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/vfs_opreg.h> 38 #include <sys/vnode.h> 39 #include <sys/file.h> 40 #include <sys/stat.h> 41 #include <sys/kmem.h> 42 #include <sys/taskq.h> 43 #include <sys/uio.h> 44 #include <sys/vmsystm.h> 45 #include <sys/atomic.h> 46 #include <sys/vm.h> 47 #include <vm/seg_vn.h> 48 #include <vm/pvn.h> 49 #include <vm/as.h> 50 #include <sys/mman.h> 51 #include <sys/pathname.h> 52 #include <sys/cmn_err.h> 53 #include <sys/errno.h> 54 #include <sys/unistd.h> 55 #include <sys/zfs_dir.h> 56 #include <sys/zfs_acl.h> 57 #include <sys/zfs_ioctl.h> 58 #include <sys/fs/zfs.h> 59 #include <sys/dmu.h> 60 #include <sys/spa.h> 61 #include <sys/txg.h> 62 #include <sys/dbuf.h> 63 #include <sys/zap.h> 64 #include <sys/dirent.h> 65 #include <sys/policy.h> 66 #include <sys/sunddi.h> 67 #include <sys/filio.h> 68 #include "fs/fs_subr.h" 69 #include <sys/zfs_ctldir.h> 70 #include <sys/zfs_fuid.h> 71 #include <sys/dnlc.h> 72 #include <sys/zfs_rlock.h> 73 #include <sys/extdirent.h> 74 #include <sys/kidmap.h> 75 #include <sys/cred_impl.h> 76 #include <sys/attr.h> 77 78 /* 79 * Programming rules. 80 * 81 * Each vnode op performs some logical unit of work. To do this, the ZPL must 82 * properly lock its in-core state, create a DMU transaction, do the work, 83 * record this work in the intent log (ZIL), commit the DMU transaction, 84 * and wait for the intent log to commit if it is a synchronous operation. 85 * Moreover, the vnode ops must work in both normal and log replay context. 86 * The ordering of events is important to avoid deadlocks and references 87 * to freed memory. The example below illustrates the following Big Rules: 88 * 89 * (1) A check must be made in each zfs thread for a mounted file system. 90 * This is done avoiding races using ZFS_ENTER(zfsvfs). 91 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes 92 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros 93 * can return EIO from the calling function. 94 * 95 * (2) VN_RELE() should always be the last thing except for zil_commit() 96 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 97 * First, if it's the last reference, the vnode/znode 98 * can be freed, so the zp may point to freed memory. Second, the last 99 * reference will call zfs_zinactive(), which may induce a lot of work -- 100 * pushing cached pages (which acquires range locks) and syncing out 101 * cached atime changes. Third, zfs_zinactive() may require a new tx, 102 * which could deadlock the system if you were already holding one. 103 * 104 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 105 * as they can span dmu_tx_assign() calls. 106 * 107 * (4) Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign(). 108 * In normal operation, this will be TXG_NOWAIT. During ZIL replay, 109 * it will be a specific txg. Either way, dmu_tx_assign() never blocks. 110 * This is critical because we don't want to block while holding locks. 111 * Note, in particular, that if a lock is sometimes acquired before 112 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 113 * use a non-blocking assign can deadlock the system. The scenario: 114 * 115 * Thread A has grabbed a lock before calling dmu_tx_assign(). 116 * Thread B is in an already-assigned tx, and blocks for this lock. 117 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 118 * forever, because the previous txg can't quiesce until B's tx commits. 119 * 120 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 121 * then drop all locks, call dmu_tx_wait(), and try again. 122 * 123 * (5) If the operation succeeded, generate the intent log entry for it 124 * before dropping locks. This ensures that the ordering of events 125 * in the intent log matches the order in which they actually occurred. 126 * 127 * (6) At the end of each vnode op, the DMU tx must always commit, 128 * regardless of whether there were any errors. 129 * 130 * (7) After dropping all locks, invoke zil_commit(zilog, seq, foid) 131 * to ensure that synchronous semantics are provided when necessary. 132 * 133 * In general, this is how things should be ordered in each vnode op: 134 * 135 * ZFS_ENTER(zfsvfs); // exit if unmounted 136 * top: 137 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 138 * rw_enter(...); // grab any other locks you need 139 * tx = dmu_tx_create(...); // get DMU tx 140 * dmu_tx_hold_*(); // hold each object you might modify 141 * error = dmu_tx_assign(tx, zfsvfs->z_assign); // try to assign 142 * if (error) { 143 * rw_exit(...); // drop locks 144 * zfs_dirent_unlock(dl); // unlock directory entry 145 * VN_RELE(...); // release held vnodes 146 * if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 147 * dmu_tx_wait(tx); 148 * dmu_tx_abort(tx); 149 * goto top; 150 * } 151 * dmu_tx_abort(tx); // abort DMU tx 152 * ZFS_EXIT(zfsvfs); // finished in zfs 153 * return (error); // really out of space 154 * } 155 * error = do_real_work(); // do whatever this VOP does 156 * if (error == 0) 157 * zfs_log_*(...); // on success, make ZIL entry 158 * dmu_tx_commit(tx); // commit DMU tx -- error or not 159 * rw_exit(...); // drop locks 160 * zfs_dirent_unlock(dl); // unlock directory entry 161 * VN_RELE(...); // release held vnodes 162 * zil_commit(zilog, seq, foid); // synchronous when necessary 163 * ZFS_EXIT(zfsvfs); // finished in zfs 164 * return (error); // done, report error 165 */ 166 167 /* ARGSUSED */ 168 static int 169 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 170 { 171 znode_t *zp = VTOZ(*vpp); 172 173 if ((flag & FWRITE) && (zp->z_phys->zp_flags & ZFS_APPENDONLY) && 174 ((flag & FAPPEND) == 0)) { 175 return (EPERM); 176 } 177 178 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 179 ZTOV(zp)->v_type == VREG && 180 !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) && 181 zp->z_phys->zp_size > 0) 182 if (fs_vscan(*vpp, cr, 0) != 0) 183 return (EACCES); 184 185 /* Keep a count of the synchronous opens in the znode */ 186 if (flag & (FSYNC | FDSYNC)) 187 atomic_inc_32(&zp->z_sync_cnt); 188 189 return (0); 190 } 191 192 /* ARGSUSED */ 193 static int 194 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 195 caller_context_t *ct) 196 { 197 znode_t *zp = VTOZ(vp); 198 199 /* Decrement the synchronous opens in the znode */ 200 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 201 atomic_dec_32(&zp->z_sync_cnt); 202 203 /* 204 * Clean up any locks held by this process on the vp. 205 */ 206 cleanlocks(vp, ddi_get_pid(), 0); 207 cleanshares(vp, ddi_get_pid()); 208 209 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 210 ZTOV(zp)->v_type == VREG && 211 !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) && 212 zp->z_phys->zp_size > 0) 213 VERIFY(fs_vscan(vp, cr, 1) == 0); 214 215 return (0); 216 } 217 218 /* 219 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 220 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 221 */ 222 static int 223 zfs_holey(vnode_t *vp, int cmd, offset_t *off) 224 { 225 znode_t *zp = VTOZ(vp); 226 uint64_t noff = (uint64_t)*off; /* new offset */ 227 uint64_t file_sz; 228 int error; 229 boolean_t hole; 230 231 file_sz = zp->z_phys->zp_size; 232 if (noff >= file_sz) { 233 return (ENXIO); 234 } 235 236 if (cmd == _FIO_SEEK_HOLE) 237 hole = B_TRUE; 238 else 239 hole = B_FALSE; 240 241 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 242 243 /* end of file? */ 244 if ((error == ESRCH) || (noff > file_sz)) { 245 /* 246 * Handle the virtual hole at the end of file. 247 */ 248 if (hole) { 249 *off = file_sz; 250 return (0); 251 } 252 return (ENXIO); 253 } 254 255 if (noff < *off) 256 return (error); 257 *off = noff; 258 return (error); 259 } 260 261 /* ARGSUSED */ 262 static int 263 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred, 264 int *rvalp, caller_context_t *ct) 265 { 266 offset_t off; 267 int error; 268 zfsvfs_t *zfsvfs; 269 znode_t *zp; 270 271 switch (com) { 272 case _FIOFFS: 273 return (zfs_sync(vp->v_vfsp, 0, cred)); 274 275 /* 276 * The following two ioctls are used by bfu. Faking out, 277 * necessary to avoid bfu errors. 278 */ 279 case _FIOGDIO: 280 case _FIOSDIO: 281 return (0); 282 283 case _FIO_SEEK_DATA: 284 case _FIO_SEEK_HOLE: 285 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 286 return (EFAULT); 287 288 zp = VTOZ(vp); 289 zfsvfs = zp->z_zfsvfs; 290 ZFS_ENTER(zfsvfs); 291 ZFS_VERIFY_ZP(zp); 292 293 /* offset parameter is in/out */ 294 error = zfs_holey(vp, com, &off); 295 ZFS_EXIT(zfsvfs); 296 if (error) 297 return (error); 298 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 299 return (EFAULT); 300 return (0); 301 } 302 return (ENOTTY); 303 } 304 305 /* 306 * When a file is memory mapped, we must keep the IO data synchronized 307 * between the DMU cache and the memory mapped pages. What this means: 308 * 309 * On Write: If we find a memory mapped page, we write to *both* 310 * the page and the dmu buffer. 311 * 312 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 313 * the file is memory mapped. 314 */ 315 static int 316 mappedwrite(vnode_t *vp, int nbytes, uio_t *uio, dmu_tx_t *tx) 317 { 318 znode_t *zp = VTOZ(vp); 319 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 320 int64_t start, off; 321 int len = nbytes; 322 int error = 0; 323 324 start = uio->uio_loffset; 325 off = start & PAGEOFFSET; 326 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 327 page_t *pp; 328 uint64_t bytes = MIN(PAGESIZE - off, len); 329 uint64_t woff = uio->uio_loffset; 330 331 /* 332 * We don't want a new page to "appear" in the middle of 333 * the file update (because it may not get the write 334 * update data), so we grab a lock to block 335 * zfs_getpage(). 336 */ 337 rw_enter(&zp->z_map_lock, RW_WRITER); 338 if (pp = page_lookup(vp, start, SE_SHARED)) { 339 caddr_t va; 340 341 rw_exit(&zp->z_map_lock); 342 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L); 343 error = uiomove(va+off, bytes, UIO_WRITE, uio); 344 if (error == 0) { 345 dmu_write(zfsvfs->z_os, zp->z_id, 346 woff, bytes, va+off, tx); 347 } 348 ppmapout(va); 349 page_unlock(pp); 350 } else { 351 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 352 uio, bytes, tx); 353 rw_exit(&zp->z_map_lock); 354 } 355 len -= bytes; 356 off = 0; 357 if (error) 358 break; 359 } 360 return (error); 361 } 362 363 /* 364 * When a file is memory mapped, we must keep the IO data synchronized 365 * between the DMU cache and the memory mapped pages. What this means: 366 * 367 * On Read: We "read" preferentially from memory mapped pages, 368 * else we default from the dmu buffer. 369 * 370 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 371 * the file is memory mapped. 372 */ 373 static int 374 mappedread(vnode_t *vp, int nbytes, uio_t *uio) 375 { 376 znode_t *zp = VTOZ(vp); 377 objset_t *os = zp->z_zfsvfs->z_os; 378 int64_t start, off; 379 int len = nbytes; 380 int error = 0; 381 382 start = uio->uio_loffset; 383 off = start & PAGEOFFSET; 384 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 385 page_t *pp; 386 uint64_t bytes = MIN(PAGESIZE - off, len); 387 388 if (pp = page_lookup(vp, start, SE_SHARED)) { 389 caddr_t va; 390 391 va = ppmapin(pp, PROT_READ, (caddr_t)-1L); 392 error = uiomove(va + off, bytes, UIO_READ, uio); 393 ppmapout(va); 394 page_unlock(pp); 395 } else { 396 error = dmu_read_uio(os, zp->z_id, uio, bytes); 397 } 398 len -= bytes; 399 off = 0; 400 if (error) 401 break; 402 } 403 return (error); 404 } 405 406 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 407 408 /* 409 * Read bytes from specified file into supplied buffer. 410 * 411 * IN: vp - vnode of file to be read from. 412 * uio - structure supplying read location, range info, 413 * and return buffer. 414 * ioflag - SYNC flags; used to provide FRSYNC semantics. 415 * cr - credentials of caller. 416 * ct - caller context 417 * 418 * OUT: uio - updated offset and range, buffer filled. 419 * 420 * RETURN: 0 if success 421 * error code if failure 422 * 423 * Side Effects: 424 * vp - atime updated if byte count > 0 425 */ 426 /* ARGSUSED */ 427 static int 428 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 429 { 430 znode_t *zp = VTOZ(vp); 431 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 432 objset_t *os; 433 ssize_t n, nbytes; 434 int error; 435 rl_t *rl; 436 437 ZFS_ENTER(zfsvfs); 438 ZFS_VERIFY_ZP(zp); 439 os = zfsvfs->z_os; 440 441 /* 442 * Validate file offset 443 */ 444 if (uio->uio_loffset < (offset_t)0) { 445 ZFS_EXIT(zfsvfs); 446 return (EINVAL); 447 } 448 449 /* 450 * Fasttrack empty reads 451 */ 452 if (uio->uio_resid == 0) { 453 ZFS_EXIT(zfsvfs); 454 return (0); 455 } 456 457 /* 458 * Check for mandatory locks 459 */ 460 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 461 if (error = chklock(vp, FREAD, 462 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 463 ZFS_EXIT(zfsvfs); 464 return (error); 465 } 466 } 467 468 /* 469 * If we're in FRSYNC mode, sync out this znode before reading it. 470 */ 471 if (ioflag & FRSYNC) 472 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 473 474 /* 475 * Lock the range against changes. 476 */ 477 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 478 479 /* 480 * If we are reading past end-of-file we can skip 481 * to the end; but we might still need to set atime. 482 */ 483 if (uio->uio_loffset >= zp->z_phys->zp_size) { 484 error = 0; 485 goto out; 486 } 487 488 ASSERT(uio->uio_loffset < zp->z_phys->zp_size); 489 n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset); 490 491 while (n > 0) { 492 nbytes = MIN(n, zfs_read_chunk_size - 493 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 494 495 if (vn_has_cached_data(vp)) 496 error = mappedread(vp, nbytes, uio); 497 else 498 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 499 if (error) 500 break; 501 502 n -= nbytes; 503 } 504 505 out: 506 zfs_range_unlock(rl); 507 508 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 509 ZFS_EXIT(zfsvfs); 510 return (error); 511 } 512 513 /* 514 * Fault in the pages of the first n bytes specified by the uio structure. 515 * 1 byte in each page is touched and the uio struct is unmodified. 516 * Any error will exit this routine as this is only a best 517 * attempt to get the pages resident. This is a copy of ufs_trans_touch(). 518 */ 519 static void 520 zfs_prefault_write(ssize_t n, struct uio *uio) 521 { 522 struct iovec *iov; 523 ulong_t cnt, incr; 524 caddr_t p; 525 uint8_t tmp; 526 527 iov = uio->uio_iov; 528 529 while (n) { 530 cnt = MIN(iov->iov_len, n); 531 if (cnt == 0) { 532 /* empty iov entry */ 533 iov++; 534 continue; 535 } 536 n -= cnt; 537 /* 538 * touch each page in this segment. 539 */ 540 p = iov->iov_base; 541 while (cnt) { 542 switch (uio->uio_segflg) { 543 case UIO_USERSPACE: 544 case UIO_USERISPACE: 545 if (fuword8(p, &tmp)) 546 return; 547 break; 548 case UIO_SYSSPACE: 549 if (kcopy(p, &tmp, 1)) 550 return; 551 break; 552 } 553 incr = MIN(cnt, PAGESIZE); 554 p += incr; 555 cnt -= incr; 556 } 557 /* 558 * touch the last byte in case it straddles a page. 559 */ 560 p--; 561 switch (uio->uio_segflg) { 562 case UIO_USERSPACE: 563 case UIO_USERISPACE: 564 if (fuword8(p, &tmp)) 565 return; 566 break; 567 case UIO_SYSSPACE: 568 if (kcopy(p, &tmp, 1)) 569 return; 570 break; 571 } 572 iov++; 573 } 574 } 575 576 /* 577 * Write the bytes to a file. 578 * 579 * IN: vp - vnode of file to be written to. 580 * uio - structure supplying write location, range info, 581 * and data buffer. 582 * ioflag - FAPPEND flag set if in append mode. 583 * cr - credentials of caller. 584 * ct - caller context (NFS/CIFS fem monitor only) 585 * 586 * OUT: uio - updated offset and range. 