zfs_vnops.c revision 273509
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 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 if (error == ESRCH) 270 return (SET_ERROR(ENXIO)); 271 272 /* 273 * We could find a hole that begins after the logical end-of-file, 274 * because dmu_offset_next() only works on whole blocks. If the 275 * EOF falls mid-block, then indicate that the "virtual hole" 276 * at the end of the file begins at the logical EOF, rather than 277 * at the end of the last block. 278 */ 279 if (noff > file_sz) { 280 ASSERT(hole); 281 noff = file_sz; 282 } 283 284 if (noff < *off) 285 return (error); 286 *off = noff; 287 return (error); 288} 289 290/* ARGSUSED */ 291static int 292zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred, 293 int *rvalp, caller_context_t *ct) 294{ 295 offset_t off; 296 int error; 297 zfsvfs_t *zfsvfs; 298 znode_t *zp; 299 300 switch (com) { 301 case _FIOFFS: 302 return (0); 303 304 /* 305 * The following two ioctls are used by bfu. Faking out, 306 * necessary to avoid bfu errors. 307 */ 308 case _FIOGDIO: 309 case _FIOSDIO: 310 return (0); 311 312 case _FIO_SEEK_DATA: 313 case _FIO_SEEK_HOLE: 314#ifdef sun 315 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 316 return (SET_ERROR(EFAULT)); 317#else 318 off = *(offset_t *)data; 319#endif 320 zp = VTOZ(vp); 321 zfsvfs = zp->z_zfsvfs; 322 ZFS_ENTER(zfsvfs); 323 ZFS_VERIFY_ZP(zp); 324 325 /* offset parameter is in/out */ 326 error = zfs_holey(vp, com, &off); 327 ZFS_EXIT(zfsvfs); 328 if (error) 329 return (error); 330#ifdef sun 331 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 332 return (SET_ERROR(EFAULT)); 333#else 334 *(offset_t *)data = off; 335#endif 336 return (0); 337 } 338 return (SET_ERROR(ENOTTY)); 339} 340 341static vm_page_t 342page_busy(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) 343{ 344 vm_object_t obj; 345 vm_page_t pp; 346 int64_t end; 347 348 /* 349 * At present vm_page_clear_dirty extends the cleared range to DEV_BSIZE 350 * aligned boundaries, if the range is not aligned. As a result a 351 * DEV_BSIZE subrange with partially dirty data may get marked as clean. 352 * It may happen that all DEV_BSIZE subranges are marked clean and thus 353 * the whole page would be considred clean despite have some dirty data. 354 * For this reason we should shrink the range to DEV_BSIZE aligned 355 * boundaries before calling vm_page_clear_dirty. 356 */ 357 end = rounddown2(off + nbytes, DEV_BSIZE); 358 off = roundup2(off, DEV_BSIZE); 359 nbytes = end - off; 360 361 obj = vp->v_object; 362 zfs_vmobject_assert_wlocked(obj); 363 364 for (;;) { 365 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 366 pp->valid) { 367 if (vm_page_xbusied(pp)) { 368 /* 369 * Reference the page before unlocking and 370 * sleeping so that the page daemon is less 371 * likely to reclaim it. 372 */ 373 vm_page_reference(pp); 374 vm_page_lock(pp); 375 zfs_vmobject_wunlock(obj); 376 vm_page_busy_sleep(pp, "zfsmwb"); 377 zfs_vmobject_wlock(obj); 378 continue; 379 } 380 vm_page_sbusy(pp); 381 } else if (pp == NULL) { 382 pp = vm_page_alloc(obj, OFF_TO_IDX(start), 383 VM_ALLOC_SYSTEM | VM_ALLOC_IFCACHED | 384 VM_ALLOC_SBUSY); 385 } else { 386 ASSERT(pp != NULL && !pp->valid); 387 pp = NULL; 388 } 389 390 if (pp != NULL) { 391 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 392 vm_object_pip_add(obj, 1); 393 pmap_remove_write(pp); 394 if (nbytes != 0) 395 vm_page_clear_dirty(pp, off, nbytes); 396 } 397 break; 398 } 399 return (pp); 400} 401 402static void 403page_unbusy(vm_page_t pp) 404{ 405 406 vm_page_sunbusy(pp); 407 vm_object_pip_subtract(pp->object, 1); 408} 409 410static vm_page_t 411page_hold(vnode_t *vp, int64_t start) 412{ 413 vm_object_t obj; 414 vm_page_t pp; 415 416 obj = vp->v_object; 417 zfs_vmobject_assert_wlocked(obj); 418 419 for (;;) { 420 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 421 pp->valid) { 422 if (vm_page_xbusied(pp)) { 423 /* 424 * Reference the page before unlocking and 425 * sleeping so that the page daemon is less 426 * likely to reclaim it. 427 */ 428 vm_page_reference(pp); 429 vm_page_lock(pp); 430 zfs_vmobject_wunlock(obj); 431 vm_page_busy_sleep(pp, "zfsmwb"); 432 zfs_vmobject_wlock(obj); 433 continue; 434 } 435 436 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 437 vm_page_lock(pp); 438 vm_page_hold(pp); 439 vm_page_unlock(pp); 440 441 } else 442 pp = NULL; 443 break; 444 } 445 return (pp); 446} 447 448static void 449page_unhold(vm_page_t pp) 450{ 451 452 vm_page_lock(pp); 453 vm_page_unhold(pp); 454 vm_page_unlock(pp); 455} 456 457/* 458 * When a file is memory mapped, we must keep the IO data synchronized 459 * between the DMU cache and the memory mapped pages. What this means: 460 * 461 * On Write: If we find a memory mapped page, we write to *both* 462 * the page and the dmu buffer. 463 */ 464static void 465update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid, 466 int segflg, dmu_tx_t *tx) 467{ 468 vm_object_t obj; 469 struct sf_buf *sf; 470 caddr_t va; 471 int off; 472 473 ASSERT(segflg != UIO_NOCOPY); 474 ASSERT(vp->v_mount != NULL); 475 obj = vp->v_object; 476 ASSERT(obj != NULL); 477 478 off = start & PAGEOFFSET; 479 zfs_vmobject_wlock(obj); 480 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 481 vm_page_t pp; 482 int nbytes = imin(PAGESIZE - off, len); 483 484 if ((pp = page_busy(vp, start, off, nbytes)) != NULL) { 485 zfs_vmobject_wunlock(obj); 486 487 va = zfs_map_page(pp, &sf); 488 (void) dmu_read(os, oid, start+off, nbytes, 489 va+off, DMU_READ_PREFETCH);; 490 zfs_unmap_page(sf); 491 492 zfs_vmobject_wlock(obj); 493 page_unbusy(pp); 494 } 495 len -= nbytes; 496 off = 0; 497 } 498 vm_object_pip_wakeupn(obj, 0); 499 zfs_vmobject_wunlock(obj); 500} 501 502/* 503 * Read with UIO_NOCOPY flag means that sendfile(2) requests 504 * ZFS to populate a range of page cache pages with data. 505 * 506 * NOTE: this function could be optimized to pre-allocate 507 * all pages in advance, drain exclusive busy on all of them, 508 * map them into contiguous KVA region and populate them 509 * in one single dmu_read() call. 510 */ 511static int 512mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio) 513{ 514 znode_t *zp = VTOZ(vp); 515 objset_t *os = zp->z_zfsvfs->z_os; 516 struct sf_buf *sf; 517 vm_object_t obj; 518 vm_page_t pp; 519 int64_t start; 520 caddr_t va; 521 int len = nbytes; 522 int off; 523 int error = 0; 524 525 ASSERT(uio->uio_segflg == UIO_NOCOPY); 526 ASSERT(vp->v_mount != NULL); 527 obj = vp->v_object; 528 ASSERT(obj != NULL); 529 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0); 530 531 zfs_vmobject_wlock(obj); 532 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) { 533 int bytes = MIN(PAGESIZE, len); 534 535 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_SBUSY | 536 VM_ALLOC_NORMAL | VM_ALLOC_IGN_SBUSY); 537 if (pp->valid == 0) { 538 zfs_vmobject_wunlock(obj); 539 va = zfs_map_page(pp, &sf); 540 error = dmu_read(os, zp->z_id, start, bytes, va, 541 DMU_READ_PREFETCH); 542 if (bytes != PAGESIZE && error == 0) 543 bzero(va + bytes, PAGESIZE - bytes); 544 zfs_unmap_page(sf); 545 zfs_vmobject_wlock(obj); 546 vm_page_sunbusy(pp); 547 vm_page_lock(pp); 548 if (error) { 549 if (pp->wire_count == 0 && pp->valid == 0 && 550 !vm_page_busied(pp)) 551 vm_page_free(pp); 552 } else { 553 pp->valid = VM_PAGE_BITS_ALL; 554 vm_page_activate(pp); 555 } 556 vm_page_unlock(pp); 557 } else { 558 ASSERT3U(pp->valid, ==, VM_PAGE_BITS_ALL); 559 vm_page_sunbusy(pp); 560 } 561 if (error) 562 break; 563 uio->uio_resid -= bytes; 564 uio->uio_offset += bytes; 565 len -= bytes; 566 } 567 zfs_vmobject_wunlock(obj); 568 return (error); 569} 570 571/* 572 * When a file is memory mapped, we must keep the IO data synchronized 573 * between the DMU cache and the memory mapped pages. What this means: 574 * 575 * On Read: We "read" preferentially from memory mapped pages, 576 * else we default from the dmu buffer. 577 * 578 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 579 * the file is memory mapped. 580 */ 581static int 582mappedread(vnode_t *vp, int nbytes, uio_t *uio) 583{ 584 znode_t *zp = VTOZ(vp); 585 vm_object_t obj; 586 int64_t start; 587 caddr_t va; 588 int len = nbytes; 589 int off; 590 int error = 0; 591 592 ASSERT(vp->v_mount != NULL); 593 obj = vp->v_object; 594 ASSERT(obj != NULL); 595 596 start = uio->uio_loffset; 597 off = start & PAGEOFFSET; 598 zfs_vmobject_wlock(obj); 599 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 600 vm_page_t pp; 601 uint64_t bytes = MIN(PAGESIZE - off, len); 602 603 if (pp = page_hold(vp, start)) { 604 struct sf_buf *sf; 605 caddr_t va; 606 607 zfs_vmobject_wunlock(obj); 608 va = zfs_map_page(pp, &sf); 609 error = uiomove(va + off, bytes, UIO_READ, uio); 610 zfs_unmap_page(sf); 611 zfs_vmobject_wlock(obj); 612 page_unhold(pp); 613 } else { 614 zfs_vmobject_wunlock(obj); 615 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 616 uio, bytes); 617 zfs_vmobject_wlock(obj); 618 } 619 len -= bytes; 620 off = 0; 621 if (error) 622 break; 623 } 624 zfs_vmobject_wunlock(obj); 625 return (error); 626} 627 628offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 629 630/* 631 * Read bytes from specified file into supplied buffer. 632 * 633 * IN: vp - vnode of file to be read from. 634 * uio - structure supplying read location, range info, 635 * and return buffer. 636 * ioflag - SYNC flags; used to provide FRSYNC semantics. 637 * cr - credentials of caller. 638 * ct - caller context 639 * 640 * OUT: uio - updated offset and range, buffer filled. 641 * 642 * RETURN: 0 on success, error code on failure. 643 * 644 * Side Effects: 645 * vp - atime updated if byte count > 0 646 */ 647/* ARGSUSED */ 648static int 649zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 650{ 651 znode_t *zp = VTOZ(vp); 652 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 653 ssize_t n, nbytes; 654 int error = 0; 655 rl_t *rl; 656 xuio_t *xuio = NULL; 657 658 ZFS_ENTER(zfsvfs); 659 ZFS_VERIFY_ZP(zp); 660 661 if (zp->z_pflags & ZFS_AV_QUARANTINED) { 662 ZFS_EXIT(zfsvfs); 663 return (SET_ERROR(EACCES)); 664 } 665 666 /* 667 * Validate file offset 668 */ 669 if (uio->uio_loffset < (offset_t)0) { 670 ZFS_EXIT(zfsvfs); 671 return (SET_ERROR(EINVAL)); 672 } 673 674 /* 675 * Fasttrack empty reads 676 */ 677 if (uio->uio_resid == 0) { 678 ZFS_EXIT(zfsvfs); 679 return (0); 680 } 681 682 /* 683 * Check for mandatory locks 684 */ 685 if (MANDMODE(zp->z_mode)) { 686 if (error = chklock(vp, FREAD, 687 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 688 ZFS_EXIT(zfsvfs); 689 return (error); 690 } 691 } 692 693 /* 694 * If we're in FRSYNC mode, sync out this znode before reading it. 695 */ 696 if (zfsvfs->z_log && 697 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)) 698 zil_commit(zfsvfs->z_log, zp->z_id); 699 700 /* 701 * Lock the range against changes. 702 */ 703 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 704 705 /* 706 * If we are reading past end-of-file we can skip 707 * to the end; but we might still need to set atime. 708 */ 709 if (uio->uio_loffset >= zp->z_size) { 710 error = 0; 711 goto out; 712 } 713 714 ASSERT(uio->uio_loffset < zp->z_size); 715 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset); 716 717#ifdef sun 718 if ((uio->uio_extflg == UIO_XUIO) && 719 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) { 720 int nblk; 721 int blksz = zp->z_blksz; 722 uint64_t offset = uio->uio_loffset; 723 724 xuio = (xuio_t *)uio; 725 if ((ISP2(blksz))) { 726 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset, 727 blksz)) / blksz; 728 } else { 729 ASSERT(offset + n <= blksz); 730 nblk = 1; 731 } 732 (void) dmu_xuio_init(xuio, nblk); 733 734 if (vn_has_cached_data(vp)) { 735 /* 736 * For simplicity, we always allocate a full buffer 737 * even if we only expect to read a portion of a block. 738 */ 739 while (--nblk >= 0) { 740 (void) dmu_xuio_add(xuio, 741 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 742 blksz), 0, blksz); 743 } 744 } 745 } 746#endif /* sun */ 747 748 while (n > 0) { 749 nbytes = MIN(n, zfs_read_chunk_size - 750 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 751 752#ifdef __FreeBSD__ 753 if (uio->uio_segflg == UIO_NOCOPY) 754 error = mappedread_sf(vp, nbytes, uio); 755 else 756#endif /* __FreeBSD__ */ 757 if (vn_has_cached_data(vp)) { 758 error = mappedread(vp, nbytes, uio); 759 } else { 760 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 761 uio, nbytes); 762 } 763 if (error) { 764 /* convert checksum errors into IO errors */ 765 if (error == ECKSUM) 766 error = SET_ERROR(EIO); 767 break; 768 } 769 770 n -= nbytes; 771 } 772out: 773 zfs_range_unlock(rl); 774 775 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 776 ZFS_EXIT(zfsvfs); 777 return (error); 778} 779 780/* 781 * Write the bytes to a file. 782 * 783 * IN: vp - vnode of file to be written to. 784 * uio - structure supplying write location, range info, 785 * and data buffer. 786 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is 787 * set if in append mode. 788 * cr - credentials of caller. 789 * ct - caller context (NFS/CIFS fem monitor only) 790 * 791 * OUT: uio - updated offset and range. 792 * 793 * RETURN: 0 on success, error code on failure. 794 * 795 * Timestamps: 796 * vp - ctime|mtime updated if byte count > 0 797 */ 798 799/* ARGSUSED */ 800static int 801zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 802{ 803 znode_t *zp = VTOZ(vp); 804 rlim64_t limit = MAXOFFSET_T; 805 ssize_t start_resid = uio->uio_resid; 806 ssize_t tx_bytes; 807 uint64_t end_size; 808 dmu_tx_t *tx; 809 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 810 zilog_t *zilog; 811 offset_t woff; 812 ssize_t n, nbytes; 813 rl_t *rl; 814 int max_blksz = zfsvfs->z_max_blksz; 815 int error = 0; 816 arc_buf_t *abuf; 817 iovec_t *aiov = NULL; 818 xuio_t *xuio = NULL; 819 int i_iov = 0; 820 int iovcnt = uio->uio_iovcnt; 821 iovec_t *iovp = uio->uio_iov; 822 int write_eof; 823 int count = 0; 824 sa_bulk_attr_t bulk[4]; 825 uint64_t mtime[2], ctime[2]; 826 827 /* 828 * Fasttrack empty write 829 */ 830 n = start_resid; 831 if (n == 0) 832 return (0); 833 834 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 835 limit = MAXOFFSET_T; 836 837 ZFS_ENTER(zfsvfs); 838 ZFS_VERIFY_ZP(zp); 839 840 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 841 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 842 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 843 &zp->z_size, 8); 844 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 845 &zp->z_pflags, 8); 846 847 /* 848 * If immutable or not appending then return EPERM 849 */ 850 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 851 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 852 (uio->uio_loffset < zp->z_size))) { 853 ZFS_EXIT(zfsvfs); 854 return (SET_ERROR(EPERM)); 855 } 856 857 zilog = zfsvfs->z_log; 858 859 /* 860 * Validate file offset 861 */ 862 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset; 863 if (woff < 0) { 864 ZFS_EXIT(zfsvfs); 865 return (SET_ERROR(EINVAL)); 866 } 867 868 /* 869 * Check for mandatory locks before calling zfs_range_lock() 870 * in order to prevent a deadlock with locks set via fcntl(). 871 */ 872 if (MANDMODE((mode_t)zp->z_mode) && 873 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 874 ZFS_EXIT(zfsvfs); 875 return (error); 876 } 877 878#ifdef sun 879 /* 880 * Pre-fault the pages to ensure slow (eg NFS) pages 881 * don't hold up txg. 882 * Skip this if uio contains loaned arc_buf. 883 */ 884 if ((uio->uio_extflg == UIO_XUIO) && 885 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) 886 xuio = (xuio_t *)uio; 887 else 888 uio_prefaultpages(MIN(n, max_blksz), uio); 889#endif /* sun */ 890 891 /* 892 * If in append mode, set the io offset pointer to eof. 893 */ 894 if (ioflag & FAPPEND) { 895 /* 896 * Obtain an appending range lock to guarantee file append 897 * semantics. We reset the write offset once we have the lock. 898 */ 899 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 900 woff = rl->r_off; 901 if (rl->r_len == UINT64_MAX) { 902 /* 903 * We overlocked the file because this write will cause 904 * the file block size to increase. 905 * Note that zp_size cannot change with this lock held. 906 */ 907 woff = zp->z_size; 908 } 909 uio->uio_loffset = woff; 910 } else { 911 /* 912 * Note that if the file block size will change as a result of 913 * this write, then this range lock will lock the entire file 914 * so that we can re-write the block safely. 915 */ 916 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 917 } 918 919 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) { 920 zfs_range_unlock(rl); 921 ZFS_EXIT(zfsvfs); 922 return (EFBIG); 923 } 924 925 if (woff >= limit) { 926 zfs_range_unlock(rl); 927 ZFS_EXIT(zfsvfs); 928 return (SET_ERROR(EFBIG)); 929 } 930 931 if ((woff + n) > limit || woff > (limit - n)) 932 n = limit - woff; 933 934 /* Will this write extend the file length? */ 935 write_eof = (woff + n > zp->z_size); 936 937 end_size = MAX(zp->z_size, woff + n); 938 939 /* 940 * Write the file in reasonable size chunks. Each chunk is written 941 * in a separate transaction; this keeps the intent log records small 942 * and allows us to do more fine-grained space accounting. 943 */ 944 while (n > 0) { 945 abuf = NULL; 946 woff = uio->uio_loffset; 947 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 948 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 949 if (abuf != NULL) 950 dmu_return_arcbuf(abuf); 951 error = SET_ERROR(EDQUOT); 952 break; 953 } 954 955 if (xuio && abuf == NULL) { 956 ASSERT(i_iov < iovcnt); 957 aiov = &iovp[i_iov]; 958 abuf = dmu_xuio_arcbuf(xuio, i_iov); 959 dmu_xuio_clear(xuio, i_iov); 960 DTRACE_PROBE3(zfs_cp_write, int, i_iov, 961 iovec_t *, aiov, arc_buf_t *, abuf); 962 ASSERT((aiov->iov_base == abuf->b_data) || 963 ((char *)aiov->iov_base - (char *)abuf->b_data + 964 aiov->iov_len == arc_buf_size(abuf))); 965 i_iov++; 966 } else if (abuf == NULL && n >= max_blksz && 967 woff >= zp->z_size && 968 P2PHASE(woff, max_blksz) == 0 && 969 zp->z_blksz == max_blksz) { 970 /* 971 * This write covers a full block. "Borrow" a buffer 972 * from the dmu so that we can fill it before we enter 973 * a transaction. This avoids the possibility of 974 * holding up the transaction if the data copy hangs 975 * up on a pagefault (e.g., from an NFS server mapping). 976 */ 977 size_t cbytes; 978 979 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 980 max_blksz); 981 ASSERT(abuf != NULL); 982 ASSERT(arc_buf_size(abuf) == max_blksz); 983 if (error = uiocopy(abuf->b_data, max_blksz, 984 UIO_WRITE, uio, &cbytes)) { 985 dmu_return_arcbuf(abuf); 986 break; 987 } 988 ASSERT(cbytes == max_blksz); 989 } 990 991 /* 992 * Start a transaction. 993 */ 994 tx = dmu_tx_create(zfsvfs->z_os); 995 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 996 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 997 zfs_sa_upgrade_txholds(tx, zp); 998 error = dmu_tx_assign(tx, TXG_WAIT); 999 if (error) { 1000 dmu_tx_abort(tx); 1001 if (abuf != NULL) 1002 dmu_return_arcbuf(abuf); 1003 break; 1004 } 1005 1006 /* 1007 * If zfs_range_lock() over-locked we grow the blocksize 1008 * and then reduce the lock range. This will only happen 1009 * on the first iteration since zfs_range_reduce() will 1010 * shrink down r_len to the appropriate size. 1011 */ 1012 if (rl->r_len == UINT64_MAX) { 1013 uint64_t new_blksz; 1014 1015 if (zp->z_blksz > max_blksz) { 1016 ASSERT(!ISP2(zp->z_blksz)); 1017 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 1018 } else { 1019 new_blksz = MIN(end_size, max_blksz); 1020 } 1021 zfs_grow_blocksize(zp, new_blksz, tx); 1022 zfs_range_reduce(rl, woff, n); 1023 } 1024 1025 /* 1026 * XXX - should we really limit each write to z_max_blksz? 1027 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 1028 */ 1029 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 1030 1031 if (woff + nbytes > zp->z_size) 1032 vnode_pager_setsize(vp, woff + nbytes); 1033 1034 if (abuf == NULL) { 1035 tx_bytes = uio->uio_resid; 1036 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), 1037 uio, nbytes, tx); 1038 tx_bytes -= uio->uio_resid; 1039 } else { 1040 tx_bytes = nbytes; 1041 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len); 1042 /* 1043 * If this is not a full block write, but we are 1044 * extending the file past EOF and this data starts 1045 * block-aligned, use assign_arcbuf(). Otherwise, 1046 * write via dmu_write(). 1047 */ 1048 if (tx_bytes < max_blksz && (!write_eof || 1049 aiov->iov_base != abuf->b_data)) { 1050 ASSERT(xuio); 1051 dmu_write(zfsvfs->z_os, zp->z_id, woff, 1052 aiov->iov_len, aiov->iov_base, tx); 1053 dmu_return_arcbuf(abuf); 1054 xuio_stat_wbuf_copied(); 1055 } else { 1056 ASSERT(xuio || tx_bytes == max_blksz); 1057 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), 1058 woff, abuf, tx); 1059 } 1060 ASSERT(tx_bytes <= uio->uio_resid); 1061 uioskip(uio, tx_bytes); 1062 } 1063 if (tx_bytes && vn_has_cached_data(vp)) { 1064 update_pages(vp, woff, tx_bytes, zfsvfs->z_os, 1065 zp->z_id, uio->uio_segflg, tx); 1066 } 1067 1068 /* 1069 * If we made no progress, we're done. If we made even 1070 * partial progress, update the znode and ZIL accordingly. 1071 */ 1072 if (tx_bytes == 0) { 1073 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 1074 (void *)&zp->z_size, sizeof (uint64_t), tx); 1075 dmu_tx_commit(tx); 1076 ASSERT(error != 0); 1077 break; 1078 } 1079 1080 /* 1081 * Clear Set-UID/Set-GID bits on successful write if not 1082 * privileged and at least one of the excute bits is set. 1083 * 1084 * It would be nice to to this after all writes have 1085 * been done, but that would still expose the ISUID/ISGID 1086 * to another app after the partial write is committed. 1087 * 1088 * Note: we don't call zfs_fuid_map_id() here because 1089 * user 0 is not an ephemeral uid. 1090 */ 1091 mutex_enter(&zp->z_acl_lock); 1092 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | 1093 (S_IXUSR >> 6))) != 0 && 1094 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && 1095 secpolicy_vnode_setid_retain(vp, cr, 1096 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) { 1097 uint64_t newmode; 1098 zp->z_mode &= ~(S_ISUID | S_ISGID); 1099 newmode = zp->z_mode; 1100 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), 1101 (void *)&newmode, sizeof (uint64_t), tx); 1102 } 1103 mutex_exit(&zp->z_acl_lock); 1104 1105 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 1106 B_TRUE); 1107 1108 /* 1109 * Update the file size (zp_size) if it has changed; 1110 * account for possible concurrent updates. 1111 */ 1112 while ((end_size = zp->z_size) < uio->uio_loffset) { 1113 (void) atomic_cas_64(&zp->z_size, end_size, 1114 uio->uio_loffset); 1115 ASSERT(error == 0); 1116 } 1117 /* 1118 * If we are replaying and eof is non zero then force 1119 * the file size to the specified eof. Note, there's no 1120 * concurrency during replay. 1121 */ 1122 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) 1123 zp->z_size = zfsvfs->z_replay_eof; 1124 1125 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1126 1127 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 1128 dmu_tx_commit(tx); 1129 1130 if (error != 0) 1131 break; 1132 ASSERT(tx_bytes == nbytes); 1133 n -= nbytes; 1134 1135#ifdef sun 1136 if (!xuio && n > 0) 1137 uio_prefaultpages(MIN(n, max_blksz), uio); 1138#endif /* sun */ 1139 } 1140 1141 zfs_range_unlock(rl); 1142 1143 /* 1144 * If we're in replay mode, or we made no progress, return error. 