zfs_vnops.c revision 175202
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)); 773 774 if (woff + nbytes > zp->z_phys->zp_size) 775 vnode_pager_setsize(vp, woff + nbytes); 776 777 rw_enter(&zp->z_map_lock, RW_READER); 778 779 tx_bytes = uio->uio_resid; 780 if (vn_has_cached_data(vp)) { 781 rw_exit(&zp->z_map_lock); 782 error = mappedwrite(vp, nbytes, uio, tx); 783 } else { 784 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 785 uio, nbytes, tx); 786 rw_exit(&zp->z_map_lock); 787 } 788 tx_bytes -= uio->uio_resid; 789 790 /* 791 * If we made no progress, we're done. If we made even 792 * partial progress, update the znode and ZIL accordingly. 793 */ 794 if (tx_bytes == 0) { 795 dmu_tx_commit(tx); 796 ASSERT(error != 0); 797 break; 798 } 799 800 /* 801 * Clear Set-UID/Set-GID bits on successful write if not 802 * privileged and at least one of the excute bits is set. 803 * 804 * It would be nice to to this after all writes have 805 * been done, but that would still expose the ISUID/ISGID 806 * to another app after the partial write is committed. 807 */ 808 mutex_enter(&zp->z_acl_lock); 809 if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) | 810 (S_IXUSR >> 6))) != 0 && 811 (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 && 812 secpolicy_vnode_setid_retain(cr, 813 (zp->z_phys->zp_mode & S_ISUID) != 0 && 814 zp->z_phys->zp_uid == 0) != 0) { 815 zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID); 816 } 817 mutex_exit(&zp->z_acl_lock); 818 819 /* 820 * Update time stamp. NOTE: This marks the bonus buffer as 821 * dirty, so we don't have to do it again for zp_size. 822 */ 823 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 824 825 /* 826 * Update the file size (zp_size) if it has changed; 827 * account for possible concurrent updates. 828 */ 829 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) 830 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 831 uio->uio_loffset); 832 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 833 dmu_tx_commit(tx); 834 835 if (error != 0) 836 break; 837 ASSERT(tx_bytes == nbytes); 838 n -= nbytes; 839 } 840 841 zfs_range_unlock(rl); 842 843 /* 844 * If we're in replay mode, or we made no progress, return error. 845 * Otherwise, it's at least a partial write, so it's successful. 846 */ 847 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 848 ZFS_EXIT(zfsvfs); 849 return (error); 850 } 851 852 if (ioflag & (FSYNC | FDSYNC)) 853 zil_commit(zilog, zp->z_last_itx, zp->z_id); 854 855 ZFS_EXIT(zfsvfs); 856 return (0); 857} 858 859void 860zfs_get_done(dmu_buf_t *db, void *vzgd) 861{ 862 zgd_t *zgd = (zgd_t *)vzgd; 863 rl_t *rl = zgd->zgd_rl; 864 vnode_t *vp = ZTOV(rl->r_zp); 865 int vfslocked; 866 867 vfslocked = VFS_LOCK_GIANT(vp->v_vfsp); 868 dmu_buf_rele(db, vzgd); 869 zfs_range_unlock(rl); 870 VN_RELE(vp); 871 zil_add_vdev(zgd->zgd_zilog, DVA_GET_VDEV(BP_IDENTITY(zgd->zgd_bp))); 872 kmem_free(zgd, sizeof (zgd_t)); 873 VFS_UNLOCK_GIANT(vfslocked); 874} 875 876/* 877 * Get data to generate a TX_WRITE intent log record. 878 */ 879int 880zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 881{ 882 zfsvfs_t *zfsvfs = arg; 883 objset_t *os = zfsvfs->z_os; 884 znode_t *zp; 885 uint64_t off = lr->lr_offset; 886 dmu_buf_t *db; 887 rl_t *rl; 888 zgd_t *zgd; 889 int dlen = lr->lr_length; /* length of user data */ 890 int error = 0; 891 892 ASSERT(zio); 893 ASSERT(dlen != 0); 894 895 /* 896 * Nothing to do if the file has been removed 897 */ 898 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 899 return (ENOENT); 900 if (zp->z_unlinked) { 901 VN_RELE(ZTOV(zp)); 902 return (ENOENT); 903 } 904 905 /* 906 * Write records come in two flavors: immediate and indirect. 907 * For small writes it's cheaper to store the data with the 908 * log record (immediate); for large writes it's cheaper to 909 * sync the data and get a pointer to it (indirect) so that 910 * we don't have to write the data twice. 911 */ 912 if (buf != NULL) { /* immediate write */ 913 rl = zfs_range_lock(zp, off, dlen, RL_READER); 914 /* test for truncation needs to be done while range locked */ 915 if (off >= zp->z_phys->zp_size) { 916 error = ENOENT; 917 goto out; 918 } 919 VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf)); 920 } else { /* indirect write */ 921 uint64_t boff; /* block starting offset */ 922 923 /* 924 * Have to lock the whole block to ensure when it's 925 * written out and it's checksum is being calculated 926 * that no one can change the data. We need to re-check 927 * blocksize after we get the lock in case it's changed! 928 */ 929 for (;;) { 930 if (ISP2(zp->z_blksz)) { 931 boff = P2ALIGN_TYPED(off, zp->z_blksz, 932 uint64_t); 933 } else { 934 boff = 0; 935 } 936 dlen = zp->z_blksz; 937 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 938 if (zp->z_blksz == dlen) 939 break; 940 zfs_range_unlock(rl); 941 } 942 /* test for truncation needs to be done while range locked */ 943 if (off >= zp->z_phys->zp_size) { 944 error = ENOENT; 945 goto out; 946 } 947 zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP); 948 zgd->zgd_rl = rl; 949 zgd->zgd_zilog = zfsvfs->z_log; 950 zgd->zgd_bp = &lr->lr_blkptr; 951 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db)); 952 ASSERT(boff == db->db_offset); 953 lr->lr_blkoff = off - boff; 954 error = dmu_sync(zio, db, &lr->lr_blkptr, 955 lr->lr_common.lrc_txg, zfs_get_done, zgd); 956 ASSERT(error == EEXIST || lr->lr_length <= zp->z_blksz); 957 if (error == 0) { 958 zil_add_vdev(zfsvfs->z_log, 959 DVA_GET_VDEV(BP_IDENTITY(&lr->lr_blkptr))); 960 } 961 /* 962 * If we get EINPROGRESS, then we need to wait for a 963 * write IO initiated by dmu_sync() to complete before 964 * we can release this dbuf. We will finish everything 965 * up in the zfs_get_done() callback. 966 */ 967 if (error == EINPROGRESS) 968 return (0); 969 dmu_buf_rele(db, zgd); 970 kmem_free(zgd, sizeof (zgd_t)); 971 } 972out: 973 zfs_range_unlock(rl); 974 VN_RELE(ZTOV(zp)); 975 return (error); 976} 977 978/*ARGSUSED*/ 979static int 980zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr) 981{ 982 znode_t *zp = VTOZ(vp); 983 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 984 int error; 985 986 ZFS_ENTER(zfsvfs); 987 error = zfs_zaccess_rwx(zp, mode, cr); 988 ZFS_EXIT(zfsvfs); 989 return (error); 990} 991 992/* 993 * Lookup an entry in a directory, or an extended attribute directory. 994 * If it exists, return a held vnode reference for it. 995 * 996 * IN: dvp - vnode of directory to search. 997 * nm - name of entry to lookup. 998 * pnp - full pathname to lookup [UNUSED]. 999 * flags - LOOKUP_XATTR set if looking for an attribute. 1000 * rdir - root directory vnode [UNUSED]. 1001 * cr - credentials of caller. 1002 * 1003 * OUT: vpp - vnode of located entry, NULL if not found. 1004 * 1005 * RETURN: 0 if success 1006 * error code if failure 1007 * 1008 * Timestamps: 1009 * NA 1010 */ 1011/* ARGSUSED */ 1012static int 1013zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 1014 int nameiop, cred_t *cr, kthread_t *td) 1015{ 1016 1017 znode_t *zdp = VTOZ(dvp); 1018 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1019 int error; 1020 1021 ZFS_ENTER(zfsvfs); 1022 1023 *vpp = NULL; 1024 1025#ifdef TODO 1026 if (flags & LOOKUP_XATTR) { 1027 /* 1028 * If the xattr property is off, refuse the lookup request. 1029 */ 1030 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1031 ZFS_EXIT(zfsvfs); 1032 return (EINVAL); 1033 } 1034 1035 /* 1036 * We don't allow recursive attributes.. 1037 * Maybe someday we will. 1038 */ 1039 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 1040 ZFS_EXIT(zfsvfs); 1041 return (EINVAL); 1042 } 1043 1044 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1045 ZFS_EXIT(zfsvfs); 1046 return (error); 1047 } 1048 1049 /* 1050 * Do we have permission to get into attribute directory? 1051 */ 1052 1053 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) { 1054 VN_RELE(*vpp); 1055 } 1056 1057 ZFS_EXIT(zfsvfs); 1058 return (error); 1059 } 1060#endif /* TODO */ 1061 1062 if (dvp->v_type != VDIR) { 1063 ZFS_EXIT(zfsvfs); 1064 return (ENOTDIR); 1065 } 1066 1067 /* 1068 * Check accessibility of directory. 1069 */ 1070 1071 if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) { 1072 ZFS_EXIT(zfsvfs); 1073 return (error); 1074 } 1075 1076 if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) { 1077 1078 /* 1079 * Convert device special files 1080 */ 1081 if (IS_DEVVP(*vpp)) { 1082 vnode_t *svp; 1083 1084 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1085 VN_RELE(*vpp); 1086 if (svp == NULL) 1087 error = ENOSYS; 1088 else 1089 *vpp = svp; 1090 } 1091 } 1092 1093 ZFS_EXIT(zfsvfs); 1094 1095 /* Translate errors and add SAVENAME when needed. */ 1096 if (cnp->cn_flags & ISLASTCN) { 1097 switch (nameiop) { 1098 case CREATE: 1099 case RENAME: 1100 if (error == ENOENT) { 1101 error = EJUSTRETURN; 1102 cnp->cn_flags |= SAVENAME; 1103 break; 1104 } 1105 /* FALLTHROUGH */ 1106 case DELETE: 1107 if (error == 0) 1108 cnp->cn_flags |= SAVENAME; 1109 break; 1110 } 1111 } 1112 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) { 1113 int ltype = 0; 1114 1115 if (cnp->cn_flags & ISDOTDOT) { 1116 ltype = VOP_ISLOCKED(dvp, td); 1117 VOP_UNLOCK(dvp, 0, td); 1118 } 1119 error = vn_lock(*vpp, cnp->cn_lkflags); 1120 if (cnp->cn_flags & ISDOTDOT) 1121 vn_lock(dvp, ltype | LK_RETRY); 1122 if (error != 0) { 1123 VN_RELE(*vpp); 1124 *vpp = NULL; 1125 return (error); 1126 } 1127 } 1128 1129#ifdef FREEBSD_NAMECACHE 1130 /* 1131 * Insert name into cache (as non-existent) if appropriate. 