587 * 588 * RETURN: 0 if success 589 * error code if failure 590 * 591 * Timestamps: 592 * vp - ctime|mtime updated if byte count > 0 593 */ 594 /* ARGSUSED */ 595 static int 596 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 597 { 598 znode_t *zp = VTOZ(vp); 599 rlim64_t limit = uio->uio_llimit; 600 ssize_t start_resid = uio->uio_resid; 601 ssize_t tx_bytes; 602 uint64_t end_size; 603 dmu_tx_t *tx; 604 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 605 zilog_t *zilog; 606 offset_t woff; 607 ssize_t n, nbytes; 608 rl_t *rl; 609 int max_blksz = zfsvfs->z_max_blksz; 610 uint64_t pflags = zp->z_phys->zp_flags; 611 int error; 612 613 /* 614 * If immutable or not appending then return EPERM 615 */ 616 if ((pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 617 ((pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 618 (uio->uio_loffset < zp->z_phys->zp_size))) 619 return (EPERM); 620 621 /* 622 * Fasttrack empty write 623 */ 624 n = start_resid; 625 if (n == 0) 626 return (0); 627 628 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 629 limit = MAXOFFSET_T; 630 631 ZFS_ENTER(zfsvfs); 632 ZFS_VERIFY_ZP(zp); 633 zilog = zfsvfs->z_log; 634 635 /* 636 * Pre-fault the pages to ensure slow (eg NFS) pages 637 * don't hold up txg. 638 */ 639 zfs_prefault_write(n, uio); 640 641 /* 642 * If in append mode, set the io offset pointer to eof. 643 */ 644 if (ioflag & FAPPEND) { 645 /* 646 * Range lock for a file append: 647 * The value for the start of range will be determined by 648 * zfs_range_lock() (to guarantee append semantics). 649 * If this write will cause the block size to increase, 650 * zfs_range_lock() will lock the entire file, so we must 651 * later reduce the range after we grow the block size. 652 */ 653 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 654 if (rl->r_len == UINT64_MAX) { 655 /* overlocked, zp_size can't change */ 656 woff = uio->uio_loffset = zp->z_phys->zp_size; 657 } else { 658 woff = uio->uio_loffset = rl->r_off; 659 } 660 } else { 661 woff = uio->uio_loffset; 662 /* 663 * Validate file offset 664 */ 665 if (woff < 0) { 666 ZFS_EXIT(zfsvfs); 667 return (EINVAL); 668 } 669 670 /* 671 * If we need to grow the block size then zfs_range_lock() 672 * will lock a wider range than we request here. 673 * Later after growing the block size we reduce the range. 674 */ 675 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 676 } 677 678 if (woff >= limit) { 679 zfs_range_unlock(rl); 680 ZFS_EXIT(zfsvfs); 681 return (EFBIG); 682 } 683 684 if ((woff + n) > limit || woff > (limit - n)) 685 n = limit - woff; 686 687 /* 688 * Check for mandatory locks 689 */ 690 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 691 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 692 zfs_range_unlock(rl); 693 ZFS_EXIT(zfsvfs); 694 return (error); 695 } 696 end_size = MAX(zp->z_phys->zp_size, woff + n); 697 698 /* 699 * Write the file in reasonable size chunks. Each chunk is written 700 * in a separate transaction; this keeps the intent log records small 701 * and allows us to do more fine-grained space accounting. 702 */ 703 while (n > 0) { 704 /* 705 * Start a transaction. 706 */ 707 woff = uio->uio_loffset; 708 tx = dmu_tx_create(zfsvfs->z_os); 709 dmu_tx_hold_bonus(tx, zp->z_id); 710 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 711 error = dmu_tx_assign(tx, zfsvfs->z_assign); 712 if (error) { 713 if (error == ERESTART && 714 zfsvfs->z_assign == TXG_NOWAIT) { 715 dmu_tx_wait(tx); 716 dmu_tx_abort(tx); 717 continue; 718 } 719 dmu_tx_abort(tx); 720 break; 721 } 722 723 /* 724 * If zfs_range_lock() over-locked we grow the blocksize 725 * and then reduce the lock range. This will only happen 726 * on the first iteration since zfs_range_reduce() will 727 * shrink down r_len to the appropriate size. 728 */ 729 if (rl->r_len == UINT64_MAX) { 730 uint64_t new_blksz; 731 732 if (zp->z_blksz > max_blksz) { 733 ASSERT(!ISP2(zp->z_blksz)); 734 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 735 } else { 736 new_blksz = MIN(end_size, max_blksz); 737 } 738 zfs_grow_blocksize(zp, new_blksz, tx); 739 zfs_range_reduce(rl, woff, n); 740 } 741 742 /* 743 * XXX - should we really limit each write to z_max_blksz? 744 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 745 */ 746 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 747 rw_enter(&zp->z_map_lock, RW_READER); 748 749 tx_bytes = uio->uio_resid; 750 if (vn_has_cached_data(vp)) { 751 rw_exit(&zp->z_map_lock); 752 error = mappedwrite(vp, nbytes, uio, tx); 753 } else { 754 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 755 uio, nbytes, tx); 756 rw_exit(&zp->z_map_lock); 757 } 758 tx_bytes -= uio->uio_resid; 759 760 /* 761 * If we made no progress, we're done. If we made even 762 * partial progress, update the znode and ZIL accordingly. 763 */ 764 if (tx_bytes == 0) { 765 dmu_tx_commit(tx); 766 ASSERT(error != 0); 767 break; 768 } 769 770 /* 771 * Clear Set-UID/Set-GID bits on successful write if not 772 * privileged and at least one of the excute bits is set. 773 * 774 * It would be nice to to this after all writes have 775 * been done, but that would still expose the ISUID/ISGID 776 * to another app after the partial write is committed. 777 * 778 * Note: we don't call zfs_fuid_map_id() here because 779 * user 0 is not an ephemeral uid. 780 */ 781 mutex_enter(&zp->z_acl_lock); 782 if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) | 783 (S_IXUSR >> 6))) != 0 && 784 (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 && 785 secpolicy_vnode_setid_retain(cr, 786 (zp->z_phys->zp_mode & S_ISUID) != 0 && 787 zp->z_phys->zp_uid == 0) != 0) { 788 zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID); 789 } 790 mutex_exit(&zp->z_acl_lock); 791 792 /* 793 * Update time stamp. NOTE: This marks the bonus buffer as 794 * dirty, so we don't have to do it again for zp_size. 795 */ 796 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 797 798 /* 799 * Update the file size (zp_size) if it has changed; 800 * account for possible concurrent updates. 801 */ 802 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) 803 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 804 uio->uio_loffset); 805 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 806 dmu_tx_commit(tx); 807 808 if (error != 0) 809 break; 810 ASSERT(tx_bytes == nbytes); 811 n -= nbytes; 812 } 813 814 zfs_range_unlock(rl); 815 816 /* 817 * If we're in replay mode, or we made no progress, return error. 818 * Otherwise, it's at least a partial write, so it's successful. 819 */ 820 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 821 ZFS_EXIT(zfsvfs); 822 return (error); 823 } 824 825 if (ioflag & (FSYNC | FDSYNC)) 826 zil_commit(zilog, zp->z_last_itx, zp->z_id); 827 828 ZFS_EXIT(zfsvfs); 829 return (0); 830 } 831 832 void 833 zfs_get_done(dmu_buf_t *db, void *vzgd) 834 { 835 zgd_t *zgd = (zgd_t *)vzgd; 836 rl_t *rl = zgd->zgd_rl; 837 vnode_t *vp = ZTOV(rl->r_zp); 838 839 dmu_buf_rele(db, vzgd); 840 zfs_range_unlock(rl); 841 VN_RELE(vp); 842 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 843 kmem_free(zgd, sizeof (zgd_t)); 844 } 845 846 /* 847 * Get data to generate a TX_WRITE intent log record. 848 */ 849 int 850 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 851 { 852 zfsvfs_t *zfsvfs = arg; 853 objset_t *os = zfsvfs->z_os; 854 znode_t *zp; 855 uint64_t off = lr->lr_offset; 856 dmu_buf_t *db; 857 rl_t *rl; 858 zgd_t *zgd; 859 int dlen = lr->lr_length; /* length of user data */ 860 int error = 0; 861 862 ASSERT(zio); 863 ASSERT(dlen != 0); 864 865 /* 866 * Nothing to do if the file has been removed 867 */ 868 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 869 return (ENOENT); 870 if (zp->z_unlinked) { 871 VN_RELE(ZTOV(zp)); 872 return (ENOENT); 873 } 874 875 /* 876 * Write records come in two flavors: immediate and indirect. 877 * For small writes it's cheaper to store the data with the 878 * log record (immediate); for large writes it's cheaper to 879 * sync the data and get a pointer to it (indirect) so that 880 * we don't have to write the data twice. 881 */ 882 if (buf != NULL) { /* immediate write */ 883 rl = zfs_range_lock(zp, off, dlen, RL_READER); 884 /* test for truncation needs to be done while range locked */ 885 if (off >= zp->z_phys->zp_size) { 886 error = ENOENT; 887 goto out; 888 } 889 VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf)); 890 } else { /* indirect write */ 891 uint64_t boff; /* block starting offset */ 892 893 /* 894 * Have to lock the whole block to ensure when it's 895 * written out and it's checksum is being calculated 896 * that no one can change the data. We need to re-check 897 * blocksize after we get the lock in case it's changed! 898 */ 899 for (;;) { 900 if (ISP2(zp->z_blksz)) { 901 boff = P2ALIGN_TYPED(off, zp->z_blksz, 902 uint64_t); 903 } else { 904 boff = 0; 905 } 906 dlen = zp->z_blksz; 907 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 908 if (zp->z_blksz == dlen) 909 break; 910 zfs_range_unlock(rl); 911 } 912 /* test for truncation needs to be done while range locked */ 913 if (off >= zp->z_phys->zp_size) { 914 error = ENOENT; 915 goto out; 916 } 917 zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP); 918 zgd->zgd_rl = rl; 919 zgd->zgd_zilog = zfsvfs->z_log; 920 zgd->zgd_bp = &lr->lr_blkptr; 921 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db)); 922 ASSERT(boff == db->db_offset); 923 lr->lr_blkoff = off - boff; 924 error = dmu_sync(zio, db, &lr->lr_blkptr, 925 lr->lr_common.lrc_txg, zfs_get_done, zgd); 926 ASSERT((error && error != EINPROGRESS) || 927 lr->lr_length <= zp->z_blksz); 928 if (error == 0) 929 zil_add_block(zfsvfs->z_log, &lr->lr_blkptr); 930 /* 931 * If we get EINPROGRESS, then we need to wait for a 932 * write IO initiated by dmu_sync() to complete before 933 * we can release this dbuf. We will finish everything 934 * up in the zfs_get_done() callback. 935 */ 936 if (error == EINPROGRESS) 937 return (0); 938 dmu_buf_rele(db, zgd); 939 kmem_free(zgd, sizeof (zgd_t)); 940 } 941 out: 942 zfs_range_unlock(rl); 943 VN_RELE(ZTOV(zp)); 944 return (error); 945 } 946 947 /*ARGSUSED*/ 948 static int 949 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 950 caller_context_t *ct) 951 { 952 znode_t *zp = VTOZ(vp); 953 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 954 int error; 955 956 ZFS_ENTER(zfsvfs); 957 ZFS_VERIFY_ZP(zp); 958 959 if (flag & V_ACE_MASK) 960 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 961 else 962 error = zfs_zaccess_rwx(zp, mode, flag, cr); 963 964 ZFS_EXIT(zfsvfs); 965 return (error); 966 } 967 968 /* 969 * Lookup an entry in a directory, or an extended attribute directory. 970 * If it exists, return a held vnode reference for it. 971 * 972 * IN: dvp - vnode of directory to search. 973 * nm - name of entry to lookup. 974 * pnp - full pathname to lookup [UNUSED]. 975 * flags - LOOKUP_XATTR set if looking for an attribute. 976 * rdir - root directory vnode [UNUSED]. 977 * cr - credentials of caller. 978 * ct - caller context 979 * direntflags - directory lookup flags 980 * realpnp - returned pathname. 981 * 982 * OUT: vpp - vnode of located entry, NULL if not found. 983 * 984 * RETURN: 0 if success 985 * error code if failure 986 * 987 * Timestamps: 988 * NA 989 */ 990 /* ARGSUSED */ 991 static int 992 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 993 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 994 int *direntflags, pathname_t *realpnp) 995 { 996 znode_t *zdp = VTOZ(dvp); 997 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 998 int error; 999 1000 ZFS_ENTER(zfsvfs); 1001 ZFS_VERIFY_ZP(zdp); 1002 1003 *vpp = NULL; 1004 1005 if (flags & LOOKUP_XATTR) { 1006 /* 1007 * If the xattr property is off, refuse the lookup request. 1008 */ 1009 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1010 ZFS_EXIT(zfsvfs); 1011 return (EINVAL); 1012 } 1013 1014 /* 1015 * We don't allow recursive attributes.. 1016 * Maybe someday we will. 1017 */ 1018 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 1019 ZFS_EXIT(zfsvfs); 1020 return (EINVAL); 1021 } 1022 1023 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1024 ZFS_EXIT(zfsvfs); 1025 return (error); 1026 } 1027 1028 /* 1029 * Do we have permission to get into attribute directory? 1030 */ 1031 1032 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1033 B_FALSE, cr)) { 1034 VN_RELE(*vpp); 1035 *vpp = NULL; 1036 } 1037 1038 ZFS_EXIT(zfsvfs); 1039 return (error); 1040 } 1041 1042 if (dvp->v_type != VDIR) { 1043 ZFS_EXIT(zfsvfs); 1044 return (ENOTDIR); 1045 } 1046 1047 /* 1048 * Check accessibility of directory. 1049 */ 1050 1051 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1052 ZFS_EXIT(zfsvfs); 1053 return (error); 1054 } 1055 1056 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1057 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1058 ZFS_EXIT(zfsvfs); 1059 return (EILSEQ); 1060 } 1061 1062 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1063 if (error == 0) { 1064 /* 1065 * Convert device special files 1066 */ 1067 if (IS_DEVVP(*vpp)) { 1068 vnode_t *svp; 1069 1070 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1071 VN_RELE(*vpp); 1072 if (svp == NULL) 1073 error = ENOSYS; 1074 else 1075 *vpp = svp; 1076 } 1077 } 1078 1079 ZFS_EXIT(zfsvfs); 1080 return (error); 1081 } 1082 1083 /* 1084 * Attempt to create a new entry in a directory. If the entry 1085 * already exists, truncate the file if permissible, else return 1086 * an error. Return the vp of the created or trunc'd file. 1087 * 1088 * IN: dvp - vnode of directory to put new file entry in. 1089 * name - name of new file entry. 1090 * vap - attributes of new file. 1091 * excl - flag indicating exclusive or non-exclusive mode. 1092 * mode - mode to open file with. 1093 * cr - credentials of caller. 1094 * flag - large file flag [UNUSED]. 1095 * ct - caller context 1096 * vsecp - ACL to be set 1097 * 1098 * OUT: vpp - vnode of created or trunc'd entry. 1099 * 1100 * RETURN: 0 if success 1101 * error code if failure 1102 * 1103 * Timestamps: 1104 * dvp - ctime|mtime updated if new entry created 1105 * vp - ctime|mtime always, atime if new 1106 */ 1107 1108 /* ARGSUSED */ 1109 static int 1110 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1111 int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct, 1112 vsecattr_t *vsecp) 1113 { 1114 znode_t *zp, *dzp = VTOZ(dvp); 1115 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1116 zilog_t *zilog; 1117 objset_t *os; 1118 zfs_dirlock_t *dl; 1119 dmu_tx_t *tx; 1120 int error; 1121 zfs_acl_t *aclp = NULL; 1122 zfs_fuid_info_t *fuidp = NULL; 1123 1124 /* 1125 * If we have an ephemeral id, ACL, or XVATTR then 1126 * make sure file system is at proper version 1127 */ 1128 1129 if (zfsvfs->z_use_fuids == B_FALSE && 1130 (vsecp || (vap->va_mask & AT_XVATTR) || 1131 IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr)))) 1132 return (EINVAL); 1133 1134 ZFS_ENTER(zfsvfs); 1135 ZFS_VERIFY_ZP(dzp); 1136 os = zfsvfs->z_os; 1137 zilog = zfsvfs->z_log; 1138 1139 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1140 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1141 ZFS_EXIT(zfsvfs); 1142 return (EILSEQ); 1143 } 1144 1145 if (vap->va_mask & AT_XVATTR) { 1146 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1147 crgetuid(cr), cr, vap->va_type)) != 0) { 1148 ZFS_EXIT(zfsvfs); 1149 return (error); 1150 } 1151 } 1152 top: 1153 *vpp = NULL; 1154 1155 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1156 vap->va_mode &= ~VSVTX; 1157 1158 if (*name == '\0') { 1159 /* 1160 * Null component name refers to the directory itself. 