 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 324 return (SET_ERROR(EFAULT)); 325#else 326 off = *(offset_t *)data; 327#endif 328 zp = VTOZ(vp); 329 zfsvfs = zp->z_zfsvfs; 330 ZFS_ENTER(zfsvfs); 331 ZFS_VERIFY_ZP(zp); 332 333 /* offset parameter is in/out */ 334 error = zfs_holey(vp, com, &off); 335 ZFS_EXIT(zfsvfs); 336 if (error) 337 return (error); 338#ifdef illumos 339 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 340 return (SET_ERROR(EFAULT)); 341#else 342 *(offset_t *)data = off; 343#endif 344 return (0); 345 } 346#ifdef illumos 347 case _FIO_COUNT_FILLED: 348 { 349 /* 350 * _FIO_COUNT_FILLED adds a new ioctl command which 351 * exposes the number of filled blocks in a 352 * ZFS object. 353 */ 354 zp = VTOZ(vp); 355 zfsvfs = zp->z_zfsvfs; 356 ZFS_ENTER(zfsvfs); 357 ZFS_VERIFY_ZP(zp); 358 359 /* 360 * Wait for all dirty blocks for this object 361 * to get synced out to disk, and the DMU info 362 * updated. 363 */ 364 error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id); 365 if (error) { 366 ZFS_EXIT(zfsvfs); 367 return (error); 368 } 369 370 /* 371 * Retrieve fill count from DMU object. 372 */ 373 error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi); 374 if (error) { 375 ZFS_EXIT(zfsvfs); 376 return (error); 377 } 378 379 ndata = doi.doi_fill_count; 380 381 ZFS_EXIT(zfsvfs); 382 if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag)) 383 return (SET_ERROR(EFAULT)); 384 return (0); 385 } 386#endif 387 } 388 return (SET_ERROR(ENOTTY)); 389} 390 391static vm_page_t 392page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) 393{ 394 vm_object_t obj; 395 vm_page_t pp; 396 int64_t end; 397 398 /* 399 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE 400 * aligned boundaries, if the range is not aligned. As a result a 401 * DEV_BSIZE subrange with partially dirty data may get marked as clean. 402 * It may happen that all DEV_BSIZE subranges are marked clean and thus 403 * the whole page would be considred clean despite have some dirty data. 404 * For this reason we should shrink the range to DEV_BSIZE aligned 405 * boundaries before calling vm_page_clear_dirty. 406 */ 407 end = rounddown2(off + nbytes, DEV_BSIZE); 408 off = roundup2(off, DEV_BSIZE); 409 nbytes = end - off; 410 411 obj = vp->v_object; 412 zfs_vmobject_assert_wlocked(obj); 413 414 for (;;) { 415 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 416 pp->valid) { 417 if (vm_page_xbusied(pp)) { 418 /* 419 * Reference the page before unlocking and 420 * sleeping so that the page daemon is less 421 * likely to reclaim it. 422 */ 423 vm_page_reference(pp); 424 vm_page_lock(pp); 425 zfs_vmobject_wunlock(obj); 426 vm_page_busy_sleep(pp, "zfsmwb"); 427 zfs_vmobject_wlock(obj); 428 continue; 429 } 430 vm_page_sbusy(pp); 431 } else if (pp == NULL) { 432 pp = vm_page_alloc(obj, OFF_TO_IDX(start), 433 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED | 434 VM_ALLOC_SBUSY); 435 } else { 436 ASSERT(pp != NULL && !pp->valid); 437 pp = NULL; 438 } 439 440 if (pp != NULL) { 441 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 442 vm_object_pip_add(obj, 1); 443 pmap_remove_write(pp); 444 if (nbytes != 0) 445 vm_page_clear_dirty(pp, off, nbytes); 446 } 447 break; 448 } 449 return (pp); 450} 451 452static void 453page_unbusy(vm_page_t pp) 454{ 455 456 vm_page_sunbusy(pp); 457 vm_object_pip_subtract(pp->object, 1); 458} 459 460static vm_page_t 461page_hold(vnode_t *vp, int64_t start) 462{ 463 vm_object_t obj; 464 vm_page_t pp; 465 466 obj = vp->v_object; 467 zfs_vmobject_assert_wlocked(obj); 468 469 for (;;) { 470 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 471 pp->valid) { 472 if (vm_page_xbusied(pp)) { 473 /* 474 * Reference the page before unlocking and 475 * sleeping so that the page daemon is less 476 * likely to reclaim it. 477 */ 478 vm_page_reference(pp); 479 vm_page_lock(pp); 480 zfs_vmobject_wunlock(obj); 481 vm_page_busy_sleep(pp, "zfsmwb"); 482 zfs_vmobject_wlock(obj); 483 continue; 484 } 485 486 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 487 vm_page_lock(pp); 488 vm_page_hold(pp); 489 vm_page_unlock(pp); 490 491 } else 492 pp = NULL; 493 break; 494 } 495 return (pp); 496} 497 498static void 499page_unhold(vm_page_t pp) 500{ 501 502 vm_page_lock(pp); 503 vm_page_unhold(pp); 504 vm_page_unlock(pp); 505} 506 507/* 508 * When a file is memory mapped, we must keep the IO data synchronized 509 * between the DMU cache and the memory mapped pages. What this means: 510 * 511 * On Write: If we find a memory mapped page, we write to *both* 512 * the page and the dmu buffer. 513 */ 514static void 515update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid, 516 int segflg, dmu_tx_t *tx) 517{ 518 vm_object_t obj; 519 struct sf_buf *sf; 520 caddr_t va; 521 int off; 522 523 ASSERT(segflg != UIO_NOCOPY); 524 ASSERT(vp->v_mount != NULL); 525 obj = vp->v_object; 526 ASSERT(obj != NULL); 527 528 off = start & PAGEOFFSET; 529 zfs_vmobject_wlock(obj); 530 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 531 vm_page_t pp; 532 int nbytes = imin(PAGESIZE - off, len); 533 534 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) { 535 zfs_vmobject_wunlock(obj); 536 537 va = zfs_map_page(pp, &sf); 538 (void) dmu_read(os, oid, start+off, nbytes, 539 va+off, DMU_READ_PREFETCH);; 540 zfs_unmap_page(sf); 541 542 zfs_vmobject_wlock(obj); 543 page_unbusy(pp); 544 } 545 len -= nbytes; 546 off = 0; 547 } 548 vm_object_pip_wakeupn(obj, 0); 549 zfs_vmobject_wunlock(obj); 550} 551 552/* 553 * Read with UIO_NOCOPY flag means that sendfile(2) requests 554 * ZFS to populate a range of page cache pages with data. 555 * 556 * NOTE: this function could be optimized to pre-allocate 557 * all pages in advance, drain exclusive busy on all of them, 558 * map them into contiguous KVA region and populate them 559 * in one single dmu_read() call. 560 */ 561static int 562mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio) 563{ 564 znode_t *zp = VTOZ(vp); 565 objset_t *os = zp->z_zfsvfs->z_os; 566 struct sf_buf *sf; 567 vm_object_t obj; 568 vm_page_t pp; 569 int64_t start; 570 caddr_t va; 571 int len = nbytes; 572 int off; 573 int error = 0; 574 575 ASSERT(uio->uio_segflg == UIO_NOCOPY); 576 ASSERT(vp->v_mount != NULL); 577 obj = vp->v_object; 578 ASSERT(obj != NULL); 579 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0); 580 581 zfs_vmobject_wlock(obj); 582 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) { 583 int bytes = MIN(PAGESIZE, len); 584 585 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY | 586 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); 587 if (pp->valid == 0) { 588 zfs_vmobject_wunlock(obj); 589 va = zfs_map_page(pp, &sf); 590 error = dmu_read(os, zp->z_id, start, bytes, va, 591 DMU_READ_PREFETCH); 592 if (bytes != PAGESIZE && error == 0) 593 bzero(va + bytes, PAGESIZE - bytes); 594 zfs_unmap_page(sf); 595 zfs_vmobject_wlock(obj); 596 vm_page_sunbusy(pp); 597 vm_page_lock(pp); 598 if (error) { 599 if (pp->wire_count == 0 && pp->valid == 0 && 600 !vm_page_busied(pp)) 601 vm_page_free(pp); 602 } else { 603 pp->valid = VM_PAGE_BITS_ALL; 604 vm_page_activate(pp); 605 } 606 vm_page_unlock(pp); 607 } else { 608 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 609 vm_page_sunbusy(pp); 610 } 611 if (error) 612 break; 613 uio->uio_resid -= bytes; 614 uio->uio_offset += bytes; 615 len -= bytes; 616 } 617 zfs_vmobject_wunlock(obj); 618 return (error); 619} 620 621/* 622 * When a file is memory mapped, we must keep the IO data synchronized 623 * between the DMU cache and the memory mapped pages. What this means: 624 * 625 * On Read: We "read" preferentially from memory mapped pages, 626 * else we default from the dmu buffer. 627 * 628 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 629 * the file is memory mapped. 630 */ 631static int 632mappedread(vnode_t *vp, int nbytes, uio_t *uio) 633{ 634 znode_t *zp = VTOZ(vp); 635 vm_object_t obj; 636 int64_t start; 637 caddr_t va; 638 int len = nbytes; 639 int off; 640 int error = 0; 641 642 ASSERT(vp->v_mount != NULL); 643 obj = vp->v_object; 644 ASSERT(obj != NULL); 645 646 start = uio->uio_loffset; 647 off = start & PAGEOFFSET; 648 zfs_vmobject_wlock(obj); 649 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 650 vm_page_t pp; 651 uint64_t bytes = MIN(PAGESIZE - off, len); 652 653 if (pp = page_hold(vp, start)) { 654 struct sf_buf *sf; 655 caddr_t va; 656 657 zfs_vmobject_wunlock(obj); 658 va = zfs_map_page(pp, &sf); 659 error = uiomove(va + off, bytes, UIO_READ, uio); 660 zfs_unmap_page(sf); 661 zfs_vmobject_wlock(obj); 662 page_unhold(pp); 663 } else { 664 zfs_vmobject_wunlock(obj); 665 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 666 uio, bytes); 667 zfs_vmobject_wlock(obj); 668 } 669 len -= bytes; 670 off = 0; 671 if (error) 672 break; 673 } 674 zfs_vmobject_wunlock(obj); 675 return (error); 676} 677 678offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 679 680/* 681 * Read bytes from specified file into supplied buffer. 682 * 683 * IN: vp - vnode of file to be read from. 684 * uio - structure supplying read location, range info, 685 * and return buffer. 686 * ioflag - SYNC flags; used to provide FRSYNC semantics. 687 * cr - credentials of caller. 688 * ct - caller context 689 * 690 * OUT: uio - updated offset and range, buffer filled. 691 * 692 * RETURN: 0 on success, error code on failure. 693 * 694 * Side Effects: 695 * vp - atime updated if byte count > 0 696 */ 697/* ARGSUSED */ 698static int 699zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 700{ 701 znode_t *zp = VTOZ(vp); 702 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 703 ssize_t n, nbytes; 704 int error = 0; 705 rl_t *rl; 706 xuio_t *xuio = NULL; 707 708 ZFS_ENTER(zfsvfs); 709 ZFS_VERIFY_ZP(zp); 710 711 if (zp->z_pflags & ZFS_AV_QUARANTINED) { 712 ZFS_EXIT(zfsvfs); 713 return (SET_ERROR(EACCES)); 714 } 715 716 /* 717 * Validate file offset 718 */ 719 if (uio->uio_loffset < (offset_t)0) { 720 ZFS_EXIT(zfsvfs); 721 return (SET_ERROR(EINVAL)); 722 } 723 724 /* 725 * Fasttrack empty reads 726 */ 727 if (uio->uio_resid == 0) { 728 ZFS_EXIT(zfsvfs); 729 return (0); 730 } 731 732 /* 733 * Check for mandatory locks 734 */ 735 if (MANDMODE(zp->z_mode)) { 736 if (error = chklock(vp, FREAD, 737 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 738 ZFS_EXIT(zfsvfs); 739 return (error); 740 } 741 } 742 743 /* 744 * If we're in FRSYNC mode, sync out this znode before reading it. 745 */ 746 if (zfsvfs->z_log && 747 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)) 748 zil_commit(zfsvfs->z_log, zp->z_id); 749 750 /* 751 * Lock the range against changes. 752 */ 753 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 754 755 /* 756 * If we are reading past end-of-file we can skip 757 * to the end; but we might still need to set atime. 758 */ 759 if (uio->uio_loffset >= zp->z_size) { 760 error = 0; 761 goto out; 762 } 763 764 ASSERT(uio->uio_loffset < zp->z_size); 765 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset); 766 767#ifdef illumos 768 if ((uio->uio_extflg == UIO_XUIO) && 769 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) { 770 int nblk; 771 int blksz = zp->z_blksz; 772 uint64_t offset = uio->uio_loffset; 773 774 xuio = (xuio_t *)uio; 775 if ((ISP2(blksz))) { 776 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset, 777 blksz)) / blksz; 778 } else { 779 ASSERT(offset + n <= blksz); 780 nblk = 1; 781 } 782 (void) dmu_xuio_init(xuio, nblk); 783 784 if (vn_has_cached_data(vp)) { 785 /* 786 * For simplicity, we always allocate a full buffer 787 * even if we only expect to read a portion of a block. 788 */ 789 while (--nblk >= 0) { 790 (void) dmu_xuio_add(xuio, 791 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 792 blksz), 0, blksz); 793 } 794 } 795 } 796#endif /* illumos */ 797 798 while (n > 0) { 799 nbytes = MIN(n, zfs_read_chunk_size - 800 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 801 802#ifdef __FreeBSD__ 803 if (uio->uio_segflg == UIO_NOCOPY) 804 error = mappedread_sf(vp, nbytes, uio); 805 else 806#endif /* __FreeBSD__ */ 807 if (vn_has_cached_data(vp)) { 808 error = mappedread(vp, nbytes, uio); 809 } else { 810 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 811 uio, nbytes); 812 } 813 if (error) { 814 /* convert checksum errors into IO errors */ 815 if (error == ECKSUM) 816 error = SET_ERROR(EIO); 817 break; 818 } 819 820 n -= nbytes; 821 } 822out: 823 zfs_range_unlock(rl); 824 825 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 826 ZFS_EXIT(zfsvfs); 827 return (error); 828} 829 830/* 831 * Write the bytes to a file. 832 * 833 * IN: vp - vnode of file to be written to. 834 * uio - structure supplying write location, range info, 835 * and data buffer. 836 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is 837 * set if in append mode. 838 * cr - credentials of caller. 839 * ct - caller context (NFS/CIFS fem monitor only) 840 * 841 * OUT: uio - updated offset and range. 842 * 843 * RETURN: 0 on success, error code on failure. 844 * 845 * Timestamps: 846 * vp - ctime|mtime updated if byte count > 0 847 */ 848 849/* ARGSUSED */ 850static int 851zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 852{ 853 znode_t *zp = VTOZ(vp); 854 rlim64_t limit = MAXOFFSET_T; 855 ssize_t start_resid = uio->uio_resid; 856 ssize_t tx_bytes; 857 uint64_t end_size; 858 dmu_tx_t *tx; 859 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 860 zilog_t *zilog; 861 offset_t woff; 862 ssize_t n, nbytes; 863 rl_t *rl; 864 int max_blksz = zfsvfs->z_max_blksz; 865 int error = 0; 866 arc_buf_t *abuf; 867 iovec_t *aiov = NULL; 868 xuio_t *xuio = NULL; 869 int i_iov = 0; 870 int iovcnt = uio->uio_iovcnt; 871 iovec_t *iovp = uio->uio_iov; 872 int write_eof; 873 int count = 0; 874 sa_bulk_attr_t bulk[4]; 875 uint64_t mtime[2], ctime[2]; 876 877 /* 878 * Fasttrack empty write 879 */ 880 n = start_resid; 881 if (n == 0) 882 return (0); 883 884 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 885 limit = MAXOFFSET_T; 886 887 ZFS_ENTER(zfsvfs); 888 ZFS_VERIFY_ZP(zp); 889 890 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 891 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 892 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 893 &zp->z_size, 8); 894 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 895 &zp->z_pflags, 8); 896 897 /* 898 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our 899 * callers might not be able to detect properly that we are read-only, 900 * so check it explicitly here. 901 */ 902 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 903 ZFS_EXIT(zfsvfs); 904 return (SET_ERROR(EROFS)); 905 } 906 907 /* 908 * If immutable or not appending then return EPERM 909 */ 910 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 911 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 912 (uio->uio_loffset < zp->z_size))) { 913 ZFS_EXIT(zfsvfs); 914 return (SET_ERROR(EPERM)); 915 } 916 917 zilog = zfsvfs->z_log; 918 919 /* 920 * Validate file offset 921 */ 922 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset; 923 if (woff < 0) { 924 ZFS_EXIT(zfsvfs); 925 return (SET_ERROR(EINVAL)); 926 } 927 928 /* 929 * Check for mandatory locks before calling zfs_range_lock() 930 * in order to prevent a deadlock with locks set via fcntl(). 931 */ 932 if (MANDMODE((mode_t)zp->z_mode) && 933 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 934 ZFS_EXIT(zfsvfs); 935 return (error); 936 } 937 938#ifdef illumos 939 /* 940 * Pre-fault the pages to ensure slow (eg NFS) pages 941 * don't hold up txg. 942 * Skip this if uio contains loaned arc_buf. 943 */ 944 if ((uio->uio_extflg == UIO_XUIO) && 945 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) 946 xuio = (xuio_t *)uio; 947 else 948 uio_prefaultpages(MIN(n, max_blksz), uio); 949#endif 950 951 /* 952 * If in append mode, set the io offset pointer to eof. 953 */ 954 if (ioflag & FAPPEND) { 955 /* 956 * Obtain an appending range lock to guarantee file append 957 * semantics. We reset the write offset once we have the lock. 958 */ 959 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 960 woff = rl->r_off; 961 if (rl->r_len == UINT64_MAX) { 962 /* 963 * We overlocked the file because this write will cause 964 * the file block size to increase. 965 * Note that zp_size cannot change with this lock held. 966 */ 967 woff = zp->z_size; 968 } 969 uio->uio_loffset = woff; 970 } else { 971 /* 972 * Note that if the file block size will change as a result of 973 * this write, then this range lock will lock the entire file 974 * so that we can re-write the block safely. 975 */ 976 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 977 } 978 979 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) { 980 zfs_range_unlock(rl); 981 ZFS_EXIT(zfsvfs); 982 return (EFBIG); 983 } 984 985 if (woff >= limit) { 986 zfs_range_unlock(rl); 987 ZFS_EXIT(zfsvfs); 988 return (SET_ERROR(EFBIG)); 989 } 990 991 if ((woff + n) > limit || woff > (limit - n)) 992 n = limit - woff; 993 994 /* Will this write extend the file length? */ 995 write_eof = (woff + n > zp->z_size); 996 997 end_size = MAX(zp->z_size, woff + n); 998 999 /* 1000 * Write the file in reasonable size chunks. Each chunk is written 1001 * in a separate transaction; this keeps the intent log records small 1002 * and allows us to do more fine-grained space accounting. 1003 */ 1004 while (n > 0) { 1005 abuf = NULL; 1006 woff = uio->uio_loffset; 1007 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 1008 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 1009 if (abuf != NULL) 1010 dmu_return_arcbuf(abuf); 1011 error = SET_ERROR(EDQUOT); 1012 break; 1013 } 1014 1015 if (xuio && abuf == NULL) { 1016 ASSERT(i_iov < iovcnt); 1017 aiov = &iovp[i_iov]; 1018 abuf = dmu_xuio_arcbuf(xuio, i_iov); 1019 dmu_xuio_clear(xuio, i_iov); 1020 DTRACE_PROBE3(zfs_cp_write, int, i_iov, 1021 iovec_t *, aiov, arc_buf_t *, abuf); 1022 ASSERT((aiov->iov_base == abuf->b_data) || 1023 ((char *)aiov->iov_base - (char *)abuf->b_data + 1024 aiov->iov_len == arc_buf_size(abuf))); 1025 i_iov++; 1026 } else if (abuf == NULL && n >= max_blksz && 1027 woff >= zp->z_size && 1028 P2PHASE(woff, max_blksz) == 0 && 1029 zp->z_blksz == max_blksz) { 1030 /* 1031 * This write covers a full block. "Borrow" a buffer 1032 * from the dmu so that we can fill it before we enter 1033 * a transaction. This avoids the possibility of 1034 * holding up the transaction if the data copy hangs 1035 * up on a pagefault (e.g., from an NFS server mapping). 1036 */ 1037 size_t cbytes; 1038 1039 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 1040 max_blksz); 1041 ASSERT(abuf != NULL); 1042 ASSERT(arc_buf_size(abuf) == max_blksz); 1043 if (error = uiocopy(abuf->b_data, max_blksz, 1044 UIO_WRITE, uio, &cbytes)) { 1045 dmu_return_arcbuf(abuf); 1046 break; 1047 } 1048 ASSERT(cbytes == max_blksz); 1049 } 1050 1051 /* 1052 * Start a transaction. 1053 */ 1054 tx = dmu_tx_create(zfsvfs->z_os); 1055 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1056 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 1057 zfs_sa_upgrade_txholds(tx, zp); 1058 error = dmu_tx_assign(tx, TXG_WAIT); 1059 if (error) { 1060 dmu_tx_abort(tx); 1061 if (abuf != NULL) 1062 dmu_return_arcbuf(abuf); 1063 break; 1064 } 1065 1066 /* 1067 * If zfs_range_lock() over-locked we grow the blocksize 1068 * and then reduce the lock range. This will only happen 1069 * on the first iteration since zfs_range_reduce() will 1070 * shrink down r_len to the appropriate size. 1071 */ 1072 if (rl->r_len == UINT64_MAX) { 1073 uint64_t new_blksz; 1074 1075 if (zp->z_blksz > max_blksz) { 1076 /* 1077 * File's blocksize is already larger than the 1078 * "recordsize" property. Only let it grow to 1079 * the next power of 2. 1080 */ 1081 ASSERT(!ISP2(zp->z_blksz)); 1082 new_blksz = MIN(end_size, 1083 1 << highbit64(zp->z_blksz)); 1084 } else { 1085 new_blksz = MIN(end_size, max_blksz); 1086 } 1087 zfs_grow_blocksize(zp, new_blksz, tx); 1088 zfs_range_reduce(rl, woff, n); 1089 } 1090 1091 /* 1092 * XXX - should we really limit each write to z_max_blksz? 1093 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 1094 */ 1095 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 1096 1097 if (woff + nbytes > zp->z_size) 1098 vnode_pager_setsize(vp, woff + nbytes); 1099 1100 if (abuf == NULL) { 1101 tx_bytes = uio->uio_resid; 1102 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), 1103 uio, nbytes, tx); 1104 tx_bytes -= uio->uio_resid; 1105 } else { 1106 tx_bytes = nbytes; 1107 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len); 1108 /* 1109 * If this is not a full block write, but we are 1110 * extending the file past EOF and this data starts 1111 * block-aligned, use assign_arcbuf(). Otherwise, 1112 * write via dmu_write(). 1113 */ 1114 if (tx_bytes < max_blksz && (!write_eof || 1115 aiov->iov_base != abuf->b_data)) { 1116 ASSERT(xuio); 1117 dmu_write(zfsvfs->z_os, zp->z_id, woff, 1118 aiov->iov_len, aiov->iov_base, tx); 1119 dmu_return_arcbuf(abuf); 1120 xuio_stat_wbuf_copied(); 1121 } else { 1122 ASSERT(xuio || tx_bytes == max_blksz); 1123 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), 1124 woff, abuf, tx); 1125 } 1126 ASSERT(tx_bytes <= uio->uio_resid); 1127 uioskip(uio, tx_bytes); 1128 } 1129 if (tx_bytes && vn_has_cached_data(vp)) { 1130 update_pages(vp, woff, tx_bytes, zfsvfs->z_os, 1131 zp->z_id, uio->uio_segflg, tx); 1132 } 1133 1134 /* 1135 * If we made no progress, we're done. If we made even 1136 * partial progress, update the znode and ZIL accordingly. 1137 */ 1138 if (tx_bytes == 0) { 1139 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 1140 (void *)&zp->z_size, sizeof (uint64_t), tx); 1141 dmu_tx_commit(tx); 1142 ASSERT(error != 0); 1143 break; 1144 } 1145 1146 /* 1147 * Clear Set-UID/Set-GID bits on successful write if not 1148 * privileged and at least one of the excute bits is set. 1149 * 1150 * It would be nice to to this after all writes have 1151 * been done, but that would still expose the ISUID/ISGID 1152 * to another app after the partial write is committed. 1153 * 1154 * Note: we don't call zfs_fuid_map_id() here because 1155 * user 0 is not an ephemeral uid. 1156 */ 1157 mutex_enter(&zp->z_acl_lock); 1158 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | 1159 (S_IXUSR >> 6))) != 0 && 1160 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && 1161 secpolicy_vnode_setid_retain(vp, cr, 1162 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) { 1163 uint64_t newmode; 1164 zp->z_mode &= ~(S_ISUID | S_ISGID); 1165 newmode = zp->z_mode; 1166 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), 1167 (void *)&newmode, sizeof (uint64_t), tx); 1168 } 1169 mutex_exit(&zp->z_acl_lock); 1170 1171 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 1172 B_TRUE); 1173 1174 /* 1175 * Update the file size (zp_size) if it has changed; 1176 * account for possible concurrent updates. 1177 */ 1178 while ((end_size = zp->z_size) < uio->uio_loffset) { 1179 (void) atomic_cas_64(&zp->z_size, end_size, 1180 uio->uio_loffset); 1181 ASSERT(error == 0); 1182 } 1183 /* 1184 * If we are replaying and eof is non zero then force 1185 * the file size to the specified eof. Note, there's no 1186 * concurrency during replay. 1187 */ 1188 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) 1189 zp->z_size = zfsvfs->z_replay_eof; 1190 1191 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1192 1193 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 1194 dmu_tx_commit(tx); 1195 1196 if (error != 0) 1197 break; 1198 ASSERT(tx_bytes == nbytes); 1199 n -= nbytes; 1200 1201#ifdef illumos 1202 if (!xuio && n > 0) 1203 uio_prefaultpages(MIN(n, max_blksz), uio); 1204#endif 1205 } 1206 1207 zfs_range_unlock(rl); 1208 1209 /* 1210 * If we're in replay mode, or we made no progress, return error. 