zfs_vnops.c revision 269061
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2013, 2014 by Delphix. All rights reserved. 24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved. 25 */ 26 27/* Portions Copyright 2007 Jeremy Teo */ 28/* Portions Copyright 2010 Robert Milkowski */ 29 30#include <sys/types.h> 31#include <sys/param.h> 32#include <sys/time.h> 33#include <sys/systm.h> 34#include <sys/sysmacros.h> 35#include <sys/resource.h> 36#include <sys/vfs.h> 37#include <sys/vm.h> 38#include <sys/vnode.h> 39#include <sys/file.h> 40#include <sys/stat.h> 41#include <sys/kmem.h> 42#include <sys/taskq.h> 43#include <sys/uio.h> 44#include <sys/atomic.h> 45#include <sys/namei.h> 46#include <sys/mman.h> 47#include <sys/cmn_err.h> 48#include <sys/errno.h> 49#include <sys/unistd.h> 50#include <sys/zfs_dir.h> 51#include <sys/zfs_ioctl.h> 52#include <sys/fs/zfs.h> 53#include <sys/dmu.h> 54#include <sys/dmu_objset.h> 55#include <sys/spa.h> 56#include <sys/txg.h> 57#include <sys/dbuf.h> 58#include <sys/zap.h> 59#include <sys/sa.h> 60#include <sys/dirent.h> 61#include <sys/policy.h> 62#include <sys/sunddi.h> 63#include <sys/filio.h> 64#include <sys/sid.h> 65#include <sys/zfs_ctldir.h> 66#include <sys/zfs_fuid.h> 67#include <sys/zfs_sa.h> 68#include <sys/dnlc.h> 69#include <sys/zfs_rlock.h> 70#include <sys/extdirent.h> 71#include <sys/kidmap.h> 72#include <sys/bio.h> 73#include <sys/buf.h> 74#include <sys/sched.h> 75#include <sys/acl.h> 76#include <vm/vm_param.h> 77#include <vm/vm_pageout.h> 78 79/* 80 * Programming rules. 81 * 82 * Each vnode op performs some logical unit of work. To do this, the ZPL must 83 * properly lock its in-core state, create a DMU transaction, do the work, 84 * record this work in the intent log (ZIL), commit the DMU transaction, 85 * and wait for the intent log to commit if it is a synchronous operation. 86 * Moreover, the vnode ops must work in both normal and log replay context. 87 * The ordering of events is important to avoid deadlocks and references 88 * to freed memory. The example below illustrates the following Big Rules: 89 * 90 * (1) A check must be made in each zfs thread for a mounted file system. 91 * This is done avoiding races using ZFS_ENTER(zfsvfs). 92 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes 93 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros 94 * can return EIO from the calling function. 95 * 96 * (2) VN_RELE() should always be the last thing except for zil_commit() 97 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 98 * First, if it's the last reference, the vnode/znode 99 * can be freed, so the zp may point to freed memory. Second, the last 100 * reference will call zfs_zinactive(), which may induce a lot of work -- 101 * pushing cached pages (which acquires range locks) and syncing out 102 * cached atime changes. Third, zfs_zinactive() may require a new tx, 103 * which could deadlock the system if you were already holding one. 104 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). 105 * 106 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 107 * as they can span dmu_tx_assign() calls. 108 * 109 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to 110 * dmu_tx_assign(). This is critical because we don't want to block 111 * while holding locks. 112 * 113 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This 114 * reduces lock contention and CPU usage when we must wait (note that if 115 * throughput is constrained by the storage, nearly every transaction 116 * must wait). 117 * 118 * Note, in particular, that if a lock is sometimes acquired before 119 * the tx assigns, and sometimes after (e.g. z_lock), then failing 120 * to use a non-blocking assign can deadlock the system. The scenario: 121 * 122 * Thread A has grabbed a lock before calling dmu_tx_assign(). 123 * Thread B is in an already-assigned tx, and blocks for this lock. 124 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 125 * forever, because the previous txg can't quiesce until B's tx commits. 126 * 127 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 128 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent 129 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT, 130 * to indicate that this operation has already called dmu_tx_wait(). 131 * This will ensure that we don't retry forever, waiting a short bit 132 * each time. 133 * 134 * (5) If the operation succeeded, generate the intent log entry for it 135 * before dropping locks. This ensures that the ordering of events 136 * in the intent log matches the order in which they actually occurred. 137 * During ZIL replay the zfs_log_* functions will update the sequence 138 * number to indicate the zil transaction has replayed. 139 * 140 * (6) At the end of each vnode op, the DMU tx must always commit, 141 * regardless of whether there were any errors. 142 * 143 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 144 * to ensure that synchronous semantics are provided when necessary. 145 * 146 * In general, this is how things should be ordered in each vnode op: 147 * 148 * ZFS_ENTER(zfsvfs); // exit if unmounted 149 * top: 150 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 151 * rw_enter(...); // grab any other locks you need 152 * tx = dmu_tx_create(...); // get DMU tx 153 * dmu_tx_hold_*(); // hold each object you might modify 154 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 155 * if (error) { 156 * rw_exit(...); // drop locks 157 * zfs_dirent_unlock(dl); // unlock directory entry 158 * VN_RELE(...); // release held vnodes 159 * if (error == ERESTART) { 160 * waited = B_TRUE; 161 * dmu_tx_wait(tx); 162 * dmu_tx_abort(tx); 163 * goto top; 164 * } 165 * dmu_tx_abort(tx); // abort DMU tx 166 * ZFS_EXIT(zfsvfs); // finished in zfs 167 * return (error); // really out of space 168 * } 169 * error = do_real_work(); // do whatever this VOP does 170 * if (error == 0) 171 * zfs_log_*(...); // on success, make ZIL entry 172 * dmu_tx_commit(tx); // commit DMU tx -- error or not 173 * rw_exit(...); // drop locks 174 * zfs_dirent_unlock(dl); // unlock directory entry 175 * VN_RELE(...); // release held vnodes 176 * zil_commit(zilog, foid); // synchronous when necessary 177 * ZFS_EXIT(zfsvfs); // finished in zfs 178 * return (error); // done, report error 179 */ 180 181/* ARGSUSED */ 182static int 183zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 184{ 185 znode_t *zp = VTOZ(*vpp); 186 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 187 188 ZFS_ENTER(zfsvfs); 189 ZFS_VERIFY_ZP(zp); 190 191 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 192 ((flag & FAPPEND) == 0)) { 193 ZFS_EXIT(zfsvfs); 194 return (SET_ERROR(EPERM)); 195 } 196 197 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 198 ZTOV(zp)->v_type == VREG && 199 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) { 200 if (fs_vscan(*vpp, cr, 0) != 0) { 201 ZFS_EXIT(zfsvfs); 202 return (SET_ERROR(EACCES)); 203 } 204 } 205 206 /* Keep a count of the synchronous opens in the znode */ 207 if (flag & (FSYNC | FDSYNC)) 208 atomic_inc_32(&zp->z_sync_cnt); 209 210 ZFS_EXIT(zfsvfs); 211 return (0); 212} 213 214/* ARGSUSED */ 215static int 216zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 217 caller_context_t *ct) 218{ 219 znode_t *zp = VTOZ(vp); 220 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 221 222 /* 223 * Clean up any locks held by this process on the vp. 224 */ 225 cleanlocks(vp, ddi_get_pid(), 0); 226 cleanshares(vp, ddi_get_pid()); 227 228 ZFS_ENTER(zfsvfs); 229 ZFS_VERIFY_ZP(zp); 230 231 /* Decrement the synchronous opens in the znode */ 232 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 233 atomic_dec_32(&zp->z_sync_cnt); 234 235 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 236 ZTOV(zp)->v_type == VREG && 237 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) 238 VERIFY(fs_vscan(vp, cr, 1) == 0); 239 240 ZFS_EXIT(zfsvfs); 241 return (0); 242} 243 244/* 245 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 246 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 247 */ 248static int 249zfs_holey(vnode_t *vp, u_long cmd, offset_t *off) 250{ 251 znode_t *zp = VTOZ(vp); 252 uint64_t noff = (uint64_t)*off; /* new offset */ 253 uint64_t file_sz; 254 int error; 255 boolean_t hole; 256 257 file_sz = zp->z_size; 258 if (noff >= file_sz) { 259 return (SET_ERROR(ENXIO)); 260 } 261 262 if (cmd == _FIO_SEEK_HOLE) 263 hole = B_TRUE; 264 else 265 hole = B_FALSE; 266 267 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 268 269 /* end of file? */ 270 if ((error == ESRCH) || (noff > file_sz)) { 271 /* 272 * Handle the virtual hole at the end of file. 273 */ 274 if (hole) { 275 *off = file_sz; 276 return (0); 277 } 278 return (SET_ERROR(ENXIO)); 279 } 280 281 if (noff < *off) 282 return (error); 283 *off = noff; 284 return (error); 285} 286 287/* ARGSUSED */ 288static int 289zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred, 290 int *rvalp, caller_context_t *ct) 291{ 292 offset_t off; 293 int error; 294 zfsvfs_t *zfsvfs; 295 znode_t *zp; 296 297 switch (com) { 298 case _FIOFFS: 299 return (0); 300 301 /* 302 * The following two ioctls are used by bfu. Faking out, 303 * necessary to avoid bfu errors. 304 */ 305 case _FIOGDIO: 306 case _FIOSDIO: 307 return (0); 308 309 case _FIO_SEEK_DATA: 310 case _FIO_SEEK_HOLE: 311#ifdef sun 312 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 313 return (SET_ERROR(EFAULT)); 314#else 315 off = *(offset_t *)data; 316#endif 317 zp = VTOZ(vp); 318 zfsvfs = zp->z_zfsvfs; 319 ZFS_ENTER(zfsvfs); 320 ZFS_VERIFY_ZP(zp); 321 322 /* offset parameter is in/out */ 323 error = zfs_holey(vp, com, &off); 324 ZFS_EXIT(zfsvfs); 325 if (error) 326 return (error); 327#ifdef sun 328 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 329 return (SET_ERROR(EFAULT)); 330#else 331 *(offset_t *)data = off; 332#endif 333 return (0); 334 } 335 return (SET_ERROR(ENOTTY)); 336} 337 338static vm_page_t 339page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) 340{ 341 vm_object_t obj; 342 vm_page_t pp; 343 int64_t end; 344 345 /* 346 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE 347 * aligned boundaries, if the range is not aligned. As a result a 348 * DEV_BSIZE subrange with partially dirty data may get marked as clean. 349 * It may happen that all DEV_BSIZE subranges are marked clean and thus 350 * the whole page would be considred clean despite have some dirty data. 351 * For this reason we should shrink the range to DEV_BSIZE aligned 352 * boundaries before calling vm_page_clear_dirty. 353 */ 354 end = rounddown2(off + nbytes, DEV_BSIZE); 355 off = roundup2(off, DEV_BSIZE); 356 nbytes = end - off; 357 358 obj = vp->v_object; 359 zfs_vmobject_assert_wlocked(obj); 360 361 for (;;) { 362 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 363 pp->valid) { 364 if (vm_page_xbusied(pp)) { 365 /* 366 * Reference the page before unlocking and 367 * sleeping so that the page daemon is less 368 * likely to reclaim it. 369 */ 370 vm_page_reference(pp); 371 vm_page_lock(pp); 372 zfs_vmobject_wunlock(obj); 373 vm_page_busy_sleep(pp, "zfsmwb"); 374 zfs_vmobject_wlock(obj); 375 continue; 376 } 377 vm_page_sbusy(pp); 378 } else if (pp == NULL) { 379 pp = vm_page_alloc(obj, OFF_TO_IDX(start), 380 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED | 381 VM_ALLOC_SBUSY); 382 } else { 383 ASSERT(pp != NULL && !pp->valid); 384 pp = NULL; 385 } 386 387 if (pp != NULL) { 388 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 389 vm_object_pip_add(obj, 1); 390 pmap_remove_write(pp); 391 if (nbytes != 0) 392 vm_page_clear_dirty(pp, off, nbytes); 393 } 394 break; 395 } 396 return (pp); 397} 398 399static void 400page_unbusy(vm_page_t pp) 401{ 402 403 vm_page_sunbusy(pp); 404 vm_object_pip_subtract(pp->object, 1); 405} 406 407static vm_page_t 408page_hold(vnode_t *vp, int64_t start) 409{ 410 vm_object_t obj; 411 vm_page_t pp; 412 413 obj = vp->v_object; 414 zfs_vmobject_assert_wlocked(obj); 415 416 for (;;) { 417 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 418 pp->valid) { 419 if (vm_page_xbusied(pp)) { 420 /* 421 * Reference the page before unlocking and 422 * sleeping so that the page daemon is less 423 * likely to reclaim it. 424 */ 425 vm_page_reference(pp); 426 vm_page_lock(pp); 427 zfs_vmobject_wunlock(obj); 428 vm_page_busy_sleep(pp, "zfsmwb"); 429 zfs_vmobject_wlock(obj); 430 continue; 431 } 432 433 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 434 vm_page_lock(pp); 435 vm_page_hold(pp); 436 vm_page_unlock(pp); 437 438 } else 439 pp = NULL; 440 break; 441 } 442 return (pp); 443} 444 445static void 446page_unhold(vm_page_t pp) 447{ 448 449 vm_page_lock(pp); 450 vm_page_unhold(pp); 451 vm_page_unlock(pp); 452} 453 454/* 455 * When a file is memory mapped, we must keep the IO data synchronized 456 * between the DMU cache and the memory mapped pages. What this means: 457 * 458 * On Write: If we find a memory mapped page, we write to *both* 459 * the page and the dmu buffer. 460 */ 461static void 462update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid, 463 int segflg, dmu_tx_t *tx) 464{ 465 vm_object_t obj; 466 struct sf_buf *sf; 467 caddr_t va; 468 int off; 469 470 ASSERT(segflg != UIO_NOCOPY); 471 ASSERT(vp->v_mount != NULL); 472 obj = vp->v_object; 473 ASSERT(obj != NULL); 474 475 off = start & PAGEOFFSET; 476 zfs_vmobject_wlock(obj); 477 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 478 vm_page_t pp; 479 int nbytes = imin(PAGESIZE - off, len); 480 481 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) { 482 zfs_vmobject_wunlock(obj); 483 484 va = zfs_map_page(pp, &sf); 485 (void) dmu_read(os, oid, start+off, nbytes, 486 va+off, DMU_READ_PREFETCH);; 487 zfs_unmap_page(sf); 488 489 zfs_vmobject_wlock(obj); 490 page_unbusy(pp); 491 } 492 len -= nbytes; 493 off = 0; 494 } 495 vm_object_pip_wakeupn(obj, 0); 496 zfs_vmobject_wunlock(obj); 497} 498 499/* 500 * Read with UIO_NOCOPY flag means that sendfile(2) requests 501 * ZFS to populate a range of page cache pages with data. 502 * 503 * NOTE: this function could be optimized to pre-allocate 504 * all pages in advance, drain exclusive busy on all of them, 505 * map them into contiguous KVA region and populate them 506 * in one single dmu_read() call. 507 */ 508static int 509mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio) 510{ 511 znode_t *zp = VTOZ(vp); 512 objset_t *os = zp->z_zfsvfs->z_os; 513 struct sf_buf *sf; 514 vm_object_t obj; 515 vm_page_t pp; 516 int64_t start; 517 caddr_t va; 518 int len = nbytes; 519 int off; 520 int error = 0; 521 522 ASSERT(uio->uio_segflg == UIO_NOCOPY); 523 ASSERT(vp->v_mount != NULL); 524 obj = vp->v_object; 525 ASSERT(obj != NULL); 526 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0); 527 528 zfs_vmobject_wlock(obj); 529 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) { 530 int bytes = MIN(PAGESIZE, len); 531 532 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY | 533 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); 534 if (pp->valid == 0) { 535 zfs_vmobject_wunlock(obj); 536 va = zfs_map_page(pp, &sf); 537 error = dmu_read(os, zp->z_id, start, bytes, va, 538 DMU_READ_PREFETCH); 539 if (bytes != PAGESIZE && error == 0) 540 bzero(va + bytes, PAGESIZE - bytes); 541 zfs_unmap_page(sf); 542 zfs_vmobject_wlock(obj); 543 vm_page_sunbusy(pp); 544 vm_page_lock(pp); 545 if (error) { 546 if (pp->wire_count == 0 && pp->valid == 0 && 547 !vm_page_busied(pp)) 548 vm_page_free(pp); 549 } else { 550 pp->valid = VM_PAGE_BITS_ALL; 551 vm_page_activate(pp); 552 } 553 vm_page_unlock(pp); 554 } else { 555 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 556 vm_page_sunbusy(pp); 557 } 558 if (error) 559 break; 560 uio->uio_resid -= bytes; 561 uio->uio_offset += bytes; 562 len -= bytes; 563 } 564 zfs_vmobject_wunlock(obj); 565 return (error); 566} 567 568/* 569 * When a file is memory mapped, we must keep the IO data synchronized 570 * between the DMU cache and the memory mapped pages. What this means: 571 * 572 * On Read: We "read" preferentially from memory mapped pages, 573 * else we default from the dmu buffer. 574 * 575 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 576 * the file is memory mapped. 577 */ 578static int 579mappedread(vnode_t *vp, int nbytes, uio_t *uio) 580{ 581 znode_t *zp = VTOZ(vp); 582 objset_t *os = zp->z_zfsvfs->z_os; 583 vm_object_t obj; 584 int64_t start; 585 caddr_t va; 586 int len = nbytes; 587 int off; 588 int error = 0; 589 590 ASSERT(vp->v_mount != NULL); 591 obj = vp->v_object; 592 ASSERT(obj != NULL); 593 594 start = uio->uio_loffset; 595 off = start & PAGEOFFSET; 596 zfs_vmobject_wlock(obj); 597 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 598 vm_page_t pp; 599 uint64_t bytes = MIN(PAGESIZE - off, len); 600 601 if (pp = page_hold(vp, start)) { 602 struct sf_buf *sf; 603 caddr_t va; 604 605 zfs_vmobject_wunlock(obj); 606 va = zfs_map_page(pp, &sf); 607 error = uiomove(va + off, bytes, UIO_READ, uio); 608 zfs_unmap_page(sf); 609 zfs_vmobject_wlock(obj); 610 page_unhold(pp); 611 } else { 612 zfs_vmobject_wunlock(obj); 613 error = dmu_read_uio(os, zp->z_id, uio, bytes); 614 zfs_vmobject_wlock(obj); 615 } 616 len -= bytes; 617 off = 0; 618 if (error) 619 break; 620 } 621 zfs_vmobject_wunlock(obj); 622 return (error); 623} 624 625offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 626 627/* 628 * Read bytes from specified file into supplied buffer. 629 * 630 * IN: vp - vnode of file to be read from. 631 * uio - structure supplying read location, range info, 632 * and return buffer. 633 * ioflag - SYNC flags; used to provide FRSYNC semantics. 634 * cr - credentials of caller. 635 * ct - caller context 636 * 637 * OUT: uio - updated offset and range, buffer filled. 638 * 639 * RETURN: 0 on success, error code on failure. 640 * 641 * Side Effects: 642 * vp - atime updated if byte count > 0 643 */ 644/* ARGSUSED */ 645static int 646zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 647{ 648 znode_t *zp = VTOZ(vp); 649 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 650 objset_t *os; 651 ssize_t n, nbytes; 652 int error = 0; 653 rl_t *rl; 654 xuio_t *xuio = NULL; 655 656 ZFS_ENTER(zfsvfs); 657 ZFS_VERIFY_ZP(zp); 658 os = zfsvfs->z_os; 659 660 if (zp->z_pflags & ZFS_AV_QUARANTINED) { 661 ZFS_EXIT(zfsvfs); 662 return (SET_ERROR(EACCES)); 663 } 664 665 /* 666 * Validate file offset 667 */ 668 if (uio->uio_loffset < (offset_t)0) { 669 ZFS_EXIT(zfsvfs); 670 return (SET_ERROR(EINVAL)); 671 } 672 673 /* 674 * Fasttrack empty reads 675 */ 676 if (uio->uio_resid == 0) { 677 ZFS_EXIT(zfsvfs); 678 return (0); 679 } 680 681 /* 682 * Check for mandatory locks 683 */ 684 if (MANDMODE(zp->z_mode)) { 685 if (error = chklock(vp, FREAD, 686 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 687 ZFS_EXIT(zfsvfs); 688 return (error); 689 } 690 } 691 692 /* 693 * If we're in FRSYNC mode, sync out this znode before reading it. 694 */ 695 if (zfsvfs->z_log && 696 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)) 697 zil_commit(zfsvfs->z_log, zp->z_id); 698 699 /* 700 * Lock the range against changes. 701 */ 702 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 703 704 /* 705 * If we are reading past end-of-file we can skip 706 * to the end; but we might still need to set atime. 707 */ 708 if (uio->uio_loffset >= zp->z_size) { 709 error = 0; 710 goto out; 711 } 712 713 ASSERT(uio->uio_loffset < zp->z_size); 714 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset); 715 716#ifdef sun 717 if ((uio->uio_extflg == UIO_XUIO) && 718 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) { 719 int nblk; 720 int blksz = zp->z_blksz; 721 uint64_t offset = uio->uio_loffset; 722 723 xuio = (xuio_t *)uio; 724 if ((ISP2(blksz))) { 725 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset, 726 blksz)) / blksz; 727 } else { 728 ASSERT(offset + n <= blksz); 729 nblk = 1; 730 } 731 (void) dmu_xuio_init(xuio, nblk); 732 733 if (vn_has_cached_data(vp)) { 734 /* 735 * For simplicity, we always allocate a full buffer 736 * even if we only expect to read a portion of a block. 737 */ 738 while (--nblk >= 0) { 739 (void) dmu_xuio_add(xuio, 740 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 741 blksz), 0, blksz); 742 } 743 } 744 } 745#endif /* sun */ 746 747 while (n > 0) { 748 nbytes = MIN(n, zfs_read_chunk_size - 749 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 750 751#ifdef __FreeBSD__ 752 if (uio->uio_segflg == UIO_NOCOPY) 753 error = mappedread_sf(vp, nbytes, uio); 754 else 755#endif /* __FreeBSD__ */ 756 if (vn_has_cached_data(vp)) 757 error = mappedread(vp, nbytes, uio); 758 else 759 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 760 if (error) { 761 /* convert checksum errors into IO errors */ 762 if (error == ECKSUM) 763 error = SET_ERROR(EIO); 764 break; 765 } 766 767 n -= nbytes; 768 } 769out: 770 zfs_range_unlock(rl); 771 772 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 773 ZFS_EXIT(zfsvfs); 774 return (error); 775} 776 777/* 778 * Write the bytes to a file. 779 * 780 * IN: vp - vnode of file to be written to. 781 * uio - structure supplying write location, range info, 782 * and data buffer. 783 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is 784 * set if in append mode. 785 * cr - credentials of caller. 786 * ct - caller context (NFS/CIFS fem monitor only) 787 * 788 * OUT: uio - updated offset and range. 789 * 790 * RETURN: 0 on success, error code on failure. 791 * 792 * Timestamps: 793 * vp - ctime|mtime updated if byte count > 0 794 */ 795 796/* ARGSUSED */ 797static int 798zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 799{ 800 znode_t *zp = VTOZ(vp); 801 rlim64_t limit = MAXOFFSET_T; 802 ssize_t start_resid = uio->uio_resid; 803 ssize_t tx_bytes; 804 uint64_t end_size; 805 dmu_tx_t *tx; 806 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 807 zilog_t *zilog; 808 offset_t woff; 809 ssize_t n, nbytes; 810 rl_t *rl; 811 int max_blksz = zfsvfs->z_max_blksz; 812 int error = 0; 813 arc_buf_t *abuf; 814 iovec_t *aiov = NULL; 815 xuio_t *xuio = NULL; 816 int i_iov = 0; 817 int iovcnt = uio->uio_iovcnt; 818 iovec_t *iovp = uio->uio_iov; 819 int write_eof; 820 int count = 0; 821 sa_bulk_attr_t bulk[4]; 822 uint64_t mtime[2], ctime[2]; 823 824 /* 825 * Fasttrack empty write 826 */ 827 n = start_resid; 828 if (n == 0) 829 return (0); 830 831 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 832 limit = MAXOFFSET_T; 833 834 ZFS_ENTER(zfsvfs); 835 ZFS_VERIFY_ZP(zp); 836 837 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 838 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 839 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 840 &zp->z_size, 8); 841 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 842 &zp->z_pflags, 8); 843 844 /* 845 * If immutable or not appending then return EPERM 846 */ 847 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 848 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 849 (uio->uio_loffset < zp->z_size))) { 850 ZFS_EXIT(zfsvfs); 851 return (SET_ERROR(EPERM)); 852 } 853 854 zilog = zfsvfs->z_log; 855 856 /* 857 * Validate file offset 858 */ 859 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset; 860 if (woff < 0) { 861 ZFS_EXIT(zfsvfs); 862 return (SET_ERROR(EINVAL)); 863 } 864 865 /* 866 * Check for mandatory locks before calling zfs_range_lock() 867 * in order to prevent a deadlock with locks set via fcntl(). 868 */ 869 if (MANDMODE((mode_t)zp->z_mode) && 870 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 871 ZFS_EXIT(zfsvfs); 872 return (error); 873 } 874 875#ifdef sun 876 /* 877 * Pre-fault the pages to ensure slow (eg NFS) pages 878 * don't hold up txg. 879 * Skip this if uio contains loaned arc_buf. 880 */ 881 if ((uio->uio_extflg == UIO_XUIO) && 882 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) 883 xuio = (xuio_t *)uio; 884 else 885 uio_prefaultpages(MIN(n, max_blksz), uio); 886#endif /* sun */ 887 888 /* 889 * If in append mode, set the io offset pointer to eof. 890 */ 891 if (ioflag & FAPPEND) { 892 /* 893 * Obtain an appending range lock to guarantee file append 894 * semantics. We reset the write offset once we have the lock. 895 */ 896 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 897 woff = rl->r_off; 898 if (rl->r_len == UINT64_MAX) { 899 /* 900 * We overlocked the file because this write will cause 901 * the file block size to increase. 902 * Note that zp_size cannot change with this lock held. 903 */ 904 woff = zp->z_size; 905 } 906 uio->uio_loffset = woff; 907 } else { 908 /* 909 * Note that if the file block size will change as a result of 910 * this write, then this range lock will lock the entire file 911 * so that we can re-write the block safely. 