1161 */ 1162 VN_HOLD(dvp); 1163 zp = dzp; 1164 dl = NULL; 1165 error = 0; 1166 } else { 1167 /* possible VN_HOLD(zp) */ 1168 int zflg = 0; 1169 1170 if (flag & FIGNORECASE) 1171 zflg |= ZCILOOK; 1172 1173 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1174 NULL, NULL); 1175 if (error) { 1176 if (strcmp(name, "..") == 0) 1177 error = EISDIR; 1178 ZFS_EXIT(zfsvfs); 1179 if (aclp) 1180 zfs_acl_free(aclp); 1181 return (error); 1182 } 1183 } 1184 if (vsecp && aclp == NULL) { 1185 error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp); 1186 if (error) { 1187 ZFS_EXIT(zfsvfs); 1188 if (dl) 1189 zfs_dirent_unlock(dl); 1190 return (error); 1191 } 1192 } 1193 1194 if (zp == NULL) { 1195 uint64_t txtype; 1196 1197 /* 1198 * Create a new file object and update the directory 1199 * to reference it. 1200 */ 1201 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1202 goto out; 1203 } 1204 1205 /* 1206 * We only support the creation of regular files in 1207 * extended attribute directories. 1208 */ 1209 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1210 (vap->va_type != VREG)) { 1211 error = EINVAL; 1212 goto out; 1213 } 1214 1215 tx = dmu_tx_create(os); 1216 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1217 if (zfsvfs->z_fuid_obj == 0) { 1218 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1219 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1220 SPA_MAXBLOCKSIZE); 1221 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 1222 } else { 1223 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 1224 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 1225 SPA_MAXBLOCKSIZE); 1226 } 1227 dmu_tx_hold_bonus(tx, dzp->z_id); 1228 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1229 if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp) { 1230 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1231 0, SPA_MAXBLOCKSIZE); 1232 } 1233 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1234 if (error) { 1235 zfs_dirent_unlock(dl); 1236 if (error == ERESTART && 1237 zfsvfs->z_assign == TXG_NOWAIT) { 1238 dmu_tx_wait(tx); 1239 dmu_tx_abort(tx); 1240 goto top; 1241 } 1242 dmu_tx_abort(tx); 1243 ZFS_EXIT(zfsvfs); 1244 if (aclp) 1245 zfs_acl_free(aclp); 1246 return (error); 1247 } 1248 zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp); 1249 (void) zfs_link_create(dl, zp, tx, ZNEW); 1250 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1251 if (flag & FIGNORECASE) 1252 txtype |= TX_CI; 1253 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1254 vsecp, fuidp, vap); 1255 if (fuidp) 1256 zfs_fuid_info_free(fuidp); 1257 dmu_tx_commit(tx); 1258 } else { 1259 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1260 1261 /* 1262 * A directory entry already exists for this name. 1263 */ 1264 /* 1265 * Can't truncate an existing file if in exclusive mode. 1266 */ 1267 if (excl == EXCL) { 1268 error = EEXIST; 1269 goto out; 1270 } 1271 /* 1272 * Can't open a directory for writing. 1273 */ 1274 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1275 error = EISDIR; 1276 goto out; 1277 } 1278 /* 1279 * Verify requested access to file. 1280 */ 1281 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1282 goto out; 1283 } 1284 1285 mutex_enter(&dzp->z_lock); 1286 dzp->z_seq++; 1287 mutex_exit(&dzp->z_lock); 1288 1289 /* 1290 * Truncate regular files if requested. 1291 */ 1292 if ((ZTOV(zp)->v_type == VREG) && 1293 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1294 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1295 if (error == ERESTART && 1296 zfsvfs->z_assign == TXG_NOWAIT) { 1297 /* NB: we already did dmu_tx_wait() */ 1298 zfs_dirent_unlock(dl); 1299 VN_RELE(ZTOV(zp)); 1300 goto top; 1301 } 1302 1303 if (error == 0) { 1304 vnevent_create(ZTOV(zp), ct); 1305 } 1306 } 1307 } 1308 out: 1309 1310 if (dl) 1311 zfs_dirent_unlock(dl); 1312 1313 if (error) { 1314 if (zp) 1315 VN_RELE(ZTOV(zp)); 1316 } else { 1317 *vpp = ZTOV(zp); 1318 /* 1319 * If vnode is for a device return a specfs vnode instead. 1320 */ 1321 if (IS_DEVVP(*vpp)) { 1322 struct vnode *svp; 1323 1324 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1325 VN_RELE(*vpp); 1326 if (svp == NULL) { 1327 error = ENOSYS; 1328 } 1329 *vpp = svp; 1330 } 1331 } 1332 if (aclp) 1333 zfs_acl_free(aclp); 1334 1335 ZFS_EXIT(zfsvfs); 1336 return (error); 1337 } 1338 1339 /* 1340 * Remove an entry from a directory. 1341 * 1342 * IN: dvp - vnode of directory to remove entry from. 1343 * name - name of entry to remove. 1344 * cr - credentials of caller. 1345 * ct - caller context 1346 * flags - case flags 1347 * 1348 * RETURN: 0 if success 1349 * error code if failure 1350 * 1351 * Timestamps: 1352 * dvp - ctime|mtime 1353 * vp - ctime (if nlink > 0) 1354 */ 1355 /*ARGSUSED*/ 1356 static int 1357 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1358 int flags) 1359 { 1360 znode_t *zp, *dzp = VTOZ(dvp); 1361 znode_t *xzp = NULL; 1362 vnode_t *vp; 1363 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1364 zilog_t *zilog; 1365 uint64_t acl_obj, xattr_obj; 1366 zfs_dirlock_t *dl; 1367 dmu_tx_t *tx; 1368 boolean_t may_delete_now, delete_now = FALSE; 1369 boolean_t unlinked; 1370 uint64_t txtype; 1371 pathname_t *realnmp = NULL; 1372 pathname_t realnm; 1373 int error; 1374 int zflg = ZEXISTS; 1375 1376 ZFS_ENTER(zfsvfs); 1377 ZFS_VERIFY_ZP(dzp); 1378 zilog = zfsvfs->z_log; 1379 1380 if (flags & FIGNORECASE) { 1381 zflg |= ZCILOOK; 1382 pn_alloc(&realnm); 1383 realnmp = &realnm; 1384 } 1385 1386 top: 1387 /* 1388 * Attempt to lock directory; fail if entry doesn't exist. 1389 */ 1390 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1391 NULL, realnmp)) { 1392 if (realnmp) 1393 pn_free(realnmp); 1394 ZFS_EXIT(zfsvfs); 1395 return (error); 1396 } 1397 1398 vp = ZTOV(zp); 1399 1400 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1401 goto out; 1402 } 1403 1404 /* 1405 * Need to use rmdir for removing directories. 1406 */ 1407 if (vp->v_type == VDIR) { 1408 error = EPERM; 1409 goto out; 1410 } 1411 1412 vnevent_remove(vp, dvp, name, ct); 1413 1414 if (realnmp) 1415 dnlc_remove(dvp, realnmp->pn_path); 1416 else 1417 dnlc_remove(dvp, name); 1418 1419 mutex_enter(&vp->v_lock); 1420 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1421 mutex_exit(&vp->v_lock); 1422 1423 /* 1424 * We may delete the znode now, or we may put it in the unlinked set; 1425 * it depends on whether we're the last link, and on whether there are 1426 * other holds on the vnode. So we dmu_tx_hold() the right things to 1427 * allow for either case. 1428 */ 1429 tx = dmu_tx_create(zfsvfs->z_os); 1430 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1431 dmu_tx_hold_bonus(tx, zp->z_id); 1432 if (may_delete_now) 1433 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 1434 1435 /* are there any extended attributes? */ 1436 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1437 /* XXX - do we need this if we are deleting? */ 1438 dmu_tx_hold_bonus(tx, xattr_obj); 1439 } 1440 1441 /* are there any additional acls */ 1442 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1443 may_delete_now) 1444 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1445 1446 /* charge as an update -- would be nice not to charge at all */ 1447 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1448 1449 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1450 if (error) { 1451 zfs_dirent_unlock(dl); 1452 VN_RELE(vp); 1453 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1454 dmu_tx_wait(tx); 1455 dmu_tx_abort(tx); 1456 goto top; 1457 } 1458 if (realnmp) 1459 pn_free(realnmp); 1460 dmu_tx_abort(tx); 1461 ZFS_EXIT(zfsvfs); 1462 return (error); 1463 } 1464 1465 /* 1466 * Remove the directory entry. 1467 */ 1468 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1469 1470 if (error) { 1471 dmu_tx_commit(tx); 1472 goto out; 1473 } 1474 1475 if (unlinked) { 1476 mutex_enter(&vp->v_lock); 1477 delete_now = may_delete_now && 1478 vp->v_count == 1 && !vn_has_cached_data(vp) && 1479 zp->z_phys->zp_xattr == xattr_obj && 1480 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1481 mutex_exit(&vp->v_lock); 1482 } 1483 1484 if (delete_now) { 1485 if (zp->z_phys->zp_xattr) { 1486 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1487 ASSERT3U(error, ==, 0); 1488 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1489 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1490 mutex_enter(&xzp->z_lock); 1491 xzp->z_unlinked = 1; 1492 xzp->z_phys->zp_links = 0; 1493 mutex_exit(&xzp->z_lock); 1494 zfs_unlinked_add(xzp, tx); 1495 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1496 } 1497 mutex_enter(&zp->z_lock); 1498 mutex_enter(&vp->v_lock); 1499 vp->v_count--; 1500 ASSERT3U(vp->v_count, ==, 0); 1501 mutex_exit(&vp->v_lock); 1502 mutex_exit(&zp->z_lock); 1503 zfs_znode_delete(zp, tx); 1504 } else if (unlinked) { 1505 zfs_unlinked_add(zp, tx); 1506 } 1507 1508 txtype = TX_REMOVE; 1509 if (flags & FIGNORECASE) 1510 txtype |= TX_CI; 1511 zfs_log_remove(zilog, tx, txtype, dzp, name); 1512 1513 dmu_tx_commit(tx); 1514 out: 1515 if (realnmp) 1516 pn_free(realnmp); 1517 1518 zfs_dirent_unlock(dl); 1519 1520 if (!delete_now) { 1521 VN_RELE(vp); 1522 } else if (xzp) { 1523 /* this rele delayed to prevent nesting transactions */ 1524 VN_RELE(ZTOV(xzp)); 1525 } 1526 1527 ZFS_EXIT(zfsvfs); 1528 return (error); 1529 } 1530 1531 /* 1532 * Create a new directory and insert it into dvp using the name 1533 * provided. Return a pointer to the inserted directory. 1534 * 1535 * IN: dvp - vnode of directory to add subdir to. 1536 * dirname - name of new directory. 1537 * vap - attributes of new directory. 1538 * cr - credentials of caller. 1539 * ct - caller context 1540 * vsecp - ACL to be set 1541 * 1542 * OUT: vpp - vnode of created directory. 1543 * 1544 * RETURN: 0 if success 1545 * error code if failure 1546 * 1547 * Timestamps: 1548 * dvp - ctime|mtime updated 1549 * vp - ctime|mtime|atime updated 1550 */ 1551 /*ARGSUSED*/ 1552 static int 1553 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 1554 caller_context_t *ct, int flags, vsecattr_t *vsecp) 1555 { 1556 znode_t *zp, *dzp = VTOZ(dvp); 1557 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1558 zilog_t *zilog; 1559 zfs_dirlock_t *dl; 1560 uint64_t txtype; 1561 dmu_tx_t *tx; 1562 int error; 1563 zfs_acl_t *aclp = NULL; 1564 zfs_fuid_info_t *fuidp = NULL; 1565 int zf = ZNEW; 1566 1567 ASSERT(vap->va_type == VDIR); 1568 1569 /* 1570 * If we have an ephemeral id, ACL, or XVATTR then 1571 * make sure file system is at proper version 1572 */ 1573 1574 if (zfsvfs->z_use_fuids == B_FALSE && 1575 (vsecp || (vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(crgetuid(cr))|| 1576 IS_EPHEMERAL(crgetgid(cr)))) 1577 return (EINVAL); 1578 1579 ZFS_ENTER(zfsvfs); 1580 ZFS_VERIFY_ZP(dzp); 1581 zilog = zfsvfs->z_log; 1582 1583 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1584 ZFS_EXIT(zfsvfs); 1585 return (EINVAL); 1586 } 1587 1588 if (zfsvfs->z_utf8 && u8_validate(dirname, 1589 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1590 ZFS_EXIT(zfsvfs); 1591 return (EILSEQ); 1592 } 1593 if (flags & FIGNORECASE) 1594 zf |= ZCILOOK; 1595 1596 if (vap->va_mask & AT_XVATTR) 1597 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1598 crgetuid(cr), cr, vap->va_type)) != 0) { 1599 ZFS_EXIT(zfsvfs); 1600 return (error); 1601 } 1602 1603 /* 1604 * First make sure the new directory doesn't exist. 1605 */ 1606 top: 1607 *vpp = NULL; 1608 1609 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 1610 NULL, NULL)) { 1611 ZFS_EXIT(zfsvfs); 1612 return (error); 1613 } 1614 1615 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 1616 zfs_dirent_unlock(dl); 1617 ZFS_EXIT(zfsvfs); 1618 return (error); 1619 } 1620 1621 if (vsecp && aclp == NULL) { 1622 error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp); 1623 if (error) { 1624 zfs_dirent_unlock(dl); 1625 ZFS_EXIT(zfsvfs); 1626 return (error); 1627 } 1628 } 1629 /* 1630 * Add a new entry to the directory. 1631 */ 1632 tx = dmu_tx_create(zfsvfs->z_os); 1633 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1634 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1635 if (zfsvfs->z_fuid_obj == 0) { 1636 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1637 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1638 SPA_MAXBLOCKSIZE); 1639 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 1640 } else { 1641 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 1642 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 1643 SPA_MAXBLOCKSIZE); 1644 } 1645 if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp) 1646 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1647 0, SPA_MAXBLOCKSIZE); 1648 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1649 if (error) { 1650 zfs_dirent_unlock(dl); 1651 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1652 dmu_tx_wait(tx); 1653 dmu_tx_abort(tx); 1654 goto top; 1655 } 1656 dmu_tx_abort(tx); 1657 ZFS_EXIT(zfsvfs); 1658 if (aclp) 1659 zfs_acl_free(aclp); 1660 return (error); 1661 } 1662 1663 /* 1664 * Create new node. 1665 */ 1666 zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp); 1667 1668 if (aclp) 1669 zfs_acl_free(aclp); 1670 1671 /* 1672 * Now put new name in parent dir. 1673 */ 1674 (void) zfs_link_create(dl, zp, tx, ZNEW); 1675 1676 *vpp = ZTOV(zp); 1677 1678 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 1679 if (flags & FIGNORECASE) 1680 txtype |= TX_CI; 1681 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, fuidp, vap); 1682 1683 if (fuidp) 1684 zfs_fuid_info_free(fuidp); 1685 dmu_tx_commit(tx); 1686 1687 zfs_dirent_unlock(dl); 1688 1689 ZFS_EXIT(zfsvfs); 1690 return (0); 1691 } 1692 1693 /* 1694 * Remove a directory subdir entry. If the current working 1695 * directory is the same as the subdir to be removed, the 1696 * remove will fail. 1697 * 1698 * IN: dvp - vnode of directory to remove from. 1699 * name - name of directory to be removed. 1700 * cwd - vnode of current working directory. 1701 * cr - credentials of caller. 1702 * ct - caller context 1703 * flags - case flags 1704 * 1705 * RETURN: 0 if success 1706 * error code if failure 1707 * 1708 * Timestamps: 1709 * dvp - ctime|mtime updated 1710 */ 1711 /*ARGSUSED*/ 1712 static int 1713 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 1714 caller_context_t *ct, int flags) 1715 { 1716 znode_t *dzp = VTOZ(dvp); 1717 znode_t *zp; 1718 vnode_t *vp; 1719 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1720 zilog_t *zilog; 1721 zfs_dirlock_t *dl; 1722 dmu_tx_t *tx; 1723 int error; 1724 int zflg = ZEXISTS; 1725 1726 ZFS_ENTER(zfsvfs); 1727 ZFS_VERIFY_ZP(dzp); 1728 zilog = zfsvfs->z_log; 1729 1730 if (flags & FIGNORECASE) 1731 zflg |= ZCILOOK; 1732 top: 1733 zp = NULL; 1734 1735 /* 1736 * Attempt to lock directory; fail if entry doesn't exist. 