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_vfsops.h> 56 #include <sys/zfs_dir.h> 57 #include <sys/zfs_acl.h> 58 #include <sys/zfs_ioctl.h> 59 #include <sys/zfs_i18n.h> 60 #include <sys/fs/zfs.h> 61 #include <sys/dmu.h> 62 #include <sys/spa.h> 63 #include <sys/txg.h> 64 #include <sys/dbuf.h> 65 #include <sys/zap.h> 66 #include <sys/dirent.h> 67 #include <sys/policy.h> 68 #include <sys/sunddi.h> 69 #include <sys/filio.h> 70 #include "fs/fs_subr.h" 71 #include <sys/zfs_ctldir.h> 72 #include <sys/zfs_fuid.h> 73 #include <sys/dnlc.h> 74 #include <sys/zfs_rlock.h> 75 #include <sys/extdirent.h> 76 #include <sys/kidmap.h> 77 #include <sys/cred_impl.h> 78 79 /* 80 * Programming rules. 81 * 82 * Each vnode op performs some logical unit of work. To do this, the ZPL must 83 * properly lock its in-core state, create a DMU transaction, do the work, 84 * record this work in the intent log (ZIL), commit the DMU transaction, 85 * and wait for the intent log to commit if it is a synchronous operation. 86 * Moreover, the vnode ops must work in both normal and log replay context. 87 * The ordering of events is important to avoid deadlocks and references 88 * to freed memory. The example below illustrates the following Big Rules: 89 * 90 * (1) A check must be made in each zfs thread for a mounted file system. 91 * This is done avoiding races using ZFS_ENTER(zfsvfs). 92 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes 93 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros 94 * can return EIO from the calling function. 95 * 96 * (2) VN_RELE() should always be the last thing except for zil_commit() 97 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 98 * First, if it's the last reference, the vnode/znode 99 * can be freed, so the zp may point to freed memory. Second, the last 100 * reference will call zfs_zinactive(), which may induce a lot of work -- 101 * pushing cached pages (which acquires range locks) and syncing out 102 * cached atime changes. Third, zfs_zinactive() may require a new tx, 103 * which could deadlock the system if you were already holding one. 104 * 105 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 106 * as they can span dmu_tx_assign() calls. 107 * 108 * (4) Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign(). 109 * In normal operation, this will be TXG_NOWAIT. During ZIL replay, 110 * it will be a specific txg. Either way, dmu_tx_assign() never blocks. 111 * This is critical because we don't want to block while holding locks. 112 * Note, in particular, that if a lock is sometimes acquired before 113 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 114 * use a non-blocking assign can deadlock the system. The scenario: 115 * 116 * Thread A has grabbed a lock before calling dmu_tx_assign(). 117 * Thread B is in an already-assigned tx, and blocks for this lock. 118 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 119 * forever, because the previous txg can't quiesce until B's tx commits. 120 * 121 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 122 * then drop all locks, call dmu_tx_wait(), and try again. 123 * 124 * (5) If the operation succeeded, generate the intent log entry for it 125 * before dropping locks. This ensures that the ordering of events 126 * in the intent log matches the order in which they actually occurred. 127 * 128 * (6) At the end of each vnode op, the DMU tx must always commit, 129 * regardless of whether there were any errors. 130 * 131 * (7) After dropping all locks, invoke zil_commit(zilog, seq, foid) 132 * to ensure that synchronous semantics are provided when necessary. 133 * 134 * In general, this is how things should be ordered in each vnode op: 135 * 136 * ZFS_ENTER(zfsvfs); // exit if unmounted 137 * top: 138 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 139 * rw_enter(...); // grab any other locks you need 140 * tx = dmu_tx_create(...); // get DMU tx 141 * dmu_tx_hold_*(); // hold each object you might modify 142 * error = dmu_tx_assign(tx, zfsvfs->z_assign); // try to assign 143 * if (error) { 144 * rw_exit(...); // drop locks 145 * zfs_dirent_unlock(dl); // unlock directory entry 146 * VN_RELE(...); // release held vnodes 147 * if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 148 * dmu_tx_wait(tx); 149 * dmu_tx_abort(tx); 150 * goto top; 151 * } 152 * dmu_tx_abort(tx); // abort DMU tx 153 * ZFS_EXIT(zfsvfs); // finished in zfs 154 * return (error); // really out of space 155 * } 156 * error = do_real_work(); // do whatever this VOP does 157 * if (error == 0) 158 * zfs_log_*(...); // on success, make ZIL entry 159 * dmu_tx_commit(tx); // commit DMU tx -- error or not 160 * rw_exit(...); // drop locks 161 * zfs_dirent_unlock(dl); // unlock directory entry 162 * VN_RELE(...); // release held vnodes 163 * zil_commit(zilog, seq, foid); // synchronous when necessary 164 * ZFS_EXIT(zfsvfs); // finished in zfs 165 * return (error); // done, report error 166 */ 167 168 /* ARGSUSED */ 169 static int 170 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 171 { 172 znode_t *zp = VTOZ(*vpp); 173 174 if ((flag & FWRITE) && (zp->z_phys->zp_flags & ZFS_APPENDONLY) && 175 ((flag & FAPPEND) == 0)) { 176 return (EPERM); 177 } 178 179 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 180 ZTOV(zp)->v_type == VREG && 181 !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) && 182 zp->z_phys->zp_size > 0) 183 if (fs_vscan(*vpp, cr, 0) != 0) 184 return (EACCES); 185 186 /* Keep a count of the synchronous opens in the znode */ 187 if (flag & (FSYNC | FDSYNC)) 188 atomic_inc_32(&zp->z_sync_cnt); 189 190 return (0); 191 } 192 193 /* ARGSUSED */ 194 static int 195 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 196 caller_context_t *ct) 197 { 198 znode_t *zp = VTOZ(vp); 199 200 /* Decrement the synchronous opens in the znode */ 201 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 202 atomic_dec_32(&zp->z_sync_cnt); 203 204 /* 205 * Clean up any locks held by this process on the vp. 206 */ 207 cleanlocks(vp, ddi_get_pid(), 0); 208 cleanshares(vp, ddi_get_pid()); 209 210 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 211 ZTOV(zp)->v_type == VREG && 212 !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) && 213 zp->z_phys->zp_size > 0) 214 VERIFY(fs_vscan(vp, cr, 1) == 0); 215 216 return (0); 217 } 218 219 /* 220 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 221 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 222 */ 223 static int 224 zfs_holey(vnode_t *vp, int cmd, offset_t *off) 225 { 226 znode_t *zp = VTOZ(vp); 227 uint64_t noff = (uint64_t)*off; /* new offset */ 228 uint64_t file_sz; 229 int error; 230 boolean_t hole; 231 232 file_sz = zp->z_phys->zp_size; 233 if (noff >= file_sz) { 234 return (ENXIO); 235 } 236 237 if (cmd == _FIO_SEEK_HOLE) 238 hole = B_TRUE; 239 else 240 hole = B_FALSE; 241 242 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 243 244 /* end of file? */ 245 if ((error == ESRCH) || (noff > file_sz)) { 246 /* 247 * Handle the virtual hole at the end of file. 248 */ 249 if (hole) { 250 *off = file_sz; 251 return (0); 252 } 253 return (ENXIO); 254 } 255 256 if (noff < *off) 257 return (error); 258 *off = noff; 259 return (error); 260 } 261 262 /* ARGSUSED */ 263 static int 264 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred, 265 int *rvalp, caller_context_t *ct) 266 { 267 offset_t off; 268 int error; 269 zfsvfs_t *zfsvfs; 270 znode_t *zp; 271 272 switch (com) { 273 case _FIOFFS: 274 return (zfs_sync(vp->v_vfsp, 0, cred)); 275 276 /* 277 * The following two ioctls are used by bfu. Faking out, 278 * necessary to avoid bfu errors. 279 */ 280 case _FIOGDIO: 281 case _FIOSDIO: 282 return (0); 283 284 case _FIO_SEEK_DATA: 285 case _FIO_SEEK_HOLE: 286 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 287 return (EFAULT); 288 289 zp = VTOZ(vp); 290 zfsvfs = zp->z_zfsvfs; 291 ZFS_ENTER(zfsvfs); 292 ZFS_VERIFY_ZP(zp); 293 294 /* offset parameter is in/out */ 295 error = zfs_holey(vp, com, &off); 296 ZFS_EXIT(zfsvfs); 297 if (error) 298 return (error); 299 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 300 return (EFAULT); 301 return (0); 302 } 303 return (ENOTTY); 304 } 305 306 /* 307 * When a file is memory mapped, we must keep the IO data synchronized 308 * between the DMU cache and the memory mapped pages. What this means: 309 * 310 * On Write: If we find a memory mapped page, we write to *both* 311 * the page and the dmu buffer. 312 * 313 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 314 * the file is memory mapped. 315 */ 316 static int 317 mappedwrite(vnode_t *vp, int nbytes, uio_t *uio, dmu_tx_t *tx) 318 { 319 znode_t *zp = VTOZ(vp); 320 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 321 int64_t start, off; 322 int len = nbytes; 323 int error = 0; 324 325 start = uio->uio_loffset; 326 off = start & PAGEOFFSET; 327 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 328 page_t *pp; 329 uint64_t bytes = MIN(PAGESIZE - off, len); 330 uint64_t woff = uio->uio_loffset; 331 332 /* 333 * We don't want a new page to "appear" in the middle of 334 * the file update (because it may not get the write 335 * update data), so we grab a lock to block 336 * zfs_getpage(). 337 */ 338 rw_enter(&zp->z_map_lock, RW_WRITER); 339 if (pp = page_lookup(vp, start, SE_SHARED)) { 340 caddr_t va; 341 342 rw_exit(&zp->z_map_lock); 343 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L); 344 error = uiomove(va+off, bytes, UIO_WRITE, uio); 345 if (error == 0) { 346 dmu_write(zfsvfs->z_os, zp->z_id, 347 woff, bytes, va+off, tx); 348 } 349 ppmapout(va); 350 page_unlock(pp); 351 } else { 352 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 353 uio, bytes, tx); 354 rw_exit(&zp->z_map_lock); 355 } 356 len -= bytes; 357 off = 0; 358 if (error) 359 break; 360 } 361 return (error); 362 } 363 364 /* 365 * When a file is memory mapped, we must keep the IO data synchronized 366 * between the DMU cache and the memory mapped pages. What this means: 367 * 368 * On Read: We "read" preferentially from memory mapped pages, 369 * else we default from the dmu buffer. 