zfs_vnops.c revision 169195
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 dirbytes += bytes; 350 } 351 len -= bytes; 352 off = 0; 353 if (error) 354 break; 355 } 356 VM_OBJECT_UNLOCK(obj); 357 if (error == 0 && dirbytes > 0) 358 error = dmu_write_uio(os, zp->z_id, uio, dirbytes, tx); 359 return (error); 360} 361 362/* 363 * When a file is memory mapped, we must keep the IO data synchronized 364 * between the DMU cache and the memory mapped pages. What this means: 365 * 366 * On Read: We "read" preferentially from memory mapped pages, 367 * else we default from the dmu buffer. 368 * 369 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 370 * the file is memory mapped. 371 */ 372static int 373mappedread(vnode_t *vp, int nbytes, uio_t *uio) 374{ 375 znode_t *zp = VTOZ(vp); 376 objset_t *os = zp->z_zfsvfs->z_os; 377 vm_object_t obj; 378 vm_page_t m; 379 struct sf_buf *sf; 380 int64_t start, off; 381 caddr_t va; 382 int len = nbytes; 383 int error = 0; 384 uint64_t dirbytes; 385 386 ASSERT(vp->v_mount != NULL); 387 obj = vp->v_object; 388 ASSERT(obj != NULL); 389 390 start = uio->uio_loffset; 391 off = start & PAGEOFFSET; 392 dirbytes = 0; 393 VM_OBJECT_LOCK(obj); 394 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 395 uint64_t bytes = MIN(PAGESIZE - off, len); 396 397again: 398 if ((m = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 399 vm_page_is_valid(m, (vm_offset_t)off, bytes)) { 400 if (vm_page_sleep_if_busy(m, FALSE, "zfsmrb")) 401 goto again; 402 vm_page_busy(m); 403 VM_OBJECT_UNLOCK(obj); 404 if (dirbytes > 0) { 405 error = dmu_read_uio(os, zp->z_id, uio, 406 dirbytes); 407 dirbytes = 0; 408 } 409 if (error == 0) { 410 sched_pin(); 411 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 412 va = (caddr_t)sf_buf_kva(sf); 413 error = uiomove(va + off, bytes, UIO_READ, uio); 414 sf_buf_free(sf); 415 sched_unpin(); 416 } 417 VM_OBJECT_LOCK(obj); 418 vm_page_wakeup(m); 419 } else if (m != NULL && uio->uio_segflg == UIO_NOCOPY) { 420 /* 421 * The code below is here to make sendfile(2) work 422 * correctly with ZFS. As pointed out by ups@ 423 * sendfile(2) should be changed to use VOP_GETPAGES(), 424 * but it pessimize performance of sendfile/UFS, that's 425 * why I handle this special case in ZFS code. 426 */ 427 if (vm_page_sleep_if_busy(m, FALSE, "zfsmrb")) 428 goto again; 429 vm_page_busy(m); 430 VM_OBJECT_UNLOCK(obj); 431 if (dirbytes > 0) { 432 error = dmu_read_uio(os, zp->z_id, uio, 433 dirbytes); 434 dirbytes = 0; 435 } 436 if (error == 0) { 437 sched_pin(); 438 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 439 va = (caddr_t)sf_buf_kva(sf); 440 error = dmu_read(os, zp->z_id, start + off, 441 bytes, (void *)(va + off)); 442 sf_buf_free(sf); 443 sched_unpin(); 444 } 445 VM_OBJECT_LOCK(obj); 446 vm_page_wakeup(m); 447 if (error == 0) 448 uio->uio_resid -= bytes; 449 } else { 450 dirbytes += bytes; 451 } 452 len -= bytes; 453 off = 0; 454 if (error) 455 break; 456 } 457 VM_OBJECT_UNLOCK(obj); 458 if (error == 0 && dirbytes > 0) 459 error = dmu_read_uio(os, zp->z_id, uio, dirbytes); 460 return (error); 461} 462 463offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 464 465/* 466 * Read bytes from specified file into supplied buffer. 467 * 468 * IN: vp - vnode of file to be read from. 469 * uio - structure supplying read location, range info, 470 * and return buffer. 471 * ioflag - SYNC flags; used to provide FRSYNC semantics. 472 * cr - credentials of caller. 473 * 474 * OUT: uio - updated offset and range, buffer filled. 475 * 476 * RETURN: 0 if success 477 * error code if failure 478 * 479 * Side Effects: 480 * vp - atime updated if byte count > 0 481 */ 482/* ARGSUSED */ 483static int 484zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 485{ 486 znode_t *zp = VTOZ(vp); 487 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 488 objset_t *os = zfsvfs->z_os; 489 ssize_t n, nbytes; 490 int error; 491 rl_t *rl; 492 493 ZFS_ENTER(zfsvfs); 494 495 /* 496 * Validate file offset 497 */ 498 if (uio->uio_loffset < (offset_t)0) { 499 ZFS_EXIT(zfsvfs); 500 return (EINVAL); 501 } 502 503 /* 504 * Fasttrack empty reads 505 */ 506 if (uio->uio_resid == 0) { 507 ZFS_EXIT(zfsvfs); 508 return (0); 509 } 510 511 /* 512 * Check for mandatory locks 513 */ 514 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 515 if (error = chklock(vp, FREAD, 516 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 517 ZFS_EXIT(zfsvfs); 518 return (error); 519 } 520 } 521 522 /* 523 * If we're in FRSYNC mode, sync out this znode before reading it. 524 */ 525 if (ioflag & FRSYNC) 526 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 527 528 /* 529 * Lock the range against changes. 530 */ 531 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 532 533 /* 534 * If we are reading past end-of-file we can skip 535 * to the end; but we might still need to set atime. 536 */ 537 if (uio->uio_loffset >= zp->z_phys->zp_size) { 538 error = 0; 539 goto out; 540 } 541 542 ASSERT(uio->uio_loffset < zp->z_phys->zp_size); 543 n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset); 544 545 while (n > 0) { 546 nbytes = MIN(n, zfs_read_chunk_size - 547 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 548 549 if (vn_has_cached_data(vp)) 550 error = mappedread(vp, nbytes, uio); 551 else 552 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 553 if (error) 554 break; 555 556 n -= nbytes; 557 } 558 559out: 560 zfs_range_unlock(rl); 561 562 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 563 ZFS_EXIT(zfsvfs); 564 return (error); 565} 566 567/* 568 * Fault in the pages of the first n bytes specified by the uio structure. 569 * 1 byte in each page is touched and the uio struct is unmodified. 570 * Any error will exit this routine as this is only a best 571 * attempt to get the pages resident. This is a copy of ufs_trans_touch(). 572 */ 573static void 574zfs_prefault_write(ssize_t n, struct uio *uio) 575{ 576 struct iovec *iov; 577 ulong_t cnt, incr; 578 caddr_t p; 579 580 if (uio->uio_segflg != UIO_USERSPACE) 581 return; 582 583 iov = uio->uio_iov; 584 585 while (n) { 586 cnt = MIN(iov->iov_len, n); 587 if (cnt == 0) { 588 /* empty iov entry */ 589 iov++; 590 continue; 591 } 592 n -= cnt; 593 /* 594 * touch each page in this segment. 595 */ 596 p = iov->iov_base; 597 while (cnt) { 598 if (fubyte(p) == -1) 599 return; 600 incr = MIN(cnt, PAGESIZE); 601 p += incr; 602 cnt -= incr; 603 } 604 /* 605 * touch the last byte in case it straddles a page. 606 */ 607 p--; 608 if (fubyte(p) == -1) 609 return; 610 iov++; 611 } 612} 613 614/* 615 * Write the bytes to a file. 616 * 617 * IN: vp - vnode of file to be written to. 618 * uio - structure supplying write location, range info, 619 * and data buffer. 620 * ioflag - IO_APPEND flag set if in append mode. 621 * cr - credentials of caller. 622 * 623 * OUT: uio - updated offset and range. 624 * 625 * RETURN: 0 if success 626 * error code if failure 627 * 628 * Timestamps: 629 * vp - ctime|mtime updated if byte count > 0 630 */ 631/* ARGSUSED */ 632static int 633zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 634{ 635 znode_t *zp = VTOZ(vp); 636 rlim64_t limit = MAXOFFSET_T; 637 ssize_t start_resid = uio->uio_resid; 638 ssize_t tx_bytes; 639 uint64_t end_size; 640 dmu_tx_t *tx; 641 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 642 zilog_t *zilog = zfsvfs->z_log; 643 offset_t woff; 644 ssize_t n, nbytes; 645 rl_t *rl; 646 int max_blksz = zfsvfs->z_max_blksz; 647 int error; 648 649 /* 650 * Fasttrack empty write 651 */ 652 n = start_resid; 653 if (n == 0) 654 return (0); 655 656 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 657 limit = MAXOFFSET_T; 658 659 ZFS_ENTER(zfsvfs); 660 661 /* 662 * Pre-fault the pages to ensure slow (eg NFS) pages 663 * don't hold up txg. 664 */ 665 zfs_prefault_write(n, uio); 666 667 /* 668 * If in append mode, set the io offset pointer to eof. 669 */ 670 if (ioflag & IO_APPEND) { 671 /* 672 * Range lock for a file append: 673 * The value for the start of range will be determined by 674 * zfs_range_lock() (to guarantee append semantics). 675 * If this write will cause the block size to increase, 676 * zfs_range_lock() will lock the entire file, so we must 677 * later reduce the range after we grow the block size. 678 */ 679 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 680 if (rl->r_len == UINT64_MAX) { 681 /* overlocked, zp_size can't change */ 682 woff = uio->uio_loffset = zp->z_phys->zp_size; 683 } else { 684 woff = uio->uio_loffset = rl->r_off; 685 } 686 } else { 687 woff = uio->uio_loffset; 688 /* 689 * Validate file offset 690 */ 691 if (woff < 0) { 692 ZFS_EXIT(zfsvfs); 693 return (EINVAL); 694 } 695 696 /* 697 * If we need to grow the block size then zfs_range_lock() 698 * will lock a wider range than we request here. 699 * Later after growing the block size we reduce the range. 700 */ 701 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 702 } 703 704 if (woff >= limit) { 705 zfs_range_unlock(rl); 706 ZFS_EXIT(zfsvfs); 707 return (EFBIG); 708 } 709 710 if ((woff + n) > limit || woff > (limit - n)) 711 n = limit - woff; 712 713 /* 714 * Check for mandatory locks 715 */ 716 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 717 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 718 zfs_range_unlock(rl); 719 ZFS_EXIT(zfsvfs); 720 return (error); 721 } 722 end_size = MAX(zp->z_phys->zp_size, woff + n); 723 724 /* 725 * Write the file in reasonable size chunks. Each chunk is written 726 * in a separate transaction; this keeps the intent log records small 727 * and allows us to do more fine-grained space accounting. 728 */ 729 while (n > 0) { 730 /* 731 * Start a transaction. 732 */ 733 woff = uio->uio_loffset; 734 tx = dmu_tx_create(zfsvfs->z_os); 735 dmu_tx_hold_bonus(tx, zp->z_id); 736 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 737 error = dmu_tx_assign(tx, zfsvfs->z_assign); 738 if (error) { 739 if (error == ERESTART && 740 zfsvfs->z_assign == TXG_NOWAIT) { 741 dmu_tx_wait(tx); 742 dmu_tx_abort(tx); 743 continue; 744 } 745 dmu_tx_abort(tx); 746 break; 747 } 748 749 /* 750 * If zfs_range_lock() over-locked we grow the blocksize 751 * and then reduce the lock range. This will only happen 752 * on the first iteration since zfs_range_reduce() will 753 * shrink down r_len to the appropriate size. 754 */ 755 if (rl->r_len == UINT64_MAX) { 756 uint64_t new_blksz; 757 758 if (zp->z_blksz > max_blksz) { 759 ASSERT(!ISP2(zp->z_blksz)); 760 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 761 } else { 762 new_blksz = MIN(end_size, max_blksz); 763 } 764 zfs_grow_blocksize(zp, new_blksz, tx); 765 zfs_range_reduce(rl, woff, n); 766 } 767 768 /* 769 * XXX - should we really limit each write to z_max_blksz? 