1737 */ 1738 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1739 NULL, NULL)) { 1740 ZFS_EXIT(zfsvfs); 1741 return (error); 1742 } 1743 1744 vp = ZTOV(zp); 1745 1746 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1747 goto out; 1748 } 1749 1750 if (vp->v_type != VDIR) { 1751 error = ENOTDIR; 1752 goto out; 1753 } 1754 1755 if (vp == cwd) { 1756 error = EINVAL; 1757 goto out; 1758 } 1759 1760 vnevent_rmdir(vp, dvp, name, ct); 1761 1762 /* 1763 * Grab a lock on the directory to make sure that noone is 1764 * trying to add (or lookup) entries while we are removing it. 1765 */ 1766 rw_enter(&zp->z_name_lock, RW_WRITER); 1767 1768 /* 1769 * Grab a lock on the parent pointer to make sure we play well 1770 * with the treewalk and directory rename code. 1771 */ 1772 rw_enter(&zp->z_parent_lock, RW_WRITER); 1773 1774 tx = dmu_tx_create(zfsvfs->z_os); 1775 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1776 dmu_tx_hold_bonus(tx, zp->z_id); 1777 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1778 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1779 if (error) { 1780 rw_exit(&zp->z_parent_lock); 1781 rw_exit(&zp->z_name_lock); 1782 zfs_dirent_unlock(dl); 1783 VN_RELE(vp); 1784 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1785 dmu_tx_wait(tx); 1786 dmu_tx_abort(tx); 1787 goto top; 1788 } 1789 dmu_tx_abort(tx); 1790 ZFS_EXIT(zfsvfs); 1791 return (error); 1792 } 1793 1794 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 1795 1796 if (error == 0) { 1797 uint64_t txtype = TX_RMDIR; 1798 if (flags & FIGNORECASE) 1799 txtype |= TX_CI; 1800 zfs_log_remove(zilog, tx, txtype, dzp, name); 1801 } 1802 1803 dmu_tx_commit(tx); 1804 1805 rw_exit(&zp->z_parent_lock); 1806 rw_exit(&zp->z_name_lock); 1807 out: 1808 zfs_dirent_unlock(dl); 1809 1810 VN_RELE(vp); 1811 1812 ZFS_EXIT(zfsvfs); 1813 return (error); 1814 } 1815 1816 /* 1817 * Read as many directory entries as will fit into the provided 1818 * buffer from the given directory cursor position (specified in 1819 * the uio structure. 1820 * 1821 * IN: vp - vnode of directory to read. 1822 * uio - structure supplying read location, range info, 1823 * and return buffer. 1824 * cr - credentials of caller. 1825 * ct - caller context 1826 * flags - case flags 1827 * 1828 * OUT: uio - updated offset and range, buffer filled. 1829 * eofp - set to true if end-of-file detected. 1830 * 1831 * RETURN: 0 if success 1832 * error code if failure 1833 * 1834 * Timestamps: 1835 * vp - atime updated 1836 * 1837 * Note that the low 4 bits of the cookie returned by zap is always zero. 1838 * This allows us to use the low range for "special" directory entries: 1839 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1840 * we use the offset 2 for the '.zfs' directory. 1841 */ 1842 /* ARGSUSED */ 1843 static int 1844 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, 1845 caller_context_t *ct, int flags) 1846 { 1847 znode_t *zp = VTOZ(vp); 1848 iovec_t *iovp; 1849 edirent_t *eodp; 1850 dirent64_t *odp; 1851 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1852 objset_t *os; 1853 caddr_t outbuf; 1854 size_t bufsize; 1855 zap_cursor_t zc; 1856 zap_attribute_t zap; 1857 uint_t bytes_wanted; 1858 uint64_t offset; /* must be unsigned; checks for < 1 */ 1859 int local_eof; 1860 int outcount; 1861 int error; 1862 uint8_t prefetch; 1863 boolean_t check_sysattrs; 1864 1865 ZFS_ENTER(zfsvfs); 1866 ZFS_VERIFY_ZP(zp); 1867 1868 /* 1869 * If we are not given an eof variable, 1870 * use a local one. 1871 */ 1872 if (eofp == NULL) 1873 eofp = &local_eof; 1874 1875 /* 1876 * Check for valid iov_len. 1877 */ 1878 if (uio->uio_iov->iov_len <= 0) { 1879 ZFS_EXIT(zfsvfs); 1880 return (EINVAL); 1881 } 1882 1883 /* 1884 * Quit if directory has been removed (posix) 1885 */ 1886 if ((*eofp = zp->z_unlinked) != 0) { 1887 ZFS_EXIT(zfsvfs); 1888 return (0); 1889 } 1890 1891 error = 0; 1892 os = zfsvfs->z_os; 1893 offset = uio->uio_loffset; 1894 prefetch = zp->z_zn_prefetch; 1895 1896 /* 1897 * Initialize the iterator cursor. 1898 */ 1899 if (offset <= 3) { 1900 /* 1901 * Start iteration from the beginning of the directory. 1902 */ 1903 zap_cursor_init(&zc, os, zp->z_id); 1904 } else { 1905 /* 1906 * The offset is a serialized cursor. 1907 */ 1908 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1909 } 1910 1911 /* 1912 * Get space to change directory entries into fs independent format. 1913 */ 1914 iovp = uio->uio_iov; 1915 bytes_wanted = iovp->iov_len; 1916 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1917 bufsize = bytes_wanted; 1918 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1919 odp = (struct dirent64 *)outbuf; 1920 } else { 1921 bufsize = bytes_wanted; 1922 odp = (struct dirent64 *)iovp->iov_base; 1923 } 1924 eodp = (struct edirent *)odp; 1925 1926 /* 1927 * If this VFS supports system attributes; and we're looking at an 1928 * extended attribute directory; and we care about normalization 1929 * conflicts on this vfs; then we must check for normalization 1930 * conflicts with the sysattr name space. 1931 */ 1932 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) && 1933 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 1934 (flags & V_RDDIR_ENTFLAGS); 1935 1936 /* 1937 * Transform to file-system independent format 1938 */ 1939 outcount = 0; 1940 while (outcount < bytes_wanted) { 1941 ino64_t objnum; 1942 ushort_t reclen; 1943 off64_t *next; 1944 1945 /* 1946 * Special case `.', `..', and `.zfs'. 1947 */ 1948 if (offset == 0) { 1949 (void) strcpy(zap.za_name, "."); 1950 zap.za_normalization_conflict = 0; 1951 objnum = zp->z_id; 1952 } else if (offset == 1) { 1953 (void) strcpy(zap.za_name, ".."); 1954 zap.za_normalization_conflict = 0; 1955 objnum = zp->z_phys->zp_parent; 1956 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1957 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1958 zap.za_normalization_conflict = 0; 1959 objnum = ZFSCTL_INO_ROOT; 1960 } else { 1961 /* 1962 * Grab next entry. 1963 */ 1964 if (error = zap_cursor_retrieve(&zc, &zap)) { 1965 if ((*eofp = (error == ENOENT)) != 0) 1966 break; 1967 else 1968 goto update; 1969 } 1970 1971 if (zap.za_integer_length != 8 || 1972 zap.za_num_integers != 1) { 1973 cmn_err(CE_WARN, "zap_readdir: bad directory " 1974 "entry, obj = %lld, offset = %lld\n", 1975 (u_longlong_t)zp->z_id, 1976 (u_longlong_t)offset); 1977 error = ENXIO; 1978 goto update; 1979 } 1980 1981 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 1982 /* 1983 * MacOS X can extract the object type here such as: 1984 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1985 */ 1986 1987 if (check_sysattrs && !zap.za_normalization_conflict) { 1988 zap.za_normalization_conflict = 1989 xattr_sysattr_casechk(zap.za_name); 1990 } 1991 } 1992 1993 if (flags & V_RDDIR_ENTFLAGS) 1994 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 1995 else 1996 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1997 1998 /* 1999 * Will this entry fit in the buffer? 2000 */ 2001 if (outcount + reclen > bufsize) { 2002 /* 2003 * Did we manage to fit anything in the buffer? 2004 */ 2005 if (!outcount) { 2006 error = EINVAL; 2007 goto update; 2008 } 2009 break; 2010 } 2011 if (flags & V_RDDIR_ENTFLAGS) { 2012 /* 2013 * Add extended flag entry: 2014 */ 2015 eodp->ed_ino = objnum; 2016 eodp->ed_reclen = reclen; 2017 /* NOTE: ed_off is the offset for the *next* entry */ 2018 next = &(eodp->ed_off); 2019 eodp->ed_eflags = zap.za_normalization_conflict ? 2020 ED_CASE_CONFLICT : 0; 2021 (void) strncpy(eodp->ed_name, zap.za_name, 2022 EDIRENT_NAMELEN(reclen)); 2023 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2024 } else { 2025 /* 2026 * Add normal entry: 2027 */ 2028 odp->d_ino = objnum; 2029 odp->d_reclen = reclen; 2030 /* NOTE: d_off is the offset for the *next* entry */ 2031 next = &(odp->d_off); 2032 (void) strncpy(odp->d_name, zap.za_name, 2033 DIRENT64_NAMELEN(reclen)); 2034 odp = (dirent64_t *)((intptr_t)odp + reclen); 2035 } 2036 outcount += reclen; 2037 2038 ASSERT(outcount <= bufsize); 2039 2040 /* Prefetch znode */ 2041 if (prefetch) 2042 dmu_prefetch(os, objnum, 0, 0); 2043 2044 /* 2045 * Move to the next entry, fill in the previous offset. 2046 */ 2047 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2048 zap_cursor_advance(&zc); 2049 offset = zap_cursor_serialize(&zc); 2050 } else { 2051 offset += 1; 2052 } 2053 *next = offset; 2054 } 2055 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2056 2057 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2058 iovp->iov_base += outcount; 2059 iovp->iov_len -= outcount; 2060 uio->uio_resid -= outcount; 2061 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2062 /* 2063 * Reset the pointer. 2064 */ 2065 offset = uio->uio_loffset; 2066 } 2067 2068 update: 2069 zap_cursor_fini(&zc); 2070 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2071 kmem_free(outbuf, bufsize); 2072 2073 if (error == ENOENT) 2074 error = 0; 2075 2076 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2077 2078 uio->uio_loffset = offset; 2079 ZFS_EXIT(zfsvfs); 2080 return (error); 2081 } 2082 2083 ulong_t zfs_fsync_sync_cnt = 4; 2084 2085 static int 2086 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2087 { 2088 znode_t *zp = VTOZ(vp); 2089 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2090 2091 /* 2092 * Regardless of whether this is required for standards conformance, 2093 * this is the logical behavior when fsync() is called on a file with 2094 * dirty pages. We use B_ASYNC since the ZIL transactions are already 2095 * going to be pushed out as part of the zil_commit(). 2096 */ 2097 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 2098 (vp->v_type == VREG) && !(IS_SWAPVP(vp))) 2099 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct); 2100 2101 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2102 2103 ZFS_ENTER(zfsvfs); 2104 ZFS_VERIFY_ZP(zp); 2105 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 2106 ZFS_EXIT(zfsvfs); 2107 return (0); 2108 } 2109 2110 2111 /* 2112 * Get the requested file attributes and place them in the provided 2113 * vattr structure. 2114 * 2115 * IN: vp - vnode of file. 2116 * vap - va_mask identifies requested attributes. 2117 * If AT_XVATTR set, then optional attrs are requested 2118 * flags - ATTR_NOACLCHECK (CIFS server context) 2119 * cr - credentials of caller. 2120 * ct - caller context 2121 * 2122 * OUT: vap - attribute values. 2123 * 2124 * RETURN: 0 (always succeeds) 2125 */ 2126 /* ARGSUSED */ 2127 static int 2128 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2129 caller_context_t *ct) 2130 { 2131 znode_t *zp = VTOZ(vp); 2132 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2133 znode_phys_t *pzp; 2134 int error = 0; 2135 uint64_t links; 2136 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2137 xoptattr_t *xoap = NULL; 2138 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2139 2140 ZFS_ENTER(zfsvfs); 2141 ZFS_VERIFY_ZP(zp); 2142 pzp = zp->z_phys; 2143 2144 mutex_enter(&zp->z_lock); 2145 2146 /* 2147 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2148 * Also, if we are the owner don't bother, since owner should 2149 * always be allowed to read basic attributes of file. 2150 */ 2151 if (!(pzp->zp_flags & ZFS_ACL_TRIVIAL) && 2152 (pzp->zp_uid != crgetuid(cr))) { 2153 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2154 skipaclchk, cr)) { 2155 mutex_exit(&zp->z_lock); 2156 ZFS_EXIT(zfsvfs); 2157 return (error); 2158 } 2159 } 2160 2161 /* 2162 * Return all attributes. It's cheaper to provide the answer 2163 * than to determine whether we were asked the question. 2164 */ 2165 2166 vap->va_type = vp->v_type; 2167 vap->va_mode = pzp->zp_mode & MODEMASK; 2168 zfs_fuid_map_ids(zp, &vap->va_uid, &vap->va_gid); 2169 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2170 vap->va_nodeid = zp->z_id; 2171 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2172 links = pzp->zp_links + 1; 2173 else 2174 links = pzp->zp_links; 2175 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */ 2176 vap->va_size = pzp->zp_size; 2177 vap->va_rdev = vp->v_rdev; 2178 vap->va_seq = zp->z_seq; 2179 2180 /* 2181 * Add in any requested optional attributes and the create time. 2182 * Also set the corresponding bits in the returned attribute bitmap. 2183 */ 2184 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2185 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2186 xoap->xoa_archive = 2187 ((pzp->zp_flags & ZFS_ARCHIVE) != 0); 2188 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2189 } 2190 2191 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2192 xoap->xoa_readonly = 2193 ((pzp->zp_flags & ZFS_READONLY) != 0); 2194 XVA_SET_RTN(xvap, XAT_READONLY); 2195 } 2196 2197 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2198 xoap->xoa_system = 2199 ((pzp->zp_flags & ZFS_SYSTEM) != 0); 2200 XVA_SET_RTN(xvap, XAT_SYSTEM); 2201 } 2202 2203 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2204 xoap->xoa_hidden = 2205 ((pzp->zp_flags & ZFS_HIDDEN) != 0); 2206 XVA_SET_RTN(xvap, XAT_HIDDEN); 2207 } 2208 2209 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2210 xoap->xoa_nounlink = 2211 ((pzp->zp_flags & ZFS_NOUNLINK) != 0); 2212 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2213 } 2214 2215 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2216 xoap->xoa_immutable = 2217 ((pzp->zp_flags & ZFS_IMMUTABLE) != 0); 2218 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2219 } 2220 2221 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2222 xoap->xoa_appendonly = 2223 ((pzp->zp_flags & ZFS_APPENDONLY) != 0); 2224 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2225 } 2226 2227 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2228 xoap->xoa_nodump = 2229 ((pzp->zp_flags & ZFS_NODUMP) != 0); 2230 XVA_SET_RTN(xvap, XAT_NODUMP); 2231 } 2232 2233 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2234 xoap->xoa_opaque = 2235 ((pzp->zp_flags & ZFS_OPAQUE) != 0); 2236 XVA_SET_RTN(xvap, XAT_OPAQUE); 2237 } 2238 2239 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2240 xoap->xoa_av_quarantined = 2241 ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0); 2242 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2243 } 2244 2245 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2246 xoap->xoa_av_modified = 2247 ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0); 2248 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2249 } 2250 2251 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2252 vp->v_type == VREG && 2253 (pzp->zp_flags & ZFS_BONUS_SCANSTAMP)) { 2254 size_t len; 2255 dmu_object_info_t doi; 2256 2257 /* 2258 * Only VREG files have anti-virus scanstamps, so we 2259 * won't conflict with symlinks in the bonus buffer. 2260 */ 2261 dmu_object_info_from_db(zp->z_dbuf, &doi); 2262 len = sizeof (xoap->xoa_av_scanstamp) + 2263 sizeof (znode_phys_t); 2264 if (len <= doi.doi_bonus_size) { 2265 /* 2266 * pzp points to the start of the 2267 * znode_phys_t. pzp + 1 points to the 2268 * first byte after the znode_phys_t. 2269 */ 2270 (void) memcpy(xoap->xoa_av_scanstamp, 2271 pzp + 1, 2272 sizeof (xoap->xoa_av_scanstamp)); 2273 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 2274 } 2275 } 2276 2277 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2278 ZFS_TIME_DECODE(&xoap->xoa_createtime, pzp->zp_crtime); 2279 XVA_SET_RTN(xvap, XAT_CREATETIME); 2280 } 2281 } 2282 2283 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 2284 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 2285 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 2286 2287 mutex_exit(&zp->z_lock); 2288 2289 dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks); 2290 2291 if (zp->z_blksz == 0) { 2292 /* 2293 * Block size hasn't been set; suggest maximal I/O transfers. 2294 */ 2295 vap->va_blksize = zfsvfs->z_max_blksz; 2296 } 2297 2298 ZFS_EXIT(zfsvfs); 2299 return (0); 2300 } 2301 2302 /* 2303 * Set the file attributes to the values contained in the 2304 * vattr structure. 2305 * 2306 * IN: vp - vnode of file to be modified. 2307 * vap - new attribute values. 2308 * If AT_XVATTR set, then optional attrs are being set 2309 * flags - ATTR_UTIME set if non-default time values provided. 