370 * 371 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 372 * the file is memory mapped. 373 */ 374 static int 375 mappedread(vnode_t *vp, int nbytes, uio_t *uio) 376 { 377 znode_t *zp = VTOZ(vp); 378 objset_t *os = zp->z_zfsvfs->z_os; 379 int64_t start, off; 380 int len = nbytes; 381 int error = 0; 382 383 start = uio->uio_loffset; 384 off = start & PAGEOFFSET; 385 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 386 page_t *pp; 387 uint64_t bytes = MIN(PAGESIZE - off, len); 388 389 if (pp = page_lookup(vp, start, SE_SHARED)) { 390 caddr_t va; 391 392 va = ppmapin(pp, PROT_READ, (caddr_t)-1L); 393 error = uiomove(va + off, bytes, UIO_READ, uio); 394 ppmapout(va); 395 page_unlock(pp); 396 } else { 397 error = dmu_read_uio(os, zp->z_id, uio, bytes); 398 } 399 len -= bytes; 400 off = 0; 401 if (error) 402 break; 403 } 404 return (error); 405 } 406 407 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 408 409 /* 410 * Read bytes from specified file into supplied buffer. 411 * 412 * IN: vp - vnode of file to be read from. 413 * uio - structure supplying read location, range info, 414 * and return buffer. 415 * ioflag - SYNC flags; used to provide FRSYNC semantics. 416 * cr - credentials of caller. 417 * ct - caller context 418 * 419 * OUT: uio - updated offset and range, buffer filled. 420 * 421 * RETURN: 0 if success 422 * error code if failure 423 * 424 * Side Effects: 425 * vp - atime updated if byte count > 0 426 */ 427 /* ARGSUSED */ 428 static int 429 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 430 { 431 znode_t *zp = VTOZ(vp); 432 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 433 objset_t *os; 434 ssize_t n, nbytes; 435 int error; 436 rl_t *rl; 437 438 ZFS_ENTER(zfsvfs); 439 ZFS_VERIFY_ZP(zp); 440 os = zfsvfs->z_os; 441 442 /* 443 * Validate file offset 444 */ 445 if (uio->uio_loffset < (offset_t)0) { 446 ZFS_EXIT(zfsvfs); 447 return (EINVAL); 448 } 449 450 /* 451 * Fasttrack empty reads 452 */ 453 if (uio->uio_resid == 0) { 454 ZFS_EXIT(zfsvfs); 455 return (0); 456 } 457 458 /* 459 * Check for mandatory locks 460 */ 461 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 462 if (error = chklock(vp, FREAD, 463 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 464 ZFS_EXIT(zfsvfs); 465 return (error); 466 } 467 } 468 469 /* 470 * If we're in FRSYNC mode, sync out this znode before reading it. 471 */ 472 if (ioflag & FRSYNC) 473 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 474 475 /* 476 * Lock the range against changes. 477 */ 478 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 479 480 /* 481 * If we are reading past end-of-file we can skip 482 * to the end; but we might still need to set atime. 483 */ 484 if (uio->uio_loffset >= zp->z_phys->zp_size) { 485 error = 0; 486 goto out; 487 } 488 489 ASSERT(uio->uio_loffset < zp->z_phys->zp_size); 490 n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset); 491 492 while (n > 0) { 493 nbytes = MIN(n, zfs_read_chunk_size - 494 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 495 496 if (vn_has_cached_data(vp)) 497 error = mappedread(vp, nbytes, uio); 498 else 499 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 500 if (error) 501 break; 502 503 n -= nbytes; 504 } 505 506 out: 507 zfs_range_unlock(rl); 508 509 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 510 ZFS_EXIT(zfsvfs); 511 return (error); 512 } 513 514 /* 515 * Fault in the pages of the first n bytes specified by the uio structure. 516 * 1 byte in each page is touched and the uio struct is unmodified. 517 * Any error will exit this routine as this is only a best 518 * attempt to get the pages resident. This is a copy of ufs_trans_touch(). 519 */ 520 static void 521 zfs_prefault_write(ssize_t n, struct uio *uio) 522 { 523 struct iovec *iov; 524 ulong_t cnt, incr; 525 caddr_t p; 526 uint8_t tmp; 527 528 iov = uio->uio_iov; 529 530 while (n) { 531 cnt = MIN(iov->iov_len, n); 532 if (cnt == 0) { 533 /* empty iov entry */ 534 iov++; 535 continue; 536 } 537 n -= cnt; 538 /* 539 * touch each page in this segment. 540 */ 541 p = iov->iov_base; 542 while (cnt) { 543 switch (uio->uio_segflg) { 544 case UIO_USERSPACE: 545 case UIO_USERISPACE: 546 if (fuword8(p, &tmp)) 547 return; 548 break; 549 case UIO_SYSSPACE: 550 if (kcopy(p, &tmp, 1)) 551 return; 552 break; 553 } 554 incr = MIN(cnt, PAGESIZE); 555 p += incr; 556 cnt -= incr; 557 } 558 /* 559 * touch the last byte in case it straddles a page. 560 */ 561 p--; 562 switch (uio->uio_segflg) { 563 case UIO_USERSPACE: 564 case UIO_USERISPACE: 565 if (fuword8(p, &tmp)) 566 return; 567 break; 568 case UIO_SYSSPACE: 569 if (kcopy(p, &tmp, 1)) 570 return; 571 break; 572 } 573 iov++; 574 } 575 } 576 577 /* 578 * Write the bytes to a file. 579 * 580 * IN: vp - vnode of file to be written to. 581 * uio - structure supplying write location, range info, 582 * and data buffer. 583 * ioflag - FAPPEND flag set if in append mode. 584 * cr - credentials of caller. 585 * ct - caller context (NFS/CIFS fem monitor only) 586 * 587 * OUT: uio - updated offset and range. 588 * 589 * RETURN: 0 if success 590 * error code if failure 591 * 592 * Timestamps: 593 * vp - ctime|mtime updated if byte count > 0 594 */ 595 /* ARGSUSED */ 596 static int 597 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 598 { 599 znode_t *zp = VTOZ(vp); 600 rlim64_t limit = uio->uio_llimit; 601 ssize_t start_resid = uio->uio_resid; 602 ssize_t tx_bytes; 603 uint64_t end_size; 604 dmu_tx_t *tx; 605 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 606 zilog_t *zilog; 607 offset_t woff; 608 ssize_t n, nbytes; 609 rl_t *rl; 610 int max_blksz = zfsvfs->z_max_blksz; 611 uint64_t pflags = zp->z_phys->zp_flags; 612 int error; 613 614 /* 615 * If immutable or not appending then return EPERM 616 */ 617 if ((pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 618 ((pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 619 (uio->uio_loffset < zp->z_phys->zp_size))) 620 return (EPERM); 621 622 /* 623 * Fasttrack empty write 624 */ 625 n = start_resid; 626 if (n == 0) 627 return (0); 628 629 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 630 limit = MAXOFFSET_T; 631 632 ZFS_ENTER(zfsvfs); 633 ZFS_VERIFY_ZP(zp); 634 zilog = zfsvfs->z_log; 635 636 /* 637 * Pre-fault the pages to ensure slow (eg NFS) pages 638 * don't hold up txg. 639 */ 640 zfs_prefault_write(n, uio); 641 642 /* 643 * If in append mode, set the io offset pointer to eof. 644 */ 645 if (ioflag & FAPPEND) { 646 /* 647 * Range lock for a file append: 648 * The value for the start of range will be determined by 649 * zfs_range_lock() (to guarantee append semantics). 650 * If this write will cause the block size to increase, 651 * zfs_range_lock() will lock the entire file, so we must 652 * later reduce the range after we grow the block size. 653 */ 654 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 655 if (rl->r_len == UINT64_MAX) { 656 /* overlocked, zp_size can't change */ 657 woff = uio->uio_loffset = zp->z_phys->zp_size; 658 } else { 659 woff = uio->uio_loffset = rl->r_off; 660 } 661 } else { 662 woff = uio->uio_loffset; 663 /* 664 * Validate file offset 665 */ 666 if (woff < 0) { 667 ZFS_EXIT(zfsvfs); 668 return (EINVAL); 669 } 670 671 /* 672 * If we need to grow the block size then zfs_range_lock() 673 * will lock a wider range than we request here. 674 * Later after growing the block size we reduce the range. 675 */ 676 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 677 } 678 679 if (woff >= limit) { 680 zfs_range_unlock(rl); 681 ZFS_EXIT(zfsvfs); 682 return (EFBIG); 683 } 684 685 if ((woff + n) > limit || woff > (limit - n)) 686 n = limit - woff; 687 688 /* 689 * Check for mandatory locks 690 */ 691 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 692 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 693 zfs_range_unlock(rl); 694 ZFS_EXIT(zfsvfs); 695 return (error); 696 } 697 end_size = MAX(zp->z_phys->zp_size, woff + n); 698 699 /* 700 * Write the file in reasonable size chunks. Each chunk is written 701 * in a separate transaction; this keeps the intent log records small 702 * and allows us to do more fine-grained space accounting. 703 */ 704 while (n > 0) { 705 /* 706 * Start a transaction. 707 */ 708 woff = uio->uio_loffset; 709 tx = dmu_tx_create(zfsvfs->z_os); 710 dmu_tx_hold_bonus(tx, zp->z_id); 711 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 712 error = dmu_tx_assign(tx, zfsvfs->z_assign); 713 if (error) { 714 if (error == ERESTART && 715 zfsvfs->z_assign == TXG_NOWAIT) { 716 dmu_tx_wait(tx); 717 dmu_tx_abort(tx); 718 continue; 719 } 720 dmu_tx_abort(tx); 721 break; 722 } 723 724 /* 725 * If zfs_range_lock() over-locked we grow the blocksize 726 * and then reduce the lock range. This will only happen 727 * on the first iteration since zfs_range_reduce() will 728 * shrink down r_len to the appropriate size. 729 */ 730 if (rl->r_len == UINT64_MAX) { 731 uint64_t new_blksz; 732 733 if (zp->z_blksz > max_blksz) { 734 ASSERT(!ISP2(zp->z_blksz)); 735 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 736 } else { 737 new_blksz = MIN(end_size, max_blksz); 738 } 739 zfs_grow_blocksize(zp, new_blksz, tx); 740 zfs_range_reduce(rl, woff, n); 741 } 742 743 /* 744 * XXX - should we really limit each write to z_max_blksz? 745 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 746 */ 747 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 748 rw_enter(&zp->z_map_lock, RW_READER); 749 750 tx_bytes = uio->uio_resid; 751 if (vn_has_cached_data(vp)) { 752 rw_exit(&zp->z_map_lock); 753 error = mappedwrite(vp, nbytes, uio, tx); 754 } else { 755 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 756 uio, nbytes, tx); 757 rw_exit(&zp->z_map_lock); 758 } 759 tx_bytes -= uio->uio_resid; 760 761 /* 762 * If we made no progress, we're done. If we made even 763 * partial progress, update the znode and ZIL accordingly. 