1211 * Otherwise, it's at least a partial write, so it's successful. 1212 */ 1213 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 1214 ZFS_EXIT(zfsvfs); 1215 return (error); 1216 } 1217 1218 if (ioflag & (FSYNC | FDSYNC) || 1219 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1220 zil_commit(zilog, zp->z_id); 1221 1222 ZFS_EXIT(zfsvfs); 1223 return (0); 1224} 1225 1226void 1227zfs_get_done(zgd_t *zgd, int error) 1228{ 1229 znode_t *zp = zgd->zgd_private; 1230 objset_t *os = zp->z_zfsvfs->z_os; 1231 1232 if (zgd->zgd_db) 1233 dmu_buf_rele(zgd->zgd_db, zgd); 1234 1235 zfs_range_unlock(zgd->zgd_rl); 1236 1237 /* 1238 * Release the vnode asynchronously as we currently have the 1239 * txg stopped from syncing. 1240 */ 1241 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1242 1243 if (error == 0 && zgd->zgd_bp) 1244 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1245 1246 kmem_free(zgd, sizeof (zgd_t)); 1247} 1248 1249#ifdef DEBUG 1250static int zil_fault_io = 0; 1251#endif 1252 1253/* 1254 * Get data to generate a TX_WRITE intent log record. 1255 */ 1256int 1257zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1258{ 1259 zfsvfs_t *zfsvfs = arg; 1260 objset_t *os = zfsvfs->z_os; 1261 znode_t *zp; 1262 uint64_t object = lr->lr_foid; 1263 uint64_t offset = lr->lr_offset; 1264 uint64_t size = lr->lr_length; 1265 blkptr_t *bp = &lr->lr_blkptr; 1266 dmu_buf_t *db; 1267 zgd_t *zgd; 1268 int error = 0; 1269 1270 ASSERT(zio != NULL); 1271 ASSERT(size != 0); 1272 1273 /* 1274 * Nothing to do if the file has been removed 1275 */ 1276 if (zfs_zget(zfsvfs, object, &zp) != 0) 1277 return (SET_ERROR(ENOENT)); 1278 if (zp->z_unlinked) { 1279 /* 1280 * Release the vnode asynchronously as we currently have the 1281 * txg stopped from syncing. 1282 */ 1283 VN_RELE_ASYNC(ZTOV(zp), 1284 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1285 return (SET_ERROR(ENOENT)); 1286 } 1287 1288 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1289 zgd->zgd_zilog = zfsvfs->z_log; 1290 zgd->zgd_private = zp; 1291 1292 /* 1293 * Write records come in two flavors: immediate and indirect. 1294 * For small writes it's cheaper to store the data with the 1295 * log record (immediate); for large writes it's cheaper to 1296 * sync the data and get a pointer to it (indirect) so that 1297 * we don't have to write the data twice. 1298 */ 1299 if (buf != NULL) { /* immediate write */ 1300 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER); 1301 /* test for truncation needs to be done while range locked */ 1302 if (offset >= zp->z_size) { 1303 error = SET_ERROR(ENOENT); 1304 } else { 1305 error = dmu_read(os, object, offset, size, buf, 1306 DMU_READ_NO_PREFETCH); 1307 } 1308 ASSERT(error == 0 || error == ENOENT); 1309 } else { /* indirect write */ 1310 /* 1311 * Have to lock the whole block to ensure when it's 1312 * written out and it's checksum is being calculated 1313 * that no one can change the data. We need to re-check 1314 * blocksize after we get the lock in case it's changed! 1315 */ 1316 for (;;) { 1317 uint64_t blkoff; 1318 size = zp->z_blksz; 1319 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; 1320 offset -= blkoff; 1321 zgd->zgd_rl = zfs_range_lock(zp, offset, size, 1322 RL_READER); 1323 if (zp->z_blksz == size) 1324 break; 1325 offset += blkoff; 1326 zfs_range_unlock(zgd->zgd_rl); 1327 } 1328 /* test for truncation needs to be done while range locked */ 1329 if (lr->lr_offset >= zp->z_size) 1330 error = SET_ERROR(ENOENT); 1331#ifdef DEBUG 1332 if (zil_fault_io) { 1333 error = SET_ERROR(EIO); 1334 zil_fault_io = 0; 1335 } 1336#endif 1337 if (error == 0) 1338 error = dmu_buf_hold(os, object, offset, zgd, &db, 1339 DMU_READ_NO_PREFETCH); 1340 1341 if (error == 0) { 1342 blkptr_t *obp = dmu_buf_get_blkptr(db); 1343 if (obp) { 1344 ASSERT(BP_IS_HOLE(bp)); 1345 *bp = *obp; 1346 } 1347 1348 zgd->zgd_db = db; 1349 zgd->zgd_bp = bp; 1350 1351 ASSERT(db->db_offset == offset); 1352 ASSERT(db->db_size == size); 1353 1354 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1355 zfs_get_done, zgd); 1356 ASSERT(error || lr->lr_length <= zp->z_blksz); 1357 1358 /* 1359 * On success, we need to wait for the write I/O 1360 * initiated by dmu_sync() to complete before we can 1361 * release this dbuf. We will finish everything up 1362 * in the zfs_get_done() callback. 1363 */ 1364 if (error == 0) 1365 return (0); 1366 1367 if (error == EALREADY) { 1368 lr->lr_common.lrc_txtype = TX_WRITE2; 1369 error = 0; 1370 } 1371 } 1372 } 1373 1374 zfs_get_done(zgd, error); 1375 1376 return (error); 1377} 1378 1379/*ARGSUSED*/ 1380static int 1381zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 1382 caller_context_t *ct) 1383{ 1384 znode_t *zp = VTOZ(vp); 1385 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1386 int error; 1387 1388 ZFS_ENTER(zfsvfs); 1389 ZFS_VERIFY_ZP(zp); 1390 1391 if (flag & V_ACE_MASK) 1392 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 1393 else 1394 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1395 1396 ZFS_EXIT(zfsvfs); 1397 return (error); 1398} 1399 1400/* 1401 * If vnode is for a device return a specfs vnode instead. 1402 */ 1403static int 1404specvp_check(vnode_t **vpp, cred_t *cr) 1405{ 1406 int error = 0; 1407 1408 if (IS_DEVVP(*vpp)) { 1409 struct vnode *svp; 1410 1411 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1412 VN_RELE(*vpp); 1413 if (svp == NULL) 1414 error = SET_ERROR(ENOSYS); 1415 *vpp = svp; 1416 } 1417 return (error); 1418} 1419 1420 1421/* 1422 * Lookup an entry in a directory, or an extended attribute directory. 1423 * If it exists, return a held vnode reference for it. 1424 * 1425 * IN: dvp - vnode of directory to search. 1426 * nm - name of entry to lookup. 1427 * pnp - full pathname to lookup [UNUSED]. 1428 * flags - LOOKUP_XATTR set if looking for an attribute. 1429 * rdir - root directory vnode [UNUSED]. 1430 * cr - credentials of caller. 1431 * ct - caller context 1432 * direntflags - directory lookup flags 1433 * realpnp - returned pathname. 1434 * 1435 * OUT: vpp - vnode of located entry, NULL if not found. 1436 * 1437 * RETURN: 0 on success, error code on failure. 1438 * 1439 * Timestamps: 1440 * NA 1441 */ 1442/* ARGSUSED */ 1443static int 1444zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 1445 int nameiop, cred_t *cr, kthread_t *td, int flags) 1446{ 1447 znode_t *zdp = VTOZ(dvp); 1448 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1449 int error = 0; 1450 int *direntflags = NULL; 1451 void *realpnp = NULL; 1452 1453 /* fast path */ 1454 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 1455 1456 if (dvp->v_type != VDIR) { 1457 return (SET_ERROR(ENOTDIR)); 1458 } else if (zdp->z_sa_hdl == NULL) { 1459 return (SET_ERROR(EIO)); 1460 } 1461 1462 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 1463 error = zfs_fastaccesschk_execute(zdp, cr); 1464 if (!error) { 1465 *vpp = dvp; 1466 VN_HOLD(*vpp); 1467 return (0); 1468 } 1469 return (error); 1470 } else { 1471 vnode_t *tvp = dnlc_lookup(dvp, nm); 1472 1473 if (tvp) { 1474 error = zfs_fastaccesschk_execute(zdp, cr); 1475 if (error) { 1476 VN_RELE(tvp); 1477 return (error); 1478 } 1479 if (tvp == DNLC_NO_VNODE) { 1480 VN_RELE(tvp); 1481 return (SET_ERROR(ENOENT)); 1482 } else { 1483 *vpp = tvp; 1484 return (specvp_check(vpp, cr)); 1485 } 1486 } 1487 } 1488 } 1489 1490 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1491 1492 ZFS_ENTER(zfsvfs); 1493 ZFS_VERIFY_ZP(zdp); 1494 1495 *vpp = NULL; 1496 1497 if (flags & LOOKUP_XATTR) { 1498#ifdef TODO 1499 /* 1500 * If the xattr property is off, refuse the lookup request. 1501 */ 1502 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1503 ZFS_EXIT(zfsvfs); 1504 return (SET_ERROR(EINVAL)); 1505 } 1506#endif 1507 1508 /* 1509 * We don't allow recursive attributes.. 1510 * Maybe someday we will. 1511 */ 1512 if (zdp->z_pflags & ZFS_XATTR) { 1513 ZFS_EXIT(zfsvfs); 1514 return (SET_ERROR(EINVAL)); 1515 } 1516 1517 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1518 ZFS_EXIT(zfsvfs); 1519 return (error); 1520 } 1521 1522 /* 1523 * Do we have permission to get into attribute directory? 1524 */ 1525 1526 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1527 B_FALSE, cr)) { 1528 VN_RELE(*vpp); 1529 *vpp = NULL; 1530 } 1531 1532 ZFS_EXIT(zfsvfs); 1533 return (error); 1534 } 1535 1536 if (dvp->v_type != VDIR) { 1537 ZFS_EXIT(zfsvfs); 1538 return (SET_ERROR(ENOTDIR)); 1539 } 1540 1541 /* 1542 * Check accessibility of directory. 1543 */ 1544 1545 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1546 ZFS_EXIT(zfsvfs); 1547 return (error); 1548 } 1549 1550 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1551 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1552 ZFS_EXIT(zfsvfs); 1553 return (SET_ERROR(EILSEQ)); 1554 } 1555 1556 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1557 if (error == 0) 1558 error = specvp_check(vpp, cr); 1559 1560 /* Translate errors and add SAVENAME when needed. */ 1561 if (cnp->cn_flags & ISLASTCN) { 1562 switch (nameiop) { 1563 case CREATE: 1564 case RENAME: 1565 if (error == ENOENT) { 1566 error = EJUSTRETURN; 1567 cnp->cn_flags |= SAVENAME; 1568 break; 1569 } 1570 /* FALLTHROUGH */ 1571 case DELETE: 1572 if (error == 0) 1573 cnp->cn_flags |= SAVENAME; 1574 break; 1575 } 1576 } 1577 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) { 1578 int ltype = 0; 1579 1580 if (cnp->cn_flags & ISDOTDOT) { 1581 ltype = VOP_ISLOCKED(dvp); 1582 VOP_UNLOCK(dvp, 0); 1583 } 1584 ZFS_EXIT(zfsvfs); 1585 error = vn_lock(*vpp, cnp->cn_lkflags); 1586 if (cnp->cn_flags & ISDOTDOT) 1587 vn_lock(dvp, ltype | LK_RETRY); 1588 if (error != 0) { 1589 VN_RELE(*vpp); 1590 *vpp = NULL; 1591 return (error); 1592 } 1593 } else { 1594 ZFS_EXIT(zfsvfs); 1595 } 1596 1597#ifdef FREEBSD_NAMECACHE 1598 /* 1599 * Insert name into cache (as non-existent) if appropriate. 1600 */ 1601 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0) 1602 cache_enter(dvp, *vpp, cnp); 1603 /* 1604 * Insert name into cache if appropriate. 1605 */ 1606 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1607 if (!(cnp->cn_flags & ISLASTCN) || 1608 (nameiop != DELETE && nameiop != RENAME)) { 1609 cache_enter(dvp, *vpp, cnp); 1610 } 1611 } 1612#endif 1613 1614 return (error); 1615} 1616 1617/* 1618 * Attempt to create a new entry in a directory. If the entry 1619 * already exists, truncate the file if permissible, else return 1620 * an error. Return the vp of the created or trunc'd file. 1621 * 1622 * IN: dvp - vnode of directory to put new file entry in. 1623 * name - name of new file entry. 1624 * vap - attributes of new file. 1625 * excl - flag indicating exclusive or non-exclusive mode. 1626 * mode - mode to open file with. 1627 * cr - credentials of caller. 1628 * flag - large file flag [UNUSED]. 1629 * ct - caller context 1630 * vsecp - ACL to be set 1631 * 1632 * OUT: vpp - vnode of created or trunc'd entry. 1633 * 1634 * RETURN: 0 on success, error code on failure. 1635 * 1636 * Timestamps: 1637 * dvp - ctime|mtime updated if new entry created 1638 * vp - ctime|mtime always, atime if new 1639 */ 1640 1641/* ARGSUSED */ 1642static int 1643zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1644 vnode_t **vpp, cred_t *cr, kthread_t *td) 1645{ 1646 znode_t *zp, *dzp = VTOZ(dvp); 1647 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1648 zilog_t *zilog; 1649 objset_t *os; 1650 zfs_dirlock_t *dl; 1651 dmu_tx_t *tx; 1652 int error; 1653 ksid_t *ksid; 1654 uid_t uid; 1655 gid_t gid = crgetgid(cr); 1656 zfs_acl_ids_t acl_ids; 1657 boolean_t fuid_dirtied; 1658 boolean_t have_acl = B_FALSE; 1659 boolean_t waited = B_FALSE; 1660 void *vsecp = NULL; 1661 int flag = 0; 1662 1663 /* 1664 * If we have an ephemeral id, ACL, or XVATTR then 1665 * make sure file system is at proper version 1666 */ 1667 1668 ksid = crgetsid(cr, KSID_OWNER); 1669 if (ksid) 1670 uid = ksid_getid(ksid); 1671 else 1672 uid = crgetuid(cr); 1673 1674 if (zfsvfs->z_use_fuids == B_FALSE && 1675 (vsecp || (vap->va_mask & AT_XVATTR) || 1676 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1677 return (SET_ERROR(EINVAL)); 1678 1679 ZFS_ENTER(zfsvfs); 1680 ZFS_VERIFY_ZP(dzp); 1681 os = zfsvfs->z_os; 1682 zilog = zfsvfs->z_log; 1683 1684 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1685 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1686 ZFS_EXIT(zfsvfs); 1687 return (SET_ERROR(EILSEQ)); 1688 } 1689 1690 if (vap->va_mask & AT_XVATTR) { 1691 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 1692 crgetuid(cr), cr, vap->va_type)) != 0) { 1693 ZFS_EXIT(zfsvfs); 1694 return (error); 1695 } 1696 } 1697 1698 getnewvnode_reserve(1); 1699 1700top: 1701 *vpp = NULL; 1702 1703 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) 1704 vap->va_mode &= ~S_ISVTX; 1705 1706 if (*name == '\0') { 1707 /* 1708 * Null component name refers to the directory itself. 1709 */ 1710 VN_HOLD(dvp); 1711 zp = dzp; 1712 dl = NULL; 1713 error = 0; 1714 } else { 1715 /* possible VN_HOLD(zp) */ 1716 int zflg = 0; 1717 1718 if (flag & FIGNORECASE) 1719 zflg |= ZCILOOK; 1720 1721 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1722 NULL, NULL); 1723 if (error) { 1724 if (have_acl) 1725 zfs_acl_ids_free(&acl_ids); 1726 if (strcmp(name, "..") == 0) 1727 error = SET_ERROR(EISDIR); 1728 getnewvnode_drop_reserve(); 1729 ZFS_EXIT(zfsvfs); 1730 return (error); 1731 } 1732 } 1733 1734 if (zp == NULL) { 1735 uint64_t txtype; 1736 1737 /* 1738 * Create a new file object and update the directory 1739 * to reference it. 1740 */ 1741 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1742 if (have_acl) 1743 zfs_acl_ids_free(&acl_ids); 1744 goto out; 1745 } 1746 1747 /* 1748 * We only support the creation of regular files in 1749 * extended attribute directories. 1750 */ 1751 1752 if ((dzp->z_pflags & ZFS_XATTR) && 1753 (vap->va_type != VREG)) { 1754 if (have_acl) 1755 zfs_acl_ids_free(&acl_ids); 1756 error = SET_ERROR(EINVAL); 1757 goto out; 1758 } 1759 1760 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 1761 cr, vsecp, &acl_ids)) != 0) 1762 goto out; 1763 have_acl = B_TRUE; 1764 1765 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1766 zfs_acl_ids_free(&acl_ids); 1767 error = SET_ERROR(EDQUOT); 1768 goto out; 1769 } 1770 1771 tx = dmu_tx_create(os); 1772 1773 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1774 ZFS_SA_BASE_ATTR_SIZE); 1775 1776 fuid_dirtied = zfsvfs->z_fuid_dirty; 1777 if (fuid_dirtied) 1778 zfs_fuid_txhold(zfsvfs, tx); 1779 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1780 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1781 if (!zfsvfs->z_use_sa && 1782 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1783 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1784 0, acl_ids.z_aclp->z_acl_bytes); 1785 } 1786 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 1787 if (error) { 1788 zfs_dirent_unlock(dl); 1789 if (error == ERESTART) { 1790 waited = B_TRUE; 1791 dmu_tx_wait(tx); 1792 dmu_tx_abort(tx); 1793 goto top; 1794 } 1795 zfs_acl_ids_free(&acl_ids); 1796 dmu_tx_abort(tx); 1797 getnewvnode_drop_reserve(); 1798 ZFS_EXIT(zfsvfs); 1799 return (error); 1800 } 1801 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1802 1803 if (fuid_dirtied) 1804 zfs_fuid_sync(zfsvfs, tx); 1805 1806 (void) zfs_link_create(dl, zp, tx, ZNEW); 1807 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1808 if (flag & FIGNORECASE) 1809 txtype |= TX_CI; 1810 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1811 vsecp, acl_ids.z_fuidp, vap); 1812 zfs_acl_ids_free(&acl_ids); 1813 dmu_tx_commit(tx); 1814 } else { 1815 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1816 1817 if (have_acl) 1818 zfs_acl_ids_free(&acl_ids); 1819 have_acl = B_FALSE; 1820 1821 /* 1822 * A directory entry already exists for this name. 1823 */ 1824 /* 1825 * Can't truncate an existing file if in exclusive mode. 1826 */ 1827 if (excl == EXCL) { 1828 error = SET_ERROR(EEXIST); 1829 goto out; 1830 } 1831 /* 1832 * Can't open a directory for writing. 1833 */ 1834 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1835 error = SET_ERROR(EISDIR); 1836 goto out; 1837 } 1838 /* 1839 * Verify requested access to file. 1840 */ 1841 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1842 goto out; 1843 } 1844 1845 mutex_enter(&dzp->z_lock); 1846 dzp->z_seq++; 1847 mutex_exit(&dzp->z_lock); 1848 1849 /* 1850 * Truncate regular files if requested. 1851 */ 1852 if ((ZTOV(zp)->v_type == VREG) && 1853 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1854 /* we can't hold any locks when calling zfs_freesp() */ 1855 zfs_dirent_unlock(dl); 1856 dl = NULL; 1857 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1858 if (error == 0) { 1859 vnevent_create(ZTOV(zp), ct); 1860 } 1861 } 1862 } 1863out: 1864 getnewvnode_drop_reserve(); 1865 if (dl) 1866 zfs_dirent_unlock(dl); 1867 1868 if (error) { 1869 if (zp) 1870 VN_RELE(ZTOV(zp)); 1871 } else { 1872 *vpp = ZTOV(zp); 1873 error = specvp_check(vpp, cr); 1874 } 1875 1876 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1877 zil_commit(zilog, 0); 1878 1879 ZFS_EXIT(zfsvfs); 1880 return (error); 1881} 1882 1883/* 1884 * Remove an entry from a directory. 1885 * 1886 * IN: dvp - vnode of directory to remove entry from. 1887 * name - name of entry to remove. 1888 * cr - credentials of caller. 1889 * ct - caller context 1890 * flags - case flags 1891 * 1892 * RETURN: 0 on success, error code on failure. 1893 * 1894 * Timestamps: 1895 * dvp - ctime|mtime 1896 * vp - ctime (if nlink > 0) 1897 */ 1898 1899uint64_t null_xattr = 0; 1900 1901/*ARGSUSED*/ 1902static int 1903zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1904 int flags) 1905{ 1906 znode_t *zp, *dzp = VTOZ(dvp); 1907 znode_t *xzp; 1908 vnode_t *vp; 1909 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1910 zilog_t *zilog; 1911 uint64_t acl_obj, xattr_obj; 1912 uint64_t xattr_obj_unlinked = 0; 1913 uint64_t obj = 0; 1914 zfs_dirlock_t *dl; 1915 dmu_tx_t *tx; 1916 boolean_t may_delete_now, delete_now = FALSE; 1917 boolean_t unlinked, toobig = FALSE; 1918 uint64_t txtype; 1919 pathname_t *realnmp = NULL; 1920 pathname_t realnm; 1921 int error; 1922 int zflg = ZEXISTS; 1923 boolean_t waited = B_FALSE; 1924 1925 ZFS_ENTER(zfsvfs); 1926 ZFS_VERIFY_ZP(dzp); 1927 zilog = zfsvfs->z_log; 1928 1929 if (flags & FIGNORECASE) { 1930 zflg |= ZCILOOK; 1931 pn_alloc(&realnm); 1932 realnmp = &realnm; 1933 } 1934 1935top: 1936 xattr_obj = 0; 1937 xzp = NULL; 1938 /* 1939 * Attempt to lock directory; fail if entry doesn't exist. 1940 */ 1941 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1942 NULL, realnmp)) { 1943 if (realnmp) 1944 pn_free(realnmp); 1945 ZFS_EXIT(zfsvfs); 1946 return (error); 1947 } 1948 1949 vp = ZTOV(zp); 1950 1951 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1952 goto out; 1953 } 1954 1955 /* 1956 * Need to use rmdir for removing directories. 1957 */ 1958 if (vp->v_type == VDIR) { 1959 error = SET_ERROR(EPERM); 1960 goto out; 1961 } 1962 1963 vnevent_remove(vp, dvp, name, ct); 1964 1965 if (realnmp) 1966 dnlc_remove(dvp, realnmp->pn_buf); 1967 else 1968 dnlc_remove(dvp, name); 1969 1970 VI_LOCK(vp); 1971 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1972 VI_UNLOCK(vp); 1973 1974 /* 1975 * We may delete the znode now, or we may put it in the unlinked set; 1976 * it depends on whether we're the last link, and on whether there are 1977 * other holds on the vnode. So we dmu_tx_hold() the right things to 1978 * allow for either case. 1979 */ 1980 obj = zp->z_id; 1981 tx = dmu_tx_create(zfsvfs->z_os); 1982 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1983 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1984 zfs_sa_upgrade_txholds(tx, zp); 1985 zfs_sa_upgrade_txholds(tx, dzp); 1986 if (may_delete_now) { 1987 toobig = 1988 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 1989 /* if the file is too big, only hold_free a token amount */ 1990 dmu_tx_hold_free(tx, zp->z_id, 0, 1991 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1992 } 1993 1994 /* are there any extended attributes? */ 1995 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1996 &xattr_obj, sizeof (xattr_obj)); 1997 if (error == 0 && xattr_obj) { 1998 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1999 ASSERT0(error); 2000 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2001 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 2002 } 2003 2004 mutex_enter(&zp->z_lock); 2005 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now) 2006 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 2007 mutex_exit(&zp->z_lock); 2008 2009 /* charge as an update -- would be nice not to charge at all */ 2010 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2011 2012 /* 2013 * Mark this transaction as typically resulting in a net free of space 2014 */ 2015 dmu_tx_mark_netfree(tx); 2016 2017 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2018 if (error) { 2019 zfs_dirent_unlock(dl); 2020 VN_RELE(vp); 2021 if (xzp) 2022 VN_RELE(ZTOV(xzp)); 2023 if (error == ERESTART) { 2024 waited = B_TRUE; 2025 dmu_tx_wait(tx); 2026 dmu_tx_abort(tx); 2027 goto top; 2028 } 2029 if (realnmp) 2030 pn_free(realnmp); 2031 dmu_tx_abort(tx); 2032 ZFS_EXIT(zfsvfs); 2033 return (error); 2034 } 2035 2036 /* 2037 * Remove the directory entry. 2038 */ 2039 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 2040 2041 if (error) { 2042 dmu_tx_commit(tx); 2043 goto out; 2044 } 2045 2046 if (unlinked) { 2047 /* 2048 * Hold z_lock so that we can make sure that the ACL obj 2049 * hasn't changed. Could have been deleted due to 2050 * zfs_sa_upgrade(). 2051 */ 2052 mutex_enter(&zp->z_lock); 2053 VI_LOCK(vp); 2054 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 2055 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked)); 2056 delete_now = may_delete_now && !toobig && 2057 vp->v_count == 1 && !vn_has_cached_data(vp) && 2058 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) == 2059 acl_obj; 2060 VI_UNLOCK(vp); 2061 } 2062 2063 if (delete_now) { 2064#ifdef __FreeBSD__ 2065 panic("zfs_remove: delete_now branch taken"); 2066#endif 2067 if (xattr_obj_unlinked) { 2068 ASSERT3U(xzp->z_links, ==, 2); 2069 mutex_enter(&xzp->z_lock); 2070 xzp->z_unlinked = 1; 2071 xzp->z_links = 0; 2072 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 2073 &xzp->z_links, sizeof (xzp->z_links), tx); 2074 ASSERT3U(error, ==, 0); 2075 mutex_exit(&xzp->z_lock); 2076 zfs_unlinked_add(xzp, tx); 2077 2078 if (zp->z_is_sa) 2079 error = sa_remove(zp->z_sa_hdl, 2080 SA_ZPL_XATTR(zfsvfs), tx); 2081 else 2082 error = sa_update(zp->z_sa_hdl, 2083 SA_ZPL_XATTR(zfsvfs), &null_xattr, 2084 sizeof (uint64_t), tx); 2085 ASSERT0(error); 2086 } 2087 VI_LOCK(vp); 2088 vp->v_count--; 2089 ASSERT0(vp->v_count); 2090 VI_UNLOCK(vp); 2091 mutex_exit(&zp->z_lock); 2092 zfs_znode_delete(zp, tx); 2093 } else if (unlinked) { 2094 mutex_exit(&zp->z_lock); 2095 zfs_unlinked_add(zp, tx); 2096#ifdef __FreeBSD__ 2097 vp->v_vflag |= VV_NOSYNC; 2098#endif 2099 } 2100 2101 txtype = TX_REMOVE; 2102 if (flags & FIGNORECASE) 2103 txtype |= TX_CI; 2104 zfs_log_remove(zilog, tx, txtype, dzp, name, obj); 2105 2106 dmu_tx_commit(tx); 2107out: 2108 if (realnmp) 2109 pn_free(realnmp); 2110 2111 zfs_dirent_unlock(dl); 2112 2113 if (!delete_now) 2114 VN_RELE(vp); 2115 if (xzp) 2116 VN_RELE(ZTOV(xzp)); 2117 2118 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2119 zil_commit(zilog, 0); 2120 2121 ZFS_EXIT(zfsvfs); 2122 return (error); 2123} 2124 2125/* 2126 * Create a new directory and insert it into dvp using the name 2127 * provided. Return a pointer to the inserted directory. 