zfs_vnops.c revision 169057
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 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26#pragma ident "%Z%%M% %I% %E% SMI" 27 28#include <sys/types.h> 29#include <sys/param.h> 30#include <sys/time.h> 31#include <sys/systm.h> 32#include <sys/sysmacros.h> 33#include <sys/resource.h> 34#include <sys/vfs.h> 35#include <sys/vnode.h> 36#include <sys/file.h> 37#include <sys/stat.h> 38#include <sys/kmem.h> 39#include <sys/taskq.h> 40#include <sys/uio.h> 41#include <sys/atomic.h> 42#include <sys/namei.h> 43#include <sys/mman.h> 44#include <sys/cmn_err.h> 45#include <sys/errno.h> 46#include <sys/unistd.h> 47#include <sys/zfs_vfsops.h> 48#include <sys/zfs_dir.h> 49#include <sys/zfs_acl.h> 50#include <sys/zfs_ioctl.h> 51#include <sys/fs/zfs.h> 52#include <sys/dmu.h> 53#include <sys/spa.h> 54#include <sys/txg.h> 55#include <sys/dbuf.h> 56#include <sys/zap.h> 57#include <sys/dirent.h> 58#include <sys/policy.h> 59#include <sys/sunddi.h> 60#include <sys/filio.h> 61#include <sys/zfs_ctldir.h> 62#include <sys/dnlc.h> 63#include <sys/zfs_rlock.h> 64#include <sys/bio.h> 65#include <sys/buf.h> 66#include <sys/sf_buf.h> 67#include <sys/sched.h> 68 69/* 70 * Programming rules. 71 * 72 * Each vnode op performs some logical unit of work. To do this, the ZPL must 73 * properly lock its in-core state, create a DMU transaction, do the work, 74 * record this work in the intent log (ZIL), commit the DMU transaction, 75 * and wait the the intent log to commit if it's is a synchronous operation. 76 * Morover, the vnode ops must work in both normal and log replay context. 77 * The ordering of events is important to avoid deadlocks and references 78 * to freed memory. The example below illustrates the following Big Rules: 79 * 80 * (1) A check must be made in each zfs thread for a mounted file system. 81 * This is done avoiding races using ZFS_ENTER(zfsvfs). 82 * A ZFS_EXIT(zfsvfs) is needed before all returns. 83 * 84 * (2) VN_RELE() should always be the last thing except for zil_commit() 85 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 86 * First, if it's the last reference, the vnode/znode 87 * can be freed, so the zp may point to freed memory. Second, the last 88 * reference will call zfs_zinactive(), which may induce a lot of work -- 89 * pushing cached pages (which acquires range locks) and syncing out 90 * cached atime changes. Third, zfs_zinactive() may require a new tx, 91 * which could deadlock the system if you were already holding one. 92 * 93 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 94 * as they can span dmu_tx_assign() calls. 95 * 96 * (4) Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign(). 97 * In normal operation, this will be TXG_NOWAIT. During ZIL replay, 98 * it will be a specific txg. Either way, dmu_tx_assign() never blocks. 99 * This is critical because we don't want to block while holding locks. 100 * Note, in particular, that if a lock is sometimes acquired before 101 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 102 * use a non-blocking assign can deadlock the system. The scenario: 103 * 104 * Thread A has grabbed a lock before calling dmu_tx_assign(). 105 * Thread B is in an already-assigned tx, and blocks for this lock. 106 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 107 * forever, because the previous txg can't quiesce until B's tx commits. 108 * 109 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 110 * then drop all locks, call dmu_tx_wait(), and try again. 111 * 112 * (5) If the operation succeeded, generate the intent log entry for it 113 * before dropping locks. This ensures that the ordering of events 114 * in the intent log matches the order in which they actually occurred. 115 * 116 * (6) At the end of each vnode op, the DMU tx must always commit, 117 * regardless of whether there were any errors. 118 * 119 * (7) After dropping all locks, invoke zil_commit(zilog, seq, foid) 120 * to ensure that synchronous semantics are provided when necessary. 121 * 122 * In general, this is how things should be ordered in each vnode op: 123 * 124 * ZFS_ENTER(zfsvfs); // exit if unmounted 125 * top: 126 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 127 * rw_enter(...); // grab any other locks you need 128 * tx = dmu_tx_create(...); // get DMU tx 129 * dmu_tx_hold_*(); // hold each object you might modify 130 * error = dmu_tx_assign(tx, zfsvfs->z_assign); // try to assign 131 * if (error) { 132 * rw_exit(...); // drop locks 133 * zfs_dirent_unlock(dl); // unlock directory entry 134 * VN_RELE(...); // release held vnodes 135 * if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 136 * dmu_tx_wait(tx); 137 * dmu_tx_abort(tx); 138 * goto top; 139 * } 140 * dmu_tx_abort(tx); // abort DMU tx 141 * ZFS_EXIT(zfsvfs); // finished in zfs 142 * return (error); // really out of space 143 * } 144 * error = do_real_work(); // do whatever this VOP does 145 * if (error == 0) 146 * zfs_log_*(...); // on success, make ZIL entry 147 * dmu_tx_commit(tx); // commit DMU tx -- error or not 148 * rw_exit(...); // drop locks 149 * zfs_dirent_unlock(dl); // unlock directory entry 150 * VN_RELE(...); // release held vnodes 151 * zil_commit(zilog, seq, foid); // synchronous when necessary 152 * ZFS_EXIT(zfsvfs); // finished in zfs 153 * return (error); // done, report error 154 */ 155/* ARGSUSED */ 156static int 157zfs_open(vnode_t **vpp, int flag, cred_t *cr) 158{ 159 znode_t *zp = VTOZ(*vpp); 160 161 /* Keep a count of the synchronous opens in the znode */ 162 if (flag & (FSYNC | FDSYNC)) 163 atomic_inc_32(&zp->z_sync_cnt); 164 return (0); 165} 166 167/* ARGSUSED */ 168static int 169zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr) 170{ 171 znode_t *zp = VTOZ(vp); 172 173 /* Decrement the synchronous opens in the znode */ 174 if (flag & (FSYNC | FDSYNC)) 175 atomic_dec_32(&zp->z_sync_cnt); 176 177 /* 178 * Clean up any locks held by this process on the vp. 179 */ 180 cleanlocks(vp, ddi_get_pid(), 0); 181 cleanshares(vp, ddi_get_pid()); 182 183 return (0); 184} 185 186/* 187 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 188 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 189 */ 190static int 191zfs_holey(vnode_t *vp, u_long cmd, offset_t *off) 192{ 193 znode_t *zp = VTOZ(vp); 194 uint64_t noff = (uint64_t)*off; /* new offset */ 195 uint64_t file_sz; 196 int error; 197 boolean_t hole; 198 199 file_sz = zp->z_phys->zp_size; 200 if (noff >= file_sz) { 201 return (ENXIO); 202 } 203 204 if (cmd == _FIO_SEEK_HOLE) 205 hole = B_TRUE; 206 else 207 hole = B_FALSE; 208 209 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 210 211 /* end of file? */ 212 if ((error == ESRCH) || (noff > file_sz)) { 213 /* 214 * Handle the virtual hole at the end of file. 215 */ 216 if (hole) { 217 *off = file_sz; 218 return (0); 219 } 220 return (ENXIO); 221 } 222 223 if (noff < *off) 224 return (error); 225 *off = noff; 226 return (error); 227} 228 229/* ARGSUSED */ 230static int 231zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred, 232 int *rvalp) 233{ 234 offset_t off; 235 int error; 236 zfsvfs_t *zfsvfs; 237 238 switch (com) { 239 case _FIOFFS: 240 return (0); 241 242 /* 243 * The following two ioctls are used by bfu. Faking out, 244 * necessary to avoid bfu errors. 245 */ 246 case _FIOGDIO: 247 case _FIOSDIO: 248 return (0); 249 250 case _FIO_SEEK_DATA: 251 case _FIO_SEEK_HOLE: 252 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 253 return (EFAULT); 254 255 zfsvfs = VTOZ(vp)->z_zfsvfs; 256 ZFS_ENTER(zfsvfs); 257 258 /* offset parameter is in/out */ 259 error = zfs_holey(vp, com, &off); 260 ZFS_EXIT(zfsvfs); 261 if (error) 262 return (error); 263 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 264 return (EFAULT); 265 return (0); 266 } 267 return (ENOTTY); 268} 269 270/* 271 * When a file is memory mapped, we must keep the IO data synchronized 272 * between the DMU cache and the memory mapped pages. What this means: 273 * 274 * On Write: If we find a memory mapped page, we write to *both* 275 * the page and the dmu buffer. 276 * 277 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 278 * the file is memory mapped. 279 */ 280static int 281mappedwrite(vnode_t *vp, int nbytes, uio_t *uio, dmu_tx_t *tx) 282{ 283 znode_t *zp = VTOZ(vp); 284 objset_t *os = zp->z_zfsvfs->z_os; 285 vm_object_t obj; 286 vm_page_t m; 287 struct sf_buf *sf; 288 int64_t start, off; 289 int len = nbytes; 290 int error = 0; 291 292 ASSERT(vp->v_mount != NULL); 293 obj = vp->v_object; 294 ASSERT(obj != NULL); 295 296 start = uio->uio_loffset; 297 off = start & PAGEOFFSET; 298 VM_OBJECT_LOCK(obj); 299 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 300 uint64_t bytes = MIN(PAGESIZE - off, len); 301 302again: 303 if ((m = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 304 vm_page_is_valid(m, (vm_offset_t)off, bytes)) { 305 uint64_t woff, dmubytes; 306 caddr_t va; 307 308 if (vm_page_sleep_if_busy(m, FALSE, "zfsmwb")) 309 goto again; 310 woff = uio->uio_loffset; 311 dmubytes = MIN(PAGESIZE, 312 obj->un_pager.vnp.vnp_size - (woff - off)); 313 vm_page_busy(m); 314 vm_page_lock_queues(); 315 vm_page_undirty(m); 316 vm_page_unlock_queues(); 317 VM_OBJECT_UNLOCK(obj); 318 sched_pin(); 319 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 320 va = (caddr_t)sf_buf_kva(sf); 321 error = uiomove(va + off, bytes, UIO_WRITE, uio); 322 dmu_write(os, zp->z_id, woff - off, dmubytes, va, tx); 323 sf_buf_free(sf); 324 sched_unpin(); 325 VM_OBJECT_LOCK(obj); 326 vm_page_wakeup(m); 327 } else { 328 VM_OBJECT_UNLOCK(obj); 329 error = dmu_write_uio(os, zp->z_id, uio, bytes, tx); 330 VM_OBJECT_LOCK(obj); 331 } 332 len -= bytes; 333 off = 0; 334 if (error) 335 break; 336 } 337 VM_OBJECT_UNLOCK(obj); 338 return (error); 339} 340 341/* 342 * When a file is memory mapped, we must keep the IO data synchronized 343 * between the DMU cache and the memory mapped pages. What this means: 344 * 345 * On Read: We "read" preferentially from memory mapped pages, 346 * else we default from the dmu buffer. 347 * 348 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 349 * the file is memory mapped. 350 */ 351static int 352mappedread(vnode_t *vp, int nbytes, uio_t *uio) 353{ 354 znode_t *zp = VTOZ(vp); 355 objset_t *os = zp->z_zfsvfs->z_os; 356 vm_object_t obj; 357 vm_page_t m; 358 struct sf_buf *sf; 359 int64_t start, off; 360 caddr_t va; 361 int len = nbytes; 362 int error = 0; 363 364 ASSERT(vp->v_mount != NULL); 365 obj = vp->v_object; 366 ASSERT(obj != NULL); 367 368 start = uio->uio_loffset; 369 off = start & PAGEOFFSET; 370 VM_OBJECT_LOCK(obj); 371 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 372 uint64_t bytes = MIN(PAGESIZE - off, len); 373 374again: 375 if ((m = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 376 vm_page_is_valid(m, (vm_offset_t)off, bytes)) { 377 if (vm_page_sleep_if_busy(m, FALSE, "zfsmrb")) 378 goto again; 379 vm_page_busy(m); 380 VM_OBJECT_UNLOCK(obj); 381 sched_pin(); 382 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 383 va = (caddr_t)sf_buf_kva(sf); 384 error = uiomove(va + off, bytes, UIO_READ, uio); 385 sf_buf_free(sf); 386 sched_unpin(); 387 VM_OBJECT_LOCK(obj); 388 vm_page_wakeup(m); 389 } else if (m != NULL && uio->uio_segflg == UIO_NOCOPY) { 390 /* 391 * The code below is here to make sendfile(2) work 392 * correctly with ZFS. As pointed out by ups@ 393 * sendfile(2) should be changed to use VOP_GETPAGES(), 394 * but it pessimize performance of sendfile/UFS, that's 395 * why I handle this special case in ZFS code. 396 */ 397 if (vm_page_sleep_if_busy(m, FALSE, "zfsmrb")) 398 goto again; 399 vm_page_busy(m); 400 VM_OBJECT_UNLOCK(obj); 401 sched_pin(); 402 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 403 va = (caddr_t)sf_buf_kva(sf); 404 error = dmu_read(os, zp->z_id, start + off, bytes, 405 (void *)(va + off)); 406 sf_buf_free(sf); 407 sched_unpin(); 408 VM_OBJECT_LOCK(obj); 409 vm_page_wakeup(m); 410 uio->uio_resid -= bytes; 411 } else { 412 VM_OBJECT_UNLOCK(obj); 413 error = dmu_read_uio(os, zp->z_id, uio, bytes); 414 VM_OBJECT_LOCK(obj); 415 } 416 len -= bytes; 417 off = 0; 418 if (error) 419 break; 420 } 421 VM_OBJECT_UNLOCK(obj); 422 return (error); 423} 424 425offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 426 427/* 428 * Read bytes from specified file into supplied buffer. 