912 */ 913 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 914 } 915 916 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) { 917 zfs_range_unlock(rl); 918 ZFS_EXIT(zfsvfs); 919 return (EFBIG); 920 } 921 922 if (woff >= limit) { 923 zfs_range_unlock(rl); 924 ZFS_EXIT(zfsvfs); 925 return (SET_ERROR(EFBIG)); 926 } 927 928 if ((woff + n) > limit || woff > (limit - n)) 929 n = limit - woff; 930 931 /* Will this write extend the file length? */ 932 write_eof = (woff + n > zp->z_size); 933 934 end_size = MAX(zp->z_size, woff + n); 935 936 /* 937 * Write the file in reasonable size chunks. Each chunk is written 938 * in a separate transaction; this keeps the intent log records small 939 * and allows us to do more fine-grained space accounting. 940 */ 941 while (n > 0) { 942 abuf = NULL; 943 woff = uio->uio_loffset; 944 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 945 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 946 if (abuf != NULL) 947 dmu_return_arcbuf(abuf); 948 error = SET_ERROR(EDQUOT); 949 break; 950 } 951 952 if (xuio && abuf == NULL) { 953 ASSERT(i_iov < iovcnt); 954 aiov = &iovp[i_iov]; 955 abuf = dmu_xuio_arcbuf(xuio, i_iov); 956 dmu_xuio_clear(xuio, i_iov); 957 DTRACE_PROBE3(zfs_cp_write, int, i_iov, 958 iovec_t *, aiov, arc_buf_t *, abuf); 959 ASSERT((aiov->iov_base == abuf->b_data) || 960 ((char *)aiov->iov_base - (char *)abuf->b_data + 961 aiov->iov_len == arc_buf_size(abuf))); 962 i_iov++; 963 } else if (abuf == NULL && n >= max_blksz && 964 woff >= zp->z_size && 965 P2PHASE(woff, max_blksz) == 0 && 966 zp->z_blksz == max_blksz) { 967 /* 968 * This write covers a full block. "Borrow" a buffer 969 * from the dmu so that we can fill it before we enter 970 * a transaction. This avoids the possibility of 971 * holding up the transaction if the data copy hangs 972 * up on a pagefault (e.g., from an NFS server mapping). 973 */ 974 size_t cbytes; 975 976 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 977 max_blksz); 978 ASSERT(abuf != NULL); 979 ASSERT(arc_buf_size(abuf) == max_blksz); 980 if (error = uiocopy(abuf->b_data, max_blksz, 981 UIO_WRITE, uio, &cbytes)) { 982 dmu_return_arcbuf(abuf); 983 break; 984 } 985 ASSERT(cbytes == max_blksz); 986 } 987 988 /* 989 * Start a transaction. 990 */ 991 tx = dmu_tx_create(zfsvfs->z_os); 992 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 993 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 994 zfs_sa_upgrade_txholds(tx, zp); 995 error = dmu_tx_assign(tx, TXG_WAIT); 996 if (error) { 997 dmu_tx_abort(tx); 998 if (abuf != NULL) 999 dmu_return_arcbuf(abuf); 1000 break; 1001 } 1002 1003 /* 1004 * If zfs_range_lock() over-locked we grow the blocksize 1005 * and then reduce the lock range. This will only happen 1006 * on the first iteration since zfs_range_reduce() will 1007 * shrink down r_len to the appropriate size. 1008 */ 1009 if (rl->r_len == UINT64_MAX) { 1010 uint64_t new_blksz; 1011 1012 if (zp->z_blksz > max_blksz) { 1013 ASSERT(!ISP2(zp->z_blksz)); 1014 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 1015 } else { 1016 new_blksz = MIN(end_size, max_blksz); 1017 } 1018 zfs_grow_blocksize(zp, new_blksz, tx); 1019 zfs_range_reduce(rl, woff, n); 1020 } 1021 1022 /* 1023 * XXX - should we really limit each write to z_max_blksz? 1024 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 1025 */ 1026 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 1027 1028 if (woff + nbytes > zp->z_size) 1029 vnode_pager_setsize(vp, woff + nbytes); 1030 1031 if (abuf == NULL) { 1032 tx_bytes = uio->uio_resid; 1033 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), 1034 uio, nbytes, tx); 1035 tx_bytes -= uio->uio_resid; 1036 } else { 1037 tx_bytes = nbytes; 1038 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len); 1039 /* 1040 * If this is not a full block write, but we are 1041 * extending the file past EOF and this data starts 1042 * block-aligned, use assign_arcbuf(). Otherwise, 1043 * write via dmu_write(). 1044 */ 1045 if (tx_bytes < max_blksz && (!write_eof || 1046 aiov->iov_base != abuf->b_data)) { 1047 ASSERT(xuio); 1048 dmu_write(zfsvfs->z_os, zp->z_id, woff, 1049 aiov->iov_len, aiov->iov_base, tx); 1050 dmu_return_arcbuf(abuf); 1051 xuio_stat_wbuf_copied(); 1052 } else { 1053 ASSERT(xuio || tx_bytes == max_blksz); 1054 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), 1055 woff, abuf, tx); 1056 } 1057 ASSERT(tx_bytes <= uio->uio_resid); 1058 uioskip(uio, tx_bytes); 1059 } 1060 if (tx_bytes && vn_has_cached_data(vp)) { 1061 update_pages(vp, woff, tx_bytes, zfsvfs->z_os, 1062 zp->z_id, uio->uio_segflg, tx); 1063 } 1064 1065 /* 1066 * If we made no progress, we're done. If we made even 1067 * partial progress, update the znode and ZIL accordingly. 1068 */ 1069 if (tx_bytes == 0) { 1070 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 1071 (void *)&zp->z_size, sizeof (uint64_t), tx); 1072 dmu_tx_commit(tx); 1073 ASSERT(error != 0); 1074 break; 1075 } 1076 1077 /* 1078 * Clear Set-UID/Set-GID bits on successful write if not 1079 * privileged and at least one of the excute bits is set. 1080 * 1081 * It would be nice to to this after all writes have 1082 * been done, but that would still expose the ISUID/ISGID 1083 * to another app after the partial write is committed. 1084 * 1085 * Note: we don't call zfs_fuid_map_id() here because 1086 * user 0 is not an ephemeral uid. 1087 */ 1088 mutex_enter(&zp->z_acl_lock); 1089 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | 1090 (S_IXUSR >> 6))) != 0 && 1091 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && 1092 secpolicy_vnode_setid_retain(vp, cr, 1093 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) { 1094 uint64_t newmode; 1095 zp->z_mode &= ~(S_ISUID | S_ISGID); 1096 newmode = zp->z_mode; 1097 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), 1098 (void *)&newmode, sizeof (uint64_t), tx); 1099 } 1100 mutex_exit(&zp->z_acl_lock); 1101 1102 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 1103 B_TRUE); 1104 1105 /* 1106 * Update the file size (zp_size) if it has changed; 1107 * account for possible concurrent updates. 1108 */ 1109 while ((end_size = zp->z_size) < uio->uio_loffset) { 1110 (void) atomic_cas_64(&zp->z_size, end_size, 1111 uio->uio_loffset); 1112 ASSERT(error == 0); 1113 } 1114 /* 1115 * If we are replaying and eof is non zero then force 1116 * the file size to the specified eof. Note, there's no 1117 * concurrency during replay. 1118 */ 1119 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) 1120 zp->z_size = zfsvfs->z_replay_eof; 1121 1122 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1123 1124 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 1125 dmu_tx_commit(tx); 1126 1127 if (error != 0) 1128 break; 1129 ASSERT(tx_bytes == nbytes); 1130 n -= nbytes; 1131 1132#ifdef sun 1133 if (!xuio && n > 0) 1134 uio_prefaultpages(MIN(n, max_blksz), uio); 1135#endif /* sun */ 1136 } 1137 1138 zfs_range_unlock(rl); 1139 1140 /* 1141 * If we're in replay mode, or we made no progress, return error. 1142 * Otherwise, it's at least a partial write, so it's successful. 1143 */ 1144 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 1145 ZFS_EXIT(zfsvfs); 1146 return (error); 1147 } 1148 1149 if (ioflag & (FSYNC | FDSYNC) || 1150 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1151 zil_commit(zilog, zp->z_id); 1152 1153 ZFS_EXIT(zfsvfs); 1154 return (0); 1155} 1156 1157void 1158zfs_get_done(zgd_t *zgd, int error) 1159{ 1160 znode_t *zp = zgd->zgd_private; 1161 objset_t *os = zp->z_zfsvfs->z_os; 1162 1163 if (zgd->zgd_db) 1164 dmu_buf_rele(zgd->zgd_db, zgd); 1165 1166 zfs_range_unlock(zgd->zgd_rl); 1167 1168 /* 1169 * Release the vnode asynchronously as we currently have the 1170 * txg stopped from syncing. 1171 */ 1172 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1173 1174 if (error == 0 && zgd->zgd_bp) 1175 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1176 1177 kmem_free(zgd, sizeof (zgd_t)); 1178} 1179 1180#ifdef DEBUG 1181static int zil_fault_io = 0; 1182#endif 1183 1184/* 1185 * Get data to generate a TX_WRITE intent log record. 1186 */ 1187int 1188zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1189{ 1190 zfsvfs_t *zfsvfs = arg; 1191 objset_t *os = zfsvfs->z_os; 1192 znode_t *zp; 1193 uint64_t object = lr->lr_foid; 1194 uint64_t offset = lr->lr_offset; 1195 uint64_t size = lr->lr_length; 1196 blkptr_t *bp = &lr->lr_blkptr; 1197 dmu_buf_t *db; 1198 zgd_t *zgd; 1199 int error = 0; 1200 1201 ASSERT(zio != NULL); 1202 ASSERT(size != 0); 1203 1204 /* 1205 * Nothing to do if the file has been removed 1206 */ 1207 if (zfs_zget(zfsvfs, object, &zp) != 0) 1208 return (SET_ERROR(ENOENT)); 1209 if (zp->z_unlinked) { 1210 /* 1211 * Release the vnode asynchronously as we currently have the 1212 * txg stopped from syncing. 1213 */ 1214 VN_RELE_ASYNC(ZTOV(zp), 1215 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1216 return (SET_ERROR(ENOENT)); 1217 } 1218 1219 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1220 zgd->zgd_zilog = zfsvfs->z_log; 1221 zgd->zgd_private = zp; 1222 1223 /* 1224 * Write records come in two flavors: immediate and indirect. 1225 * For small writes it's cheaper to store the data with the 1226 * log record (immediate); for large writes it's cheaper to 1227 * sync the data and get a pointer to it (indirect) so that 1228 * we don't have to write the data twice. 1229 */ 1230 if (buf != NULL) { /* immediate write */ 1231 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER); 1232 /* test for truncation needs to be done while range locked */ 1233 if (offset >= zp->z_size) { 1234 error = SET_ERROR(ENOENT); 1235 } else { 1236 error = dmu_read(os, object, offset, size, buf, 1237 DMU_READ_NO_PREFETCH); 1238 } 1239 ASSERT(error == 0 || error == ENOENT); 1240 } else { /* indirect write */ 1241 /* 1242 * Have to lock the whole block to ensure when it's 1243 * written out and it's checksum is being calculated 1244 * that no one can change the data. We need to re-check 1245 * blocksize after we get the lock in case it's changed! 1246 */ 1247 for (;;) { 1248 uint64_t blkoff; 1249 size = zp->z_blksz; 1250 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; 1251 offset -= blkoff; 1252 zgd->zgd_rl = zfs_range_lock(zp, offset, size, 1253 RL_READER); 1254 if (zp->z_blksz == size) 1255 break; 1256 offset += blkoff; 1257 zfs_range_unlock(zgd->zgd_rl); 1258 } 1259 /* test for truncation needs to be done while range locked */ 1260 if (lr->lr_offset >= zp->z_size) 1261 error = SET_ERROR(ENOENT); 1262#ifdef DEBUG 1263 if (zil_fault_io) { 1264 error = SET_ERROR(EIO); 1265 zil_fault_io = 0; 1266 } 1267#endif 1268 if (error == 0) 1269 error = dmu_buf_hold(os, object, offset, zgd, &db, 1270 DMU_READ_NO_PREFETCH); 1271 1272 if (error == 0) { 1273 blkptr_t *obp = dmu_buf_get_blkptr(db); 1274 if (obp) { 1275 ASSERT(BP_IS_HOLE(bp)); 1276 *bp = *obp; 1277 } 1278 1279 zgd->zgd_db = db; 1280 zgd->zgd_bp = bp; 1281 1282 ASSERT(db->db_offset == offset); 1283 ASSERT(db->db_size == size); 1284 1285 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1286 zfs_get_done, zgd); 1287 ASSERT(error || lr->lr_length <= zp->z_blksz); 1288 1289 /* 1290 * On success, we need to wait for the write I/O 1291 * initiated by dmu_sync() to complete before we can 1292 * release this dbuf. We will finish everything up 1293 * in the zfs_get_done() callback. 1294 */ 1295 if (error == 0) 1296 return (0); 1297 1298 if (error == EALREADY) { 1299 lr->lr_common.lrc_txtype = TX_WRITE2; 1300 error = 0; 1301 } 1302 } 1303 } 1304 1305 zfs_get_done(zgd, error); 1306 1307 return (error); 1308} 1309 1310/*ARGSUSED*/ 1311static int 1312zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 1313 caller_context_t *ct) 1314{ 1315 znode_t *zp = VTOZ(vp); 1316 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1317 int error; 1318 1319 ZFS_ENTER(zfsvfs); 1320 ZFS_VERIFY_ZP(zp); 1321 1322 if (flag & V_ACE_MASK) 1323 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 1324 else 1325 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1326 1327 ZFS_EXIT(zfsvfs); 1328 return (error); 1329} 1330 1331/* 1332 * If vnode is for a device return a specfs vnode instead. 1333 */ 1334static int 1335specvp_check(vnode_t **vpp, cred_t *cr) 1336{ 1337 int error = 0; 1338 1339 if (IS_DEVVP(*vpp)) { 1340 struct vnode *svp; 1341 1342 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1343 VN_RELE(*vpp); 1344 if (svp == NULL) 1345 error = SET_ERROR(ENOSYS); 1346 *vpp = svp; 1347 } 1348 return (error); 1349} 1350 1351 1352/* 1353 * Lookup an entry in a directory, or an extended attribute directory. 1354 * If it exists, return a held vnode reference for it. 1355 * 1356 * IN: dvp - vnode of directory to search. 1357 * nm - name of entry to lookup. 1358 * pnp - full pathname to lookup [UNUSED]. 1359 * flags - LOOKUP_XATTR set if looking for an attribute. 1360 * rdir - root directory vnode [UNUSED]. 1361 * cr - credentials of caller. 1362 * ct - caller context 1363 * direntflags - directory lookup flags 1364 * realpnp - returned pathname. 1365 * 1366 * OUT: vpp - vnode of located entry, NULL if not found. 1367 * 1368 * RETURN: 0 on success, error code on failure. 1369 * 1370 * Timestamps: 1371 * NA 1372 */ 1373/* ARGSUSED */ 1374static int 1375zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 1376 int nameiop, cred_t *cr, kthread_t *td, int flags) 1377{ 1378 znode_t *zdp = VTOZ(dvp); 1379 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1380 int error = 0; 1381 int *direntflags = NULL; 1382 void *realpnp = NULL; 1383 1384 /* fast path */ 1385 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 1386 1387 if (dvp->v_type != VDIR) { 1388 return (SET_ERROR(ENOTDIR)); 1389 } else if (zdp->z_sa_hdl == NULL) { 1390 return (SET_ERROR(EIO)); 1391 } 1392 1393 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 1394 error = zfs_fastaccesschk_execute(zdp, cr); 1395 if (!error) { 1396 *vpp = dvp; 1397 VN_HOLD(*vpp); 1398 return (0); 1399 } 1400 return (error); 1401 } else { 1402 vnode_t *tvp = dnlc_lookup(dvp, nm); 1403 1404 if (tvp) { 1405 error = zfs_fastaccesschk_execute(zdp, cr); 1406 if (error) { 1407 VN_RELE(tvp); 1408 return (error); 1409 } 1410 if (tvp == DNLC_NO_VNODE) { 1411 VN_RELE(tvp); 1412 return (SET_ERROR(ENOENT)); 1413 } else { 1414 *vpp = tvp; 1415 return (specvp_check(vpp, cr)); 1416 } 1417 } 1418 } 1419 } 1420 1421 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1422 1423 ZFS_ENTER(zfsvfs); 1424 ZFS_VERIFY_ZP(zdp); 1425 1426 *vpp = NULL; 1427 1428 if (flags & LOOKUP_XATTR) { 1429#ifdef TODO 1430 /* 1431 * If the xattr property is off, refuse the lookup request. 1432 */ 1433 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1434 ZFS_EXIT(zfsvfs); 1435 return (SET_ERROR(EINVAL)); 1436 } 1437#endif 1438 1439 /* 1440 * We don't allow recursive attributes.. 1441 * Maybe someday we will. 1442 */ 1443 if (zdp->z_pflags & ZFS_XATTR) { 1444 ZFS_EXIT(zfsvfs); 1445 return (SET_ERROR(EINVAL)); 1446 } 1447 1448 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1449 ZFS_EXIT(zfsvfs); 1450 return (error); 1451 } 1452 1453 /* 1454 * Do we have permission to get into attribute directory? 1455 */ 1456 1457 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1458 B_FALSE, cr)) { 1459 VN_RELE(*vpp); 1460 *vpp = NULL; 1461 } 1462 1463 ZFS_EXIT(zfsvfs); 1464 return (error); 1465 } 1466 1467 if (dvp->v_type != VDIR) { 1468 ZFS_EXIT(zfsvfs); 1469 return (SET_ERROR(ENOTDIR)); 1470 } 1471 1472 /* 1473 * Check accessibility of directory. 1474 */ 1475 1476 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1477 ZFS_EXIT(zfsvfs); 1478 return (error); 1479 } 1480 1481 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1482 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1483 ZFS_EXIT(zfsvfs); 1484 return (SET_ERROR(EILSEQ)); 1485 } 1486 1487 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1488 if (error == 0) 1489 error = specvp_check(vpp, cr); 1490 1491 /* Translate errors and add SAVENAME when needed. */ 1492 if (cnp->cn_flags & ISLASTCN) { 1493 switch (nameiop) { 1494 case CREATE: 1495 case RENAME: 1496 if (error == ENOENT) { 1497 error = EJUSTRETURN; 1498 cnp->cn_flags |= SAVENAME; 1499 break; 1500 } 1501 /* FALLTHROUGH */ 1502 case DELETE: 1503 if (error == 0) 1504 cnp->cn_flags |= SAVENAME; 1505 break; 1506 } 1507 } 1508 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) { 1509 int ltype = 0; 1510 1511 if (cnp->cn_flags & ISDOTDOT) { 1512 ltype = VOP_ISLOCKED(dvp); 1513 VOP_UNLOCK(dvp, 0); 1514 } 1515 ZFS_EXIT(zfsvfs); 1516 error = vn_lock(*vpp, cnp->cn_lkflags); 1517 if (cnp->cn_flags & ISDOTDOT) 1518 vn_lock(dvp, ltype | LK_RETRY); 1519 if (error != 0) { 1520 VN_RELE(*vpp); 1521 *vpp = NULL; 1522 return (error); 1523 } 1524 } else { 1525 ZFS_EXIT(zfsvfs); 1526 } 1527 1528#ifdef FREEBSD_NAMECACHE 1529 /* 1530 * Insert name into cache (as non-existent) if appropriate. 1531 */ 1532 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE) 1533 cache_enter(dvp, *vpp, cnp); 1534 /* 1535 * Insert name into cache if appropriate. 1536 */ 1537 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1538 if (!(cnp->cn_flags & ISLASTCN) || 1539 (nameiop != DELETE && nameiop != RENAME)) { 1540 cache_enter(dvp, *vpp, cnp); 1541 } 1542 } 1543#endif 1544 1545 return (error); 1546} 1547 1548/* 1549 * Attempt to create a new entry in a directory. If the entry 1550 * already exists, truncate the file if permissible, else return 1551 * an error. Return the vp of the created or trunc'd file. 1552 * 1553 * IN: dvp - vnode of directory to put new file entry in. 1554 * name - name of new file entry. 1555 * vap - attributes of new file. 1556 * excl - flag indicating exclusive or non-exclusive mode. 1557 * mode - mode to open file with. 1558 * cr - credentials of caller. 1559 * flag - large file flag [UNUSED]. 1560 * ct - caller context 1561 * vsecp - ACL to be set 1562 * 1563 * OUT: vpp - vnode of created or trunc'd entry. 1564 * 1565 * RETURN: 0 on success, error code on failure. 1566 * 1567 * Timestamps: 1568 * dvp - ctime|mtime updated if new entry created 1569 * vp - ctime|mtime always, atime if new 1570 */ 1571 1572/* ARGSUSED */ 1573static int 1574zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1575 vnode_t **vpp, cred_t *cr, kthread_t *td) 1576{ 1577 znode_t *zp, *dzp = VTOZ(dvp); 1578 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1579 zilog_t *zilog; 1580 objset_t *os; 1581 zfs_dirlock_t *dl; 1582 dmu_tx_t *tx; 1583 int error; 1584 ksid_t *ksid; 1585 uid_t uid; 1586 gid_t gid = crgetgid(cr); 1587 zfs_acl_ids_t acl_ids; 1588 boolean_t fuid_dirtied; 1589 boolean_t have_acl = B_FALSE; 1590 boolean_t waited = B_FALSE; 1591 void *vsecp = NULL; 1592 int flag = 0; 1593 1594 /* 1595 * If we have an ephemeral id, ACL, or XVATTR then 1596 * make sure file system is at proper version 1597 */ 1598 1599 ksid = crgetsid(cr, KSID_OWNER); 1600 if (ksid) 1601 uid = ksid_getid(ksid); 1602 else 1603 uid = crgetuid(cr); 1604 1605 if (zfsvfs->z_use_fuids == B_FALSE && 1606 (vsecp || (vap->va_mask & AT_XVATTR) || 1607 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1608 return (SET_ERROR(EINVAL)); 1609 1610 ZFS_ENTER(zfsvfs); 1611 ZFS_VERIFY_ZP(dzp); 1612 os = zfsvfs->z_os; 1613 zilog = zfsvfs->z_log; 1614 1615 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1616 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1617 ZFS_EXIT(zfsvfs); 1618 return (SET_ERROR(EILSEQ)); 1619 } 1620 1621 if (vap->va_mask & AT_XVATTR) { 1622 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 1623 crgetuid(cr), cr, vap->va_type)) != 0) { 1624 ZFS_EXIT(zfsvfs); 1625 return (error); 1626 } 1627 } 1628 1629 getnewvnode_reserve(1); 1630 1631top: 1632 *vpp = NULL; 1633 1634 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) 1635 vap->va_mode &= ~S_ISVTX; 1636 1637 if (*name == '\0') { 1638 /* 1639 * Null component name refers to the directory itself. 1640 */ 1641 VN_HOLD(dvp); 1642 zp = dzp; 1643 dl = NULL; 1644 error = 0; 1645 } else { 1646 /* possible VN_HOLD(zp) */ 1647 int zflg = 0; 1648 1649 if (flag & FIGNORECASE) 1650 zflg |= ZCILOOK; 1651 1652 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1653 NULL, NULL); 1654 if (error) { 1655 if (have_acl) 1656 zfs_acl_ids_free(&acl_ids); 1657 if (strcmp(name, "..") == 0) 1658 error = SET_ERROR(EISDIR); 1659 getnewvnode_drop_reserve(); 1660 ZFS_EXIT(zfsvfs); 1661 return (error); 1662 } 1663 } 1664 1665 if (zp == NULL) { 1666 uint64_t txtype; 1667 1668 /* 1669 * Create a new file object and update the directory 1670 * to reference it. 1671 */ 1672 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1673 if (have_acl) 1674 zfs_acl_ids_free(&acl_ids); 1675 goto out; 1676 } 1677 1678 /* 1679 * We only support the creation of regular files in 1680 * extended attribute directories. 1681 */ 1682 1683 if ((dzp->z_pflags & ZFS_XATTR) && 1684 (vap->va_type != VREG)) { 1685 if (have_acl) 1686 zfs_acl_ids_free(&acl_ids); 1687 error = SET_ERROR(EINVAL); 1688 goto out; 1689 } 1690 1691 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 1692 cr, vsecp, &acl_ids)) != 0) 1693 goto out; 1694 have_acl = B_TRUE; 1695 1696 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1697 zfs_acl_ids_free(&acl_ids); 1698 error = SET_ERROR(EDQUOT); 1699 goto out; 1700 } 1701 1702 tx = dmu_tx_create(os); 1703 1704 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1705 ZFS_SA_BASE_ATTR_SIZE); 1706 1707 fuid_dirtied = zfsvfs->z_fuid_dirty; 1708 if (fuid_dirtied) 1709 zfs_fuid_txhold(zfsvfs, tx); 1710 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1711 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1712 if (!zfsvfs->z_use_sa && 1713 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1714 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1715 0, acl_ids.z_aclp->z_acl_bytes); 1716 } 1717 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 1718 if (error) { 1719 zfs_dirent_unlock(dl); 1720 if (error == ERESTART) { 1721 waited = B_TRUE; 1722 dmu_tx_wait(tx); 1723 dmu_tx_abort(tx); 1724 goto top; 1725 } 1726 zfs_acl_ids_free(&acl_ids); 1727 dmu_tx_abort(tx); 1728 getnewvnode_drop_reserve(); 1729 ZFS_EXIT(zfsvfs); 1730 return (error); 1731 } 1732 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1733 1734 if (fuid_dirtied) 1735 zfs_fuid_sync(zfsvfs, tx); 1736 1737 (void) zfs_link_create(dl, zp, tx, ZNEW); 1738 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1739 if (flag & FIGNORECASE) 1740 txtype |= TX_CI; 1741 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1742 vsecp, acl_ids.z_fuidp, vap); 1743 zfs_acl_ids_free(&acl_ids); 1744 dmu_tx_commit(tx); 1745 } else { 1746 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1747 1748 if (have_acl) 1749 zfs_acl_ids_free(&acl_ids); 1750 have_acl = B_FALSE; 1751 1752 /* 1753 * A directory entry already exists for this name. 1754 */ 1755 /* 1756 * Can't truncate an existing file if in exclusive mode. 1757 */ 1758 if (excl == EXCL) { 1759 error = SET_ERROR(EEXIST); 1760 goto out; 1761 } 1762 /* 1763 * Can't open a directory for writing. 1764 */ 1765 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1766 error = SET_ERROR(EISDIR); 1767 goto out; 1768 } 1769 /* 1770 * Verify requested access to file. 1771 */ 1772 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1773 goto out; 1774 } 1775 1776 mutex_enter(&dzp->z_lock); 1777 dzp->z_seq++; 1778 mutex_exit(&dzp->z_lock); 1779 1780 /* 1781 * Truncate regular files if requested. 1782 */ 1783 if ((ZTOV(zp)->v_type == VREG) && 1784 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1785 /* we can't hold any locks when calling zfs_freesp() */ 1786 zfs_dirent_unlock(dl); 1787 dl = NULL; 1788 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1789 if (error == 0) { 1790 vnevent_create(ZTOV(zp), ct); 1791 } 1792 } 1793 } 1794out: 1795 getnewvnode_drop_reserve(); 1796 if (dl) 1797 zfs_dirent_unlock(dl); 1798 1799 if (error) { 1800 if (zp) 1801 VN_RELE(ZTOV(zp)); 1802 } else { 1803 *vpp = ZTOV(zp); 1804 error = specvp_check(vpp, cr); 1805 } 1806 1807 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1808 zil_commit(zilog, 0); 1809 1810 ZFS_EXIT(zfsvfs); 1811 return (error); 1812} 1813 1814/* 1815 * Remove an entry from a directory. 1816 * 1817 * IN: dvp - vnode of directory to remove entry from. 1818 * name - name of entry to remove. 1819 * cr - credentials of caller. 1820 * ct - caller context 1821 * flags - case flags 1822 * 1823 * RETURN: 0 on success, error code on failure. 