zfs_vnops.c revision 177633
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/* Portions Copyright 2007 Jeremy Teo */ 27 28#pragma ident "%Z%%M% %I% %E% SMI" 29 30#include <sys/types.h> 31#include <sys/param.h> 32#include <sys/time.h> 33#include <sys/systm.h> 34#include <sys/sysmacros.h> 35#include <sys/resource.h> 36#include <sys/vfs.h> 37#include <sys/vnode.h> 38#include <sys/file.h> 39#include <sys/stat.h> 40#include <sys/kmem.h> 41#include <sys/taskq.h> 42#include <sys/uio.h> 43#include <sys/atomic.h> 44#include <sys/namei.h> 45#include <sys/mman.h> 46#include <sys/cmn_err.h> 47#include <sys/errno.h> 48#include <sys/unistd.h> 49#include <sys/zfs_vfsops.h> 50#include <sys/zfs_dir.h> 51#include <sys/zfs_acl.h> 52#include <sys/zfs_ioctl.h> 53#include <sys/fs/zfs.h> 54#include <sys/dmu.h> 55#include <sys/spa.h> 56#include <sys/txg.h> 57#include <sys/dbuf.h> 58#include <sys/zap.h> 59#include <sys/dirent.h> 60#include <sys/policy.h> 61#include <sys/sunddi.h> 62#include <sys/filio.h> 63#include <sys/zfs_ctldir.h> 64#include <sys/dnlc.h> 65#include <sys/zfs_rlock.h> 66#include <sys/bio.h> 67#include <sys/buf.h> 68#include <sys/sf_buf.h> 69#include <sys/sched.h> 70 71/* 72 * Programming rules. 73 * 74 * Each vnode op performs some logical unit of work. To do this, the ZPL must 75 * properly lock its in-core state, create a DMU transaction, do the work, 76 * record this work in the intent log (ZIL), commit the DMU transaction, 77 * and wait the the intent log to commit if it's is a synchronous operation. 78 * Morover, the vnode ops must work in both normal and log replay context. 79 * The ordering of events is important to avoid deadlocks and references 80 * to freed memory. The example below illustrates the following Big Rules: 81 * 82 * (1) A check must be made in each zfs thread for a mounted file system. 83 * This is done avoiding races using ZFS_ENTER(zfsvfs). 84 * A ZFS_EXIT(zfsvfs) is needed before all returns. 85 * 86 * (2) VN_RELE() should always be the last thing except for zil_commit() 87 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 88 * First, if it's the last reference, the vnode/znode 89 * can be freed, so the zp may point to freed memory. Second, the last 90 * reference will call zfs_zinactive(), which may induce a lot of work -- 91 * pushing cached pages (which acquires range locks) and syncing out 92 * cached atime changes. Third, zfs_zinactive() may require a new tx, 93 * which could deadlock the system if you were already holding one. 94 * 95 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 96 * as they can span dmu_tx_assign() calls. 97 * 98 * (4) Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign(). 99 * In normal operation, this will be TXG_NOWAIT. During ZIL replay, 100 * it will be a specific txg. Either way, dmu_tx_assign() never blocks. 101 * This is critical because we don't want to block while holding locks. 102 * Note, in particular, that if a lock is sometimes acquired before 103 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 104 * use a non-blocking assign can deadlock the system. The scenario: 105 * 106 * Thread A has grabbed a lock before calling dmu_tx_assign(). 107 * Thread B is in an already-assigned tx, and blocks for this lock. 108 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 109 * forever, because the previous txg can't quiesce until B's tx commits. 110 * 111 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 112 * then drop all locks, call dmu_tx_wait(), and try again. 113 * 114 * (5) If the operation succeeded, generate the intent log entry for it 115 * before dropping locks. This ensures that the ordering of events 116 * in the intent log matches the order in which they actually occurred. 117 * 118 * (6) At the end of each vnode op, the DMU tx must always commit, 119 * regardless of whether there were any errors. 120 * 121 * (7) After dropping all locks, invoke zil_commit(zilog, seq, foid) 122 * to ensure that synchronous semantics are provided when necessary. 123 * 124 * In general, this is how things should be ordered in each vnode op: 125 * 126 * ZFS_ENTER(zfsvfs); // exit if unmounted 127 * top: 128 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 129 * rw_enter(...); // grab any other locks you need 130 * tx = dmu_tx_create(...); // get DMU tx 131 * dmu_tx_hold_*(); // hold each object you might modify 132 * error = dmu_tx_assign(tx, zfsvfs->z_assign); // try to assign 133 * if (error) { 134 * rw_exit(...); // drop locks 135 * zfs_dirent_unlock(dl); // unlock directory entry 136 * VN_RELE(...); // release held vnodes 137 * if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 138 * dmu_tx_wait(tx); 139 * dmu_tx_abort(tx); 140 * goto top; 141 * } 142 * dmu_tx_abort(tx); // abort DMU tx 143 * ZFS_EXIT(zfsvfs); // finished in zfs 144 * return (error); // really out of space 145 * } 146 * error = do_real_work(); // do whatever this VOP does 147 * if (error == 0) 148 * zfs_log_*(...); // on success, make ZIL entry 149 * dmu_tx_commit(tx); // commit DMU tx -- error or not 150 * rw_exit(...); // drop locks 151 * zfs_dirent_unlock(dl); // unlock directory entry 152 * VN_RELE(...); // release held vnodes 153 * zil_commit(zilog, seq, foid); // synchronous when necessary 154 * ZFS_EXIT(zfsvfs); // finished in zfs 155 * return (error); // done, report error 156 */ 157/* ARGSUSED */ 158static int 159zfs_open(vnode_t **vpp, int flag, cred_t *cr) 160{ 161 znode_t *zp = VTOZ(*vpp); 162 163 /* Keep a count of the synchronous opens in the znode */ 164 if (flag & (FSYNC | FDSYNC)) 165 atomic_inc_32(&zp->z_sync_cnt); 166 return (0); 167} 168 169/* ARGSUSED */ 170static int 171zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr) 172{ 173 znode_t *zp = VTOZ(vp); 174 175 /* Decrement the synchronous opens in the znode */ 176 if (flag & (FSYNC | FDSYNC)) 177 atomic_dec_32(&zp->z_sync_cnt); 178 179 /* 180 * Clean up any locks held by this process on the vp. 181 */ 182 cleanlocks(vp, ddi_get_pid(), 0); 183 cleanshares(vp, ddi_get_pid()); 184 185 return (0); 186} 187 188/* 189 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 190 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 191 */ 192static int 193zfs_holey(vnode_t *vp, u_long cmd, offset_t *off) 194{ 195 znode_t *zp = VTOZ(vp); 196 uint64_t noff = (uint64_t)*off; /* new offset */ 197 uint64_t file_sz; 198 int error; 199 boolean_t hole; 200 201 file_sz = zp->z_phys->zp_size; 202 if (noff >= file_sz) { 203 return (ENXIO); 204 } 205 206 if (cmd == _FIO_SEEK_HOLE) 207 hole = B_TRUE; 208 else 209 hole = B_FALSE; 210 211 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 212 213 /* end of file? */ 214 if ((error == ESRCH) || (noff > file_sz)) { 215 /* 216 * Handle the virtual hole at the end of file. 217 */ 218 if (hole) { 219 *off = file_sz; 220 return (0); 221 } 222 return (ENXIO); 223 } 224 225 if (noff < *off) 226 return (error); 227 *off = noff; 228 return (error); 229} 230 231/* ARGSUSED */ 232static int 233zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred, 234 int *rvalp) 235{ 236 offset_t off; 237 int error; 238 zfsvfs_t *zfsvfs; 239 240 switch (com) { 241 case _FIOFFS: 242 return (0); 243 244 /* 245 * The following two ioctls are used by bfu. Faking out, 246 * necessary to avoid bfu errors. 247 */ 248 case _FIOGDIO: 249 case _FIOSDIO: 250 return (0); 251 252 case _FIO_SEEK_DATA: 253 case _FIO_SEEK_HOLE: 254 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 255 return (EFAULT); 256 257 zfsvfs = VTOZ(vp)->z_zfsvfs; 258 ZFS_ENTER(zfsvfs); 259 260 /* offset parameter is in/out */ 261 error = zfs_holey(vp, com, &off); 262 ZFS_EXIT(zfsvfs); 263 if (error) 264 return (error); 265 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 266 return (EFAULT); 267 return (0); 268 } 269 return (ENOTTY); 270} 271 272/* 273 * When a file is memory mapped, we must keep the IO data synchronized 274 * between the DMU cache and the memory mapped pages. What this means: 275 * 276 * On Write: If we find a memory mapped page, we write to *both* 277 * the page and the dmu buffer. 278 * 279 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 280 * the file is memory mapped. 281 */ 282static int 283mappedwrite(vnode_t *vp, int nbytes, uio_t *uio, dmu_tx_t *tx) 284{ 285 znode_t *zp = VTOZ(vp); 286 objset_t *os = zp->z_zfsvfs->z_os; 287 vm_object_t obj; 288 vm_page_t m; 289 struct sf_buf *sf; 290 int64_t start, off; 291 int len = nbytes; 292 int error = 0; 293 uint64_t dirbytes; 294 295 ASSERT(vp->v_mount != NULL); 296 obj = vp->v_object; 297 ASSERT(obj != NULL); 298 299 start = uio->uio_loffset; 300 off = start & PAGEOFFSET; 301 dirbytes = 0; 302 VM_OBJECT_LOCK(obj); 303 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 304 uint64_t bytes = MIN(PAGESIZE - off, len); 305 uint64_t fsize; 306 307again: 308 if ((m = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 309 vm_page_is_valid(m, (vm_offset_t)off, bytes)) { 310 uint64_t woff; 311 caddr_t va; 312 313 if (vm_page_sleep_if_busy(m, FALSE, "zfsmwb")) 314 goto again; 315 fsize = obj->un_pager.vnp.vnp_size; 316 vm_page_busy(m); 317 vm_page_lock_queues(); 318 vm_page_undirty(m); 319 vm_page_unlock_queues(); 320 VM_OBJECT_UNLOCK(obj); 321 if (dirbytes > 0) { 322 error = dmu_write_uio(os, zp->z_id, uio, 323 dirbytes, tx); 324 dirbytes = 0; 325 } 326 if (error == 0) { 327 sched_pin(); 328 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 329 va = (caddr_t)sf_buf_kva(sf); 330 woff = uio->uio_loffset - off; 331 error = uiomove(va + off, bytes, UIO_WRITE, uio); 332 /* 333 * The uiomove() above could have been partially 334 * successful, that's why we call dmu_write() 335 * below unconditionally. The page was marked 336 * non-dirty above and we would lose the changes 337 * without doing so. If the uiomove() failed 338 * entirely, well, we just write what we got 339 * before one more time. 340 */ 341 dmu_write(os, zp->z_id, woff, 342 MIN(PAGESIZE, fsize - woff), va, tx); 343 sf_buf_free(sf); 344 sched_unpin(); 345 } 346 VM_OBJECT_LOCK(obj); 347 vm_page_wakeup(m); 348 } else { 349 if (__predict_false(obj->cache != NULL)) { 350 vm_page_cache_free(obj, OFF_TO_IDX(start), 351 OFF_TO_IDX(start) + 1); 352 } 353 dirbytes += bytes; 354 } 355 len -= bytes; 356 off = 0; 357 if (error) 358 break; 359 } 360 VM_OBJECT_UNLOCK(obj); 361 if (error == 0 && dirbytes > 0) 362 error = dmu_write_uio(os, zp->z_id, uio, dirbytes, tx); 363 return (error); 364} 365 366/* 367 * When a file is memory mapped, we must keep the IO data synchronized 368 * between the DMU cache and the memory mapped pages. What this means: 369 * 370 * On Read: We "read" preferentially from memory mapped pages, 371 * else we default from the dmu buffer. 372 * 373 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 374 * the file is memory mapped. 