764 */ 765 if (tx_bytes == 0) { 766 dmu_tx_commit(tx); 767 ASSERT(error != 0); 768 break; 769 } 770 771 /* 772 * Clear Set-UID/Set-GID bits on successful write if not 773 * privileged and at least one of the excute bits is set. 774 * 775 * It would be nice to to this after all writes have 776 * been done, but that would still expose the ISUID/ISGID 777 * to another app after the partial write is committed. 778 * 779 * Note: we don't call zfs_fuid_map_id() here because 780 * user 0 is not an ephemeral uid. 781 */ 782 mutex_enter(&zp->z_acl_lock); 783 if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) | 784 (S_IXUSR >> 6))) != 0 && 785 (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 && 786 secpolicy_vnode_setid_retain(cr, 787 (zp->z_phys->zp_mode & S_ISUID) != 0 && 788 zp->z_phys->zp_uid == 0) != 0) { 789 zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID); 790 } 791 mutex_exit(&zp->z_acl_lock); 792 793 /* 794 * Update time stamp. NOTE: This marks the bonus buffer as 795 * dirty, so we don't have to do it again for zp_size. 796 */ 797 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 798 799 /* 800 * Update the file size (zp_size) if it has changed; 801 * account for possible concurrent updates. 802 */ 803 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) 804 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 805 uio->uio_loffset); 806 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 807 dmu_tx_commit(tx); 808 809 if (error != 0) 810 break; 811 ASSERT(tx_bytes == nbytes); 812 n -= nbytes; 813 } 814 815 zfs_range_unlock(rl); 816 817 /* 818 * If we're in replay mode, or we made no progress, return error. 819 * Otherwise, it's at least a partial write, so it's successful. 820 */ 821 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 822 ZFS_EXIT(zfsvfs); 823 return (error); 824 } 825 826 if (ioflag & (FSYNC | FDSYNC)) 827 zil_commit(zilog, zp->z_last_itx, zp->z_id); 828 829 ZFS_EXIT(zfsvfs); 830 return (0); 831 } 832 833 void 834 zfs_get_done(dmu_buf_t *db, void *vzgd) 835 { 836 zgd_t *zgd = (zgd_t *)vzgd; 837 rl_t *rl = zgd->zgd_rl; 838 vnode_t *vp = ZTOV(rl->r_zp); 839 840 dmu_buf_rele(db, vzgd); 841 zfs_range_unlock(rl); 842 VN_RELE(vp); 843 zil_add_vdev(zgd->zgd_zilog, DVA_GET_VDEV(BP_IDENTITY(zgd->zgd_bp))); 844 kmem_free(zgd, sizeof (zgd_t)); 845 } 846 847 /* 848 * Get data to generate a TX_WRITE intent log record. 849 */ 850 int 851 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 852 { 853 zfsvfs_t *zfsvfs = arg; 854 objset_t *os = zfsvfs->z_os; 855 znode_t *zp; 856 uint64_t off = lr->lr_offset; 857 dmu_buf_t *db; 858 rl_t *rl; 859 zgd_t *zgd; 860 int dlen = lr->lr_length; /* length of user data */ 861 int error = 0; 862 863 ASSERT(zio); 864 ASSERT(dlen != 0); 865 866 /* 867 * Nothing to do if the file has been removed 868 */ 869 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 870 return (ENOENT); 871 if (zp->z_unlinked) { 872 VN_RELE(ZTOV(zp)); 873 return (ENOENT); 874 } 875 876 /* 877 * Write records come in two flavors: immediate and indirect. 878 * For small writes it's cheaper to store the data with the 879 * log record (immediate); for large writes it's cheaper to 880 * sync the data and get a pointer to it (indirect) so that 881 * we don't have to write the data twice. 882 */ 883 if (buf != NULL) { /* immediate write */ 884 rl = zfs_range_lock(zp, off, dlen, RL_READER); 885 /* test for truncation needs to be done while range locked */ 886 if (off >= zp->z_phys->zp_size) { 887 error = ENOENT; 888 goto out; 889 } 890 VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf)); 891 } else { /* indirect write */ 892 uint64_t boff; /* block starting offset */ 893 894 /* 895 * Have to lock the whole block to ensure when it's 896 * written out and it's checksum is being calculated 897 * that no one can change the data. We need to re-check 898 * blocksize after we get the lock in case it's changed! 899 */ 900 for (;;) { 901 if (ISP2(zp->z_blksz)) { 902 boff = P2ALIGN_TYPED(off, zp->z_blksz, 903 uint64_t); 904 } else { 905 boff = 0; 906 } 907 dlen = zp->z_blksz; 908 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 909 if (zp->z_blksz == dlen) 910 break; 911 zfs_range_unlock(rl); 912 } 913 /* test for truncation needs to be done while range locked */ 914 if (off >= zp->z_phys->zp_size) { 915 error = ENOENT; 916 goto out; 917 } 918 zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP); 919 zgd->zgd_rl = rl; 920 zgd->zgd_zilog = zfsvfs->z_log; 921 zgd->zgd_bp = &lr->lr_blkptr; 922 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db)); 923 ASSERT(boff == db->db_offset); 924 lr->lr_blkoff = off - boff; 925 error = dmu_sync(zio, db, &lr->lr_blkptr, 926 lr->lr_common.lrc_txg, zfs_get_done, zgd); 927 ASSERT((error && error != EINPROGRESS) || 928 lr->lr_length <= zp->z_blksz); 929 if (error == 0) { 930 zil_add_vdev(zfsvfs->z_log, 931 DVA_GET_VDEV(BP_IDENTITY(&lr->lr_blkptr))); 932 } 933 /* 934 * If we get EINPROGRESS, then we need to wait for a 935 * write IO initiated by dmu_sync() to complete before 936 * we can release this dbuf. We will finish everything 937 * up in the zfs_get_done() callback. 938 */ 939 if (error == EINPROGRESS) 940 return (0); 941 dmu_buf_rele(db, zgd); 942 kmem_free(zgd, sizeof (zgd_t)); 943 } 944 out: 945 zfs_range_unlock(rl); 946 VN_RELE(ZTOV(zp)); 947 return (error); 948 } 949 950 /*ARGSUSED*/ 951 static int 952 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 953 caller_context_t *ct) 954 { 955 znode_t *zp = VTOZ(vp); 956 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 957 int error; 958 959 ZFS_ENTER(zfsvfs); 960 ZFS_VERIFY_ZP(zp); 961 962 if (flag & V_ACE_MASK) 963 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 964 else 965 error = zfs_zaccess_rwx(zp, mode, flag, cr); 966 967 ZFS_EXIT(zfsvfs); 968 return (error); 969 } 970 971 /* 972 * Lookup an entry in a directory, or an extended attribute directory. 973 * If it exists, return a held vnode reference for it. 974 * 975 * IN: dvp - vnode of directory to search. 976 * nm - name of entry to lookup. 977 * pnp - full pathname to lookup [UNUSED]. 978 * flags - LOOKUP_XATTR set if looking for an attribute. 979 * rdir - root directory vnode [UNUSED]. 980 * cr - credentials of caller. 981 * ct - caller context 982 * direntflags - directory lookup flags 983 * realpnp - returned pathname. 984 * 985 * OUT: vpp - vnode of located entry, NULL if not found. 986 * 987 * RETURN: 0 if success 988 * error code if failure 989 * 990 * Timestamps: 991 * NA 992 */ 993 /* ARGSUSED */ 994 static int 995 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 996 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 997 int *direntflags, pathname_t *realpnp) 998 { 999 znode_t *zdp = VTOZ(dvp); 1000 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1001 int error; 1002 1003 ZFS_ENTER(zfsvfs); 1004 ZFS_VERIFY_ZP(zdp); 1005 1006 *vpp = NULL; 1007 1008 if (flags & LOOKUP_XATTR) { 1009 /* 1010 * If the xattr property is off, refuse the lookup request. 1011 */ 1012 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1013 ZFS_EXIT(zfsvfs); 1014 return (EINVAL); 1015 } 1016 1017 /* 1018 * We don't allow recursive attributes.. 1019 * Maybe someday we will. 1020 */ 1021 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 1022 ZFS_EXIT(zfsvfs); 1023 return (EINVAL); 1024 } 1025 1026 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1027 ZFS_EXIT(zfsvfs); 1028 return (error); 1029 } 1030 1031 /* 1032 * Do we have permission to get into attribute directory? 1033 */ 1034 1035 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1036 B_FALSE, cr)) { 1037 VN_RELE(*vpp); 1038 *vpp = NULL; 1039 } 1040 1041 ZFS_EXIT(zfsvfs); 1042 return (error); 1043 } 1044 1045 if (dvp->v_type != VDIR) { 1046 ZFS_EXIT(zfsvfs); 1047 return (ENOTDIR); 1048 } 1049 1050 /* 1051 * Check accessibility of directory. 1052 */ 1053 1054 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1055 ZFS_EXIT(zfsvfs); 1056 return (error); 1057 } 1058 1059 if (zfsvfs->z_case & ZFS_UTF8_ONLY && u8_validate(nm, strlen(nm), 1060 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1061 ZFS_EXIT(zfsvfs); 1062 return (EILSEQ); 1063 } 1064 1065 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1066 if (error == 0) { 1067 /* 1068 * Convert device special files 1069 */ 1070 if (IS_DEVVP(*vpp)) { 1071 vnode_t *svp; 1072 1073 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1074 VN_RELE(*vpp); 1075 if (svp == NULL) 1076 error = ENOSYS; 1077 else 1078 *vpp = svp; 1079 } 1080 } 1081 1082 ZFS_EXIT(zfsvfs); 1083 return (error); 1084 } 1085 1086 /* 1087 * Attempt to create a new entry in a directory. If the entry 1088 * already exists, truncate the file if permissible, else return 1089 * an error. Return the vp of the created or trunc'd file. 1090 * 1091 * IN: dvp - vnode of directory to put new file entry in. 1092 * name - name of new file entry. 1093 * vap - attributes of new file. 1094 * excl - flag indicating exclusive or non-exclusive mode. 1095 * mode - mode to open file with. 1096 * cr - credentials of caller. 1097 * flag - large file flag [UNUSED]. 1098 * ct - caller context 1099 * vsecp - ACL to be set 1100 * 1101 * OUT: vpp - vnode of created or trunc'd entry. 1102 * 1103 * RETURN: 0 if success 1104 * error code if failure 1105 * 1106 * Timestamps: 1107 * dvp - ctime|mtime updated if new entry created 1108 * vp - ctime|mtime always, atime if new 1109 */ 1110 1111 /* ARGSUSED */ 1112 static int 1113 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1114 int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct, 1115 vsecattr_t *vsecp) 1116 { 1117 znode_t *zp, *dzp = VTOZ(dvp); 1118 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1119 zilog_t *zilog; 1120 objset_t *os; 1121 zfs_dirlock_t *dl; 1122 dmu_tx_t *tx; 1123 int error; 1124 uint64_t zoid; 1125 zfs_acl_t *aclp = NULL; 1126 zfs_fuid_info_t *fuidp = NULL; 1127 1128 /* 1129 * If we have an ephemeral id, ACL, or XVATTR then 1130 * make sure file system is at proper version 1131 */ 1132 1133 if (zfsvfs->z_use_fuids == B_FALSE && 1134 (vsecp || (vap->va_mask & AT_XVATTR) || 1135 IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr)))) 1136 return (EINVAL); 1137 1138 ZFS_ENTER(zfsvfs); 1139 ZFS_VERIFY_ZP(dzp); 1140 os = zfsvfs->z_os; 1141 zilog = zfsvfs->z_log; 1142 1143 if (zfsvfs->z_case & ZFS_UTF8_ONLY && u8_validate(name, strlen(name), 1144 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1145 ZFS_EXIT(zfsvfs); 1146 return (EILSEQ); 1147 } 1148 1149 if (vap->va_mask & AT_XVATTR) { 1150 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1151 crgetuid(cr), cr, vap->va_type)) != 0) { 1152 ZFS_EXIT(zfsvfs); 1153 return (error); 1154 } 1155 } 1156 top: 1157 *vpp = NULL; 1158 1159 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1160 vap->va_mode &= ~VSVTX; 1161 1162 if (*name == '\0') { 1163 /* 1164 * Null component name refers to the directory itself. 