zfs_vnops.c revision 304121
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 79/* 80 * Programming rules. 81 * 82 * Each vnode op performs some logical unit of work. To do this, the ZPL must 83 * properly lock its in-core state, create a DMU transaction, do the work, 84 * record this work in the intent log (ZIL), commit the DMU transaction, 85 * and wait for the intent log to commit if it is a synchronous operation. 86 * Moreover, the vnode ops must work in both normal and log replay context. 87 * The ordering of events is important to avoid deadlocks and references 88 * to freed memory. The example below illustrates the following Big Rules: 89 * 90 * (1) A check must be made in each zfs thread for a mounted file system. 91 * This is done avoiding races using ZFS_ENTER(zfsvfs). 92 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes 93 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros 94 * can return EIO from the calling function. 95 * 96 * (2) VN_RELE() should always be the last thing except for zil_commit() 97 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 98 * First, if it's the last reference, the vnode/znode 99 * can be freed, so the zp may point to freed memory. Second, the last 100 * reference will call zfs_zinactive(), which may induce a lot of work -- 101 * pushing cached pages (which acquires range locks) and syncing out 102 * cached atime changes. Third, zfs_zinactive() may require a new tx, 103 * which could deadlock the system if you were already holding one. 104 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). 105 * 106 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 107 * as they can span dmu_tx_assign() calls. 108 * 109 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to 110 * dmu_tx_assign(). This is critical because we don't want to block 111 * while holding locks. 112 * 113 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This 114 * reduces lock contention and CPU usage when we must wait (note that if 115 * throughput is constrained by the storage, nearly every transaction 116 * must wait). 117 * 118 * Note, in particular, that if a lock is sometimes acquired before 119 * the tx assigns, and sometimes after (e.g. z_lock), then failing 120 * to use a non-blocking assign can deadlock the system. The scenario: 121 * 122 * Thread A has grabbed a lock before calling dmu_tx_assign(). 123 * Thread B is in an already-assigned tx, and blocks for this lock. 124 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 125 * forever, because the previous txg can't quiesce until B's tx commits. 126 * 127 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 128 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent 129 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT, 130 * to indicate that this operation has already called dmu_tx_wait(). 131 * This will ensure that we don't retry forever, waiting a short bit 132 * each time. 133 * 134 * (5) If the operation succeeded, generate the intent log entry for it 135 * before dropping locks. This ensures that the ordering of events 136 * in the intent log matches the order in which they actually occurred. 137 * During ZIL replay the zfs_log_* functions will update the sequence 138 * number to indicate the zil transaction has replayed. 139 * 140 * (6) At the end of each vnode op, the DMU tx must always commit, 141 * regardless of whether there were any errors. 142 * 143 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 144 * to ensure that synchronous semantics are provided when necessary. 145 * 146 * In general, this is how things should be ordered in each vnode op: 147 * 148 * ZFS_ENTER(zfsvfs); // exit if unmounted 149 * top: 150 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 151 * rw_enter(...); // grab any other locks you need 152 * tx = dmu_tx_create(...); // get DMU tx 153 * dmu_tx_hold_*(); // hold each object you might modify 154 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 155 * if (error) { 156 * rw_exit(...); // drop locks 157 * zfs_dirent_unlock(dl); // unlock directory entry 158 * VN_RELE(...); // release held vnodes 159 * if (error == ERESTART) { 160 * waited = B_TRUE; 161 * dmu_tx_wait(tx); 162 * dmu_tx_abort(tx); 163 * goto top; 164 * } 165 * dmu_tx_abort(tx); // abort DMU tx 166 * ZFS_EXIT(zfsvfs); // finished in zfs 167 * return (error); // really out of space 168 * } 169 * error = do_real_work(); // do whatever this VOP does 170 * if (error == 0) 171 * zfs_log_*(...); // on success, make ZIL entry 172 * dmu_tx_commit(tx); // commit DMU tx -- error or not 173 * rw_exit(...); // drop locks 174 * zfs_dirent_unlock(dl); // unlock directory entry 175 * VN_RELE(...); // release held vnodes 176 * zil_commit(zilog, foid); // synchronous when necessary 177 * ZFS_EXIT(zfsvfs); // finished in zfs 178 * return (error); // done, report error 179 */ 180 181/* ARGSUSED */ 182static int 183zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 184{ 185 znode_t *zp = VTOZ(*vpp); 186 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 187 188 ZFS_ENTER(zfsvfs); 189 ZFS_VERIFY_ZP(zp); 190 191 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 192 ((flag & FAPPEND) == 0)) { 193 ZFS_EXIT(zfsvfs); 194 return (SET_ERROR(EPERM)); 195 } 196 197 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 198 ZTOV(zp)->v_type == VREG && 199 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) { 200 if (fs_vscan(*vpp, cr, 0) != 0) { 201 ZFS_EXIT(zfsvfs); 202 return (SET_ERROR(EACCES)); 203 } 204 } 205 206 /* Keep a count of the synchronous opens in the znode */ 207 if (flag & (FSYNC | FDSYNC)) 208 atomic_inc_32(&zp->z_sync_cnt); 209 210 ZFS_EXIT(zfsvfs); 211 return (0); 212} 213 214/* ARGSUSED */ 215static int 216zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 217 caller_context_t *ct) 218{ 219 znode_t *zp = VTOZ(vp); 220 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 221 222 /* 223 * Clean up any locks held by this process on the vp. 224 */ 225 cleanlocks(vp, ddi_get_pid(), 0); 226 cleanshares(vp, ddi_get_pid()); 227 228 ZFS_ENTER(zfsvfs); 229 ZFS_VERIFY_ZP(zp); 230 231 /* Decrement the synchronous opens in the znode */ 232 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 233 atomic_dec_32(&zp->z_sync_cnt); 234 235 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 236 ZTOV(zp)->v_type == VREG && 237 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) 238 VERIFY(fs_vscan(vp, cr, 1) == 0); 239 240 ZFS_EXIT(zfsvfs); 241 return (0); 242} 243 244/* 245 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 246 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 247 */ 248static int 249zfs_holey(vnode_t *vp, u_long cmd, offset_t *off) 250{ 251 znode_t *zp = VTOZ(vp); 252 uint64_t noff = (uint64_t)*off; /* new offset */ 253 uint64_t file_sz; 254 int error; 255 boolean_t hole; 256 257 file_sz = zp->z_size; 258 if (noff >= file_sz) { 259 return (SET_ERROR(ENXIO)); 260 } 261 262 if (cmd == _FIO_SEEK_HOLE) 263 hole = B_TRUE; 264 else 265 hole = B_FALSE; 266 267 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 268 269 if (error == ESRCH) 270 return (SET_ERROR(ENXIO)); 271 272 /* 273 * We could find a hole that begins after the logical end-of-file, 274 * because dmu_offset_next() only works on whole blocks. If the 275 * EOF falls mid-block, then indicate that the "virtual hole" 276 * at the end of the file begins at the logical EOF, rather than 277 * at the end of the last block. 278 */ 279 if (noff > file_sz) { 280 ASSERT(hole); 281 noff = file_sz; 282 } 283 284 if (noff < *off) 285 return (error); 286 *off = noff; 287 return (error); 288} 289 290/* ARGSUSED */ 291static int 292zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred, 293 int *rvalp, caller_context_t *ct) 294{ 295 offset_t off; 296 offset_t ndata; 297 dmu_object_info_t doi; 298 int error; 299 zfsvfs_t *zfsvfs; 300 znode_t *zp; 301 302 switch (com) { 303 case _FIOFFS: 304 { 305 return (0); 306 307 /* 308 * The following two ioctls are used by bfu. Faking out, 309 * necessary to avoid bfu errors. 310 */ 311 } 312 case _FIOGDIO: 313 case _FIOSDIO: 314 { 315 return (0); 316 } 317 318 case _FIO_SEEK_DATA: 319 case _FIO_SEEK_HOLE: 320 { 321#ifdef illumos 322 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 323 return (SET_ERROR(EFAULT)); 324#else 325 off = *(offset_t *)data; 326#endif 327 zp = VTOZ(vp); 328 zfsvfs = zp->z_zfsvfs; 329 ZFS_ENTER(zfsvfs); 330 ZFS_VERIFY_ZP(zp); 331 332 /* offset parameter is in/out */ 333 error = zfs_holey(vp, com, &off); 334 ZFS_EXIT(zfsvfs); 335 if (error) 336 return (error); 337#ifdef illumos 338 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 339 return (SET_ERROR(EFAULT)); 340#else 341 *(offset_t *)data = off; 342#endif 343 return (0); 344 } 345#ifdef illumos 346 case _FIO_COUNT_FILLED: 347 { 348 /* 349 * _FIO_COUNT_FILLED adds a new ioctl command which 350 * exposes the number of filled blocks in a 351 * ZFS object. 352 */ 353 zp = VTOZ(vp); 354 zfsvfs = zp->z_zfsvfs; 355 ZFS_ENTER(zfsvfs); 356 ZFS_VERIFY_ZP(zp); 357 358 /* 359 * Wait for all dirty blocks for this object 360 * to get synced out to disk, and the DMU info 361 * updated. 362 */ 363 error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id); 364 if (error) { 365 ZFS_EXIT(zfsvfs); 366 return (error); 367 } 368 369 /* 370 * Retrieve fill count from DMU object. 371 */ 372 error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi); 373 if (error) { 374 ZFS_EXIT(zfsvfs); 375 return (error); 376 } 377 378 ndata = doi.doi_fill_count; 379 380 ZFS_EXIT(zfsvfs); 381 if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag)) 382 return (SET_ERROR(EFAULT)); 383 return (0); 384 } 385#endif 386 } 387 return (SET_ERROR(ENOTTY)); 388} 389 390static vm_page_t 391page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) 392{ 393 vm_object_t obj; 394 vm_page_t pp; 395 int64_t end; 396 397 /* 398 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE 399 * aligned boundaries, if the range is not aligned. As a result a 400 * DEV_BSIZE subrange with partially dirty data may get marked as clean. 401 * It may happen that all DEV_BSIZE subranges are marked clean and thus 402 * the whole page would be considred clean despite have some dirty data. 403 * For this reason we should shrink the range to DEV_BSIZE aligned 404 * boundaries before calling vm_page_clear_dirty. 405 */ 406 end = rounddown2(off + nbytes, DEV_BSIZE); 407 off = roundup2(off, DEV_BSIZE); 408 nbytes = end - off; 409 410 obj = vp->v_object; 411 zfs_vmobject_assert_wlocked(obj); 412 413 for (;;) { 414 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 415 pp->valid) { 416 if (vm_page_xbusied(pp)) { 417 /* 418 * Reference the page before unlocking and 419 * sleeping so that the page daemon is less 420 * likely to reclaim it. 421 */ 422 vm_page_reference(pp); 423 vm_page_lock(pp); 424 zfs_vmobject_wunlock(obj); 425 vm_page_busy_sleep(pp, "zfsmwb"); 426 zfs_vmobject_wlock(obj); 427 continue; 428 } 429 vm_page_sbusy(pp); 430 } else if (pp == NULL) { 431 pp = vm_page_alloc(obj, OFF_TO_IDX(start), 432 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED | 433 VM_ALLOC_SBUSY); 434 } else { 435 ASSERT(pp != NULL && !pp->valid); 436 pp = NULL; 437 } 438 439 if (pp != NULL) { 440 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 441 vm_object_pip_add(obj, 1); 442 pmap_remove_write(pp); 443 if (nbytes != 0) 444 vm_page_clear_dirty(pp, off, nbytes); 445 } 446 break; 447 } 448 return (pp); 449} 450 451static void 452page_unbusy(vm_page_t pp) 453{ 454 455 vm_page_sunbusy(pp); 456 vm_object_pip_subtract(pp->object, 1); 457} 458 459static vm_page_t 460page_hold(vnode_t *vp, int64_t start) 461{ 462 vm_object_t obj; 463 vm_page_t pp; 464 465 obj = vp->v_object; 466 zfs_vmobject_assert_wlocked(obj); 467 468 for (;;) { 469 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 470 pp->valid) { 471 if (vm_page_xbusied(pp)) { 472 /* 473 * Reference the page before unlocking and 474 * sleeping so that the page daemon is less 475 * likely to reclaim it. 476 */ 477 vm_page_reference(pp); 478 vm_page_lock(pp); 479 zfs_vmobject_wunlock(obj); 480 vm_page_busy_sleep(pp, "zfsmwb"); 481 zfs_vmobject_wlock(obj); 482 continue; 483 } 484 485 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 486 vm_page_lock(pp); 487 vm_page_hold(pp); 488 vm_page_unlock(pp); 489 490 } else 491 pp = NULL; 492 break; 493 } 494 return (pp); 495} 496 497static void 498page_unhold(vm_page_t pp) 499{ 500 501 vm_page_lock(pp); 502 vm_page_unhold(pp); 503 vm_page_unlock(pp); 504} 505 506/* 507 * When a file is memory mapped, we must keep the IO data synchronized 508 * between the DMU cache and the memory mapped pages. What this means: 509 * 510 * On Write: If we find a memory mapped page, we write to *both* 511 * the page and the dmu buffer. 512 */ 513static void 514update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid, 515 int segflg, dmu_tx_t *tx) 516{ 517 vm_object_t obj; 518 struct sf_buf *sf; 519 caddr_t va; 520 int off; 521 522 ASSERT(segflg != UIO_NOCOPY); 523 ASSERT(vp->v_mount != NULL); 524 obj = vp->v_object; 525 ASSERT(obj != NULL); 526 527 off = start & PAGEOFFSET; 528 zfs_vmobject_wlock(obj); 529 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 530 vm_page_t pp; 531 int nbytes = imin(PAGESIZE - off, len); 532 533 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) { 534 zfs_vmobject_wunlock(obj); 535 536 va = zfs_map_page(pp, &sf); 537 (void) dmu_read(os, oid, start+off, nbytes, 538 va+off, DMU_READ_PREFETCH);; 539 zfs_unmap_page(sf); 540 541 zfs_vmobject_wlock(obj); 542 page_unbusy(pp); 543 } 544 len -= nbytes; 545 off = 0; 546 } 547 vm_object_pip_wakeupn(obj, 0); 548 zfs_vmobject_wunlock(obj); 549} 550 551/* 552 * Read with UIO_NOCOPY flag means that sendfile(2) requests 553 * ZFS to populate a range of page cache pages with data. 554 * 555 * NOTE: this function could be optimized to pre-allocate 556 * all pages in advance, drain exclusive busy on all of them, 557 * map them into contiguous KVA region and populate them 558 * in one single dmu_read() call. 559 */ 560static int 561mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio) 562{ 563 znode_t *zp = VTOZ(vp); 564 objset_t *os = zp->z_zfsvfs->z_os; 565 struct sf_buf *sf; 566 vm_object_t obj; 567 vm_page_t pp; 568 int64_t start; 569 caddr_t va; 570 int len = nbytes; 571 int off; 572 int error = 0; 573 574 ASSERT(uio->uio_segflg == UIO_NOCOPY); 575 ASSERT(vp->v_mount != NULL); 576 obj = vp->v_object; 577 ASSERT(obj != NULL); 578 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0); 579 580 zfs_vmobject_wlock(obj); 581 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) { 582 int bytes = MIN(PAGESIZE, len); 583 584 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY | 585 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); 586 if (pp->valid == 0) { 587 zfs_vmobject_wunlock(obj); 588 va = zfs_map_page(pp, &sf); 589 error = dmu_read(os, zp->z_id, start, bytes, va, 590 DMU_READ_PREFETCH); 591 if (bytes != PAGESIZE && error == 0) 592 bzero(va + bytes, PAGESIZE - bytes); 593 zfs_unmap_page(sf); 594 zfs_vmobject_wlock(obj); 595 vm_page_sunbusy(pp); 596 vm_page_lock(pp); 597 if (error) { 598 if (pp->wire_count == 0 && pp->valid == 0 && 599 !vm_page_busied(pp)) 600 vm_page_free(pp); 601 } else { 602 pp->valid = VM_PAGE_BITS_ALL; 603 vm_page_activate(pp); 604 } 605 vm_page_unlock(pp); 606 } else { 607 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 608 vm_page_sunbusy(pp); 609 } 610 if (error) 611 break; 612 uio->uio_resid -= bytes; 613 uio->uio_offset += bytes; 614 len -= bytes; 615 } 616 zfs_vmobject_wunlock(obj); 617 return (error); 618} 619 620/* 621 * When a file is memory mapped, we must keep the IO data synchronized 622 * between the DMU cache and the memory mapped pages. What this means: 623 * 624 * On Read: We "read" preferentially from memory mapped pages, 625 * else we default from the dmu buffer. 626 * 627 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 628 * the file is memory mapped. 629 */ 630static int 631mappedread(vnode_t *vp, int nbytes, uio_t *uio) 632{ 633 znode_t *zp = VTOZ(vp); 634 vm_object_t obj; 635 int64_t start; 636 caddr_t va; 637 int len = nbytes; 638 int off; 639 int error = 0; 640 641 ASSERT(vp->v_mount != NULL); 642 obj = vp->v_object; 643 ASSERT(obj != NULL); 644 645 start = uio->uio_loffset; 646 off = start & PAGEOFFSET; 647 zfs_vmobject_wlock(obj); 648 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 649 vm_page_t pp; 650 uint64_t bytes = MIN(PAGESIZE - off, len); 651 652 if (pp = page_hold(vp, start)) { 653 struct sf_buf *sf; 654 caddr_t va; 655 656 zfs_vmobject_wunlock(obj); 657 va = zfs_map_page(pp, &sf); 658#ifdef illumos 659 error = uiomove(va + off, bytes, UIO_READ, uio); 660#else 661 error = vn_io_fault_uiomove(va + off, bytes, uio); 662#endif 663 zfs_unmap_page(sf); 664 zfs_vmobject_wlock(obj); 665 page_unhold(pp); 666 } else { 667 zfs_vmobject_wunlock(obj); 668 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 669 uio, bytes); 670 zfs_vmobject_wlock(obj); 671 } 672 len -= bytes; 673 off = 0; 674 if (error) 675 break; 676 } 677 zfs_vmobject_wunlock(obj); 678 return (error); 679} 680 681offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 682 683/* 684 * Read bytes from specified file into supplied buffer. 685 * 686 * IN: vp - vnode of file to be read from. 687 * uio - structure supplying read location, range info, 688 * and return buffer. 689 * ioflag - SYNC flags; used to provide FRSYNC semantics. 690 * cr - credentials of caller. 691 * ct - caller context 692 * 693 * OUT: uio - updated offset and range, buffer filled. 694 * 695 * RETURN: 0 on success, error code on failure. 696 * 697 * Side Effects: 698 * vp - atime updated if byte count > 0 699 */ 700/* ARGSUSED */ 701static int 702zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 703{ 704 znode_t *zp = VTOZ(vp); 705 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 706 ssize_t n, nbytes; 707 int error = 0; 708 rl_t *rl; 709 xuio_t *xuio = NULL; 710 711 ZFS_ENTER(zfsvfs); 712 ZFS_VERIFY_ZP(zp); 713 714 if (zp->z_pflags & ZFS_AV_QUARANTINED) { 715 ZFS_EXIT(zfsvfs); 716 return (SET_ERROR(EACCES)); 717 } 718 719 /* 720 * Validate file offset 721 */ 722 if (uio->uio_loffset < (offset_t)0) { 723 ZFS_EXIT(zfsvfs); 724 return (SET_ERROR(EINVAL)); 725 } 726 727 /* 728 * Fasttrack empty reads 729 */ 730 if (uio->uio_resid == 0) { 731 ZFS_EXIT(zfsvfs); 732 return (0); 733 } 734 735 /* 736 * Check for mandatory locks 737 */ 738 if (MANDMODE(zp->z_mode)) { 739 if (error = chklock(vp, FREAD, 740 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 741 ZFS_EXIT(zfsvfs); 742 return (error); 743 } 744 } 745 746 /* 747 * If we're in FRSYNC mode, sync out this znode before reading it. 748 */ 749 if (zfsvfs->z_log && 750 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)) 751 zil_commit(zfsvfs->z_log, zp->z_id); 752 753 /* 754 * Lock the range against changes. 755 */ 756 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 757 758 /* 759 * If we are reading past end-of-file we can skip 760 * to the end; but we might still need to set atime. 761 */ 762 if (uio->uio_loffset >= zp->z_size) { 763 error = 0; 764 goto out; 765 } 766 767 ASSERT(uio->uio_loffset < zp->z_size); 768 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset); 769 770#ifdef illumos 771 if ((uio->uio_extflg == UIO_XUIO) && 772 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) { 773 int nblk; 774 int blksz = zp->z_blksz; 775 uint64_t offset = uio->uio_loffset; 776 777 xuio = (xuio_t *)uio; 778 if ((ISP2(blksz))) { 779 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset, 780 blksz)) / blksz; 781 } else { 782 ASSERT(offset + n <= blksz); 783 nblk = 1; 784 } 785 (void) dmu_xuio_init(xuio, nblk); 786 787 if (vn_has_cached_data(vp)) { 788 /* 789 * For simplicity, we always allocate a full buffer 790 * even if we only expect to read a portion of a block. 791 */ 792 while (--nblk >= 0) { 793 (void) dmu_xuio_add(xuio, 794 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 795 blksz), 0, blksz); 796 } 797 } 798 } 799#endif /* illumos */ 800 801 while (n > 0) { 802 nbytes = MIN(n, zfs_read_chunk_size - 803 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 804 805#ifdef __FreeBSD__ 806 if (uio->uio_segflg == UIO_NOCOPY) 807 error = mappedread_sf(vp, nbytes, uio); 808 else 809#endif /* __FreeBSD__ */ 810 if (vn_has_cached_data(vp)) { 811 error = mappedread(vp, nbytes, uio); 812 } else { 813 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 814 uio, nbytes); 815 } 816 if (error) { 817 /* convert checksum errors into IO errors */ 818 if (error == ECKSUM) 819 error = SET_ERROR(EIO); 820 break; 821 } 822 823 n -= nbytes; 824 } 825out: 826 zfs_range_unlock(rl); 827 828 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 829 ZFS_EXIT(zfsvfs); 830 return (error); 831} 832 833/* 834 * Write the bytes to a file. 835 * 836 * IN: vp - vnode of file to be written to. 837 * uio - structure supplying write location, range info, 838 * and data buffer. 839 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is 840 * set if in append mode. 841 * cr - credentials of caller. 842 * ct - caller context (NFS/CIFS fem monitor only) 843 * 844 * OUT: uio - updated offset and range. 845 * 846 * RETURN: 0 on success, error code on failure. 847 * 848 * Timestamps: 849 * vp - ctime|mtime updated if byte count > 0 850 */ 851 852/* ARGSUSED */ 853static int 854zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 855{ 856 znode_t *zp = VTOZ(vp); 857 rlim64_t limit = MAXOFFSET_T; 858 ssize_t start_resid = uio->uio_resid; 859 ssize_t tx_bytes; 860 uint64_t end_size; 861 dmu_tx_t *tx; 862 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 863 zilog_t *zilog; 864 offset_t woff; 865 ssize_t n, nbytes; 866 rl_t *rl; 867 int max_blksz = zfsvfs->z_max_blksz; 868 int error = 0; 869 arc_buf_t *abuf; 870 iovec_t *aiov = NULL; 871 xuio_t *xuio = NULL; 872 int i_iov = 0; 873 int iovcnt = uio->uio_iovcnt; 874 iovec_t *iovp = uio->uio_iov; 875 int write_eof; 876 int count = 0; 877 sa_bulk_attr_t bulk[4]; 878 uint64_t mtime[2], ctime[2]; 879 880 /* 881 * Fasttrack empty write 882 */ 883 n = start_resid; 884 if (n == 0) 885 return (0); 886 887 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 888 limit = MAXOFFSET_T; 889 890 ZFS_ENTER(zfsvfs); 891 ZFS_VERIFY_ZP(zp); 892 893 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 894 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 895 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 896 &zp->z_size, 8); 897 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 898 &zp->z_pflags, 8); 899 900 /* 901 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our 902 * callers might not be able to detect properly that we are read-only, 903 * so check it explicitly here. 904 */ 905 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 906 ZFS_EXIT(zfsvfs); 907 return (SET_ERROR(EROFS)); 908 } 909 910 /* 911 * If immutable or not appending then return EPERM 912 */ 913 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 914 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 915 (uio->uio_loffset < zp->z_size))) { 916 ZFS_EXIT(zfsvfs); 917 return (SET_ERROR(EPERM)); 918 } 919 920 zilog = zfsvfs->z_log; 921 922 /* 923 * Validate file offset 924 */ 925 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset; 926 if (woff < 0) { 927 ZFS_EXIT(zfsvfs); 928 return (SET_ERROR(EINVAL)); 929 } 930 931 /* 932 * Check for mandatory locks before calling zfs_range_lock() 933 * in order to prevent a deadlock with locks set via fcntl(). 934 */ 935 if (MANDMODE((mode_t)zp->z_mode) && 936 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 937 ZFS_EXIT(zfsvfs); 938 return (error); 939 } 940 941#ifdef illumos 942 /* 943 * Pre-fault the pages to ensure slow (eg NFS) pages 944 * don't hold up txg. 945 * Skip this if uio contains loaned arc_buf. 946 */ 947 if ((uio->uio_extflg == UIO_XUIO) && 948 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) 949 xuio = (xuio_t *)uio; 950 else 951 uio_prefaultpages(MIN(n, max_blksz), uio); 952#endif 953 954 /* 955 * If in append mode, set the io offset pointer to eof. 956 */ 957 if (ioflag & FAPPEND) { 958 /* 959 * Obtain an appending range lock to guarantee file append 960 * semantics. We reset the write offset once we have the lock. 961 */ 962 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 963 woff = rl->r_off; 964 if (rl->r_len == UINT64_MAX) { 965 /* 966 * We overlocked the file because this write will cause 967 * the file block size to increase. 968 * Note that zp_size cannot change with this lock held. 969 */ 970 woff = zp->z_size; 971 } 972 uio->uio_loffset = woff; 973 } else { 974 /* 975 * Note that if the file block size will change as a result of 976 * this write, then this range lock will lock the entire file 977 * so that we can re-write the block safely. 