1132 */ 1133 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE) 1134 cache_enter(dvp, *vpp, cnp); 1135 /* 1136 * Insert name into cache if appropriate. 1137 */ 1138 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1139 if (!(cnp->cn_flags & ISLASTCN) || 1140 (nameiop != DELETE && nameiop != RENAME)) { 1141 cache_enter(dvp, *vpp, cnp); 1142 } 1143 } 1144#endif 1145 1146 return (error); 1147} 1148 1149/* 1150 * Attempt to create a new entry in a directory. If the entry 1151 * already exists, truncate the file if permissible, else return 1152 * an error. Return the vp of the created or trunc'd file. 1153 * 1154 * IN: dvp - vnode of directory to put new file entry in. 1155 * name - name of new file entry. 1156 * vap - attributes of new file. 1157 * excl - flag indicating exclusive or non-exclusive mode. 1158 * mode - mode to open file with. 1159 * cr - credentials of caller. 1160 * flag - large file flag [UNUSED]. 1161 * 1162 * OUT: vpp - vnode of created or trunc'd entry. 1163 * 1164 * RETURN: 0 if success 1165 * error code if failure 1166 * 1167 * Timestamps: 1168 * dvp - ctime|mtime updated if new entry created 1169 * vp - ctime|mtime always, atime if new 1170 */ 1171/* ARGSUSED */ 1172static int 1173zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1174 vnode_t **vpp, cred_t *cr, kthread_t *td) 1175{ 1176 znode_t *zp, *dzp = VTOZ(dvp); 1177 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1178 zilog_t *zilog = zfsvfs->z_log; 1179 objset_t *os = zfsvfs->z_os; 1180 zfs_dirlock_t *dl; 1181 dmu_tx_t *tx; 1182 int error; 1183 uint64_t zoid; 1184 1185 ZFS_ENTER(zfsvfs); 1186 1187top: 1188 *vpp = NULL; 1189 1190 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1191 vap->va_mode &= ~VSVTX; 1192 1193 if (*name == '\0') { 1194 /* 1195 * Null component name refers to the directory itself. 1196 */ 1197 VN_HOLD(dvp); 1198 zp = dzp; 1199 dl = NULL; 1200 error = 0; 1201 } else { 1202 /* possible VN_HOLD(zp) */ 1203 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) { 1204 if (strcmp(name, "..") == 0) 1205 error = EISDIR; 1206 ZFS_EXIT(zfsvfs); 1207 return (error); 1208 } 1209 } 1210 1211 zoid = zp ? zp->z_id : -1ULL; 1212 1213 if (zp == NULL) { 1214 /* 1215 * Create a new file object and update the directory 1216 * to reference it. 1217 */ 1218 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 1219 goto out; 1220 } 1221 1222 /* 1223 * We only support the creation of regular files in 1224 * extended attribute directories. 1225 */ 1226 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1227 (vap->va_type != VREG)) { 1228 error = EINVAL; 1229 goto out; 1230 } 1231 1232 tx = dmu_tx_create(os); 1233 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1234 dmu_tx_hold_bonus(tx, dzp->z_id); 1235 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1236 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1237 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1238 0, SPA_MAXBLOCKSIZE); 1239 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1240 if (error) { 1241 zfs_dirent_unlock(dl); 1242 if (error == ERESTART && 1243 zfsvfs->z_assign == TXG_NOWAIT) { 1244 dmu_tx_wait(tx); 1245 dmu_tx_abort(tx); 1246 goto top; 1247 } 1248 dmu_tx_abort(tx); 1249 ZFS_EXIT(zfsvfs); 1250 return (error); 1251 } 1252 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1253 ASSERT(zp->z_id == zoid); 1254 (void) zfs_link_create(dl, zp, tx, ZNEW); 1255 zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name); 1256 dmu_tx_commit(tx); 1257 } else { 1258 /* 1259 * A directory entry already exists for this name. 1260 */ 1261 /* 1262 * Can't truncate an existing file if in exclusive mode. 1263 */ 1264 if (excl == EXCL) { 1265 error = EEXIST; 1266 goto out; 1267 } 1268 /* 1269 * Can't open a directory for writing. 1270 */ 1271 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1272 error = EISDIR; 1273 goto out; 1274 } 1275 /* 1276 * Verify requested access to file. 1277 */ 1278 if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) { 1279 goto out; 1280 } 1281 1282 mutex_enter(&dzp->z_lock); 1283 dzp->z_seq++; 1284 mutex_exit(&dzp->z_lock); 1285 1286 /* 1287 * Truncate regular files if requested. 1288 */ 1289 if ((ZTOV(zp)->v_type == VREG) && 1290 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1291 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1292 if (error == ERESTART && 1293 zfsvfs->z_assign == TXG_NOWAIT) { 1294 /* NB: we already did dmu_tx_wait() */ 1295 zfs_dirent_unlock(dl); 1296 VN_RELE(ZTOV(zp)); 1297 goto top; 1298 } 1299 } 1300 } 1301out: 1302 1303 if (error == 0) { 1304 *vpp = ZTOV(zp); 1305 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1306 } 1307 1308 if (dl) 1309 zfs_dirent_unlock(dl); 1310 1311 if (error) { 1312 if (zp) 1313 VN_RELE(ZTOV(zp)); 1314 } else { 1315 *vpp = ZTOV(zp); 1316 /* 1317 * If vnode is for a device return a specfs vnode instead. 1318 */ 1319 if (IS_DEVVP(*vpp)) { 1320 struct vnode *svp; 1321 1322 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1323 VN_RELE(*vpp); 1324 if (svp == NULL) { 1325 error = ENOSYS; 1326 } 1327 *vpp = svp; 1328 } 1329 } 1330 1331 ZFS_EXIT(zfsvfs); 1332 return (error); 1333} 1334 1335/* 1336 * Remove an entry from a directory. 1337 * 1338 * IN: dvp - vnode of directory to remove entry from. 1339 * name - name of entry to remove. 1340 * cr - credentials of caller. 1341 * 1342 * RETURN: 0 if success 1343 * error code if failure 1344 * 1345 * Timestamps: 1346 * dvp - ctime|mtime 1347 * vp - ctime (if nlink > 0) 1348 */ 1349static int 1350zfs_remove(vnode_t *dvp, char *name, cred_t *cr) 1351{ 1352 znode_t *zp, *dzp = VTOZ(dvp); 1353 znode_t *xzp = NULL; 1354 vnode_t *vp; 1355 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1356 zilog_t *zilog = zfsvfs->z_log; 1357 uint64_t acl_obj, xattr_obj; 1358 zfs_dirlock_t *dl; 1359 dmu_tx_t *tx; 1360 boolean_t may_delete_now, delete_now = FALSE; 1361 boolean_t unlinked; 1362 int error; 1363 1364 ZFS_ENTER(zfsvfs); 1365 1366top: 1367 /* 1368 * Attempt to lock directory; fail if entry doesn't exist. 1369 */ 1370 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1371 ZFS_EXIT(zfsvfs); 1372 return (error); 1373 } 1374 1375 vp = ZTOV(zp); 1376 1377 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1378 goto out; 1379 } 1380 1381 /* 1382 * Need to use rmdir for removing directories. 1383 */ 1384 if (vp->v_type == VDIR) { 1385 error = EPERM; 1386 goto out; 1387 } 1388 1389 vnevent_remove(vp); 1390 1391 dnlc_remove(dvp, name); 1392 1393 may_delete_now = FALSE; 1394 1395 /* 1396 * We may delete the znode now, or we may put it in the unlinked set; 1397 * it depends on whether we're the last link, and on whether there are 1398 * other holds on the vnode. So we dmu_tx_hold() the right things to 1399 * allow for either case. 1400 */ 1401 tx = dmu_tx_create(zfsvfs->z_os); 1402 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1403 dmu_tx_hold_bonus(tx, zp->z_id); 1404 if (may_delete_now) 1405 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 1406 1407 /* are there any extended attributes? */ 1408 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1409 /* XXX - do we need this if we are deleting? */ 1410 dmu_tx_hold_bonus(tx, xattr_obj); 1411 } 1412 1413 /* are there any additional acls */ 1414 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1415 may_delete_now) 1416 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1417 1418 /* charge as an update -- would be nice not to charge at all */ 1419 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1420 1421 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1422 if (error) { 1423 zfs_dirent_unlock(dl); 1424 VN_RELE(vp); 1425 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1426 dmu_tx_wait(tx); 1427 dmu_tx_abort(tx); 1428 goto top; 1429 } 1430 dmu_tx_abort(tx); 1431 ZFS_EXIT(zfsvfs); 1432 return (error); 1433 } 1434 1435 /* 1436 * Remove the directory entry. 