1145 * Otherwise, it's at least a partial write, so it's successful. 1146 */ 1147 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 1148 ZFS_EXIT(zfsvfs); 1149 return (error); 1150 } 1151 1152 if (ioflag & (FSYNC | FDSYNC) || 1153 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1154 zil_commit(zilog, zp->z_id); 1155 1156 ZFS_EXIT(zfsvfs); 1157 return (0); 1158} 1159 1160void 1161zfs_get_done(zgd_t *zgd, int error) 1162{ 1163 znode_t *zp = zgd->zgd_private; 1164 objset_t *os = zp->z_zfsvfs->z_os; 1165 1166 if (zgd->zgd_db) 1167 dmu_buf_rele(zgd->zgd_db, zgd); 1168 1169 zfs_range_unlock(zgd->zgd_rl); 1170 1171 /* 1172 * Release the vnode asynchronously as we currently have the 1173 * txg stopped from syncing. 1174 */ 1175 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1176 1177 if (error == 0 && zgd->zgd_bp) 1178 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1179 1180 kmem_free(zgd, sizeof (zgd_t)); 1181} 1182 1183#ifdef DEBUG 1184static int zil_fault_io = 0; 1185#endif 1186 1187/* 1188 * Get data to generate a TX_WRITE intent log record. 1189 */ 1190int 1191zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1192{ 1193 zfsvfs_t *zfsvfs = arg; 1194 objset_t *os = zfsvfs->z_os; 1195 znode_t *zp; 1196 uint64_t object = lr->lr_foid; 1197 uint64_t offset = lr->lr_offset; 1198 uint64_t size = lr->lr_length; 1199 blkptr_t *bp = &lr->lr_blkptr; 1200 dmu_buf_t *db; 1201 zgd_t *zgd; 1202 int error = 0; 1203 1204 ASSERT(zio != NULL); 1205 ASSERT(size != 0); 1206 1207 /* 1208 * Nothing to do if the file has been removed 1209 */ 1210 if (zfs_zget(zfsvfs, object, &zp) != 0) 1211 return (SET_ERROR(ENOENT)); 1212 if (zp->z_unlinked) { 1213 /* 1214 * Release the vnode asynchronously as we currently have the 1215 * txg stopped from syncing. 1216 */ 1217 VN_RELE_ASYNC(ZTOV(zp), 1218 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1219 return (SET_ERROR(ENOENT)); 1220 } 1221 1222 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1223 zgd->zgd_zilog = zfsvfs->z_log; 1224 zgd->zgd_private = zp; 1225 1226 /* 1227 * Write records come in two flavors: immediate and indirect. 1228 * For small writes it's cheaper to store the data with the 1229 * log record (immediate); for large writes it's cheaper to 1230 * sync the data and get a pointer to it (indirect) so that 1231 * we don't have to write the data twice. 1232 */ 1233 if (buf != NULL) { /* immediate write */ 1234 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER); 1235 /* test for truncation needs to be done while range locked */ 1236 if (offset >= zp->z_size) { 1237 error = SET_ERROR(ENOENT); 1238 } else { 1239 error = dmu_read(os, object, offset, size, buf, 1240 DMU_READ_NO_PREFETCH); 1241 } 1242 ASSERT(error == 0 || error == ENOENT); 1243 } else { /* indirect write */ 1244 /* 1245 * Have to lock the whole block to ensure when it's 1246 * written out and it's checksum is being calculated 1247 * that no one can change the data. We need to re-check 1248 * blocksize after we get the lock in case it's changed! 1249 */ 1250 for (;;) { 1251 uint64_t blkoff; 1252 size = zp->z_blksz; 1253 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; 1254 offset -= blkoff; 1255 zgd->zgd_rl = zfs_range_lock(zp, offset, size, 1256 RL_READER); 1257 if (zp->z_blksz == size) 1258 break; 1259 offset += blkoff; 1260 zfs_range_unlock(zgd->zgd_rl); 1261 } 1262 /* test for truncation needs to be done while range locked */ 1263 if (lr->lr_offset >= zp->z_size) 1264 error = SET_ERROR(ENOENT); 1265#ifdef DEBUG 1266 if (zil_fault_io) { 1267 error = SET_ERROR(EIO); 1268 zil_fault_io = 0; 1269 } 1270#endif 1271 if (error == 0) 1272 error = dmu_buf_hold(os, object, offset, zgd, &db, 1273 DMU_READ_NO_PREFETCH); 1274 1275 if (error == 0) { 1276 blkptr_t *obp = dmu_buf_get_blkptr(db); 1277 if (obp) { 1278 ASSERT(BP_IS_HOLE(bp)); 1279 *bp = *obp; 1280 } 1281 1282 zgd->zgd_db = db; 1283 zgd->zgd_bp = bp; 1284 1285 ASSERT(db->db_offset == offset); 1286 ASSERT(db->db_size == size); 1287 1288 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1289 zfs_get_done, zgd); 1290 ASSERT(error || lr->lr_length <= zp->z_blksz); 1291 1292 /* 1293 * On success, we need to wait for the write I/O 1294 * initiated by dmu_sync() to complete before we can 1295 * release this dbuf. We will finish everything up 1296 * in the zfs_get_done() callback. 1297 */ 1298 if (error == 0) 1299 return (0); 1300 1301 if (error == EALREADY) { 1302 lr->lr_common.lrc_txtype = TX_WRITE2; 1303 error = 0; 1304 } 1305 } 1306 } 1307 1308 zfs_get_done(zgd, error); 1309 1310 return (error); 1311} 1312 1313/*ARGSUSED*/ 1314static int 1315zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 1316 caller_context_t *ct) 1317{ 1318 znode_t *zp = VTOZ(vp); 1319 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1320 int error; 1321 1322 ZFS_ENTER(zfsvfs); 1323 ZFS_VERIFY_ZP(zp); 1324 1325 if (flag & V_ACE_MASK) 1326 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 1327 else 1328 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1329 1330 ZFS_EXIT(zfsvfs); 1331 return (error); 1332} 1333 1334/* 1335 * If vnode is for a device return a specfs vnode instead. 1336 */ 1337static int 1338specvp_check(vnode_t **vpp, cred_t *cr) 1339{ 1340 int error = 0; 1341 1342 if (IS_DEVVP(*vpp)) { 1343 struct vnode *svp; 1344 1345 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1346 VN_RELE(*vpp); 1347 if (svp == NULL) 1348 error = SET_ERROR(ENOSYS); 1349 *vpp = svp; 1350 } 1351 return (error); 1352} 1353 1354 1355/* 1356 * Lookup an entry in a directory, or an extended attribute directory. 1357 * If it exists, return a held vnode reference for it. 1358 * 1359 * IN: dvp - vnode of directory to search. 1360 * nm - name of entry to lookup. 1361 * pnp - full pathname to lookup [UNUSED]. 1362 * flags - LOOKUP_XATTR set if looking for an attribute. 1363 * rdir - root directory vnode [UNUSED]. 1364 * cr - credentials of caller. 1365 * ct - caller context 1366 * direntflags - directory lookup flags 1367 * realpnp - returned pathname. 1368 * 1369 * OUT: vpp - vnode of located entry, NULL if not found. 1370 * 1371 * RETURN: 0 on success, error code on failure. 1372 * 1373 * Timestamps: 1374 * NA 1375 */ 1376/* ARGSUSED */ 1377static int 1378zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 1379 int nameiop, cred_t *cr, kthread_t *td, int flags) 1380{ 1381 znode_t *zdp = VTOZ(dvp); 1382 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1383 int error = 0; 1384 int *direntflags = NULL; 1385 void *realpnp = NULL; 1386 1387 /* fast path */ 1388 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 1389 1390 if (dvp->v_type != VDIR) { 1391 return (SET_ERROR(ENOTDIR)); 1392 } else if (zdp->z_sa_hdl == NULL) { 1393 return (SET_ERROR(EIO)); 1394 } 1395 1396 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 1397 error = zfs_fastaccesschk_execute(zdp, cr); 1398 if (!error) { 1399 *vpp = dvp; 1400 VN_HOLD(*vpp); 1401 return (0); 1402 } 1403 return (error); 1404 } else { 1405 vnode_t *tvp = dnlc_lookup(dvp, nm); 1406 1407 if (tvp) { 1408 error = zfs_fastaccesschk_execute(zdp, cr); 1409 if (error) { 1410 VN_RELE(tvp); 1411 return (error); 1412 } 1413 if (tvp == DNLC_NO_VNODE) { 1414 VN_RELE(tvp); 1415 return (SET_ERROR(ENOENT)); 1416 } else { 1417 *vpp = tvp; 1418 return (specvp_check(vpp, cr)); 1419 } 1420 } 1421 } 1422 } 1423 1424 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1425 1426 ZFS_ENTER(zfsvfs); 1427 ZFS_VERIFY_ZP(zdp); 1428 1429 *vpp = NULL; 1430 1431 if (flags & LOOKUP_XATTR) { 1432#ifdef TODO 1433 /* 1434 * If the xattr property is off, refuse the lookup request. 1435 */ 1436 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1437 ZFS_EXIT(zfsvfs); 1438 return (SET_ERROR(EINVAL)); 1439 } 1440#endif 1441 1442 /* 1443 * We don't allow recursive attributes.. 1444 * Maybe someday we will. 1445 */ 1446 if (zdp->z_pflags & ZFS_XATTR) { 1447 ZFS_EXIT(zfsvfs); 1448 return (SET_ERROR(EINVAL)); 1449 } 1450 1451 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1452 ZFS_EXIT(zfsvfs); 1453 return (error); 1454 } 1455 1456 /* 1457 * Do we have permission to get into attribute directory? 1458 */ 1459 1460 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1461 B_FALSE, cr)) { 1462 VN_RELE(*vpp); 1463 *vpp = NULL; 1464 } 1465 1466 ZFS_EXIT(zfsvfs); 1467 return (error); 1468 } 1469 1470 if (dvp->v_type != VDIR) { 1471 ZFS_EXIT(zfsvfs); 1472 return (SET_ERROR(ENOTDIR)); 1473 } 1474 1475 /* 1476 * Check accessibility of directory. 1477 */ 1478 1479 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1480 ZFS_EXIT(zfsvfs); 1481 return (error); 1482 } 1483 1484 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1485 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1486 ZFS_EXIT(zfsvfs); 1487 return (SET_ERROR(EILSEQ)); 1488 } 1489 1490 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1491 if (error == 0) 1492 error = specvp_check(vpp, cr); 1493 1494 /* Translate errors and add SAVENAME when needed. */ 1495 if (cnp->cn_flags & ISLASTCN) { 1496 switch (nameiop) { 1497 case CREATE: 1498 case RENAME: 1499 if (error == ENOENT) { 1500 error = EJUSTRETURN; 1501 cnp->cn_flags |= SAVENAME; 1502 break; 1503 } 1504 /* FALLTHROUGH */ 1505 case DELETE: 1506 if (error == 0) 1507 cnp->cn_flags |= SAVENAME; 1508 break; 1509 } 1510 } 1511 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) { 1512 int ltype = 0; 1513 1514 if (cnp->cn_flags & ISDOTDOT) { 1515 ltype = VOP_ISLOCKED(dvp); 1516 VOP_UNLOCK(dvp, 0); 1517 } 1518 ZFS_EXIT(zfsvfs); 1519 error = vn_lock(*vpp, cnp->cn_lkflags); 1520 if (cnp->cn_flags & ISDOTDOT) 1521 vn_lock(dvp, ltype | LK_RETRY); 1522 if (error != 0) { 1523 VN_RELE(*vpp); 1524 *vpp = NULL; 1525 return (error); 1526 } 1527 } else { 1528 ZFS_EXIT(zfsvfs); 1529 } 1530 1531#ifdef FREEBSD_NAMECACHE 1532 /* 1533 * Insert name into cache (as non-existent) if appropriate. 1534 */ 1535 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE) 1536 cache_enter(dvp, *vpp, cnp); 1537 /* 1538 * Insert name into cache if appropriate. 1539 */ 1540 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1541 if (!(cnp->cn_flags & ISLASTCN) || 1542 (nameiop != DELETE && nameiop != RENAME)) { 1543 cache_enter(dvp, *vpp, cnp); 1544 } 1545 } 1546#endif 1547 1548 return (error); 1549} 1550 1551/* 1552 * Attempt to create a new entry in a directory. If the entry 1553 * already exists, truncate the file if permissible, else return 1554 * an error. Return the vp of the created or trunc'd file. 1555 * 1556 * IN: dvp - vnode of directory to put new file entry in. 1557 * name - name of new file entry. 1558 * vap - attributes of new file. 1559 * excl - flag indicating exclusive or non-exclusive mode. 1560 * mode - mode to open file with. 1561 * cr - credentials of caller. 1562 * flag - large file flag [UNUSED]. 1563 * ct - caller context 1564 * vsecp - ACL to be set 1565 * 1566 * OUT: vpp - vnode of created or trunc'd entry. 1567 * 1568 * RETURN: 0 on success, error code on failure. 1569 * 1570 * Timestamps: 1571 * dvp - ctime|mtime updated if new entry created 1572 * vp - ctime|mtime always, atime if new 1573 */ 1574 1575/* ARGSUSED */ 1576static int 1577zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1578 vnode_t **vpp, cred_t *cr, kthread_t *td) 1579{ 1580 znode_t *zp, *dzp = VTOZ(dvp); 1581 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1582 zilog_t *zilog; 1583 objset_t *os; 1584 zfs_dirlock_t *dl; 1585 dmu_tx_t *tx; 1586 int error; 1587 ksid_t *ksid; 1588 uid_t uid; 1589 gid_t gid = crgetgid(cr); 1590 zfs_acl_ids_t acl_ids; 1591 boolean_t fuid_dirtied; 1592 boolean_t have_acl = B_FALSE; 1593 boolean_t waited = B_FALSE; 1594 void *vsecp = NULL; 1595 int flag = 0; 1596 1597 /* 1598 * If we have an ephemeral id, ACL, or XVATTR then 1599 * make sure file system is at proper version 1600 */ 1601 1602 ksid = crgetsid(cr, KSID_OWNER); 1603 if (ksid) 1604 uid = ksid_getid(ksid); 1605 else 1606 uid = crgetuid(cr); 1607 1608 if (zfsvfs->z_use_fuids == B_FALSE && 1609 (vsecp || (vap->va_mask & AT_XVATTR) || 1610 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1611 return (SET_ERROR(EINVAL)); 1612 1613 ZFS_ENTER(zfsvfs); 1614 ZFS_VERIFY_ZP(dzp); 1615 os = zfsvfs->z_os; 1616 zilog = zfsvfs->z_log; 1617 1618 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1619 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1620 ZFS_EXIT(zfsvfs); 1621 return (SET_ERROR(EILSEQ)); 1622 } 1623 1624 if (vap->va_mask & AT_XVATTR) { 1625 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 1626 crgetuid(cr), cr, vap->va_type)) != 0) { 1627 ZFS_EXIT(zfsvfs); 1628 return (error); 1629 } 1630 } 1631 1632 getnewvnode_reserve(1); 1633 1634top: 1635 *vpp = NULL; 1636 1637 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) 1638 vap->va_mode &= ~S_ISVTX; 1639 1640 if (*name == '\0') { 1641 /* 1642 * Null component name refers to the directory itself. 1643 */ 1644 VN_HOLD(dvp); 1645 zp = dzp; 1646 dl = NULL; 1647 error = 0; 1648 } else { 1649 /* possible VN_HOLD(zp) */ 1650 int zflg = 0; 1651 1652 if (flag & FIGNORECASE) 1653 zflg |= ZCILOOK; 1654 1655 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1656 NULL, NULL); 1657 if (error) { 1658 if (have_acl) 1659 zfs_acl_ids_free(&acl_ids); 1660 if (strcmp(name, "..") == 0) 1661 error = SET_ERROR(EISDIR); 1662 getnewvnode_drop_reserve(); 1663 ZFS_EXIT(zfsvfs); 1664 return (error); 1665 } 1666 } 1667 1668 if (zp == NULL) { 1669 uint64_t txtype; 1670 1671 /* 1672 * Create a new file object and update the directory 1673 * to reference it. 1674 */ 1675 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1676 if (have_acl) 1677 zfs_acl_ids_free(&acl_ids); 1678 goto out; 1679 } 1680 1681 /* 1682 * We only support the creation of regular files in 1683 * extended attribute directories. 1684 */ 1685 1686 if ((dzp->z_pflags & ZFS_XATTR) && 1687 (vap->va_type != VREG)) { 1688 if (have_acl) 1689 zfs_acl_ids_free(&acl_ids); 1690 error = SET_ERROR(EINVAL); 1691 goto out; 1692 } 1693 1694 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 1695 cr, vsecp, &acl_ids)) != 0) 1696 goto out; 1697 have_acl = B_TRUE; 1698 1699 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1700 zfs_acl_ids_free(&acl_ids); 1701 error = SET_ERROR(EDQUOT); 1702 goto out; 1703 } 1704 1705 tx = dmu_tx_create(os); 1706 1707 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1708 ZFS_SA_BASE_ATTR_SIZE); 1709 1710 fuid_dirtied = zfsvfs->z_fuid_dirty; 1711 if (fuid_dirtied) 1712 zfs_fuid_txhold(zfsvfs, tx); 1713 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1714 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1715 if (!zfsvfs->z_use_sa && 1716 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1717 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1718 0, acl_ids.z_aclp->z_acl_bytes); 1719 } 1720 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 1721 if (error) { 1722 zfs_dirent_unlock(dl); 1723 if (error == ERESTART) { 1724 waited = B_TRUE; 1725 dmu_tx_wait(tx); 1726 dmu_tx_abort(tx); 1727 goto top; 1728 } 1729 zfs_acl_ids_free(&acl_ids); 1730 dmu_tx_abort(tx); 1731 getnewvnode_drop_reserve(); 1732 ZFS_EXIT(zfsvfs); 1733 return (error); 1734 } 1735 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1736 1737 if (fuid_dirtied) 1738 zfs_fuid_sync(zfsvfs, tx); 1739 1740 (void) zfs_link_create(dl, zp, tx, ZNEW); 1741 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1742 if (flag & FIGNORECASE) 1743 txtype |= TX_CI; 1744 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1745 vsecp, acl_ids.z_fuidp, vap); 1746 zfs_acl_ids_free(&acl_ids); 1747 dmu_tx_commit(tx); 1748 } else { 1749 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1750 1751 if (have_acl) 1752 zfs_acl_ids_free(&acl_ids); 1753 have_acl = B_FALSE; 1754 1755 /* 1756 * A directory entry already exists for this name. 1757 */ 1758 /* 1759 * Can't truncate an existing file if in exclusive mode. 1760 */ 1761 if (excl == EXCL) { 1762 error = SET_ERROR(EEXIST); 1763 goto out; 1764 } 1765 /* 1766 * Can't open a directory for writing. 1767 */ 1768 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1769 error = SET_ERROR(EISDIR); 1770 goto out; 1771 } 1772 /* 1773 * Verify requested access to file. 1774 */ 1775 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1776 goto out; 1777 } 1778 1779 mutex_enter(&dzp->z_lock); 1780 dzp->z_seq++; 1781 mutex_exit(&dzp->z_lock); 1782 1783 /* 1784 * Truncate regular files if requested. 1785 */ 1786 if ((ZTOV(zp)->v_type == VREG) && 1787 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1788 /* we can't hold any locks when calling zfs_freesp() */ 1789 zfs_dirent_unlock(dl); 1790 dl = NULL; 1791 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1792 if (error == 0) { 1793 vnevent_create(ZTOV(zp), ct); 1794 } 1795 } 1796 } 1797out: 1798 getnewvnode_drop_reserve(); 1799 if (dl) 1800 zfs_dirent_unlock(dl); 1801 1802 if (error) { 1803 if (zp) 1804 VN_RELE(ZTOV(zp)); 1805 } else { 1806 *vpp = ZTOV(zp); 1807 error = specvp_check(vpp, cr); 1808 } 1809 1810 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1811 zil_commit(zilog, 0); 1812 1813 ZFS_EXIT(zfsvfs); 1814 return (error); 1815} 1816 1817/* 1818 * Remove an entry from a directory. 1819 * 1820 * IN: dvp - vnode of directory to remove entry from. 1821 * name - name of entry to remove. 1822 * cr - credentials of caller. 1823 * ct - caller context 1824 * flags - case flags 1825 * 1826 * RETURN: 0 on success, error code on failure. 1827 * 1828 * Timestamps: 1829 * dvp - ctime|mtime 1830 * vp - ctime (if nlink > 0) 1831 */ 1832 1833uint64_t null_xattr = 0; 1834 1835/*ARGSUSED*/ 1836static int 1837zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1838 int flags) 1839{ 1840 znode_t *zp, *dzp = VTOZ(dvp); 1841 znode_t *xzp; 1842 vnode_t *vp; 1843 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1844 zilog_t *zilog; 1845 uint64_t acl_obj, xattr_obj; 1846 uint64_t xattr_obj_unlinked = 0; 1847 uint64_t obj = 0; 1848 zfs_dirlock_t *dl; 1849 dmu_tx_t *tx; 1850 boolean_t may_delete_now, delete_now = FALSE; 1851 boolean_t unlinked, toobig = FALSE; 1852 uint64_t txtype; 1853 pathname_t *realnmp = NULL; 1854 pathname_t realnm; 1855 int error; 1856 int zflg = ZEXISTS; 1857 boolean_t waited = B_FALSE; 1858 1859 ZFS_ENTER(zfsvfs); 1860 ZFS_VERIFY_ZP(dzp); 1861 zilog = zfsvfs->z_log; 1862 1863 if (flags & FIGNORECASE) { 1864 zflg |= ZCILOOK; 1865 pn_alloc(&realnm); 1866 realnmp = &realnm; 1867 } 1868 1869top: 1870 xattr_obj = 0; 1871 xzp = NULL; 1872 /* 1873 * Attempt to lock directory; fail if entry doesn't exist. 1874 */ 1875 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1876 NULL, realnmp)) { 1877 if (realnmp) 1878 pn_free(realnmp); 1879 ZFS_EXIT(zfsvfs); 1880 return (error); 1881 } 1882 1883 vp = ZTOV(zp); 1884 1885 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1886 goto out; 1887 } 1888 1889 /* 1890 * Need to use rmdir for removing directories. 1891 */ 1892 if (vp->v_type == VDIR) { 1893 error = SET_ERROR(EPERM); 1894 goto out; 1895 } 1896 1897 vnevent_remove(vp, dvp, name, ct); 1898 1899 if (realnmp) 1900 dnlc_remove(dvp, realnmp->pn_buf); 1901 else 1902 dnlc_remove(dvp, name); 1903 1904 VI_LOCK(vp); 1905 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1906 VI_UNLOCK(vp); 1907 1908 /* 1909 * We may delete the znode now, or we may put it in the unlinked set; 1910 * it depends on whether we're the last link, and on whether there are 1911 * other holds on the vnode. So we dmu_tx_hold() the right things to 1912 * allow for either case. 1913 */ 1914 obj = zp->z_id; 1915 tx = dmu_tx_create(zfsvfs->z_os); 1916 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1917 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1918 zfs_sa_upgrade_txholds(tx, zp); 1919 zfs_sa_upgrade_txholds(tx, dzp); 1920 if (may_delete_now) { 1921 toobig = 1922 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 1923 /* if the file is too big, only hold_free a token amount */ 1924 dmu_tx_hold_free(tx, zp->z_id, 0, 1925 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1926 } 1927 1928 /* are there any extended attributes? */ 1929 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1930 &xattr_obj, sizeof (xattr_obj)); 1931 if (error == 0 && xattr_obj) { 1932 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1933 ASSERT0(error); 1934 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1935 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1936 } 1937 1938 mutex_enter(&zp->z_lock); 1939 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now) 1940 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1941 mutex_exit(&zp->z_lock); 1942 1943 /* charge as an update -- would be nice not to charge at all */ 1944 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1945 1946 /* 1947 * Mark this transaction as typically resulting in a net free of 1948 * space, unless object removal will be delayed indefinitely 1949 * (due to active holds on the vnode due to the file being open). 1950 */ 1951 if (may_delete_now) 1952 dmu_tx_mark_netfree(tx); 1953 1954 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 1955 if (error) { 1956 zfs_dirent_unlock(dl); 1957 VN_RELE(vp); 1958 if (xzp) 1959 VN_RELE(ZTOV(xzp)); 1960 if (error == ERESTART) { 1961 waited = B_TRUE; 1962 dmu_tx_wait(tx); 1963 dmu_tx_abort(tx); 1964 goto top; 1965 } 1966 if (realnmp) 1967 pn_free(realnmp); 1968 dmu_tx_abort(tx); 1969 ZFS_EXIT(zfsvfs); 1970 return (error); 1971 } 1972 1973 /* 1974 * Remove the directory entry. 1975 */ 1976 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1977 1978 if (error) { 1979 dmu_tx_commit(tx); 1980 goto out; 1981 } 1982 1983 if (unlinked) { 1984 /* 1985 * Hold z_lock so that we can make sure that the ACL obj 1986 * hasn't changed. Could have been deleted due to 1987 * zfs_sa_upgrade(). 1988 */ 1989 mutex_enter(&zp->z_lock); 1990 VI_LOCK(vp); 1991 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1992 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked)); 1993 delete_now = may_delete_now && !toobig && 1994 vp->v_count == 1 && !vn_has_cached_data(vp) && 1995 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) == 1996 acl_obj; 1997 VI_UNLOCK(vp); 1998 } 1999 2000 if (delete_now) { 2001#ifdef __FreeBSD__ 2002 panic("zfs_remove: delete_now branch taken"); 2003#endif 2004 if (xattr_obj_unlinked) { 2005 ASSERT3U(xzp->z_links, ==, 2); 2006 mutex_enter(&xzp->z_lock); 2007 xzp->z_unlinked = 1; 2008 xzp->z_links = 0; 2009 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 2010 &xzp->z_links, sizeof (xzp->z_links), tx); 2011 ASSERT3U(error, ==, 0); 2012 mutex_exit(&xzp->z_lock); 2013 zfs_unlinked_add(xzp, tx); 2014 2015 if (zp->z_is_sa) 2016 error = sa_remove(zp->z_sa_hdl, 2017 SA_ZPL_XATTR(zfsvfs), tx); 2018 else 2019 error = sa_update(zp->z_sa_hdl, 2020 SA_ZPL_XATTR(zfsvfs), &null_xattr, 2021 sizeof (uint64_t), tx); 2022 ASSERT0(error); 2023 } 2024 VI_LOCK(vp); 2025 vp->v_count--; 2026 ASSERT0(vp->v_count); 2027 VI_UNLOCK(vp); 2028 mutex_exit(&zp->z_lock); 2029 zfs_znode_delete(zp, tx); 2030 } else if (unlinked) { 2031 mutex_exit(&zp->z_lock); 2032 zfs_unlinked_add(zp, tx); 2033#ifdef __FreeBSD__ 2034 vp->v_vflag |= VV_NOSYNC; 2035#endif 2036 } 2037 2038 txtype = TX_REMOVE; 2039 if (flags & FIGNORECASE) 2040 txtype |= TX_CI; 2041 zfs_log_remove(zilog, tx, txtype, dzp, name, obj); 2042 2043 dmu_tx_commit(tx); 2044out: 2045 if (realnmp) 2046 pn_free(realnmp); 2047 2048 zfs_dirent_unlock(dl); 2049 2050 if (!delete_now) 2051 VN_RELE(vp); 2052 if (xzp) 2053 VN_RELE(ZTOV(xzp)); 2054 2055 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2056 zil_commit(zilog, 0); 2057 2058 ZFS_EXIT(zfsvfs); 2059 return (error); 2060} 2061 2062/* 2063 * Create a new directory and insert it into dvp using the name 2064 * provided. Return a pointer to the inserted directory. 