429 * 430 * IN: vp - vnode of file to be read from. 431 * uio - structure supplying read location, range info, 432 * and return buffer. 433 * ioflag - SYNC flags; used to provide FRSYNC semantics. 434 * cr - credentials of caller. 435 * 436 * OUT: uio - updated offset and range, buffer filled. 437 * 438 * RETURN: 0 if success 439 * error code if failure 440 * 441 * Side Effects: 442 * vp - atime updated if byte count > 0 443 */ 444/* ARGSUSED */ 445static int 446zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 447{ 448 znode_t *zp = VTOZ(vp); 449 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 450 objset_t *os = zfsvfs->z_os; 451 ssize_t n, nbytes; 452 int error; 453 rl_t *rl; 454 455 ZFS_ENTER(zfsvfs); 456 457 /* 458 * Validate file offset 459 */ 460 if (uio->uio_loffset < (offset_t)0) { 461 ZFS_EXIT(zfsvfs); 462 return (EINVAL); 463 } 464 465 /* 466 * Fasttrack empty reads 467 */ 468 if (uio->uio_resid == 0) { 469 ZFS_EXIT(zfsvfs); 470 return (0); 471 } 472 473 /* 474 * Check for mandatory locks 475 */ 476 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 477 if (error = chklock(vp, FREAD, 478 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 479 ZFS_EXIT(zfsvfs); 480 return (error); 481 } 482 } 483 484 /* 485 * If we're in FRSYNC mode, sync out this znode before reading it. 486 */ 487 if (ioflag & FRSYNC) 488 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 489 490 /* 491 * Lock the range against changes. 492 */ 493 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 494 495 /* 496 * If we are reading past end-of-file we can skip 497 * to the end; but we might still need to set atime. 498 */ 499 if (uio->uio_loffset >= zp->z_phys->zp_size) { 500 error = 0; 501 goto out; 502 } 503 504 ASSERT(uio->uio_loffset < zp->z_phys->zp_size); 505 n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset); 506 507 while (n > 0) { 508 nbytes = MIN(n, zfs_read_chunk_size - 509 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 510 511 if (vn_has_cached_data(vp)) 512 error = mappedread(vp, nbytes, uio); 513 else 514 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 515 if (error) 516 break; 517 518 n -= nbytes; 519 } 520 521out: 522 zfs_range_unlock(rl); 523 524 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 525 ZFS_EXIT(zfsvfs); 526 return (error); 527} 528 529/* 530 * Fault in the pages of the first n bytes specified by the uio structure. 531 * 1 byte in each page is touched and the uio struct is unmodified. 532 * Any error will exit this routine as this is only a best 533 * attempt to get the pages resident. This is a copy of ufs_trans_touch(). 534 */ 535static void 536zfs_prefault_write(ssize_t n, struct uio *uio) 537{ 538 struct iovec *iov; 539 ulong_t cnt, incr; 540 caddr_t p; 541 542 if (uio->uio_segflg != UIO_USERSPACE) 543 return; 544 545 iov = uio->uio_iov; 546 547 while (n) { 548 cnt = MIN(iov->iov_len, n); 549 if (cnt == 0) { 550 /* empty iov entry */ 551 iov++; 552 continue; 553 } 554 n -= cnt; 555 /* 556 * touch each page in this segment. 557 */ 558 p = iov->iov_base; 559 while (cnt) { 560 if (fubyte(p) == -1) 561 return; 562 incr = MIN(cnt, PAGESIZE); 563 p += incr; 564 cnt -= incr; 565 } 566 /* 567 * touch the last byte in case it straddles a page. 568 */ 569 p--; 570 if (fubyte(p) == -1) 571 return; 572 iov++; 573 } 574} 575 576/* 577 * Write the bytes to a file. 578 * 579 * IN: vp - vnode of file to be written to. 580 * uio - structure supplying write location, range info, 581 * and data buffer. 582 * ioflag - IO_APPEND flag set if in append mode. 583 * cr - credentials of caller. 584 * 585 * OUT: uio - updated offset and range. 586 * 587 * RETURN: 0 if success 588 * error code if failure 589 * 590 * Timestamps: 591 * vp - ctime|mtime updated if byte count > 0 592 */ 593/* ARGSUSED */ 594static int 595zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 596{ 597 znode_t *zp = VTOZ(vp); 598 rlim64_t limit = MAXOFFSET_T; 599 ssize_t start_resid = uio->uio_resid; 600 ssize_t tx_bytes; 601 uint64_t end_size; 602 dmu_tx_t *tx; 603 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 604 zilog_t *zilog = zfsvfs->z_log; 605 offset_t woff; 606 ssize_t n, nbytes; 607 rl_t *rl; 608 int max_blksz = zfsvfs->z_max_blksz; 609 int error; 610 611 /* 612 * Fasttrack empty write 613 */ 614 n = start_resid; 615 if (n == 0) 616 return (0); 617 618 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 619 limit = MAXOFFSET_T; 620 621 ZFS_ENTER(zfsvfs); 622 623 /* 624 * Pre-fault the pages to ensure slow (eg NFS) pages 625 * don't hold up txg. 626 */ 627 zfs_prefault_write(n, uio); 628 629 /* 630 * If in append mode, set the io offset pointer to eof. 631 */ 632 if (ioflag & IO_APPEND) { 633 /* 634 * Range lock for a file append: 635 * The value for the start of range will be determined by 636 * zfs_range_lock() (to guarantee append semantics). 637 * If this write will cause the block size to increase, 638 * zfs_range_lock() will lock the entire file, so we must 639 * later reduce the range after we grow the block size. 640 */ 641 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 642 if (rl->r_len == UINT64_MAX) { 643 /* overlocked, zp_size can't change */ 644 woff = uio->uio_loffset = zp->z_phys->zp_size; 645 } else { 646 woff = uio->uio_loffset = rl->r_off; 647 } 648 } else { 649 woff = uio->uio_loffset; 650 /* 651 * Validate file offset 652 */ 653 if (woff < 0) { 654 ZFS_EXIT(zfsvfs); 655 return (EINVAL); 656 } 657 658 /* 659 * If we need to grow the block size then zfs_range_lock() 660 * will lock a wider range than we request here. 661 * Later after growing the block size we reduce the range. 662 */ 663 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 664 } 665 666 if (woff >= limit) { 667 zfs_range_unlock(rl); 668 ZFS_EXIT(zfsvfs); 669 return (EFBIG); 670 } 671 672 if ((woff + n) > limit || woff > (limit - n)) 673 n = limit - woff; 674 675 /* 676 * Check for mandatory locks 677 */ 678 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 679 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 680 zfs_range_unlock(rl); 681 ZFS_EXIT(zfsvfs); 682 return (error); 683 } 684 end_size = MAX(zp->z_phys->zp_size, woff + n); 685 686 /* 687 * Write the file in reasonable size chunks. Each chunk is written 688 * in a separate transaction; this keeps the intent log records small 689 * and allows us to do more fine-grained space accounting. 690 */ 691 while (n > 0) { 692 /* 693 * Start a transaction. 694 */ 695 woff = uio->uio_loffset; 696 tx = dmu_tx_create(zfsvfs->z_os); 697 dmu_tx_hold_bonus(tx, zp->z_id); 698 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 699 error = dmu_tx_assign(tx, zfsvfs->z_assign); 700 if (error) { 701 if (error == ERESTART && 702 zfsvfs->z_assign == TXG_NOWAIT) { 703 dmu_tx_wait(tx); 704 dmu_tx_abort(tx); 705 continue; 706 } 707 dmu_tx_abort(tx); 708 break; 709 } 710 711 /* 712 * If zfs_range_lock() over-locked we grow the blocksize 713 * and then reduce the lock range. This will only happen 714 * on the first iteration since zfs_range_reduce() will 715 * shrink down r_len to the appropriate size. 716 */ 717 if (rl->r_len == UINT64_MAX) { 718 uint64_t new_blksz; 719 720 if (zp->z_blksz > max_blksz) { 721 ASSERT(!ISP2(zp->z_blksz)); 722 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 723 } else { 724 new_blksz = MIN(end_size, max_blksz); 725 } 726 zfs_grow_blocksize(zp, new_blksz, tx); 727 zfs_range_reduce(rl, woff, n); 728 } 729 730 /* 731 * XXX - should we really limit each write to z_max_blksz? 732 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 733 */ 734 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 735 rw_enter(&zp->z_map_lock, RW_READER); 736 737 if (woff + nbytes > zp->z_phys->zp_size) 738 vnode_pager_setsize(vp, woff + nbytes); 739 740 tx_bytes = uio->uio_resid; 741 if (vn_has_cached_data(vp)) { 742 rw_exit(&zp->z_map_lock); 743 error = mappedwrite(vp, nbytes, uio, tx); 744 } else { 745 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 746 uio, nbytes, tx); 747 rw_exit(&zp->z_map_lock); 748 } 749 tx_bytes -= uio->uio_resid; 750 751 /* 752 * If we made no progress, we're done. If we made even 753 * partial progress, update the znode and ZIL accordingly. 754 */ 755 if (tx_bytes == 0) { 756 dmu_tx_commit(tx); 757 ASSERT(error != 0); 758 break; 759 } 760 761 /* 762 * Clear Set-UID/Set-GID bits on successful write if not 763 * privileged and at least one of the excute bits is set. 764 * 765 * It would be nice to to this after all writes have 766 * been done, but that would still expose the ISUID/ISGID 767 * to another app after the partial write is committed. 768 */ 769 mutex_enter(&zp->z_acl_lock); 770 if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) | 771 (S_IXUSR >> 6))) != 0 && 772 (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 && 773 secpolicy_vnode_setid_retain(cr, 774 (zp->z_phys->zp_mode & S_ISUID) != 0 && 775 zp->z_phys->zp_uid == 0) != 0) { 776 zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID); 777 } 778 mutex_exit(&zp->z_acl_lock); 779 780 /* 781 * Update time stamp. NOTE: This marks the bonus buffer as 782 * dirty, so we don't have to do it again for zp_size. 783 */ 784 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 785 786 /* 787 * Update the file size (zp_size) if it has changed; 788 * account for possible concurrent updates. 789 */ 790 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) 791 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 792 uio->uio_loffset); 793 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 794 dmu_tx_commit(tx); 795 796 if (error != 0) 797 break; 798 ASSERT(tx_bytes == nbytes); 799 n -= nbytes; 800 } 801 802 zfs_range_unlock(rl); 803 804 /* 805 * If we're in replay mode, or we made no progress, return error. 806 * Otherwise, it's at least a partial write, so it's successful. 807 */ 808 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 809 ZFS_EXIT(zfsvfs); 810 return (error); 811 } 812 813 if (ioflag & (FSYNC | FDSYNC)) 814 zil_commit(zilog, zp->z_last_itx, zp->z_id); 815 816 ZFS_EXIT(zfsvfs); 817 return (0); 818} 819 820void 821zfs_get_done(dmu_buf_t *db, void *vzgd) 822{ 823 zgd_t *zgd = (zgd_t *)vzgd; 824 rl_t *rl = zgd->zgd_rl; 825 vnode_t *vp = ZTOV(rl->r_zp); 826 int vfslocked; 827 828 vfslocked = VFS_LOCK_GIANT(vp->v_vfsp); 829 dmu_buf_rele(db, vzgd); 830 zfs_range_unlock(rl); 831 VN_RELE(vp); 832 zil_add_vdev(zgd->zgd_zilog, DVA_GET_VDEV(BP_IDENTITY(zgd->zgd_bp))); 833 kmem_free(zgd, sizeof (zgd_t)); 834 VFS_UNLOCK_GIANT(vfslocked); 835} 836 837/* 838 * Get data to generate a TX_WRITE intent log record. 