1437 */ 1438 error = zfs_link_destroy(dl, zp, tx, 0, &unlinked); 1439 1440 if (error) { 1441 dmu_tx_commit(tx); 1442 goto out; 1443 } 1444 1445 if (0 && unlinked) { 1446 VI_LOCK(vp); 1447 delete_now = may_delete_now && 1448 vp->v_count == 1 && !vn_has_cached_data(vp) && 1449 zp->z_phys->zp_xattr == xattr_obj && 1450 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1451 VI_UNLOCK(vp); 1452 } 1453 1454 if (delete_now) { 1455 if (zp->z_phys->zp_xattr) { 1456 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1457 ASSERT3U(error, ==, 0); 1458 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1459 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1460 mutex_enter(&xzp->z_lock); 1461 xzp->z_unlinked = 1; 1462 xzp->z_phys->zp_links = 0; 1463 mutex_exit(&xzp->z_lock); 1464 zfs_unlinked_add(xzp, tx); 1465 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1466 } 1467 mutex_enter(&zp->z_lock); 1468 VI_LOCK(vp); 1469 vp->v_count--; 1470 ASSERT3U(vp->v_count, ==, 0); 1471 VI_UNLOCK(vp); 1472 mutex_exit(&zp->z_lock); 1473 zfs_znode_delete(zp, tx); 1474 VFS_RELE(zfsvfs->z_vfs); 1475 } else if (unlinked) { 1476 zfs_unlinked_add(zp, tx); 1477 } 1478 1479 zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name); 1480 1481 dmu_tx_commit(tx); 1482out: 1483 zfs_dirent_unlock(dl); 1484 1485 if (!delete_now) { 1486 VN_RELE(vp); 1487 } else if (xzp) { 1488 /* this rele delayed to prevent nesting transactions */ 1489 VN_RELE(ZTOV(xzp)); 1490 } 1491 1492 ZFS_EXIT(zfsvfs); 1493 return (error); 1494} 1495 1496/* 1497 * Create a new directory and insert it into dvp using the name 1498 * provided. Return a pointer to the inserted directory. 1499 * 1500 * IN: dvp - vnode of directory to add subdir to. 1501 * dirname - name of new directory. 1502 * vap - attributes of new directory. 1503 * cr - credentials of caller. 1504 * 1505 * OUT: vpp - vnode of created directory. 1506 * 1507 * RETURN: 0 if success 1508 * error code if failure 1509 * 1510 * Timestamps: 1511 * dvp - ctime|mtime updated 1512 * vp - ctime|mtime|atime updated 1513 */ 1514static int 1515zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr) 1516{ 1517 znode_t *zp, *dzp = VTOZ(dvp); 1518 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1519 zilog_t *zilog = zfsvfs->z_log; 1520 zfs_dirlock_t *dl; 1521 uint64_t zoid = 0; 1522 dmu_tx_t *tx; 1523 int error; 1524 1525 ASSERT(vap->va_type == VDIR); 1526 1527 ZFS_ENTER(zfsvfs); 1528 1529 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1530 ZFS_EXIT(zfsvfs); 1531 return (EINVAL); 1532 } 1533top: 1534 *vpp = NULL; 1535 1536 /* 1537 * First make sure the new directory doesn't exist. 1538 */ 1539 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) { 1540 ZFS_EXIT(zfsvfs); 1541 return (error); 1542 } 1543 1544 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) { 1545 zfs_dirent_unlock(dl); 1546 ZFS_EXIT(zfsvfs); 1547 return (error); 1548 } 1549 1550 /* 1551 * Add a new entry to the directory. 1552 */ 1553 tx = dmu_tx_create(zfsvfs->z_os); 1554 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1555 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1556 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1557 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1558 0, SPA_MAXBLOCKSIZE); 1559 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1560 if (error) { 1561 zfs_dirent_unlock(dl); 1562 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1563 dmu_tx_wait(tx); 1564 dmu_tx_abort(tx); 1565 goto top; 1566 } 1567 dmu_tx_abort(tx); 1568 ZFS_EXIT(zfsvfs); 1569 return (error); 1570 } 1571 1572 /* 1573 * Create new node. 1574 */ 1575 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1576 1577 /* 1578 * Now put new name in parent dir. 1579 */ 1580 (void) zfs_link_create(dl, zp, tx, ZNEW); 1581 1582 *vpp = ZTOV(zp); 1583 1584 zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname); 1585 dmu_tx_commit(tx); 1586 1587 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1588 1589 zfs_dirent_unlock(dl); 1590 1591 ZFS_EXIT(zfsvfs); 1592 return (0); 1593} 1594 1595/* 1596 * Remove a directory subdir entry. If the current working 1597 * directory is the same as the subdir to be removed, the 1598 * remove will fail. 1599 * 1600 * IN: dvp - vnode of directory to remove from. 1601 * name - name of directory to be removed. 1602 * cwd - vnode of current working directory. 1603 * cr - credentials of caller. 1604 * 1605 * RETURN: 0 if success 1606 * error code if failure 1607 * 1608 * Timestamps: 1609 * dvp - ctime|mtime updated 1610 */ 1611static int 1612zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr) 1613{ 1614 znode_t *dzp = VTOZ(dvp); 1615 znode_t *zp; 1616 vnode_t *vp; 1617 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1618 zilog_t *zilog = zfsvfs->z_log; 1619 zfs_dirlock_t *dl; 1620 dmu_tx_t *tx; 1621 int error; 1622 1623 ZFS_ENTER(zfsvfs); 1624 1625top: 1626 zp = NULL; 1627 1628 /* 1629 * Attempt to lock directory; fail if entry doesn't exist. 1630 */ 1631 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1632 ZFS_EXIT(zfsvfs); 1633 return (error); 1634 } 1635 1636 vp = ZTOV(zp); 1637 1638 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1639 goto out; 1640 } 1641 1642 if (vp->v_type != VDIR) { 1643 error = ENOTDIR; 1644 goto out; 1645 } 1646 1647 if (vp == cwd) { 1648 error = EINVAL; 1649 goto out; 1650 } 1651 1652 vnevent_rmdir(vp); 1653 1654 /* 1655 * Grab a lock on the directory to make sure that noone is 1656 * trying to add (or lookup) entries while we are removing it. 1657 */ 1658 rw_enter(&zp->z_name_lock, RW_WRITER); 1659 1660 /* 1661 * Grab a lock on the parent pointer to make sure we play well 1662 * with the treewalk and directory rename code. 1663 */ 1664 rw_enter(&zp->z_parent_lock, RW_WRITER); 1665 1666 tx = dmu_tx_create(zfsvfs->z_os); 1667 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1668 dmu_tx_hold_bonus(tx, zp->z_id); 1669 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1670 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1671 if (error) { 1672 rw_exit(&zp->z_parent_lock); 1673 rw_exit(&zp->z_name_lock); 1674 zfs_dirent_unlock(dl); 1675 VN_RELE(vp); 1676 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1677 dmu_tx_wait(tx); 1678 dmu_tx_abort(tx); 1679 goto top; 1680 } 1681 dmu_tx_abort(tx); 1682 ZFS_EXIT(zfsvfs); 1683 return (error); 1684 } 1685 1686#ifdef FREEBSD_NAMECACHE 1687 cache_purge(dvp); 1688#endif 1689 1690 error = zfs_link_destroy(dl, zp, tx, 0, NULL); 1691 1692 if (error == 0) 1693 zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name); 1694 1695 dmu_tx_commit(tx); 1696 1697 rw_exit(&zp->z_parent_lock); 1698 rw_exit(&zp->z_name_lock); 1699#ifdef FREEBSD_NAMECACHE 1700 cache_purge(vp); 1701#endif 1702out: 1703 zfs_dirent_unlock(dl); 1704 1705 VN_RELE(vp); 1706 1707 ZFS_EXIT(zfsvfs); 1708 return (error); 1709} 1710 1711/* 1712 * Read as many directory entries as will fit into the provided 1713 * buffer from the given directory cursor position (specified in 1714 * the uio structure. 1715 * 1716 * IN: vp - vnode of directory to read. 1717 * uio - structure supplying read location, range info, 1718 * and return buffer. 1719 * cr - credentials of caller. 1720 * 1721 * OUT: uio - updated offset and range, buffer filled. 1722 * eofp - set to true if end-of-file detected. 1723 * 1724 * RETURN: 0 if success 1725 * error code if failure 1726 * 1727 * Timestamps: 1728 * vp - atime updated 1729 * 1730 * Note that the low 4 bits of the cookie returned by zap is always zero. 1731 * This allows us to use the low range for "special" directory entries: 1732 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1733 * we use the offset 2 for the '.zfs' directory. 1734 */ 1735/* ARGSUSED */ 1736static int 1737zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 1738{ 1739 znode_t *zp = VTOZ(vp); 1740 iovec_t *iovp; 1741 dirent64_t *odp; 1742 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1743 objset_t *os; 1744 caddr_t outbuf; 1745 size_t bufsize; 1746 zap_cursor_t zc; 1747 zap_attribute_t zap; 1748 uint_t bytes_wanted; 1749 uint64_t offset; /* must be unsigned; checks for < 1 */ 1750 int local_eof; 1751 int outcount; 1752 int error; 1753 uint8_t prefetch; 1754 uint8_t type; 1755 int ncooks; 1756 u_long *cooks = NULL; 1757 1758 ZFS_ENTER(zfsvfs); 1759 1760 /* 1761 * If we are not given an eof variable, 1762 * use a local one. 1763 */ 1764 if (eofp == NULL) 1765 eofp = &local_eof; 1766 1767 /* 1768 * Check for valid iov_len. 1769 */ 1770 if (uio->uio_iov->iov_len <= 0) { 1771 ZFS_EXIT(zfsvfs); 1772 return (EINVAL); 1773 } 1774 1775 /* 1776 * Quit if directory has been removed (posix) 1777 */ 1778 if ((*eofp = zp->z_unlinked) != 0) { 1779 ZFS_EXIT(zfsvfs); 1780 return (0); 1781 } 1782 1783 error = 0; 1784 os = zfsvfs->z_os; 1785 offset = uio->uio_loffset; 1786 prefetch = zp->z_zn_prefetch; 1787 1788 /* 1789 * Initialize the iterator cursor. 1790 */ 1791 if (offset <= 3) { 1792 /* 1793 * Start iteration from the beginning of the directory. 1794 */ 1795 zap_cursor_init(&zc, os, zp->z_id); 1796 } else { 1797 /* 1798 * The offset is a serialized cursor. 1799 */ 1800 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1801 } 1802 1803 /* 1804 * Get space to change directory entries into fs independent format. 1805 */ 1806 iovp = uio->uio_iov; 1807 bytes_wanted = iovp->iov_len; 1808 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1809 bufsize = bytes_wanted; 1810 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1811 odp = (struct dirent64 *)outbuf; 1812 } else { 1813 bufsize = bytes_wanted; 1814 odp = (struct dirent64 *)iovp->iov_base; 1815 } 1816 1817 if (ncookies != NULL) { 1818 /* 1819 * Minimum entry size is dirent size and 1 byte for a file name. 1820 */ 1821 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 1822 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 1823 *cookies = cooks; 1824 *ncookies = ncooks; 1825 } 1826 1827 /* 1828 * Transform to file-system independent format 1829 */ 1830 outcount = 0; 1831 while (outcount < bytes_wanted) { 1832 ino64_t objnum; 1833 ushort_t reclen; 1834 1835 /* 1836 * Special case `.', `..', and `.zfs'. 