2065 * 2066 * IN: dvp - vnode of directory to add subdir to. 2067 * dirname - name of new directory. 2068 * vap - attributes of new directory. 2069 * cr - credentials of caller. 2070 * ct - caller context 2071 * flags - case flags 2072 * vsecp - ACL to be set 2073 * 2074 * OUT: vpp - vnode of created directory. 2075 * 2076 * RETURN: 0 on success, error code on failure. 2077 * 2078 * Timestamps: 2079 * dvp - ctime|mtime updated 2080 * vp - ctime|mtime|atime updated 2081 */ 2082/*ARGSUSED*/ 2083static int 2084zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 2085 caller_context_t *ct, int flags, vsecattr_t *vsecp) 2086{ 2087 znode_t *zp, *dzp = VTOZ(dvp); 2088 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2089 zilog_t *zilog; 2090 zfs_dirlock_t *dl; 2091 uint64_t txtype; 2092 dmu_tx_t *tx; 2093 int error; 2094 int zf = ZNEW; 2095 ksid_t *ksid; 2096 uid_t uid; 2097 gid_t gid = crgetgid(cr); 2098 zfs_acl_ids_t acl_ids; 2099 boolean_t fuid_dirtied; 2100 boolean_t waited = B_FALSE; 2101 2102 ASSERT(vap->va_type == VDIR); 2103 2104 /* 2105 * If we have an ephemeral id, ACL, or XVATTR then 2106 * make sure file system is at proper version 2107 */ 2108 2109 ksid = crgetsid(cr, KSID_OWNER); 2110 if (ksid) 2111 uid = ksid_getid(ksid); 2112 else 2113 uid = crgetuid(cr); 2114 if (zfsvfs->z_use_fuids == B_FALSE && 2115 (vsecp || (vap->va_mask & AT_XVATTR) || 2116 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 2117 return (SET_ERROR(EINVAL)); 2118 2119 ZFS_ENTER(zfsvfs); 2120 ZFS_VERIFY_ZP(dzp); 2121 zilog = zfsvfs->z_log; 2122 2123 if (dzp->z_pflags & ZFS_XATTR) { 2124 ZFS_EXIT(zfsvfs); 2125 return (SET_ERROR(EINVAL)); 2126 } 2127 2128 if (zfsvfs->z_utf8 && u8_validate(dirname, 2129 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2130 ZFS_EXIT(zfsvfs); 2131 return (SET_ERROR(EILSEQ)); 2132 } 2133 if (flags & FIGNORECASE) 2134 zf |= ZCILOOK; 2135 2136 if (vap->va_mask & AT_XVATTR) { 2137 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 2138 crgetuid(cr), cr, vap->va_type)) != 0) { 2139 ZFS_EXIT(zfsvfs); 2140 return (error); 2141 } 2142 } 2143 2144 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 2145 vsecp, &acl_ids)) != 0) { 2146 ZFS_EXIT(zfsvfs); 2147 return (error); 2148 } 2149 2150 getnewvnode_reserve(1); 2151 2152 /* 2153 * First make sure the new directory doesn't exist. 2154 * 2155 * Existence is checked first to make sure we don't return 2156 * EACCES instead of EEXIST which can cause some applications 2157 * to fail. 2158 */ 2159top: 2160 *vpp = NULL; 2161 2162 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 2163 NULL, NULL)) { 2164 zfs_acl_ids_free(&acl_ids); 2165 getnewvnode_drop_reserve(); 2166 ZFS_EXIT(zfsvfs); 2167 return (error); 2168 } 2169 2170 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 2171 zfs_acl_ids_free(&acl_ids); 2172 zfs_dirent_unlock(dl); 2173 getnewvnode_drop_reserve(); 2174 ZFS_EXIT(zfsvfs); 2175 return (error); 2176 } 2177 2178 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 2179 zfs_acl_ids_free(&acl_ids); 2180 zfs_dirent_unlock(dl); 2181 getnewvnode_drop_reserve(); 2182 ZFS_EXIT(zfsvfs); 2183 return (SET_ERROR(EDQUOT)); 2184 } 2185 2186 /* 2187 * Add a new entry to the directory. 2188 */ 2189 tx = dmu_tx_create(zfsvfs->z_os); 2190 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 2191 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2192 fuid_dirtied = zfsvfs->z_fuid_dirty; 2193 if (fuid_dirtied) 2194 zfs_fuid_txhold(zfsvfs, tx); 2195 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2196 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2197 acl_ids.z_aclp->z_acl_bytes); 2198 } 2199 2200 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 2201 ZFS_SA_BASE_ATTR_SIZE); 2202 2203 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2204 if (error) { 2205 zfs_dirent_unlock(dl); 2206 if (error == ERESTART) { 2207 waited = B_TRUE; 2208 dmu_tx_wait(tx); 2209 dmu_tx_abort(tx); 2210 goto top; 2211 } 2212 zfs_acl_ids_free(&acl_ids); 2213 dmu_tx_abort(tx); 2214 getnewvnode_drop_reserve(); 2215 ZFS_EXIT(zfsvfs); 2216 return (error); 2217 } 2218 2219 /* 2220 * Create new node. 2221 */ 2222 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 2223 2224 if (fuid_dirtied) 2225 zfs_fuid_sync(zfsvfs, tx); 2226 2227 /* 2228 * Now put new name in parent dir. 2229 */ 2230 (void) zfs_link_create(dl, zp, tx, ZNEW); 2231 2232 *vpp = ZTOV(zp); 2233 2234 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 2235 if (flags & FIGNORECASE) 2236 txtype |= TX_CI; 2237 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 2238 acl_ids.z_fuidp, vap); 2239 2240 zfs_acl_ids_free(&acl_ids); 2241 2242 dmu_tx_commit(tx); 2243 2244 getnewvnode_drop_reserve(); 2245 2246 zfs_dirent_unlock(dl); 2247 2248 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2249 zil_commit(zilog, 0); 2250 2251 ZFS_EXIT(zfsvfs); 2252 return (0); 2253} 2254 2255/* 2256 * Remove a directory subdir entry. If the current working 2257 * directory is the same as the subdir to be removed, the 2258 * remove will fail. 2259 * 2260 * IN: dvp - vnode of directory to remove from. 2261 * name - name of directory to be removed. 2262 * cwd - vnode of current working directory. 2263 * cr - credentials of caller. 2264 * ct - caller context 2265 * flags - case flags 2266 * 2267 * RETURN: 0 on success, error code on failure. 2268 * 2269 * Timestamps: 2270 * dvp - ctime|mtime updated 2271 */ 2272/*ARGSUSED*/ 2273static int 2274zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 2275 caller_context_t *ct, int flags) 2276{ 2277 znode_t *dzp = VTOZ(dvp); 2278 znode_t *zp; 2279 vnode_t *vp; 2280 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2281 zilog_t *zilog; 2282 zfs_dirlock_t *dl; 2283 dmu_tx_t *tx; 2284 int error; 2285 int zflg = ZEXISTS; 2286 boolean_t waited = B_FALSE; 2287 2288 ZFS_ENTER(zfsvfs); 2289 ZFS_VERIFY_ZP(dzp); 2290 zilog = zfsvfs->z_log; 2291 2292 if (flags & FIGNORECASE) 2293 zflg |= ZCILOOK; 2294top: 2295 zp = NULL; 2296 2297 /* 2298 * Attempt to lock directory; fail if entry doesn't exist. 2299 */ 2300 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 2301 NULL, NULL)) { 2302 ZFS_EXIT(zfsvfs); 2303 return (error); 2304 } 2305 2306 vp = ZTOV(zp); 2307 2308 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 2309 goto out; 2310 } 2311 2312 if (vp->v_type != VDIR) { 2313 error = SET_ERROR(ENOTDIR); 2314 goto out; 2315 } 2316 2317 if (vp == cwd) { 2318 error = SET_ERROR(EINVAL); 2319 goto out; 2320 } 2321 2322 vnevent_rmdir(vp, dvp, name, ct); 2323 2324 /* 2325 * Grab a lock on the directory to make sure that noone is 2326 * trying to add (or lookup) entries while we are removing it. 2327 */ 2328 rw_enter(&zp->z_name_lock, RW_WRITER); 2329 2330 /* 2331 * Grab a lock on the parent pointer to make sure we play well 2332 * with the treewalk and directory rename code. 2333 */ 2334 rw_enter(&zp->z_parent_lock, RW_WRITER); 2335 2336 tx = dmu_tx_create(zfsvfs->z_os); 2337 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2338 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2339 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2340 zfs_sa_upgrade_txholds(tx, zp); 2341 zfs_sa_upgrade_txholds(tx, dzp); 2342 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 2343 if (error) { 2344 rw_exit(&zp->z_parent_lock); 2345 rw_exit(&zp->z_name_lock); 2346 zfs_dirent_unlock(dl); 2347 VN_RELE(vp); 2348 if (error == ERESTART) { 2349 waited = B_TRUE; 2350 dmu_tx_wait(tx); 2351 dmu_tx_abort(tx); 2352 goto top; 2353 } 2354 dmu_tx_abort(tx); 2355 ZFS_EXIT(zfsvfs); 2356 return (error); 2357 } 2358 2359#ifdef FREEBSD_NAMECACHE 2360 cache_purge(dvp); 2361#endif 2362 2363 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 2364 2365 if (error == 0) { 2366 uint64_t txtype = TX_RMDIR; 2367 if (flags & FIGNORECASE) 2368 txtype |= TX_CI; 2369 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT); 2370 } 2371 2372 dmu_tx_commit(tx); 2373 2374 rw_exit(&zp->z_parent_lock); 2375 rw_exit(&zp->z_name_lock); 2376#ifdef FREEBSD_NAMECACHE 2377 cache_purge(vp); 2378#endif 2379out: 2380 zfs_dirent_unlock(dl); 2381 2382 VN_RELE(vp); 2383 2384 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2385 zil_commit(zilog, 0); 2386 2387 ZFS_EXIT(zfsvfs); 2388 return (error); 2389} 2390 2391/* 2392 * Read as many directory entries as will fit into the provided 2393 * buffer from the given directory cursor position (specified in 2394 * the uio structure). 2395 * 2396 * IN: vp - vnode of directory to read. 2397 * uio - structure supplying read location, range info, 2398 * and return buffer. 2399 * cr - credentials of caller. 2400 * ct - caller context 2401 * flags - case flags 2402 * 2403 * OUT: uio - updated offset and range, buffer filled. 2404 * eofp - set to true if end-of-file detected. 2405 * 2406 * RETURN: 0 on success, error code on failure. 2407 * 2408 * Timestamps: 2409 * vp - atime updated 2410 * 2411 * Note that the low 4 bits of the cookie returned by zap is always zero. 2412 * This allows us to use the low range for "special" directory entries: 2413 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 2414 * we use the offset 2 for the '.zfs' directory. 2415 */ 2416/* ARGSUSED */ 2417static int 2418zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 2419{ 2420 znode_t *zp = VTOZ(vp); 2421 iovec_t *iovp; 2422 edirent_t *eodp; 2423 dirent64_t *odp; 2424 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2425 objset_t *os; 2426 caddr_t outbuf; 2427 size_t bufsize; 2428 zap_cursor_t zc; 2429 zap_attribute_t zap; 2430 uint_t bytes_wanted; 2431 uint64_t offset; /* must be unsigned; checks for < 1 */ 2432 uint64_t parent; 2433 int local_eof; 2434 int outcount; 2435 int error; 2436 uint8_t prefetch; 2437 boolean_t check_sysattrs; 2438 uint8_t type; 2439 int ncooks; 2440 u_long *cooks = NULL; 2441 int flags = 0; 2442 2443 ZFS_ENTER(zfsvfs); 2444 ZFS_VERIFY_ZP(zp); 2445 2446 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 2447 &parent, sizeof (parent))) != 0) { 2448 ZFS_EXIT(zfsvfs); 2449 return (error); 2450 } 2451 2452 /* 2453 * If we are not given an eof variable, 2454 * use a local one. 2455 */ 2456 if (eofp == NULL) 2457 eofp = &local_eof; 2458 2459 /* 2460 * Check for valid iov_len. 2461 */ 2462 if (uio->uio_iov->iov_len <= 0) { 2463 ZFS_EXIT(zfsvfs); 2464 return (SET_ERROR(EINVAL)); 2465 } 2466 2467 /* 2468 * Quit if directory has been removed (posix) 2469 */ 2470 if ((*eofp = zp->z_unlinked) != 0) { 2471 ZFS_EXIT(zfsvfs); 2472 return (0); 2473 } 2474 2475 error = 0; 2476 os = zfsvfs->z_os; 2477 offset = uio->uio_loffset; 2478 prefetch = zp->z_zn_prefetch; 2479 2480 /* 2481 * Initialize the iterator cursor. 2482 */ 2483 if (offset <= 3) { 2484 /* 2485 * Start iteration from the beginning of the directory. 2486 */ 2487 zap_cursor_init(&zc, os, zp->z_id); 2488 } else { 2489 /* 2490 * The offset is a serialized cursor. 2491 */ 2492 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 2493 } 2494 2495 /* 2496 * Get space to change directory entries into fs independent format. 2497 */ 2498 iovp = uio->uio_iov; 2499 bytes_wanted = iovp->iov_len; 2500 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 2501 bufsize = bytes_wanted; 2502 outbuf = kmem_alloc(bufsize, KM_SLEEP); 2503 odp = (struct dirent64 *)outbuf; 2504 } else { 2505 bufsize = bytes_wanted; 2506 outbuf = NULL; 2507 odp = (struct dirent64 *)iovp->iov_base; 2508 } 2509 eodp = (struct edirent *)odp; 2510 2511 if (ncookies != NULL) { 2512 /* 2513 * Minimum entry size is dirent size and 1 byte for a file name. 2514 */ 2515 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 2516 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 2517 *cookies = cooks; 2518 *ncookies = ncooks; 2519 } 2520 /* 2521 * If this VFS supports the system attribute view interface; and 2522 * we're looking at an extended attribute directory; and we care 2523 * about normalization conflicts on this vfs; then we must check 2524 * for normalization conflicts with the sysattr name space. 2525 */ 2526#ifdef TODO 2527 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 2528 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 2529 (flags & V_RDDIR_ENTFLAGS); 2530#else 2531 check_sysattrs = 0; 2532#endif 2533 2534 /* 2535 * Transform to file-system independent format 2536 */ 2537 outcount = 0; 2538 while (outcount < bytes_wanted) { 2539 ino64_t objnum; 2540 ushort_t reclen; 2541 off64_t *next = NULL; 2542 2543 /* 2544 * Special case `.', `..', and `.zfs'. 2545 */ 2546 if (offset == 0) { 2547 (void) strcpy(zap.za_name, "."); 2548 zap.za_normalization_conflict = 0; 2549 objnum = zp->z_id; 2550 type = DT_DIR; 2551 } else if (offset == 1) { 2552 (void) strcpy(zap.za_name, ".."); 2553 zap.za_normalization_conflict = 0; 2554 objnum = parent; 2555 type = DT_DIR; 2556 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2557 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2558 zap.za_normalization_conflict = 0; 2559 objnum = ZFSCTL_INO_ROOT; 2560 type = DT_DIR; 2561 } else { 2562 /* 2563 * Grab next entry. 2564 */ 2565 if (error = zap_cursor_retrieve(&zc, &zap)) { 2566 if ((*eofp = (error == ENOENT)) != 0) 2567 break; 2568 else 2569 goto update; 2570 } 2571 2572 if (zap.za_integer_length != 8 || 2573 zap.za_num_integers != 1) { 2574 cmn_err(CE_WARN, "zap_readdir: bad directory " 2575 "entry, obj = %lld, offset = %lld\n", 2576 (u_longlong_t)zp->z_id, 2577 (u_longlong_t)offset); 2578 error = SET_ERROR(ENXIO); 2579 goto update; 2580 } 2581 2582 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2583 /* 2584 * MacOS X can extract the object type here such as: 2585 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2586 */ 2587 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2588 2589 if (check_sysattrs && !zap.za_normalization_conflict) { 2590#ifdef TODO 2591 zap.za_normalization_conflict = 2592 xattr_sysattr_casechk(zap.za_name); 2593#else 2594 panic("%s:%u: TODO", __func__, __LINE__); 2595#endif 2596 } 2597 } 2598 2599 if (flags & V_RDDIR_ACCFILTER) { 2600 /* 2601 * If we have no access at all, don't include 2602 * this entry in the returned information 2603 */ 2604 znode_t *ezp; 2605 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2606 goto skip_entry; 2607 if (!zfs_has_access(ezp, cr)) { 2608 VN_RELE(ZTOV(ezp)); 2609 goto skip_entry; 2610 } 2611 VN_RELE(ZTOV(ezp)); 2612 } 2613 2614 if (flags & V_RDDIR_ENTFLAGS) 2615 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2616 else 2617 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2618 2619 /* 2620 * Will this entry fit in the buffer? 2621 */ 2622 if (outcount + reclen > bufsize) { 2623 /* 2624 * Did we manage to fit anything in the buffer? 2625 */ 2626 if (!outcount) { 2627 error = SET_ERROR(EINVAL); 2628 goto update; 2629 } 2630 break; 2631 } 2632 if (flags & V_RDDIR_ENTFLAGS) { 2633 /* 2634 * Add extended flag entry: 2635 */ 2636 eodp->ed_ino = objnum; 2637 eodp->ed_reclen = reclen; 2638 /* NOTE: ed_off is the offset for the *next* entry */ 2639 next = &(eodp->ed_off); 2640 eodp->ed_eflags = zap.za_normalization_conflict ? 2641 ED_CASE_CONFLICT : 0; 2642 (void) strncpy(eodp->ed_name, zap.za_name, 2643 EDIRENT_NAMELEN(reclen)); 2644 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2645 } else { 2646 /* 2647 * Add normal entry: 2648 */ 2649 odp->d_ino = objnum; 2650 odp->d_reclen = reclen; 2651 odp->d_namlen = strlen(zap.za_name); 2652 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 2653 odp->d_type = type; 2654 odp = (dirent64_t *)((intptr_t)odp + reclen); 2655 } 2656 outcount += reclen; 2657 2658 ASSERT(outcount <= bufsize); 2659 2660 /* Prefetch znode */ 2661 if (prefetch) 2662 dmu_prefetch(os, objnum, 0, 0); 2663 2664 skip_entry: 2665 /* 2666 * Move to the next entry, fill in the previous offset. 2667 */ 2668 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2669 zap_cursor_advance(&zc); 2670 offset = zap_cursor_serialize(&zc); 2671 } else { 2672 offset += 1; 2673 } 2674 2675 if (cooks != NULL) { 2676 *cooks++ = offset; 2677 ncooks--; 2678 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 2679 } 2680 } 2681 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2682 2683 /* Subtract unused cookies */ 2684 if (ncookies != NULL) 2685 *ncookies -= ncooks; 2686 2687 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2688 iovp->iov_base += outcount; 2689 iovp->iov_len -= outcount; 2690 uio->uio_resid -= outcount; 2691 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2692 /* 2693 * Reset the pointer. 2694 */ 2695 offset = uio->uio_loffset; 2696 } 2697 2698update: 2699 zap_cursor_fini(&zc); 2700 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2701 kmem_free(outbuf, bufsize); 2702 2703 if (error == ENOENT) 2704 error = 0; 2705 2706 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2707 2708 uio->uio_loffset = offset; 2709 ZFS_EXIT(zfsvfs); 2710 if (error != 0 && cookies != NULL) { 2711 free(*cookies, M_TEMP); 2712 *cookies = NULL; 2713 *ncookies = 0; 2714 } 2715 return (error); 2716} 2717 2718ulong_t zfs_fsync_sync_cnt = 4; 2719 2720static int 2721zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2722{ 2723 znode_t *zp = VTOZ(vp); 2724 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2725 2726 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2727 2728 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { 2729 ZFS_ENTER(zfsvfs); 2730 ZFS_VERIFY_ZP(zp); 2731 zil_commit(zfsvfs->z_log, zp->z_id); 2732 ZFS_EXIT(zfsvfs); 2733 } 2734 return (0); 2735} 2736 2737 2738/* 2739 * Get the requested file attributes and place them in the provided 2740 * vattr structure. 2741 * 2742 * IN: vp - vnode of file. 2743 * vap - va_mask identifies requested attributes. 2744 * If AT_XVATTR set, then optional attrs are requested 2745 * flags - ATTR_NOACLCHECK (CIFS server context) 2746 * cr - credentials of caller. 2747 * ct - caller context 2748 * 2749 * OUT: vap - attribute values. 2750 * 2751 * RETURN: 0 (always succeeds). 2752 */ 2753/* ARGSUSED */ 2754static int 2755zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2756 caller_context_t *ct) 2757{ 2758 znode_t *zp = VTOZ(vp); 2759 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2760 int error = 0; 2761 uint32_t blksize; 2762 u_longlong_t nblocks; 2763 uint64_t links; 2764 uint64_t mtime[2], ctime[2], crtime[2], rdev; 2765 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2766 xoptattr_t *xoap = NULL; 2767 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2768 sa_bulk_attr_t bulk[4]; 2769 int count = 0; 2770 2771 ZFS_ENTER(zfsvfs); 2772 ZFS_VERIFY_ZP(zp); 2773 2774 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2775 2776 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 2777 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 2778 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &crtime, 16); 2779 if (vp->v_type == VBLK || vp->v_type == VCHR) 2780 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, 2781 &rdev, 8); 2782 2783 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 2784 ZFS_EXIT(zfsvfs); 2785 return (error); 2786 } 2787 2788 /* 2789 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2790 * Also, if we are the owner don't bother, since owner should 2791 * always be allowed to read basic attributes of file. 2792 */ 2793 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 2794 (vap->va_uid != crgetuid(cr))) { 2795 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2796 skipaclchk, cr)) { 2797 ZFS_EXIT(zfsvfs); 2798 return (error); 2799 } 2800 } 2801 2802 /* 2803 * Return all attributes. It's cheaper to provide the answer 2804 * than to determine whether we were asked the question. 2805 */ 2806 2807 mutex_enter(&zp->z_lock); 2808 vap->va_type = IFTOVT(zp->z_mode); 2809 vap->va_mode = zp->z_mode & ~S_IFMT; 2810#ifdef sun 2811 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2812#else 2813 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 2814#endif 2815 vap->va_nodeid = zp->z_id; 2816 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2817 links = zp->z_links + 1; 2818 else 2819 links = zp->z_links; 2820 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */ 2821 vap->va_size = zp->z_size; 2822#ifdef sun 2823 vap->va_rdev = vp->v_rdev; 2824#else 2825 if (vp->v_type == VBLK || vp->v_type == VCHR) 2826 vap->va_rdev = zfs_cmpldev(rdev); 2827#endif 2828 vap->va_seq = zp->z_seq; 2829 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2830 vap->va_filerev = zp->z_seq; 2831 2832 /* 2833 * Add in any requested optional attributes and the create time. 2834 * Also set the corresponding bits in the returned attribute bitmap. 2835 */ 2836 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2837 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2838 xoap->xoa_archive = 2839 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2840 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2841 } 2842 2843 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2844 xoap->xoa_readonly = 2845 ((zp->z_pflags & ZFS_READONLY) != 0); 2846 XVA_SET_RTN(xvap, XAT_READONLY); 2847 } 2848 2849 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2850 xoap->xoa_system = 2851 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2852 XVA_SET_RTN(xvap, XAT_SYSTEM); 2853 } 2854 2855 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2856 xoap->xoa_hidden = 2857 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2858 XVA_SET_RTN(xvap, XAT_HIDDEN); 2859 } 2860 2861 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2862 xoap->xoa_nounlink = 2863 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2864 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2865 } 2866 2867 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2868 xoap->xoa_immutable = 2869 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2870 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2871 } 2872 2873 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2874 xoap->xoa_appendonly = 2875 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2876 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2877 } 2878 2879 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2880 xoap->xoa_nodump = 2881 ((zp->z_pflags & ZFS_NODUMP) != 0); 2882 XVA_SET_RTN(xvap, XAT_NODUMP); 2883 } 2884 2885 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2886 xoap->xoa_opaque = 2887 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2888 XVA_SET_RTN(xvap, XAT_OPAQUE); 2889 } 2890 2891 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2892 xoap->xoa_av_quarantined = 2893 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2894 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2895 } 2896 2897 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2898 xoap->xoa_av_modified = 2899 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2900 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2901 } 2902 2903 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2904 vp->v_type == VREG) { 2905 zfs_sa_get_scanstamp(zp, xvap); 2906 } 2907 2908 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2909 uint64_t times[2]; 2910 2911 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 2912 times, sizeof (times)); 2913 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 2914 XVA_SET_RTN(xvap, XAT_CREATETIME); 2915 } 2916 2917 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2918 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2919 XVA_SET_RTN(xvap, XAT_REPARSE); 2920 } 2921 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2922 xoap->xoa_generation = zp->z_gen; 2923 XVA_SET_RTN(xvap, XAT_GEN); 2924 } 2925 2926 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2927 xoap->xoa_offline = 2928 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2929 XVA_SET_RTN(xvap, XAT_OFFLINE); 2930 } 2931 2932 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2933 xoap->xoa_sparse = 2934 ((zp->z_pflags & ZFS_SPARSE) != 0); 2935 XVA_SET_RTN(xvap, XAT_SPARSE); 2936 } 2937 } 2938 2939 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2940 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2941 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2942 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 2943 2944 mutex_exit(&zp->z_lock); 2945 2946 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 2947 vap->va_blksize = blksize; 2948 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2949 2950 if (zp->z_blksz == 0) { 2951 /* 2952 * Block size hasn't been set; suggest maximal I/O transfers. 