978 */ 979 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 980 } 981 982 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) { 983 zfs_range_unlock(rl); 984 ZFS_EXIT(zfsvfs); 985 return (EFBIG); 986 } 987 988 if (woff >= limit) { 989 zfs_range_unlock(rl); 990 ZFS_EXIT(zfsvfs); 991 return (SET_ERROR(EFBIG)); 992 } 993 994 if ((woff + n) > limit || woff > (limit - n)) 995 n = limit - woff; 996 997 /* Will this write extend the file length? */ 998 write_eof = (woff + n > zp->z_size); 999 1000 end_size = MAX(zp->z_size, woff + n); 1001 1002 /* 1003 * Write the file in reasonable size chunks. Each chunk is written 1004 * in a separate transaction; this keeps the intent log records small 1005 * and allows us to do more fine-grained space accounting. 1006 */ 1007 while (n > 0) { 1008 abuf = NULL; 1009 woff = uio->uio_loffset; 1010 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 1011 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 1012 if (abuf != NULL) 1013 dmu_return_arcbuf(abuf); 1014 error = SET_ERROR(EDQUOT); 1015 break; 1016 } 1017 1018 if (xuio && abuf == NULL) { 1019 ASSERT(i_iov < iovcnt); 1020 aiov = &iovp[i_iov]; 1021 abuf = dmu_xuio_arcbuf(xuio, i_iov); 1022 dmu_xuio_clear(xuio, i_iov); 1023 DTRACE_PROBE3(zfs_cp_write, int, i_iov, 1024 iovec_t *, aiov, arc_buf_t *, abuf); 1025 ASSERT((aiov->iov_base == abuf->b_data) || 1026 ((char *)aiov->iov_base - (char *)abuf->b_data + 1027 aiov->iov_len == arc_buf_size(abuf))); 1028 i_iov++; 1029 } else if (abuf == NULL && n >= max_blksz && 1030 woff >= zp->z_size && 1031 P2PHASE(woff, max_blksz) == 0 && 1032 zp->z_blksz == max_blksz) { 1033 /* 1034 * This write covers a full block. "Borrow" a buffer 1035 * from the dmu so that we can fill it before we enter 1036 * a transaction. This avoids the possibility of 1037 * holding up the transaction if the data copy hangs 1038 * up on a pagefault (e.g., from an NFS server mapping). 1039 */ 1040#ifdef illumos 1041 size_t cbytes; 1042#endif 1043 1044 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 1045 max_blksz); 1046 ASSERT(abuf != NULL); 1047 ASSERT(arc_buf_size(abuf) == max_blksz); 1048#ifdef illumos 1049 if (error = uiocopy(abuf->b_data, max_blksz, 1050 UIO_WRITE, uio, &cbytes)) { 1051 dmu_return_arcbuf(abuf); 1052 break; 1053 } 1054 ASSERT(cbytes == max_blksz); 1055#else 1056 ssize_t resid = uio->uio_resid; 1057 error = vn_io_fault_uiomove(abuf->b_data, max_blksz, uio); 1058 if (error != 0) { 1059 uio->uio_offset -= resid - uio->uio_resid; 1060 uio->uio_resid = resid; 1061 dmu_return_arcbuf(abuf); 1062 break; 1063 } 1064#endif 1065 } 1066 1067 /* 1068 * Start a transaction. 1069 */ 1070 tx = dmu_tx_create(zfsvfs->z_os); 1071 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1072 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 1073 zfs_sa_upgrade_txholds(tx, zp); 1074 error = dmu_tx_assign(tx, TXG_WAIT); 1075 if (error) { 1076 dmu_tx_abort(tx); 1077 if (abuf != NULL) 1078 dmu_return_arcbuf(abuf); 1079 break; 1080 } 1081 1082 /* 1083 * If zfs_range_lock() over-locked we grow the blocksize 1084 * and then reduce the lock range. This will only happen 1085 * on the first iteration since zfs_range_reduce() will 1086 * shrink down r_len to the appropriate size. 1087 */ 1088 if (rl->r_len == UINT64_MAX) { 1089 uint64_t new_blksz; 1090 1091 if (zp->z_blksz > max_blksz) { 1092 /* 1093 * File's blocksize is already larger than the 1094 * "recordsize" property. Only let it grow to 1095 * the next power of 2. 1096 */ 1097 ASSERT(!ISP2(zp->z_blksz)); 1098 new_blksz = MIN(end_size, 1099 1 << highbit64(zp->z_blksz)); 1100 } else { 1101 new_blksz = MIN(end_size, max_blksz); 1102 } 1103 zfs_grow_blocksize(zp, new_blksz, tx); 1104 zfs_range_reduce(rl, woff, n); 1105 } 1106 1107 /* 1108 * XXX - should we really limit each write to z_max_blksz? 1109 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 1110 */ 1111 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 1112 1113 if (woff + nbytes > zp->z_size) 1114 vnode_pager_setsize(vp, woff + nbytes); 1115 1116 if (abuf == NULL) { 1117 tx_bytes = uio->uio_resid; 1118 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), 1119 uio, nbytes, tx); 1120 tx_bytes -= uio->uio_resid; 1121 } else { 1122 tx_bytes = nbytes; 1123 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len); 1124 /* 1125 * If this is not a full block write, but we are 1126 * extending the file past EOF and this data starts 1127 * block-aligned, use assign_arcbuf(). Otherwise, 1128 * write via dmu_write(). 1129 */ 1130 if (tx_bytes < max_blksz && (!write_eof || 1131 aiov->iov_base != abuf->b_data)) { 1132 ASSERT(xuio); 1133 dmu_write(zfsvfs->z_os, zp->z_id, woff, 1134 aiov->iov_len, aiov->iov_base, tx); 1135 dmu_return_arcbuf(abuf); 1136 xuio_stat_wbuf_copied(); 1137 } else { 1138 ASSERT(xuio || tx_bytes == max_blksz); 1139 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), 1140 woff, abuf, tx); 1141 } 1142#ifdef illumos 1143 ASSERT(tx_bytes <= uio->uio_resid); 1144 uioskip(uio, tx_bytes); 1145#endif 1146 } 1147 if (tx_bytes && vn_has_cached_data(vp)) { 1148 update_pages(vp, woff, tx_bytes, zfsvfs->z_os, 1149 zp->z_id, uio->uio_segflg, tx); 1150 } 1151 1152 /* 1153 * If we made no progress, we're done. If we made even 1154 * partial progress, update the znode and ZIL accordingly. 1155 */ 1156 if (tx_bytes == 0) { 1157 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 1158 (void *)&zp->z_size, sizeof (uint64_t), tx); 1159 dmu_tx_commit(tx); 1160 ASSERT(error != 0); 1161 break; 1162 } 1163 1164 /* 1165 * Clear Set-UID/Set-GID bits on successful write if not 1166 * privileged and at least one of the excute bits is set. 1167 * 1168 * It would be nice to to this after all writes have 1169 * been done, but that would still expose the ISUID/ISGID 1170 * to another app after the partial write is committed. 1171 * 1172 * Note: we don't call zfs_fuid_map_id() here because 1173 * user 0 is not an ephemeral uid. 1174 */ 1175 mutex_enter(&zp->z_acl_lock); 1176 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | 1177 (S_IXUSR >> 6))) != 0 && 1178 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && 1179 secpolicy_vnode_setid_retain(vp, cr, 1180 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) { 1181 uint64_t newmode; 1182 zp->z_mode &= ~(S_ISUID | S_ISGID); 1183 newmode = zp->z_mode; 1184 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), 1185 (void *)&newmode, sizeof (uint64_t), tx); 1186 } 1187 mutex_exit(&zp->z_acl_lock); 1188 1189 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 1190 B_TRUE); 1191 1192 /* 1193 * Update the file size (zp_size) if it has changed; 1194 * account for possible concurrent updates. 1195 */ 1196 while ((end_size = zp->z_size) < uio->uio_loffset) { 1197 (void) atomic_cas_64(&zp->z_size, end_size, 1198 uio->uio_loffset); 1199#ifdef illumos 1200 ASSERT(error == 0); 1201#else 1202 ASSERT(error == 0 || error == EFAULT); 1203#endif 1204 } 1205 /* 1206 * If we are replaying and eof is non zero then force 1207 * the file size to the specified eof. Note, there's no 1208 * concurrency during replay. 1209 */ 1210 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) 1211 zp->z_size = zfsvfs->z_replay_eof; 1212 1213 if (error == 0) 1214 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1215 else 1216 (void) sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1217 1218 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 1219 dmu_tx_commit(tx); 1220 1221 if (error != 0) 1222 break; 1223 ASSERT(tx_bytes == nbytes); 1224 n -= nbytes; 1225 1226#ifdef illumos 1227 if (!xuio && n > 0) 1228 uio_prefaultpages(MIN(n, max_blksz), uio); 1229#endif 1230 } 1231 1232 zfs_range_unlock(rl); 1233 1234 /* 1235 * If we're in replay mode, or we made no progress, return error. 1236 * Otherwise, it's at least a partial write, so it's successful. 1237 */ 1238 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 1239 ZFS_EXIT(zfsvfs); 1240 return (error); 1241 } 1242 1243#ifdef __FreeBSD__ 1244 /* 1245 * EFAULT means that at least one page of the source buffer was not 1246 * available. VFS will re-try remaining I/O upon this error. 1247 */ 1248 if (error == EFAULT) { 1249 ZFS_EXIT(zfsvfs); 1250 return (error); 1251 } 1252#endif 1253 1254 if (ioflag & (FSYNC | FDSYNC) || 1255 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1256 zil_commit(zilog, zp->z_id); 1257 1258 ZFS_EXIT(zfsvfs); 1259 return (0); 1260} 1261 1262void 1263zfs_get_done(zgd_t *zgd, int error) 1264{ 1265 znode_t *zp = zgd->zgd_private; 1266 objset_t *os = zp->z_zfsvfs->z_os; 1267 1268 if (zgd->zgd_db) 1269 dmu_buf_rele(zgd->zgd_db, zgd); 1270 1271 zfs_range_unlock(zgd->zgd_rl); 1272 1273 /* 1274 * Release the vnode asynchronously as we currently have the 1275 * txg stopped from syncing. 1276 */ 1277 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1278 1279 if (error == 0 && zgd->zgd_bp) 1280 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1281 1282 kmem_free(zgd, sizeof (zgd_t)); 1283} 1284 1285#ifdef DEBUG 1286static int zil_fault_io = 0; 1287#endif 1288 1289/* 1290 * Get data to generate a TX_WRITE intent log record. 1291 */ 1292int 1293zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1294{ 1295 zfsvfs_t *zfsvfs = arg; 1296 objset_t *os = zfsvfs->z_os; 1297 znode_t *zp; 1298 uint64_t object = lr->lr_foid; 1299 uint64_t offset = lr->lr_offset; 1300 uint64_t size = lr->lr_length; 1301 blkptr_t *bp = &lr->lr_blkptr; 1302 dmu_buf_t *db; 1303 zgd_t *zgd; 1304 int error = 0; 1305 1306 ASSERT(zio != NULL); 1307 ASSERT(size != 0); 1308 1309 /* 1310 * Nothing to do if the file has been removed 1311 */ 1312 if (zfs_zget(zfsvfs, object, &zp) != 0) 1313 return (SET_ERROR(ENOENT)); 1314 if (zp->z_unlinked) { 1315 /* 1316 * Release the vnode asynchronously as we currently have the 1317 * txg stopped from syncing. 1318 */ 1319 VN_RELE_ASYNC(ZTOV(zp), 1320 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1321 return (SET_ERROR(ENOENT)); 1322 } 1323 1324 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1325 zgd->zgd_zilog = zfsvfs->z_log; 1326 zgd->zgd_private = zp; 1327 1328 /* 1329 * Write records come in two flavors: immediate and indirect. 1330 * For small writes it's cheaper to store the data with the 1331 * log record (immediate); for large writes it's cheaper to 1332 * sync the data and get a pointer to it (indirect) so that 1333 * we don't have to write the data twice. 1334 */ 1335 if (buf != NULL) { /* immediate write */ 1336 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER); 1337 /* test for truncation needs to be done while range locked */ 1338 if (offset >= zp->z_size) { 1339 error = SET_ERROR(ENOENT); 1340 } else { 1341 error = dmu_read(os, object, offset, size, buf, 1342 DMU_READ_NO_PREFETCH); 1343 } 1344 ASSERT(error == 0 || error == ENOENT); 1345 } else { /* indirect write */ 1346 /* 1347 * Have to lock the whole block to ensure when it's 1348 * written out and it's checksum is being calculated 1349 * that no one can change the data. We need to re-check 1350 * blocksize after we get the lock in case it's changed! 1351 */ 1352 for (;;) { 1353 uint64_t blkoff; 1354 size = zp->z_blksz; 1355 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; 1356 offset -= blkoff; 1357 zgd->zgd_rl = zfs_range_lock(zp, offset, size, 1358 RL_READER); 1359 if (zp->z_blksz == size) 1360 break; 1361 offset += blkoff; 1362 zfs_range_unlock(zgd->zgd_rl); 1363 } 1364 /* test for truncation needs to be done while range locked */ 1365 if (lr->lr_offset >= zp->z_size) 1366 error = SET_ERROR(ENOENT); 1367#ifdef DEBUG 1368 if (zil_fault_io) { 1369 error = SET_ERROR(EIO); 1370 zil_fault_io = 0; 1371 } 1372#endif 1373 if (error == 0) 1374 error = dmu_buf_hold(os, object, offset, zgd, &db, 1375 DMU_READ_NO_PREFETCH); 1376 1377 if (error == 0) { 1378 blkptr_t *obp = dmu_buf_get_blkptr(db); 1379 if (obp) { 1380 ASSERT(BP_IS_HOLE(bp)); 1381 *bp = *obp; 1382 } 1383 1384 zgd->zgd_db = db; 1385 zgd->zgd_bp = bp; 1386 1387 ASSERT(db->db_offset == offset); 1388 ASSERT(db->db_size == size); 1389 1390 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1391 zfs_get_done, zgd); 1392 ASSERT(error || lr->lr_length <= zp->z_blksz); 1393 1394 /* 1395 * On success, we need to wait for the write I/O 1396 * initiated by dmu_sync() to complete before we can 1397 * release this dbuf. We will finish everything up 1398 * in the zfs_get_done() callback. 1399 */ 1400 if (error == 0) 1401 return (0); 1402 1403 if (error == EALREADY) { 1404 lr->lr_common.lrc_txtype = TX_WRITE2; 1405 error = 0; 1406 } 1407 } 1408 } 1409 1410 zfs_get_done(zgd, error); 1411 1412 return (error); 1413} 1414 1415/*ARGSUSED*/ 1416static int 1417zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 1418 caller_context_t *ct) 1419{ 1420 znode_t *zp = VTOZ(vp); 1421 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1422 int error; 1423 1424 ZFS_ENTER(zfsvfs); 1425 ZFS_VERIFY_ZP(zp); 1426 1427 if (flag & V_ACE_MASK) 1428 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 1429 else 1430 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1431 1432 ZFS_EXIT(zfsvfs); 1433 return (error); 1434} 1435 1436/* 1437 * If vnode is for a device return a specfs vnode instead. 1438 */ 1439static int 1440specvp_check(vnode_t **vpp, cred_t *cr) 1441{ 1442 int error = 0; 1443 1444 if (IS_DEVVP(*vpp)) { 1445 struct vnode *svp; 1446 1447 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1448 VN_RELE(*vpp); 1449 if (svp == NULL) 1450 error = SET_ERROR(ENOSYS); 1451 *vpp = svp; 1452 } 1453 return (error); 1454} 1455 1456 1457/* 1458 * Lookup an entry in a directory, or an extended attribute directory. 1459 * If it exists, return a held vnode reference for it. 1460 * 1461 * IN: dvp - vnode of directory to search. 1462 * nm - name of entry to lookup. 1463 * pnp - full pathname to lookup [UNUSED]. 1464 * flags - LOOKUP_XATTR set if looking for an attribute. 1465 * rdir - root directory vnode [UNUSED]. 1466 * cr - credentials of caller. 1467 * ct - caller context 1468 * direntflags - directory lookup flags 1469 * realpnp - returned pathname. 1470 * 1471 * OUT: vpp - vnode of located entry, NULL if not found. 1472 * 1473 * RETURN: 0 on success, error code on failure. 1474 * 1475 * Timestamps: 1476 * NA 1477 */ 1478/* ARGSUSED */ 1479static int 1480zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 1481 int nameiop, cred_t *cr, kthread_t *td, int flags) 1482{ 1483 znode_t *zdp = VTOZ(dvp); 1484 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1485 int error = 0; 1486 int *direntflags = NULL; 1487 void *realpnp = NULL; 1488 1489 /* fast path */ 1490 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 1491 1492 if (dvp->v_type != VDIR) { 1493 return (SET_ERROR(ENOTDIR)); 1494 } else if (zdp->z_sa_hdl == NULL) { 1495 return (SET_ERROR(EIO)); 1496 } 1497 1498 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 1499 error = zfs_fastaccesschk_execute(zdp, cr); 1500 if (!error) { 1501 *vpp = dvp; 1502 VN_HOLD(*vpp); 1503 return (0); 1504 } 1505 return (error); 1506 } else { 1507 vnode_t *tvp = dnlc_lookup(dvp, nm); 1508 1509 if (tvp) { 1510 error = zfs_fastaccesschk_execute(zdp, cr); 1511 if (error) { 1512 VN_RELE(tvp); 1513 return (error); 1514 } 1515 if (tvp == DNLC_NO_VNODE) { 1516 VN_RELE(tvp); 1517 return (SET_ERROR(ENOENT)); 1518 } else { 1519 *vpp = tvp; 1520 return (specvp_check(vpp, cr)); 1521 } 1522 } 1523 } 1524 } 1525 1526 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1527 1528 ZFS_ENTER(zfsvfs); 1529 ZFS_VERIFY_ZP(zdp); 1530 1531 *vpp = NULL; 1532 1533 if (flags & LOOKUP_XATTR) { 1534#ifdef TODO 1535 /* 1536 * If the xattr property is off, refuse the lookup request. 1537 */ 1538 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1539 ZFS_EXIT(zfsvfs); 1540 return (SET_ERROR(EINVAL)); 1541 } 1542#endif 1543 1544 /* 1545 * We don't allow recursive attributes.. 1546 * Maybe someday we will. 1547 */ 1548 if (zdp->z_pflags & ZFS_XATTR) { 1549 ZFS_EXIT(zfsvfs); 1550 return (SET_ERROR(EINVAL)); 1551 } 1552 1553 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1554 ZFS_EXIT(zfsvfs); 1555 return (error); 1556 } 1557 1558 /* 1559 * Do we have permission to get into attribute directory? 1560 */ 1561 1562 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1563 B_FALSE, cr)) { 1564 VN_RELE(*vpp); 1565 *vpp = NULL; 1566 } 1567 1568 ZFS_EXIT(zfsvfs); 1569 return (error); 1570 } 1571 1572 if (dvp->v_type != VDIR) { 1573 ZFS_EXIT(zfsvfs); 1574 return (SET_ERROR(ENOTDIR)); 1575 } 1576 1577 /* 1578 * Check accessibility of directory. 1579 */ 1580 1581 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1582 ZFS_EXIT(zfsvfs); 1583 return (error); 1584 } 1585 1586 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1587 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1588 ZFS_EXIT(zfsvfs); 1589 return (SET_ERROR(EILSEQ)); 1590 } 1591 1592 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1593 if (error == 0) 1594 error = specvp_check(vpp, cr); 1595 1596 /* Translate errors and add SAVENAME when needed. */ 1597 if (cnp->cn_flags & ISLASTCN) { 1598 switch (nameiop) { 1599 case CREATE: 1600 case RENAME: 1601 if (error == ENOENT) { 1602 error = EJUSTRETURN; 1603 cnp->cn_flags |= SAVENAME; 1604 break; 1605 } 1606 /* FALLTHROUGH */ 1607 case DELETE: 1608 if (error == 0) 1609 cnp->cn_flags |= SAVENAME; 1610 break; 1611 } 1612 } 1613 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) { 1614 int ltype = 0; 1615 1616 if (cnp->cn_flags & ISDOTDOT) { 1617 ltype = VOP_ISLOCKED(dvp); 1618 VOP_UNLOCK(dvp, 0); 1619 } 1620 ZFS_EXIT(zfsvfs); 1621 error = vn_lock(*vpp, cnp->cn_lkflags); 1622 if (cnp->cn_flags & ISDOTDOT) 1623 vn_lock(dvp, ltype | LK_RETRY); 1624 if (error != 0) { 1625 VN_RELE(*vpp); 1626 *vpp = NULL; 1627 return (error); 1628 } 1629 } else { 1630 ZFS_EXIT(zfsvfs); 1631 } 1632 1633#ifdef FREEBSD_NAMECACHE 1634 /* 1635 * Insert name into cache (as non-existent) if appropriate. 1636 */ 1637 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) != 0) 1638 cache_enter(dvp, *vpp, cnp); 1639 /* 1640 * Insert name into cache if appropriate. 1641 */ 1642 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1643 if (!(cnp->cn_flags & ISLASTCN) || 1644 (nameiop != DELETE && nameiop != RENAME)) { 1645 cache_enter(dvp, *vpp, cnp); 1646 } 1647 } 1648#endif 1649 1650 return (error); 1651} 1652 1653/* 1654 * Attempt to create a new entry in a directory. If the entry 1655 * already exists, truncate the file if permissible, else return 1656 * an error. Return the vp of the created or trunc'd file. 1657 * 1658 * IN: dvp - vnode of directory to put new file entry in. 1659 * name - name of new file entry. 1660 * vap - attributes of new file. 1661 * excl - flag indicating exclusive or non-exclusive mode. 1662 * mode - mode to open file with. 1663 * cr - credentials of caller. 1664 * flag - large file flag [UNUSED]. 1665 * ct - caller context 1666 * vsecp - ACL to be set 1667 * 1668 * OUT: vpp - vnode of created or trunc'd entry. 1669 * 1670 * RETURN: 0 on success, error code on failure. 1671 * 1672 * Timestamps: 1673 * dvp - ctime|mtime updated if new entry created 1674 * vp - ctime|mtime always, atime if new 1675 */ 1676 1677/* ARGSUSED */ 1678static int 1679zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1680 vnode_t **vpp, cred_t *cr, kthread_t *td) 1681{ 1682 znode_t *zp, *dzp = VTOZ(dvp); 1683 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1684 zilog_t *zilog; 1685 objset_t *os; 1686 zfs_dirlock_t *dl; 1687 dmu_tx_t *tx; 1688 int error; 1689 ksid_t *ksid; 1690 uid_t uid; 1691 gid_t gid = crgetgid(cr); 1692 zfs_acl_ids_t acl_ids; 1693 boolean_t fuid_dirtied; 1694 boolean_t have_acl = B_FALSE; 1695 boolean_t waited = B_FALSE; 1696 void *vsecp = NULL; 1697 int flag = 0; 1698 1699 /* 1700 * If we have an ephemeral id, ACL, or XVATTR then 1701 * make sure file system is at proper version 1702 */ 1703 1704 ksid = crgetsid(cr, KSID_OWNER); 1705 if (ksid) 1706 uid = ksid_getid(ksid); 1707 else 1708 uid = crgetuid(cr); 1709 1710 if (zfsvfs->z_use_fuids == B_FALSE && 1711 (vsecp || (vap->va_mask & AT_XVATTR) || 1712 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1713 return (SET_ERROR(EINVAL)); 1714 1715 ZFS_ENTER(zfsvfs); 1716 ZFS_VERIFY_ZP(dzp); 1717 os = zfsvfs->z_os; 1718 zilog = zfsvfs->z_log; 1719 1720 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1721 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1722 ZFS_EXIT(zfsvfs); 1723 return (SET_ERROR(EILSEQ)); 1724 } 1725 1726 if (vap->va_mask & AT_XVATTR) { 1727 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 1728 crgetuid(cr), cr, vap->va_type)) != 0) { 1729 ZFS_EXIT(zfsvfs); 1730 return (error); 1731 } 1732 } 1733 1734 getnewvnode_reserve(1); 1735 1736top: 1737 *vpp = NULL; 1738 1739 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) 1740 vap->va_mode &= ~S_ISVTX; 1741 1742 if (*name == '\0') { 1743 /* 1744 * Null component name refers to the directory itself. 1745 */ 1746 VN_HOLD(dvp); 1747 zp = dzp; 1748 dl = NULL; 1749 error = 0; 1750 } else { 1751 /* possible VN_HOLD(zp) */ 1752 int zflg = 0; 1753 1754 if (flag & FIGNORECASE) 1755 zflg |= ZCILOOK; 1756 1757 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1758 NULL, NULL); 1759 if (error) { 1760 if (have_acl) 1761 zfs_acl_ids_free(&acl_ids); 1762 if (strcmp(name, "..") == 0) 1763 error = SET_ERROR(EISDIR); 1764 getnewvnode_drop_reserve(); 1765 ZFS_EXIT(zfsvfs); 1766 return (error); 1767 } 1768 } 1769 1770 if (zp == NULL) { 1771 uint64_t txtype; 1772 1773 /* 1774 * Create a new file object and update the directory 1775 * to reference it. 1776 */ 1777 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1778 if (have_acl) 1779 zfs_acl_ids_free(&acl_ids); 1780 goto out; 1781 } 1782 1783 /* 1784 * We only support the creation of regular files in 1785 * extended attribute directories. 1786 */ 1787 1788 if ((dzp->z_pflags & ZFS_XATTR) && 1789 (vap->va_type != VREG)) { 1790 if (have_acl) 1791 zfs_acl_ids_free(&acl_ids); 1792 error = SET_ERROR(EINVAL); 1793 goto out; 1794 } 1795 1796 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 1797 cr, vsecp, &acl_ids)) != 0) 1798 goto out; 1799 have_acl = B_TRUE; 1800 1801 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1802 zfs_acl_ids_free(&acl_ids); 1803 error = SET_ERROR(EDQUOT); 1804 goto out; 1805 } 1806 1807 tx = dmu_tx_create(os); 1808 1809 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1810 ZFS_SA_BASE_ATTR_SIZE); 1811 1812 fuid_dirtied = zfsvfs->z_fuid_dirty; 1813 if (fuid_dirtied) 1814 zfs_fuid_txhold(zfsvfs, tx); 1815 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1816 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1817 if (!zfsvfs->z_use_sa && 1818 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1819 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1820 0, acl_ids.z_aclp->z_acl_bytes); 1821 } 1822 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 1823 if (error) { 1824 zfs_dirent_unlock(dl); 1825 if (error == ERESTART) { 1826 waited = B_TRUE; 1827 dmu_tx_wait(tx); 1828 dmu_tx_abort(tx); 1829 goto top; 1830 } 1831 zfs_acl_ids_free(&acl_ids); 1832 dmu_tx_abort(tx); 1833 getnewvnode_drop_reserve(); 1834 ZFS_EXIT(zfsvfs); 1835 return (error); 1836 } 1837 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1838 1839 if (fuid_dirtied) 1840 zfs_fuid_sync(zfsvfs, tx); 1841 1842 (void) zfs_link_create(dl, zp, tx, ZNEW); 1843 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1844 if (flag & FIGNORECASE) 1845 txtype |= TX_CI; 1846 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1847 vsecp, acl_ids.z_fuidp, vap); 1848 zfs_acl_ids_free(&acl_ids); 1849 dmu_tx_commit(tx); 1850 } else { 1851 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1852 1853 if (have_acl) 1854 zfs_acl_ids_free(&acl_ids); 1855 have_acl = B_FALSE; 1856 1857 /* 1858 * A directory entry already exists for this name. 1859 */ 1860 /* 1861 * Can't truncate an existing file if in exclusive mode. 1862 */ 1863 if (excl == EXCL) { 1864 error = SET_ERROR(EEXIST); 1865 goto out; 1866 } 1867 /* 1868 * Can't open a directory for writing. 1869 */ 1870 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1871 error = SET_ERROR(EISDIR); 1872 goto out; 1873 } 1874 /* 1875 * Verify requested access to file. 1876 */ 1877 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1878 goto out; 1879 } 1880 1881 mutex_enter(&dzp->z_lock); 1882 dzp->z_seq++; 1883 mutex_exit(&dzp->z_lock); 1884 1885 /* 1886 * Truncate regular files if requested. 1887 */ 1888 if ((ZTOV(zp)->v_type == VREG) && 1889 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1890 /* we can't hold any locks when calling zfs_freesp() */ 1891 zfs_dirent_unlock(dl); 1892 dl = NULL; 1893 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1894 if (error == 0) { 1895 vnevent_create(ZTOV(zp), ct); 1896 } 1897 } 1898 } 1899out: 1900 getnewvnode_drop_reserve(); 1901 if (dl) 1902 zfs_dirent_unlock(dl); 1903 1904 if (error) { 1905 if (zp) 1906 VN_RELE(ZTOV(zp)); 1907 } else { 1908 *vpp = ZTOV(zp); 1909 error = specvp_check(vpp, cr); 1910 } 1911 1912 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1913 zil_commit(zilog, 0); 1914 1915 ZFS_EXIT(zfsvfs); 1916 return (error); 1917} 1918 1919/* 1920 * Remove an entry from a directory. 