2128 * 2129 * IN: dvp - vnode of directory to add subdir to. 2130 * dirname - name of new directory. 2131 * vap - attributes of new directory. 2132 * cr - credentials of caller. 2133 * ct - caller context 2134 * flags - case flags 2135 * vsecp - ACL to be set 2136 * 2137 * OUT: vpp - vnode of created directory. 2138 * 2139 * RETURN: 0 on success, error code on failure. 2140 * 2141 * Timestamps: 2142 * dvp - ctime|mtime updated 2143 * vp - ctime|mtime|atime updated 2144 */ 2145/*ARGSUSED*/ 2146static int 2147zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 2148 caller_context_t *ct, int flags, vsecattr_t *vsecp) 2149{ 2150 znode_t *zp, *dzp = VTOZ(dvp); 2151 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2152 zilog_t *zilog; 2153 zfs_dirlock_t *dl; 2154 uint64_t txtype; 2155 dmu_tx_t *tx; 2156 int error; 2157 int zf = ZNEW; 2158 ksid_t *ksid; 2159 uid_t uid; 2160 gid_t gid = crgetgid(cr); 2161 zfs_acl_ids_t acl_ids; 2162 boolean_t fuid_dirtied; 2163 boolean_t waited = B_FALSE; 2164 2165 ASSERT(vap->va_type == VDIR); 2166 2167 /* 2168 * If we have an ephemeral id, ACL, or XVATTR then 2169 * make sure file system is at proper version 2170 */ 2171 2172 ksid = crgetsid(cr, KSID_OWNER); 2173 if (ksid) 2174 uid = ksid_getid(ksid); 2175 else 2176 uid = crgetuid(cr); 2177 if (zfsvfs->z_use_fuids == B_FALSE && 2178 (vsecp || (vap->va_mask & AT_XVATTR) || 2179 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 2180 return (SET_ERROR(EINVAL)); 2181 2182 ZFS_ENTER(zfsvfs); 2183 ZFS_VERIFY_ZP(dzp); 2184 zilog = zfsvfs->z_log; 2185 2186 if (dzp->z_pflags & ZFS_XATTR) { 2187 ZFS_EXIT(zfsvfs); 2188 return (SET_ERROR(EINVAL)); 2189 } 2190 2191 if (zfsvfs->z_utf8 && u8_validate(dirname, 2192 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2193 ZFS_EXIT(zfsvfs); 2194 return (SET_ERROR(EILSEQ)); 2195 } 2196 if (flags & FIGNORECASE) 2197 zf |= ZCILOOK; 2198 2199 if (vap->va_mask & AT_XVATTR) { 2200 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 2201 crgetuid(cr), cr, vap->va_type)) != 0) { 2202 ZFS_EXIT(zfsvfs); 2203 return (error); 2204 } 2205 } 2206 2207 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 2208 vsecp, &acl_ids)) != 0) { 2209 ZFS_EXIT(zfsvfs); 2210 return (error); 2211 } 2212 2213 getnewvnode_reserve(1); 2214 2215 /* 2216 * First make sure the new directory doesn't exist. 2217 * 2218 * Existence is checked first to make sure we don't return 2219 * EACCES instead of EEXIST which can cause some applications 2220 * to fail. 2221 */ 2222top: 2223 *vpp = NULL; 2224 2225 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 2226 NULL, NULL)) { 2227 zfs_acl_ids_free(&acl_ids); 2228 getnewvnode_drop_reserve(); 2229 ZFS_EXIT(zfsvfs); 2230 return (error); 2231 } 2232 2233 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 2234 zfs_acl_ids_free(&acl_ids); 2235 zfs_dirent_unlock(dl); 2236 getnewvnode_drop_reserve(); 2237 ZFS_EXIT(zfsvfs); 2238 return (error); 2239 } 2240 2241 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 2242 zfs_acl_ids_free(&acl_ids); 2243 zfs_dirent_unlock(dl); 2244 getnewvnode_drop_reserve(); 2245 ZFS_EXIT(zfsvfs); 2246 return (SET_ERROR(EDQUOT)); 2247 } 2248 2249 /* 2250 * Add a new entry to the directory. 2251 */ 2252 tx = dmu_tx_create(zfsvfs->z_os); 2253 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 2254 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2255 fuid_dirtied = zfsvfs->z_fuid_dirty; 2256 if (fuid_dirtied) 2257 zfs_fuid_txhold(zfsvfs, tx); 2258 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2259 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2260 acl_ids.z_aclp->z_acl_bytes); 2261 } 2262 2263 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 2264 ZFS_SA_BASE_ATTR_SIZE); 2265 2266 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2267 if (error) { 2268 zfs_dirent_unlock(dl); 2269 if (error == ERESTART) { 2270 waited = B_TRUE; 2271 dmu_tx_wait(tx); 2272 dmu_tx_abort(tx); 2273 goto top; 2274 } 2275 zfs_acl_ids_free(&acl_ids); 2276 dmu_tx_abort(tx); 2277 getnewvnode_drop_reserve(); 2278 ZFS_EXIT(zfsvfs); 2279 return (error); 2280 } 2281 2282 /* 2283 * Create new node. 2284 */ 2285 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 2286 2287 if (fuid_dirtied) 2288 zfs_fuid_sync(zfsvfs, tx); 2289 2290 /* 2291 * Now put new name in parent dir. 2292 */ 2293 (void) zfs_link_create(dl, zp, tx, ZNEW); 2294 2295 *vpp = ZTOV(zp); 2296 2297 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 2298 if (flags & FIGNORECASE) 2299 txtype |= TX_CI; 2300 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 2301 acl_ids.z_fuidp, vap); 2302 2303 zfs_acl_ids_free(&acl_ids); 2304 2305 dmu_tx_commit(tx); 2306 2307 getnewvnode_drop_reserve(); 2308 2309 zfs_dirent_unlock(dl); 2310 2311 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2312 zil_commit(zilog, 0); 2313 2314 ZFS_EXIT(zfsvfs); 2315 return (0); 2316} 2317 2318/* 2319 * Remove a directory subdir entry. If the current working 2320 * directory is the same as the subdir to be removed, the 2321 * remove will fail. 2322 * 2323 * IN: dvp - vnode of directory to remove from. 2324 * name - name of directory to be removed. 2325 * cwd - vnode of current working directory. 2326 * cr - credentials of caller. 2327 * ct - caller context 2328 * flags - case flags 2329 * 2330 * RETURN: 0 on success, error code on failure. 2331 * 2332 * Timestamps: 2333 * dvp - ctime|mtime updated 2334 */ 2335/*ARGSUSED*/ 2336static int 2337zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 2338 caller_context_t *ct, int flags) 2339{ 2340 znode_t *dzp = VTOZ(dvp); 2341 znode_t *zp; 2342 vnode_t *vp; 2343 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2344 zilog_t *zilog; 2345 zfs_dirlock_t *dl; 2346 dmu_tx_t *tx; 2347 int error; 2348 int zflg = ZEXISTS; 2349 boolean_t waited = B_FALSE; 2350 2351 ZFS_ENTER(zfsvfs); 2352 ZFS_VERIFY_ZP(dzp); 2353 zilog = zfsvfs->z_log; 2354 2355 if (flags & FIGNORECASE) 2356 zflg |= ZCILOOK; 2357top: 2358 zp = NULL; 2359 2360 /* 2361 * Attempt to lock directory; fail if entry doesn't exist. 2362 */ 2363 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 2364 NULL, NULL)) { 2365 ZFS_EXIT(zfsvfs); 2366 return (error); 2367 } 2368 2369 vp = ZTOV(zp); 2370 2371 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 2372 goto out; 2373 } 2374 2375 if (vp->v_type != VDIR) { 2376 error = SET_ERROR(ENOTDIR); 2377 goto out; 2378 } 2379 2380 if (vp == cwd) { 2381 error = SET_ERROR(EINVAL); 2382 goto out; 2383 } 2384 2385 vnevent_rmdir(vp, dvp, name, ct); 2386 2387 /* 2388 * Grab a lock on the directory to make sure that noone is 2389 * trying to add (or lookup) entries while we are removing it. 2390 */ 2391 rw_enter(&zp->z_name_lock, RW_WRITER); 2392 2393 /* 2394 * Grab a lock on the parent pointer to make sure we play well 2395 * with the treewalk and directory rename code. 2396 */ 2397 rw_enter(&zp->z_parent_lock, RW_WRITER); 2398 2399 tx = dmu_tx_create(zfsvfs->z_os); 2400 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2401 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2402 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2403 zfs_sa_upgrade_txholds(tx, zp); 2404 zfs_sa_upgrade_txholds(tx, dzp); 2405 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2406 if (error) { 2407 rw_exit(&zp->z_parent_lock); 2408 rw_exit(&zp->z_name_lock); 2409 zfs_dirent_unlock(dl); 2410 VN_RELE(vp); 2411 if (error == ERESTART) { 2412 waited = B_TRUE; 2413 dmu_tx_wait(tx); 2414 dmu_tx_abort(tx); 2415 goto top; 2416 } 2417 dmu_tx_abort(tx); 2418 ZFS_EXIT(zfsvfs); 2419 return (error); 2420 } 2421 2422#ifdef FREEBSD_NAMECACHE 2423 cache_purge(dvp); 2424#endif 2425 2426 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 2427 2428 if (error == 0) { 2429 uint64_t txtype = TX_RMDIR; 2430 if (flags & FIGNORECASE) 2431 txtype |= TX_CI; 2432 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT); 2433 } 2434 2435 dmu_tx_commit(tx); 2436 2437 rw_exit(&zp->z_parent_lock); 2438 rw_exit(&zp->z_name_lock); 2439#ifdef FREEBSD_NAMECACHE 2440 cache_purge(vp); 2441#endif 2442out: 2443 zfs_dirent_unlock(dl); 2444 2445 VN_RELE(vp); 2446 2447 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2448 zil_commit(zilog, 0); 2449 2450 ZFS_EXIT(zfsvfs); 2451 return (error); 2452} 2453 2454/* 2455 * Read as many directory entries as will fit into the provided 2456 * buffer from the given directory cursor position (specified in 2457 * the uio structure). 2458 * 2459 * IN: vp - vnode of directory to read. 2460 * uio - structure supplying read location, range info, 2461 * and return buffer. 2462 * cr - credentials of caller. 2463 * ct - caller context 2464 * flags - case flags 2465 * 2466 * OUT: uio - updated offset and range, buffer filled. 2467 * eofp - set to true if end-of-file detected. 2468 * 2469 * RETURN: 0 on success, error code on failure. 2470 * 2471 * Timestamps: 2472 * vp - atime updated 2473 * 2474 * Note that the low 4 bits of the cookie returned by zap is always zero. 2475 * This allows us to use the low range for "special" directory entries: 2476 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 2477 * we use the offset 2 for the '.zfs' directory. 2478 */ 2479/* ARGSUSED */ 2480static int 2481zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 2482{ 2483 znode_t *zp = VTOZ(vp); 2484 iovec_t *iovp; 2485 edirent_t *eodp; 2486 dirent64_t *odp; 2487 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2488 objset_t *os; 2489 caddr_t outbuf; 2490 size_t bufsize; 2491 zap_cursor_t zc; 2492 zap_attribute_t zap; 2493 uint_t bytes_wanted; 2494 uint64_t offset; /* must be unsigned; checks for < 1 */ 2495 uint64_t parent; 2496 int local_eof; 2497 int outcount; 2498 int error; 2499 uint8_t prefetch; 2500 boolean_t check_sysattrs; 2501 uint8_t type; 2502 int ncooks; 2503 u_long *cooks = NULL; 2504 int flags = 0; 2505 2506 ZFS_ENTER(zfsvfs); 2507 ZFS_VERIFY_ZP(zp); 2508 2509 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 2510 &parent, sizeof (parent))) != 0) { 2511 ZFS_EXIT(zfsvfs); 2512 return (error); 2513 } 2514 2515 /* 2516 * If we are not given an eof variable, 2517 * use a local one. 2518 */ 2519 if (eofp == NULL) 2520 eofp = &local_eof; 2521 2522 /* 2523 * Check for valid iov_len. 2524 */ 2525 if (uio->uio_iov->iov_len <= 0) { 2526 ZFS_EXIT(zfsvfs); 2527 return (SET_ERROR(EINVAL)); 2528 } 2529 2530 /* 2531 * Quit if directory has been removed (posix) 2532 */ 2533 if ((*eofp = zp->z_unlinked) != 0) { 2534 ZFS_EXIT(zfsvfs); 2535 return (0); 2536 } 2537 2538 error = 0; 2539 os = zfsvfs->z_os; 2540 offset = uio->uio_loffset; 2541 prefetch = zp->z_zn_prefetch; 2542 2543 /* 2544 * Initialize the iterator cursor. 2545 */ 2546 if (offset <= 3) { 2547 /* 2548 * Start iteration from the beginning of the directory. 2549 */ 2550 zap_cursor_init(&zc, os, zp->z_id); 2551 } else { 2552 /* 2553 * The offset is a serialized cursor. 2554 */ 2555 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 2556 } 2557 2558 /* 2559 * Get space to change directory entries into fs independent format. 2560 */ 2561 iovp = uio->uio_iov; 2562 bytes_wanted = iovp->iov_len; 2563 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 2564 bufsize = bytes_wanted; 2565 outbuf = kmem_alloc(bufsize, KM_SLEEP); 2566 odp = (struct dirent64 *)outbuf; 2567 } else { 2568 bufsize = bytes_wanted; 2569 outbuf = NULL; 2570 odp = (struct dirent64 *)iovp->iov_base; 2571 } 2572 eodp = (struct edirent *)odp; 2573 2574 if (ncookies != NULL) { 2575 /* 2576 * Minimum entry size is dirent size and 1 byte for a file name. 2577 */ 2578 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 2579 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 2580 *cookies = cooks; 2581 *ncookies = ncooks; 2582 } 2583 /* 2584 * If this VFS supports the system attribute view interface; and 2585 * we're looking at an extended attribute directory; and we care 2586 * about normalization conflicts on this vfs; then we must check 2587 * for normalization conflicts with the sysattr name space. 2588 */ 2589#ifdef TODO 2590 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 2591 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 2592 (flags & V_RDDIR_ENTFLAGS); 2593#else 2594 check_sysattrs = 0; 2595#endif 2596 2597 /* 2598 * Transform to file-system independent format 2599 */ 2600 outcount = 0; 2601 while (outcount < bytes_wanted) { 2602 ino64_t objnum; 2603 ushort_t reclen; 2604 off64_t *next = NULL; 2605 2606 /* 2607 * Special case `.', `..', and `.zfs'. 2608 */ 2609 if (offset == 0) { 2610 (void) strcpy(zap.za_name, "."); 2611 zap.za_normalization_conflict = 0; 2612 objnum = zp->z_id; 2613 type = DT_DIR; 2614 } else if (offset == 1) { 2615 (void) strcpy(zap.za_name, ".."); 2616 zap.za_normalization_conflict = 0; 2617 objnum = parent; 2618 type = DT_DIR; 2619 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2620 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2621 zap.za_normalization_conflict = 0; 2622 objnum = ZFSCTL_INO_ROOT; 2623 type = DT_DIR; 2624 } else { 2625 /* 2626 * Grab next entry. 2627 */ 2628 if (error = zap_cursor_retrieve(&zc, &zap)) { 2629 if ((*eofp = (error == ENOENT)) != 0) 2630 break; 2631 else 2632 goto update; 2633 } 2634 2635 if (zap.za_integer_length != 8 || 2636 zap.za_num_integers != 1) { 2637 cmn_err(CE_WARN, "zap_readdir: bad directory " 2638 "entry, obj = %lld, offset = %lld\n", 2639 (u_longlong_t)zp->z_id, 2640 (u_longlong_t)offset); 2641 error = SET_ERROR(ENXIO); 2642 goto update; 2643 } 2644 2645 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2646 /* 2647 * MacOS X can extract the object type here such as: 2648 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2649 */ 2650 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2651 2652 if (check_sysattrs && !zap.za_normalization_conflict) { 2653#ifdef TODO 2654 zap.za_normalization_conflict = 2655 xattr_sysattr_casechk(zap.za_name); 2656#else 2657 panic("%s:%u: TODO", __func__, __LINE__); 2658#endif 2659 } 2660 } 2661 2662 if (flags & V_RDDIR_ACCFILTER) { 2663 /* 2664 * If we have no access at all, don't include 2665 * this entry in the returned information 2666 */ 2667 znode_t *ezp; 2668 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2669 goto skip_entry; 2670 if (!zfs_has_access(ezp, cr)) { 2671 VN_RELE(ZTOV(ezp)); 2672 goto skip_entry; 2673 } 2674 VN_RELE(ZTOV(ezp)); 2675 } 2676 2677 if (flags & V_RDDIR_ENTFLAGS) 2678 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2679 else 2680 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2681 2682 /* 2683 * Will this entry fit in the buffer? 2684 */ 2685 if (outcount + reclen > bufsize) { 2686 /* 2687 * Did we manage to fit anything in the buffer? 2688 */ 2689 if (!outcount) { 2690 error = SET_ERROR(EINVAL); 2691 goto update; 2692 } 2693 break; 2694 } 2695 if (flags & V_RDDIR_ENTFLAGS) { 2696 /* 2697 * Add extended flag entry: 2698 */ 2699 eodp->ed_ino = objnum; 2700 eodp->ed_reclen = reclen; 2701 /* NOTE: ed_off is the offset for the *next* entry */ 2702 next = &(eodp->ed_off); 2703 eodp->ed_eflags = zap.za_normalization_conflict ? 2704 ED_CASE_CONFLICT : 0; 2705 (void) strncpy(eodp->ed_name, zap.za_name, 2706 EDIRENT_NAMELEN(reclen)); 2707 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2708 } else { 2709 /* 2710 * Add normal entry: 2711 */ 2712 odp->d_ino = objnum; 2713 odp->d_reclen = reclen; 2714 odp->d_namlen = strlen(zap.za_name); 2715 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 2716 odp->d_type = type; 2717 odp = (dirent64_t *)((intptr_t)odp + reclen); 2718 } 2719 outcount += reclen; 2720 2721 ASSERT(outcount <= bufsize); 2722 2723 /* Prefetch znode */ 2724 if (prefetch) 2725 dmu_prefetch(os, objnum, 0, 0, 0, 2726 ZIO_PRIORITY_SYNC_READ); 2727 2728 skip_entry: 2729 /* 2730 * Move to the next entry, fill in the previous offset. 2731 */ 2732 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2733 zap_cursor_advance(&zc); 2734 offset = zap_cursor_serialize(&zc); 2735 } else { 2736 offset += 1; 2737 } 2738 2739 if (cooks != NULL) { 2740 *cooks++ = offset; 2741 ncooks--; 2742 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 2743 } 2744 } 2745 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2746 2747 /* Subtract unused cookies */ 2748 if (ncookies != NULL) 2749 *ncookies -= ncooks; 2750 2751 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2752 iovp->iov_base += outcount; 2753 iovp->iov_len -= outcount; 2754 uio->uio_resid -= outcount; 2755 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2756 /* 2757 * Reset the pointer. 2758 */ 2759 offset = uio->uio_loffset; 2760 } 2761 2762update: 2763 zap_cursor_fini(&zc); 2764 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2765 kmem_free(outbuf, bufsize); 2766 2767 if (error == ENOENT) 2768 error = 0; 2769 2770 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2771 2772 uio->uio_loffset = offset; 2773 ZFS_EXIT(zfsvfs); 2774 if (error != 0 && cookies != NULL) { 2775 free(*cookies, M_TEMP); 2776 *cookies = NULL; 2777 *ncookies = 0; 2778 } 2779 return (error); 2780} 2781 2782ulong_t zfs_fsync_sync_cnt = 4; 2783 2784static int 2785zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2786{ 2787 znode_t *zp = VTOZ(vp); 2788 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2789 2790 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2791 2792 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { 2793 ZFS_ENTER(zfsvfs); 2794 ZFS_VERIFY_ZP(zp); 2795 zil_commit(zfsvfs->z_log, zp->z_id); 2796 ZFS_EXIT(zfsvfs); 2797 } 2798 return (0); 2799} 2800 2801 2802/* 2803 * Get the requested file attributes and place them in the provided 2804 * vattr structure. 2805 * 2806 * IN: vp - vnode of file. 2807 * vap - va_mask identifies requested attributes. 2808 * If AT_XVATTR set, then optional attrs are requested 2809 * flags - ATTR_NOACLCHECK (CIFS server context) 2810 * cr - credentials of caller. 2811 * ct - caller context 2812 * 2813 * OUT: vap - attribute values. 2814 * 2815 * RETURN: 0 (always succeeds). 2816 */ 2817/* ARGSUSED */ 2818static int 2819zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2820 caller_context_t *ct) 2821{ 2822 znode_t *zp = VTOZ(vp); 2823 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2824 int error = 0; 2825 uint32_t blksize; 2826 u_longlong_t nblocks; 2827 uint64_t links; 2828 uint64_t mtime[2], ctime[2], crtime[2], rdev; 2829 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2830 xoptattr_t *xoap = NULL; 2831 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2832 sa_bulk_attr_t bulk[4]; 2833 int count = 0; 2834 2835 ZFS_ENTER(zfsvfs); 2836 ZFS_VERIFY_ZP(zp); 2837 2838 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2839 2840 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 2841 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 2842 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16); 2843 if (vp->v_type == VBLK || vp->v_type == VCHR) 2844 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, 2845 &rdev, 8); 2846 2847 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 2848 ZFS_EXIT(zfsvfs); 2849 return (error); 2850 } 2851 2852 /* 2853 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2854 * Also, if we are the owner don't bother, since owner should 2855 * always be allowed to read basic attributes of file. 2856 */ 2857 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 2858 (vap->va_uid != crgetuid(cr))) { 2859 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2860 skipaclchk, cr)) { 2861 ZFS_EXIT(zfsvfs); 2862 return (error); 2863 } 2864 } 2865 2866 /* 2867 * Return all attributes. It's cheaper to provide the answer 2868 * than to determine whether we were asked the question. 2869 */ 2870 2871 mutex_enter(&zp->z_lock); 2872 vap->va_type = IFTOVT(zp->z_mode); 2873 vap->va_mode = zp->z_mode & ~S_IFMT; 2874#ifdef illumos 2875 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2876#else 2877 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 2878#endif 2879 vap->va_nodeid = zp->z_id; 2880 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2881 links = zp->z_links + 1; 2882 else 2883 links = zp->z_links; 2884 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */ 2885 vap->va_size = zp->z_size; 2886#ifdef illumos 2887 vap->va_rdev = vp->v_rdev; 2888#else 2889 if (vp->v_type == VBLK || vp->v_type == VCHR) 2890 vap->va_rdev = zfs_cmpldev(rdev); 2891#endif 2892 vap->va_seq = zp->z_seq; 2893 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2894 vap->va_filerev = zp->z_seq; 2895 2896 /* 2897 * Add in any requested optional attributes and the create time. 2898 * Also set the corresponding bits in the returned attribute bitmap. 2899 */ 2900 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2901 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2902 xoap->xoa_archive = 2903 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2904 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2905 } 2906 2907 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2908 xoap->xoa_readonly = 2909 ((zp->z_pflags & ZFS_READONLY) != 0); 2910 XVA_SET_RTN(xvap, XAT_READONLY); 2911 } 2912 2913 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2914 xoap->xoa_system = 2915 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2916 XVA_SET_RTN(xvap, XAT_SYSTEM); 2917 } 2918 2919 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2920 xoap->xoa_hidden = 2921 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2922 XVA_SET_RTN(xvap, XAT_HIDDEN); 2923 } 2924 2925 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2926 xoap->xoa_nounlink = 2927 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2928 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2929 } 2930 2931 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2932 xoap->xoa_immutable = 2933 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2934 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2935 } 2936 2937 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2938 xoap->xoa_appendonly = 2939 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2940 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2941 } 2942 2943 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2944 xoap->xoa_nodump = 2945 ((zp->z_pflags & ZFS_NODUMP) != 0); 2946 XVA_SET_RTN(xvap, XAT_NODUMP); 2947 } 2948 2949 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2950 xoap->xoa_opaque = 2951 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2952 XVA_SET_RTN(xvap, XAT_OPAQUE); 2953 } 2954 2955 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2956 xoap->xoa_av_quarantined = 2957 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2958 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2959 } 2960 2961 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2962 xoap->xoa_av_modified = 2963 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2964 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2965 } 2966 2967 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2968 vp->v_type == VREG) { 2969 zfs_sa_get_scanstamp(zp, xvap); 2970 } 2971 2972 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2973 uint64_t times[2]; 2974 2975 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 2976 times, sizeof (times)); 2977 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 2978 XVA_SET_RTN(xvap, XAT_CREATETIME); 2979 } 2980 2981 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2982 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2983 XVA_SET_RTN(xvap, XAT_REPARSE); 2984 } 2985 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2986 xoap->xoa_generation = zp->z_gen; 2987 XVA_SET_RTN(xvap, XAT_GEN); 2988 } 2989 2990 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2991 xoap->xoa_offline = 2992 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2993 XVA_SET_RTN(xvap, XAT_OFFLINE); 2994 } 2995 2996 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2997 xoap->xoa_sparse = 2998 ((zp->z_pflags & ZFS_SPARSE) != 0); 2999 XVA_SET_RTN(xvap, XAT_SPARSE); 3000 } 3001 } 3002 3003 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 3004 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 3005 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 3006 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 3007 3008 mutex_exit(&zp->z_lock); 3009 3010 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 3011 vap->va_blksize = blksize; 3012 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 3013 3014 if (zp->z_blksz == 0) { 3015 /* 3016 * Block size hasn't been set; suggest maximal I/O transfers. 