1824 * 1825 * Timestamps: 1826 * dvp - ctime|mtime 1827 * vp - ctime (if nlink > 0) 1828 */ 1829 1830uint64_t null_xattr = 0; 1831 1832/*ARGSUSED*/ 1833static int 1834zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1835 int flags) 1836{ 1837 znode_t *zp, *dzp = VTOZ(dvp); 1838 znode_t *xzp; 1839 vnode_t *vp; 1840 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1841 zilog_t *zilog; 1842 uint64_t acl_obj, xattr_obj; 1843 uint64_t xattr_obj_unlinked = 0; 1844 uint64_t obj = 0; 1845 zfs_dirlock_t *dl; 1846 dmu_tx_t *tx; 1847 boolean_t may_delete_now, delete_now = FALSE; 1848 boolean_t unlinked, toobig = FALSE; 1849 uint64_t txtype; 1850 pathname_t *realnmp = NULL; 1851 pathname_t realnm; 1852 int error; 1853 int zflg = ZEXISTS; 1854 boolean_t waited = B_FALSE; 1855 1856 ZFS_ENTER(zfsvfs); 1857 ZFS_VERIFY_ZP(dzp); 1858 zilog = zfsvfs->z_log; 1859 1860 if (flags & FIGNORECASE) { 1861 zflg |= ZCILOOK; 1862 pn_alloc(&realnm); 1863 realnmp = &realnm; 1864 } 1865 1866top: 1867 xattr_obj = 0; 1868 xzp = NULL; 1869 /* 1870 * Attempt to lock directory; fail if entry doesn't exist. 1871 */ 1872 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1873 NULL, realnmp)) { 1874 if (realnmp) 1875 pn_free(realnmp); 1876 ZFS_EXIT(zfsvfs); 1877 return (error); 1878 } 1879 1880 vp = ZTOV(zp); 1881 1882 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1883 goto out; 1884 } 1885 1886 /* 1887 * Need to use rmdir for removing directories. 1888 */ 1889 if (vp->v_type == VDIR) { 1890 error = SET_ERROR(EPERM); 1891 goto out; 1892 } 1893 1894 vnevent_remove(vp, dvp, name, ct); 1895 1896 if (realnmp) 1897 dnlc_remove(dvp, realnmp->pn_buf); 1898 else 1899 dnlc_remove(dvp, name); 1900 1901 VI_LOCK(vp); 1902 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1903 VI_UNLOCK(vp); 1904 1905 /* 1906 * We may delete the znode now, or we may put it in the unlinked set; 1907 * it depends on whether we're the last link, and on whether there are 1908 * other holds on the vnode. So we dmu_tx_hold() the right things to 1909 * allow for either case. 1910 */ 1911 obj = zp->z_id; 1912 tx = dmu_tx_create(zfsvfs->z_os); 1913 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1914 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1915 zfs_sa_upgrade_txholds(tx, zp); 1916 zfs_sa_upgrade_txholds(tx, dzp); 1917 if (may_delete_now) { 1918 toobig = 1919 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 1920 /* if the file is too big, only hold_free a token amount */ 1921 dmu_tx_hold_free(tx, zp->z_id, 0, 1922 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1923 } 1924 1925 /* are there any extended attributes? */ 1926 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1927 &xattr_obj, sizeof (xattr_obj)); 1928 if (error == 0 && xattr_obj) { 1929 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1930 ASSERT0(error); 1931 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1932 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1933 } 1934 1935 mutex_enter(&zp->z_lock); 1936 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now) 1937 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1938 mutex_exit(&zp->z_lock); 1939 1940 /* charge as an update -- would be nice not to charge at all */ 1941 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1942 1943 /* 1944 * Mark this transaction as typically resulting in a net free of 1945 * space, unless object removal will be delayed indefinitely 1946 * (due to active holds on the vnode due to the file being open). 1947 */ 1948 if (may_delete_now) 1949 dmu_tx_mark_netfree(tx); 1950 1951 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 1952 if (error) { 1953 zfs_dirent_unlock(dl); 1954 VN_RELE(vp); 1955 if (xzp) 1956 VN_RELE(ZTOV(xzp)); 1957 if (error == ERESTART) { 1958 waited = B_TRUE; 1959 dmu_tx_wait(tx); 1960 dmu_tx_abort(tx); 1961 goto top; 1962 } 1963 if (realnmp) 1964 pn_free(realnmp); 1965 dmu_tx_abort(tx); 1966 ZFS_EXIT(zfsvfs); 1967 return (error); 1968 } 1969 1970 /* 1971 * Remove the directory entry. 1972 */ 1973 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1974 1975 if (error) { 1976 dmu_tx_commit(tx); 1977 goto out; 1978 } 1979 1980 if (unlinked) { 1981 /* 1982 * Hold z_lock so that we can make sure that the ACL obj 1983 * hasn't changed. Could have been deleted due to 1984 * zfs_sa_upgrade(). 1985 */ 1986 mutex_enter(&zp->z_lock); 1987 VI_LOCK(vp); 1988 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1989 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked)); 1990 delete_now = may_delete_now && !toobig && 1991 vp->v_count == 1 && !vn_has_cached_data(vp) && 1992 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) == 1993 acl_obj; 1994 VI_UNLOCK(vp); 1995 } 1996 1997 if (delete_now) { 1998#ifdef __FreeBSD__ 1999 panic("zfs_remove: delete_now branch taken"); 2000#endif 2001 if (xattr_obj_unlinked) { 2002 ASSERT3U(xzp->z_links, ==, 2); 2003 mutex_enter(&xzp->z_lock); 2004 xzp->z_unlinked = 1; 2005 xzp->z_links = 0; 2006 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 2007 &xzp->z_links, sizeof (xzp->z_links), tx); 2008 ASSERT3U(error, ==, 0); 2009 mutex_exit(&xzp->z_lock); 2010 zfs_unlinked_add(xzp, tx); 2011 2012 if (zp->z_is_sa) 2013 error = sa_remove(zp->z_sa_hdl, 2014 SA_ZPL_XATTR(zfsvfs), tx); 2015 else 2016 error = sa_update(zp->z_sa_hdl, 2017 SA_ZPL_XATTR(zfsvfs), &null_xattr, 2018 sizeof (uint64_t), tx); 2019 ASSERT0(error); 2020 } 2021 VI_LOCK(vp); 2022 vp->v_count--; 2023 ASSERT0(vp->v_count); 2024 VI_UNLOCK(vp); 2025 mutex_exit(&zp->z_lock); 2026 zfs_znode_delete(zp, tx); 2027 } else if (unlinked) { 2028 mutex_exit(&zp->z_lock); 2029 zfs_unlinked_add(zp, tx); 2030#ifdef __FreeBSD__ 2031 vp->v_vflag |= VV_NOSYNC; 2032#endif 2033 } 2034 2035 txtype = TX_REMOVE; 2036 if (flags & FIGNORECASE) 2037 txtype |= TX_CI; 2038 zfs_log_remove(zilog, tx, txtype, dzp, name, obj); 2039 2040 dmu_tx_commit(tx); 2041out: 2042 if (realnmp) 2043 pn_free(realnmp); 2044 2045 zfs_dirent_unlock(dl); 2046 2047 if (!delete_now) 2048 VN_RELE(vp); 2049 if (xzp) 2050 VN_RELE(ZTOV(xzp)); 2051 2052 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2053 zil_commit(zilog, 0); 2054 2055 ZFS_EXIT(zfsvfs); 2056 return (error); 2057} 2058 2059/* 2060 * Create a new directory and insert it into dvp using the name 2061 * provided. Return a pointer to the inserted directory. 2062 * 2063 * IN: dvp - vnode of directory to add subdir to. 2064 * dirname - name of new directory. 2065 * vap - attributes of new directory. 2066 * cr - credentials of caller. 2067 * ct - caller context 2068 * flags - case flags 2069 * vsecp - ACL to be set 2070 * 2071 * OUT: vpp - vnode of created directory. 2072 * 2073 * RETURN: 0 on success, error code on failure. 2074 * 2075 * Timestamps: 2076 * dvp - ctime|mtime updated 2077 * vp - ctime|mtime|atime updated 2078 */ 2079/*ARGSUSED*/ 2080static int 2081zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 2082 caller_context_t *ct, int flags, vsecattr_t *vsecp) 2083{ 2084 znode_t *zp, *dzp = VTOZ(dvp); 2085 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2086 zilog_t *zilog; 2087 zfs_dirlock_t *dl; 2088 uint64_t txtype; 2089 dmu_tx_t *tx; 2090 int error; 2091 int zf = ZNEW; 2092 ksid_t *ksid; 2093 uid_t uid; 2094 gid_t gid = crgetgid(cr); 2095 zfs_acl_ids_t acl_ids; 2096 boolean_t fuid_dirtied; 2097 boolean_t waited = B_FALSE; 2098 2099 ASSERT(vap->va_type == VDIR); 2100 2101 /* 2102 * If we have an ephemeral id, ACL, or XVATTR then 2103 * make sure file system is at proper version 2104 */ 2105 2106 ksid = crgetsid(cr, KSID_OWNER); 2107 if (ksid) 2108 uid = ksid_getid(ksid); 2109 else 2110 uid = crgetuid(cr); 2111 if (zfsvfs->z_use_fuids == B_FALSE && 2112 (vsecp || (vap->va_mask & AT_XVATTR) || 2113 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 2114 return (SET_ERROR(EINVAL)); 2115 2116 ZFS_ENTER(zfsvfs); 2117 ZFS_VERIFY_ZP(dzp); 2118 zilog = zfsvfs->z_log; 2119 2120 if (dzp->z_pflags & ZFS_XATTR) { 2121 ZFS_EXIT(zfsvfs); 2122 return (SET_ERROR(EINVAL)); 2123 } 2124 2125 if (zfsvfs->z_utf8 && u8_validate(dirname, 2126 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2127 ZFS_EXIT(zfsvfs); 2128 return (SET_ERROR(EILSEQ)); 2129 } 2130 if (flags & FIGNORECASE) 2131 zf |= ZCILOOK; 2132 2133 if (vap->va_mask & AT_XVATTR) { 2134 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 2135 crgetuid(cr), cr, vap->va_type)) != 0) { 2136 ZFS_EXIT(zfsvfs); 2137 return (error); 2138 } 2139 } 2140 2141 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 2142 vsecp, &acl_ids)) != 0) { 2143 ZFS_EXIT(zfsvfs); 2144 return (error); 2145 } 2146 2147 getnewvnode_reserve(1); 2148 2149 /* 2150 * First make sure the new directory doesn't exist. 2151 * 2152 * Existence is checked first to make sure we don't return 2153 * EACCES instead of EEXIST which can cause some applications 2154 * to fail. 2155 */ 2156top: 2157 *vpp = NULL; 2158 2159 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 2160 NULL, NULL)) { 2161 zfs_acl_ids_free(&acl_ids); 2162 getnewvnode_drop_reserve(); 2163 ZFS_EXIT(zfsvfs); 2164 return (error); 2165 } 2166 2167 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 2168 zfs_acl_ids_free(&acl_ids); 2169 zfs_dirent_unlock(dl); 2170 getnewvnode_drop_reserve(); 2171 ZFS_EXIT(zfsvfs); 2172 return (error); 2173 } 2174 2175 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 2176 zfs_acl_ids_free(&acl_ids); 2177 zfs_dirent_unlock(dl); 2178 getnewvnode_drop_reserve(); 2179 ZFS_EXIT(zfsvfs); 2180 return (SET_ERROR(EDQUOT)); 2181 } 2182 2183 /* 2184 * Add a new entry to the directory. 2185 */ 2186 tx = dmu_tx_create(zfsvfs->z_os); 2187 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 2188 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2189 fuid_dirtied = zfsvfs->z_fuid_dirty; 2190 if (fuid_dirtied) 2191 zfs_fuid_txhold(zfsvfs, tx); 2192 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2193 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2194 acl_ids.z_aclp->z_acl_bytes); 2195 } 2196 2197 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 2198 ZFS_SA_BASE_ATTR_SIZE); 2199 2200 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2201 if (error) { 2202 zfs_dirent_unlock(dl); 2203 if (error == ERESTART) { 2204 waited = B_TRUE; 2205 dmu_tx_wait(tx); 2206 dmu_tx_abort(tx); 2207 goto top; 2208 } 2209 zfs_acl_ids_free(&acl_ids); 2210 dmu_tx_abort(tx); 2211 getnewvnode_drop_reserve(); 2212 ZFS_EXIT(zfsvfs); 2213 return (error); 2214 } 2215 2216 /* 2217 * Create new node. 2218 */ 2219 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 2220 2221 if (fuid_dirtied) 2222 zfs_fuid_sync(zfsvfs, tx); 2223 2224 /* 2225 * Now put new name in parent dir. 2226 */ 2227 (void) zfs_link_create(dl, zp, tx, ZNEW); 2228 2229 *vpp = ZTOV(zp); 2230 2231 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 2232 if (flags & FIGNORECASE) 2233 txtype |= TX_CI; 2234 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 2235 acl_ids.z_fuidp, vap); 2236 2237 zfs_acl_ids_free(&acl_ids); 2238 2239 dmu_tx_commit(tx); 2240 2241 getnewvnode_drop_reserve(); 2242 2243 zfs_dirent_unlock(dl); 2244 2245 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2246 zil_commit(zilog, 0); 2247 2248 ZFS_EXIT(zfsvfs); 2249 return (0); 2250} 2251 2252/* 2253 * Remove a directory subdir entry. If the current working 2254 * directory is the same as the subdir to be removed, the 2255 * remove will fail. 2256 * 2257 * IN: dvp - vnode of directory to remove from. 2258 * name - name of directory to be removed. 2259 * cwd - vnode of current working directory. 2260 * cr - credentials of caller. 2261 * ct - caller context 2262 * flags - case flags 2263 * 2264 * RETURN: 0 on success, error code on failure. 2265 * 2266 * Timestamps: 2267 * dvp - ctime|mtime updated 2268 */ 2269/*ARGSUSED*/ 2270static int 2271zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 2272 caller_context_t *ct, int flags) 2273{ 2274 znode_t *dzp = VTOZ(dvp); 2275 znode_t *zp; 2276 vnode_t *vp; 2277 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2278 zilog_t *zilog; 2279 zfs_dirlock_t *dl; 2280 dmu_tx_t *tx; 2281 int error; 2282 int zflg = ZEXISTS; 2283 boolean_t waited = B_FALSE; 2284 2285 ZFS_ENTER(zfsvfs); 2286 ZFS_VERIFY_ZP(dzp); 2287 zilog = zfsvfs->z_log; 2288 2289 if (flags & FIGNORECASE) 2290 zflg |= ZCILOOK; 2291top: 2292 zp = NULL; 2293 2294 /* 2295 * Attempt to lock directory; fail if entry doesn't exist. 2296 */ 2297 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 2298 NULL, NULL)) { 2299 ZFS_EXIT(zfsvfs); 2300 return (error); 2301 } 2302 2303 vp = ZTOV(zp); 2304 2305 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 2306 goto out; 2307 } 2308 2309 if (vp->v_type != VDIR) { 2310 error = SET_ERROR(ENOTDIR); 2311 goto out; 2312 } 2313 2314 if (vp == cwd) { 2315 error = SET_ERROR(EINVAL); 2316 goto out; 2317 } 2318 2319 vnevent_rmdir(vp, dvp, name, ct); 2320 2321 /* 2322 * Grab a lock on the directory to make sure that noone is 2323 * trying to add (or lookup) entries while we are removing it. 2324 */ 2325 rw_enter(&zp->z_name_lock, RW_WRITER); 2326 2327 /* 2328 * Grab a lock on the parent pointer to make sure we play well 2329 * with the treewalk and directory rename code. 2330 */ 2331 rw_enter(&zp->z_parent_lock, RW_WRITER); 2332 2333 tx = dmu_tx_create(zfsvfs->z_os); 2334 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2335 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2336 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2337 zfs_sa_upgrade_txholds(tx, zp); 2338 zfs_sa_upgrade_txholds(tx, dzp); 2339 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2340 if (error) { 2341 rw_exit(&zp->z_parent_lock); 2342 rw_exit(&zp->z_name_lock); 2343 zfs_dirent_unlock(dl); 2344 VN_RELE(vp); 2345 if (error == ERESTART) { 2346 waited = B_TRUE; 2347 dmu_tx_wait(tx); 2348 dmu_tx_abort(tx); 2349 goto top; 2350 } 2351 dmu_tx_abort(tx); 2352 ZFS_EXIT(zfsvfs); 2353 return (error); 2354 } 2355 2356#ifdef FREEBSD_NAMECACHE 2357 cache_purge(dvp); 2358#endif 2359 2360 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 2361 2362 if (error == 0) { 2363 uint64_t txtype = TX_RMDIR; 2364 if (flags & FIGNORECASE) 2365 txtype |= TX_CI; 2366 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT); 2367 } 2368 2369 dmu_tx_commit(tx); 2370 2371 rw_exit(&zp->z_parent_lock); 2372 rw_exit(&zp->z_name_lock); 2373#ifdef FREEBSD_NAMECACHE 2374 cache_purge(vp); 2375#endif 2376out: 2377 zfs_dirent_unlock(dl); 2378 2379 VN_RELE(vp); 2380 2381 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2382 zil_commit(zilog, 0); 2383 2384 ZFS_EXIT(zfsvfs); 2385 return (error); 2386} 2387 2388/* 2389 * Read as many directory entries as will fit into the provided 2390 * buffer from the given directory cursor position (specified in 2391 * the uio structure). 2392 * 2393 * IN: vp - vnode of directory to read. 2394 * uio - structure supplying read location, range info, 2395 * and return buffer. 2396 * cr - credentials of caller. 2397 * ct - caller context 2398 * flags - case flags 2399 * 2400 * OUT: uio - updated offset and range, buffer filled. 2401 * eofp - set to true if end-of-file detected. 2402 * 2403 * RETURN: 0 on success, error code on failure. 2404 * 2405 * Timestamps: 2406 * vp - atime updated 2407 * 2408 * Note that the low 4 bits of the cookie returned by zap is always zero. 2409 * This allows us to use the low range for "special" directory entries: 2410 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 2411 * we use the offset 2 for the '.zfs' directory. 2412 */ 2413/* ARGSUSED */ 2414static int 2415zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 2416{ 2417 znode_t *zp = VTOZ(vp); 2418 iovec_t *iovp; 2419 edirent_t *eodp; 2420 dirent64_t *odp; 2421 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2422 objset_t *os; 2423 caddr_t outbuf; 2424 size_t bufsize; 2425 zap_cursor_t zc; 2426 zap_attribute_t zap; 2427 uint_t bytes_wanted; 2428 uint64_t offset; /* must be unsigned; checks for < 1 */ 2429 uint64_t parent; 2430 int local_eof; 2431 int outcount; 2432 int error; 2433 uint8_t prefetch; 2434 boolean_t check_sysattrs; 2435 uint8_t type; 2436 int ncooks; 2437 u_long *cooks = NULL; 2438 int flags = 0; 2439 2440 ZFS_ENTER(zfsvfs); 2441 ZFS_VERIFY_ZP(zp); 2442 2443 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 2444 &parent, sizeof (parent))) != 0) { 2445 ZFS_EXIT(zfsvfs); 2446 return (error); 2447 } 2448 2449 /* 2450 * If we are not given an eof variable, 2451 * use a local one. 2452 */ 2453 if (eofp == NULL) 2454 eofp = &local_eof; 2455 2456 /* 2457 * Check for valid iov_len. 2458 */ 2459 if (uio->uio_iov->iov_len <= 0) { 2460 ZFS_EXIT(zfsvfs); 2461 return (SET_ERROR(EINVAL)); 2462 } 2463 2464 /* 2465 * Quit if directory has been removed (posix) 2466 */ 2467 if ((*eofp = zp->z_unlinked) != 0) { 2468 ZFS_EXIT(zfsvfs); 2469 return (0); 2470 } 2471 2472 error = 0; 2473 os = zfsvfs->z_os; 2474 offset = uio->uio_loffset; 2475 prefetch = zp->z_zn_prefetch; 2476 2477 /* 2478 * Initialize the iterator cursor. 2479 */ 2480 if (offset <= 3) { 2481 /* 2482 * Start iteration from the beginning of the directory. 2483 */ 2484 zap_cursor_init(&zc, os, zp->z_id); 2485 } else { 2486 /* 2487 * The offset is a serialized cursor. 2488 */ 2489 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 2490 } 2491 2492 /* 2493 * Get space to change directory entries into fs independent format. 2494 */ 2495 iovp = uio->uio_iov; 2496 bytes_wanted = iovp->iov_len; 2497 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 2498 bufsize = bytes_wanted; 2499 outbuf = kmem_alloc(bufsize, KM_SLEEP); 2500 odp = (struct dirent64 *)outbuf; 2501 } else { 2502 bufsize = bytes_wanted; 2503 outbuf = NULL; 2504 odp = (struct dirent64 *)iovp->iov_base; 2505 } 2506 eodp = (struct edirent *)odp; 2507 2508 if (ncookies != NULL) { 2509 /* 2510 * Minimum entry size is dirent size and 1 byte for a file name. 2511 */ 2512 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 2513 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 2514 *cookies = cooks; 2515 *ncookies = ncooks; 2516 } 2517 /* 2518 * If this VFS supports the system attribute view interface; and 2519 * we're looking at an extended attribute directory; and we care 2520 * about normalization conflicts on this vfs; then we must check 2521 * for normalization conflicts with the sysattr name space. 2522 */ 2523#ifdef TODO 2524 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 2525 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 2526 (flags & V_RDDIR_ENTFLAGS); 2527#else 2528 check_sysattrs = 0; 2529#endif 2530 2531 /* 2532 * Transform to file-system independent format 2533 */ 2534 outcount = 0; 2535 while (outcount < bytes_wanted) { 2536 ino64_t objnum; 2537 ushort_t reclen; 2538 off64_t *next = NULL; 2539 2540 /* 2541 * Special case `.', `..', and `.zfs'. 2542 */ 2543 if (offset == 0) { 2544 (void) strcpy(zap.za_name, "."); 2545 zap.za_normalization_conflict = 0; 2546 objnum = zp->z_id; 2547 type = DT_DIR; 2548 } else if (offset == 1) { 2549 (void) strcpy(zap.za_name, ".."); 2550 zap.za_normalization_conflict = 0; 2551 objnum = parent; 2552 type = DT_DIR; 2553 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2554 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2555 zap.za_normalization_conflict = 0; 2556 objnum = ZFSCTL_INO_ROOT; 2557 type = DT_DIR; 2558 } else { 2559 /* 2560 * Grab next entry. 2561 */ 2562 if (error = zap_cursor_retrieve(&zc, &zap)) { 2563 if ((*eofp = (error == ENOENT)) != 0) 2564 break; 2565 else 2566 goto update; 2567 } 2568 2569 if (zap.za_integer_length != 8 || 2570 zap.za_num_integers != 1) { 2571 cmn_err(CE_WARN, "zap_readdir: bad directory " 2572 "entry, obj = %lld, offset = %lld\n", 2573 (u_longlong_t)zp->z_id, 2574 (u_longlong_t)offset); 2575 error = SET_ERROR(ENXIO); 2576 goto update; 2577 } 2578 2579 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2580 /* 2581 * MacOS X can extract the object type here such as: 2582 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2583 */ 2584 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2585 2586 if (check_sysattrs && !zap.za_normalization_conflict) { 2587#ifdef TODO 2588 zap.za_normalization_conflict = 2589 xattr_sysattr_casechk(zap.za_name); 2590#else 2591 panic("%s:%u: TODO", __func__, __LINE__); 2592#endif 2593 } 2594 } 2595 2596 if (flags & V_RDDIR_ACCFILTER) { 2597 /* 2598 * If we have no access at all, don't include 2599 * this entry in the returned information 2600 */ 2601 znode_t *ezp; 2602 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2603 goto skip_entry; 2604 if (!zfs_has_access(ezp, cr)) { 2605 VN_RELE(ZTOV(ezp)); 2606 goto skip_entry; 2607 } 2608 VN_RELE(ZTOV(ezp)); 2609 } 2610 2611 if (flags & V_RDDIR_ENTFLAGS) 2612 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2613 else 2614 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2615 2616 /* 2617 * Will this entry fit in the buffer? 2618 */ 2619 if (outcount + reclen > bufsize) { 2620 /* 2621 * Did we manage to fit anything in the buffer? 2622 */ 2623 if (!outcount) { 2624 error = SET_ERROR(EINVAL); 2625 goto update; 2626 } 2627 break; 2628 } 2629 if (flags & V_RDDIR_ENTFLAGS) { 2630 /* 2631 * Add extended flag entry: 2632 */ 2633 eodp->ed_ino = objnum; 2634 eodp->ed_reclen = reclen; 2635 /* NOTE: ed_off is the offset for the *next* entry */ 2636 next = &(eodp->ed_off); 2637 eodp->ed_eflags = zap.za_normalization_conflict ? 2638 ED_CASE_CONFLICT : 0; 2639 (void) strncpy(eodp->ed_name, zap.za_name, 2640 EDIRENT_NAMELEN(reclen)); 2641 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2642 } else { 2643 /* 2644 * Add normal entry: 2645 */ 2646 odp->d_ino = objnum; 2647 odp->d_reclen = reclen; 2648 odp->d_namlen = strlen(zap.za_name); 2649 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 2650 odp->d_type = type; 2651 odp = (dirent64_t *)((intptr_t)odp + reclen); 2652 } 2653 outcount += reclen; 2654 2655 ASSERT(outcount <= bufsize); 2656 2657 /* Prefetch znode */ 2658 if (prefetch) 2659 dmu_prefetch(os, objnum, 0, 0); 2660 2661 skip_entry: 2662 /* 2663 * Move to the next entry, fill in the previous offset. 2664 */ 2665 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2666 zap_cursor_advance(&zc); 2667 offset = zap_cursor_serialize(&zc); 2668 } else { 2669 offset += 1; 2670 } 2671 2672 if (cooks != NULL) { 2673 *cooks++ = offset; 2674 ncooks--; 2675 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 2676 } 2677 } 2678 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2679 2680 /* Subtract unused cookies */ 2681 if (ncookies != NULL) 2682 *ncookies -= ncooks; 2683 2684 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2685 iovp->iov_base += outcount; 2686 iovp->iov_len -= outcount; 2687 uio->uio_resid -= outcount; 2688 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2689 /* 2690 * Reset the pointer. 2691 */ 2692 offset = uio->uio_loffset; 2693 } 2694 2695update: 2696 zap_cursor_fini(&zc); 2697 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2698 kmem_free(outbuf, bufsize); 2699 2700 if (error == ENOENT) 2701 error = 0; 2702 2703 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2704 2705 uio->uio_loffset = offset; 2706 ZFS_EXIT(zfsvfs); 2707 if (error != 0 && cookies != NULL) { 2708 free(*cookies, M_TEMP); 2709 *cookies = NULL; 2710 *ncookies = 0; 2711 } 2712 return (error); 2713} 2714 2715ulong_t zfs_fsync_sync_cnt = 4; 2716 2717static int 2718zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2719{ 2720 znode_t *zp = VTOZ(vp); 2721 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2722 2723 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2724 2725 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { 2726 ZFS_ENTER(zfsvfs); 2727 ZFS_VERIFY_ZP(zp); 2728 zil_commit(zfsvfs->z_log, zp->z_id); 2729 ZFS_EXIT(zfsvfs); 2730 } 2731 return (0); 2732} 2733 2734 2735/* 2736 * Get the requested file attributes and place them in the provided 2737 * vattr structure. 2738 * 2739 * IN: vp - vnode of file. 2740 * vap - va_mask identifies requested attributes. 2741 * If AT_XVATTR set, then optional attrs are requested 2742 * flags - ATTR_NOACLCHECK (CIFS server context) 2743 * cr - credentials of caller. 2744 * ct - caller context 2745 * 2746 * OUT: vap - attribute values. 2747 * 2748 * RETURN: 0 (always succeeds). 