1837 */ 1838 if (offset == 0) { 1839 (void) strcpy(zap.za_name, "."); 1840 objnum = zp->z_id; 1841 type = DT_DIR; 1842 } else if (offset == 1) { 1843 (void) strcpy(zap.za_name, ".."); 1844 objnum = zp->z_phys->zp_parent; 1845 type = DT_DIR; 1846 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1847 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1848 objnum = ZFSCTL_INO_ROOT; 1849 type = DT_DIR; 1850 } else { 1851 /* 1852 * Grab next entry. 1853 */ 1854 if (error = zap_cursor_retrieve(&zc, &zap)) { 1855 if ((*eofp = (error == ENOENT)) != 0) 1856 break; 1857 else 1858 goto update; 1859 } 1860 1861 if (zap.za_integer_length != 8 || 1862 zap.za_num_integers != 1) { 1863 cmn_err(CE_WARN, "zap_readdir: bad directory " 1864 "entry, obj = %lld, offset = %lld\n", 1865 (u_longlong_t)zp->z_id, 1866 (u_longlong_t)offset); 1867 error = ENXIO; 1868 goto update; 1869 } 1870 1871 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 1872 /* 1873 * MacOS X can extract the object type here such as: 1874 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1875 */ 1876 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1877 } 1878 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1879 1880 /* 1881 * Will this entry fit in the buffer? 1882 */ 1883 if (outcount + reclen > bufsize) { 1884 /* 1885 * Did we manage to fit anything in the buffer? 1886 */ 1887 if (!outcount) { 1888 error = EINVAL; 1889 goto update; 1890 } 1891 break; 1892 } 1893 /* 1894 * Add this entry: 1895 */ 1896 odp->d_ino = objnum; 1897 odp->d_reclen = reclen; 1898 odp->d_namlen = strlen(zap.za_name); 1899 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 1900 odp->d_type = type; 1901 outcount += reclen; 1902 odp = (dirent64_t *)((intptr_t)odp + reclen); 1903 1904 ASSERT(outcount <= bufsize); 1905 1906 /* Prefetch znode */ 1907 if (prefetch) 1908 dmu_prefetch(os, objnum, 0, 0); 1909 1910 /* 1911 * Move to the next entry, fill in the previous offset. 1912 */ 1913 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1914 zap_cursor_advance(&zc); 1915 offset = zap_cursor_serialize(&zc); 1916 } else { 1917 offset += 1; 1918 } 1919 1920 if (cooks != NULL) { 1921 *cooks++ = offset; 1922 ncooks--; 1923 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 1924 } 1925 } 1926 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1927 1928 /* Subtract unused cookies */ 1929 if (ncookies != NULL) 1930 *ncookies -= ncooks; 1931 1932 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 1933 iovp->iov_base += outcount; 1934 iovp->iov_len -= outcount; 1935 uio->uio_resid -= outcount; 1936 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 1937 /* 1938 * Reset the pointer. 1939 */ 1940 offset = uio->uio_loffset; 1941 } 1942 1943update: 1944 zap_cursor_fini(&zc); 1945 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 1946 kmem_free(outbuf, bufsize); 1947 1948 if (error == ENOENT) 1949 error = 0; 1950 1951 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1952 1953 uio->uio_loffset = offset; 1954 ZFS_EXIT(zfsvfs); 1955 if (error != 0 && cookies != NULL) { 1956 free(*cookies, M_TEMP); 1957 *cookies = NULL; 1958 *ncookies = 0; 1959 } 1960 return (error); 1961} 1962 1963static int 1964zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr) 1965{ 1966 znode_t *zp = VTOZ(vp); 1967 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1968 1969 ZFS_ENTER(zfsvfs); 1970 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 1971 ZFS_EXIT(zfsvfs); 1972 return (0); 1973} 1974 1975/* 1976 * Get the requested file attributes and place them in the provided 1977 * vattr structure. 1978 * 1979 * IN: vp - vnode of file. 1980 * vap - va_mask identifies requested attributes. 1981 * flags - [UNUSED] 1982 * cr - credentials of caller. 1983 * 1984 * OUT: vap - attribute values. 1985 * 1986 * RETURN: 0 (always succeeds) 1987 */ 1988/* ARGSUSED */ 1989static int 1990zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1991{ 1992 znode_t *zp = VTOZ(vp); 1993 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1994 znode_phys_t *pzp = zp->z_phys; 1995 uint32_t blksize; 1996 u_longlong_t nblocks; 1997 int error; 1998 1999 ZFS_ENTER(zfsvfs); 2000 2001 /* 2002 * Return all attributes. It's cheaper to provide the answer 2003 * than to determine whether we were asked the question. 2004 */ 2005 mutex_enter(&zp->z_lock); 2006 2007 vap->va_type = IFTOVT(pzp->zp_mode); 2008 vap->va_mode = pzp->zp_mode & ~S_IFMT; 2009 vap->va_uid = zp->z_phys->zp_uid; 2010 vap->va_gid = zp->z_phys->zp_gid; 2011 vap->va_nodeid = zp->z_id; 2012 vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX); /* nlink_t limit! */ 2013 vap->va_size = pzp->zp_size; 2014 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 2015 vap->va_rdev = zfs_cmpldev(pzp->zp_rdev); 2016 vap->va_seq = zp->z_seq; 2017 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2018 2019 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 2020 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 2021 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 2022 ZFS_TIME_DECODE(&vap->va_birthtime, pzp->zp_crtime); 2023 2024 /* 2025 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2026 * Also, if we are the owner don't bother, since owner should 2027 * always be allowed to read basic attributes of file. 2028 */ 2029 if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) && 2030 (zp->z_phys->zp_uid != crgetuid(cr))) { 2031 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) { 2032 mutex_exit(&zp->z_lock); 2033 ZFS_EXIT(zfsvfs); 2034 return (error); 2035 } 2036 } 2037 2038 mutex_exit(&zp->z_lock); 2039 2040 dmu_object_size_from_db(zp->z_dbuf, &blksize, &nblocks); 2041 vap->va_blksize = blksize; 2042 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2043 2044 if (zp->z_blksz == 0) { 2045 /* 2046 * Block size hasn't been set; suggest maximal I/O transfers. 2047 */ 2048 vap->va_blksize = zfsvfs->z_max_blksz; 2049 } 2050 2051 ZFS_EXIT(zfsvfs); 2052 return (0); 2053} 2054 2055/* 2056 * Set the file attributes to the values contained in the 2057 * vattr structure. 2058 * 2059 * IN: vp - vnode of file to be modified. 2060 * vap - new attribute values. 2061 * flags - ATTR_UTIME set if non-default time values provided. 2062 * cr - credentials of caller. 2063 * 2064 * RETURN: 0 if success 2065 * error code if failure 2066 * 2067 * Timestamps: 2068 * vp - ctime updated, mtime updated if size changed. 2069 */ 2070/* ARGSUSED */ 2071static int 2072zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2073 caller_context_t *ct) 2074{ 2075 struct znode *zp = VTOZ(vp); 2076 znode_phys_t *pzp = zp->z_phys; 2077 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2078 zilog_t *zilog = zfsvfs->z_log; 2079 dmu_tx_t *tx; 2080 vattr_t oldva; 2081 uint_t mask = vap->va_mask; 2082 uint_t saved_mask; 2083 int trim_mask = 0; 2084 uint64_t new_mode; 2085 znode_t *attrzp; 2086 int need_policy = FALSE; 2087 int err; 2088 2089 if (mask == 0) 2090 return (0); 2091 2092 if (mask & AT_NOSET) 2093 return (EINVAL); 2094 2095 if (mask & AT_SIZE && vp->v_type == VDIR) 2096 return (EISDIR); 2097 2098 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) 2099 return (EINVAL); 2100 2101 ZFS_ENTER(zfsvfs); 2102 2103top: 2104 attrzp = NULL; 2105 2106 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2107 ZFS_EXIT(zfsvfs); 2108 return (EROFS); 2109 } 2110 2111 /* 2112 * First validate permissions 2113 */ 2114 2115 if (mask & AT_SIZE) { 2116 err = zfs_zaccess(zp, ACE_WRITE_DATA, cr); 2117 if (err) { 2118 ZFS_EXIT(zfsvfs); 2119 return (err); 2120 } 2121 /* 2122 * XXX - Note, we are not providing any open 2123 * mode flags here (like FNDELAY), so we may 2124 * block if there are locks present... this 2125 * should be addressed in openat(). 2126 */ 2127 do { 2128 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2129 /* NB: we already did dmu_tx_wait() if necessary */ 2130 } while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 2131 if (err) { 2132 ZFS_EXIT(zfsvfs); 2133 return (err); 2134 } 2135 } 2136 2137 if (mask & (AT_ATIME|AT_MTIME)) 2138 need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr); 2139 2140 if (mask & (AT_UID|AT_GID)) { 2141 int idmask = (mask & (AT_UID|AT_GID)); 2142 int take_owner; 2143 int take_group; 2144 2145 /* 2146 * NOTE: even if a new mode is being set, 2147 * we may clear S_ISUID/S_ISGID bits. 2148 */ 2149 2150 if (!(mask & AT_MODE)) 2151 vap->va_mode = pzp->zp_mode; 2152 2153 /* 2154 * Take ownership or chgrp to group we are a member of 2155 */ 2156 2157 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2158 take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr); 2159 2160 /* 2161 * If both AT_UID and AT_GID are set then take_owner and 2162 * take_group must both be set in order to allow taking 2163 * ownership. 