3017 */ 3018 vap->va_blksize = zfsvfs->z_max_blksz; 3019 } 3020 3021 ZFS_EXIT(zfsvfs); 3022 return (0); 3023} 3024 3025/* 3026 * Set the file attributes to the values contained in the 3027 * vattr structure. 3028 * 3029 * IN: vp - vnode of file to be modified. 3030 * vap - new attribute values. 3031 * If AT_XVATTR set, then optional attrs are being set 3032 * flags - ATTR_UTIME set if non-default time values provided. 3033 * - ATTR_NOACLCHECK (CIFS context only). 3034 * cr - credentials of caller. 3035 * ct - caller context 3036 * 3037 * RETURN: 0 on success, error code on failure. 3038 * 3039 * Timestamps: 3040 * vp - ctime updated, mtime updated if size changed. 3041 */ 3042/* ARGSUSED */ 3043static int 3044zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 3045 caller_context_t *ct) 3046{ 3047 znode_t *zp = VTOZ(vp); 3048 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3049 zilog_t *zilog; 3050 dmu_tx_t *tx; 3051 vattr_t oldva; 3052 xvattr_t tmpxvattr; 3053 uint_t mask = vap->va_mask; 3054 uint_t saved_mask = 0; 3055 uint64_t saved_mode; 3056 int trim_mask = 0; 3057 uint64_t new_mode; 3058 uint64_t new_uid, new_gid; 3059 uint64_t xattr_obj; 3060 uint64_t mtime[2], ctime[2]; 3061 znode_t *attrzp; 3062 int need_policy = FALSE; 3063 int err, err2; 3064 zfs_fuid_info_t *fuidp = NULL; 3065 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 3066 xoptattr_t *xoap; 3067 zfs_acl_t *aclp; 3068 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 3069 boolean_t fuid_dirtied = B_FALSE; 3070 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 3071 int count = 0, xattr_count = 0; 3072 3073 if (mask == 0) 3074 return (0); 3075 3076 if (mask & AT_NOSET) 3077 return (SET_ERROR(EINVAL)); 3078 3079 ZFS_ENTER(zfsvfs); 3080 ZFS_VERIFY_ZP(zp); 3081 3082 zilog = zfsvfs->z_log; 3083 3084 /* 3085 * Make sure that if we have ephemeral uid/gid or xvattr specified 3086 * that file system is at proper version level 3087 */ 3088 3089 if (zfsvfs->z_use_fuids == B_FALSE && 3090 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 3091 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 3092 (mask & AT_XVATTR))) { 3093 ZFS_EXIT(zfsvfs); 3094 return (SET_ERROR(EINVAL)); 3095 } 3096 3097 if (mask & AT_SIZE && vp->v_type == VDIR) { 3098 ZFS_EXIT(zfsvfs); 3099 return (SET_ERROR(EISDIR)); 3100 } 3101 3102 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 3103 ZFS_EXIT(zfsvfs); 3104 return (SET_ERROR(EINVAL)); 3105 } 3106 3107 /* 3108 * If this is an xvattr_t, then get a pointer to the structure of 3109 * optional attributes. If this is NULL, then we have a vattr_t. 3110 */ 3111 xoap = xva_getxoptattr(xvap); 3112 3113 xva_init(&tmpxvattr); 3114 3115 /* 3116 * Immutable files can only alter immutable bit and atime 3117 */ 3118 if ((zp->z_pflags & ZFS_IMMUTABLE) && 3119 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 3120 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 3121 ZFS_EXIT(zfsvfs); 3122 return (SET_ERROR(EPERM)); 3123 } 3124 3125 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 3126 ZFS_EXIT(zfsvfs); 3127 return (SET_ERROR(EPERM)); 3128 } 3129 3130 /* 3131 * Verify timestamps doesn't overflow 32 bits. 3132 * ZFS can handle large timestamps, but 32bit syscalls can't 3133 * handle times greater than 2039. This check should be removed 3134 * once large timestamps are fully supported. 3135 */ 3136 if (mask & (AT_ATIME | AT_MTIME)) { 3137 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 3138 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 3139 ZFS_EXIT(zfsvfs); 3140 return (SET_ERROR(EOVERFLOW)); 3141 } 3142 } 3143 3144top: 3145 attrzp = NULL; 3146 aclp = NULL; 3147 3148 /* Can this be moved to before the top label? */ 3149 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 3150 ZFS_EXIT(zfsvfs); 3151 return (SET_ERROR(EROFS)); 3152 } 3153 3154 /* 3155 * First validate permissions 3156 */ 3157 3158 if (mask & AT_SIZE) { 3159 /* 3160 * XXX - Note, we are not providing any open 3161 * mode flags here (like FNDELAY), so we may 3162 * block if there are locks present... this 3163 * should be addressed in openat(). 3164 */ 3165 /* XXX - would it be OK to generate a log record here? */ 3166 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 3167 if (err) { 3168 ZFS_EXIT(zfsvfs); 3169 return (err); 3170 } 3171 } 3172 3173 if (mask & (AT_ATIME|AT_MTIME) || 3174 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 3175 XVA_ISSET_REQ(xvap, XAT_READONLY) || 3176 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 3177 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 3178 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 3179 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 3180 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 3181 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 3182 skipaclchk, cr); 3183 } 3184 3185 if (mask & (AT_UID|AT_GID)) { 3186 int idmask = (mask & (AT_UID|AT_GID)); 3187 int take_owner; 3188 int take_group; 3189 3190 /* 3191 * NOTE: even if a new mode is being set, 3192 * we may clear S_ISUID/S_ISGID bits. 3193 */ 3194 3195 if (!(mask & AT_MODE)) 3196 vap->va_mode = zp->z_mode; 3197 3198 /* 3199 * Take ownership or chgrp to group we are a member of 3200 */ 3201 3202 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 3203 take_group = (mask & AT_GID) && 3204 zfs_groupmember(zfsvfs, vap->va_gid, cr); 3205 3206 /* 3207 * If both AT_UID and AT_GID are set then take_owner and 3208 * take_group must both be set in order to allow taking 3209 * ownership. 3210 * 3211 * Otherwise, send the check through secpolicy_vnode_setattr() 3212 * 3213 */ 3214 3215 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 3216 ((idmask == AT_UID) && take_owner) || 3217 ((idmask == AT_GID) && take_group)) { 3218 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 3219 skipaclchk, cr) == 0) { 3220 /* 3221 * Remove setuid/setgid for non-privileged users 3222 */ 3223 secpolicy_setid_clear(vap, vp, cr); 3224 trim_mask = (mask & (AT_UID|AT_GID)); 3225 } else { 3226 need_policy = TRUE; 3227 } 3228 } else { 3229 need_policy = TRUE; 3230 } 3231 } 3232 3233 mutex_enter(&zp->z_lock); 3234 oldva.va_mode = zp->z_mode; 3235 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 3236 if (mask & AT_XVATTR) { 3237 /* 3238 * Update xvattr mask to include only those attributes 3239 * that are actually changing. 3240 * 3241 * the bits will be restored prior to actually setting 3242 * the attributes so the caller thinks they were set. 3243 */ 3244 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 3245 if (xoap->xoa_appendonly != 3246 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 3247 need_policy = TRUE; 3248 } else { 3249 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 3250 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 3251 } 3252 } 3253 3254 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 3255 if (xoap->xoa_nounlink != 3256 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 3257 need_policy = TRUE; 3258 } else { 3259 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 3260 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 3261 } 3262 } 3263 3264 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 3265 if (xoap->xoa_immutable != 3266 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 3267 need_policy = TRUE; 3268 } else { 3269 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 3270 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 3271 } 3272 } 3273 3274 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 3275 if (xoap->xoa_nodump != 3276 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 3277 need_policy = TRUE; 3278 } else { 3279 XVA_CLR_REQ(xvap, XAT_NODUMP); 3280 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 3281 } 3282 } 3283 3284 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 3285 if (xoap->xoa_av_modified != 3286 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 3287 need_policy = TRUE; 3288 } else { 3289 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 3290 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 3291 } 3292 } 3293 3294 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 3295 if ((vp->v_type != VREG && 3296 xoap->xoa_av_quarantined) || 3297 xoap->xoa_av_quarantined != 3298 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 3299 need_policy = TRUE; 3300 } else { 3301 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 3302 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 3303 } 3304 } 3305 3306 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 3307 mutex_exit(&zp->z_lock); 3308 ZFS_EXIT(zfsvfs); 3309 return (SET_ERROR(EPERM)); 3310 } 3311 3312 if (need_policy == FALSE && 3313 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 3314 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 3315 need_policy = TRUE; 3316 } 3317 } 3318 3319 mutex_exit(&zp->z_lock); 3320 3321 if (mask & AT_MODE) { 3322 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 3323 err = secpolicy_setid_setsticky_clear(vp, vap, 3324 &oldva, cr); 3325 if (err) { 3326 ZFS_EXIT(zfsvfs); 3327 return (err); 3328 } 3329 trim_mask |= AT_MODE; 3330 } else { 3331 need_policy = TRUE; 3332 } 3333 } 3334 3335 if (need_policy) { 3336 /* 3337 * If trim_mask is set then take ownership 3338 * has been granted or write_acl is present and user 3339 * has the ability to modify mode. In that case remove 3340 * UID|GID and or MODE from mask so that 3341 * secpolicy_vnode_setattr() doesn't revoke it. 3342 */ 3343 3344 if (trim_mask) { 3345 saved_mask = vap->va_mask; 3346 vap->va_mask &= ~trim_mask; 3347 if (trim_mask & AT_MODE) { 3348 /* 3349 * Save the mode, as secpolicy_vnode_setattr() 3350 * will overwrite it with ova.va_mode. 3351 */ 3352 saved_mode = vap->va_mode; 3353 } 3354 } 3355 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 3356 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 3357 if (err) { 3358 ZFS_EXIT(zfsvfs); 3359 return (err); 3360 } 3361 3362 if (trim_mask) { 3363 vap->va_mask |= saved_mask; 3364 if (trim_mask & AT_MODE) { 3365 /* 3366 * Recover the mode after 3367 * secpolicy_vnode_setattr(). 3368 */ 3369 vap->va_mode = saved_mode; 3370 } 3371 } 3372 } 3373 3374 /* 3375 * secpolicy_vnode_setattr, or take ownership may have 3376 * changed va_mask 3377 */ 3378 mask = vap->va_mask; 3379 3380 if ((mask & (AT_UID | AT_GID))) { 3381 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 3382 &xattr_obj, sizeof (xattr_obj)); 3383 3384 if (err == 0 && xattr_obj) { 3385 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 3386 if (err) 3387 goto out2; 3388 } 3389 if (mask & AT_UID) { 3390 new_uid = zfs_fuid_create(zfsvfs, 3391 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 3392 if (new_uid != zp->z_uid && 3393 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 3394 if (attrzp) 3395 VN_RELE(ZTOV(attrzp)); 3396 err = SET_ERROR(EDQUOT); 3397 goto out2; 3398 } 3399 } 3400 3401 if (mask & AT_GID) { 3402 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 3403 cr, ZFS_GROUP, &fuidp); 3404 if (new_gid != zp->z_gid && 3405 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 3406 if (attrzp) 3407 VN_RELE(ZTOV(attrzp)); 3408 err = SET_ERROR(EDQUOT); 3409 goto out2; 3410 } 3411 } 3412 } 3413 tx = dmu_tx_create(zfsvfs->z_os); 3414 3415 if (mask & AT_MODE) { 3416 uint64_t pmode = zp->z_mode; 3417 uint64_t acl_obj; 3418 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 3419 3420 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && 3421 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 3422 err = SET_ERROR(EPERM); 3423 goto out; 3424 } 3425 3426 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 3427 goto out; 3428 3429 mutex_enter(&zp->z_lock); 3430 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 3431 /* 3432 * Are we upgrading ACL from old V0 format 3433 * to V1 format? 3434 */ 3435 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 3436 zfs_znode_acl_version(zp) == 3437 ZFS_ACL_VERSION_INITIAL) { 3438 dmu_tx_hold_free(tx, acl_obj, 0, 3439 DMU_OBJECT_END); 3440 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3441 0, aclp->z_acl_bytes); 3442 } else { 3443 dmu_tx_hold_write(tx, acl_obj, 0, 3444 aclp->z_acl_bytes); 3445 } 3446 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3447 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3448 0, aclp->z_acl_bytes); 3449 } 3450 mutex_exit(&zp->z_lock); 3451 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3452 } else { 3453 if ((mask & AT_XVATTR) && 3454 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3455 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3456 else 3457 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3458 } 3459 3460 if (attrzp) { 3461 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 3462 } 3463 3464 fuid_dirtied = zfsvfs->z_fuid_dirty; 3465 if (fuid_dirtied) 3466 zfs_fuid_txhold(zfsvfs, tx); 3467 3468 zfs_sa_upgrade_txholds(tx, zp); 3469 3470 err = dmu_tx_assign(tx, TXG_WAIT); 3471 if (err) 3472 goto out; 3473 3474 count = 0; 3475 /* 3476 * Set each attribute requested. 3477 * We group settings according to the locks they need to acquire. 3478 * 3479 * Note: you cannot set ctime directly, although it will be 3480 * updated as a side-effect of calling this function. 3481 */ 3482 3483 3484 if (mask & (AT_UID|AT_GID|AT_MODE)) 3485 mutex_enter(&zp->z_acl_lock); 3486 mutex_enter(&zp->z_lock); 3487 3488 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3489 &zp->z_pflags, sizeof (zp->z_pflags)); 3490 3491 if (attrzp) { 3492 if (mask & (AT_UID|AT_GID|AT_MODE)) 3493 mutex_enter(&attrzp->z_acl_lock); 3494 mutex_enter(&attrzp->z_lock); 3495 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3496 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3497 sizeof (attrzp->z_pflags)); 3498 } 3499 3500 if (mask & (AT_UID|AT_GID)) { 3501 3502 if (mask & AT_UID) { 3503 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3504 &new_uid, sizeof (new_uid)); 3505 zp->z_uid = new_uid; 3506 if (attrzp) { 3507 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3508 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3509 sizeof (new_uid)); 3510 attrzp->z_uid = new_uid; 3511 } 3512 } 3513 3514 if (mask & AT_GID) { 3515 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3516 NULL, &new_gid, sizeof (new_gid)); 3517 zp->z_gid = new_gid; 3518 if (attrzp) { 3519 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3520 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 3521 sizeof (new_gid)); 3522 attrzp->z_gid = new_gid; 3523 } 3524 } 3525 if (!(mask & AT_MODE)) { 3526 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 3527 NULL, &new_mode, sizeof (new_mode)); 3528 new_mode = zp->z_mode; 3529 } 3530 err = zfs_acl_chown_setattr(zp); 3531 ASSERT(err == 0); 3532 if (attrzp) { 3533 err = zfs_acl_chown_setattr(attrzp); 3534 ASSERT(err == 0); 3535 } 3536 } 3537 3538 if (mask & AT_MODE) { 3539 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 3540 &new_mode, sizeof (new_mode)); 3541 zp->z_mode = new_mode; 3542 ASSERT3U((uintptr_t)aclp, !=, 0); 3543 err = zfs_aclset_common(zp, aclp, cr, tx); 3544 ASSERT0(err); 3545 if (zp->z_acl_cached) 3546 zfs_acl_free(zp->z_acl_cached); 3547 zp->z_acl_cached = aclp; 3548 aclp = NULL; 3549 } 3550 3551 3552 if (mask & AT_ATIME) { 3553 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 3554 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 3555 &zp->z_atime, sizeof (zp->z_atime)); 3556 } 3557 3558 if (mask & AT_MTIME) { 3559 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 3560 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 3561 mtime, sizeof (mtime)); 3562 } 3563 3564 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 3565 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 3566 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 3567 NULL, mtime, sizeof (mtime)); 3568 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3569 &ctime, sizeof (ctime)); 3570 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 3571 B_TRUE); 3572 } else if (mask != 0) { 3573 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3574 &ctime, sizeof (ctime)); 3575 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime, 3576 B_TRUE); 3577 if (attrzp) { 3578 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3579 SA_ZPL_CTIME(zfsvfs), NULL, 3580 &ctime, sizeof (ctime)); 3581 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 3582 mtime, ctime, B_TRUE); 3583 } 3584 } 3585 /* 3586 * Do this after setting timestamps to prevent timestamp 3587 * update from toggling bit 3588 */ 3589 3590 if (xoap && (mask & AT_XVATTR)) { 3591 3592 /* 3593 * restore trimmed off masks 3594 * so that return masks can be set for caller. 3595 */ 3596 3597 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 3598 XVA_SET_REQ(xvap, XAT_APPENDONLY); 3599 } 3600 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 3601 XVA_SET_REQ(xvap, XAT_NOUNLINK); 3602 } 3603 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 3604 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 3605 } 3606 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 3607 XVA_SET_REQ(xvap, XAT_NODUMP); 3608 } 3609 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 3610 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 3611 } 3612 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 3613 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 3614 } 3615 3616 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3617 ASSERT(vp->v_type == VREG); 3618 3619 zfs_xvattr_set(zp, xvap, tx); 3620 } 3621 3622 if (fuid_dirtied) 3623 zfs_fuid_sync(zfsvfs, tx); 3624 3625 if (mask != 0) 3626 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 3627 3628 mutex_exit(&zp->z_lock); 3629 if (mask & (AT_UID|AT_GID|AT_MODE)) 3630 mutex_exit(&zp->z_acl_lock); 3631 3632 if (attrzp) { 3633 if (mask & (AT_UID|AT_GID|AT_MODE)) 3634 mutex_exit(&attrzp->z_acl_lock); 3635 mutex_exit(&attrzp->z_lock); 3636 } 3637out: 3638 if (err == 0 && attrzp) { 3639 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 3640 xattr_count, tx); 3641 ASSERT(err2 == 0); 3642 } 3643 3644 if (attrzp) 3645 VN_RELE(ZTOV(attrzp)); 3646 3647 if (aclp) 3648 zfs_acl_free(aclp); 3649 3650 if (fuidp) { 3651 zfs_fuid_info_free(fuidp); 3652 fuidp = NULL; 3653 } 3654 3655 if (err) { 3656 dmu_tx_abort(tx); 3657 if (err == ERESTART) 3658 goto top; 3659 } else { 3660 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3661 dmu_tx_commit(tx); 3662 } 3663 3664out2: 3665 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3666 zil_commit(zilog, 0); 3667 3668 ZFS_EXIT(zfsvfs); 3669 return (err); 3670} 3671 3672typedef struct zfs_zlock { 3673 krwlock_t *zl_rwlock; /* lock we acquired */ 3674 znode_t *zl_znode; /* znode we held */ 3675 struct zfs_zlock *zl_next; /* next in list */ 3676} zfs_zlock_t; 3677 3678/* 3679 * Drop locks and release vnodes that were held by zfs_rename_lock(). 3680 */ 3681static void 3682zfs_rename_unlock(zfs_zlock_t **zlpp) 3683{ 3684 zfs_zlock_t *zl; 3685 3686 while ((zl = *zlpp) != NULL) { 3687 if (zl->zl_znode != NULL) 3688 VN_RELE(ZTOV(zl->zl_znode)); 3689 rw_exit(zl->zl_rwlock); 3690 *zlpp = zl->zl_next; 3691 kmem_free(zl, sizeof (*zl)); 3692 } 3693} 3694 3695/* 3696 * Search back through the directory tree, using the ".." entries. 3697 * Lock each directory in the chain to prevent concurrent renames. 3698 * Fail any attempt to move a directory into one of its own descendants. 3699 * XXX - z_parent_lock can overlap with map or grow locks 3700 */ 3701static int 3702zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 3703{ 3704 zfs_zlock_t *zl; 3705 znode_t *zp = tdzp; 3706 uint64_t rootid = zp->z_zfsvfs->z_root; 3707 uint64_t oidp = zp->z_id; 3708 krwlock_t *rwlp = &szp->z_parent_lock; 3709 krw_t rw = RW_WRITER; 3710 3711 /* 3712 * First pass write-locks szp and compares to zp->z_id. 3713 * Later passes read-lock zp and compare to zp->z_parent. 3714 */ 3715 do { 3716 if (!rw_tryenter(rwlp, rw)) { 3717 /* 3718 * Another thread is renaming in this path. 3719 * Note that if we are a WRITER, we don't have any 3720 * parent_locks held yet. 3721 */ 3722 if (rw == RW_READER && zp->z_id > szp->z_id) { 3723 /* 3724 * Drop our locks and restart 3725 */ 3726 zfs_rename_unlock(&zl); 3727 *zlpp = NULL; 3728 zp = tdzp; 3729 oidp = zp->z_id; 3730 rwlp = &szp->z_parent_lock; 3731 rw = RW_WRITER; 3732 continue; 3733 } else { 3734 /* 3735 * Wait for other thread to drop its locks 3736 */ 3737 rw_enter(rwlp, rw); 3738 } 3739 } 3740 3741 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 3742 zl->zl_rwlock = rwlp; 3743 zl->zl_znode = NULL; 3744 zl->zl_next = *zlpp; 3745 *zlpp = zl; 3746 3747 if (oidp == szp->z_id) /* We're a descendant of szp */ 3748 return (SET_ERROR(EINVAL)); 3749 3750 if (oidp == rootid) /* We've hit the top */ 3751 return (0); 3752 3753 if (rw == RW_READER) { /* i.e. not the first pass */ 3754 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp); 3755 if (error) 3756 return (error); 3757 zl->zl_znode = zp; 3758 } 3759 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs), 3760 &oidp, sizeof (oidp)); 3761 rwlp = &zp->z_parent_lock; 3762 rw = RW_READER; 3763 3764 } while (zp->z_id != sdzp->z_id); 3765 3766 return (0); 3767} 3768 3769/* 3770 * Move an entry from the provided source directory to the target 3771 * directory. Change the entry name as indicated. 3772 * 3773 * IN: sdvp - Source directory containing the "old entry". 3774 * snm - Old entry name. 3775 * tdvp - Target directory to contain the "new entry". 3776 * tnm - New entry name. 3777 * cr - credentials of caller. 3778 * ct - caller context 3779 * flags - case flags 3780 * 3781 * RETURN: 0 on success, error code on failure. 3782 * 3783 * Timestamps: 3784 * sdvp,tdvp - ctime|mtime updated 3785 */ 3786/*ARGSUSED*/ 3787static int 3788zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 3789 caller_context_t *ct, int flags) 3790{ 3791 znode_t *tdzp, *sdzp, *szp, *tzp; 3792 zfsvfs_t *zfsvfs; 3793 zilog_t *zilog; 3794 vnode_t *realvp; 3795 zfs_dirlock_t *sdl, *tdl; 3796 dmu_tx_t *tx; 3797 zfs_zlock_t *zl; 3798 int cmp, serr, terr; 3799 int error = 0; 3800 int zflg = 0; 3801 boolean_t waited = B_FALSE; 3802 3803 tdzp = VTOZ(tdvp); 3804 ZFS_VERIFY_ZP(tdzp); 3805 zfsvfs = tdzp->z_zfsvfs; 3806 ZFS_ENTER(zfsvfs); 3807 zilog = zfsvfs->z_log; 3808 sdzp = VTOZ(sdvp); 3809 3810 /* 3811 * In case sdzp is not valid, let's be sure to exit from the right 3812 * zfsvfs_t. 3813 */ 3814 if (sdzp->z_sa_hdl == NULL) { 3815 ZFS_EXIT(zfsvfs); 3816 return (SET_ERROR(EIO)); 3817 } 3818 3819 /* 3820 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 3821 * ctldir appear to have the same v_vfsp. 3822 */ 3823 if (sdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) { 3824 ZFS_EXIT(zfsvfs); 3825 return (SET_ERROR(EXDEV)); 3826 } 3827 3828 if (zfsvfs->z_utf8 && u8_validate(tnm, 3829 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3830 ZFS_EXIT(zfsvfs); 3831 return (SET_ERROR(EILSEQ)); 3832 } 3833 3834 if (flags & FIGNORECASE) 3835 zflg |= ZCILOOK; 3836 3837top: 3838 szp = NULL; 3839 tzp = NULL; 3840 zl = NULL; 3841 3842 /* 3843 * This is to prevent the creation of links into attribute space 3844 * by renaming a linked file into/outof an attribute directory. 