2953 */ 2954 vap->va_blksize = zfsvfs->z_max_blksz; 2955 } 2956 2957 ZFS_EXIT(zfsvfs); 2958 return (0); 2959} 2960 2961/* 2962 * Set the file attributes to the values contained in the 2963 * vattr structure. 2964 * 2965 * IN: vp - vnode of file to be modified. 2966 * vap - new attribute values. 2967 * If AT_XVATTR set, then optional attrs are being set 2968 * flags - ATTR_UTIME set if non-default time values provided. 2969 * - ATTR_NOACLCHECK (CIFS context only). 2970 * cr - credentials of caller. 2971 * ct - caller context 2972 * 2973 * RETURN: 0 on success, error code on failure. 2974 * 2975 * Timestamps: 2976 * vp - ctime updated, mtime updated if size changed. 2977 */ 2978/* ARGSUSED */ 2979static int 2980zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2981 caller_context_t *ct) 2982{ 2983 znode_t *zp = VTOZ(vp); 2984 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2985 zilog_t *zilog; 2986 dmu_tx_t *tx; 2987 vattr_t oldva; 2988 xvattr_t tmpxvattr; 2989 uint_t mask = vap->va_mask; 2990 uint_t saved_mask = 0; 2991 uint64_t saved_mode; 2992 int trim_mask = 0; 2993 uint64_t new_mode; 2994 uint64_t new_uid, new_gid; 2995 uint64_t xattr_obj; 2996 uint64_t mtime[2], ctime[2]; 2997 znode_t *attrzp; 2998 int need_policy = FALSE; 2999 int err, err2; 3000 zfs_fuid_info_t *fuidp = NULL; 3001 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 3002 xoptattr_t *xoap; 3003 zfs_acl_t *aclp; 3004 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 3005 boolean_t fuid_dirtied = B_FALSE; 3006 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 3007 int count = 0, xattr_count = 0; 3008 3009 if (mask == 0) 3010 return (0); 3011 3012 if (mask & AT_NOSET) 3013 return (SET_ERROR(EINVAL)); 3014 3015 ZFS_ENTER(zfsvfs); 3016 ZFS_VERIFY_ZP(zp); 3017 3018 zilog = zfsvfs->z_log; 3019 3020 /* 3021 * Make sure that if we have ephemeral uid/gid or xvattr specified 3022 * that file system is at proper version level 3023 */ 3024 3025 if (zfsvfs->z_use_fuids == B_FALSE && 3026 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 3027 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 3028 (mask & AT_XVATTR))) { 3029 ZFS_EXIT(zfsvfs); 3030 return (SET_ERROR(EINVAL)); 3031 } 3032 3033 if (mask & AT_SIZE && vp->v_type == VDIR) { 3034 ZFS_EXIT(zfsvfs); 3035 return (SET_ERROR(EISDIR)); 3036 } 3037 3038 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 3039 ZFS_EXIT(zfsvfs); 3040 return (SET_ERROR(EINVAL)); 3041 } 3042 3043 /* 3044 * If this is an xvattr_t, then get a pointer to the structure of 3045 * optional attributes. If this is NULL, then we have a vattr_t. 3046 */ 3047 xoap = xva_getxoptattr(xvap); 3048 3049 xva_init(&tmpxvattr); 3050 3051 /* 3052 * Immutable files can only alter immutable bit and atime 3053 */ 3054 if ((zp->z_pflags & ZFS_IMMUTABLE) && 3055 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 3056 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 3057 ZFS_EXIT(zfsvfs); 3058 return (SET_ERROR(EPERM)); 3059 } 3060 3061 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 3062 ZFS_EXIT(zfsvfs); 3063 return (SET_ERROR(EPERM)); 3064 } 3065 3066 /* 3067 * Verify timestamps doesn't overflow 32 bits. 3068 * ZFS can handle large timestamps, but 32bit syscalls can't 3069 * handle times greater than 2039. This check should be removed 3070 * once large timestamps are fully supported. 3071 */ 3072 if (mask & (AT_ATIME | AT_MTIME)) { 3073 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 3074 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 3075 ZFS_EXIT(zfsvfs); 3076 return (SET_ERROR(EOVERFLOW)); 3077 } 3078 } 3079 3080top: 3081 attrzp = NULL; 3082 aclp = NULL; 3083 3084 /* Can this be moved to before the top label? */ 3085 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 3086 ZFS_EXIT(zfsvfs); 3087 return (SET_ERROR(EROFS)); 3088 } 3089 3090 /* 3091 * First validate permissions 3092 */ 3093 3094 if (mask & AT_SIZE) { 3095 /* 3096 * XXX - Note, we are not providing any open 3097 * mode flags here (like FNDELAY), so we may 3098 * block if there are locks present... this 3099 * should be addressed in openat(). 3100 */ 3101 /* XXX - would it be OK to generate a log record here? */ 3102 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 3103 if (err) { 3104 ZFS_EXIT(zfsvfs); 3105 return (err); 3106 } 3107 } 3108 3109 if (mask & (AT_ATIME|AT_MTIME) || 3110 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 3111 XVA_ISSET_REQ(xvap, XAT_READONLY) || 3112 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 3113 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 3114 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 3115 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 3116 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 3117 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 3118 skipaclchk, cr); 3119 } 3120 3121 if (mask & (AT_UID|AT_GID)) { 3122 int idmask = (mask & (AT_UID|AT_GID)); 3123 int take_owner; 3124 int take_group; 3125 3126 /* 3127 * NOTE: even if a new mode is being set, 3128 * we may clear S_ISUID/S_ISGID bits. 3129 */ 3130 3131 if (!(mask & AT_MODE)) 3132 vap->va_mode = zp->z_mode; 3133 3134 /* 3135 * Take ownership or chgrp to group we are a member of 3136 */ 3137 3138 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 3139 take_group = (mask & AT_GID) && 3140 zfs_groupmember(zfsvfs, vap->va_gid, cr); 3141 3142 /* 3143 * If both AT_UID and AT_GID are set then take_owner and 3144 * take_group must both be set in order to allow taking 3145 * ownership. 3146 * 3147 * Otherwise, send the check through secpolicy_vnode_setattr() 3148 * 3149 */ 3150 3151 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 3152 ((idmask == AT_UID) && take_owner) || 3153 ((idmask == AT_GID) && take_group)) { 3154 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 3155 skipaclchk, cr) == 0) { 3156 /* 3157 * Remove setuid/setgid for non-privileged users 3158 */ 3159 secpolicy_setid_clear(vap, vp, cr); 3160 trim_mask = (mask & (AT_UID|AT_GID)); 3161 } else { 3162 need_policy = TRUE; 3163 } 3164 } else { 3165 need_policy = TRUE; 3166 } 3167 } 3168 3169 mutex_enter(&zp->z_lock); 3170 oldva.va_mode = zp->z_mode; 3171 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 3172 if (mask & AT_XVATTR) { 3173 /* 3174 * Update xvattr mask to include only those attributes 3175 * that are actually changing. 3176 * 3177 * the bits will be restored prior to actually setting 3178 * the attributes so the caller thinks they were set. 3179 */ 3180 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 3181 if (xoap->xoa_appendonly != 3182 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 3183 need_policy = TRUE; 3184 } else { 3185 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 3186 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 3187 } 3188 } 3189 3190 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 3191 if (xoap->xoa_nounlink != 3192 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 3193 need_policy = TRUE; 3194 } else { 3195 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 3196 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 3197 } 3198 } 3199 3200 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 3201 if (xoap->xoa_immutable != 3202 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 3203 need_policy = TRUE; 3204 } else { 3205 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 3206 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 3207 } 3208 } 3209 3210 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 3211 if (xoap->xoa_nodump != 3212 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 3213 need_policy = TRUE; 3214 } else { 3215 XVA_CLR_REQ(xvap, XAT_NODUMP); 3216 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 3217 } 3218 } 3219 3220 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 3221 if (xoap->xoa_av_modified != 3222 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 3223 need_policy = TRUE; 3224 } else { 3225 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 3226 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 3227 } 3228 } 3229 3230 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 3231 if ((vp->v_type != VREG && 3232 xoap->xoa_av_quarantined) || 3233 xoap->xoa_av_quarantined != 3234 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 3235 need_policy = TRUE; 3236 } else { 3237 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 3238 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 3239 } 3240 } 3241 3242 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 3243 mutex_exit(&zp->z_lock); 3244 ZFS_EXIT(zfsvfs); 3245 return (SET_ERROR(EPERM)); 3246 } 3247 3248 if (need_policy == FALSE && 3249 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 3250 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 3251 need_policy = TRUE; 3252 } 3253 } 3254 3255 mutex_exit(&zp->z_lock); 3256 3257 if (mask & AT_MODE) { 3258 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 3259 err = secpolicy_setid_setsticky_clear(vp, vap, 3260 &oldva, cr); 3261 if (err) { 3262 ZFS_EXIT(zfsvfs); 3263 return (err); 3264 } 3265 trim_mask |= AT_MODE; 3266 } else { 3267 need_policy = TRUE; 3268 } 3269 } 3270 3271 if (need_policy) { 3272 /* 3273 * If trim_mask is set then take ownership 3274 * has been granted or write_acl is present and user 3275 * has the ability to modify mode. In that case remove 3276 * UID|GID and or MODE from mask so that 3277 * secpolicy_vnode_setattr() doesn't revoke it. 3278 */ 3279 3280 if (trim_mask) { 3281 saved_mask = vap->va_mask; 3282 vap->va_mask &= ~trim_mask; 3283 if (trim_mask & AT_MODE) { 3284 /* 3285 * Save the mode, as secpolicy_vnode_setattr() 3286 * will overwrite it with ova.va_mode. 3287 */ 3288 saved_mode = vap->va_mode; 3289 } 3290 } 3291 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 3292 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 3293 if (err) { 3294 ZFS_EXIT(zfsvfs); 3295 return (err); 3296 } 3297 3298 if (trim_mask) { 3299 vap->va_mask |= saved_mask; 3300 if (trim_mask & AT_MODE) { 3301 /* 3302 * Recover the mode after 3303 * secpolicy_vnode_setattr(). 3304 */ 3305 vap->va_mode = saved_mode; 3306 } 3307 } 3308 } 3309 3310 /* 3311 * secpolicy_vnode_setattr, or take ownership may have 3312 * changed va_mask 3313 */ 3314 mask = vap->va_mask; 3315 3316 if ((mask & (AT_UID | AT_GID))) { 3317 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 3318 &xattr_obj, sizeof (xattr_obj)); 3319 3320 if (err == 0 && xattr_obj) { 3321 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 3322 if (err) 3323 goto out2; 3324 } 3325 if (mask & AT_UID) { 3326 new_uid = zfs_fuid_create(zfsvfs, 3327 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 3328 if (new_uid != zp->z_uid && 3329 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 3330 if (attrzp) 3331 VN_RELE(ZTOV(attrzp)); 3332 err = SET_ERROR(EDQUOT); 3333 goto out2; 3334 } 3335 } 3336 3337 if (mask & AT_GID) { 3338 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 3339 cr, ZFS_GROUP, &fuidp); 3340 if (new_gid != zp->z_gid && 3341 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 3342 if (attrzp) 3343 VN_RELE(ZTOV(attrzp)); 3344 err = SET_ERROR(EDQUOT); 3345 goto out2; 3346 } 3347 } 3348 } 3349 tx = dmu_tx_create(zfsvfs->z_os); 3350 3351 if (mask & AT_MODE) { 3352 uint64_t pmode = zp->z_mode; 3353 uint64_t acl_obj; 3354 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 3355 3356 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && 3357 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 3358 err = SET_ERROR(EPERM); 3359 goto out; 3360 } 3361 3362 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 3363 goto out; 3364 3365 mutex_enter(&zp->z_lock); 3366 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 3367 /* 3368 * Are we upgrading ACL from old V0 format 3369 * to V1 format? 3370 */ 3371 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 3372 zfs_znode_acl_version(zp) == 3373 ZFS_ACL_VERSION_INITIAL) { 3374 dmu_tx_hold_free(tx, acl_obj, 0, 3375 DMU_OBJECT_END); 3376 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3377 0, aclp->z_acl_bytes); 3378 } else { 3379 dmu_tx_hold_write(tx, acl_obj, 0, 3380 aclp->z_acl_bytes); 3381 } 3382 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3383 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3384 0, aclp->z_acl_bytes); 3385 } 3386 mutex_exit(&zp->z_lock); 3387 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3388 } else { 3389 if ((mask & AT_XVATTR) && 3390 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3391 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3392 else 3393 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3394 } 3395 3396 if (attrzp) { 3397 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 3398 } 3399 3400 fuid_dirtied = zfsvfs->z_fuid_dirty; 3401 if (fuid_dirtied) 3402 zfs_fuid_txhold(zfsvfs, tx); 3403 3404 zfs_sa_upgrade_txholds(tx, zp); 3405 3406 err = dmu_tx_assign(tx, TXG_WAIT); 3407 if (err) 3408 goto out; 3409 3410 count = 0; 3411 /* 3412 * Set each attribute requested. 3413 * We group settings according to the locks they need to acquire. 3414 * 3415 * Note: you cannot set ctime directly, although it will be 3416 * updated as a side-effect of calling this function. 3417 */ 3418 3419 3420 if (mask & (AT_UID|AT_GID|AT_MODE)) 3421 mutex_enter(&zp->z_acl_lock); 3422 mutex_enter(&zp->z_lock); 3423 3424 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3425 &zp->z_pflags, sizeof (zp->z_pflags)); 3426 3427 if (attrzp) { 3428 if (mask & (AT_UID|AT_GID|AT_MODE)) 3429 mutex_enter(&attrzp->z_acl_lock); 3430 mutex_enter(&attrzp->z_lock); 3431 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3432 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3433 sizeof (attrzp->z_pflags)); 3434 } 3435 3436 if (mask & (AT_UID|AT_GID)) { 3437 3438 if (mask & AT_UID) { 3439 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3440 &new_uid, sizeof (new_uid)); 3441 zp->z_uid = new_uid; 3442 if (attrzp) { 3443 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3444 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3445 sizeof (new_uid)); 3446 attrzp->z_uid = new_uid; 3447 } 3448 } 3449 3450 if (mask & AT_GID) { 3451 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3452 NULL, &new_gid, sizeof (new_gid)); 3453 zp->z_gid = new_gid; 3454 if (attrzp) { 3455 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3456 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 3457 sizeof (new_gid)); 3458 attrzp->z_gid = new_gid; 3459 } 3460 } 3461 if (!(mask & AT_MODE)) { 3462 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 3463 NULL, &new_mode, sizeof (new_mode)); 3464 new_mode = zp->z_mode; 3465 } 3466 err = zfs_acl_chown_setattr(zp); 3467 ASSERT(err == 0); 3468 if (attrzp) { 3469 err = zfs_acl_chown_setattr(attrzp); 3470 ASSERT(err == 0); 3471 } 3472 } 3473 3474 if (mask & AT_MODE) { 3475 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 3476 &new_mode, sizeof (new_mode)); 3477 zp->z_mode = new_mode; 3478 ASSERT3U((uintptr_t)aclp, !=, 0); 3479 err = zfs_aclset_common(zp, aclp, cr, tx); 3480 ASSERT0(err); 3481 if (zp->z_acl_cached) 3482 zfs_acl_free(zp->z_acl_cached); 3483 zp->z_acl_cached = aclp; 3484 aclp = NULL; 3485 } 3486 3487 3488 if (mask & AT_ATIME) { 3489 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 3490 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 3491 &zp->z_atime, sizeof (zp->z_atime)); 3492 } 3493 3494 if (mask & AT_MTIME) { 3495 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 3496 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 3497 mtime, sizeof (mtime)); 3498 } 3499 3500 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 3501 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 3502 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 3503 NULL, mtime, sizeof (mtime)); 3504 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3505 &ctime, sizeof (ctime)); 3506 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 3507 B_TRUE); 3508 } else if (mask != 0) { 3509 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3510 &ctime, sizeof (ctime)); 3511 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime, 3512 B_TRUE); 3513 if (attrzp) { 3514 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3515 SA_ZPL_CTIME(zfsvfs), NULL, 3516 &ctime, sizeof (ctime)); 3517 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 3518 mtime, ctime, B_TRUE); 3519 } 3520 } 3521 /* 3522 * Do this after setting timestamps to prevent timestamp 3523 * update from toggling bit 3524 */ 3525 3526 if (xoap && (mask & AT_XVATTR)) { 3527 3528 /* 3529 * restore trimmed off masks 3530 * so that return masks can be set for caller. 3531 */ 3532 3533 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 3534 XVA_SET_REQ(xvap, XAT_APPENDONLY); 3535 } 3536 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 3537 XVA_SET_REQ(xvap, XAT_NOUNLINK); 3538 } 3539 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 3540 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 3541 } 3542 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 3543 XVA_SET_REQ(xvap, XAT_NODUMP); 3544 } 3545 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 3546 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 3547 } 3548 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 3549 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 3550 } 3551 3552 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3553 ASSERT(vp->v_type == VREG); 3554 3555 zfs_xvattr_set(zp, xvap, tx); 3556 } 3557 3558 if (fuid_dirtied) 3559 zfs_fuid_sync(zfsvfs, tx); 3560 3561 if (mask != 0) 3562 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 3563 3564 mutex_exit(&zp->z_lock); 3565 if (mask & (AT_UID|AT_GID|AT_MODE)) 3566 mutex_exit(&zp->z_acl_lock); 3567 3568 if (attrzp) { 3569 if (mask & (AT_UID|AT_GID|AT_MODE)) 3570 mutex_exit(&attrzp->z_acl_lock); 3571 mutex_exit(&attrzp->z_lock); 3572 } 3573out: 3574 if (err == 0 && attrzp) { 3575 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 3576 xattr_count, tx); 3577 ASSERT(err2 == 0); 3578 } 3579 3580 if (attrzp) 3581 VN_RELE(ZTOV(attrzp)); 3582 3583 if (aclp) 3584 zfs_acl_free(aclp); 3585 3586 if (fuidp) { 3587 zfs_fuid_info_free(fuidp); 3588 fuidp = NULL; 3589 } 3590 3591 if (err) { 3592 dmu_tx_abort(tx); 3593 if (err == ERESTART) 3594 goto top; 3595 } else { 3596 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3597 dmu_tx_commit(tx); 3598 } 3599 3600out2: 3601 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3602 zil_commit(zilog, 0); 3603 3604 ZFS_EXIT(zfsvfs); 3605 return (err); 3606} 3607 3608typedef struct zfs_zlock { 3609 krwlock_t *zl_rwlock; /* lock we acquired */ 3610 znode_t *zl_znode; /* znode we held */ 3611 struct zfs_zlock *zl_next; /* next in list */ 3612} zfs_zlock_t; 3613 3614/* 3615 * Drop locks and release vnodes that were held by zfs_rename_lock(). 3616 */ 3617static void 3618zfs_rename_unlock(zfs_zlock_t **zlpp) 3619{ 3620 zfs_zlock_t *zl; 3621 3622 while ((zl = *zlpp) != NULL) { 3623 if (zl->zl_znode != NULL) 3624 VN_RELE(ZTOV(zl->zl_znode)); 3625 rw_exit(zl->zl_rwlock); 3626 *zlpp = zl->zl_next; 3627 kmem_free(zl, sizeof (*zl)); 3628 } 3629} 3630 3631/* 3632 * Search back through the directory tree, using the ".." entries. 3633 * Lock each directory in the chain to prevent concurrent renames. 3634 * Fail any attempt to move a directory into one of its own descendants. 3635 * XXX - z_parent_lock can overlap with map or grow locks 3636 */ 3637static int 3638zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 3639{ 3640 zfs_zlock_t *zl; 3641 znode_t *zp = tdzp; 3642 uint64_t rootid = zp->z_zfsvfs->z_root; 3643 uint64_t oidp = zp->z_id; 3644 krwlock_t *rwlp = &szp->z_parent_lock; 3645 krw_t rw = RW_WRITER; 3646 3647 /* 3648 * First pass write-locks szp and compares to zp->z_id. 3649 * Later passes read-lock zp and compare to zp->z_parent. 3650 */ 3651 do { 3652 if (!rw_tryenter(rwlp, rw)) { 3653 /* 3654 * Another thread is renaming in this path. 3655 * Note that if we are a WRITER, we don't have any 3656 * parent_locks held yet. 3657 */ 3658 if (rw == RW_READER && zp->z_id > szp->z_id) { 3659 /* 3660 * Drop our locks and restart 3661 */ 3662 zfs_rename_unlock(&zl); 3663 *zlpp = NULL; 3664 zp = tdzp; 3665 oidp = zp->z_id; 3666 rwlp = &szp->z_parent_lock; 3667 rw = RW_WRITER; 3668 continue; 3669 } else { 3670 /* 3671 * Wait for other thread to drop its locks 3672 */ 3673 rw_enter(rwlp, rw); 3674 } 3675 } 3676 3677 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 3678 zl->zl_rwlock = rwlp; 3679 zl->zl_znode = NULL; 3680 zl->zl_next = *zlpp; 3681 *zlpp = zl; 3682 3683 if (oidp == szp->z_id) /* We're a descendant of szp */ 3684 return (SET_ERROR(EINVAL)); 3685 3686 if (oidp == rootid) /* We've hit the top */ 3687 return (0); 3688 3689 if (rw == RW_READER) { /* i.e. not the first pass */ 3690 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp); 3691 if (error) 3692 return (error); 3693 zl->zl_znode = zp; 3694 } 3695 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs), 3696 &oidp, sizeof (oidp)); 3697 rwlp = &zp->z_parent_lock; 3698 rw = RW_READER; 3699 3700 } while (zp->z_id != sdzp->z_id); 3701 3702 return (0); 3703} 3704 3705/* 3706 * Move an entry from the provided source directory to the target 3707 * directory. Change the entry name as indicated. 3708 * 3709 * IN: sdvp - Source directory containing the "old entry". 3710 * snm - Old entry name. 3711 * tdvp - Target directory to contain the "new entry". 3712 * tnm - New entry name. 3713 * cr - credentials of caller. 3714 * ct - caller context 3715 * flags - case flags 3716 * 3717 * RETURN: 0 on success, error code on failure. 3718 * 3719 * Timestamps: 3720 * sdvp,tdvp - ctime|mtime updated 3721 */ 3722/*ARGSUSED*/ 3723static int 3724zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 3725 caller_context_t *ct, int flags) 3726{ 3727 znode_t *tdzp, *szp, *tzp; 3728 znode_t *sdzp = VTOZ(sdvp); 3729 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 3730 zilog_t *zilog; 3731 vnode_t *realvp; 3732 zfs_dirlock_t *sdl, *tdl; 3733 dmu_tx_t *tx; 3734 zfs_zlock_t *zl; 3735 int cmp, serr, terr; 3736 int error = 0; 3737 int zflg = 0; 3738 boolean_t waited = B_FALSE; 3739 3740 ZFS_ENTER(zfsvfs); 3741 ZFS_VERIFY_ZP(sdzp); 3742 zilog = zfsvfs->z_log; 3743 3744 /* 3745 * Make sure we have the real vp for the target directory. 