2164 * 2165 * Otherwise, send the check through secpolicy_vnode_setattr() 2166 * 2167 */ 2168 2169 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2170 ((idmask == AT_UID) && take_owner) || 2171 ((idmask == AT_GID) && take_group)) { 2172 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) { 2173 /* 2174 * Remove setuid/setgid for non-privileged users 2175 */ 2176 secpolicy_setid_clear(vap, cr); 2177 trim_mask = (mask & (AT_UID|AT_GID)); 2178 } else { 2179 need_policy = TRUE; 2180 } 2181 } else { 2182 need_policy = TRUE; 2183 } 2184 } 2185 2186 mutex_enter(&zp->z_lock); 2187 oldva.va_mode = pzp->zp_mode; 2188 oldva.va_uid = zp->z_phys->zp_uid; 2189 oldva.va_gid = zp->z_phys->zp_gid; 2190 mutex_exit(&zp->z_lock); 2191 2192 if (mask & AT_MODE) { 2193 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_ACL, cr) == 0) { 2194 err = secpolicy_setid_setsticky_clear(vp, vap, 2195 &oldva, cr); 2196 if (err) { 2197 ZFS_EXIT(zfsvfs); 2198 return (err); 2199 } 2200 trim_mask |= AT_MODE; 2201 } else { 2202 need_policy = TRUE; 2203 } 2204 } 2205 2206 if (need_policy) { 2207 /* 2208 * If trim_mask is set then take ownership 2209 * has been granted or write_acl is present and user 2210 * has the ability to modify mode. In that case remove 2211 * UID|GID and or MODE from mask so that 2212 * secpolicy_vnode_setattr() doesn't revoke it. 2213 */ 2214 2215 if (trim_mask) { 2216 saved_mask = vap->va_mask; 2217 vap->va_mask &= ~trim_mask; 2218 2219 } 2220 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2221 (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp); 2222 if (err) { 2223 ZFS_EXIT(zfsvfs); 2224 return (err); 2225 } 2226 2227 if (trim_mask) 2228 vap->va_mask |= saved_mask; 2229 } 2230 2231 /* 2232 * secpolicy_vnode_setattr, or take ownership may have 2233 * changed va_mask 2234 */ 2235 mask = vap->va_mask; 2236 2237 tx = dmu_tx_create(zfsvfs->z_os); 2238 dmu_tx_hold_bonus(tx, zp->z_id); 2239 2240 if (mask & AT_MODE) { 2241 uint64_t pmode = pzp->zp_mode; 2242 2243 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2244 2245 if (zp->z_phys->zp_acl.z_acl_extern_obj) 2246 dmu_tx_hold_write(tx, 2247 pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE); 2248 else 2249 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2250 0, ZFS_ACL_SIZE(MAX_ACL_SIZE)); 2251 } 2252 2253 if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) { 2254 err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp); 2255 if (err) { 2256 dmu_tx_abort(tx); 2257 ZFS_EXIT(zfsvfs); 2258 return (err); 2259 } 2260 dmu_tx_hold_bonus(tx, attrzp->z_id); 2261 } 2262 2263 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2264 if (err) { 2265 if (attrzp) 2266 VN_RELE(ZTOV(attrzp)); 2267 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2268 dmu_tx_wait(tx); 2269 dmu_tx_abort(tx); 2270 goto top; 2271 } 2272 dmu_tx_abort(tx); 2273 ZFS_EXIT(zfsvfs); 2274 return (err); 2275 } 2276 2277 dmu_buf_will_dirty(zp->z_dbuf, tx); 2278 2279 /* 2280 * Set each attribute requested. 2281 * We group settings according to the locks they need to acquire. 2282 * 2283 * Note: you cannot set ctime directly, although it will be 2284 * updated as a side-effect of calling this function. 2285 */ 2286 2287 mutex_enter(&zp->z_lock); 2288 2289 if (mask & AT_MODE) { 2290 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2291 ASSERT3U(err, ==, 0); 2292 } 2293 2294 if (attrzp) 2295 mutex_enter(&attrzp->z_lock); 2296 2297 if (mask & AT_UID) { 2298 zp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2299 if (attrzp) { 2300 attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2301 } 2302 } 2303 2304 if (mask & AT_GID) { 2305 zp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2306 if (attrzp) 2307 attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2308 } 2309 2310 if (attrzp) 2311 mutex_exit(&attrzp->z_lock); 2312 2313 if (mask & AT_ATIME) 2314 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2315 2316 if (mask & AT_MTIME) 2317 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2318 2319 if (mask & AT_SIZE) 2320 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2321 else if (mask != 0) 2322 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2323 2324 if (mask != 0) 2325 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask); 2326 2327 mutex_exit(&zp->z_lock); 2328 2329 if (attrzp) 2330 VN_RELE(ZTOV(attrzp)); 2331 2332 dmu_tx_commit(tx); 2333 2334 ZFS_EXIT(zfsvfs); 2335 return (err); 2336} 2337 2338typedef struct zfs_zlock { 2339 krwlock_t *zl_rwlock; /* lock we acquired */ 2340 znode_t *zl_znode; /* znode we held */ 2341 struct zfs_zlock *zl_next; /* next in list */ 2342} zfs_zlock_t; 2343 2344/* 2345 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2346 */ 2347static void 2348zfs_rename_unlock(zfs_zlock_t **zlpp) 2349{ 2350 zfs_zlock_t *zl; 2351 2352 while ((zl = *zlpp) != NULL) { 2353 if (zl->zl_znode != NULL) 2354 VN_RELE(ZTOV(zl->zl_znode)); 2355 rw_exit(zl->zl_rwlock); 2356 *zlpp = zl->zl_next; 2357 kmem_free(zl, sizeof (*zl)); 2358 } 2359} 2360 2361/* 2362 * Search back through the directory tree, using the ".." entries. 2363 * Lock each directory in the chain to prevent concurrent renames. 2364 * Fail any attempt to move a directory into one of its own descendants. 2365 * XXX - z_parent_lock can overlap with map or grow locks 2366 */ 2367static int 2368zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2369{ 2370 zfs_zlock_t *zl; 2371 znode_t *zp = tdzp; 2372 uint64_t rootid = zp->z_zfsvfs->z_root; 2373 uint64_t *oidp = &zp->z_id; 2374 krwlock_t *rwlp = &szp->z_parent_lock; 2375 krw_t rw = RW_WRITER; 2376 2377 /* 2378 * First pass write-locks szp and compares to zp->z_id. 2379 * Later passes read-lock zp and compare to zp->z_parent. 2380 */ 2381 do { 2382 if (!rw_tryenter(rwlp, rw)) { 2383 /* 2384 * Another thread is renaming in this path. 2385 * Note that if we are a WRITER, we don't have any 2386 * parent_locks held yet. 2387 */ 2388 if (rw == RW_READER && zp->z_id > szp->z_id) { 2389 /* 2390 * Drop our locks and restart 2391 */ 2392 zfs_rename_unlock(&zl); 2393 *zlpp = NULL; 2394 zp = tdzp; 2395 oidp = &zp->z_id; 2396 rwlp = &szp->z_parent_lock; 2397 rw = RW_WRITER; 2398 continue; 2399 } else { 2400 /* 2401 * Wait for other thread to drop its locks 2402 */ 2403 rw_enter(rwlp, rw); 2404 } 2405 } 2406 2407 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2408 zl->zl_rwlock = rwlp; 2409 zl->zl_znode = NULL; 2410 zl->zl_next = *zlpp; 2411 *zlpp = zl; 2412 2413 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2414 return (EINVAL); 2415 2416 if (*oidp == rootid) /* We've hit the top */ 2417 return (0); 2418 2419 if (rw == RW_READER) { /* i.e. not the first pass */ 2420 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2421 if (error) 2422 return (error); 2423 zl->zl_znode = zp; 2424 } 2425 oidp = &zp->z_phys->zp_parent; 2426 rwlp = &zp->z_parent_lock; 2427 rw = RW_READER; 2428 2429 } while (zp->z_id != sdzp->z_id); 2430 2431 return (0); 2432} 2433 2434/* 2435 * Move an entry from the provided source directory to the target 2436 * directory. Change the entry name as indicated. 2437 * 2438 * IN: sdvp - Source directory containing the "old entry". 2439 * snm - Old entry name. 2440 * tdvp - Target directory to contain the "new entry". 2441 * tnm - New entry name. 2442 * cr - credentials of caller. 2443 * 2444 * RETURN: 0 if success 2445 * error code if failure 2446 * 2447 * Timestamps: 2448 * sdvp,tdvp - ctime|mtime updated 2449 */ 2450static int 2451zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr) 2452{ 2453 znode_t *tdzp, *szp, *tzp; 2454 znode_t *sdzp = VTOZ(sdvp); 2455 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2456 zilog_t *zilog = zfsvfs->z_log; 2457 vnode_t *realvp; 2458 zfs_dirlock_t *sdl, *tdl; 2459 dmu_tx_t *tx; 2460 zfs_zlock_t *zl; 2461 int cmp, serr, terr, error; 2462 2463 ZFS_ENTER(zfsvfs); 2464 2465 /* 2466 * Make sure we have the real vp for the target directory. 2467 */ 2468 if (VOP_REALVP(tdvp, &realvp) == 0) 2469 tdvp = realvp; 2470 2471 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2472 ZFS_EXIT(zfsvfs); 2473 return (EXDEV); 2474 } 2475 2476 tdzp = VTOZ(tdvp); 2477top: 2478 szp = NULL; 2479 tzp = NULL; 2480 zl = NULL; 2481 2482 /* 2483 * This is to prevent the creation of links into attribute space 2484 * by renaming a linked file into/outof an attribute directory. 2485 * See the comment in zfs_link() for why this is considered bad. 2486 */ 2487 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2488 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2489 ZFS_EXIT(zfsvfs); 2490 return (EINVAL); 2491 } 2492 2493 /* 2494 * Lock source and target directory entries. To prevent deadlock, 2495 * a lock ordering must be defined. We lock the directory with 2496 * the smallest object id first, or if it's a tie, the one with 2497 * the lexically first name. 2498 */ 2499 if (sdzp->z_id < tdzp->z_id) { 2500 cmp = -1; 2501 } else if (sdzp->z_id > tdzp->z_id) { 2502 cmp = 1; 2503 } else { 2504 cmp = strcmp(snm, tnm); 2505 if (cmp == 0) { 2506 /* 2507 * POSIX: "If the old argument and the new argument 2508 * both refer to links to the same existing file, 2509 * the rename() function shall return successfully 2510 * and perform no other action." 