3845 * See the comment in zfs_link() for why this is considered bad. 3846 */ 3847 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3848 ZFS_EXIT(zfsvfs); 3849 return (SET_ERROR(EINVAL)); 3850 } 3851 3852 /* 3853 * Lock source and target directory entries. To prevent deadlock, 3854 * a lock ordering must be defined. We lock the directory with 3855 * the smallest object id first, or if it's a tie, the one with 3856 * the lexically first name. 3857 */ 3858 if (sdzp->z_id < tdzp->z_id) { 3859 cmp = -1; 3860 } else if (sdzp->z_id > tdzp->z_id) { 3861 cmp = 1; 3862 } else { 3863 /* 3864 * First compare the two name arguments without 3865 * considering any case folding. 3866 */ 3867 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 3868 3869 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 3870 ASSERT(error == 0 || !zfsvfs->z_utf8); 3871 if (cmp == 0) { 3872 /* 3873 * POSIX: "If the old argument and the new argument 3874 * both refer to links to the same existing file, 3875 * the rename() function shall return successfully 3876 * and perform no other action." 3877 */ 3878 ZFS_EXIT(zfsvfs); 3879 return (0); 3880 } 3881 /* 3882 * If the file system is case-folding, then we may 3883 * have some more checking to do. A case-folding file 3884 * system is either supporting mixed case sensitivity 3885 * access or is completely case-insensitive. Note 3886 * that the file system is always case preserving. 3887 * 3888 * In mixed sensitivity mode case sensitive behavior 3889 * is the default. FIGNORECASE must be used to 3890 * explicitly request case insensitive behavior. 3891 * 3892 * If the source and target names provided differ only 3893 * by case (e.g., a request to rename 'tim' to 'Tim'), 3894 * we will treat this as a special case in the 3895 * case-insensitive mode: as long as the source name 3896 * is an exact match, we will allow this to proceed as 3897 * a name-change request. 3898 */ 3899 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3900 (zfsvfs->z_case == ZFS_CASE_MIXED && 3901 flags & FIGNORECASE)) && 3902 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3903 &error) == 0) { 3904 /* 3905 * case preserving rename request, require exact 3906 * name matches 3907 */ 3908 zflg |= ZCIEXACT; 3909 zflg &= ~ZCILOOK; 3910 } 3911 } 3912 3913 /* 3914 * If the source and destination directories are the same, we should 3915 * grab the z_name_lock of that directory only once. 3916 */ 3917 if (sdzp == tdzp) { 3918 zflg |= ZHAVELOCK; 3919 rw_enter(&sdzp->z_name_lock, RW_READER); 3920 } 3921 3922 if (cmp < 0) { 3923 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3924 ZEXISTS | zflg, NULL, NULL); 3925 terr = zfs_dirent_lock(&tdl, 3926 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3927 } else { 3928 terr = zfs_dirent_lock(&tdl, 3929 tdzp, tnm, &tzp, zflg, NULL, NULL); 3930 serr = zfs_dirent_lock(&sdl, 3931 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3932 NULL, NULL); 3933 } 3934 3935 if (serr) { 3936 /* 3937 * Source entry invalid or not there. 3938 */ 3939 if (!terr) { 3940 zfs_dirent_unlock(tdl); 3941 if (tzp) 3942 VN_RELE(ZTOV(tzp)); 3943 } 3944 3945 if (sdzp == tdzp) 3946 rw_exit(&sdzp->z_name_lock); 3947 3948 /* 3949 * FreeBSD: In OpenSolaris they only check if rename source is 3950 * ".." here, because "." is handled in their lookup. This is 3951 * not the case for FreeBSD, so we check for "." explicitly. 3952 */ 3953 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 3954 serr = SET_ERROR(EINVAL); 3955 ZFS_EXIT(zfsvfs); 3956 return (serr); 3957 } 3958 if (terr) { 3959 zfs_dirent_unlock(sdl); 3960 VN_RELE(ZTOV(szp)); 3961 3962 if (sdzp == tdzp) 3963 rw_exit(&sdzp->z_name_lock); 3964 3965 if (strcmp(tnm, "..") == 0) 3966 terr = SET_ERROR(EINVAL); 3967 ZFS_EXIT(zfsvfs); 3968 return (terr); 3969 } 3970 3971 /* 3972 * Must have write access at the source to remove the old entry 3973 * and write access at the target to create the new entry. 3974 * Note that if target and source are the same, this can be 3975 * done in a single check. 3976 */ 3977 3978 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3979 goto out; 3980 3981 if (ZTOV(szp)->v_type == VDIR) { 3982 /* 3983 * Check to make sure rename is valid. 3984 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3985 */ 3986 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3987 goto out; 3988 } 3989 3990 /* 3991 * Does target exist? 3992 */ 3993 if (tzp) { 3994 /* 3995 * Source and target must be the same type. 3996 */ 3997 if (ZTOV(szp)->v_type == VDIR) { 3998 if (ZTOV(tzp)->v_type != VDIR) { 3999 error = SET_ERROR(ENOTDIR); 4000 goto out; 4001 } 4002 } else { 4003 if (ZTOV(tzp)->v_type == VDIR) { 4004 error = SET_ERROR(EISDIR); 4005 goto out; 4006 } 4007 } 4008 /* 4009 * POSIX dictates that when the source and target 4010 * entries refer to the same file object, rename 4011 * must do nothing and exit without error. 4012 */ 4013 if (szp->z_id == tzp->z_id) { 4014 error = 0; 4015 goto out; 4016 } 4017 } 4018 4019 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 4020 if (tzp) 4021 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 4022 4023 /* 4024 * notify the target directory if it is not the same 4025 * as source directory. 4026 */ 4027 if (tdvp != sdvp) { 4028 vnevent_rename_dest_dir(tdvp, ct); 4029 } 4030 4031 tx = dmu_tx_create(zfsvfs->z_os); 4032 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4033 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 4034 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 4035 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 4036 if (sdzp != tdzp) { 4037 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 4038 zfs_sa_upgrade_txholds(tx, tdzp); 4039 } 4040 if (tzp) { 4041 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 4042 zfs_sa_upgrade_txholds(tx, tzp); 4043 } 4044 4045 zfs_sa_upgrade_txholds(tx, szp); 4046 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 4047 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4048 if (error) { 4049 if (zl != NULL) 4050 zfs_rename_unlock(&zl); 4051 zfs_dirent_unlock(sdl); 4052 zfs_dirent_unlock(tdl); 4053 4054 if (sdzp == tdzp) 4055 rw_exit(&sdzp->z_name_lock); 4056 4057 VN_RELE(ZTOV(szp)); 4058 if (tzp) 4059 VN_RELE(ZTOV(tzp)); 4060 if (error == ERESTART) { 4061 waited = B_TRUE; 4062 dmu_tx_wait(tx); 4063 dmu_tx_abort(tx); 4064 goto top; 4065 } 4066 dmu_tx_abort(tx); 4067 ZFS_EXIT(zfsvfs); 4068 return (error); 4069 } 4070 4071 if (tzp) /* Attempt to remove the existing target */ 4072 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 4073 4074 if (error == 0) { 4075 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 4076 if (error == 0) { 4077 szp->z_pflags |= ZFS_AV_MODIFIED; 4078 4079 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 4080 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 4081 ASSERT0(error); 4082 4083 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 4084 if (error == 0) { 4085 zfs_log_rename(zilog, tx, TX_RENAME | 4086 (flags & FIGNORECASE ? TX_CI : 0), sdzp, 4087 sdl->dl_name, tdzp, tdl->dl_name, szp); 4088 4089 /* 4090 * Update path information for the target vnode 4091 */ 4092 vn_renamepath(tdvp, ZTOV(szp), tnm, 4093 strlen(tnm)); 4094 } else { 4095 /* 4096 * At this point, we have successfully created 4097 * the target name, but have failed to remove 4098 * the source name. Since the create was done 4099 * with the ZRENAMING flag, there are 4100 * complications; for one, the link count is 4101 * wrong. The easiest way to deal with this 4102 * is to remove the newly created target, and 4103 * return the original error. This must 4104 * succeed; fortunately, it is very unlikely to 4105 * fail, since we just created it. 4106 */ 4107 VERIFY3U(zfs_link_destroy(tdl, szp, tx, 4108 ZRENAMING, NULL), ==, 0); 4109 } 4110 } 4111#ifdef FREEBSD_NAMECACHE 4112 if (error == 0) { 4113 cache_purge(sdvp); 4114 cache_purge(tdvp); 4115 cache_purge(ZTOV(szp)); 4116 if (tzp) 4117 cache_purge(ZTOV(tzp)); 4118 } 4119#endif 4120 } 4121 4122 dmu_tx_commit(tx); 4123out: 4124 if (zl != NULL) 4125 zfs_rename_unlock(&zl); 4126 4127 zfs_dirent_unlock(sdl); 4128 zfs_dirent_unlock(tdl); 4129 4130 if (sdzp == tdzp) 4131 rw_exit(&sdzp->z_name_lock); 4132 4133 4134 VN_RELE(ZTOV(szp)); 4135 if (tzp) 4136 VN_RELE(ZTOV(tzp)); 4137 4138 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4139 zil_commit(zilog, 0); 4140 4141 ZFS_EXIT(zfsvfs); 4142 4143 return (error); 4144} 4145 4146/* 4147 * Insert the indicated symbolic reference entry into the directory. 4148 * 4149 * IN: dvp - Directory to contain new symbolic link. 4150 * link - Name for new symlink entry. 4151 * vap - Attributes of new entry. 4152 * cr - credentials of caller. 4153 * ct - caller context 4154 * flags - case flags 4155 * 4156 * RETURN: 0 on success, error code on failure. 4157 * 4158 * Timestamps: 4159 * dvp - ctime|mtime updated 4160 */ 4161/*ARGSUSED*/ 4162static int 4163zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, 4164 cred_t *cr, kthread_t *td) 4165{ 4166 znode_t *zp, *dzp = VTOZ(dvp); 4167 zfs_dirlock_t *dl; 4168 dmu_tx_t *tx; 4169 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4170 zilog_t *zilog; 4171 uint64_t len = strlen(link); 4172 int error; 4173 int zflg = ZNEW; 4174 zfs_acl_ids_t acl_ids; 4175 boolean_t fuid_dirtied; 4176 uint64_t txtype = TX_SYMLINK; 4177 boolean_t waited = B_FALSE; 4178 int flags = 0; 4179 4180 ASSERT(vap->va_type == VLNK); 4181 4182 ZFS_ENTER(zfsvfs); 4183 ZFS_VERIFY_ZP(dzp); 4184 zilog = zfsvfs->z_log; 4185 4186 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 4187 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4188 ZFS_EXIT(zfsvfs); 4189 return (SET_ERROR(EILSEQ)); 4190 } 4191 if (flags & FIGNORECASE) 4192 zflg |= ZCILOOK; 4193 4194 if (len > MAXPATHLEN) { 4195 ZFS_EXIT(zfsvfs); 4196 return (SET_ERROR(ENAMETOOLONG)); 4197 } 4198 4199 if ((error = zfs_acl_ids_create(dzp, 0, 4200 vap, cr, NULL, &acl_ids)) != 0) { 4201 ZFS_EXIT(zfsvfs); 4202 return (error); 4203 } 4204 4205 getnewvnode_reserve(1); 4206 4207top: 4208 /* 4209 * Attempt to lock directory; fail if entry already exists. 4210 */ 4211 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 4212 if (error) { 4213 zfs_acl_ids_free(&acl_ids); 4214 getnewvnode_drop_reserve(); 4215 ZFS_EXIT(zfsvfs); 4216 return (error); 4217 } 4218 4219 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4220 zfs_acl_ids_free(&acl_ids); 4221 zfs_dirent_unlock(dl); 4222 getnewvnode_drop_reserve(); 4223 ZFS_EXIT(zfsvfs); 4224 return (error); 4225 } 4226 4227 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 4228 zfs_acl_ids_free(&acl_ids); 4229 zfs_dirent_unlock(dl); 4230 getnewvnode_drop_reserve(); 4231 ZFS_EXIT(zfsvfs); 4232 return (SET_ERROR(EDQUOT)); 4233 } 4234 tx = dmu_tx_create(zfsvfs->z_os); 4235 fuid_dirtied = zfsvfs->z_fuid_dirty; 4236 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 4237 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4238 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 4239 ZFS_SA_BASE_ATTR_SIZE + len); 4240 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 4241 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 4242 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 4243 acl_ids.z_aclp->z_acl_bytes); 4244 } 4245 if (fuid_dirtied) 4246 zfs_fuid_txhold(zfsvfs, tx); 4247 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4248 if (error) { 4249 zfs_dirent_unlock(dl); 4250 if (error == ERESTART) { 4251 waited = B_TRUE; 4252 dmu_tx_wait(tx); 4253 dmu_tx_abort(tx); 4254 goto top; 4255 } 4256 zfs_acl_ids_free(&acl_ids); 4257 dmu_tx_abort(tx); 4258 getnewvnode_drop_reserve(); 4259 ZFS_EXIT(zfsvfs); 4260 return (error); 4261 } 4262 4263 /* 4264 * Create a new object for the symlink. 4265 * for version 4 ZPL datsets the symlink will be an SA attribute 4266 */ 4267 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 4268 4269 if (fuid_dirtied) 4270 zfs_fuid_sync(zfsvfs, tx); 4271 4272 mutex_enter(&zp->z_lock); 4273 if (zp->z_is_sa) 4274 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 4275 link, len, tx); 4276 else 4277 zfs_sa_symlink(zp, link, len, tx); 4278 mutex_exit(&zp->z_lock); 4279 4280 zp->z_size = len; 4281 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 4282 &zp->z_size, sizeof (zp->z_size), tx); 4283 /* 4284 * Insert the new object into the directory. 4285 */ 4286 (void) zfs_link_create(dl, zp, tx, ZNEW); 4287 4288 if (flags & FIGNORECASE) 4289 txtype |= TX_CI; 4290 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 4291 *vpp = ZTOV(zp); 4292 4293 zfs_acl_ids_free(&acl_ids); 4294 4295 dmu_tx_commit(tx); 4296 4297 getnewvnode_drop_reserve(); 4298 4299 zfs_dirent_unlock(dl); 4300 4301 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4302 zil_commit(zilog, 0); 4303 4304 ZFS_EXIT(zfsvfs); 4305 return (error); 4306} 4307 4308/* 4309 * Return, in the buffer contained in the provided uio structure, 4310 * the symbolic path referred to by vp. 4311 * 4312 * IN: vp - vnode of symbolic link. 4313 * uio - structure to contain the link path. 4314 * cr - credentials of caller. 4315 * ct - caller context 4316 * 4317 * OUT: uio - structure containing the link path. 4318 * 4319 * RETURN: 0 on success, error code on failure. 4320 * 4321 * Timestamps: 4322 * vp - atime updated 4323 */ 4324/* ARGSUSED */ 4325static int 4326zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 4327{ 4328 znode_t *zp = VTOZ(vp); 4329 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4330 int error; 4331 4332 ZFS_ENTER(zfsvfs); 4333 ZFS_VERIFY_ZP(zp); 4334 4335 mutex_enter(&zp->z_lock); 4336 if (zp->z_is_sa) 4337 error = sa_lookup_uio(zp->z_sa_hdl, 4338 SA_ZPL_SYMLINK(zfsvfs), uio); 4339 else 4340 error = zfs_sa_readlink(zp, uio); 4341 mutex_exit(&zp->z_lock); 4342 4343 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4344 4345 ZFS_EXIT(zfsvfs); 4346 return (error); 4347} 4348 4349/* 4350 * Insert a new entry into directory tdvp referencing svp. 4351 * 4352 * IN: tdvp - Directory to contain new entry. 4353 * svp - vnode of new entry. 4354 * name - name of new entry. 4355 * cr - credentials of caller. 4356 * ct - caller context 4357 * 4358 * RETURN: 0 on success, error code on failure. 4359 * 4360 * Timestamps: 4361 * tdvp - ctime|mtime updated 4362 * svp - ctime updated 4363 */ 4364/* ARGSUSED */ 4365static int 4366zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 4367 caller_context_t *ct, int flags) 4368{ 4369 znode_t *dzp = VTOZ(tdvp); 4370 znode_t *tzp, *szp; 4371 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4372 zilog_t *zilog; 4373 zfs_dirlock_t *dl; 4374 dmu_tx_t *tx; 4375 vnode_t *realvp; 4376 int error; 4377 int zf = ZNEW; 4378 uint64_t parent; 4379 uid_t owner; 4380 boolean_t waited = B_FALSE; 4381 4382 ASSERT(tdvp->v_type == VDIR); 4383 4384 ZFS_ENTER(zfsvfs); 4385 ZFS_VERIFY_ZP(dzp); 4386 zilog = zfsvfs->z_log; 4387 4388 if (VOP_REALVP(svp, &realvp, ct) == 0) 4389 svp = realvp; 4390 4391 /* 4392 * POSIX dictates that we return EPERM here. 4393 * Better choices include ENOTSUP or EISDIR. 4394 */ 4395 if (svp->v_type == VDIR) { 4396 ZFS_EXIT(zfsvfs); 4397 return (SET_ERROR(EPERM)); 4398 } 4399 4400 szp = VTOZ(svp); 4401 ZFS_VERIFY_ZP(szp); 4402 4403 if (szp->z_pflags & (ZFS_APPENDONLY | ZFS_IMMUTABLE | ZFS_READONLY)) { 4404 ZFS_EXIT(zfsvfs); 4405 return (SET_ERROR(EPERM)); 4406 } 4407 4408 /* 4409 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 4410 * ctldir appear to have the same v_vfsp. 4411 */ 4412 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) { 4413 ZFS_EXIT(zfsvfs); 4414 return (SET_ERROR(EXDEV)); 4415 } 4416 4417 /* Prevent links to .zfs/shares files */ 4418 4419 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 4420 &parent, sizeof (uint64_t))) != 0) { 4421 ZFS_EXIT(zfsvfs); 4422 return (error); 4423 } 4424 if (parent == zfsvfs->z_shares_dir) { 4425 ZFS_EXIT(zfsvfs); 4426 return (SET_ERROR(EPERM)); 4427 } 4428 4429 if (zfsvfs->z_utf8 && u8_validate(name, 4430 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4431 ZFS_EXIT(zfsvfs); 4432 return (SET_ERROR(EILSEQ)); 4433 } 4434 if (flags & FIGNORECASE) 4435 zf |= ZCILOOK; 4436 4437 /* 4438 * We do not support links between attributes and non-attributes 4439 * because of the potential security risk of creating links 4440 * into "normal" file space in order to circumvent restrictions 4441 * imposed in attribute space. 4442 */ 4443 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 4444 ZFS_EXIT(zfsvfs); 4445 return (SET_ERROR(EINVAL)); 4446 } 4447 4448 4449 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 4450 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) { 4451 ZFS_EXIT(zfsvfs); 4452 return (SET_ERROR(EPERM)); 4453 } 4454 4455 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4456 ZFS_EXIT(zfsvfs); 4457 return (error); 4458 } 4459 4460top: 4461 /* 4462 * Attempt to lock directory; fail if entry already exists. 4463 */ 4464 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 4465 if (error) { 4466 ZFS_EXIT(zfsvfs); 4467 return (error); 4468 } 4469 4470 tx = dmu_tx_create(zfsvfs->z_os); 4471 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4472 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4473 zfs_sa_upgrade_txholds(tx, szp); 4474 zfs_sa_upgrade_txholds(tx, dzp); 4475 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4476 if (error) { 4477 zfs_dirent_unlock(dl); 4478 if (error == ERESTART) { 4479 waited = B_TRUE; 4480 dmu_tx_wait(tx); 4481 dmu_tx_abort(tx); 4482 goto top; 4483 } 4484 dmu_tx_abort(tx); 4485 ZFS_EXIT(zfsvfs); 4486 return (error); 4487 } 4488 4489 error = zfs_link_create(dl, szp, tx, 0); 4490 4491 if (error == 0) { 4492 uint64_t txtype = TX_LINK; 4493 if (flags & FIGNORECASE) 4494 txtype |= TX_CI; 4495 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 4496 } 4497 4498 dmu_tx_commit(tx); 4499 4500 zfs_dirent_unlock(dl); 4501 4502 if (error == 0) { 4503 vnevent_link(svp, ct); 4504 } 4505 4506 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4507 zil_commit(zilog, 0); 4508 4509 ZFS_EXIT(zfsvfs); 4510 return (error); 4511} 4512 4513#ifdef illumos 4514/* 4515 * zfs_null_putapage() is used when the file system has been force 4516 * unmounted. It just drops the pages. 4517 */ 4518/* ARGSUSED */ 4519static int 4520zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4521 size_t *lenp, int flags, cred_t *cr) 4522{ 4523 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 4524 return (0); 4525} 4526 4527/* 4528 * Push a page out to disk, klustering if possible. 4529 * 4530 * IN: vp - file to push page to. 4531 * pp - page to push. 4532 * flags - additional flags. 4533 * cr - credentials of caller. 4534 * 4535 * OUT: offp - start of range pushed. 4536 * lenp - len of range pushed. 4537 * 4538 * RETURN: 0 on success, error code on failure. 4539 * 4540 * NOTE: callers must have locked the page to be pushed. On 4541 * exit, the page (and all other pages in the kluster) must be 4542 * unlocked. 4543 */ 4544/* ARGSUSED */ 4545static int 4546zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4547 size_t *lenp, int flags, cred_t *cr) 4548{ 4549 znode_t *zp = VTOZ(vp); 4550 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4551 dmu_tx_t *tx; 4552 u_offset_t off, koff; 4553 size_t len, klen; 4554 int err; 4555 4556 off = pp->p_offset; 4557 len = PAGESIZE; 4558 /* 4559 * If our blocksize is bigger than the page size, try to kluster 4560 * multiple pages so that we write a full block (thus avoiding 4561 * a read-modify-write). 4562 */ 4563 if (off < zp->z_size && zp->z_blksz > PAGESIZE) { 4564 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 4565 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 4566 ASSERT(koff <= zp->z_size); 4567 if (koff + klen > zp->z_size) 4568 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); 4569 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 4570 } 4571 ASSERT3U(btop(len), ==, btopr(len)); 4572 4573 /* 4574 * Can't push pages past end-of-file. 4575 */ 4576 if (off >= zp->z_size) { 4577 /* ignore all pages */ 4578 err = 0; 4579 goto out; 4580 } else if (off + len > zp->z_size) { 4581 int npages = btopr(zp->z_size - off); 4582 page_t *trunc; 4583 4584 page_list_break(&pp, &trunc, npages); 4585 /* ignore pages past end of file */ 4586 if (trunc) 4587 pvn_write_done(trunc, flags); 4588 len = zp->z_size - off; 4589 } 4590 4591 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4592 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4593 err = SET_ERROR(EDQUOT); 4594 goto out; 4595 } 4596 tx = dmu_tx_create(zfsvfs->z_os); 4597 dmu_tx_hold_write(tx, zp->z_id, off, len); 4598 4599 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4600 zfs_sa_upgrade_txholds(tx, zp); 4601 err = dmu_tx_assign(tx, TXG_WAIT); 4602 if (err != 0) { 4603 dmu_tx_abort(tx); 4604 goto out; 4605 } 4606 4607 if (zp->z_blksz <= PAGESIZE) { 4608 caddr_t va = zfs_map_page(pp, S_READ); 4609 ASSERT3U(len, <=, PAGESIZE); 4610 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 4611 zfs_unmap_page(pp, va); 4612 } else { 4613 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 4614 } 4615 4616 if (err == 0) { 4617 uint64_t mtime[2], ctime[2]; 4618 sa_bulk_attr_t bulk[3]; 4619 int count = 0; 4620 4621 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4622 &mtime, 16); 4623 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4624 &ctime, 16); 4625 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4626 &zp->z_pflags, 8); 4627 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4628 B_TRUE); 4629 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4630 } 4631 dmu_tx_commit(tx); 4632 4633out: 4634 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 4635 if (offp) 4636 *offp = off; 4637 if (lenp) 4638 *lenp = len; 4639 4640 return (err); 4641} 4642 4643/* 4644 * Copy the portion of the file indicated from pages into the file. 4645 * The pages are stored in a page list attached to the files vnode. 4646 * 4647 * IN: vp - vnode of file to push page data to. 4648 * off - position in file to put data. 4649 * len - amount of data to write. 4650 * flags - flags to control the operation. 4651 * cr - credentials of caller. 4652 * ct - caller context. 4653 * 4654 * RETURN: 0 on success, error code on failure. 4655 * 4656 * Timestamps: 4657 * vp - ctime|mtime updated 4658 */ 4659/*ARGSUSED*/ 4660static int 4661zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4662 caller_context_t *ct) 4663{ 4664 znode_t *zp = VTOZ(vp); 4665 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4666 page_t *pp; 4667 size_t io_len; 4668 u_offset_t io_off; 4669 uint_t blksz; 4670 rl_t *rl; 4671 int error = 0; 4672 4673 ZFS_ENTER(zfsvfs); 4674 ZFS_VERIFY_ZP(zp); 4675 4676 /* 4677 * Align this request to the file block size in case we kluster. 4678 * XXX - this can result in pretty aggresive locking, which can 4679 * impact simultanious read/write access. One option might be 4680 * to break up long requests (len == 0) into block-by-block 4681 * operations to get narrower locking. 