839 */ 840int 841zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 842{ 843 zfsvfs_t *zfsvfs = arg; 844 objset_t *os = zfsvfs->z_os; 845 znode_t *zp; 846 uint64_t off = lr->lr_offset; 847 dmu_buf_t *db; 848 rl_t *rl; 849 zgd_t *zgd; 850 int dlen = lr->lr_length; /* length of user data */ 851 int error = 0; 852 853 ASSERT(zio); 854 ASSERT(dlen != 0); 855 856 /* 857 * Nothing to do if the file has been removed 858 */ 859 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 860 return (ENOENT); 861 if (zp->z_unlinked) { 862 VN_RELE(ZTOV(zp)); 863 return (ENOENT); 864 } 865 866 /* 867 * Write records come in two flavors: immediate and indirect. 868 * For small writes it's cheaper to store the data with the 869 * log record (immediate); for large writes it's cheaper to 870 * sync the data and get a pointer to it (indirect) so that 871 * we don't have to write the data twice. 872 */ 873 if (buf != NULL) { /* immediate write */ 874 rl = zfs_range_lock(zp, off, dlen, RL_READER); 875 /* test for truncation needs to be done while range locked */ 876 if (off >= zp->z_phys->zp_size) { 877 error = ENOENT; 878 goto out; 879 } 880 VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf)); 881 } else { /* indirect write */ 882 uint64_t boff; /* block starting offset */ 883 884 /* 885 * Have to lock the whole block to ensure when it's 886 * written out and it's checksum is being calculated 887 * that no one can change the data. We need to re-check 888 * blocksize after we get the lock in case it's changed! 889 */ 890 for (;;) { 891 if (ISP2(zp->z_blksz)) { 892 boff = P2ALIGN_TYPED(off, zp->z_blksz, 893 uint64_t); 894 } else { 895 boff = 0; 896 } 897 dlen = zp->z_blksz; 898 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 899 if (zp->z_blksz == dlen) 900 break; 901 zfs_range_unlock(rl); 902 } 903 /* test for truncation needs to be done while range locked */ 904 if (off >= zp->z_phys->zp_size) { 905 error = ENOENT; 906 goto out; 907 } 908 zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP); 909 zgd->zgd_rl = rl; 910 zgd->zgd_zilog = zfsvfs->z_log; 911 zgd->zgd_bp = &lr->lr_blkptr; 912 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db)); 913 ASSERT(boff == db->db_offset); 914 lr->lr_blkoff = off - boff; 915 error = dmu_sync(zio, db, &lr->lr_blkptr, 916 lr->lr_common.lrc_txg, zfs_get_done, zgd); 917 ASSERT(error == EEXIST || lr->lr_length <= zp->z_blksz); 918 if (error == 0) { 919 zil_add_vdev(zfsvfs->z_log, 920 DVA_GET_VDEV(BP_IDENTITY(&lr->lr_blkptr))); 921 } 922 /* 923 * If we get EINPROGRESS, then we need to wait for a 924 * write IO initiated by dmu_sync() to complete before 925 * we can release this dbuf. We will finish everything 926 * up in the zfs_get_done() callback. 927 */ 928 if (error == EINPROGRESS) 929 return (0); 930 dmu_buf_rele(db, zgd); 931 kmem_free(zgd, sizeof (zgd_t)); 932 } 933out: 934 zfs_range_unlock(rl); 935 VN_RELE(ZTOV(zp)); 936 return (error); 937} 938 939/*ARGSUSED*/ 940static int 941zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr) 942{ 943 znode_t *zp = VTOZ(vp); 944 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 945 int error; 946 947 ZFS_ENTER(zfsvfs); 948 error = zfs_zaccess_rwx(zp, mode, cr); 949 ZFS_EXIT(zfsvfs); 950 return (error); 951} 952 953/* 954 * Lookup an entry in a directory, or an extended attribute directory. 955 * If it exists, return a held vnode reference for it. 956 * 957 * IN: dvp - vnode of directory to search. 958 * nm - name of entry to lookup. 959 * pnp - full pathname to lookup [UNUSED]. 960 * flags - LOOKUP_XATTR set if looking for an attribute. 961 * rdir - root directory vnode [UNUSED]. 962 * cr - credentials of caller. 963 * 964 * OUT: vpp - vnode of located entry, NULL if not found. 965 * 966 * RETURN: 0 if success 967 * error code if failure 968 * 969 * Timestamps: 970 * NA 971 */ 972/* ARGSUSED */ 973static int 974zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 975 int nameiop, cred_t *cr, kthread_t *td) 976{ 977 978 znode_t *zdp = VTOZ(dvp); 979 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 980 int error; 981 982 ZFS_ENTER(zfsvfs); 983 984 *vpp = NULL; 985 986#ifdef TODO 987 if (flags & LOOKUP_XATTR) { 988 /* 989 * If the xattr property is off, refuse the lookup request. 990 */ 991 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 992 ZFS_EXIT(zfsvfs); 993 return (EINVAL); 994 } 995 996 /* 997 * We don't allow recursive attributes.. 998 * Maybe someday we will. 999 */ 1000 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 1001 ZFS_EXIT(zfsvfs); 1002 return (EINVAL); 1003 } 1004 1005 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1006 ZFS_EXIT(zfsvfs); 1007 return (error); 1008 } 1009 1010 /* 1011 * Do we have permission to get into attribute directory? 1012 */ 1013 1014 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) { 1015 VN_RELE(*vpp); 1016 } 1017 1018 ZFS_EXIT(zfsvfs); 1019 return (error); 1020 } 1021#endif /* TODO */ 1022 1023 if (dvp->v_type != VDIR) { 1024 ZFS_EXIT(zfsvfs); 1025 return (ENOTDIR); 1026 } 1027 1028 /* 1029 * Check accessibility of directory. 1030 */ 1031 1032 if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) { 1033 ZFS_EXIT(zfsvfs); 1034 return (error); 1035 } 1036 1037 if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) { 1038 1039 /* 1040 * Convert device special files 1041 */ 1042 if (IS_DEVVP(*vpp)) { 1043 vnode_t *svp; 1044 1045 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1046 VN_RELE(*vpp); 1047 if (svp == NULL) 1048 error = ENOSYS; 1049 else 1050 *vpp = svp; 1051 } 1052 } 1053 1054 ZFS_EXIT(zfsvfs); 1055 1056 /* Translate errors and add SAVENAME when needed. */ 1057 if (cnp->cn_flags & ISLASTCN) { 1058 switch (nameiop) { 1059 case CREATE: 1060 case RENAME: 1061 if (error == ENOENT) { 1062 error = EJUSTRETURN; 1063 cnp->cn_flags |= SAVENAME; 1064 break; 1065 } 1066 /* FALLTHROUGH */ 1067 case DELETE: 1068 if (error == 0) 1069 cnp->cn_flags |= SAVENAME; 1070 break; 1071 } 1072 } 1073 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) { 1074 if (cnp->cn_flags & ISDOTDOT) 1075 VOP_UNLOCK(dvp, 0, td); 1076 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, td); 1077 if (cnp->cn_flags & ISDOTDOT) 1078 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td); 1079 } 1080 1081#ifdef FREEBSD_NAMECACHE 1082 /* 1083 * Insert name into cache (as non-existent) if appropriate. 1084 */ 1085 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE) 1086 cache_enter(dvp, *vpp, cnp); 1087 /* 1088 * Insert name into cache if appropriate. 1089 */ 1090 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1091 if (!(cnp->cn_flags & ISLASTCN) || 1092 (nameiop != DELETE && nameiop != RENAME)) { 1093 cache_enter(dvp, *vpp, cnp); 1094 } 1095 } 1096#endif 1097 1098 return (error); 1099} 1100 1101/* 1102 * Attempt to create a new entry in a directory. If the entry 1103 * already exists, truncate the file if permissible, else return 1104 * an error. Return the vp of the created or trunc'd file. 1105 * 1106 * IN: dvp - vnode of directory to put new file entry in. 1107 * name - name of new file entry. 1108 * vap - attributes of new file. 1109 * excl - flag indicating exclusive or non-exclusive mode. 1110 * mode - mode to open file with. 1111 * cr - credentials of caller. 1112 * flag - large file flag [UNUSED]. 1113 * 1114 * OUT: vpp - vnode of created or trunc'd entry. 1115 * 1116 * RETURN: 0 if success 1117 * error code if failure 1118 * 1119 * Timestamps: 1120 * dvp - ctime|mtime updated if new entry created 1121 * vp - ctime|mtime always, atime if new 1122 */ 1123/* ARGSUSED */ 1124static int 1125zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1126 vnode_t **vpp, cred_t *cr, kthread_t *td) 1127{ 1128 znode_t *zp, *dzp = VTOZ(dvp); 1129 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1130 zilog_t *zilog = zfsvfs->z_log; 1131 objset_t *os = zfsvfs->z_os; 1132 zfs_dirlock_t *dl; 1133 dmu_tx_t *tx; 1134 int error; 1135 uint64_t zoid; 1136 1137 ZFS_ENTER(zfsvfs); 1138 1139top: 1140 *vpp = NULL; 1141 1142 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1143 vap->va_mode &= ~VSVTX; 1144 1145 if (*name == '\0') { 1146 /* 1147 * Null component name refers to the directory itself. 1148 */ 1149 VN_HOLD(dvp); 1150 zp = dzp; 1151 dl = NULL; 1152 error = 0; 1153 } else { 1154 /* possible VN_HOLD(zp) */ 1155 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) { 1156 if (strcmp(name, "..") == 0) 1157 error = EISDIR; 1158 ZFS_EXIT(zfsvfs); 1159 return (error); 1160 } 1161 } 1162 1163 zoid = zp ? zp->z_id : -1ULL; 1164 1165 if (zp == NULL) { 1166 /* 1167 * Create a new file object and update the directory 1168 * to reference it. 1169 */ 1170 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 1171 goto out; 1172 } 1173 1174 /* 1175 * We only support the creation of regular files in 1176 * extended attribute directories. 1177 */ 1178 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1179 (vap->va_type != VREG)) { 1180 error = EINVAL; 1181 goto out; 1182 } 1183 1184 tx = dmu_tx_create(os); 1185 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1186 dmu_tx_hold_bonus(tx, dzp->z_id); 1187 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1188 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1189 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1190 0, SPA_MAXBLOCKSIZE); 1191 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1192 if (error) { 1193 zfs_dirent_unlock(dl); 1194 if (error == ERESTART && 1195 zfsvfs->z_assign == TXG_NOWAIT) { 1196 dmu_tx_wait(tx); 1197 dmu_tx_abort(tx); 1198 goto top; 1199 } 1200 dmu_tx_abort(tx); 1201 ZFS_EXIT(zfsvfs); 1202 return (error); 1203 } 1204 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1205 ASSERT(zp->z_id == zoid); 1206 (void) zfs_link_create(dl, zp, tx, ZNEW); 1207 zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name); 1208 dmu_tx_commit(tx); 1209 } else { 1210 /* 1211 * A directory entry already exists for this name. 1212 */ 1213 /* 1214 * Can't truncate an existing file if in exclusive mode. 1215 */ 1216 if (excl == EXCL) { 1217 error = EEXIST; 1218 goto out; 1219 } 1220 /* 1221 * Can't open a directory for writing. 1222 */ 1223 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1224 error = EISDIR; 1225 goto out; 1226 } 1227 /* 1228 * Verify requested access to file. 1229 */ 1230 if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) { 1231 goto out; 1232 } 1233 1234 mutex_enter(&dzp->z_lock); 1235 dzp->z_seq++; 1236 mutex_exit(&dzp->z_lock); 1237 1238 /* 1239 * Truncate regular files if requested. 1240 */ 1241 if ((ZTOV(zp)->v_type == VREG) && 1242 (zp->z_phys->zp_size != 0) && 1243 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1244 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1245 if (error == ERESTART && 1246 zfsvfs->z_assign == TXG_NOWAIT) { 1247 /* NB: we already did dmu_tx_wait() */ 1248 zfs_dirent_unlock(dl); 1249 VN_RELE(ZTOV(zp)); 1250 goto top; 1251 } 1252 } 1253 } 1254out: 1255 1256 if (error == 0) { 1257 *vpp = ZTOV(zp); 1258 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, td); 1259 } 1260 1261 if (dl) 1262 zfs_dirent_unlock(dl); 1263 1264 if (error) { 1265 if (zp) 1266 VN_RELE(ZTOV(zp)); 1267 } else { 1268 *vpp = ZTOV(zp); 1269 /* 1270 * If vnode is for a device return a specfs vnode instead. 1271 */ 1272 if (IS_DEVVP(*vpp)) { 1273 struct vnode *svp; 1274 1275 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1276 VN_RELE(*vpp); 1277 if (svp == NULL) { 1278 error = ENOSYS; 1279 } 1280 *vpp = svp; 1281 } 1282 } 1283 1284 ZFS_EXIT(zfsvfs); 1285 return (error); 1286} 1287 1288/* 1289 * Remove an entry from a directory. 