1921 * 1922 * IN: dvp - vnode of directory to remove entry from. 1923 * name - name of entry to remove. 1924 * cr - credentials of caller. 1925 * ct - caller context 1926 * flags - case flags 1927 * 1928 * RETURN: 0 on success, error code on failure. 1929 * 1930 * Timestamps: 1931 * dvp - ctime|mtime 1932 * vp - ctime (if nlink > 0) 1933 */ 1934 1935uint64_t null_xattr = 0; 1936 1937/*ARGSUSED*/ 1938static int 1939zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1940 int flags) 1941{ 1942 znode_t *zp, *dzp = VTOZ(dvp); 1943 znode_t *xzp; 1944 vnode_t *vp; 1945 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1946 zilog_t *zilog; 1947 uint64_t acl_obj, xattr_obj; 1948 uint64_t xattr_obj_unlinked = 0; 1949 uint64_t obj = 0; 1950 zfs_dirlock_t *dl; 1951 dmu_tx_t *tx; 1952 boolean_t may_delete_now, delete_now = FALSE; 1953 boolean_t unlinked, toobig = FALSE; 1954 uint64_t txtype; 1955 pathname_t *realnmp = NULL; 1956 pathname_t realnm; 1957 int error; 1958 int zflg = ZEXISTS; 1959 boolean_t waited = B_FALSE; 1960 1961 ZFS_ENTER(zfsvfs); 1962 ZFS_VERIFY_ZP(dzp); 1963 zilog = zfsvfs->z_log; 1964 1965 if (flags & FIGNORECASE) { 1966 zflg |= ZCILOOK; 1967 pn_alloc(&realnm); 1968 realnmp = &realnm; 1969 } 1970 1971top: 1972 xattr_obj = 0; 1973 xzp = NULL; 1974 /* 1975 * Attempt to lock directory; fail if entry doesn't exist. 1976 */ 1977 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1978 NULL, realnmp)) { 1979 if (realnmp) 1980 pn_free(realnmp); 1981 ZFS_EXIT(zfsvfs); 1982 return (error); 1983 } 1984 1985 vp = ZTOV(zp); 1986 1987 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1988 goto out; 1989 } 1990 1991 /* 1992 * Need to use rmdir for removing directories. 1993 */ 1994 if (vp->v_type == VDIR) { 1995 error = SET_ERROR(EPERM); 1996 goto out; 1997 } 1998 1999 vnevent_remove(vp, dvp, name, ct); 2000 2001 if (realnmp) 2002 dnlc_remove(dvp, realnmp->pn_buf); 2003 else 2004 dnlc_remove(dvp, name); 2005 2006 VI_LOCK(vp); 2007 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 2008 VI_UNLOCK(vp); 2009 2010 /* 2011 * We may delete the znode now, or we may put it in the unlinked set; 2012 * it depends on whether we're the last link, and on whether there are 2013 * other holds on the vnode. So we dmu_tx_hold() the right things to 2014 * allow for either case. 2015 */ 2016 obj = zp->z_id; 2017 tx = dmu_tx_create(zfsvfs->z_os); 2018 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2019 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2020 zfs_sa_upgrade_txholds(tx, zp); 2021 zfs_sa_upgrade_txholds(tx, dzp); 2022 if (may_delete_now) { 2023 toobig = 2024 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 2025 /* if the file is too big, only hold_free a token amount */ 2026 dmu_tx_hold_free(tx, zp->z_id, 0, 2027 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 2028 } 2029 2030 /* are there any extended attributes? */ 2031 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 2032 &xattr_obj, sizeof (xattr_obj)); 2033 if (error == 0 && xattr_obj) { 2034 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 2035 ASSERT0(error); 2036 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2037 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 2038 } 2039 2040 mutex_enter(&zp->z_lock); 2041 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now) 2042 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 2043 mutex_exit(&zp->z_lock); 2044 2045 /* charge as an update -- would be nice not to charge at all */ 2046 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2047 2048 /* 2049 * Mark this transaction as typically resulting in a net free of space 2050 */ 2051 dmu_tx_mark_netfree(tx); 2052 2053 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2054 if (error) { 2055 zfs_dirent_unlock(dl); 2056 VN_RELE(vp); 2057 if (xzp) 2058 VN_RELE(ZTOV(xzp)); 2059 if (error == ERESTART) { 2060 waited = B_TRUE; 2061 dmu_tx_wait(tx); 2062 dmu_tx_abort(tx); 2063 goto top; 2064 } 2065 if (realnmp) 2066 pn_free(realnmp); 2067 dmu_tx_abort(tx); 2068 ZFS_EXIT(zfsvfs); 2069 return (error); 2070 } 2071 2072 /* 2073 * Remove the directory entry. 2074 */ 2075 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 2076 2077 if (error) { 2078 dmu_tx_commit(tx); 2079 goto out; 2080 } 2081 2082 if (unlinked) { 2083 /* 2084 * Hold z_lock so that we can make sure that the ACL obj 2085 * hasn't changed. Could have been deleted due to 2086 * zfs_sa_upgrade(). 2087 */ 2088 mutex_enter(&zp->z_lock); 2089 VI_LOCK(vp); 2090 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 2091 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked)); 2092 delete_now = may_delete_now && !toobig && 2093 vp->v_count == 1 && !vn_has_cached_data(vp) && 2094 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) == 2095 acl_obj; 2096 VI_UNLOCK(vp); 2097 } 2098 2099 if (delete_now) { 2100#ifdef __FreeBSD__ 2101 panic("zfs_remove: delete_now branch taken"); 2102#endif 2103 if (xattr_obj_unlinked) { 2104 ASSERT3U(xzp->z_links, ==, 2); 2105 mutex_enter(&xzp->z_lock); 2106 xzp->z_unlinked = 1; 2107 xzp->z_links = 0; 2108 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 2109 &xzp->z_links, sizeof (xzp->z_links), tx); 2110 ASSERT3U(error, ==, 0); 2111 mutex_exit(&xzp->z_lock); 2112 zfs_unlinked_add(xzp, tx); 2113 2114 if (zp->z_is_sa) 2115 error = sa_remove(zp->z_sa_hdl, 2116 SA_ZPL_XATTR(zfsvfs), tx); 2117 else 2118 error = sa_update(zp->z_sa_hdl, 2119 SA_ZPL_XATTR(zfsvfs), &null_xattr, 2120 sizeof (uint64_t), tx); 2121 ASSERT0(error); 2122 } 2123 VI_LOCK(vp); 2124 vp->v_count--; 2125 ASSERT0(vp->v_count); 2126 VI_UNLOCK(vp); 2127 mutex_exit(&zp->z_lock); 2128 zfs_znode_delete(zp, tx); 2129 } else if (unlinked) { 2130 mutex_exit(&zp->z_lock); 2131 zfs_unlinked_add(zp, tx); 2132#ifdef __FreeBSD__ 2133 vp->v_vflag |= VV_NOSYNC; 2134#endif 2135 } 2136 2137 txtype = TX_REMOVE; 2138 if (flags & FIGNORECASE) 2139 txtype |= TX_CI; 2140 zfs_log_remove(zilog, tx, txtype, dzp, name, obj); 2141 2142 dmu_tx_commit(tx); 2143out: 2144 if (realnmp) 2145 pn_free(realnmp); 2146 2147 zfs_dirent_unlock(dl); 2148 2149 if (!delete_now) 2150 VN_RELE(vp); 2151 if (xzp) 2152 VN_RELE(ZTOV(xzp)); 2153 2154 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2155 zil_commit(zilog, 0); 2156 2157 ZFS_EXIT(zfsvfs); 2158 return (error); 2159} 2160 2161/* 2162 * Create a new directory and insert it into dvp using the name 2163 * provided. Return a pointer to the inserted directory. 2164 * 2165 * IN: dvp - vnode of directory to add subdir to. 2166 * dirname - name of new directory. 2167 * vap - attributes of new directory. 2168 * cr - credentials of caller. 2169 * ct - caller context 2170 * flags - case flags 2171 * vsecp - ACL to be set 2172 * 2173 * OUT: vpp - vnode of created directory. 2174 * 2175 * RETURN: 0 on success, error code on failure. 2176 * 2177 * Timestamps: 2178 * dvp - ctime|mtime updated 2179 * vp - ctime|mtime|atime updated 2180 */ 2181/*ARGSUSED*/ 2182static int 2183zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 2184 caller_context_t *ct, int flags, vsecattr_t *vsecp) 2185{ 2186 znode_t *zp, *dzp = VTOZ(dvp); 2187 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2188 zilog_t *zilog; 2189 zfs_dirlock_t *dl; 2190 uint64_t txtype; 2191 dmu_tx_t *tx; 2192 int error; 2193 int zf = ZNEW; 2194 ksid_t *ksid; 2195 uid_t uid; 2196 gid_t gid = crgetgid(cr); 2197 zfs_acl_ids_t acl_ids; 2198 boolean_t fuid_dirtied; 2199 boolean_t waited = B_FALSE; 2200 2201 ASSERT(vap->va_type == VDIR); 2202 2203 /* 2204 * If we have an ephemeral id, ACL, or XVATTR then 2205 * make sure file system is at proper version 2206 */ 2207 2208 ksid = crgetsid(cr, KSID_OWNER); 2209 if (ksid) 2210 uid = ksid_getid(ksid); 2211 else 2212 uid = crgetuid(cr); 2213 if (zfsvfs->z_use_fuids == B_FALSE && 2214 (vsecp || (vap->va_mask & AT_XVATTR) || 2215 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 2216 return (SET_ERROR(EINVAL)); 2217 2218 ZFS_ENTER(zfsvfs); 2219 ZFS_VERIFY_ZP(dzp); 2220 zilog = zfsvfs->z_log; 2221 2222 if (dzp->z_pflags & ZFS_XATTR) { 2223 ZFS_EXIT(zfsvfs); 2224 return (SET_ERROR(EINVAL)); 2225 } 2226 2227 if (zfsvfs->z_utf8 && u8_validate(dirname, 2228 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2229 ZFS_EXIT(zfsvfs); 2230 return (SET_ERROR(EILSEQ)); 2231 } 2232 if (flags & FIGNORECASE) 2233 zf |= ZCILOOK; 2234 2235 if (vap->va_mask & AT_XVATTR) { 2236 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 2237 crgetuid(cr), cr, vap->va_type)) != 0) { 2238 ZFS_EXIT(zfsvfs); 2239 return (error); 2240 } 2241 } 2242 2243 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 2244 vsecp, &acl_ids)) != 0) { 2245 ZFS_EXIT(zfsvfs); 2246 return (error); 2247 } 2248 2249 getnewvnode_reserve(1); 2250 2251 /* 2252 * First make sure the new directory doesn't exist. 2253 * 2254 * Existence is checked first to make sure we don't return 2255 * EACCES instead of EEXIST which can cause some applications 2256 * to fail. 2257 */ 2258top: 2259 *vpp = NULL; 2260 2261 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 2262 NULL, NULL)) { 2263 zfs_acl_ids_free(&acl_ids); 2264 getnewvnode_drop_reserve(); 2265 ZFS_EXIT(zfsvfs); 2266 return (error); 2267 } 2268 2269 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 2270 zfs_acl_ids_free(&acl_ids); 2271 zfs_dirent_unlock(dl); 2272 getnewvnode_drop_reserve(); 2273 ZFS_EXIT(zfsvfs); 2274 return (error); 2275 } 2276 2277 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 2278 zfs_acl_ids_free(&acl_ids); 2279 zfs_dirent_unlock(dl); 2280 getnewvnode_drop_reserve(); 2281 ZFS_EXIT(zfsvfs); 2282 return (SET_ERROR(EDQUOT)); 2283 } 2284 2285 /* 2286 * Add a new entry to the directory. 2287 */ 2288 tx = dmu_tx_create(zfsvfs->z_os); 2289 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 2290 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2291 fuid_dirtied = zfsvfs->z_fuid_dirty; 2292 if (fuid_dirtied) 2293 zfs_fuid_txhold(zfsvfs, tx); 2294 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2295 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2296 acl_ids.z_aclp->z_acl_bytes); 2297 } 2298 2299 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 2300 ZFS_SA_BASE_ATTR_SIZE); 2301 2302 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2303 if (error) { 2304 zfs_dirent_unlock(dl); 2305 if (error == ERESTART) { 2306 waited = B_TRUE; 2307 dmu_tx_wait(tx); 2308 dmu_tx_abort(tx); 2309 goto top; 2310 } 2311 zfs_acl_ids_free(&acl_ids); 2312 dmu_tx_abort(tx); 2313 getnewvnode_drop_reserve(); 2314 ZFS_EXIT(zfsvfs); 2315 return (error); 2316 } 2317 2318 /* 2319 * Create new node. 2320 */ 2321 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 2322 2323 if (fuid_dirtied) 2324 zfs_fuid_sync(zfsvfs, tx); 2325 2326 /* 2327 * Now put new name in parent dir. 2328 */ 2329 (void) zfs_link_create(dl, zp, tx, ZNEW); 2330 2331 *vpp = ZTOV(zp); 2332 2333 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 2334 if (flags & FIGNORECASE) 2335 txtype |= TX_CI; 2336 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 2337 acl_ids.z_fuidp, vap); 2338 2339 zfs_acl_ids_free(&acl_ids); 2340 2341 dmu_tx_commit(tx); 2342 2343 getnewvnode_drop_reserve(); 2344 2345 zfs_dirent_unlock(dl); 2346 2347 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2348 zil_commit(zilog, 0); 2349 2350 ZFS_EXIT(zfsvfs); 2351 return (0); 2352} 2353 2354/* 2355 * Remove a directory subdir entry. If the current working 2356 * directory is the same as the subdir to be removed, the 2357 * remove will fail. 2358 * 2359 * IN: dvp - vnode of directory to remove from. 2360 * name - name of directory to be removed. 2361 * cwd - vnode of current working directory. 2362 * cr - credentials of caller. 2363 * ct - caller context 2364 * flags - case flags 2365 * 2366 * RETURN: 0 on success, error code on failure. 2367 * 2368 * Timestamps: 2369 * dvp - ctime|mtime updated 2370 */ 2371/*ARGSUSED*/ 2372static int 2373zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 2374 caller_context_t *ct, int flags) 2375{ 2376 znode_t *dzp = VTOZ(dvp); 2377 znode_t *zp; 2378 vnode_t *vp; 2379 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2380 zilog_t *zilog; 2381 zfs_dirlock_t *dl; 2382 dmu_tx_t *tx; 2383 int error; 2384 int zflg = ZEXISTS; 2385 boolean_t waited = B_FALSE; 2386 2387 ZFS_ENTER(zfsvfs); 2388 ZFS_VERIFY_ZP(dzp); 2389 zilog = zfsvfs->z_log; 2390 2391 if (flags & FIGNORECASE) 2392 zflg |= ZCILOOK; 2393top: 2394 zp = NULL; 2395 2396 /* 2397 * Attempt to lock directory; fail if entry doesn't exist. 2398 */ 2399 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 2400 NULL, NULL)) { 2401 ZFS_EXIT(zfsvfs); 2402 return (error); 2403 } 2404 2405 vp = ZTOV(zp); 2406 2407 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 2408 goto out; 2409 } 2410 2411 if (vp->v_type != VDIR) { 2412 error = SET_ERROR(ENOTDIR); 2413 goto out; 2414 } 2415 2416 if (vp == cwd) { 2417 error = SET_ERROR(EINVAL); 2418 goto out; 2419 } 2420 2421 vnevent_rmdir(vp, dvp, name, ct); 2422 2423 /* 2424 * Grab a lock on the directory to make sure that noone is 2425 * trying to add (or lookup) entries while we are removing it. 2426 */ 2427 rw_enter(&zp->z_name_lock, RW_WRITER); 2428 2429 /* 2430 * Grab a lock on the parent pointer to make sure we play well 2431 * with the treewalk and directory rename code. 2432 */ 2433 rw_enter(&zp->z_parent_lock, RW_WRITER); 2434 2435 tx = dmu_tx_create(zfsvfs->z_os); 2436 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2437 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2438 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2439 zfs_sa_upgrade_txholds(tx, zp); 2440 zfs_sa_upgrade_txholds(tx, dzp); 2441 dmu_tx_mark_netfree(tx); 2442 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2443 if (error) { 2444 rw_exit(&zp->z_parent_lock); 2445 rw_exit(&zp->z_name_lock); 2446 zfs_dirent_unlock(dl); 2447 VN_RELE(vp); 2448 if (error == ERESTART) { 2449 waited = B_TRUE; 2450 dmu_tx_wait(tx); 2451 dmu_tx_abort(tx); 2452 goto top; 2453 } 2454 dmu_tx_abort(tx); 2455 ZFS_EXIT(zfsvfs); 2456 return (error); 2457 } 2458 2459#ifdef FREEBSD_NAMECACHE 2460 cache_purge(dvp); 2461#endif 2462 2463 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 2464 2465 if (error == 0) { 2466 uint64_t txtype = TX_RMDIR; 2467 if (flags & FIGNORECASE) 2468 txtype |= TX_CI; 2469 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT); 2470 } 2471 2472 dmu_tx_commit(tx); 2473 2474 rw_exit(&zp->z_parent_lock); 2475 rw_exit(&zp->z_name_lock); 2476#ifdef FREEBSD_NAMECACHE 2477 cache_purge(vp); 2478#endif 2479out: 2480 zfs_dirent_unlock(dl); 2481 2482 VN_RELE(vp); 2483 2484 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2485 zil_commit(zilog, 0); 2486 2487 ZFS_EXIT(zfsvfs); 2488 return (error); 2489} 2490 2491/* 2492 * Read as many directory entries as will fit into the provided 2493 * buffer from the given directory cursor position (specified in 2494 * the uio structure). 2495 * 2496 * IN: vp - vnode of directory to read. 2497 * uio - structure supplying read location, range info, 2498 * and return buffer. 2499 * cr - credentials of caller. 2500 * ct - caller context 2501 * flags - case flags 2502 * 2503 * OUT: uio - updated offset and range, buffer filled. 2504 * eofp - set to true if end-of-file detected. 2505 * 2506 * RETURN: 0 on success, error code on failure. 2507 * 2508 * Timestamps: 2509 * vp - atime updated 2510 * 2511 * Note that the low 4 bits of the cookie returned by zap is always zero. 2512 * This allows us to use the low range for "special" directory entries: 2513 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 2514 * we use the offset 2 for the '.zfs' directory. 2515 */ 2516/* ARGSUSED */ 2517static int 2518zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 2519{ 2520 znode_t *zp = VTOZ(vp); 2521 iovec_t *iovp; 2522 edirent_t *eodp; 2523 dirent64_t *odp; 2524 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2525 objset_t *os; 2526 caddr_t outbuf; 2527 size_t bufsize; 2528 zap_cursor_t zc; 2529 zap_attribute_t zap; 2530 uint_t bytes_wanted; 2531 uint64_t offset; /* must be unsigned; checks for < 1 */ 2532 uint64_t parent; 2533 int local_eof; 2534 int outcount; 2535 int error; 2536 uint8_t prefetch; 2537 boolean_t check_sysattrs; 2538 uint8_t type; 2539 int ncooks; 2540 u_long *cooks = NULL; 2541 int flags = 0; 2542 2543 ZFS_ENTER(zfsvfs); 2544 ZFS_VERIFY_ZP(zp); 2545 2546 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 2547 &parent, sizeof (parent))) != 0) { 2548 ZFS_EXIT(zfsvfs); 2549 return (error); 2550 } 2551 2552 /* 2553 * If we are not given an eof variable, 2554 * use a local one. 2555 */ 2556 if (eofp == NULL) 2557 eofp = &local_eof; 2558 2559 /* 2560 * Check for valid iov_len. 2561 */ 2562 if (uio->uio_iov->iov_len <= 0) { 2563 ZFS_EXIT(zfsvfs); 2564 return (SET_ERROR(EINVAL)); 2565 } 2566 2567 /* 2568 * Quit if directory has been removed (posix) 2569 */ 2570 if ((*eofp = zp->z_unlinked) != 0) { 2571 ZFS_EXIT(zfsvfs); 2572 return (0); 2573 } 2574 2575 error = 0; 2576 os = zfsvfs->z_os; 2577 offset = uio->uio_loffset; 2578 prefetch = zp->z_zn_prefetch; 2579 2580 /* 2581 * Initialize the iterator cursor. 2582 */ 2583 if (offset <= 3) { 2584 /* 2585 * Start iteration from the beginning of the directory. 2586 */ 2587 zap_cursor_init(&zc, os, zp->z_id); 2588 } else { 2589 /* 2590 * The offset is a serialized cursor. 2591 */ 2592 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 2593 } 2594 2595 /* 2596 * Get space to change directory entries into fs independent format. 2597 */ 2598 iovp = uio->uio_iov; 2599 bytes_wanted = iovp->iov_len; 2600 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 2601 bufsize = bytes_wanted; 2602 outbuf = kmem_alloc(bufsize, KM_SLEEP); 2603 odp = (struct dirent64 *)outbuf; 2604 } else { 2605 bufsize = bytes_wanted; 2606 outbuf = NULL; 2607 odp = (struct dirent64 *)iovp->iov_base; 2608 } 2609 eodp = (struct edirent *)odp; 2610 2611 if (ncookies != NULL) { 2612 /* 2613 * Minimum entry size is dirent size and 1 byte for a file name. 2614 */ 2615 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 2616 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 2617 *cookies = cooks; 2618 *ncookies = ncooks; 2619 } 2620 /* 2621 * If this VFS supports the system attribute view interface; and 2622 * we're looking at an extended attribute directory; and we care 2623 * about normalization conflicts on this vfs; then we must check 2624 * for normalization conflicts with the sysattr name space. 2625 */ 2626#ifdef TODO 2627 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 2628 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 2629 (flags & V_RDDIR_ENTFLAGS); 2630#else 2631 check_sysattrs = 0; 2632#endif 2633 2634 /* 2635 * Transform to file-system independent format 2636 */ 2637 outcount = 0; 2638 while (outcount < bytes_wanted) { 2639 ino64_t objnum; 2640 ushort_t reclen; 2641 off64_t *next = NULL; 2642 2643 /* 2644 * Special case `.', `..', and `.zfs'. 2645 */ 2646 if (offset == 0) { 2647 (void) strcpy(zap.za_name, "."); 2648 zap.za_normalization_conflict = 0; 2649 objnum = zp->z_id; 2650 type = DT_DIR; 2651 } else if (offset == 1) { 2652 (void) strcpy(zap.za_name, ".."); 2653 zap.za_normalization_conflict = 0; 2654 objnum = parent; 2655 type = DT_DIR; 2656 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2657 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2658 zap.za_normalization_conflict = 0; 2659 objnum = ZFSCTL_INO_ROOT; 2660 type = DT_DIR; 2661 } else { 2662 /* 2663 * Grab next entry. 2664 */ 2665 if (error = zap_cursor_retrieve(&zc, &zap)) { 2666 if ((*eofp = (error == ENOENT)) != 0) 2667 break; 2668 else 2669 goto update; 2670 } 2671 2672 if (zap.za_integer_length != 8 || 2673 zap.za_num_integers != 1) { 2674 cmn_err(CE_WARN, "zap_readdir: bad directory " 2675 "entry, obj = %lld, offset = %lld\n", 2676 (u_longlong_t)zp->z_id, 2677 (u_longlong_t)offset); 2678 error = SET_ERROR(ENXIO); 2679 goto update; 2680 } 2681 2682 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2683 /* 2684 * MacOS X can extract the object type here such as: 2685 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2686 */ 2687 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2688 2689 if (check_sysattrs && !zap.za_normalization_conflict) { 2690#ifdef TODO 2691 zap.za_normalization_conflict = 2692 xattr_sysattr_casechk(zap.za_name); 2693#else 2694 panic("%s:%u: TODO", __func__, __LINE__); 2695#endif 2696 } 2697 } 2698 2699 if (flags & V_RDDIR_ACCFILTER) { 2700 /* 2701 * If we have no access at all, don't include 2702 * this entry in the returned information 2703 */ 2704 znode_t *ezp; 2705 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2706 goto skip_entry; 2707 if (!zfs_has_access(ezp, cr)) { 2708 VN_RELE(ZTOV(ezp)); 2709 goto skip_entry; 2710 } 2711 VN_RELE(ZTOV(ezp)); 2712 } 2713 2714 if (flags & V_RDDIR_ENTFLAGS) 2715 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2716 else 2717 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2718 2719 /* 2720 * Will this entry fit in the buffer? 2721 */ 2722 if (outcount + reclen > bufsize) { 2723 /* 2724 * Did we manage to fit anything in the buffer? 2725 */ 2726 if (!outcount) { 2727 error = SET_ERROR(EINVAL); 2728 goto update; 2729 } 2730 break; 2731 } 2732 if (flags & V_RDDIR_ENTFLAGS) { 2733 /* 2734 * Add extended flag entry: 2735 */ 2736 eodp->ed_ino = objnum; 2737 eodp->ed_reclen = reclen; 2738 /* NOTE: ed_off is the offset for the *next* entry */ 2739 next = &(eodp->ed_off); 2740 eodp->ed_eflags = zap.za_normalization_conflict ? 2741 ED_CASE_CONFLICT : 0; 2742 (void) strncpy(eodp->ed_name, zap.za_name, 2743 EDIRENT_NAMELEN(reclen)); 2744 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2745 } else { 2746 /* 2747 * Add normal entry: 2748 */ 2749 odp->d_ino = objnum; 2750 odp->d_reclen = reclen; 2751 odp->d_namlen = strlen(zap.za_name); 2752 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 2753 odp->d_type = type; 2754 odp = (dirent64_t *)((intptr_t)odp + reclen); 2755 } 2756 outcount += reclen; 2757 2758 ASSERT(outcount <= bufsize); 2759 2760 /* Prefetch znode */ 2761 if (prefetch) 2762 dmu_prefetch(os, objnum, 0, 0, 0, 2763 ZIO_PRIORITY_SYNC_READ); 2764 2765 skip_entry: 2766 /* 2767 * Move to the next entry, fill in the previous offset. 2768 */ 2769 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2770 zap_cursor_advance(&zc); 2771 offset = zap_cursor_serialize(&zc); 2772 } else { 2773 offset += 1; 2774 } 2775 2776 if (cooks != NULL) { 2777 *cooks++ = offset; 2778 ncooks--; 2779 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 2780 } 2781 } 2782 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2783 2784 /* Subtract unused cookies */ 2785 if (ncookies != NULL) 2786 *ncookies -= ncooks; 2787 2788 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2789 iovp->iov_base += outcount; 2790 iovp->iov_len -= outcount; 2791 uio->uio_resid -= outcount; 2792 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2793 /* 2794 * Reset the pointer. 2795 */ 2796 offset = uio->uio_loffset; 2797 } 2798 2799update: 2800 zap_cursor_fini(&zc); 2801 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2802 kmem_free(outbuf, bufsize); 2803 2804 if (error == ENOENT) 2805 error = 0; 2806 2807 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2808 2809 uio->uio_loffset = offset; 2810 ZFS_EXIT(zfsvfs); 2811 if (error != 0 && cookies != NULL) { 2812 free(*cookies, M_TEMP); 2813 *cookies = NULL; 2814 *ncookies = 0; 2815 } 2816 return (error); 2817} 2818 2819ulong_t zfs_fsync_sync_cnt = 4; 2820 2821static int 2822zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2823{ 2824 znode_t *zp = VTOZ(vp); 2825 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2826 2827 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2828 2829 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { 2830 ZFS_ENTER(zfsvfs); 2831 ZFS_VERIFY_ZP(zp); 2832 zil_commit(zfsvfs->z_log, zp->z_id); 2833 ZFS_EXIT(zfsvfs); 2834 } 2835 return (0); 2836} 2837 2838 2839/* 2840 * Get the requested file attributes and place them in the provided 2841 * vattr structure. 2842 * 2843 * IN: vp - vnode of file. 2844 * vap - va_mask identifies requested attributes. 2845 * If AT_XVATTR set, then optional attrs are requested 2846 * flags - ATTR_NOACLCHECK (CIFS server context) 2847 * cr - credentials of caller. 2848 * ct - caller context 2849 * 2850 * OUT: vap - attribute values. 2851 * 2852 * RETURN: 0 (always succeeds). 