3746 */ 3747 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 3748 tdvp = realvp; 3749 3750 tdzp = VTOZ(tdvp); 3751 ZFS_VERIFY_ZP(tdzp); 3752 3753 /* 3754 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 3755 * ctldir appear to have the same v_vfsp. 3756 */ 3757 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) { 3758 ZFS_EXIT(zfsvfs); 3759 return (SET_ERROR(EXDEV)); 3760 } 3761 3762 if (zfsvfs->z_utf8 && u8_validate(tnm, 3763 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3764 ZFS_EXIT(zfsvfs); 3765 return (SET_ERROR(EILSEQ)); 3766 } 3767 3768 if (flags & FIGNORECASE) 3769 zflg |= ZCILOOK; 3770 3771top: 3772 szp = NULL; 3773 tzp = NULL; 3774 zl = NULL; 3775 3776 /* 3777 * This is to prevent the creation of links into attribute space 3778 * by renaming a linked file into/outof an attribute directory. 3779 * See the comment in zfs_link() for why this is considered bad. 3780 */ 3781 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3782 ZFS_EXIT(zfsvfs); 3783 return (SET_ERROR(EINVAL)); 3784 } 3785 3786 /* 3787 * Lock source and target directory entries. To prevent deadlock, 3788 * a lock ordering must be defined. We lock the directory with 3789 * the smallest object id first, or if it's a tie, the one with 3790 * the lexically first name. 3791 */ 3792 if (sdzp->z_id < tdzp->z_id) { 3793 cmp = -1; 3794 } else if (sdzp->z_id > tdzp->z_id) { 3795 cmp = 1; 3796 } else { 3797 /* 3798 * First compare the two name arguments without 3799 * considering any case folding. 3800 */ 3801 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 3802 3803 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 3804 ASSERT(error == 0 || !zfsvfs->z_utf8); 3805 if (cmp == 0) { 3806 /* 3807 * POSIX: "If the old argument and the new argument 3808 * both refer to links to the same existing file, 3809 * the rename() function shall return successfully 3810 * and perform no other action." 3811 */ 3812 ZFS_EXIT(zfsvfs); 3813 return (0); 3814 } 3815 /* 3816 * If the file system is case-folding, then we may 3817 * have some more checking to do. A case-folding file 3818 * system is either supporting mixed case sensitivity 3819 * access or is completely case-insensitive. Note 3820 * that the file system is always case preserving. 3821 * 3822 * In mixed sensitivity mode case sensitive behavior 3823 * is the default. FIGNORECASE must be used to 3824 * explicitly request case insensitive behavior. 3825 * 3826 * If the source and target names provided differ only 3827 * by case (e.g., a request to rename 'tim' to 'Tim'), 3828 * we will treat this as a special case in the 3829 * case-insensitive mode: as long as the source name 3830 * is an exact match, we will allow this to proceed as 3831 * a name-change request. 3832 */ 3833 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3834 (zfsvfs->z_case == ZFS_CASE_MIXED && 3835 flags & FIGNORECASE)) && 3836 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3837 &error) == 0) { 3838 /* 3839 * case preserving rename request, require exact 3840 * name matches 3841 */ 3842 zflg |= ZCIEXACT; 3843 zflg &= ~ZCILOOK; 3844 } 3845 } 3846 3847 /* 3848 * If the source and destination directories are the same, we should 3849 * grab the z_name_lock of that directory only once. 3850 */ 3851 if (sdzp == tdzp) { 3852 zflg |= ZHAVELOCK; 3853 rw_enter(&sdzp->z_name_lock, RW_READER); 3854 } 3855 3856 if (cmp < 0) { 3857 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3858 ZEXISTS | zflg, NULL, NULL); 3859 terr = zfs_dirent_lock(&tdl, 3860 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3861 } else { 3862 terr = zfs_dirent_lock(&tdl, 3863 tdzp, tnm, &tzp, zflg, NULL, NULL); 3864 serr = zfs_dirent_lock(&sdl, 3865 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3866 NULL, NULL); 3867 } 3868 3869 if (serr) { 3870 /* 3871 * Source entry invalid or not there. 3872 */ 3873 if (!terr) { 3874 zfs_dirent_unlock(tdl); 3875 if (tzp) 3876 VN_RELE(ZTOV(tzp)); 3877 } 3878 3879 if (sdzp == tdzp) 3880 rw_exit(&sdzp->z_name_lock); 3881 3882 /* 3883 * FreeBSD: In OpenSolaris they only check if rename source is 3884 * ".." here, because "." is handled in their lookup. This is 3885 * not the case for FreeBSD, so we check for "." explicitly. 3886 */ 3887 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 3888 serr = SET_ERROR(EINVAL); 3889 ZFS_EXIT(zfsvfs); 3890 return (serr); 3891 } 3892 if (terr) { 3893 zfs_dirent_unlock(sdl); 3894 VN_RELE(ZTOV(szp)); 3895 3896 if (sdzp == tdzp) 3897 rw_exit(&sdzp->z_name_lock); 3898 3899 if (strcmp(tnm, "..") == 0) 3900 terr = SET_ERROR(EINVAL); 3901 ZFS_EXIT(zfsvfs); 3902 return (terr); 3903 } 3904 3905 /* 3906 * Must have write access at the source to remove the old entry 3907 * and write access at the target to create the new entry. 3908 * Note that if target and source are the same, this can be 3909 * done in a single check. 3910 */ 3911 3912 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3913 goto out; 3914 3915 if (ZTOV(szp)->v_type == VDIR) { 3916 /* 3917 * Check to make sure rename is valid. 3918 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3919 */ 3920 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3921 goto out; 3922 } 3923 3924 /* 3925 * Does target exist? 3926 */ 3927 if (tzp) { 3928 /* 3929 * Source and target must be the same type. 3930 */ 3931 if (ZTOV(szp)->v_type == VDIR) { 3932 if (ZTOV(tzp)->v_type != VDIR) { 3933 error = SET_ERROR(ENOTDIR); 3934 goto out; 3935 } 3936 } else { 3937 if (ZTOV(tzp)->v_type == VDIR) { 3938 error = SET_ERROR(EISDIR); 3939 goto out; 3940 } 3941 } 3942 /* 3943 * POSIX dictates that when the source and target 3944 * entries refer to the same file object, rename 3945 * must do nothing and exit without error. 3946 */ 3947 if (szp->z_id == tzp->z_id) { 3948 error = 0; 3949 goto out; 3950 } 3951 } 3952 3953 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3954 if (tzp) 3955 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3956 3957 /* 3958 * notify the target directory if it is not the same 3959 * as source directory. 3960 */ 3961 if (tdvp != sdvp) { 3962 vnevent_rename_dest_dir(tdvp, ct); 3963 } 3964 3965 tx = dmu_tx_create(zfsvfs->z_os); 3966 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3967 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 3968 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3969 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3970 if (sdzp != tdzp) { 3971 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 3972 zfs_sa_upgrade_txholds(tx, tdzp); 3973 } 3974 if (tzp) { 3975 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 3976 zfs_sa_upgrade_txholds(tx, tzp); 3977 } 3978 3979 zfs_sa_upgrade_txholds(tx, szp); 3980 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3981 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 3982 if (error) { 3983 if (zl != NULL) 3984 zfs_rename_unlock(&zl); 3985 zfs_dirent_unlock(sdl); 3986 zfs_dirent_unlock(tdl); 3987 3988 if (sdzp == tdzp) 3989 rw_exit(&sdzp->z_name_lock); 3990 3991 VN_RELE(ZTOV(szp)); 3992 if (tzp) 3993 VN_RELE(ZTOV(tzp)); 3994 if (error == ERESTART) { 3995 waited = B_TRUE; 3996 dmu_tx_wait(tx); 3997 dmu_tx_abort(tx); 3998 goto top; 3999 } 4000 dmu_tx_abort(tx); 4001 ZFS_EXIT(zfsvfs); 4002 return (error); 4003 } 4004 4005 if (tzp) /* Attempt to remove the existing target */ 4006 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 4007 4008 if (error == 0) { 4009 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 4010 if (error == 0) { 4011 szp->z_pflags |= ZFS_AV_MODIFIED; 4012 4013 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 4014 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 4015 ASSERT0(error); 4016 4017 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 4018 if (error == 0) { 4019 zfs_log_rename(zilog, tx, TX_RENAME | 4020 (flags & FIGNORECASE ? TX_CI : 0), sdzp, 4021 sdl->dl_name, tdzp, tdl->dl_name, szp); 4022 4023 /* 4024 * Update path information for the target vnode 4025 */ 4026 vn_renamepath(tdvp, ZTOV(szp), tnm, 4027 strlen(tnm)); 4028 } else { 4029 /* 4030 * At this point, we have successfully created 4031 * the target name, but have failed to remove 4032 * the source name. Since the create was done 4033 * with the ZRENAMING flag, there are 4034 * complications; for one, the link count is 4035 * wrong. The easiest way to deal with this 4036 * is to remove the newly created target, and 4037 * return the original error. This must 4038 * succeed; fortunately, it is very unlikely to 4039 * fail, since we just created it. 4040 */ 4041 VERIFY3U(zfs_link_destroy(tdl, szp, tx, 4042 ZRENAMING, NULL), ==, 0); 4043 } 4044 } 4045#ifdef FREEBSD_NAMECACHE 4046 if (error == 0) { 4047 cache_purge(sdvp); 4048 cache_purge(tdvp); 4049 cache_purge(ZTOV(szp)); 4050 if (tzp) 4051 cache_purge(ZTOV(tzp)); 4052 } 4053#endif 4054 } 4055 4056 dmu_tx_commit(tx); 4057out: 4058 if (zl != NULL) 4059 zfs_rename_unlock(&zl); 4060 4061 zfs_dirent_unlock(sdl); 4062 zfs_dirent_unlock(tdl); 4063 4064 if (sdzp == tdzp) 4065 rw_exit(&sdzp->z_name_lock); 4066 4067 4068 VN_RELE(ZTOV(szp)); 4069 if (tzp) 4070 VN_RELE(ZTOV(tzp)); 4071 4072 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4073 zil_commit(zilog, 0); 4074 4075 ZFS_EXIT(zfsvfs); 4076 4077 return (error); 4078} 4079 4080/* 4081 * Insert the indicated symbolic reference entry into the directory. 4082 * 4083 * IN: dvp - Directory to contain new symbolic link. 4084 * link - Name for new symlink entry. 4085 * vap - Attributes of new entry. 4086 * cr - credentials of caller. 4087 * ct - caller context 4088 * flags - case flags 4089 * 4090 * RETURN: 0 on success, error code on failure. 4091 * 4092 * Timestamps: 4093 * dvp - ctime|mtime updated 4094 */ 4095/*ARGSUSED*/ 4096static int 4097zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, 4098 cred_t *cr, kthread_t *td) 4099{ 4100 znode_t *zp, *dzp = VTOZ(dvp); 4101 zfs_dirlock_t *dl; 4102 dmu_tx_t *tx; 4103 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4104 zilog_t *zilog; 4105 uint64_t len = strlen(link); 4106 int error; 4107 int zflg = ZNEW; 4108 zfs_acl_ids_t acl_ids; 4109 boolean_t fuid_dirtied; 4110 uint64_t txtype = TX_SYMLINK; 4111 boolean_t waited = B_FALSE; 4112 int flags = 0; 4113 4114 ASSERT(vap->va_type == VLNK); 4115 4116 ZFS_ENTER(zfsvfs); 4117 ZFS_VERIFY_ZP(dzp); 4118 zilog = zfsvfs->z_log; 4119 4120 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 4121 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4122 ZFS_EXIT(zfsvfs); 4123 return (SET_ERROR(EILSEQ)); 4124 } 4125 if (flags & FIGNORECASE) 4126 zflg |= ZCILOOK; 4127 4128 if (len > MAXPATHLEN) { 4129 ZFS_EXIT(zfsvfs); 4130 return (SET_ERROR(ENAMETOOLONG)); 4131 } 4132 4133 if ((error = zfs_acl_ids_create(dzp, 0, 4134 vap, cr, NULL, &acl_ids)) != 0) { 4135 ZFS_EXIT(zfsvfs); 4136 return (error); 4137 } 4138 4139 getnewvnode_reserve(1); 4140 4141top: 4142 /* 4143 * Attempt to lock directory; fail if entry already exists. 4144 */ 4145 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 4146 if (error) { 4147 zfs_acl_ids_free(&acl_ids); 4148 getnewvnode_drop_reserve(); 4149 ZFS_EXIT(zfsvfs); 4150 return (error); 4151 } 4152 4153 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4154 zfs_acl_ids_free(&acl_ids); 4155 zfs_dirent_unlock(dl); 4156 getnewvnode_drop_reserve(); 4157 ZFS_EXIT(zfsvfs); 4158 return (error); 4159 } 4160 4161 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 4162 zfs_acl_ids_free(&acl_ids); 4163 zfs_dirent_unlock(dl); 4164 getnewvnode_drop_reserve(); 4165 ZFS_EXIT(zfsvfs); 4166 return (SET_ERROR(EDQUOT)); 4167 } 4168 tx = dmu_tx_create(zfsvfs->z_os); 4169 fuid_dirtied = zfsvfs->z_fuid_dirty; 4170 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 4171 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4172 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 4173 ZFS_SA_BASE_ATTR_SIZE + len); 4174 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 4175 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 4176 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 4177 acl_ids.z_aclp->z_acl_bytes); 4178 } 4179 if (fuid_dirtied) 4180 zfs_fuid_txhold(zfsvfs, tx); 4181 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4182 if (error) { 4183 zfs_dirent_unlock(dl); 4184 if (error == ERESTART) { 4185 waited = B_TRUE; 4186 dmu_tx_wait(tx); 4187 dmu_tx_abort(tx); 4188 goto top; 4189 } 4190 zfs_acl_ids_free(&acl_ids); 4191 dmu_tx_abort(tx); 4192 getnewvnode_drop_reserve(); 4193 ZFS_EXIT(zfsvfs); 4194 return (error); 4195 } 4196 4197 /* 4198 * Create a new object for the symlink. 4199 * for version 4 ZPL datsets the symlink will be an SA attribute 4200 */ 4201 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 4202 4203 if (fuid_dirtied) 4204 zfs_fuid_sync(zfsvfs, tx); 4205 4206 mutex_enter(&zp->z_lock); 4207 if (zp->z_is_sa) 4208 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 4209 link, len, tx); 4210 else 4211 zfs_sa_symlink(zp, link, len, tx); 4212 mutex_exit(&zp->z_lock); 4213 4214 zp->z_size = len; 4215 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 4216 &zp->z_size, sizeof (zp->z_size), tx); 4217 /* 4218 * Insert the new object into the directory. 4219 */ 4220 (void) zfs_link_create(dl, zp, tx, ZNEW); 4221 4222 if (flags & FIGNORECASE) 4223 txtype |= TX_CI; 4224 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 4225 *vpp = ZTOV(zp); 4226 4227 zfs_acl_ids_free(&acl_ids); 4228 4229 dmu_tx_commit(tx); 4230 4231 getnewvnode_drop_reserve(); 4232 4233 zfs_dirent_unlock(dl); 4234 4235 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4236 zil_commit(zilog, 0); 4237 4238 ZFS_EXIT(zfsvfs); 4239 return (error); 4240} 4241 4242/* 4243 * Return, in the buffer contained in the provided uio structure, 4244 * the symbolic path referred to by vp. 4245 * 4246 * IN: vp - vnode of symbolic link. 4247 * uio - structure to contain the link path. 4248 * cr - credentials of caller. 4249 * ct - caller context 4250 * 4251 * OUT: uio - structure containing the link path. 4252 * 4253 * RETURN: 0 on success, error code on failure. 4254 * 4255 * Timestamps: 4256 * vp - atime updated 4257 */ 4258/* ARGSUSED */ 4259static int 4260zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 4261{ 4262 znode_t *zp = VTOZ(vp); 4263 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4264 int error; 4265 4266 ZFS_ENTER(zfsvfs); 4267 ZFS_VERIFY_ZP(zp); 4268 4269 mutex_enter(&zp->z_lock); 4270 if (zp->z_is_sa) 4271 error = sa_lookup_uio(zp->z_sa_hdl, 4272 SA_ZPL_SYMLINK(zfsvfs), uio); 4273 else 4274 error = zfs_sa_readlink(zp, uio); 4275 mutex_exit(&zp->z_lock); 4276 4277 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4278 4279 ZFS_EXIT(zfsvfs); 4280 return (error); 4281} 4282 4283/* 4284 * Insert a new entry into directory tdvp referencing svp. 4285 * 4286 * IN: tdvp - Directory to contain new entry. 4287 * svp - vnode of new entry. 4288 * name - name of new entry. 4289 * cr - credentials of caller. 4290 * ct - caller context 4291 * 4292 * RETURN: 0 on success, error code on failure. 4293 * 4294 * Timestamps: 4295 * tdvp - ctime|mtime updated 4296 * svp - ctime updated 4297 */ 4298/* ARGSUSED */ 4299static int 4300zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 4301 caller_context_t *ct, int flags) 4302{ 4303 znode_t *dzp = VTOZ(tdvp); 4304 znode_t *tzp, *szp; 4305 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4306 zilog_t *zilog; 4307 zfs_dirlock_t *dl; 4308 dmu_tx_t *tx; 4309 vnode_t *realvp; 4310 int error; 4311 int zf = ZNEW; 4312 uint64_t parent; 4313 uid_t owner; 4314 boolean_t waited = B_FALSE; 4315 4316 ASSERT(tdvp->v_type == VDIR); 4317 4318 ZFS_ENTER(zfsvfs); 4319 ZFS_VERIFY_ZP(dzp); 4320 zilog = zfsvfs->z_log; 4321 4322 if (VOP_REALVP(svp, &realvp, ct) == 0) 4323 svp = realvp; 4324 4325 /* 4326 * POSIX dictates that we return EPERM here. 4327 * Better choices include ENOTSUP or EISDIR. 4328 */ 4329 if (svp->v_type == VDIR) { 4330 ZFS_EXIT(zfsvfs); 4331 return (SET_ERROR(EPERM)); 4332 } 4333 4334 szp = VTOZ(svp); 4335 ZFS_VERIFY_ZP(szp); 4336 4337 /* 4338 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 4339 * ctldir appear to have the same v_vfsp. 4340 */ 4341 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) { 4342 ZFS_EXIT(zfsvfs); 4343 return (SET_ERROR(EXDEV)); 4344 } 4345 4346 /* Prevent links to .zfs/shares files */ 4347 4348 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 4349 &parent, sizeof (uint64_t))) != 0) { 4350 ZFS_EXIT(zfsvfs); 4351 return (error); 4352 } 4353 if (parent == zfsvfs->z_shares_dir) { 4354 ZFS_EXIT(zfsvfs); 4355 return (SET_ERROR(EPERM)); 4356 } 4357 4358 if (zfsvfs->z_utf8 && u8_validate(name, 4359 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4360 ZFS_EXIT(zfsvfs); 4361 return (SET_ERROR(EILSEQ)); 4362 } 4363 if (flags & FIGNORECASE) 4364 zf |= ZCILOOK; 4365 4366 /* 4367 * We do not support links between attributes and non-attributes 4368 * because of the potential security risk of creating links 4369 * into "normal" file space in order to circumvent restrictions 4370 * imposed in attribute space. 4371 */ 4372 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 4373 ZFS_EXIT(zfsvfs); 4374 return (SET_ERROR(EINVAL)); 4375 } 4376 4377 4378 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 4379 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) { 4380 ZFS_EXIT(zfsvfs); 4381 return (SET_ERROR(EPERM)); 4382 } 4383 4384 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4385 ZFS_EXIT(zfsvfs); 4386 return (error); 4387 } 4388 4389top: 4390 /* 4391 * Attempt to lock directory; fail if entry already exists. 4392 */ 4393 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 4394 if (error) { 4395 ZFS_EXIT(zfsvfs); 4396 return (error); 4397 } 4398 4399 tx = dmu_tx_create(zfsvfs->z_os); 4400 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4401 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4402 zfs_sa_upgrade_txholds(tx, szp); 4403 zfs_sa_upgrade_txholds(tx, dzp); 4404 error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT); 4405 if (error) { 4406 zfs_dirent_unlock(dl); 4407 if (error == ERESTART) { 4408 waited = B_TRUE; 4409 dmu_tx_wait(tx); 4410 dmu_tx_abort(tx); 4411 goto top; 4412 } 4413 dmu_tx_abort(tx); 4414 ZFS_EXIT(zfsvfs); 4415 return (error); 4416 } 4417 4418 error = zfs_link_create(dl, szp, tx, 0); 4419 4420 if (error == 0) { 4421 uint64_t txtype = TX_LINK; 4422 if (flags & FIGNORECASE) 4423 txtype |= TX_CI; 4424 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 4425 } 4426 4427 dmu_tx_commit(tx); 4428 4429 zfs_dirent_unlock(dl); 4430 4431 if (error == 0) { 4432 vnevent_link(svp, ct); 4433 } 4434 4435 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4436 zil_commit(zilog, 0); 4437 4438 ZFS_EXIT(zfsvfs); 4439 return (error); 4440} 4441 4442#ifdef sun 4443/* 4444 * zfs_null_putapage() is used when the file system has been force 4445 * unmounted. It just drops the pages. 4446 */ 4447/* ARGSUSED */ 4448static int 4449zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4450 size_t *lenp, int flags, cred_t *cr) 4451{ 4452 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 4453 return (0); 4454} 4455 4456/* 4457 * Push a page out to disk, klustering if possible. 4458 * 4459 * IN: vp - file to push page to. 4460 * pp - page to push. 4461 * flags - additional flags. 4462 * cr - credentials of caller. 4463 * 4464 * OUT: offp - start of range pushed. 4465 * lenp - len of range pushed. 4466 * 4467 * RETURN: 0 on success, error code on failure. 4468 * 4469 * NOTE: callers must have locked the page to be pushed. On 4470 * exit, the page (and all other pages in the kluster) must be 4471 * unlocked. 4472 */ 4473/* ARGSUSED */ 4474static int 4475zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4476 size_t *lenp, int flags, cred_t *cr) 4477{ 4478 znode_t *zp = VTOZ(vp); 4479 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4480 dmu_tx_t *tx; 4481 u_offset_t off, koff; 4482 size_t len, klen; 4483 int err; 4484 4485 off = pp->p_offset; 4486 len = PAGESIZE; 4487 /* 4488 * If our blocksize is bigger than the page size, try to kluster 4489 * multiple pages so that we write a full block (thus avoiding 4490 * a read-modify-write). 4491 */ 4492 if (off < zp->z_size && zp->z_blksz > PAGESIZE) { 4493 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 4494 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 4495 ASSERT(koff <= zp->z_size); 4496 if (koff + klen > zp->z_size) 4497 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); 4498 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 4499 } 4500 ASSERT3U(btop(len), ==, btopr(len)); 4501 4502 /* 4503 * Can't push pages past end-of-file. 4504 */ 4505 if (off >= zp->z_size) { 4506 /* ignore all pages */ 4507 err = 0; 4508 goto out; 4509 } else if (off + len > zp->z_size) { 4510 int npages = btopr(zp->z_size - off); 4511 page_t *trunc; 4512 4513 page_list_break(&pp, &trunc, npages); 4514 /* ignore pages past end of file */ 4515 if (trunc) 4516 pvn_write_done(trunc, flags); 4517 len = zp->z_size - off; 4518 } 4519 4520 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4521 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4522 err = SET_ERROR(EDQUOT); 4523 goto out; 4524 } 4525 tx = dmu_tx_create(zfsvfs->z_os); 4526 dmu_tx_hold_write(tx, zp->z_id, off, len); 4527 4528 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4529 zfs_sa_upgrade_txholds(tx, zp); 4530 err = dmu_tx_assign(tx, TXG_WAIT); 4531 if (err != 0) { 4532 dmu_tx_abort(tx); 4533 goto out; 4534 } 4535 4536 if (zp->z_blksz <= PAGESIZE) { 4537 caddr_t va = zfs_map_page(pp, S_READ); 4538 ASSERT3U(len, <=, PAGESIZE); 4539 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 4540 zfs_unmap_page(pp, va); 4541 } else { 4542 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 4543 } 4544 4545 if (err == 0) { 4546 uint64_t mtime[2], ctime[2]; 4547 sa_bulk_attr_t bulk[3]; 4548 int count = 0; 4549 4550 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4551 &mtime, 16); 4552 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4553 &ctime, 16); 4554 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4555 &zp->z_pflags, 8); 4556 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4557 B_TRUE); 4558 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4559 } 4560 dmu_tx_commit(tx); 4561 4562out: 4563 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 4564 if (offp) 4565 *offp = off; 4566 if (lenp) 4567 *lenp = len; 4568 4569 return (err); 4570} 4571 4572/* 4573 * Copy the portion of the file indicated from pages into the file. 4574 * The pages are stored in a page list attached to the files vnode. 4575 * 4576 * IN: vp - vnode of file to push page data to. 4577 * off - position in file to put data. 4578 * len - amount of data to write. 4579 * flags - flags to control the operation. 4580 * cr - credentials of caller. 4581 * ct - caller context. 4582 * 4583 * RETURN: 0 on success, error code on failure. 4584 * 4585 * Timestamps: 4586 * vp - ctime|mtime updated 4587 */ 4588/*ARGSUSED*/ 4589static int 4590zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4591 caller_context_t *ct) 4592{ 4593 znode_t *zp = VTOZ(vp); 4594 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4595 page_t *pp; 4596 size_t io_len; 4597 u_offset_t io_off; 4598 uint_t blksz; 4599 rl_t *rl; 4600 int error = 0; 4601 4602 ZFS_ENTER(zfsvfs); 4603 ZFS_VERIFY_ZP(zp); 4604 4605 /* 4606 * Align this request to the file block size in case we kluster. 4607 * XXX - this can result in pretty aggresive locking, which can 4608 * impact simultanious read/write access. One option might be 4609 * to break up long requests (len == 0) into block-by-block 4610 * operations to get narrower locking. 