1290 * 1291 * IN: dvp - vnode of directory to remove entry from. 1292 * name - name of entry to remove. 1293 * cr - credentials of caller. 1294 * 1295 * RETURN: 0 if success 1296 * error code if failure 1297 * 1298 * Timestamps: 1299 * dvp - ctime|mtime 1300 * vp - ctime (if nlink > 0) 1301 */ 1302static int 1303zfs_remove(vnode_t *dvp, char *name, cred_t *cr) 1304{ 1305 znode_t *zp, *dzp = VTOZ(dvp); 1306 znode_t *xzp = NULL; 1307 vnode_t *vp; 1308 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1309 zilog_t *zilog = zfsvfs->z_log; 1310 uint64_t acl_obj, xattr_obj; 1311 zfs_dirlock_t *dl; 1312 dmu_tx_t *tx; 1313 boolean_t may_delete_now, delete_now = FALSE; 1314 boolean_t unlinked; 1315 int error; 1316 1317 ZFS_ENTER(zfsvfs); 1318 1319top: 1320 /* 1321 * Attempt to lock directory; fail if entry doesn't exist. 1322 */ 1323 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1324 ZFS_EXIT(zfsvfs); 1325 return (error); 1326 } 1327 1328 vp = ZTOV(zp); 1329 1330 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1331 goto out; 1332 } 1333 1334 /* 1335 * Need to use rmdir for removing directories. 1336 */ 1337 if (vp->v_type == VDIR) { 1338 error = EPERM; 1339 goto out; 1340 } 1341 1342 vnevent_remove(vp); 1343 1344 dnlc_remove(dvp, name); 1345 1346 may_delete_now = FALSE; 1347 1348 /* 1349 * We may delete the znode now, or we may put it in the unlinked set; 1350 * it depends on whether we're the last link, and on whether there are 1351 * other holds on the vnode. So we dmu_tx_hold() the right things to 1352 * allow for either case. 1353 */ 1354 tx = dmu_tx_create(zfsvfs->z_os); 1355 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1356 dmu_tx_hold_bonus(tx, zp->z_id); 1357 if (may_delete_now) 1358 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 1359 1360 /* are there any extended attributes? */ 1361 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1362 /* XXX - do we need this if we are deleting? */ 1363 dmu_tx_hold_bonus(tx, xattr_obj); 1364 } 1365 1366 /* are there any additional acls */ 1367 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1368 may_delete_now) 1369 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1370 1371 /* charge as an update -- would be nice not to charge at all */ 1372 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1373 1374 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1375 if (error) { 1376 zfs_dirent_unlock(dl); 1377 VN_RELE(vp); 1378 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1379 dmu_tx_wait(tx); 1380 dmu_tx_abort(tx); 1381 goto top; 1382 } 1383 dmu_tx_abort(tx); 1384 ZFS_EXIT(zfsvfs); 1385 return (error); 1386 } 1387 1388 /* 1389 * Remove the directory entry. 1390 */ 1391 error = zfs_link_destroy(dl, zp, tx, 0, &unlinked); 1392 1393 if (error) { 1394 dmu_tx_commit(tx); 1395 goto out; 1396 } 1397 1398 if (0 && unlinked) { 1399 VI_LOCK(vp); 1400 delete_now = may_delete_now && 1401 vp->v_count == 1 && !vn_has_cached_data(vp) && 1402 zp->z_phys->zp_xattr == xattr_obj && 1403 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1404 VI_UNLOCK(vp); 1405 } 1406 1407 if (delete_now) { 1408 if (zp->z_phys->zp_xattr) { 1409 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1410 ASSERT3U(error, ==, 0); 1411 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1412 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1413 mutex_enter(&xzp->z_lock); 1414 xzp->z_unlinked = 1; 1415 xzp->z_phys->zp_links = 0; 1416 mutex_exit(&xzp->z_lock); 1417 zfs_unlinked_add(xzp, tx); 1418 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1419 } 1420 mutex_enter(&zp->z_lock); 1421 VI_LOCK(vp); 1422 vp->v_count--; 1423 ASSERT3U(vp->v_count, ==, 0); 1424 VI_UNLOCK(vp); 1425 mutex_exit(&zp->z_lock); 1426 zfs_znode_delete(zp, tx); 1427 VFS_RELE(zfsvfs->z_vfs); 1428 } else if (unlinked) { 1429 zfs_unlinked_add(zp, tx); 1430 } 1431 1432 zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name); 1433 1434 dmu_tx_commit(tx); 1435out: 1436 zfs_dirent_unlock(dl); 1437 1438 if (!delete_now) { 1439 VN_RELE(vp); 1440 } else if (xzp) { 1441 /* this rele delayed to prevent nesting transactions */ 1442 VN_RELE(ZTOV(xzp)); 1443 } 1444 1445 ZFS_EXIT(zfsvfs); 1446 return (error); 1447} 1448 1449/* 1450 * Create a new directory and insert it into dvp using the name 1451 * provided. Return a pointer to the inserted directory. 1452 * 1453 * IN: dvp - vnode of directory to add subdir to. 1454 * dirname - name of new directory. 1455 * vap - attributes of new directory. 1456 * cr - credentials of caller. 1457 * 1458 * OUT: vpp - vnode of created directory. 1459 * 1460 * RETURN: 0 if success 1461 * error code if failure 1462 * 1463 * Timestamps: 1464 * dvp - ctime|mtime updated 1465 * vp - ctime|mtime|atime updated 1466 */ 1467static int 1468zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr) 1469{ 1470 znode_t *zp, *dzp = VTOZ(dvp); 1471 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1472 zilog_t *zilog = zfsvfs->z_log; 1473 zfs_dirlock_t *dl; 1474 uint64_t zoid = 0; 1475 dmu_tx_t *tx; 1476 int error; 1477 1478 ASSERT(vap->va_type == VDIR); 1479 1480 ZFS_ENTER(zfsvfs); 1481 1482 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1483 ZFS_EXIT(zfsvfs); 1484 return (EINVAL); 1485 } 1486top: 1487 *vpp = NULL; 1488 1489 /* 1490 * First make sure the new directory doesn't exist. 1491 */ 1492 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) { 1493 ZFS_EXIT(zfsvfs); 1494 return (error); 1495 } 1496 1497 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) { 1498 zfs_dirent_unlock(dl); 1499 ZFS_EXIT(zfsvfs); 1500 return (error); 1501 } 1502 1503 /* 1504 * Add a new entry to the directory. 1505 */ 1506 tx = dmu_tx_create(zfsvfs->z_os); 1507 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1508 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1509 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1510 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1511 0, SPA_MAXBLOCKSIZE); 1512 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1513 if (error) { 1514 zfs_dirent_unlock(dl); 1515 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1516 dmu_tx_wait(tx); 1517 dmu_tx_abort(tx); 1518 goto top; 1519 } 1520 dmu_tx_abort(tx); 1521 ZFS_EXIT(zfsvfs); 1522 return (error); 1523 } 1524 1525 /* 1526 * Create new node. 1527 */ 1528 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1529 1530 /* 1531 * Now put new name in parent dir. 1532 */ 1533 (void) zfs_link_create(dl, zp, tx, ZNEW); 1534 1535 *vpp = ZTOV(zp); 1536 1537 zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname); 1538 dmu_tx_commit(tx); 1539 1540 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, curthread); 1541 1542 zfs_dirent_unlock(dl); 1543 1544 ZFS_EXIT(zfsvfs); 1545 return (0); 1546} 1547 1548/* 1549 * Remove a directory subdir entry. If the current working 1550 * directory is the same as the subdir to be removed, the 1551 * remove will fail. 1552 * 1553 * IN: dvp - vnode of directory to remove from. 1554 * name - name of directory to be removed. 1555 * cwd - vnode of current working directory. 1556 * cr - credentials of caller. 1557 * 1558 * RETURN: 0 if success 1559 * error code if failure 1560 * 1561 * Timestamps: 1562 * dvp - ctime|mtime updated 1563 */ 1564static int 1565zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr) 1566{ 1567 znode_t *dzp = VTOZ(dvp); 1568 znode_t *zp; 1569 vnode_t *vp; 1570 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1571 zilog_t *zilog = zfsvfs->z_log; 1572 zfs_dirlock_t *dl; 1573 dmu_tx_t *tx; 1574 int error; 1575 1576 ZFS_ENTER(zfsvfs); 1577 1578top: 1579 zp = NULL; 1580 1581 /* 1582 * Attempt to lock directory; fail if entry doesn't exist. 1583 */ 1584 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1585 ZFS_EXIT(zfsvfs); 1586 return (error); 1587 } 1588 1589 vp = ZTOV(zp); 1590 1591 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1592 goto out; 1593 } 1594 1595 if (vp->v_type != VDIR) { 1596 error = ENOTDIR; 1597 goto out; 1598 } 1599 1600 if (vp == cwd) { 1601 error = EINVAL; 1602 goto out; 1603 } 1604 1605 vnevent_rmdir(vp); 1606 1607 /* 1608 * Grab a lock on the directory to make sure that noone is 1609 * trying to add (or lookup) entries while we are removing it. 1610 */ 1611 rw_enter(&zp->z_name_lock, RW_WRITER); 1612 1613 /* 1614 * Grab a lock on the parent pointer to make sure we play well 1615 * with the treewalk and directory rename code. 1616 */ 1617 rw_enter(&zp->z_parent_lock, RW_WRITER); 1618 1619 tx = dmu_tx_create(zfsvfs->z_os); 1620 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1621 dmu_tx_hold_bonus(tx, zp->z_id); 1622 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1623 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1624 if (error) { 1625 rw_exit(&zp->z_parent_lock); 1626 rw_exit(&zp->z_name_lock); 1627 zfs_dirent_unlock(dl); 1628 VN_RELE(vp); 1629 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1630 dmu_tx_wait(tx); 1631 dmu_tx_abort(tx); 1632 goto top; 1633 } 1634 dmu_tx_abort(tx); 1635 ZFS_EXIT(zfsvfs); 1636 return (error); 1637 } 1638 1639#ifdef FREEBSD_NAMECACHE 1640 cache_purge(dvp); 1641#endif 1642 1643 error = zfs_link_destroy(dl, zp, tx, 0, NULL); 1644 1645 if (error == 0) 1646 zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name); 1647 1648 dmu_tx_commit(tx); 1649 1650 rw_exit(&zp->z_parent_lock); 1651 rw_exit(&zp->z_name_lock); 1652#ifdef FREEBSD_NAMECACHE 1653 cache_purge(vp); 1654#endif 1655out: 1656 zfs_dirent_unlock(dl); 1657 1658 VN_RELE(vp); 1659 1660 ZFS_EXIT(zfsvfs); 1661 return (error); 1662} 1663 1664/* 1665 * Read as many directory entries as will fit into the provided 1666 * buffer from the given directory cursor position (specified in 1667 * the uio structure. 1668 * 1669 * IN: vp - vnode of directory to read. 1670 * uio - structure supplying read location, range info, 1671 * and return buffer. 1672 * cr - credentials of caller. 1673 * 1674 * OUT: uio - updated offset and range, buffer filled. 1675 * eofp - set to true if end-of-file detected. 1676 * 1677 * RETURN: 0 if success 1678 * error code if failure 1679 * 1680 * Timestamps: 1681 * vp - atime updated 1682 * 1683 * Note that the low 4 bits of the cookie returned by zap is always zero. 1684 * This allows us to use the low range for "special" directory entries: 1685 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1686 * we use the offset 2 for the '.zfs' directory. 1687 */ 1688/* ARGSUSED */ 1689static int 1690zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 1691{ 1692 znode_t *zp = VTOZ(vp); 1693 iovec_t *iovp; 1694 dirent64_t *odp; 1695 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1696 objset_t *os; 1697 caddr_t outbuf; 1698 size_t bufsize; 1699 zap_cursor_t zc; 1700 zap_attribute_t zap; 1701 uint_t bytes_wanted; 1702 uint64_t offset; /* must be unsigned; checks for < 1 */ 1703 int local_eof; 1704 int outcount; 1705 int error; 1706 uint8_t prefetch; 1707 uint8_t type; 1708 int ncooks; 1709 u_long *cooks = NULL; 1710 1711 ZFS_ENTER(zfsvfs); 1712 1713 /* 1714 * If we are not given an eof variable, 1715 * use a local one. 1716 */ 1717 if (eofp == NULL) 1718 eofp = &local_eof; 1719 1720 /* 1721 * Check for valid iov_len. 1722 */ 1723 if (uio->uio_iov->iov_len <= 0) { 1724 ZFS_EXIT(zfsvfs); 1725 return (EINVAL); 1726 } 1727 1728 /* 1729 * Quit if directory has been removed (posix) 1730 */ 1731 if ((*eofp = zp->z_unlinked) != 0) { 1732 ZFS_EXIT(zfsvfs); 1733 return (0); 1734 } 1735 1736 error = 0; 1737 os = zfsvfs->z_os; 1738 offset = uio->uio_loffset; 1739 prefetch = zp->z_zn_prefetch; 1740 1741 /* 1742 * Initialize the iterator cursor. 