2853 */ 2854/* ARGSUSED */ 2855static int 2856zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2857 caller_context_t *ct) 2858{ 2859 znode_t *zp = VTOZ(vp); 2860 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2861 int error = 0; 2862 uint32_t blksize; 2863 u_longlong_t nblocks; 2864 uint64_t links; 2865 uint64_t mtime[2], ctime[2], crtime[2], rdev; 2866 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2867 xoptattr_t *xoap = NULL; 2868 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2869 sa_bulk_attr_t bulk[4]; 2870 int count = 0; 2871 2872 ZFS_ENTER(zfsvfs); 2873 ZFS_VERIFY_ZP(zp); 2874 2875 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2876 2877 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 2878 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 2879 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16); 2880 if (vp->v_type == VBLK || vp->v_type == VCHR) 2881 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, 2882 &rdev, 8); 2883 2884 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 2885 ZFS_EXIT(zfsvfs); 2886 return (error); 2887 } 2888 2889 /* 2890 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2891 * Also, if we are the owner don't bother, since owner should 2892 * always be allowed to read basic attributes of file. 2893 */ 2894 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 2895 (vap->va_uid != crgetuid(cr))) { 2896 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2897 skipaclchk, cr)) { 2898 ZFS_EXIT(zfsvfs); 2899 return (error); 2900 } 2901 } 2902 2903 /* 2904 * Return all attributes. It's cheaper to provide the answer 2905 * than to determine whether we were asked the question. 2906 */ 2907 2908 mutex_enter(&zp->z_lock); 2909 vap->va_type = IFTOVT(zp->z_mode); 2910 vap->va_mode = zp->z_mode & ~S_IFMT; 2911#ifdef illumos 2912 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2913#else 2914 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 2915#endif 2916 vap->va_nodeid = zp->z_id; 2917 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2918 links = zp->z_links + 1; 2919 else 2920 links = zp->z_links; 2921 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */ 2922 vap->va_size = zp->z_size; 2923#ifdef illumos 2924 vap->va_rdev = vp->v_rdev; 2925#else 2926 if (vp->v_type == VBLK || vp->v_type == VCHR) 2927 vap->va_rdev = zfs_cmpldev(rdev); 2928#endif 2929 vap->va_seq = zp->z_seq; 2930 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2931 vap->va_filerev = zp->z_seq; 2932 2933 /* 2934 * Add in any requested optional attributes and the create time. 2935 * Also set the corresponding bits in the returned attribute bitmap. 2936 */ 2937 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2938 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2939 xoap->xoa_archive = 2940 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2941 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2942 } 2943 2944 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2945 xoap->xoa_readonly = 2946 ((zp->z_pflags & ZFS_READONLY) != 0); 2947 XVA_SET_RTN(xvap, XAT_READONLY); 2948 } 2949 2950 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2951 xoap->xoa_system = 2952 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2953 XVA_SET_RTN(xvap, XAT_SYSTEM); 2954 } 2955 2956 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2957 xoap->xoa_hidden = 2958 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2959 XVA_SET_RTN(xvap, XAT_HIDDEN); 2960 } 2961 2962 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2963 xoap->xoa_nounlink = 2964 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2965 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2966 } 2967 2968 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2969 xoap->xoa_immutable = 2970 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2971 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2972 } 2973 2974 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2975 xoap->xoa_appendonly = 2976 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2977 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2978 } 2979 2980 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2981 xoap->xoa_nodump = 2982 ((zp->z_pflags & ZFS_NODUMP) != 0); 2983 XVA_SET_RTN(xvap, XAT_NODUMP); 2984 } 2985 2986 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2987 xoap->xoa_opaque = 2988 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2989 XVA_SET_RTN(xvap, XAT_OPAQUE); 2990 } 2991 2992 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2993 xoap->xoa_av_quarantined = 2994 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2995 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2996 } 2997 2998 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2999 xoap->xoa_av_modified = 3000 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 3001 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 3002 } 3003 3004 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 3005 vp->v_type == VREG) { 3006 zfs_sa_get_scanstamp(zp, xvap); 3007 } 3008 3009 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 3010 uint64_t times[2]; 3011 3012 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 3013 times, sizeof (times)); 3014 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 3015 XVA_SET_RTN(xvap, XAT_CREATETIME); 3016 } 3017 3018 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 3019 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 3020 XVA_SET_RTN(xvap, XAT_REPARSE); 3021 } 3022 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 3023 xoap->xoa_generation = zp->z_gen; 3024 XVA_SET_RTN(xvap, XAT_GEN); 3025 } 3026 3027 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 3028 xoap->xoa_offline = 3029 ((zp->z_pflags & ZFS_OFFLINE) != 0); 3030 XVA_SET_RTN(xvap, XAT_OFFLINE); 3031 } 3032 3033 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 3034 xoap->xoa_sparse = 3035 ((zp->z_pflags & ZFS_SPARSE) != 0); 3036 XVA_SET_RTN(xvap, XAT_SPARSE); 3037 } 3038 } 3039 3040 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 3041 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 3042 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 3043 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 3044 3045 mutex_exit(&zp->z_lock); 3046 3047 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 3048 vap->va_blksize = blksize; 3049 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 3050 3051 if (zp->z_blksz == 0) { 3052 /* 3053 * Block size hasn't been set; suggest maximal I/O transfers. 3054 */ 3055 vap->va_blksize = zfsvfs->z_max_blksz; 3056 } 3057 3058 ZFS_EXIT(zfsvfs); 3059 return (0); 3060} 3061 3062/* 3063 * Set the file attributes to the values contained in the 3064 * vattr structure. 3065 * 3066 * IN: vp - vnode of file to be modified. 3067 * vap - new attribute values. 3068 * If AT_XVATTR set, then optional attrs are being set 3069 * flags - ATTR_UTIME set if non-default time values provided. 3070 * - ATTR_NOACLCHECK (CIFS context only). 3071 * cr - credentials of caller. 3072 * ct - caller context 3073 * 3074 * RETURN: 0 on success, error code on failure. 3075 * 3076 * Timestamps: 3077 * vp - ctime updated, mtime updated if size changed. 3078 */ 3079/* ARGSUSED */ 3080static int 3081zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 3082 caller_context_t *ct) 3083{ 3084 znode_t *zp = VTOZ(vp); 3085 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3086 zilog_t *zilog; 3087 dmu_tx_t *tx; 3088 vattr_t oldva; 3089 xvattr_t tmpxvattr; 3090 uint_t mask = vap->va_mask; 3091 uint_t saved_mask = 0; 3092 uint64_t saved_mode; 3093 int trim_mask = 0; 3094 uint64_t new_mode; 3095 uint64_t new_uid, new_gid; 3096 uint64_t xattr_obj; 3097 uint64_t mtime[2], ctime[2]; 3098 znode_t *attrzp; 3099 int need_policy = FALSE; 3100 int err, err2; 3101 zfs_fuid_info_t *fuidp = NULL; 3102 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 3103 xoptattr_t *xoap; 3104 zfs_acl_t *aclp; 3105 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 3106 boolean_t fuid_dirtied = B_FALSE; 3107 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 3108 int count = 0, xattr_count = 0; 3109 3110 if (mask == 0) 3111 return (0); 3112 3113 if (mask & AT_NOSET) 3114 return (SET_ERROR(EINVAL)); 3115 3116 ZFS_ENTER(zfsvfs); 3117 ZFS_VERIFY_ZP(zp); 3118 3119 zilog = zfsvfs->z_log; 3120 3121 /* 3122 * Make sure that if we have ephemeral uid/gid or xvattr specified 3123 * that file system is at proper version level 3124 */ 3125 3126 if (zfsvfs->z_use_fuids == B_FALSE && 3127 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 3128 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 3129 (mask & AT_XVATTR))) { 3130 ZFS_EXIT(zfsvfs); 3131 return (SET_ERROR(EINVAL)); 3132 } 3133 3134 if (mask & AT_SIZE && vp->v_type == VDIR) { 3135 ZFS_EXIT(zfsvfs); 3136 return (SET_ERROR(EISDIR)); 3137 } 3138 3139 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 3140 ZFS_EXIT(zfsvfs); 3141 return (SET_ERROR(EINVAL)); 3142 } 3143 3144 /* 3145 * If this is an xvattr_t, then get a pointer to the structure of 3146 * optional attributes. If this is NULL, then we have a vattr_t. 3147 */ 3148 xoap = xva_getxoptattr(xvap); 3149 3150 xva_init(&tmpxvattr); 3151 3152 /* 3153 * Immutable files can only alter immutable bit and atime 3154 */ 3155 if ((zp->z_pflags & ZFS_IMMUTABLE) && 3156 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 3157 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 3158 ZFS_EXIT(zfsvfs); 3159 return (SET_ERROR(EPERM)); 3160 } 3161 3162 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 3163 ZFS_EXIT(zfsvfs); 3164 return (SET_ERROR(EPERM)); 3165 } 3166 3167 /* 3168 * Verify timestamps doesn't overflow 32 bits. 3169 * ZFS can handle large timestamps, but 32bit syscalls can't 3170 * handle times greater than 2039. This check should be removed 3171 * once large timestamps are fully supported. 3172 */ 3173 if (mask & (AT_ATIME | AT_MTIME)) { 3174 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 3175 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 3176 ZFS_EXIT(zfsvfs); 3177 return (SET_ERROR(EOVERFLOW)); 3178 } 3179 } 3180 3181top: 3182 attrzp = NULL; 3183 aclp = NULL; 3184 3185 /* Can this be moved to before the top label? */ 3186 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 3187 ZFS_EXIT(zfsvfs); 3188 return (SET_ERROR(EROFS)); 3189 } 3190 3191 /* 3192 * First validate permissions 3193 */ 3194 3195 if (mask & AT_SIZE) { 3196 /* 3197 * XXX - Note, we are not providing any open 3198 * mode flags here (like FNDELAY), so we may 3199 * block if there are locks present... this 3200 * should be addressed in openat(). 3201 */ 3202 /* XXX - would it be OK to generate a log record here? */ 3203 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 3204 if (err) { 3205 ZFS_EXIT(zfsvfs); 3206 return (err); 3207 } 3208 } 3209 3210 if (mask & (AT_ATIME|AT_MTIME) || 3211 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 3212 XVA_ISSET_REQ(xvap, XAT_READONLY) || 3213 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 3214 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 3215 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 3216 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 3217 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 3218 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 3219 skipaclchk, cr); 3220 } 3221 3222 if (mask & (AT_UID|AT_GID)) { 3223 int idmask = (mask & (AT_UID|AT_GID)); 3224 int take_owner; 3225 int take_group; 3226 3227 /* 3228 * NOTE: even if a new mode is being set, 3229 * we may clear S_ISUID/S_ISGID bits. 3230 */ 3231 3232 if (!(mask & AT_MODE)) 3233 vap->va_mode = zp->z_mode; 3234 3235 /* 3236 * Take ownership or chgrp to group we are a member of 3237 */ 3238 3239 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 3240 take_group = (mask & AT_GID) && 3241 zfs_groupmember(zfsvfs, vap->va_gid, cr); 3242 3243 /* 3244 * If both AT_UID and AT_GID are set then take_owner and 3245 * take_group must both be set in order to allow taking 3246 * ownership. 3247 * 3248 * Otherwise, send the check through secpolicy_vnode_setattr() 3249 * 3250 */ 3251 3252 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 3253 ((idmask == AT_UID) && take_owner) || 3254 ((idmask == AT_GID) && take_group)) { 3255 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 3256 skipaclchk, cr) == 0) { 3257 /* 3258 * Remove setuid/setgid for non-privileged users 3259 */ 3260 secpolicy_setid_clear(vap, vp, cr); 3261 trim_mask = (mask & (AT_UID|AT_GID)); 3262 } else { 3263 need_policy = TRUE; 3264 } 3265 } else { 3266 need_policy = TRUE; 3267 } 3268 } 3269 3270 mutex_enter(&zp->z_lock); 3271 oldva.va_mode = zp->z_mode; 3272 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 3273 if (mask & AT_XVATTR) { 3274 /* 3275 * Update xvattr mask to include only those attributes 3276 * that are actually changing. 3277 * 3278 * the bits will be restored prior to actually setting 3279 * the attributes so the caller thinks they were set. 3280 */ 3281 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 3282 if (xoap->xoa_appendonly != 3283 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 3284 need_policy = TRUE; 3285 } else { 3286 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 3287 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 3288 } 3289 } 3290 3291 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 3292 if (xoap->xoa_nounlink != 3293 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 3294 need_policy = TRUE; 3295 } else { 3296 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 3297 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 3298 } 3299 } 3300 3301 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 3302 if (xoap->xoa_immutable != 3303 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 3304 need_policy = TRUE; 3305 } else { 3306 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 3307 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 3308 } 3309 } 3310 3311 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 3312 if (xoap->xoa_nodump != 3313 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 3314 need_policy = TRUE; 3315 } else { 3316 XVA_CLR_REQ(xvap, XAT_NODUMP); 3317 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 3318 } 3319 } 3320 3321 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 3322 if (xoap->xoa_av_modified != 3323 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 3324 need_policy = TRUE; 3325 } else { 3326 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 3327 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 3328 } 3329 } 3330 3331 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 3332 if ((vp->v_type != VREG && 3333 xoap->xoa_av_quarantined) || 3334 xoap->xoa_av_quarantined != 3335 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 3336 need_policy = TRUE; 3337 } else { 3338 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 3339 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 3340 } 3341 } 3342 3343 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 3344 mutex_exit(&zp->z_lock); 3345 ZFS_EXIT(zfsvfs); 3346 return (SET_ERROR(EPERM)); 3347 } 3348 3349 if (need_policy == FALSE && 3350 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 3351 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 3352 need_policy = TRUE; 3353 } 3354 } 3355 3356 mutex_exit(&zp->z_lock); 3357 3358 if (mask & AT_MODE) { 3359 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 3360 err = secpolicy_setid_setsticky_clear(vp, vap, 3361 &oldva, cr); 3362 if (err) { 3363 ZFS_EXIT(zfsvfs); 3364 return (err); 3365 } 3366 trim_mask |= AT_MODE; 3367 } else { 3368 need_policy = TRUE; 3369 } 3370 } 3371 3372 if (need_policy) { 3373 /* 3374 * If trim_mask is set then take ownership 3375 * has been granted or write_acl is present and user 3376 * has the ability to modify mode. In that case remove 3377 * UID|GID and or MODE from mask so that 3378 * secpolicy_vnode_setattr() doesn't revoke it. 3379 */ 3380 3381 if (trim_mask) { 3382 saved_mask = vap->va_mask; 3383 vap->va_mask &= ~trim_mask; 3384 if (trim_mask & AT_MODE) { 3385 /* 3386 * Save the mode, as secpolicy_vnode_setattr() 3387 * will overwrite it with ova.va_mode. 3388 */ 3389 saved_mode = vap->va_mode; 3390 } 3391 } 3392 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 3393 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 3394 if (err) { 3395 ZFS_EXIT(zfsvfs); 3396 return (err); 3397 } 3398 3399 if (trim_mask) { 3400 vap->va_mask |= saved_mask; 3401 if (trim_mask & AT_MODE) { 3402 /* 3403 * Recover the mode after 3404 * secpolicy_vnode_setattr(). 3405 */ 3406 vap->va_mode = saved_mode; 3407 } 3408 } 3409 } 3410 3411 /* 3412 * secpolicy_vnode_setattr, or take ownership may have 3413 * changed va_mask 3414 */ 3415 mask = vap->va_mask; 3416 3417 if ((mask & (AT_UID | AT_GID))) { 3418 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 3419 &xattr_obj, sizeof (xattr_obj)); 3420 3421 if (err == 0 && xattr_obj) { 3422 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 3423 if (err) 3424 goto out2; 3425 } 3426 if (mask & AT_UID) { 3427 new_uid = zfs_fuid_create(zfsvfs, 3428 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 3429 if (new_uid != zp->z_uid && 3430 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 3431 if (attrzp) 3432 VN_RELE(ZTOV(attrzp)); 3433 err = SET_ERROR(EDQUOT); 3434 goto out2; 3435 } 3436 } 3437 3438 if (mask & AT_GID) { 3439 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 3440 cr, ZFS_GROUP, &fuidp); 3441 if (new_gid != zp->z_gid && 3442 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 3443 if (attrzp) 3444 VN_RELE(ZTOV(attrzp)); 3445 err = SET_ERROR(EDQUOT); 3446 goto out2; 3447 } 3448 } 3449 } 3450 tx = dmu_tx_create(zfsvfs->z_os); 3451 3452 if (mask & AT_MODE) { 3453 uint64_t pmode = zp->z_mode; 3454 uint64_t acl_obj; 3455 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 3456 3457 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && 3458 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 3459 err = SET_ERROR(EPERM); 3460 goto out; 3461 } 3462 3463 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 3464 goto out; 3465 3466 mutex_enter(&zp->z_lock); 3467 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 3468 /* 3469 * Are we upgrading ACL from old V0 format 3470 * to V1 format? 3471 */ 3472 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 3473 zfs_znode_acl_version(zp) == 3474 ZFS_ACL_VERSION_INITIAL) { 3475 dmu_tx_hold_free(tx, acl_obj, 0, 3476 DMU_OBJECT_END); 3477 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3478 0, aclp->z_acl_bytes); 3479 } else { 3480 dmu_tx_hold_write(tx, acl_obj, 0, 3481 aclp->z_acl_bytes); 3482 } 3483 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3484 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3485 0, aclp->z_acl_bytes); 3486 } 3487 mutex_exit(&zp->z_lock); 3488 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3489 } else { 3490 if ((mask & AT_XVATTR) && 3491 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3492 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3493 else 3494 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3495 } 3496 3497 if (attrzp) { 3498 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 3499 } 3500 3501 fuid_dirtied = zfsvfs->z_fuid_dirty; 3502 if (fuid_dirtied) 3503 zfs_fuid_txhold(zfsvfs, tx); 3504 3505 zfs_sa_upgrade_txholds(tx, zp); 3506 3507 err = dmu_tx_assign(tx, TXG_WAIT); 3508 if (err) 3509 goto out; 3510 3511 count = 0; 3512 /* 3513 * Set each attribute requested. 3514 * We group settings according to the locks they need to acquire. 3515 * 3516 * Note: you cannot set ctime directly, although it will be 3517 * updated as a side-effect of calling this function. 3518 */ 3519 3520 3521 if (mask & (AT_UID|AT_GID|AT_MODE)) 3522 mutex_enter(&zp->z_acl_lock); 3523 mutex_enter(&zp->z_lock); 3524 3525 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3526 &zp->z_pflags, sizeof (zp->z_pflags)); 3527 3528 if (attrzp) { 3529 if (mask & (AT_UID|AT_GID|AT_MODE)) 3530 mutex_enter(&attrzp->z_acl_lock); 3531 mutex_enter(&attrzp->z_lock); 3532 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3533 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3534 sizeof (attrzp->z_pflags)); 3535 } 3536 3537 if (mask & (AT_UID|AT_GID)) { 3538 3539 if (mask & AT_UID) { 3540 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3541 &new_uid, sizeof (new_uid)); 3542 zp->z_uid = new_uid; 3543 if (attrzp) { 3544 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3545 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3546 sizeof (new_uid)); 3547 attrzp->z_uid = new_uid; 3548 } 3549 } 3550 3551 if (mask & AT_GID) { 3552 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3553 NULL, &new_gid, sizeof (new_gid)); 3554 zp->z_gid = new_gid; 3555 if (attrzp) { 3556 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3557 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 3558 sizeof (new_gid)); 3559 attrzp->z_gid = new_gid; 3560 } 3561 } 3562 if (!(mask & AT_MODE)) { 3563 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 3564 NULL, &new_mode, sizeof (new_mode)); 3565 new_mode = zp->z_mode; 3566 } 3567 err = zfs_acl_chown_setattr(zp); 3568 ASSERT(err == 0); 3569 if (attrzp) { 3570 err = zfs_acl_chown_setattr(attrzp); 3571 ASSERT(err == 0); 3572 } 3573 } 3574 3575 if (mask & AT_MODE) { 3576 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 3577 &new_mode, sizeof (new_mode)); 3578 zp->z_mode = new_mode; 3579 ASSERT3U((uintptr_t)aclp, !=, 0); 3580 err = zfs_aclset_common(zp, aclp, cr, tx); 3581 ASSERT0(err); 3582 if (zp->z_acl_cached) 3583 zfs_acl_free(zp->z_acl_cached); 3584 zp->z_acl_cached = aclp; 3585 aclp = NULL; 3586 } 3587 3588 3589 if (mask & AT_ATIME) { 3590 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 3591 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 3592 &zp->z_atime, sizeof (zp->z_atime)); 3593 } 3594 3595 if (mask & AT_MTIME) { 3596 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 3597 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 3598 mtime, sizeof (mtime)); 3599 } 3600 3601 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 3602 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 3603 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 3604 NULL, mtime, sizeof (mtime)); 3605 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3606 &ctime, sizeof (ctime)); 3607 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 3608 B_TRUE); 3609 } else if (mask != 0) { 3610 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3611 &ctime, sizeof (ctime)); 3612 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime, 3613 B_TRUE); 3614 if (attrzp) { 3615 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3616 SA_ZPL_CTIME(zfsvfs), NULL, 3617 &ctime, sizeof (ctime)); 3618 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 3619 mtime, ctime, B_TRUE); 3620 } 3621 } 3622 /* 3623 * Do this after setting timestamps to prevent timestamp 3624 * update from toggling bit 3625 */ 3626 3627 if (xoap && (mask & AT_XVATTR)) { 3628 3629 /* 3630 * restore trimmed off masks 3631 * so that return masks can be set for caller. 3632 */ 3633 3634 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 3635 XVA_SET_REQ(xvap, XAT_APPENDONLY); 3636 } 3637 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 3638 XVA_SET_REQ(xvap, XAT_NOUNLINK); 3639 } 3640 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 3641 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 3642 } 3643 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 3644 XVA_SET_REQ(xvap, XAT_NODUMP); 3645 } 3646 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 3647 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 3648 } 3649 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 3650 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 3651 } 3652 3653 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3654 ASSERT(vp->v_type == VREG); 3655 3656 zfs_xvattr_set(zp, xvap, tx); 3657 } 3658 3659 if (fuid_dirtied) 3660 zfs_fuid_sync(zfsvfs, tx); 3661 3662 if (mask != 0) 3663 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 3664 3665 mutex_exit(&zp->z_lock); 3666 if (mask & (AT_UID|AT_GID|AT_MODE)) 3667 mutex_exit(&zp->z_acl_lock); 3668 3669 if (attrzp) { 3670 if (mask & (AT_UID|AT_GID|AT_MODE)) 3671 mutex_exit(&attrzp->z_acl_lock); 3672 mutex_exit(&attrzp->z_lock); 3673 } 3674out: 3675 if (err == 0 && attrzp) { 3676 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 3677 xattr_count, tx); 3678 ASSERT(err2 == 0); 3679 } 3680 3681 if (attrzp) 3682 VN_RELE(ZTOV(attrzp)); 3683 3684 if (aclp) 3685 zfs_acl_free(aclp); 3686 3687 if (fuidp) { 3688 zfs_fuid_info_free(fuidp); 3689 fuidp = NULL; 3690 } 3691 3692 if (err) { 3693 dmu_tx_abort(tx); 3694 if (err == ERESTART) 3695 goto top; 3696 } else { 3697 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3698 dmu_tx_commit(tx); 3699 } 3700 3701out2: 3702 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3703 zil_commit(zilog, 0); 3704 3705 ZFS_EXIT(zfsvfs); 3706 return (err); 3707} 3708 3709typedef struct zfs_zlock { 3710 krwlock_t *zl_rwlock; /* lock we acquired */ 3711 znode_t *zl_znode; /* znode we held */ 3712 struct zfs_zlock *zl_next; /* next in list */ 3713} zfs_zlock_t; 3714 3715/* 3716 * Drop locks and release vnodes that were held by zfs_rename_lock(). 