375 */ 376static int 377mappedread(vnode_t *vp, int nbytes, uio_t *uio) 378{ 379 znode_t *zp = VTOZ(vp); 380 objset_t *os = zp->z_zfsvfs->z_os; 381 vm_object_t obj; 382 vm_page_t m; 383 struct sf_buf *sf; 384 int64_t start, off; 385 caddr_t va; 386 int len = nbytes; 387 int error = 0; 388 uint64_t dirbytes; 389 390 ASSERT(vp->v_mount != NULL); 391 obj = vp->v_object; 392 ASSERT(obj != NULL); 393 394 start = uio->uio_loffset; 395 off = start & PAGEOFFSET; 396 dirbytes = 0; 397 VM_OBJECT_LOCK(obj); 398 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 399 uint64_t bytes = MIN(PAGESIZE - off, len); 400 401again: 402 if ((m = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 403 vm_page_is_valid(m, (vm_offset_t)off, bytes)) { 404 if (vm_page_sleep_if_busy(m, FALSE, "zfsmrb")) 405 goto again; 406 vm_page_busy(m); 407 VM_OBJECT_UNLOCK(obj); 408 if (dirbytes > 0) { 409 error = dmu_read_uio(os, zp->z_id, uio, 410 dirbytes); 411 dirbytes = 0; 412 } 413 if (error == 0) { 414 sched_pin(); 415 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 416 va = (caddr_t)sf_buf_kva(sf); 417 error = uiomove(va + off, bytes, UIO_READ, uio); 418 sf_buf_free(sf); 419 sched_unpin(); 420 } 421 VM_OBJECT_LOCK(obj); 422 vm_page_wakeup(m); 423 } else if (m != NULL && uio->uio_segflg == UIO_NOCOPY) { 424 /* 425 * The code below is here to make sendfile(2) work 426 * correctly with ZFS. As pointed out by ups@ 427 * sendfile(2) should be changed to use VOP_GETPAGES(), 428 * but it pessimize performance of sendfile/UFS, that's 429 * why I handle this special case in ZFS code. 430 */ 431 if (vm_page_sleep_if_busy(m, FALSE, "zfsmrb")) 432 goto again; 433 vm_page_busy(m); 434 VM_OBJECT_UNLOCK(obj); 435 if (dirbytes > 0) { 436 error = dmu_read_uio(os, zp->z_id, uio, 437 dirbytes); 438 dirbytes = 0; 439 } 440 if (error == 0) { 441 sched_pin(); 442 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 443 va = (caddr_t)sf_buf_kva(sf); 444 error = dmu_read(os, zp->z_id, start + off, 445 bytes, (void *)(va + off)); 446 sf_buf_free(sf); 447 sched_unpin(); 448 } 449 VM_OBJECT_LOCK(obj); 450 vm_page_wakeup(m); 451 if (error == 0) 452 uio->uio_resid -= bytes; 453 } else { 454 dirbytes += bytes; 455 } 456 len -= bytes; 457 off = 0; 458 if (error) 459 break; 460 } 461 VM_OBJECT_UNLOCK(obj); 462 if (error == 0 && dirbytes > 0) 463 error = dmu_read_uio(os, zp->z_id, uio, dirbytes); 464 return (error); 465} 466 467offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 468 469/* 470 * Read bytes from specified file into supplied buffer. 471 * 472 * IN: vp - vnode of file to be read from. 473 * uio - structure supplying read location, range info, 474 * and return buffer. 475 * ioflag - SYNC flags; used to provide FRSYNC semantics. 476 * cr - credentials of caller. 477 * 478 * OUT: uio - updated offset and range, buffer filled. 479 * 480 * RETURN: 0 if success 481 * error code if failure 482 * 483 * Side Effects: 484 * vp - atime updated if byte count > 0 485 */ 486/* ARGSUSED */ 487static int 488zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 489{ 490 znode_t *zp = VTOZ(vp); 491 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 492 objset_t *os = zfsvfs->z_os; 493 ssize_t n, nbytes; 494 int error; 495 rl_t *rl; 496 497 ZFS_ENTER(zfsvfs); 498 499 /* 500 * Validate file offset 501 */ 502 if (uio->uio_loffset < (offset_t)0) { 503 ZFS_EXIT(zfsvfs); 504 return (EINVAL); 505 } 506 507 /* 508 * Fasttrack empty reads 509 */ 510 if (uio->uio_resid == 0) { 511 ZFS_EXIT(zfsvfs); 512 return (0); 513 } 514 515 /* 516 * Check for mandatory locks 517 */ 518 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 519 if (error = chklock(vp, FREAD, 520 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 521 ZFS_EXIT(zfsvfs); 522 return (error); 523 } 524 } 525 526 /* 527 * If we're in FRSYNC mode, sync out this znode before reading it. 528 */ 529 if (ioflag & FRSYNC) 530 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 531 532 /* 533 * Lock the range against changes. 534 */ 535 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 536 537 /* 538 * If we are reading past end-of-file we can skip 539 * to the end; but we might still need to set atime. 540 */ 541 if (uio->uio_loffset >= zp->z_phys->zp_size) { 542 error = 0; 543 goto out; 544 } 545 546 ASSERT(uio->uio_loffset < zp->z_phys->zp_size); 547 n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset); 548 549 while (n > 0) { 550 nbytes = MIN(n, zfs_read_chunk_size - 551 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 552 553 if (vn_has_cached_data(vp)) 554 error = mappedread(vp, nbytes, uio); 555 else 556 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 557 if (error) 558 break; 559 560 n -= nbytes; 561 } 562 563out: 564 zfs_range_unlock(rl); 565 566 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 567 ZFS_EXIT(zfsvfs); 568 return (error); 569} 570 571/* 572 * Fault in the pages of the first n bytes specified by the uio structure. 573 * 1 byte in each page is touched and the uio struct is unmodified. 574 * Any error will exit this routine as this is only a best 575 * attempt to get the pages resident. This is a copy of ufs_trans_touch(). 576 */ 577static void 578zfs_prefault_write(ssize_t n, struct uio *uio) 579{ 580 struct iovec *iov; 581 ulong_t cnt, incr; 582 caddr_t p; 583 584 if (uio->uio_segflg != UIO_USERSPACE) 585 return; 586 587 iov = uio->uio_iov; 588 589 while (n) { 590 cnt = MIN(iov->iov_len, n); 591 if (cnt == 0) { 592 /* empty iov entry */ 593 iov++; 594 continue; 595 } 596 n -= cnt; 597 /* 598 * touch each page in this segment. 599 */ 600 p = iov->iov_base; 601 while (cnt) { 602 if (fubyte(p) == -1) 603 return; 604 incr = MIN(cnt, PAGESIZE); 605 p += incr; 606 cnt -= incr; 607 } 608 /* 609 * touch the last byte in case it straddles a page. 610 */ 611 p--; 612 if (fubyte(p) == -1) 613 return; 614 iov++; 615 } 616} 617 618/* 619 * Write the bytes to a file. 620 * 621 * IN: vp - vnode of file to be written to. 622 * uio - structure supplying write location, range info, 623 * and data buffer. 624 * ioflag - IO_APPEND flag set if in append mode. 625 * cr - credentials of caller. 626 * 627 * OUT: uio - updated offset and range. 628 * 629 * RETURN: 0 if success 630 * error code if failure 631 * 632 * Timestamps: 633 * vp - ctime|mtime updated if byte count > 0 634 */ 635/* ARGSUSED */ 636static int 637zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 638{ 639 znode_t *zp = VTOZ(vp); 640 rlim64_t limit = MAXOFFSET_T; 641 ssize_t start_resid = uio->uio_resid; 642 ssize_t tx_bytes; 643 uint64_t end_size; 644 dmu_tx_t *tx; 645 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 646 zilog_t *zilog = zfsvfs->z_log; 647 offset_t woff; 648 ssize_t n, nbytes; 649 rl_t *rl; 650 int max_blksz = zfsvfs->z_max_blksz; 651 int error; 652 653 /* 654 * Fasttrack empty write 655 */ 656 n = start_resid; 657 if (n == 0) 658 return (0); 659 660 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 661 limit = MAXOFFSET_T; 662 663 ZFS_ENTER(zfsvfs); 664 665 /* 666 * Pre-fault the pages to ensure slow (eg NFS) pages 667 * don't hold up txg. 668 */ 669 zfs_prefault_write(n, uio); 670 671 /* 672 * If in append mode, set the io offset pointer to eof. 673 */ 674 if (ioflag & IO_APPEND) { 675 /* 676 * Range lock for a file append: 677 * The value for the start of range will be determined by 678 * zfs_range_lock() (to guarantee append semantics). 679 * If this write will cause the block size to increase, 680 * zfs_range_lock() will lock the entire file, so we must 681 * later reduce the range after we grow the block size. 682 */ 683 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 684 if (rl->r_len == UINT64_MAX) { 685 /* overlocked, zp_size can't change */ 686 woff = uio->uio_loffset = zp->z_phys->zp_size; 687 } else { 688 woff = uio->uio_loffset = rl->r_off; 689 } 690 } else { 691 woff = uio->uio_loffset; 692 /* 693 * Validate file offset 694 */ 695 if (woff < 0) { 696 ZFS_EXIT(zfsvfs); 697 return (EINVAL); 698 } 699 700 /* 701 * If we need to grow the block size then zfs_range_lock() 702 * will lock a wider range than we request here. 703 * Later after growing the block size we reduce the range. 704 */ 705 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 706 } 707 708 if (woff >= limit) { 709 zfs_range_unlock(rl); 710 ZFS_EXIT(zfsvfs); 711 return (EFBIG); 712 } 713 714 if ((woff + n) > limit || woff > (limit - n)) 715 n = limit - woff; 716 717 /* 718 * Check for mandatory locks 719 */ 720 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 721 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 722 zfs_range_unlock(rl); 723 ZFS_EXIT(zfsvfs); 724 return (error); 725 } 726 end_size = MAX(zp->z_phys->zp_size, woff + n); 727 728 /* 729 * Write the file in reasonable size chunks. Each chunk is written 730 * in a separate transaction; this keeps the intent log records small 731 * and allows us to do more fine-grained space accounting. 732 */ 733 while (n > 0) { 734 /* 735 * Start a transaction. 736 */ 737 woff = uio->uio_loffset; 738 tx = dmu_tx_create(zfsvfs->z_os); 739 dmu_tx_hold_bonus(tx, zp->z_id); 740 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 741 error = dmu_tx_assign(tx, zfsvfs->z_assign); 742 if (error) { 743 if (error == ERESTART && 744 zfsvfs->z_assign == TXG_NOWAIT) { 745 dmu_tx_wait(tx); 746 dmu_tx_abort(tx); 747 continue; 748 } 749 dmu_tx_abort(tx); 750 break; 751 } 752 753 /* 754 * If zfs_range_lock() over-locked we grow the blocksize 755 * and then reduce the lock range. This will only happen 756 * on the first iteration since zfs_range_reduce() will 757 * shrink down r_len to the appropriate size. 758 */ 759 if (rl->r_len == UINT64_MAX) { 760 uint64_t new_blksz; 761 762 if (zp->z_blksz > max_blksz) { 763 ASSERT(!ISP2(zp->z_blksz)); 764 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 765 } else { 766 new_blksz = MIN(end_size, max_blksz); 767 } 768 zfs_grow_blocksize(zp, new_blksz, tx); 769 zfs_range_reduce(rl, woff, n); 770 } 771 772 /* 773 * XXX - should we really limit each write to z_max_blksz? 774 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 775 */ 776 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 777 778 if (woff + nbytes > zp->z_phys->zp_size) 779 vnode_pager_setsize(vp, woff + nbytes); 780 781 rw_enter(&zp->z_map_lock, RW_READER); 782 783 tx_bytes = uio->uio_resid; 784 if (vn_has_cached_data(vp)) { 785 rw_exit(&zp->z_map_lock); 786 error = mappedwrite(vp, nbytes, uio, tx); 787 } else { 788 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 789 uio, nbytes, tx); 790 rw_exit(&zp->z_map_lock); 791 } 792 tx_bytes -= uio->uio_resid; 793 794 /* 795 * If we made no progress, we're done. If we made even 796 * partial progress, update the znode and ZIL accordingly. 797 */ 798 if (tx_bytes == 0) { 799 dmu_tx_commit(tx); 800 ASSERT(error != 0); 801 break; 802 } 803 804 /* 805 * Clear Set-UID/Set-GID bits on successful write if not 806 * privileged and at least one of the excute bits is set. 807 * 808 * It would be nice to to this after all writes have 809 * been done, but that would still expose the ISUID/ISGID 810 * to another app after the partial write is committed. 811 */ 812 mutex_enter(&zp->z_acl_lock); 813 if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) | 814 (S_IXUSR >> 6))) != 0 && 815 (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 && 816 secpolicy_vnode_setid_retain(cr, 817 (zp->z_phys->zp_mode & S_ISUID) != 0 && 818 zp->z_phys->zp_uid == 0) != 0) { 819 zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID); 820 } 821 mutex_exit(&zp->z_acl_lock); 822 823 /* 824 * Update time stamp. NOTE: This marks the bonus buffer as 825 * dirty, so we don't have to do it again for zp_size. 826 */ 827 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 828 829 /* 830 * Update the file size (zp_size) if it has changed; 831 * account for possible concurrent updates. 832 */ 833 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) 834 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 835 uio->uio_loffset); 836 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 837 dmu_tx_commit(tx); 838 839 if (error != 0) 840 break; 841 ASSERT(tx_bytes == nbytes); 842 n -= nbytes; 843 } 844 845 zfs_range_unlock(rl); 846 847 /* 848 * If we're in replay mode, or we made no progress, return error. 849 * Otherwise, it's at least a partial write, so it's successful. 