2310 * - ATTR_NOACLCHECK (CIFS context only). 2311 * cr - credentials of caller. 2312 * ct - caller context 2313 * 2314 * RETURN: 0 if success 2315 * error code if failure 2316 * 2317 * Timestamps: 2318 * vp - ctime updated, mtime updated if size changed. 2319 */ 2320 /* ARGSUSED */ 2321 static int 2322 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2323 caller_context_t *ct) 2324 { 2325 znode_t *zp = VTOZ(vp); 2326 znode_phys_t *pzp; 2327 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2328 zilog_t *zilog; 2329 dmu_tx_t *tx; 2330 vattr_t oldva; 2331 uint_t mask = vap->va_mask; 2332 uint_t saved_mask; 2333 int trim_mask = 0; 2334 uint64_t new_mode; 2335 znode_t *attrzp; 2336 int need_policy = FALSE; 2337 int err; 2338 zfs_fuid_info_t *fuidp = NULL; 2339 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2340 xoptattr_t *xoap; 2341 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2342 2343 if (mask == 0) 2344 return (0); 2345 2346 if (mask & AT_NOSET) 2347 return (EINVAL); 2348 2349 ZFS_ENTER(zfsvfs); 2350 ZFS_VERIFY_ZP(zp); 2351 2352 pzp = zp->z_phys; 2353 zilog = zfsvfs->z_log; 2354 2355 /* 2356 * Make sure that if we have ephemeral uid/gid or xvattr specified 2357 * that file system is at proper version level 2358 */ 2359 2360 if (zfsvfs->z_use_fuids == B_FALSE && 2361 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2362 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2363 (mask & AT_XVATTR))) { 2364 ZFS_EXIT(zfsvfs); 2365 return (EINVAL); 2366 } 2367 2368 if (mask & AT_SIZE && vp->v_type == VDIR) { 2369 ZFS_EXIT(zfsvfs); 2370 return (EISDIR); 2371 } 2372 2373 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 2374 ZFS_EXIT(zfsvfs); 2375 return (EINVAL); 2376 } 2377 2378 /* 2379 * If this is an xvattr_t, then get a pointer to the structure of 2380 * optional attributes. If this is NULL, then we have a vattr_t. 2381 */ 2382 xoap = xva_getxoptattr(xvap); 2383 2384 /* 2385 * Immutable files can only alter immutable bit and atime 2386 */ 2387 if ((pzp->zp_flags & ZFS_IMMUTABLE) && 2388 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2389 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 2390 ZFS_EXIT(zfsvfs); 2391 return (EPERM); 2392 } 2393 2394 if ((mask & AT_SIZE) && (pzp->zp_flags & ZFS_READONLY)) { 2395 ZFS_EXIT(zfsvfs); 2396 return (EPERM); 2397 } 2398 2399 top: 2400 attrzp = NULL; 2401 2402 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2403 ZFS_EXIT(zfsvfs); 2404 return (EROFS); 2405 } 2406 2407 /* 2408 * First validate permissions 2409 */ 2410 2411 if (mask & AT_SIZE) { 2412 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr); 2413 if (err) { 2414 ZFS_EXIT(zfsvfs); 2415 return (err); 2416 } 2417 /* 2418 * XXX - Note, we are not providing any open 2419 * mode flags here (like FNDELAY), so we may 2420 * block if there are locks present... this 2421 * should be addressed in openat(). 2422 */ 2423 do { 2424 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2425 /* NB: we already did dmu_tx_wait() if necessary */ 2426 } while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 2427 if (err) { 2428 ZFS_EXIT(zfsvfs); 2429 return (err); 2430 } 2431 } 2432 2433 if (mask & (AT_ATIME|AT_MTIME) || 2434 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2435 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2436 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2437 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2438 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) 2439 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2440 skipaclchk, cr); 2441 2442 if (mask & (AT_UID|AT_GID)) { 2443 int idmask = (mask & (AT_UID|AT_GID)); 2444 int take_owner; 2445 int take_group; 2446 2447 /* 2448 * NOTE: even if a new mode is being set, 2449 * we may clear S_ISUID/S_ISGID bits. 2450 */ 2451 2452 if (!(mask & AT_MODE)) 2453 vap->va_mode = pzp->zp_mode; 2454 2455 /* 2456 * Take ownership or chgrp to group we are a member of 2457 */ 2458 2459 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2460 take_group = (mask & AT_GID) && 2461 zfs_groupmember(zfsvfs, vap->va_gid, cr); 2462 2463 /* 2464 * If both AT_UID and AT_GID are set then take_owner and 2465 * take_group must both be set in order to allow taking 2466 * ownership. 2467 * 2468 * Otherwise, send the check through secpolicy_vnode_setattr() 2469 * 2470 */ 2471 2472 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2473 ((idmask == AT_UID) && take_owner) || 2474 ((idmask == AT_GID) && take_group)) { 2475 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 2476 skipaclchk, cr) == 0) { 2477 /* 2478 * Remove setuid/setgid for non-privileged users 2479 */ 2480 secpolicy_setid_clear(vap, cr); 2481 trim_mask = (mask & (AT_UID|AT_GID)); 2482 } else { 2483 need_policy = TRUE; 2484 } 2485 } else { 2486 need_policy = TRUE; 2487 } 2488 } 2489 2490 mutex_enter(&zp->z_lock); 2491 oldva.va_mode = pzp->zp_mode; 2492 zfs_fuid_map_ids(zp, &oldva.va_uid, &oldva.va_gid); 2493 if (mask & AT_XVATTR) { 2494 if ((need_policy == FALSE) && 2495 (XVA_ISSET_REQ(xvap, XAT_APPENDONLY) && 2496 xoap->xoa_appendonly != 2497 ((pzp->zp_flags & ZFS_APPENDONLY) != 0)) || 2498 (XVA_ISSET_REQ(xvap, XAT_NOUNLINK) && 2499 xoap->xoa_nounlink != 2500 ((pzp->zp_flags & ZFS_NOUNLINK) != 0)) || 2501 (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE) && 2502 xoap->xoa_immutable != 2503 ((pzp->zp_flags & ZFS_IMMUTABLE) != 0)) || 2504 (XVA_ISSET_REQ(xvap, XAT_NODUMP) && 2505 xoap->xoa_nodump != 2506 ((pzp->zp_flags & ZFS_NODUMP) != 0)) || 2507 (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED) && 2508 xoap->xoa_av_modified != 2509 ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0)) || 2510 ((XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED) && 2511 ((vp->v_type != VREG && xoap->xoa_av_quarantined) || 2512 xoap->xoa_av_quarantined != 2513 ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0)))) || 2514 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) || 2515 (XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 2516 need_policy = TRUE; 2517 } 2518 } 2519 2520 mutex_exit(&zp->z_lock); 2521 2522 if (mask & AT_MODE) { 2523 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 2524 err = secpolicy_setid_setsticky_clear(vp, vap, 2525 &oldva, cr); 2526 if (err) { 2527 ZFS_EXIT(zfsvfs); 2528 return (err); 2529 } 2530 trim_mask |= AT_MODE; 2531 } else { 2532 need_policy = TRUE; 2533 } 2534 } 2535 2536 if (need_policy) { 2537 /* 2538 * If trim_mask is set then take ownership 2539 * has been granted or write_acl is present and user 2540 * has the ability to modify mode. In that case remove 2541 * UID|GID and or MODE from mask so that 2542 * secpolicy_vnode_setattr() doesn't revoke it. 2543 */ 2544 2545 if (trim_mask) { 2546 saved_mask = vap->va_mask; 2547 vap->va_mask &= ~trim_mask; 2548 } 2549 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2550 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 2551 if (err) { 2552 ZFS_EXIT(zfsvfs); 2553 return (err); 2554 } 2555 2556 if (trim_mask) 2557 vap->va_mask |= saved_mask; 2558 } 2559 2560 /* 2561 * secpolicy_vnode_setattr, or take ownership may have 2562 * changed va_mask 2563 */ 2564 mask = vap->va_mask; 2565 2566 tx = dmu_tx_create(zfsvfs->z_os); 2567 dmu_tx_hold_bonus(tx, zp->z_id); 2568 if (zfsvfs->z_fuid_obj == 0) { 2569 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 2570 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2571 SPA_MAXBLOCKSIZE); 2572 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 2573 } else { 2574 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 2575 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 2576 SPA_MAXBLOCKSIZE); 2577 } 2578 2579 if (mask & AT_MODE) { 2580 uint64_t pmode = pzp->zp_mode; 2581 2582 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2583 2584 if (pzp->zp_acl.z_acl_extern_obj) { 2585 /* Are we upgrading ACL from old V0 format to new V1 */ 2586 if (zfsvfs->z_version <= ZPL_VERSION_FUID && 2587 pzp->zp_acl.z_acl_version == 2588 ZFS_ACL_VERSION_INITIAL) { 2589 dmu_tx_hold_free(tx, 2590 pzp->zp_acl.z_acl_extern_obj, 0, 2591 DMU_OBJECT_END); 2592 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2593 0, sizeof (zfs_object_ace_t) * 2048 + 6); 2594 } else { 2595 dmu_tx_hold_write(tx, 2596 pzp->zp_acl.z_acl_extern_obj, 0, 2597 SPA_MAXBLOCKSIZE); 2598 } 2599 } else { 2600 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2601 0, sizeof (zfs_object_ace_t) * 2048 + 6); 2602 } 2603 } 2604 2605 if ((mask & (AT_UID | AT_GID)) && pzp->zp_xattr != 0) { 2606 err = zfs_zget(zp->z_zfsvfs, pzp->zp_xattr, &attrzp); 2607 if (err) { 2608 dmu_tx_abort(tx); 2609 ZFS_EXIT(zfsvfs); 2610 return (err); 2611 } 2612 dmu_tx_hold_bonus(tx, attrzp->z_id); 2613 } 2614 2615 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2616 if (err) { 2617 if (attrzp) 2618 VN_RELE(ZTOV(attrzp)); 2619 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2620 dmu_tx_wait(tx); 2621 dmu_tx_abort(tx); 2622 goto top; 2623 } 2624 dmu_tx_abort(tx); 2625 ZFS_EXIT(zfsvfs); 2626 return (err); 2627 } 2628 2629 dmu_buf_will_dirty(zp->z_dbuf, tx); 2630 2631 /* 2632 * Set each attribute requested. 2633 * We group settings according to the locks they need to acquire. 2634 * 2635 * Note: you cannot set ctime directly, although it will be 2636 * updated as a side-effect of calling this function. 2637 */ 2638 2639 mutex_enter(&zp->z_lock); 2640 2641 if (mask & AT_MODE) { 2642 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2643 ASSERT3U(err, ==, 0); 2644 } 2645 2646 if (attrzp) 2647 mutex_enter(&attrzp->z_lock); 2648 2649 if (mask & AT_UID) { 2650 pzp->zp_uid = zfs_fuid_create(zfsvfs, 2651 vap->va_uid, ZFS_OWNER, tx, &fuidp); 2652 if (attrzp) { 2653 attrzp->z_phys->zp_uid = zfs_fuid_create(zfsvfs, 2654 vap->va_uid, ZFS_OWNER, tx, &fuidp); 2655 } 2656 } 2657 2658 if (mask & AT_GID) { 2659 pzp->zp_gid = zfs_fuid_create(zfsvfs, vap->va_gid, 2660 ZFS_GROUP, tx, &fuidp); 2661 if (attrzp) 2662 attrzp->z_phys->zp_gid = zfs_fuid_create(zfsvfs, 2663 vap->va_gid, ZFS_GROUP, tx, &fuidp); 2664 } 2665 2666 if (attrzp) 2667 mutex_exit(&attrzp->z_lock); 2668 2669 if (mask & AT_ATIME) 2670 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2671 2672 if (mask & AT_MTIME) 2673 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2674 2675 if (mask & AT_SIZE) 2676 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2677 else if (mask != 0) 2678 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2679 /* 2680 * Do this after setting timestamps to prevent timestamp 2681 * update from toggling bit 2682 */ 2683 2684 if (xoap && (mask & AT_XVATTR)) { 2685 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 2686 size_t len; 2687 dmu_object_info_t doi; 2688 2689 ASSERT(vp->v_type == VREG); 2690 2691 /* Grow the bonus buffer if necessary. */ 2692 dmu_object_info_from_db(zp->z_dbuf, &doi); 2693 len = sizeof (xoap->xoa_av_scanstamp) + 2694 sizeof (znode_phys_t); 2695 if (len > doi.doi_bonus_size) 2696 VERIFY(dmu_set_bonus(zp->z_dbuf, len, tx) == 0); 2697 } 2698 zfs_xvattr_set(zp, xvap); 2699 } 2700 2701 if (mask != 0) 2702 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 2703 2704 if (fuidp) 2705 zfs_fuid_info_free(fuidp); 2706 mutex_exit(&zp->z_lock); 2707 2708 if (attrzp) 2709 VN_RELE(ZTOV(attrzp)); 2710 2711 dmu_tx_commit(tx); 2712 2713 ZFS_EXIT(zfsvfs); 2714 return (err); 2715 } 2716 2717 typedef struct zfs_zlock { 2718 krwlock_t *zl_rwlock; /* lock we acquired */ 2719 znode_t *zl_znode; /* znode we held */ 2720 struct zfs_zlock *zl_next; /* next in list */ 2721 } zfs_zlock_t; 2722 2723 /* 2724 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2725 */ 2726 static void 2727 zfs_rename_unlock(zfs_zlock_t **zlpp) 2728 { 2729 zfs_zlock_t *zl; 2730 2731 while ((zl = *zlpp) != NULL) { 2732 if (zl->zl_znode != NULL) 2733 VN_RELE(ZTOV(zl->zl_znode)); 2734 rw_exit(zl->zl_rwlock); 2735 *zlpp = zl->zl_next; 2736 kmem_free(zl, sizeof (*zl)); 2737 } 2738 } 2739 2740 /* 2741 * Search back through the directory tree, using the ".." entries. 2742 * Lock each directory in the chain to prevent concurrent renames. 2743 * Fail any attempt to move a directory into one of its own descendants. 2744 * XXX - z_parent_lock can overlap with map or grow locks 2745 */ 2746 static int 2747 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2748 { 2749 zfs_zlock_t *zl; 2750 znode_t *zp = tdzp; 2751 uint64_t rootid = zp->z_zfsvfs->z_root; 2752 uint64_t *oidp = &zp->z_id; 2753 krwlock_t *rwlp = &szp->z_parent_lock; 2754 krw_t rw = RW_WRITER; 2755 2756 /* 2757 * First pass write-locks szp and compares to zp->z_id. 2758 * Later passes read-lock zp and compare to zp->z_parent. 2759 */ 2760 do { 2761 if (!rw_tryenter(rwlp, rw)) { 2762 /* 2763 * Another thread is renaming in this path. 2764 * Note that if we are a WRITER, we don't have any 2765 * parent_locks held yet. 2766 */ 2767 if (rw == RW_READER && zp->z_id > szp->z_id) { 2768 /* 2769 * Drop our locks and restart 2770 */ 2771 zfs_rename_unlock(&zl); 2772 *zlpp = NULL; 2773 zp = tdzp; 2774 oidp = &zp->z_id; 2775 rwlp = &szp->z_parent_lock; 2776 rw = RW_WRITER; 2777 continue; 2778 } else { 2779 /* 2780 * Wait for other thread to drop its locks 2781 */ 2782 rw_enter(rwlp, rw); 2783 } 2784 } 2785 2786 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2787 zl->zl_rwlock = rwlp; 2788 zl->zl_znode = NULL; 2789 zl->zl_next = *zlpp; 2790 *zlpp = zl; 2791 2792 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2793 return (EINVAL); 2794 2795 if (*oidp == rootid) /* We've hit the top */ 2796 return (0); 2797 2798 if (rw == RW_READER) { /* i.e. not the first pass */ 2799 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2800 if (error) 2801 return (error); 2802 zl->zl_znode = zp; 2803 } 2804 oidp = &zp->z_phys->zp_parent; 2805 rwlp = &zp->z_parent_lock; 2806 rw = RW_READER; 2807 2808 } while (zp->z_id != sdzp->z_id); 2809 2810 return (0); 2811 } 2812 2813 /* 2814 * Move an entry from the provided source directory to the target 2815 * directory. Change the entry name as indicated. 2816 * 2817 * IN: sdvp - Source directory containing the "old entry". 2818 * snm - Old entry name. 2819 * tdvp - Target directory to contain the "new entry". 2820 * tnm - New entry name. 2821 * cr - credentials of caller. 2822 * ct - caller context 2823 * flags - case flags 2824 * 2825 * RETURN: 0 if success 2826 * error code if failure 2827 * 2828 * Timestamps: 2829 * sdvp,tdvp - ctime|mtime updated 2830 */ 2831 /*ARGSUSED*/ 2832 static int 2833 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 2834 caller_context_t *ct, int flags) 2835 { 2836 znode_t *tdzp, *szp, *tzp; 2837 znode_t *sdzp = VTOZ(sdvp); 2838 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2839 zilog_t *zilog; 2840 vnode_t *realvp; 2841 zfs_dirlock_t *sdl, *tdl; 2842 dmu_tx_t *tx; 2843 zfs_zlock_t *zl; 2844 int cmp, serr, terr; 2845 int error = 0; 2846 int zflg = 0; 2847 2848 ZFS_ENTER(zfsvfs); 2849 ZFS_VERIFY_ZP(sdzp); 2850 zilog = zfsvfs->z_log; 2851 2852 /* 2853 * Make sure we have the real vp for the target directory. 2854 */ 2855 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 2856 tdvp = realvp; 2857 2858 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2859 ZFS_EXIT(zfsvfs); 2860 return (EXDEV); 2861 } 2862 2863 tdzp = VTOZ(tdvp); 2864 ZFS_VERIFY_ZP(tdzp); 2865 if (zfsvfs->z_utf8 && u8_validate(tnm, 2866 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2867 ZFS_EXIT(zfsvfs); 2868 return (EILSEQ); 2869 } 2870 2871 if (flags & FIGNORECASE) 2872 zflg |= ZCILOOK; 2873 2874 top: 2875 szp = NULL; 2876 tzp = NULL; 2877 zl = NULL; 2878 2879 /* 2880 * This is to prevent the creation of links into attribute space 2881 * by renaming a linked file into/outof an attribute directory. 