2749 */ 2750/* ARGSUSED */ 2751static int 2752zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2753 caller_context_t *ct) 2754{ 2755 znode_t *zp = VTOZ(vp); 2756 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2757 int error = 0; 2758 uint32_t blksize; 2759 u_longlong_t nblocks; 2760 uint64_t links; 2761 uint64_t mtime[2], ctime[2], crtime[2], rdev; 2762 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2763 xoptattr_t *xoap = NULL; 2764 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2765 sa_bulk_attr_t bulk[4]; 2766 int count = 0; 2767 2768 ZFS_ENTER(zfsvfs); 2769 ZFS_VERIFY_ZP(zp); 2770 2771 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2772 2773 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 2774 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 2775 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16); 2776 if (vp->v_type == VBLK || vp->v_type == VCHR) 2777 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, 2778 &rdev, 8); 2779 2780 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 2781 ZFS_EXIT(zfsvfs); 2782 return (error); 2783 } 2784 2785 /* 2786 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2787 * Also, if we are the owner don't bother, since owner should 2788 * always be allowed to read basic attributes of file. 2789 */ 2790 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 2791 (vap->va_uid != crgetuid(cr))) { 2792 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2793 skipaclchk, cr)) { 2794 ZFS_EXIT(zfsvfs); 2795 return (error); 2796 } 2797 } 2798 2799 /* 2800 * Return all attributes. It's cheaper to provide the answer 2801 * than to determine whether we were asked the question. 2802 */ 2803 2804 mutex_enter(&zp->z_lock); 2805 vap->va_type = IFTOVT(zp->z_mode); 2806 vap->va_mode = zp->z_mode & ~S_IFMT; 2807#ifdef sun 2808 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2809#else 2810 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 2811#endif 2812 vap->va_nodeid = zp->z_id; 2813 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2814 links = zp->z_links + 1; 2815 else 2816 links = zp->z_links; 2817 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */ 2818 vap->va_size = zp->z_size; 2819#ifdef sun 2820 vap->va_rdev = vp->v_rdev; 2821#else 2822 if (vp->v_type == VBLK || vp->v_type == VCHR) 2823 vap->va_rdev = zfs_cmpldev(rdev); 2824#endif 2825 vap->va_seq = zp->z_seq; 2826 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2827 2828 /* 2829 * Add in any requested optional attributes and the create time. 2830 * Also set the corresponding bits in the returned attribute bitmap. 2831 */ 2832 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2833 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2834 xoap->xoa_archive = 2835 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2836 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2837 } 2838 2839 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2840 xoap->xoa_readonly = 2841 ((zp->z_pflags & ZFS_READONLY) != 0); 2842 XVA_SET_RTN(xvap, XAT_READONLY); 2843 } 2844 2845 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2846 xoap->xoa_system = 2847 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2848 XVA_SET_RTN(xvap, XAT_SYSTEM); 2849 } 2850 2851 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2852 xoap->xoa_hidden = 2853 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2854 XVA_SET_RTN(xvap, XAT_HIDDEN); 2855 } 2856 2857 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2858 xoap->xoa_nounlink = 2859 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2860 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2861 } 2862 2863 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2864 xoap->xoa_immutable = 2865 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2866 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2867 } 2868 2869 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2870 xoap->xoa_appendonly = 2871 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2872 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2873 } 2874 2875 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2876 xoap->xoa_nodump = 2877 ((zp->z_pflags & ZFS_NODUMP) != 0); 2878 XVA_SET_RTN(xvap, XAT_NODUMP); 2879 } 2880 2881 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2882 xoap->xoa_opaque = 2883 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2884 XVA_SET_RTN(xvap, XAT_OPAQUE); 2885 } 2886 2887 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2888 xoap->xoa_av_quarantined = 2889 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2890 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2891 } 2892 2893 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2894 xoap->xoa_av_modified = 2895 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2896 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2897 } 2898 2899 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2900 vp->v_type == VREG) { 2901 zfs_sa_get_scanstamp(zp, xvap); 2902 } 2903 2904 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2905 uint64_t times[2]; 2906 2907 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 2908 times, sizeof (times)); 2909 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 2910 XVA_SET_RTN(xvap, XAT_CREATETIME); 2911 } 2912 2913 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2914 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2915 XVA_SET_RTN(xvap, XAT_REPARSE); 2916 } 2917 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2918 xoap->xoa_generation = zp->z_gen; 2919 XVA_SET_RTN(xvap, XAT_GEN); 2920 } 2921 2922 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2923 xoap->xoa_offline = 2924 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2925 XVA_SET_RTN(xvap, XAT_OFFLINE); 2926 } 2927 2928 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2929 xoap->xoa_sparse = 2930 ((zp->z_pflags & ZFS_SPARSE) != 0); 2931 XVA_SET_RTN(xvap, XAT_SPARSE); 2932 } 2933 } 2934 2935 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2936 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2937 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2938 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 2939 2940 mutex_exit(&zp->z_lock); 2941 2942 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 2943 vap->va_blksize = blksize; 2944 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2945 2946 if (zp->z_blksz == 0) { 2947 /* 2948 * Block size hasn't been set; suggest maximal I/O transfers. 2949 */ 2950 vap->va_blksize = zfsvfs->z_max_blksz; 2951 } 2952 2953 ZFS_EXIT(zfsvfs); 2954 return (0); 2955} 2956 2957/* 2958 * Set the file attributes to the values contained in the 2959 * vattr structure. 2960 * 2961 * IN: vp - vnode of file to be modified. 2962 * vap - new attribute values. 2963 * If AT_XVATTR set, then optional attrs are being set 2964 * flags - ATTR_UTIME set if non-default time values provided. 2965 * - ATTR_NOACLCHECK (CIFS context only). 2966 * cr - credentials of caller. 2967 * ct - caller context 2968 * 2969 * RETURN: 0 on success, error code on failure. 2970 * 2971 * Timestamps: 2972 * vp - ctime updated, mtime updated if size changed. 2973 */ 2974/* ARGSUSED */ 2975static int 2976zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2977 caller_context_t *ct) 2978{ 2979 znode_t *zp = VTOZ(vp); 2980 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2981 zilog_t *zilog; 2982 dmu_tx_t *tx; 2983 vattr_t oldva; 2984 xvattr_t tmpxvattr; 2985 uint_t mask = vap->va_mask; 2986 uint_t saved_mask = 0; 2987 uint64_t saved_mode; 2988 int trim_mask = 0; 2989 uint64_t new_mode; 2990 uint64_t new_uid, new_gid; 2991 uint64_t xattr_obj; 2992 uint64_t mtime[2], ctime[2]; 2993 znode_t *attrzp; 2994 int need_policy = FALSE; 2995 int err, err2; 2996 zfs_fuid_info_t *fuidp = NULL; 2997 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2998 xoptattr_t *xoap; 2999 zfs_acl_t *aclp; 3000 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 3001 boolean_t fuid_dirtied = B_FALSE; 3002 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 3003 int count = 0, xattr_count = 0; 3004 3005 if (mask == 0) 3006 return (0); 3007 3008 if (mask & AT_NOSET) 3009 return (SET_ERROR(EINVAL)); 3010 3011 ZFS_ENTER(zfsvfs); 3012 ZFS_VERIFY_ZP(zp); 3013 3014 zilog = zfsvfs->z_log; 3015 3016 /* 3017 * Make sure that if we have ephemeral uid/gid or xvattr specified 3018 * that file system is at proper version level 3019 */ 3020 3021 if (zfsvfs->z_use_fuids == B_FALSE && 3022 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 3023 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 3024 (mask & AT_XVATTR))) { 3025 ZFS_EXIT(zfsvfs); 3026 return (SET_ERROR(EINVAL)); 3027 } 3028 3029 if (mask & AT_SIZE && vp->v_type == VDIR) { 3030 ZFS_EXIT(zfsvfs); 3031 return (SET_ERROR(EISDIR)); 3032 } 3033 3034 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 3035 ZFS_EXIT(zfsvfs); 3036 return (SET_ERROR(EINVAL)); 3037 } 3038 3039 /* 3040 * If this is an xvattr_t, then get a pointer to the structure of 3041 * optional attributes. If this is NULL, then we have a vattr_t. 3042 */ 3043 xoap = xva_getxoptattr(xvap); 3044 3045 xva_init(&tmpxvattr); 3046 3047 /* 3048 * Immutable files can only alter immutable bit and atime 3049 */ 3050 if ((zp->z_pflags & ZFS_IMMUTABLE) && 3051 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 3052 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 3053 ZFS_EXIT(zfsvfs); 3054 return (SET_ERROR(EPERM)); 3055 } 3056 3057 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 3058 ZFS_EXIT(zfsvfs); 3059 return (SET_ERROR(EPERM)); 3060 } 3061 3062 /* 3063 * Verify timestamps doesn't overflow 32 bits. 3064 * ZFS can handle large timestamps, but 32bit syscalls can't 3065 * handle times greater than 2039. This check should be removed 3066 * once large timestamps are fully supported. 3067 */ 3068 if (mask & (AT_ATIME | AT_MTIME)) { 3069 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 3070 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 3071 ZFS_EXIT(zfsvfs); 3072 return (SET_ERROR(EOVERFLOW)); 3073 } 3074 } 3075 3076top: 3077 attrzp = NULL; 3078 aclp = NULL; 3079 3080 /* Can this be moved to before the top label? */ 3081 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 3082 ZFS_EXIT(zfsvfs); 3083 return (SET_ERROR(EROFS)); 3084 } 3085 3086 /* 3087 * First validate permissions 3088 */ 3089 3090 if (mask & AT_SIZE) { 3091 /* 3092 * XXX - Note, we are not providing any open 3093 * mode flags here (like FNDELAY), so we may 3094 * block if there are locks present... this 3095 * should be addressed in openat(). 3096 */ 3097 /* XXX - would it be OK to generate a log record here? */ 3098 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 3099 if (err) { 3100 ZFS_EXIT(zfsvfs); 3101 return (err); 3102 } 3103 } 3104 3105 if (mask & (AT_ATIME|AT_MTIME) || 3106 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 3107 XVA_ISSET_REQ(xvap, XAT_READONLY) || 3108 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 3109 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 3110 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 3111 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 3112 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 3113 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 3114 skipaclchk, cr); 3115 } 3116 3117 if (mask & (AT_UID|AT_GID)) { 3118 int idmask = (mask & (AT_UID|AT_GID)); 3119 int take_owner; 3120 int take_group; 3121 3122 /* 3123 * NOTE: even if a new mode is being set, 3124 * we may clear S_ISUID/S_ISGID bits. 3125 */ 3126 3127 if (!(mask & AT_MODE)) 3128 vap->va_mode = zp->z_mode; 3129 3130 /* 3131 * Take ownership or chgrp to group we are a member of 3132 */ 3133 3134 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 3135 take_group = (mask & AT_GID) && 3136 zfs_groupmember(zfsvfs, vap->va_gid, cr); 3137 3138 /* 3139 * If both AT_UID and AT_GID are set then take_owner and 3140 * take_group must both be set in order to allow taking 3141 * ownership. 3142 * 3143 * Otherwise, send the check through secpolicy_vnode_setattr() 3144 * 3145 */ 3146 3147 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 3148 ((idmask == AT_UID) && take_owner) || 3149 ((idmask == AT_GID) && take_group)) { 3150 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 3151 skipaclchk, cr) == 0) { 3152 /* 3153 * Remove setuid/setgid for non-privileged users 3154 */ 3155 secpolicy_setid_clear(vap, vp, cr); 3156 trim_mask = (mask & (AT_UID|AT_GID)); 3157 } else { 3158 need_policy = TRUE; 3159 } 3160 } else { 3161 need_policy = TRUE; 3162 } 3163 } 3164 3165 mutex_enter(&zp->z_lock); 3166 oldva.va_mode = zp->z_mode; 3167 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 3168 if (mask & AT_XVATTR) { 3169 /* 3170 * Update xvattr mask to include only those attributes 3171 * that are actually changing. 3172 * 3173 * the bits will be restored prior to actually setting 3174 * the attributes so the caller thinks they were set. 3175 */ 3176 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 3177 if (xoap->xoa_appendonly != 3178 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 3179 need_policy = TRUE; 3180 } else { 3181 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 3182 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 3183 } 3184 } 3185 3186 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 3187 if (xoap->xoa_nounlink != 3188 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 3189 need_policy = TRUE; 3190 } else { 3191 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 3192 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 3193 } 3194 } 3195 3196 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 3197 if (xoap->xoa_immutable != 3198 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 3199 need_policy = TRUE; 3200 } else { 3201 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 3202 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 3203 } 3204 } 3205 3206 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 3207 if (xoap->xoa_nodump != 3208 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 3209 need_policy = TRUE; 3210 } else { 3211 XVA_CLR_REQ(xvap, XAT_NODUMP); 3212 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 3213 } 3214 } 3215 3216 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 3217 if (xoap->xoa_av_modified != 3218 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 3219 need_policy = TRUE; 3220 } else { 3221 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 3222 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 3223 } 3224 } 3225 3226 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 3227 if ((vp->v_type != VREG && 3228 xoap->xoa_av_quarantined) || 3229 xoap->xoa_av_quarantined != 3230 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 3231 need_policy = TRUE; 3232 } else { 3233 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 3234 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 3235 } 3236 } 3237 3238 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 3239 mutex_exit(&zp->z_lock); 3240 ZFS_EXIT(zfsvfs); 3241 return (SET_ERROR(EPERM)); 3242 } 3243 3244 if (need_policy == FALSE && 3245 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 3246 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 3247 need_policy = TRUE; 3248 } 3249 } 3250 3251 mutex_exit(&zp->z_lock); 3252 3253 if (mask & AT_MODE) { 3254 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 3255 err = secpolicy_setid_setsticky_clear(vp, vap, 3256 &oldva, cr); 3257 if (err) { 3258 ZFS_EXIT(zfsvfs); 3259 return (err); 3260 } 3261 trim_mask |= AT_MODE; 3262 } else { 3263 need_policy = TRUE; 3264 } 3265 } 3266 3267 if (need_policy) { 3268 /* 3269 * If trim_mask is set then take ownership 3270 * has been granted or write_acl is present and user 3271 * has the ability to modify mode. In that case remove 3272 * UID|GID and or MODE from mask so that 3273 * secpolicy_vnode_setattr() doesn't revoke it. 3274 */ 3275 3276 if (trim_mask) { 3277 saved_mask = vap->va_mask; 3278 vap->va_mask &= ~trim_mask; 3279 if (trim_mask & AT_MODE) { 3280 /* 3281 * Save the mode, as secpolicy_vnode_setattr() 3282 * will overwrite it with ova.va_mode. 3283 */ 3284 saved_mode = vap->va_mode; 3285 } 3286 } 3287 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 3288 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 3289 if (err) { 3290 ZFS_EXIT(zfsvfs); 3291 return (err); 3292 } 3293 3294 if (trim_mask) { 3295 vap->va_mask |= saved_mask; 3296 if (trim_mask & AT_MODE) { 3297 /* 3298 * Recover the mode after 3299 * secpolicy_vnode_setattr(). 3300 */ 3301 vap->va_mode = saved_mode; 3302 } 3303 } 3304 } 3305 3306 /* 3307 * secpolicy_vnode_setattr, or take ownership may have 3308 * changed va_mask 3309 */ 3310 mask = vap->va_mask; 3311 3312 if ((mask & (AT_UID | AT_GID))) { 3313 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 3314 &xattr_obj, sizeof (xattr_obj)); 3315 3316 if (err == 0 && xattr_obj) { 3317 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 3318 if (err) 3319 goto out2; 3320 } 3321 if (mask & AT_UID) { 3322 new_uid = zfs_fuid_create(zfsvfs, 3323 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 3324 if (new_uid != zp->z_uid && 3325 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 3326 if (attrzp) 3327 VN_RELE(ZTOV(attrzp)); 3328 err = SET_ERROR(EDQUOT); 3329 goto out2; 3330 } 3331 } 3332 3333 if (mask & AT_GID) { 3334 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 3335 cr, ZFS_GROUP, &fuidp); 3336 if (new_gid != zp->z_gid && 3337 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 3338 if (attrzp) 3339 VN_RELE(ZTOV(attrzp)); 3340 err = SET_ERROR(EDQUOT); 3341 goto out2; 3342 } 3343 } 3344 } 3345 tx = dmu_tx_create(zfsvfs->z_os); 3346 3347 if (mask & AT_MODE) { 3348 uint64_t pmode = zp->z_mode; 3349 uint64_t acl_obj; 3350 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 3351 3352 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && 3353 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 3354 err = SET_ERROR(EPERM); 3355 goto out; 3356 } 3357 3358 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 3359 goto out; 3360 3361 mutex_enter(&zp->z_lock); 3362 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 3363 /* 3364 * Are we upgrading ACL from old V0 format 3365 * to V1 format? 3366 */ 3367 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 3368 zfs_znode_acl_version(zp) == 3369 ZFS_ACL_VERSION_INITIAL) { 3370 dmu_tx_hold_free(tx, acl_obj, 0, 3371 DMU_OBJECT_END); 3372 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3373 0, aclp->z_acl_bytes); 3374 } else { 3375 dmu_tx_hold_write(tx, acl_obj, 0, 3376 aclp->z_acl_bytes); 3377 } 3378 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3379 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3380 0, aclp->z_acl_bytes); 3381 } 3382 mutex_exit(&zp->z_lock); 3383 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3384 } else { 3385 if ((mask & AT_XVATTR) && 3386 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3387 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3388 else 3389 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3390 } 3391 3392 if (attrzp) { 3393 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 3394 } 3395 3396 fuid_dirtied = zfsvfs->z_fuid_dirty; 3397 if (fuid_dirtied) 3398 zfs_fuid_txhold(zfsvfs, tx); 3399 3400 zfs_sa_upgrade_txholds(tx, zp); 3401 3402 err = dmu_tx_assign(tx, TXG_WAIT); 3403 if (err) 3404 goto out; 3405 3406 count = 0; 3407 /* 3408 * Set each attribute requested. 3409 * We group settings according to the locks they need to acquire. 3410 * 3411 * Note: you cannot set ctime directly, although it will be 3412 * updated as a side-effect of calling this function. 3413 */ 3414 3415 3416 if (mask & (AT_UID|AT_GID|AT_MODE)) 3417 mutex_enter(&zp->z_acl_lock); 3418 mutex_enter(&zp->z_lock); 3419 3420 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3421 &zp->z_pflags, sizeof (zp->z_pflags)); 3422 3423 if (attrzp) { 3424 if (mask & (AT_UID|AT_GID|AT_MODE)) 3425 mutex_enter(&attrzp->z_acl_lock); 3426 mutex_enter(&attrzp->z_lock); 3427 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3428 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3429 sizeof (attrzp->z_pflags)); 3430 } 3431 3432 if (mask & (AT_UID|AT_GID)) { 3433 3434 if (mask & AT_UID) { 3435 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3436 &new_uid, sizeof (new_uid)); 3437 zp->z_uid = new_uid; 3438 if (attrzp) { 3439 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3440 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3441 sizeof (new_uid)); 3442 attrzp->z_uid = new_uid; 3443 } 3444 } 3445 3446 if (mask & AT_GID) { 3447 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3448 NULL, &new_gid, sizeof (new_gid)); 3449 zp->z_gid = new_gid; 3450 if (attrzp) { 3451 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3452 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 3453 sizeof (new_gid)); 3454 attrzp->z_gid = new_gid; 3455 } 3456 } 3457 if (!(mask & AT_MODE)) { 3458 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 3459 NULL, &new_mode, sizeof (new_mode)); 3460 new_mode = zp->z_mode; 3461 } 3462 err = zfs_acl_chown_setattr(zp); 3463 ASSERT(err == 0); 3464 if (attrzp) { 3465 err = zfs_acl_chown_setattr(attrzp); 3466 ASSERT(err == 0); 3467 } 3468 } 3469 3470 if (mask & AT_MODE) { 3471 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 3472 &new_mode, sizeof (new_mode)); 3473 zp->z_mode = new_mode; 3474 ASSERT3U((uintptr_t)aclp, !=, 0); 3475 err = zfs_aclset_common(zp, aclp, cr, tx); 3476 ASSERT0(err); 3477 if (zp->z_acl_cached) 3478 zfs_acl_free(zp->z_acl_cached); 3479 zp->z_acl_cached = aclp; 3480 aclp = NULL; 3481 } 3482 3483 3484 if (mask & AT_ATIME) { 3485 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 3486 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 3487 &zp->z_atime, sizeof (zp->z_atime)); 3488 } 3489 3490 if (mask & AT_MTIME) { 3491 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 3492 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 3493 mtime, sizeof (mtime)); 3494 } 3495 3496 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 3497 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 3498 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 3499 NULL, mtime, sizeof (mtime)); 3500 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3501 &ctime, sizeof (ctime)); 3502 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 3503 B_TRUE); 3504 } else if (mask != 0) { 3505 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3506 &ctime, sizeof (ctime)); 3507 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime, 3508 B_TRUE); 3509 if (attrzp) { 3510 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3511 SA_ZPL_CTIME(zfsvfs), NULL, 3512 &ctime, sizeof (ctime)); 3513 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 3514 mtime, ctime, B_TRUE); 3515 } 3516 } 3517 /* 3518 * Do this after setting timestamps to prevent timestamp 3519 * update from toggling bit 3520 */ 3521 3522 if (xoap && (mask & AT_XVATTR)) { 3523 3524 /* 3525 * restore trimmed off masks 3526 * so that return masks can be set for caller. 3527 */ 3528 3529 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 3530 XVA_SET_REQ(xvap, XAT_APPENDONLY); 3531 } 3532 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 3533 XVA_SET_REQ(xvap, XAT_NOUNLINK); 3534 } 3535 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 3536 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 3537 } 3538 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 3539 XVA_SET_REQ(xvap, XAT_NODUMP); 3540 } 3541 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 3542 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 3543 } 3544 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 3545 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 3546 } 3547 3548 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3549 ASSERT(vp->v_type == VREG); 3550 3551 zfs_xvattr_set(zp, xvap, tx); 3552 } 3553 3554 if (fuid_dirtied) 3555 zfs_fuid_sync(zfsvfs, tx); 3556 3557 if (mask != 0) 3558 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 3559 3560 mutex_exit(&zp->z_lock); 3561 if (mask & (AT_UID|AT_GID|AT_MODE)) 3562 mutex_exit(&zp->z_acl_lock); 3563 3564 if (attrzp) { 3565 if (mask & (AT_UID|AT_GID|AT_MODE)) 3566 mutex_exit(&attrzp->z_acl_lock); 3567 mutex_exit(&attrzp->z_lock); 3568 } 3569out: 3570 if (err == 0 && attrzp) { 3571 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 3572 xattr_count, tx); 3573 ASSERT(err2 == 0); 3574 } 3575 3576 if (attrzp) 3577 VN_RELE(ZTOV(attrzp)); 3578 3579 if (aclp) 3580 zfs_acl_free(aclp); 3581 3582 if (fuidp) { 3583 zfs_fuid_info_free(fuidp); 3584 fuidp = NULL; 3585 } 3586 3587 if (err) { 3588 dmu_tx_abort(tx); 3589 if (err == ERESTART) 3590 goto top; 3591 } else { 3592 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3593 dmu_tx_commit(tx); 3594 } 3595 3596out2: 3597 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3598 zil_commit(zilog, 0); 3599 3600 ZFS_EXIT(zfsvfs); 3601 return (err); 3602} 3603 3604typedef struct zfs_zlock { 3605 krwlock_t *zl_rwlock; /* lock we acquired */ 3606 znode_t *zl_znode; /* znode we held */ 3607 struct zfs_zlock *zl_next; /* next in list */ 3608} zfs_zlock_t; 3609 3610/* 3611 * Drop locks and release vnodes that were held by zfs_rename_lock(). 