4682 */ 4683 blksz = zp->z_blksz; 4684 if (ISP2(blksz)) 4685 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 4686 else 4687 io_off = 0; 4688 if (len > 0 && ISP2(blksz)) 4689 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 4690 else 4691 io_len = 0; 4692 4693 if (io_len == 0) { 4694 /* 4695 * Search the entire vp list for pages >= io_off. 4696 */ 4697 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 4698 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 4699 goto out; 4700 } 4701 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 4702 4703 if (off > zp->z_size) { 4704 /* past end of file */ 4705 zfs_range_unlock(rl); 4706 ZFS_EXIT(zfsvfs); 4707 return (0); 4708 } 4709 4710 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); 4711 4712 for (off = io_off; io_off < off + len; io_off += io_len) { 4713 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 4714 pp = page_lookup(vp, io_off, 4715 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 4716 } else { 4717 pp = page_lookup_nowait(vp, io_off, 4718 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 4719 } 4720 4721 if (pp != NULL && pvn_getdirty(pp, flags)) { 4722 int err; 4723 4724 /* 4725 * Found a dirty page to push 4726 */ 4727 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 4728 if (err) 4729 error = err; 4730 } else { 4731 io_len = PAGESIZE; 4732 } 4733 } 4734out: 4735 zfs_range_unlock(rl); 4736 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4737 zil_commit(zfsvfs->z_log, zp->z_id); 4738 ZFS_EXIT(zfsvfs); 4739 return (error); 4740} 4741#endif /* illumos */ 4742 4743/*ARGSUSED*/ 4744void 4745zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4746{ 4747 znode_t *zp = VTOZ(vp); 4748 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4749 int error; 4750 4751 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4752 if (zp->z_sa_hdl == NULL) { 4753 /* 4754 * The fs has been unmounted, or we did a 4755 * suspend/resume and this file no longer exists. 4756 */ 4757 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4758 vrecycle(vp); 4759 return; 4760 } 4761 4762 mutex_enter(&zp->z_lock); 4763 if (zp->z_unlinked) { 4764 /* 4765 * Fast path to recycle a vnode of a removed file. 4766 */ 4767 mutex_exit(&zp->z_lock); 4768 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4769 vrecycle(vp); 4770 return; 4771 } 4772 mutex_exit(&zp->z_lock); 4773 4774 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4775 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4776 4777 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4778 zfs_sa_upgrade_txholds(tx, zp); 4779 error = dmu_tx_assign(tx, TXG_WAIT); 4780 if (error) { 4781 dmu_tx_abort(tx); 4782 } else { 4783 mutex_enter(&zp->z_lock); 4784 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4785 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4786 zp->z_atime_dirty = 0; 4787 mutex_exit(&zp->z_lock); 4788 dmu_tx_commit(tx); 4789 } 4790 } 4791 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4792} 4793 4794#ifdef illumos 4795/* 4796 * Bounds-check the seek operation. 4797 * 4798 * IN: vp - vnode seeking within 4799 * ooff - old file offset 4800 * noffp - pointer to new file offset 4801 * ct - caller context 4802 * 4803 * RETURN: 0 on success, EINVAL if new offset invalid. 4804 */ 4805/* ARGSUSED */ 4806static int 4807zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 4808 caller_context_t *ct) 4809{ 4810 if (vp->v_type == VDIR) 4811 return (0); 4812 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4813} 4814 4815/* 4816 * Pre-filter the generic locking function to trap attempts to place 4817 * a mandatory lock on a memory mapped file. 4818 */ 4819static int 4820zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 4821 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 4822{ 4823 znode_t *zp = VTOZ(vp); 4824 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4825 4826 ZFS_ENTER(zfsvfs); 4827 ZFS_VERIFY_ZP(zp); 4828 4829 /* 4830 * We are following the UFS semantics with respect to mapcnt 4831 * here: If we see that the file is mapped already, then we will 4832 * return an error, but we don't worry about races between this 4833 * function and zfs_map(). 4834 */ 4835 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { 4836 ZFS_EXIT(zfsvfs); 4837 return (SET_ERROR(EAGAIN)); 4838 } 4839 ZFS_EXIT(zfsvfs); 4840 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4841} 4842 4843/* 4844 * If we can't find a page in the cache, we will create a new page 4845 * and fill it with file data. For efficiency, we may try to fill 4846 * multiple pages at once (klustering) to fill up the supplied page 4847 * list. Note that the pages to be filled are held with an exclusive 4848 * lock to prevent access by other threads while they are being filled. 4849 */ 4850static int 4851zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 4852 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 4853{ 4854 znode_t *zp = VTOZ(vp); 4855 page_t *pp, *cur_pp; 4856 objset_t *os = zp->z_zfsvfs->z_os; 4857 u_offset_t io_off, total; 4858 size_t io_len; 4859 int err; 4860 4861 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4862 /* 4863 * We only have a single page, don't bother klustering 4864 */ 4865 io_off = off; 4866 io_len = PAGESIZE; 4867 pp = page_create_va(vp, io_off, io_len, 4868 PG_EXCL | PG_WAIT, seg, addr); 4869 } else { 4870 /* 4871 * Try to find enough pages to fill the page list 4872 */ 4873 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4874 &io_len, off, plsz, 0); 4875 } 4876 if (pp == NULL) { 4877 /* 4878 * The page already exists, nothing to do here. 4879 */ 4880 *pl = NULL; 4881 return (0); 4882 } 4883 4884 /* 4885 * Fill the pages in the kluster. 4886 */ 4887 cur_pp = pp; 4888 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4889 caddr_t va; 4890 4891 ASSERT3U(io_off, ==, cur_pp->p_offset); 4892 va = zfs_map_page(cur_pp, S_WRITE); 4893 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4894 DMU_READ_PREFETCH); 4895 zfs_unmap_page(cur_pp, va); 4896 if (err) { 4897 /* On error, toss the entire kluster */ 4898 pvn_read_done(pp, B_ERROR); 4899 /* convert checksum errors into IO errors */ 4900 if (err == ECKSUM) 4901 err = SET_ERROR(EIO); 4902 return (err); 4903 } 4904 cur_pp = cur_pp->p_next; 4905 } 4906 4907 /* 4908 * Fill in the page list array from the kluster starting 4909 * from the desired offset `off'. 4910 * NOTE: the page list will always be null terminated. 4911 */ 4912 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4913 ASSERT(pl == NULL || (*pl)->p_offset == off); 4914 4915 return (0); 4916} 4917 4918/* 4919 * Return pointers to the pages for the file region [off, off + len] 4920 * in the pl array. If plsz is greater than len, this function may 4921 * also return page pointers from after the specified region 4922 * (i.e. the region [off, off + plsz]). These additional pages are 4923 * only returned if they are already in the cache, or were created as 4924 * part of a klustered read. 4925 * 4926 * IN: vp - vnode of file to get data from. 4927 * off - position in file to get data from. 4928 * len - amount of data to retrieve. 4929 * plsz - length of provided page list. 4930 * seg - segment to obtain pages for. 4931 * addr - virtual address of fault. 4932 * rw - mode of created pages. 4933 * cr - credentials of caller. 4934 * ct - caller context. 4935 * 4936 * OUT: protp - protection mode of created pages. 4937 * pl - list of pages created. 4938 * 4939 * RETURN: 0 on success, error code on failure. 4940 * 4941 * Timestamps: 4942 * vp - atime updated 4943 */ 4944/* ARGSUSED */ 4945static int 4946zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4947 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4948 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4949{ 4950 znode_t *zp = VTOZ(vp); 4951 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4952 page_t **pl0 = pl; 4953 int err = 0; 4954 4955 /* we do our own caching, faultahead is unnecessary */ 4956 if (pl == NULL) 4957 return (0); 4958 else if (len > plsz) 4959 len = plsz; 4960 else 4961 len = P2ROUNDUP(len, PAGESIZE); 4962 ASSERT(plsz >= len); 4963 4964 ZFS_ENTER(zfsvfs); 4965 ZFS_VERIFY_ZP(zp); 4966 4967 if (protp) 4968 *protp = PROT_ALL; 4969 4970 /* 4971 * Loop through the requested range [off, off + len) looking 4972 * for pages. If we don't find a page, we will need to create 4973 * a new page and fill it with data from the file. 4974 */ 4975 while (len > 0) { 4976 if (*pl = page_lookup(vp, off, SE_SHARED)) 4977 *(pl+1) = NULL; 4978 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 4979 goto out; 4980 while (*pl) { 4981 ASSERT3U((*pl)->p_offset, ==, off); 4982 off += PAGESIZE; 4983 addr += PAGESIZE; 4984 if (len > 0) { 4985 ASSERT3U(len, >=, PAGESIZE); 4986 len -= PAGESIZE; 4987 } 4988 ASSERT3U(plsz, >=, PAGESIZE); 4989 plsz -= PAGESIZE; 4990 pl++; 4991 } 4992 } 4993 4994 /* 4995 * Fill out the page array with any pages already in the cache. 4996 */ 4997 while (plsz > 0 && 4998 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 4999 off += PAGESIZE; 5000 plsz -= PAGESIZE; 5001 } 5002out: 5003 if (err) { 5004 /* 5005 * Release any pages we have previously locked. 5006 */ 5007 while (pl > pl0) 5008 page_unlock(*--pl); 5009 } else { 5010 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 5011 } 5012 5013 *pl = NULL; 5014 5015 ZFS_EXIT(zfsvfs); 5016 return (err); 5017} 5018 5019/* 5020 * Request a memory map for a section of a file. This code interacts 5021 * with common code and the VM system as follows: 5022 * 5023 * - common code calls mmap(), which ends up in smmap_common() 5024 * - this calls VOP_MAP(), which takes you into (say) zfs 5025 * - zfs_map() calls as_map(), passing segvn_create() as the callback 5026 * - segvn_create() creates the new segment and calls VOP_ADDMAP() 5027 * - zfs_addmap() updates z_mapcnt 5028 */ 5029/*ARGSUSED*/ 5030static int 5031zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 5032 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 5033 caller_context_t *ct) 5034{ 5035 znode_t *zp = VTOZ(vp); 5036 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5037 segvn_crargs_t vn_a; 5038 int error; 5039 5040 ZFS_ENTER(zfsvfs); 5041 ZFS_VERIFY_ZP(zp); 5042 5043 if ((prot & PROT_WRITE) && (zp->z_pflags & 5044 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 5045 ZFS_EXIT(zfsvfs); 5046 return (SET_ERROR(EPERM)); 5047 } 5048 5049 if ((prot & (PROT_READ | PROT_EXEC)) && 5050 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 5051 ZFS_EXIT(zfsvfs); 5052 return (SET_ERROR(EACCES)); 5053 } 5054 5055 if (vp->v_flag & VNOMAP) { 5056 ZFS_EXIT(zfsvfs); 5057 return (SET_ERROR(ENOSYS)); 5058 } 5059 5060 if (off < 0 || len > MAXOFFSET_T - off) { 5061 ZFS_EXIT(zfsvfs); 5062 return (SET_ERROR(ENXIO)); 5063 } 5064 5065 if (vp->v_type != VREG) { 5066 ZFS_EXIT(zfsvfs); 5067 return (SET_ERROR(ENODEV)); 5068 } 5069 5070 /* 5071 * If file is locked, disallow mapping. 5072 */ 5073 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { 5074 ZFS_EXIT(zfsvfs); 5075 return (SET_ERROR(EAGAIN)); 5076 } 5077 5078 as_rangelock(as); 5079 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 5080 if (error != 0) { 5081 as_rangeunlock(as); 5082 ZFS_EXIT(zfsvfs); 5083 return (error); 5084 } 5085 5086 vn_a.vp = vp; 5087 vn_a.offset = (u_offset_t)off; 5088 vn_a.type = flags & MAP_TYPE; 5089 vn_a.prot = prot; 5090 vn_a.maxprot = maxprot; 5091 vn_a.cred = cr; 5092 vn_a.amp = NULL; 5093 vn_a.flags = flags & ~MAP_TYPE; 5094 vn_a.szc = 0; 5095 vn_a.lgrp_mem_policy_flags = 0; 5096 5097 error = as_map(as, *addrp, len, segvn_create, &vn_a); 5098 5099 as_rangeunlock(as); 5100 ZFS_EXIT(zfsvfs); 5101 return (error); 5102} 5103 5104/* ARGSUSED */ 5105static int 5106zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5107 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 5108 caller_context_t *ct) 5109{ 5110 uint64_t pages = btopr(len); 5111 5112 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 5113 return (0); 5114} 5115 5116/* 5117 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 5118 * more accurate mtime for the associated file. Since we don't have a way of 5119 * detecting when the data was actually modified, we have to resort to 5120 * heuristics. If an explicit msync() is done, then we mark the mtime when the 5121 * last page is pushed. The problem occurs when the msync() call is omitted, 5122 * which by far the most common case: 5123 * 5124 * open() 5125 * mmap() 5126 * <modify memory> 5127 * munmap() 5128 * close() 5129 * <time lapse> 5130 * putpage() via fsflush 5131 * 5132 * If we wait until fsflush to come along, we can have a modification time that 5133 * is some arbitrary point in the future. In order to prevent this in the 5134 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 5135 * torn down. 5136 */ 5137/* ARGSUSED */ 5138static int 5139zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5140 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 5141 caller_context_t *ct) 5142{ 5143 uint64_t pages = btopr(len); 5144 5145 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 5146 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 5147 5148 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 5149 vn_has_cached_data(vp)) 5150 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 5151 5152 return (0); 5153} 5154 5155/* 5156 * Free or allocate space in a file. Currently, this function only 5157 * supports the `F_FREESP' command. However, this command is somewhat 5158 * misnamed, as its functionality includes the ability to allocate as 5159 * well as free space. 5160 * 5161 * IN: vp - vnode of file to free data in. 5162 * cmd - action to take (only F_FREESP supported). 5163 * bfp - section of file to free/alloc. 5164 * flag - current file open mode flags. 5165 * offset - current file offset. 5166 * cr - credentials of caller [UNUSED]. 5167 * ct - caller context. 5168 * 5169 * RETURN: 0 on success, error code on failure. 5170 * 5171 * Timestamps: 5172 * vp - ctime|mtime updated 5173 */ 5174/* ARGSUSED */ 5175static int 5176zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 5177 offset_t offset, cred_t *cr, caller_context_t *ct) 5178{ 5179 znode_t *zp = VTOZ(vp); 5180 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5181 uint64_t off, len; 5182 int error; 5183 5184 ZFS_ENTER(zfsvfs); 5185 ZFS_VERIFY_ZP(zp); 5186 5187 if (cmd != F_FREESP) { 5188 ZFS_EXIT(zfsvfs); 5189 return (SET_ERROR(EINVAL)); 5190 } 5191 5192 /* 5193 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our 5194 * callers might not be able to detect properly that we are read-only, 5195 * so check it explicitly here. 5196 */ 5197 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 5198 ZFS_EXIT(zfsvfs); 5199 return (SET_ERROR(EROFS)); 5200 } 5201 5202 if (error = convoff(vp, bfp, 0, offset)) { 5203 ZFS_EXIT(zfsvfs); 5204 return (error); 5205 } 5206 5207 if (bfp->l_len < 0) { 5208 ZFS_EXIT(zfsvfs); 5209 return (SET_ERROR(EINVAL)); 5210 } 5211 5212 off = bfp->l_start; 5213 len = bfp->l_len; /* 0 means from off to end of file */ 5214 5215 error = zfs_freesp(zp, off, len, flag, TRUE); 5216 5217 ZFS_EXIT(zfsvfs); 5218 return (error); 5219} 5220#endif /* illumos */ 5221 5222CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 5223CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 5224 5225/*ARGSUSED*/ 5226static int 5227zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 5228{ 5229 znode_t *zp = VTOZ(vp); 5230 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5231 uint32_t gen; 5232 uint64_t gen64; 5233 uint64_t object = zp->z_id; 5234 zfid_short_t *zfid; 5235 int size, i, error; 5236 5237 ZFS_ENTER(zfsvfs); 5238 ZFS_VERIFY_ZP(zp); 5239 5240 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 5241 &gen64, sizeof (uint64_t))) != 0) { 5242 ZFS_EXIT(zfsvfs); 5243 return (error); 5244 } 5245 5246 gen = (uint32_t)gen64; 5247 5248 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 5249 5250#ifdef illumos 5251 if (fidp->fid_len < size) { 5252 fidp->fid_len = size; 5253 ZFS_EXIT(zfsvfs); 5254 return (SET_ERROR(ENOSPC)); 5255 } 5256#else 5257 fidp->fid_len = size; 5258#endif 5259 5260 zfid = (zfid_short_t *)fidp; 5261 5262 zfid->zf_len = size; 5263 5264 for (i = 0; i < sizeof (zfid->zf_object); i++) 5265 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 5266 5267 /* Must have a non-zero generation number to distinguish from .zfs */ 5268 if (gen == 0) 5269 gen = 1; 5270 for (i = 0; i < sizeof (zfid->zf_gen); i++) 5271 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 5272 5273 if (size == LONG_FID_LEN) { 5274 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 5275 zfid_long_t *zlfid; 5276 5277 zlfid = (zfid_long_t *)fidp; 5278 5279 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 5280 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 5281 5282 /* XXX - this should be the generation number for the objset */ 5283 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 5284 zlfid->zf_setgen[i] = 0; 5285 } 5286 5287 ZFS_EXIT(zfsvfs); 5288 return (0); 5289} 5290 5291static int 5292zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 5293 caller_context_t *ct) 5294{ 5295 znode_t *zp, *xzp; 5296 zfsvfs_t *zfsvfs; 5297 zfs_dirlock_t *dl; 5298 int error; 5299 5300 switch (cmd) { 5301 case _PC_LINK_MAX: 5302 *valp = INT_MAX; 5303 return (0); 5304 5305 case _PC_FILESIZEBITS: 5306 *valp = 64; 5307 return (0); 5308#ifdef illumos 5309 case _PC_XATTR_EXISTS: 5310 zp = VTOZ(vp); 5311 zfsvfs = zp->z_zfsvfs; 5312 ZFS_ENTER(zfsvfs); 5313 ZFS_VERIFY_ZP(zp); 5314 *valp = 0; 5315 error = zfs_dirent_lock(&dl, zp, "", &xzp, 5316 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 5317 if (error == 0) { 5318 zfs_dirent_unlock(dl); 5319 if (!zfs_dirempty(xzp)) 5320 *valp = 1; 5321 VN_RELE(ZTOV(xzp)); 5322 } else if (error == ENOENT) { 5323 /* 5324 * If there aren't extended attributes, it's the 5325 * same as having zero of them. 5326 */ 5327 error = 0; 5328 } 5329 ZFS_EXIT(zfsvfs); 5330 return (error); 5331 5332 case _PC_SATTR_ENABLED: 5333 case _PC_SATTR_EXISTS: 5334 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 5335 (vp->v_type == VREG || vp->v_type == VDIR); 5336 return (0); 5337 5338 case _PC_ACCESS_FILTERING: 5339 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 5340 vp->v_type == VDIR; 5341 return (0); 5342 5343 case _PC_ACL_ENABLED: 5344 *valp = _ACL_ACE_ENABLED; 5345 return (0); 5346#endif /* illumos */ 5347 case _PC_MIN_HOLE_SIZE: 5348 *valp = (int)SPA_MINBLOCKSIZE; 5349 return (0); 5350#ifdef illumos 5351 case _PC_TIMESTAMP_RESOLUTION: 5352 /* nanosecond timestamp resolution */ 5353 *valp = 1L; 5354 return (0); 5355#endif 5356 case _PC_ACL_EXTENDED: 5357 *valp = 0; 5358 return (0); 5359 5360 case _PC_ACL_NFS4: 5361 *valp = 1; 5362 return (0); 5363 5364 case _PC_ACL_PATH_MAX: 5365 *valp = ACL_MAX_ENTRIES; 5366 return (0); 5367 5368 default: 5369 return (EOPNOTSUPP); 5370 } 5371} 5372 5373/*ARGSUSED*/ 5374static int 5375zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5376 caller_context_t *ct) 5377{ 5378 znode_t *zp = VTOZ(vp); 5379 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5380 int error; 5381 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5382 5383 ZFS_ENTER(zfsvfs); 5384 ZFS_VERIFY_ZP(zp); 5385 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 5386 ZFS_EXIT(zfsvfs); 5387 5388 return (error); 5389} 5390 5391/*ARGSUSED*/ 5392int 5393zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5394 caller_context_t *ct) 5395{ 5396 znode_t *zp = VTOZ(vp); 5397 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5398 int error; 5399 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5400 zilog_t *zilog = zfsvfs->z_log; 5401 5402 ZFS_ENTER(zfsvfs); 5403 ZFS_VERIFY_ZP(zp); 5404 5405 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 5406 5407 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 5408 zil_commit(zilog, 0); 5409 5410 ZFS_EXIT(zfsvfs); 5411 return (error); 5412} 5413 5414#ifdef illumos 5415/* 5416 * The smallest read we may consider to loan out an arcbuf. 5417 * This must be a power of 2. 5418 */ 5419int zcr_blksz_min = (1 << 10); /* 1K */ 5420/* 5421 * If set to less than the file block size, allow loaning out of an 5422 * arcbuf for a partial block read. This must be a power of 2. 5423 */ 5424int zcr_blksz_max = (1 << 17); /* 128K */ 5425 5426/*ARGSUSED*/ 5427static int 5428zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr, 5429 caller_context_t *ct) 5430{ 5431 znode_t *zp = VTOZ(vp); 5432 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5433 int max_blksz = zfsvfs->z_max_blksz; 5434 uio_t *uio = &xuio->xu_uio; 5435 ssize_t size = uio->uio_resid; 5436 offset_t offset = uio->uio_loffset; 5437 int blksz; 5438 int fullblk, i; 5439 arc_buf_t *abuf; 5440 ssize_t maxsize; 5441 int preamble, postamble; 5442 5443 if (xuio->xu_type != UIOTYPE_ZEROCOPY) 5444 return (SET_ERROR(EINVAL)); 5445 5446 ZFS_ENTER(zfsvfs); 5447 ZFS_VERIFY_ZP(zp); 5448 switch (ioflag) { 5449 case UIO_WRITE: 5450 /* 5451 * Loan out an arc_buf for write if write size is bigger than 5452 * max_blksz, and the file's block size is also max_blksz. 5453 */ 5454 blksz = max_blksz; 5455 if (size < blksz || zp->z_blksz != blksz) { 5456 ZFS_EXIT(zfsvfs); 5457 return (SET_ERROR(EINVAL)); 5458 } 5459 /* 5460 * Caller requests buffers for write before knowing where the 5461 * write offset might be (e.g. NFS TCP write). 5462 */ 5463 if (offset == -1) { 5464 preamble = 0; 5465 } else { 5466 preamble = P2PHASE(offset, blksz); 5467 if (preamble) { 5468 preamble = blksz - preamble; 5469 size -= preamble; 5470 } 5471 } 5472 5473 postamble = P2PHASE(size, blksz); 5474 size -= postamble; 5475 5476 fullblk = size / blksz; 5477 (void) dmu_xuio_init(xuio, 5478 (preamble != 0) + fullblk + (postamble != 0)); 5479 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble, 5480 int, postamble, int, 5481 (preamble != 0) + fullblk + (postamble != 0)); 5482 5483 /* 5484 * Have to fix iov base/len for partial buffers. They 5485 * currently represent full arc_buf's. 5486 */ 5487 if (preamble) { 5488 /* data begins in the middle of the arc_buf */ 5489 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5490 blksz); 5491 ASSERT(abuf); 5492 (void) dmu_xuio_add(xuio, abuf, 5493 blksz - preamble, preamble); 5494 } 5495 5496 for (i = 0; i < fullblk; i++) { 5497 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5498 blksz); 5499 ASSERT(abuf); 5500 (void) dmu_xuio_add(xuio, abuf, 0, blksz); 5501 } 5502 5503 if (postamble) { 5504 /* data ends in the middle of the arc_buf */ 5505 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5506 blksz); 5507 ASSERT(abuf); 5508 (void) dmu_xuio_add(xuio, abuf, 0, postamble); 5509 } 5510 break; 5511 case UIO_READ: 5512 /* 5513 * Loan out an arc_buf for read if the read size is larger than 5514 * the current file block size. Block alignment is not 5515 * considered. Partial arc_buf will be loaned out for read. 5516 */ 5517 blksz = zp->z_blksz; 5518 if (blksz < zcr_blksz_min) 5519 blksz = zcr_blksz_min; 5520 if (blksz > zcr_blksz_max) 5521 blksz = zcr_blksz_max; 5522 /* avoid potential complexity of dealing with it */ 5523 if (blksz > max_blksz) { 5524 ZFS_EXIT(zfsvfs); 5525 return (SET_ERROR(EINVAL)); 5526 } 5527 5528 maxsize = zp->z_size - uio->uio_loffset; 5529 if (size > maxsize) 5530 size = maxsize; 5531 5532 if (size < blksz || vn_has_cached_data(vp)) { 5533 ZFS_EXIT(zfsvfs); 5534 return (SET_ERROR(EINVAL)); 5535 } 5536 break; 5537 default: 5538 ZFS_EXIT(zfsvfs); 5539 return (SET_ERROR(EINVAL)); 5540 } 5541 5542 uio->uio_extflg = UIO_XUIO; 5543 XUIO_XUZC_RW(xuio) = ioflag; 5544 ZFS_EXIT(zfsvfs); 5545 return (0); 5546} 5547 5548/*ARGSUSED*/ 5549static int 5550zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct) 5551{ 5552 int i; 5553 arc_buf_t *abuf; 5554 int ioflag = XUIO_XUZC_RW(xuio); 5555 5556 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY); 5557 5558 i = dmu_xuio_cnt(xuio); 5559 while (i-- > 0) { 5560 abuf = dmu_xuio_arcbuf(xuio, i); 5561 /* 5562 * if abuf == NULL, it must be a write buffer 5563 * that has been returned in zfs_write(). 