850 */ 851 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 852 ZFS_EXIT(zfsvfs); 853 return (error); 854 } 855 856 if (ioflag & (FSYNC | FDSYNC)) 857 zil_commit(zilog, zp->z_last_itx, zp->z_id); 858 859 ZFS_EXIT(zfsvfs); 860 return (0); 861} 862 863void 864zfs_get_done(dmu_buf_t *db, void *vzgd) 865{ 866 zgd_t *zgd = (zgd_t *)vzgd; 867 rl_t *rl = zgd->zgd_rl; 868 vnode_t *vp = ZTOV(rl->r_zp); 869 int vfslocked; 870 871 vfslocked = VFS_LOCK_GIANT(vp->v_vfsp); 872 dmu_buf_rele(db, vzgd); 873 zfs_range_unlock(rl); 874 VN_RELE(vp); 875 zil_add_vdev(zgd->zgd_zilog, DVA_GET_VDEV(BP_IDENTITY(zgd->zgd_bp))); 876 kmem_free(zgd, sizeof (zgd_t)); 877 VFS_UNLOCK_GIANT(vfslocked); 878} 879 880/* 881 * Get data to generate a TX_WRITE intent log record. 882 */ 883int 884zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 885{ 886 zfsvfs_t *zfsvfs = arg; 887 objset_t *os = zfsvfs->z_os; 888 znode_t *zp; 889 uint64_t off = lr->lr_offset; 890 dmu_buf_t *db; 891 rl_t *rl; 892 zgd_t *zgd; 893 int dlen = lr->lr_length; /* length of user data */ 894 int error = 0; 895 896 ASSERT(zio); 897 ASSERT(dlen != 0); 898 899 /* 900 * Nothing to do if the file has been removed 901 */ 902 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 903 return (ENOENT); 904 if (zp->z_unlinked) { 905 VN_RELE(ZTOV(zp)); 906 return (ENOENT); 907 } 908 909 /* 910 * Write records come in two flavors: immediate and indirect. 911 * For small writes it's cheaper to store the data with the 912 * log record (immediate); for large writes it's cheaper to 913 * sync the data and get a pointer to it (indirect) so that 914 * we don't have to write the data twice. 915 */ 916 if (buf != NULL) { /* immediate write */ 917 rl = zfs_range_lock(zp, off, dlen, RL_READER); 918 /* test for truncation needs to be done while range locked */ 919 if (off >= zp->z_phys->zp_size) { 920 error = ENOENT; 921 goto out; 922 } 923 VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf)); 924 } else { /* indirect write */ 925 uint64_t boff; /* block starting offset */ 926 927 /* 928 * Have to lock the whole block to ensure when it's 929 * written out and it's checksum is being calculated 930 * that no one can change the data. We need to re-check 931 * blocksize after we get the lock in case it's changed! 932 */ 933 for (;;) { 934 if (ISP2(zp->z_blksz)) { 935 boff = P2ALIGN_TYPED(off, zp->z_blksz, 936 uint64_t); 937 } else { 938 boff = 0; 939 } 940 dlen = zp->z_blksz; 941 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 942 if (zp->z_blksz == dlen) 943 break; 944 zfs_range_unlock(rl); 945 } 946 /* test for truncation needs to be done while range locked */ 947 if (off >= zp->z_phys->zp_size) { 948 error = ENOENT; 949 goto out; 950 } 951 zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP); 952 zgd->zgd_rl = rl; 953 zgd->zgd_zilog = zfsvfs->z_log; 954 zgd->zgd_bp = &lr->lr_blkptr; 955 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db)); 956 ASSERT(boff == db->db_offset); 957 lr->lr_blkoff = off - boff; 958 error = dmu_sync(zio, db, &lr->lr_blkptr, 959 lr->lr_common.lrc_txg, zfs_get_done, zgd); 960 ASSERT(error == EEXIST || lr->lr_length <= zp->z_blksz); 961 if (error == 0) { 962 zil_add_vdev(zfsvfs->z_log, 963 DVA_GET_VDEV(BP_IDENTITY(&lr->lr_blkptr))); 964 } 965 /* 966 * If we get EINPROGRESS, then we need to wait for a 967 * write IO initiated by dmu_sync() to complete before 968 * we can release this dbuf. We will finish everything 969 * up in the zfs_get_done() callback. 970 */ 971 if (error == EINPROGRESS) 972 return (0); 973 dmu_buf_rele(db, zgd); 974 kmem_free(zgd, sizeof (zgd_t)); 975 } 976out: 977 zfs_range_unlock(rl); 978 VN_RELE(ZTOV(zp)); 979 return (error); 980} 981 982/*ARGSUSED*/ 983static int 984zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr) 985{ 986 znode_t *zp = VTOZ(vp); 987 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 988 int error; 989 990 ZFS_ENTER(zfsvfs); 991 error = zfs_zaccess_rwx(zp, mode, cr); 992 ZFS_EXIT(zfsvfs); 993 return (error); 994} 995 996/* 997 * Lookup an entry in a directory, or an extended attribute directory. 998 * If it exists, return a held vnode reference for it. 999 * 1000 * IN: dvp - vnode of directory to search. 1001 * nm - name of entry to lookup. 1002 * pnp - full pathname to lookup [UNUSED]. 1003 * flags - LOOKUP_XATTR set if looking for an attribute. 1004 * rdir - root directory vnode [UNUSED]. 1005 * cr - credentials of caller. 1006 * 1007 * OUT: vpp - vnode of located entry, NULL if not found. 1008 * 1009 * RETURN: 0 if success 1010 * error code if failure 1011 * 1012 * Timestamps: 1013 * NA 1014 */ 1015/* ARGSUSED */ 1016static int 1017zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 1018 int nameiop, cred_t *cr, kthread_t *td) 1019{ 1020 1021 znode_t *zdp = VTOZ(dvp); 1022 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1023 int error; 1024 1025 ZFS_ENTER(zfsvfs); 1026 1027 *vpp = NULL; 1028 1029#ifdef TODO 1030 if (flags & LOOKUP_XATTR) { 1031 /* 1032 * If the xattr property is off, refuse the lookup request. 1033 */ 1034 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1035 ZFS_EXIT(zfsvfs); 1036 return (EINVAL); 1037 } 1038 1039 /* 1040 * We don't allow recursive attributes.. 1041 * Maybe someday we will. 1042 */ 1043 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 1044 ZFS_EXIT(zfsvfs); 1045 return (EINVAL); 1046 } 1047 1048 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1049 ZFS_EXIT(zfsvfs); 1050 return (error); 1051 } 1052 1053 /* 1054 * Do we have permission to get into attribute directory? 1055 */ 1056 1057 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) { 1058 VN_RELE(*vpp); 1059 } 1060 1061 ZFS_EXIT(zfsvfs); 1062 return (error); 1063 } 1064#endif /* TODO */ 1065 1066 if (dvp->v_type != VDIR) { 1067 ZFS_EXIT(zfsvfs); 1068 return (ENOTDIR); 1069 } 1070 1071 /* 1072 * Check accessibility of directory. 1073 */ 1074 1075 if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) { 1076 ZFS_EXIT(zfsvfs); 1077 return (error); 1078 } 1079 1080 if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) { 1081 1082 /* 1083 * Convert device special files 1084 */ 1085 if (IS_DEVVP(*vpp)) { 1086 vnode_t *svp; 1087 1088 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1089 VN_RELE(*vpp); 1090 if (svp == NULL) 1091 error = ENOSYS; 1092 else 1093 *vpp = svp; 1094 } 1095 } 1096 1097 ZFS_EXIT(zfsvfs); 1098 1099 /* Translate errors and add SAVENAME when needed. */ 1100 if (cnp->cn_flags & ISLASTCN) { 1101 switch (nameiop) { 1102 case CREATE: 1103 case RENAME: 1104 if (error == ENOENT) { 1105 error = EJUSTRETURN; 1106 cnp->cn_flags |= SAVENAME; 1107 break; 1108 } 1109 /* FALLTHROUGH */ 1110 case DELETE: 1111 if (error == 0) 1112 cnp->cn_flags |= SAVENAME; 1113 break; 1114 } 1115 } 1116 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) { 1117 int ltype = 0; 1118 1119 if (cnp->cn_flags & ISDOTDOT) { 1120 ltype = VOP_ISLOCKED(dvp); 1121 VOP_UNLOCK(dvp, 0); 1122 } 1123 error = vn_lock(*vpp, cnp->cn_lkflags); 1124 if (cnp->cn_flags & ISDOTDOT) 1125 vn_lock(dvp, ltype | LK_RETRY); 1126 if (error != 0) { 1127 VN_RELE(*vpp); 1128 *vpp = NULL; 1129 return (error); 1130 } 1131 } 1132 1133#ifdef FREEBSD_NAMECACHE 1134 /* 1135 * Insert name into cache (as non-existent) if appropriate. 1136 */ 1137 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE) 1138 cache_enter(dvp, *vpp, cnp); 1139 /* 1140 * Insert name into cache if appropriate. 1141 */ 1142 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1143 if (!(cnp->cn_flags & ISLASTCN) || 1144 (nameiop != DELETE && nameiop != RENAME)) { 1145 cache_enter(dvp, *vpp, cnp); 1146 } 1147 } 1148#endif 1149 1150 return (error); 1151} 1152 1153/* 1154 * Attempt to create a new entry in a directory. If the entry 1155 * already exists, truncate the file if permissible, else return 1156 * an error. Return the vp of the created or trunc'd file. 1157 * 1158 * IN: dvp - vnode of directory to put new file entry in. 1159 * name - name of new file entry. 1160 * vap - attributes of new file. 1161 * excl - flag indicating exclusive or non-exclusive mode. 1162 * mode - mode to open file with. 1163 * cr - credentials of caller. 1164 * flag - large file flag [UNUSED]. 1165 * 1166 * OUT: vpp - vnode of created or trunc'd entry. 1167 * 1168 * RETURN: 0 if success 1169 * error code if failure 1170 * 1171 * Timestamps: 1172 * dvp - ctime|mtime updated if new entry created 1173 * vp - ctime|mtime always, atime if new 1174 */ 1175/* ARGSUSED */ 1176static int 1177zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1178 vnode_t **vpp, cred_t *cr) 1179{ 1180 znode_t *zp, *dzp = VTOZ(dvp); 1181 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1182 zilog_t *zilog = zfsvfs->z_log; 1183 objset_t *os = zfsvfs->z_os; 1184 zfs_dirlock_t *dl; 1185 dmu_tx_t *tx; 1186 int error; 1187 uint64_t zoid; 1188 1189 ZFS_ENTER(zfsvfs); 1190 1191top: 1192 *vpp = NULL; 1193 1194 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1195 vap->va_mode &= ~VSVTX; 1196 1197 if (*name == '\0') { 1198 /* 1199 * Null component name refers to the directory itself. 1200 */ 1201 VN_HOLD(dvp); 1202 zp = dzp; 1203 dl = NULL; 1204 error = 0; 1205 } else { 1206 /* possible VN_HOLD(zp) */ 1207 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) { 1208 if (strcmp(name, "..") == 0) 1209 error = EISDIR; 1210 ZFS_EXIT(zfsvfs); 1211 return (error); 1212 } 1213 } 1214 1215 zoid = zp ? zp->z_id : -1ULL; 1216 1217 if (zp == NULL) { 1218 /* 1219 * Create a new file object and update the directory 1220 * to reference it. 1221 */ 1222 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 1223 goto out; 1224 } 1225 1226 /* 1227 * We only support the creation of regular files in 1228 * extended attribute directories. 1229 */ 1230 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1231 (vap->va_type != VREG)) { 1232 error = EINVAL; 1233 goto out; 1234 } 1235 1236 tx = dmu_tx_create(os); 1237 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1238 dmu_tx_hold_bonus(tx, dzp->z_id); 1239 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1240 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1241 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1242 0, SPA_MAXBLOCKSIZE); 1243 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1244 if (error) { 1245 zfs_dirent_unlock(dl); 1246 if (error == ERESTART && 1247 zfsvfs->z_assign == TXG_NOWAIT) { 1248 dmu_tx_wait(tx); 1249 dmu_tx_abort(tx); 1250 goto top; 1251 } 1252 dmu_tx_abort(tx); 1253 ZFS_EXIT(zfsvfs); 1254 return (error); 1255 } 1256 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1257 ASSERT(zp->z_id == zoid); 1258 (void) zfs_link_create(dl, zp, tx, ZNEW); 1259 zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name); 1260 dmu_tx_commit(tx); 1261 } else { 1262 /* 1263 * A directory entry already exists for this name. 1264 */ 1265 /* 1266 * Can't truncate an existing file if in exclusive mode. 1267 */ 1268 if (excl == EXCL) { 1269 error = EEXIST; 1270 goto out; 1271 } 1272 /* 1273 * Can't open a directory for writing. 1274 */ 1275 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1276 error = EISDIR; 1277 goto out; 1278 } 1279 /* 1280 * Verify requested access to file. 1281 */ 1282 if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) { 1283 goto out; 1284 } 1285 1286 mutex_enter(&dzp->z_lock); 1287 dzp->z_seq++; 1288 mutex_exit(&dzp->z_lock); 1289 1290 /* 1291 * Truncate regular files if requested. 1292 */ 1293 if ((ZTOV(zp)->v_type == VREG) && 1294 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1295 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1296 if (error == ERESTART && 1297 zfsvfs->z_assign == TXG_NOWAIT) { 1298 /* NB: we already did dmu_tx_wait() */ 1299 zfs_dirent_unlock(dl); 1300 VN_RELE(ZTOV(zp)); 1301 goto top; 1302 } 1303 } 1304 } 1305out: 1306 1307 if (error == 0) { 1308 *vpp = ZTOV(zp); 1309 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1310 } 1311 1312 if (dl) 1313 zfs_dirent_unlock(dl); 1314 1315 if (error) { 1316 if (zp) 1317 VN_RELE(ZTOV(zp)); 1318 } else { 1319 *vpp = ZTOV(zp); 1320 /* 1321 * If vnode is for a device return a specfs vnode instead. 