3612 */ 3613static void 3614zfs_rename_unlock(zfs_zlock_t **zlpp) 3615{ 3616 zfs_zlock_t *zl; 3617 3618 while ((zl = *zlpp) != NULL) { 3619 if (zl->zl_znode != NULL) 3620 VN_RELE(ZTOV(zl->zl_znode)); 3621 rw_exit(zl->zl_rwlock); 3622 *zlpp = zl->zl_next; 3623 kmem_free(zl, sizeof (*zl)); 3624 } 3625} 3626 3627/* 3628 * Search back through the directory tree, using the ".." entries. 3629 * Lock each directory in the chain to prevent concurrent renames. 3630 * Fail any attempt to move a directory into one of its own descendants. 3631 * XXX - z_parent_lock can overlap with map or grow locks 3632 */ 3633static int 3634zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 3635{ 3636 zfs_zlock_t *zl; 3637 znode_t *zp = tdzp; 3638 uint64_t rootid = zp->z_zfsvfs->z_root; 3639 uint64_t oidp = zp->z_id; 3640 krwlock_t *rwlp = &szp->z_parent_lock; 3641 krw_t rw = RW_WRITER; 3642 3643 /* 3644 * First pass write-locks szp and compares to zp->z_id. 3645 * Later passes read-lock zp and compare to zp->z_parent. 3646 */ 3647 do { 3648 if (!rw_tryenter(rwlp, rw)) { 3649 /* 3650 * Another thread is renaming in this path. 3651 * Note that if we are a WRITER, we don't have any 3652 * parent_locks held yet. 3653 */ 3654 if (rw == RW_READER && zp->z_id > szp->z_id) { 3655 /* 3656 * Drop our locks and restart 3657 */ 3658 zfs_rename_unlock(&zl); 3659 *zlpp = NULL; 3660 zp = tdzp; 3661 oidp = zp->z_id; 3662 rwlp = &szp->z_parent_lock; 3663 rw = RW_WRITER; 3664 continue; 3665 } else { 3666 /* 3667 * Wait for other thread to drop its locks 3668 */ 3669 rw_enter(rwlp, rw); 3670 } 3671 } 3672 3673 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 3674 zl->zl_rwlock = rwlp; 3675 zl->zl_znode = NULL; 3676 zl->zl_next = *zlpp; 3677 *zlpp = zl; 3678 3679 if (oidp == szp->z_id) /* We're a descendant of szp */ 3680 return (SET_ERROR(EINVAL)); 3681 3682 if (oidp == rootid) /* We've hit the top */ 3683 return (0); 3684 3685 if (rw == RW_READER) { /* i.e. not the first pass */ 3686 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp); 3687 if (error) 3688 return (error); 3689 zl->zl_znode = zp; 3690 } 3691 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs), 3692 &oidp, sizeof (oidp)); 3693 rwlp = &zp->z_parent_lock; 3694 rw = RW_READER; 3695 3696 } while (zp->z_id != sdzp->z_id); 3697 3698 return (0); 3699} 3700 3701/* 3702 * Move an entry from the provided source directory to the target 3703 * directory. Change the entry name as indicated. 3704 * 3705 * IN: sdvp - Source directory containing the "old entry". 3706 * snm - Old entry name. 3707 * tdvp - Target directory to contain the "new entry". 3708 * tnm - New entry name. 3709 * cr - credentials of caller. 3710 * ct - caller context 3711 * flags - case flags 3712 * 3713 * RETURN: 0 on success, error code on failure. 3714 * 3715 * Timestamps: 3716 * sdvp,tdvp - ctime|mtime updated 3717 */ 3718/*ARGSUSED*/ 3719static int 3720zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 3721 caller_context_t *ct, int flags) 3722{ 3723 znode_t *tdzp, *szp, *tzp; 3724 znode_t *sdzp = VTOZ(sdvp); 3725 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 3726 zilog_t *zilog; 3727 vnode_t *realvp; 3728 zfs_dirlock_t *sdl, *tdl; 3729 dmu_tx_t *tx; 3730 zfs_zlock_t *zl; 3731 int cmp, serr, terr; 3732 int error = 0; 3733 int zflg = 0; 3734 boolean_t waited = B_FALSE; 3735 3736 ZFS_ENTER(zfsvfs); 3737 ZFS_VERIFY_ZP(sdzp); 3738 zilog = zfsvfs->z_log; 3739 3740 /* 3741 * Make sure we have the real vp for the target directory. 3742 */ 3743 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 3744 tdvp = realvp; 3745 3746 tdzp = VTOZ(tdvp); 3747 ZFS_VERIFY_ZP(tdzp); 3748 3749 /* 3750 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 3751 * ctldir appear to have the same v_vfsp. 3752 */ 3753 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) { 3754 ZFS_EXIT(zfsvfs); 3755 return (SET_ERROR(EXDEV)); 3756 } 3757 3758 if (zfsvfs->z_utf8 && u8_validate(tnm, 3759 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3760 ZFS_EXIT(zfsvfs); 3761 return (SET_ERROR(EILSEQ)); 3762 } 3763 3764 if (flags & FIGNORECASE) 3765 zflg |= ZCILOOK; 3766 3767top: 3768 szp = NULL; 3769 tzp = NULL; 3770 zl = NULL; 3771 3772 /* 3773 * This is to prevent the creation of links into attribute space 3774 * by renaming a linked file into/outof an attribute directory. 3775 * See the comment in zfs_link() for why this is considered bad. 3776 */ 3777 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3778 ZFS_EXIT(zfsvfs); 3779 return (SET_ERROR(EINVAL)); 3780 } 3781 3782 /* 3783 * Lock source and target directory entries. To prevent deadlock, 3784 * a lock ordering must be defined. We lock the directory with 3785 * the smallest object id first, or if it's a tie, the one with 3786 * the lexically first name. 3787 */ 3788 if (sdzp->z_id < tdzp->z_id) { 3789 cmp = -1; 3790 } else if (sdzp->z_id > tdzp->z_id) { 3791 cmp = 1; 3792 } else { 3793 /* 3794 * First compare the two name arguments without 3795 * considering any case folding. 3796 */ 3797 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 3798 3799 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 3800 ASSERT(error == 0 || !zfsvfs->z_utf8); 3801 if (cmp == 0) { 3802 /* 3803 * POSIX: "If the old argument and the new argument 3804 * both refer to links to the same existing file, 3805 * the rename() function shall return successfully 3806 * and perform no other action." 3807 */ 3808 ZFS_EXIT(zfsvfs); 3809 return (0); 3810 } 3811 /* 3812 * If the file system is case-folding, then we may 3813 * have some more checking to do. A case-folding file 3814 * system is either supporting mixed case sensitivity 3815 * access or is completely case-insensitive. Note 3816 * that the file system is always case preserving. 3817 * 3818 * In mixed sensitivity mode case sensitive behavior 3819 * is the default. FIGNORECASE must be used to 3820 * explicitly request case insensitive behavior. 3821 * 3822 * If the source and target names provided differ only 3823 * by case (e.g., a request to rename 'tim' to 'Tim'), 3824 * we will treat this as a special case in the 3825 * case-insensitive mode: as long as the source name 3826 * is an exact match, we will allow this to proceed as 3827 * a name-change request. 3828 */ 3829 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3830 (zfsvfs->z_case == ZFS_CASE_MIXED && 3831 flags & FIGNORECASE)) && 3832 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3833 &error) == 0) { 3834 /* 3835 * case preserving rename request, require exact 3836 * name matches 3837 */ 3838 zflg |= ZCIEXACT; 3839 zflg &= ~ZCILOOK; 3840 } 3841 } 3842 3843 /* 3844 * If the source and destination directories are the same, we should 3845 * grab the z_name_lock of that directory only once. 3846 */ 3847 if (sdzp == tdzp) { 3848 zflg |= ZHAVELOCK; 3849 rw_enter(&sdzp->z_name_lock, RW_READER); 3850 } 3851 3852 if (cmp < 0) { 3853 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3854 ZEXISTS | zflg, NULL, NULL); 3855 terr = zfs_dirent_lock(&tdl, 3856 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3857 } else { 3858 terr = zfs_dirent_lock(&tdl, 3859 tdzp, tnm, &tzp, zflg, NULL, NULL); 3860 serr = zfs_dirent_lock(&sdl, 3861 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3862 NULL, NULL); 3863 } 3864 3865 if (serr) { 3866 /* 3867 * Source entry invalid or not there. 3868 */ 3869 if (!terr) { 3870 zfs_dirent_unlock(tdl); 3871 if (tzp) 3872 VN_RELE(ZTOV(tzp)); 3873 } 3874 3875 if (sdzp == tdzp) 3876 rw_exit(&sdzp->z_name_lock); 3877 3878 /* 3879 * FreeBSD: In OpenSolaris they only check if rename source is 3880 * ".." here, because "." is handled in their lookup. This is 3881 * not the case for FreeBSD, so we check for "." explicitly. 3882 */ 3883 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 3884 serr = SET_ERROR(EINVAL); 3885 ZFS_EXIT(zfsvfs); 3886 return (serr); 3887 } 3888 if (terr) { 3889 zfs_dirent_unlock(sdl); 3890 VN_RELE(ZTOV(szp)); 3891 3892 if (sdzp == tdzp) 3893 rw_exit(&sdzp->z_name_lock); 3894 3895 if (strcmp(tnm, "..") == 0) 3896 terr = SET_ERROR(EINVAL); 3897 ZFS_EXIT(zfsvfs); 3898 return (terr); 3899 } 3900 3901 /* 3902 * Must have write access at the source to remove the old entry 3903 * and write access at the target to create the new entry. 3904 * Note that if target and source are the same, this can be 3905 * done in a single check. 3906 */ 3907 3908 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3909 goto out; 3910 3911 if (ZTOV(szp)->v_type == VDIR) { 3912 /* 3913 * Check to make sure rename is valid. 3914 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3915 */ 3916 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3917 goto out; 3918 } 3919 3920 /* 3921 * Does target exist? 3922 */ 3923 if (tzp) { 3924 /* 3925 * Source and target must be the same type. 3926 */ 3927 if (ZTOV(szp)->v_type == VDIR) { 3928 if (ZTOV(tzp)->v_type != VDIR) { 3929 error = SET_ERROR(ENOTDIR); 3930 goto out; 3931 } 3932 } else { 3933 if (ZTOV(tzp)->v_type == VDIR) { 3934 error = SET_ERROR(EISDIR); 3935 goto out; 3936 } 3937 } 3938 /* 3939 * POSIX dictates that when the source and target 3940 * entries refer to the same file object, rename 3941 * must do nothing and exit without error. 3942 */ 3943 if (szp->z_id == tzp->z_id) { 3944 error = 0; 3945 goto out; 3946 } 3947 } 3948 3949 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3950 if (tzp) 3951 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3952 3953 /* 3954 * notify the target directory if it is not the same 3955 * as source directory. 3956 */ 3957 if (tdvp != sdvp) { 3958 vnevent_rename_dest_dir(tdvp, ct); 3959 } 3960 3961 tx = dmu_tx_create(zfsvfs->z_os); 3962 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3963 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 3964 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3965 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3966 if (sdzp != tdzp) { 3967 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 3968 zfs_sa_upgrade_txholds(tx, tdzp); 3969 } 3970 if (tzp) { 3971 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 3972 zfs_sa_upgrade_txholds(tx, tzp); 3973 } 3974 3975 zfs_sa_upgrade_txholds(tx, szp); 3976 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3977 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 3978 if (error) { 3979 if (zl != NULL) 3980 zfs_rename_unlock(&zl); 3981 zfs_dirent_unlock(sdl); 3982 zfs_dirent_unlock(tdl); 3983 3984 if (sdzp == tdzp) 3985 rw_exit(&sdzp->z_name_lock); 3986 3987 VN_RELE(ZTOV(szp)); 3988 if (tzp) 3989 VN_RELE(ZTOV(tzp)); 3990 if (error == ERESTART) { 3991 waited = B_TRUE; 3992 dmu_tx_wait(tx); 3993 dmu_tx_abort(tx); 3994 goto top; 3995 } 3996 dmu_tx_abort(tx); 3997 ZFS_EXIT(zfsvfs); 3998 return (error); 3999 } 4000 4001 if (tzp) /* Attempt to remove the existing target */ 4002 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 4003 4004 if (error == 0) { 4005 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 4006 if (error == 0) { 4007 szp->z_pflags |= ZFS_AV_MODIFIED; 4008 4009 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 4010 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 4011 ASSERT0(error); 4012 4013 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 4014 if (error == 0) { 4015 zfs_log_rename(zilog, tx, TX_RENAME | 4016 (flags & FIGNORECASE ? TX_CI : 0), sdzp, 4017 sdl->dl_name, tdzp, tdl->dl_name, szp); 4018 4019 /* 4020 * Update path information for the target vnode 4021 */ 4022 vn_renamepath(tdvp, ZTOV(szp), tnm, 4023 strlen(tnm)); 4024 } else { 4025 /* 4026 * At this point, we have successfully created 4027 * the target name, but have failed to remove 4028 * the source name. Since the create was done 4029 * with the ZRENAMING flag, there are 4030 * complications; for one, the link count is 4031 * wrong. The easiest way to deal with this 4032 * is to remove the newly created target, and 4033 * return the original error. This must 4034 * succeed; fortunately, it is very unlikely to 4035 * fail, since we just created it. 4036 */ 4037 VERIFY3U(zfs_link_destroy(tdl, szp, tx, 4038 ZRENAMING, NULL), ==, 0); 4039 } 4040 } 4041#ifdef FREEBSD_NAMECACHE 4042 if (error == 0) { 4043 cache_purge(sdvp); 4044 cache_purge(tdvp); 4045 cache_purge(ZTOV(szp)); 4046 if (tzp) 4047 cache_purge(ZTOV(tzp)); 4048 } 4049#endif 4050 } 4051 4052 dmu_tx_commit(tx); 4053out: 4054 if (zl != NULL) 4055 zfs_rename_unlock(&zl); 4056 4057 zfs_dirent_unlock(sdl); 4058 zfs_dirent_unlock(tdl); 4059 4060 if (sdzp == tdzp) 4061 rw_exit(&sdzp->z_name_lock); 4062 4063 4064 VN_RELE(ZTOV(szp)); 4065 if (tzp) 4066 VN_RELE(ZTOV(tzp)); 4067 4068 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4069 zil_commit(zilog, 0); 4070 4071 ZFS_EXIT(zfsvfs); 4072 4073 return (error); 4074} 4075 4076/* 4077 * Insert the indicated symbolic reference entry into the directory. 4078 * 4079 * IN: dvp - Directory to contain new symbolic link. 4080 * link - Name for new symlink entry. 4081 * vap - Attributes of new entry. 4082 * cr - credentials of caller. 4083 * ct - caller context 4084 * flags - case flags 4085 * 4086 * RETURN: 0 on success, error code on failure. 4087 * 4088 * Timestamps: 4089 * dvp - ctime|mtime updated 4090 */ 4091/*ARGSUSED*/ 4092static int 4093zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, 4094 cred_t *cr, kthread_t *td) 4095{ 4096 znode_t *zp, *dzp = VTOZ(dvp); 4097 zfs_dirlock_t *dl; 4098 dmu_tx_t *tx; 4099 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4100 zilog_t *zilog; 4101 uint64_t len = strlen(link); 4102 int error; 4103 int zflg = ZNEW; 4104 zfs_acl_ids_t acl_ids; 4105 boolean_t fuid_dirtied; 4106 uint64_t txtype = TX_SYMLINK; 4107 boolean_t waited = B_FALSE; 4108 int flags = 0; 4109 4110 ASSERT(vap->va_type == VLNK); 4111 4112 ZFS_ENTER(zfsvfs); 4113 ZFS_VERIFY_ZP(dzp); 4114 zilog = zfsvfs->z_log; 4115 4116 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 4117 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4118 ZFS_EXIT(zfsvfs); 4119 return (SET_ERROR(EILSEQ)); 4120 } 4121 if (flags & FIGNORECASE) 4122 zflg |= ZCILOOK; 4123 4124 if (len > MAXPATHLEN) { 4125 ZFS_EXIT(zfsvfs); 4126 return (SET_ERROR(ENAMETOOLONG)); 4127 } 4128 4129 if ((error = zfs_acl_ids_create(dzp, 0, 4130 vap, cr, NULL, &acl_ids)) != 0) { 4131 ZFS_EXIT(zfsvfs); 4132 return (error); 4133 } 4134 4135 getnewvnode_reserve(1); 4136 4137top: 4138 /* 4139 * Attempt to lock directory; fail if entry already exists. 4140 */ 4141 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 4142 if (error) { 4143 zfs_acl_ids_free(&acl_ids); 4144 getnewvnode_drop_reserve(); 4145 ZFS_EXIT(zfsvfs); 4146 return (error); 4147 } 4148 4149 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4150 zfs_acl_ids_free(&acl_ids); 4151 zfs_dirent_unlock(dl); 4152 getnewvnode_drop_reserve(); 4153 ZFS_EXIT(zfsvfs); 4154 return (error); 4155 } 4156 4157 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 4158 zfs_acl_ids_free(&acl_ids); 4159 zfs_dirent_unlock(dl); 4160 getnewvnode_drop_reserve(); 4161 ZFS_EXIT(zfsvfs); 4162 return (SET_ERROR(EDQUOT)); 4163 } 4164 tx = dmu_tx_create(zfsvfs->z_os); 4165 fuid_dirtied = zfsvfs->z_fuid_dirty; 4166 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 4167 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4168 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 4169 ZFS_SA_BASE_ATTR_SIZE + len); 4170 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 4171 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 4172 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 4173 acl_ids.z_aclp->z_acl_bytes); 4174 } 4175 if (fuid_dirtied) 4176 zfs_fuid_txhold(zfsvfs, tx); 4177 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4178 if (error) { 4179 zfs_dirent_unlock(dl); 4180 if (error == ERESTART) { 4181 waited = B_TRUE; 4182 dmu_tx_wait(tx); 4183 dmu_tx_abort(tx); 4184 goto top; 4185 } 4186 zfs_acl_ids_free(&acl_ids); 4187 dmu_tx_abort(tx); 4188 getnewvnode_drop_reserve(); 4189 ZFS_EXIT(zfsvfs); 4190 return (error); 4191 } 4192 4193 /* 4194 * Create a new object for the symlink. 4195 * for version 4 ZPL datsets the symlink will be an SA attribute 4196 */ 4197 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 4198 4199 if (fuid_dirtied) 4200 zfs_fuid_sync(zfsvfs, tx); 4201 4202 mutex_enter(&zp->z_lock); 4203 if (zp->z_is_sa) 4204 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 4205 link, len, tx); 4206 else 4207 zfs_sa_symlink(zp, link, len, tx); 4208 mutex_exit(&zp->z_lock); 4209 4210 zp->z_size = len; 4211 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 4212 &zp->z_size, sizeof (zp->z_size), tx); 4213 /* 4214 * Insert the new object into the directory. 4215 */ 4216 (void) zfs_link_create(dl, zp, tx, ZNEW); 4217 4218 if (flags & FIGNORECASE) 4219 txtype |= TX_CI; 4220 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 4221 *vpp = ZTOV(zp); 4222 4223 zfs_acl_ids_free(&acl_ids); 4224 4225 dmu_tx_commit(tx); 4226 4227 getnewvnode_drop_reserve(); 4228 4229 zfs_dirent_unlock(dl); 4230 4231 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4232 zil_commit(zilog, 0); 4233 4234 ZFS_EXIT(zfsvfs); 4235 return (error); 4236} 4237 4238/* 4239 * Return, in the buffer contained in the provided uio structure, 4240 * the symbolic path referred to by vp. 4241 * 4242 * IN: vp - vnode of symbolic link. 4243 * uio - structure to contain the link path. 4244 * cr - credentials of caller. 4245 * ct - caller context 4246 * 4247 * OUT: uio - structure containing the link path. 4248 * 4249 * RETURN: 0 on success, error code on failure. 4250 * 4251 * Timestamps: 4252 * vp - atime updated 4253 */ 4254/* ARGSUSED */ 4255static int 4256zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 4257{ 4258 znode_t *zp = VTOZ(vp); 4259 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4260 int error; 4261 4262 ZFS_ENTER(zfsvfs); 4263 ZFS_VERIFY_ZP(zp); 4264 4265 mutex_enter(&zp->z_lock); 4266 if (zp->z_is_sa) 4267 error = sa_lookup_uio(zp->z_sa_hdl, 4268 SA_ZPL_SYMLINK(zfsvfs), uio); 4269 else 4270 error = zfs_sa_readlink(zp, uio); 4271 mutex_exit(&zp->z_lock); 4272 4273 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4274 4275 ZFS_EXIT(zfsvfs); 4276 return (error); 4277} 4278 4279/* 4280 * Insert a new entry into directory tdvp referencing svp. 4281 * 4282 * IN: tdvp - Directory to contain new entry. 4283 * svp - vnode of new entry. 4284 * name - name of new entry. 4285 * cr - credentials of caller. 4286 * ct - caller context 4287 * 4288 * RETURN: 0 on success, error code on failure. 4289 * 4290 * Timestamps: 4291 * tdvp - ctime|mtime updated 4292 * svp - ctime updated 4293 */ 4294/* ARGSUSED */ 4295static int 4296zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 4297 caller_context_t *ct, int flags) 4298{ 4299 znode_t *dzp = VTOZ(tdvp); 4300 znode_t *tzp, *szp; 4301 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4302 zilog_t *zilog; 4303 zfs_dirlock_t *dl; 4304 dmu_tx_t *tx; 4305 vnode_t *realvp; 4306 int error; 4307 int zf = ZNEW; 4308 uint64_t parent; 4309 uid_t owner; 4310 boolean_t waited = B_FALSE; 4311 4312 ASSERT(tdvp->v_type == VDIR); 4313 4314 ZFS_ENTER(zfsvfs); 4315 ZFS_VERIFY_ZP(dzp); 4316 zilog = zfsvfs->z_log; 4317 4318 if (VOP_REALVP(svp, &realvp, ct) == 0) 4319 svp = realvp; 4320 4321 /* 4322 * POSIX dictates that we return EPERM here. 4323 * Better choices include ENOTSUP or EISDIR. 4324 */ 4325 if (svp->v_type == VDIR) { 4326 ZFS_EXIT(zfsvfs); 4327 return (SET_ERROR(EPERM)); 4328 } 4329 4330 szp = VTOZ(svp); 4331 ZFS_VERIFY_ZP(szp); 4332 4333 /* 4334 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 4335 * ctldir appear to have the same v_vfsp. 4336 */ 4337 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) { 4338 ZFS_EXIT(zfsvfs); 4339 return (SET_ERROR(EXDEV)); 4340 } 4341 4342 /* Prevent links to .zfs/shares files */ 4343 4344 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 4345 &parent, sizeof (uint64_t))) != 0) { 4346 ZFS_EXIT(zfsvfs); 4347 return (error); 4348 } 4349 if (parent == zfsvfs->z_shares_dir) { 4350 ZFS_EXIT(zfsvfs); 4351 return (SET_ERROR(EPERM)); 4352 } 4353 4354 if (zfsvfs->z_utf8 && u8_validate(name, 4355 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4356 ZFS_EXIT(zfsvfs); 4357 return (SET_ERROR(EILSEQ)); 4358 } 4359 if (flags & FIGNORECASE) 4360 zf |= ZCILOOK; 4361 4362 /* 4363 * We do not support links between attributes and non-attributes 4364 * because of the potential security risk of creating links 4365 * into "normal" file space in order to circumvent restrictions 4366 * imposed in attribute space. 4367 */ 4368 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 4369 ZFS_EXIT(zfsvfs); 4370 return (SET_ERROR(EINVAL)); 4371 } 4372 4373 4374 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 4375 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) { 4376 ZFS_EXIT(zfsvfs); 4377 return (SET_ERROR(EPERM)); 4378 } 4379 4380 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4381 ZFS_EXIT(zfsvfs); 4382 return (error); 4383 } 4384 4385top: 4386 /* 4387 * Attempt to lock directory; fail if entry already exists. 4388 */ 4389 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 4390 if (error) { 4391 ZFS_EXIT(zfsvfs); 4392 return (error); 4393 } 4394 4395 tx = dmu_tx_create(zfsvfs->z_os); 4396 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4397 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4398 zfs_sa_upgrade_txholds(tx, szp); 4399 zfs_sa_upgrade_txholds(tx, dzp); 4400 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4401 if (error) { 4402 zfs_dirent_unlock(dl); 4403 if (error == ERESTART) { 4404 waited = B_TRUE; 4405 dmu_tx_wait(tx); 4406 dmu_tx_abort(tx); 4407 goto top; 4408 } 4409 dmu_tx_abort(tx); 4410 ZFS_EXIT(zfsvfs); 4411 return (error); 4412 } 4413 4414 error = zfs_link_create(dl, szp, tx, 0); 4415 4416 if (error == 0) { 4417 uint64_t txtype = TX_LINK; 4418 if (flags & FIGNORECASE) 4419 txtype |= TX_CI; 4420 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 4421 } 4422 4423 dmu_tx_commit(tx); 4424 4425 zfs_dirent_unlock(dl); 4426 4427 if (error == 0) { 4428 vnevent_link(svp, ct); 4429 } 4430 4431 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4432 zil_commit(zilog, 0); 4433 4434 ZFS_EXIT(zfsvfs); 4435 return (error); 4436} 4437 4438#ifdef sun 4439/* 4440 * zfs_null_putapage() is used when the file system has been force 4441 * unmounted. It just drops the pages. 4442 */ 4443/* ARGSUSED */ 4444static int 4445zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4446 size_t *lenp, int flags, cred_t *cr) 4447{ 4448 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 4449 return (0); 4450} 4451 4452/* 4453 * Push a page out to disk, klustering if possible. 4454 * 4455 * IN: vp - file to push page to. 4456 * pp - page to push. 4457 * flags - additional flags. 4458 * cr - credentials of caller. 4459 * 4460 * OUT: offp - start of range pushed. 4461 * lenp - len of range pushed. 4462 * 4463 * RETURN: 0 on success, error code on failure. 4464 * 4465 * NOTE: callers must have locked the page to be pushed. On 4466 * exit, the page (and all other pages in the kluster) must be 4467 * unlocked. 4468 */ 4469/* ARGSUSED */ 4470static int 4471zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4472 size_t *lenp, int flags, cred_t *cr) 4473{ 4474 znode_t *zp = VTOZ(vp); 4475 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4476 dmu_tx_t *tx; 4477 u_offset_t off, koff; 4478 size_t len, klen; 4479 int err; 4480 4481 off = pp->p_offset; 4482 len = PAGESIZE; 4483 /* 4484 * If our blocksize is bigger than the page size, try to kluster 4485 * multiple pages so that we write a full block (thus avoiding 4486 * a read-modify-write). 