3717 */ 3718static void 3719zfs_rename_unlock(zfs_zlock_t **zlpp) 3720{ 3721 zfs_zlock_t *zl; 3722 3723 while ((zl = *zlpp) != NULL) { 3724 if (zl->zl_znode != NULL) 3725 VN_RELE(ZTOV(zl->zl_znode)); 3726 rw_exit(zl->zl_rwlock); 3727 *zlpp = zl->zl_next; 3728 kmem_free(zl, sizeof (*zl)); 3729 } 3730} 3731 3732/* 3733 * Search back through the directory tree, using the ".." entries. 3734 * Lock each directory in the chain to prevent concurrent renames. 3735 * Fail any attempt to move a directory into one of its own descendants. 3736 * XXX - z_parent_lock can overlap with map or grow locks 3737 */ 3738static int 3739zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 3740{ 3741 zfs_zlock_t *zl; 3742 znode_t *zp = tdzp; 3743 uint64_t rootid = zp->z_zfsvfs->z_root; 3744 uint64_t oidp = zp->z_id; 3745 krwlock_t *rwlp = &szp->z_parent_lock; 3746 krw_t rw = RW_WRITER; 3747 3748 /* 3749 * First pass write-locks szp and compares to zp->z_id. 3750 * Later passes read-lock zp and compare to zp->z_parent. 3751 */ 3752 do { 3753 if (!rw_tryenter(rwlp, rw)) { 3754 /* 3755 * Another thread is renaming in this path. 3756 * Note that if we are a WRITER, we don't have any 3757 * parent_locks held yet. 3758 */ 3759 if (rw == RW_READER && zp->z_id > szp->z_id) { 3760 /* 3761 * Drop our locks and restart 3762 */ 3763 zfs_rename_unlock(&zl); 3764 *zlpp = NULL; 3765 zp = tdzp; 3766 oidp = zp->z_id; 3767 rwlp = &szp->z_parent_lock; 3768 rw = RW_WRITER; 3769 continue; 3770 } else { 3771 /* 3772 * Wait for other thread to drop its locks 3773 */ 3774 rw_enter(rwlp, rw); 3775 } 3776 } 3777 3778 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 3779 zl->zl_rwlock = rwlp; 3780 zl->zl_znode = NULL; 3781 zl->zl_next = *zlpp; 3782 *zlpp = zl; 3783 3784 if (oidp == szp->z_id) /* We're a descendant of szp */ 3785 return (SET_ERROR(EINVAL)); 3786 3787 if (oidp == rootid) /* We've hit the top */ 3788 return (0); 3789 3790 if (rw == RW_READER) { /* i.e. not the first pass */ 3791 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp); 3792 if (error) 3793 return (error); 3794 zl->zl_znode = zp; 3795 } 3796 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs), 3797 &oidp, sizeof (oidp)); 3798 rwlp = &zp->z_parent_lock; 3799 rw = RW_READER; 3800 3801 } while (zp->z_id != sdzp->z_id); 3802 3803 return (0); 3804} 3805 3806/* 3807 * Move an entry from the provided source directory to the target 3808 * directory. Change the entry name as indicated. 3809 * 3810 * IN: sdvp - Source directory containing the "old entry". 3811 * snm - Old entry name. 3812 * tdvp - Target directory to contain the "new entry". 3813 * tnm - New entry name. 3814 * cr - credentials of caller. 3815 * ct - caller context 3816 * flags - case flags 3817 * 3818 * RETURN: 0 on success, error code on failure. 3819 * 3820 * Timestamps: 3821 * sdvp,tdvp - ctime|mtime updated 3822 */ 3823/*ARGSUSED*/ 3824static int 3825zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 3826 caller_context_t *ct, int flags) 3827{ 3828 znode_t *tdzp, *sdzp, *szp, *tzp; 3829 zfsvfs_t *zfsvfs; 3830 zilog_t *zilog; 3831 vnode_t *realvp; 3832 zfs_dirlock_t *sdl, *tdl; 3833 dmu_tx_t *tx; 3834 zfs_zlock_t *zl; 3835 int cmp, serr, terr; 3836 int error = 0; 3837 int zflg = 0; 3838 boolean_t waited = B_FALSE; 3839 3840 tdzp = VTOZ(tdvp); 3841 ZFS_VERIFY_ZP(tdzp); 3842 zfsvfs = tdzp->z_zfsvfs; 3843 ZFS_ENTER(zfsvfs); 3844 zilog = zfsvfs->z_log; 3845 sdzp = VTOZ(sdvp); 3846 3847 /* 3848 * In case sdzp is not valid, let's be sure to exit from the right 3849 * zfsvfs_t. 3850 */ 3851 if (sdzp->z_sa_hdl == NULL) { 3852 ZFS_EXIT(zfsvfs); 3853 return (SET_ERROR(EIO)); 3854 } 3855 3856 /* 3857 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 3858 * ctldir appear to have the same v_vfsp. 3859 */ 3860 if (sdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) { 3861 ZFS_EXIT(zfsvfs); 3862 return (SET_ERROR(EXDEV)); 3863 } 3864 3865 if (zfsvfs->z_utf8 && u8_validate(tnm, 3866 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3867 ZFS_EXIT(zfsvfs); 3868 return (SET_ERROR(EILSEQ)); 3869 } 3870 3871 if (flags & FIGNORECASE) 3872 zflg |= ZCILOOK; 3873 3874top: 3875 szp = NULL; 3876 tzp = NULL; 3877 zl = NULL; 3878 3879 /* 3880 * This is to prevent the creation of links into attribute space 3881 * by renaming a linked file into/outof an attribute directory. 3882 * See the comment in zfs_link() for why this is considered bad. 3883 */ 3884 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3885 ZFS_EXIT(zfsvfs); 3886 return (SET_ERROR(EINVAL)); 3887 } 3888 3889 /* 3890 * Lock source and target directory entries. To prevent deadlock, 3891 * a lock ordering must be defined. We lock the directory with 3892 * the smallest object id first, or if it's a tie, the one with 3893 * the lexically first name. 3894 */ 3895 if (sdzp->z_id < tdzp->z_id) { 3896 cmp = -1; 3897 } else if (sdzp->z_id > tdzp->z_id) { 3898 cmp = 1; 3899 } else { 3900 /* 3901 * First compare the two name arguments without 3902 * considering any case folding. 3903 */ 3904 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 3905 3906 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 3907 ASSERT(error == 0 || !zfsvfs->z_utf8); 3908 if (cmp == 0) { 3909 /* 3910 * POSIX: "If the old argument and the new argument 3911 * both refer to links to the same existing file, 3912 * the rename() function shall return successfully 3913 * and perform no other action." 3914 */ 3915 ZFS_EXIT(zfsvfs); 3916 return (0); 3917 } 3918 /* 3919 * If the file system is case-folding, then we may 3920 * have some more checking to do. A case-folding file 3921 * system is either supporting mixed case sensitivity 3922 * access or is completely case-insensitive. Note 3923 * that the file system is always case preserving. 3924 * 3925 * In mixed sensitivity mode case sensitive behavior 3926 * is the default. FIGNORECASE must be used to 3927 * explicitly request case insensitive behavior. 3928 * 3929 * If the source and target names provided differ only 3930 * by case (e.g., a request to rename 'tim' to 'Tim'), 3931 * we will treat this as a special case in the 3932 * case-insensitive mode: as long as the source name 3933 * is an exact match, we will allow this to proceed as 3934 * a name-change request. 3935 */ 3936 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3937 (zfsvfs->z_case == ZFS_CASE_MIXED && 3938 flags & FIGNORECASE)) && 3939 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3940 &error) == 0) { 3941 /* 3942 * case preserving rename request, require exact 3943 * name matches 3944 */ 3945 zflg |= ZCIEXACT; 3946 zflg &= ~ZCILOOK; 3947 } 3948 } 3949 3950 /* 3951 * If the source and destination directories are the same, we should 3952 * grab the z_name_lock of that directory only once. 3953 */ 3954 if (sdzp == tdzp) { 3955 zflg |= ZHAVELOCK; 3956 rw_enter(&sdzp->z_name_lock, RW_READER); 3957 } 3958 3959 if (cmp < 0) { 3960 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3961 ZEXISTS | zflg, NULL, NULL); 3962 terr = zfs_dirent_lock(&tdl, 3963 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3964 } else { 3965 terr = zfs_dirent_lock(&tdl, 3966 tdzp, tnm, &tzp, zflg, NULL, NULL); 3967 serr = zfs_dirent_lock(&sdl, 3968 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3969 NULL, NULL); 3970 } 3971 3972 if (serr) { 3973 /* 3974 * Source entry invalid or not there. 3975 */ 3976 if (!terr) { 3977 zfs_dirent_unlock(tdl); 3978 if (tzp) 3979 VN_RELE(ZTOV(tzp)); 3980 } 3981 3982 if (sdzp == tdzp) 3983 rw_exit(&sdzp->z_name_lock); 3984 3985 /* 3986 * FreeBSD: In OpenSolaris they only check if rename source is 3987 * ".." here, because "." is handled in their lookup. This is 3988 * not the case for FreeBSD, so we check for "." explicitly. 3989 */ 3990 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 3991 serr = SET_ERROR(EINVAL); 3992 ZFS_EXIT(zfsvfs); 3993 return (serr); 3994 } 3995 if (terr) { 3996 zfs_dirent_unlock(sdl); 3997 VN_RELE(ZTOV(szp)); 3998 3999 if (sdzp == tdzp) 4000 rw_exit(&sdzp->z_name_lock); 4001 4002 if (strcmp(tnm, "..") == 0) 4003 terr = SET_ERROR(EINVAL); 4004 ZFS_EXIT(zfsvfs); 4005 return (terr); 4006 } 4007 4008 /* 4009 * Must have write access at the source to remove the old entry 4010 * and write access at the target to create the new entry. 4011 * Note that if target and source are the same, this can be 4012 * done in a single check. 4013 */ 4014 4015 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 4016 goto out; 4017 4018 if (ZTOV(szp)->v_type == VDIR) { 4019 /* 4020 * Check to make sure rename is valid. 4021 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 4022 */ 4023 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 4024 goto out; 4025 } 4026 4027 /* 4028 * Does target exist? 4029 */ 4030 if (tzp) { 4031 /* 4032 * Source and target must be the same type. 4033 */ 4034 if (ZTOV(szp)->v_type == VDIR) { 4035 if (ZTOV(tzp)->v_type != VDIR) { 4036 error = SET_ERROR(ENOTDIR); 4037 goto out; 4038 } 4039 } else { 4040 if (ZTOV(tzp)->v_type == VDIR) { 4041 error = SET_ERROR(EISDIR); 4042 goto out; 4043 } 4044 } 4045 /* 4046 * POSIX dictates that when the source and target 4047 * entries refer to the same file object, rename 4048 * must do nothing and exit without error. 4049 */ 4050 if (szp->z_id == tzp->z_id) { 4051 error = 0; 4052 goto out; 4053 } 4054 } 4055 4056 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 4057 if (tzp) 4058 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 4059 4060 /* 4061 * notify the target directory if it is not the same 4062 * as source directory. 4063 */ 4064 if (tdvp != sdvp) { 4065 vnevent_rename_dest_dir(tdvp, ct); 4066 } 4067 4068 tx = dmu_tx_create(zfsvfs->z_os); 4069 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4070 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 4071 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 4072 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 4073 if (sdzp != tdzp) { 4074 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 4075 zfs_sa_upgrade_txholds(tx, tdzp); 4076 } 4077 if (tzp) { 4078 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 4079 zfs_sa_upgrade_txholds(tx, tzp); 4080 } 4081 4082 zfs_sa_upgrade_txholds(tx, szp); 4083 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 4084 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4085 if (error) { 4086 if (zl != NULL) 4087 zfs_rename_unlock(&zl); 4088 zfs_dirent_unlock(sdl); 4089 zfs_dirent_unlock(tdl); 4090 4091 if (sdzp == tdzp) 4092 rw_exit(&sdzp->z_name_lock); 4093 4094 VN_RELE(ZTOV(szp)); 4095 if (tzp) 4096 VN_RELE(ZTOV(tzp)); 4097 if (error == ERESTART) { 4098 waited = B_TRUE; 4099 dmu_tx_wait(tx); 4100 dmu_tx_abort(tx); 4101 goto top; 4102 } 4103 dmu_tx_abort(tx); 4104 ZFS_EXIT(zfsvfs); 4105 return (error); 4106 } 4107 4108 if (tzp) /* Attempt to remove the existing target */ 4109 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 4110 4111 if (error == 0) { 4112 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 4113 if (error == 0) { 4114 szp->z_pflags |= ZFS_AV_MODIFIED; 4115 4116 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 4117 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 4118 ASSERT0(error); 4119 4120 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 4121 if (error == 0) { 4122 zfs_log_rename(zilog, tx, TX_RENAME | 4123 (flags & FIGNORECASE ? TX_CI : 0), sdzp, 4124 sdl->dl_name, tdzp, tdl->dl_name, szp); 4125 4126 /* 4127 * Update path information for the target vnode 4128 */ 4129 vn_renamepath(tdvp, ZTOV(szp), tnm, 4130 strlen(tnm)); 4131 } else { 4132 /* 4133 * At this point, we have successfully created 4134 * the target name, but have failed to remove 4135 * the source name. Since the create was done 4136 * with the ZRENAMING flag, there are 4137 * complications; for one, the link count is 4138 * wrong. The easiest way to deal with this 4139 * is to remove the newly created target, and 4140 * return the original error. This must 4141 * succeed; fortunately, it is very unlikely to 4142 * fail, since we just created it. 4143 */ 4144 VERIFY3U(zfs_link_destroy(tdl, szp, tx, 4145 ZRENAMING, NULL), ==, 0); 4146 } 4147 } 4148#ifdef FREEBSD_NAMECACHE 4149 if (error == 0) { 4150 cache_purge(sdvp); 4151 cache_purge(tdvp); 4152 cache_purge(ZTOV(szp)); 4153 if (tzp) 4154 cache_purge(ZTOV(tzp)); 4155 } 4156#endif 4157 } 4158 4159 dmu_tx_commit(tx); 4160out: 4161 if (zl != NULL) 4162 zfs_rename_unlock(&zl); 4163 4164 zfs_dirent_unlock(sdl); 4165 zfs_dirent_unlock(tdl); 4166 4167 if (sdzp == tdzp) 4168 rw_exit(&sdzp->z_name_lock); 4169 4170 4171 VN_RELE(ZTOV(szp)); 4172 if (tzp) 4173 VN_RELE(ZTOV(tzp)); 4174 4175 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4176 zil_commit(zilog, 0); 4177 4178 ZFS_EXIT(zfsvfs); 4179 4180 return (error); 4181} 4182 4183/* 4184 * Insert the indicated symbolic reference entry into the directory. 4185 * 4186 * IN: dvp - Directory to contain new symbolic link. 4187 * link - Name for new symlink entry. 4188 * vap - Attributes of new entry. 4189 * cr - credentials of caller. 4190 * ct - caller context 4191 * flags - case flags 4192 * 4193 * RETURN: 0 on success, error code on failure. 4194 * 4195 * Timestamps: 4196 * dvp - ctime|mtime updated 4197 */ 4198/*ARGSUSED*/ 4199static int 4200zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, 4201 cred_t *cr, kthread_t *td) 4202{ 4203 znode_t *zp, *dzp = VTOZ(dvp); 4204 zfs_dirlock_t *dl; 4205 dmu_tx_t *tx; 4206 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4207 zilog_t *zilog; 4208 uint64_t len = strlen(link); 4209 int error; 4210 int zflg = ZNEW; 4211 zfs_acl_ids_t acl_ids; 4212 boolean_t fuid_dirtied; 4213 uint64_t txtype = TX_SYMLINK; 4214 boolean_t waited = B_FALSE; 4215 int flags = 0; 4216 4217 ASSERT(vap->va_type == VLNK); 4218 4219 ZFS_ENTER(zfsvfs); 4220 ZFS_VERIFY_ZP(dzp); 4221 zilog = zfsvfs->z_log; 4222 4223 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 4224 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4225 ZFS_EXIT(zfsvfs); 4226 return (SET_ERROR(EILSEQ)); 4227 } 4228 if (flags & FIGNORECASE) 4229 zflg |= ZCILOOK; 4230 4231 if (len > MAXPATHLEN) { 4232 ZFS_EXIT(zfsvfs); 4233 return (SET_ERROR(ENAMETOOLONG)); 4234 } 4235 4236 if ((error = zfs_acl_ids_create(dzp, 0, 4237 vap, cr, NULL, &acl_ids)) != 0) { 4238 ZFS_EXIT(zfsvfs); 4239 return (error); 4240 } 4241 4242 getnewvnode_reserve(1); 4243 4244top: 4245 /* 4246 * Attempt to lock directory; fail if entry already exists. 4247 */ 4248 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 4249 if (error) { 4250 zfs_acl_ids_free(&acl_ids); 4251 getnewvnode_drop_reserve(); 4252 ZFS_EXIT(zfsvfs); 4253 return (error); 4254 } 4255 4256 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4257 zfs_acl_ids_free(&acl_ids); 4258 zfs_dirent_unlock(dl); 4259 getnewvnode_drop_reserve(); 4260 ZFS_EXIT(zfsvfs); 4261 return (error); 4262 } 4263 4264 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 4265 zfs_acl_ids_free(&acl_ids); 4266 zfs_dirent_unlock(dl); 4267 getnewvnode_drop_reserve(); 4268 ZFS_EXIT(zfsvfs); 4269 return (SET_ERROR(EDQUOT)); 4270 } 4271 tx = dmu_tx_create(zfsvfs->z_os); 4272 fuid_dirtied = zfsvfs->z_fuid_dirty; 4273 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 4274 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4275 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 4276 ZFS_SA_BASE_ATTR_SIZE + len); 4277 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 4278 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 4279 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 4280 acl_ids.z_aclp->z_acl_bytes); 4281 } 4282 if (fuid_dirtied) 4283 zfs_fuid_txhold(zfsvfs, tx); 4284 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4285 if (error) { 4286 zfs_dirent_unlock(dl); 4287 if (error == ERESTART) { 4288 waited = B_TRUE; 4289 dmu_tx_wait(tx); 4290 dmu_tx_abort(tx); 4291 goto top; 4292 } 4293 zfs_acl_ids_free(&acl_ids); 4294 dmu_tx_abort(tx); 4295 getnewvnode_drop_reserve(); 4296 ZFS_EXIT(zfsvfs); 4297 return (error); 4298 } 4299 4300 /* 4301 * Create a new object for the symlink. 4302 * for version 4 ZPL datsets the symlink will be an SA attribute 4303 */ 4304 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 4305 4306 if (fuid_dirtied) 4307 zfs_fuid_sync(zfsvfs, tx); 4308 4309 mutex_enter(&zp->z_lock); 4310 if (zp->z_is_sa) 4311 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 4312 link, len, tx); 4313 else 4314 zfs_sa_symlink(zp, link, len, tx); 4315 mutex_exit(&zp->z_lock); 4316 4317 zp->z_size = len; 4318 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 4319 &zp->z_size, sizeof (zp->z_size), tx); 4320 /* 4321 * Insert the new object into the directory. 4322 */ 4323 (void) zfs_link_create(dl, zp, tx, ZNEW); 4324 4325 if (flags & FIGNORECASE) 4326 txtype |= TX_CI; 4327 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 4328 *vpp = ZTOV(zp); 4329 4330 zfs_acl_ids_free(&acl_ids); 4331 4332 dmu_tx_commit(tx); 4333 4334 getnewvnode_drop_reserve(); 4335 4336 zfs_dirent_unlock(dl); 4337 4338 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4339 zil_commit(zilog, 0); 4340 4341 ZFS_EXIT(zfsvfs); 4342 return (error); 4343} 4344 4345/* 4346 * Return, in the buffer contained in the provided uio structure, 4347 * the symbolic path referred to by vp. 4348 * 4349 * IN: vp - vnode of symbolic link. 4350 * uio - structure to contain the link path. 4351 * cr - credentials of caller. 4352 * ct - caller context 4353 * 4354 * OUT: uio - structure containing the link path. 4355 * 4356 * RETURN: 0 on success, error code on failure. 4357 * 4358 * Timestamps: 4359 * vp - atime updated 4360 */ 4361/* ARGSUSED */ 4362static int 4363zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 4364{ 4365 znode_t *zp = VTOZ(vp); 4366 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4367 int error; 4368 4369 ZFS_ENTER(zfsvfs); 4370 ZFS_VERIFY_ZP(zp); 4371 4372 mutex_enter(&zp->z_lock); 4373 if (zp->z_is_sa) 4374 error = sa_lookup_uio(zp->z_sa_hdl, 4375 SA_ZPL_SYMLINK(zfsvfs), uio); 4376 else 4377 error = zfs_sa_readlink(zp, uio); 4378 mutex_exit(&zp->z_lock); 4379 4380 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4381 4382 ZFS_EXIT(zfsvfs); 4383 return (error); 4384} 4385 4386/* 4387 * Insert a new entry into directory tdvp referencing svp. 4388 * 4389 * IN: tdvp - Directory to contain new entry. 4390 * svp - vnode of new entry. 4391 * name - name of new entry. 4392 * cr - credentials of caller. 4393 * ct - caller context 4394 * 4395 * RETURN: 0 on success, error code on failure. 4396 * 4397 * Timestamps: 4398 * tdvp - ctime|mtime updated 4399 * svp - ctime updated 4400 */ 4401/* ARGSUSED */ 4402static int 4403zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 4404 caller_context_t *ct, int flags) 4405{ 4406 znode_t *dzp = VTOZ(tdvp); 4407 znode_t *tzp, *szp; 4408 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4409 zilog_t *zilog; 4410 zfs_dirlock_t *dl; 4411 dmu_tx_t *tx; 4412 vnode_t *realvp; 4413 int error; 4414 int zf = ZNEW; 4415 uint64_t parent; 4416 uid_t owner; 4417 boolean_t waited = B_FALSE; 4418 4419 ASSERT(tdvp->v_type == VDIR); 4420 4421 ZFS_ENTER(zfsvfs); 4422 ZFS_VERIFY_ZP(dzp); 4423 zilog = zfsvfs->z_log; 4424 4425 if (VOP_REALVP(svp, &realvp, ct) == 0) 4426 svp = realvp; 4427 4428 /* 4429 * POSIX dictates that we return EPERM here. 4430 * Better choices include ENOTSUP or EISDIR. 4431 */ 4432 if (svp->v_type == VDIR) { 4433 ZFS_EXIT(zfsvfs); 4434 return (SET_ERROR(EPERM)); 4435 } 4436 4437 szp = VTOZ(svp); 4438 ZFS_VERIFY_ZP(szp); 4439 4440 if (szp->z_pflags & (ZFS_APPENDONLY | ZFS_IMMUTABLE | ZFS_READONLY)) { 4441 ZFS_EXIT(zfsvfs); 4442 return (SET_ERROR(EPERM)); 4443 } 4444 4445 /* 4446 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 4447 * ctldir appear to have the same v_vfsp. 4448 */ 4449 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) { 4450 ZFS_EXIT(zfsvfs); 4451 return (SET_ERROR(EXDEV)); 4452 } 4453 4454 /* Prevent links to .zfs/shares files */ 4455 4456 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 4457 &parent, sizeof (uint64_t))) != 0) { 4458 ZFS_EXIT(zfsvfs); 4459 return (error); 4460 } 4461 if (parent == zfsvfs->z_shares_dir) { 4462 ZFS_EXIT(zfsvfs); 4463 return (SET_ERROR(EPERM)); 4464 } 4465 4466 if (zfsvfs->z_utf8 && u8_validate(name, 4467 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4468 ZFS_EXIT(zfsvfs); 4469 return (SET_ERROR(EILSEQ)); 4470 } 4471 if (flags & FIGNORECASE) 4472 zf |= ZCILOOK; 4473 4474 /* 4475 * We do not support links between attributes and non-attributes 4476 * because of the potential security risk of creating links 4477 * into "normal" file space in order to circumvent restrictions 4478 * imposed in attribute space. 4479 */ 4480 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 4481 ZFS_EXIT(zfsvfs); 4482 return (SET_ERROR(EINVAL)); 4483 } 4484 4485 4486 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 4487 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) { 4488 ZFS_EXIT(zfsvfs); 4489 return (SET_ERROR(EPERM)); 4490 } 4491 4492 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4493 ZFS_EXIT(zfsvfs); 4494 return (error); 4495 } 4496 4497top: 4498 /* 4499 * Attempt to lock directory; fail if entry already exists. 4500 */ 4501 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 4502 if (error) { 4503 ZFS_EXIT(zfsvfs); 4504 return (error); 4505 } 4506 4507 tx = dmu_tx_create(zfsvfs->z_os); 4508 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4509 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4510 zfs_sa_upgrade_txholds(tx, szp); 4511 zfs_sa_upgrade_txholds(tx, dzp); 4512 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4513 if (error) { 4514 zfs_dirent_unlock(dl); 4515 if (error == ERESTART) { 4516 waited = B_TRUE; 4517 dmu_tx_wait(tx); 4518 dmu_tx_abort(tx); 4519 goto top; 4520 } 4521 dmu_tx_abort(tx); 4522 ZFS_EXIT(zfsvfs); 4523 return (error); 4524 } 4525 4526 error = zfs_link_create(dl, szp, tx, 0); 4527 4528 if (error == 0) { 4529 uint64_t txtype = TX_LINK; 4530 if (flags & FIGNORECASE) 4531 txtype |= TX_CI; 4532 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 4533 } 4534 4535 dmu_tx_commit(tx); 4536 4537 zfs_dirent_unlock(dl); 4538 4539 if (error == 0) { 4540 vnevent_link(svp, ct); 4541 } 4542 4543 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4544 zil_commit(zilog, 0); 4545 4546 ZFS_EXIT(zfsvfs); 4547 return (error); 4548} 4549 4550#ifdef illumos 4551/* 4552 * zfs_null_putapage() is used when the file system has been force 4553 * unmounted. It just drops the pages. 4554 */ 4555/* ARGSUSED */ 4556static int 4557zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4558 size_t *lenp, int flags, cred_t *cr) 4559{ 4560 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 4561 return (0); 4562} 4563 4564/* 4565 * Push a page out to disk, klustering if possible. 4566 * 4567 * IN: vp - file to push page to. 4568 * pp - page to push. 4569 * flags - additional flags. 4570 * cr - credentials of caller. 4571 * 4572 * OUT: offp - start of range pushed. 4573 * lenp - len of range pushed. 4574 * 4575 * RETURN: 0 on success, error code on failure. 4576 * 4577 * NOTE: callers must have locked the page to be pushed. On 4578 * exit, the page (and all other pages in the kluster) must be 4579 * unlocked. 4580 */ 4581/* ARGSUSED */ 4582static int 4583zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4584 size_t *lenp, int flags, cred_t *cr) 4585{ 4586 znode_t *zp = VTOZ(vp); 4587 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4588 dmu_tx_t *tx; 4589 u_offset_t off, koff; 4590 size_t len, klen; 4591 int err; 4592 4593 off = pp->p_offset; 4594 len = PAGESIZE; 4595 /* 4596 * If our blocksize is bigger than the page size, try to kluster 4597 * multiple pages so that we write a full block (thus avoiding 4598 * a read-modify-write). 