1165 */ 1166 VN_HOLD(dvp); 1167 zp = dzp; 1168 dl = NULL; 1169 error = 0; 1170 } else { 1171 /* possible VN_HOLD(zp) */ 1172 int zflg = 0; 1173 1174 if (flag & FIGNORECASE) 1175 zflg |= ZCILOOK; 1176 1177 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1178 NULL, NULL); 1179 if (error) { 1180 if (strcmp(name, "..") == 0) 1181 error = EISDIR; 1182 ZFS_EXIT(zfsvfs); 1183 if (aclp) 1184 zfs_acl_free(aclp); 1185 return (error); 1186 } 1187 } 1188 if (vsecp && aclp == NULL) { 1189 error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp); 1190 if (error) { 1191 ZFS_EXIT(zfsvfs); 1192 if (dl) 1193 zfs_dirent_unlock(dl); 1194 return (error); 1195 } 1196 } 1197 zoid = zp ? zp->z_id : -1ULL; 1198 1199 if (zp == NULL) { 1200 uint64_t txtype; 1201 1202 /* 1203 * Create a new file object and update the directory 1204 * to reference it. 1205 */ 1206 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1207 goto out; 1208 } 1209 1210 /* 1211 * We only support the creation of regular files in 1212 * extended attribute directories. 1213 */ 1214 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1215 (vap->va_type != VREG)) { 1216 error = EINVAL; 1217 goto out; 1218 } 1219 1220 tx = dmu_tx_create(os); 1221 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1222 if (zfsvfs->z_fuid_obj == 0) { 1223 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1224 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1225 SPA_MAXBLOCKSIZE); 1226 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 1227 } else { 1228 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 1229 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 1230 SPA_MAXBLOCKSIZE); 1231 } 1232 dmu_tx_hold_bonus(tx, dzp->z_id); 1233 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1234 if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp) { 1235 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1236 0, SPA_MAXBLOCKSIZE); 1237 } 1238 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1239 if (error) { 1240 zfs_dirent_unlock(dl); 1241 if (error == ERESTART && 1242 zfsvfs->z_assign == TXG_NOWAIT) { 1243 dmu_tx_wait(tx); 1244 dmu_tx_abort(tx); 1245 goto top; 1246 } 1247 dmu_tx_abort(tx); 1248 ZFS_EXIT(zfsvfs); 1249 if (aclp) 1250 zfs_acl_free(aclp); 1251 return (error); 1252 } 1253 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0, aclp, &fuidp); 1254 ASSERT(zp->z_id == zoid); 1255 (void) zfs_link_create(dl, zp, tx, ZNEW); 1256 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1257 if (flag & FIGNORECASE) 1258 txtype |= TX_CI; 1259 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1260 vsecp, fuidp, vap); 1261 if (fuidp) 1262 zfs_fuid_info_free(fuidp); 1263 dmu_tx_commit(tx); 1264 } else { 1265 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1266 1267 /* 1268 * A directory entry already exists for this name. 1269 */ 1270 /* 1271 * Can't truncate an existing file if in exclusive mode. 1272 */ 1273 if (excl == EXCL) { 1274 error = EEXIST; 1275 goto out; 1276 } 1277 /* 1278 * Can't open a directory for writing. 1279 */ 1280 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1281 error = EISDIR; 1282 goto out; 1283 } 1284 /* 1285 * Verify requested access to file. 1286 */ 1287 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1288 goto out; 1289 } 1290 1291 mutex_enter(&dzp->z_lock); 1292 dzp->z_seq++; 1293 mutex_exit(&dzp->z_lock); 1294 1295 /* 1296 * Truncate regular files if requested. 1297 */ 1298 if ((ZTOV(zp)->v_type == VREG) && 1299 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1300 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1301 if (error == ERESTART && 1302 zfsvfs->z_assign == TXG_NOWAIT) { 1303 /* NB: we already did dmu_tx_wait() */ 1304 zfs_dirent_unlock(dl); 1305 VN_RELE(ZTOV(zp)); 1306 goto top; 1307 } 1308 1309 if (error == 0) { 1310 vnevent_create(ZTOV(zp), ct); 1311 } 1312 } 1313 } 1314 out: 1315 1316 if (dl) 1317 zfs_dirent_unlock(dl); 1318 1319 if (error) { 1320 if (zp) 1321 VN_RELE(ZTOV(zp)); 1322 } else { 1323 *vpp = ZTOV(zp); 1324 /* 1325 * If vnode is for a device return a specfs vnode instead. 1326 */ 1327 if (IS_DEVVP(*vpp)) { 1328 struct vnode *svp; 1329 1330 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1331 VN_RELE(*vpp); 1332 if (svp == NULL) { 1333 error = ENOSYS; 1334 } 1335 *vpp = svp; 1336 } 1337 } 1338 if (aclp) 1339 zfs_acl_free(aclp); 1340 1341 ZFS_EXIT(zfsvfs); 1342 return (error); 1343 } 1344 1345 /* 1346 * Remove an entry from a directory. 1347 * 1348 * IN: dvp - vnode of directory to remove entry from. 1349 * name - name of entry to remove. 1350 * cr - credentials of caller. 1351 * ct - caller context 1352 * flags - case flags 1353 * 1354 * RETURN: 0 if success 1355 * error code if failure 1356 * 1357 * Timestamps: 1358 * dvp - ctime|mtime 1359 * vp - ctime (if nlink > 0) 1360 */ 1361 /*ARGSUSED*/ 1362 static int 1363 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1364 int flags) 1365 { 1366 znode_t *zp, *dzp = VTOZ(dvp); 1367 znode_t *xzp = NULL; 1368 vnode_t *vp; 1369 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1370 zilog_t *zilog; 1371 uint64_t acl_obj, xattr_obj; 1372 zfs_dirlock_t *dl; 1373 dmu_tx_t *tx; 1374 boolean_t may_delete_now, delete_now = FALSE; 1375 boolean_t unlinked; 1376 uint64_t txtype; 1377 pathname_t *realnmp = NULL; 1378 pathname_t realnm; 1379 int error; 1380 int zflg = ZEXISTS; 1381 1382 ZFS_ENTER(zfsvfs); 1383 ZFS_VERIFY_ZP(dzp); 1384 zilog = zfsvfs->z_log; 1385 1386 if (flags & FIGNORECASE) { 1387 zflg |= ZCILOOK; 1388 pn_alloc(&realnm); 1389 realnmp = &realnm; 1390 } 1391 1392 top: 1393 /* 1394 * Attempt to lock directory; fail if entry doesn't exist. 1395 */ 1396 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1397 NULL, realnmp)) { 1398 if (realnmp) 1399 pn_free(realnmp); 1400 ZFS_EXIT(zfsvfs); 1401 return (error); 1402 } 1403 1404 vp = ZTOV(zp); 1405 1406 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1407 goto out; 1408 } 1409 1410 /* 1411 * Need to use rmdir for removing directories. 1412 */ 1413 if (vp->v_type == VDIR) { 1414 error = EPERM; 1415 goto out; 1416 } 1417 1418 vnevent_remove(vp, dvp, name, ct); 1419 1420 if (realnmp) 1421 dnlc_remove(dvp, realnmp->pn_path); 1422 else 1423 dnlc_remove(dvp, name); 1424 1425 mutex_enter(&vp->v_lock); 1426 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1427 mutex_exit(&vp->v_lock); 1428 1429 /* 1430 * We may delete the znode now, or we may put it in the unlinked set; 1431 * it depends on whether we're the last link, and on whether there are 1432 * other holds on the vnode. So we dmu_tx_hold() the right things to 1433 * allow for either case. 1434 */ 1435 tx = dmu_tx_create(zfsvfs->z_os); 1436 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1437 dmu_tx_hold_bonus(tx, zp->z_id); 1438 if (may_delete_now) 1439 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 1440 1441 /* are there any extended attributes? */ 1442 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1443 /* XXX - do we need this if we are deleting? */ 1444 dmu_tx_hold_bonus(tx, xattr_obj); 1445 } 1446 1447 /* are there any additional acls */ 1448 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1449 may_delete_now) 1450 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1451 1452 /* charge as an update -- would be nice not to charge at all */ 1453 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1454 1455 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1456 if (error) { 1457 zfs_dirent_unlock(dl); 1458 VN_RELE(vp); 1459 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1460 dmu_tx_wait(tx); 1461 dmu_tx_abort(tx); 1462 goto top; 1463 } 1464 if (realnmp) 1465 pn_free(realnmp); 1466 dmu_tx_abort(tx); 1467 ZFS_EXIT(zfsvfs); 1468 return (error); 1469 } 1470 1471 /* 1472 * Remove the directory entry. 1473 */ 1474 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1475 1476 if (error) { 1477 dmu_tx_commit(tx); 1478 goto out; 1479 } 1480 1481 if (unlinked) { 1482 mutex_enter(&vp->v_lock); 1483 delete_now = may_delete_now && 1484 vp->v_count == 1 && !vn_has_cached_data(vp) && 1485 zp->z_phys->zp_xattr == xattr_obj && 1486 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1487 mutex_exit(&vp->v_lock); 1488 } 1489 1490 if (delete_now) { 1491 if (zp->z_phys->zp_xattr) { 1492 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1493 ASSERT3U(error, ==, 0); 1494 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1495 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1496 mutex_enter(&xzp->z_lock); 1497 xzp->z_unlinked = 1; 1498 xzp->z_phys->zp_links = 0; 1499 mutex_exit(&xzp->z_lock); 1500 zfs_unlinked_add(xzp, tx); 1501 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1502 } 1503 mutex_enter(&zp->z_lock); 1504 mutex_enter(&vp->v_lock); 1505 vp->v_count--; 1506 ASSERT3U(vp->v_count, ==, 0); 1507 mutex_exit(&vp->v_lock); 1508 mutex_exit(&zp->z_lock); 1509 zfs_znode_delete(zp, tx); 1510 VFS_RELE(zfsvfs->z_vfs); 1511 } else if (unlinked) { 1512 zfs_unlinked_add(zp, tx); 1513 } 1514 1515 txtype = TX_REMOVE; 1516 if (flags & FIGNORECASE) 1517 txtype |= TX_CI; 1518 zfs_log_remove(zilog, tx, txtype, dzp, name); 1519 1520 dmu_tx_commit(tx); 1521 out: 1522 if (realnmp) 1523 pn_free(realnmp); 1524 1525 zfs_dirent_unlock(dl); 1526 1527 if (!delete_now) { 1528 VN_RELE(vp); 1529 } else if (xzp) { 1530 /* this rele delayed to prevent nesting transactions */ 1531 VN_RELE(ZTOV(xzp)); 1532 } 1533 1534 ZFS_EXIT(zfsvfs); 1535 return (error); 1536 } 1537 1538 /* 1539 * Create a new directory and insert it into dvp using the name 1540 * provided. Return a pointer to the inserted directory. 