1322 */ 1323 if (IS_DEVVP(*vpp)) { 1324 struct vnode *svp; 1325 1326 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1327 VN_RELE(*vpp); 1328 if (svp == NULL) { 1329 error = ENOSYS; 1330 } 1331 *vpp = svp; 1332 } 1333 } 1334 1335 ZFS_EXIT(zfsvfs); 1336 return (error); 1337} 1338 1339/* 1340 * Remove an entry from a directory. 1341 * 1342 * IN: dvp - vnode of directory to remove entry from. 1343 * name - name of entry to remove. 1344 * cr - credentials of caller. 1345 * 1346 * RETURN: 0 if success 1347 * error code if failure 1348 * 1349 * Timestamps: 1350 * dvp - ctime|mtime 1351 * vp - ctime (if nlink > 0) 1352 */ 1353static int 1354zfs_remove(vnode_t *dvp, char *name, cred_t *cr) 1355{ 1356 znode_t *zp, *dzp = VTOZ(dvp); 1357 znode_t *xzp = NULL; 1358 vnode_t *vp; 1359 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1360 zilog_t *zilog = zfsvfs->z_log; 1361 uint64_t acl_obj, xattr_obj; 1362 zfs_dirlock_t *dl; 1363 dmu_tx_t *tx; 1364 boolean_t may_delete_now, delete_now = FALSE; 1365 boolean_t unlinked; 1366 int error; 1367 1368 ZFS_ENTER(zfsvfs); 1369 1370top: 1371 /* 1372 * Attempt to lock directory; fail if entry doesn't exist. 1373 */ 1374 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1375 ZFS_EXIT(zfsvfs); 1376 return (error); 1377 } 1378 1379 vp = ZTOV(zp); 1380 1381 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1382 goto out; 1383 } 1384 1385 /* 1386 * Need to use rmdir for removing directories. 1387 */ 1388 if (vp->v_type == VDIR) { 1389 error = EPERM; 1390 goto out; 1391 } 1392 1393 vnevent_remove(vp); 1394 1395 dnlc_remove(dvp, name); 1396 1397 may_delete_now = FALSE; 1398 1399 /* 1400 * We may delete the znode now, or we may put it in the unlinked set; 1401 * it depends on whether we're the last link, and on whether there are 1402 * other holds on the vnode. So we dmu_tx_hold() the right things to 1403 * allow for either case. 1404 */ 1405 tx = dmu_tx_create(zfsvfs->z_os); 1406 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1407 dmu_tx_hold_bonus(tx, zp->z_id); 1408 if (may_delete_now) 1409 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 1410 1411 /* are there any extended attributes? */ 1412 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1413 /* XXX - do we need this if we are deleting? */ 1414 dmu_tx_hold_bonus(tx, xattr_obj); 1415 } 1416 1417 /* are there any additional acls */ 1418 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1419 may_delete_now) 1420 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1421 1422 /* charge as an update -- would be nice not to charge at all */ 1423 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1424 1425 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1426 if (error) { 1427 zfs_dirent_unlock(dl); 1428 VN_RELE(vp); 1429 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1430 dmu_tx_wait(tx); 1431 dmu_tx_abort(tx); 1432 goto top; 1433 } 1434 dmu_tx_abort(tx); 1435 ZFS_EXIT(zfsvfs); 1436 return (error); 1437 } 1438 1439 /* 1440 * Remove the directory entry. 1441 */ 1442 error = zfs_link_destroy(dl, zp, tx, 0, &unlinked); 1443 1444 if (error) { 1445 dmu_tx_commit(tx); 1446 goto out; 1447 } 1448 1449 if (0 && unlinked) { 1450 VI_LOCK(vp); 1451 delete_now = may_delete_now && 1452 vp->v_count == 1 && !vn_has_cached_data(vp) && 1453 zp->z_phys->zp_xattr == xattr_obj && 1454 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1455 VI_UNLOCK(vp); 1456 } 1457 1458 if (delete_now) { 1459 if (zp->z_phys->zp_xattr) { 1460 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1461 ASSERT3U(error, ==, 0); 1462 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1463 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1464 mutex_enter(&xzp->z_lock); 1465 xzp->z_unlinked = 1; 1466 xzp->z_phys->zp_links = 0; 1467 mutex_exit(&xzp->z_lock); 1468 zfs_unlinked_add(xzp, tx); 1469 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1470 } 1471 mutex_enter(&zp->z_lock); 1472 VI_LOCK(vp); 1473 vp->v_count--; 1474 ASSERT3U(vp->v_count, ==, 0); 1475 VI_UNLOCK(vp); 1476 mutex_exit(&zp->z_lock); 1477 zfs_znode_delete(zp, tx); 1478 VFS_RELE(zfsvfs->z_vfs); 1479 } else if (unlinked) { 1480 zfs_unlinked_add(zp, tx); 1481 } 1482 1483 zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name); 1484 1485 dmu_tx_commit(tx); 1486out: 1487 zfs_dirent_unlock(dl); 1488 1489 if (!delete_now) { 1490 VN_RELE(vp); 1491 } else if (xzp) { 1492 /* this rele delayed to prevent nesting transactions */ 1493 VN_RELE(ZTOV(xzp)); 1494 } 1495 1496 ZFS_EXIT(zfsvfs); 1497 return (error); 1498} 1499 1500/* 1501 * Create a new directory and insert it into dvp using the name 1502 * provided. Return a pointer to the inserted directory. 1503 * 1504 * IN: dvp - vnode of directory to add subdir to. 1505 * dirname - name of new directory. 1506 * vap - attributes of new directory. 1507 * cr - credentials of caller. 1508 * 1509 * OUT: vpp - vnode of created directory. 1510 * 1511 * RETURN: 0 if success 1512 * error code if failure 1513 * 1514 * Timestamps: 1515 * dvp - ctime|mtime updated 1516 * vp - ctime|mtime|atime updated 1517 */ 1518static int 1519zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr) 1520{ 1521 znode_t *zp, *dzp = VTOZ(dvp); 1522 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1523 zilog_t *zilog = zfsvfs->z_log; 1524 zfs_dirlock_t *dl; 1525 uint64_t zoid = 0; 1526 dmu_tx_t *tx; 1527 int error; 1528 1529 ASSERT(vap->va_type == VDIR); 1530 1531 ZFS_ENTER(zfsvfs); 1532 1533 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1534 ZFS_EXIT(zfsvfs); 1535 return (EINVAL); 1536 } 1537top: 1538 *vpp = NULL; 1539 1540 /* 1541 * First make sure the new directory doesn't exist. 1542 */ 1543 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) { 1544 ZFS_EXIT(zfsvfs); 1545 return (error); 1546 } 1547 1548 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) { 1549 zfs_dirent_unlock(dl); 1550 ZFS_EXIT(zfsvfs); 1551 return (error); 1552 } 1553 1554 /* 1555 * Add a new entry to the directory. 1556 */ 1557 tx = dmu_tx_create(zfsvfs->z_os); 1558 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1559 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1560 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1561 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1562 0, SPA_MAXBLOCKSIZE); 1563 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1564 if (error) { 1565 zfs_dirent_unlock(dl); 1566 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1567 dmu_tx_wait(tx); 1568 dmu_tx_abort(tx); 1569 goto top; 1570 } 1571 dmu_tx_abort(tx); 1572 ZFS_EXIT(zfsvfs); 1573 return (error); 1574 } 1575 1576 /* 1577 * Create new node. 1578 */ 1579 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1580 1581 /* 1582 * Now put new name in parent dir. 1583 */ 1584 (void) zfs_link_create(dl, zp, tx, ZNEW); 1585 1586 *vpp = ZTOV(zp); 1587 1588 zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname); 1589 dmu_tx_commit(tx); 1590 1591 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1592 1593 zfs_dirent_unlock(dl); 1594 1595 ZFS_EXIT(zfsvfs); 1596 return (0); 1597} 1598 1599/* 1600 * Remove a directory subdir entry. If the current working 1601 * directory is the same as the subdir to be removed, the 1602 * remove will fail. 1603 * 1604 * IN: dvp - vnode of directory to remove from. 1605 * name - name of directory to be removed. 1606 * cwd - vnode of current working directory. 1607 * cr - credentials of caller. 1608 * 1609 * RETURN: 0 if success 1610 * error code if failure 1611 * 1612 * Timestamps: 1613 * dvp - ctime|mtime updated 1614 */ 1615static int 1616zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr) 1617{ 1618 znode_t *dzp = VTOZ(dvp); 1619 znode_t *zp; 1620 vnode_t *vp; 1621 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1622 zilog_t *zilog = zfsvfs->z_log; 1623 zfs_dirlock_t *dl; 1624 dmu_tx_t *tx; 1625 int error; 1626 1627 ZFS_ENTER(zfsvfs); 1628 1629top: 1630 zp = NULL; 1631 1632 /* 1633 * Attempt to lock directory; fail if entry doesn't exist. 1634 */ 1635 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1636 ZFS_EXIT(zfsvfs); 1637 return (error); 1638 } 1639 1640 vp = ZTOV(zp); 1641 1642 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1643 goto out; 1644 } 1645 1646 if (vp->v_type != VDIR) { 1647 error = ENOTDIR; 1648 goto out; 1649 } 1650 1651 if (vp == cwd) { 1652 error = EINVAL; 1653 goto out; 1654 } 1655 1656 vnevent_rmdir(vp); 1657 1658 /* 1659 * Grab a lock on the directory to make sure that noone is 1660 * trying to add (or lookup) entries while we are removing it. 1661 */ 1662 rw_enter(&zp->z_name_lock, RW_WRITER); 1663 1664 /* 1665 * Grab a lock on the parent pointer to make sure we play well 1666 * with the treewalk and directory rename code. 1667 */ 1668 rw_enter(&zp->z_parent_lock, RW_WRITER); 1669 1670 tx = dmu_tx_create(zfsvfs->z_os); 1671 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1672 dmu_tx_hold_bonus(tx, zp->z_id); 1673 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1674 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1675 if (error) { 1676 rw_exit(&zp->z_parent_lock); 1677 rw_exit(&zp->z_name_lock); 1678 zfs_dirent_unlock(dl); 1679 VN_RELE(vp); 1680 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1681 dmu_tx_wait(tx); 1682 dmu_tx_abort(tx); 1683 goto top; 1684 } 1685 dmu_tx_abort(tx); 1686 ZFS_EXIT(zfsvfs); 1687 return (error); 1688 } 1689 1690#ifdef FREEBSD_NAMECACHE 1691 cache_purge(dvp); 1692#endif 1693 1694 error = zfs_link_destroy(dl, zp, tx, 0, NULL); 1695 1696 if (error == 0) 1697 zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name); 1698 1699 dmu_tx_commit(tx); 1700 1701 rw_exit(&zp->z_parent_lock); 1702 rw_exit(&zp->z_name_lock); 1703#ifdef FREEBSD_NAMECACHE 1704 cache_purge(vp); 1705#endif 1706out: 1707 zfs_dirent_unlock(dl); 1708 1709 VN_RELE(vp); 1710 1711 ZFS_EXIT(zfsvfs); 1712 return (error); 1713} 1714 1715/* 1716 * Read as many directory entries as will fit into the provided 1717 * buffer from the given directory cursor position (specified in 1718 * the uio structure. 1719 * 1720 * IN: vp - vnode of directory to read. 1721 * uio - structure supplying read location, range info, 1722 * and return buffer. 1723 * cr - credentials of caller. 1724 * 1725 * OUT: uio - updated offset and range, buffer filled. 1726 * eofp - set to true if end-of-file detected. 1727 * 1728 * RETURN: 0 if success 1729 * error code if failure 1730 * 1731 * Timestamps: 1732 * vp - atime updated 1733 * 1734 * Note that the low 4 bits of the cookie returned by zap is always zero. 1735 * This allows us to use the low range for "special" directory entries: 1736 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1737 * we use the offset 2 for the '.zfs' directory. 1738 */ 1739/* ARGSUSED */ 1740static int 1741zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 1742{ 1743 znode_t *zp = VTOZ(vp); 1744 iovec_t *iovp; 1745 dirent64_t *odp; 1746 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1747 objset_t *os; 1748 caddr_t outbuf; 1749 size_t bufsize; 1750 zap_cursor_t zc; 1751 zap_attribute_t zap; 1752 uint_t bytes_wanted; 1753 uint64_t offset; /* must be unsigned; checks for < 1 */ 1754 int local_eof; 1755 int outcount; 1756 int error; 1757 uint8_t prefetch; 1758 uint8_t type; 1759 int ncooks; 1760 u_long *cooks = NULL; 1761 1762 ZFS_ENTER(zfsvfs); 1763 1764 /* 1765 * If we are not given an eof variable, 1766 * use a local one. 1767 */ 1768 if (eofp == NULL) 1769 eofp = &local_eof; 1770 1771 /* 1772 * Check for valid iov_len. 1773 */ 1774 if (uio->uio_iov->iov_len <= 0) { 1775 ZFS_EXIT(zfsvfs); 1776 return (EINVAL); 1777 } 1778 1779 /* 1780 * Quit if directory has been removed (posix) 1781 */ 1782 if ((*eofp = zp->z_unlinked) != 0) { 1783 ZFS_EXIT(zfsvfs); 1784 return (0); 1785 } 1786 1787 error = 0; 1788 os = zfsvfs->z_os; 1789 offset = uio->uio_loffset; 1790 prefetch = zp->z_zn_prefetch; 1791 1792 /* 1793 * Initialize the iterator cursor. 