4599 */ 4600 if (off < zp->z_size && zp->z_blksz > PAGESIZE) { 4601 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 4602 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 4603 ASSERT(koff <= zp->z_size); 4604 if (koff + klen > zp->z_size) 4605 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); 4606 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 4607 } 4608 ASSERT3U(btop(len), ==, btopr(len)); 4609 4610 /* 4611 * Can't push pages past end-of-file. 4612 */ 4613 if (off >= zp->z_size) { 4614 /* ignore all pages */ 4615 err = 0; 4616 goto out; 4617 } else if (off + len > zp->z_size) { 4618 int npages = btopr(zp->z_size - off); 4619 page_t *trunc; 4620 4621 page_list_break(&pp, &trunc, npages); 4622 /* ignore pages past end of file */ 4623 if (trunc) 4624 pvn_write_done(trunc, flags); 4625 len = zp->z_size - off; 4626 } 4627 4628 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4629 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4630 err = SET_ERROR(EDQUOT); 4631 goto out; 4632 } 4633 tx = dmu_tx_create(zfsvfs->z_os); 4634 dmu_tx_hold_write(tx, zp->z_id, off, len); 4635 4636 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4637 zfs_sa_upgrade_txholds(tx, zp); 4638 err = dmu_tx_assign(tx, TXG_WAIT); 4639 if (err != 0) { 4640 dmu_tx_abort(tx); 4641 goto out; 4642 } 4643 4644 if (zp->z_blksz <= PAGESIZE) { 4645 caddr_t va = zfs_map_page(pp, S_READ); 4646 ASSERT3U(len, <=, PAGESIZE); 4647 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 4648 zfs_unmap_page(pp, va); 4649 } else { 4650 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 4651 } 4652 4653 if (err == 0) { 4654 uint64_t mtime[2], ctime[2]; 4655 sa_bulk_attr_t bulk[3]; 4656 int count = 0; 4657 4658 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4659 &mtime, 16); 4660 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4661 &ctime, 16); 4662 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4663 &zp->z_pflags, 8); 4664 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4665 B_TRUE); 4666 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4667 } 4668 dmu_tx_commit(tx); 4669 4670out: 4671 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 4672 if (offp) 4673 *offp = off; 4674 if (lenp) 4675 *lenp = len; 4676 4677 return (err); 4678} 4679 4680/* 4681 * Copy the portion of the file indicated from pages into the file. 4682 * The pages are stored in a page list attached to the files vnode. 4683 * 4684 * IN: vp - vnode of file to push page data to. 4685 * off - position in file to put data. 4686 * len - amount of data to write. 4687 * flags - flags to control the operation. 4688 * cr - credentials of caller. 4689 * ct - caller context. 4690 * 4691 * RETURN: 0 on success, error code on failure. 4692 * 4693 * Timestamps: 4694 * vp - ctime|mtime updated 4695 */ 4696/*ARGSUSED*/ 4697static int 4698zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4699 caller_context_t *ct) 4700{ 4701 znode_t *zp = VTOZ(vp); 4702 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4703 page_t *pp; 4704 size_t io_len; 4705 u_offset_t io_off; 4706 uint_t blksz; 4707 rl_t *rl; 4708 int error = 0; 4709 4710 ZFS_ENTER(zfsvfs); 4711 ZFS_VERIFY_ZP(zp); 4712 4713 /* 4714 * Align this request to the file block size in case we kluster. 4715 * XXX - this can result in pretty aggresive locking, which can 4716 * impact simultanious read/write access. One option might be 4717 * to break up long requests (len == 0) into block-by-block 4718 * operations to get narrower locking. 4719 */ 4720 blksz = zp->z_blksz; 4721 if (ISP2(blksz)) 4722 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 4723 else 4724 io_off = 0; 4725 if (len > 0 && ISP2(blksz)) 4726 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 4727 else 4728 io_len = 0; 4729 4730 if (io_len == 0) { 4731 /* 4732 * Search the entire vp list for pages >= io_off. 4733 */ 4734 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 4735 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 4736 goto out; 4737 } 4738 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 4739 4740 if (off > zp->z_size) { 4741 /* past end of file */ 4742 zfs_range_unlock(rl); 4743 ZFS_EXIT(zfsvfs); 4744 return (0); 4745 } 4746 4747 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); 4748 4749 for (off = io_off; io_off < off + len; io_off += io_len) { 4750 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 4751 pp = page_lookup(vp, io_off, 4752 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 4753 } else { 4754 pp = page_lookup_nowait(vp, io_off, 4755 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 4756 } 4757 4758 if (pp != NULL && pvn_getdirty(pp, flags)) { 4759 int err; 4760 4761 /* 4762 * Found a dirty page to push 4763 */ 4764 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 4765 if (err) 4766 error = err; 4767 } else { 4768 io_len = PAGESIZE; 4769 } 4770 } 4771out: 4772 zfs_range_unlock(rl); 4773 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4774 zil_commit(zfsvfs->z_log, zp->z_id); 4775 ZFS_EXIT(zfsvfs); 4776 return (error); 4777} 4778#endif /* illumos */ 4779 4780/*ARGSUSED*/ 4781void 4782zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4783{ 4784 znode_t *zp = VTOZ(vp); 4785 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4786 int error; 4787 4788 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4789 if (zp->z_sa_hdl == NULL) { 4790 /* 4791 * The fs has been unmounted, or we did a 4792 * suspend/resume and this file no longer exists. 4793 */ 4794 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4795 vrecycle(vp); 4796 return; 4797 } 4798 4799 mutex_enter(&zp->z_lock); 4800 if (zp->z_unlinked) { 4801 /* 4802 * Fast path to recycle a vnode of a removed file. 4803 */ 4804 mutex_exit(&zp->z_lock); 4805 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4806 vrecycle(vp); 4807 return; 4808 } 4809 mutex_exit(&zp->z_lock); 4810 4811 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4812 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4813 4814 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4815 zfs_sa_upgrade_txholds(tx, zp); 4816 error = dmu_tx_assign(tx, TXG_WAIT); 4817 if (error) { 4818 dmu_tx_abort(tx); 4819 } else { 4820 mutex_enter(&zp->z_lock); 4821 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4822 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4823 zp->z_atime_dirty = 0; 4824 mutex_exit(&zp->z_lock); 4825 dmu_tx_commit(tx); 4826 } 4827 } 4828 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4829} 4830 4831#ifdef illumos 4832/* 4833 * Bounds-check the seek operation. 4834 * 4835 * IN: vp - vnode seeking within 4836 * ooff - old file offset 4837 * noffp - pointer to new file offset 4838 * ct - caller context 4839 * 4840 * RETURN: 0 on success, EINVAL if new offset invalid. 4841 */ 4842/* ARGSUSED */ 4843static int 4844zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 4845 caller_context_t *ct) 4846{ 4847 if (vp->v_type == VDIR) 4848 return (0); 4849 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4850} 4851 4852/* 4853 * Pre-filter the generic locking function to trap attempts to place 4854 * a mandatory lock on a memory mapped file. 4855 */ 4856static int 4857zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 4858 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 4859{ 4860 znode_t *zp = VTOZ(vp); 4861 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4862 4863 ZFS_ENTER(zfsvfs); 4864 ZFS_VERIFY_ZP(zp); 4865 4866 /* 4867 * We are following the UFS semantics with respect to mapcnt 4868 * here: If we see that the file is mapped already, then we will 4869 * return an error, but we don't worry about races between this 4870 * function and zfs_map(). 4871 */ 4872 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { 4873 ZFS_EXIT(zfsvfs); 4874 return (SET_ERROR(EAGAIN)); 4875 } 4876 ZFS_EXIT(zfsvfs); 4877 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4878} 4879 4880/* 4881 * If we can't find a page in the cache, we will create a new page 4882 * and fill it with file data. For efficiency, we may try to fill 4883 * multiple pages at once (klustering) to fill up the supplied page 4884 * list. Note that the pages to be filled are held with an exclusive 4885 * lock to prevent access by other threads while they are being filled. 4886 */ 4887static int 4888zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 4889 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 4890{ 4891 znode_t *zp = VTOZ(vp); 4892 page_t *pp, *cur_pp; 4893 objset_t *os = zp->z_zfsvfs->z_os; 4894 u_offset_t io_off, total; 4895 size_t io_len; 4896 int err; 4897 4898 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4899 /* 4900 * We only have a single page, don't bother klustering 4901 */ 4902 io_off = off; 4903 io_len = PAGESIZE; 4904 pp = page_create_va(vp, io_off, io_len, 4905 PG_EXCL | PG_WAIT, seg, addr); 4906 } else { 4907 /* 4908 * Try to find enough pages to fill the page list 4909 */ 4910 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4911 &io_len, off, plsz, 0); 4912 } 4913 if (pp == NULL) { 4914 /* 4915 * The page already exists, nothing to do here. 4916 */ 4917 *pl = NULL; 4918 return (0); 4919 } 4920 4921 /* 4922 * Fill the pages in the kluster. 4923 */ 4924 cur_pp = pp; 4925 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4926 caddr_t va; 4927 4928 ASSERT3U(io_off, ==, cur_pp->p_offset); 4929 va = zfs_map_page(cur_pp, S_WRITE); 4930 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4931 DMU_READ_PREFETCH); 4932 zfs_unmap_page(cur_pp, va); 4933 if (err) { 4934 /* On error, toss the entire kluster */ 4935 pvn_read_done(pp, B_ERROR); 4936 /* convert checksum errors into IO errors */ 4937 if (err == ECKSUM) 4938 err = SET_ERROR(EIO); 4939 return (err); 4940 } 4941 cur_pp = cur_pp->p_next; 4942 } 4943 4944 /* 4945 * Fill in the page list array from the kluster starting 4946 * from the desired offset `off'. 4947 * NOTE: the page list will always be null terminated. 4948 */ 4949 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4950 ASSERT(pl == NULL || (*pl)->p_offset == off); 4951 4952 return (0); 4953} 4954 4955/* 4956 * Return pointers to the pages for the file region [off, off + len] 4957 * in the pl array. If plsz is greater than len, this function may 4958 * also return page pointers from after the specified region 4959 * (i.e. the region [off, off + plsz]). These additional pages are 4960 * only returned if they are already in the cache, or were created as 4961 * part of a klustered read. 4962 * 4963 * IN: vp - vnode of file to get data from. 4964 * off - position in file to get data from. 4965 * len - amount of data to retrieve. 4966 * plsz - length of provided page list. 4967 * seg - segment to obtain pages for. 4968 * addr - virtual address of fault. 4969 * rw - mode of created pages. 4970 * cr - credentials of caller. 4971 * ct - caller context. 4972 * 4973 * OUT: protp - protection mode of created pages. 4974 * pl - list of pages created. 4975 * 4976 * RETURN: 0 on success, error code on failure. 4977 * 4978 * Timestamps: 4979 * vp - atime updated 4980 */ 4981/* ARGSUSED */ 4982static int 4983zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4984 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4985 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4986{ 4987 znode_t *zp = VTOZ(vp); 4988 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4989 page_t **pl0 = pl; 4990 int err = 0; 4991 4992 /* we do our own caching, faultahead is unnecessary */ 4993 if (pl == NULL) 4994 return (0); 4995 else if (len > plsz) 4996 len = plsz; 4997 else 4998 len = P2ROUNDUP(len, PAGESIZE); 4999 ASSERT(plsz >= len); 5000 5001 ZFS_ENTER(zfsvfs); 5002 ZFS_VERIFY_ZP(zp); 5003 5004 if (protp) 5005 *protp = PROT_ALL; 5006 5007 /* 5008 * Loop through the requested range [off, off + len) looking 5009 * for pages. If we don't find a page, we will need to create 5010 * a new page and fill it with data from the file. 5011 */ 5012 while (len > 0) { 5013 if (*pl = page_lookup(vp, off, SE_SHARED)) 5014 *(pl+1) = NULL; 5015 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 5016 goto out; 5017 while (*pl) { 5018 ASSERT3U((*pl)->p_offset, ==, off); 5019 off += PAGESIZE; 5020 addr += PAGESIZE; 5021 if (len > 0) { 5022 ASSERT3U(len, >=, PAGESIZE); 5023 len -= PAGESIZE; 5024 } 5025 ASSERT3U(plsz, >=, PAGESIZE); 5026 plsz -= PAGESIZE; 5027 pl++; 5028 } 5029 } 5030 5031 /* 5032 * Fill out the page array with any pages already in the cache. 5033 */ 5034 while (plsz > 0 && 5035 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 5036 off += PAGESIZE; 5037 plsz -= PAGESIZE; 5038 } 5039out: 5040 if (err) { 5041 /* 5042 * Release any pages we have previously locked. 5043 */ 5044 while (pl > pl0) 5045 page_unlock(*--pl); 5046 } else { 5047 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 5048 } 5049 5050 *pl = NULL; 5051 5052 ZFS_EXIT(zfsvfs); 5053 return (err); 5054} 5055 5056/* 5057 * Request a memory map for a section of a file. This code interacts 5058 * with common code and the VM system as follows: 5059 * 5060 * - common code calls mmap(), which ends up in smmap_common() 5061 * - this calls VOP_MAP(), which takes you into (say) zfs 5062 * - zfs_map() calls as_map(), passing segvn_create() as the callback 5063 * - segvn_create() creates the new segment and calls VOP_ADDMAP() 5064 * - zfs_addmap() updates z_mapcnt 5065 */ 5066/*ARGSUSED*/ 5067static int 5068zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 5069 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 5070 caller_context_t *ct) 5071{ 5072 znode_t *zp = VTOZ(vp); 5073 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5074 segvn_crargs_t vn_a; 5075 int error; 5076 5077 ZFS_ENTER(zfsvfs); 5078 ZFS_VERIFY_ZP(zp); 5079 5080 if ((prot & PROT_WRITE) && (zp->z_pflags & 5081 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 5082 ZFS_EXIT(zfsvfs); 5083 return (SET_ERROR(EPERM)); 5084 } 5085 5086 if ((prot & (PROT_READ | PROT_EXEC)) && 5087 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 5088 ZFS_EXIT(zfsvfs); 5089 return (SET_ERROR(EACCES)); 5090 } 5091 5092 if (vp->v_flag & VNOMAP) { 5093 ZFS_EXIT(zfsvfs); 5094 return (SET_ERROR(ENOSYS)); 5095 } 5096 5097 if (off < 0 || len > MAXOFFSET_T - off) { 5098 ZFS_EXIT(zfsvfs); 5099 return (SET_ERROR(ENXIO)); 5100 } 5101 5102 if (vp->v_type != VREG) { 5103 ZFS_EXIT(zfsvfs); 5104 return (SET_ERROR(ENODEV)); 5105 } 5106 5107 /* 5108 * If file is locked, disallow mapping. 5109 */ 5110 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { 5111 ZFS_EXIT(zfsvfs); 5112 return (SET_ERROR(EAGAIN)); 5113 } 5114 5115 as_rangelock(as); 5116 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 5117 if (error != 0) { 5118 as_rangeunlock(as); 5119 ZFS_EXIT(zfsvfs); 5120 return (error); 5121 } 5122 5123 vn_a.vp = vp; 5124 vn_a.offset = (u_offset_t)off; 5125 vn_a.type = flags & MAP_TYPE; 5126 vn_a.prot = prot; 5127 vn_a.maxprot = maxprot; 5128 vn_a.cred = cr; 5129 vn_a.amp = NULL; 5130 vn_a.flags = flags & ~MAP_TYPE; 5131 vn_a.szc = 0; 5132 vn_a.lgrp_mem_policy_flags = 0; 5133 5134 error = as_map(as, *addrp, len, segvn_create, &vn_a); 5135 5136 as_rangeunlock(as); 5137 ZFS_EXIT(zfsvfs); 5138 return (error); 5139} 5140 5141/* ARGSUSED */ 5142static int 5143zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5144 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 5145 caller_context_t *ct) 5146{ 5147 uint64_t pages = btopr(len); 5148 5149 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 5150 return (0); 5151} 5152 5153/* 5154 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 5155 * more accurate mtime for the associated file. Since we don't have a way of 5156 * detecting when the data was actually modified, we have to resort to 5157 * heuristics. If an explicit msync() is done, then we mark the mtime when the 5158 * last page is pushed. The problem occurs when the msync() call is omitted, 5159 * which by far the most common case: 5160 * 5161 * open() 5162 * mmap() 5163 * <modify memory> 5164 * munmap() 5165 * close() 5166 * <time lapse> 5167 * putpage() via fsflush 5168 * 5169 * If we wait until fsflush to come along, we can have a modification time that 5170 * is some arbitrary point in the future. In order to prevent this in the 5171 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 5172 * torn down. 5173 */ 5174/* ARGSUSED */ 5175static int 5176zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5177 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 5178 caller_context_t *ct) 5179{ 5180 uint64_t pages = btopr(len); 5181 5182 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 5183 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 5184 5185 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 5186 vn_has_cached_data(vp)) 5187 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 5188 5189 return (0); 5190} 5191 5192/* 5193 * Free or allocate space in a file. Currently, this function only 5194 * supports the `F_FREESP' command. However, this command is somewhat 5195 * misnamed, as its functionality includes the ability to allocate as 5196 * well as free space. 5197 * 5198 * IN: vp - vnode of file to free data in. 5199 * cmd - action to take (only F_FREESP supported). 5200 * bfp - section of file to free/alloc. 5201 * flag - current file open mode flags. 5202 * offset - current file offset. 5203 * cr - credentials of caller [UNUSED]. 5204 * ct - caller context. 5205 * 5206 * RETURN: 0 on success, error code on failure. 5207 * 5208 * Timestamps: 5209 * vp - ctime|mtime updated 5210 */ 5211/* ARGSUSED */ 5212static int 5213zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 5214 offset_t offset, cred_t *cr, caller_context_t *ct) 5215{ 5216 znode_t *zp = VTOZ(vp); 5217 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5218 uint64_t off, len; 5219 int error; 5220 5221 ZFS_ENTER(zfsvfs); 5222 ZFS_VERIFY_ZP(zp); 5223 5224 if (cmd != F_FREESP) { 5225 ZFS_EXIT(zfsvfs); 5226 return (SET_ERROR(EINVAL)); 5227 } 5228 5229 /* 5230 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our 5231 * callers might not be able to detect properly that we are read-only, 5232 * so check it explicitly here. 5233 */ 5234 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 5235 ZFS_EXIT(zfsvfs); 5236 return (SET_ERROR(EROFS)); 5237 } 5238 5239 if (error = convoff(vp, bfp, 0, offset)) { 5240 ZFS_EXIT(zfsvfs); 5241 return (error); 5242 } 5243 5244 if (bfp->l_len < 0) { 5245 ZFS_EXIT(zfsvfs); 5246 return (SET_ERROR(EINVAL)); 5247 } 5248 5249 off = bfp->l_start; 5250 len = bfp->l_len; /* 0 means from off to end of file */ 5251 5252 error = zfs_freesp(zp, off, len, flag, TRUE); 5253 5254 ZFS_EXIT(zfsvfs); 5255 return (error); 5256} 5257#endif /* illumos */ 5258 5259CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 5260CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 5261 5262/*ARGSUSED*/ 5263static int 5264zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 5265{ 5266 znode_t *zp = VTOZ(vp); 5267 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5268 uint32_t gen; 5269 uint64_t gen64; 5270 uint64_t object = zp->z_id; 5271 zfid_short_t *zfid; 5272 int size, i, error; 5273 5274 ZFS_ENTER(zfsvfs); 5275 ZFS_VERIFY_ZP(zp); 5276 5277 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 5278 &gen64, sizeof (uint64_t))) != 0) { 5279 ZFS_EXIT(zfsvfs); 5280 return (error); 5281 } 5282 5283 gen = (uint32_t)gen64; 5284 5285 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 5286 5287#ifdef illumos 5288 if (fidp->fid_len < size) { 5289 fidp->fid_len = size; 5290 ZFS_EXIT(zfsvfs); 5291 return (SET_ERROR(ENOSPC)); 5292 } 5293#else 5294 fidp->fid_len = size; 5295#endif 5296 5297 zfid = (zfid_short_t *)fidp; 5298 5299 zfid->zf_len = size; 5300 5301 for (i = 0; i < sizeof (zfid->zf_object); i++) 5302 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 5303 5304 /* Must have a non-zero generation number to distinguish from .zfs */ 5305 if (gen == 0) 5306 gen = 1; 5307 for (i = 0; i < sizeof (zfid->zf_gen); i++) 5308 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 5309 5310 if (size == LONG_FID_LEN) { 5311 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 5312 zfid_long_t *zlfid; 5313 5314 zlfid = (zfid_long_t *)fidp; 5315 5316 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 5317 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 5318 5319 /* XXX - this should be the generation number for the objset */ 5320 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 5321 zlfid->zf_setgen[i] = 0; 5322 } 5323 5324 ZFS_EXIT(zfsvfs); 5325 return (0); 5326} 5327 5328static int 5329zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 5330 caller_context_t *ct) 5331{ 5332 znode_t *zp, *xzp; 5333 zfsvfs_t *zfsvfs; 5334 zfs_dirlock_t *dl; 5335 int error; 5336 5337 switch (cmd) { 5338 case _PC_LINK_MAX: 5339 *valp = INT_MAX; 5340 return (0); 5341 5342 case _PC_FILESIZEBITS: 5343 *valp = 64; 5344 return (0); 5345#ifdef illumos 5346 case _PC_XATTR_EXISTS: 5347 zp = VTOZ(vp); 5348 zfsvfs = zp->z_zfsvfs; 5349 ZFS_ENTER(zfsvfs); 5350 ZFS_VERIFY_ZP(zp); 5351 *valp = 0; 5352 error = zfs_dirent_lock(&dl, zp, "", &xzp, 5353 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 5354 if (error == 0) { 5355 zfs_dirent_unlock(dl); 5356 if (!zfs_dirempty(xzp)) 5357 *valp = 1; 5358 VN_RELE(ZTOV(xzp)); 5359 } else if (error == ENOENT) { 5360 /* 5361 * If there aren't extended attributes, it's the 5362 * same as having zero of them. 5363 */ 5364 error = 0; 5365 } 5366 ZFS_EXIT(zfsvfs); 5367 return (error); 5368 5369 case _PC_SATTR_ENABLED: 5370 case _PC_SATTR_EXISTS: 5371 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 5372 (vp->v_type == VREG || vp->v_type == VDIR); 5373 return (0); 5374 5375 case _PC_ACCESS_FILTERING: 5376 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 5377 vp->v_type == VDIR; 5378 return (0); 5379 5380 case _PC_ACL_ENABLED: 5381 *valp = _ACL_ACE_ENABLED; 5382 return (0); 5383#endif /* illumos */ 5384 case _PC_MIN_HOLE_SIZE: 5385 *valp = (int)SPA_MINBLOCKSIZE; 5386 return (0); 5387#ifdef illumos 5388 case _PC_TIMESTAMP_RESOLUTION: 5389 /* nanosecond timestamp resolution */ 5390 *valp = 1L; 5391 return (0); 5392#endif 5393 case _PC_ACL_EXTENDED: 5394 *valp = 0; 5395 return (0); 5396 5397 case _PC_ACL_NFS4: 5398 *valp = 1; 5399 return (0); 5400 5401 case _PC_ACL_PATH_MAX: 5402 *valp = ACL_MAX_ENTRIES; 5403 return (0); 5404 5405 default: 5406 return (EOPNOTSUPP); 5407 } 5408} 5409 5410/*ARGSUSED*/ 5411static int 5412zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5413 caller_context_t *ct) 5414{ 5415 znode_t *zp = VTOZ(vp); 5416 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5417 int error; 5418 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5419 5420 ZFS_ENTER(zfsvfs); 5421 ZFS_VERIFY_ZP(zp); 5422 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 5423 ZFS_EXIT(zfsvfs); 5424 5425 return (error); 5426} 5427 5428/*ARGSUSED*/ 5429int 5430zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5431 caller_context_t *ct) 5432{ 5433 znode_t *zp = VTOZ(vp); 5434 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5435 int error; 5436 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5437 zilog_t *zilog = zfsvfs->z_log; 5438 5439 ZFS_ENTER(zfsvfs); 5440 ZFS_VERIFY_ZP(zp); 5441 5442 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 5443 5444 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 5445 zil_commit(zilog, 0); 5446 5447 ZFS_EXIT(zfsvfs); 5448 return (error); 5449} 5450 5451#ifdef illumos 5452/* 5453 * The smallest read we may consider to loan out an arcbuf. 5454 * This must be a power of 2. 5455 */ 5456int zcr_blksz_min = (1 << 10); /* 1K */ 5457/* 5458 * If set to less than the file block size, allow loaning out of an 5459 * arcbuf for a partial block read. This must be a power of 2. 5460 */ 5461int zcr_blksz_max = (1 << 17); /* 128K */ 5462 5463/*ARGSUSED*/ 5464static int 5465zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr, 5466 caller_context_t *ct) 5467{ 5468 znode_t *zp = VTOZ(vp); 5469 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5470 int max_blksz = zfsvfs->z_max_blksz; 5471 uio_t *uio = &xuio->xu_uio; 5472 ssize_t size = uio->uio_resid; 5473 offset_t offset = uio->uio_loffset; 5474 int blksz; 5475 int fullblk, i; 5476 arc_buf_t *abuf; 5477 ssize_t maxsize; 5478 int preamble, postamble; 5479 5480 if (xuio->xu_type != UIOTYPE_ZEROCOPY) 5481 return (SET_ERROR(EINVAL)); 5482 5483 ZFS_ENTER(zfsvfs); 5484 ZFS_VERIFY_ZP(zp); 5485 switch (ioflag) { 5486 case UIO_WRITE: 5487 /* 5488 * Loan out an arc_buf for write if write size is bigger than 5489 * max_blksz, and the file's block size is also max_blksz. 5490 */ 5491 blksz = max_blksz; 5492 if (size < blksz || zp->z_blksz != blksz) { 5493 ZFS_EXIT(zfsvfs); 5494 return (SET_ERROR(EINVAL)); 5495 } 5496 /* 5497 * Caller requests buffers for write before knowing where the 5498 * write offset might be (e.g. NFS TCP write). 