770 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 771 */ 772 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 773 rw_enter(&zp->z_map_lock, RW_READER); 774 775 if (woff + nbytes > zp->z_phys->zp_size) 776 vnode_pager_setsize(vp, woff + nbytes); 777 778 tx_bytes = uio->uio_resid; 779 if (vn_has_cached_data(vp)) { 780 rw_exit(&zp->z_map_lock); 781 error = mappedwrite(vp, nbytes, uio, tx); 782 } else { 783 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 784 uio, nbytes, tx); 785 rw_exit(&zp->z_map_lock); 786 } 787 tx_bytes -= uio->uio_resid; 788 789 /* 790 * If we made no progress, we're done. If we made even 791 * partial progress, update the znode and ZIL accordingly. 792 */ 793 if (tx_bytes == 0) { 794 dmu_tx_commit(tx); 795 ASSERT(error != 0); 796 break; 797 } 798 799 /* 800 * Clear Set-UID/Set-GID bits on successful write if not 801 * privileged and at least one of the excute bits is set. 802 * 803 * It would be nice to to this after all writes have 804 * been done, but that would still expose the ISUID/ISGID 805 * to another app after the partial write is committed. 806 */ 807 mutex_enter(&zp->z_acl_lock); 808 if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) | 809 (S_IXUSR >> 6))) != 0 && 810 (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 && 811 secpolicy_vnode_setid_retain(cr, 812 (zp->z_phys->zp_mode & S_ISUID) != 0 && 813 zp->z_phys->zp_uid == 0) != 0) { 814 zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID); 815 } 816 mutex_exit(&zp->z_acl_lock); 817 818 /* 819 * Update time stamp. NOTE: This marks the bonus buffer as 820 * dirty, so we don't have to do it again for zp_size. 821 */ 822 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 823 824 /* 825 * Update the file size (zp_size) if it has changed; 826 * account for possible concurrent updates. 827 */ 828 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) 829 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 830 uio->uio_loffset); 831 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 832 dmu_tx_commit(tx); 833 834 if (error != 0) 835 break; 836 ASSERT(tx_bytes == nbytes); 837 n -= nbytes; 838 } 839 840 zfs_range_unlock(rl); 841 842 /* 843 * If we're in replay mode, or we made no progress, return error. 844 * Otherwise, it's at least a partial write, so it's successful. 845 */ 846 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 847 ZFS_EXIT(zfsvfs); 848 return (error); 849 } 850 851 if (ioflag & (FSYNC | FDSYNC)) 852 zil_commit(zilog, zp->z_last_itx, zp->z_id); 853 854 ZFS_EXIT(zfsvfs); 855 return (0); 856} 857 858void 859zfs_get_done(dmu_buf_t *db, void *vzgd) 860{ 861 zgd_t *zgd = (zgd_t *)vzgd; 862 rl_t *rl = zgd->zgd_rl; 863 vnode_t *vp = ZTOV(rl->r_zp); 864 int vfslocked; 865 866 vfslocked = VFS_LOCK_GIANT(vp->v_vfsp); 867 dmu_buf_rele(db, vzgd); 868 zfs_range_unlock(rl); 869 VN_RELE(vp); 870 zil_add_vdev(zgd->zgd_zilog, DVA_GET_VDEV(BP_IDENTITY(zgd->zgd_bp))); 871 kmem_free(zgd, sizeof (zgd_t)); 872 VFS_UNLOCK_GIANT(vfslocked); 873} 874 875/* 876 * Get data to generate a TX_WRITE intent log record. 877 */ 878int 879zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 880{ 881 zfsvfs_t *zfsvfs = arg; 882 objset_t *os = zfsvfs->z_os; 883 znode_t *zp; 884 uint64_t off = lr->lr_offset; 885 dmu_buf_t *db; 886 rl_t *rl; 887 zgd_t *zgd; 888 int dlen = lr->lr_length; /* length of user data */ 889 int error = 0; 890 891 ASSERT(zio); 892 ASSERT(dlen != 0); 893 894 /* 895 * Nothing to do if the file has been removed 896 */ 897 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 898 return (ENOENT); 899 if (zp->z_unlinked) { 900 VN_RELE(ZTOV(zp)); 901 return (ENOENT); 902 } 903 904 /* 905 * Write records come in two flavors: immediate and indirect. 906 * For small writes it's cheaper to store the data with the 907 * log record (immediate); for large writes it's cheaper to 908 * sync the data and get a pointer to it (indirect) so that 909 * we don't have to write the data twice. 910 */ 911 if (buf != NULL) { /* immediate write */ 912 rl = zfs_range_lock(zp, off, dlen, RL_READER); 913 /* test for truncation needs to be done while range locked */ 914 if (off >= zp->z_phys->zp_size) { 915 error = ENOENT; 916 goto out; 917 } 918 VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf)); 919 } else { /* indirect write */ 920 uint64_t boff; /* block starting offset */ 921 922 /* 923 * Have to lock the whole block to ensure when it's 924 * written out and it's checksum is being calculated 925 * that no one can change the data. We need to re-check 926 * blocksize after we get the lock in case it's changed! 927 */ 928 for (;;) { 929 if (ISP2(zp->z_blksz)) { 930 boff = P2ALIGN_TYPED(off, zp->z_blksz, 931 uint64_t); 932 } else { 933 boff = 0; 934 } 935 dlen = zp->z_blksz; 936 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 937 if (zp->z_blksz == dlen) 938 break; 939 zfs_range_unlock(rl); 940 } 941 /* test for truncation needs to be done while range locked */ 942 if (off >= zp->z_phys->zp_size) { 943 error = ENOENT; 944 goto out; 945 } 946 zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP); 947 zgd->zgd_rl = rl; 948 zgd->zgd_zilog = zfsvfs->z_log; 949 zgd->zgd_bp = &lr->lr_blkptr; 950 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db)); 951 ASSERT(boff == db->db_offset); 952 lr->lr_blkoff = off - boff; 953 error = dmu_sync(zio, db, &lr->lr_blkptr, 954 lr->lr_common.lrc_txg, zfs_get_done, zgd); 955 ASSERT(error == EEXIST || lr->lr_length <= zp->z_blksz); 956 if (error == 0) { 957 zil_add_vdev(zfsvfs->z_log, 958 DVA_GET_VDEV(BP_IDENTITY(&lr->lr_blkptr))); 959 } 960 /* 961 * If we get EINPROGRESS, then we need to wait for a 962 * write IO initiated by dmu_sync() to complete before 963 * we can release this dbuf. We will finish everything 964 * up in the zfs_get_done() callback. 965 */ 966 if (error == EINPROGRESS) 967 return (0); 968 dmu_buf_rele(db, zgd); 969 kmem_free(zgd, sizeof (zgd_t)); 970 } 971out: 972 zfs_range_unlock(rl); 973 VN_RELE(ZTOV(zp)); 974 return (error); 975} 976 977/*ARGSUSED*/ 978static int 979zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr) 980{ 981 znode_t *zp = VTOZ(vp); 982 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 983 int error; 984 985 ZFS_ENTER(zfsvfs); 986 error = zfs_zaccess_rwx(zp, mode, cr); 987 ZFS_EXIT(zfsvfs); 988 return (error); 989} 990 991/* 992 * Lookup an entry in a directory, or an extended attribute directory. 993 * If it exists, return a held vnode reference for it. 994 * 995 * IN: dvp - vnode of directory to search. 996 * nm - name of entry to lookup. 997 * pnp - full pathname to lookup [UNUSED]. 998 * flags - LOOKUP_XATTR set if looking for an attribute. 999 * rdir - root directory vnode [UNUSED]. 1000 * cr - credentials of caller. 1001 * 1002 * OUT: vpp - vnode of located entry, NULL if not found. 1003 * 1004 * RETURN: 0 if success 1005 * error code if failure 1006 * 1007 * Timestamps: 1008 * NA 1009 */ 1010/* ARGSUSED */ 1011static int 1012zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 1013 int nameiop, cred_t *cr, kthread_t *td) 1014{ 1015 1016 znode_t *zdp = VTOZ(dvp); 1017 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1018 int error; 1019 1020 ZFS_ENTER(zfsvfs); 1021 1022 *vpp = NULL; 1023 1024#ifdef TODO 1025 if (flags & LOOKUP_XATTR) { 1026 /* 1027 * If the xattr property is off, refuse the lookup request. 1028 */ 1029 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1030 ZFS_EXIT(zfsvfs); 1031 return (EINVAL); 1032 } 1033 1034 /* 1035 * We don't allow recursive attributes.. 1036 * Maybe someday we will. 1037 */ 1038 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 1039 ZFS_EXIT(zfsvfs); 1040 return (EINVAL); 1041 } 1042 1043 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1044 ZFS_EXIT(zfsvfs); 1045 return (error); 1046 } 1047 1048 /* 1049 * Do we have permission to get into attribute directory? 1050 */ 1051 1052 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) { 1053 VN_RELE(*vpp); 1054 } 1055 1056 ZFS_EXIT(zfsvfs); 1057 return (error); 1058 } 1059#endif /* TODO */ 1060 1061 if (dvp->v_type != VDIR) { 1062 ZFS_EXIT(zfsvfs); 1063 return (ENOTDIR); 1064 } 1065 1066 /* 1067 * Check accessibility of directory. 1068 */ 1069 1070 if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) { 1071 ZFS_EXIT(zfsvfs); 1072 return (error); 1073 } 1074 1075 if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) { 1076 1077 /* 1078 * Convert device special files 1079 */ 1080 if (IS_DEVVP(*vpp)) { 1081 vnode_t *svp; 1082 1083 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1084 VN_RELE(*vpp); 1085 if (svp == NULL) 1086 error = ENOSYS; 1087 else 1088 *vpp = svp; 1089 } 1090 } 1091 1092 ZFS_EXIT(zfsvfs); 1093 1094 /* Translate errors and add SAVENAME when needed. */ 1095 if (cnp->cn_flags & ISLASTCN) { 1096 switch (nameiop) { 1097 case CREATE: 1098 case RENAME: 1099 if (error == ENOENT) { 1100 error = EJUSTRETURN; 1101 cnp->cn_flags |= SAVENAME; 1102 break; 1103 } 1104 /* FALLTHROUGH */ 1105 case DELETE: 1106 if (error == 0) 1107 cnp->cn_flags |= SAVENAME; 1108 break; 1109 } 1110 } 1111 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) { 1112 if (cnp->cn_flags & ISDOTDOT) 1113 VOP_UNLOCK(dvp, 0, td); 1114 error = vn_lock(*vpp, LK_EXCLUSIVE, td); 1115 if (cnp->cn_flags & ISDOTDOT) 1116 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td); 1117 if (error != 0) { 1118 VN_RELE(*vpp); 1119 *vpp = NULL; 1120 return (error); 1121 } 1122 } 1123 1124#ifdef FREEBSD_NAMECACHE 1125 /* 1126 * Insert name into cache (as non-existent) if appropriate. 1127 */ 1128 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE) 1129 cache_enter(dvp, *vpp, cnp); 1130 /* 1131 * Insert name into cache if appropriate. 1132 */ 1133 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1134 if (!(cnp->cn_flags & ISLASTCN) || 1135 (nameiop != DELETE && nameiop != RENAME)) { 1136 cache_enter(dvp, *vpp, cnp); 1137 } 1138 } 1139#endif 1140 1141 return (error); 1142} 1143 1144/* 1145 * Attempt to create a new entry in a directory. If the entry 1146 * already exists, truncate the file if permissible, else return 1147 * an error. Return the vp of the created or trunc'd file. 1148 * 1149 * IN: dvp - vnode of directory to put new file entry in. 1150 * name - name of new file entry. 