1541 * 1542 * IN: dvp - vnode of directory to add subdir to. 1543 * dirname - name of new directory. 1544 * vap - attributes of new directory. 1545 * cr - credentials of caller. 1546 * ct - caller context 1547 * vsecp - ACL to be set 1548 * 1549 * OUT: vpp - vnode of created directory. 1550 * 1551 * RETURN: 0 if success 1552 * error code if failure 1553 * 1554 * Timestamps: 1555 * dvp - ctime|mtime updated 1556 * vp - ctime|mtime|atime updated 1557 */ 1558 /*ARGSUSED*/ 1559 static int 1560 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 1561 caller_context_t *ct, int flags, vsecattr_t *vsecp) 1562 { 1563 znode_t *zp, *dzp = VTOZ(dvp); 1564 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1565 zilog_t *zilog; 1566 zfs_dirlock_t *dl; 1567 uint64_t zoid = 0; 1568 uint64_t txtype; 1569 dmu_tx_t *tx; 1570 int error; 1571 zfs_acl_t *aclp = NULL; 1572 zfs_fuid_info_t *fuidp = NULL; 1573 int zf = ZNEW; 1574 1575 ASSERT(vap->va_type == VDIR); 1576 1577 /* 1578 * If we have an ephemeral id, ACL, or XVATTR then 1579 * make sure file system is at proper version 1580 */ 1581 1582 if (zfsvfs->z_use_fuids == B_FALSE && 1583 (vsecp || (vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(crgetuid(cr))|| 1584 IS_EPHEMERAL(crgetgid(cr)))) 1585 return (EINVAL); 1586 1587 ZFS_ENTER(zfsvfs); 1588 ZFS_VERIFY_ZP(dzp); 1589 zilog = zfsvfs->z_log; 1590 1591 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1592 ZFS_EXIT(zfsvfs); 1593 return (EINVAL); 1594 } 1595 1596 if (zfsvfs->z_case & ZFS_UTF8_ONLY && u8_validate(dirname, 1597 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1598 ZFS_EXIT(zfsvfs); 1599 return (EILSEQ); 1600 } 1601 if (flags & FIGNORECASE) 1602 zf |= ZCILOOK; 1603 1604 if (vap->va_mask & AT_XVATTR) 1605 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1606 crgetuid(cr), cr, vap->va_type)) != 0) { 1607 ZFS_EXIT(zfsvfs); 1608 return (error); 1609 } 1610 1611 /* 1612 * First make sure the new directory doesn't exist. 1613 */ 1614 top: 1615 *vpp = NULL; 1616 1617 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 1618 NULL, NULL)) { 1619 ZFS_EXIT(zfsvfs); 1620 return (error); 1621 } 1622 1623 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 1624 zfs_dirent_unlock(dl); 1625 ZFS_EXIT(zfsvfs); 1626 return (error); 1627 } 1628 1629 if (vsecp && aclp == NULL) { 1630 error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp); 1631 if (error) { 1632 zfs_dirent_unlock(dl); 1633 ZFS_EXIT(zfsvfs); 1634 return (error); 1635 } 1636 } 1637 /* 1638 * Add a new entry to the directory. 1639 */ 1640 tx = dmu_tx_create(zfsvfs->z_os); 1641 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1642 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1643 if (zfsvfs->z_fuid_obj == 0) { 1644 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1645 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1646 SPA_MAXBLOCKSIZE); 1647 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 1648 } else { 1649 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 1650 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 1651 SPA_MAXBLOCKSIZE); 1652 } 1653 if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp) 1654 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1655 0, SPA_MAXBLOCKSIZE); 1656 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1657 if (error) { 1658 zfs_dirent_unlock(dl); 1659 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1660 dmu_tx_wait(tx); 1661 dmu_tx_abort(tx); 1662 goto top; 1663 } 1664 dmu_tx_abort(tx); 1665 ZFS_EXIT(zfsvfs); 1666 if (aclp) 1667 zfs_acl_free(aclp); 1668 return (error); 1669 } 1670 1671 /* 1672 * Create new node. 1673 */ 1674 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0, aclp, &fuidp); 1675 1676 if (aclp) 1677 zfs_acl_free(aclp); 1678 1679 /* 1680 * Now put new name in parent dir. 1681 */ 1682 (void) zfs_link_create(dl, zp, tx, ZNEW); 1683 1684 *vpp = ZTOV(zp); 1685 1686 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 1687 if (flags & FIGNORECASE) 1688 txtype |= TX_CI; 1689 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, fuidp, vap); 1690 1691 if (fuidp) 1692 zfs_fuid_info_free(fuidp); 1693 dmu_tx_commit(tx); 1694 1695 zfs_dirent_unlock(dl); 1696 1697 ZFS_EXIT(zfsvfs); 1698 return (0); 1699 } 1700 1701 /* 1702 * Remove a directory subdir entry. If the current working 1703 * directory is the same as the subdir to be removed, the 1704 * remove will fail. 1705 * 1706 * IN: dvp - vnode of directory to remove from. 1707 * name - name of directory to be removed. 1708 * cwd - vnode of current working directory. 1709 * cr - credentials of caller. 1710 * ct - caller context 1711 * flags - case flags 1712 * 1713 * RETURN: 0 if success 1714 * error code if failure 1715 * 1716 * Timestamps: 1717 * dvp - ctime|mtime updated 1718 */ 1719 /*ARGSUSED*/ 1720 static int 1721 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 1722 caller_context_t *ct, int flags) 1723 { 1724 znode_t *dzp = VTOZ(dvp); 1725 znode_t *zp; 1726 vnode_t *vp; 1727 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1728 zilog_t *zilog; 1729 zfs_dirlock_t *dl; 1730 dmu_tx_t *tx; 1731 int error; 1732 int zflg = ZEXISTS; 1733 1734 ZFS_ENTER(zfsvfs); 1735 ZFS_VERIFY_ZP(dzp); 1736 zilog = zfsvfs->z_log; 1737 1738 if (flags & FIGNORECASE) 1739 zflg |= ZCILOOK; 1740 top: 1741 zp = NULL; 1742 1743 /* 1744 * Attempt to lock directory; fail if entry doesn't exist. 1745 */ 1746 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1747 NULL, NULL)) { 1748 ZFS_EXIT(zfsvfs); 1749 return (error); 1750 } 1751 1752 vp = ZTOV(zp); 1753 1754 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1755 goto out; 1756 } 1757 1758 if (vp->v_type != VDIR) { 1759 error = ENOTDIR; 1760 goto out; 1761 } 1762 1763 if (vp == cwd) { 1764 error = EINVAL; 1765 goto out; 1766 } 1767 1768 vnevent_rmdir(vp, dvp, name, ct); 1769 1770 /* 1771 * Grab a lock on the directory to make sure that noone is 1772 * trying to add (or lookup) entries while we are removing it. 1773 */ 1774 rw_enter(&zp->z_name_lock, RW_WRITER); 1775 1776 /* 1777 * Grab a lock on the parent pointer to make sure we play well 1778 * with the treewalk and directory rename code. 1779 */ 1780 rw_enter(&zp->z_parent_lock, RW_WRITER); 1781 1782 tx = dmu_tx_create(zfsvfs->z_os); 1783 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1784 dmu_tx_hold_bonus(tx, zp->z_id); 1785 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1786 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1787 if (error) { 1788 rw_exit(&zp->z_parent_lock); 1789 rw_exit(&zp->z_name_lock); 1790 zfs_dirent_unlock(dl); 1791 VN_RELE(vp); 1792 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1793 dmu_tx_wait(tx); 1794 dmu_tx_abort(tx); 1795 goto top; 1796 } 1797 dmu_tx_abort(tx); 1798 ZFS_EXIT(zfsvfs); 1799 return (error); 1800 } 1801 1802 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 1803 1804 if (error == 0) { 1805 uint64_t txtype = TX_RMDIR; 1806 if (flags & FIGNORECASE) 1807 txtype |= TX_CI; 1808 zfs_log_remove(zilog, tx, txtype, dzp, name); 1809 } 1810 1811 dmu_tx_commit(tx); 1812 1813 rw_exit(&zp->z_parent_lock); 1814 rw_exit(&zp->z_name_lock); 1815 out: 1816 zfs_dirent_unlock(dl); 1817 1818 VN_RELE(vp); 1819 1820 ZFS_EXIT(zfsvfs); 1821 return (error); 1822 } 1823 1824 /* 1825 * Read as many directory entries as will fit into the provided 1826 * buffer from the given directory cursor position (specified in 1827 * the uio structure. 1828 * 1829 * IN: vp - vnode of directory to read. 1830 * uio - structure supplying read location, range info, 1831 * and return buffer. 1832 * cr - credentials of caller. 1833 * ct - caller context 1834 * flags - case flags 1835 * 1836 * OUT: uio - updated offset and range, buffer filled. 1837 * eofp - set to true if end-of-file detected. 1838 * 1839 * RETURN: 0 if success 1840 * error code if failure 1841 * 1842 * Timestamps: 1843 * vp - atime updated 1844 * 1845 * Note that the low 4 bits of the cookie returned by zap is always zero. 1846 * This allows us to use the low range for "special" directory entries: 1847 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1848 * we use the offset 2 for the '.zfs' directory. 1849 */ 1850 /* ARGSUSED */ 1851 static int 1852 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, 1853 caller_context_t *ct, int flags) 1854 { 1855 znode_t *zp = VTOZ(vp); 1856 iovec_t *iovp; 1857 edirent_t *eodp; 1858 dirent64_t *odp; 1859 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1860 objset_t *os; 1861 caddr_t outbuf; 1862 size_t bufsize; 1863 zap_cursor_t zc; 1864 zap_attribute_t zap; 1865 uint_t bytes_wanted; 1866 uint64_t offset; /* must be unsigned; checks for < 1 */ 1867 int local_eof; 1868 int outcount; 1869 int error; 1870 uint8_t prefetch; 1871 1872 ZFS_ENTER(zfsvfs); 1873 ZFS_VERIFY_ZP(zp); 1874 1875 /* 1876 * If we are not given an eof variable, 1877 * use a local one. 1878 */ 1879 if (eofp == NULL) 1880 eofp = &local_eof; 1881 1882 /* 1883 * Check for valid iov_len. 1884 */ 1885 if (uio->uio_iov->iov_len <= 0) { 1886 ZFS_EXIT(zfsvfs); 1887 return (EINVAL); 1888 } 1889 1890 /* 1891 * Quit if directory has been removed (posix) 1892 */ 1893 if ((*eofp = zp->z_unlinked) != 0) { 1894 ZFS_EXIT(zfsvfs); 1895 return (0); 1896 } 1897 1898 error = 0; 1899 os = zfsvfs->z_os; 1900 offset = uio->uio_loffset; 1901 prefetch = zp->z_zn_prefetch; 1902 1903 /* 1904 * Initialize the iterator cursor. 1905 */ 1906 if (offset <= 3) { 1907 /* 1908 * Start iteration from the beginning of the directory. 1909 */ 1910 zap_cursor_init(&zc, os, zp->z_id); 1911 } else { 1912 /* 1913 * The offset is a serialized cursor. 1914 */ 1915 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1916 } 1917 1918 /* 1919 * Get space to change directory entries into fs independent format. 