5499 */ 5500 if (offset == -1) { 5501 preamble = 0; 5502 } else { 5503 preamble = P2PHASE(offset, blksz); 5504 if (preamble) { 5505 preamble = blksz - preamble; 5506 size -= preamble; 5507 } 5508 } 5509 5510 postamble = P2PHASE(size, blksz); 5511 size -= postamble; 5512 5513 fullblk = size / blksz; 5514 (void) dmu_xuio_init(xuio, 5515 (preamble != 0) + fullblk + (postamble != 0)); 5516 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble, 5517 int, postamble, int, 5518 (preamble != 0) + fullblk + (postamble != 0)); 5519 5520 /* 5521 * Have to fix iov base/len for partial buffers. They 5522 * currently represent full arc_buf's. 5523 */ 5524 if (preamble) { 5525 /* data begins in the middle of the arc_buf */ 5526 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5527 blksz); 5528 ASSERT(abuf); 5529 (void) dmu_xuio_add(xuio, abuf, 5530 blksz - preamble, preamble); 5531 } 5532 5533 for (i = 0; i < fullblk; i++) { 5534 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5535 blksz); 5536 ASSERT(abuf); 5537 (void) dmu_xuio_add(xuio, abuf, 0, blksz); 5538 } 5539 5540 if (postamble) { 5541 /* data ends in the middle of the arc_buf */ 5542 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5543 blksz); 5544 ASSERT(abuf); 5545 (void) dmu_xuio_add(xuio, abuf, 0, postamble); 5546 } 5547 break; 5548 case UIO_READ: 5549 /* 5550 * Loan out an arc_buf for read if the read size is larger than 5551 * the current file block size. Block alignment is not 5552 * considered. Partial arc_buf will be loaned out for read. 5553 */ 5554 blksz = zp->z_blksz; 5555 if (blksz < zcr_blksz_min) 5556 blksz = zcr_blksz_min; 5557 if (blksz > zcr_blksz_max) 5558 blksz = zcr_blksz_max; 5559 /* avoid potential complexity of dealing with it */ 5560 if (blksz > max_blksz) { 5561 ZFS_EXIT(zfsvfs); 5562 return (SET_ERROR(EINVAL)); 5563 } 5564 5565 maxsize = zp->z_size - uio->uio_loffset; 5566 if (size > maxsize) 5567 size = maxsize; 5568 5569 if (size < blksz || vn_has_cached_data(vp)) { 5570 ZFS_EXIT(zfsvfs); 5571 return (SET_ERROR(EINVAL)); 5572 } 5573 break; 5574 default: 5575 ZFS_EXIT(zfsvfs); 5576 return (SET_ERROR(EINVAL)); 5577 } 5578 5579 uio->uio_extflg = UIO_XUIO; 5580 XUIO_XUZC_RW(xuio) = ioflag; 5581 ZFS_EXIT(zfsvfs); 5582 return (0); 5583} 5584 5585/*ARGSUSED*/ 5586static int 5587zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct) 5588{ 5589 int i; 5590 arc_buf_t *abuf; 5591 int ioflag = XUIO_XUZC_RW(xuio); 5592 5593 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY); 5594 5595 i = dmu_xuio_cnt(xuio); 5596 while (i-- > 0) { 5597 abuf = dmu_xuio_arcbuf(xuio, i); 5598 /* 5599 * if abuf == NULL, it must be a write buffer 5600 * that has been returned in zfs_write(). 5601 */ 5602 if (abuf) 5603 dmu_return_arcbuf(abuf); 5604 ASSERT(abuf || ioflag == UIO_WRITE); 5605 } 5606 5607 dmu_xuio_fini(xuio); 5608 return (0); 5609} 5610 5611/* 5612 * Predeclare these here so that the compiler assumes that 5613 * this is an "old style" function declaration that does 5614 * not include arguments => we won't get type mismatch errors 5615 * in the initializations that follow. 5616 */ 5617static int zfs_inval(); 5618static int zfs_isdir(); 5619 5620static int 5621zfs_inval() 5622{ 5623 return (SET_ERROR(EINVAL)); 5624} 5625 5626static int 5627zfs_isdir() 5628{ 5629 return (SET_ERROR(EISDIR)); 5630} 5631/* 5632 * Directory vnode operations template 5633 */ 5634vnodeops_t *zfs_dvnodeops; 5635const fs_operation_def_t zfs_dvnodeops_template[] = { 5636 VOPNAME_OPEN, { .vop_open = zfs_open }, 5637 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5638 VOPNAME_READ, { .error = zfs_isdir }, 5639 VOPNAME_WRITE, { .error = zfs_isdir }, 5640 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5641 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5642 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5643 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5644 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5645 VOPNAME_CREATE, { .vop_create = zfs_create }, 5646 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5647 VOPNAME_LINK, { .vop_link = zfs_link }, 5648 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5649 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 5650 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5651 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5652 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 5653 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5654 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5655 VOPNAME_FID, { .vop_fid = zfs_fid }, 5656 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5657 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5658 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5659 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5660 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5661 NULL, NULL 5662}; 5663 5664/* 5665 * Regular file vnode operations template 5666 */ 5667vnodeops_t *zfs_fvnodeops; 5668const fs_operation_def_t zfs_fvnodeops_template[] = { 5669 VOPNAME_OPEN, { .vop_open = zfs_open }, 5670 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5671 VOPNAME_READ, { .vop_read = zfs_read }, 5672 VOPNAME_WRITE, { .vop_write = zfs_write }, 5673 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5674 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5675 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5676 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5677 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5678 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5679 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5680 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5681 VOPNAME_FID, { .vop_fid = zfs_fid }, 5682 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5683 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 5684 VOPNAME_SPACE, { .vop_space = zfs_space }, 5685 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 5686 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 5687 VOPNAME_MAP, { .vop_map = zfs_map }, 5688 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 5689 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 5690 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5691 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5692 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5693 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5694 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf }, 5695 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf }, 5696 NULL, NULL 5697}; 5698 5699/* 5700 * Symbolic link vnode operations template 5701 */ 5702vnodeops_t *zfs_symvnodeops; 5703const fs_operation_def_t zfs_symvnodeops_template[] = { 5704 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5705 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5706 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5707 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5708 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 5709 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5710 VOPNAME_FID, { .vop_fid = zfs_fid }, 5711 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5712 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5713 NULL, NULL 5714}; 5715 5716/* 5717 * special share hidden files vnode operations template 5718 */ 5719vnodeops_t *zfs_sharevnodeops; 5720const fs_operation_def_t zfs_sharevnodeops_template[] = { 5721 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5722 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5723 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5724 VOPNAME_FID, { .vop_fid = zfs_fid }, 5725 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5726 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5727 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5728 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5729 NULL, NULL 5730}; 5731 5732/* 5733 * Extended attribute directory vnode operations template 5734 * 5735 * This template is identical to the directory vnodes 5736 * operation template except for restricted operations: 5737 * VOP_MKDIR() 5738 * VOP_SYMLINK() 5739 * 5740 * Note that there are other restrictions embedded in: 5741 * zfs_create() - restrict type to VREG 5742 * zfs_link() - no links into/out of attribute space 5743 * zfs_rename() - no moves into/out of attribute space 5744 */ 5745vnodeops_t *zfs_xdvnodeops; 5746const fs_operation_def_t zfs_xdvnodeops_template[] = { 5747 VOPNAME_OPEN, { .vop_open = zfs_open }, 5748 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5749 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5750 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5751 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5752 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5753 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5754 VOPNAME_CREATE, { .vop_create = zfs_create }, 5755 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5756 VOPNAME_LINK, { .vop_link = zfs_link }, 5757 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5758 VOPNAME_MKDIR, { .error = zfs_inval }, 5759 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5760 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5761 VOPNAME_SYMLINK, { .error = zfs_inval }, 5762 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5763 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5764 VOPNAME_FID, { .vop_fid = zfs_fid }, 5765 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5766 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5767 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5768 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5769 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5770 NULL, NULL 5771}; 5772 5773/* 5774 * Error vnode operations template 5775 */ 5776vnodeops_t *zfs_evnodeops; 5777const fs_operation_def_t zfs_evnodeops_template[] = { 5778 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5779 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5780 NULL, NULL 5781}; 5782#endif /* illumos */ 5783 5784static int 5785ioflags(int ioflags) 5786{ 5787 int flags = 0; 5788 5789 if (ioflags & IO_APPEND) 5790 flags |= FAPPEND; 5791 if (ioflags & IO_NDELAY) 5792 flags |= FNONBLOCK; 5793 if (ioflags & IO_SYNC) 5794 flags |= (FSYNC | FDSYNC | FRSYNC); 5795 5796 return (flags); 5797} 5798 5799static int 5800zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage) 5801{ 5802 znode_t *zp = VTOZ(vp); 5803 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5804 objset_t *os = zp->z_zfsvfs->z_os; 5805 vm_page_t mfirst, mlast, mreq; 5806 vm_object_t object; 5807 caddr_t va; 5808 struct sf_buf *sf; 5809 off_t startoff, endoff; 5810 int i, error; 5811 vm_pindex_t reqstart, reqend; 5812 int pcount, lsize, reqsize, size; 5813 5814 ZFS_ENTER(zfsvfs); 5815 ZFS_VERIFY_ZP(zp); 5816 5817 pcount = OFF_TO_IDX(round_page(count)); 5818 mreq = m[reqpage]; 5819 object = mreq->object; 5820 error = 0; 5821 5822 KASSERT(vp->v_object == object, ("mismatching object")); 5823 5824 if (pcount > 1 && zp->z_blksz > PAGESIZE) { 5825 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz); 5826 reqstart = OFF_TO_IDX(round_page(startoff)); 5827 if (reqstart < m[0]->pindex) 5828 reqstart = 0; 5829 else 5830 reqstart = reqstart - m[0]->pindex; 5831 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE, 5832 zp->z_blksz); 5833 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1; 5834 if (reqend > m[pcount - 1]->pindex) 5835 reqend = m[pcount - 1]->pindex; 5836 reqsize = reqend - m[reqstart]->pindex + 1; 5837 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize, 5838 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds")); 5839 } else { 5840 reqstart = reqpage; 5841 reqsize = 1; 5842 } 5843 mfirst = m[reqstart]; 5844 mlast = m[reqstart + reqsize - 1]; 5845 5846 zfs_vmobject_wlock(object); 5847 5848 for (i = 0; i < reqstart; i++) { 5849 vm_page_lock(m[i]); 5850 vm_page_free(m[i]); 5851 vm_page_unlock(m[i]); 5852 } 5853 for (i = reqstart + reqsize; i < pcount; i++) { 5854 vm_page_lock(m[i]); 5855 vm_page_free(m[i]); 5856 vm_page_unlock(m[i]); 5857 } 5858 5859 if (mreq->valid && reqsize == 1) { 5860 if (mreq->valid != VM_PAGE_BITS_ALL) 5861 vm_page_zero_invalid(mreq, TRUE); 5862 zfs_vmobject_wunlock(object); 5863 ZFS_EXIT(zfsvfs); 5864 return (zfs_vm_pagerret_ok); 5865 } 5866 5867 PCPU_INC(cnt.v_vnodein); 5868 PCPU_ADD(cnt.v_vnodepgsin, reqsize); 5869 5870 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) { 5871 for (i = reqstart; i < reqstart + reqsize; i++) { 5872 if (i != reqpage) { 5873 vm_page_lock(m[i]); 5874 vm_page_free(m[i]); 5875 vm_page_unlock(m[i]); 5876 } 5877 } 5878 zfs_vmobject_wunlock(object); 5879 ZFS_EXIT(zfsvfs); 5880 return (zfs_vm_pagerret_bad); 5881 } 5882 5883 lsize = PAGE_SIZE; 5884 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size) 5885 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex); 5886 5887 zfs_vmobject_wunlock(object); 5888 5889 for (i = reqstart; i < reqstart + reqsize; i++) { 5890 size = PAGE_SIZE; 5891 if (i == (reqstart + reqsize - 1)) 5892 size = lsize; 5893 va = zfs_map_page(m[i], &sf); 5894 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex), 5895 size, va, DMU_READ_PREFETCH); 5896 if (size != PAGE_SIZE) 5897 bzero(va + size, PAGE_SIZE - size); 5898 zfs_unmap_page(sf); 5899 if (error != 0) 5900 break; 5901 } 5902 5903 zfs_vmobject_wlock(object); 5904 5905 for (i = reqstart; i < reqstart + reqsize; i++) { 5906 if (!error) 5907 m[i]->valid = VM_PAGE_BITS_ALL; 5908 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i])); 5909 if (i != reqpage) 5910 vm_page_readahead_finish(m[i]); 5911 } 5912 5913 zfs_vmobject_wunlock(object); 5914 5915 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 5916 ZFS_EXIT(zfsvfs); 5917 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok); 5918} 5919 5920static int 5921zfs_freebsd_getpages(ap) 5922 struct vop_getpages_args /* { 5923 struct vnode *a_vp; 5924 vm_page_t *a_m; 5925 int a_count; 5926 int a_reqpage; 5927 vm_ooffset_t a_offset; 5928 } */ *ap; 5929{ 5930 5931 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage)); 5932} 5933 5934static int 5935zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags, 5936 int *rtvals) 5937{ 5938 znode_t *zp = VTOZ(vp); 5939 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5940 rl_t *rl; 5941 dmu_tx_t *tx; 5942 struct sf_buf *sf; 5943 vm_object_t object; 5944 vm_page_t m; 5945 caddr_t va; 5946 size_t tocopy; 5947 size_t lo_len; 5948 vm_ooffset_t lo_off; 5949 vm_ooffset_t off; 5950 uint_t blksz; 5951 int ncount; 5952 int pcount; 5953 int err; 5954 int i; 5955 5956 ZFS_ENTER(zfsvfs); 5957 ZFS_VERIFY_ZP(zp); 5958 5959 object = vp->v_object; 5960 pcount = btoc(len); 5961 ncount = pcount; 5962 5963 KASSERT(ma[0]->object == object, ("mismatching object")); 5964 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length")); 5965 5966 for (i = 0; i < pcount; i++) 5967 rtvals[i] = zfs_vm_pagerret_error; 5968 5969 off = IDX_TO_OFF(ma[0]->pindex); 5970 blksz = zp->z_blksz; 5971 lo_off = rounddown(off, blksz); 5972 lo_len = roundup(len + (off - lo_off), blksz); 5973 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER); 5974 5975 zfs_vmobject_wlock(object); 5976 if (len + off > object->un_pager.vnp.vnp_size) { 5977 if (object->un_pager.vnp.vnp_size > off) { 5978 int pgoff; 5979 5980 len = object->un_pager.vnp.vnp_size - off; 5981 ncount = btoc(len); 5982 if ((pgoff = (int)len & PAGE_MASK) != 0) { 5983 /* 5984 * If the object is locked and the following 5985 * conditions hold, then the page's dirty 5986 * field cannot be concurrently changed by a 5987 * pmap operation. 5988 */ 5989 m = ma[ncount - 1]; 5990 vm_page_assert_sbusied(m); 5991 KASSERT(!pmap_page_is_write_mapped(m), 5992 ("zfs_putpages: page %p is not read-only", m)); 5993 vm_page_clear_dirty(m, pgoff, PAGE_SIZE - 5994 pgoff); 5995 } 5996 } else { 5997 len = 0; 5998 ncount = 0; 5999 } 6000 if (ncount < pcount) { 6001 for (i = ncount; i < pcount; i++) { 6002 rtvals[i] = zfs_vm_pagerret_bad; 6003 } 6004 } 6005 } 6006 zfs_vmobject_wunlock(object); 6007 6008 if (ncount == 0) 6009 goto out; 6010 6011 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 6012 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 6013 goto out; 6014 } 6015 6016top: 6017 tx = dmu_tx_create(zfsvfs->z_os); 6018 dmu_tx_hold_write(tx, zp->z_id, off, len); 6019 6020 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 6021 zfs_sa_upgrade_txholds(tx, zp); 6022 err = dmu_tx_assign(tx, TXG_NOWAIT); 6023 if (err != 0) { 6024 if (err == ERESTART) { 6025 dmu_tx_wait(tx); 6026 dmu_tx_abort(tx); 6027 goto top; 6028 } 6029 dmu_tx_abort(tx); 6030 goto out; 6031 } 6032 6033 if (zp->z_blksz < PAGE_SIZE) { 6034 i = 0; 6035 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) { 6036 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len; 6037 va = zfs_map_page(ma[i], &sf); 6038 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx); 6039 zfs_unmap_page(sf); 6040 } 6041 } else { 6042 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx); 6043 } 6044 6045 if (err == 0) { 6046 uint64_t mtime[2], ctime[2]; 6047 sa_bulk_attr_t bulk[3]; 6048 int count = 0; 6049 6050 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 6051 &mtime, 16); 6052 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 6053 &ctime, 16); 6054 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 6055 &zp->z_pflags, 8); 6056 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 6057 B_TRUE); 6058 (void)sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 6059 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 6060 6061 zfs_vmobject_wlock(object); 6062 for (i = 0; i < ncount; i++) { 6063 rtvals[i] = zfs_vm_pagerret_ok; 6064 vm_page_undirty(ma[i]); 6065 } 6066 zfs_vmobject_wunlock(object); 6067 PCPU_INC(cnt.v_vnodeout); 6068 PCPU_ADD(cnt.v_vnodepgsout, ncount); 6069 } 6070 dmu_tx_commit(tx); 6071 6072out: 6073 zfs_range_unlock(rl); 6074 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 || 6075 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 6076 zil_commit(zfsvfs->z_log, zp->z_id); 6077 ZFS_EXIT(zfsvfs); 6078 return (rtvals[0]); 6079} 6080 6081int 6082zfs_freebsd_putpages(ap) 6083 struct vop_putpages_args /* { 6084 struct vnode *a_vp; 6085 vm_page_t *a_m; 6086 int a_count; 6087 int a_sync; 6088 int *a_rtvals; 6089 vm_ooffset_t a_offset; 6090 } */ *ap; 6091{ 6092 6093 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync, 6094 ap->a_rtvals)); 6095} 6096 6097static int 6098zfs_freebsd_bmap(ap) 6099 struct vop_bmap_args /* { 6100 struct vnode *a_vp; 6101 daddr_t a_bn; 6102 struct bufobj **a_bop; 6103 daddr_t *a_bnp; 6104 int *a_runp; 6105 int *a_runb; 6106 } */ *ap; 6107{ 6108 6109 if (ap->a_bop != NULL) 6110 *ap->a_bop = &ap->a_vp->v_bufobj; 6111 if (ap->a_bnp != NULL) 6112 *ap->a_bnp = ap->a_bn; 6113 if (ap->a_runp != NULL) 6114 *ap->a_runp = 0; 6115 if (ap->a_runb != NULL) 6116 *ap->a_runb = 0; 6117 6118 return (0); 6119} 6120 6121static int 6122zfs_freebsd_open(ap) 6123 struct vop_open_args /* { 6124 struct vnode *a_vp; 6125 int a_mode; 6126 struct ucred *a_cred; 6127 struct thread *a_td; 6128 } */ *ap; 6129{ 6130 vnode_t *vp = ap->a_vp; 6131 znode_t *zp = VTOZ(vp); 6132 int error; 6133 6134 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL); 6135 if (error == 0) 6136 vnode_create_vobject(vp, zp->z_size, ap->a_td); 6137 return (error); 6138} 6139 6140static int 6141zfs_freebsd_close(ap) 6142 struct vop_close_args /* { 6143 struct vnode *a_vp; 6144 int a_fflag; 6145 struct ucred *a_cred; 6146 struct thread *a_td; 6147 } */ *ap; 6148{ 6149 6150 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL)); 6151} 6152 6153static int 6154zfs_freebsd_ioctl(ap) 6155 struct vop_ioctl_args /* { 6156 struct vnode *a_vp; 6157 u_long a_command; 6158 caddr_t a_data; 6159 int a_fflag; 6160 struct ucred *cred; 6161 struct thread *td; 6162 } */ *ap; 6163{ 6164 6165 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 6166 ap->a_fflag, ap->a_cred, NULL, NULL)); 6167} 6168 6169static int 6170zfs_freebsd_read(ap) 6171 struct vop_read_args /* { 6172 struct vnode *a_vp; 6173 struct uio *a_uio; 6174 int a_ioflag; 6175 struct ucred *a_cred; 6176 } */ *ap; 6177{ 6178 6179 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 6180 ap->a_cred, NULL)); 6181} 6182 6183static int 6184zfs_freebsd_write(ap) 6185 struct vop_write_args /* { 6186 struct vnode *a_vp; 6187 struct uio *a_uio; 6188 int a_ioflag; 6189 struct ucred *a_cred; 6190 } */ *ap; 6191{ 6192 6193 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 6194 ap->a_cred, NULL)); 6195} 6196 6197static int 6198zfs_freebsd_access(ap) 6199 struct vop_access_args /* { 6200 struct vnode *a_vp; 6201 accmode_t a_accmode; 6202 struct ucred *a_cred; 6203 struct thread *a_td; 6204 } */ *ap; 6205{ 6206 vnode_t *vp = ap->a_vp; 6207 znode_t *zp = VTOZ(vp); 6208 accmode_t accmode; 6209 int error = 0; 6210 6211 /* 6212 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 6213 */ 6214 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 6215 if (accmode != 0) 6216 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL); 6217 6218 /* 6219 * VADMIN has to be handled by vaccess(). 6220 */ 6221 if (error == 0) { 6222 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 6223 if (accmode != 0) { 6224 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 6225 zp->z_gid, accmode, ap->a_cred, NULL); 6226 } 6227 } 6228 6229 /* 6230 * For VEXEC, ensure that at least one execute bit is set for 6231 * non-directories. 6232 */ 6233 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 6234 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 6235 error = EACCES; 6236 } 6237 6238 return (error); 6239} 6240 6241static int 6242zfs_freebsd_lookup(ap) 6243 struct vop_lookup_args /* { 6244 struct vnode *a_dvp; 6245 struct vnode **a_vpp; 6246 struct componentname *a_cnp; 6247 } */ *ap; 6248{ 6249 struct componentname *cnp = ap->a_cnp; 6250 char nm[NAME_MAX + 1]; 6251 6252 ASSERT(cnp->cn_namelen < sizeof(nm)); 6253 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 6254 6255 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 6256 cnp->cn_cred, cnp->cn_thread, 0)); 6257} 6258 6259static int 6260zfs_freebsd_create(ap) 6261 struct vop_create_args /* { 6262 struct vnode *a_dvp; 6263 struct vnode **a_vpp; 6264 struct componentname *a_cnp; 6265 struct vattr *a_vap; 6266 } */ *ap; 6267{ 6268 struct componentname *cnp = ap->a_cnp; 6269 vattr_t *vap = ap->a_vap; 6270 int error, mode; 6271 6272 ASSERT(cnp->cn_flags & SAVENAME); 6273 6274 vattr_init_mask(vap); 6275 mode = vap->va_mode & ALLPERMS; 6276 6277 error = zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 6278 ap->a_vpp, cnp->cn_cred, cnp->cn_thread); 6279#ifdef FREEBSD_NAMECACHE 6280 if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0) 6281 cache_enter(ap->a_dvp, *ap->a_vpp, cnp); 6282#endif 6283 return (error); 6284} 6285 6286static int 6287zfs_freebsd_remove(ap) 6288 struct vop_remove_args /* { 6289 struct vnode *a_dvp; 6290 struct vnode *a_vp; 6291 struct componentname *a_cnp; 6292 } */ *ap; 6293{ 6294 6295 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6296 6297 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 6298 ap->a_cnp->cn_cred, NULL, 0)); 6299} 6300 6301static int 6302zfs_freebsd_mkdir(ap) 6303 struct vop_mkdir_args /* { 6304 struct vnode *a_dvp; 6305 struct vnode **a_vpp; 6306 struct componentname *a_cnp; 6307 struct vattr *a_vap; 6308 } */ *ap; 6309{ 6310 vattr_t *vap = ap->a_vap; 6311 6312 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6313 6314 vattr_init_mask(vap); 6315 6316 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 6317 ap->a_cnp->cn_cred, NULL, 0, NULL)); 6318} 6319 6320static int 6321zfs_freebsd_rmdir(ap) 6322 struct vop_rmdir_args /* { 6323 struct vnode *a_dvp; 6324 struct vnode *a_vp; 6325 struct componentname *a_cnp; 6326 } */ *ap; 6327{ 6328 struct componentname *cnp = ap->a_cnp; 6329 6330 ASSERT(cnp->cn_flags & SAVENAME); 6331 6332 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0)); 6333} 6334 6335static int 6336zfs_freebsd_readdir(ap) 6337 struct vop_readdir_args /* { 6338 struct vnode *a_vp; 6339 struct uio *a_uio; 6340 struct ucred *a_cred; 6341 int *a_eofflag; 6342 int *a_ncookies; 6343 u_long **a_cookies; 6344 } */ *ap; 6345{ 6346 6347 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 6348 ap->a_ncookies, ap->a_cookies)); 6349} 6350 6351static int 6352zfs_freebsd_fsync(ap) 6353 struct vop_fsync_args /* { 6354 struct vnode *a_vp; 6355 int a_waitfor; 6356 struct thread *a_td; 6357 } */ *ap; 6358{ 6359 6360 vop_stdfsync(ap); 6361 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL)); 6362} 6363 6364static int 6365zfs_freebsd_getattr(ap) 6366 struct vop_getattr_args /* { 6367 struct vnode *a_vp; 6368 struct vattr *a_vap; 6369 struct ucred *a_cred; 6370 } */ *ap; 6371{ 6372 vattr_t *vap = ap->a_vap; 6373 xvattr_t xvap; 6374 u_long fflags = 0; 6375 int error; 6376 6377 xva_init(&xvap); 6378 xvap.xva_vattr = *vap; 6379 xvap.xva_vattr.va_mask |= AT_XVATTR; 6380 6381 /* Convert chflags into ZFS-type flags. */ 6382 /* XXX: what about SF_SETTABLE?. */ 6383 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 6384 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 6385 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 6386 XVA_SET_REQ(&xvap, XAT_NODUMP); 6387 XVA_SET_REQ(&xvap, XAT_READONLY); 6388 XVA_SET_REQ(&xvap, XAT_ARCHIVE); 6389 XVA_SET_REQ(&xvap, XAT_SYSTEM); 6390 XVA_SET_REQ(&xvap, XAT_HIDDEN); 6391 XVA_SET_REQ(&xvap, XAT_REPARSE); 6392 XVA_SET_REQ(&xvap, XAT_OFFLINE); 6393 XVA_SET_REQ(&xvap, XAT_SPARSE); 6394 6395 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL); 6396 if (error != 0) 6397 return (error); 6398 6399 /* Convert ZFS xattr into chflags. */ 6400#define FLAG_CHECK(fflag, xflag, xfield) do { \ 6401 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 6402 fflags |= (fflag); \ 6403} while (0) 6404 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 6405 xvap.xva_xoptattrs.xoa_immutable); 6406 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 6407 xvap.xva_xoptattrs.xoa_appendonly); 6408 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 6409 xvap.xva_xoptattrs.xoa_nounlink); 6410 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE, 6411 xvap.xva_xoptattrs.xoa_archive); 6412 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 6413 xvap.xva_xoptattrs.xoa_nodump); 6414 FLAG_CHECK(UF_READONLY, XAT_READONLY, 6415 xvap.xva_xoptattrs.xoa_readonly); 6416 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM, 6417 xvap.xva_xoptattrs.xoa_system); 6418 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN, 6419 xvap.xva_xoptattrs.xoa_hidden); 6420 FLAG_CHECK(UF_REPARSE, XAT_REPARSE, 6421 xvap.xva_xoptattrs.xoa_reparse); 6422 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE, 6423 xvap.xva_xoptattrs.xoa_offline); 6424 FLAG_CHECK(UF_SPARSE, XAT_SPARSE, 6425 xvap.xva_xoptattrs.xoa_sparse); 6426 6427#undef FLAG_CHECK 6428 *vap = xvap.xva_vattr; 6429 vap->va_flags = fflags; 6430 return (0); 6431} 6432 6433static int 6434zfs_freebsd_setattr(ap) 6435 struct vop_setattr_args /* { 6436 struct vnode *a_vp; 6437 struct vattr *a_vap; 6438 struct ucred *a_cred; 6439 } */ *ap; 6440{ 6441 vnode_t *vp = ap->a_vp; 6442 vattr_t *vap = ap->a_vap; 6443 cred_t *cred = ap->a_cred; 6444 xvattr_t xvap; 6445 u_long fflags; 6446 uint64_t zflags; 6447 6448 vattr_init_mask(vap); 6449 vap->va_mask &= ~AT_NOSET; 6450 6451 xva_init(&xvap); 6452 xvap.xva_vattr = *vap; 6453 6454 zflags = VTOZ(vp)->z_pflags; 6455 6456 if (vap->va_flags != VNOVAL) { 6457 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 6458 int error; 6459 6460 if (zfsvfs->z_use_fuids == B_FALSE) 6461 return (EOPNOTSUPP); 6462 6463 fflags = vap->va_flags; 6464 /* 6465 * XXX KDM 6466 * We need to figure out whether it makes sense to allow 6467 * UF_REPARSE through, since we don't really have other 6468 * facilities to handle reparse points and zfs_setattr() 6469 * doesn't currently allow setting that attribute anyway. 