5564 */ 5565 if (abuf) 5566 dmu_return_arcbuf(abuf); 5567 ASSERT(abuf || ioflag == UIO_WRITE); 5568 } 5569 5570 dmu_xuio_fini(xuio); 5571 return (0); 5572} 5573 5574/* 5575 * Predeclare these here so that the compiler assumes that 5576 * this is an "old style" function declaration that does 5577 * not include arguments => we won't get type mismatch errors 5578 * in the initializations that follow. 5579 */ 5580static int zfs_inval(); 5581static int zfs_isdir(); 5582 5583static int 5584zfs_inval() 5585{ 5586 return (SET_ERROR(EINVAL)); 5587} 5588 5589static int 5590zfs_isdir() 5591{ 5592 return (SET_ERROR(EISDIR)); 5593} 5594/* 5595 * Directory vnode operations template 5596 */ 5597vnodeops_t *zfs_dvnodeops; 5598const fs_operation_def_t zfs_dvnodeops_template[] = { 5599 VOPNAME_OPEN, { .vop_open = zfs_open }, 5600 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5601 VOPNAME_READ, { .error = zfs_isdir }, 5602 VOPNAME_WRITE, { .error = zfs_isdir }, 5603 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5604 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5605 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5606 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5607 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5608 VOPNAME_CREATE, { .vop_create = zfs_create }, 5609 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5610 VOPNAME_LINK, { .vop_link = zfs_link }, 5611 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5612 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 5613 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5614 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5615 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 5616 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5617 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5618 VOPNAME_FID, { .vop_fid = zfs_fid }, 5619 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5620 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5621 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5622 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5623 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5624 NULL, NULL 5625}; 5626 5627/* 5628 * Regular file vnode operations template 5629 */ 5630vnodeops_t *zfs_fvnodeops; 5631const fs_operation_def_t zfs_fvnodeops_template[] = { 5632 VOPNAME_OPEN, { .vop_open = zfs_open }, 5633 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5634 VOPNAME_READ, { .vop_read = zfs_read }, 5635 VOPNAME_WRITE, { .vop_write = zfs_write }, 5636 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5637 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5638 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5639 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5640 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5641 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5642 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5643 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5644 VOPNAME_FID, { .vop_fid = zfs_fid }, 5645 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5646 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 5647 VOPNAME_SPACE, { .vop_space = zfs_space }, 5648 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 5649 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 5650 VOPNAME_MAP, { .vop_map = zfs_map }, 5651 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 5652 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 5653 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5654 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5655 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5656 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5657 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf }, 5658 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf }, 5659 NULL, NULL 5660}; 5661 5662/* 5663 * Symbolic link vnode operations template 5664 */ 5665vnodeops_t *zfs_symvnodeops; 5666const fs_operation_def_t zfs_symvnodeops_template[] = { 5667 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5668 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5669 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5670 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5671 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 5672 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5673 VOPNAME_FID, { .vop_fid = zfs_fid }, 5674 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5675 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5676 NULL, NULL 5677}; 5678 5679/* 5680 * special share hidden files vnode operations template 5681 */ 5682vnodeops_t *zfs_sharevnodeops; 5683const fs_operation_def_t zfs_sharevnodeops_template[] = { 5684 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5685 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5686 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5687 VOPNAME_FID, { .vop_fid = zfs_fid }, 5688 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5689 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5690 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5691 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5692 NULL, NULL 5693}; 5694 5695/* 5696 * Extended attribute directory vnode operations template 5697 * 5698 * This template is identical to the directory vnodes 5699 * operation template except for restricted operations: 5700 * VOP_MKDIR() 5701 * VOP_SYMLINK() 5702 * 5703 * Note that there are other restrictions embedded in: 5704 * zfs_create() - restrict type to VREG 5705 * zfs_link() - no links into/out of attribute space 5706 * zfs_rename() - no moves into/out of attribute space 5707 */ 5708vnodeops_t *zfs_xdvnodeops; 5709const fs_operation_def_t zfs_xdvnodeops_template[] = { 5710 VOPNAME_OPEN, { .vop_open = zfs_open }, 5711 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5712 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5713 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5714 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5715 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5716 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5717 VOPNAME_CREATE, { .vop_create = zfs_create }, 5718 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5719 VOPNAME_LINK, { .vop_link = zfs_link }, 5720 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5721 VOPNAME_MKDIR, { .error = zfs_inval }, 5722 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5723 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5724 VOPNAME_SYMLINK, { .error = zfs_inval }, 5725 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5726 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5727 VOPNAME_FID, { .vop_fid = zfs_fid }, 5728 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5729 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5730 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5731 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5732 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5733 NULL, NULL 5734}; 5735 5736/* 5737 * Error vnode operations template 5738 */ 5739vnodeops_t *zfs_evnodeops; 5740const fs_operation_def_t zfs_evnodeops_template[] = { 5741 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5742 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5743 NULL, NULL 5744}; 5745#endif /* illumos */ 5746 5747static int 5748ioflags(int ioflags) 5749{ 5750 int flags = 0; 5751 5752 if (ioflags & IO_APPEND) 5753 flags |= FAPPEND; 5754 if (ioflags & IO_NDELAY) 5755 flags |= FNONBLOCK; 5756 if (ioflags & IO_SYNC) 5757 flags |= (FSYNC | FDSYNC | FRSYNC); 5758 5759 return (flags); 5760} 5761 5762static int 5763zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage) 5764{ 5765 znode_t *zp = VTOZ(vp); 5766 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5767 objset_t *os = zp->z_zfsvfs->z_os; 5768 vm_page_t mfirst, mlast, mreq; 5769 vm_object_t object; 5770 caddr_t va; 5771 struct sf_buf *sf; 5772 off_t startoff, endoff; 5773 int i, error; 5774 vm_pindex_t reqstart, reqend; 5775 int pcount, lsize, reqsize, size; 5776 5777 ZFS_ENTER(zfsvfs); 5778 ZFS_VERIFY_ZP(zp); 5779 5780 pcount = OFF_TO_IDX(round_page(count)); 5781 mreq = m[reqpage]; 5782 object = mreq->object; 5783 error = 0; 5784 5785 KASSERT(vp->v_object == object, ("mismatching object")); 5786 5787 if (pcount > 1 && zp->z_blksz > PAGESIZE) { 5788 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz); 5789 reqstart = OFF_TO_IDX(round_page(startoff)); 5790 if (reqstart < m[0]->pindex) 5791 reqstart = 0; 5792 else 5793 reqstart = reqstart - m[0]->pindex; 5794 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE, 5795 zp->z_blksz); 5796 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1; 5797 if (reqend > m[pcount - 1]->pindex) 5798 reqend = m[pcount - 1]->pindex; 5799 reqsize = reqend - m[reqstart]->pindex + 1; 5800 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize, 5801 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds")); 5802 } else { 5803 reqstart = reqpage; 5804 reqsize = 1; 5805 } 5806 mfirst = m[reqstart]; 5807 mlast = m[reqstart + reqsize - 1]; 5808 5809 zfs_vmobject_wlock(object); 5810 5811 for (i = 0; i < reqstart; i++) { 5812 vm_page_lock(m[i]); 5813 vm_page_free(m[i]); 5814 vm_page_unlock(m[i]); 5815 } 5816 for (i = reqstart + reqsize; i < pcount; i++) { 5817 vm_page_lock(m[i]); 5818 vm_page_free(m[i]); 5819 vm_page_unlock(m[i]); 5820 } 5821 5822 if (mreq->valid && reqsize == 1) { 5823 if (mreq->valid != VM_PAGE_BITS_ALL) 5824 vm_page_zero_invalid(mreq, TRUE); 5825 zfs_vmobject_wunlock(object); 5826 ZFS_EXIT(zfsvfs); 5827 return (zfs_vm_pagerret_ok); 5828 } 5829 5830 PCPU_INC(cnt.v_vnodein); 5831 PCPU_ADD(cnt.v_vnodepgsin, reqsize); 5832 5833 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) { 5834 for (i = reqstart; i < reqstart + reqsize; i++) { 5835 if (i != reqpage) { 5836 vm_page_lock(m[i]); 5837 vm_page_free(m[i]); 5838 vm_page_unlock(m[i]); 5839 } 5840 } 5841 zfs_vmobject_wunlock(object); 5842 ZFS_EXIT(zfsvfs); 5843 return (zfs_vm_pagerret_bad); 5844 } 5845 5846 lsize = PAGE_SIZE; 5847 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size) 5848 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex); 5849 5850 zfs_vmobject_wunlock(object); 5851 5852 for (i = reqstart; i < reqstart + reqsize; i++) { 5853 size = PAGE_SIZE; 5854 if (i == (reqstart + reqsize - 1)) 5855 size = lsize; 5856 va = zfs_map_page(m[i], &sf); 5857 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex), 5858 size, va, DMU_READ_PREFETCH); 5859 if (size != PAGE_SIZE) 5860 bzero(va + size, PAGE_SIZE - size); 5861 zfs_unmap_page(sf); 5862 if (error != 0) 5863 break; 5864 } 5865 5866 zfs_vmobject_wlock(object); 5867 5868 for (i = reqstart; i < reqstart + reqsize; i++) { 5869 if (!error) 5870 m[i]->valid = VM_PAGE_BITS_ALL; 5871 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i])); 5872 if (i != reqpage) 5873 vm_page_readahead_finish(m[i]); 5874 } 5875 5876 zfs_vmobject_wunlock(object); 5877 5878 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 5879 ZFS_EXIT(zfsvfs); 5880 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok); 5881} 5882 5883static int 5884zfs_freebsd_getpages(ap) 5885 struct vop_getpages_args /* { 5886 struct vnode *a_vp; 5887 vm_page_t *a_m; 5888 int a_count; 5889 int a_reqpage; 5890 vm_ooffset_t a_offset; 5891 } */ *ap; 5892{ 5893 5894 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage)); 5895} 5896 5897static int 5898zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags, 5899 int *rtvals) 5900{ 5901 znode_t *zp = VTOZ(vp); 5902 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5903 rl_t *rl; 5904 dmu_tx_t *tx; 5905 struct sf_buf *sf; 5906 vm_object_t object; 5907 vm_page_t m; 5908 caddr_t va; 5909 size_t tocopy; 5910 size_t lo_len; 5911 vm_ooffset_t lo_off; 5912 vm_ooffset_t off; 5913 uint_t blksz; 5914 int ncount; 5915 int pcount; 5916 int err; 5917 int i; 5918 5919 ZFS_ENTER(zfsvfs); 5920 ZFS_VERIFY_ZP(zp); 5921 5922 object = vp->v_object; 5923 pcount = btoc(len); 5924 ncount = pcount; 5925 5926 KASSERT(ma[0]->object == object, ("mismatching object")); 5927 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length")); 5928 5929 for (i = 0; i < pcount; i++) 5930 rtvals[i] = zfs_vm_pagerret_error; 5931 5932 off = IDX_TO_OFF(ma[0]->pindex); 5933 blksz = zp->z_blksz; 5934 lo_off = rounddown(off, blksz); 5935 lo_len = roundup(len + (off - lo_off), blksz); 5936 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER); 5937 5938 zfs_vmobject_wlock(object); 5939 if (len + off > object->un_pager.vnp.vnp_size) { 5940 if (object->un_pager.vnp.vnp_size > off) { 5941 int pgoff; 5942 5943 len = object->un_pager.vnp.vnp_size - off; 5944 ncount = btoc(len); 5945 if ((pgoff = (int)len & PAGE_MASK) != 0) { 5946 /* 5947 * If the object is locked and the following 5948 * conditions hold, then the page's dirty 5949 * field cannot be concurrently changed by a 5950 * pmap operation. 5951 */ 5952 m = ma[ncount - 1]; 5953 vm_page_assert_sbusied(m); 5954 KASSERT(!pmap_page_is_write_mapped(m), 5955 ("zfs_putpages: page %p is not read-only", m)); 5956 vm_page_clear_dirty(m, pgoff, PAGE_SIZE - 5957 pgoff); 5958 } 5959 } else { 5960 len = 0; 5961 ncount = 0; 5962 } 5963 if (ncount < pcount) { 5964 for (i = ncount; i < pcount; i++) { 5965 rtvals[i] = zfs_vm_pagerret_bad; 5966 } 5967 } 5968 } 5969 zfs_vmobject_wunlock(object); 5970 5971 if (ncount == 0) 5972 goto out; 5973 5974 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 5975 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 5976 goto out; 5977 } 5978 5979top: 5980 tx = dmu_tx_create(zfsvfs->z_os); 5981 dmu_tx_hold_write(tx, zp->z_id, off, len); 5982 5983 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 5984 zfs_sa_upgrade_txholds(tx, zp); 5985 err = dmu_tx_assign(tx, TXG_NOWAIT); 5986 if (err != 0) { 5987 if (err == ERESTART) { 5988 dmu_tx_wait(tx); 5989 dmu_tx_abort(tx); 5990 goto top; 5991 } 5992 dmu_tx_abort(tx); 5993 goto out; 5994 } 5995 5996 if (zp->z_blksz < PAGE_SIZE) { 5997 i = 0; 5998 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) { 5999 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len; 6000 va = zfs_map_page(ma[i], &sf); 6001 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx); 6002 zfs_unmap_page(sf); 6003 } 6004 } else { 6005 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx); 6006 } 6007 6008 if (err == 0) { 6009 uint64_t mtime[2], ctime[2]; 6010 sa_bulk_attr_t bulk[3]; 6011 int count = 0; 6012 6013 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 6014 &mtime, 16); 6015 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 6016 &ctime, 16); 6017 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 6018 &zp->z_pflags, 8); 6019 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 6020 B_TRUE); 6021 (void)sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 6022 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 6023 6024 zfs_vmobject_wlock(object); 6025 for (i = 0; i < ncount; i++) { 6026 rtvals[i] = zfs_vm_pagerret_ok; 6027 vm_page_undirty(ma[i]); 6028 } 6029 zfs_vmobject_wunlock(object); 6030 PCPU_INC(cnt.v_vnodeout); 6031 PCPU_ADD(cnt.v_vnodepgsout, ncount); 6032 } 6033 dmu_tx_commit(tx); 6034 6035out: 6036 zfs_range_unlock(rl); 6037 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 || 6038 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 6039 zil_commit(zfsvfs->z_log, zp->z_id); 6040 ZFS_EXIT(zfsvfs); 6041 return (rtvals[0]); 6042} 6043 6044int 6045zfs_freebsd_putpages(ap) 6046 struct vop_putpages_args /* { 6047 struct vnode *a_vp; 6048 vm_page_t *a_m; 6049 int a_count; 6050 int a_sync; 6051 int *a_rtvals; 6052 vm_ooffset_t a_offset; 6053 } */ *ap; 6054{ 6055 6056 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync, 6057 ap->a_rtvals)); 6058} 6059 6060static int 6061zfs_freebsd_bmap(ap) 6062 struct vop_bmap_args /* { 6063 struct vnode *a_vp; 6064 daddr_t a_bn; 6065 struct bufobj **a_bop; 6066 daddr_t *a_bnp; 6067 int *a_runp; 6068 int *a_runb; 6069 } */ *ap; 6070{ 6071 6072 if (ap->a_bop != NULL) 6073 *ap->a_bop = &ap->a_vp->v_bufobj; 6074 if (ap->a_bnp != NULL) 6075 *ap->a_bnp = ap->a_bn; 6076 if (ap->a_runp != NULL) 6077 *ap->a_runp = 0; 6078 if (ap->a_runb != NULL) 6079 *ap->a_runb = 0; 6080 6081 return (0); 6082} 6083 6084static int 6085zfs_freebsd_open(ap) 6086 struct vop_open_args /* { 6087 struct vnode *a_vp; 6088 int a_mode; 6089 struct ucred *a_cred; 6090 struct thread *a_td; 6091 } */ *ap; 6092{ 6093 vnode_t *vp = ap->a_vp; 6094 znode_t *zp = VTOZ(vp); 6095 int error; 6096 6097 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL); 6098 if (error == 0) 6099 vnode_create_vobject(vp, zp->z_size, ap->a_td); 6100 return (error); 6101} 6102 6103static int 6104zfs_freebsd_close(ap) 6105 struct vop_close_args /* { 6106 struct vnode *a_vp; 6107 int a_fflag; 6108 struct ucred *a_cred; 6109 struct thread *a_td; 6110 } */ *ap; 6111{ 6112 6113 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL)); 6114} 6115 6116static int 6117zfs_freebsd_ioctl(ap) 6118 struct vop_ioctl_args /* { 6119 struct vnode *a_vp; 6120 u_long a_command; 6121 caddr_t a_data; 6122 int a_fflag; 6123 struct ucred *cred; 6124 struct thread *td; 6125 } */ *ap; 6126{ 6127 6128 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 6129 ap->a_fflag, ap->a_cred, NULL, NULL)); 6130} 6131 6132static int 6133zfs_freebsd_read(ap) 6134 struct vop_read_args /* { 6135 struct vnode *a_vp; 6136 struct uio *a_uio; 6137 int a_ioflag; 6138 struct ucred *a_cred; 6139 } */ *ap; 6140{ 6141 6142 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 6143 ap->a_cred, NULL)); 6144} 6145 6146static int 6147zfs_freebsd_write(ap) 6148 struct vop_write_args /* { 6149 struct vnode *a_vp; 6150 struct uio *a_uio; 6151 int a_ioflag; 6152 struct ucred *a_cred; 6153 } */ *ap; 6154{ 6155 6156 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 6157 ap->a_cred, NULL)); 6158} 6159 6160static int 6161zfs_freebsd_access(ap) 6162 struct vop_access_args /* { 6163 struct vnode *a_vp; 6164 accmode_t a_accmode; 6165 struct ucred *a_cred; 6166 struct thread *a_td; 6167 } */ *ap; 6168{ 6169 vnode_t *vp = ap->a_vp; 6170 znode_t *zp = VTOZ(vp); 6171 accmode_t accmode; 6172 int error = 0; 6173 6174 /* 6175 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 6176 */ 6177 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 6178 if (accmode != 0) 6179 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL); 6180 6181 /* 6182 * VADMIN has to be handled by vaccess(). 6183 */ 6184 if (error == 0) { 6185 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 6186 if (accmode != 0) { 6187 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 6188 zp->z_gid, accmode, ap->a_cred, NULL); 6189 } 6190 } 6191 6192 /* 6193 * For VEXEC, ensure that at least one execute bit is set for 6194 * non-directories. 6195 */ 6196 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 6197 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 6198 error = EACCES; 6199 } 6200 6201 return (error); 6202} 6203 6204static int 6205zfs_freebsd_lookup(ap) 6206 struct vop_lookup_args /* { 6207 struct vnode *a_dvp; 6208 struct vnode **a_vpp; 6209 struct componentname *a_cnp; 6210 } */ *ap; 6211{ 6212 struct componentname *cnp = ap->a_cnp; 6213 char nm[NAME_MAX + 1]; 6214 6215 ASSERT(cnp->cn_namelen < sizeof(nm)); 6216 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 6217 6218 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 6219 cnp->cn_cred, cnp->cn_thread, 0)); 6220} 6221 6222static int 6223zfs_freebsd_create(ap) 6224 struct vop_create_args /* { 6225 struct vnode *a_dvp; 6226 struct vnode **a_vpp; 6227 struct componentname *a_cnp; 6228 struct vattr *a_vap; 6229 } */ *ap; 6230{ 6231 struct componentname *cnp = ap->a_cnp; 6232 vattr_t *vap = ap->a_vap; 6233 int error, mode; 6234 6235 ASSERT(cnp->cn_flags & SAVENAME); 6236 6237 vattr_init_mask(vap); 6238 mode = vap->va_mode & ALLPERMS; 6239 6240 error = zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 6241 ap->a_vpp, cnp->cn_cred, cnp->cn_thread); 6242#ifdef FREEBSD_NAMECACHE 6243 if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0) 6244 cache_enter(ap->a_dvp, *ap->a_vpp, cnp); 6245#endif 6246 return (error); 6247} 6248 6249static int 6250zfs_freebsd_remove(ap) 6251 struct vop_remove_args /* { 6252 struct vnode *a_dvp; 6253 struct vnode *a_vp; 6254 struct componentname *a_cnp; 6255 } */ *ap; 6256{ 6257 6258 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6259 6260 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 6261 ap->a_cnp->cn_cred, NULL, 0)); 6262} 6263 6264static int 6265zfs_freebsd_mkdir(ap) 6266 struct vop_mkdir_args /* { 6267 struct vnode *a_dvp; 6268 struct vnode **a_vpp; 6269 struct componentname *a_cnp; 6270 struct vattr *a_vap; 6271 } */ *ap; 6272{ 6273 vattr_t *vap = ap->a_vap; 6274 6275 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6276 6277 vattr_init_mask(vap); 6278 6279 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 6280 ap->a_cnp->cn_cred, NULL, 0, NULL)); 6281} 6282 6283static int 6284zfs_freebsd_rmdir(ap) 6285 struct vop_rmdir_args /* { 6286 struct vnode *a_dvp; 6287 struct vnode *a_vp; 6288 struct componentname *a_cnp; 6289 } */ *ap; 6290{ 6291 struct componentname *cnp = ap->a_cnp; 6292 6293 ASSERT(cnp->cn_flags & SAVENAME); 6294 6295 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0)); 6296} 6297 6298static int 6299zfs_freebsd_readdir(ap) 6300 struct vop_readdir_args /* { 6301 struct vnode *a_vp; 6302 struct uio *a_uio; 6303 struct ucred *a_cred; 6304 int *a_eofflag; 6305 int *a_ncookies; 6306 u_long **a_cookies; 6307 } */ *ap; 6308{ 6309 6310 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 6311 ap->a_ncookies, ap->a_cookies)); 6312} 6313 6314static int 6315zfs_freebsd_fsync(ap) 6316 struct vop_fsync_args /* { 6317 struct vnode *a_vp; 6318 int a_waitfor; 6319 struct thread *a_td; 6320 } */ *ap; 6321{ 6322 6323 vop_stdfsync(ap); 6324 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL)); 6325} 6326 6327static int 6328zfs_freebsd_getattr(ap) 6329 struct vop_getattr_args /* { 6330 struct vnode *a_vp; 6331 struct vattr *a_vap; 6332 struct ucred *a_cred; 6333 } */ *ap; 6334{ 6335 vattr_t *vap = ap->a_vap; 6336 xvattr_t xvap; 6337 u_long fflags = 0; 6338 int error; 6339 6340 xva_init(&xvap); 6341 xvap.xva_vattr = *vap; 6342 xvap.xva_vattr.va_mask |= AT_XVATTR; 6343 6344 /* Convert chflags into ZFS-type flags. */ 6345 /* XXX: what about SF_SETTABLE?. */ 6346 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 6347 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 6348 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 6349 XVA_SET_REQ(&xvap, XAT_NODUMP); 6350 XVA_SET_REQ(&xvap, XAT_READONLY); 6351 XVA_SET_REQ(&xvap, XAT_ARCHIVE); 6352 XVA_SET_REQ(&xvap, XAT_SYSTEM); 6353 XVA_SET_REQ(&xvap, XAT_HIDDEN); 6354 XVA_SET_REQ(&xvap, XAT_REPARSE); 6355 XVA_SET_REQ(&xvap, XAT_OFFLINE); 6356 XVA_SET_REQ(&xvap, XAT_SPARSE); 6357 6358 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL); 6359 if (error != 0) 6360 return (error); 6361 6362 /* Convert ZFS xattr into chflags. */ 6363#define FLAG_CHECK(fflag, xflag, xfield) do { \ 6364 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 6365 fflags |= (fflag); \ 6366} while (0) 6367 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 6368 xvap.xva_xoptattrs.xoa_immutable); 6369 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 6370 xvap.xva_xoptattrs.xoa_appendonly); 6371 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 6372 xvap.xva_xoptattrs.xoa_nounlink); 6373 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE, 6374 xvap.xva_xoptattrs.xoa_archive); 6375 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 6376 xvap.xva_xoptattrs.xoa_nodump); 6377 FLAG_CHECK(UF_READONLY, XAT_READONLY, 6378 xvap.xva_xoptattrs.xoa_readonly); 6379 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM, 6380 xvap.xva_xoptattrs.xoa_system); 6381 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN, 6382 xvap.xva_xoptattrs.xoa_hidden); 6383 FLAG_CHECK(UF_REPARSE, XAT_REPARSE, 6384 xvap.xva_xoptattrs.xoa_reparse); 6385 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE, 6386 xvap.xva_xoptattrs.xoa_offline); 6387 FLAG_CHECK(UF_SPARSE, XAT_SPARSE, 6388 xvap.xva_xoptattrs.xoa_sparse); 6389 6390#undef FLAG_CHECK 6391 *vap = xvap.xva_vattr; 6392 vap->va_flags = fflags; 6393 return (0); 6394} 6395 6396static int 6397zfs_freebsd_setattr(ap) 6398 struct vop_setattr_args /* { 6399 struct vnode *a_vp; 6400 struct vattr *a_vap; 6401 struct ucred *a_cred; 6402 } */ *ap; 6403{ 6404 vnode_t *vp = ap->a_vp; 6405 vattr_t *vap = ap->a_vap; 6406 cred_t *cred = ap->a_cred; 6407 xvattr_t xvap; 6408 u_long fflags; 6409 uint64_t zflags; 6410 6411 vattr_init_mask(vap); 6412 vap->va_mask &= ~AT_NOSET; 6413 6414 xva_init(&xvap); 6415 xvap.xva_vattr = *vap; 6416 6417 zflags = VTOZ(vp)->z_pflags; 6418 6419 if (vap->va_flags != VNOVAL) { 6420 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 6421 int error; 6422 6423 if (zfsvfs->z_use_fuids == B_FALSE) 6424 return (EOPNOTSUPP); 6425 6426 fflags = vap->va_flags; 6427 /* 6428 * XXX KDM 6429 * We need to figure out whether it makes sense to allow 6430 * UF_REPARSE through, since we don't really have other 6431 * facilities to handle reparse points and zfs_setattr() 6432 * doesn't currently allow setting that attribute anyway. 