1920 */ 1921 iovp = uio->uio_iov; 1922 bytes_wanted = iovp->iov_len; 1923 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1924 bufsize = bytes_wanted; 1925 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1926 odp = (struct dirent64 *)outbuf; 1927 } else { 1928 bufsize = bytes_wanted; 1929 odp = (struct dirent64 *)iovp->iov_base; 1930 } 1931 eodp = (struct edirent *)odp; 1932 1933 /* 1934 * Transform to file-system independent format 1935 */ 1936 outcount = 0; 1937 while (outcount < bytes_wanted) { 1938 ino64_t objnum; 1939 ushort_t reclen; 1940 off64_t *next; 1941 1942 /* 1943 * Special case `.', `..', and `.zfs'. 1944 */ 1945 if (offset == 0) { 1946 (void) strcpy(zap.za_name, "."); 1947 zap.za_normalization_conflict = 0; 1948 objnum = zp->z_id; 1949 } else if (offset == 1) { 1950 (void) strcpy(zap.za_name, ".."); 1951 zap.za_normalization_conflict = 0; 1952 objnum = zp->z_phys->zp_parent; 1953 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1954 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1955 zap.za_normalization_conflict = 0; 1956 objnum = ZFSCTL_INO_ROOT; 1957 } else { 1958 /* 1959 * Grab next entry. 1960 */ 1961 if (error = zap_cursor_retrieve(&zc, &zap)) { 1962 if ((*eofp = (error == ENOENT)) != 0) 1963 break; 1964 else 1965 goto update; 1966 } 1967 1968 if (zap.za_integer_length != 8 || 1969 zap.za_num_integers != 1) { 1970 cmn_err(CE_WARN, "zap_readdir: bad directory " 1971 "entry, obj = %lld, offset = %lld\n", 1972 (u_longlong_t)zp->z_id, 1973 (u_longlong_t)offset); 1974 error = ENXIO; 1975 goto update; 1976 } 1977 1978 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 1979 /* 1980 * MacOS X can extract the object type here such as: 1981 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1982 */ 1983 } 1984 1985 if (flags & V_RDDIR_ENTFLAGS) 1986 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 1987 else 1988 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1989 1990 /* 1991 * Will this entry fit in the buffer? 1992 */ 1993 if (outcount + reclen > bufsize) { 1994 /* 1995 * Did we manage to fit anything in the buffer? 1996 */ 1997 if (!outcount) { 1998 error = EINVAL; 1999 goto update; 2000 } 2001 break; 2002 } 2003 if (flags & V_RDDIR_ENTFLAGS) { 2004 /* 2005 * Add extended flag entry: 2006 */ 2007 eodp->ed_ino = objnum; 2008 eodp->ed_reclen = reclen; 2009 /* NOTE: ed_off is the offset for the *next* entry */ 2010 next = &(eodp->ed_off); 2011 eodp->ed_eflags = zap.za_normalization_conflict ? 2012 ED_CASE_CONFLICT : 0; 2013 (void) strncpy(eodp->ed_name, zap.za_name, 2014 EDIRENT_NAMELEN(reclen)); 2015 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2016 } else { 2017 /* 2018 * Add normal entry: 2019 */ 2020 odp->d_ino = objnum; 2021 odp->d_reclen = reclen; 2022 /* NOTE: d_off is the offset for the *next* entry */ 2023 next = &(odp->d_off); 2024 (void) strncpy(odp->d_name, zap.za_name, 2025 DIRENT64_NAMELEN(reclen)); 2026 odp = (dirent64_t *)((intptr_t)odp + reclen); 2027 } 2028 outcount += reclen; 2029 2030 ASSERT(outcount <= bufsize); 2031 2032 /* Prefetch znode */ 2033 if (prefetch) 2034 dmu_prefetch(os, objnum, 0, 0); 2035 2036 /* 2037 * Move to the next entry, fill in the previous offset. 2038 */ 2039 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2040 zap_cursor_advance(&zc); 2041 offset = zap_cursor_serialize(&zc); 2042 } else { 2043 offset += 1; 2044 } 2045 *next = offset; 2046 } 2047 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2048 2049 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2050 iovp->iov_base += outcount; 2051 iovp->iov_len -= outcount; 2052 uio->uio_resid -= outcount; 2053 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2054 /* 2055 * Reset the pointer. 2056 */ 2057 offset = uio->uio_loffset; 2058 } 2059 2060 update: 2061 zap_cursor_fini(&zc); 2062 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2063 kmem_free(outbuf, bufsize); 2064 2065 if (error == ENOENT) 2066 error = 0; 2067 2068 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2069 2070 uio->uio_loffset = offset; 2071 ZFS_EXIT(zfsvfs); 2072 return (error); 2073 } 2074 2075 ulong_t zfs_fsync_sync_cnt = 4; 2076 2077 static int 2078 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2079 { 2080 znode_t *zp = VTOZ(vp); 2081 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2082 2083 /* 2084 * Regardless of whether this is required for standards conformance, 2085 * this is the logical behavior when fsync() is called on a file with 2086 * dirty pages. We use B_ASYNC since the ZIL transactions are already 2087 * going to be pushed out as part of the zil_commit(). 2088 */ 2089 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 2090 (vp->v_type == VREG) && !(IS_SWAPVP(vp))) 2091 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct); 2092 2093 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2094 2095 ZFS_ENTER(zfsvfs); 2096 ZFS_VERIFY_ZP(zp); 2097 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 2098 ZFS_EXIT(zfsvfs); 2099 return (0); 2100 } 2101 2102 2103 /* 2104 * Get the requested file attributes and place them in the provided 2105 * vattr structure. 2106 * 2107 * IN: vp - vnode of file. 2108 * vap - va_mask identifies requested attributes. 2109 * If AT_XVATTR set, then optional attrs are requested 2110 * flags - ATTR_NOACLCHECK (CIFS server context) 2111 * cr - credentials of caller. 2112 * ct - caller context 2113 * 2114 * OUT: vap - attribute values. 2115 * 2116 * RETURN: 0 (always succeeds) 2117 */ 2118 /* ARGSUSED */ 2119 static int 2120 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2121 caller_context_t *ct) 2122 { 2123 znode_t *zp = VTOZ(vp); 2124 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2125 znode_phys_t *pzp; 2126 int error = 0; 2127 uint64_t links; 2128 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2129 xoptattr_t *xoap = NULL; 2130 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2131 2132 ZFS_ENTER(zfsvfs); 2133 ZFS_VERIFY_ZP(zp); 2134 pzp = zp->z_phys; 2135 2136 mutex_enter(&zp->z_lock); 2137 2138 /* 2139 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2140 * Also, if we are the owner don't bother, since owner should 2141 * always be allowed to read basic attributes of file. 2142 */ 2143 if (!(pzp->zp_flags & ZFS_ACL_TRIVIAL) && 2144 (pzp->zp_uid != crgetuid(cr))) { 2145 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2146 skipaclchk, cr)) { 2147 mutex_exit(&zp->z_lock); 2148 ZFS_EXIT(zfsvfs); 2149 return (error); 2150 } 2151 } 2152 2153 /* 2154 * Return all attributes. It's cheaper to provide the answer 2155 * than to determine whether we were asked the question. 2156 */ 2157 2158 vap->va_type = vp->v_type; 2159 vap->va_mode = pzp->zp_mode & MODEMASK; 2160 zfs_fuid_map_ids(zp, &vap->va_uid, &vap->va_gid); 2161 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2162 vap->va_nodeid = zp->z_id; 2163 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2164 links = pzp->zp_links + 1; 2165 else 2166 links = pzp->zp_links; 2167 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */ 2168 vap->va_size = pzp->zp_size; 2169 vap->va_rdev = vp->v_rdev; 2170 vap->va_seq = zp->z_seq; 2171 2172 /* 2173 * Add in any requested optional attributes and the create time. 2174 * Also set the corresponding bits in the returned attribute bitmap. 2175 */ 2176 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2177 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2178 xoap->xoa_archive = 2179 ((pzp->zp_flags & ZFS_ARCHIVE) != 0); 2180 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2181 } 2182 2183 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2184 xoap->xoa_readonly = 2185 ((pzp->zp_flags & ZFS_READONLY) != 0); 2186 XVA_SET_RTN(xvap, XAT_READONLY); 2187 } 2188 2189 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2190 xoap->xoa_system = 2191 ((pzp->zp_flags & ZFS_SYSTEM) != 0); 2192 XVA_SET_RTN(xvap, XAT_SYSTEM); 2193 } 2194 2195 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2196 xoap->xoa_hidden = 2197 ((pzp->zp_flags & ZFS_HIDDEN) != 0); 2198 XVA_SET_RTN(xvap, XAT_HIDDEN); 2199 } 2200 2201 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2202 xoap->xoa_nounlink = 2203 ((pzp->zp_flags & ZFS_NOUNLINK) != 0); 2204 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2205 } 2206 2207 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2208 xoap->xoa_immutable = 2209 ((pzp->zp_flags & ZFS_IMMUTABLE) != 0); 2210 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2211 } 2212 2213 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2214 xoap->xoa_appendonly = 2215 ((pzp->zp_flags & ZFS_APPENDONLY) != 0); 2216 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2217 } 2218 2219 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2220 xoap->xoa_nodump = 2221 ((pzp->zp_flags & ZFS_NODUMP) != 0); 2222 XVA_SET_RTN(xvap, XAT_NODUMP); 2223 } 2224 2225 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2226 xoap->xoa_opaque = 2227 ((pzp->zp_flags & ZFS_OPAQUE) != 0); 2228 XVA_SET_RTN(xvap, XAT_OPAQUE); 2229 } 2230 2231 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2232 xoap->xoa_av_quarantined = 2233 ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0); 2234 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2235 } 2236 2237 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2238 xoap->xoa_av_modified = 2239 ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0); 2240 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2241 } 2242 2243 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2244 vp->v_type == VREG && 2245 (pzp->zp_flags & ZFS_BONUS_SCANSTAMP)) { 2246 size_t len; 2247 dmu_object_info_t doi; 2248 2249 /* 2250 * Only VREG files have anti-virus scanstamps, so we 2251 * won't conflict with symlinks in the bonus buffer. 