1743 */ 1744 if (offset <= 3) { 1745 /* 1746 * Start iteration from the beginning of the directory. 1747 */ 1748 zap_cursor_init(&zc, os, zp->z_id); 1749 } else { 1750 /* 1751 * The offset is a serialized cursor. 1752 */ 1753 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1754 } 1755 1756 /* 1757 * Get space to change directory entries into fs independent format. 1758 */ 1759 iovp = uio->uio_iov; 1760 bytes_wanted = iovp->iov_len; 1761 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1762 bufsize = bytes_wanted; 1763 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1764 odp = (struct dirent64 *)outbuf; 1765 } else { 1766 bufsize = bytes_wanted; 1767 odp = (struct dirent64 *)iovp->iov_base; 1768 } 1769 1770 if (ncookies != NULL) { 1771 /* 1772 * Minimum entry size is dirent size and 1 byte for a file name. 1773 */ 1774 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 1775 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 1776 *cookies = cooks; 1777 *ncookies = ncooks; 1778 } 1779 1780 /* 1781 * Transform to file-system independent format 1782 */ 1783 outcount = 0; 1784 while (outcount < bytes_wanted) { 1785 ino64_t objnum; 1786 ushort_t reclen; 1787 1788 /* 1789 * Special case `.', `..', and `.zfs'. 1790 */ 1791 if (offset == 0) { 1792 (void) strcpy(zap.za_name, "."); 1793 objnum = zp->z_id; 1794 } else if (offset == 1) { 1795 (void) strcpy(zap.za_name, ".."); 1796 objnum = zp->z_phys->zp_parent; 1797 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1798 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1799 objnum = ZFSCTL_INO_ROOT; 1800 } else { 1801 /* 1802 * Grab next entry. 1803 */ 1804 if (error = zap_cursor_retrieve(&zc, &zap)) { 1805 if ((*eofp = (error == ENOENT)) != 0) 1806 break; 1807 else 1808 goto update; 1809 } 1810 1811 if (zap.za_integer_length != 8 || 1812 zap.za_num_integers != 1) { 1813 cmn_err(CE_WARN, "zap_readdir: bad directory " 1814 "entry, obj = %lld, offset = %lld\n", 1815 (u_longlong_t)zp->z_id, 1816 (u_longlong_t)offset); 1817 error = ENXIO; 1818 goto update; 1819 } 1820 1821 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 1822 /* 1823 * MacOS X can extract the object type here such as: 1824 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1825 */ 1826 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1827 } 1828 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1829 1830 /* 1831 * Will this entry fit in the buffer? 1832 */ 1833 if (outcount + reclen > bufsize) { 1834 /* 1835 * Did we manage to fit anything in the buffer? 1836 */ 1837 if (!outcount) { 1838 error = EINVAL; 1839 goto update; 1840 } 1841 break; 1842 } 1843 /* 1844 * Add this entry: 1845 */ 1846 odp->d_ino = objnum; 1847 odp->d_reclen = reclen; 1848 odp->d_namlen = strlen(zap.za_name); 1849 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 1850 odp->d_type = type; 1851 outcount += reclen; 1852 odp = (dirent64_t *)((intptr_t)odp + reclen); 1853 1854 ASSERT(outcount <= bufsize); 1855 1856 /* Prefetch znode */ 1857 if (prefetch) 1858 dmu_prefetch(os, objnum, 0, 0); 1859 1860 /* 1861 * Move to the next entry, fill in the previous offset. 1862 */ 1863 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1864 zap_cursor_advance(&zc); 1865 offset = zap_cursor_serialize(&zc); 1866 } else { 1867 offset += 1; 1868 } 1869 1870 if (cooks != NULL) { 1871 *cooks++ = offset; 1872 ncooks--; 1873 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 1874 } 1875 } 1876 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1877 1878 /* Subtract unused cookies */ 1879 if (ncookies != NULL) 1880 *ncookies -= ncooks; 1881 1882 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 1883 iovp->iov_base += outcount; 1884 iovp->iov_len -= outcount; 1885 uio->uio_resid -= outcount; 1886 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 1887 /* 1888 * Reset the pointer. 1889 */ 1890 offset = uio->uio_loffset; 1891 } 1892 1893update: 1894 zap_cursor_fini(&zc); 1895 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 1896 kmem_free(outbuf, bufsize); 1897 1898 if (error == ENOENT) 1899 error = 0; 1900 1901 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1902 1903 uio->uio_loffset = offset; 1904 ZFS_EXIT(zfsvfs); 1905 if (error != 0) { 1906 free(*cookies, M_TEMP); 1907 *cookies = NULL; 1908 *ncookies = 0; 1909 } 1910 return (error); 1911} 1912 1913static int 1914zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr) 1915{ 1916 znode_t *zp = VTOZ(vp); 1917 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1918 1919 ZFS_ENTER(zfsvfs); 1920 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 1921 ZFS_EXIT(zfsvfs); 1922 return (0); 1923} 1924 1925/* 1926 * Get the requested file attributes and place them in the provided 1927 * vattr structure. 1928 * 1929 * IN: vp - vnode of file. 1930 * vap - va_mask identifies requested attributes. 1931 * flags - [UNUSED] 1932 * cr - credentials of caller. 1933 * 1934 * OUT: vap - attribute values. 1935 * 1936 * RETURN: 0 (always succeeds) 1937 */ 1938/* ARGSUSED */ 1939static int 1940zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1941{ 1942 znode_t *zp = VTOZ(vp); 1943 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1944 znode_phys_t *pzp = zp->z_phys; 1945 uint32_t blksize; 1946 u_longlong_t nblocks; 1947 int error; 1948 1949 ZFS_ENTER(zfsvfs); 1950 1951 /* 1952 * Return all attributes. It's cheaper to provide the answer 1953 * than to determine whether we were asked the question. 1954 */ 1955 mutex_enter(&zp->z_lock); 1956 1957 vap->va_type = IFTOVT(pzp->zp_mode); 1958 vap->va_mode = pzp->zp_mode & ~S_IFMT; 1959 vap->va_uid = zp->z_phys->zp_uid; 1960 vap->va_gid = zp->z_phys->zp_gid; 1961 vap->va_nodeid = zp->z_id; 1962 vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX); /* nlink_t limit! */ 1963 vap->va_size = pzp->zp_size; 1964 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 1965 vap->va_rdev = zfs_cmpldev(pzp->zp_rdev); 1966 vap->va_seq = zp->z_seq; 1967 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 1968 1969 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 1970 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 1971 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 1972 ZFS_TIME_DECODE(&vap->va_birthtime, pzp->zp_crtime); 1973 1974 /* 1975 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 1976 * Also, if we are the owner don't bother, since owner should 1977 * always be allowed to read basic attributes of file. 1978 */ 1979 if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) && 1980 (zp->z_phys->zp_uid != crgetuid(cr))) { 1981 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) { 1982 mutex_exit(&zp->z_lock); 1983 ZFS_EXIT(zfsvfs); 1984 return (error); 1985 } 1986 } 1987 1988 mutex_exit(&zp->z_lock); 1989 1990 dmu_object_size_from_db(zp->z_dbuf, &blksize, &nblocks); 1991 vap->va_blksize = blksize; 1992 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 1993 1994 if (zp->z_blksz == 0) { 1995 /* 1996 * Block size hasn't been set; suggest maximal I/O transfers. 1997 */ 1998 vap->va_blksize = zfsvfs->z_max_blksz; 1999 } 2000 2001 ZFS_EXIT(zfsvfs); 2002 return (0); 2003} 2004 2005/* 2006 * Set the file attributes to the values contained in the 2007 * vattr structure. 2008 * 2009 * IN: vp - vnode of file to be modified. 2010 * vap - new attribute values. 2011 * flags - ATTR_UTIME set if non-default time values provided. 2012 * cr - credentials of caller. 2013 * 2014 * RETURN: 0 if success 2015 * error code if failure 2016 * 2017 * Timestamps: 2018 * vp - ctime updated, mtime updated if size changed. 2019 */ 2020/* ARGSUSED */ 2021static int 2022zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2023 caller_context_t *ct) 2024{ 2025 struct znode *zp = VTOZ(vp); 2026 znode_phys_t *pzp = zp->z_phys; 2027 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2028 zilog_t *zilog = zfsvfs->z_log; 2029 dmu_tx_t *tx; 2030 vattr_t oldva; 2031 uint_t mask = vap->va_mask; 2032 uint_t saved_mask; 2033 int trim_mask = 0; 2034 uint64_t new_mode; 2035 znode_t *attrzp; 2036 int need_policy = FALSE; 2037 int err; 2038 2039 if (mask == 0) 2040 return (0); 2041 2042 if (mask & AT_NOSET) 2043 return (EINVAL); 2044 2045 if (mask & AT_SIZE && vp->v_type == VDIR) 2046 return (EISDIR); 2047 2048 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) 2049 return (EINVAL); 2050 2051 ZFS_ENTER(zfsvfs); 2052 2053top: 2054 attrzp = NULL; 2055 2056 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2057 ZFS_EXIT(zfsvfs); 2058 return (EROFS); 2059 } 2060 2061 /* 2062 * First validate permissions 2063 */ 2064 2065 if (mask & AT_SIZE) { 2066 err = zfs_zaccess(zp, ACE_WRITE_DATA, cr); 2067 if (err) { 2068 ZFS_EXIT(zfsvfs); 2069 return (err); 2070 } 2071 /* 2072 * XXX - Note, we are not providing any open 2073 * mode flags here (like FNDELAY), so we may 2074 * block if there are locks present... this 2075 * should be addressed in openat(). 2076 */ 2077 do { 2078 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2079 /* NB: we already did dmu_tx_wait() if necessary */ 2080 } while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 2081 if (err) { 2082 ZFS_EXIT(zfsvfs); 2083 return (err); 2084 } 2085 } 2086 2087 if (mask & (AT_ATIME|AT_MTIME)) 2088 need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr); 2089 2090 if (mask & (AT_UID|AT_GID)) { 2091 int idmask = (mask & (AT_UID|AT_GID)); 2092 int take_owner; 2093 int take_group; 2094 2095 /* 2096 * NOTE: even if a new mode is being set, 2097 * we may clear S_ISUID/S_ISGID bits. 2098 */ 2099 2100 if (!(mask & AT_MODE)) 2101 vap->va_mode = pzp->zp_mode; 2102 2103 /* 2104 * Take ownership or chgrp to group we are a member of 2105 */ 2106 2107 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2108 take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr); 2109 2110 /* 2111 * If both AT_UID and AT_GID are set then take_owner and 2112 * take_group must both be set in order to allow taking 2113 * ownership. 2114 * 2115 * Otherwise, send the check through secpolicy_vnode_setattr() 2116 * 2117 */ 2118 2119 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2120 ((idmask == AT_UID) && take_owner) || 2121 ((idmask == AT_GID) && take_group)) { 2122 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) { 2123 /* 2124 * Remove setuid/setgid for non-privileged users 2125 */ 2126 secpolicy_setid_clear(vap, cr); 2127 trim_mask = (mask & (AT_UID|AT_GID)); 2128 } else { 2129 need_policy = TRUE; 2130 } 2131 } else { 2132 need_policy = TRUE; 2133 } 2134 } 2135 2136 mutex_enter(&zp->z_lock); 2137 oldva.va_mode = pzp->zp_mode; 2138 oldva.va_uid = zp->z_phys->zp_uid; 2139 oldva.va_gid = zp->z_phys->zp_gid; 2140 mutex_exit(&zp->z_lock); 2141 2142 if (mask & AT_MODE) { 2143 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_ACL, cr) == 0) { 2144 err = secpolicy_setid_setsticky_clear(vp, vap, 2145 &oldva, cr); 2146 if (err) { 2147 ZFS_EXIT(zfsvfs); 2148 return (err); 2149 } 2150 trim_mask |= AT_MODE; 2151 } else { 2152 need_policy = TRUE; 2153 } 2154 } 2155 2156 if (need_policy) { 2157 /* 2158 * If trim_mask is set then take ownership 2159 * has been granted or write_acl is present and user 2160 * has the ability to modify mode. In that case remove 2161 * UID|GID and or MODE from mask so that 2162 * secpolicy_vnode_setattr() doesn't revoke it. 2163 */ 2164 2165 if (trim_mask) { 2166 saved_mask = vap->va_mask; 2167 vap->va_mask &= ~trim_mask; 2168 2169 } 2170 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2171 (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp); 2172 if (err) { 2173 ZFS_EXIT(zfsvfs); 2174 return (err); 2175 } 2176 2177 if (trim_mask) 2178 vap->va_mask |= saved_mask; 2179 } 2180 2181 /* 2182 * secpolicy_vnode_setattr, or take ownership may have 2183 * changed va_mask 2184 */ 2185 mask = vap->va_mask; 2186 2187 tx = dmu_tx_create(zfsvfs->z_os); 2188 dmu_tx_hold_bonus(tx, zp->z_id); 2189 2190 if (mask & AT_MODE) { 2191 uint64_t pmode = pzp->zp_mode; 2192 2193 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2194 2195 if (zp->z_phys->zp_acl.z_acl_extern_obj) 2196 dmu_tx_hold_write(tx, 2197 pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE); 2198 else 2199 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2200 0, ZFS_ACL_SIZE(MAX_ACL_SIZE)); 2201 } 2202 2203 if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) { 2204 err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp); 2205 if (err) { 2206 dmu_tx_abort(tx); 2207 ZFS_EXIT(zfsvfs); 2208 return (err); 2209 } 2210 dmu_tx_hold_bonus(tx, attrzp->z_id); 2211 } 2212 2213 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2214 if (err) { 2215 if (attrzp) 2216 VN_RELE(ZTOV(attrzp)); 2217 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2218 dmu_tx_wait(tx); 2219 dmu_tx_abort(tx); 2220 goto top; 2221 } 2222 dmu_tx_abort(tx); 2223 ZFS_EXIT(zfsvfs); 2224 return (err); 2225 } 2226 2227 dmu_buf_will_dirty(zp->z_dbuf, tx); 2228 2229 /* 2230 * Set each attribute requested. 