1794 */ 1795 if (offset <= 3) { 1796 /* 1797 * Start iteration from the beginning of the directory. 1798 */ 1799 zap_cursor_init(&zc, os, zp->z_id); 1800 } else { 1801 /* 1802 * The offset is a serialized cursor. 1803 */ 1804 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1805 } 1806 1807 /* 1808 * Get space to change directory entries into fs independent format. 1809 */ 1810 iovp = uio->uio_iov; 1811 bytes_wanted = iovp->iov_len; 1812 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1813 bufsize = bytes_wanted; 1814 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1815 odp = (struct dirent64 *)outbuf; 1816 } else { 1817 bufsize = bytes_wanted; 1818 odp = (struct dirent64 *)iovp->iov_base; 1819 } 1820 1821 if (ncookies != NULL) { 1822 /* 1823 * Minimum entry size is dirent size and 1 byte for a file name. 1824 */ 1825 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 1826 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 1827 *cookies = cooks; 1828 *ncookies = ncooks; 1829 } 1830 1831 /* 1832 * Transform to file-system independent format 1833 */ 1834 outcount = 0; 1835 while (outcount < bytes_wanted) { 1836 ino64_t objnum; 1837 ushort_t reclen; 1838 1839 /* 1840 * Special case `.', `..', and `.zfs'. 1841 */ 1842 if (offset == 0) { 1843 (void) strcpy(zap.za_name, "."); 1844 objnum = zp->z_id; 1845 type = DT_DIR; 1846 } else if (offset == 1) { 1847 (void) strcpy(zap.za_name, ".."); 1848 objnum = zp->z_phys->zp_parent; 1849 type = DT_DIR; 1850 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1851 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1852 objnum = ZFSCTL_INO_ROOT; 1853 type = DT_DIR; 1854 } else { 1855 /* 1856 * Grab next entry. 1857 */ 1858 if (error = zap_cursor_retrieve(&zc, &zap)) { 1859 if ((*eofp = (error == ENOENT)) != 0) 1860 break; 1861 else 1862 goto update; 1863 } 1864 1865 if (zap.za_integer_length != 8 || 1866 zap.za_num_integers != 1) { 1867 cmn_err(CE_WARN, "zap_readdir: bad directory " 1868 "entry, obj = %lld, offset = %lld\n", 1869 (u_longlong_t)zp->z_id, 1870 (u_longlong_t)offset); 1871 error = ENXIO; 1872 goto update; 1873 } 1874 1875 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 1876 /* 1877 * MacOS X can extract the object type here such as: 1878 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1879 */ 1880 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1881 } 1882 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1883 1884 /* 1885 * Will this entry fit in the buffer? 1886 */ 1887 if (outcount + reclen > bufsize) { 1888 /* 1889 * Did we manage to fit anything in the buffer? 1890 */ 1891 if (!outcount) { 1892 error = EINVAL; 1893 goto update; 1894 } 1895 break; 1896 } 1897 /* 1898 * Add this entry: 1899 */ 1900 odp->d_ino = objnum; 1901 odp->d_reclen = reclen; 1902 odp->d_namlen = strlen(zap.za_name); 1903 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 1904 odp->d_type = type; 1905 outcount += reclen; 1906 odp = (dirent64_t *)((intptr_t)odp + reclen); 1907 1908 ASSERT(outcount <= bufsize); 1909 1910 /* Prefetch znode */ 1911 if (prefetch) 1912 dmu_prefetch(os, objnum, 0, 0); 1913 1914 /* 1915 * Move to the next entry, fill in the previous offset. 1916 */ 1917 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1918 zap_cursor_advance(&zc); 1919 offset = zap_cursor_serialize(&zc); 1920 } else { 1921 offset += 1; 1922 } 1923 1924 if (cooks != NULL) { 1925 *cooks++ = offset; 1926 ncooks--; 1927 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 1928 } 1929 } 1930 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1931 1932 /* Subtract unused cookies */ 1933 if (ncookies != NULL) 1934 *ncookies -= ncooks; 1935 1936 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 1937 iovp->iov_base += outcount; 1938 iovp->iov_len -= outcount; 1939 uio->uio_resid -= outcount; 1940 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 1941 /* 1942 * Reset the pointer. 1943 */ 1944 offset = uio->uio_loffset; 1945 } 1946 1947update: 1948 zap_cursor_fini(&zc); 1949 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 1950 kmem_free(outbuf, bufsize); 1951 1952 if (error == ENOENT) 1953 error = 0; 1954 1955 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1956 1957 uio->uio_loffset = offset; 1958 ZFS_EXIT(zfsvfs); 1959 if (error != 0 && cookies != NULL) { 1960 free(*cookies, M_TEMP); 1961 *cookies = NULL; 1962 *ncookies = 0; 1963 } 1964 return (error); 1965} 1966 1967static int 1968zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr) 1969{ 1970 znode_t *zp = VTOZ(vp); 1971 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1972 1973 ZFS_ENTER(zfsvfs); 1974 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 1975 ZFS_EXIT(zfsvfs); 1976 return (0); 1977} 1978 1979/* 1980 * Get the requested file attributes and place them in the provided 1981 * vattr structure. 1982 * 1983 * IN: vp - vnode of file. 1984 * vap - va_mask identifies requested attributes. 1985 * flags - [UNUSED] 1986 * cr - credentials of caller. 1987 * 1988 * OUT: vap - attribute values. 1989 * 1990 * RETURN: 0 (always succeeds) 1991 */ 1992/* ARGSUSED */ 1993static int 1994zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1995{ 1996 znode_t *zp = VTOZ(vp); 1997 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1998 znode_phys_t *pzp = zp->z_phys; 1999 uint32_t blksize; 2000 u_longlong_t nblocks; 2001 int error; 2002 2003 ZFS_ENTER(zfsvfs); 2004 2005 /* 2006 * Return all attributes. It's cheaper to provide the answer 2007 * than to determine whether we were asked the question. 2008 */ 2009 mutex_enter(&zp->z_lock); 2010 2011 vap->va_type = IFTOVT(pzp->zp_mode); 2012 vap->va_mode = pzp->zp_mode & ~S_IFMT; 2013 vap->va_uid = zp->z_phys->zp_uid; 2014 vap->va_gid = zp->z_phys->zp_gid; 2015 vap->va_nodeid = zp->z_id; 2016 vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX); /* nlink_t limit! */ 2017 vap->va_size = pzp->zp_size; 2018 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 2019 vap->va_rdev = zfs_cmpldev(pzp->zp_rdev); 2020 vap->va_seq = zp->z_seq; 2021 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2022 2023 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 2024 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 2025 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 2026 ZFS_TIME_DECODE(&vap->va_birthtime, pzp->zp_crtime); 2027 2028 /* 2029 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2030 * Also, if we are the owner don't bother, since owner should 2031 * always be allowed to read basic attributes of file. 2032 */ 2033 if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) && 2034 (zp->z_phys->zp_uid != crgetuid(cr))) { 2035 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) { 2036 mutex_exit(&zp->z_lock); 2037 ZFS_EXIT(zfsvfs); 2038 return (error); 2039 } 2040 } 2041 2042 mutex_exit(&zp->z_lock); 2043 2044 dmu_object_size_from_db(zp->z_dbuf, &blksize, &nblocks); 2045 vap->va_blksize = blksize; 2046 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2047 2048 if (zp->z_blksz == 0) { 2049 /* 2050 * Block size hasn't been set; suggest maximal I/O transfers. 2051 */ 2052 vap->va_blksize = zfsvfs->z_max_blksz; 2053 } 2054 2055 ZFS_EXIT(zfsvfs); 2056 return (0); 2057} 2058 2059/* 2060 * Set the file attributes to the values contained in the 2061 * vattr structure. 2062 * 2063 * IN: vp - vnode of file to be modified. 2064 * vap - new attribute values. 2065 * flags - ATTR_UTIME set if non-default time values provided. 2066 * cr - credentials of caller. 2067 * 2068 * RETURN: 0 if success 2069 * error code if failure 2070 * 2071 * Timestamps: 2072 * vp - ctime updated, mtime updated if size changed. 2073 */ 2074/* ARGSUSED */ 2075static int 2076zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2077 caller_context_t *ct) 2078{ 2079 struct znode *zp = VTOZ(vp); 2080 znode_phys_t *pzp = zp->z_phys; 2081 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2082 zilog_t *zilog = zfsvfs->z_log; 2083 dmu_tx_t *tx; 2084 vattr_t oldva; 2085 uint_t mask = vap->va_mask; 2086 uint_t saved_mask; 2087 int trim_mask = 0; 2088 uint64_t new_mode; 2089 znode_t *attrzp; 2090 int need_policy = FALSE; 2091 int err; 2092 2093 if (mask == 0) 2094 return (0); 2095 2096 if (mask & AT_NOSET) 2097 return (EINVAL); 2098 2099 if (mask & AT_SIZE && vp->v_type == VDIR) 2100 return (EISDIR); 2101 2102 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) 2103 return (EINVAL); 2104 2105 ZFS_ENTER(zfsvfs); 2106 2107top: 2108 attrzp = NULL; 2109 2110 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2111 ZFS_EXIT(zfsvfs); 2112 return (EROFS); 2113 } 2114 2115 /* 2116 * First validate permissions 2117 */ 2118 2119 if (mask & AT_SIZE) { 2120 err = zfs_zaccess(zp, ACE_WRITE_DATA, cr); 2121 if (err) { 2122 ZFS_EXIT(zfsvfs); 2123 return (err); 2124 } 2125 /* 2126 * XXX - Note, we are not providing any open 2127 * mode flags here (like FNDELAY), so we may 2128 * block if there are locks present... this 2129 * should be addressed in openat(). 2130 */ 2131 do { 2132 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2133 /* NB: we already did dmu_tx_wait() if necessary */ 2134 } while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 2135 if (err) { 2136 ZFS_EXIT(zfsvfs); 2137 return (err); 2138 } 2139 } 2140 2141 if (mask & (AT_ATIME|AT_MTIME)) 2142 need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr); 2143 2144 if (mask & (AT_UID|AT_GID)) { 2145 int idmask = (mask & (AT_UID|AT_GID)); 2146 int take_owner; 2147 int take_group; 2148 2149 /* 2150 * NOTE: even if a new mode is being set, 2151 * we may clear S_ISUID/S_ISGID bits. 2152 */ 2153 2154 if (!(mask & AT_MODE)) 2155 vap->va_mode = pzp->zp_mode; 2156 2157 /* 2158 * Take ownership or chgrp to group we are a member of 2159 */ 2160 2161 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2162 take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr); 2163 2164 /* 2165 * If both AT_UID and AT_GID are set then take_owner and 2166 * take_group must both be set in order to allow taking 2167 * ownership. 