2252 */ 2253 dmu_object_info_from_db(zp->z_dbuf, &doi); 2254 len = sizeof (xoap->xoa_av_scanstamp) + 2255 sizeof (znode_phys_t); 2256 if (len <= doi.doi_bonus_size) { 2257 /* 2258 * pzp points to the start of the 2259 * znode_phys_t. pzp + 1 points to the 2260 * first byte after the znode_phys_t. 2261 */ 2262 (void) memcpy(xoap->xoa_av_scanstamp, 2263 pzp + 1, 2264 sizeof (xoap->xoa_av_scanstamp)); 2265 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 2266 } 2267 } 2268 2269 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2270 ZFS_TIME_DECODE(&xoap->xoa_createtime, pzp->zp_crtime); 2271 XVA_SET_RTN(xvap, XAT_CREATETIME); 2272 } 2273 } 2274 2275 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 2276 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 2277 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 2278 2279 mutex_exit(&zp->z_lock); 2280 2281 dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks); 2282 2283 if (zp->z_blksz == 0) { 2284 /* 2285 * Block size hasn't been set; suggest maximal I/O transfers. 2286 */ 2287 vap->va_blksize = zfsvfs->z_max_blksz; 2288 } 2289 2290 ZFS_EXIT(zfsvfs); 2291 return (0); 2292 } 2293 2294 /* 2295 * Set the file attributes to the values contained in the 2296 * vattr structure. 2297 * 2298 * IN: vp - vnode of file to be modified. 2299 * vap - new attribute values. 2300 * If AT_XVATTR set, then optional attrs are being set 2301 * flags - ATTR_UTIME set if non-default time values provided. 2302 * - ATTR_NOACLCHECK (CIFS context only). 2303 * cr - credentials of caller. 2304 * ct - caller context 2305 * 2306 * RETURN: 0 if success 2307 * error code if failure 2308 * 2309 * Timestamps: 2310 * vp - ctime updated, mtime updated if size changed. 2311 */ 2312 /* ARGSUSED */ 2313 static int 2314 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2315 caller_context_t *ct) 2316 { 2317 znode_t *zp = VTOZ(vp); 2318 znode_phys_t *pzp; 2319 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2320 zilog_t *zilog; 2321 dmu_tx_t *tx; 2322 vattr_t oldva; 2323 uint_t mask = vap->va_mask; 2324 uint_t saved_mask; 2325 int trim_mask = 0; 2326 uint64_t new_mode; 2327 znode_t *attrzp; 2328 int need_policy = FALSE; 2329 int err; 2330 zfs_fuid_info_t *fuidp = NULL; 2331 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2332 xoptattr_t *xoap; 2333 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2334 2335 if (mask == 0) 2336 return (0); 2337 2338 if (mask & AT_NOSET) 2339 return (EINVAL); 2340 2341 ZFS_ENTER(zfsvfs); 2342 ZFS_VERIFY_ZP(zp); 2343 2344 pzp = zp->z_phys; 2345 zilog = zfsvfs->z_log; 2346 2347 /* 2348 * Make sure that if we have ephemeral uid/gid or xvattr specified 2349 * that file system is at proper version level 2350 */ 2351 2352 if (zfsvfs->z_use_fuids == B_FALSE && 2353 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2354 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2355 (mask & AT_XVATTR))) 2356 return (EINVAL); 2357 2358 if (mask & AT_SIZE && vp->v_type == VDIR) 2359 return (EISDIR); 2360 2361 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) 2362 return (EINVAL); 2363 2364 /* 2365 * If this is an xvattr_t, then get a pointer to the structure of 2366 * optional attributes. If this is NULL, then we have a vattr_t. 2367 */ 2368 xoap = xva_getxoptattr(xvap); 2369 2370 /* 2371 * Immutable files can only alter immutable bit and atime 2372 */ 2373 if ((pzp->zp_flags & ZFS_IMMUTABLE) && 2374 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2375 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) 2376 return (EPERM); 2377 2378 if ((mask & AT_SIZE) && (pzp->zp_flags & ZFS_READONLY)) 2379 return (EPERM); 2380 2381 top: 2382 attrzp = NULL; 2383 2384 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2385 ZFS_EXIT(zfsvfs); 2386 return (EROFS); 2387 } 2388 2389 /* 2390 * First validate permissions 2391 */ 2392 2393 if (mask & AT_SIZE) { 2394 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr); 2395 if (err) { 2396 ZFS_EXIT(zfsvfs); 2397 return (err); 2398 } 2399 /* 2400 * XXX - Note, we are not providing any open 2401 * mode flags here (like FNDELAY), so we may 2402 * block if there are locks present... this 2403 * should be addressed in openat(). 2404 */ 2405 do { 2406 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2407 /* NB: we already did dmu_tx_wait() if necessary */ 2408 } while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 2409 if (err) { 2410 ZFS_EXIT(zfsvfs); 2411 return (err); 2412 } 2413 } 2414 2415 if (mask & (AT_ATIME|AT_MTIME) || 2416 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2417 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2418 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2419 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2420 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) 2421 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2422 skipaclchk, cr); 2423 2424 if (mask & (AT_UID|AT_GID)) { 2425 int idmask = (mask & (AT_UID|AT_GID)); 2426 int take_owner; 2427 int take_group; 2428 2429 /* 2430 * NOTE: even if a new mode is being set, 2431 * we may clear S_ISUID/S_ISGID bits. 2432 */ 2433 2434 if (!(mask & AT_MODE)) 2435 vap->va_mode = pzp->zp_mode; 2436 2437 /* 2438 * Take ownership or chgrp to group we are a member of 2439 */ 2440 2441 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2442 take_group = (mask & AT_GID) && 2443 zfs_groupmember(zfsvfs, vap->va_gid, cr); 2444 2445 /* 2446 * If both AT_UID and AT_GID are set then take_owner and 2447 * take_group must both be set in order to allow taking 2448 * ownership. 2449 * 2450 * Otherwise, send the check through secpolicy_vnode_setattr() 2451 * 2452 */ 2453 2454 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2455 ((idmask == AT_UID) && take_owner) || 2456 ((idmask == AT_GID) && take_group)) { 2457 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 2458 skipaclchk, cr) == 0) { 2459 /* 2460 * Remove setuid/setgid for non-privileged users 2461 */ 2462 secpolicy_setid_clear(vap, cr); 2463 trim_mask = (mask & (AT_UID|AT_GID)); 2464 } else { 2465 need_policy = TRUE; 2466 } 2467 } else { 2468 need_policy = TRUE; 2469 } 2470 } 2471 2472 mutex_enter(&zp->z_lock); 2473 oldva.va_mode = pzp->zp_mode; 2474 zfs_fuid_map_ids(zp, &oldva.va_uid, &oldva.va_gid); 2475 if (mask & AT_XVATTR) { 2476 if ((need_policy == FALSE) && 2477 (XVA_ISSET_REQ(xvap, XAT_APPENDONLY) && 2478 xoap->xoa_appendonly != 2479 ((pzp->zp_flags & ZFS_APPENDONLY) != 0)) || 2480 (XVA_ISSET_REQ(xvap, XAT_NOUNLINK) && 2481 xoap->xoa_nounlink != 2482 ((pzp->zp_flags & ZFS_NOUNLINK) != 0)) || 2483 (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE) && 2484 xoap->xoa_immutable != 2485 ((pzp->zp_flags & ZFS_IMMUTABLE) != 0)) || 2486 (XVA_ISSET_REQ(xvap, XAT_NODUMP) && 2487 xoap->xoa_nodump != 2488 ((pzp->zp_flags & ZFS_NODUMP) != 0)) || 2489 (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED) && 2490 xoap->xoa_av_modified != 2491 ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0)) || 2492 (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED) && 2493 xoap->xoa_av_quarantined != 2494 ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0)) || 2495 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) || 2496 (XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 2497 need_policy = TRUE; 2498 } 2499 } 2500 2501 mutex_exit(&zp->z_lock); 2502 2503 if (mask & AT_MODE) { 2504 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 2505 err = secpolicy_setid_setsticky_clear(vp, vap, 2506 &oldva, cr); 2507 if (err) { 2508 ZFS_EXIT(zfsvfs); 2509 return (err); 2510 } 2511 trim_mask |= AT_MODE; 2512 } else { 2513 need_policy = TRUE; 2514 } 2515 } 2516 2517 if (need_policy) { 2518 /* 2519 * If trim_mask is set then take ownership 2520 * has been granted or write_acl is present and user 2521 * has the ability to modify mode. In that case remove 2522 * UID|GID and or MODE from mask so that 2523 * secpolicy_vnode_setattr() doesn't revoke it. 2524 */ 2525 2526 if (trim_mask) { 2527 saved_mask = vap->va_mask; 2528 vap->va_mask &= ~trim_mask; 2529 } 2530 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2531 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 2532 if (err) { 2533 ZFS_EXIT(zfsvfs); 2534 return (err); 2535 } 2536 2537 if (trim_mask) 2538 vap->va_mask |= saved_mask; 2539 } 2540 2541 /* 2542 * secpolicy_vnode_setattr, or take ownership may have 2543 * changed va_mask 2544 */ 2545 mask = vap->va_mask; 2546 2547 tx = dmu_tx_create(zfsvfs->z_os); 2548 dmu_tx_hold_bonus(tx, zp->z_id); 2549 if (zfsvfs->z_fuid_obj == 0) { 2550 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 2551 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2552 SPA_MAXBLOCKSIZE); 2553 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 2554 } else { 2555 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 2556 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 2557 SPA_MAXBLOCKSIZE); 2558 } 2559 2560 if (mask & AT_MODE) { 2561 uint64_t pmode = pzp->zp_mode; 2562 2563 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2564 2565 if (pzp->zp_acl.z_acl_extern_obj) { 2566 /* Are we upgrading ACL from old V0 format to new V1 */ 2567 if (zfsvfs->z_version <= ZPL_VERSION_FUID && 2568 pzp->zp_acl.z_acl_version == 2569 ZFS_ACL_VERSION_INITIAL) { 2570 dmu_tx_hold_free(tx, 2571 pzp->zp_acl.z_acl_extern_obj, 0, 2572 DMU_OBJECT_END); 2573 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2574 0, sizeof (zfs_object_ace_t) * 2048 + 6); 2575 } else { 2576 dmu_tx_hold_write(tx, 2577 pzp->zp_acl.z_acl_extern_obj, 0, 2578 SPA_MAXBLOCKSIZE); 2579 } 2580 } else { 2581 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2582 0, sizeof (zfs_object_ace_t) * 2048 + 6); 2583 } 2584 } 2585 2586 if ((mask & (AT_UID | AT_GID)) && pzp->zp_xattr != 0) { 2587 err = zfs_zget(zp->z_zfsvfs, pzp->zp_xattr, &attrzp); 2588 if (err) { 2589 dmu_tx_abort(tx); 2590 ZFS_EXIT(zfsvfs); 2591 return (err); 2592 } 2593 dmu_tx_hold_bonus(tx, attrzp->z_id); 2594 } 2595 2596 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2597 if (err) { 2598 if (attrzp) 2599 VN_RELE(ZTOV(attrzp)); 2600 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2601 dmu_tx_wait(tx); 2602 dmu_tx_abort(tx); 2603 goto top; 2604 } 2605 dmu_tx_abort(tx); 2606 ZFS_EXIT(zfsvfs); 2607 return (err); 2608 } 2609 2610 dmu_buf_will_dirty(zp->z_dbuf, tx); 2611 2612 /* 2613 * Set each attribute requested. 