2231 * We group settings according to the locks they need to acquire. 2232 * 2233 * Note: you cannot set ctime directly, although it will be 2234 * updated as a side-effect of calling this function. 2235 */ 2236 2237 mutex_enter(&zp->z_lock); 2238 2239 if (mask & AT_MODE) { 2240 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2241 ASSERT3U(err, ==, 0); 2242 } 2243 2244 if (attrzp) 2245 mutex_enter(&attrzp->z_lock); 2246 2247 if (mask & AT_UID) { 2248 zp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2249 if (attrzp) { 2250 attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2251 } 2252 } 2253 2254 if (mask & AT_GID) { 2255 zp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2256 if (attrzp) 2257 attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2258 } 2259 2260 if (attrzp) 2261 mutex_exit(&attrzp->z_lock); 2262 2263 if (mask & AT_ATIME) 2264 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2265 2266 if (mask & AT_MTIME) 2267 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2268 2269 if (mask & AT_SIZE) 2270 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2271 else if (mask != 0) 2272 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2273 2274 if (mask != 0) 2275 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask); 2276 2277 mutex_exit(&zp->z_lock); 2278 2279 if (attrzp) 2280 VN_RELE(ZTOV(attrzp)); 2281 2282 dmu_tx_commit(tx); 2283 2284 ZFS_EXIT(zfsvfs); 2285 return (err); 2286} 2287 2288typedef struct zfs_zlock { 2289 krwlock_t *zl_rwlock; /* lock we acquired */ 2290 znode_t *zl_znode; /* znode we held */ 2291 struct zfs_zlock *zl_next; /* next in list */ 2292} zfs_zlock_t; 2293 2294/* 2295 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2296 */ 2297static void 2298zfs_rename_unlock(zfs_zlock_t **zlpp) 2299{ 2300 zfs_zlock_t *zl; 2301 2302 while ((zl = *zlpp) != NULL) { 2303 if (zl->zl_znode != NULL) 2304 VN_RELE(ZTOV(zl->zl_znode)); 2305 rw_exit(zl->zl_rwlock); 2306 *zlpp = zl->zl_next; 2307 kmem_free(zl, sizeof (*zl)); 2308 } 2309} 2310 2311/* 2312 * Search back through the directory tree, using the ".." entries. 2313 * Lock each directory in the chain to prevent concurrent renames. 2314 * Fail any attempt to move a directory into one of its own descendants. 2315 * XXX - z_parent_lock can overlap with map or grow locks 2316 */ 2317static int 2318zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2319{ 2320 zfs_zlock_t *zl; 2321 znode_t *zp = tdzp; 2322 uint64_t rootid = zp->z_zfsvfs->z_root; 2323 uint64_t *oidp = &zp->z_id; 2324 krwlock_t *rwlp = &szp->z_parent_lock; 2325 krw_t rw = RW_WRITER; 2326 2327 /* 2328 * First pass write-locks szp and compares to zp->z_id. 2329 * Later passes read-lock zp and compare to zp->z_parent. 2330 */ 2331 do { 2332 if (!rw_tryenter(rwlp, rw)) { 2333 /* 2334 * Another thread is renaming in this path. 2335 * Note that if we are a WRITER, we don't have any 2336 * parent_locks held yet. 2337 */ 2338 if (rw == RW_READER && zp->z_id > szp->z_id) { 2339 /* 2340 * Drop our locks and restart 2341 */ 2342 zfs_rename_unlock(&zl); 2343 *zlpp = NULL; 2344 zp = tdzp; 2345 oidp = &zp->z_id; 2346 rwlp = &szp->z_parent_lock; 2347 rw = RW_WRITER; 2348 continue; 2349 } else { 2350 /* 2351 * Wait for other thread to drop its locks 2352 */ 2353 rw_enter(rwlp, rw); 2354 } 2355 } 2356 2357 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2358 zl->zl_rwlock = rwlp; 2359 zl->zl_znode = NULL; 2360 zl->zl_next = *zlpp; 2361 *zlpp = zl; 2362 2363 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2364 return (EINVAL); 2365 2366 if (*oidp == rootid) /* We've hit the top */ 2367 return (0); 2368 2369 if (rw == RW_READER) { /* i.e. not the first pass */ 2370 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2371 if (error) 2372 return (error); 2373 zl->zl_znode = zp; 2374 } 2375 oidp = &zp->z_phys->zp_parent; 2376 rwlp = &zp->z_parent_lock; 2377 rw = RW_READER; 2378 2379 } while (zp->z_id != sdzp->z_id); 2380 2381 return (0); 2382} 2383 2384/* 2385 * Move an entry from the provided source directory to the target 2386 * directory. Change the entry name as indicated. 2387 * 2388 * IN: sdvp - Source directory containing the "old entry". 2389 * snm - Old entry name. 2390 * tdvp - Target directory to contain the "new entry". 2391 * tnm - New entry name. 2392 * cr - credentials of caller. 2393 * 2394 * RETURN: 0 if success 2395 * error code if failure 2396 * 2397 * Timestamps: 2398 * sdvp,tdvp - ctime|mtime updated 2399 */ 2400static int 2401zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr) 2402{ 2403 znode_t *tdzp, *szp, *tzp; 2404 znode_t *sdzp = VTOZ(sdvp); 2405 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2406 zilog_t *zilog = zfsvfs->z_log; 2407 vnode_t *realvp; 2408 zfs_dirlock_t *sdl, *tdl; 2409 dmu_tx_t *tx; 2410 zfs_zlock_t *zl; 2411 int cmp, serr, terr, error; 2412 2413 ZFS_ENTER(zfsvfs); 2414 2415 /* 2416 * Make sure we have the real vp for the target directory. 2417 */ 2418 if (VOP_REALVP(tdvp, &realvp) == 0) 2419 tdvp = realvp; 2420 2421 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2422 ZFS_EXIT(zfsvfs); 2423 return (EXDEV); 2424 } 2425 2426 tdzp = VTOZ(tdvp); 2427top: 2428 szp = NULL; 2429 tzp = NULL; 2430 zl = NULL; 2431 2432 /* 2433 * This is to prevent the creation of links into attribute space 2434 * by renaming a linked file into/outof an attribute directory. 2435 * See the comment in zfs_link() for why this is considered bad. 2436 */ 2437 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2438 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2439 ZFS_EXIT(zfsvfs); 2440 return (EINVAL); 2441 } 2442 2443 /* 2444 * Lock source and target directory entries. To prevent deadlock, 2445 * a lock ordering must be defined. We lock the directory with 2446 * the smallest object id first, or if it's a tie, the one with 2447 * the lexically first name. 2448 */ 2449 if (sdzp->z_id < tdzp->z_id) { 2450 cmp = -1; 2451 } else if (sdzp->z_id > tdzp->z_id) { 2452 cmp = 1; 2453 } else { 2454 cmp = strcmp(snm, tnm); 2455 if (cmp == 0) { 2456 /* 2457 * POSIX: "If the old argument and the new argument 2458 * both refer to links to the same existing file, 2459 * the rename() function shall return successfully 2460 * and perform no other action." 2461 */ 2462 ZFS_EXIT(zfsvfs); 2463 return (0); 2464 } 2465 } 2466 if (cmp < 0) { 2467 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2468 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2469 } else { 2470 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2471 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2472 } 2473 2474 if (serr) { 2475 /* 2476 * Source entry invalid or not there. 2477 */ 2478 if (!terr) { 2479 zfs_dirent_unlock(tdl); 2480 if (tzp) 2481 VN_RELE(ZTOV(tzp)); 2482 } 2483 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 2484 serr = EINVAL; 2485 ZFS_EXIT(zfsvfs); 2486 return (serr); 2487 } 2488 if (terr) { 2489 zfs_dirent_unlock(sdl); 2490 VN_RELE(ZTOV(szp)); 2491 if (strcmp(tnm, "..") == 0) 2492 terr = EINVAL; 2493 ZFS_EXIT(zfsvfs); 2494 return (terr); 2495 } 2496 2497 /* 2498 * Must have write access at the source to remove the old entry 2499 * and write access at the target to create the new entry. 2500 * Note that if target and source are the same, this can be 2501 * done in a single check. 2502 */ 2503 2504 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2505 goto out; 2506 2507 if (ZTOV(szp)->v_type == VDIR) { 2508 /* 2509 * Check to make sure rename is valid. 2510 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2511 */ 2512 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 2513 goto out; 2514 } 2515 2516 /* 2517 * Does target exist? 2518 */ 2519 if (tzp) { 2520 /* 2521 * Source and target must be the same type. 2522 */ 2523 if (ZTOV(szp)->v_type == VDIR) { 2524 if (ZTOV(tzp)->v_type != VDIR) { 2525 error = ENOTDIR; 2526 goto out; 2527 } 2528 } else { 2529 if (ZTOV(tzp)->v_type == VDIR) { 2530 error = EISDIR; 2531 goto out; 2532 } 2533 } 2534 /* 2535 * POSIX dictates that when the source and target 2536 * entries refer to the same file object, rename 2537 * must do nothing and exit without error. 2538 */ 2539 if (szp->z_id == tzp->z_id) { 2540 error = 0; 2541 goto out; 2542 } 2543 } 2544 2545 vnevent_rename_src(ZTOV(szp)); 2546 if (tzp) 2547 vnevent_rename_dest(ZTOV(tzp)); 2548 2549 tx = dmu_tx_create(zfsvfs->z_os); 2550 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 2551 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 2552 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 2553 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 2554 if (sdzp != tdzp) 2555 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 2556 if (tzp) 2557 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 2558 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2559 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2560 if (error) { 2561 if (zl != NULL) 2562 zfs_rename_unlock(&zl); 2563 zfs_dirent_unlock(sdl); 2564 zfs_dirent_unlock(tdl); 2565 VN_RELE(ZTOV(szp)); 2566 if (tzp) 2567 VN_RELE(ZTOV(tzp)); 2568 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2569 dmu_tx_wait(tx); 2570 dmu_tx_abort(tx); 2571 goto top; 2572 } 2573 dmu_tx_abort(tx); 2574 ZFS_EXIT(zfsvfs); 2575 return (error); 2576 } 2577 2578 if (tzp) /* Attempt to remove the existing target */ 2579 error = zfs_link_destroy(tdl, tzp, tx, 0, NULL); 2580 2581 if (error == 0) { 2582 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 2583 if (error == 0) { 2584 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 2585 ASSERT(error == 0); 2586 zfs_log_rename(zilog, tx, TX_RENAME, sdzp, 2587 sdl->dl_name, tdzp, tdl->dl_name, szp); 2588 } 2589#ifdef FREEBSD_NAMECACHE 2590 if (error == 0) { 2591 cache_purge(sdvp); 2592 cache_purge(tdvp); 2593 } 2594#endif 2595 } 2596 2597 dmu_tx_commit(tx); 2598out: 2599 if (zl != NULL) 2600 zfs_rename_unlock(&zl); 2601 2602 zfs_dirent_unlock(sdl); 2603 zfs_dirent_unlock(tdl); 2604 2605 VN_RELE(ZTOV(szp)); 2606 if (tzp) 2607 VN_RELE(ZTOV(tzp)); 2608 2609 ZFS_EXIT(zfsvfs); 2610 2611 return (error); 2612} 2613 2614/* 2615 * Insert the indicated symbolic reference entry into the directory. 