2168 * 2169 * Otherwise, send the check through secpolicy_vnode_setattr() 2170 * 2171 */ 2172 2173 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2174 ((idmask == AT_UID) && take_owner) || 2175 ((idmask == AT_GID) && take_group)) { 2176 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) { 2177 /* 2178 * Remove setuid/setgid for non-privileged users 2179 */ 2180 secpolicy_setid_clear(vap, cr); 2181 trim_mask = (mask & (AT_UID|AT_GID)); 2182 } else { 2183 need_policy = TRUE; 2184 } 2185 } else { 2186 need_policy = TRUE; 2187 } 2188 } 2189 2190 mutex_enter(&zp->z_lock); 2191 oldva.va_mode = pzp->zp_mode; 2192 oldva.va_uid = zp->z_phys->zp_uid; 2193 oldva.va_gid = zp->z_phys->zp_gid; 2194 mutex_exit(&zp->z_lock); 2195 2196 if (mask & AT_MODE) { 2197 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_ACL, cr) == 0) { 2198 err = secpolicy_setid_setsticky_clear(vp, vap, 2199 &oldva, cr); 2200 if (err) { 2201 ZFS_EXIT(zfsvfs); 2202 return (err); 2203 } 2204 trim_mask |= AT_MODE; 2205 } else { 2206 need_policy = TRUE; 2207 } 2208 } 2209 2210 if (need_policy) { 2211 /* 2212 * If trim_mask is set then take ownership 2213 * has been granted or write_acl is present and user 2214 * has the ability to modify mode. In that case remove 2215 * UID|GID and or MODE from mask so that 2216 * secpolicy_vnode_setattr() doesn't revoke it. 2217 */ 2218 2219 if (trim_mask) { 2220 saved_mask = vap->va_mask; 2221 vap->va_mask &= ~trim_mask; 2222 2223 } 2224 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2225 (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp); 2226 if (err) { 2227 ZFS_EXIT(zfsvfs); 2228 return (err); 2229 } 2230 2231 if (trim_mask) 2232 vap->va_mask |= saved_mask; 2233 } 2234 2235 /* 2236 * secpolicy_vnode_setattr, or take ownership may have 2237 * changed va_mask 2238 */ 2239 mask = vap->va_mask; 2240 2241 tx = dmu_tx_create(zfsvfs->z_os); 2242 dmu_tx_hold_bonus(tx, zp->z_id); 2243 2244 if (mask & AT_MODE) { 2245 uint64_t pmode = pzp->zp_mode; 2246 2247 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2248 2249 if (zp->z_phys->zp_acl.z_acl_extern_obj) 2250 dmu_tx_hold_write(tx, 2251 pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE); 2252 else 2253 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2254 0, ZFS_ACL_SIZE(MAX_ACL_SIZE)); 2255 } 2256 2257 if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) { 2258 err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp); 2259 if (err) { 2260 dmu_tx_abort(tx); 2261 ZFS_EXIT(zfsvfs); 2262 return (err); 2263 } 2264 dmu_tx_hold_bonus(tx, attrzp->z_id); 2265 } 2266 2267 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2268 if (err) { 2269 if (attrzp) 2270 VN_RELE(ZTOV(attrzp)); 2271 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2272 dmu_tx_wait(tx); 2273 dmu_tx_abort(tx); 2274 goto top; 2275 } 2276 dmu_tx_abort(tx); 2277 ZFS_EXIT(zfsvfs); 2278 return (err); 2279 } 2280 2281 dmu_buf_will_dirty(zp->z_dbuf, tx); 2282 2283 /* 2284 * Set each attribute requested. 2285 * We group settings according to the locks they need to acquire. 2286 * 2287 * Note: you cannot set ctime directly, although it will be 2288 * updated as a side-effect of calling this function. 2289 */ 2290 2291 mutex_enter(&zp->z_lock); 2292 2293 if (mask & AT_MODE) { 2294 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2295 ASSERT3U(err, ==, 0); 2296 } 2297 2298 if (attrzp) 2299 mutex_enter(&attrzp->z_lock); 2300 2301 if (mask & AT_UID) { 2302 zp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2303 if (attrzp) { 2304 attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2305 } 2306 } 2307 2308 if (mask & AT_GID) { 2309 zp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2310 if (attrzp) 2311 attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2312 } 2313 2314 if (attrzp) 2315 mutex_exit(&attrzp->z_lock); 2316 2317 if (mask & AT_ATIME) 2318 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2319 2320 if (mask & AT_MTIME) 2321 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2322 2323 if (mask & AT_SIZE) 2324 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2325 else if (mask != 0) 2326 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2327 2328 if (mask != 0) 2329 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask); 2330 2331 mutex_exit(&zp->z_lock); 2332 2333 if (attrzp) 2334 VN_RELE(ZTOV(attrzp)); 2335 2336 dmu_tx_commit(tx); 2337 2338 ZFS_EXIT(zfsvfs); 2339 return (err); 2340} 2341 2342typedef struct zfs_zlock { 2343 krwlock_t *zl_rwlock; /* lock we acquired */ 2344 znode_t *zl_znode; /* znode we held */ 2345 struct zfs_zlock *zl_next; /* next in list */ 2346} zfs_zlock_t; 2347 2348/* 2349 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2350 */ 2351static void 2352zfs_rename_unlock(zfs_zlock_t **zlpp) 2353{ 2354 zfs_zlock_t *zl; 2355 2356 while ((zl = *zlpp) != NULL) { 2357 if (zl->zl_znode != NULL) 2358 VN_RELE(ZTOV(zl->zl_znode)); 2359 rw_exit(zl->zl_rwlock); 2360 *zlpp = zl->zl_next; 2361 kmem_free(zl, sizeof (*zl)); 2362 } 2363} 2364 2365/* 2366 * Search back through the directory tree, using the ".." entries. 2367 * Lock each directory in the chain to prevent concurrent renames. 2368 * Fail any attempt to move a directory into one of its own descendants. 2369 * XXX - z_parent_lock can overlap with map or grow locks 2370 */ 2371static int 2372zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2373{ 2374 zfs_zlock_t *zl; 2375 znode_t *zp = tdzp; 2376 uint64_t rootid = zp->z_zfsvfs->z_root; 2377 uint64_t *oidp = &zp->z_id; 2378 krwlock_t *rwlp = &szp->z_parent_lock; 2379 krw_t rw = RW_WRITER; 2380 2381 /* 2382 * First pass write-locks szp and compares to zp->z_id. 2383 * Later passes read-lock zp and compare to zp->z_parent. 2384 */ 2385 do { 2386 if (!rw_tryenter(rwlp, rw)) { 2387 /* 2388 * Another thread is renaming in this path. 2389 * Note that if we are a WRITER, we don't have any 2390 * parent_locks held yet. 2391 */ 2392 if (rw == RW_READER && zp->z_id > szp->z_id) { 2393 /* 2394 * Drop our locks and restart 2395 */ 2396 zfs_rename_unlock(&zl); 2397 *zlpp = NULL; 2398 zp = tdzp; 2399 oidp = &zp->z_id; 2400 rwlp = &szp->z_parent_lock; 2401 rw = RW_WRITER; 2402 continue; 2403 } else { 2404 /* 2405 * Wait for other thread to drop its locks 2406 */ 2407 rw_enter(rwlp, rw); 2408 } 2409 } 2410 2411 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2412 zl->zl_rwlock = rwlp; 2413 zl->zl_znode = NULL; 2414 zl->zl_next = *zlpp; 2415 *zlpp = zl; 2416 2417 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2418 return (EINVAL); 2419 2420 if (*oidp == rootid) /* We've hit the top */ 2421 return (0); 2422 2423 if (rw == RW_READER) { /* i.e. not the first pass */ 2424 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2425 if (error) 2426 return (error); 2427 zl->zl_znode = zp; 2428 } 2429 oidp = &zp->z_phys->zp_parent; 2430 rwlp = &zp->z_parent_lock; 2431 rw = RW_READER; 2432 2433 } while (zp->z_id != sdzp->z_id); 2434 2435 return (0); 2436} 2437 2438/* 2439 * Move an entry from the provided source directory to the target 2440 * directory. Change the entry name as indicated. 2441 * 2442 * IN: sdvp - Source directory containing the "old entry". 2443 * snm - Old entry name. 2444 * tdvp - Target directory to contain the "new entry". 2445 * tnm - New entry name. 2446 * cr - credentials of caller. 2447 * 2448 * RETURN: 0 if success 2449 * error code if failure 2450 * 2451 * Timestamps: 2452 * sdvp,tdvp - ctime|mtime updated 2453 */ 2454static int 2455zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr) 2456{ 2457 znode_t *tdzp, *szp, *tzp; 2458 znode_t *sdzp = VTOZ(sdvp); 2459 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2460 zilog_t *zilog = zfsvfs->z_log; 2461 vnode_t *realvp; 2462 zfs_dirlock_t *sdl, *tdl; 2463 dmu_tx_t *tx; 2464 zfs_zlock_t *zl; 2465 int cmp, serr, terr, error; 2466 2467 ZFS_ENTER(zfsvfs); 2468 2469 /* 2470 * Make sure we have the real vp for the target directory. 2471 */ 2472 if (VOP_REALVP(tdvp, &realvp) == 0) 2473 tdvp = realvp; 2474 2475 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2476 ZFS_EXIT(zfsvfs); 2477 return (EXDEV); 2478 } 2479 2480 tdzp = VTOZ(tdvp); 2481top: 2482 szp = NULL; 2483 tzp = NULL; 2484 zl = NULL; 2485 2486 /* 2487 * This is to prevent the creation of links into attribute space 2488 * by renaming a linked file into/outof an attribute directory. 2489 * See the comment in zfs_link() for why this is considered bad. 2490 */ 2491 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2492 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2493 ZFS_EXIT(zfsvfs); 2494 return (EINVAL); 2495 } 2496 2497 /* 2498 * Lock source and target directory entries. To prevent deadlock, 2499 * a lock ordering must be defined. We lock the directory with 2500 * the smallest object id first, or if it's a tie, the one with 2501 * the lexically first name. 2502 */ 2503 if (sdzp->z_id < tdzp->z_id) { 2504 cmp = -1; 2505 } else if (sdzp->z_id > tdzp->z_id) { 2506 cmp = 1; 2507 } else { 2508 cmp = strcmp(snm, tnm); 2509 if (cmp == 0) { 2510 /* 2511 * POSIX: "If the old argument and the new argument 2512 * both refer to links to the same existing file, 2513 * the rename() function shall return successfully 2514 * and perform no other action." 2515 */ 2516 ZFS_EXIT(zfsvfs); 2517 return (0); 2518 } 2519 } 2520 if (cmp < 0) { 2521 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2522 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2523 } else { 2524 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2525 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2526 } 2527 2528 if (serr) { 2529 /* 2530 * Source entry invalid or not there. 2531 */ 2532 if (!terr) { 2533 zfs_dirent_unlock(tdl); 2534 if (tzp) 2535 VN_RELE(ZTOV(tzp)); 2536 } 2537 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 2538 serr = EINVAL; 2539 ZFS_EXIT(zfsvfs); 2540 return (serr); 2541 } 2542 if (terr) { 2543 zfs_dirent_unlock(sdl); 2544 VN_RELE(ZTOV(szp)); 2545 if (strcmp(tnm, "..") == 0) 2546 terr = EINVAL; 2547 ZFS_EXIT(zfsvfs); 2548 return (terr); 2549 } 2550 2551 /* 2552 * Must have write access at the source to remove the old entry 2553 * and write access at the target to create the new entry. 2554 * Note that if target and source are the same, this can be 2555 * done in a single check. 2556 */ 2557 2558 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2559 goto out; 2560 2561 if (ZTOV(szp)->v_type == VDIR) { 2562 /* 2563 * Check to make sure rename is valid. 2564 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2565 */ 2566 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 2567 goto out; 2568 } 2569 2570 /* 2571 * Does target exist? 2572 */ 2573 if (tzp) { 2574 /* 2575 * Source and target must be the same type. 2576 */ 2577 if (ZTOV(szp)->v_type == VDIR) { 2578 if (ZTOV(tzp)->v_type != VDIR) { 2579 error = ENOTDIR; 2580 goto out; 2581 } 2582 } else { 2583 if (ZTOV(tzp)->v_type == VDIR) { 2584 error = EISDIR; 2585 goto out; 2586 } 2587 } 2588 /* 2589 * POSIX dictates that when the source and target 2590 * entries refer to the same file object, rename 2591 * must do nothing and exit without error. 2592 */ 2593 if (szp->z_id == tzp->z_id) { 2594 error = 0; 2595 goto out; 2596 } 2597 } 2598 2599 vnevent_rename_src(ZTOV(szp)); 2600 if (tzp) 2601 vnevent_rename_dest(ZTOV(tzp)); 2602 2603 tx = dmu_tx_create(zfsvfs->z_os); 2604 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 2605 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 2606 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 2607 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 2608 if (sdzp != tdzp) 2609 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 2610 if (tzp) 2611 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 2612 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2613 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2614 if (error) { 2615 if (zl != NULL) 2616 zfs_rename_unlock(&zl); 2617 zfs_dirent_unlock(sdl); 2618 zfs_dirent_unlock(tdl); 2619 VN_RELE(ZTOV(szp)); 2620 if (tzp) 2621 VN_RELE(ZTOV(tzp)); 2622 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2623 dmu_tx_wait(tx); 2624 dmu_tx_abort(tx); 2625 goto top; 2626 } 2627 dmu_tx_abort(tx); 2628 ZFS_EXIT(zfsvfs); 2629 return (error); 2630 } 2631 2632 if (tzp) /* Attempt to remove the existing target */ 2633 error = zfs_link_destroy(tdl, tzp, tx, 0, NULL); 2634 2635 if (error == 0) { 2636 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 2637 if (error == 0) { 2638 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 2639 ASSERT(error == 0); 2640 zfs_log_rename(zilog, tx, TX_RENAME, sdzp, 2641 sdl->dl_name, tdzp, tdl->dl_name, szp); 2642 } 2643#ifdef FREEBSD_NAMECACHE 2644 if (error == 0) { 2645 cache_purge(sdvp); 2646 cache_purge(tdvp); 2647 } 2648#endif 2649 } 2650 2651 dmu_tx_commit(tx); 2652out: 2653 if (zl != NULL) 2654 zfs_rename_unlock(&zl); 2655 2656 zfs_dirent_unlock(sdl); 2657 zfs_dirent_unlock(tdl); 2658 2659 VN_RELE(ZTOV(szp)); 2660 if (tzp) 2661 VN_RELE(ZTOV(tzp)); 2662 2663 ZFS_EXIT(zfsvfs); 2664 2665 return (error); 2666} 2667 2668/* 2669 * Insert the indicated symbolic reference entry into the directory. 