6470 */ 6471 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE| 6472 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE| 6473 UF_OFFLINE|UF_SPARSE)) != 0) 6474 return (EOPNOTSUPP); 6475 /* 6476 * Unprivileged processes are not permitted to unset system 6477 * flags, or modify flags if any system flags are set. 6478 * Privileged non-jail processes may not modify system flags 6479 * if securelevel > 0 and any existing system flags are set. 6480 * Privileged jail processes behave like privileged non-jail 6481 * processes if the security.jail.chflags_allowed sysctl is 6482 * is non-zero; otherwise, they behave like unprivileged 6483 * processes. 6484 */ 6485 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 6486 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) { 6487 if (zflags & 6488 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6489 error = securelevel_gt(cred, 0); 6490 if (error != 0) 6491 return (error); 6492 } 6493 } else { 6494 /* 6495 * Callers may only modify the file flags on objects they 6496 * have VADMIN rights for. 6497 */ 6498 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0) 6499 return (error); 6500 if (zflags & 6501 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6502 return (EPERM); 6503 } 6504 if (fflags & 6505 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 6506 return (EPERM); 6507 } 6508 } 6509 6510#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 6511 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 6512 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 6513 XVA_SET_REQ(&xvap, (xflag)); \ 6514 (xfield) = ((fflags & (fflag)) != 0); \ 6515 } \ 6516} while (0) 6517 /* Convert chflags into ZFS-type flags. */ 6518 /* XXX: what about SF_SETTABLE?. */ 6519 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 6520 xvap.xva_xoptattrs.xoa_immutable); 6521 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 6522 xvap.xva_xoptattrs.xoa_appendonly); 6523 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 6524 xvap.xva_xoptattrs.xoa_nounlink); 6525 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE, 6526 xvap.xva_xoptattrs.xoa_archive); 6527 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 6528 xvap.xva_xoptattrs.xoa_nodump); 6529 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY, 6530 xvap.xva_xoptattrs.xoa_readonly); 6531 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM, 6532 xvap.xva_xoptattrs.xoa_system); 6533 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN, 6534 xvap.xva_xoptattrs.xoa_hidden); 6535 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE, 6536 xvap.xva_xoptattrs.xoa_hidden); 6537 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE, 6538 xvap.xva_xoptattrs.xoa_offline); 6539 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE, 6540 xvap.xva_xoptattrs.xoa_sparse); 6541#undef FLAG_CHANGE 6542 } 6543 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL)); 6544} 6545 6546static int 6547zfs_freebsd_rename(ap) 6548 struct vop_rename_args /* { 6549 struct vnode *a_fdvp; 6550 struct vnode *a_fvp; 6551 struct componentname *a_fcnp; 6552 struct vnode *a_tdvp; 6553 struct vnode *a_tvp; 6554 struct componentname *a_tcnp; 6555 } */ *ap; 6556{ 6557 vnode_t *fdvp = ap->a_fdvp; 6558 vnode_t *fvp = ap->a_fvp; 6559 vnode_t *tdvp = ap->a_tdvp; 6560 vnode_t *tvp = ap->a_tvp; 6561 int error; 6562 6563 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 6564 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 6565 6566 /* 6567 * Check for cross-device rename. 6568 */ 6569 if ((fdvp->v_mount != tdvp->v_mount) || 6570 (tvp && (fdvp->v_mount != tvp->v_mount))) 6571 error = EXDEV; 6572 else 6573 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 6574 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0); 6575 if (tdvp == tvp) 6576 VN_RELE(tdvp); 6577 else 6578 VN_URELE(tdvp); 6579 if (tvp) 6580 VN_URELE(tvp); 6581 VN_RELE(fdvp); 6582 VN_RELE(fvp); 6583 6584 return (error); 6585} 6586 6587static int 6588zfs_freebsd_symlink(ap) 6589 struct vop_symlink_args /* { 6590 struct vnode *a_dvp; 6591 struct vnode **a_vpp; 6592 struct componentname *a_cnp; 6593 struct vattr *a_vap; 6594 char *a_target; 6595 } */ *ap; 6596{ 6597 struct componentname *cnp = ap->a_cnp; 6598 vattr_t *vap = ap->a_vap; 6599 6600 ASSERT(cnp->cn_flags & SAVENAME); 6601 6602 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 6603 vattr_init_mask(vap); 6604 6605 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 6606 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 6607} 6608 6609static int 6610zfs_freebsd_readlink(ap) 6611 struct vop_readlink_args /* { 6612 struct vnode *a_vp; 6613 struct uio *a_uio; 6614 struct ucred *a_cred; 6615 } */ *ap; 6616{ 6617 6618 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL)); 6619} 6620 6621static int 6622zfs_freebsd_link(ap) 6623 struct vop_link_args /* { 6624 struct vnode *a_tdvp; 6625 struct vnode *a_vp; 6626 struct componentname *a_cnp; 6627 } */ *ap; 6628{ 6629 struct componentname *cnp = ap->a_cnp; 6630 vnode_t *vp = ap->a_vp; 6631 vnode_t *tdvp = ap->a_tdvp; 6632 6633 if (tdvp->v_mount != vp->v_mount) 6634 return (EXDEV); 6635 6636 ASSERT(cnp->cn_flags & SAVENAME); 6637 6638 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0)); 6639} 6640 6641static int 6642zfs_freebsd_inactive(ap) 6643 struct vop_inactive_args /* { 6644 struct vnode *a_vp; 6645 struct thread *a_td; 6646 } */ *ap; 6647{ 6648 vnode_t *vp = ap->a_vp; 6649 6650 zfs_inactive(vp, ap->a_td->td_ucred, NULL); 6651 return (0); 6652} 6653 6654static int 6655zfs_freebsd_reclaim(ap) 6656 struct vop_reclaim_args /* { 6657 struct vnode *a_vp; 6658 struct thread *a_td; 6659 } */ *ap; 6660{ 6661 vnode_t *vp = ap->a_vp; 6662 znode_t *zp = VTOZ(vp); 6663 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6664 6665 ASSERT(zp != NULL); 6666 6667 /* Destroy the vm object and flush associated pages. */ 6668 vnode_destroy_vobject(vp); 6669 6670 /* 6671 * z_teardown_inactive_lock protects from a race with 6672 * zfs_znode_dmu_fini in zfsvfs_teardown during 6673 * force unmount. 6674 */ 6675 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 6676 if (zp->z_sa_hdl == NULL) 6677 zfs_znode_free(zp); 6678 else 6679 zfs_zinactive(zp); 6680 rw_exit(&zfsvfs->z_teardown_inactive_lock); 6681 6682 vp->v_data = NULL; 6683 return (0); 6684} 6685 6686static int 6687zfs_freebsd_fid(ap) 6688 struct vop_fid_args /* { 6689 struct vnode *a_vp; 6690 struct fid *a_fid; 6691 } */ *ap; 6692{ 6693 6694 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 6695} 6696 6697static int 6698zfs_freebsd_pathconf(ap) 6699 struct vop_pathconf_args /* { 6700 struct vnode *a_vp; 6701 int a_name; 6702 register_t *a_retval; 6703 } */ *ap; 6704{ 6705 ulong_t val; 6706 int error; 6707 6708 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); 6709 if (error == 0) 6710 *ap->a_retval = val; 6711 else if (error == EOPNOTSUPP) 6712 error = vop_stdpathconf(ap); 6713 return (error); 6714} 6715 6716static int 6717zfs_freebsd_fifo_pathconf(ap) 6718 struct vop_pathconf_args /* { 6719 struct vnode *a_vp; 6720 int a_name; 6721 register_t *a_retval; 6722 } */ *ap; 6723{ 6724 6725 switch (ap->a_name) { 6726 case _PC_ACL_EXTENDED: 6727 case _PC_ACL_NFS4: 6728 case _PC_ACL_PATH_MAX: 6729 case _PC_MAC_PRESENT: 6730 return (zfs_freebsd_pathconf(ap)); 6731 default: 6732 return (fifo_specops.vop_pathconf(ap)); 6733 } 6734} 6735 6736/* 6737 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 6738 * extended attribute name: 6739 * 6740 * NAMESPACE PREFIX 6741 * system freebsd:system: 6742 * user (none, can be used to access ZFS fsattr(5) attributes 6743 * created on Solaris) 6744 */ 6745static int 6746zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 6747 size_t size) 6748{ 6749 const char *namespace, *prefix, *suffix; 6750 6751 /* We don't allow '/' character in attribute name. */ 6752 if (strchr(name, '/') != NULL) 6753 return (EINVAL); 6754 /* We don't allow attribute names that start with "freebsd:" string. */ 6755 if (strncmp(name, "freebsd:", 8) == 0) 6756 return (EINVAL); 6757 6758 bzero(attrname, size); 6759 6760 switch (attrnamespace) { 6761 case EXTATTR_NAMESPACE_USER: 6762#if 0 6763 prefix = "freebsd:"; 6764 namespace = EXTATTR_NAMESPACE_USER_STRING; 6765 suffix = ":"; 6766#else 6767 /* 6768 * This is the default namespace by which we can access all 6769 * attributes created on Solaris. 6770 */ 6771 prefix = namespace = suffix = ""; 6772#endif 6773 break; 6774 case EXTATTR_NAMESPACE_SYSTEM: 6775 prefix = "freebsd:"; 6776 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 6777 suffix = ":"; 6778 break; 6779 case EXTATTR_NAMESPACE_EMPTY: 6780 default: 6781 return (EINVAL); 6782 } 6783 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 6784 name) >= size) { 6785 return (ENAMETOOLONG); 6786 } 6787 return (0); 6788} 6789 6790/* 6791 * Vnode operating to retrieve a named extended attribute. 6792 */ 6793static int 6794zfs_getextattr(struct vop_getextattr_args *ap) 6795/* 6796vop_getextattr { 6797 IN struct vnode *a_vp; 6798 IN int a_attrnamespace; 6799 IN const char *a_name; 6800 INOUT struct uio *a_uio; 6801 OUT size_t *a_size; 6802 IN struct ucred *a_cred; 6803 IN struct thread *a_td; 6804}; 6805*/ 6806{ 6807 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6808 struct thread *td = ap->a_td; 6809 struct nameidata nd; 6810 char attrname[255]; 6811 struct vattr va; 6812 vnode_t *xvp = NULL, *vp; 6813 int error, flags; 6814 6815 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6816 ap->a_cred, ap->a_td, VREAD); 6817 if (error != 0) 6818 return (error); 6819 6820 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6821 sizeof(attrname)); 6822 if (error != 0) 6823 return (error); 6824 6825 ZFS_ENTER(zfsvfs); 6826 6827 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6828 LOOKUP_XATTR); 6829 if (error != 0) { 6830 ZFS_EXIT(zfsvfs); 6831 return (error); 6832 } 6833 6834 flags = FREAD; 6835 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6836 xvp, td); 6837 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL); 6838 vp = nd.ni_vp; 6839 NDFREE(&nd, NDF_ONLY_PNBUF); 6840 if (error != 0) { 6841 ZFS_EXIT(zfsvfs); 6842 if (error == ENOENT) 6843 error = ENOATTR; 6844 return (error); 6845 } 6846 6847 if (ap->a_size != NULL) { 6848 error = VOP_GETATTR(vp, &va, ap->a_cred); 6849 if (error == 0) 6850 *ap->a_size = (size_t)va.va_size; 6851 } else if (ap->a_uio != NULL) 6852 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6853 6854 VOP_UNLOCK(vp, 0); 6855 vn_close(vp, flags, ap->a_cred, td); 6856 ZFS_EXIT(zfsvfs); 6857 6858 return (error); 6859} 6860 6861/* 6862 * Vnode operation to remove a named attribute. 6863 */ 6864int 6865zfs_deleteextattr(struct vop_deleteextattr_args *ap) 6866/* 6867vop_deleteextattr { 6868 IN struct vnode *a_vp; 6869 IN int a_attrnamespace; 6870 IN const char *a_name; 6871 IN struct ucred *a_cred; 6872 IN struct thread *a_td; 6873}; 6874*/ 6875{ 6876 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6877 struct thread *td = ap->a_td; 6878 struct nameidata nd; 6879 char attrname[255]; 6880 struct vattr va; 6881 vnode_t *xvp = NULL, *vp; 6882 int error, flags; 6883 6884 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6885 ap->a_cred, ap->a_td, VWRITE); 6886 if (error != 0) 6887 return (error); 6888 6889 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6890 sizeof(attrname)); 6891 if (error != 0) 6892 return (error); 6893 6894 ZFS_ENTER(zfsvfs); 6895 6896 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6897 LOOKUP_XATTR); 6898 if (error != 0) { 6899 ZFS_EXIT(zfsvfs); 6900 return (error); 6901 } 6902 6903 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 6904 UIO_SYSSPACE, attrname, xvp, td); 6905 error = namei(&nd); 6906 vp = nd.ni_vp; 6907 if (error != 0) { 6908 ZFS_EXIT(zfsvfs); 6909 NDFREE(&nd, NDF_ONLY_PNBUF); 6910 if (error == ENOENT) 6911 error = ENOATTR; 6912 return (error); 6913 } 6914 6915 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 6916 NDFREE(&nd, NDF_ONLY_PNBUF); 6917 6918 vput(nd.ni_dvp); 6919 if (vp == nd.ni_dvp) 6920 vrele(vp); 6921 else 6922 vput(vp); 6923 ZFS_EXIT(zfsvfs); 6924 6925 return (error); 6926} 6927 6928/* 6929 * Vnode operation to set a named attribute. 6930 */ 6931static int 6932zfs_setextattr(struct vop_setextattr_args *ap) 6933/* 6934vop_setextattr { 6935 IN struct vnode *a_vp; 6936 IN int a_attrnamespace; 6937 IN const char *a_name; 6938 INOUT struct uio *a_uio; 6939 IN struct ucred *a_cred; 6940 IN struct thread *a_td; 6941}; 6942*/ 6943{ 6944 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6945 struct thread *td = ap->a_td; 6946 struct nameidata nd; 6947 char attrname[255]; 6948 struct vattr va; 6949 vnode_t *xvp = NULL, *vp; 6950 int error, flags; 6951 6952 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6953 ap->a_cred, ap->a_td, VWRITE); 6954 if (error != 0) 6955 return (error); 6956 6957 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6958 sizeof(attrname)); 6959 if (error != 0) 6960 return (error); 6961 6962 ZFS_ENTER(zfsvfs); 6963 6964 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6965 LOOKUP_XATTR | CREATE_XATTR_DIR); 6966 if (error != 0) { 6967 ZFS_EXIT(zfsvfs); 6968 return (error); 6969 } 6970 6971 flags = FFLAGS(O_WRONLY | O_CREAT); 6972 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6973 xvp, td); 6974 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL); 6975 vp = nd.ni_vp; 6976 NDFREE(&nd, NDF_ONLY_PNBUF); 6977 if (error != 0) { 6978 ZFS_EXIT(zfsvfs); 6979 return (error); 6980 } 6981 6982 VATTR_NULL(&va); 6983 va.va_size = 0; 6984 error = VOP_SETATTR(vp, &va, ap->a_cred); 6985 if (error == 0) 6986 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6987 6988 VOP_UNLOCK(vp, 0); 6989 vn_close(vp, flags, ap->a_cred, td); 6990 ZFS_EXIT(zfsvfs); 6991 6992 return (error); 6993} 6994 6995/* 6996 * Vnode operation to retrieve extended attributes on a vnode. 6997 */ 6998static int 6999zfs_listextattr(struct vop_listextattr_args *ap) 7000/* 7001vop_listextattr { 7002 IN struct vnode *a_vp; 7003 IN int a_attrnamespace; 7004 INOUT struct uio *a_uio; 7005 OUT size_t *a_size; 7006 IN struct ucred *a_cred; 7007 IN struct thread *a_td; 7008}; 7009*/ 7010{ 7011 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 7012 struct thread *td = ap->a_td; 7013 struct nameidata nd; 7014 char attrprefix[16]; 7015 u_char dirbuf[sizeof(struct dirent)]; 7016 struct dirent *dp; 7017 struct iovec aiov; 7018 struct uio auio, *uio = ap->a_uio; 7019 size_t *sizep = ap->a_size; 7020 size_t plen; 7021 vnode_t *xvp = NULL, *vp; 7022 int done, error, eof, pos; 7023 7024 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 7025 ap->a_cred, ap->a_td, VREAD); 7026 if (error != 0) 7027 return (error); 7028 7029 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 7030 sizeof(attrprefix)); 7031 if (error != 0) 7032 return (error); 7033 plen = strlen(attrprefix); 7034 7035 ZFS_ENTER(zfsvfs); 7036 7037 if (sizep != NULL) 7038 *sizep = 0; 7039 7040 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 7041 LOOKUP_XATTR); 7042 if (error != 0) { 7043 ZFS_EXIT(zfsvfs); 7044 /* 7045 * ENOATTR means that the EA directory does not yet exist, 7046 * i.e. there are no extended attributes there. 7047 */ 7048 if (error == ENOATTR) 7049 error = 0; 7050 return (error); 7051 } 7052 7053 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 7054 UIO_SYSSPACE, ".", xvp, td); 7055 error = namei(&nd); 7056 vp = nd.ni_vp; 7057 NDFREE(&nd, NDF_ONLY_PNBUF); 7058 if (error != 0) { 7059 ZFS_EXIT(zfsvfs); 7060 return (error); 7061 } 7062 7063 auio.uio_iov = &aiov; 7064 auio.uio_iovcnt = 1; 7065 auio.uio_segflg = UIO_SYSSPACE; 7066 auio.uio_td = td; 7067 auio.uio_rw = UIO_READ; 7068 auio.uio_offset = 0; 7069 7070 do { 7071 u_char nlen; 7072 7073 aiov.iov_base = (void *)dirbuf; 7074 aiov.iov_len = sizeof(dirbuf); 7075 auio.uio_resid = sizeof(dirbuf); 7076 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 7077 done = sizeof(dirbuf) - auio.uio_resid; 7078 if (error != 0) 7079 break; 7080 for (pos = 0; pos < done;) { 7081 dp = (struct dirent *)(dirbuf + pos); 7082 pos += dp->d_reclen; 7083 /* 7084 * XXX: Temporarily we also accept DT_UNKNOWN, as this 7085 * is what we get when attribute was created on Solaris. 7086 */ 7087 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 7088 continue; 7089 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0) 7090 continue; 7091 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 7092 continue; 7093 nlen = dp->d_namlen - plen; 7094 if (sizep != NULL) 7095 *sizep += 1 + nlen; 7096 else if (uio != NULL) { 7097 /* 7098 * Format of extattr name entry is one byte for 7099 * length and the rest for name. 7100 */ 7101 error = uiomove(&nlen, 1, uio->uio_rw, uio); 7102 if (error == 0) { 7103 error = uiomove(dp->d_name + plen, nlen, 7104 uio->uio_rw, uio); 7105 } 7106 if (error != 0) 7107 break; 7108 } 7109 } 7110 } while (!eof && error == 0); 7111 7112 vput(vp); 7113 ZFS_EXIT(zfsvfs); 7114 7115 return (error); 7116} 7117 7118int 7119zfs_freebsd_getacl(ap) 7120 struct vop_getacl_args /* { 7121 struct vnode *vp; 7122 acl_type_t type; 7123 struct acl *aclp; 7124 struct ucred *cred; 7125 struct thread *td; 7126 } */ *ap; 7127{ 7128 int error; 7129 vsecattr_t vsecattr; 7130 7131 if (ap->a_type != ACL_TYPE_NFS4) 7132 return (EINVAL); 7133 7134 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 7135 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL)) 7136 return (error); 7137 7138 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt); 7139 if (vsecattr.vsa_aclentp != NULL) 7140 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 7141 7142 return (error); 7143} 7144 7145int 7146zfs_freebsd_setacl(ap) 7147 struct vop_setacl_args /* { 7148 struct vnode *vp; 7149 acl_type_t type; 7150 struct acl *aclp; 7151 struct ucred *cred; 7152 struct thread *td; 7153 } */ *ap; 7154{ 7155 int error; 7156 vsecattr_t vsecattr; 7157 int aclbsize; /* size of acl list in bytes */ 7158 aclent_t *aaclp; 7159 7160 if (ap->a_type != ACL_TYPE_NFS4) 7161 return (EINVAL); 7162 7163 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 7164 return (EINVAL); 7165 7166 /* 7167 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 7168 * splitting every entry into two and appending "canonical six" 7169 * entries at the end. Don't allow for setting an ACL that would 7170 * cause chmod(2) to run out of ACL entries. 7171 */ 7172 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 7173 return (ENOSPC); 7174 7175 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 7176 if (error != 0) 7177 return (error); 7178 7179 vsecattr.vsa_mask = VSA_ACE; 7180 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t); 7181 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 7182 aaclp = vsecattr.vsa_aclentp; 7183 vsecattr.vsa_aclentsz = aclbsize; 7184 7185 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 7186 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL); 7187 kmem_free(aaclp, aclbsize); 7188 7189 return (error); 7190} 7191 7192int 7193zfs_freebsd_aclcheck(ap) 7194 struct vop_aclcheck_args /* { 7195 struct vnode *vp; 7196 acl_type_t type; 7197 struct acl *aclp; 7198 struct ucred *cred; 7199 struct thread *td; 7200 } */ *ap; 7201{ 7202 7203 return (EOPNOTSUPP); 7204} 7205 7206static int 7207zfs_vptocnp(struct vop_vptocnp_args *ap) 7208{ 7209 vnode_t *covered_vp; 7210 vnode_t *vp = ap->a_vp;; 7211 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 7212 znode_t *zp = VTOZ(vp); 7213 uint64_t parent; 7214 int ltype; 7215 int error; 7216 7217 ZFS_ENTER(zfsvfs); 7218 ZFS_VERIFY_ZP(zp); 7219 7220 /* 7221 * If we are a snapshot mounted under .zfs, run the operation 7222 * on the covered vnode. 7223 */ 7224 if ((error = sa_lookup(zp->z_sa_hdl, 7225 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) { 7226 ZFS_EXIT(zfsvfs); 7227 return (error); 7228 } 7229 7230 if (zp->z_id != parent || zfsvfs->z_parent == zfsvfs) { 7231 ZFS_EXIT(zfsvfs); 7232 return (vop_stdvptocnp(ap)); 7233 } 7234 ZFS_EXIT(zfsvfs); 7235 7236 covered_vp = vp->v_mount->mnt_vnodecovered; 7237 vhold(covered_vp); 7238 ltype = VOP_ISLOCKED(vp); 7239 VOP_UNLOCK(vp, 0); 7240 error = vget(covered_vp, LK_EXCLUSIVE, curthread); 7241 vdrop(covered_vp); 7242 if (error == 0) { 7243 error = VOP_VPTOCNP(covered_vp, ap->a_vpp, ap->a_cred, 7244 ap->a_buf, ap->a_buflen); 7245 vput(covered_vp); 7246 } 7247 vn_lock(vp, ltype | LK_RETRY); 7248 if ((vp->v_iflag & VI_DOOMED) != 0) 7249 error = SET_ERROR(ENOENT); 7250 return (error); 7251} 7252 7253struct vop_vector zfs_vnodeops; 7254struct vop_vector zfs_fifoops; 7255struct vop_vector zfs_shareops; 7256 7257struct vop_vector zfs_vnodeops = { 7258 .vop_default = &default_vnodeops, 7259 .vop_inactive = zfs_freebsd_inactive, 7260 .vop_reclaim = zfs_freebsd_reclaim, 7261 .vop_access = zfs_freebsd_access, 7262#ifdef FREEBSD_NAMECACHE 7263 .vop_lookup = vfs_cache_lookup, 7264 .vop_cachedlookup = zfs_freebsd_lookup, 7265#else 7266 .vop_lookup = zfs_freebsd_lookup, 7267#endif 7268 .vop_getattr = zfs_freebsd_getattr, 7269 .vop_setattr = zfs_freebsd_setattr, 7270 .vop_create = zfs_freebsd_create, 7271 .vop_mknod = zfs_freebsd_create, 7272 .vop_mkdir = zfs_freebsd_mkdir, 7273 .vop_readdir = zfs_freebsd_readdir, 7274 .vop_fsync = zfs_freebsd_fsync, 7275 .vop_open = zfs_freebsd_open, 7276 .vop_close = zfs_freebsd_close, 7277 .vop_rmdir = zfs_freebsd_rmdir, 7278 .vop_ioctl = zfs_freebsd_ioctl, 7279 .vop_link = zfs_freebsd_link, 7280 .vop_symlink = zfs_freebsd_symlink, 7281 .vop_readlink = zfs_freebsd_readlink, 7282 .vop_read = zfs_freebsd_read, 7283 .vop_write = zfs_freebsd_write, 7284 .vop_remove = zfs_freebsd_remove, 7285 .vop_rename = zfs_freebsd_rename, 7286 .vop_pathconf = zfs_freebsd_pathconf, 7287 .vop_bmap = zfs_freebsd_bmap, 7288 .vop_fid = zfs_freebsd_fid, 7289 .vop_getextattr = zfs_getextattr, 7290 .vop_deleteextattr = zfs_deleteextattr, 7291 .vop_setextattr = zfs_setextattr, 7292 .vop_listextattr = zfs_listextattr, 7293 .vop_getacl = zfs_freebsd_getacl, 7294 .vop_setacl = zfs_freebsd_setacl, 7295 .vop_aclcheck = zfs_freebsd_aclcheck, 7296 .vop_getpages = zfs_freebsd_getpages, 7297 .vop_putpages = zfs_freebsd_putpages, 7298 .vop_vptocnp = zfs_vptocnp, 7299}; 7300 7301struct vop_vector zfs_fifoops = { 7302 .vop_default = &fifo_specops, 7303 .vop_fsync = zfs_freebsd_fsync, 7304 .vop_access = zfs_freebsd_access, 7305 .vop_getattr = zfs_freebsd_getattr, 7306 .vop_inactive = zfs_freebsd_inactive, 7307 .vop_read = VOP_PANIC, 7308 .vop_reclaim = zfs_freebsd_reclaim, 7309 .vop_setattr = zfs_freebsd_setattr, 7310 .vop_write = VOP_PANIC, 7311 .vop_pathconf = zfs_freebsd_fifo_pathconf, 7312 .vop_fid = zfs_freebsd_fid, 7313 .vop_getacl = zfs_freebsd_getacl, 7314 .vop_setacl = zfs_freebsd_setacl, 7315 .vop_aclcheck = zfs_freebsd_aclcheck, 7316}; 7317 7318/* 7319 * special share hidden files vnode operations template 7320 */ 7321struct vop_vector zfs_shareops = { 7322 .vop_default = &default_vnodeops, 7323 .vop_access = zfs_freebsd_access, 7324 .vop_inactive = zfs_freebsd_inactive, 7325 .vop_reclaim = zfs_freebsd_reclaim, 7326 .vop_fid = zfs_freebsd_fid, 7327 .vop_pathconf = zfs_freebsd_pathconf, 7328}; 7329