4611 */ 4612 blksz = zp->z_blksz; 4613 if (ISP2(blksz)) 4614 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 4615 else 4616 io_off = 0; 4617 if (len > 0 && ISP2(blksz)) 4618 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 4619 else 4620 io_len = 0; 4621 4622 if (io_len == 0) { 4623 /* 4624 * Search the entire vp list for pages >= io_off. 4625 */ 4626 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 4627 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 4628 goto out; 4629 } 4630 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 4631 4632 if (off > zp->z_size) { 4633 /* past end of file */ 4634 zfs_range_unlock(rl); 4635 ZFS_EXIT(zfsvfs); 4636 return (0); 4637 } 4638 4639 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); 4640 4641 for (off = io_off; io_off < off + len; io_off += io_len) { 4642 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 4643 pp = page_lookup(vp, io_off, 4644 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 4645 } else { 4646 pp = page_lookup_nowait(vp, io_off, 4647 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 4648 } 4649 4650 if (pp != NULL && pvn_getdirty(pp, flags)) { 4651 int err; 4652 4653 /* 4654 * Found a dirty page to push 4655 */ 4656 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 4657 if (err) 4658 error = err; 4659 } else { 4660 io_len = PAGESIZE; 4661 } 4662 } 4663out: 4664 zfs_range_unlock(rl); 4665 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4666 zil_commit(zfsvfs->z_log, zp->z_id); 4667 ZFS_EXIT(zfsvfs); 4668 return (error); 4669} 4670#endif /* sun */ 4671 4672/*ARGSUSED*/ 4673void 4674zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4675{ 4676 znode_t *zp = VTOZ(vp); 4677 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4678 int error; 4679 4680 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4681 if (zp->z_sa_hdl == NULL) { 4682 /* 4683 * The fs has been unmounted, or we did a 4684 * suspend/resume and this file no longer exists. 4685 */ 4686 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4687 vrecycle(vp); 4688 return; 4689 } 4690 4691 mutex_enter(&zp->z_lock); 4692 if (zp->z_unlinked) { 4693 /* 4694 * Fast path to recycle a vnode of a removed file. 4695 */ 4696 mutex_exit(&zp->z_lock); 4697 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4698 vrecycle(vp); 4699 return; 4700 } 4701 mutex_exit(&zp->z_lock); 4702 4703 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4704 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4705 4706 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4707 zfs_sa_upgrade_txholds(tx, zp); 4708 error = dmu_tx_assign(tx, TXG_WAIT); 4709 if (error) { 4710 dmu_tx_abort(tx); 4711 } else { 4712 mutex_enter(&zp->z_lock); 4713 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4714 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4715 zp->z_atime_dirty = 0; 4716 mutex_exit(&zp->z_lock); 4717 dmu_tx_commit(tx); 4718 } 4719 } 4720 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4721} 4722 4723#ifdef sun 4724/* 4725 * Bounds-check the seek operation. 4726 * 4727 * IN: vp - vnode seeking within 4728 * ooff - old file offset 4729 * noffp - pointer to new file offset 4730 * ct - caller context 4731 * 4732 * RETURN: 0 on success, EINVAL if new offset invalid. 4733 */ 4734/* ARGSUSED */ 4735static int 4736zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 4737 caller_context_t *ct) 4738{ 4739 if (vp->v_type == VDIR) 4740 return (0); 4741 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4742} 4743 4744/* 4745 * Pre-filter the generic locking function to trap attempts to place 4746 * a mandatory lock on a memory mapped file. 4747 */ 4748static int 4749zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 4750 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 4751{ 4752 znode_t *zp = VTOZ(vp); 4753 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4754 4755 ZFS_ENTER(zfsvfs); 4756 ZFS_VERIFY_ZP(zp); 4757 4758 /* 4759 * We are following the UFS semantics with respect to mapcnt 4760 * here: If we see that the file is mapped already, then we will 4761 * return an error, but we don't worry about races between this 4762 * function and zfs_map(). 4763 */ 4764 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { 4765 ZFS_EXIT(zfsvfs); 4766 return (SET_ERROR(EAGAIN)); 4767 } 4768 ZFS_EXIT(zfsvfs); 4769 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4770} 4771 4772/* 4773 * If we can't find a page in the cache, we will create a new page 4774 * and fill it with file data. For efficiency, we may try to fill 4775 * multiple pages at once (klustering) to fill up the supplied page 4776 * list. Note that the pages to be filled are held with an exclusive 4777 * lock to prevent access by other threads while they are being filled. 4778 */ 4779static int 4780zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 4781 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 4782{ 4783 znode_t *zp = VTOZ(vp); 4784 page_t *pp, *cur_pp; 4785 objset_t *os = zp->z_zfsvfs->z_os; 4786 u_offset_t io_off, total; 4787 size_t io_len; 4788 int err; 4789 4790 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4791 /* 4792 * We only have a single page, don't bother klustering 4793 */ 4794 io_off = off; 4795 io_len = PAGESIZE; 4796 pp = page_create_va(vp, io_off, io_len, 4797 PG_EXCL | PG_WAIT, seg, addr); 4798 } else { 4799 /* 4800 * Try to find enough pages to fill the page list 4801 */ 4802 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4803 &io_len, off, plsz, 0); 4804 } 4805 if (pp == NULL) { 4806 /* 4807 * The page already exists, nothing to do here. 4808 */ 4809 *pl = NULL; 4810 return (0); 4811 } 4812 4813 /* 4814 * Fill the pages in the kluster. 4815 */ 4816 cur_pp = pp; 4817 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4818 caddr_t va; 4819 4820 ASSERT3U(io_off, ==, cur_pp->p_offset); 4821 va = zfs_map_page(cur_pp, S_WRITE); 4822 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4823 DMU_READ_PREFETCH); 4824 zfs_unmap_page(cur_pp, va); 4825 if (err) { 4826 /* On error, toss the entire kluster */ 4827 pvn_read_done(pp, B_ERROR); 4828 /* convert checksum errors into IO errors */ 4829 if (err == ECKSUM) 4830 err = SET_ERROR(EIO); 4831 return (err); 4832 } 4833 cur_pp = cur_pp->p_next; 4834 } 4835 4836 /* 4837 * Fill in the page list array from the kluster starting 4838 * from the desired offset `off'. 4839 * NOTE: the page list will always be null terminated. 4840 */ 4841 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4842 ASSERT(pl == NULL || (*pl)->p_offset == off); 4843 4844 return (0); 4845} 4846 4847/* 4848 * Return pointers to the pages for the file region [off, off + len] 4849 * in the pl array. If plsz is greater than len, this function may 4850 * also return page pointers from after the specified region 4851 * (i.e. the region [off, off + plsz]). These additional pages are 4852 * only returned if they are already in the cache, or were created as 4853 * part of a klustered read. 4854 * 4855 * IN: vp - vnode of file to get data from. 4856 * off - position in file to get data from. 4857 * len - amount of data to retrieve. 4858 * plsz - length of provided page list. 4859 * seg - segment to obtain pages for. 4860 * addr - virtual address of fault. 4861 * rw - mode of created pages. 4862 * cr - credentials of caller. 4863 * ct - caller context. 4864 * 4865 * OUT: protp - protection mode of created pages. 4866 * pl - list of pages created. 4867 * 4868 * RETURN: 0 on success, error code on failure. 4869 * 4870 * Timestamps: 4871 * vp - atime updated 4872 */ 4873/* ARGSUSED */ 4874static int 4875zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4876 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4877 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4878{ 4879 znode_t *zp = VTOZ(vp); 4880 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4881 page_t **pl0 = pl; 4882 int err = 0; 4883 4884 /* we do our own caching, faultahead is unnecessary */ 4885 if (pl == NULL) 4886 return (0); 4887 else if (len > plsz) 4888 len = plsz; 4889 else 4890 len = P2ROUNDUP(len, PAGESIZE); 4891 ASSERT(plsz >= len); 4892 4893 ZFS_ENTER(zfsvfs); 4894 ZFS_VERIFY_ZP(zp); 4895 4896 if (protp) 4897 *protp = PROT_ALL; 4898 4899 /* 4900 * Loop through the requested range [off, off + len) looking 4901 * for pages. If we don't find a page, we will need to create 4902 * a new page and fill it with data from the file. 4903 */ 4904 while (len > 0) { 4905 if (*pl = page_lookup(vp, off, SE_SHARED)) 4906 *(pl+1) = NULL; 4907 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 4908 goto out; 4909 while (*pl) { 4910 ASSERT3U((*pl)->p_offset, ==, off); 4911 off += PAGESIZE; 4912 addr += PAGESIZE; 4913 if (len > 0) { 4914 ASSERT3U(len, >=, PAGESIZE); 4915 len -= PAGESIZE; 4916 } 4917 ASSERT3U(plsz, >=, PAGESIZE); 4918 plsz -= PAGESIZE; 4919 pl++; 4920 } 4921 } 4922 4923 /* 4924 * Fill out the page array with any pages already in the cache. 4925 */ 4926 while (plsz > 0 && 4927 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 4928 off += PAGESIZE; 4929 plsz -= PAGESIZE; 4930 } 4931out: 4932 if (err) { 4933 /* 4934 * Release any pages we have previously locked. 4935 */ 4936 while (pl > pl0) 4937 page_unlock(*--pl); 4938 } else { 4939 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4940 } 4941 4942 *pl = NULL; 4943 4944 ZFS_EXIT(zfsvfs); 4945 return (err); 4946} 4947 4948/* 4949 * Request a memory map for a section of a file. This code interacts 4950 * with common code and the VM system as follows: 4951 * 4952 * - common code calls mmap(), which ends up in smmap_common() 4953 * - this calls VOP_MAP(), which takes you into (say) zfs 4954 * - zfs_map() calls as_map(), passing segvn_create() as the callback 4955 * - segvn_create() creates the new segment and calls VOP_ADDMAP() 4956 * - zfs_addmap() updates z_mapcnt 4957 */ 4958/*ARGSUSED*/ 4959static int 4960zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4961 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4962 caller_context_t *ct) 4963{ 4964 znode_t *zp = VTOZ(vp); 4965 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4966 segvn_crargs_t vn_a; 4967 int error; 4968 4969 ZFS_ENTER(zfsvfs); 4970 ZFS_VERIFY_ZP(zp); 4971 4972 if ((prot & PROT_WRITE) && (zp->z_pflags & 4973 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 4974 ZFS_EXIT(zfsvfs); 4975 return (SET_ERROR(EPERM)); 4976 } 4977 4978 if ((prot & (PROT_READ | PROT_EXEC)) && 4979 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 4980 ZFS_EXIT(zfsvfs); 4981 return (SET_ERROR(EACCES)); 4982 } 4983 4984 if (vp->v_flag & VNOMAP) { 4985 ZFS_EXIT(zfsvfs); 4986 return (SET_ERROR(ENOSYS)); 4987 } 4988 4989 if (off < 0 || len > MAXOFFSET_T - off) { 4990 ZFS_EXIT(zfsvfs); 4991 return (SET_ERROR(ENXIO)); 4992 } 4993 4994 if (vp->v_type != VREG) { 4995 ZFS_EXIT(zfsvfs); 4996 return (SET_ERROR(ENODEV)); 4997 } 4998 4999 /* 5000 * If file is locked, disallow mapping. 5001 */ 5002 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { 5003 ZFS_EXIT(zfsvfs); 5004 return (SET_ERROR(EAGAIN)); 5005 } 5006 5007 as_rangelock(as); 5008 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 5009 if (error != 0) { 5010 as_rangeunlock(as); 5011 ZFS_EXIT(zfsvfs); 5012 return (error); 5013 } 5014 5015 vn_a.vp = vp; 5016 vn_a.offset = (u_offset_t)off; 5017 vn_a.type = flags & MAP_TYPE; 5018 vn_a.prot = prot; 5019 vn_a.maxprot = maxprot; 5020 vn_a.cred = cr; 5021 vn_a.amp = NULL; 5022 vn_a.flags = flags & ~MAP_TYPE; 5023 vn_a.szc = 0; 5024 vn_a.lgrp_mem_policy_flags = 0; 5025 5026 error = as_map(as, *addrp, len, segvn_create, &vn_a); 5027 5028 as_rangeunlock(as); 5029 ZFS_EXIT(zfsvfs); 5030 return (error); 5031} 5032 5033/* ARGSUSED */ 5034static int 5035zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5036 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 5037 caller_context_t *ct) 5038{ 5039 uint64_t pages = btopr(len); 5040 5041 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 5042 return (0); 5043} 5044 5045/* 5046 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 5047 * more accurate mtime for the associated file. Since we don't have a way of 5048 * detecting when the data was actually modified, we have to resort to 5049 * heuristics. If an explicit msync() is done, then we mark the mtime when the 5050 * last page is pushed. The problem occurs when the msync() call is omitted, 5051 * which by far the most common case: 5052 * 5053 * open() 5054 * mmap() 5055 * <modify memory> 5056 * munmap() 5057 * close() 5058 * <time lapse> 5059 * putpage() via fsflush 5060 * 5061 * If we wait until fsflush to come along, we can have a modification time that 5062 * is some arbitrary point in the future. In order to prevent this in the 5063 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 5064 * torn down. 5065 */ 5066/* ARGSUSED */ 5067static int 5068zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 5069 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 5070 caller_context_t *ct) 5071{ 5072 uint64_t pages = btopr(len); 5073 5074 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 5075 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 5076 5077 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 5078 vn_has_cached_data(vp)) 5079 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 5080 5081 return (0); 5082} 5083 5084/* 5085 * Free or allocate space in a file. Currently, this function only 5086 * supports the `F_FREESP' command. However, this command is somewhat 5087 * misnamed, as its functionality includes the ability to allocate as 5088 * well as free space. 5089 * 5090 * IN: vp - vnode of file to free data in. 5091 * cmd - action to take (only F_FREESP supported). 5092 * bfp - section of file to free/alloc. 5093 * flag - current file open mode flags. 5094 * offset - current file offset. 5095 * cr - credentials of caller [UNUSED]. 5096 * ct - caller context. 5097 * 5098 * RETURN: 0 on success, error code on failure. 5099 * 5100 * Timestamps: 5101 * vp - ctime|mtime updated 5102 */ 5103/* ARGSUSED */ 5104static int 5105zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 5106 offset_t offset, cred_t *cr, caller_context_t *ct) 5107{ 5108 znode_t *zp = VTOZ(vp); 5109 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5110 uint64_t off, len; 5111 int error; 5112 5113 ZFS_ENTER(zfsvfs); 5114 ZFS_VERIFY_ZP(zp); 5115 5116 if (cmd != F_FREESP) { 5117 ZFS_EXIT(zfsvfs); 5118 return (SET_ERROR(EINVAL)); 5119 } 5120 5121 if (error = convoff(vp, bfp, 0, offset)) { 5122 ZFS_EXIT(zfsvfs); 5123 return (error); 5124 } 5125 5126 if (bfp->l_len < 0) { 5127 ZFS_EXIT(zfsvfs); 5128 return (SET_ERROR(EINVAL)); 5129 } 5130 5131 off = bfp->l_start; 5132 len = bfp->l_len; /* 0 means from off to end of file */ 5133 5134 error = zfs_freesp(zp, off, len, flag, TRUE); 5135 5136 ZFS_EXIT(zfsvfs); 5137 return (error); 5138} 5139#endif /* sun */ 5140 5141CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 5142CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 5143 5144/*ARGSUSED*/ 5145static int 5146zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 5147{ 5148 znode_t *zp = VTOZ(vp); 5149 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5150 uint32_t gen; 5151 uint64_t gen64; 5152 uint64_t object = zp->z_id; 5153 zfid_short_t *zfid; 5154 int size, i, error; 5155 5156 ZFS_ENTER(zfsvfs); 5157 ZFS_VERIFY_ZP(zp); 5158 5159 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 5160 &gen64, sizeof (uint64_t))) != 0) { 5161 ZFS_EXIT(zfsvfs); 5162 return (error); 5163 } 5164 5165 gen = (uint32_t)gen64; 5166 5167 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 5168 5169#ifdef illumos 5170 if (fidp->fid_len < size) { 5171 fidp->fid_len = size; 5172 ZFS_EXIT(zfsvfs); 5173 return (SET_ERROR(ENOSPC)); 5174 } 5175#else 5176 fidp->fid_len = size; 5177#endif 5178 5179 zfid = (zfid_short_t *)fidp; 5180 5181 zfid->zf_len = size; 5182 5183 for (i = 0; i < sizeof (zfid->zf_object); i++) 5184 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 5185 5186 /* Must have a non-zero generation number to distinguish from .zfs */ 5187 if (gen == 0) 5188 gen = 1; 5189 for (i = 0; i < sizeof (zfid->zf_gen); i++) 5190 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 5191 5192 if (size == LONG_FID_LEN) { 5193 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 5194 zfid_long_t *zlfid; 5195 5196 zlfid = (zfid_long_t *)fidp; 5197 5198 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 5199 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 5200 5201 /* XXX - this should be the generation number for the objset */ 5202 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 5203 zlfid->zf_setgen[i] = 0; 5204 } 5205 5206 ZFS_EXIT(zfsvfs); 5207 return (0); 5208} 5209 5210static int 5211zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 5212 caller_context_t *ct) 5213{ 5214 znode_t *zp, *xzp; 5215 zfsvfs_t *zfsvfs; 5216 zfs_dirlock_t *dl; 5217 int error; 5218 5219 switch (cmd) { 5220 case _PC_LINK_MAX: 5221 *valp = INT_MAX; 5222 return (0); 5223 5224 case _PC_FILESIZEBITS: 5225 *valp = 64; 5226 return (0); 5227#ifdef sun 5228 case _PC_XATTR_EXISTS: 5229 zp = VTOZ(vp); 5230 zfsvfs = zp->z_zfsvfs; 5231 ZFS_ENTER(zfsvfs); 5232 ZFS_VERIFY_ZP(zp); 5233 *valp = 0; 5234 error = zfs_dirent_lock(&dl, zp, "", &xzp, 5235 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 5236 if (error == 0) { 5237 zfs_dirent_unlock(dl); 5238 if (!zfs_dirempty(xzp)) 5239 *valp = 1; 5240 VN_RELE(ZTOV(xzp)); 5241 } else if (error == ENOENT) { 5242 /* 5243 * If there aren't extended attributes, it's the 5244 * same as having zero of them. 5245 */ 5246 error = 0; 5247 } 5248 ZFS_EXIT(zfsvfs); 5249 return (error); 5250 5251 case _PC_SATTR_ENABLED: 5252 case _PC_SATTR_EXISTS: 5253 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 5254 (vp->v_type == VREG || vp->v_type == VDIR); 5255 return (0); 5256 5257 case _PC_ACCESS_FILTERING: 5258 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 5259 vp->v_type == VDIR; 5260 return (0); 5261 5262 case _PC_ACL_ENABLED: 5263 *valp = _ACL_ACE_ENABLED; 5264 return (0); 5265#endif /* sun */ 5266 case _PC_MIN_HOLE_SIZE: 5267 *valp = (int)SPA_MINBLOCKSIZE; 5268 return (0); 5269#ifdef sun 5270 case _PC_TIMESTAMP_RESOLUTION: 5271 /* nanosecond timestamp resolution */ 5272 *valp = 1L; 5273 return (0); 5274#endif /* sun */ 5275 case _PC_ACL_EXTENDED: 5276 *valp = 0; 5277 return (0); 5278 5279 case _PC_ACL_NFS4: 5280 *valp = 1; 5281 return (0); 5282 5283 case _PC_ACL_PATH_MAX: 5284 *valp = ACL_MAX_ENTRIES; 5285 return (0); 5286 5287 default: 5288 return (EOPNOTSUPP); 5289 } 5290} 5291 5292/*ARGSUSED*/ 5293static int 5294zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5295 caller_context_t *ct) 5296{ 5297 znode_t *zp = VTOZ(vp); 5298 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5299 int error; 5300 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5301 5302 ZFS_ENTER(zfsvfs); 5303 ZFS_VERIFY_ZP(zp); 5304 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 5305 ZFS_EXIT(zfsvfs); 5306 5307 return (error); 5308} 5309 5310/*ARGSUSED*/ 5311int 5312zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5313 caller_context_t *ct) 5314{ 5315 znode_t *zp = VTOZ(vp); 5316 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5317 int error; 5318 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5319 zilog_t *zilog = zfsvfs->z_log; 5320 5321 ZFS_ENTER(zfsvfs); 5322 ZFS_VERIFY_ZP(zp); 5323 5324 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 5325 5326 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 5327 zil_commit(zilog, 0); 5328 5329 ZFS_EXIT(zfsvfs); 5330 return (error); 5331} 5332 5333#ifdef sun 5334/* 5335 * The smallest read we may consider to loan out an arcbuf. 5336 * This must be a power of 2. 5337 */ 5338int zcr_blksz_min = (1 << 10); /* 1K */ 5339/* 5340 * If set to less than the file block size, allow loaning out of an 5341 * arcbuf for a partial block read. This must be a power of 2. 5342 */ 5343int zcr_blksz_max = (1 << 17); /* 128K */ 5344 5345/*ARGSUSED*/ 5346static int 5347zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr, 5348 caller_context_t *ct) 5349{ 5350 znode_t *zp = VTOZ(vp); 5351 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5352 int max_blksz = zfsvfs->z_max_blksz; 5353 uio_t *uio = &xuio->xu_uio; 5354 ssize_t size = uio->uio_resid; 5355 offset_t offset = uio->uio_loffset; 5356 int blksz; 5357 int fullblk, i; 5358 arc_buf_t *abuf; 5359 ssize_t maxsize; 5360 int preamble, postamble; 5361 5362 if (xuio->xu_type != UIOTYPE_ZEROCOPY) 5363 return (SET_ERROR(EINVAL)); 5364 5365 ZFS_ENTER(zfsvfs); 5366 ZFS_VERIFY_ZP(zp); 5367 switch (ioflag) { 5368 case UIO_WRITE: 5369 /* 5370 * Loan out an arc_buf for write if write size is bigger than 5371 * max_blksz, and the file's block size is also max_blksz. 5372 */ 5373 blksz = max_blksz; 5374 if (size < blksz || zp->z_blksz != blksz) { 5375 ZFS_EXIT(zfsvfs); 5376 return (SET_ERROR(EINVAL)); 5377 } 5378 /* 5379 * Caller requests buffers for write before knowing where the 5380 * write offset might be (e.g. NFS TCP write). 5381 */ 5382 if (offset == -1) { 5383 preamble = 0; 5384 } else { 5385 preamble = P2PHASE(offset, blksz); 5386 if (preamble) { 5387 preamble = blksz - preamble; 5388 size -= preamble; 5389 } 5390 } 5391 5392 postamble = P2PHASE(size, blksz); 5393 size -= postamble; 5394 5395 fullblk = size / blksz; 5396 (void) dmu_xuio_init(xuio, 5397 (preamble != 0) + fullblk + (postamble != 0)); 5398 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble, 5399 int, postamble, int, 5400 (preamble != 0) + fullblk + (postamble != 0)); 5401 5402 /* 5403 * Have to fix iov base/len for partial buffers. They 5404 * currently represent full arc_buf's. 5405 */ 5406 if (preamble) { 5407 /* data begins in the middle of the arc_buf */ 5408 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5409 blksz); 5410 ASSERT(abuf); 5411 (void) dmu_xuio_add(xuio, abuf, 5412 blksz - preamble, preamble); 5413 } 5414 5415 for (i = 0; i < fullblk; i++) { 5416 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5417 blksz); 5418 ASSERT(abuf); 5419 (void) dmu_xuio_add(xuio, abuf, 0, blksz); 5420 } 5421 5422 if (postamble) { 5423 /* data ends in the middle of the arc_buf */ 5424 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5425 blksz); 5426 ASSERT(abuf); 5427 (void) dmu_xuio_add(xuio, abuf, 0, postamble); 5428 } 5429 break; 5430 case UIO_READ: 5431 /* 5432 * Loan out an arc_buf for read if the read size is larger than 5433 * the current file block size. Block alignment is not 5434 * considered. Partial arc_buf will be loaned out for read. 5435 */ 5436 blksz = zp->z_blksz; 5437 if (blksz < zcr_blksz_min) 5438 blksz = zcr_blksz_min; 5439 if (blksz > zcr_blksz_max) 5440 blksz = zcr_blksz_max; 5441 /* avoid potential complexity of dealing with it */ 5442 if (blksz > max_blksz) { 5443 ZFS_EXIT(zfsvfs); 5444 return (SET_ERROR(EINVAL)); 5445 } 5446 5447 maxsize = zp->z_size - uio->uio_loffset; 5448 if (size > maxsize) 5449 size = maxsize; 5450 5451 if (size < blksz || vn_has_cached_data(vp)) { 5452 ZFS_EXIT(zfsvfs); 5453 return (SET_ERROR(EINVAL)); 5454 } 5455 break; 5456 default: 5457 ZFS_EXIT(zfsvfs); 5458 return (SET_ERROR(EINVAL)); 5459 } 5460 5461 uio->uio_extflg = UIO_XUIO; 5462 XUIO_XUZC_RW(xuio) = ioflag; 5463 ZFS_EXIT(zfsvfs); 5464 return (0); 5465} 5466 5467/*ARGSUSED*/ 5468static int 5469zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct) 5470{ 5471 int i; 5472 arc_buf_t *abuf; 5473 int ioflag = XUIO_XUZC_RW(xuio); 5474 5475 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY); 5476 5477 i = dmu_xuio_cnt(xuio); 5478 while (i-- > 0) { 5479 abuf = dmu_xuio_arcbuf(xuio, i); 5480 /* 5481 * if abuf == NULL, it must be a write buffer 5482 * that has been returned in zfs_write(). 