2882 * See the comment in zfs_link() for why this is considered bad. 2883 */ 2884 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2885 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2886 ZFS_EXIT(zfsvfs); 2887 return (EINVAL); 2888 } 2889 2890 /* 2891 * Lock source and target directory entries. To prevent deadlock, 2892 * a lock ordering must be defined. We lock the directory with 2893 * the smallest object id first, or if it's a tie, the one with 2894 * the lexically first name. 2895 */ 2896 if (sdzp->z_id < tdzp->z_id) { 2897 cmp = -1; 2898 } else if (sdzp->z_id > tdzp->z_id) { 2899 cmp = 1; 2900 } else { 2901 /* 2902 * First compare the two name arguments without 2903 * considering any case folding. 2904 */ 2905 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 2906 2907 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 2908 ASSERT(error == 0 || !zfsvfs->z_utf8); 2909 if (cmp == 0) { 2910 /* 2911 * POSIX: "If the old argument and the new argument 2912 * both refer to links to the same existing file, 2913 * the rename() function shall return successfully 2914 * and perform no other action." 2915 */ 2916 ZFS_EXIT(zfsvfs); 2917 return (0); 2918 } 2919 /* 2920 * If the file system is case-folding, then we may 2921 * have some more checking to do. A case-folding file 2922 * system is either supporting mixed case sensitivity 2923 * access or is completely case-insensitive. Note 2924 * that the file system is always case preserving. 2925 * 2926 * In mixed sensitivity mode case sensitive behavior 2927 * is the default. FIGNORECASE must be used to 2928 * explicitly request case insensitive behavior. 2929 * 2930 * If the source and target names provided differ only 2931 * by case (e.g., a request to rename 'tim' to 'Tim'), 2932 * we will treat this as a special case in the 2933 * case-insensitive mode: as long as the source name 2934 * is an exact match, we will allow this to proceed as 2935 * a name-change request. 2936 */ 2937 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 2938 (zfsvfs->z_case == ZFS_CASE_MIXED && 2939 flags & FIGNORECASE)) && 2940 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 2941 &error) == 0) { 2942 /* 2943 * case preserving rename request, require exact 2944 * name matches 2945 */ 2946 zflg |= ZCIEXACT; 2947 zflg &= ~ZCILOOK; 2948 } 2949 } 2950 2951 if (cmp < 0) { 2952 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 2953 ZEXISTS | zflg, NULL, NULL); 2954 terr = zfs_dirent_lock(&tdl, 2955 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 2956 } else { 2957 terr = zfs_dirent_lock(&tdl, 2958 tdzp, tnm, &tzp, zflg, NULL, NULL); 2959 serr = zfs_dirent_lock(&sdl, 2960 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 2961 NULL, NULL); 2962 } 2963 2964 if (serr) { 2965 /* 2966 * Source entry invalid or not there. 2967 */ 2968 if (!terr) { 2969 zfs_dirent_unlock(tdl); 2970 if (tzp) 2971 VN_RELE(ZTOV(tzp)); 2972 } 2973 if (strcmp(snm, "..") == 0) 2974 serr = EINVAL; 2975 ZFS_EXIT(zfsvfs); 2976 return (serr); 2977 } 2978 if (terr) { 2979 zfs_dirent_unlock(sdl); 2980 VN_RELE(ZTOV(szp)); 2981 if (strcmp(tnm, "..") == 0) 2982 terr = EINVAL; 2983 ZFS_EXIT(zfsvfs); 2984 return (terr); 2985 } 2986 2987 /* 2988 * Must have write access at the source to remove the old entry 2989 * and write access at the target to create the new entry. 2990 * Note that if target and source are the same, this can be 2991 * done in a single check. 2992 */ 2993 2994 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2995 goto out; 2996 2997 if (ZTOV(szp)->v_type == VDIR) { 2998 /* 2999 * Check to make sure rename is valid. 3000 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3001 */ 3002 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3003 goto out; 3004 } 3005 3006 /* 3007 * Does target exist? 3008 */ 3009 if (tzp) { 3010 /* 3011 * Source and target must be the same type. 3012 */ 3013 if (ZTOV(szp)->v_type == VDIR) { 3014 if (ZTOV(tzp)->v_type != VDIR) { 3015 error = ENOTDIR; 3016 goto out; 3017 } 3018 } else { 3019 if (ZTOV(tzp)->v_type == VDIR) { 3020 error = EISDIR; 3021 goto out; 3022 } 3023 } 3024 /* 3025 * POSIX dictates that when the source and target 3026 * entries refer to the same file object, rename 3027 * must do nothing and exit without error. 3028 */ 3029 if (szp->z_id == tzp->z_id) { 3030 error = 0; 3031 goto out; 3032 } 3033 } 3034 3035 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3036 if (tzp) 3037 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3038 3039 /* 3040 * notify the target directory if it is not the same 3041 * as source directory. 3042 */ 3043 if (tdvp != sdvp) { 3044 vnevent_rename_dest_dir(tdvp, ct); 3045 } 3046 3047 tx = dmu_tx_create(zfsvfs->z_os); 3048 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 3049 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 3050 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3051 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3052 if (sdzp != tdzp) 3053 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 3054 if (tzp) 3055 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 3056 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3057 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3058 if (error) { 3059 if (zl != NULL) 3060 zfs_rename_unlock(&zl); 3061 zfs_dirent_unlock(sdl); 3062 zfs_dirent_unlock(tdl); 3063 VN_RELE(ZTOV(szp)); 3064 if (tzp) 3065 VN_RELE(ZTOV(tzp)); 3066 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3067 dmu_tx_wait(tx); 3068 dmu_tx_abort(tx); 3069 goto top; 3070 } 3071 dmu_tx_abort(tx); 3072 ZFS_EXIT(zfsvfs); 3073 return (error); 3074 } 3075 3076 if (tzp) /* Attempt to remove the existing target */ 3077 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 3078 3079 if (error == 0) { 3080 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 3081 if (error == 0) { 3082 szp->z_phys->zp_flags |= ZFS_AV_MODIFIED; 3083 3084 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 3085 ASSERT(error == 0); 3086 3087 zfs_log_rename(zilog, tx, 3088 TX_RENAME | (flags & FIGNORECASE ? TX_CI : 0), 3089 sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp); 3090 } 3091 } 3092 3093 dmu_tx_commit(tx); 3094 out: 3095 if (zl != NULL) 3096 zfs_rename_unlock(&zl); 3097 3098 zfs_dirent_unlock(sdl); 3099 zfs_dirent_unlock(tdl); 3100 3101 VN_RELE(ZTOV(szp)); 3102 if (tzp) 3103 VN_RELE(ZTOV(tzp)); 3104 3105 ZFS_EXIT(zfsvfs); 3106 return (error); 3107 } 3108 3109 /* 3110 * Insert the indicated symbolic reference entry into the directory. 3111 * 3112 * IN: dvp - Directory to contain new symbolic link. 3113 * link - Name for new symlink entry. 3114 * vap - Attributes of new entry. 3115 * target - Target path of new symlink. 3116 * cr - credentials of caller. 3117 * ct - caller context 3118 * flags - case flags 3119 * 3120 * RETURN: 0 if success 3121 * error code if failure 3122 * 3123 * Timestamps: 3124 * dvp - ctime|mtime updated 3125 */ 3126 /*ARGSUSED*/ 3127 static int 3128 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr, 3129 caller_context_t *ct, int flags) 3130 { 3131 znode_t *zp, *dzp = VTOZ(dvp); 3132 zfs_dirlock_t *dl; 3133 dmu_tx_t *tx; 3134 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3135 zilog_t *zilog; 3136 int len = strlen(link); 3137 int error; 3138 int zflg = ZNEW; 3139 zfs_fuid_info_t *fuidp = NULL; 3140 3141 ASSERT(vap->va_type == VLNK); 3142 3143 ZFS_ENTER(zfsvfs); 3144 ZFS_VERIFY_ZP(dzp); 3145 zilog = zfsvfs->z_log; 3146 3147 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 3148 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3149 ZFS_EXIT(zfsvfs); 3150 return (EILSEQ); 3151 } 3152 if (flags & FIGNORECASE) 3153 zflg |= ZCILOOK; 3154 top: 3155 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3156 ZFS_EXIT(zfsvfs); 3157 return (error); 3158 } 3159 3160 if (len > MAXPATHLEN) { 3161 ZFS_EXIT(zfsvfs); 3162 return (ENAMETOOLONG); 3163 } 3164 3165 /* 3166 * Attempt to lock directory; fail if entry already exists. 3167 */ 3168 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 3169 if (error) { 3170 ZFS_EXIT(zfsvfs); 3171 return (error); 3172 } 3173 3174 tx = dmu_tx_create(zfsvfs->z_os); 3175 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 3176 dmu_tx_hold_bonus(tx, dzp->z_id); 3177 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3178 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 3179 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 3180 if (zfsvfs->z_fuid_obj == 0) { 3181 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 3182 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 3183 SPA_MAXBLOCKSIZE); 3184 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 3185 } else { 3186 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 3187 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 3188 SPA_MAXBLOCKSIZE); 3189 } 3190 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3191 if (error) { 3192 zfs_dirent_unlock(dl); 3193 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3194 dmu_tx_wait(tx); 3195 dmu_tx_abort(tx); 3196 goto top; 3197 } 3198 dmu_tx_abort(tx); 3199 ZFS_EXIT(zfsvfs); 3200 return (error); 3201 } 3202 3203 dmu_buf_will_dirty(dzp->z_dbuf, tx); 3204 3205 /* 3206 * Create a new object for the symlink. 3207 * Put the link content into bonus buffer if it will fit; 3208 * otherwise, store it just like any other file data. 3209 */ 3210 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 3211 zfs_mknode(dzp, vap, tx, cr, 0, &zp, len, NULL, &fuidp); 3212 if (len != 0) 3213 bcopy(link, zp->z_phys + 1, len); 3214 } else { 3215 dmu_buf_t *dbp; 3216 3217 zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, NULL, &fuidp); 3218 /* 3219 * Nothing can access the znode yet so no locking needed 3220 * for growing the znode's blocksize. 3221 */ 3222 zfs_grow_blocksize(zp, len, tx); 3223 3224 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, 3225 zp->z_id, 0, FTAG, &dbp)); 3226 dmu_buf_will_dirty(dbp, tx); 3227 3228 ASSERT3U(len, <=, dbp->db_size); 3229 bcopy(link, dbp->db_data, len); 3230 dmu_buf_rele(dbp, FTAG); 3231 } 3232 zp->z_phys->zp_size = len; 3233 3234 /* 3235 * Insert the new object into the directory. 3236 */ 3237 (void) zfs_link_create(dl, zp, tx, ZNEW); 3238 out: 3239 if (error == 0) { 3240 uint64_t txtype = TX_SYMLINK; 3241 if (flags & FIGNORECASE) 3242 txtype |= TX_CI; 3243 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 3244 } 3245 if (fuidp) 3246 zfs_fuid_info_free(fuidp); 3247 3248 dmu_tx_commit(tx); 3249 3250 zfs_dirent_unlock(dl); 3251 3252 VN_RELE(ZTOV(zp)); 3253 3254 ZFS_EXIT(zfsvfs); 3255 return (error); 3256 } 3257 3258 /* 3259 * Return, in the buffer contained in the provided uio structure, 3260 * the symbolic path referred to by vp. 3261 * 3262 * IN: vp - vnode of symbolic link. 3263 * uoip - structure to contain the link path. 3264 * cr - credentials of caller. 3265 * ct - caller context 3266 * 3267 * OUT: uio - structure to contain the link path. 3268 * 3269 * RETURN: 0 if success 3270 * error code if failure 3271 * 3272 * Timestamps: 3273 * vp - atime updated 3274 */ 3275 /* ARGSUSED */ 3276 static int 3277 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 3278 { 3279 znode_t *zp = VTOZ(vp); 3280 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3281 size_t bufsz; 3282 int error; 3283 3284 ZFS_ENTER(zfsvfs); 3285 ZFS_VERIFY_ZP(zp); 3286 3287 bufsz = (size_t)zp->z_phys->zp_size; 3288 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 3289 error = uiomove(zp->z_phys + 1, 3290 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 3291 } else { 3292 dmu_buf_t *dbp; 3293 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 3294 if (error) { 3295 ZFS_EXIT(zfsvfs); 3296 return (error); 3297 } 3298 error = uiomove(dbp->db_data, 3299 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 3300 dmu_buf_rele(dbp, FTAG); 3301 } 3302 3303 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3304 ZFS_EXIT(zfsvfs); 3305 return (error); 3306 } 3307 3308 /* 3309 * Insert a new entry into directory tdvp referencing svp. 3310 * 3311 * IN: tdvp - Directory to contain new entry. 3312 * svp - vnode of new entry. 3313 * name - name of new entry. 3314 * cr - credentials of caller. 3315 * ct - caller context 3316 * 3317 * RETURN: 0 if success 3318 * error code if failure 3319 * 3320 * Timestamps: 3321 * tdvp - ctime|mtime updated 3322 * svp - ctime updated 3323 */ 3324 /* ARGSUSED */ 3325 static int 3326 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 3327 caller_context_t *ct, int flags) 3328 { 3329 znode_t *dzp = VTOZ(tdvp); 3330 znode_t *tzp, *szp; 3331 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3332 zilog_t *zilog; 3333 zfs_dirlock_t *dl; 3334 dmu_tx_t *tx; 3335 vnode_t *realvp; 3336 int error; 3337 int zf = ZNEW; 3338 uid_t owner; 3339 3340 ASSERT(tdvp->v_type == VDIR); 3341 3342 ZFS_ENTER(zfsvfs); 3343 ZFS_VERIFY_ZP(dzp); 3344 zilog = zfsvfs->z_log; 3345 3346 if (VOP_REALVP(svp, &realvp, ct) == 0) 3347 svp = realvp; 3348 3349 if (svp->v_vfsp != tdvp->v_vfsp) { 3350 ZFS_EXIT(zfsvfs); 3351 return (EXDEV); 3352 } 3353 szp = VTOZ(svp); 3354 ZFS_VERIFY_ZP(szp); 3355 3356 if (zfsvfs->z_utf8 && u8_validate(name, 3357 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3358 ZFS_EXIT(zfsvfs); 3359 return (EILSEQ); 3360 } 3361 if (flags & FIGNORECASE) 3362 zf |= ZCILOOK; 3363 3364 top: 3365 /* 3366 * We do not support links between attributes and non-attributes 3367 * because of the potential security risk of creating links 3368 * into "normal" file space in order to circumvent restrictions 3369 * imposed in attribute space. 3370 */ 3371 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 3372 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 3373 ZFS_EXIT(zfsvfs); 3374 return (EINVAL); 3375 } 3376 3377 /* 3378 * POSIX dictates that we return EPERM here. 3379 * Better choices include ENOTSUP or EISDIR. 3380 */ 3381 if (svp->v_type == VDIR) { 3382 ZFS_EXIT(zfsvfs); 3383 return (EPERM); 3384 } 3385 3386 zfs_fuid_map_id(zfsvfs, szp->z_phys->zp_uid, ZFS_OWNER, &owner); 3387 if (owner != crgetuid(cr) && 3388 secpolicy_basic_link(cr) != 0) { 3389 ZFS_EXIT(zfsvfs); 3390 return (EPERM); 3391 } 3392 3393 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3394 ZFS_EXIT(zfsvfs); 3395 return (error); 3396 } 3397 3398 /* 3399 * Attempt to lock directory; fail if entry already exists. 3400 */ 3401 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 3402 if (error) { 3403 ZFS_EXIT(zfsvfs); 3404 return (error); 3405 } 3406 3407 tx = dmu_tx_create(zfsvfs->z_os); 3408 dmu_tx_hold_bonus(tx, szp->z_id); 3409 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3410 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3411 if (error) { 3412 zfs_dirent_unlock(dl); 3413 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3414 dmu_tx_wait(tx); 3415 dmu_tx_abort(tx); 3416 goto top; 3417 } 3418 dmu_tx_abort(tx); 3419 ZFS_EXIT(zfsvfs); 3420 return (error); 3421 } 3422 3423 error = zfs_link_create(dl, szp, tx, 0); 3424 3425 if (error == 0) { 3426 uint64_t txtype = TX_LINK; 3427 if (flags & FIGNORECASE) 3428 txtype |= TX_CI; 3429 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 3430 } 3431 3432 dmu_tx_commit(tx); 3433 3434 zfs_dirent_unlock(dl); 3435 3436 if (error == 0) { 3437 vnevent_link(svp, ct); 3438 } 3439 3440 ZFS_EXIT(zfsvfs); 3441 return (error); 3442 } 3443 3444 /* 3445 * zfs_null_putapage() is used when the file system has been force 3446 * unmounted. It just drops the pages. 