6433 */ 6434 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE| 6435 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE| 6436 UF_OFFLINE|UF_SPARSE)) != 0) 6437 return (EOPNOTSUPP); 6438 /* 6439 * Unprivileged processes are not permitted to unset system 6440 * flags, or modify flags if any system flags are set. 6441 * Privileged non-jail processes may not modify system flags 6442 * if securelevel > 0 and any existing system flags are set. 6443 * Privileged jail processes behave like privileged non-jail 6444 * processes if the security.jail.chflags_allowed sysctl is 6445 * is non-zero; otherwise, they behave like unprivileged 6446 * processes. 6447 */ 6448 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 6449 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) { 6450 if (zflags & 6451 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6452 error = securelevel_gt(cred, 0); 6453 if (error != 0) 6454 return (error); 6455 } 6456 } else { 6457 /* 6458 * Callers may only modify the file flags on objects they 6459 * have VADMIN rights for. 6460 */ 6461 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0) 6462 return (error); 6463 if (zflags & 6464 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6465 return (EPERM); 6466 } 6467 if (fflags & 6468 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 6469 return (EPERM); 6470 } 6471 } 6472 6473#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 6474 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 6475 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 6476 XVA_SET_REQ(&xvap, (xflag)); \ 6477 (xfield) = ((fflags & (fflag)) != 0); \ 6478 } \ 6479} while (0) 6480 /* Convert chflags into ZFS-type flags. */ 6481 /* XXX: what about SF_SETTABLE?. */ 6482 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 6483 xvap.xva_xoptattrs.xoa_immutable); 6484 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 6485 xvap.xva_xoptattrs.xoa_appendonly); 6486 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 6487 xvap.xva_xoptattrs.xoa_nounlink); 6488 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE, 6489 xvap.xva_xoptattrs.xoa_archive); 6490 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 6491 xvap.xva_xoptattrs.xoa_nodump); 6492 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY, 6493 xvap.xva_xoptattrs.xoa_readonly); 6494 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM, 6495 xvap.xva_xoptattrs.xoa_system); 6496 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN, 6497 xvap.xva_xoptattrs.xoa_hidden); 6498 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE, 6499 xvap.xva_xoptattrs.xoa_hidden); 6500 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE, 6501 xvap.xva_xoptattrs.xoa_offline); 6502 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE, 6503 xvap.xva_xoptattrs.xoa_sparse); 6504#undef FLAG_CHANGE 6505 } 6506 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL)); 6507} 6508 6509static int 6510zfs_freebsd_rename(ap) 6511 struct vop_rename_args /* { 6512 struct vnode *a_fdvp; 6513 struct vnode *a_fvp; 6514 struct componentname *a_fcnp; 6515 struct vnode *a_tdvp; 6516 struct vnode *a_tvp; 6517 struct componentname *a_tcnp; 6518 } */ *ap; 6519{ 6520 vnode_t *fdvp = ap->a_fdvp; 6521 vnode_t *fvp = ap->a_fvp; 6522 vnode_t *tdvp = ap->a_tdvp; 6523 vnode_t *tvp = ap->a_tvp; 6524 int error; 6525 6526 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 6527 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 6528 6529 /* 6530 * Check for cross-device rename. 6531 */ 6532 if ((fdvp->v_mount != tdvp->v_mount) || 6533 (tvp && (fdvp->v_mount != tvp->v_mount))) 6534 error = EXDEV; 6535 else 6536 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 6537 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0); 6538 if (tdvp == tvp) 6539 VN_RELE(tdvp); 6540 else 6541 VN_URELE(tdvp); 6542 if (tvp) 6543 VN_URELE(tvp); 6544 VN_RELE(fdvp); 6545 VN_RELE(fvp); 6546 6547 return (error); 6548} 6549 6550static int 6551zfs_freebsd_symlink(ap) 6552 struct vop_symlink_args /* { 6553 struct vnode *a_dvp; 6554 struct vnode **a_vpp; 6555 struct componentname *a_cnp; 6556 struct vattr *a_vap; 6557 char *a_target; 6558 } */ *ap; 6559{ 6560 struct componentname *cnp = ap->a_cnp; 6561 vattr_t *vap = ap->a_vap; 6562 6563 ASSERT(cnp->cn_flags & SAVENAME); 6564 6565 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 6566 vattr_init_mask(vap); 6567 6568 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 6569 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 6570} 6571 6572static int 6573zfs_freebsd_readlink(ap) 6574 struct vop_readlink_args /* { 6575 struct vnode *a_vp; 6576 struct uio *a_uio; 6577 struct ucred *a_cred; 6578 } */ *ap; 6579{ 6580 6581 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL)); 6582} 6583 6584static int 6585zfs_freebsd_link(ap) 6586 struct vop_link_args /* { 6587 struct vnode *a_tdvp; 6588 struct vnode *a_vp; 6589 struct componentname *a_cnp; 6590 } */ *ap; 6591{ 6592 struct componentname *cnp = ap->a_cnp; 6593 vnode_t *vp = ap->a_vp; 6594 vnode_t *tdvp = ap->a_tdvp; 6595 6596 if (tdvp->v_mount != vp->v_mount) 6597 return (EXDEV); 6598 6599 ASSERT(cnp->cn_flags & SAVENAME); 6600 6601 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0)); 6602} 6603 6604static int 6605zfs_freebsd_inactive(ap) 6606 struct vop_inactive_args /* { 6607 struct vnode *a_vp; 6608 struct thread *a_td; 6609 } */ *ap; 6610{ 6611 vnode_t *vp = ap->a_vp; 6612 6613 zfs_inactive(vp, ap->a_td->td_ucred, NULL); 6614 return (0); 6615} 6616 6617static int 6618zfs_freebsd_reclaim(ap) 6619 struct vop_reclaim_args /* { 6620 struct vnode *a_vp; 6621 struct thread *a_td; 6622 } */ *ap; 6623{ 6624 vnode_t *vp = ap->a_vp; 6625 znode_t *zp = VTOZ(vp); 6626 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6627 6628 ASSERT(zp != NULL); 6629 6630 /* Destroy the vm object and flush associated pages. */ 6631 vnode_destroy_vobject(vp); 6632 6633 /* 6634 * z_teardown_inactive_lock protects from a race with 6635 * zfs_znode_dmu_fini in zfsvfs_teardown during 6636 * force unmount. 6637 */ 6638 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 6639 if (zp->z_sa_hdl == NULL) 6640 zfs_znode_free(zp); 6641 else 6642 zfs_zinactive(zp); 6643 rw_exit(&zfsvfs->z_teardown_inactive_lock); 6644 6645 vp->v_data = NULL; 6646 return (0); 6647} 6648 6649static int 6650zfs_freebsd_fid(ap) 6651 struct vop_fid_args /* { 6652 struct vnode *a_vp; 6653 struct fid *a_fid; 6654 } */ *ap; 6655{ 6656 6657 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 6658} 6659 6660static int 6661zfs_freebsd_pathconf(ap) 6662 struct vop_pathconf_args /* { 6663 struct vnode *a_vp; 6664 int a_name; 6665 register_t *a_retval; 6666 } */ *ap; 6667{ 6668 ulong_t val; 6669 int error; 6670 6671 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); 6672 if (error == 0) 6673 *ap->a_retval = val; 6674 else if (error == EOPNOTSUPP) 6675 error = vop_stdpathconf(ap); 6676 return (error); 6677} 6678 6679static int 6680zfs_freebsd_fifo_pathconf(ap) 6681 struct vop_pathconf_args /* { 6682 struct vnode *a_vp; 6683 int a_name; 6684 register_t *a_retval; 6685 } */ *ap; 6686{ 6687 6688 switch (ap->a_name) { 6689 case _PC_ACL_EXTENDED: 6690 case _PC_ACL_NFS4: 6691 case _PC_ACL_PATH_MAX: 6692 case _PC_MAC_PRESENT: 6693 return (zfs_freebsd_pathconf(ap)); 6694 default: 6695 return (fifo_specops.vop_pathconf(ap)); 6696 } 6697} 6698 6699/* 6700 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 6701 * extended attribute name: 6702 * 6703 * NAMESPACE PREFIX 6704 * system freebsd:system: 6705 * user (none, can be used to access ZFS fsattr(5) attributes 6706 * created on Solaris) 6707 */ 6708static int 6709zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 6710 size_t size) 6711{ 6712 const char *namespace, *prefix, *suffix; 6713 6714 /* We don't allow '/' character in attribute name. */ 6715 if (strchr(name, '/') != NULL) 6716 return (EINVAL); 6717 /* We don't allow attribute names that start with "freebsd:" string. */ 6718 if (strncmp(name, "freebsd:", 8) == 0) 6719 return (EINVAL); 6720 6721 bzero(attrname, size); 6722 6723 switch (attrnamespace) { 6724 case EXTATTR_NAMESPACE_USER: 6725#if 0 6726 prefix = "freebsd:"; 6727 namespace = EXTATTR_NAMESPACE_USER_STRING; 6728 suffix = ":"; 6729#else 6730 /* 6731 * This is the default namespace by which we can access all 6732 * attributes created on Solaris. 6733 */ 6734 prefix = namespace = suffix = ""; 6735#endif 6736 break; 6737 case EXTATTR_NAMESPACE_SYSTEM: 6738 prefix = "freebsd:"; 6739 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 6740 suffix = ":"; 6741 break; 6742 case EXTATTR_NAMESPACE_EMPTY: 6743 default: 6744 return (EINVAL); 6745 } 6746 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 6747 name) >= size) { 6748 return (ENAMETOOLONG); 6749 } 6750 return (0); 6751} 6752 6753/* 6754 * Vnode operating to retrieve a named extended attribute. 6755 */ 6756static int 6757zfs_getextattr(struct vop_getextattr_args *ap) 6758/* 6759vop_getextattr { 6760 IN struct vnode *a_vp; 6761 IN int a_attrnamespace; 6762 IN const char *a_name; 6763 INOUT struct uio *a_uio; 6764 OUT size_t *a_size; 6765 IN struct ucred *a_cred; 6766 IN struct thread *a_td; 6767}; 6768*/ 6769{ 6770 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6771 struct thread *td = ap->a_td; 6772 struct nameidata nd; 6773 char attrname[255]; 6774 struct vattr va; 6775 vnode_t *xvp = NULL, *vp; 6776 int error, flags; 6777 6778 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6779 ap->a_cred, ap->a_td, VREAD); 6780 if (error != 0) 6781 return (error); 6782 6783 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6784 sizeof(attrname)); 6785 if (error != 0) 6786 return (error); 6787 6788 ZFS_ENTER(zfsvfs); 6789 6790 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6791 LOOKUP_XATTR); 6792 if (error != 0) { 6793 ZFS_EXIT(zfsvfs); 6794 return (error); 6795 } 6796 6797 flags = FREAD; 6798 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6799 xvp, td); 6800 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL); 6801 vp = nd.ni_vp; 6802 NDFREE(&nd, NDF_ONLY_PNBUF); 6803 if (error != 0) { 6804 ZFS_EXIT(zfsvfs); 6805 if (error == ENOENT) 6806 error = ENOATTR; 6807 return (error); 6808 } 6809 6810 if (ap->a_size != NULL) { 6811 error = VOP_GETATTR(vp, &va, ap->a_cred); 6812 if (error == 0) 6813 *ap->a_size = (size_t)va.va_size; 6814 } else if (ap->a_uio != NULL) 6815 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6816 6817 VOP_UNLOCK(vp, 0); 6818 vn_close(vp, flags, ap->a_cred, td); 6819 ZFS_EXIT(zfsvfs); 6820 6821 return (error); 6822} 6823 6824/* 6825 * Vnode operation to remove a named attribute. 6826 */ 6827int 6828zfs_deleteextattr(struct vop_deleteextattr_args *ap) 6829/* 6830vop_deleteextattr { 6831 IN struct vnode *a_vp; 6832 IN int a_attrnamespace; 6833 IN const char *a_name; 6834 IN struct ucred *a_cred; 6835 IN struct thread *a_td; 6836}; 6837*/ 6838{ 6839 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6840 struct thread *td = ap->a_td; 6841 struct nameidata nd; 6842 char attrname[255]; 6843 struct vattr va; 6844 vnode_t *xvp = NULL, *vp; 6845 int error, flags; 6846 6847 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6848 ap->a_cred, ap->a_td, VWRITE); 6849 if (error != 0) 6850 return (error); 6851 6852 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6853 sizeof(attrname)); 6854 if (error != 0) 6855 return (error); 6856 6857 ZFS_ENTER(zfsvfs); 6858 6859 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6860 LOOKUP_XATTR); 6861 if (error != 0) { 6862 ZFS_EXIT(zfsvfs); 6863 return (error); 6864 } 6865 6866 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 6867 UIO_SYSSPACE, attrname, xvp, td); 6868 error = namei(&nd); 6869 vp = nd.ni_vp; 6870 if (error != 0) { 6871 ZFS_EXIT(zfsvfs); 6872 NDFREE(&nd, NDF_ONLY_PNBUF); 6873 if (error == ENOENT) 6874 error = ENOATTR; 6875 return (error); 6876 } 6877 6878 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 6879 NDFREE(&nd, NDF_ONLY_PNBUF); 6880 6881 vput(nd.ni_dvp); 6882 if (vp == nd.ni_dvp) 6883 vrele(vp); 6884 else 6885 vput(vp); 6886 ZFS_EXIT(zfsvfs); 6887 6888 return (error); 6889} 6890 6891/* 6892 * Vnode operation to set a named attribute. 6893 */ 6894static int 6895zfs_setextattr(struct vop_setextattr_args *ap) 6896/* 6897vop_setextattr { 6898 IN struct vnode *a_vp; 6899 IN int a_attrnamespace; 6900 IN const char *a_name; 6901 INOUT struct uio *a_uio; 6902 IN struct ucred *a_cred; 6903 IN struct thread *a_td; 6904}; 6905*/ 6906{ 6907 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6908 struct thread *td = ap->a_td; 6909 struct nameidata nd; 6910 char attrname[255]; 6911 struct vattr va; 6912 vnode_t *xvp = NULL, *vp; 6913 int error, flags; 6914 6915 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6916 ap->a_cred, ap->a_td, VWRITE); 6917 if (error != 0) 6918 return (error); 6919 6920 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6921 sizeof(attrname)); 6922 if (error != 0) 6923 return (error); 6924 6925 ZFS_ENTER(zfsvfs); 6926 6927 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6928 LOOKUP_XATTR | CREATE_XATTR_DIR); 6929 if (error != 0) { 6930 ZFS_EXIT(zfsvfs); 6931 return (error); 6932 } 6933 6934 flags = FFLAGS(O_WRONLY | O_CREAT); 6935 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6936 xvp, td); 6937 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL); 6938 vp = nd.ni_vp; 6939 NDFREE(&nd, NDF_ONLY_PNBUF); 6940 if (error != 0) { 6941 ZFS_EXIT(zfsvfs); 6942 return (error); 6943 } 6944 6945 VATTR_NULL(&va); 6946 va.va_size = 0; 6947 error = VOP_SETATTR(vp, &va, ap->a_cred); 6948 if (error == 0) 6949 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6950 6951 VOP_UNLOCK(vp, 0); 6952 vn_close(vp, flags, ap->a_cred, td); 6953 ZFS_EXIT(zfsvfs); 6954 6955 return (error); 6956} 6957 6958/* 6959 * Vnode operation to retrieve extended attributes on a vnode. 6960 */ 6961static int 6962zfs_listextattr(struct vop_listextattr_args *ap) 6963/* 6964vop_listextattr { 6965 IN struct vnode *a_vp; 6966 IN int a_attrnamespace; 6967 INOUT struct uio *a_uio; 6968 OUT size_t *a_size; 6969 IN struct ucred *a_cred; 6970 IN struct thread *a_td; 6971}; 6972*/ 6973{ 6974 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6975 struct thread *td = ap->a_td; 6976 struct nameidata nd; 6977 char attrprefix[16]; 6978 u_char dirbuf[sizeof(struct dirent)]; 6979 struct dirent *dp; 6980 struct iovec aiov; 6981 struct uio auio, *uio = ap->a_uio; 6982 size_t *sizep = ap->a_size; 6983 size_t plen; 6984 vnode_t *xvp = NULL, *vp; 6985 int done, error, eof, pos; 6986 6987 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6988 ap->a_cred, ap->a_td, VREAD); 6989 if (error != 0) 6990 return (error); 6991 6992 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 6993 sizeof(attrprefix)); 6994 if (error != 0) 6995 return (error); 6996 plen = strlen(attrprefix); 6997 6998 ZFS_ENTER(zfsvfs); 6999 7000 if (sizep != NULL) 7001 *sizep = 0; 7002 7003 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 7004 LOOKUP_XATTR); 7005 if (error != 0) { 7006 ZFS_EXIT(zfsvfs); 7007 /* 7008 * ENOATTR means that the EA directory does not yet exist, 7009 * i.e. there are no extended attributes there. 7010 */ 7011 if (error == ENOATTR) 7012 error = 0; 7013 return (error); 7014 } 7015 7016 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 7017 UIO_SYSSPACE, ".", xvp, td); 7018 error = namei(&nd); 7019 vp = nd.ni_vp; 7020 NDFREE(&nd, NDF_ONLY_PNBUF); 7021 if (error != 0) { 7022 ZFS_EXIT(zfsvfs); 7023 return (error); 7024 } 7025 7026 auio.uio_iov = &aiov; 7027 auio.uio_iovcnt = 1; 7028 auio.uio_segflg = UIO_SYSSPACE; 7029 auio.uio_td = td; 7030 auio.uio_rw = UIO_READ; 7031 auio.uio_offset = 0; 7032 7033 do { 7034 u_char nlen; 7035 7036 aiov.iov_base = (void *)dirbuf; 7037 aiov.iov_len = sizeof(dirbuf); 7038 auio.uio_resid = sizeof(dirbuf); 7039 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 7040 done = sizeof(dirbuf) - auio.uio_resid; 7041 if (error != 0) 7042 break; 7043 for (pos = 0; pos < done;) { 7044 dp = (struct dirent *)(dirbuf + pos); 7045 pos += dp->d_reclen; 7046 /* 7047 * XXX: Temporarily we also accept DT_UNKNOWN, as this 7048 * is what we get when attribute was created on Solaris. 7049 */ 7050 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 7051 continue; 7052 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0) 7053 continue; 7054 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 7055 continue; 7056 nlen = dp->d_namlen - plen; 7057 if (sizep != NULL) 7058 *sizep += 1 + nlen; 7059 else if (uio != NULL) { 7060 /* 7061 * Format of extattr name entry is one byte for 7062 * length and the rest for name. 7063 */ 7064 error = uiomove(&nlen, 1, uio->uio_rw, uio); 7065 if (error == 0) { 7066 error = uiomove(dp->d_name + plen, nlen, 7067 uio->uio_rw, uio); 7068 } 7069 if (error != 0) 7070 break; 7071 } 7072 } 7073 } while (!eof && error == 0); 7074 7075 vput(vp); 7076 ZFS_EXIT(zfsvfs); 7077 7078 return (error); 7079} 7080 7081int 7082zfs_freebsd_getacl(ap) 7083 struct vop_getacl_args /* { 7084 struct vnode *vp; 7085 acl_type_t type; 7086 struct acl *aclp; 7087 struct ucred *cred; 7088 struct thread *td; 7089 } */ *ap; 7090{ 7091 int error; 7092 vsecattr_t vsecattr; 7093 7094 if (ap->a_type != ACL_TYPE_NFS4) 7095 return (EINVAL); 7096 7097 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 7098 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL)) 7099 return (error); 7100 7101 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt); 7102 if (vsecattr.vsa_aclentp != NULL) 7103 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 7104 7105 return (error); 7106} 7107 7108int 7109zfs_freebsd_setacl(ap) 7110 struct vop_setacl_args /* { 7111 struct vnode *vp; 7112 acl_type_t type; 7113 struct acl *aclp; 7114 struct ucred *cred; 7115 struct thread *td; 7116 } */ *ap; 7117{ 7118 int error; 7119 vsecattr_t vsecattr; 7120 int aclbsize; /* size of acl list in bytes */ 7121 aclent_t *aaclp; 7122 7123 if (ap->a_type != ACL_TYPE_NFS4) 7124 return (EINVAL); 7125 7126 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 7127 return (EINVAL); 7128 7129 /* 7130 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 7131 * splitting every entry into two and appending "canonical six" 7132 * entries at the end. Don't allow for setting an ACL that would 7133 * cause chmod(2) to run out of ACL entries. 7134 */ 7135 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 7136 return (ENOSPC); 7137 7138 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 7139 if (error != 0) 7140 return (error); 7141 7142 vsecattr.vsa_mask = VSA_ACE; 7143 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t); 7144 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 7145 aaclp = vsecattr.vsa_aclentp; 7146 vsecattr.vsa_aclentsz = aclbsize; 7147 7148 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 7149 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL); 7150 kmem_free(aaclp, aclbsize); 7151 7152 return (error); 7153} 7154 7155int 7156zfs_freebsd_aclcheck(ap) 7157 struct vop_aclcheck_args /* { 7158 struct vnode *vp; 7159 acl_type_t type; 7160 struct acl *aclp; 7161 struct ucred *cred; 7162 struct thread *td; 7163 } */ *ap; 7164{ 7165 7166 return (EOPNOTSUPP); 7167} 7168 7169struct vop_vector zfs_vnodeops; 7170struct vop_vector zfs_fifoops; 7171struct vop_vector zfs_shareops; 7172 7173struct vop_vector zfs_vnodeops = { 7174 .vop_default = &default_vnodeops, 7175 .vop_inactive = zfs_freebsd_inactive, 7176 .vop_reclaim = zfs_freebsd_reclaim, 7177 .vop_access = zfs_freebsd_access, 7178#ifdef FREEBSD_NAMECACHE 7179 .vop_lookup = vfs_cache_lookup, 7180 .vop_cachedlookup = zfs_freebsd_lookup, 7181#else 7182 .vop_lookup = zfs_freebsd_lookup, 7183#endif 7184 .vop_getattr = zfs_freebsd_getattr, 7185 .vop_setattr = zfs_freebsd_setattr, 7186 .vop_create = zfs_freebsd_create, 7187 .vop_mknod = zfs_freebsd_create, 7188 .vop_mkdir = zfs_freebsd_mkdir, 7189 .vop_readdir = zfs_freebsd_readdir, 7190 .vop_fsync = zfs_freebsd_fsync, 7191 .vop_open = zfs_freebsd_open, 7192 .vop_close = zfs_freebsd_close, 7193 .vop_rmdir = zfs_freebsd_rmdir, 7194 .vop_ioctl = zfs_freebsd_ioctl, 7195 .vop_link = zfs_freebsd_link, 7196 .vop_symlink = zfs_freebsd_symlink, 7197 .vop_readlink = zfs_freebsd_readlink, 7198 .vop_read = zfs_freebsd_read, 7199 .vop_write = zfs_freebsd_write, 7200 .vop_remove = zfs_freebsd_remove, 7201 .vop_rename = zfs_freebsd_rename, 7202 .vop_pathconf = zfs_freebsd_pathconf, 7203 .vop_bmap = zfs_freebsd_bmap, 7204 .vop_fid = zfs_freebsd_fid, 7205 .vop_getextattr = zfs_getextattr, 7206 .vop_deleteextattr = zfs_deleteextattr, 7207 .vop_setextattr = zfs_setextattr, 7208 .vop_listextattr = zfs_listextattr, 7209 .vop_getacl = zfs_freebsd_getacl, 7210 .vop_setacl = zfs_freebsd_setacl, 7211 .vop_aclcheck = zfs_freebsd_aclcheck, 7212 .vop_getpages = zfs_freebsd_getpages, 7213 .vop_putpages = zfs_freebsd_putpages, 7214}; 7215 7216struct vop_vector zfs_fifoops = { 7217 .vop_default = &fifo_specops, 7218 .vop_fsync = zfs_freebsd_fsync, 7219 .vop_access = zfs_freebsd_access, 7220 .vop_getattr = zfs_freebsd_getattr, 7221 .vop_inactive = zfs_freebsd_inactive, 7222 .vop_read = VOP_PANIC, 7223 .vop_reclaim = zfs_freebsd_reclaim, 7224 .vop_setattr = zfs_freebsd_setattr, 7225 .vop_write = VOP_PANIC, 7226 .vop_pathconf = zfs_freebsd_fifo_pathconf, 7227 .vop_fid = zfs_freebsd_fid, 7228 .vop_getacl = zfs_freebsd_getacl, 7229 .vop_setacl = zfs_freebsd_setacl, 7230 .vop_aclcheck = zfs_freebsd_aclcheck, 7231}; 7232 7233/* 7234 * special share hidden files vnode operations template 7235 */ 7236struct vop_vector zfs_shareops = { 7237 .vop_default = &default_vnodeops, 7238 .vop_access = zfs_freebsd_access, 7239 .vop_inactive = zfs_freebsd_inactive, 7240 .vop_reclaim = zfs_freebsd_reclaim, 7241 .vop_fid = zfs_freebsd_fid, 7242 .vop_pathconf = zfs_freebsd_pathconf, 7243}; 7244