5483 */ 5484 if (abuf) 5485 dmu_return_arcbuf(abuf); 5486 ASSERT(abuf || ioflag == UIO_WRITE); 5487 } 5488 5489 dmu_xuio_fini(xuio); 5490 return (0); 5491} 5492 5493/* 5494 * Predeclare these here so that the compiler assumes that 5495 * this is an "old style" function declaration that does 5496 * not include arguments => we won't get type mismatch errors 5497 * in the initializations that follow. 5498 */ 5499static int zfs_inval(); 5500static int zfs_isdir(); 5501 5502static int 5503zfs_inval() 5504{ 5505 return (SET_ERROR(EINVAL)); 5506} 5507 5508static int 5509zfs_isdir() 5510{ 5511 return (SET_ERROR(EISDIR)); 5512} 5513/* 5514 * Directory vnode operations template 5515 */ 5516vnodeops_t *zfs_dvnodeops; 5517const fs_operation_def_t zfs_dvnodeops_template[] = { 5518 VOPNAME_OPEN, { .vop_open = zfs_open }, 5519 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5520 VOPNAME_READ, { .error = zfs_isdir }, 5521 VOPNAME_WRITE, { .error = zfs_isdir }, 5522 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5523 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5524 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5525 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5526 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5527 VOPNAME_CREATE, { .vop_create = zfs_create }, 5528 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5529 VOPNAME_LINK, { .vop_link = zfs_link }, 5530 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5531 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 5532 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5533 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5534 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 5535 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5536 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5537 VOPNAME_FID, { .vop_fid = zfs_fid }, 5538 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5539 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5540 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5541 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5542 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5543 NULL, NULL 5544}; 5545 5546/* 5547 * Regular file vnode operations template 5548 */ 5549vnodeops_t *zfs_fvnodeops; 5550const fs_operation_def_t zfs_fvnodeops_template[] = { 5551 VOPNAME_OPEN, { .vop_open = zfs_open }, 5552 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5553 VOPNAME_READ, { .vop_read = zfs_read }, 5554 VOPNAME_WRITE, { .vop_write = zfs_write }, 5555 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5556 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5557 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5558 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5559 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5560 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5561 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5562 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5563 VOPNAME_FID, { .vop_fid = zfs_fid }, 5564 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5565 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 5566 VOPNAME_SPACE, { .vop_space = zfs_space }, 5567 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 5568 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 5569 VOPNAME_MAP, { .vop_map = zfs_map }, 5570 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 5571 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 5572 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5573 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5574 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5575 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5576 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf }, 5577 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf }, 5578 NULL, NULL 5579}; 5580 5581/* 5582 * Symbolic link vnode operations template 5583 */ 5584vnodeops_t *zfs_symvnodeops; 5585const fs_operation_def_t zfs_symvnodeops_template[] = { 5586 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5587 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5588 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5589 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5590 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 5591 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5592 VOPNAME_FID, { .vop_fid = zfs_fid }, 5593 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5594 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5595 NULL, NULL 5596}; 5597 5598/* 5599 * special share hidden files vnode operations template 5600 */ 5601vnodeops_t *zfs_sharevnodeops; 5602const fs_operation_def_t zfs_sharevnodeops_template[] = { 5603 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5604 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5605 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5606 VOPNAME_FID, { .vop_fid = zfs_fid }, 5607 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5608 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5609 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5610 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5611 NULL, NULL 5612}; 5613 5614/* 5615 * Extended attribute directory vnode operations template 5616 * 5617 * This template is identical to the directory vnodes 5618 * operation template except for restricted operations: 5619 * VOP_MKDIR() 5620 * VOP_SYMLINK() 5621 * 5622 * Note that there are other restrictions embedded in: 5623 * zfs_create() - restrict type to VREG 5624 * zfs_link() - no links into/out of attribute space 5625 * zfs_rename() - no moves into/out of attribute space 5626 */ 5627vnodeops_t *zfs_xdvnodeops; 5628const fs_operation_def_t zfs_xdvnodeops_template[] = { 5629 VOPNAME_OPEN, { .vop_open = zfs_open }, 5630 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5631 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5632 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5633 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5634 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5635 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5636 VOPNAME_CREATE, { .vop_create = zfs_create }, 5637 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5638 VOPNAME_LINK, { .vop_link = zfs_link }, 5639 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5640 VOPNAME_MKDIR, { .error = zfs_inval }, 5641 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5642 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5643 VOPNAME_SYMLINK, { .error = zfs_inval }, 5644 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5645 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5646 VOPNAME_FID, { .vop_fid = zfs_fid }, 5647 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5648 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5649 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5650 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5651 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5652 NULL, NULL 5653}; 5654 5655/* 5656 * Error vnode operations template 5657 */ 5658vnodeops_t *zfs_evnodeops; 5659const fs_operation_def_t zfs_evnodeops_template[] = { 5660 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5661 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5662 NULL, NULL 5663}; 5664#endif /* sun */ 5665 5666static int 5667ioflags(int ioflags) 5668{ 5669 int flags = 0; 5670 5671 if (ioflags & IO_APPEND) 5672 flags |= FAPPEND; 5673 if (ioflags & IO_NDELAY) 5674 flags |= FNONBLOCK; 5675 if (ioflags & IO_SYNC) 5676 flags |= (FSYNC | FDSYNC | FRSYNC); 5677 5678 return (flags); 5679} 5680 5681static int 5682zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage) 5683{ 5684 znode_t *zp = VTOZ(vp); 5685 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5686 objset_t *os = zp->z_zfsvfs->z_os; 5687 vm_page_t mfirst, mlast, mreq; 5688 vm_object_t object; 5689 caddr_t va; 5690 struct sf_buf *sf; 5691 off_t startoff, endoff; 5692 int i, error; 5693 vm_pindex_t reqstart, reqend; 5694 int pcount, lsize, reqsize, size; 5695 5696 ZFS_ENTER(zfsvfs); 5697 ZFS_VERIFY_ZP(zp); 5698 5699 pcount = OFF_TO_IDX(round_page(count)); 5700 mreq = m[reqpage]; 5701 object = mreq->object; 5702 error = 0; 5703 5704 KASSERT(vp->v_object == object, ("mismatching object")); 5705 5706 if (pcount > 1 && zp->z_blksz > PAGESIZE) { 5707 startoff = rounddown(IDX_TO_OFF(mreq->pindex), zp->z_blksz); 5708 reqstart = OFF_TO_IDX(round_page(startoff)); 5709 if (reqstart < m[0]->pindex) 5710 reqstart = 0; 5711 else 5712 reqstart = reqstart - m[0]->pindex; 5713 endoff = roundup(IDX_TO_OFF(mreq->pindex) + PAGE_SIZE, 5714 zp->z_blksz); 5715 reqend = OFF_TO_IDX(trunc_page(endoff)) - 1; 5716 if (reqend > m[pcount - 1]->pindex) 5717 reqend = m[pcount - 1]->pindex; 5718 reqsize = reqend - m[reqstart]->pindex + 1; 5719 KASSERT(reqstart <= reqpage && reqpage < reqstart + reqsize, 5720 ("reqpage beyond [reqstart, reqstart + reqsize[ bounds")); 5721 } else { 5722 reqstart = reqpage; 5723 reqsize = 1; 5724 } 5725 mfirst = m[reqstart]; 5726 mlast = m[reqstart + reqsize - 1]; 5727 5728 zfs_vmobject_wlock(object); 5729 5730 for (i = 0; i < reqstart; i++) { 5731 vm_page_lock(m[i]); 5732 vm_page_free(m[i]); 5733 vm_page_unlock(m[i]); 5734 } 5735 for (i = reqstart + reqsize; i < pcount; i++) { 5736 vm_page_lock(m[i]); 5737 vm_page_free(m[i]); 5738 vm_page_unlock(m[i]); 5739 } 5740 5741 if (mreq->valid && reqsize == 1) { 5742 if (mreq->valid != VM_PAGE_BITS_ALL) 5743 vm_page_zero_invalid(mreq, TRUE); 5744 zfs_vmobject_wunlock(object); 5745 ZFS_EXIT(zfsvfs); 5746 return (zfs_vm_pagerret_ok); 5747 } 5748 5749 PCPU_INC(cnt.v_vnodein); 5750 PCPU_ADD(cnt.v_vnodepgsin, reqsize); 5751 5752 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) { 5753 for (i = reqstart; i < reqstart + reqsize; i++) { 5754 if (i != reqpage) { 5755 vm_page_lock(m[i]); 5756 vm_page_free(m[i]); 5757 vm_page_unlock(m[i]); 5758 } 5759 } 5760 zfs_vmobject_wunlock(object); 5761 ZFS_EXIT(zfsvfs); 5762 return (zfs_vm_pagerret_bad); 5763 } 5764 5765 lsize = PAGE_SIZE; 5766 if (IDX_TO_OFF(mlast->pindex) + lsize > object->un_pager.vnp.vnp_size) 5767 lsize = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mlast->pindex); 5768 5769 zfs_vmobject_wunlock(object); 5770 5771 for (i = reqstart; i < reqstart + reqsize; i++) { 5772 size = PAGE_SIZE; 5773 if (i == (reqstart + reqsize - 1)) 5774 size = lsize; 5775 va = zfs_map_page(m[i], &sf); 5776 error = dmu_read(os, zp->z_id, IDX_TO_OFF(m[i]->pindex), 5777 size, va, DMU_READ_PREFETCH); 5778 if (size != PAGE_SIZE) 5779 bzero(va + size, PAGE_SIZE - size); 5780 zfs_unmap_page(sf); 5781 if (error != 0) 5782 break; 5783 } 5784 5785 zfs_vmobject_wlock(object); 5786 5787 for (i = reqstart; i < reqstart + reqsize; i++) { 5788 if (!error) 5789 m[i]->valid = VM_PAGE_BITS_ALL; 5790 KASSERT(m[i]->dirty == 0, ("zfs_getpages: page %p is dirty", m[i])); 5791 if (i != reqpage) 5792 vm_page_readahead_finish(m[i]); 5793 } 5794 5795 zfs_vmobject_wunlock(object); 5796 5797 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 5798 ZFS_EXIT(zfsvfs); 5799 return (error ? zfs_vm_pagerret_error : zfs_vm_pagerret_ok); 5800} 5801 5802static int 5803zfs_freebsd_getpages(ap) 5804 struct vop_getpages_args /* { 5805 struct vnode *a_vp; 5806 vm_page_t *a_m; 5807 int a_count; 5808 int a_reqpage; 5809 vm_ooffset_t a_offset; 5810 } */ *ap; 5811{ 5812 5813 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage)); 5814} 5815 5816static int 5817zfs_putpages(struct vnode *vp, vm_page_t *ma, size_t len, int flags, 5818 int *rtvals) 5819{ 5820 znode_t *zp = VTOZ(vp); 5821 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5822 rl_t *rl; 5823 dmu_tx_t *tx; 5824 struct sf_buf *sf; 5825 vm_object_t object; 5826 vm_page_t m; 5827 caddr_t va; 5828 size_t tocopy; 5829 size_t lo_len; 5830 vm_ooffset_t lo_off; 5831 vm_ooffset_t off; 5832 uint_t blksz; 5833 int ncount; 5834 int pcount; 5835 int err; 5836 int i; 5837 5838 ZFS_ENTER(zfsvfs); 5839 ZFS_VERIFY_ZP(zp); 5840 5841 object = vp->v_object; 5842 pcount = btoc(len); 5843 ncount = pcount; 5844 5845 KASSERT(ma[0]->object == object, ("mismatching object")); 5846 KASSERT(len > 0 && (len & PAGE_MASK) == 0, ("unexpected length")); 5847 5848 for (i = 0; i < pcount; i++) 5849 rtvals[i] = zfs_vm_pagerret_error; 5850 5851 off = IDX_TO_OFF(ma[0]->pindex); 5852 blksz = zp->z_blksz; 5853 lo_off = rounddown(off, blksz); 5854 lo_len = roundup(len + (off - lo_off), blksz); 5855 rl = zfs_range_lock(zp, lo_off, lo_len, RL_WRITER); 5856 5857 zfs_vmobject_wlock(object); 5858 if (len + off > object->un_pager.vnp.vnp_size) { 5859 if (object->un_pager.vnp.vnp_size > off) { 5860 int pgoff; 5861 5862 len = object->un_pager.vnp.vnp_size - off; 5863 ncount = btoc(len); 5864 if ((pgoff = (int)len & PAGE_MASK) != 0) { 5865 /* 5866 * If the object is locked and the following 5867 * conditions hold, then the page's dirty 5868 * field cannot be concurrently changed by a 5869 * pmap operation. 5870 */ 5871 m = ma[ncount - 1]; 5872 vm_page_assert_sbusied(m); 5873 KASSERT(!pmap_page_is_write_mapped(m), 5874 ("zfs_putpages: page %p is not read-only", m)); 5875 vm_page_clear_dirty(m, pgoff, PAGE_SIZE - 5876 pgoff); 5877 } 5878 } else { 5879 len = 0; 5880 ncount = 0; 5881 } 5882 if (ncount < pcount) { 5883 for (i = ncount; i < pcount; i++) { 5884 rtvals[i] = zfs_vm_pagerret_bad; 5885 } 5886 } 5887 } 5888 zfs_vmobject_wunlock(object); 5889 5890 if (ncount == 0) 5891 goto out; 5892 5893 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 5894 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 5895 goto out; 5896 } 5897 5898top: 5899 tx = dmu_tx_create(zfsvfs->z_os); 5900 dmu_tx_hold_write(tx, zp->z_id, off, len); 5901 5902 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 5903 zfs_sa_upgrade_txholds(tx, zp); 5904 err = dmu_tx_assign(tx, TXG_NOWAIT); 5905 if (err != 0) { 5906 if (err == ERESTART) { 5907 dmu_tx_wait(tx); 5908 dmu_tx_abort(tx); 5909 goto top; 5910 } 5911 dmu_tx_abort(tx); 5912 goto out; 5913 } 5914 5915 if (zp->z_blksz < PAGE_SIZE) { 5916 i = 0; 5917 for (i = 0; len > 0; off += tocopy, len -= tocopy, i++) { 5918 tocopy = len > PAGE_SIZE ? PAGE_SIZE : len; 5919 va = zfs_map_page(ma[i], &sf); 5920 dmu_write(zfsvfs->z_os, zp->z_id, off, tocopy, va, tx); 5921 zfs_unmap_page(sf); 5922 } 5923 } else { 5924 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, ma, tx); 5925 } 5926 5927 if (err == 0) { 5928 uint64_t mtime[2], ctime[2]; 5929 sa_bulk_attr_t bulk[3]; 5930 int count = 0; 5931 5932 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 5933 &mtime, 16); 5934 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 5935 &ctime, 16); 5936 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 5937 &zp->z_pflags, 8); 5938 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 5939 B_TRUE); 5940 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 5941 5942 zfs_vmobject_wlock(object); 5943 for (i = 0; i < ncount; i++) { 5944 rtvals[i] = zfs_vm_pagerret_ok; 5945 vm_page_undirty(ma[i]); 5946 } 5947 zfs_vmobject_wunlock(object); 5948 PCPU_INC(cnt.v_vnodeout); 5949 PCPU_ADD(cnt.v_vnodepgsout, ncount); 5950 } 5951 dmu_tx_commit(tx); 5952 5953out: 5954 zfs_range_unlock(rl); 5955 if ((flags & (zfs_vm_pagerput_sync | zfs_vm_pagerput_inval)) != 0 || 5956 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 5957 zil_commit(zfsvfs->z_log, zp->z_id); 5958 ZFS_EXIT(zfsvfs); 5959 return (rtvals[0]); 5960} 5961 5962int 5963zfs_freebsd_putpages(ap) 5964 struct vop_putpages_args /* { 5965 struct vnode *a_vp; 5966 vm_page_t *a_m; 5967 int a_count; 5968 int a_sync; 5969 int *a_rtvals; 5970 vm_ooffset_t a_offset; 5971 } */ *ap; 5972{ 5973 5974 return (zfs_putpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_sync, 5975 ap->a_rtvals)); 5976} 5977 5978static int 5979zfs_freebsd_bmap(ap) 5980 struct vop_bmap_args /* { 5981 struct vnode *a_vp; 5982 daddr_t a_bn; 5983 struct bufobj **a_bop; 5984 daddr_t *a_bnp; 5985 int *a_runp; 5986 int *a_runb; 5987 } */ *ap; 5988{ 5989 5990 if (ap->a_bop != NULL) 5991 *ap->a_bop = &ap->a_vp->v_bufobj; 5992 if (ap->a_bnp != NULL) 5993 *ap->a_bnp = ap->a_bn; 5994 if (ap->a_runp != NULL) 5995 *ap->a_runp = 0; 5996 if (ap->a_runb != NULL) 5997 *ap->a_runb = 0; 5998 5999 return (0); 6000} 6001 6002static int 6003zfs_freebsd_open(ap) 6004 struct vop_open_args /* { 6005 struct vnode *a_vp; 6006 int a_mode; 6007 struct ucred *a_cred; 6008 struct thread *a_td; 6009 } */ *ap; 6010{ 6011 vnode_t *vp = ap->a_vp; 6012 znode_t *zp = VTOZ(vp); 6013 int error; 6014 6015 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL); 6016 if (error == 0) 6017 vnode_create_vobject(vp, zp->z_size, ap->a_td); 6018 return (error); 6019} 6020 6021static int 6022zfs_freebsd_close(ap) 6023 struct vop_close_args /* { 6024 struct vnode *a_vp; 6025 int a_fflag; 6026 struct ucred *a_cred; 6027 struct thread *a_td; 6028 } */ *ap; 6029{ 6030 6031 return (zfs_close(ap->a_vp, ap->a_fflag, 1, 0, ap->a_cred, NULL)); 6032} 6033 6034static int 6035zfs_freebsd_ioctl(ap) 6036 struct vop_ioctl_args /* { 6037 struct vnode *a_vp; 6038 u_long a_command; 6039 caddr_t a_data; 6040 int a_fflag; 6041 struct ucred *cred; 6042 struct thread *td; 6043 } */ *ap; 6044{ 6045 6046 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 6047 ap->a_fflag, ap->a_cred, NULL, NULL)); 6048} 6049 6050static int 6051zfs_freebsd_read(ap) 6052 struct vop_read_args /* { 6053 struct vnode *a_vp; 6054 struct uio *a_uio; 6055 int a_ioflag; 6056 struct ucred *a_cred; 6057 } */ *ap; 6058{ 6059 6060 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 6061 ap->a_cred, NULL)); 6062} 6063 6064static int 6065zfs_freebsd_write(ap) 6066 struct vop_write_args /* { 6067 struct vnode *a_vp; 6068 struct uio *a_uio; 6069 int a_ioflag; 6070 struct ucred *a_cred; 6071 } */ *ap; 6072{ 6073 6074 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 6075 ap->a_cred, NULL)); 6076} 6077 6078static int 6079zfs_freebsd_access(ap) 6080 struct vop_access_args /* { 6081 struct vnode *a_vp; 6082 accmode_t a_accmode; 6083 struct ucred *a_cred; 6084 struct thread *a_td; 6085 } */ *ap; 6086{ 6087 vnode_t *vp = ap->a_vp; 6088 znode_t *zp = VTOZ(vp); 6089 accmode_t accmode; 6090 int error = 0; 6091 6092 /* 6093 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 6094 */ 6095 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 6096 if (accmode != 0) 6097 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL); 6098 6099 /* 6100 * VADMIN has to be handled by vaccess(). 6101 */ 6102 if (error == 0) { 6103 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 6104 if (accmode != 0) { 6105 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 6106 zp->z_gid, accmode, ap->a_cred, NULL); 6107 } 6108 } 6109 6110 /* 6111 * For VEXEC, ensure that at least one execute bit is set for 6112 * non-directories. 6113 */ 6114 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 6115 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 6116 error = EACCES; 6117 } 6118 6119 return (error); 6120} 6121 6122static int 6123zfs_freebsd_lookup(ap) 6124 struct vop_lookup_args /* { 6125 struct vnode *a_dvp; 6126 struct vnode **a_vpp; 6127 struct componentname *a_cnp; 6128 } */ *ap; 6129{ 6130 struct componentname *cnp = ap->a_cnp; 6131 char nm[NAME_MAX + 1]; 6132 6133 ASSERT(cnp->cn_namelen < sizeof(nm)); 6134 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 6135 6136 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 6137 cnp->cn_cred, cnp->cn_thread, 0)); 6138} 6139 6140static int 6141zfs_freebsd_create(ap) 6142 struct vop_create_args /* { 6143 struct vnode *a_dvp; 6144 struct vnode **a_vpp; 6145 struct componentname *a_cnp; 6146 struct vattr *a_vap; 6147 } */ *ap; 6148{ 6149 struct componentname *cnp = ap->a_cnp; 6150 vattr_t *vap = ap->a_vap; 6151 int mode; 6152 6153 ASSERT(cnp->cn_flags & SAVENAME); 6154 6155 vattr_init_mask(vap); 6156 mode = vap->va_mode & ALLPERMS; 6157 6158 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 6159 ap->a_vpp, cnp->cn_cred, cnp->cn_thread)); 6160} 6161 6162static int 6163zfs_freebsd_remove(ap) 6164 struct vop_remove_args /* { 6165 struct vnode *a_dvp; 6166 struct vnode *a_vp; 6167 struct componentname *a_cnp; 6168 } */ *ap; 6169{ 6170 6171 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6172 6173 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 6174 ap->a_cnp->cn_cred, NULL, 0)); 6175} 6176 6177static int 6178zfs_freebsd_mkdir(ap) 6179 struct vop_mkdir_args /* { 6180 struct vnode *a_dvp; 6181 struct vnode **a_vpp; 6182 struct componentname *a_cnp; 6183 struct vattr *a_vap; 6184 } */ *ap; 6185{ 6186 vattr_t *vap = ap->a_vap; 6187 6188 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 6189 6190 vattr_init_mask(vap); 6191 6192 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 6193 ap->a_cnp->cn_cred, NULL, 0, NULL)); 6194} 6195 6196static int 6197zfs_freebsd_rmdir(ap) 6198 struct vop_rmdir_args /* { 6199 struct vnode *a_dvp; 6200 struct vnode *a_vp; 6201 struct componentname *a_cnp; 6202 } */ *ap; 6203{ 6204 struct componentname *cnp = ap->a_cnp; 6205 6206 ASSERT(cnp->cn_flags & SAVENAME); 6207 6208 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0)); 6209} 6210 6211static int 6212zfs_freebsd_readdir(ap) 6213 struct vop_readdir_args /* { 6214 struct vnode *a_vp; 6215 struct uio *a_uio; 6216 struct ucred *a_cred; 6217 int *a_eofflag; 6218 int *a_ncookies; 6219 u_long **a_cookies; 6220 } */ *ap; 6221{ 6222 6223 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 6224 ap->a_ncookies, ap->a_cookies)); 6225} 6226 6227static int 6228zfs_freebsd_fsync(ap) 6229 struct vop_fsync_args /* { 6230 struct vnode *a_vp; 6231 int a_waitfor; 6232 struct thread *a_td; 6233 } */ *ap; 6234{ 6235 6236 vop_stdfsync(ap); 6237 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL)); 6238} 6239 6240static int 6241zfs_freebsd_getattr(ap) 6242 struct vop_getattr_args /* { 6243 struct vnode *a_vp; 6244 struct vattr *a_vap; 6245 struct ucred *a_cred; 6246 } */ *ap; 6247{ 6248 vattr_t *vap = ap->a_vap; 6249 xvattr_t xvap; 6250 u_long fflags = 0; 6251 int error; 6252 6253 xva_init(&xvap); 6254 xvap.xva_vattr = *vap; 6255 xvap.xva_vattr.va_mask |= AT_XVATTR; 6256 6257 /* Convert chflags into ZFS-type flags. */ 6258 /* XXX: what about SF_SETTABLE?. */ 6259 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 6260 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 6261 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 6262 XVA_SET_REQ(&xvap, XAT_NODUMP); 6263 XVA_SET_REQ(&xvap, XAT_READONLY); 6264 XVA_SET_REQ(&xvap, XAT_ARCHIVE); 6265 XVA_SET_REQ(&xvap, XAT_SYSTEM); 6266 XVA_SET_REQ(&xvap, XAT_HIDDEN); 6267 XVA_SET_REQ(&xvap, XAT_REPARSE); 6268 XVA_SET_REQ(&xvap, XAT_OFFLINE); 6269 XVA_SET_REQ(&xvap, XAT_SPARSE); 6270 6271 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL); 6272 if (error != 0) 6273 return (error); 6274 6275 /* Convert ZFS xattr into chflags. */ 6276#define FLAG_CHECK(fflag, xflag, xfield) do { \ 6277 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 6278 fflags |= (fflag); \ 6279} while (0) 6280 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 6281 xvap.xva_xoptattrs.xoa_immutable); 6282 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 6283 xvap.xva_xoptattrs.xoa_appendonly); 6284 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 6285 xvap.xva_xoptattrs.xoa_nounlink); 6286 FLAG_CHECK(UF_ARCHIVE, XAT_ARCHIVE, 6287 xvap.xva_xoptattrs.xoa_archive); 6288 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 6289 xvap.xva_xoptattrs.xoa_nodump); 6290 FLAG_CHECK(UF_READONLY, XAT_READONLY, 6291 xvap.xva_xoptattrs.xoa_readonly); 6292 FLAG_CHECK(UF_SYSTEM, XAT_SYSTEM, 6293 xvap.xva_xoptattrs.xoa_system); 6294 FLAG_CHECK(UF_HIDDEN, XAT_HIDDEN, 6295 xvap.xva_xoptattrs.xoa_hidden); 6296 FLAG_CHECK(UF_REPARSE, XAT_REPARSE, 6297 xvap.xva_xoptattrs.xoa_reparse); 6298 FLAG_CHECK(UF_OFFLINE, XAT_OFFLINE, 6299 xvap.xva_xoptattrs.xoa_offline); 6300 FLAG_CHECK(UF_SPARSE, XAT_SPARSE, 6301 xvap.xva_xoptattrs.xoa_sparse); 6302 6303#undef FLAG_CHECK 6304 *vap = xvap.xva_vattr; 6305 vap->va_flags = fflags; 6306 return (0); 6307} 6308 6309static int 6310zfs_freebsd_setattr(ap) 6311 struct vop_setattr_args /* { 6312 struct vnode *a_vp; 6313 struct vattr *a_vap; 6314 struct ucred *a_cred; 6315 } */ *ap; 6316{ 6317 vnode_t *vp = ap->a_vp; 6318 vattr_t *vap = ap->a_vap; 6319 cred_t *cred = ap->a_cred; 6320 xvattr_t xvap; 6321 u_long fflags; 6322 uint64_t zflags; 6323 6324 vattr_init_mask(vap); 6325 vap->va_mask &= ~AT_NOSET; 6326 6327 xva_init(&xvap); 6328 xvap.xva_vattr = *vap; 6329 6330 zflags = VTOZ(vp)->z_pflags; 6331 6332 if (vap->va_flags != VNOVAL) { 6333 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 6334 int error; 6335 6336 if (zfsvfs->z_use_fuids == B_FALSE) 6337 return (EOPNOTSUPP); 6338 6339 fflags = vap->va_flags; 6340 /* 6341 * XXX KDM 6342 * We need to figure out whether it makes sense to allow 6343 * UF_REPARSE through, since we don't really have other 6344 * facilities to handle reparse points and zfs_setattr() 6345 * doesn't currently allow setting that attribute anyway. 