1151 * vap - attributes of new file. 1152 * excl - flag indicating exclusive or non-exclusive mode. 1153 * mode - mode to open file with. 1154 * cr - credentials of caller. 1155 * flag - large file flag [UNUSED]. 1156 * 1157 * OUT: vpp - vnode of created or trunc'd entry. 1158 * 1159 * RETURN: 0 if success 1160 * error code if failure 1161 * 1162 * Timestamps: 1163 * dvp - ctime|mtime updated if new entry created 1164 * vp - ctime|mtime always, atime if new 1165 */ 1166/* ARGSUSED */ 1167static int 1168zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1169 vnode_t **vpp, cred_t *cr, kthread_t *td) 1170{ 1171 znode_t *zp, *dzp = VTOZ(dvp); 1172 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1173 zilog_t *zilog = zfsvfs->z_log; 1174 objset_t *os = zfsvfs->z_os; 1175 zfs_dirlock_t *dl; 1176 dmu_tx_t *tx; 1177 int error; 1178 uint64_t zoid; 1179 1180 ZFS_ENTER(zfsvfs); 1181 1182top: 1183 *vpp = NULL; 1184 1185 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1186 vap->va_mode &= ~VSVTX; 1187 1188 if (*name == '\0') { 1189 /* 1190 * Null component name refers to the directory itself. 1191 */ 1192 VN_HOLD(dvp); 1193 zp = dzp; 1194 dl = NULL; 1195 error = 0; 1196 } else { 1197 /* possible VN_HOLD(zp) */ 1198 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) { 1199 if (strcmp(name, "..") == 0) 1200 error = EISDIR; 1201 ZFS_EXIT(zfsvfs); 1202 return (error); 1203 } 1204 } 1205 1206 zoid = zp ? zp->z_id : -1ULL; 1207 1208 if (zp == NULL) { 1209 /* 1210 * Create a new file object and update the directory 1211 * to reference it. 1212 */ 1213 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 1214 goto out; 1215 } 1216 1217 /* 1218 * We only support the creation of regular files in 1219 * extended attribute directories. 1220 */ 1221 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1222 (vap->va_type != VREG)) { 1223 error = EINVAL; 1224 goto out; 1225 } 1226 1227 tx = dmu_tx_create(os); 1228 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1229 dmu_tx_hold_bonus(tx, dzp->z_id); 1230 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1231 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1232 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1233 0, SPA_MAXBLOCKSIZE); 1234 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1235 if (error) { 1236 zfs_dirent_unlock(dl); 1237 if (error == ERESTART && 1238 zfsvfs->z_assign == TXG_NOWAIT) { 1239 dmu_tx_wait(tx); 1240 dmu_tx_abort(tx); 1241 goto top; 1242 } 1243 dmu_tx_abort(tx); 1244 ZFS_EXIT(zfsvfs); 1245 return (error); 1246 } 1247 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1248 ASSERT(zp->z_id == zoid); 1249 (void) zfs_link_create(dl, zp, tx, ZNEW); 1250 zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name); 1251 dmu_tx_commit(tx); 1252 } else { 1253 /* 1254 * A directory entry already exists for this name. 1255 */ 1256 /* 1257 * Can't truncate an existing file if in exclusive mode. 1258 */ 1259 if (excl == EXCL) { 1260 error = EEXIST; 1261 goto out; 1262 } 1263 /* 1264 * Can't open a directory for writing. 1265 */ 1266 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1267 error = EISDIR; 1268 goto out; 1269 } 1270 /* 1271 * Verify requested access to file. 1272 */ 1273 if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) { 1274 goto out; 1275 } 1276 1277 mutex_enter(&dzp->z_lock); 1278 dzp->z_seq++; 1279 mutex_exit(&dzp->z_lock); 1280 1281 /* 1282 * Truncate regular files if requested. 1283 */ 1284 if ((ZTOV(zp)->v_type == VREG) && 1285 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1286 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1287 if (error == ERESTART && 1288 zfsvfs->z_assign == TXG_NOWAIT) { 1289 /* NB: we already did dmu_tx_wait() */ 1290 zfs_dirent_unlock(dl); 1291 VN_RELE(ZTOV(zp)); 1292 goto top; 1293 } 1294 } 1295 } 1296out: 1297 1298 if (error == 0) { 1299 *vpp = ZTOV(zp); 1300 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, td); 1301 } 1302 1303 if (dl) 1304 zfs_dirent_unlock(dl); 1305 1306 if (error) { 1307 if (zp) 1308 VN_RELE(ZTOV(zp)); 1309 } else { 1310 *vpp = ZTOV(zp); 1311 /* 1312 * If vnode is for a device return a specfs vnode instead. 1313 */ 1314 if (IS_DEVVP(*vpp)) { 1315 struct vnode *svp; 1316 1317 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1318 VN_RELE(*vpp); 1319 if (svp == NULL) { 1320 error = ENOSYS; 1321 } 1322 *vpp = svp; 1323 } 1324 } 1325 1326 ZFS_EXIT(zfsvfs); 1327 return (error); 1328} 1329 1330/* 1331 * Remove an entry from a directory. 1332 * 1333 * IN: dvp - vnode of directory to remove entry from. 1334 * name - name of entry to remove. 1335 * cr - credentials of caller. 1336 * 1337 * RETURN: 0 if success 1338 * error code if failure 1339 * 1340 * Timestamps: 1341 * dvp - ctime|mtime 1342 * vp - ctime (if nlink > 0) 1343 */ 1344static int 1345zfs_remove(vnode_t *dvp, char *name, cred_t *cr) 1346{ 1347 znode_t *zp, *dzp = VTOZ(dvp); 1348 znode_t *xzp = NULL; 1349 vnode_t *vp; 1350 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1351 zilog_t *zilog = zfsvfs->z_log; 1352 uint64_t acl_obj, xattr_obj; 1353 zfs_dirlock_t *dl; 1354 dmu_tx_t *tx; 1355 boolean_t may_delete_now, delete_now = FALSE; 1356 boolean_t unlinked; 1357 int error; 1358 1359 ZFS_ENTER(zfsvfs); 1360 1361top: 1362 /* 1363 * Attempt to lock directory; fail if entry doesn't exist. 1364 */ 1365 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1366 ZFS_EXIT(zfsvfs); 1367 return (error); 1368 } 1369 1370 vp = ZTOV(zp); 1371 1372 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1373 goto out; 1374 } 1375 1376 /* 1377 * Need to use rmdir for removing directories. 1378 */ 1379 if (vp->v_type == VDIR) { 1380 error = EPERM; 1381 goto out; 1382 } 1383 1384 vnevent_remove(vp); 1385 1386 dnlc_remove(dvp, name); 1387 1388 may_delete_now = FALSE; 1389 1390 /* 1391 * We may delete the znode now, or we may put it in the unlinked set; 1392 * it depends on whether we're the last link, and on whether there are 1393 * other holds on the vnode. So we dmu_tx_hold() the right things to 1394 * allow for either case. 1395 */ 1396 tx = dmu_tx_create(zfsvfs->z_os); 1397 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1398 dmu_tx_hold_bonus(tx, zp->z_id); 1399 if (may_delete_now) 1400 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 1401 1402 /* are there any extended attributes? */ 1403 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1404 /* XXX - do we need this if we are deleting? */ 1405 dmu_tx_hold_bonus(tx, xattr_obj); 1406 } 1407 1408 /* are there any additional acls */ 1409 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1410 may_delete_now) 1411 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1412 1413 /* charge as an update -- would be nice not to charge at all */ 1414 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1415 1416 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1417 if (error) { 1418 zfs_dirent_unlock(dl); 1419 VN_RELE(vp); 1420 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1421 dmu_tx_wait(tx); 1422 dmu_tx_abort(tx); 1423 goto top; 1424 } 1425 dmu_tx_abort(tx); 1426 ZFS_EXIT(zfsvfs); 1427 return (error); 1428 } 1429 1430 /* 1431 * Remove the directory entry. 1432 */ 1433 error = zfs_link_destroy(dl, zp, tx, 0, &unlinked); 1434 1435 if (error) { 1436 dmu_tx_commit(tx); 1437 goto out; 1438 } 1439 1440 if (0 && unlinked) { 1441 VI_LOCK(vp); 1442 delete_now = may_delete_now && 1443 vp->v_count == 1 && !vn_has_cached_data(vp) && 1444 zp->z_phys->zp_xattr == xattr_obj && 1445 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1446 VI_UNLOCK(vp); 1447 } 1448 1449 if (delete_now) { 1450 if (zp->z_phys->zp_xattr) { 1451 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1452 ASSERT3U(error, ==, 0); 1453 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1454 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1455 mutex_enter(&xzp->z_lock); 1456 xzp->z_unlinked = 1; 1457 xzp->z_phys->zp_links = 0; 1458 mutex_exit(&xzp->z_lock); 1459 zfs_unlinked_add(xzp, tx); 1460 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1461 } 1462 mutex_enter(&zp->z_lock); 1463 VI_LOCK(vp); 1464 vp->v_count--; 1465 ASSERT3U(vp->v_count, ==, 0); 1466 VI_UNLOCK(vp); 1467 mutex_exit(&zp->z_lock); 1468 zfs_znode_delete(zp, tx); 1469 VFS_RELE(zfsvfs->z_vfs); 1470 } else if (unlinked) { 1471 zfs_unlinked_add(zp, tx); 1472 } 1473 1474 zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name); 1475 1476 dmu_tx_commit(tx); 1477out: 1478 zfs_dirent_unlock(dl); 1479 1480 if (!delete_now) { 1481 VN_RELE(vp); 1482 } else if (xzp) { 1483 /* this rele delayed to prevent nesting transactions */ 1484 VN_RELE(ZTOV(xzp)); 1485 } 1486 1487 ZFS_EXIT(zfsvfs); 1488 return (error); 1489} 1490 1491/* 1492 * Create a new directory and insert it into dvp using the name 1493 * provided. Return a pointer to the inserted directory. 1494 * 1495 * IN: dvp - vnode of directory to add subdir to. 1496 * dirname - name of new directory. 1497 * vap - attributes of new directory. 1498 * cr - credentials of caller. 1499 * 1500 * OUT: vpp - vnode of created directory. 1501 * 1502 * RETURN: 0 if success 1503 * error code if failure 1504 * 1505 * Timestamps: 1506 * dvp - ctime|mtime updated 1507 * vp - ctime|mtime|atime updated 1508 */ 1509static int 1510zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr) 1511{ 1512 znode_t *zp, *dzp = VTOZ(dvp); 1513 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1514 zilog_t *zilog = zfsvfs->z_log; 1515 zfs_dirlock_t *dl; 1516 uint64_t zoid = 0; 1517 dmu_tx_t *tx; 1518 int error; 1519 1520 ASSERT(vap->va_type == VDIR); 1521 1522 ZFS_ENTER(zfsvfs); 1523 1524 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1525 ZFS_EXIT(zfsvfs); 1526 return (EINVAL); 1527 } 1528top: 1529 *vpp = NULL; 1530 1531 /* 1532 * First make sure the new directory doesn't exist. 1533 */ 1534 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) { 1535 ZFS_EXIT(zfsvfs); 1536 return (error); 1537 } 1538 1539 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) { 1540 zfs_dirent_unlock(dl); 1541 ZFS_EXIT(zfsvfs); 1542 return (error); 1543 } 1544 1545 /* 1546 * Add a new entry to the directory. 