2670 * 2671 * IN: dvp - Directory to contain new symbolic link. 2672 * link - Name for new symlink entry. 2673 * vap - Attributes of new entry. 2674 * target - Target path of new symlink. 2675 * cr - credentials of caller. 2676 * 2677 * RETURN: 0 if success 2678 * error code if failure 2679 * 2680 * Timestamps: 2681 * dvp - ctime|mtime updated 2682 */ 2683static int 2684zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, cred_t *cr, kthread_t *td) 2685{ 2686 znode_t *zp, *dzp = VTOZ(dvp); 2687 zfs_dirlock_t *dl; 2688 dmu_tx_t *tx; 2689 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2690 zilog_t *zilog = zfsvfs->z_log; 2691 uint64_t zoid; 2692 int len = strlen(link); 2693 int error; 2694 2695 ASSERT(vap->va_type == VLNK); 2696 2697 ZFS_ENTER(zfsvfs); 2698top: 2699 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2700 ZFS_EXIT(zfsvfs); 2701 return (error); 2702 } 2703 2704 if (len > MAXPATHLEN) { 2705 ZFS_EXIT(zfsvfs); 2706 return (ENAMETOOLONG); 2707 } 2708 2709 /* 2710 * Attempt to lock directory; fail if entry already exists. 2711 */ 2712 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) { 2713 ZFS_EXIT(zfsvfs); 2714 return (error); 2715 } 2716 2717 tx = dmu_tx_create(zfsvfs->z_os); 2718 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 2719 dmu_tx_hold_bonus(tx, dzp->z_id); 2720 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2721 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 2722 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 2723 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2724 if (error) { 2725 zfs_dirent_unlock(dl); 2726 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2727 dmu_tx_wait(tx); 2728 dmu_tx_abort(tx); 2729 goto top; 2730 } 2731 dmu_tx_abort(tx); 2732 ZFS_EXIT(zfsvfs); 2733 return (error); 2734 } 2735 2736 dmu_buf_will_dirty(dzp->z_dbuf, tx); 2737 2738 /* 2739 * Create a new object for the symlink. 2740 * Put the link content into bonus buffer if it will fit; 2741 * otherwise, store it just like any other file data. 2742 */ 2743 zoid = 0; 2744 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 2745 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len); 2746 if (len != 0) 2747 bcopy(link, zp->z_phys + 1, len); 2748 } else { 2749 dmu_buf_t *dbp; 2750 2751 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 2752 2753 /* 2754 * Nothing can access the znode yet so no locking needed 2755 * for growing the znode's blocksize. 2756 */ 2757 zfs_grow_blocksize(zp, len, tx); 2758 2759 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp)); 2760 dmu_buf_will_dirty(dbp, tx); 2761 2762 ASSERT3U(len, <=, dbp->db_size); 2763 bcopy(link, dbp->db_data, len); 2764 dmu_buf_rele(dbp, FTAG); 2765 } 2766 zp->z_phys->zp_size = len; 2767 2768 /* 2769 * Insert the new object into the directory. 2770 */ 2771 (void) zfs_link_create(dl, zp, tx, ZNEW); 2772out: 2773 if (error == 0) { 2774 zfs_log_symlink(zilog, tx, TX_SYMLINK, dzp, zp, name, link); 2775 *vpp = ZTOV(zp); 2776 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 2777 } 2778 2779 dmu_tx_commit(tx); 2780 2781 zfs_dirent_unlock(dl); 2782 2783 ZFS_EXIT(zfsvfs); 2784 return (error); 2785} 2786 2787/* 2788 * Return, in the buffer contained in the provided uio structure, 2789 * the symbolic path referred to by vp. 2790 * 2791 * IN: vp - vnode of symbolic link. 2792 * uoip - structure to contain the link path. 2793 * cr - credentials of caller. 2794 * 2795 * OUT: uio - structure to contain the link path. 2796 * 2797 * RETURN: 0 if success 2798 * error code if failure 2799 * 2800 * Timestamps: 2801 * vp - atime updated 2802 */ 2803/* ARGSUSED */ 2804static int 2805zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr) 2806{ 2807 znode_t *zp = VTOZ(vp); 2808 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2809 size_t bufsz; 2810 int error; 2811 2812 ZFS_ENTER(zfsvfs); 2813 2814 bufsz = (size_t)zp->z_phys->zp_size; 2815 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 2816 error = uiomove(zp->z_phys + 1, 2817 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2818 } else { 2819 dmu_buf_t *dbp; 2820 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 2821 if (error) { 2822 ZFS_EXIT(zfsvfs); 2823 return (error); 2824 } 2825 error = uiomove(dbp->db_data, 2826 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2827 dmu_buf_rele(dbp, FTAG); 2828 } 2829 2830 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2831 ZFS_EXIT(zfsvfs); 2832 return (error); 2833} 2834 2835/* 2836 * Insert a new entry into directory tdvp referencing svp. 2837 * 2838 * IN: tdvp - Directory to contain new entry. 2839 * svp - vnode of new entry. 2840 * name - name of new entry. 2841 * cr - credentials of caller. 2842 * 2843 * RETURN: 0 if success 2844 * error code if failure 2845 * 2846 * Timestamps: 2847 * tdvp - ctime|mtime updated 2848 * svp - ctime updated 2849 */ 2850/* ARGSUSED */ 2851static int 2852zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr) 2853{ 2854 znode_t *dzp = VTOZ(tdvp); 2855 znode_t *tzp, *szp; 2856 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2857 zilog_t *zilog = zfsvfs->z_log; 2858 zfs_dirlock_t *dl; 2859 dmu_tx_t *tx; 2860 vnode_t *realvp; 2861 int error; 2862 2863 ASSERT(tdvp->v_type == VDIR); 2864 2865 ZFS_ENTER(zfsvfs); 2866 2867 if (VOP_REALVP(svp, &realvp) == 0) 2868 svp = realvp; 2869 2870 if (svp->v_vfsp != tdvp->v_vfsp) { 2871 ZFS_EXIT(zfsvfs); 2872 return (EXDEV); 2873 } 2874 2875 szp = VTOZ(svp); 2876top: 2877 /* 2878 * We do not support links between attributes and non-attributes 2879 * because of the potential security risk of creating links 2880 * into "normal" file space in order to circumvent restrictions 2881 * imposed in attribute space. 2882 */ 2883 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 2884 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 2885 ZFS_EXIT(zfsvfs); 2886 return (EINVAL); 2887 } 2888 2889 /* 2890 * POSIX dictates that we return EPERM here. 2891 * Better choices include ENOTSUP or EISDIR. 2892 */ 2893 if (svp->v_type == VDIR) { 2894 ZFS_EXIT(zfsvfs); 2895 return (EPERM); 2896 } 2897 2898 if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) && 2899 secpolicy_basic_link(cr) != 0) { 2900 ZFS_EXIT(zfsvfs); 2901 return (EPERM); 2902 } 2903 2904 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2905 ZFS_EXIT(zfsvfs); 2906 return (error); 2907 } 2908 2909 /* 2910 * Attempt to lock directory; fail if entry already exists. 2911 */ 2912 if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) { 2913 ZFS_EXIT(zfsvfs); 2914 return (error); 2915 } 2916 2917 tx = dmu_tx_create(zfsvfs->z_os); 2918 dmu_tx_hold_bonus(tx, szp->z_id); 2919 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2920 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2921 if (error) { 2922 zfs_dirent_unlock(dl); 2923 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2924 dmu_tx_wait(tx); 2925 dmu_tx_abort(tx); 2926 goto top; 2927 } 2928 dmu_tx_abort(tx); 2929 ZFS_EXIT(zfsvfs); 2930 return (error); 2931 } 2932 2933 error = zfs_link_create(dl, szp, tx, 0); 2934 2935 if (error == 0) 2936 zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name); 2937 2938 dmu_tx_commit(tx); 2939 2940 zfs_dirent_unlock(dl); 2941 2942 ZFS_EXIT(zfsvfs); 2943 return (error); 2944} 2945 2946void 2947zfs_inactive(vnode_t *vp, cred_t *cr) 2948{ 2949 znode_t *zp = VTOZ(vp); 2950 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2951 int error; 2952 2953 rw_enter(&zfsvfs->z_um_lock, RW_READER); 2954 if (zfsvfs->z_unmounted2) { 2955 ASSERT(zp->z_dbuf_held == 0); 2956 2957 mutex_enter(&zp->z_lock); 2958 VI_LOCK(vp); 2959 vp->v_count = 0; /* count arrives as 1 */ 2960 VI_UNLOCK(vp); 2961 if (zp->z_dbuf == NULL) { 2962 mutex_exit(&zp->z_lock); 2963 zfs_znode_free(zp); 2964 } else { 2965 mutex_exit(&zp->z_lock); 2966 } 2967 rw_exit(&zfsvfs->z_um_lock); 2968 VFS_RELE(zfsvfs->z_vfs); 2969 return; 2970 } 2971 2972 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 2973 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 2974 2975 dmu_tx_hold_bonus(tx, zp->z_id); 2976 error = dmu_tx_assign(tx, TXG_WAIT); 2977 if (error) { 2978 dmu_tx_abort(tx); 2979 } else { 2980 dmu_buf_will_dirty(zp->z_dbuf, tx); 2981 mutex_enter(&zp->z_lock); 2982 zp->z_atime_dirty = 0; 2983 mutex_exit(&zp->z_lock); 2984 dmu_tx_commit(tx); 2985 } 2986 } 2987 2988 zfs_zinactive(zp); 2989 rw_exit(&zfsvfs->z_um_lock); 2990} 2991 2992CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 2993CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 2994 2995static int 2996zfs_fid(vnode_t *vp, fid_t *fidp) 2997{ 2998 znode_t *zp = VTOZ(vp); 2999 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3000 uint32_t gen = (uint32_t)zp->z_phys->zp_gen; 3001 uint64_t object = zp->z_id; 3002 zfid_short_t *zfid; 3003 int size, i; 3004 3005 ZFS_ENTER(zfsvfs); 3006 3007 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 3008 fidp->fid_len = size; 3009 3010 zfid = (zfid_short_t *)fidp; 3011 3012 zfid->zf_len = size; 3013 3014 for (i = 0; i < sizeof (zfid->zf_object); i++) 3015 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 3016 3017 /* Must have a non-zero generation number to distinguish from .zfs */ 3018 if (gen == 0) 3019 gen = 1; 3020 for (i = 0; i < sizeof (zfid->zf_gen); i++) 3021 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 3022 3023 if (size == LONG_FID_LEN) { 3024 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 3025 zfid_long_t *zlfid; 3026 3027 zlfid = (zfid_long_t *)fidp; 3028 3029 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 3030 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 3031 3032 /* XXX - this should be the generation number for the objset */ 3033 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 3034 zlfid->zf_setgen[i] = 0; 3035 } 3036 3037 ZFS_EXIT(zfsvfs); 3038 return (0); 3039} 3040 3041static int 3042zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr) 3043{ 3044 znode_t *zp, *xzp; 3045 zfsvfs_t *zfsvfs; 3046 zfs_dirlock_t *dl; 3047 int error; 3048 3049 switch (cmd) { 3050 case _PC_LINK_MAX: 3051 *valp = INT_MAX; 3052 return (0); 3053 3054 case _PC_FILESIZEBITS: 3055 *valp = 64; 3056 return (0); 3057 3058#if 0 3059 case _PC_XATTR_EXISTS: 3060 zp = VTOZ(vp); 3061 zfsvfs = zp->z_zfsvfs; 3062 ZFS_ENTER(zfsvfs); 3063 *valp = 0; 3064 error = zfs_dirent_lock(&dl, zp, "", &xzp, 3065 ZXATTR | ZEXISTS | ZSHARED); 3066 if (error == 0) { 3067 zfs_dirent_unlock(dl); 3068 if (!zfs_dirempty(xzp)) 3069 *valp = 1; 3070 VN_RELE(ZTOV(xzp)); 3071 } else if (error == ENOENT) { 3072 /* 3073 * If there aren't extended attributes, it's the 3074 * same as having zero of them. 3075 */ 3076 error = 0; 3077 } 3078 ZFS_EXIT(zfsvfs); 3079 return (error); 3080#endif 3081 3082 case _PC_ACL_EXTENDED: 3083 *valp = 0; /* TODO */ 3084 return (0); 3085 3086 case _PC_MIN_HOLE_SIZE: 3087 *valp = (int)SPA_MINBLOCKSIZE; 3088 return (0); 3089 3090 default: 3091 return (EOPNOTSUPP); 3092 } 3093} 3094 3095#ifdef TODO 3096/*ARGSUSED*/ 3097static int 3098zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3099{ 3100 znode_t *zp = VTOZ(vp); 3101 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3102 int error; 3103 3104 ZFS_ENTER(zfsvfs); 3105 error = zfs_getacl(zp, vsecp, cr); 3106 ZFS_EXIT(zfsvfs); 3107 3108 return (error); 3109} 3110#endif /* TODO */ 3111 3112#ifdef TODO 3113/*ARGSUSED*/ 3114static int 3115zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3116{ 3117 znode_t *zp = VTOZ(vp); 3118 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3119 int error; 3120 3121 ZFS_ENTER(zfsvfs); 3122 error = zfs_setacl(zp, vsecp, cr); 3123 ZFS_EXIT(zfsvfs); 3124 return (error); 3125} 3126#endif /* TODO */ 3127 3128static int 3129zfs_freebsd_open(ap) 3130 