6346 */ 6347 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_ARCHIVE| 6348 UF_NODUMP|UF_SYSTEM|UF_HIDDEN|UF_READONLY|UF_REPARSE| 6349 UF_OFFLINE|UF_SPARSE)) != 0) 6350 return (EOPNOTSUPP); 6351 /* 6352 * Unprivileged processes are not permitted to unset system 6353 * flags, or modify flags if any system flags are set. 6354 * Privileged non-jail processes may not modify system flags 6355 * if securelevel > 0 and any existing system flags are set. 6356 * Privileged jail processes behave like privileged non-jail 6357 * processes if the security.jail.chflags_allowed sysctl is 6358 * is non-zero; otherwise, they behave like unprivileged 6359 * processes. 6360 */ 6361 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 6362 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) { 6363 if (zflags & 6364 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6365 error = securelevel_gt(cred, 0); 6366 if (error != 0) 6367 return (error); 6368 } 6369 } else { 6370 /* 6371 * Callers may only modify the file flags on objects they 6372 * have VADMIN rights for. 6373 */ 6374 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0) 6375 return (error); 6376 if (zflags & 6377 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 6378 return (EPERM); 6379 } 6380 if (fflags & 6381 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 6382 return (EPERM); 6383 } 6384 } 6385 6386#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 6387 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 6388 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 6389 XVA_SET_REQ(&xvap, (xflag)); \ 6390 (xfield) = ((fflags & (fflag)) != 0); \ 6391 } \ 6392} while (0) 6393 /* Convert chflags into ZFS-type flags. */ 6394 /* XXX: what about SF_SETTABLE?. */ 6395 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 6396 xvap.xva_xoptattrs.xoa_immutable); 6397 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 6398 xvap.xva_xoptattrs.xoa_appendonly); 6399 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 6400 xvap.xva_xoptattrs.xoa_nounlink); 6401 FLAG_CHANGE(UF_ARCHIVE, ZFS_ARCHIVE, XAT_ARCHIVE, 6402 xvap.xva_xoptattrs.xoa_archive); 6403 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 6404 xvap.xva_xoptattrs.xoa_nodump); 6405 FLAG_CHANGE(UF_READONLY, ZFS_READONLY, XAT_READONLY, 6406 xvap.xva_xoptattrs.xoa_readonly); 6407 FLAG_CHANGE(UF_SYSTEM, ZFS_SYSTEM, XAT_SYSTEM, 6408 xvap.xva_xoptattrs.xoa_system); 6409 FLAG_CHANGE(UF_HIDDEN, ZFS_HIDDEN, XAT_HIDDEN, 6410 xvap.xva_xoptattrs.xoa_hidden); 6411 FLAG_CHANGE(UF_REPARSE, ZFS_REPARSE, XAT_REPARSE, 6412 xvap.xva_xoptattrs.xoa_hidden); 6413 FLAG_CHANGE(UF_OFFLINE, ZFS_OFFLINE, XAT_OFFLINE, 6414 xvap.xva_xoptattrs.xoa_offline); 6415 FLAG_CHANGE(UF_SPARSE, ZFS_SPARSE, XAT_SPARSE, 6416 xvap.xva_xoptattrs.xoa_sparse); 6417#undef FLAG_CHANGE 6418 } 6419 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL)); 6420} 6421 6422static int 6423zfs_freebsd_rename(ap) 6424 struct vop_rename_args /* { 6425 struct vnode *a_fdvp; 6426 struct vnode *a_fvp; 6427 struct componentname *a_fcnp; 6428 struct vnode *a_tdvp; 6429 struct vnode *a_tvp; 6430 struct componentname *a_tcnp; 6431 } */ *ap; 6432{ 6433 vnode_t *fdvp = ap->a_fdvp; 6434 vnode_t *fvp = ap->a_fvp; 6435 vnode_t *tdvp = ap->a_tdvp; 6436 vnode_t *tvp = ap->a_tvp; 6437 int error; 6438 6439 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 6440 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 6441 6442 /* 6443 * Check for cross-device rename. 6444 */ 6445 if ((fdvp->v_mount != tdvp->v_mount) || 6446 (tvp && (fdvp->v_mount != tvp->v_mount))) 6447 error = EXDEV; 6448 else 6449 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 6450 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0); 6451 if (tdvp == tvp) 6452 VN_RELE(tdvp); 6453 else 6454 VN_URELE(tdvp); 6455 if (tvp) 6456 VN_URELE(tvp); 6457 VN_RELE(fdvp); 6458 VN_RELE(fvp); 6459 6460 return (error); 6461} 6462 6463static int 6464zfs_freebsd_symlink(ap) 6465 struct vop_symlink_args /* { 6466 struct vnode *a_dvp; 6467 struct vnode **a_vpp; 6468 struct componentname *a_cnp; 6469 struct vattr *a_vap; 6470 char *a_target; 6471 } */ *ap; 6472{ 6473 struct componentname *cnp = ap->a_cnp; 6474 vattr_t *vap = ap->a_vap; 6475 6476 ASSERT(cnp->cn_flags & SAVENAME); 6477 6478 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 6479 vattr_init_mask(vap); 6480 6481 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 6482 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 6483} 6484 6485static int 6486zfs_freebsd_readlink(ap) 6487 struct vop_readlink_args /* { 6488 struct vnode *a_vp; 6489 struct uio *a_uio; 6490 struct ucred *a_cred; 6491 } */ *ap; 6492{ 6493 6494 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL)); 6495} 6496 6497static int 6498zfs_freebsd_link(ap) 6499 struct vop_link_args /* { 6500 struct vnode *a_tdvp; 6501 struct vnode *a_vp; 6502 struct componentname *a_cnp; 6503 } */ *ap; 6504{ 6505 struct componentname *cnp = ap->a_cnp; 6506 vnode_t *vp = ap->a_vp; 6507 vnode_t *tdvp = ap->a_tdvp; 6508 6509 if (tdvp->v_mount != vp->v_mount) 6510 return (EXDEV); 6511 6512 ASSERT(cnp->cn_flags & SAVENAME); 6513 6514 return (zfs_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0)); 6515} 6516 6517static int 6518zfs_freebsd_inactive(ap) 6519 struct vop_inactive_args /* { 6520 struct vnode *a_vp; 6521 struct thread *a_td; 6522 } */ *ap; 6523{ 6524 vnode_t *vp = ap->a_vp; 6525 6526 zfs_inactive(vp, ap->a_td->td_ucred, NULL); 6527 return (0); 6528} 6529 6530static int 6531zfs_freebsd_reclaim(ap) 6532 struct vop_reclaim_args /* { 6533 struct vnode *a_vp; 6534 struct thread *a_td; 6535 } */ *ap; 6536{ 6537 vnode_t *vp = ap->a_vp; 6538 znode_t *zp = VTOZ(vp); 6539 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6540 6541 ASSERT(zp != NULL); 6542 6543 /* Destroy the vm object and flush associated pages. */ 6544 vnode_destroy_vobject(vp); 6545 6546 /* 6547 * z_teardown_inactive_lock protects from a race with 6548 * zfs_znode_dmu_fini in zfsvfs_teardown during 6549 * force unmount. 6550 */ 6551 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 6552 if (zp->z_sa_hdl == NULL) 6553 zfs_znode_free(zp); 6554 else 6555 zfs_zinactive(zp); 6556 rw_exit(&zfsvfs->z_teardown_inactive_lock); 6557 6558 vp->v_data = NULL; 6559 return (0); 6560} 6561 6562static int 6563zfs_freebsd_fid(ap) 6564 struct vop_fid_args /* { 6565 struct vnode *a_vp; 6566 struct fid *a_fid; 6567 } */ *ap; 6568{ 6569 6570 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 6571} 6572 6573static int 6574zfs_freebsd_pathconf(ap) 6575 struct vop_pathconf_args /* { 6576 struct vnode *a_vp; 6577 int a_name; 6578 register_t *a_retval; 6579 } */ *ap; 6580{ 6581 ulong_t val; 6582 int error; 6583 6584 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); 6585 if (error == 0) 6586 *ap->a_retval = val; 6587 else if (error == EOPNOTSUPP) 6588 error = vop_stdpathconf(ap); 6589 return (error); 6590} 6591 6592static int 6593zfs_freebsd_fifo_pathconf(ap) 6594 struct vop_pathconf_args /* { 6595 struct vnode *a_vp; 6596 int a_name; 6597 register_t *a_retval; 6598 } */ *ap; 6599{ 6600 6601 switch (ap->a_name) { 6602 case _PC_ACL_EXTENDED: 6603 case _PC_ACL_NFS4: 6604 case _PC_ACL_PATH_MAX: 6605 case _PC_MAC_PRESENT: 6606 return (zfs_freebsd_pathconf(ap)); 6607 default: 6608 return (fifo_specops.vop_pathconf(ap)); 6609 } 6610} 6611 6612/* 6613 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 6614 * extended attribute name: 6615 * 6616 * NAMESPACE PREFIX 6617 * system freebsd:system: 6618 * user (none, can be used to access ZFS fsattr(5) attributes 6619 * created on Solaris) 6620 */ 6621static int 6622zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 6623 size_t size) 6624{ 6625 const char *namespace, *prefix, *suffix; 6626 6627 /* We don't allow '/' character in attribute name. */ 6628 if (strchr(name, '/') != NULL) 6629 return (EINVAL); 6630 /* We don't allow attribute names that start with "freebsd:" string. */ 6631 if (strncmp(name, "freebsd:", 8) == 0) 6632 return (EINVAL); 6633 6634 bzero(attrname, size); 6635 6636 switch (attrnamespace) { 6637 case EXTATTR_NAMESPACE_USER: 6638#if 0 6639 prefix = "freebsd:"; 6640 namespace = EXTATTR_NAMESPACE_USER_STRING; 6641 suffix = ":"; 6642#else 6643 /* 6644 * This is the default namespace by which we can access all 6645 * attributes created on Solaris. 6646 */ 6647 prefix = namespace = suffix = ""; 6648#endif 6649 break; 6650 case EXTATTR_NAMESPACE_SYSTEM: 6651 prefix = "freebsd:"; 6652 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 6653 suffix = ":"; 6654 break; 6655 case EXTATTR_NAMESPACE_EMPTY: 6656 default: 6657 return (EINVAL); 6658 } 6659 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 6660 name) >= size) { 6661 return (ENAMETOOLONG); 6662 } 6663 return (0); 6664} 6665 6666/* 6667 * Vnode operating to retrieve a named extended attribute. 6668 */ 6669static int 6670zfs_getextattr(struct vop_getextattr_args *ap) 6671/* 6672vop_getextattr { 6673 IN struct vnode *a_vp; 6674 IN int a_attrnamespace; 6675 IN const char *a_name; 6676 INOUT struct uio *a_uio; 6677 OUT size_t *a_size; 6678 IN struct ucred *a_cred; 6679 IN struct thread *a_td; 6680}; 6681*/ 6682{ 6683 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6684 struct thread *td = ap->a_td; 6685 struct nameidata nd; 6686 char attrname[255]; 6687 struct vattr va; 6688 vnode_t *xvp = NULL, *vp; 6689 int error, flags; 6690 6691 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6692 ap->a_cred, ap->a_td, VREAD); 6693 if (error != 0) 6694 return (error); 6695 6696 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6697 sizeof(attrname)); 6698 if (error != 0) 6699 return (error); 6700 6701 ZFS_ENTER(zfsvfs); 6702 6703 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6704 LOOKUP_XATTR); 6705 if (error != 0) { 6706 ZFS_EXIT(zfsvfs); 6707 return (error); 6708 } 6709 6710 flags = FREAD; 6711 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6712 xvp, td); 6713 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL); 6714 vp = nd.ni_vp; 6715 NDFREE(&nd, NDF_ONLY_PNBUF); 6716 if (error != 0) { 6717 ZFS_EXIT(zfsvfs); 6718 if (error == ENOENT) 6719 error = ENOATTR; 6720 return (error); 6721 } 6722 6723 if (ap->a_size != NULL) { 6724 error = VOP_GETATTR(vp, &va, ap->a_cred); 6725 if (error == 0) 6726 *ap->a_size = (size_t)va.va_size; 6727 } else if (ap->a_uio != NULL) 6728 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6729 6730 VOP_UNLOCK(vp, 0); 6731 vn_close(vp, flags, ap->a_cred, td); 6732 ZFS_EXIT(zfsvfs); 6733 6734 return (error); 6735} 6736 6737/* 6738 * Vnode operation to remove a named attribute. 6739 */ 6740int 6741zfs_deleteextattr(struct vop_deleteextattr_args *ap) 6742/* 6743vop_deleteextattr { 6744 IN struct vnode *a_vp; 6745 IN int a_attrnamespace; 6746 IN const char *a_name; 6747 IN struct ucred *a_cred; 6748 IN struct thread *a_td; 6749}; 6750*/ 6751{ 6752 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6753 struct thread *td = ap->a_td; 6754 struct nameidata nd; 6755 char attrname[255]; 6756 struct vattr va; 6757 vnode_t *xvp = NULL, *vp; 6758 int error, flags; 6759 6760 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6761 ap->a_cred, ap->a_td, VWRITE); 6762 if (error != 0) 6763 return (error); 6764 6765 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6766 sizeof(attrname)); 6767 if (error != 0) 6768 return (error); 6769 6770 ZFS_ENTER(zfsvfs); 6771 6772 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6773 LOOKUP_XATTR); 6774 if (error != 0) { 6775 ZFS_EXIT(zfsvfs); 6776 return (error); 6777 } 6778 6779 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF, 6780 UIO_SYSSPACE, attrname, xvp, td); 6781 error = namei(&nd); 6782 vp = nd.ni_vp; 6783 if (error != 0) { 6784 ZFS_EXIT(zfsvfs); 6785 NDFREE(&nd, NDF_ONLY_PNBUF); 6786 if (error == ENOENT) 6787 error = ENOATTR; 6788 return (error); 6789 } 6790 6791 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 6792 NDFREE(&nd, NDF_ONLY_PNBUF); 6793 6794 vput(nd.ni_dvp); 6795 if (vp == nd.ni_dvp) 6796 vrele(vp); 6797 else 6798 vput(vp); 6799 ZFS_EXIT(zfsvfs); 6800 6801 return (error); 6802} 6803 6804/* 6805 * Vnode operation to set a named attribute. 6806 */ 6807static int 6808zfs_setextattr(struct vop_setextattr_args *ap) 6809/* 6810vop_setextattr { 6811 IN struct vnode *a_vp; 6812 IN int a_attrnamespace; 6813 IN const char *a_name; 6814 INOUT struct uio *a_uio; 6815 IN struct ucred *a_cred; 6816 IN struct thread *a_td; 6817}; 6818*/ 6819{ 6820 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6821 struct thread *td = ap->a_td; 6822 struct nameidata nd; 6823 char attrname[255]; 6824 struct vattr va; 6825 vnode_t *xvp = NULL, *vp; 6826 int error, flags; 6827 6828 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6829 ap->a_cred, ap->a_td, VWRITE); 6830 if (error != 0) 6831 return (error); 6832 6833 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6834 sizeof(attrname)); 6835 if (error != 0) 6836 return (error); 6837 6838 ZFS_ENTER(zfsvfs); 6839 6840 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6841 LOOKUP_XATTR | CREATE_XATTR_DIR); 6842 if (error != 0) { 6843 ZFS_EXIT(zfsvfs); 6844 return (error); 6845 } 6846 6847 flags = FFLAGS(O_WRONLY | O_CREAT); 6848 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, attrname, 6849 xvp, td); 6850 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL); 6851 vp = nd.ni_vp; 6852 NDFREE(&nd, NDF_ONLY_PNBUF); 6853 if (error != 0) { 6854 ZFS_EXIT(zfsvfs); 6855 return (error); 6856 } 6857 6858 VATTR_NULL(&va); 6859 va.va_size = 0; 6860 error = VOP_SETATTR(vp, &va, ap->a_cred); 6861 if (error == 0) 6862 VOP_WRITE(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6863 6864 VOP_UNLOCK(vp, 0); 6865 vn_close(vp, flags, ap->a_cred, td); 6866 ZFS_EXIT(zfsvfs); 6867 6868 return (error); 6869} 6870 6871/* 6872 * Vnode operation to retrieve extended attributes on a vnode. 6873 */ 6874static int 6875zfs_listextattr(struct vop_listextattr_args *ap) 6876/* 6877vop_listextattr { 6878 IN struct vnode *a_vp; 6879 IN int a_attrnamespace; 6880 INOUT struct uio *a_uio; 6881 OUT size_t *a_size; 6882 IN struct ucred *a_cred; 6883 IN struct thread *a_td; 6884}; 6885*/ 6886{ 6887 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6888 struct thread *td = ap->a_td; 6889 struct nameidata nd; 6890 char attrprefix[16]; 6891 u_char dirbuf[sizeof(struct dirent)]; 6892 struct dirent *dp; 6893 struct iovec aiov; 6894 struct uio auio, *uio = ap->a_uio; 6895 size_t *sizep = ap->a_size; 6896 size_t plen; 6897 vnode_t *xvp = NULL, *vp; 6898 int done, error, eof, pos; 6899 6900 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6901 ap->a_cred, ap->a_td, VREAD); 6902 if (error != 0) 6903 return (error); 6904 6905 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 6906 sizeof(attrprefix)); 6907 if (error != 0) 6908 return (error); 6909 plen = strlen(attrprefix); 6910 6911 ZFS_ENTER(zfsvfs); 6912 6913 if (sizep != NULL) 6914 *sizep = 0; 6915 6916 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6917 LOOKUP_XATTR); 6918 if (error != 0) { 6919 ZFS_EXIT(zfsvfs); 6920 /* 6921 * ENOATTR means that the EA directory does not yet exist, 6922 * i.e. there are no extended attributes there. 6923 */ 6924 if (error == ENOATTR) 6925 error = 0; 6926 return (error); 6927 } 6928 6929 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED, 6930 UIO_SYSSPACE, ".", xvp, td); 6931 error = namei(&nd); 6932 vp = nd.ni_vp; 6933 NDFREE(&nd, NDF_ONLY_PNBUF); 6934 if (error != 0) { 6935 ZFS_EXIT(zfsvfs); 6936 return (error); 6937 } 6938 6939 auio.uio_iov = &aiov; 6940 auio.uio_iovcnt = 1; 6941 auio.uio_segflg = UIO_SYSSPACE; 6942 auio.uio_td = td; 6943 auio.uio_rw = UIO_READ; 6944 auio.uio_offset = 0; 6945 6946 do { 6947 u_char nlen; 6948 6949 aiov.iov_base = (void *)dirbuf; 6950 aiov.iov_len = sizeof(dirbuf); 6951 auio.uio_resid = sizeof(dirbuf); 6952 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 6953 done = sizeof(dirbuf) - auio.uio_resid; 6954 if (error != 0) 6955 break; 6956 for (pos = 0; pos < done;) { 6957 dp = (struct dirent *)(dirbuf + pos); 6958 pos += dp->d_reclen; 6959 /* 6960 * XXX: Temporarily we also accept DT_UNKNOWN, as this 6961 * is what we get when attribute was created on Solaris. 6962 */ 6963 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 6964 continue; 6965 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0) 6966 continue; 6967 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 6968 continue; 6969 nlen = dp->d_namlen - plen; 6970 if (sizep != NULL) 6971 *sizep += 1 + nlen; 6972 else if (uio != NULL) { 6973 /* 6974 * Format of extattr name entry is one byte for 6975 * length and the rest for name. 6976 */ 6977 error = uiomove(&nlen, 1, uio->uio_rw, uio); 6978 if (error == 0) { 6979 error = uiomove(dp->d_name + plen, nlen, 6980 uio->uio_rw, uio); 6981 } 6982 if (error != 0) 6983 break; 6984 } 6985 } 6986 } while (!eof && error == 0); 6987 6988 vput(vp); 6989 ZFS_EXIT(zfsvfs); 6990 6991 return (error); 6992} 6993 6994int 6995zfs_freebsd_getacl(ap) 6996 struct vop_getacl_args /* { 6997 struct vnode *vp; 6998 acl_type_t type; 6999 struct acl *aclp; 7000 struct ucred *cred; 7001 struct thread *td; 7002 } */ *ap; 7003{ 7004 int error; 7005 vsecattr_t vsecattr; 7006 7007 if (ap->a_type != ACL_TYPE_NFS4) 7008 return (EINVAL); 7009 7010 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 7011 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL)) 7012 return (error); 7013 7014 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt); 7015 if (vsecattr.vsa_aclentp != NULL) 7016 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 7017 7018 return (error); 7019} 7020 7021int 7022zfs_freebsd_setacl(ap) 7023 struct vop_setacl_args /* { 7024 struct vnode *vp; 7025 acl_type_t type; 7026 struct acl *aclp; 7027 struct ucred *cred; 7028 struct thread *td; 7029 } */ *ap; 7030{ 7031 int error; 7032 vsecattr_t vsecattr; 7033 int aclbsize; /* size of acl list in bytes */ 7034 aclent_t *aaclp; 7035 7036 if (ap->a_type != ACL_TYPE_NFS4) 7037 return (EINVAL); 7038 7039 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 7040 return (EINVAL); 7041 7042 /* 7043 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 7044 * splitting every entry into two and appending "canonical six" 7045 * entries at the end. Don't allow for setting an ACL that would 7046 * cause chmod(2) to run out of ACL entries. 7047 */ 7048 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 7049 return (ENOSPC); 7050 7051 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 7052 if (error != 0) 7053 return (error); 7054 7055 vsecattr.vsa_mask = VSA_ACE; 7056 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t); 7057 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 7058 aaclp = vsecattr.vsa_aclentp; 7059 vsecattr.vsa_aclentsz = aclbsize; 7060 7061 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 7062 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL); 7063 kmem_free(aaclp, aclbsize); 7064 7065 return (error); 7066} 7067 7068int 7069zfs_freebsd_aclcheck(ap) 7070 struct vop_aclcheck_args /* { 7071 struct vnode *vp; 7072 acl_type_t type; 7073 struct acl *aclp; 7074 struct ucred *cred; 7075 struct thread *td; 7076 } */ *ap; 7077{ 7078 7079 return (EOPNOTSUPP); 7080} 7081 7082struct vop_vector zfs_vnodeops; 7083struct vop_vector zfs_fifoops; 7084struct vop_vector zfs_shareops; 7085 7086struct vop_vector zfs_vnodeops = { 7087 .vop_default = &default_vnodeops, 7088 .vop_inactive = zfs_freebsd_inactive, 7089 .vop_reclaim = zfs_freebsd_reclaim, 7090 .vop_access = zfs_freebsd_access, 7091#ifdef FREEBSD_NAMECACHE 7092 .vop_lookup = vfs_cache_lookup, 7093 .vop_cachedlookup = zfs_freebsd_lookup, 7094#else 7095 .vop_lookup = zfs_freebsd_lookup, 7096#endif 7097 .vop_getattr = zfs_freebsd_getattr, 7098 .vop_setattr = zfs_freebsd_setattr, 7099 .vop_create = zfs_freebsd_create, 7100 .vop_mknod = zfs_freebsd_create, 7101 .vop_mkdir = zfs_freebsd_mkdir, 7102 .vop_readdir = zfs_freebsd_readdir, 7103 .vop_fsync = zfs_freebsd_fsync, 7104 .vop_open = zfs_freebsd_open, 7105 .vop_close = zfs_freebsd_close, 7106 .vop_rmdir = zfs_freebsd_rmdir, 7107 .vop_ioctl = zfs_freebsd_ioctl, 7108 .vop_link = zfs_freebsd_link, 7109 .vop_symlink = zfs_freebsd_symlink, 7110 .vop_readlink = zfs_freebsd_readlink, 7111 .vop_read = zfs_freebsd_read, 7112 .vop_write = zfs_freebsd_write, 7113 .vop_remove = zfs_freebsd_remove, 7114 .vop_rename = zfs_freebsd_rename, 7115 .vop_pathconf = zfs_freebsd_pathconf, 7116 .vop_bmap = zfs_freebsd_bmap, 7117 .vop_fid = zfs_freebsd_fid, 7118 .vop_getextattr = zfs_getextattr, 7119 .vop_deleteextattr = zfs_deleteextattr, 7120 .vop_setextattr = zfs_setextattr, 7121 .vop_listextattr = zfs_listextattr, 7122 .vop_getacl = zfs_freebsd_getacl, 7123 .vop_setacl = zfs_freebsd_setacl, 7124 .vop_aclcheck = zfs_freebsd_aclcheck, 7125 .vop_getpages = zfs_freebsd_getpages, 7126 .vop_putpages = zfs_freebsd_putpages, 7127}; 7128 7129struct vop_vector zfs_fifoops = { 7130 .vop_default = &fifo_specops, 7131 .vop_fsync = zfs_freebsd_fsync, 7132 .vop_access = zfs_freebsd_access, 7133 .vop_getattr = zfs_freebsd_getattr, 7134 .vop_inactive = zfs_freebsd_inactive, 7135 .vop_read = VOP_PANIC, 7136 .vop_reclaim = zfs_freebsd_reclaim, 7137 .vop_setattr = zfs_freebsd_setattr, 7138 .vop_write = VOP_PANIC, 7139 .vop_pathconf = zfs_freebsd_fifo_pathconf, 7140 .vop_fid = zfs_freebsd_fid, 7141 .vop_getacl = zfs_freebsd_getacl, 7142 .vop_setacl = zfs_freebsd_setacl, 7143 .vop_aclcheck = zfs_freebsd_aclcheck, 7144}; 7145 7146/* 7147 * special share hidden files vnode operations template 7148 */ 7149struct vop_vector zfs_shareops = { 7150 .vop_default = &default_vnodeops, 7151 .vop_access = zfs_freebsd_access, 7152 .vop_inactive = zfs_freebsd_inactive, 7153 .vop_reclaim = zfs_freebsd_reclaim, 7154 .vop_fid = zfs_freebsd_fid, 7155 .vop_pathconf = zfs_freebsd_pathconf, 7156}; 7157