1547 */ 1548 tx = dmu_tx_create(zfsvfs->z_os); 1549 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1550 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1551 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1552 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1553 0, SPA_MAXBLOCKSIZE); 1554 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1555 if (error) { 1556 zfs_dirent_unlock(dl); 1557 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1558 dmu_tx_wait(tx); 1559 dmu_tx_abort(tx); 1560 goto top; 1561 } 1562 dmu_tx_abort(tx); 1563 ZFS_EXIT(zfsvfs); 1564 return (error); 1565 } 1566 1567 /* 1568 * Create new node. 1569 */ 1570 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1571 1572 /* 1573 * Now put new name in parent dir. 1574 */ 1575 (void) zfs_link_create(dl, zp, tx, ZNEW); 1576 1577 *vpp = ZTOV(zp); 1578 1579 zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname); 1580 dmu_tx_commit(tx); 1581 1582 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, curthread); 1583 1584 zfs_dirent_unlock(dl); 1585 1586 ZFS_EXIT(zfsvfs); 1587 return (0); 1588} 1589 1590/* 1591 * Remove a directory subdir entry. If the current working 1592 * directory is the same as the subdir to be removed, the 1593 * remove will fail. 1594 * 1595 * IN: dvp - vnode of directory to remove from. 1596 * name - name of directory to be removed. 1597 * cwd - vnode of current working directory. 1598 * cr - credentials of caller. 1599 * 1600 * RETURN: 0 if success 1601 * error code if failure 1602 * 1603 * Timestamps: 1604 * dvp - ctime|mtime updated 1605 */ 1606static int 1607zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr) 1608{ 1609 znode_t *dzp = VTOZ(dvp); 1610 znode_t *zp; 1611 vnode_t *vp; 1612 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1613 zilog_t *zilog = zfsvfs->z_log; 1614 zfs_dirlock_t *dl; 1615 dmu_tx_t *tx; 1616 int error; 1617 1618 ZFS_ENTER(zfsvfs); 1619 1620top: 1621 zp = NULL; 1622 1623 /* 1624 * Attempt to lock directory; fail if entry doesn't exist. 1625 */ 1626 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1627 ZFS_EXIT(zfsvfs); 1628 return (error); 1629 } 1630 1631 vp = ZTOV(zp); 1632 1633 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1634 goto out; 1635 } 1636 1637 if (vp->v_type != VDIR) { 1638 error = ENOTDIR; 1639 goto out; 1640 } 1641 1642 if (vp == cwd) { 1643 error = EINVAL; 1644 goto out; 1645 } 1646 1647 vnevent_rmdir(vp); 1648 1649 /* 1650 * Grab a lock on the directory to make sure that noone is 1651 * trying to add (or lookup) entries while we are removing it. 1652 */ 1653 rw_enter(&zp->z_name_lock, RW_WRITER); 1654 1655 /* 1656 * Grab a lock on the parent pointer to make sure we play well 1657 * with the treewalk and directory rename code. 1658 */ 1659 rw_enter(&zp->z_parent_lock, RW_WRITER); 1660 1661 tx = dmu_tx_create(zfsvfs->z_os); 1662 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1663 dmu_tx_hold_bonus(tx, zp->z_id); 1664 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1665 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1666 if (error) { 1667 rw_exit(&zp->z_parent_lock); 1668 rw_exit(&zp->z_name_lock); 1669 zfs_dirent_unlock(dl); 1670 VN_RELE(vp); 1671 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1672 dmu_tx_wait(tx); 1673 dmu_tx_abort(tx); 1674 goto top; 1675 } 1676 dmu_tx_abort(tx); 1677 ZFS_EXIT(zfsvfs); 1678 return (error); 1679 } 1680 1681#ifdef FREEBSD_NAMECACHE 1682 cache_purge(dvp); 1683#endif 1684 1685 error = zfs_link_destroy(dl, zp, tx, 0, NULL); 1686 1687 if (error == 0) 1688 zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name); 1689 1690 dmu_tx_commit(tx); 1691 1692 rw_exit(&zp->z_parent_lock); 1693 rw_exit(&zp->z_name_lock); 1694#ifdef FREEBSD_NAMECACHE 1695 cache_purge(vp); 1696#endif 1697out: 1698 zfs_dirent_unlock(dl); 1699 1700 VN_RELE(vp); 1701 1702 ZFS_EXIT(zfsvfs); 1703 return (error); 1704} 1705 1706/* 1707 * Read as many directory entries as will fit into the provided 1708 * buffer from the given directory cursor position (specified in 1709 * the uio structure. 1710 * 1711 * IN: vp - vnode of directory to read. 1712 * uio - structure supplying read location, range info, 1713 * and return buffer. 1714 * cr - credentials of caller. 1715 * 1716 * OUT: uio - updated offset and range, buffer filled. 1717 * eofp - set to true if end-of-file detected. 1718 * 1719 * RETURN: 0 if success 1720 * error code if failure 1721 * 1722 * Timestamps: 1723 * vp - atime updated 1724 * 1725 * Note that the low 4 bits of the cookie returned by zap is always zero. 1726 * This allows us to use the low range for "special" directory entries: 1727 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1728 * we use the offset 2 for the '.zfs' directory. 1729 */ 1730/* ARGSUSED */ 1731static int 1732zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 1733{ 1734 znode_t *zp = VTOZ(vp); 1735 iovec_t *iovp; 1736 dirent64_t *odp; 1737 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1738 objset_t *os; 1739 caddr_t outbuf; 1740 size_t bufsize; 1741 zap_cursor_t zc; 1742 zap_attribute_t zap; 1743 uint_t bytes_wanted; 1744 uint64_t offset; /* must be unsigned; checks for < 1 */ 1745 int local_eof; 1746 int outcount; 1747 int error; 1748 uint8_t prefetch; 1749 uint8_t type; 1750 int ncooks; 1751 u_long *cooks = NULL; 1752 1753 ZFS_ENTER(zfsvfs); 1754 1755 /* 1756 * If we are not given an eof variable, 1757 * use a local one. 1758 */ 1759 if (eofp == NULL) 1760 eofp = &local_eof; 1761 1762 /* 1763 * Check for valid iov_len. 1764 */ 1765 if (uio->uio_iov->iov_len <= 0) { 1766 ZFS_EXIT(zfsvfs); 1767 return (EINVAL); 1768 } 1769 1770 /* 1771 * Quit if directory has been removed (posix) 1772 */ 1773 if ((*eofp = zp->z_unlinked) != 0) { 1774 ZFS_EXIT(zfsvfs); 1775 return (0); 1776 } 1777 1778 error = 0; 1779 os = zfsvfs->z_os; 1780 offset = uio->uio_loffset; 1781 prefetch = zp->z_zn_prefetch; 1782 1783 /* 1784 * Initialize the iterator cursor. 1785 */ 1786 if (offset <= 3) { 1787 /* 1788 * Start iteration from the beginning of the directory. 1789 */ 1790 zap_cursor_init(&zc, os, zp->z_id); 1791 } else { 1792 /* 1793 * The offset is a serialized cursor. 1794 */ 1795 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1796 } 1797 1798 /* 1799 * Get space to change directory entries into fs independent format. 1800 */ 1801 iovp = uio->uio_iov; 1802 bytes_wanted = iovp->iov_len; 1803 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1804 bufsize = bytes_wanted; 1805 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1806 odp = (struct dirent64 *)outbuf; 1807 } else { 1808 bufsize = bytes_wanted; 1809 odp = (struct dirent64 *)iovp->iov_base; 1810 } 1811 1812 if (ncookies != NULL) { 1813 /* 1814 * Minimum entry size is dirent size and 1 byte for a file name. 1815 */ 1816 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 1817 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 1818 *cookies = cooks; 1819 *ncookies = ncooks; 1820 } 1821 1822 /* 1823 * Transform to file-system independent format 1824 */ 1825 outcount = 0; 1826 while (outcount < bytes_wanted) { 1827 ino64_t objnum; 1828 ushort_t reclen; 1829 1830 /* 1831 * Special case `.', `..', and `.zfs'. 1832 */ 1833 if (offset == 0) { 1834 (void) strcpy(zap.za_name, "."); 1835 objnum = zp->z_id; 1836 type = DT_DIR; 1837 } else if (offset == 1) { 1838 (void) strcpy(zap.za_name, ".."); 1839 objnum = zp->z_phys->zp_parent; 1840 type = DT_DIR; 1841 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1842 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1843 objnum = ZFSCTL_INO_ROOT; 1844 type = DT_DIR; 1845 } else { 1846 /* 1847 * Grab next entry. 1848 */ 1849 if (error = zap_cursor_retrieve(&zc, &zap)) { 1850 if ((*eofp = (error == ENOENT)) != 0) 1851 break; 1852 else 1853 goto update; 1854 } 1855 1856 if (zap.za_integer_length != 8 || 1857 zap.za_num_integers != 1) { 1858 cmn_err(CE_WARN, "zap_readdir: bad directory " 1859 "entry, obj = %lld, offset = %lld\n", 1860 (u_longlong_t)zp->z_id, 1861 (u_longlong_t)offset); 1862 error = ENXIO; 1863 goto update; 1864 } 1865 1866 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 1867 /* 1868 * MacOS X can extract the object type here such as: 1869 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1870 */ 1871 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1872 } 1873 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1874 1875 /* 1876 * Will this entry fit in the buffer? 1877 */ 1878 if (outcount + reclen > bufsize) { 1879 /* 1880 * Did we manage to fit anything in the buffer? 1881 */ 1882 if (!outcount) { 1883 error = EINVAL; 1884 goto update; 1885 } 1886 break; 1887 } 1888 /* 1889 * Add this entry: 1890 */ 1891 odp->d_ino = objnum; 1892 odp->d_reclen = reclen; 1893 odp->d_namlen = strlen(zap.za_name); 1894 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 1895 odp->d_type = type; 1896 outcount += reclen; 1897 odp = (dirent64_t *)((intptr_t)odp + reclen); 1898 1899 ASSERT(outcount <= bufsize); 1900 1901 /* Prefetch znode */ 1902 if (prefetch) 1903 dmu_prefetch(os, objnum, 0, 0); 1904 1905 /* 1906 * Move to the next entry, fill in the previous offset. 1907 */ 1908 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1909 zap_cursor_advance(&zc); 1910 offset = zap_cursor_serialize(&zc); 1911 } else { 1912 offset += 1; 1913 } 1914 1915 if (cooks != NULL) { 1916 *cooks++ = offset; 1917 ncooks--; 1918 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 1919 } 1920 } 1921 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1922 1923 /* Subtract unused cookies */ 1924 if (ncookies != NULL) 1925 *ncookies -= ncooks; 1926 1927 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 1928 iovp->iov_base += outcount; 1929 iovp->iov_len -= outcount; 1930 uio->uio_resid -= outcount; 1931 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 1932 /* 1933 * Reset the pointer. 1934 */ 1935 offset = uio->uio_loffset; 1936 } 1937 1938update: 1939 zap_cursor_fini(&zc); 1940 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 1941 kmem_free(outbuf, bufsize); 1942 1943 if (error == ENOENT) 1944 error = 0; 1945 1946 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1947 1948 uio->uio_loffset = offset; 1949 ZFS_EXIT(zfsvfs); 1950 if (error != 0 && cookies != NULL) { 1951 free(*cookies, M_TEMP); 1952 *cookies = NULL; 1953 *ncookies = 0; 1954 } 1955 return (error); 1956} 1957 1958static int 1959zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr) 1960{ 1961 znode_t *zp = VTOZ(vp); 1962 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1963 1964 ZFS_ENTER(zfsvfs); 1965 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 1966 ZFS_EXIT(zfsvfs); 1967 return (0); 1968} 1969 1970/* 1971 * Get the requested file attributes and place them in the provided 1972 * vattr structure. 