3447 */ 3448 /* ARGSUSED */ 3449 static int 3450 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 3451 size_t *lenp, int flags, cred_t *cr) 3452 { 3453 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 3454 return (0); 3455 } 3456 3457 /* 3458 * Push a page out to disk, klustering if possible. 3459 * 3460 * IN: vp - file to push page to. 3461 * pp - page to push. 3462 * flags - additional flags. 3463 * cr - credentials of caller. 3464 * 3465 * OUT: offp - start of range pushed. 3466 * lenp - len of range pushed. 3467 * 3468 * RETURN: 0 if success 3469 * error code if failure 3470 * 3471 * NOTE: callers must have locked the page to be pushed. On 3472 * exit, the page (and all other pages in the kluster) must be 3473 * unlocked. 3474 */ 3475 /* ARGSUSED */ 3476 static int 3477 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 3478 size_t *lenp, int flags, cred_t *cr) 3479 { 3480 znode_t *zp = VTOZ(vp); 3481 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3482 zilog_t *zilog = zfsvfs->z_log; 3483 dmu_tx_t *tx; 3484 rl_t *rl; 3485 u_offset_t off, koff; 3486 size_t len, klen; 3487 uint64_t filesz; 3488 int err; 3489 3490 filesz = zp->z_phys->zp_size; 3491 off = pp->p_offset; 3492 len = PAGESIZE; 3493 /* 3494 * If our blocksize is bigger than the page size, try to kluster 3495 * muiltiple pages so that we write a full block (thus avoiding 3496 * a read-modify-write). 3497 */ 3498 if (off < filesz && zp->z_blksz > PAGESIZE) { 3499 if (!ISP2(zp->z_blksz)) { 3500 /* Only one block in the file. */ 3501 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3502 koff = 0; 3503 } else { 3504 klen = zp->z_blksz; 3505 koff = P2ALIGN(off, (u_offset_t)klen); 3506 } 3507 ASSERT(koff <= filesz); 3508 if (koff + klen > filesz) 3509 klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE); 3510 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 3511 } 3512 ASSERT3U(btop(len), ==, btopr(len)); 3513 top: 3514 rl = zfs_range_lock(zp, off, len, RL_WRITER); 3515 /* 3516 * Can't push pages past end-of-file. 3517 */ 3518 filesz = zp->z_phys->zp_size; 3519 if (off >= filesz) { 3520 /* ignore all pages */ 3521 err = 0; 3522 goto out; 3523 } else if (off + len > filesz) { 3524 int npages = btopr(filesz - off); 3525 page_t *trunc; 3526 3527 page_list_break(&pp, &trunc, npages); 3528 /* ignore pages past end of file */ 3529 if (trunc) 3530 pvn_write_done(trunc, flags); 3531 len = filesz - off; 3532 } 3533 3534 tx = dmu_tx_create(zfsvfs->z_os); 3535 dmu_tx_hold_write(tx, zp->z_id, off, len); 3536 dmu_tx_hold_bonus(tx, zp->z_id); 3537 err = dmu_tx_assign(tx, zfsvfs->z_assign); 3538 if (err != 0) { 3539 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3540 zfs_range_unlock(rl); 3541 dmu_tx_wait(tx); 3542 dmu_tx_abort(tx); 3543 err = 0; 3544 goto top; 3545 } 3546 dmu_tx_abort(tx); 3547 goto out; 3548 } 3549 3550 if (zp->z_blksz <= PAGESIZE) { 3551 caddr_t va = ppmapin(pp, PROT_READ, (caddr_t)-1); 3552 ASSERT3U(len, <=, PAGESIZE); 3553 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 3554 ppmapout(va); 3555 } else { 3556 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 3557 } 3558 3559 if (err == 0) { 3560 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 3561 zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0); 3562 dmu_tx_commit(tx); 3563 } 3564 3565 out: 3566 zfs_range_unlock(rl); 3567 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 3568 if (offp) 3569 *offp = off; 3570 if (lenp) 3571 *lenp = len; 3572 3573 return (err); 3574 } 3575 3576 /* 3577 * Copy the portion of the file indicated from pages into the file. 3578 * The pages are stored in a page list attached to the files vnode. 3579 * 3580 * IN: vp - vnode of file to push page data to. 3581 * off - position in file to put data. 3582 * len - amount of data to write. 3583 * flags - flags to control the operation. 3584 * cr - credentials of caller. 3585 * ct - caller context. 3586 * 3587 * RETURN: 0 if success 3588 * error code if failure 3589 * 3590 * Timestamps: 3591 * vp - ctime|mtime updated 3592 */ 3593 /*ARGSUSED*/ 3594 static int 3595 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 3596 caller_context_t *ct) 3597 { 3598 znode_t *zp = VTOZ(vp); 3599 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3600 page_t *pp; 3601 size_t io_len; 3602 u_offset_t io_off; 3603 uint64_t filesz; 3604 int error = 0; 3605 3606 ZFS_ENTER(zfsvfs); 3607 ZFS_VERIFY_ZP(zp); 3608 3609 if (len == 0) { 3610 /* 3611 * Search the entire vp list for pages >= off. 3612 */ 3613 error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage, 3614 flags, cr); 3615 goto out; 3616 } 3617 3618 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3619 if (off > filesz) { 3620 /* past end of file */ 3621 ZFS_EXIT(zfsvfs); 3622 return (0); 3623 } 3624 3625 len = MIN(len, filesz - off); 3626 3627 for (io_off = off; io_off < off + len; io_off += io_len) { 3628 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 3629 pp = page_lookup(vp, io_off, 3630 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 3631 } else { 3632 pp = page_lookup_nowait(vp, io_off, 3633 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 3634 } 3635 3636 if (pp != NULL && pvn_getdirty(pp, flags)) { 3637 int err; 3638 3639 /* 3640 * Found a dirty page to push 3641 */ 3642 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 3643 if (err) 3644 error = err; 3645 } else { 3646 io_len = PAGESIZE; 3647 } 3648 } 3649 out: 3650 if ((flags & B_ASYNC) == 0) 3651 zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id); 3652 ZFS_EXIT(zfsvfs); 3653 return (error); 3654 } 3655 3656 /*ARGSUSED*/ 3657 void 3658 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 3659 { 3660 znode_t *zp = VTOZ(vp); 3661 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3662 int error; 3663 3664 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 3665 if (zp->z_dbuf == NULL) { 3666 /* 3667 * The fs has been unmounted, or we did a 3668 * suspend/resume and this file no longer exists. 3669 */ 3670 if (vn_has_cached_data(vp)) { 3671 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 3672 B_INVAL, cr); 3673 } 3674 3675 mutex_enter(&zp->z_lock); 3676 vp->v_count = 0; /* count arrives as 1 */ 3677 mutex_exit(&zp->z_lock); 3678 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3679 zfs_znode_free(zp); 3680 return; 3681 } 3682 3683 /* 3684 * Attempt to push any data in the page cache. If this fails 3685 * we will get kicked out later in zfs_zinactive(). 3686 */ 3687 if (vn_has_cached_data(vp)) { 3688 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 3689 cr); 3690 } 3691 3692 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 3693 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 3694 3695 dmu_tx_hold_bonus(tx, zp->z_id); 3696 error = dmu_tx_assign(tx, TXG_WAIT); 3697 if (error) { 3698 dmu_tx_abort(tx); 3699 } else { 3700 dmu_buf_will_dirty(zp->z_dbuf, tx); 3701 mutex_enter(&zp->z_lock); 3702 zp->z_atime_dirty = 0; 3703 mutex_exit(&zp->z_lock); 3704 dmu_tx_commit(tx); 3705 } 3706 } 3707 3708 zfs_zinactive(zp); 3709 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3710 } 3711 3712 /* 3713 * Bounds-check the seek operation. 3714 * 3715 * IN: vp - vnode seeking within 3716 * ooff - old file offset 3717 * noffp - pointer to new file offset 3718 * ct - caller context 3719 * 3720 * RETURN: 0 if success 3721 * EINVAL if new offset invalid 3722 */ 3723 /* ARGSUSED */ 3724 static int 3725 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 3726 caller_context_t *ct) 3727 { 3728 if (vp->v_type == VDIR) 3729 return (0); 3730 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 3731 } 3732 3733 /* 3734 * Pre-filter the generic locking function to trap attempts to place 3735 * a mandatory lock on a memory mapped file. 3736 */ 3737 static int 3738 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 3739 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 3740 { 3741 znode_t *zp = VTOZ(vp); 3742 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3743 int error; 3744 3745 ZFS_ENTER(zfsvfs); 3746 ZFS_VERIFY_ZP(zp); 3747 3748 /* 3749 * We are following the UFS semantics with respect to mapcnt 3750 * here: If we see that the file is mapped already, then we will 3751 * return an error, but we don't worry about races between this 3752 * function and zfs_map(). 3753 */ 3754 if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) { 3755 ZFS_EXIT(zfsvfs); 3756 return (EAGAIN); 3757 } 3758 error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct); 3759 ZFS_EXIT(zfsvfs); 3760 return (error); 3761 } 3762 3763 /* 3764 * If we can't find a page in the cache, we will create a new page 3765 * and fill it with file data. For efficiency, we may try to fill 3766 * multiple pages at once (klustering). 3767 */ 3768 static int 3769 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 3770 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 3771 { 3772 znode_t *zp = VTOZ(vp); 3773 page_t *pp, *cur_pp; 3774 objset_t *os = zp->z_zfsvfs->z_os; 3775 caddr_t va; 3776 u_offset_t io_off, total; 3777 uint64_t oid = zp->z_id; 3778 size_t io_len; 3779 uint64_t filesz; 3780 int err; 3781 3782 /* 3783 * If we are only asking for a single page don't bother klustering. 3784 */ 3785 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3786 if (off >= filesz) 3787 return (EFAULT); 3788 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 3789 io_off = off; 3790 io_len = PAGESIZE; 3791 pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr); 3792 } else { 3793 /* 3794 * Try to fill a kluster of pages (a blocks worth). 3795 */ 3796 size_t klen; 3797 u_offset_t koff; 3798 3799 if (!ISP2(zp->z_blksz)) { 3800 /* Only one block in the file. */ 3801 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3802 koff = 0; 3803 } else { 3804 /* 3805 * It would be ideal to align our offset to the 3806 * blocksize but doing so has resulted in some 3807 * strange application crashes. For now, we 3808 * leave the offset as is and only adjust the 3809 * length if we are off the end of the file. 3810 */ 3811 koff = off; 3812 klen = plsz; 3813 } 3814 ASSERT(koff <= filesz); 3815 if (koff + klen > filesz) 3816 klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff; 3817 ASSERT3U(off, >=, koff); 3818 ASSERT3U(off, <, koff + klen); 3819 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 3820 &io_len, koff, klen, 0); 3821 } 3822 if (pp == NULL) { 3823 /* 3824 * Some other thread entered the page before us. 3825 * Return to zfs_getpage to retry the lookup. 3826 */ 3827 *pl = NULL; 3828 return (0); 3829 } 3830 3831 /* 3832 * Fill the pages in the kluster. 3833 */ 3834 cur_pp = pp; 3835 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 3836 ASSERT3U(io_off, ==, cur_pp->p_offset); 3837 va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 3838 err = dmu_read(os, oid, io_off, PAGESIZE, va); 3839 ppmapout(va); 3840 if (err) { 3841 /* On error, toss the entire kluster */ 3842 pvn_read_done(pp, B_ERROR); 3843 return (err); 3844 } 3845 cur_pp = cur_pp->p_next; 3846 } 3847 out: 3848 /* 3849 * Fill in the page list array from the kluster. If 3850 * there are too many pages in the kluster, return 3851 * as many pages as possible starting from the desired 3852 * offset `off'. 3853 * NOTE: the page list will always be null terminated. 3854 */ 3855 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 3856 3857 return (0); 3858 } 3859 3860 /* 3861 * Return pointers to the pages for the file region [off, off + len] 3862 * in the pl array. If plsz is greater than len, this function may 3863 * also return page pointers from before or after the specified 3864 * region (i.e. some region [off', off' + plsz]). These additional 3865 * pages are only returned if they are already in the cache, or were 3866 * created as part of a klustered read. 3867 * 3868 * IN: vp - vnode of file to get data from. 3869 * off - position in file to get data from. 3870 * len - amount of data to retrieve. 3871 * plsz - length of provided page list. 3872 * seg - segment to obtain pages for. 3873 * addr - virtual address of fault. 3874 * rw - mode of created pages. 3875 * cr - credentials of caller. 3876 * ct - caller context. 3877 * 3878 * OUT: protp - protection mode of created pages. 3879 * pl - list of pages created. 3880 * 3881 * RETURN: 0 if success 3882 * error code if failure 3883 * 3884 * Timestamps: 3885 * vp - atime updated 3886 */ 3887 /* ARGSUSED */ 3888 static int 3889 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 3890 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 3891 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 3892 { 3893 znode_t *zp = VTOZ(vp); 3894 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3895 page_t *pp, **pl0 = pl; 3896 int need_unlock = 0, err = 0; 3897 offset_t orig_off; 3898 3899 ZFS_ENTER(zfsvfs); 3900 ZFS_VERIFY_ZP(zp); 3901 3902 if (protp) 3903 *protp = PROT_ALL; 3904 3905 /* no faultahead (for now) */ 3906 if (pl == NULL) { 3907 ZFS_EXIT(zfsvfs); 3908 return (0); 3909 } 3910 3911 /* can't fault past EOF */ 3912 if (off >= zp->z_phys->zp_size) { 3913 ZFS_EXIT(zfsvfs); 3914 return (EFAULT); 3915 } 3916 orig_off = off; 3917 3918 /* 3919 * If we already own the lock, then we must be page faulting 3920 * in the middle of a write to this file (i.e., we are writing 3921 * to this file using data from a mapped region of the file). 3922 */ 3923 if (rw_owner(&zp->z_map_lock) != curthread) { 3924 rw_enter(&zp->z_map_lock, RW_WRITER); 3925 need_unlock = TRUE; 3926 } 3927 3928 /* 3929 * Loop through the requested range [off, off + len] looking 3930 * for pages. If we don't find a page, we will need to create 3931 * a new page and fill it with data from the file. 3932 */ 3933 while (len > 0) { 3934 if (plsz < PAGESIZE) 3935 break; 3936 if (pp = page_lookup(vp, off, SE_SHARED)) { 3937 *pl++ = pp; 3938 off += PAGESIZE; 3939 addr += PAGESIZE; 3940 len -= PAGESIZE; 3941 plsz -= PAGESIZE; 3942 } else { 3943 err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw); 3944 if (err) 3945 goto out; 3946 /* 3947 * klustering may have changed our region 3948 * to be block aligned. 3949 */ 3950 if (((pp = *pl) != 0) && (off != pp->p_offset)) { 3951 int delta = off - pp->p_offset; 3952 len += delta; 3953 off -= delta; 3954 addr -= delta; 3955 } 3956 while (*pl) { 3957 pl++; 3958 off += PAGESIZE; 3959 addr += PAGESIZE; 3960 plsz -= PAGESIZE; 3961 if (len > PAGESIZE) 3962 len -= PAGESIZE; 3963 else 3964 len = 0; 3965 } 3966 } 3967 } 3968 3969 /* 3970 * Fill out the page array with any pages already in the cache. 3971 */ 3972 while (plsz > 0) { 3973 pp = page_lookup_nowait(vp, off, SE_SHARED); 3974 if (pp == NULL) 3975 break; 3976 *pl++ = pp; 3977 off += PAGESIZE; 3978 plsz -= PAGESIZE; 3979 } 3980 3981 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3982 out: 3983 /* 3984 * We can't grab the range lock for the page as reader which would 3985 * stop truncation as this leads to deadlock. So we need to recheck 3986 * the file size. 3987 */ 3988 if (orig_off >= zp->z_phys->zp_size) 3989 err = EFAULT; 3990 if (err) { 3991 /* 3992 * Release any pages we have previously locked. 