struct vop_open_args /* { 3131 struct vnode *a_vp; 3132 int a_mode; 3133 struct ucred *a_cred; 3134 struct thread *a_td; 3135 } */ *ap; 3136{ 3137 vnode_t *vp = ap->a_vp; 3138 znode_t *zp = VTOZ(vp); 3139 int error; 3140 3141 error = zfs_open(&vp, ap->a_mode, ap->a_cred); 3142 if (error == 0) 3143 vnode_create_vobject(vp, zp->z_phys->zp_size, ap->a_td); 3144 return (error); 3145} 3146 3147static int 3148zfs_freebsd_close(ap) 3149 struct vop_close_args /* { 3150 struct vnode *a_vp; 3151 int a_fflag; 3152 struct ucred *a_cred; 3153 struct thread *a_td; 3154 } */ *ap; 3155{ 3156 3157 return (zfs_close(ap->a_vp, ap->a_fflag, 0, 0, ap->a_cred)); 3158} 3159 3160static int 3161zfs_freebsd_ioctl(ap) 3162 struct vop_ioctl_args /* { 3163 struct vnode *a_vp; 3164 u_long a_command; 3165 caddr_t a_data; 3166 int a_fflag; 3167 struct ucred *cred; 3168 struct thread *td; 3169 } */ *ap; 3170{ 3171 3172 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 3173 ap->a_fflag, ap->a_cred, NULL)); 3174} 3175 3176static int 3177zfs_freebsd_read(ap) 3178 struct vop_read_args /* { 3179 struct vnode *a_vp; 3180 struct uio *a_uio; 3181 int a_ioflag; 3182 struct ucred *a_cred; 3183 } */ *ap; 3184{ 3185 3186 return (zfs_read(ap->a_vp, ap->a_uio, ap->a_ioflag, ap->a_cred, NULL)); 3187} 3188 3189static int 3190zfs_freebsd_write(ap) 3191 struct vop_write_args /* { 3192 struct vnode *a_vp; 3193 struct uio *a_uio; 3194 int a_ioflag; 3195 struct ucred *a_cred; 3196 } */ *ap; 3197{ 3198 3199 return (zfs_write(ap->a_vp, ap->a_uio, ap->a_ioflag, ap->a_cred, NULL)); 3200} 3201 3202static int 3203zfs_freebsd_access(ap) 3204 struct vop_access_args /* { 3205 struct vnode *a_vp; 3206 int a_mode; 3207 struct ucred *a_cred; 3208 struct thread *a_td; 3209 } */ *ap; 3210{ 3211 3212 return (zfs_access(ap->a_vp, ap->a_mode, 0, ap->a_cred)); 3213} 3214 3215static int 3216zfs_freebsd_lookup(ap) 3217 struct vop_lookup_args /* { 3218 struct vnode *a_dvp; 3219 struct vnode **a_vpp; 3220 struct componentname *a_cnp; 3221 } */ *ap; 3222{ 3223 struct componentname *cnp = ap->a_cnp; 3224 char nm[NAME_MAX + 1]; 3225 3226 ASSERT(cnp->cn_namelen < sizeof(nm)); 3227 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 3228 3229 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 3230 cnp->cn_cred, cnp->cn_thread)); 3231} 3232 3233static int 3234zfs_freebsd_create(ap) 3235 struct vop_create_args /* { 3236 struct vnode *a_dvp; 3237 struct vnode **a_vpp; 3238 struct componentname *a_cnp; 3239 struct vattr *a_vap; 3240 } */ *ap; 3241{ 3242 struct componentname *cnp = ap->a_cnp; 3243 vattr_t *vap = ap->a_vap; 3244 int mode; 3245 3246 ASSERT(cnp->cn_flags & SAVENAME); 3247 3248 vattr_init_mask(vap); 3249 mode = vap->va_mode & ALLPERMS; 3250 3251 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 3252 ap->a_vpp, cnp->cn_cred)); 3253} 3254 3255static int 3256zfs_freebsd_remove(ap) 3257 struct vop_remove_args /* { 3258 struct vnode *a_dvp; 3259 struct vnode *a_vp; 3260 struct componentname *a_cnp; 3261 } */ *ap; 3262{ 3263 3264 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 3265 3266 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 3267 ap->a_cnp->cn_cred)); 3268} 3269 3270static int 3271zfs_freebsd_mkdir(ap) 3272 struct vop_mkdir_args /* { 3273 struct vnode *a_dvp; 3274 struct vnode **a_vpp; 3275 struct componentname *a_cnp; 3276 struct vattr *a_vap; 3277 } */ *ap; 3278{ 3279 vattr_t *vap = ap->a_vap; 3280 3281 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 3282 3283 vattr_init_mask(vap); 3284 3285 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 3286 ap->a_cnp->cn_cred)); 3287} 3288 3289static int 3290zfs_freebsd_rmdir(ap) 3291 struct vop_rmdir_args /* { 3292 struct vnode *a_dvp; 3293 struct vnode *a_vp; 3294 struct componentname *a_cnp; 3295 } */ *ap; 3296{ 3297 struct componentname *cnp = ap->a_cnp; 3298 3299 ASSERT(cnp->cn_flags & SAVENAME); 3300 3301 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred)); 3302} 3303 3304static int 3305zfs_freebsd_readdir(ap) 3306 struct vop_readdir_args /* { 3307 struct vnode *a_vp; 3308 struct uio *a_uio; 3309 struct ucred *a_cred; 3310 int *a_eofflag; 3311 int *a_ncookies; 3312 u_long **a_cookies; 3313 } */ *ap; 3314{ 3315 3316 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 3317 ap->a_ncookies, ap->a_cookies)); 3318} 3319 3320static int 3321zfs_freebsd_fsync(ap) 3322 struct vop_fsync_args /* { 3323 struct vnode *a_vp; 3324 int a_waitfor; 3325 struct thread *a_td; 3326 } */ *ap; 3327{ 3328 3329 vop_stdfsync(ap); 3330 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred)); 3331} 3332 3333static int 3334zfs_freebsd_getattr(ap) 3335 struct vop_getattr_args /* { 3336 struct vnode *a_vp; 3337 struct vattr *a_vap; 3338 struct ucred *a_cred; 3339 struct thread *a_td; 3340 } */ *ap; 3341{ 3342 3343 return (zfs_getattr(ap->a_vp, ap->a_vap, 0, ap->a_cred)); 3344} 3345 3346static int 3347zfs_freebsd_setattr(ap) 3348 struct vop_setattr_args /* { 3349 struct vnode *a_vp; 3350 struct vattr *a_vap; 3351 struct ucred *a_cred; 3352 struct thread *a_td; 3353 } */ *ap; 3354{ 3355 vattr_t *vap = ap->a_vap; 3356 3357 /* No support for FreeBSD's chflags(2). */ 3358 if (vap->va_flags != VNOVAL) 3359 return (EOPNOTSUPP); 3360 3361 vattr_init_mask(vap); 3362 vap->va_mask &= ~AT_NOSET; 3363 3364 return (zfs_setattr(ap->a_vp, vap, 0, ap->a_cred, NULL)); 3365} 3366 3367static int 3368zfs_freebsd_rename(ap) 3369 struct vop_rename_args /* { 3370 struct vnode *a_fdvp; 3371 struct vnode *a_fvp; 3372 struct componentname *a_fcnp; 3373 struct vnode *a_tdvp; 3374 struct vnode *a_tvp; 3375 struct componentname *a_tcnp; 3376 } */ *ap; 3377{ 3378 vnode_t *fdvp = ap->a_fdvp; 3379 vnode_t *fvp = ap->a_fvp; 3380 vnode_t *tdvp = ap->a_tdvp; 3381 vnode_t *tvp = ap->a_tvp; 3382 int error; 3383 3384 ASSERT(ap->a_fcnp->cn_flags & SAVENAME); 3385 ASSERT(ap->a_tcnp->cn_flags & SAVENAME); 3386 3387 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 3388 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred); 3389 3390 if (tdvp == tvp) 3391 VN_RELE(tdvp); 3392 else 3393 VN_URELE(tdvp); 3394 if (tvp) 3395 VN_URELE(tvp); 3396 VN_RELE(fdvp); 3397 VN_RELE(fvp); 3398 3399 return (error); 3400} 3401 3402static int 3403zfs_freebsd_symlink(ap) 3404 struct vop_symlink_args /* { 3405 struct vnode *a_dvp; 3406 struct vnode **a_vpp; 3407 struct componentname *a_cnp; 3408 struct vattr *a_vap; 3409 char *a_target; 3410 } */ *ap; 3411{ 3412 struct componentname *cnp = ap->a_cnp; 3413 vattr_t *vap = ap->a_vap; 3414 3415 ASSERT(cnp->cn_flags & SAVENAME); 3416 3417 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 3418 vattr_init_mask(vap); 3419 3420 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 3421 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 3422} 3423 3424static int 3425zfs_freebsd_readlink(ap) 3426 struct vop_readlink_args /* { 3427 struct vnode *a_vp; 3428 struct uio *a_uio; 3429 struct ucred *a_cred; 3430 } */ *ap; 3431{ 3432 3433 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred)); 3434} 3435 3436static int 3437zfs_freebsd_link(ap) 3438 struct vop_link_args /* { 3439 struct vnode *a_tdvp; 3440 struct vnode *a_vp; 3441 struct componentname *a_cnp; 3442 } */ *ap; 3443{ 3444 struct componentname *cnp = ap->a_cnp; 3445 3446 ASSERT(cnp->cn_flags & SAVENAME); 3447 3448 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred)); 3449} 3450 3451static int 3452zfs_freebsd_inactive(ap) 3453 struct vop_inactive_args /* { 3454 struct vnode *a_vp; 3455 struct thread *a_td; 3456 } */ *ap; 3457{ 3458 vnode_t *vp = ap->a_vp; 3459 3460 zfs_inactive(vp, ap->a_td->td_ucred); 3461 return (0); 3462} 3463 3464static int 3465zfs_freebsd_reclaim(ap) 3466 struct vop_reclaim_args /* { 3467 struct vnode *a_vp; 3468 struct thread *a_td; 3469 } */ *ap; 3470{ 3471 vnode_t *vp = ap->a_vp; 3472 znode_t *zp = VTOZ(vp); 3473 zfsvfs_t *zfsvfs; 3474 int rele = 1; 3475 3476 ASSERT(zp != NULL); 3477 3478 /* 3479 * Destroy the vm object and flush associated pages. 3480 */ 3481 vnode_destroy_vobject(vp); 3482 3483 mutex_enter(&zp->z_lock); 3484 ASSERT(zp->z_phys); 3485 ASSERT(zp->z_dbuf_held); 3486 zfsvfs = zp->z_zfsvfs; 3487 if (!zp->z_unlinked) { 3488 zp->z_dbuf_held = 0; 3489 ZTOV(zp) = NULL; 3490 mutex_exit(&zp->z_lock); 3491 dmu_buf_rele(zp->z_dbuf, NULL); 3492 } else { 3493 mutex_exit(&zp->z_lock); 3494 } 3495 VI_LOCK(vp); 3496 if (vp->v_count > 0) 3497 rele = 0; 3498 vp->v_data = NULL; 3499 ASSERT(vp->v_holdcnt >= 1); 3500 VI_UNLOCK(vp); 3501 if (!zp->z_unlinked && rele) 3502 VFS_RELE(zfsvfs->z_vfs); 3503 return (0); 3504} 3505 3506static int 3507zfs_freebsd_fid(ap) 3508 struct vop_fid_args /* { 3509 struct vnode *a_vp; 3510 struct fid *a_fid; 3511 } */ *ap; 3512{ 3513 3514 return (zfs_fid(ap->a_vp, (void *)ap->a_fid)); 3515} 3516 3517static int 3518zfs_freebsd_pathconf(ap) 3519 struct vop_pathconf_args /* { 3520 struct vnode *a_vp; 3521 int a_name; 3522 register_t *a_retval; 3523 } */ *ap; 3524{ 3525 ulong_t val; 3526 int error; 3527 3528 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred); 3529 if (error == 0) 3530 *ap->a_retval = val; 3531 else if (error == EOPNOTSUPP) 3532 error = vop_stdpathconf(ap); 3533 return (error); 3534} 3535 3536/* 3537 * Advisory record locking support 3538 */ 3539static int 3540zfs_freebsd_advlock(ap) 3541 struct vop_advlock_args /* { 3542 struct vnode *a_vp; 3543 caddr_t a_id; 3544 int a_op; 3545 struct flock *a_fl; 3546 int a_flags; 3547 } */ *ap; 3548{ 3549 znode_t *zp = VTOZ(ap->a_vp); 3550 3551 return (lf_advlock(ap, &(zp->z_lockf), zp->z_phys->zp_size)); 3552} 3553 3554/* 3555 * Advisory record locking support 3556 */ 3557static int 3558zfs_freebsd_advlockasync(ap) 3559 struct vop_advlockasync_args /* { 3560 struct vnode *a_vp; 3561 caddr_t a_id; 3562 int a_op; 3563 struct flock *a_fl; 3564 int a_flags; 3565 struct task *a_task; 3566 } */ *ap; 3567{ 3568 znode_t *zp = VTOZ(ap->a_vp); 3569 3570 return (lf_advlockasync(ap, &(zp->z_lockf), zp->z_phys->zp_size)); 3571} 3572 3573struct vop_vector zfs_vnodeops; 3574struct vop_vector zfs_fifoops; 3575 3576struct vop_vector zfs_vnodeops = { 3577 .vop_default = &default_vnodeops, 3578 .vop_inactive = zfs_freebsd_inactive, 3579 .vop_reclaim = zfs_freebsd_reclaim, 3580 .vop_access = zfs_freebsd_access, 3581#ifdef FREEBSD_NAMECACHE 3582 .vop_lookup = vfs_cache_lookup, 3583 .vop_cachedlookup = zfs_freebsd_lookup, 3584#else 3585 .vop_lookup = zfs_freebsd_lookup, 3586#endif 3587 .vop_getattr = zfs_freebsd_getattr, 3588 .vop_setattr = zfs_freebsd_setattr, 3589 .vop_create = zfs_freebsd_create, 3590 .vop_mknod = zfs_freebsd_create, 3591 .vop_mkdir = zfs_freebsd_mkdir, 3592 .vop_readdir = zfs_freebsd_readdir, 3593 .vop_fsync = zfs_freebsd_fsync, 3594 .vop_open = zfs_freebsd_open, 3595 .vop_close = zfs_freebsd_close, 3596 .vop_rmdir = zfs_freebsd_rmdir, 3597 .vop_ioctl = zfs_freebsd_ioctl, 3598 .vop_link = zfs_freebsd_link, 3599 .vop_symlink = zfs_freebsd_symlink, 3600 .vop_readlink = zfs_freebsd_readlink, 3601 .vop_read = zfs_freebsd_read, 3602 .vop_write = zfs_freebsd_write, 3603 .vop_remove = zfs_freebsd_remove, 3604 .vop_rename = zfs_freebsd_rename, 3605 .vop_advlock = zfs_freebsd_advlock, 3606 .vop_advlockasync = zfs_freebsd_advlockasync, 3607 .vop_pathconf = zfs_freebsd_pathconf, 3608 .vop_bmap = VOP_EOPNOTSUPP, 3609 .vop_fid = zfs_freebsd_fid, 3610}; 3611 3612struct vop_vector zfs_fifoops = { 3613 .vop_default = &fifo_specops, 3614 .vop_fsync = VOP_PANIC, 3615 .vop_access = zfs_freebsd_access, 3616 .vop_getattr = zfs_freebsd_getattr, 3617 .vop_inactive = zfs_freebsd_inactive, 3618 .vop_read = VOP_PANIC, 3619 .vop_reclaim = zfs_freebsd_reclaim, 3620 .vop_setattr = zfs_freebsd_setattr, 3621 .vop_write = VOP_PANIC, 3622 .vop_fid = zfs_freebsd_fid, 3623}; 3624