4487 */ 4488 if (off < zp->z_size && zp->z_blksz > PAGESIZE) { 4489 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 4490 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 4491 ASSERT(koff <= zp->z_size); 4492 if (koff + klen > zp->z_size) 4493 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); 4494 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 4495 } 4496 ASSERT3U(btop(len), ==, btopr(len)); 4497 4498 /* 4499 * Can't push pages past end-of-file. 4500 */ 4501 if (off >= zp->z_size) { 4502 /* ignore all pages */ 4503 err = 0; 4504 goto out; 4505 } else if (off + len > zp->z_size) { 4506 int npages = btopr(zp->z_size - off); 4507 page_t *trunc; 4508 4509 page_list_break(&pp, &trunc, npages); 4510 /* ignore pages past end of file */ 4511 if (trunc) 4512 pvn_write_done(trunc, flags); 4513 len = zp->z_size - off; 4514 } 4515 4516 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4517 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4518 err = SET_ERROR(EDQUOT); 4519 goto out; 4520 } 4521 tx = dmu_tx_create(zfsvfs->z_os); 4522 dmu_tx_hold_write(tx, zp->z_id, off, len); 4523 4524 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4525 zfs_sa_upgrade_txholds(tx, zp); 4526 err = dmu_tx_assign(tx, TXG_WAIT); 4527 if (err != 0) { 4528 dmu_tx_abort(tx); 4529 goto out; 4530 } 4531 4532 if (zp->z_blksz <= PAGESIZE) { 4533 caddr_t va = zfs_map_page(pp, S_READ); 4534 ASSERT3U(len, <=, PAGESIZE); 4535 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 4536 zfs_unmap_page(pp, va); 4537 } else { 4538 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 4539 } 4540 4541 if (err == 0) { 4542 uint64_t mtime[2], ctime[2]; 4543 sa_bulk_attr_t bulk[3]; 4544 int count = 0; 4545 4546 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4547 &mtime, 16); 4548 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4549 &ctime, 16); 4550 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4551 &zp->z_pflags, 8); 4552 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4553 B_TRUE); 4554 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4555 } 4556 dmu_tx_commit(tx); 4557 4558out: 4559 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 4560 if (offp) 4561 *offp = off; 4562 if (lenp) 4563 *lenp = len; 4564 4565 return (err); 4566} 4567 4568/* 4569 * Copy the portion of the file indicated from pages into the file. 4570 * The pages are stored in a page list attached to the files vnode. 4571 * 4572 * IN: vp - vnode of file to push page data to. 4573 * off - position in file to put data. 4574 * len - amount of data to write. 4575 * flags - flags to control the operation. 4576 * cr - credentials of caller. 4577 * ct - caller context. 4578 * 4579 * RETURN: 0 on success, error code on failure. 4580 * 4581 * Timestamps: 4582 * vp - ctime|mtime updated 4583 */ 4584/*ARGSUSED*/ 4585static int 4586zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4587 caller_context_t *ct) 4588{ 4589 znode_t *zp = VTOZ(vp); 4590 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4591 page_t *pp; 4592 size_t io_len; 4593 u_offset_t io_off; 4594 uint_t blksz; 4595 rl_t *rl; 4596 int error = 0; 4597 4598 ZFS_ENTER(zfsvfs); 4599 ZFS_VERIFY_ZP(zp); 4600 4601 /* 4602 * Align this request to the file block size in case we kluster. 4603 * XXX - this can result in pretty aggresive locking, which can 4604 * impact simultanious read/write access. One option might be 4605 * to break up long requests (len == 0) into block-by-block 4606 * operations to get narrower locking. 4607 */ 4608 blksz = zp->z_blksz; 4609 if (ISP2(blksz)) 4610 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 4611 else 4612 io_off = 0; 4613 if (len > 0 && ISP2(blksz)) 4614 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 4615 else 4616 io_len = 0; 4617 4618 if (io_len == 0) { 4619 /* 4620 * Search the entire vp list for pages >= io_off. 4621 */ 4622 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 4623 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 4624 goto out; 4625 } 4626 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 4627 4628 if (off > zp->z_size) { 4629 /* past end of file */ 4630 zfs_range_unlock(rl); 4631 ZFS_EXIT(zfsvfs); 4632 return (0); 4633 } 4634 4635 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); 4636 4637 for (off = io_off; io_off < off + len; io_off += io_len) { 4638 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 4639 pp = page_lookup(vp, io_off, 4640 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 4641 } else { 4642 pp = page_lookup_nowait(vp, io_off, 4643 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 4644 } 4645 4646 if (pp != NULL && pvn_getdirty(pp, flags)) { 4647 int err; 4648 4649 /* 4650 * Found a dirty page to push 4651 */ 4652 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 4653 if (err) 4654 error = err; 4655 } else { 4656 io_len = PAGESIZE; 4657 } 4658 } 4659out: 4660 zfs_range_unlock(rl); 4661 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4662 zil_commit(zfsvfs->z_log, zp->z_id); 4663 ZFS_EXIT(zfsvfs); 4664 return (error); 4665} 4666#endif /* sun */ 4667 4668/*ARGSUSED*/ 4669void 4670zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4671{ 4672 znode_t *zp = VTOZ(vp); 4673 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4674 int error; 4675 4676 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4677 if (zp->z_sa_hdl == NULL) { 4678 /* 4679 * The fs has been unmounted, or we did a 4680 * suspend/resume and this file no longer exists. 4681 */ 4682 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4683 vrecycle(vp); 4684 return; 4685 } 4686 4687 mutex_enter(&zp->z_lock); 4688 if (zp->z_unlinked) { 4689 /* 4690 * Fast path to recycle a vnode of a removed file. 4691 */ 4692 mutex_exit(&zp->z_lock); 4693 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4694 vrecycle(vp); 4695 return; 4696 } 4697 mutex_exit(&zp->z_lock); 4698 4699 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4700 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4701 4702 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4703 zfs_sa_upgrade_txholds(tx, zp); 4704 error = dmu_tx_assign(tx, TXG_WAIT); 4705 if (error) { 4706 dmu_tx_abort(tx); 4707 } else { 4708 mutex_enter(&zp->z_lock); 4709 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4710 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4711 zp->z_atime_dirty = 0; 4712 mutex_exit(&zp->z_lock); 4713 dmu_tx_commit(tx); 4714 } 4715 } 4716 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4717} 4718 4719#ifdef sun 4720/* 4721 * Bounds-check the seek operation. 4722 * 4723 * IN: vp - vnode seeking within 4724 * ooff - old file offset 4725 * noffp - pointer to new file offset 4726 * ct - caller context 4727 * 4728 * RETURN: 0 on success, EINVAL if new offset invalid. 4729 */ 4730/* ARGSUSED */ 4731static int 4732zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 4733 caller_context_t *ct) 4734{ 4735 if (vp->v_type == VDIR) 4736 return (0); 4737 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4738} 4739 4740/* 4741 * Pre-filter the generic locking function to trap attempts to place 4742 * a mandatory lock on a memory mapped file. 4743 */ 4744static int 4745zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 4746 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 4747{ 4748 znode_t *zp = VTOZ(vp); 4749 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4750 4751 ZFS_ENTER(zfsvfs); 4752 ZFS_VERIFY_ZP(zp); 4753 4754 /* 4755 * We are following the UFS semantics with respect to mapcnt 4756 * here: If we see that the file is mapped already, then we will 4757 * return an error, but we don't worry about races between this 4758 * function and zfs_map(). 4759 */ 4760 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { 4761 ZFS_EXIT(zfsvfs); 4762 return (SET_ERROR(EAGAIN)); 4763 } 4764 ZFS_EXIT(zfsvfs); 4765 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4766} 4767 4768/* 4769 * If we can't find a page in the cache, we will create a new page 4770 * and fill it with file data. For efficiency, we may try to fill 4771 * multiple pages at once (klustering) to fill up the supplied page 4772 * list. Note that the pages to be filled are held with an exclusive 4773 * lock to prevent access by other threads while they are being filled. 4774 */ 4775static int 4776zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 4777 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 4778{ 4779 znode_t *zp = VTOZ(vp); 4780 page_t *pp, *cur_pp; 4781 objset_t *os = zp->z_zfsvfs->z_os; 4782 u_offset_t io_off, total; 4783 size_t io_len; 4784 int err; 4785 4786 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4787 /* 4788 * We only have a single page, don't bother klustering 4789 */ 4790 io_off = off; 4791 io_len = PAGESIZE; 4792 pp = page_create_va(vp, io_off, io_len, 4793 PG_EXCL | PG_WAIT, seg, addr); 4794 } else { 4795 /* 4796 * Try to find enough pages to fill the page list 4797 */ 4798 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4799 &io_len, off, plsz, 0); 4800 } 4801 if (pp == NULL) { 4802 /* 4803 * The page already exists, nothing to do here. 4804 */ 4805 *pl = NULL; 4806 return (0); 4807 } 4808 4809 /* 4810 * Fill the pages in the kluster. 4811 */ 4812 cur_pp = pp; 4813 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4814 caddr_t va; 4815 4816 ASSERT3U(io_off, ==, cur_pp->p_offset); 4817 va = zfs_map_page(cur_pp, S_WRITE); 4818 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4819 DMU_READ_PREFETCH); 4820 zfs_unmap_page(cur_pp, va); 4821 if (err) { 4822 /* On error, toss the entire kluster */ 4823 pvn_read_done(pp, B_ERROR); 4824 /* convert checksum errors into IO errors */ 4825 if (err == ECKSUM) 4826 err = SET_ERROR(EIO); 4827 return (err); 4828 } 4829 cur_pp = cur_pp->p_next; 4830 } 4831 4832 /* 4833 * Fill in the page list array from the kluster starting 4834 * from the desired offset `off'. 4835 * NOTE: the page list will always be null terminated. 4836 */ 4837 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4838 ASSERT(pl == NULL || (*pl)->p_offset == off); 4839 4840 return (0); 4841} 4842 4843/* 4844 * Return pointers to the pages for the file region [off, off + len] 4845 * in the pl array. If plsz is greater than len, this function may 4846 * also return page pointers from after the specified region 4847 * (i.e. the region [off, off + plsz]). These additional pages are 4848 * only returned if they are already in the cache, or were created as 4849 * part of a klustered read. 4850 * 4851 * IN: vp - vnode of file to get data from. 4852 * off - position in file to get data from. 4853 * len - amount of data to retrieve. 4854 * plsz - length of provided page list. 4855 * seg - segment to obtain pages for. 4856 * addr - virtual address of fault. 4857 * rw - mode of created pages. 4858 * cr - credentials of caller. 4859 * ct - caller context. 4860 * 4861 * OUT: protp - protection mode of created pages. 4862 * pl - list of pages created. 4863 * 4864 * RETURN: 0 on success, error code on failure. 4865 * 4866 * Timestamps: 4867 * vp - atime updated 4868 */ 4869/* ARGSUSED */ 4870static int 4871zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4872 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4873 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4874{ 4875 znode_t *zp = VTOZ(vp); 4876 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4877 page_t **pl0 = pl; 4878 int err = 0; 4879 4880 /* we do our own caching, faultahead is unnecessary */ 4881 if (pl == NULL) 4882 return (0); 4883 else if (len > plsz) 4884 len = plsz; 4885 else 4886 len = P2ROUNDUP(len, PAGESIZE); 4887 ASSERT(plsz >= len); 4888 4889 ZFS_ENTER(zfsvfs); 4890 ZFS_VERIFY_ZP(zp); 4891 4892 if (protp) 4893 *protp = PROT_ALL; 4894 4895 /* 4896 * Loop through the requested range [off, off + len) looking 4897 * for pages. If we don't find a page, we will need to create 4898 * a new page and fill it with data from the file. 4899 */ 4900 while (len > 0) { 4901 if (*pl = page_lookup(vp, off, SE_SHARED)) 4902 *(pl+1) = NULL; 4903 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 4904 goto out; 4905 while (*pl) { 4906 ASSERT3U((*pl)->p_offset, ==, off); 4907 off += PAGESIZE; 4908 addr += PAGESIZE; 4909 if (len > 0) { 4910 ASSERT3U(len, >=, PAGESIZE); 4911 len -= PAGESIZE; 4912 } 4913 ASSERT3U(plsz, >=, PAGESIZE); 4914 plsz -= PAGESIZE; 4915 pl++; 4916 } 4917 } 4918 4919 /* 4920 * Fill out the page array with any pages already in the cache. 4921 */ 4922 while (plsz > 0 && 4923 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 4924 off += PAGESIZE; 4925 plsz -= PAGESIZE; 4926 } 4927out: 4928 if (err) { 4929 /* 4930 * Release any pages we have previously locked. 4931 */ 4932 while (pl > pl0) 4933 page_unlock(*--pl); 4934 } else { 4935 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4936 } 4937 4938 *pl = NULL; 4939 4940 ZFS_EXIT(zfsvfs); 4941 return (err); 4942} 4943 4944/* 4945 * Request a memory map for a section of a file. This code interacts 4946 * with common code and the VM system as follows: 4947 * 4948 * - common code calls mmap(), which ends up in smmap_common() 4949 * - this calls VOP_MAP(), which takes you into (say) zfs 4950 * - zfs_map() calls as_map(), passing segvn_create() as the callback 4951 * - segvn_create() creates the new segment and calls VOP_ADDMAP() 4952 * - zfs_addmap() updates z_mapcnt 4953 */ 4954/*ARGSUSED*/ 4955static int 4956zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4957 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4958 caller_context_t *ct) 4959{ 4960 znode_t *zp = VTOZ(vp); 4961 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4962 segvn_crargs_t vn_a; 4963 int error; 4964 4965 ZFS_ENTER(zfsvfs); 4966 ZFS_VERIFY_ZP(zp); 4967 4968 if ((prot & PROT_WRITE) && (zp->z_pflags & 4969 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 4970 ZFS_EXIT(zfsvfs); 4971 return (SET_ERROR(EPERM)); 4972 } 4973 4974 if ((prot & (PROT_READ | PROT_EXEC)) && 4975 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 4976 ZFS_EXIT(zfsvfs); 4977 return (SET_ERROR(EACCES)); 4978 } 4979 4980 if (vp->v_flag & VNOMAP) { 4981 ZFS_EXIT(zfsvfs); 4982 return (SET_ERROR(ENOSYS)); 4983 } 4984 4985 if (off < 0 || len > MAXOFFSET_T - off) { 4986 ZFS_EXIT(zfsvfs); 4987 return (SET_ERROR(ENXIO)); 4988 } 4989 4990 if (vp->v_type != VREG) { 4991 ZFS_EXIT(zfsvfs); 4992 return (SET_ERROR(ENODEV)); 4993 } 4994 4995 /* 4996 * If file is locked, disallow mapping. 4997 */ 4998 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { 4999 ZFS_EXIT(zfsvfs); 5000 return (SET_ERROR(EAGAIN)); 5001 } 5002 5003 as_rangelock(as); 5004 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 5005 if (error != 0) { 5006 as_rangeunlock(as); 5007 ZFS_EXIT(zfsvfs); 5008 return (error); 5009 } 5010 5011 vn_a.vp = vp; 5012 vn_a.offset = (u_offset_t)off; 5013 vn_a.type = flags & MAP_TYPE; 5014 vn_a.prot = prot; 5015 vn_a.maxprot = maxprot; 5016 vn_a.cred = cr; 5017 vn_a.amp = NULL; 5018 vn_a.flags = flags & ~MAP_TYPE; 5019 vn_a.szc = 0; 5020 vn_a.lgrp_mem_policy_flags = 0; 5021 5022 error = as_map(as, *addrp, len, segvn_create, &vn_a); 5023 5024 as_rangeunlock(as); 5025 ZFS_EXIT(zfsvfs); 5026 return (error); 5027} 5028 5029/* ARGSUSED */ 5030static int 5031zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5032 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 5033 caller_context_t *ct) 5034{ 5035 uint64_t pages = btopr(len); 5036 5037 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 5038 return (0); 5039} 5040 5041/* 5042 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 5043 * more accurate mtime for the associated file. Since we don't have a way of 5044 * detecting when the data was actually modified, we have to resort to 5045 * heuristics. If an explicit msync() is done, then we mark the mtime when the 5046 * last page is pushed. The problem occurs when the msync() call is omitted, 5047 * which by far the most common case: 5048 * 5049 * open() 5050 * mmap() 5051 * <modify memory> 5052 * munmap() 5053 * close() 5054 * <time lapse> 5055 * putpage() via fsflush 5056 * 5057 * If we wait until fsflush to come along, we can have a modification time that 5058 * is some arbitrary point in the future. In order to prevent this in the 5059 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 5060 * torn down. 5061 */ 5062/* ARGSUSED */ 5063static int 5064zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5065 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 5066 caller_context_t *ct) 5067{ 5068 uint64_t pages = btopr(len); 5069 5070 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 5071 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 5072 5073 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 5074 vn_has_cached_data(vp)) 5075 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 5076 5077 return (0); 5078} 5079 5080/* 5081 * Free or allocate space in a file. Currently, this function only 5082 * supports the `F_FREESP' command. However, this command is somewhat 5083 * misnamed, as its functionality includes the ability to allocate as 5084 * well as free space. 5085 * 5086 * IN: vp - vnode of file to free data in. 5087 * cmd - action to take (only F_FREESP supported). 5088 * bfp - section of file to free/alloc. 5089 * flag - current file open mode flags. 5090 * offset - current file offset. 5091 * cr - credentials of caller [UNUSED]. 5092 * ct - caller context. 5093 * 5094 * RETURN: 0 on success, error code on failure. 5095 * 5096 * Timestamps: 5097 * vp - ctime|mtime updated 5098 */ 5099/* ARGSUSED */ 5100static int 5101zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 5102 offset_t offset, cred_t *cr, caller_context_t *ct) 5103{ 5104 znode_t *zp = VTOZ(vp); 5105 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5106 uint64_t off, len; 5107 int error; 5108 5109 ZFS_ENTER(zfsvfs); 5110 ZFS_VERIFY_ZP(zp); 5111 5112 if (cmd != F_FREESP) { 5113 ZFS_EXIT(zfsvfs); 5114 return (SET_ERROR(EINVAL)); 5115 } 5116 5117 if (error = convoff(vp, bfp, 0, offset)) { 5118 ZFS_EXIT(zfsvfs); 5119 return (error); 5120 } 5121 5122 if (bfp->l_len < 0) { 5123 ZFS_EXIT(zfsvfs); 5124 return (SET_ERROR(EINVAL)); 5125 } 5126 5127 off = bfp->l_start; 5128 len = bfp->l_len; /* 0 means from off to end of file */ 5129 5130 error = zfs_freesp(zp, off, len, flag, TRUE); 5131 5132 ZFS_EXIT(zfsvfs); 5133 return (error); 5134} 5135#endif /* sun */ 5136 5137CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 5138CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 5139 5140/*ARGSUSED*/ 5141static int 5142zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 5143{ 5144 znode_t *zp = VTOZ(vp); 5145 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5146 uint32_t gen; 5147 uint64_t gen64; 5148 uint64_t object = zp->z_id; 5149 zfid_short_t *zfid; 5150 int size, i, error; 5151 5152 ZFS_ENTER(zfsvfs); 5153 ZFS_VERIFY_ZP(zp); 5154 5155 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 5156 &gen64, sizeof (uint64_t))) != 0) { 5157 ZFS_EXIT(zfsvfs); 5158 return (error); 5159 } 5160 5161 gen = (uint32_t)gen64; 5162 5163 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 5164 5165#ifdef illumos 5166 if (fidp->fid_len < size) { 5167 fidp->fid_len = size; 5168 ZFS_EXIT(zfsvfs); 5169 return (SET_ERROR(ENOSPC)); 5170 } 5171#else 5172 fidp->fid_len = size; 5173#endif 5174 5175 zfid = (zfid_short_t *)fidp; 5176 5177 zfid->zf_len = size; 5178 5179 for (i = 0; i < sizeof (zfid->zf_object); i++) 5180 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 5181 5182 /* Must have a non-zero generation number to distinguish from .zfs */ 5183 if (gen == 0) 5184 gen = 1; 5185 for (i = 0; i < sizeof (zfid->zf_gen); i++) 5186 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 5187 5188 if (size == LONG_FID_LEN) { 5189 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 5190 zfid_long_t *zlfid; 5191 5192 zlfid = (zfid_long_t *)fidp; 5193 5194 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 5195 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 5196 5197 /* XXX - this should be the generation number for the objset */ 5198 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 5199 zlfid->zf_setgen[i] = 0; 5200 } 5201 5202 ZFS_EXIT(zfsvfs); 5203 return (0); 5204} 5205 5206static int 5207zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 5208 caller_context_t *ct) 5209{ 5210 znode_t *zp, *xzp; 5211 zfsvfs_t *zfsvfs; 5212 zfs_dirlock_t *dl; 5213 int error; 5214 5215 switch (cmd) { 5216 case _PC_LINK_MAX: 5217 *valp = INT_MAX; 5218 return (0); 5219 5220 case _PC_FILESIZEBITS: 5221 *valp = 64; 5222 return (0); 5223#ifdef sun 5224 case _PC_XATTR_EXISTS: 5225 zp = VTOZ(vp); 5226 zfsvfs = zp->z_zfsvfs; 5227 ZFS_ENTER(zfsvfs); 5228 ZFS_VERIFY_ZP(zp); 5229 *valp = 0; 5230 error = zfs_dirent_lock(&dl, zp, "", &xzp, 5231 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 5232 if (error == 0) { 5233 zfs_dirent_unlock(dl); 5234 if (!zfs_dirempty(xzp)) 5235 *valp = 1; 5236 VN_RELE(ZTOV(xzp)); 5237 } else if (error == ENOENT) { 5238 /* 5239 * If there aren't extended attributes, it's the 5240 * same as having zero of them. 5241 */ 5242 error = 0; 5243 } 5244 ZFS_EXIT(zfsvfs); 5245 return (error); 5246 5247 case _PC_SATTR_ENABLED: 5248 case _PC_SATTR_EXISTS: 5249 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 5250 (vp->v_type == VREG || vp->v_type == VDIR); 5251 return (0); 5252 5253 case _PC_ACCESS_FILTERING: 5254 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 5255 vp->v_type == VDIR; 5256 return (0); 5257 5258 case _PC_ACL_ENABLED: 5259 *valp = _ACL_ACE_ENABLED; 5260 return (0); 5261#endif /* sun */ 5262 case _PC_MIN_HOLE_SIZE: 5263 *valp = (int)SPA_MINBLOCKSIZE; 5264 return (0); 5265#ifdef sun 5266 case _PC_TIMESTAMP_RESOLUTION: 5267 /* nanosecond timestamp resolution */ 5268 *valp = 1L; 5269 return (0); 5270#endif /* sun */ 5271 case _PC_ACL_EXTENDED: 5272 *valp = 0; 5273 return (0); 5274 5275 case _PC_ACL_NFS4: 5276 *valp = 1; 5277 return (0); 5278 5279 case _PC_ACL_PATH_MAX: 5280 *valp = ACL_MAX_ENTRIES; 5281 return (0); 5282 5283 default: 5284 return (EOPNOTSUPP); 5285 } 5286} 5287 5288/*ARGSUSED*/ 5289static int 5290zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5291 caller_context_t *ct) 5292{ 5293 znode_t *zp = VTOZ(vp); 5294 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5295 int error; 5296 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5297 5298 ZFS_ENTER(zfsvfs); 5299 ZFS_VERIFY_ZP(zp); 5300 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 5301 ZFS_EXIT(zfsvfs); 5302 5303 return (error); 5304} 5305 5306/*ARGSUSED*/ 5307int 5308zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5309 caller_context_t *ct) 5310{ 5311 znode_t *zp = VTOZ(vp); 5312 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5313 int error; 5314 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5315 zilog_t *zilog = zfsvfs->z_log; 5316 5317 ZFS_ENTER(zfsvfs); 5318 ZFS_VERIFY_ZP(zp); 5319 5320 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 5321 5322 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 5323 zil_commit(zilog, 0); 5324 5325 ZFS_EXIT(zfsvfs); 5326 return (error); 5327} 5328 5329#ifdef sun 5330/* 5331 * The smallest read we may consider to loan out an arcbuf. 5332 * This must be a power of 2. 5333 */ 5334int zcr_blksz_min = (1 << 10); /* 1K */ 5335/* 5336 * If set to less than the file block size, allow loaning out of an 5337 * arcbuf for a partial block read. This must be a power of 2. 5338 */ 5339int zcr_blksz_max = (1 << 17); /* 128K */ 5340 5341/*ARGSUSED*/ 5342static int 5343zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr, 5344 caller_context_t *ct) 5345{ 5346 znode_t *zp = VTOZ(vp); 5347 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5348 int max_blksz = zfsvfs->z_max_blksz; 5349 uio_t *uio = &xuio->xu_uio; 5350 ssize_t size = uio->uio_resid; 5351 offset_t offset = uio->uio_loffset; 5352 int blksz; 5353 int fullblk, i; 5354 arc_buf_t *abuf; 5355 ssize_t maxsize; 5356 int preamble, postamble; 5357 5358 if (xuio->xu_type != UIOTYPE_ZEROCOPY) 5359 return (SET_ERROR(EINVAL)); 5360 5361 ZFS_ENTER(zfsvfs); 5362 ZFS_VERIFY_ZP(zp); 5363 switch (ioflag) { 5364 case UIO_WRITE: 5365 /* 5366 * Loan out an arc_buf for write if write size is bigger than 5367 * max_blksz, and the file's block size is also max_blksz. 5368 */ 5369 blksz = max_blksz; 5370 if (size < blksz || zp->z_blksz != blksz) { 5371 ZFS_EXIT(zfsvfs); 5372 return (SET_ERROR(EINVAL)); 5373 } 5374 /* 5375 * Caller requests buffers for write before knowing where the 5376 * write offset might be (e.g. NFS TCP write). 