2511 */ 2512 ZFS_EXIT(zfsvfs); 2513 return (0); 2514 } 2515 } 2516 if (cmp < 0) { 2517 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2518 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2519 } else { 2520 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2521 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2522 } 2523 2524 if (serr) { 2525 /* 2526 * Source entry invalid or not there. 2527 */ 2528 if (!terr) { 2529 zfs_dirent_unlock(tdl); 2530 if (tzp) 2531 VN_RELE(ZTOV(tzp)); 2532 } 2533 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 2534 serr = EINVAL; 2535 ZFS_EXIT(zfsvfs); 2536 return (serr); 2537 } 2538 if (terr) { 2539 zfs_dirent_unlock(sdl); 2540 VN_RELE(ZTOV(szp)); 2541 if (strcmp(tnm, "..") == 0) 2542 terr = EINVAL; 2543 ZFS_EXIT(zfsvfs); 2544 return (terr); 2545 } 2546 2547 /* 2548 * Must have write access at the source to remove the old entry 2549 * and write access at the target to create the new entry. 2550 * Note that if target and source are the same, this can be 2551 * done in a single check. 2552 */ 2553 2554 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2555 goto out; 2556 2557 if (ZTOV(szp)->v_type == VDIR) { 2558 /* 2559 * Check to make sure rename is valid. 2560 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2561 */ 2562 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 2563 goto out; 2564 } 2565 2566 /* 2567 * Does target exist? 2568 */ 2569 if (tzp) { 2570 /* 2571 * Source and target must be the same type. 2572 */ 2573 if (ZTOV(szp)->v_type == VDIR) { 2574 if (ZTOV(tzp)->v_type != VDIR) { 2575 error = ENOTDIR; 2576 goto out; 2577 } 2578 } else { 2579 if (ZTOV(tzp)->v_type == VDIR) { 2580 error = EISDIR; 2581 goto out; 2582 } 2583 } 2584 /* 2585 * POSIX dictates that when the source and target 2586 * entries refer to the same file object, rename 2587 * must do nothing and exit without error. 2588 */ 2589 if (szp->z_id == tzp->z_id) { 2590 error = 0; 2591 goto out; 2592 } 2593 } 2594 2595 vnevent_rename_src(ZTOV(szp)); 2596 if (tzp) 2597 vnevent_rename_dest(ZTOV(tzp)); 2598 2599 tx = dmu_tx_create(zfsvfs->z_os); 2600 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 2601 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 2602 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 2603 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 2604 if (sdzp != tdzp) 2605 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 2606 if (tzp) 2607 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 2608 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2609 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2610 if (error) { 2611 if (zl != NULL) 2612 zfs_rename_unlock(&zl); 2613 zfs_dirent_unlock(sdl); 2614 zfs_dirent_unlock(tdl); 2615 VN_RELE(ZTOV(szp)); 2616 if (tzp) 2617 VN_RELE(ZTOV(tzp)); 2618 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2619 dmu_tx_wait(tx); 2620 dmu_tx_abort(tx); 2621 goto top; 2622 } 2623 dmu_tx_abort(tx); 2624 ZFS_EXIT(zfsvfs); 2625 return (error); 2626 } 2627 2628 if (tzp) /* Attempt to remove the existing target */ 2629 error = zfs_link_destroy(tdl, tzp, tx, 0, NULL); 2630 2631 if (error == 0) { 2632 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 2633 if (error == 0) { 2634 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 2635 ASSERT(error == 0); 2636 zfs_log_rename(zilog, tx, TX_RENAME, sdzp, 2637 sdl->dl_name, tdzp, tdl->dl_name, szp); 2638 } 2639#ifdef FREEBSD_NAMECACHE 2640 if (error == 0) { 2641 cache_purge(sdvp); 2642 cache_purge(tdvp); 2643 } 2644#endif 2645 } 2646 2647 dmu_tx_commit(tx); 2648out: 2649 if (zl != NULL) 2650 zfs_rename_unlock(&zl); 2651 2652 zfs_dirent_unlock(sdl); 2653 zfs_dirent_unlock(tdl); 2654 2655 VN_RELE(ZTOV(szp)); 2656 if (tzp) 2657 VN_RELE(ZTOV(tzp)); 2658 2659 ZFS_EXIT(zfsvfs); 2660 2661 return (error); 2662} 2663 2664/* 2665 * Insert the indicated symbolic reference entry into the directory. 2666 * 2667 * IN: dvp - Directory to contain new symbolic link. 2668 * link - Name for new symlink entry. 2669 * vap - Attributes of new entry. 2670 * target - Target path of new symlink. 2671 * cr - credentials of caller. 2672 * 2673 * RETURN: 0 if success 2674 * error code if failure 2675 * 2676 * Timestamps: 2677 * dvp - ctime|mtime updated 2678 */ 2679static int 2680zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, cred_t *cr, kthread_t *td) 2681{ 2682 znode_t *zp, *dzp = VTOZ(dvp); 2683 zfs_dirlock_t *dl; 2684 dmu_tx_t *tx; 2685 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2686 zilog_t *zilog = zfsvfs->z_log; 2687 uint64_t zoid; 2688 int len = strlen(link); 2689 int error; 2690 2691 ASSERT(vap->va_type == VLNK); 2692 2693 ZFS_ENTER(zfsvfs); 2694top: 2695 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2696 ZFS_EXIT(zfsvfs); 2697 return (error); 2698 } 2699 2700 if (len > MAXPATHLEN) { 2701 ZFS_EXIT(zfsvfs); 2702 return (ENAMETOOLONG); 2703 } 2704 2705 /* 2706 * Attempt to lock directory; fail if entry already exists. 2707 */ 2708 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) { 2709 ZFS_EXIT(zfsvfs); 2710 return (error); 2711 } 2712 2713 tx = dmu_tx_create(zfsvfs->z_os); 2714 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 2715 dmu_tx_hold_bonus(tx, dzp->z_id); 2716 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2717 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 2718 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 2719 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2720 if (error) { 2721 zfs_dirent_unlock(dl); 2722 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2723 dmu_tx_wait(tx); 2724 dmu_tx_abort(tx); 2725 goto top; 2726 } 2727 dmu_tx_abort(tx); 2728 ZFS_EXIT(zfsvfs); 2729 return (error); 2730 } 2731 2732 dmu_buf_will_dirty(dzp->z_dbuf, tx); 2733 2734 /* 2735 * Create a new object for the symlink. 2736 * Put the link content into bonus buffer if it will fit; 2737 * otherwise, store it just like any other file data. 2738 */ 2739 zoid = 0; 2740 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 2741 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len); 2742 if (len != 0) 2743 bcopy(link, zp->z_phys + 1, len); 2744 } else { 2745 dmu_buf_t *dbp; 2746 2747 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 2748 2749 /* 2750 * Nothing can access the znode yet so no locking needed 2751 * for growing the znode's blocksize. 2752 */ 2753 zfs_grow_blocksize(zp, len, tx); 2754 2755 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp)); 2756 dmu_buf_will_dirty(dbp, tx); 2757 2758 ASSERT3U(len, <=, dbp->db_size); 2759 bcopy(link, dbp->db_data, len); 2760 dmu_buf_rele(dbp, FTAG); 2761 } 2762 zp->z_phys->zp_size = len; 2763 2764 /* 2765 * Insert the new object into the directory. 2766 */ 2767 (void) zfs_link_create(dl, zp, tx, ZNEW); 2768out: 2769 if (error == 0) { 2770 zfs_log_symlink(zilog, tx, TX_SYMLINK, dzp, zp, name, link); 2771 *vpp = ZTOV(zp); 2772 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 2773 } 2774 2775 dmu_tx_commit(tx); 2776 2777 zfs_dirent_unlock(dl); 2778 2779 ZFS_EXIT(zfsvfs); 2780 return (error); 2781} 2782 2783/* 2784 * Return, in the buffer contained in the provided uio structure, 2785 * the symbolic path referred to by vp. 2786 * 2787 * IN: vp - vnode of symbolic link. 2788 * uoip - structure to contain the link path. 2789 * cr - credentials of caller. 2790 * 2791 * OUT: uio - structure to contain the link path. 2792 * 2793 * RETURN: 0 if success 2794 * error code if failure 2795 * 2796 * Timestamps: 2797 * vp - atime updated 2798 */ 2799/* ARGSUSED */ 2800static int 2801zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr) 2802{ 2803 znode_t *zp = VTOZ(vp); 2804 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2805 size_t bufsz; 2806 int error; 2807 2808 ZFS_ENTER(zfsvfs); 2809 2810 bufsz = (size_t)zp->z_phys->zp_size; 2811 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 2812 error = uiomove(zp->z_phys + 1, 2813 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2814 } else { 2815 dmu_buf_t *dbp; 2816 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 2817 if (error) { 2818 ZFS_EXIT(zfsvfs); 2819 return (error); 2820 } 2821 error = uiomove(dbp->db_data, 2822 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2823 dmu_buf_rele(dbp, FTAG); 2824 } 2825 2826 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2827 ZFS_EXIT(zfsvfs); 2828 return (error); 2829} 2830 2831/* 2832 * Insert a new entry into directory tdvp referencing svp. 2833 * 2834 * IN: tdvp - Directory to contain new entry. 2835 * svp - vnode of new entry. 2836 * name - name of new entry. 2837 * cr - credentials of caller. 