1973 * 1974 * IN: vp - vnode of file. 1975 * vap - va_mask identifies requested attributes. 1976 * flags - [UNUSED] 1977 * cr - credentials of caller. 1978 * 1979 * OUT: vap - attribute values. 1980 * 1981 * RETURN: 0 (always succeeds) 1982 */ 1983/* ARGSUSED */ 1984static int 1985zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1986{ 1987 znode_t *zp = VTOZ(vp); 1988 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1989 znode_phys_t *pzp = zp->z_phys; 1990 uint32_t blksize; 1991 u_longlong_t nblocks; 1992 int error; 1993 1994 ZFS_ENTER(zfsvfs); 1995 1996 /* 1997 * Return all attributes. It's cheaper to provide the answer 1998 * than to determine whether we were asked the question. 1999 */ 2000 mutex_enter(&zp->z_lock); 2001 2002 vap->va_type = IFTOVT(pzp->zp_mode); 2003 vap->va_mode = pzp->zp_mode & ~S_IFMT; 2004 vap->va_uid = zp->z_phys->zp_uid; 2005 vap->va_gid = zp->z_phys->zp_gid; 2006 vap->va_nodeid = zp->z_id; 2007 vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX); /* nlink_t limit! */ 2008 vap->va_size = pzp->zp_size; 2009 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 2010 vap->va_rdev = zfs_cmpldev(pzp->zp_rdev); 2011 vap->va_seq = zp->z_seq; 2012 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2013 2014 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 2015 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 2016 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 2017 ZFS_TIME_DECODE(&vap->va_birthtime, pzp->zp_crtime); 2018 2019 /* 2020 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2021 * Also, if we are the owner don't bother, since owner should 2022 * always be allowed to read basic attributes of file. 2023 */ 2024 if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) && 2025 (zp->z_phys->zp_uid != crgetuid(cr))) { 2026 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) { 2027 mutex_exit(&zp->z_lock); 2028 ZFS_EXIT(zfsvfs); 2029 return (error); 2030 } 2031 } 2032 2033 mutex_exit(&zp->z_lock); 2034 2035 dmu_object_size_from_db(zp->z_dbuf, &blksize, &nblocks); 2036 vap->va_blksize = blksize; 2037 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2038 2039 if (zp->z_blksz == 0) { 2040 /* 2041 * Block size hasn't been set; suggest maximal I/O transfers. 2042 */ 2043 vap->va_blksize = zfsvfs->z_max_blksz; 2044 } 2045 2046 ZFS_EXIT(zfsvfs); 2047 return (0); 2048} 2049 2050/* 2051 * Set the file attributes to the values contained in the 2052 * vattr structure. 2053 * 2054 * IN: vp - vnode of file to be modified. 2055 * vap - new attribute values. 2056 * flags - ATTR_UTIME set if non-default time values provided. 2057 * cr - credentials of caller. 2058 * 2059 * RETURN: 0 if success 2060 * error code if failure 2061 * 2062 * Timestamps: 2063 * vp - ctime updated, mtime updated if size changed. 2064 */ 2065/* ARGSUSED */ 2066static int 2067zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2068 caller_context_t *ct) 2069{ 2070 struct znode *zp = VTOZ(vp); 2071 znode_phys_t *pzp = zp->z_phys; 2072 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2073 zilog_t *zilog = zfsvfs->z_log; 2074 dmu_tx_t *tx; 2075 vattr_t oldva; 2076 uint_t mask = vap->va_mask; 2077 uint_t saved_mask; 2078 int trim_mask = 0; 2079 uint64_t new_mode; 2080 znode_t *attrzp; 2081 int need_policy = FALSE; 2082 int err; 2083 2084 if (mask == 0) 2085 return (0); 2086 2087 if (mask & AT_NOSET) 2088 return (EINVAL); 2089 2090 if (mask & AT_SIZE && vp->v_type == VDIR) 2091 return (EISDIR); 2092 2093 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) 2094 return (EINVAL); 2095 2096 ZFS_ENTER(zfsvfs); 2097 2098top: 2099 attrzp = NULL; 2100 2101 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2102 ZFS_EXIT(zfsvfs); 2103 return (EROFS); 2104 } 2105 2106 /* 2107 * First validate permissions 2108 */ 2109 2110 if (mask & AT_SIZE) { 2111 err = zfs_zaccess(zp, ACE_WRITE_DATA, cr); 2112 if (err) { 2113 ZFS_EXIT(zfsvfs); 2114 return (err); 2115 } 2116 /* 2117 * XXX - Note, we are not providing any open 2118 * mode flags here (like FNDELAY), so we may 2119 * block if there are locks present... this 2120 * should be addressed in openat(). 2121 */ 2122 do { 2123 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2124 /* NB: we already did dmu_tx_wait() if necessary */ 2125 } while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 2126 if (err) { 2127 ZFS_EXIT(zfsvfs); 2128 return (err); 2129 } 2130 } 2131 2132 if (mask & (AT_ATIME|AT_MTIME)) 2133 need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr); 2134 2135 if (mask & (AT_UID|AT_GID)) { 2136 int idmask = (mask & (AT_UID|AT_GID)); 2137 int take_owner; 2138 int take_group; 2139 2140 /* 2141 * NOTE: even if a new mode is being set, 2142 * we may clear S_ISUID/S_ISGID bits. 2143 */ 2144 2145 if (!(mask & AT_MODE)) 2146 vap->va_mode = pzp->zp_mode; 2147 2148 /* 2149 * Take ownership or chgrp to group we are a member of 2150 */ 2151 2152 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2153 take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr); 2154 2155 /* 2156 * If both AT_UID and AT_GID are set then take_owner and 2157 * take_group must both be set in order to allow taking 2158 * ownership. 2159 * 2160 * Otherwise, send the check through secpolicy_vnode_setattr() 2161 * 2162 */ 2163 2164 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2165 ((idmask == AT_UID) && take_owner) || 2166 ((idmask == AT_GID) && take_group)) { 2167 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) { 2168 /* 2169 * Remove setuid/setgid for non-privileged users 2170 */ 2171 secpolicy_setid_clear(vap, cr); 2172 trim_mask = (mask & (AT_UID|AT_GID)); 2173 } else { 2174 need_policy = TRUE; 2175 } 2176 } else { 2177 need_policy = TRUE; 2178 } 2179 } 2180 2181 mutex_enter(&zp->z_lock); 2182 oldva.va_mode = pzp->zp_mode; 2183 oldva.va_uid = zp->z_phys->zp_uid; 2184 oldva.va_gid = zp->z_phys->zp_gid; 2185 mutex_exit(&zp->z_lock); 2186 2187 if (mask & AT_MODE) { 2188 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_ACL, cr) == 0) { 2189 err = secpolicy_setid_setsticky_clear(vp, vap, 2190 &oldva, cr); 2191 if (err) { 2192 ZFS_EXIT(zfsvfs); 2193 return (err); 2194 } 2195 trim_mask |= AT_MODE; 2196 } else { 2197 need_policy = TRUE; 2198 } 2199 } 2200 2201 if (need_policy) { 2202 /* 2203 * If trim_mask is set then take ownership 2204 * has been granted or write_acl is present and user 2205 * has the ability to modify mode. In that case remove 2206 * UID|GID and or MODE from mask so that 2207 * secpolicy_vnode_setattr() doesn't revoke it. 2208 */ 2209 2210 if (trim_mask) { 2211 saved_mask = vap->va_mask; 2212 vap->va_mask &= ~trim_mask; 2213 2214 } 2215 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2216 (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp); 2217 if (err) { 2218 ZFS_EXIT(zfsvfs); 2219 return (err); 2220 } 2221 2222 if (trim_mask) 2223 vap->va_mask |= saved_mask; 2224 } 2225 2226 /* 2227 * secpolicy_vnode_setattr, or take ownership may have 2228 * changed va_mask 2229 */ 2230 mask = vap->va_mask; 2231 2232 tx = dmu_tx_create(zfsvfs->z_os); 2233 dmu_tx_hold_bonus(tx, zp->z_id); 2234 2235 if (mask & AT_MODE) { 2236 uint64_t pmode = pzp->zp_mode; 2237 2238 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2239 2240 if (zp->z_phys->zp_acl.z_acl_extern_obj) 2241 dmu_tx_hold_write(tx, 2242 pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE); 2243 else 2244 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2245 0, ZFS_ACL_SIZE(MAX_ACL_SIZE)); 2246 } 2247 2248 if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) { 2249 err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp); 2250 if (err) { 2251 dmu_tx_abort(tx); 2252 ZFS_EXIT(zfsvfs); 2253 return (err); 2254 } 2255 dmu_tx_hold_bonus(tx, attrzp->z_id); 2256 } 2257 2258 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2259 if (err) { 2260 if (attrzp) 2261 VN_RELE(ZTOV(attrzp)); 2262 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2263 dmu_tx_wait(tx); 2264 dmu_tx_abort(tx); 2265 goto top; 2266 } 2267 dmu_tx_abort(tx); 2268 ZFS_EXIT(zfsvfs); 2269 return (err); 2270 } 2271 2272 dmu_buf_will_dirty(zp->z_dbuf, tx); 2273 2274 /* 2275 * Set each attribute requested. 2276 * We group settings according to the locks they need to acquire. 2277 * 2278 * Note: you cannot set ctime directly, although it will be 2279 * updated as a side-effect of calling this function. 2280 */ 2281 2282 mutex_enter(&zp->z_lock); 2283 2284 if (mask & AT_MODE) { 2285 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2286 ASSERT3U(err, ==, 0); 2287 } 2288 2289 if (attrzp) 2290 mutex_enter(&attrzp->z_lock); 2291 2292 if (mask & AT_UID) { 2293 zp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2294 if (attrzp) { 2295 attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2296 } 2297 } 2298 2299 if (mask & AT_GID) { 2300 zp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2301 if (attrzp) 2302 attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2303 } 2304 2305 if (attrzp) 2306 mutex_exit(&attrzp->z_lock); 2307 2308 if (mask & AT_ATIME) 2309 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2310 2311 if (mask & AT_MTIME) 2312 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2313 2314 if (mask & AT_SIZE) 2315 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2316 else if (mask != 0) 2317 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2318 2319 if (mask != 0) 2320 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask); 2321 2322 mutex_exit(&zp->z_lock); 2323 2324 if (attrzp) 2325 VN_RELE(ZTOV(attrzp)); 2326 2327 dmu_tx_commit(tx); 2328 2329 ZFS_EXIT(zfsvfs); 2330 return (err); 2331} 2332 2333typedef struct zfs_zlock { 2334 krwlock_t *zl_rwlock; /* lock we acquired */ 2335 znode_t *zl_znode; /* znode we held */ 2336 struct zfs_zlock *zl_next; /* next in list */ 2337} zfs_zlock_t; 2338 2339/* 2340 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2341 */ 2342static void 2343zfs_rename_unlock(zfs_zlock_t **zlpp) 2344{ 2345 zfs_zlock_t *zl; 2346 2347 while ((zl = *zlpp) != NULL) { 2348 if (zl->zl_znode != NULL) 2349 VN_RELE(ZTOV(zl->zl_znode)); 2350 rw_exit(zl->zl_rwlock); 2351 *zlpp = zl->zl_next; 2352 kmem_free(zl, sizeof (*zl)); 2353 } 2354} 2355 2356/* 2357 * Search back through the directory tree, using the ".." entries. 