5377 */ 5378 if (offset == -1) { 5379 preamble = 0; 5380 } else { 5381 preamble = P2PHASE(offset, blksz); 5382 if (preamble) { 5383 preamble = blksz - preamble; 5384 size -= preamble; 5385 } 5386 } 5387 5388 postamble = P2PHASE(size, blksz); 5389 size -= postamble; 5390 5391 fullblk = size / blksz; 5392 (void) dmu_xuio_init(xuio, 5393 (preamble != 0) + fullblk + (postamble != 0)); 5394 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble, 5395 int, postamble, int, 5396 (preamble != 0) + fullblk + (postamble != 0)); 5397 5398 /* 5399 * Have to fix iov base/len for partial buffers. They 5400 * currently represent full arc_buf's. 5401 */ 5402 if (preamble) { 5403 /* data begins in the middle of the arc_buf */ 5404 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5405 blksz); 5406 ASSERT(abuf); 5407 (void) dmu_xuio_add(xuio, abuf, 5408 blksz - preamble, preamble); 5409 } 5410 5411 for (i = 0; i < fullblk; i++) { 5412 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5413 blksz); 5414 ASSERT(abuf); 5415 (void) dmu_xuio_add(xuio, abuf, 0, blksz); 5416 } 5417 5418 if (postamble) { 5419 /* data ends in the middle of the arc_buf */ 5420 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5421 blksz); 5422 ASSERT(abuf); 5423 (void) dmu_xuio_add(xuio, abuf, 0, postamble); 5424 } 5425 break; 5426 case UIO_READ: 5427 /* 5428 * Loan out an arc_buf for read if the read size is larger than 5429 * the current file block size. Block alignment is not 5430 * considered. Partial arc_buf will be loaned out for read. 5431 */ 5432 blksz = zp->z_blksz; 5433 if (blksz < zcr_blksz_min) 5434 blksz = zcr_blksz_min; 5435 if (blksz > zcr_blksz_max) 5436 blksz = zcr_blksz_max; 5437 /* avoid potential complexity of dealing with it */ 5438 if (blksz > max_blksz) { 5439 ZFS_EXIT(zfsvfs); 5440 return (SET_ERROR(EINVAL)); 5441 } 5442 5443 maxsize = zp->z_size - uio->uio_loffset; 5444 if (size > maxsize) 5445 size = maxsize; 5446 5447 if (size < blksz || vn_has_cached_data(vp)) { 5448 ZFS_EXIT(zfsvfs); 5449 return (SET_ERROR(EINVAL)); 5450 } 5451 break; 5452 default: 5453 ZFS_EXIT(zfsvfs); 5454 return (SET_ERROR(EINVAL)); 5455 } 5456 5457 uio->uio_extflg = UIO_XUIO; 5458 XUIO_XUZC_RW(xuio) = ioflag; 5459 ZFS_EXIT(zfsvfs); 5460 return (0); 5461} 5462 5463/*ARGSUSED*/ 5464static int 5465zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct) 5466{ 5467 int i; 5468 arc_buf_t *abuf; 5469 int ioflag = XUIO_XUZC_RW(xuio); 5470 5471 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY); 5472 5473 i = dmu_xuio_cnt(xuio); 5474 while (i-- > 0) { 5475 abuf = dmu_xuio_arcbuf(xuio, i); 5476 /* 5477 * if abuf == NULL, it must be a write buffer 5478 * that has been returned in zfs_write(). 5479 */ 5480 if (abuf) 5481 dmu_return_arcbuf(abuf); 5482 ASSERT(abuf || ioflag == UIO_WRITE); 5483 } 5484 5485 dmu_xuio_fini(xuio); 5486 return (0); 5487} 5488 5489/* 5490 * Predeclare these here so that the compiler assumes that 5491 * this is an "old style" function declaration that does 5492 * not include arguments => we won't get type mismatch errors 5493 * in the initializations that follow. 5494 */ 5495static int zfs_inval(); 5496static int zfs_isdir(); 5497 5498static int 5499zfs_inval() 5500{ 5501 return (SET_ERROR(EINVAL)); 5502} 5503 5504static int 5505zfs_isdir() 5506{ 5507 return (SET_ERROR(EISDIR)); 5508} 5509/* 5510 * Directory vnode operations template 5511 */ 5512vnodeops_t *zfs_dvnodeops; 5513const fs_operation_def_t zfs_dvnodeops_template[] = { 5514 VOPNAME_OPEN, { .vop_open = zfs_open }, 5515 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5516 VOPNAME_READ, { .error = zfs_isdir }, 5517 VOPNAME_WRITE, { .error = zfs_isdir }, 5518 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5519 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5520 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5521 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5522 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5523 VOPNAME_CREATE, { .vop_create = zfs_create }, 5524 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5525 VOPNAME_LINK, { .vop_link = zfs_link }, 5526 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5527 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 5528 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5529 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5530 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 5531 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5532 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5533 VOPNAME_FID, { .vop_fid = zfs_fid }, 5534 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5535 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5536 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5537 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5538 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5539 NULL, NULL 5540}; 5541 5542/* 5543 * Regular file vnode operations template 5544 */ 5545vnodeops_t *zfs_fvnodeops; 5546const fs_operation_def_t zfs_fvnodeops_template[] = { 5547 VOPNAME_OPEN, { .vop_open = zfs_open }, 5548 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5549 VOPNAME_READ, { .vop_read = zfs_read }, 5550 VOPNAME_WRITE, { .vop_write = zfs_write }, 5551 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5552 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5553 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5554 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5555 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5556 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5557 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5558 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5559 VOPNAME_FID, { .vop_fid = zfs_fid }, 5560 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5561 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 5562 VOPNAME_SPACE, { .vop_space = zfs_space }, 5563 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 5564 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 5565 VOPNAME_MAP, { .vop_map = zfs_map }, 5566 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 5567 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 5568 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5569 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5570 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5571 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5572 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf }, 5573 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf }, 5574 NULL, NULL 5575}; 5576 5577/* 5578 * Symbolic link vnode operations template 5579 */ 5580vnodeops_t *zfs_symvnodeops; 5581const fs_operation_def_t zfs_symvnodeops_template[] = { 5582 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5583 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5584 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5585 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5586 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 5587 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5588 VOPNAME_FID, { .vop_fid = zfs_fid }, 5589 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5590 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5591 NULL, NULL 5592}; 5593 5594/* 5595 * special share hidden files vnode operations template 5596 */ 5597vnodeops_t *zfs_sharevnodeops; 5598const fs_operation_def_t zfs_sharevnodeops_template[] = { 5599 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5600 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5601 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5602 VOPNAME_FID, { .vop_fid = zfs_fid }, 5603 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5604 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5605 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5606 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5607 NULL, NULL 5608}; 5609 5610/* 5611 * Extended attribute directory vnode operations template 5612 * 5613 * This template is identical to the directory vnodes 5614 * operation template except for restricted operations: 5615 * VOP_MKDIR() 5616 * VOP_SYMLINK() 5617 * 5618 * Note that there are other restrictions embedded in: 5619 * zfs_create() - restrict type to VREG 5620 * zfs_link() - no links into/out of attribute space 5621 * zfs_rename() - no moves into/out of attribute space 5622 */ 5623vnodeops_t *zfs_xdvnodeops; 5624const fs_operation_def_t zfs_xdvnodeops_template[] = { 5625 VOPNAME_OPEN, { .vop_open = zfs_open }, 5626 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5627 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5628 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5629 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5630 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5631 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5632 VOPNAME_CREATE, { .vop_create = zfs_create }, 5633 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5634 VOPNAME_LINK, { .vop_link = zfs_link }, 5635 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5636 VOPNAME_MKDIR, { .error = zfs_inval }, 5637 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5638 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5639 VOPNAME_SYMLINK, { .error = zfs_inval }, 5640 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5641 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5642 VOPNAME_FID, { .vop_fid = zfs_fid }, 5643 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5644 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5645 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5646 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5647 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5648 NULL, NULL 5649}; 5650 5651/* 5652 * Error vnode operations template 5653 */ 5654vnodeops_t *zfs_evnodeops; 5655const fs_operation_def_t zfs_evnodeops_template[] = { 5656 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5657 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5658 NULL, NULL 5659}; 5660#endif /* sun */ 5661 5662static int 5663ioflags(int ioflags) 5664{ 5665 int flags = 0; 5666 5667 if (ioflags & IO_APPEND) 5668 flags |= FAPPEND; 5669 if (ioflags & IO_NDELAY) 5670 flags |= FNONBLOCK; 5671 if (ioflags & IO_SYNC) 5672 flags |= (FSYNC | FDSYNC | FRSYNC); 5673 5674 return (flags); 5675} 5676 5677static int 5678zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage) 5679{ 5680 znode_t *zp = VTOZ(vp); 5681 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5682 objset_t *os = zp->z_zfsvfs->z_os; 5683 vm_page_t mfirst, mlast, mreq; 5684 vm_object_t object; 5685 caddr_t va; 5686 struct sf_buf *sf; 5687 off_t startoff, endoff; 5688 int i, error; 5689 vm_pindex_t reqstart, reqend; 5690 int pcount, lsize, reqsize, size; 5691 5692 ZFS_ENTER(zfsvfs); 5693 ZFS_VERIFY_ZP(zp); 5694 5695 pcount = OFF_TO_IDX(round_page(count)); 5696 mreq = m[reqpage]; 5697 object = mreq->object; 5698 error = 0; 5699 5700 KASSERT(vp->v_object == object, ("mismatching object")); 5701 5702 if (pcount > 1 && zp->z_blksz > PAGESIZE) { 5703 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz); 5704 reqstart = OFF_TO_IDX(round_page(startoff)); 5705 if (reqstart < m[0]->pindex) 5706 reqstart = 0; 5707 else 5708 reqstart = reqstart - m[0]->pindex; 5709 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE, 5710 zp->z_blksz); 5711 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1; 5712 if (reqend > m[pcount - 1]->pindex) 5713 reqend = m[pcount - 1]->pindex; 5714 reqsize = reqend - m[reqstart]->pindex + 1; 5715 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize, 5716 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds")); 5717 } else { 5718 reqstart = reqpage; 5719 reqsize = 1; 5720 } 5721 mfirst = m[reqstart]; 5722 mlast = m[reqstart + reqsize - 1]; 5723 5724 zfs_vmobject_wlock(object); 5725 5726 for (i = 0; i < reqstart; i++) { 5727 vm_page_lock(m[i]); 5728 vm_page_free(m[i]); 5729 vm_page_unlock(m[i]); 5730 } 5731 for (i = reqstart + reqsize; i < pcount; i++) { 5732 vm_page_lock(m[i]); 5733 vm_page_free(m[i]); 5734 vm_page_unlock(m[i]); 5735 } 5736 5737 if (mreq->valid && reqsize == 1) { 5738 if (mreq->valid != VM_PAGE_BITS_ALL) 5739 vm_page_zero_invalid(mreq, TRUE); 5740 zfs_vmobject_wunlock(object); 5741 ZFS_EXIT(zfsvfs); 5742 return (zfs_vm_pagerret_ok); 5743 } 5744 5745 PCPU_INC(cnt.v_vnodein); 5746 PCPU_ADD(cnt.v_vnodepgsin, reqsize); 5747 5748 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) { 5749 for (i = reqstart; i < reqstart + reqsize; i++) { 5750 if (i != reqpage) { 5751 vm_page_lock(m[i]); 5752 vm_page_free(m[i]); 5753 vm_page_unlock(m[i]); 5754 } 5755 } 5756 zfs_vmobject_wunlock(object); 5757 ZFS_EXIT(zfsvfs); 5758 return (zfs_vm_pagerret_bad); 5759 } 5760 5761 lsize = PAGE_SIZE; 5762 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size) 5763 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex); 5764 5765 zfs_vmobject_wunlock(object); 5766 5767 for (i = reqstart; i < reqstart + reqsize; i++) { 5768 size = PAGE_SIZE; 5769 if (i == (reqstart + reqsize - 1)) 5770 size = lsize; 5771 va = zfs_map_page(m[i], &sf); 5772 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex), 5773 size, va, DMU_READ_PREFETCH); 5774 if (size != PAGE_SIZE) 5775 bzero(va + size, PAGE_SIZE - size); 5776 zfs_unmap_page(sf); 5777 if (error != 0) 5778 break; 5779 } 5780 5781 zfs_vmobject_wlock(object); 5782 5783 for (i = reqstart; i < reqstart + reqsize; i++) { 5784 if (!error) 5785 m[i]->valid = VM_PAGE_BITS_ALL; 5786 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i])); 5787 if (i != reqpage) 5788 vm_page_readahead_finish(m[i]); 5789 } 5790 5791 zfs_vmobject_wunlock(object); 5792 5793 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 5794 ZFS_EXIT(zfsvfs); 5795 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok); 5796} 5797 5798static int 5799zfs_freebsd_getpages(ap) 5800 struct vop_getpages_args /* { 5801 struct vnode *a_vp; 5802 vm_page_t *a_m; 5803 int a_count; 5804 int a_reqpage; 5805 vm_ooffset_t a_offset; 5806 } */ *ap; 5807{ 5808 5809 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage)); 5810} 5811 5812static int 5813zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags, 5814 int *rtvals) 5815{ 5816 znode_t *zp = VTOZ(vp); 5817 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5818 rl_t *rl; 5819 dmu_tx_t *tx; 5820 struct sf_buf *sf; 5821 vm_object_t object; 5822 vm_page_t m; 5823 caddr_t va; 5824 size_t tocopy; 5825 size_t lo_len; 5826 vm_ooffset_t lo_off; 5827 vm_ooffset_t off; 5828 uint_t blksz; 5829 int ncount; 5830 int pcount; 5831 int err; 5832 int i; 5833 5834 ZFS_ENTER(zfsvfs); 5835 ZFS_VERIFY_ZP(zp); 5836 5837 object = vp->v_object; 5838 pcount = btoc(len); 5839 ncount = pcount; 5840 5841 KASSERT(ma[0]->object == object, ("mismatching object")); 5842 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length")); 5843 5844 for (i = 0; i < pcount; i++) 5845 rtvals[i] = zfs_vm_pagerret_error; 5846 5847 off = IDX_TO_OFF(ma[0]->pindex); 5848 blksz = zp->z_blksz; 5849 lo_off = rounddown(off, blksz); 5850 lo_len = roundup(len + (off - lo_off), blksz); 5851 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER); 5852 5853 zfs_vmobject_wlock(object); 5854 if (len + off > object->un_pager.vnp.vnp_size) { 5855 if (object->un_pager.vnp.vnp_size > off) { 5856 int pgoff; 5857 5858 len = object->un_pager.vnp.vnp_size - off; 5859 ncount = btoc(len); 5860 if ((pgoff = (int)len & PAGE_MASK) != 0) { 5861 /* 5862 * If the object is locked and the following 5863 * conditions hold, then the page's dirty 5864 * field cannot be concurrently changed by a 5865 * pmap operation. 5866 */ 5867 m = ma[ncount - 1]; 5868 vm_page_assert_sbusied(m); 5869 KASSERT(!pmap_page_is_write_mapped(m), 5870 ("zfs_putpages: page %p is not read-only", m)); 5871 vm_page_clear_dirty(m, pgoff, PAGE_SIZE - 5872 pgoff); 5873 } 5874 } else { 5875 len = 0; 5876 ncount = 0; 5877 } 5878 if (ncount < pcount) { 5879 for (i = ncount; i < pcount; i++) { 5880 rtvals[i] = zfs_vm_pagerret_bad; 5881 } 5882 } 5883 } 5884 zfs_vmobject_wunlock(object); 5885 5886 if (ncount == 0) 5887 goto out; 5888 5889 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 5890 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 5891 goto out; 5892 } 5893 5894top: 5895 tx = dmu_tx_create(zfsvfs->z_os); 5896 dmu_tx_hold_write(tx, zp->z_id, off, len); 5897 5898 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 5899 zfs_sa_upgrade_txholds(tx, zp); 5900 err = dmu_tx_assign(tx, TXG_NOWAIT); 5901 if (err != 0) { 5902 if (err == ERESTART) { 5903 dmu_tx_wait(tx); 5904 dmu_tx_abort(tx); 5905 goto top; 5906 } 5907 dmu_tx_abort(tx); 5908 goto out; 5909 } 5910 5911 if (zp->z_blksz < PAGE_SIZE) { 5912 i = 0; 5913 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) { 5914 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len; 5915 va = zfs_map_page(ma[i], &sf); 5916 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx); 5917 zfs_unmap_page(sf); 5918 } 5919 } else { 5920 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx); 5921 } 5922 5923 if (err == 0) { 5924 uint64_t mtime[2], ctime[2]; 5925 sa_bulk_attr_t bulk[3]; 5926 int count = 0; 5927 5928 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 5929 &mtime, 16); 5930 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 5931 &ctime, 16); 5932 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 5933 &zp->z_pflags, 8); 5934 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 5935 B_TRUE); 5936 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 5937 5938 zfs_vmobject_wlock(object); 5939 for (i = 0; i < ncount; i++) { 5940 rtvals[i] = zfs_vm_pagerret_ok; 5941 vm_page_undirty(ma[i]); 5942 } 5943 zfs_vmobject_wunlock(object); 5944 PCPU_INC(cnt.v_vnodeout); 5945 PCPU_ADD(cnt.v_vnodepgsout, ncount); 5946 } 5947 dmu_tx_commit(tx); 5948 5949out: 5950 zfs_range_unlock(rl); 5951 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 || 5952 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 5953 zil_commit(zfsvfs->z_log, zp->z_id); 5954 ZFS_EXIT(zfsvfs); 5955 return (rtvals[0]); 5956} 5957 5958int 5959zfs_freebsd_putpages(ap) 5960 struct vop_putpages_args /* { 5961 struct vnode *a_vp; 5962 vm_page_t *a_m; 5963 int a_count; 5964 int a_sync; 5965 int *a_rtvals; 5966 vm_ooffset_t a_offset; 5967 } */ *ap; 5968{ 5969 5970 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync, 5971 ap->a_rtvals)); 5972} 5973 5974static int 5975zfs_freebsd_bmap(ap) 5976 struct vop_bmap_args /* { 5977 struct vnode *a_vp; 5978 daddr_t a_bn; 5979 struct bufobj **a_bop; 5980 daddr_t *a_bnp; 5981 int *a_runp; 5982 int *a_runb; 5983 } */ *ap; 5984{ 5985 5986 if (ap->a_bop != NULL) 5987 *ap->a_bop = &ap->a_vp->v_bufobj; 5988 if (ap->a_bnp != NULL) 5989 *ap->a_bnp = ap->a_bn; 5990 if (ap->a_runp != NULL) 5991 *ap->a_runp = 0; 5992 if (ap->a_runb != NULL) 5993 *ap->a_runb = 0; 5994 5995 return (0); 5996} 5997 5998static int 5999zfs_freebsd_open(ap) 6000 struct vop_open_args /* { 6001 struct vnode *a_vp; 6002 int a_mode; 6003 struct ucred *a_cred; 6004 struct thread *a_td; 6005 } */ *ap; 6006{ 6007 vnode_t *vp = ap->a_vp; 6008 znode_t *zp = VTOZ(vp); 6009 int error; 6010 6011 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL); 6012 if (error == 0) 6013 vnode_create_vobject(vp, zp->z_size, ap->a_td); 6014 return (error); 6015} 6016 6017static int 6018zfs_freebsd_close(ap) 6019 struct vop_close_args /* { 6020 struct vnode *a_vp; 6021 int a_fflag; 6022 struct ucred *a_cred; 6023 struct thread *a_td; 6024 } */ *ap; 6025{ 6026 6027 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL)); 6028} 6029 6030static int 6031zfs_freebsd_ioctl(ap) 6032 struct vop_ioctl_args /* { 6033 struct vnode *a_vp; 6034 u_long a_command; 6035 caddr_t a_data; 6036 int a_fflag; 6037 struct ucred *cred; 6038 struct thread *td; 6039 } */ *ap; 6040{ 6041 6042 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 6043 ap->a_fflag, ap->a_cred, NULL, NULL)); 6044} 6045 6046static int 6047zfs_freebsd_read(ap) 6048 struct vop_read_args /* { 6049 struct vnode *a_vp; 6050 struct uio *a_uio; 6051 int a_ioflag; 6052 struct ucred *a_cred; 6053 } */ *ap; 6054{ 6055 6056 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 6057 ap->a_cred, NULL)); 6058} 6059 6060static int 6061zfs_freebsd_write(ap) 6062 struct vop_write_args /* { 6063 struct vnode *a_vp; 6064 struct uio *a_uio; 6065 int a_ioflag; 6066 struct ucred *a_cred; 6067 } */ *ap; 6068{ 6069 6070 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 6071 ap->a_cred, NULL)); 6072} 6073 6074static int 6075zfs_freebsd_access(ap) 6076 struct vop_access_args /* { 6077 struct vnode *a_vp; 6078 accmode_t a_accmode; 6079 struct ucred *a_cred; 6080 struct thread *a_td; 6081 } */ *ap; 6082{ 6083 vnode_t *vp = ap->a_vp; 6084 znode_t *zp = VTOZ(vp); 6085 accmode_t accmode; 6086 int error = 0; 6087 6088 /* 6089 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 6090 */ 6091 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 6092 if (accmode != 0) 6093 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL); 6094 6095 /* 6096 * VADMIN has to be handled by vaccess(). 6097 */ 6098 if (error == 0) { 6099 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 6100 if (accmode != 0) { 6101 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 6102 zp->z_gid, accmode, ap->a_cred, NULL); 6103 } 6104 } 6105 6106 /* 6107 * For VEXEC, ensure that at least one execute bit is set for 6108 * non-directories. 6109 */ 6110 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 6111 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 6112 error = EACCES; 6113 } 6114 6115 return (error); 6116} 6117 6118static int 6119zfs_freebsd_lookup(ap) 6120 struct vop_lookup_args /* { 6121 struct vnode *a_dvp; 6122 struct vnode **a_vpp; 6123 struct componentname *a_cnp; 6124 } */ *ap; 6125{ 6126 struct componentname *cnp = ap->a_cnp; 6127 char nm[NAME_MAX + 1]; 6128 6129 ASSERT(cnp->cn_namelen < sizeof(nm)); 6130 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 6131 6132 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 6133 cnp->cn_cred, cnp->cn_thread, 0)); 6134} 6135 6136static int 6137zfs_freebsd_create(ap) 6138 struct vop_create_args /* { 6139 struct vnode *a_dvp; 6140 struct vnode **a_vpp; 6141 struct componentname *a_cnp; 6142 struct vattr *a_vap; 6143 } */ *ap; 6144{ 6145 struct componentname *cnp = ap->a_cnp; 6146 vattr_t *vap = ap->a_vap; 6147 int mode; 6148 6149 ASSERT(cnp->cn_flags & SAVENAME); 6150 6151 vattr_init_mask(vap); 6152 mode = vap->va_mode & ALLPERMS; 6153 6154 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 6155 ap->a_vpp, cnp->cn_cred, cnp->cn_thread)); 6156} 6157 6158static int 6159zfs_freebsd_remove(ap) 6160 struct vop_remove_args /* { 6161 struct vnode *a_dvp; 6162 struct vnode *a_vp; 6163 struct componentname *a_cnp; 6164 } */ *ap; 6165{ 6166 6167 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6168 6169 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 6170 ap->a_cnp->cn_cred, NULL, 0)); 6171} 6172 6173static int 6174zfs_freebsd_mkdir(ap) 6175 struct vop_mkdir_args /* { 6176 struct vnode *a_dvp; 6177 struct vnode **a_vpp; 6178 struct componentname *a_cnp; 6179 struct vattr *a_vap; 6180 } */ *ap; 6181{ 6182 vattr_t *vap = ap->a_vap; 6183 6184 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6185 6186 vattr_init_mask(vap); 6187 6188 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 6189 ap->a_cnp->cn_cred, NULL, 0, NULL)); 6190} 6191 6192static int 6193zfs_freebsd_rmdir(ap) 6194 struct vop_rmdir_args /* { 6195 struct vnode *a_dvp; 6196 struct vnode *a_vp; 6197 struct componentname *a_cnp; 6198 } */ *ap; 6199{ 6200 struct componentname *cnp = ap->a_cnp; 6201 6202 ASSERT(cnp->cn_flags & SAVENAME); 6203 6204 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0)); 6205} 6206 6207static int 6208zfs_freebsd_readdir(ap) 6209 struct vop_readdir_args /* { 6210 struct vnode *a_vp; 6211 struct uio *a_uio; 6212 struct ucred *a_cred; 6213 int *a_eofflag; 6214 int *a_ncookies; 6215 u_long **a_cookies; 6216 } */ *ap; 6217{ 6218 6219 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 6220 ap->a_ncookies, ap->a_cookies)); 6221} 6222 6223static int 6224zfs_freebsd_fsync(ap) 6225 struct vop_fsync_args /* { 6226 struct vnode *a_vp; 6227 int a_waitfor; 6228 struct thread *a_td; 6229 } */ *ap; 6230{ 6231 6232 vop_stdfsync(ap); 6233 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL)); 6234} 6235 6236static int 6237zfs_freebsd_getattr(ap) 6238 struct vop_getattr_args /* { 6239 struct vnode *a_vp; 6240 struct vattr *a_vap; 6241 struct ucred *a_cred; 6242 } */ *ap; 6243{ 6244 vattr_t *vap = ap->a_vap; 6245 xvattr_t xvap; 6246 u_long fflags = 0; 6247 int error; 6248 6249 xva_init(&xvap); 6250 xvap.xva_vattr = *vap; 6251 xvap.xva_vattr.va_mask |= AT_XVATTR; 6252 6253 /* Convert chflags into ZFS-type flags. */ 6254 /* XXX: what about SF_SETTABLE?. */ 6255 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 6256 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 6257 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 6258 XVA_SET_REQ(&xvap, XAT_NODUMP); 6259 XVA_SET_REQ(&xvap, XAT_READONLY); 6260 XVA_SET_REQ(&xvap, XAT_ARCHIVE); 6261 XVA_SET_REQ(&xvap, XAT_SYSTEM); 6262 XVA_SET_REQ(&xvap, XAT_HIDDEN); 6263 XVA_SET_REQ(&xvap, XAT_REPARSE); 6264 XVA_SET_REQ(&xvap, XAT_OFFLINE); 6265 XVA_SET_REQ(&xvap, XAT_SPARSE); 6266 6267 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL); 6268 if (error != 0) 6269 return (error); 6270 6271 /* Convert ZFS xattr into chflags. */ 6272#define FLAG_CHECK(fflag, xflag, xfield) do { \ 6273 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 6274 fflags |= (fflag); \ 6275} while (0) 6276 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 6277 xvap.xva_xoptattrs.xoa_immutable); 6278 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 6279 xvap.xva_xoptattrs.xoa_appendonly); 6280 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 6281 xvap.xva_xoptattrs.xoa_nounlink); 6282 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE, 6283 xvap.xva_xoptattrs.xoa_archive); 6284 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 6285 xvap.xva_xoptattrs.xoa_nodump); 6286 FLAG_CHECK(UF_READONLY, XAT_READONLY, 6287 xvap.xva_xoptattrs.xoa_readonly); 6288 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM, 6289 xvap.xva_xoptattrs.xoa_system); 6290 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN, 6291 xvap.xva_xoptattrs.xoa_hidden); 6292 FLAG_CHECK(UF_REPARSE, XAT_REPARSE, 6293 xvap.xva_xoptattrs.xoa_reparse); 6294 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE, 6295 xvap.xva_xoptattrs.xoa_offline); 6296 FLAG_CHECK(UF_SPARSE, XAT_SPARSE, 6297 xvap.xva_xoptattrs.xoa_sparse); 6298 6299#undef FLAG_CHECK 6300 *vap = xvap.xva_vattr; 6301 vap->va_flags = fflags; 6302 return (0); 6303} 6304 6305static int 6306zfs_freebsd_setattr(ap) 6307 struct vop_setattr_args /* { 6308 struct vnode *a_vp; 6309 struct vattr *a_vap; 6310 struct ucred *a_cred; 6311 } */ *ap; 6312{ 6313 vnode_t *vp = ap->a_vp; 6314 vattr_t *vap = ap->a_vap; 6315 cred_t *cred = ap->a_cred; 6316 xvattr_t xvap; 6317 u_long fflags; 6318 uint64_t zflags; 6319 6320 vattr_init_mask(vap); 6321 vap->va_mask &= ~AT_NOSET; 6322 6323 xva_init(&xvap); 6324 xvap.xva_vattr = *vap; 6325 6326 zflags = VTOZ(vp)->z_pflags; 6327 6328 if (vap->va_flags != VNOVAL) { 6329 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 6330 int error; 6331 6332 if (zfsvfs->z_use_fuids == B_FALSE) 6333 return (EOPNOTSUPP); 6334 6335 fflags = vap->va_flags; 6336 /* 6337 * XXX KDM 6338 * We need to figure out whether it makes sense to allow 6339 * UF_REPARSE through, since we don't really have other 6340 * facilities to handle reparse points and zfs_setattr() 6341 * doesn't currently allow setting that attribute anyway. 