2838 * 2839 * RETURN: 0 if success 2840 * error code if failure 2841 * 2842 * Timestamps: 2843 * tdvp - ctime|mtime updated 2844 * svp - ctime updated 2845 */ 2846/* ARGSUSED */ 2847static int 2848zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr) 2849{ 2850 znode_t *dzp = VTOZ(tdvp); 2851 znode_t *tzp, *szp; 2852 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2853 zilog_t *zilog = zfsvfs->z_log; 2854 zfs_dirlock_t *dl; 2855 dmu_tx_t *tx; 2856 vnode_t *realvp; 2857 int error; 2858 2859 ASSERT(tdvp->v_type == VDIR); 2860 2861 ZFS_ENTER(zfsvfs); 2862 2863 if (VOP_REALVP(svp, &realvp) == 0) 2864 svp = realvp; 2865 2866 if (svp->v_vfsp != tdvp->v_vfsp) { 2867 ZFS_EXIT(zfsvfs); 2868 return (EXDEV); 2869 } 2870 2871 szp = VTOZ(svp); 2872top: 2873 /* 2874 * We do not support links between attributes and non-attributes 2875 * because of the potential security risk of creating links 2876 * into "normal" file space in order to circumvent restrictions 2877 * imposed in attribute space. 2878 */ 2879 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 2880 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 2881 ZFS_EXIT(zfsvfs); 2882 return (EINVAL); 2883 } 2884 2885 /* 2886 * POSIX dictates that we return EPERM here. 2887 * Better choices include ENOTSUP or EISDIR. 2888 */ 2889 if (svp->v_type == VDIR) { 2890 ZFS_EXIT(zfsvfs); 2891 return (EPERM); 2892 } 2893 2894 if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) && 2895 secpolicy_basic_link(cr) != 0) { 2896 ZFS_EXIT(zfsvfs); 2897 return (EPERM); 2898 } 2899 2900 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2901 ZFS_EXIT(zfsvfs); 2902 return (error); 2903 } 2904 2905 /* 2906 * Attempt to lock directory; fail if entry already exists. 2907 */ 2908 if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) { 2909 ZFS_EXIT(zfsvfs); 2910 return (error); 2911 } 2912 2913 tx = dmu_tx_create(zfsvfs->z_os); 2914 dmu_tx_hold_bonus(tx, szp->z_id); 2915 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2916 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2917 if (error) { 2918 zfs_dirent_unlock(dl); 2919 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2920 dmu_tx_wait(tx); 2921 dmu_tx_abort(tx); 2922 goto top; 2923 } 2924 dmu_tx_abort(tx); 2925 ZFS_EXIT(zfsvfs); 2926 return (error); 2927 } 2928 2929 error = zfs_link_create(dl, szp, tx, 0); 2930 2931 if (error == 0) 2932 zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name); 2933 2934 dmu_tx_commit(tx); 2935 2936 zfs_dirent_unlock(dl); 2937 2938 ZFS_EXIT(zfsvfs); 2939 return (error); 2940} 2941 2942void 2943zfs_inactive(vnode_t *vp, cred_t *cr) 2944{ 2945 znode_t *zp = VTOZ(vp); 2946 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2947 int error; 2948 2949 rw_enter(&zfsvfs->z_um_lock, RW_READER); 2950 if (zfsvfs->z_unmounted2) { 2951 ASSERT(zp->z_dbuf_held == 0); 2952 2953 mutex_enter(&zp->z_lock); 2954 VI_LOCK(vp); 2955 vp->v_count = 0; /* count arrives as 1 */ 2956 VI_UNLOCK(vp); 2957 if (zp->z_dbuf == NULL) { 2958 mutex_exit(&zp->z_lock); 2959 zfs_znode_free(zp); 2960 } else { 2961 mutex_exit(&zp->z_lock); 2962 } 2963 rw_exit(&zfsvfs->z_um_lock); 2964 VFS_RELE(zfsvfs->z_vfs); 2965 return; 2966 } 2967 2968 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 2969 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 2970 2971 dmu_tx_hold_bonus(tx, zp->z_id); 2972 error = dmu_tx_assign(tx, TXG_WAIT); 2973 if (error) { 2974 dmu_tx_abort(tx); 2975 } else { 2976 dmu_buf_will_dirty(zp->z_dbuf, tx); 2977 mutex_enter(&zp->z_lock); 2978 zp->z_atime_dirty = 0; 2979 mutex_exit(&zp->z_lock); 2980 dmu_tx_commit(tx); 2981 } 2982 } 2983 2984 zfs_zinactive(zp); 2985 rw_exit(&zfsvfs->z_um_lock); 2986} 2987 2988CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 2989CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 2990 2991static int 2992zfs_fid(vnode_t *vp, fid_t *fidp) 2993{ 2994 znode_t *zp = VTOZ(vp); 2995 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2996 uint32_t gen = (uint32_t)zp->z_phys->zp_gen; 2997 uint64_t object = zp->z_id; 2998 zfid_short_t *zfid; 2999 int size, i; 3000 3001 ZFS_ENTER(zfsvfs); 3002 3003 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 3004 fidp->fid_len = size; 3005 3006 zfid = (zfid_short_t *)fidp; 3007 3008 zfid->zf_len = size; 3009 3010 for (i = 0; i < sizeof (zfid->zf_object); i++) 3011 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 3012 3013 /* Must have a non-zero generation number to distinguish from .zfs */ 3014 if (gen == 0) 3015 gen = 1; 3016 for (i = 0; i < sizeof (zfid->zf_gen); i++) 3017 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 3018 3019 if (size == LONG_FID_LEN) { 3020 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 3021 zfid_long_t *zlfid; 3022 3023 zlfid = (zfid_long_t *)fidp; 3024 3025 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 3026 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 3027 3028 /* XXX - this should be the generation number for the objset */ 3029 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 3030 zlfid->zf_setgen[i] = 0; 3031 } 3032 3033 ZFS_EXIT(zfsvfs); 3034 return (0); 3035} 3036 3037static int 3038zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr) 3039{ 3040 znode_t *zp, *xzp; 3041 zfsvfs_t *zfsvfs; 3042 zfs_dirlock_t *dl; 3043 int error; 3044 3045 switch (cmd) { 3046 case _PC_LINK_MAX: 3047 *valp = INT_MAX; 3048 return (0); 3049 3050 case _PC_FILESIZEBITS: 3051 *valp = 64; 3052 return (0); 3053 3054#if 0 3055 case _PC_XATTR_EXISTS: 3056 zp = VTOZ(vp); 3057 zfsvfs = zp->z_zfsvfs; 3058 ZFS_ENTER(zfsvfs); 3059 *valp = 0; 3060 error = zfs_dirent_lock(&dl, zp, "", &xzp, 3061 ZXATTR | ZEXISTS | ZSHARED); 3062 if (error == 0) { 3063 zfs_dirent_unlock(dl); 3064 if (!zfs_dirempty(xzp)) 3065 *valp = 1; 3066 VN_RELE(ZTOV(xzp)); 3067 } else if (error == ENOENT) { 3068 /* 3069 * If there aren't extended attributes, it's the 3070 * same as having zero of them. 3071 */ 3072 error = 0; 3073 } 3074 ZFS_EXIT(zfsvfs); 3075 return (error); 3076#endif 3077 3078 case _PC_ACL_EXTENDED: 3079 *valp = 0; /* TODO */ 3080 return (0); 3081 3082 case _PC_MIN_HOLE_SIZE: 3083 *valp = (int)SPA_MINBLOCKSIZE; 3084 return (0); 3085 3086 default: 3087 return (EOPNOTSUPP); 3088 } 3089} 3090 3091#ifdef TODO 3092/*ARGSUSED*/ 3093static int 3094zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3095{ 3096 znode_t *zp = VTOZ(vp); 3097 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3098 int error; 3099 3100 ZFS_ENTER(zfsvfs); 3101 error = zfs_getacl(zp, vsecp, cr); 3102 ZFS_EXIT(zfsvfs); 3103 3104 return (error); 3105} 3106#endif /* TODO */ 3107 3108#ifdef TODO 3109/*ARGSUSED*/ 3110static int 3111zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3112{ 3113 znode_t *zp = VTOZ(vp); 3114 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3115 int error; 3116 3117 ZFS_ENTER(zfsvfs); 3118 error = zfs_setacl(zp, vsecp, cr); 3119 ZFS_EXIT(zfsvfs); 3120 return (error); 3121} 3122#endif /* TODO */ 3123 3124static int 3125zfs_freebsd_open(ap) 3126 struct vop_open_args /* { 3127 struct vnode *a_vp; 3128 int a_mode; 3129 struct ucred *a_cred; 3130 struct thread *a_td; 3131 } */ *ap; 3132{ 3133 vnode_t *vp = ap->a_vp; 3134 znode_t *zp = VTOZ(vp); 3135 int error; 3136 3137 error = zfs_open(&vp, ap->a_mode, ap->a_cred); 3138 if (error == 0) 3139 vnode_create_vobject(vp, zp->z_phys->zp_size, ap->a_td); 3140 return (error); 3141} 3142 3143static int 3144zfs_freebsd_close(ap) 3145 struct vop_close_args /* { 3146 struct vnode *a_vp; 3147 int a_fflag; 3148 struct ucred *a_cred; 3149 struct thread *a_td; 3150 } */ *ap; 3151{ 3152 3153 return (zfs_close(ap->a_vp, ap->a_fflag, 0, 0, ap->a_cred)); 3154} 3155 3156static int 3157zfs_freebsd_ioctl(ap) 3158 struct vop_ioctl_args /* { 3159 struct vnode *a_vp; 3160 u_long a_command; 3161 caddr_t a_data; 3162 int a_fflag; 3163 struct ucred *cred; 3164 struct thread *td; 3165 } */ *ap; 3166{ 3167 3168 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 3169 ap->a_fflag, ap->a_cred, NULL)); 3170} 3171 3172static int 3173zfs_freebsd_read(ap) 3174 struct vop_read_args /* { 3175 struct vnode *a_vp; 3176 struct uio *a_uio; 3177 int a_ioflag; 3178 struct ucred *a_cred; 3179 } */ *ap; 3180{ 3181 3182 return (zfs_read(ap->a_vp, ap->a_uio, ap->a_ioflag, ap->a_cred, NULL)); 3183} 3184 3185static int 3186zfs_freebsd_write(ap) 3187 struct vop_write_args /* { 3188 struct vnode *a_vp; 3189 struct uio *a_uio; 3190 int a_ioflag; 3191 struct ucred *a_cred; 3192 } */ *ap; 3193{ 3194 3195 return (zfs_write(ap->a_vp, ap->a_uio, ap->a_ioflag, ap->a_cred, NULL)); 3196} 3197 3198static int 3199zfs_freebsd_access(ap) 3200 struct vop_access_args /* { 3201 struct vnode *a_vp; 3202 int a_mode; 3203 struct ucred *a_cred; 3204 struct thread *a_td; 3205 } */ *ap; 3206{ 3207 3208 return (zfs_access(ap->a_vp, ap->a_mode, 0, ap->a_cred)); 3209} 3210 3211static int 3212zfs_freebsd_lookup(ap) 3213 struct vop_lookup_args /* { 3214 struct vnode *a_dvp; 3215 struct vnode **a_vpp; 3216 struct componentname *a_cnp; 3217 } */ *ap; 3218{ 3219 struct componentname *cnp = ap->a_cnp; 3220 char nm[NAME_MAX + 1]; 3221 3222 ASSERT(cnp->cn_namelen < sizeof(nm)); 3223 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 3224 3225 