3993 */ 3994 while (pl > pl0) 3995 page_unlock(*--pl); 3996 } 3997 3998 *pl = NULL; 3999 4000 if (need_unlock) 4001 rw_exit(&zp->z_map_lock); 4002 4003 ZFS_EXIT(zfsvfs); 4004 return (err); 4005 } 4006 4007 /* 4008 * Request a memory map for a section of a file. This code interacts 4009 * with common code and the VM system as follows: 4010 * 4011 * common code calls mmap(), which ends up in smmap_common() 4012 * 4013 * this calls VOP_MAP(), which takes you into (say) zfs 4014 * 4015 * zfs_map() calls as_map(), passing segvn_create() as the callback 4016 * 4017 * segvn_create() creates the new segment and calls VOP_ADDMAP() 4018 * 4019 * zfs_addmap() updates z_mapcnt 4020 */ 4021 /*ARGSUSED*/ 4022 static int 4023 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4024 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4025 caller_context_t *ct) 4026 { 4027 znode_t *zp = VTOZ(vp); 4028 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4029 segvn_crargs_t vn_a; 4030 int error; 4031 4032 if ((prot & PROT_WRITE) && 4033 (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_READONLY | 4034 ZFS_APPENDONLY))) 4035 return (EPERM); 4036 4037 ZFS_ENTER(zfsvfs); 4038 ZFS_VERIFY_ZP(zp); 4039 4040 if (vp->v_flag & VNOMAP) { 4041 ZFS_EXIT(zfsvfs); 4042 return (ENOSYS); 4043 } 4044 4045 if (off < 0 || len > MAXOFFSET_T - off) { 4046 ZFS_EXIT(zfsvfs); 4047 return (ENXIO); 4048 } 4049 4050 if (vp->v_type != VREG) { 4051 ZFS_EXIT(zfsvfs); 4052 return (ENODEV); 4053 } 4054 4055 /* 4056 * If file is locked, disallow mapping. 4057 */ 4058 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) { 4059 ZFS_EXIT(zfsvfs); 4060 return (EAGAIN); 4061 } 4062 4063 as_rangelock(as); 4064 if ((flags & MAP_FIXED) == 0) { 4065 map_addr(addrp, len, off, 1, flags); 4066 if (*addrp == NULL) { 4067 as_rangeunlock(as); 4068 ZFS_EXIT(zfsvfs); 4069 return (ENOMEM); 4070 } 4071 } else { 4072 /* 4073 * User specified address - blow away any previous mappings 4074 */ 4075 (void) as_unmap(as, *addrp, len); 4076 } 4077 4078 vn_a.vp = vp; 4079 vn_a.offset = (u_offset_t)off; 4080 vn_a.type = flags & MAP_TYPE; 4081 vn_a.prot = prot; 4082 vn_a.maxprot = maxprot; 4083 vn_a.cred = cr; 4084 vn_a.amp = NULL; 4085 vn_a.flags = flags & ~MAP_TYPE; 4086 vn_a.szc = 0; 4087 vn_a.lgrp_mem_policy_flags = 0; 4088 4089 error = as_map(as, *addrp, len, segvn_create, &vn_a); 4090 4091 as_rangeunlock(as); 4092 ZFS_EXIT(zfsvfs); 4093 return (error); 4094 } 4095 4096 /* ARGSUSED */ 4097 static int 4098 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4099 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4100 caller_context_t *ct) 4101 { 4102 uint64_t pages = btopr(len); 4103 4104 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 4105 return (0); 4106 } 4107 4108 /* 4109 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 4110 * more accurate mtime for the associated file. Since we don't have a way of 4111 * detecting when the data was actually modified, we have to resort to 4112 * heuristics. If an explicit msync() is done, then we mark the mtime when the 4113 * last page is pushed. The problem occurs when the msync() call is omitted, 4114 * which by far the most common case: 4115 * 4116 * open() 4117 * mmap() 4118 * <modify memory> 4119 * munmap() 4120 * close() 4121 * <time lapse> 4122 * putpage() via fsflush 4123 * 4124 * If we wait until fsflush to come along, we can have a modification time that 4125 * is some arbitrary point in the future. In order to prevent this in the 4126 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 4127 * torn down. 4128 */ 4129 /* ARGSUSED */ 4130 static int 4131 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4132 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 4133 caller_context_t *ct) 4134 { 4135 uint64_t pages = btopr(len); 4136 4137 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 4138 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 4139 4140 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 4141 vn_has_cached_data(vp)) 4142 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 4143 4144 return (0); 4145 } 4146 4147 /* 4148 * Free or allocate space in a file. Currently, this function only 4149 * supports the `F_FREESP' command. However, this command is somewhat 4150 * misnamed, as its functionality includes the ability to allocate as 4151 * well as free space. 4152 * 4153 * IN: vp - vnode of file to free data in. 4154 * cmd - action to take (only F_FREESP supported). 4155 * bfp - section of file to free/alloc. 4156 * flag - current file open mode flags. 4157 * offset - current file offset. 4158 * cr - credentials of caller [UNUSED]. 4159 * ct - caller context. 4160 * 4161 * RETURN: 0 if success 4162 * error code if failure 4163 * 4164 * Timestamps: 4165 * vp - ctime|mtime updated 4166 */ 4167 /* ARGSUSED */ 4168 static int 4169 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 4170 offset_t offset, cred_t *cr, caller_context_t *ct) 4171 { 4172 znode_t *zp = VTOZ(vp); 4173 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4174 uint64_t off, len; 4175 int error; 4176 4177 ZFS_ENTER(zfsvfs); 4178 ZFS_VERIFY_ZP(zp); 4179 4180 top: 4181 if (cmd != F_FREESP) { 4182 ZFS_EXIT(zfsvfs); 4183 return (EINVAL); 4184 } 4185 4186 if (error = convoff(vp, bfp, 0, offset)) { 4187 ZFS_EXIT(zfsvfs); 4188 return (error); 4189 } 4190 4191 if (bfp->l_len < 0) { 4192 ZFS_EXIT(zfsvfs); 4193 return (EINVAL); 4194 } 4195 4196 off = bfp->l_start; 4197 len = bfp->l_len; /* 0 means from off to end of file */ 4198 4199 do { 4200 error = zfs_freesp(zp, off, len, flag, TRUE); 4201 /* NB: we already did dmu_tx_wait() if necessary */ 4202 } while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 4203 4204 ZFS_EXIT(zfsvfs); 4205 return (error); 4206 } 4207 4208 /*ARGSUSED*/ 4209 static int 4210 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 4211 { 4212 znode_t *zp = VTOZ(vp); 4213 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4214 uint32_t gen; 4215 uint64_t object = zp->z_id; 4216 zfid_short_t *zfid; 4217 int size, i; 4218 4219 ZFS_ENTER(zfsvfs); 4220 ZFS_VERIFY_ZP(zp); 4221 gen = (uint32_t)zp->z_gen; 4222 4223 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 4224 if (fidp->fid_len < size) { 4225 fidp->fid_len = size; 4226 ZFS_EXIT(zfsvfs); 4227 return (ENOSPC); 4228 } 4229 4230 zfid = (zfid_short_t *)fidp; 4231 4232 zfid->zf_len = size; 4233 4234 for (i = 0; i < sizeof (zfid->zf_object); i++) 4235 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 4236 4237 /* Must have a non-zero generation number to distinguish from .zfs */ 4238 if (gen == 0) 4239 gen = 1; 4240 for (i = 0; i < sizeof (zfid->zf_gen); i++) 4241 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 4242 4243 if (size == LONG_FID_LEN) { 4244 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 4245 zfid_long_t *zlfid; 4246 4247 zlfid = (zfid_long_t *)fidp; 4248 4249 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 4250 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 4251 4252 /* XXX - this should be the generation number for the objset */ 4253 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 4254 zlfid->zf_setgen[i] = 0; 4255 } 4256 4257 ZFS_EXIT(zfsvfs); 4258 return (0); 4259 } 4260 4261 static int 4262 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 4263 caller_context_t *ct) 4264 { 4265 znode_t *zp, *xzp; 4266 zfsvfs_t *zfsvfs; 4267 zfs_dirlock_t *dl; 4268 int error; 4269 4270 switch (cmd) { 4271 case _PC_LINK_MAX: 4272 *valp = ULONG_MAX; 4273 return (0); 4274 4275 case _PC_FILESIZEBITS: 4276 *valp = 64; 4277 return (0); 4278 4279 case _PC_XATTR_EXISTS: 4280 zp = VTOZ(vp); 4281 zfsvfs = zp->z_zfsvfs; 4282 ZFS_ENTER(zfsvfs); 4283 ZFS_VERIFY_ZP(zp); 4284 *valp = 0; 4285 error = zfs_dirent_lock(&dl, zp, "", &xzp, 4286 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 4287 if (error == 0) { 4288 zfs_dirent_unlock(dl); 4289 if (!zfs_dirempty(xzp)) 4290 *valp = 1; 4291 VN_RELE(ZTOV(xzp)); 4292 } else if (error == ENOENT) { 4293 /* 4294 * If there aren't extended attributes, it's the 4295 * same as having zero of them. 4296 */ 4297 error = 0; 4298 } 4299 ZFS_EXIT(zfsvfs); 4300 return (error); 4301 4302 case _PC_SATTR_ENABLED: 4303 case _PC_SATTR_EXISTS: 4304 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) && 4305 (vp->v_type == VREG || vp->v_type == VDIR); 4306 return (0); 4307 4308 case _PC_ACL_ENABLED: 4309 *valp = _ACL_ACE_ENABLED; 4310 return (0); 4311 4312 case _PC_MIN_HOLE_SIZE: 4313 *valp = (ulong_t)SPA_MINBLOCKSIZE; 4314 return (0); 4315 4316 default: 4317 return (fs_pathconf(vp, cmd, valp, cr, ct)); 4318 } 4319 } 4320 4321 /*ARGSUSED*/ 4322 static int 4323 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4324 caller_context_t *ct) 4325 { 4326 znode_t *zp = VTOZ(vp); 4327 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4328 int error; 4329 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4330 4331 ZFS_ENTER(zfsvfs); 4332 ZFS_VERIFY_ZP(zp); 4333 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 4334 ZFS_EXIT(zfsvfs); 4335 4336 return (error); 4337 } 4338 4339 /*ARGSUSED*/ 4340 static int 4341 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4342 caller_context_t *ct) 4343 { 4344 znode_t *zp = VTOZ(vp); 4345 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4346 int error; 4347 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4348 4349 ZFS_ENTER(zfsvfs); 4350 ZFS_VERIFY_ZP(zp); 4351 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 4352 ZFS_EXIT(zfsvfs); 4353 return (error); 4354 } 4355 4356 /* 4357 * Predeclare these here so that the compiler assumes that 4358 * this is an "old style" function declaration that does 4359 * not include arguments => we won't get type mismatch errors 4360 * in the initializations that follow. 4361 */ 4362 static int zfs_inval(); 4363 static int zfs_isdir(); 4364 4365 static int 4366 zfs_inval() 4367 { 4368 return (EINVAL); 4369 } 4370 4371 static int 4372 zfs_isdir() 4373 { 4374 return (EISDIR); 4375 } 4376 /* 4377 * Directory vnode operations template 4378 */ 4379 vnodeops_t *zfs_dvnodeops; 4380 const fs_operation_def_t zfs_dvnodeops_template[] = { 4381 VOPNAME_OPEN, { .vop_open = zfs_open }, 4382 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4383 VOPNAME_READ, { .error = zfs_isdir }, 4384 VOPNAME_WRITE, { .error = zfs_isdir }, 4385 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4386 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4387 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4388 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4389 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4390 VOPNAME_CREATE, { .vop_create = zfs_create }, 4391 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 4392 VOPNAME_LINK, { .vop_link = zfs_link }, 4393 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4394 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 4395 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 4396 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 4397 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 4398 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4399 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4400 VOPNAME_FID, { .vop_fid = zfs_fid }, 4401 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4402 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4403 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4404 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4405 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4406 NULL, NULL 4407 }; 4408 4409 /* 4410 * Regular file vnode operations template 4411 */ 4412 vnodeops_t *zfs_fvnodeops; 4413 const fs_operation_def_t zfs_fvnodeops_template[] = { 4414 VOPNAME_OPEN, { .vop_open = zfs_open }, 4415 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4416 VOPNAME_READ, { .vop_read = zfs_read }, 4417 VOPNAME_WRITE, { .vop_write = zfs_write }, 4418 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4419 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4420 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4421 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4422 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4423 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4424 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4425 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4426 VOPNAME_FID, { .vop_fid = zfs_fid }, 4427 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4428 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 4429 VOPNAME_SPACE, { .vop_space = zfs_space }, 4430 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 4431 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 4432 VOPNAME_MAP, { .vop_map = zfs_map }, 4433 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 4434 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 4435 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4436 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4437 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4438 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4439 NULL, NULL 4440 }; 4441 4442 /* 4443 * Symbolic link vnode operations template 4444 */ 4445 vnodeops_t *zfs_symvnodeops; 4446 const fs_operation_def_t zfs_symvnodeops_template[] = { 4447 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4448 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4449 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4450 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4451 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 4452 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4453 VOPNAME_FID, { .vop_fid = zfs_fid }, 4454 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4455 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4456 NULL, NULL 4457 }; 4458 4459 /* 4460 * Extended attribute directory vnode operations template 4461 * This template is identical to the directory vnodes 4462 * operation template except for restricted operations: 4463 * VOP_MKDIR() 4464 * VOP_SYMLINK() 4465 * Note that there are other restrictions embedded in: 4466 * zfs_create() - restrict type to VREG 4467 * zfs_link() - no links into/out of attribute space 4468 * zfs_rename() - no moves into/out of attribute space 4469 */ 4470 vnodeops_t *zfs_xdvnodeops; 4471 const fs_operation_def_t zfs_xdvnodeops_template[] = { 4472 VOPNAME_OPEN, { .vop_open = zfs_open }, 4473 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4474 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4475 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4476 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4477 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4478 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4479 VOPNAME_CREATE, { .vop_create = zfs_create }, 4480 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 4481 VOPNAME_LINK, { .vop_link = zfs_link }, 4482 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4483 VOPNAME_MKDIR, { .error = zfs_inval }, 4484 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 4485 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 4486 VOPNAME_SYMLINK, { .error = zfs_inval }, 4487 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4488 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4489 VOPNAME_FID, { .vop_fid = zfs_fid }, 4490 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4491 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4492 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4493 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4494 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4495 NULL, NULL 4496 }; 4497 4498 /* 4499 * Error vnode operations template 4500 */ 4501 vnodeops_t *zfs_evnodeops; 4502 const fs_operation_def_t zfs_evnodeops_template[] = { 4503 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4504 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4505 NULL, NULL 4506 }; 4507