6342 */ 6343 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE| 6344 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE| 6345 UF_OFFLINE|UF_SPARSE)) != 0) 6346 return (EOPNOTSUPP); 6347 /* 6348 * Unprivileged processes are not permitted to unset system 6349 * flags, or modify flags if any system flags are set. 6350 * Privileged non-jail processes may not modify system flags 6351 * if securelevel > 0 and any existing system flags are set. 6352 * Privileged jail processes behave like privileged non-jail 6353 * processes if the security.jail.chflags_allowed sysctl is 6354 * is non-zero; otherwise, they behave like unprivileged 6355 * processes. 6356 */ 6357 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 6358 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) { 6359 if (zflags & 6360 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6361 error = securelevel_gt(cred, 0); 6362 if (error != 0) 6363 return (error); 6364 } 6365 } else { 6366 /* 6367 * Callers may only modify the file flags on objects they 6368 * have VADMIN rights for. 6369 */ 6370 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0) 6371 return (error); 6372 if (zflags & 6373 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6374 return (EPERM); 6375 } 6376 if (fflags & 6377 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 6378 return (EPERM); 6379 } 6380 } 6381 6382#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 6383 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 6384 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 6385 XVA_SET_REQ(&xvap, (xflag)); \ 6386 (xfield) = ((fflags & (fflag)) != 0); \ 6387 } \ 6388} while (0) 6389 /* Convert chflags into ZFS-type flags. */ 6390 /* XXX: what about SF_SETTABLE?. */ 6391 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 6392 xvap.xva_xoptattrs.xoa_immutable); 6393 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 6394 xvap.xva_xoptattrs.xoa_appendonly); 6395 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 6396 xvap.xva_xoptattrs.xoa_nounlink); 6397 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE, 6398 xvap.xva_xoptattrs.xoa_archive); 6399 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 6400 xvap.xva_xoptattrs.xoa_nodump); 6401 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY, 6402 xvap.xva_xoptattrs.xoa_readonly); 6403 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM, 6404 xvap.xva_xoptattrs.xoa_system); 6405 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN, 6406 xvap.xva_xoptattrs.xoa_hidden); 6407 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE, 6408 xvap.xva_xoptattrs.xoa_hidden); 6409 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE, 6410 xvap.xva_xoptattrs.xoa_offline); 6411 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE, 6412 xvap.xva_xoptattrs.xoa_sparse); 6413#undef FLAG_CHANGE 6414 } 6415 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL)); 6416} 6417 6418static int 6419zfs_freebsd_rename(ap) 6420 struct vop_rename_args /* { 6421 struct vnode *a_fdvp; 6422 struct vnode *a_fvp; 6423 struct componentname *a_fcnp; 6424 struct vnode *a_tdvp; 6425 struct vnode *a_tvp; 6426 struct componentname *a_tcnp; 6427 } */ *ap; 6428{ 6429 vnode_t *fdvp = ap->a_fdvp; 6430 vnode_t *fvp = ap->a_fvp; 6431 vnode_t *tdvp = ap->a_tdvp; 6432 vnode_t *tvp = ap->a_tvp; 6433 int error; 6434 6435 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 6436 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 6437 6438 /* 6439 * Check for cross-device rename. 6440 */ 6441 if ((fdvp->v_mount != tdvp->v_mount) || 6442 (tvp && (fdvp->v_mount != tvp->v_mount))) 6443 error = EXDEV; 6444 else 6445 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 6446 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0); 6447 if (tdvp == tvp) 6448 VN_RELE(tdvp); 6449 else 6450 VN_URELE(tdvp); 6451 if (tvp) 6452 VN_URELE(tvp); 6453 VN_RELE(fdvp); 6454 VN_RELE(fvp); 6455 6456 return (error); 6457} 6458 6459static int 6460zfs_freebsd_symlink(ap) 6461 struct vop_symlink_args /* { 6462 struct vnode *a_dvp; 6463 struct vnode **a_vpp; 6464 struct componentname *a_cnp; 6465 struct vattr *a_vap; 6466 char *a_target; 6467 } */ *ap; 6468{ 6469 struct componentname *cnp = ap->a_cnp; 6470 vattr_t *vap = ap->a_vap; 6471 6472 ASSERT(cnp->cn_flags & SAVENAME); 6473 6474 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 6475 vattr_init_mask(vap); 6476 6477 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 6478 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 6479} 6480 6481static int 6482zfs_freebsd_readlink(ap) 6483 struct vop_readlink_args /* { 6484 struct vnode *a_vp; 6485 struct uio *a_uio; 6486 struct ucred *a_cred; 6487 } */ *ap; 6488{ 6489 6490 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL)); 6491} 6492 6493static int 6494zfs_freebsd_link(ap) 6495 struct vop_link_args /* { 6496 struct vnode *a_tdvp; 6497 struct vnode *a_vp; 6498 struct componentname *a_cnp; 6499 } */ *ap; 6500{ 6501 struct componentname *cnp = ap->a_cnp; 6502 vnode_t *vp = ap->a_vp; 6503 vnode_t *tdvp = ap->a_tdvp; 6504 6505 if (tdvp->v_mount != vp->v_mount) 6506 return (EXDEV); 6507 6508 ASSERT(cnp->cn_flags & SAVENAME); 6509 6510 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0)); 6511} 6512 6513static int 6514zfs_freebsd_inactive(ap) 6515 struct vop_inactive_args /* { 6516 struct vnode *a_vp; 6517 struct thread *a_td; 6518 } */ *ap; 6519{ 6520 vnode_t *vp = ap->a_vp; 6521 6522 zfs_inactive(vp, ap->a_td->td_ucred, NULL); 6523 return (0); 6524} 6525 6526static int 6527zfs_freebsd_reclaim(ap) 6528 struct vop_reclaim_args /* { 6529 struct vnode *a_vp; 6530 struct thread *a_td; 6531 } */ *ap; 6532{ 6533 vnode_t *vp = ap->a_vp; 6534 znode_t *zp = VTOZ(vp); 6535 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6536 6537 ASSERT(zp != NULL); 6538 6539 /* Destroy the vm object and flush associated pages. */ 6540 vnode_destroy_vobject(vp); 6541 6542 /* 6543 * z_teardown_inactive_lock protects from a race with 6544 * zfs_znode_dmu_fini in zfsvfs_teardown during 6545 * force unmount. 6546 */ 6547 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 6548 if (zp->z_sa_hdl == NULL) 6549 zfs_znode_free(zp); 6550 else 6551 zfs_zinactive(zp); 6552 rw_exit(&zfsvfs->z_teardown_inactive_lock); 6553 6554 vp->v_data = NULL; 6555 return (0); 6556} 6557 6558static int 6559zfs_freebsd_fid(ap) 6560 struct vop_fid_args /* { 6561 struct vnode *a_vp; 6562 struct fid *a_fid; 6563 } */ *ap; 6564{ 6565 6566 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 6567} 6568 6569static int 6570zfs_freebsd_pathconf(ap) 6571 struct vop_pathconf_args /* { 6572 struct vnode *a_vp; 6573 int a_name; 6574 register_t *a_retval; 6575 } */ *ap; 6576{ 6577 ulong_t val; 6578 int error; 6579 6580 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); 6581 if (error == 0) 6582 *ap->a_retval = val; 6583 else if (error == EOPNOTSUPP) 6584 error = vop_stdpathconf(ap); 6585 return (error); 6586} 6587 6588static int 6589zfs_freebsd_fifo_pathconf(ap) 6590 struct vop_pathconf_args /* { 6591 struct vnode *a_vp; 6592 int a_name; 6593 register_t *a_retval; 6594 } */ *ap; 6595{ 6596 6597 switch (ap->a_name) { 6598 case _PC_ACL_EXTENDED: 6599 case _PC_ACL_NFS4: 6600 case _PC_ACL_PATH_MAX: 6601 case _PC_MAC_PRESENT: 6602 return (zfs_freebsd_pathconf(ap)); 6603 default: 6604 return (fifo_specops.vop_pathconf(ap)); 6605 } 6606} 6607 6608/* 6609 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 6610 * extended attribute name: 6611 * 6612 * NAMESPACE PREFIX 6613 * system freebsd:system: 6614 * user (none, can be used to access ZFS fsattr(5) attributes 6615 * created on Solaris) 6616 */ 6617static int 6618zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 6619 size_t size) 6620{ 6621 const char *namespace, *prefix, *suffix; 6622 6623 /* We don't allow '/' character in attribute name. */ 6624 if (strchr(name, '/') != NULL) 6625 return (EINVAL); 6626 /* We don't allow attribute names that start with "freebsd:" string. */ 6627 if (strncmp(name, "freebsd:", 8) == 0) 6628 return (EINVAL); 6629 6630 bzero(attrname, size); 6631 6632 switch (attrnamespace) { 6633 case EXTATTR_NAMESPACE_USER: 6634#if 0 6635 prefix = "freebsd:"; 6636 namespace = EXTATTR_NAMESPACE_USER_STRING; 6637 suffix = ":"; 6638#else 6639 /* 6640 * This is the default namespace by which we can access all 6641 * attributes created on Solaris. 6642 */ 6643 prefix = namespace = suffix = ""; 6644#endif 6645 break; 6646 case EXTATTR_NAMESPACE_SYSTEM: 6647 prefix = "freebsd:"; 6648 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 6649 suffix = ":"; 6650 break; 6651 case EXTATTR_NAMESPACE_EMPTY: 6652 default: 6653 return (EINVAL); 6654 } 6655 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 6656 name) >= size) { 6657 return (ENAMETOOLONG); 6658 } 6659 return (0); 6660} 6661 6662/* 6663 * Vnode operating to retrieve a named extended attribute. 6664 */ 6665static int 6666zfs_getextattr(struct vop_getextattr_args *ap) 6667/* 6668vop_getextattr { 6669 IN struct vnode *a_vp; 6670 IN int a_attrnamespace; 6671 IN const char *a_name; 6672 INOUT struct uio *a_uio; 6673 OUT size_t *a_size; 6674 IN struct ucred *a_cred; 6675 IN struct thread *a_td; 6676}; 6677*/ 6678{ 6679 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6680 struct thread *td = ap->a_td; 6681 struct nameidata nd; 6682 char attrname[255]; 6683 struct vattr va; 6684 vnode_t *xvp = NULL, *vp; 6685 int error, flags; 6686 6687 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6688 ap->a_cred, ap->a_td, VREAD); 6689 if (error != 0) 6690 return (error); 6691 6692 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6693 sizeof(attrname)); 6694 if (error != 0) 6695 return (error); 6696 6697 ZFS_ENTER(zfsvfs); 6698 6699 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6700 LOOKUP_XATTR); 6701 if (error != 0) { 6702 ZFS_EXIT(zfsvfs); 6703 return (error); 6704 } 6705 6706 flags = FREAD; 6707 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6708 xvp, td); 6709 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL); 6710 vp = nd.ni_vp; 6711 NDFREE(&nd, NDF_ONLY_PNBUF); 6712 if (error != 0) { 6713 ZFS_EXIT(zfsvfs); 6714 if (error == ENOENT) 6715 error = ENOATTR; 6716 return (error); 6717 } 6718 6719 if (ap->a_size != NULL) { 6720 error = VOP_GETATTR(vp, &va, ap->a_cred); 6721 if (error == 0) 6722 *ap->a_size = (size_t)va.va_size; 6723 } else if (ap->a_uio != NULL) 6724 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6725 6726 VOP_UNLOCK(vp, 0); 6727 vn_close(vp, flags, ap->a_cred, td); 6728 ZFS_EXIT(zfsvfs); 6729 6730 return (error); 6731} 6732 6733/* 6734 * Vnode operation to remove a named attribute. 6735 */ 6736int 6737zfs_deleteextattr(struct vop_deleteextattr_args *ap) 6738/* 6739vop_deleteextattr { 6740 IN struct vnode *a_vp; 6741 IN int a_attrnamespace; 6742 IN const char *a_name; 6743 IN struct ucred *a_cred; 6744 IN struct thread *a_td; 6745}; 6746*/ 6747{ 6748 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6749 struct thread *td = ap->a_td; 6750 struct nameidata nd; 6751 char attrname[255]; 6752 struct vattr va; 6753 vnode_t *xvp = NULL, *vp; 6754 int error, flags; 6755 6756 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6757 ap->a_cred, ap->a_td, VWRITE); 6758 if (error != 0) 6759 return (error); 6760 6761 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6762 sizeof(attrname)); 6763 if (error != 0) 6764 return (error); 6765 6766 ZFS_ENTER(zfsvfs); 6767 6768 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6769 LOOKUP_XATTR); 6770 if (error != 0) { 6771 ZFS_EXIT(zfsvfs); 6772 return (error); 6773 } 6774 6775 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 6776 UIO_SYSSPACE, attrname, xvp, td); 6777 error = namei(&nd); 6778 vp = nd.ni_vp; 6779 if (error != 0) { 6780 ZFS_EXIT(zfsvfs); 6781 NDFREE(&nd, NDF_ONLY_PNBUF); 6782 if (error == ENOENT) 6783 error = ENOATTR; 6784 return (error); 6785 } 6786 6787 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 6788 NDFREE(&nd, NDF_ONLY_PNBUF); 6789 6790 vput(nd.ni_dvp); 6791 if (vp == nd.ni_dvp) 6792 vrele(vp); 6793 else 6794 vput(vp); 6795 ZFS_EXIT(zfsvfs); 6796 6797 return (error); 6798} 6799 6800/* 6801 * Vnode operation to set a named attribute. 6802 */ 6803static int 6804zfs_setextattr(struct vop_setextattr_args *ap) 6805/* 6806vop_setextattr { 6807 IN struct vnode *a_vp; 6808 IN int a_attrnamespace; 6809 IN const char *a_name; 6810 INOUT struct uio *a_uio; 6811 IN struct ucred *a_cred; 6812 IN struct thread *a_td; 6813}; 6814*/ 6815{ 6816 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6817 struct thread *td = ap->a_td; 6818 struct nameidata nd; 6819 char attrname[255]; 6820 struct vattr va; 6821 vnode_t *xvp = NULL, *vp; 6822 int error, flags; 6823 6824 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6825 ap->a_cred, ap->a_td, VWRITE); 6826 if (error != 0) 6827 return (error); 6828 6829 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6830 sizeof(attrname)); 6831 if (error != 0) 6832 return (error); 6833 6834 ZFS_ENTER(zfsvfs); 6835 6836 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6837 LOOKUP_XATTR | CREATE_XATTR_DIR); 6838 if (error != 0) { 6839 ZFS_EXIT(zfsvfs); 6840 return (error); 6841 } 6842 6843 flags = FFLAGS(O_WRONLY | O_CREAT); 6844 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6845 xvp, td); 6846 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL); 6847 vp = nd.ni_vp; 6848 NDFREE(&nd, NDF_ONLY_PNBUF); 6849 if (error != 0) { 6850 ZFS_EXIT(zfsvfs); 6851 return (error); 6852 } 6853 6854 VATTR_NULL(&va); 6855 va.va_size = 0; 6856 error = VOP_SETATTR(vp, &va, ap->a_cred); 6857 if (error == 0) 6858 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6859 6860 VOP_UNLOCK(vp, 0); 6861 vn_close(vp, flags, ap->a_cred, td); 6862 ZFS_EXIT(zfsvfs); 6863 6864 return (error); 6865} 6866 6867/* 6868 * Vnode operation to retrieve extended attributes on a vnode. 6869 */ 6870static int 6871zfs_listextattr(struct vop_listextattr_args *ap) 6872/* 6873vop_listextattr { 6874 IN struct vnode *a_vp; 6875 IN int a_attrnamespace; 6876 INOUT struct uio *a_uio; 6877 OUT size_t *a_size; 6878 IN struct ucred *a_cred; 6879 IN struct thread *a_td; 6880}; 6881*/ 6882{ 6883 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6884 struct thread *td = ap->a_td; 6885 struct nameidata nd; 6886 char attrprefix[16]; 6887 u_char dirbuf[sizeof(struct dirent)]; 6888 struct dirent *dp; 6889 struct iovec aiov; 6890 struct uio auio, *uio = ap->a_uio; 6891 size_t *sizep = ap->a_size; 6892 size_t plen; 6893 vnode_t *xvp = NULL, *vp; 6894 int done, error, eof, pos; 6895 6896 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6897 ap->a_cred, ap->a_td, VREAD); 6898 if (error != 0) 6899 return (error); 6900 6901 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 6902 sizeof(attrprefix)); 6903 if (error != 0) 6904 return (error); 6905 plen = strlen(attrprefix); 6906 6907 ZFS_ENTER(zfsvfs); 6908 6909 if (sizep != NULL) 6910 *sizep = 0; 6911 6912 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6913 LOOKUP_XATTR); 6914 if (error != 0) { 6915 ZFS_EXIT(zfsvfs); 6916 /* 6917 * ENOATTR means that the EA directory does not yet exist, 6918 * i.e. there are no extended attributes there. 6919 */ 6920 if (error == ENOATTR) 6921 error = 0; 6922 return (error); 6923 } 6924 6925 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 6926 UIO_SYSSPACE, ".", xvp, td); 6927 error = namei(&nd); 6928 vp = nd.ni_vp; 6929 NDFREE(&nd, NDF_ONLY_PNBUF); 6930 if (error != 0) { 6931 ZFS_EXIT(zfsvfs); 6932 return (error); 6933 } 6934 6935 auio.uio_iov = &aiov; 6936 auio.uio_iovcnt = 1; 6937 auio.uio_segflg = UIO_SYSSPACE; 6938 auio.uio_td = td; 6939 auio.uio_rw = UIO_READ; 6940 auio.uio_offset = 0; 6941 6942 do { 6943 u_char nlen; 6944 6945 aiov.iov_base = (void *)dirbuf; 6946 aiov.iov_len = sizeof(dirbuf); 6947 auio.uio_resid = sizeof(dirbuf); 6948 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 6949 done = sizeof(dirbuf) - auio.uio_resid; 6950 if (error != 0) 6951 break; 6952 for (pos = 0; pos < done;) { 6953 dp = (struct dirent *)(dirbuf + pos); 6954 pos += dp->d_reclen; 6955 /* 6956 * XXX: Temporarily we also accept DT_UNKNOWN, as this 6957 * is what we get when attribute was created on Solaris. 6958 */ 6959 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 6960 continue; 6961 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0) 6962 continue; 6963 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 6964 continue; 6965 nlen = dp->d_namlen - plen; 6966 if (sizep != NULL) 6967 *sizep += 1 + nlen; 6968 else if (uio != NULL) { 6969 /* 6970 * Format of extattr name entry is one byte for 6971 * length and the rest for name. 6972 */ 6973 error = uiomove(&nlen, 1, uio->uio_rw, uio); 6974 if (error == 0) { 6975 error = uiomove(dp->d_name + plen, nlen, 6976 uio->uio_rw, uio); 6977 } 6978 if (error != 0) 6979 break; 6980 } 6981 } 6982 } while (!eof && error == 0); 6983 6984 vput(vp); 6985 ZFS_EXIT(zfsvfs); 6986 6987 return (error); 6988} 6989 6990int 6991zfs_freebsd_getacl(ap) 6992 struct vop_getacl_args /* { 6993 struct vnode *vp; 6994 acl_type_t type; 6995 struct acl *aclp; 6996 struct ucred *cred; 6997 struct thread *td; 6998 } */ *ap; 6999{ 7000 int error; 7001 vsecattr_t vsecattr; 7002 7003 if (ap->a_type != ACL_TYPE_NFS4) 7004 return (EINVAL); 7005 7006 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 7007 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL)) 7008 return (error); 7009 7010 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt); 7011 if (vsecattr.vsa_aclentp != NULL) 7012 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 7013 7014 return (error); 7015} 7016 7017int 7018zfs_freebsd_setacl(ap) 7019 struct vop_setacl_args /* { 7020 struct vnode *vp; 7021 acl_type_t type; 7022 struct acl *aclp; 7023 struct ucred *cred; 7024 struct thread *td; 7025 } */ *ap; 7026{ 7027 int error; 7028 vsecattr_t vsecattr; 7029 int aclbsize; /* size of acl list in bytes */ 7030 aclent_t *aaclp; 7031 7032 if (ap->a_type != ACL_TYPE_NFS4) 7033 return (EINVAL); 7034 7035 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 7036 return (EINVAL); 7037 7038 /* 7039 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 7040 * splitting every entry into two and appending "canonical six" 7041 * entries at the end. Don't allow for setting an ACL that would 7042 * cause chmod(2) to run out of ACL entries. 7043 */ 7044 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 7045 return (ENOSPC); 7046 7047 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 7048 if (error != 0) 7049 return (error); 7050 7051 vsecattr.vsa_mask = VSA_ACE; 7052 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t); 7053 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 7054 aaclp = vsecattr.vsa_aclentp; 7055 vsecattr.vsa_aclentsz = aclbsize; 7056 7057 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 7058 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL); 7059 kmem_free(aaclp, aclbsize); 7060 7061 return (error); 7062} 7063 7064int 7065zfs_freebsd_aclcheck(ap) 7066 struct vop_aclcheck_args /* { 7067 struct vnode *vp; 7068 acl_type_t type; 7069 struct acl *aclp; 7070 struct ucred *cred; 7071 struct thread *td; 7072 } */ *ap; 7073{ 7074 7075 return (EOPNOTSUPP); 7076} 7077 7078struct vop_vector zfs_vnodeops; 7079struct vop_vector zfs_fifoops; 7080struct vop_vector zfs_shareops; 7081 7082struct vop_vector zfs_vnodeops = { 7083 .vop_default = &default_vnodeops, 7084 .vop_inactive = zfs_freebsd_inactive, 7085 .vop_reclaim = zfs_freebsd_reclaim, 7086 .vop_access = zfs_freebsd_access, 7087#ifdef FREEBSD_NAMECACHE 7088 .vop_lookup = vfs_cache_lookup, 7089 .vop_cachedlookup = zfs_freebsd_lookup, 7090#else 7091 .vop_lookup = zfs_freebsd_lookup, 7092#endif 7093 .vop_getattr = zfs_freebsd_getattr, 7094 .vop_setattr = zfs_freebsd_setattr, 7095 .vop_create = zfs_freebsd_create, 7096 .vop_mknod = zfs_freebsd_create, 7097 .vop_mkdir = zfs_freebsd_mkdir, 7098 .vop_readdir = zfs_freebsd_readdir, 7099 .vop_fsync = zfs_freebsd_fsync, 7100 .vop_open = zfs_freebsd_open, 7101 .vop_close = zfs_freebsd_close, 7102 .vop_rmdir = zfs_freebsd_rmdir, 7103 .vop_ioctl = zfs_freebsd_ioctl, 7104 .vop_link = zfs_freebsd_link, 7105 .vop_symlink = zfs_freebsd_symlink, 7106 .vop_readlink = zfs_freebsd_readlink, 7107 .vop_read = zfs_freebsd_read, 7108 .vop_write = zfs_freebsd_write, 7109 .vop_remove = zfs_freebsd_remove, 7110 .vop_rename = zfs_freebsd_rename, 7111 .vop_pathconf = zfs_freebsd_pathconf, 7112 .vop_bmap = zfs_freebsd_bmap, 7113 .vop_fid = zfs_freebsd_fid, 7114 .vop_getextattr = zfs_getextattr, 7115 .vop_deleteextattr = zfs_deleteextattr, 7116 .vop_setextattr = zfs_setextattr, 7117 .vop_listextattr = zfs_listextattr, 7118 .vop_getacl = zfs_freebsd_getacl, 7119 .vop_setacl = zfs_freebsd_setacl, 7120 .vop_aclcheck = zfs_freebsd_aclcheck, 7121 .vop_getpages = zfs_freebsd_getpages, 7122 .vop_putpages = zfs_freebsd_putpages, 7123}; 7124 7125struct vop_vector zfs_fifoops = { 7126 .vop_default = &fifo_specops, 7127 .vop_fsync = zfs_freebsd_fsync, 7128 .vop_access = zfs_freebsd_access, 7129 .vop_getattr = zfs_freebsd_getattr, 7130 .vop_inactive = zfs_freebsd_inactive, 7131 .vop_read = VOP_PANIC, 7132 .vop_reclaim = zfs_freebsd_reclaim, 7133 .vop_setattr = zfs_freebsd_setattr, 7134 .vop_write = VOP_PANIC, 7135 .vop_pathconf = zfs_freebsd_fifo_pathconf, 7136 .vop_fid = zfs_freebsd_fid, 7137 .vop_getacl = zfs_freebsd_getacl, 7138 .vop_setacl = zfs_freebsd_setacl, 7139 .vop_aclcheck = zfs_freebsd_aclcheck, 7140}; 7141 7142/* 7143 * special share hidden files vnode operations template 7144 */ 7145struct vop_vector zfs_shareops = { 7146 .vop_default = &default_vnodeops, 7147 .vop_access = zfs_freebsd_access, 7148 .vop_inactive = zfs_freebsd_inactive, 7149 .vop_reclaim = zfs_freebsd_reclaim, 7150 .vop_fid = zfs_freebsd_fid, 7151 .vop_pathconf = zfs_freebsd_pathconf, 7152}; 7153