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 3226 cnp->cn_cred, cnp->cn_thread)); 3227} 3228 3229static int 3230zfs_freebsd_create(ap) 3231 struct vop_create_args /* { 3232 struct vnode *a_dvp; 3233 struct vnode **a_vpp; 3234 struct componentname *a_cnp; 3235 struct vattr *a_vap; 3236 } */ *ap; 3237{ 3238 struct componentname *cnp = ap->a_cnp; 3239 vattr_t *vap = ap->a_vap; 3240 int mode; 3241 3242 ASSERT(cnp->cn_flags & SAVENAME); 3243 3244 vattr_init_mask(vap); 3245 mode = vap->va_mode & ALLPERMS; 3246 3247 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 3248 ap->a_vpp, cnp->cn_cred, cnp->cn_thread)); 3249} 3250 3251static int 3252zfs_freebsd_remove(ap) 3253 struct vop_remove_args /* { 3254 struct vnode *a_dvp; 3255 struct vnode *a_vp; 3256 struct componentname *a_cnp; 3257 } */ *ap; 3258{ 3259 3260 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 3261 3262 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 3263 ap->a_cnp->cn_cred)); 3264} 3265 3266static int 3267zfs_freebsd_mkdir(ap) 3268 struct vop_mkdir_args /* { 3269 struct vnode *a_dvp; 3270 struct vnode **a_vpp; 3271 struct componentname *a_cnp; 3272 struct vattr *a_vap; 3273 } */ *ap; 3274{ 3275 vattr_t *vap = ap->a_vap; 3276 3277 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 3278 3279 vattr_init_mask(vap); 3280 3281 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 3282 ap->a_cnp->cn_cred)); 3283} 3284 3285static int 3286zfs_freebsd_rmdir(ap) 3287 struct vop_rmdir_args /* { 3288 struct vnode *a_dvp; 3289 struct vnode *a_vp; 3290 struct componentname *a_cnp; 3291 } */ *ap; 3292{ 3293 struct componentname *cnp = ap->a_cnp; 3294 3295 ASSERT(cnp->cn_flags & SAVENAME); 3296 3297 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred)); 3298} 3299 3300static int 3301zfs_freebsd_readdir(ap) 3302 struct vop_readdir_args /* { 3303 struct vnode *a_vp; 3304 struct uio *a_uio; 3305 struct ucred *a_cred; 3306 int *a_eofflag; 3307 int *a_ncookies; 3308 u_long **a_cookies; 3309 } */ *ap; 3310{ 3311 3312 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 3313 ap->a_ncookies, ap->a_cookies)); 3314} 3315 3316static int 3317zfs_freebsd_fsync(ap) 3318 struct vop_fsync_args /* { 3319 struct vnode *a_vp; 3320 int a_waitfor; 3321 struct thread *a_td; 3322 } */ *ap; 3323{ 3324 3325 vop_stdfsync(ap); 3326 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred)); 3327} 3328 3329static int 3330zfs_freebsd_getattr(ap) 3331 struct vop_getattr_args /* { 3332 struct vnode *a_vp; 3333 struct vattr *a_vap; 3334 struct ucred *a_cred; 3335 struct thread *a_td; 3336 } */ *ap; 3337{ 3338 3339 return (zfs_getattr(ap->a_vp, ap->a_vap, 0, ap->a_cred)); 3340} 3341 3342static int 3343zfs_freebsd_setattr(ap) 3344 struct vop_setattr_args /* { 3345 struct vnode *a_vp; 3346 struct vattr *a_vap; 3347 struct ucred *a_cred; 3348 struct thread *a_td; 3349 } */ *ap; 3350{ 3351 vattr_t *vap = ap->a_vap; 3352 3353 /* No support for FreeBSD's chflags(2). */ 3354 if (vap->va_flags != VNOVAL) 3355 return (EOPNOTSUPP); 3356 3357 vattr_init_mask(vap); 3358 vap->va_mask &= ~AT_NOSET; 3359 3360 return (zfs_setattr(ap->a_vp, vap, 0, ap->a_cred, NULL)); 3361} 3362 3363static int 3364zfs_freebsd_rename(ap) 3365 struct vop_rename_args /* { 3366 struct vnode *a_fdvp; 3367 struct vnode *a_fvp; 3368 struct componentname *a_fcnp; 3369 struct vnode *a_tdvp; 3370 struct vnode *a_tvp; 3371 struct componentname *a_tcnp; 3372 } */ *ap; 3373{ 3374 vnode_t *fdvp = ap->a_fdvp; 3375 vnode_t *fvp = ap->a_fvp; 3376 vnode_t *tdvp = ap->a_tdvp; 3377 vnode_t *tvp = ap->a_tvp; 3378 int error; 3379 3380 ASSERT(ap->a_fcnp->cn_flags & SAVENAME); 3381 ASSERT(ap->a_tcnp->cn_flags & SAVENAME); 3382 3383 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 3384 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred); 3385 3386 if (tdvp == tvp) 3387 VN_RELE(tdvp); 3388 else 3389 VN_URELE(tdvp); 3390 if (tvp) 3391 VN_URELE(tvp); 3392 VN_RELE(fdvp); 3393 VN_RELE(fvp); 3394 3395 return (error); 3396} 3397 3398static int 3399zfs_freebsd_symlink(ap) 3400 struct vop_symlink_args /* { 3401 struct vnode *a_dvp; 3402 struct vnode **a_vpp; 3403 struct componentname *a_cnp; 3404 struct vattr *a_vap; 3405 char *a_target; 3406 } */ *ap; 3407{ 3408 struct componentname *cnp = ap->a_cnp; 3409 vattr_t *vap = ap->a_vap; 3410 3411 ASSERT(cnp->cn_flags & SAVENAME); 3412 3413 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 3414 vattr_init_mask(vap); 3415 3416 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 3417 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 3418} 3419 3420static int 3421zfs_freebsd_readlink(ap) 3422 struct vop_readlink_args /* { 3423 struct vnode *a_vp; 3424 struct uio *a_uio; 3425 struct ucred *a_cred; 3426 } */ *ap; 3427{ 3428 3429 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred)); 3430} 3431 3432static int 3433zfs_freebsd_link(ap) 3434 struct vop_link_args /* { 3435 struct vnode *a_tdvp; 3436 struct vnode *a_vp; 3437 struct componentname *a_cnp; 3438 } */ *ap; 3439{ 3440 struct componentname *cnp = ap->a_cnp; 3441 3442 ASSERT(cnp->cn_flags & SAVENAME); 3443 3444 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred)); 3445} 3446 3447static int 3448zfs_freebsd_inactive(ap) 3449 struct vop_inactive_args /* { 3450 struct vnode *a_vp; 3451 struct thread *a_td; 3452 } */ *ap; 3453{ 3454 vnode_t *vp = ap->a_vp; 3455 3456 zfs_inactive(vp, ap->a_td->td_ucred); 3457 return (0); 3458} 3459 3460static int 3461zfs_freebsd_reclaim(ap) 3462 struct vop_reclaim_args /* { 3463 struct vnode *a_vp; 3464 struct thread *a_td; 3465 } */ *ap; 3466{ 3467 vnode_t *vp = ap->a_vp; 3468 znode_t *zp = VTOZ(vp); 3469 zfsvfs_t *zfsvfs; 3470 int rele = 1; 3471 3472 ASSERT(zp != NULL); 3473 3474 /* 3475 * Destroy the vm object and flush associated pages. 3476 */ 3477 vnode_destroy_vobject(vp); 3478 3479 mutex_enter(&zp->z_lock); 3480 ASSERT(zp->z_phys); 3481 ASSERT(zp->z_dbuf_held); 3482 zfsvfs = zp->z_zfsvfs; 3483 if (!zp->z_unlinked) { 3484 zp->z_dbuf_held = 0; 3485 ZTOV(zp) = NULL; 3486 mutex_exit(&zp->z_lock); 3487 dmu_buf_rele(zp->z_dbuf, NULL); 3488 } else { 3489 mutex_exit(&zp->z_lock); 3490 } 3491 VI_LOCK(vp); 3492 if (vp->v_count > 0) 3493 rele = 0; 3494 vp->v_data = NULL; 3495 ASSERT(vp->v_holdcnt >= 1); 3496 VI_UNLOCK(vp); 3497 if (!zp->z_unlinked && rele) 3498 VFS_RELE(zfsvfs->z_vfs); 3499 return (0); 3500} 3501 3502static int 3503zfs_freebsd_fid(ap) 3504 struct vop_fid_args /* { 3505 struct vnode *a_vp; 3506 struct fid *a_fid; 3507 } */ *ap; 3508{ 3509 3510 return (zfs_fid(ap->a_vp, (void *)ap->a_fid)); 3511} 3512 3513static int 3514zfs_freebsd_pathconf(ap) 3515 struct vop_pathconf_args /* { 3516 struct vnode *a_vp; 3517 int a_name; 3518 register_t *a_retval; 3519 } */ *ap; 3520{ 3521 ulong_t val; 3522 int error; 3523 3524 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred); 3525 if (error == 0) 3526 *ap->a_retval = val; 3527 else if (error == EOPNOTSUPP) 3528 error = vop_stdpathconf(ap); 3529 return (error); 3530} 3531 3532/* 3533 * Advisory record locking support 3534 */ 3535static int 3536zfs_freebsd_advlock(ap) 3537 struct vop_advlock_args /* { 3538 struct vnode *a_vp; 3539 caddr_t a_id; 3540 int a_op; 3541 struct flock *a_fl; 3542 int a_flags; 3543 } */ *ap; 3544{ 3545 znode_t *zp = VTOZ(ap->a_vp); 3546 3547 return (lf_advlock(ap, &(zp->z_lockf), zp->z_phys->zp_size)); 3548} 3549 3550struct vop_vector zfs_vnodeops; 3551struct vop_vector zfs_fifoops; 3552 3553struct vop_vector zfs_vnodeops = { 3554 .vop_default = &default_vnodeops, 3555 .vop_inactive = zfs_freebsd_inactive, 3556 .vop_reclaim = zfs_freebsd_reclaim, 3557 .vop_access = zfs_freebsd_access, 3558#ifdef FREEBSD_NAMECACHE 3559 .vop_lookup = vfs_cache_lookup, 3560 .vop_cachedlookup = zfs_freebsd_lookup, 3561#else 3562 .vop_lookup = zfs_freebsd_lookup, 3563#endif 3564 .vop_getattr = zfs_freebsd_getattr, 3565 .vop_setattr = zfs_freebsd_setattr, 3566 .vop_create = zfs_freebsd_create, 3567 .vop_mknod = zfs_freebsd_create, 3568 .vop_mkdir = zfs_freebsd_mkdir, 3569 .vop_readdir = zfs_freebsd_readdir, 3570 .vop_fsync = zfs_freebsd_fsync, 3571 .vop_open = zfs_freebsd_open, 3572 .vop_close = zfs_freebsd_close, 3573 .vop_rmdir = zfs_freebsd_rmdir, 3574 .vop_ioctl = zfs_freebsd_ioctl, 3575 .vop_link = zfs_freebsd_link, 3576 .vop_symlink = zfs_freebsd_symlink, 3577 .vop_readlink = zfs_freebsd_readlink, 3578 .vop_read = zfs_freebsd_read, 3579 .vop_write = zfs_freebsd_write, 3580 .vop_remove = zfs_freebsd_remove, 3581 .vop_rename = zfs_freebsd_rename, 3582 .vop_advlock = zfs_freebsd_advlock, 3583 .vop_pathconf = zfs_freebsd_pathconf, 3584 .vop_bmap = VOP_EOPNOTSUPP, 3585 .vop_fid = zfs_freebsd_fid, 3586}; 3587 3588struct vop_vector zfs_fifoops = { 3589 .vop_default = &fifo_specops, 3590 .vop_fsync = VOP_PANIC, 3591 .vop_access = zfs_freebsd_access, 3592 .vop_getattr = zfs_freebsd_getattr, 3593 .vop_inactive = zfs_freebsd_inactive, 3594 .vop_read = VOP_PANIC, 3595 .vop_reclaim = zfs_freebsd_reclaim, 3596 .vop_setattr = zfs_freebsd_setattr, 3597 .vop_write = VOP_PANIC, 3598 .vop_fid = zfs_freebsd_fid, 3599}; 3600