2358 * Lock each directory in the chain to prevent concurrent renames. 2359 * Fail any attempt to move a directory into one of its own descendants. 2360 * XXX - z_parent_lock can overlap with map or grow locks 2361 */ 2362static int 2363zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2364{ 2365 zfs_zlock_t *zl; 2366 znode_t *zp = tdzp; 2367 uint64_t rootid = zp->z_zfsvfs->z_root; 2368 uint64_t *oidp = &zp->z_id; 2369 krwlock_t *rwlp = &szp->z_parent_lock; 2370 krw_t rw = RW_WRITER; 2371 2372 /* 2373 * First pass write-locks szp and compares to zp->z_id. 2374 * Later passes read-lock zp and compare to zp->z_parent. 2375 */ 2376 do { 2377 if (!rw_tryenter(rwlp, rw)) { 2378 /* 2379 * Another thread is renaming in this path. 2380 * Note that if we are a WRITER, we don't have any 2381 * parent_locks held yet. 2382 */ 2383 if (rw == RW_READER && zp->z_id > szp->z_id) { 2384 /* 2385 * Drop our locks and restart 2386 */ 2387 zfs_rename_unlock(&zl); 2388 *zlpp = NULL; 2389 zp = tdzp; 2390 oidp = &zp->z_id; 2391 rwlp = &szp->z_parent_lock; 2392 rw = RW_WRITER; 2393 continue; 2394 } else { 2395 /* 2396 * Wait for other thread to drop its locks 2397 */ 2398 rw_enter(rwlp, rw); 2399 } 2400 } 2401 2402 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2403 zl->zl_rwlock = rwlp; 2404 zl->zl_znode = NULL; 2405 zl->zl_next = *zlpp; 2406 *zlpp = zl; 2407 2408 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2409 return (EINVAL); 2410 2411 if (*oidp == rootid) /* We've hit the top */ 2412 return (0); 2413 2414 if (rw == RW_READER) { /* i.e. not the first pass */ 2415 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2416 if (error) 2417 return (error); 2418 zl->zl_znode = zp; 2419 } 2420 oidp = &zp->z_phys->zp_parent; 2421 rwlp = &zp->z_parent_lock; 2422 rw = RW_READER; 2423 2424 } while (zp->z_id != sdzp->z_id); 2425 2426 return (0); 2427} 2428 2429/* 2430 * Move an entry from the provided source directory to the target 2431 * directory. Change the entry name as indicated. 2432 * 2433 * IN: sdvp - Source directory containing the "old entry". 2434 * snm - Old entry name. 2435 * tdvp - Target directory to contain the "new entry". 2436 * tnm - New entry name. 2437 * cr - credentials of caller. 2438 * 2439 * RETURN: 0 if success 2440 * error code if failure 2441 * 2442 * Timestamps: 2443 * sdvp,tdvp - ctime|mtime updated 2444 */ 2445static int 2446zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr) 2447{ 2448 znode_t *tdzp, *szp, *tzp; 2449 znode_t *sdzp = VTOZ(sdvp); 2450 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2451 zilog_t *zilog = zfsvfs->z_log; 2452 vnode_t *realvp; 2453 zfs_dirlock_t *sdl, *tdl; 2454 dmu_tx_t *tx; 2455 zfs_zlock_t *zl; 2456 int cmp, serr, terr, error; 2457 2458 ZFS_ENTER(zfsvfs); 2459 2460 /* 2461 * Make sure we have the real vp for the target directory. 2462 */ 2463 if (VOP_REALVP(tdvp, &realvp) == 0) 2464 tdvp = realvp; 2465 2466 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2467 ZFS_EXIT(zfsvfs); 2468 return (EXDEV); 2469 } 2470 2471 tdzp = VTOZ(tdvp); 2472top: 2473 szp = NULL; 2474 tzp = NULL; 2475 zl = NULL; 2476 2477 /* 2478 * This is to prevent the creation of links into attribute space 2479 * by renaming a linked file into/outof an attribute directory. 2480 * See the comment in zfs_link() for why this is considered bad. 2481 */ 2482 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2483 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2484 ZFS_EXIT(zfsvfs); 2485 return (EINVAL); 2486 } 2487 2488 /* 2489 * Lock source and target directory entries. To prevent deadlock, 2490 * a lock ordering must be defined. We lock the directory with 2491 * the smallest object id first, or if it's a tie, the one with 2492 * the lexically first name. 2493 */ 2494 if (sdzp->z_id < tdzp->z_id) { 2495 cmp = -1; 2496 } else if (sdzp->z_id > tdzp->z_id) { 2497 cmp = 1; 2498 } else { 2499 cmp = strcmp(snm, tnm); 2500 if (cmp == 0) { 2501 /* 2502 * POSIX: "If the old argument and the new argument 2503 * both refer to links to the same existing file, 2504 * the rename() function shall return successfully 2505 * and perform no other action." 2506 */ 2507 ZFS_EXIT(zfsvfs); 2508 return (0); 2509 } 2510 } 2511 if (cmp < 0) { 2512 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2513 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2514 } else { 2515 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2516 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2517 } 2518 2519 if (serr) { 2520 /* 2521 * Source entry invalid or not there. 2522 */ 2523 if (!terr) { 2524 zfs_dirent_unlock(tdl); 2525 if (tzp) 2526 VN_RELE(ZTOV(tzp)); 2527 } 2528 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 2529 serr = EINVAL; 2530 ZFS_EXIT(zfsvfs); 2531 return (serr); 2532 } 2533 if (terr) { 2534 zfs_dirent_unlock(sdl); 2535 VN_RELE(ZTOV(szp)); 2536 if (strcmp(tnm, "..") == 0) 2537 terr = EINVAL; 2538 ZFS_EXIT(zfsvfs); 2539 return (terr); 2540 } 2541 2542 /* 2543 * Must have write access at the source to remove the old entry 2544 * and write access at the target to create the new entry. 2545 * Note that if target and source are the same, this can be 2546 * done in a single check. 2547 */ 2548 2549 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2550 goto out; 2551 2552 if (ZTOV(szp)->v_type == VDIR) { 2553 /* 2554 * Check to make sure rename is valid. 2555 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2556 */ 2557 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 2558 goto out; 2559 } 2560 2561 /* 2562 * Does target exist? 2563 */ 2564 if (tzp) { 2565 /* 2566 * Source and target must be the same type. 2567 */ 2568 if (ZTOV(szp)->v_type == VDIR) { 2569 if (ZTOV(tzp)->v_type != VDIR) { 2570 error = ENOTDIR; 2571 goto out; 2572 } 2573 } else { 2574 if (ZTOV(tzp)->v_type == VDIR) { 2575 error = EISDIR; 2576 goto out; 2577 } 2578 } 2579 /* 2580 * POSIX dictates that when the source and target 2581 * entries refer to the same file object, rename 2582 * must do nothing and exit without error. 2583 */ 2584 if (szp->z_id == tzp->z_id) { 2585 error = 0; 2586 goto out; 2587 } 2588 } 2589 2590 vnevent_rename_src(ZTOV(szp)); 2591 if (tzp) 2592 vnevent_rename_dest(ZTOV(tzp)); 2593 2594 tx = dmu_tx_create(zfsvfs->z_os); 2595 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 2596 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 2597 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 2598 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 2599 if (sdzp != tdzp) 2600 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 2601 if (tzp) 2602 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 2603 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2604 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2605 if (error) { 2606 if (zl != NULL) 2607 zfs_rename_unlock(&zl); 2608 zfs_dirent_unlock(sdl); 2609 zfs_dirent_unlock(tdl); 2610 VN_RELE(ZTOV(szp)); 2611 if (tzp) 2612 VN_RELE(ZTOV(tzp)); 2613 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2614 dmu_tx_wait(tx); 2615 dmu_tx_abort(tx); 2616 goto top; 2617 } 2618 dmu_tx_abort(tx); 2619 ZFS_EXIT(zfsvfs); 2620 return (error); 2621 } 2622 2623 if (tzp) /* Attempt to remove the existing target */ 2624 error = zfs_link_destroy(tdl, tzp, tx, 0, NULL); 2625 2626 if (error == 0) { 2627 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 2628 if (error == 0) { 2629 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 2630 ASSERT(error == 0); 2631 zfs_log_rename(zilog, tx, TX_RENAME, sdzp, 2632 sdl->dl_name, tdzp, tdl->dl_name, szp); 2633 } 2634#ifdef FREEBSD_NAMECACHE 2635 if (error == 0) { 2636 cache_purge(sdvp); 2637 cache_purge(tdvp); 2638 } 2639#endif 2640 } 2641 2642 dmu_tx_commit(tx); 2643out: 2644 if (zl != NULL) 2645 zfs_rename_unlock(&zl); 2646 2647 zfs_dirent_unlock(sdl); 2648 zfs_dirent_unlock(tdl); 2649 2650 VN_RELE(ZTOV(szp)); 2651 if (tzp) 2652 VN_RELE(ZTOV(tzp)); 2653 2654 ZFS_EXIT(zfsvfs); 2655 2656 return (error); 2657} 2658 2659/* 2660 * Insert the indicated symbolic reference entry into the directory. 2661 * 2662 * IN: dvp - Directory to contain new symbolic link. 2663 * link - Name for new symlink entry. 2664 * vap - Attributes of new entry. 2665 * target - Target path of new symlink. 2666 * cr - credentials of caller. 2667 * 2668 * RETURN: 0 if success 2669 * error code if failure 2670 * 2671 * Timestamps: 2672 * dvp - ctime|mtime updated 2673 */ 2674static int 2675zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, cred_t *cr, kthread_t *td) 2676{ 2677 znode_t *zp, *dzp = VTOZ(dvp); 2678 zfs_dirlock_t *dl; 2679 dmu_tx_t *tx; 2680 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2681 zilog_t *zilog = zfsvfs->z_log; 2682 uint64_t zoid; 2683 int len = strlen(link); 2684 int error; 2685 2686 ASSERT(vap->va_type == VLNK); 2687 2688 ZFS_ENTER(zfsvfs); 2689top: 2690 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2691 ZFS_EXIT(zfsvfs); 2692 return (error); 2693 } 2694 2695 if (len > MAXPATHLEN) { 2696 ZFS_EXIT(zfsvfs); 2697 return (ENAMETOOLONG); 2698 } 2699 2700 /* 2701 * Attempt to lock directory; fail if entry already exists. 2702 */ 2703 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) { 2704 ZFS_EXIT(zfsvfs); 2705 return (error); 2706 } 2707 2708 tx = dmu_tx_create(zfsvfs->z_os); 2709 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 2710 dmu_tx_hold_bonus(tx, dzp->z_id); 2711 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2712 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 2713 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 2714 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2715 if (error) { 2716 zfs_dirent_unlock(dl); 2717 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2718 dmu_tx_wait(tx); 2719 dmu_tx_abort(tx); 2720 goto top; 2721 } 2722 dmu_tx_abort(tx); 2723 ZFS_EXIT(zfsvfs); 2724 return (error); 2725 } 2726 2727 dmu_buf_will_dirty(dzp->z_dbuf, tx); 2728 2729 /* 2730 * Create a new object for the symlink. 