2616 * 2617 * IN: dvp - Directory to contain new symbolic link. 2618 * link - Name for new symlink entry. 2619 * vap - Attributes of new entry. 2620 * target - Target path of new symlink. 2621 * cr - credentials of caller. 2622 * 2623 * RETURN: 0 if success 2624 * error code if failure 2625 * 2626 * Timestamps: 2627 * dvp - ctime|mtime updated 2628 */ 2629static int 2630zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, cred_t *cr, kthread_t *td) 2631{ 2632 znode_t *zp, *dzp = VTOZ(dvp); 2633 zfs_dirlock_t *dl; 2634 dmu_tx_t *tx; 2635 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2636 zilog_t *zilog = zfsvfs->z_log; 2637 uint64_t zoid; 2638 int len = strlen(link); 2639 int error; 2640 2641 ASSERT(vap->va_type == VLNK); 2642 2643 ZFS_ENTER(zfsvfs); 2644top: 2645 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2646 ZFS_EXIT(zfsvfs); 2647 return (error); 2648 } 2649 2650 if (len > MAXPATHLEN) { 2651 ZFS_EXIT(zfsvfs); 2652 return (ENAMETOOLONG); 2653 } 2654 2655 /* 2656 * Attempt to lock directory; fail if entry already exists. 2657 */ 2658 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) { 2659 ZFS_EXIT(zfsvfs); 2660 return (error); 2661 } 2662 2663 tx = dmu_tx_create(zfsvfs->z_os); 2664 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 2665 dmu_tx_hold_bonus(tx, dzp->z_id); 2666 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2667 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 2668 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 2669 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2670 if (error) { 2671 zfs_dirent_unlock(dl); 2672 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2673 dmu_tx_wait(tx); 2674 dmu_tx_abort(tx); 2675 goto top; 2676 } 2677 dmu_tx_abort(tx); 2678 ZFS_EXIT(zfsvfs); 2679 return (error); 2680 } 2681 2682 dmu_buf_will_dirty(dzp->z_dbuf, tx); 2683 2684 /* 2685 * Create a new object for the symlink. 2686 * Put the link content into bonus buffer if it will fit; 2687 * otherwise, store it just like any other file data. 2688 */ 2689 zoid = 0; 2690 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 2691 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len); 2692 if (len != 0) 2693 bcopy(link, zp->z_phys + 1, len); 2694 } else { 2695 dmu_buf_t *dbp; 2696 2697 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 2698 2699 /* 2700 * Nothing can access the znode yet so no locking needed 2701 * for growing the znode's blocksize. 2702 */ 2703 zfs_grow_blocksize(zp, len, tx); 2704 2705 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp)); 2706 dmu_buf_will_dirty(dbp, tx); 2707 2708 ASSERT3U(len, <=, dbp->db_size); 2709 bcopy(link, dbp->db_data, len); 2710 dmu_buf_rele(dbp, FTAG); 2711 } 2712 zp->z_phys->zp_size = len; 2713 2714 /* 2715 * Insert the new object into the directory. 2716 */ 2717 (void) zfs_link_create(dl, zp, tx, ZNEW); 2718out: 2719 if (error == 0) { 2720 zfs_log_symlink(zilog, tx, TX_SYMLINK, dzp, zp, name, link); 2721 *vpp = ZTOV(zp); 2722 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, td); 2723 } 2724 2725 dmu_tx_commit(tx); 2726 2727 zfs_dirent_unlock(dl); 2728 2729 ZFS_EXIT(zfsvfs); 2730 return (error); 2731} 2732 2733/* 2734 * Return, in the buffer contained in the provided uio structure, 2735 * the symbolic path referred to by vp. 2736 * 2737 * IN: vp - vnode of symbolic link. 2738 * uoip - structure to contain the link path. 2739 * cr - credentials of caller. 2740 * 2741 * OUT: uio - structure to contain the link path. 2742 * 2743 * RETURN: 0 if success 2744 * error code if failure 2745 * 2746 * Timestamps: 2747 * vp - atime updated 2748 */ 2749/* ARGSUSED */ 2750static int 2751zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr) 2752{ 2753 znode_t *zp = VTOZ(vp); 2754 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2755 size_t bufsz; 2756 int error; 2757 2758 ZFS_ENTER(zfsvfs); 2759 2760 bufsz = (size_t)zp->z_phys->zp_size; 2761 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 2762 error = uiomove(zp->z_phys + 1, 2763 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2764 } else { 2765 dmu_buf_t *dbp; 2766 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 2767 if (error) { 2768 ZFS_EXIT(zfsvfs); 2769 return (error); 2770 } 2771 error = uiomove(dbp->db_data, 2772 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2773 dmu_buf_rele(dbp, FTAG); 2774 } 2775 2776 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2777 ZFS_EXIT(zfsvfs); 2778 return (error); 2779} 2780 2781/* 2782 * Insert a new entry into directory tdvp referencing svp. 2783 * 2784 * IN: tdvp - Directory to contain new entry. 2785 * svp - vnode of new entry. 2786 * name - name of new entry. 2787 * cr - credentials of caller. 2788 * 2789 * RETURN: 0 if success 2790 * error code if failure 2791 * 2792 * Timestamps: 2793 * tdvp - ctime|mtime updated 2794 * svp - ctime updated 2795 */ 2796/* ARGSUSED */ 2797static int 2798zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr) 2799{ 2800 znode_t *dzp = VTOZ(tdvp); 2801 znode_t *tzp, *szp; 2802 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2803 zilog_t *zilog = zfsvfs->z_log; 2804 zfs_dirlock_t *dl; 2805 dmu_tx_t *tx; 2806 vnode_t *realvp; 2807 int error; 2808 2809 ASSERT(tdvp->v_type == VDIR); 2810 2811 ZFS_ENTER(zfsvfs); 2812 2813 if (VOP_REALVP(svp, &realvp) == 0) 2814 svp = realvp; 2815 2816 if (svp->v_vfsp != tdvp->v_vfsp) { 2817 ZFS_EXIT(zfsvfs); 2818 return (EXDEV); 2819 } 2820 2821 szp = VTOZ(svp); 2822top: 2823 /* 2824 * We do not support links between attributes and non-attributes 2825 * because of the potential security risk of creating links 2826 * into "normal" file space in order to circumvent restrictions 2827 * imposed in attribute space. 2828 */ 2829 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 2830 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 2831 ZFS_EXIT(zfsvfs); 2832 return (EINVAL); 2833 } 2834 2835 /* 2836 * POSIX dictates that we return EPERM here. 2837 * Better choices include ENOTSUP or EISDIR. 2838 */ 2839 if (svp->v_type == VDIR) { 2840 ZFS_EXIT(zfsvfs); 2841 return (EPERM); 2842 } 2843 2844 if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) && 2845 secpolicy_basic_link(cr) != 0) { 2846 ZFS_EXIT(zfsvfs); 2847 return (EPERM); 2848 } 2849 2850 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2851 ZFS_EXIT(zfsvfs); 2852 return (error); 2853 } 2854 2855 /* 2856 * Attempt to lock directory; fail if entry already exists. 2857 */ 2858 if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) { 2859 ZFS_EXIT(zfsvfs); 2860 return (error); 2861 } 2862 2863 tx = dmu_tx_create(zfsvfs->z_os); 2864 dmu_tx_hold_bonus(tx, szp->z_id); 2865 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2866 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2867 if (error) { 2868 zfs_dirent_unlock(dl); 2869 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2870 dmu_tx_wait(tx); 2871 dmu_tx_abort(tx); 2872 goto top; 2873 } 2874 dmu_tx_abort(tx); 2875 ZFS_EXIT(zfsvfs); 2876 return (error); 2877 } 2878 2879 error = zfs_link_create(dl, szp, tx, 0); 2880 2881 if (error == 0) 2882 zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name); 2883 2884 dmu_tx_commit(tx); 2885 2886 zfs_dirent_unlock(dl); 2887 2888 ZFS_EXIT(zfsvfs); 2889 return (error); 2890} 2891 2892void 2893zfs_inactive(vnode_t *vp, cred_t *cr) 2894{ 2895 znode_t *zp = VTOZ(vp); 2896 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2897 int error; 2898 2899 rw_enter(&zfsvfs->z_um_lock, RW_READER); 2900 if (zfsvfs->z_unmounted2) { 2901 ASSERT(zp->z_dbuf_held == 0); 2902 2903 mutex_enter(&zp->z_lock); 2904 VI_LOCK(vp); 2905 vp->v_count = 0; /* count arrives as 1 */ 2906 VI_UNLOCK(vp); 2907 if (zp->z_dbuf == NULL) { 2908 mutex_exit(&zp->z_lock); 2909 zfs_znode_free(zp); 2910 } else { 2911 mutex_exit(&zp->z_lock); 2912 } 2913 rw_exit(&zfsvfs->z_um_lock); 2914 VFS_RELE(zfsvfs->z_vfs); 2915 return; 2916 } 2917 2918 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 2919 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 2920 2921 dmu_tx_hold_bonus(tx, zp->z_id); 2922 error = dmu_tx_assign(tx, TXG_WAIT); 2923 if (error) { 2924 dmu_tx_abort(tx); 2925 } else { 2926 dmu_buf_will_dirty(zp->z_dbuf, tx); 2927 mutex_enter(&zp->z_lock); 2928 zp->z_atime_dirty = 0; 2929 mutex_exit(&zp->z_lock); 2930 dmu_tx_commit(tx); 2931 } 2932 } 2933 2934 zfs_zinactive(zp); 2935 rw_exit(&zfsvfs->z_um_lock); 2936} 2937 2938CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 2939CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 2940 2941static int 2942zfs_fid(vnode_t *vp, fid_t *fidp) 2943{ 2944 znode_t *zp = VTOZ(vp); 2945 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2946 uint32_t gen = (uint32_t)zp->z_phys->zp_gen; 2947 uint64_t object = zp->z_id; 2948 zfid_short_t *zfid; 2949 int size, i; 2950 2951 ZFS_ENTER(zfsvfs); 2952 2953 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 2954 fidp->fid_len = size; 2955 2956 zfid = (zfid_short_t *)fidp; 2957 2958 zfid->zf_len = size; 2959 2960 for (i = 0; i < sizeof (zfid->zf_object); i++) 2961 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 2962 2963 /* Must have a non-zero generation number to distinguish from .zfs */ 2964 if (gen == 0) 2965 gen = 1; 2966 for (i = 0; i < sizeof (zfid->zf_gen); i++) 2967 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 2968 2969 if (size == LONG_FID_LEN) { 2970 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 2971 zfid_long_t *zlfid; 2972 2973 zlfid = (zfid_long_t *)fidp; 2974 2975 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 2976 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 2977 2978 /* XXX - this should be the generation number for the objset */ 2979 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 2980 zlfid->zf_setgen[i] = 0; 2981 } 2982 2983 ZFS_EXIT(zfsvfs); 2984 return (0); 2985} 2986 2987static int 2988zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr) 2989{ 2990 znode_t *zp, *xzp; 2991 zfsvfs_t *zfsvfs; 2992 zfs_dirlock_t *dl; 2993 int error; 2994 2995 switch (cmd) { 2996 case _PC_LINK_MAX: 2997 *valp = INT_MAX; 2998 return (0); 2999 3000 case _PC_FILESIZEBITS: 3001 *valp = 64; 3002 return (0); 3003 3004#if 0 3005 case _PC_XATTR_EXISTS: 3006 zp = VTOZ(vp); 3007 zfsvfs = zp->z_zfsvfs; 3008 ZFS_ENTER(zfsvfs); 3009 *valp = 0; 3010 error = zfs_dirent_lock(&dl, zp, "", &xzp, 3011 ZXATTR | ZEXISTS | ZSHARED); 3012 if (error == 0) { 3013 zfs_dirent_unlock(dl); 3014 if (!zfs_dirempty(xzp)) 3015 *valp = 1; 3016 VN_RELE(ZTOV(xzp)); 3017 } else if (error == ENOENT) { 3018 /* 3019 * If there aren't extended attributes, it's the 3020 * same as having zero of them. 3021 */ 3022 error = 0; 3023 } 3024 ZFS_EXIT(zfsvfs); 3025 return (error); 3026#endif 3027 3028 case _PC_ACL_EXTENDED: 3029 *valp = 0; /* TODO */ 3030 return (0); 3031 3032 case _PC_MIN_HOLE_SIZE: 3033 *valp = (int)SPA_MINBLOCKSIZE; 3034 return (0); 3035 3036 default: 3037 return (EOPNOTSUPP); 3038 } 3039} 3040 3041#ifdef TODO 3042/*ARGSUSED*/ 3043static int 3044zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3045{ 3046 znode_t *zp = VTOZ(vp); 3047 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3048 int error; 3049 3050 ZFS_ENTER(zfsvfs); 3051 error = zfs_getacl(zp, vsecp, cr); 3052 ZFS_EXIT(zfsvfs); 3053 3054 return (error); 3055} 3056#endif /* TODO */ 3057 3058#ifdef TODO 3059/*ARGSUSED*/ 3060static int 3061zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3062{ 3063 znode_t *zp = VTOZ(vp); 3064 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3065 int error; 3066 3067 ZFS_ENTER(zfsvfs); 3068 error = zfs_setacl(zp, vsecp, cr); 3069 ZFS_EXIT(zfsvfs); 3070 return (error); 3071} 3072#endif /* TODO */ 3073 3074static int 3075zfs_freebsd_open(ap) 3076 struct vop_open_args /* { 3077 struct vnode *a_vp; 3078 int a_mode; 3079 struct ucred *a_cred; 3080 struct thread *a_td; 3081 } */ *ap; 3082{ 3083 vnode_t *vp = ap->a_vp; 3084 znode_t *zp = VTOZ(vp); 3085 int error; 3086 3087 error = zfs_open(&vp, ap->a_mode, ap->a_cred); 3088 if (error == 0) 3089 vnode_create_vobject(vp, zp->z_phys->zp_size, ap->a_td); 3090 return (error); 3091} 3092 3093static int 3094zfs_freebsd_close(ap) 3095 struct vop_close_args /* { 3096 struct vnode *a_vp; 3097 int a_fflag; 3098 struct ucred *a_cred; 3099 struct thread *a_td; 3100 } */ *ap; 3101{ 3102 3103 return (zfs_close(ap->a_vp, ap->a_fflag, 0, 0, ap->a_cred)); 3104} 3105 3106static int 3107zfs_freebsd_ioctl(ap) 3108 struct vop_ioctl_args /* { 3109 struct vnode *a_vp; 3110 u_long a_command; 3111 caddr_t a_data; 3112 int a_fflag; 3113 struct ucred *cred; 3114 struct thread *td; 3115 } */ *ap; 3116{ 3117 3118 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 3119 ap->a_fflag, ap->a_cred, NULL)); 3120} 3121 3122static int 3123zfs_freebsd_read(ap) 3124 struct vop_read_args /* { 3125 struct vnode *a_vp; 3126 struct uio *a_uio; 3127 int a_ioflag; 3128 struct ucred *a_cred; 3129 } */ *ap; 3130{ 3131 3132 return (zfs_read(ap->a_vp, ap->a_uio, ap->a_ioflag, ap->a_cred, NULL)); 3133} 3134 3135static int 3136zfs_freebsd_write(ap) 3137 struct vop_write_args /* { 3138 struct vnode *a_vp; 3139 struct uio *a_uio; 3140 int a_ioflag; 3141 struct ucred *a_cred; 3142 } */ *ap; 3143{ 3144 3145 return (zfs_write(ap->a_vp, ap->a_uio, ap->a_ioflag, ap->a_cred, NULL)); 3146} 3147 3148static int 3149zfs_freebsd_access(ap) 3150 struct vop_access_args /* { 3151 struct vnode *a_vp; 3152 int a_mode; 3153 struct ucred *a_cred; 3154 struct thread *a_td; 3155 } */ *ap; 3156{ 3157 3158 return (zfs_access(ap->a_vp, ap->a_mode, 0, ap->a_cred)); 3159} 3160 3161static int 3162zfs_freebsd_lookup(ap) 3163 struct vop_lookup_args /* { 3164 struct vnode *a_dvp; 3165 struct vnode **a_vpp; 3166 struct componentname *a_cnp; 3167 } */ *ap; 3168{ 3169 struct componentname *cnp = ap->a_cnp; 3170 char nm[NAME_MAX + 1]; 3171 3172 ASSERT(cnp->cn_namelen < sizeof(nm)); 3173 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 3174 3175 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 3176 cnp->cn_cred, cnp->cn_thread)); 3177} 3178 3179static int 3180zfs_freebsd_create(ap) 3181 struct vop_create_args /* { 3182 struct vnode *a_dvp; 3183 struct vnode **a_vpp; 3184 struct componentname *a_cnp; 3185 struct vattr *a_vap; 3186 } */ *ap; 3187{ 3188 struct componentname *cnp = ap->a_cnp; 3189 vattr_t *vap = ap->a_vap; 3190 int mode; 3191 3192 ASSERT(cnp->cn_flags & SAVENAME); 3193 3194 vattr_init_mask(vap); 3195 mode = vap->va_mode & ALLPERMS; 3196 3197 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 3198 ap->a_vpp, cnp->cn_cred, cnp->cn_thread)); 3199} 3200 3201static int 3202zfs_freebsd_remove(ap) 3203 struct vop_remove_args /* { 3204 struct vnode *a_dvp; 3205 struct vnode *a_vp; 3206 struct componentname *a_cnp; 3207 } */ *ap; 3208{ 3209 3210 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 3211 3212 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 3213 ap->a_cnp->cn_cred)); 3214} 3215 3216static int 3217zfs_freebsd_mkdir(ap) 3218 struct vop_mkdir_args /* { 3219 struct vnode *a_dvp; 3220 struct vnode **a_vpp; 3221 struct componentname *a_cnp; 3222 struct vattr *a_vap; 3223 } */ *ap; 3224{ 3225 vattr_t *vap = ap->a_vap; 3226 3227 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 3228 3229 vattr_init_mask(vap); 3230 3231 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 3232 ap->a_cnp->cn_cred)); 3233} 3234 3235static int 3236zfs_freebsd_rmdir(ap) 3237 struct vop_rmdir_args /* { 3238 struct vnode *a_dvp; 3239 struct vnode *a_vp; 3240 struct componentname *a_cnp; 3241 } */ *ap; 3242{ 3243 struct componentname *cnp = ap->a_cnp; 3244 3245 ASSERT(cnp->cn_flags & SAVENAME); 3246 3247 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred)); 3248} 3249 3250static int 3251zfs_freebsd_readdir(ap) 3252 struct vop_readdir_args /* { 3253 struct vnode *a_vp; 3254 struct uio *a_uio; 3255 struct ucred *a_cred; 3256 int *a_eofflag; 3257 int *a_ncookies; 3258 u_long **a_cookies; 3259 } */ *ap; 3260{ 3261 3262 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 3263 ap->a_ncookies, ap->a_cookies)); 3264} 3265 3266static int 3267zfs_freebsd_fsync(ap) 3268 struct vop_fsync_args /* { 3269 struct vnode *a_vp; 3270 int a_waitfor; 3271 struct thread *a_td; 3272 } */ *ap; 3273{ 3274 3275 vop_stdfsync(ap); 3276 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred)); 3277} 3278 3279static int 3280zfs_freebsd_getattr(ap) 3281 struct vop_getattr_args /* { 3282 struct vnode *a_vp; 3283 struct vattr *a_vap; 3284 struct ucred *a_cred; 3285 struct thread *a_td; 3286 } */ *ap; 3287{ 3288 3289 return (zfs_getattr(ap->a_vp, ap->a_vap, 0, ap->a_cred)); 3290} 3291 3292static int 3293zfs_freebsd_setattr(ap) 3294 struct vop_setattr_args /* { 3295 struct vnode *a_vp; 3296 struct vattr *a_vap; 3297 struct ucred *a_cred; 3298 struct thread *a_td; 3299 } */ *ap; 3300{ 3301 vattr_t *vap = ap->a_vap; 3302 3303 /* No support for FreeBSD's chflags(2). */ 3304 if (vap->va_flags != VNOVAL) 3305 return (EOPNOTSUPP); 3306 3307 vattr_init_mask(vap); 3308 3309 return (zfs_setattr(ap->a_vp, vap, 0, ap->a_cred, NULL)); 3310} 3311 3312static int 3313zfs_freebsd_rename(ap) 3314 struct vop_rename_args /* { 3315 struct vnode *a_fdvp; 3316 struct vnode *a_fvp; 3317 struct componentname *a_fcnp; 3318 struct vnode *a_tdvp; 3319 struct vnode *a_tvp; 3320 struct componentname *a_tcnp; 3321 } */ *ap; 3322{ 3323 vnode_t *fdvp = ap->a_fdvp; 3324 vnode_t *fvp = ap->a_fvp; 3325 vnode_t *tdvp = ap->a_tdvp; 3326 vnode_t *tvp = ap->a_tvp; 3327 int error; 3328 3329 ASSERT(ap->a_fcnp->cn_flags & SAVENAME); 3330 ASSERT(ap->a_tcnp->cn_flags & SAVENAME); 3331 3332 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 3333 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred); 3334 3335 if (tdvp == tvp) 3336 VN_RELE(tdvp); 3337 else 3338 VN_URELE(tdvp); 3339 if (tvp) 3340 VN_URELE(tvp); 3341 VN_RELE(fdvp); 3342 VN_RELE(fvp); 3343 3344 return (error); 3345} 3346 3347static int 3348zfs_freebsd_symlink(ap) 3349 struct vop_symlink_args /* { 3350 struct vnode *a_dvp; 3351 struct vnode **a_vpp; 3352 struct componentname *a_cnp; 3353 struct vattr *a_vap; 3354 char *a_target; 3355 } */ *ap; 3356{ 3357 struct componentname *cnp = ap->a_cnp; 3358 vattr_t *vap = ap->a_vap; 3359 3360 ASSERT(cnp->cn_flags & SAVENAME); 3361 3362 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 3363 vattr_init_mask(vap); 3364 3365 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 3366 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 3367} 3368 3369static int 3370zfs_freebsd_readlink(ap) 3371 struct vop_readlink_args /* { 3372 struct vnode *a_vp; 3373 struct uio *a_uio; 3374 struct ucred *a_cred; 3375 } */ *ap; 3376{ 3377 3378 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred)); 3379} 3380 3381static int 3382zfs_freebsd_link(ap) 3383 struct vop_link_args /* { 3384 struct vnode *a_tdvp; 3385 struct vnode *a_vp; 3386 struct componentname *a_cnp; 3387 } */ *ap; 3388{ 3389 struct componentname *cnp = ap->a_cnp; 3390 3391 ASSERT(cnp->cn_flags & SAVENAME); 3392 3393 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred)); 3394} 3395 3396static int 3397zfs_freebsd_inactive(ap) 3398 struct vop_inactive_args /* { 3399 struct vnode *a_vp; 3400 struct thread *a_td; 3401 } */ *ap; 3402{ 3403 vnode_t *vp = ap->a_vp; 3404 3405 zfs_inactive(vp, ap->a_td->td_ucred); 3406 return (0); 3407} 3408 3409static int 3410zfs_freebsd_reclaim(ap) 3411 struct vop_reclaim_args /* { 3412 struct vnode *a_vp; 3413 struct thread *a_td; 3414 } */ *ap; 3415{ 3416 vnode_t *vp = ap->a_vp; 3417 znode_t *zp = VTOZ(vp); 3418 zfsvfs_t *zfsvfs; 3419 int rele = 1; 3420 3421 ASSERT(zp != NULL); 3422 3423 /* 3424 * Destroy the vm object and flush associated pages. 3425 */ 3426 vnode_destroy_vobject(vp); 3427 3428 mutex_enter(&zp->z_lock); 3429 ASSERT(zp->z_phys); 3430 ASSERT(zp->z_dbuf_held); 3431 zfsvfs = zp->z_zfsvfs; 3432 if (!zp->z_unlinked) { 3433 zp->z_dbuf_held = 0; 3434 ZTOV(zp) = NULL; 3435 mutex_exit(&zp->z_lock); 3436 dmu_buf_rele(zp->z_dbuf, NULL); 3437 } else { 3438 mutex_exit(&zp->z_lock); 3439 } 3440 VI_LOCK(vp); 3441 if (vp->v_count > 0) 3442 rele = 0; 3443 vp->v_data = NULL; 3444 ASSERT(vp->v_holdcnt > 1); 3445 vdropl(vp); 3446 if (!zp->z_unlinked && rele) 3447 VFS_RELE(zfsvfs->z_vfs); 3448 return (0); 3449} 3450 3451static int 3452zfs_freebsd_fid(ap) 3453 struct vop_fid_args /* { 3454 struct vnode *a_vp; 3455 struct fid *a_fid; 3456 } */ *ap; 3457{ 3458 3459 return (zfs_fid(ap->a_vp, (void *)ap->a_fid)); 3460} 3461 3462static int 3463zfs_freebsd_pathconf(ap) 3464 struct vop_pathconf_args /* { 3465 struct vnode *a_vp; 3466 int a_name; 3467 register_t *a_retval; 3468 } */ *ap; 3469{ 3470 ulong_t val; 3471 int error; 3472 3473 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred); 3474 if (error == 0) 3475 *ap->a_retval = val; 3476 else if (error == EOPNOTSUPP) 3477 error = vop_stdpathconf(ap); 3478 return (error); 3479} 3480 3481/* 3482 * Advisory record locking support 3483 */ 3484static int 3485zfs_freebsd_advlock(ap) 3486 struct vop_advlock_args /* { 3487 struct vnode *a_vp; 3488 caddr_t a_id; 3489 int a_op; 3490 struct flock *a_fl; 3491 int a_flags; 3492 } */ *ap; 3493{ 3494 znode_t *zp = VTOZ(ap->a_vp); 3495 3496 return (lf_advlock(ap, &(zp->z_lockf), zp->z_phys->zp_size)); 3497} 3498 3499struct vop_vector zfs_vnodeops; 3500struct vop_vector zfs_fifoops; 3501 3502struct vop_vector zfs_vnodeops = { 3503 .vop_default = &default_vnodeops, 3504 .vop_inactive = zfs_freebsd_inactive, 3505 .vop_reclaim = zfs_freebsd_reclaim, 3506 .vop_access = zfs_freebsd_access, 3507#ifdef FREEBSD_NAMECACHE 3508 .vop_lookup = vfs_cache_lookup, 3509 .vop_cachedlookup = zfs_freebsd_lookup, 3510#else 3511 .vop_lookup = zfs_freebsd_lookup, 3512#endif 3513 .vop_getattr = zfs_freebsd_getattr, 3514 .vop_setattr = zfs_freebsd_setattr, 3515 .vop_create = zfs_freebsd_create, 3516 .vop_mknod = zfs_freebsd_create, 3517 .vop_mkdir = zfs_freebsd_mkdir, 3518 .vop_readdir = zfs_freebsd_readdir, 3519 .vop_fsync = zfs_freebsd_fsync, 3520 .vop_open = zfs_freebsd_open, 3521 .vop_close = zfs_freebsd_close, 3522 .vop_rmdir = zfs_freebsd_rmdir, 3523 .vop_ioctl = zfs_freebsd_ioctl, 3524 .vop_link = zfs_freebsd_link, 3525 .vop_symlink = zfs_freebsd_symlink, 3526 .vop_readlink = zfs_freebsd_readlink, 3527 .vop_read = zfs_freebsd_read, 3528 .vop_write = zfs_freebsd_write, 3529 .vop_remove = zfs_freebsd_remove, 3530 .vop_rename = zfs_freebsd_rename, 3531 .vop_advlock = zfs_freebsd_advlock, 3532 .vop_pathconf = zfs_freebsd_pathconf, 3533 .vop_bmap = VOP_EOPNOTSUPP, 3534 .vop_fid = zfs_freebsd_fid, 3535}; 3536 3537struct vop_vector zfs_fifoops = { 3538 .vop_default = &fifo_specops, 3539 .vop_fsync = VOP_PANIC, 3540 .vop_access = zfs_freebsd_access, 3541 .vop_getattr = zfs_freebsd_getattr, 3542 .vop_inactive = zfs_freebsd_inactive, 3543 .vop_read = VOP_PANIC, 3544 .vop_reclaim = zfs_freebsd_reclaim, 3545 .vop_setattr = zfs_freebsd_setattr, 3546 .vop_write = VOP_PANIC, 3547 .vop_fid = zfs_freebsd_fid, 3548}; 3549