2731 * Put the link content into bonus buffer if it will fit; 2732 * otherwise, store it just like any other file data. 2733 */ 2734 zoid = 0; 2735 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 2736 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len); 2737 if (len != 0) 2738 bcopy(link, zp->z_phys + 1, len); 2739 } else { 2740 dmu_buf_t *dbp; 2741 2742 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 2743 2744 /* 2745 * Nothing can access the znode yet so no locking needed 2746 * for growing the znode's blocksize. 2747 */ 2748 zfs_grow_blocksize(zp, len, tx); 2749 2750 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp)); 2751 dmu_buf_will_dirty(dbp, tx); 2752 2753 ASSERT3U(len, <=, dbp->db_size); 2754 bcopy(link, dbp->db_data, len); 2755 dmu_buf_rele(dbp, FTAG); 2756 } 2757 zp->z_phys->zp_size = len; 2758 2759 /* 2760 * Insert the new object into the directory. 2761 */ 2762 (void) zfs_link_create(dl, zp, tx, ZNEW); 2763out: 2764 if (error == 0) { 2765 zfs_log_symlink(zilog, tx, TX_SYMLINK, dzp, zp, name, link); 2766 *vpp = ZTOV(zp); 2767 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, td); 2768 } 2769 2770 dmu_tx_commit(tx); 2771 2772 zfs_dirent_unlock(dl); 2773 2774 ZFS_EXIT(zfsvfs); 2775 return (error); 2776} 2777 2778/* 2779 * Return, in the buffer contained in the provided uio structure, 2780 * the symbolic path referred to by vp. 2781 * 2782 * IN: vp - vnode of symbolic link. 2783 * uoip - structure to contain the link path. 2784 * cr - credentials of caller. 2785 * 2786 * OUT: uio - structure to contain the link path. 2787 * 2788 * RETURN: 0 if success 2789 * error code if failure 2790 * 2791 * Timestamps: 2792 * vp - atime updated 2793 */ 2794/* ARGSUSED */ 2795static int 2796zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr) 2797{ 2798 znode_t *zp = VTOZ(vp); 2799 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2800 size_t bufsz; 2801 int error; 2802 2803 ZFS_ENTER(zfsvfs); 2804 2805 bufsz = (size_t)zp->z_phys->zp_size; 2806 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 2807 error = uiomove(zp->z_phys + 1, 2808 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2809 } else { 2810 dmu_buf_t *dbp; 2811 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 2812 if (error) { 2813 ZFS_EXIT(zfsvfs); 2814 return (error); 2815 } 2816 error = uiomove(dbp->db_data, 2817 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2818 dmu_buf_rele(dbp, FTAG); 2819 } 2820 2821 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2822 ZFS_EXIT(zfsvfs); 2823 return (error); 2824} 2825 2826/* 2827 * Insert a new entry into directory tdvp referencing svp. 2828 * 2829 * IN: tdvp - Directory to contain new entry. 2830 * svp - vnode of new entry. 2831 * name - name of new entry. 2832 * cr - credentials of caller. 2833 * 2834 * RETURN: 0 if success 2835 * error code if failure 2836 * 2837 * Timestamps: 2838 * tdvp - ctime|mtime updated 2839 * svp - ctime updated 2840 */ 2841/* ARGSUSED */ 2842static int 2843zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr) 2844{ 2845 znode_t *dzp = VTOZ(tdvp); 2846 znode_t *tzp, *szp; 2847 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2848 zilog_t *zilog = zfsvfs->z_log; 2849 zfs_dirlock_t *dl; 2850 dmu_tx_t *tx; 2851 vnode_t *realvp; 2852 int error; 2853 2854 ASSERT(tdvp->v_type == VDIR); 2855 2856 ZFS_ENTER(zfsvfs); 2857 2858 if (VOP_REALVP(svp, &realvp) == 0) 2859 svp = realvp; 2860 2861 if (svp->v_vfsp != tdvp->v_vfsp) { 2862 ZFS_EXIT(zfsvfs); 2863 return (EXDEV); 2864 } 2865 2866 szp = VTOZ(svp); 2867top: 2868 /* 2869 * We do not support links between attributes and non-attributes 2870 * because of the potential security risk of creating links 2871 * into "normal" file space in order to circumvent restrictions 2872 * imposed in attribute space. 2873 */ 2874 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 2875 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 2876 ZFS_EXIT(zfsvfs); 2877 return (EINVAL); 2878 } 2879 2880 /* 2881 * POSIX dictates that we return EPERM here. 2882 * Better choices include ENOTSUP or EISDIR. 2883 */ 2884 if (svp->v_type == VDIR) { 2885 ZFS_EXIT(zfsvfs); 2886 return (EPERM); 2887 } 2888 2889 if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) && 2890 secpolicy_basic_link(cr) != 0) { 2891 ZFS_EXIT(zfsvfs); 2892 return (EPERM); 2893 } 2894 2895 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2896 ZFS_EXIT(zfsvfs); 2897 return (error); 2898 } 2899 2900 /* 2901 * Attempt to lock directory; fail if entry already exists. 2902 */ 2903 if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) { 2904 ZFS_EXIT(zfsvfs); 2905 return (error); 2906 } 2907 2908 tx = dmu_tx_create(zfsvfs->z_os); 2909 dmu_tx_hold_bonus(tx, szp->z_id); 2910 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2911 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2912 if (error) { 2913 zfs_dirent_unlock(dl); 2914 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2915 dmu_tx_wait(tx); 2916 dmu_tx_abort(tx); 2917 goto top; 2918 } 2919 dmu_tx_abort(tx); 2920 ZFS_EXIT(zfsvfs); 2921 return (error); 2922 } 2923 2924 error = zfs_link_create(dl, szp, tx, 0); 2925 2926 if (error == 0) 2927 zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name); 2928 2929 dmu_tx_commit(tx); 2930 2931 zfs_dirent_unlock(dl); 2932 2933 ZFS_EXIT(zfsvfs); 2934 return (error); 2935} 2936 2937void 2938zfs_inactive(vnode_t *vp, cred_t *cr) 2939{ 2940 znode_t *zp = VTOZ(vp); 2941 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2942 int error; 2943 2944 rw_enter(&zfsvfs->z_um_lock, RW_READER); 2945 if (zfsvfs->z_unmounted2) { 2946 ASSERT(zp->z_dbuf_held == 0); 2947 2948 mutex_enter(&zp->z_lock); 2949 VI_LOCK(vp); 2950 vp->v_count = 0; /* count arrives as 1 */ 2951 VI_UNLOCK(vp); 2952 if (zp->z_dbuf == NULL) { 2953 mutex_exit(&zp->z_lock); 2954 zfs_znode_free(zp); 2955 } else { 2956 mutex_exit(&zp->z_lock); 2957 } 2958 rw_exit(&zfsvfs->z_um_lock); 2959 VFS_RELE(zfsvfs->z_vfs); 2960 return; 2961 } 2962 2963 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 2964 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 2965 2966 dmu_tx_hold_bonus(tx, zp->z_id); 2967 error = dmu_tx_assign(tx, TXG_WAIT); 2968 if (error) { 2969 dmu_tx_abort(tx); 2970 } else { 2971 dmu_buf_will_dirty(zp->z_dbuf, tx); 2972 mutex_enter(&zp->z_lock); 2973 zp->z_atime_dirty = 0; 2974 mutex_exit(&zp->z_lock); 2975 dmu_tx_commit(tx); 2976 } 2977 } 2978 2979 zfs_zinactive(zp); 2980 rw_exit(&zfsvfs->z_um_lock); 2981} 2982 2983CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 2984CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 2985 2986static int 2987zfs_fid(vnode_t *vp, fid_t *fidp) 2988{ 2989 znode_t *zp = VTOZ(vp); 2990 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2991 uint32_t gen = (uint32_t)zp->z_phys->zp_gen; 2992 uint64_t object = zp->z_id; 2993 zfid_short_t *zfid; 2994 int size, i; 2995 2996 ZFS_ENTER(zfsvfs); 2997 2998 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 2999 fidp->fid_len = size; 3000 3001 zfid = (zfid_short_t *)fidp; 3002 3003 zfid->zf_len = size; 3004 3005 for (i = 0; i < sizeof (zfid->zf_object); i++) 3006 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 3007 3008 /* Must have a non-zero generation number to distinguish from .zfs */ 3009 if (gen == 0) 3010 gen = 1; 3011 for (i = 0; i < sizeof (zfid->zf_gen); i++) 3012 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 3013 3014 if (size == LONG_FID_LEN) { 3015 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 3016 zfid_long_t *zlfid; 3017 3018 zlfid = (zfid_long_t *)fidp; 3019 3020 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 3021 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 3022 3023 /* XXX - this should be the generation number for the objset */ 3024 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 3025 zlfid->zf_setgen[i] = 0; 3026 } 3027 3028 ZFS_EXIT(zfsvfs); 3029 return (0); 3030} 3031 3032static int 3033zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr) 3034{ 3035 znode_t *zp, *xzp; 3036 zfsvfs_t *zfsvfs; 3037 zfs_dirlock_t *dl; 3038 int error; 3039 3040 switch (cmd) { 3041 case _PC_LINK_MAX: 3042 *valp = INT_MAX; 3043 return (0); 3044 3045 case _PC_FILESIZEBITS: 3046 *valp = 64; 3047 return (0); 3048 3049#if 0 3050 case _PC_XATTR_EXISTS: 3051 zp = VTOZ(vp); 3052 zfsvfs = zp->z_zfsvfs; 3053 ZFS_ENTER(zfsvfs); 3054 *valp = 0; 3055 error = zfs_dirent_lock(&dl, zp, "", &xzp, 3056 ZXATTR | ZEXISTS | ZSHARED); 3057 if (error == 0) { 3058 zfs_dirent_unlock(dl); 3059 if (!zfs_dirempty(xzp)) 3060 *valp = 1; 3061 VN_RELE(ZTOV(xzp)); 3062 } else if (error == ENOENT) { 3063 /* 3064 * If there aren't extended attributes, it's the 3065 * same as having zero of them. 3066 */ 3067 error = 0; 3068 } 3069 ZFS_EXIT(zfsvfs); 3070 return (error); 3071#endif 3072 3073 case _PC_ACL_EXTENDED: 3074 *valp = 0; /* TODO */ 3075 return (0); 3076 3077 case _PC_MIN_HOLE_SIZE: 3078 *valp = (int)SPA_MINBLOCKSIZE; 3079 return (0); 3080 3081 default: 3082 return (EOPNOTSUPP); 3083 } 3084} 3085 3086#ifdef TODO 3087/*ARGSUSED*/ 3088static int 3089zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3090{ 3091 znode_t *zp = VTOZ(vp); 3092 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3093 int error; 3094 3095 ZFS_ENTER(zfsvfs); 3096 error = zfs_getacl(zp, vsecp, cr); 3097 ZFS_EXIT(zfsvfs); 3098 3099 return (error); 3100} 3101#endif /* TODO */ 3102 3103#ifdef TODO 3104/*ARGSUSED*/ 3105static int 3106zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3107{ 3108 znode_t *zp = VTOZ(vp); 3109 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3110 int error; 3111 3112 ZFS_ENTER(zfsvfs); 3113 error = zfs_setacl(zp, vsecp, cr); 3114 ZFS_EXIT(zfsvfs); 3115 return (error); 3116} 3117#endif /* TODO */ 3118 3119static int 3120zfs_freebsd_open(ap) 3121 struct vop_open_args /* { 3122 struct vnode *a_vp; 3123 int a_mode; 3124 struct ucred *a_cred; 3125 struct thread *a_td; 3126 } */ *ap; 3127{ 3128 vnode_t *vp = ap->a_vp; 3129 znode_t *zp = VTOZ(vp); 3130 int error; 3131 3132 error = zfs_open(&vp, ap->a_mode, ap->a_cred); 3133 if (error == 0) 3134 vnode_create_vobject(vp, zp->z_phys->zp_size, ap->a_td); 3135 return (error); 3136} 3137 3138static int 3139zfs_freebsd_close(ap) 3140 struct vop_close_args /* { 3141 struct vnode *a_vp; 3142 int a_fflag; 3143 struct ucred *a_cred; 3144 struct thread *a_td; 3145 } */ *ap; 3146{ 3147 3148 return (zfs_close(ap->a_vp, ap->a_fflag, 0, 0, ap->a_cred)); 3149} 3150 3151static int 3152zfs_freebsd_ioctl(ap) 3153 struct vop_ioctl_args /* { 3154 struct vnode *a_vp; 3155 u_long a_command; 3156 caddr_t a_data; 3157 int a_fflag; 3158 struct ucred *cred; 3159 struct thread *td; 3160 } */ *ap; 3161{ 3162 3163 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 3164 ap->a_fflag, ap->a_cred, NULL)); 3165} 3166 3167static int 3168zfs_freebsd_read(ap) 3169 struct vop_read_args /* { 3170 struct vnode *a_vp; 3171 struct uio *a_uio; 3172 int a_ioflag; 3173 struct ucred *a_cred; 3174 } */ *ap; 3175{ 3176 3177 return (zfs_read(ap->a_vp, ap->a_uio, ap->a_ioflag, ap->a_cred, NULL)); 3178} 3179 3180static int 3181zfs_freebsd_write(ap) 3182 struct vop_write_args /* { 3183 struct vnode *a_vp; 3184 struct uio *a_uio; 3185 int a_ioflag; 3186 struct ucred *a_cred; 3187 } */ *ap; 3188{ 3189 3190 return (zfs_write(ap->a_vp, ap->a_uio, ap->a_ioflag, ap->a_cred, NULL)); 3191} 3192 3193static int 3194zfs_freebsd_access(ap) 3195 struct vop_access_args /* { 3196 struct vnode *a_vp; 3197 int a_mode; 3198 struct ucred *a_cred; 3199 struct thread *a_td; 3200 } */ *ap; 3201{ 3202 3203 return (zfs_access(ap->a_vp, ap->a_mode, 0, ap->a_cred)); 3204} 3205 3206static int 3207zfs_freebsd_lookup(ap) 3208 struct vop_lookup_args /* { 3209 struct vnode *a_dvp; 3210 struct vnode **a_vpp; 3211 struct componentname *a_cnp; 3212 } */ *ap; 3213{ 3214 struct componentname *cnp = ap->a_cnp; 3215 char nm[NAME_MAX + 1]; 3216 3217 ASSERT(cnp->cn_namelen < sizeof(nm)); 3218 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 3219 3220 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 3221 cnp->cn_cred, cnp->cn_thread)); 3222} 3223 3224static int 3225zfs_freebsd_create(ap) 3226 struct vop_create_args /* { 3227 struct vnode *a_dvp; 3228 struct vnode **a_vpp; 3229 struct componentname *a_cnp; 3230 struct vattr *a_vap; 3231 } */ *ap; 3232{ 3233 struct componentname *cnp = ap->a_cnp; 3234 vattr_t *vap = ap->a_vap; 3235 int mode; 3236 3237 ASSERT(cnp->cn_flags & SAVENAME); 3238 3239 vattr_init_mask(vap); 3240 mode = vap->va_mode & ALLPERMS; 3241 3242 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 3243 ap->a_vpp, cnp->cn_cred, cnp->cn_thread)); 3244} 3245 3246static int 3247zfs_freebsd_remove(ap) 3248 struct vop_remove_args /* { 3249 struct vnode *a_dvp; 3250 struct vnode *a_vp; 3251 struct componentname *a_cnp; 3252 } */ *ap; 3253{ 3254 3255 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 3256 3257 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 3258 ap->a_cnp->cn_cred)); 3259} 3260 3261static int 3262zfs_freebsd_mkdir(ap) 3263 struct vop_mkdir_args /* { 3264 struct vnode *a_dvp; 3265 struct vnode **a_vpp; 3266 struct componentname *a_cnp; 3267 struct vattr *a_vap; 3268 } */ *ap; 3269{ 3270 vattr_t *vap = ap->a_vap; 3271 3272 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 3273 3274 vattr_init_mask(vap); 3275 3276 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 3277 ap->a_cnp->cn_cred)); 3278} 3279 3280static int 3281zfs_freebsd_rmdir(ap) 3282 struct vop_rmdir_args /* { 3283 struct vnode *a_dvp; 3284 struct vnode *a_vp; 3285 struct componentname *a_cnp; 3286 } */ *ap; 3287{ 3288 struct componentname *cnp = ap->a_cnp; 3289 3290 ASSERT(cnp->cn_flags & SAVENAME); 3291 3292 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred)); 3293} 3294 3295static int 3296zfs_freebsd_readdir(ap) 3297 struct vop_readdir_args /* { 3298 struct vnode *a_vp; 3299 struct uio *a_uio; 3300 struct ucred *a_cred; 3301 int *a_eofflag; 3302 int *a_ncookies; 3303 u_long **a_cookies; 3304 } */ *ap; 3305{ 3306 3307 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 3308 ap->a_ncookies, ap->a_cookies)); 3309} 3310 3311static int 3312zfs_freebsd_fsync(ap) 3313 struct vop_fsync_args /* { 3314 struct vnode *a_vp; 3315 int a_waitfor; 3316 struct thread *a_td; 3317 } */ *ap; 3318{ 3319 3320 vop_stdfsync(ap); 3321 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred)); 3322} 3323 3324static int 3325zfs_freebsd_getattr(ap) 3326 struct vop_getattr_args /* { 3327 struct vnode *a_vp; 3328 struct vattr *a_vap; 3329 struct ucred *a_cred; 3330 struct thread *a_td; 3331 } */ *ap; 3332{ 3333 3334 return (zfs_getattr(ap->a_vp, ap->a_vap, 0, ap->a_cred)); 3335} 3336 3337static int 3338zfs_freebsd_setattr(ap) 3339 struct vop_setattr_args /* { 3340 struct vnode *a_vp; 3341 struct vattr *a_vap; 3342 struct ucred *a_cred; 3343 struct thread *a_td; 3344 } */ *ap; 3345{ 3346 vattr_t *vap = ap->a_vap; 3347 3348 /* No support for FreeBSD's chflags(2). */ 3349 if (vap->va_flags != VNOVAL) 3350 return (EOPNOTSUPP); 3351 3352 vattr_init_mask(vap); 3353 3354 return (zfs_setattr(ap->a_vp, vap, 0, ap->a_cred, NULL)); 3355} 3356 3357static int 3358zfs_freebsd_rename(ap) 3359 struct vop_rename_args /* { 3360 struct vnode *a_fdvp; 3361 struct vnode *a_fvp; 3362 struct componentname *a_fcnp; 3363 struct vnode *a_tdvp; 3364 struct vnode *a_tvp; 3365 struct componentname *a_tcnp; 3366 } */ *ap; 3367{ 3368 vnode_t *fdvp = ap->a_fdvp; 3369 vnode_t *fvp = ap->a_fvp; 3370 vnode_t *tdvp = ap->a_tdvp; 3371 vnode_t *tvp = ap->a_tvp; 3372 int error; 3373 3374 ASSERT(ap->a_fcnp->cn_flags & SAVENAME); 3375 ASSERT(ap->a_tcnp->cn_flags & SAVENAME); 3376 3377 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 3378 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred); 3379 3380 if (tdvp == tvp) 3381 VN_RELE(tdvp); 3382 else 3383 VN_URELE(tdvp); 3384 if (tvp) 3385 VN_URELE(tvp); 3386 VN_RELE(fdvp); 3387 VN_RELE(fvp); 3388 3389 return (error); 3390} 3391 3392static int 3393zfs_freebsd_symlink(ap) 3394 struct vop_symlink_args /* { 3395 struct vnode *a_dvp; 3396 struct vnode **a_vpp; 3397 struct componentname *a_cnp; 3398 struct vattr *a_vap; 3399 char *a_target; 3400 } */ *ap; 3401{ 3402 struct componentname *cnp = ap->a_cnp; 3403 vattr_t *vap = ap->a_vap; 3404 3405 ASSERT(cnp->cn_flags & SAVENAME); 3406 3407 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 3408 vattr_init_mask(vap); 3409 3410 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 3411 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 3412} 3413 3414static int 3415zfs_freebsd_readlink(ap) 3416 struct vop_readlink_args /* { 3417 struct vnode *a_vp; 3418 struct uio *a_uio; 3419 struct ucred *a_cred; 3420 } */ *ap; 3421{ 3422 3423 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred)); 3424} 3425 3426static int 3427zfs_freebsd_link(ap) 3428 struct vop_link_args /* { 3429 struct vnode *a_tdvp; 3430 struct vnode *a_vp; 3431 struct componentname *a_cnp; 3432 } */ *ap; 3433{ 3434 struct componentname *cnp = ap->a_cnp; 3435 3436 ASSERT(cnp->cn_flags & SAVENAME); 3437 3438 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred)); 3439} 3440 3441static int 3442zfs_freebsd_inactive(ap) 3443 struct vop_inactive_args /* { 3444 struct vnode *a_vp; 3445 struct thread *a_td; 3446 } */ *ap; 3447{ 3448 vnode_t *vp = ap->a_vp; 3449 3450 zfs_inactive(vp, ap->a_td->td_ucred); 3451 return (0); 3452} 3453 3454static int 3455zfs_freebsd_reclaim(ap) 3456 struct vop_reclaim_args /* { 3457 struct vnode *a_vp; 3458 struct thread *a_td; 3459 } */ *ap; 3460{ 3461 vnode_t *vp = ap->a_vp; 3462 znode_t *zp = VTOZ(vp); 3463 zfsvfs_t *zfsvfs; 3464 int rele = 1; 3465 3466 ASSERT(zp != NULL); 3467 3468 /* 3469 * Destroy the vm object and flush associated pages. 3470 */ 3471 vnode_destroy_vobject(vp); 3472 3473 mutex_enter(&zp->z_lock); 3474 ASSERT(zp->z_phys); 3475 ASSERT(zp->z_dbuf_held); 3476 zfsvfs = zp->z_zfsvfs; 3477 if (!zp->z_unlinked) { 3478 zp->z_dbuf_held = 0; 3479 ZTOV(zp) = NULL; 3480 mutex_exit(&zp->z_lock); 3481 dmu_buf_rele(zp->z_dbuf, NULL); 3482 } else { 3483 mutex_exit(&zp->z_lock); 3484 } 3485 VI_LOCK(vp); 3486 if (vp->v_count > 0) 3487 rele = 0; 3488 vp->v_data = NULL; 3489 ASSERT(vp->v_holdcnt > 1); 3490 vdropl(vp); 3491 if (!zp->z_unlinked && rele) 3492 VFS_RELE(zfsvfs->z_vfs); 3493 return (0); 3494} 3495 3496static int 3497zfs_freebsd_fid(ap) 3498 struct vop_fid_args /* { 3499 struct vnode *a_vp; 3500 struct fid *a_fid; 3501 } */ *ap; 3502{ 3503 3504 return (zfs_fid(ap->a_vp, (void *)ap->a_fid)); 3505} 3506 3507static int 3508zfs_freebsd_pathconf(ap) 3509 struct vop_pathconf_args /* { 3510 struct vnode *a_vp; 3511 int a_name; 3512 register_t *a_retval; 3513 } */ *ap; 3514{ 3515 ulong_t val; 3516 int error; 3517 3518 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred); 3519 if (error == 0) 3520 *ap->a_retval = val; 3521 else if (error == EOPNOTSUPP) 3522 error = vop_stdpathconf(ap); 3523 return (error); 3524} 3525 3526/* 3527 * Advisory record locking support 3528 */ 3529static int 3530zfs_freebsd_advlock(ap) 3531 struct vop_advlock_args /* { 3532 struct vnode *a_vp; 3533 caddr_t a_id; 3534 int a_op; 3535 struct flock *a_fl; 3536 int a_flags; 3537 } */ *ap; 3538{ 3539 znode_t *zp = VTOZ(ap->a_vp); 3540 3541 return (lf_advlock(ap, &(zp->z_lockf), zp->z_phys->zp_size)); 3542} 3543 3544struct vop_vector zfs_vnodeops; 3545struct vop_vector zfs_fifoops; 3546 3547struct vop_vector zfs_vnodeops = { 3548 .vop_default = &default_vnodeops, 3549 .vop_inactive = zfs_freebsd_inactive, 3550 .vop_reclaim = zfs_freebsd_reclaim, 3551 .vop_access = zfs_freebsd_access, 3552#ifdef FREEBSD_NAMECACHE 3553 .vop_lookup = vfs_cache_lookup, 3554 .vop_cachedlookup = zfs_freebsd_lookup, 3555#else 3556 .vop_lookup = zfs_freebsd_lookup, 3557#endif 3558 .vop_getattr = zfs_freebsd_getattr, 3559 .vop_setattr = zfs_freebsd_setattr, 3560 .vop_create = zfs_freebsd_create, 3561 .vop_mknod = zfs_freebsd_create, 3562 .vop_mkdir = zfs_freebsd_mkdir, 3563 .vop_readdir = zfs_freebsd_readdir, 3564 .vop_fsync = zfs_freebsd_fsync, 3565 .vop_open = zfs_freebsd_open, 3566 .vop_close = zfs_freebsd_close, 3567 .vop_rmdir = zfs_freebsd_rmdir, 3568 .vop_ioctl = zfs_freebsd_ioctl, 3569 .vop_link = zfs_freebsd_link, 3570 .vop_symlink = zfs_freebsd_symlink, 3571 .vop_readlink = zfs_freebsd_readlink, 3572 .vop_read = zfs_freebsd_read, 3573 .vop_write = zfs_freebsd_write, 3574 .vop_remove = zfs_freebsd_remove, 3575 .vop_rename = zfs_freebsd_rename, 3576 .vop_advlock = zfs_freebsd_advlock, 3577 .vop_pathconf = zfs_freebsd_pathconf, 3578 .vop_bmap = VOP_EOPNOTSUPP, 3579 .vop_fid = zfs_freebsd_fid, 3580}; 3581 3582struct vop_vector zfs_fifoops = { 3583 .vop_default = &fifo_specops, 3584 .vop_fsync = VOP_PANIC, 3585 .vop_access = zfs_freebsd_access, 3586 .vop_getattr = zfs_freebsd_getattr, 3587 .vop_inactive = zfs_freebsd_inactive, 3588 .vop_read = VOP_PANIC, 3589 .vop_reclaim = zfs_freebsd_reclaim, 3590 .vop_setattr = zfs_freebsd_setattr, 3591 .vop_write = VOP_PANIC, 3592 .vop_fid = zfs_freebsd_fid, 3593}; 3594