2614 * We group settings according to the locks they need to acquire. 2615 * 2616 * Note: you cannot set ctime directly, although it will be 2617 * updated as a side-effect of calling this function. 2618 */ 2619 2620 mutex_enter(&zp->z_lock); 2621 2622 if (mask & AT_MODE) { 2623 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2624 ASSERT3U(err, ==, 0); 2625 } 2626 2627 if (attrzp) 2628 mutex_enter(&attrzp->z_lock); 2629 2630 if (mask & AT_UID) { 2631 pzp->zp_uid = zfs_fuid_create(zfsvfs, 2632 vap->va_uid, ZFS_OWNER, tx, &fuidp); 2633 if (attrzp) { 2634 attrzp->z_phys->zp_uid = zfs_fuid_create(zfsvfs, 2635 vap->va_uid, ZFS_OWNER, tx, &fuidp); 2636 } 2637 } 2638 2639 if (mask & AT_GID) { 2640 pzp->zp_gid = zfs_fuid_create(zfsvfs, vap->va_gid, 2641 ZFS_GROUP, tx, &fuidp); 2642 if (attrzp) 2643 attrzp->z_phys->zp_gid = zfs_fuid_create(zfsvfs, 2644 vap->va_gid, ZFS_GROUP, tx, &fuidp); 2645 } 2646 2647 if (attrzp) 2648 mutex_exit(&attrzp->z_lock); 2649 2650 if (mask & AT_ATIME) 2651 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2652 2653 if (mask & AT_MTIME) 2654 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2655 2656 if (mask & AT_SIZE) 2657 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2658 else if (mask != 0) 2659 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2660 /* 2661 * Do this after setting timestamps to prevent timestamp 2662 * update from toggling bit 2663 */ 2664 2665 if (xoap && (mask & AT_XVATTR)) { 2666 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 2667 size_t len; 2668 dmu_object_info_t doi; 2669 2670 ASSERT(vp->v_type == VREG); 2671 2672 /* Grow the bonus buffer if necessary. */ 2673 dmu_object_info_from_db(zp->z_dbuf, &doi); 2674 len = sizeof (xoap->xoa_av_scanstamp) + 2675 sizeof (znode_phys_t); 2676 if (len > doi.doi_bonus_size) 2677 VERIFY(dmu_set_bonus(zp->z_dbuf, len, tx) == 0); 2678 } 2679 zfs_xvattr_set(zp, xvap); 2680 } 2681 2682 if (mask != 0) 2683 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 2684 2685 if (fuidp) 2686 zfs_fuid_info_free(fuidp); 2687 mutex_exit(&zp->z_lock); 2688 2689 if (attrzp) 2690 VN_RELE(ZTOV(attrzp)); 2691 2692 dmu_tx_commit(tx); 2693 2694 ZFS_EXIT(zfsvfs); 2695 return (err); 2696 } 2697 2698 typedef struct zfs_zlock { 2699 krwlock_t *zl_rwlock; /* lock we acquired */ 2700 znode_t *zl_znode; /* znode we held */ 2701 struct zfs_zlock *zl_next; /* next in list */ 2702 } zfs_zlock_t; 2703 2704 /* 2705 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2706 */ 2707 static void 2708 zfs_rename_unlock(zfs_zlock_t **zlpp) 2709 { 2710 zfs_zlock_t *zl; 2711 2712 while ((zl = *zlpp) != NULL) { 2713 if (zl->zl_znode != NULL) 2714 VN_RELE(ZTOV(zl->zl_znode)); 2715 rw_exit(zl->zl_rwlock); 2716 *zlpp = zl->zl_next; 2717 kmem_free(zl, sizeof (*zl)); 2718 } 2719 } 2720 2721 /* 2722 * Search back through the directory tree, using the ".." entries. 2723 * Lock each directory in the chain to prevent concurrent renames. 2724 * Fail any attempt to move a directory into one of its own descendants. 2725 * XXX - z_parent_lock can overlap with map or grow locks 2726 */ 2727 static int 2728 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2729 { 2730 zfs_zlock_t *zl; 2731 znode_t *zp = tdzp; 2732 uint64_t rootid = zp->z_zfsvfs->z_root; 2733 uint64_t *oidp = &zp->z_id; 2734 krwlock_t *rwlp = &szp->z_parent_lock; 2735 krw_t rw = RW_WRITER; 2736 2737 /* 2738 * First pass write-locks szp and compares to zp->z_id. 2739 * Later passes read-lock zp and compare to zp->z_parent. 2740 */ 2741 do { 2742 if (!rw_tryenter(rwlp, rw)) { 2743 /* 2744 * Another thread is renaming in this path. 2745 * Note that if we are a WRITER, we don't have any 2746 * parent_locks held yet. 2747 */ 2748 if (rw == RW_READER && zp->z_id > szp->z_id) { 2749 /* 2750 * Drop our locks and restart 2751 */ 2752 zfs_rename_unlock(&zl); 2753 *zlpp = NULL; 2754 zp = tdzp; 2755 oidp = &zp->z_id; 2756 rwlp = &szp->z_parent_lock; 2757 rw = RW_WRITER; 2758 continue; 2759 } else { 2760 /* 2761 * Wait for other thread to drop its locks 2762 */ 2763 rw_enter(rwlp, rw); 2764 } 2765 } 2766 2767 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2768 zl->zl_rwlock = rwlp; 2769 zl->zl_znode = NULL; 2770 zl->zl_next = *zlpp; 2771 *zlpp = zl; 2772 2773 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2774 return (EINVAL); 2775 2776 if (*oidp == rootid) /* We've hit the top */ 2777 return (0); 2778 2779 if (rw == RW_READER) { /* i.e. not the first pass */ 2780 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2781 if (error) 2782 return (error); 2783 zl->zl_znode = zp; 2784 } 2785 oidp = &zp->z_phys->zp_parent; 2786 rwlp = &zp->z_parent_lock; 2787 rw = RW_READER; 2788 2789 } while (zp->z_id != sdzp->z_id); 2790 2791 return (0); 2792 } 2793 2794 /* 2795 * Move an entry from the provided source directory to the target 2796 * directory. Change the entry name as indicated. 2797 * 2798 * IN: sdvp - Source directory containing the "old entry". 2799 * snm - Old entry name. 2800 * tdvp - Target directory to contain the "new entry". 2801 * tnm - New entry name. 2802 * cr - credentials of caller. 2803 * ct - caller context 2804 * flags - case flags 2805 * 2806 * RETURN: 0 if success 2807 * error code if failure 2808 * 2809 * Timestamps: 2810 * sdvp,tdvp - ctime|mtime updated 2811 */ 2812 /*ARGSUSED*/ 2813 static int 2814 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 2815 caller_context_t *ct, int flags) 2816 { 2817 znode_t *tdzp, *szp, *tzp; 2818 znode_t *sdzp = VTOZ(sdvp); 2819 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2820 zilog_t *zilog; 2821 vnode_t *realvp; 2822 zfs_dirlock_t *sdl, *tdl; 2823 dmu_tx_t *tx; 2824 zfs_zlock_t *zl; 2825 int cmp, serr, terr; 2826 int error = 0; 2827 int zflg = 0; 2828 2829 ZFS_ENTER(zfsvfs); 2830 ZFS_VERIFY_ZP(sdzp); 2831 zilog = zfsvfs->z_log; 2832 2833 /* 2834 * Make sure we have the real vp for the target directory. 2835 */ 2836 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 2837 tdvp = realvp; 2838 2839 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2840 ZFS_EXIT(zfsvfs); 2841 return (EXDEV); 2842 } 2843 2844 tdzp = VTOZ(tdvp); 2845 ZFS_VERIFY_ZP(tdzp); 2846 if (zfsvfs->z_case & ZFS_UTF8_ONLY && u8_validate(tnm, 2847 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2848 ZFS_EXIT(zfsvfs); 2849 return (EILSEQ); 2850 } 2851 2852 if (flags & FIGNORECASE) 2853 zflg |= ZCILOOK; 2854 2855 top: 2856 szp = NULL; 2857 tzp = NULL; 2858 zl = NULL; 2859 2860 /* 2861 * This is to prevent the creation of links into attribute space 2862 * by renaming a linked file into/outof an attribute directory. 2863 * See the comment in zfs_link() for why this is considered bad. 2864 */ 2865 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2866 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2867 ZFS_EXIT(zfsvfs); 2868 return (EINVAL); 2869 } 2870 2871 /* 2872 * Lock source and target directory entries. To prevent deadlock, 2873 * a lock ordering must be defined. We lock the directory with 2874 * the smallest object id first, or if it's a tie, the one with 2875 * the lexically first name. 2876 */ 2877 if (sdzp->z_id < tdzp->z_id) { 2878 cmp = -1; 2879 } else if (sdzp->z_id > tdzp->z_id) { 2880 cmp = 1; 2881 } else { 2882 /* 2883 * First compare the two name arguments without 2884 * considering any case folding. 2885 */ 2886 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 2887 2888 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 2889 ASSERT(error == 0 || !(zfsvfs->z_case & ZFS_UTF8_ONLY)); 2890 if (cmp == 0) { 2891 /* 2892 * POSIX: "If the old argument and the new argument 2893 * both refer to links to the same existing file, 2894 * the rename() function shall return successfully 2895 * and perform no other action." 2896 */ 2897 ZFS_EXIT(zfsvfs); 2898 return (0); 2899 } 2900 /* 2901 * If the file system is case-folding, then we may 2902 * have some more checking to do. A case-folding file 2903 * system is either supporting mixed case sensitivity 2904 * access or is completely case-insensitive. Note 2905 * that the file system is always case preserving. 2906 * 2907 * In mixed sensitivity mode case sensitive behavior 2908 * is the default. FIGNORECASE must be used to 2909 * explicitly request case insensitive behavior. 2910 * 2911 * If the source and target names provided differ only 2912 * by case (e.g., a request to rename 'tim' to 'Tim'), 2913 * we will treat this as a special case in the 2914 * case-insensitive mode: as long as the source name 2915 * is an exact match, we will allow this to proceed as 2916 * a name-change request. 2917 */ 2918 if ((zfsvfs->z_case & ZFS_CI_ONLY || 2919 (zfsvfs->z_case & ZFS_CI_MIXD && flags & FIGNORECASE)) && 2920 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 2921 &error) == 0) { 2922 /* 2923 * case preserving rename request, require exact 2924 * name matches 2925 */ 2926 zflg |= ZCIEXACT; 2927 zflg &= ~ZCILOOK; 2928 } 2929 } 2930 2931 if (cmp < 0) { 2932 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 2933 ZEXISTS | zflg, NULL, NULL); 2934 terr = zfs_dirent_lock(&tdl, 2935 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 2936 } else { 2937 terr = zfs_dirent_lock(&tdl, 2938 tdzp, tnm, &tzp, zflg, NULL, NULL); 2939 serr = zfs_dirent_lock(&sdl, 2940 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 2941 NULL, NULL); 2942 } 2943 2944 if (serr) { 2945 /* 2946 * Source entry invalid or not there. 2947 */ 2948 if (!terr) { 2949 zfs_dirent_unlock(tdl); 2950 if (tzp) 2951 VN_RELE(ZTOV(tzp)); 2952 } 2953 if (strcmp(snm, "..") == 0) 2954 serr = EINVAL; 2955 ZFS_EXIT(zfsvfs); 2956 return (serr); 2957 } 2958 if (terr) { 2959 zfs_dirent_unlock(sdl); 2960 VN_RELE(ZTOV(szp)); 2961 if (strcmp(tnm, "..") == 0) 2962 terr = EINVAL; 2963 ZFS_EXIT(zfsvfs); 2964 return (terr); 2965 } 2966 2967 /* 2968 * Must have write access at the source to remove the old entry 2969 * and write access at the target to create the new entry. 2970 * Note that if target and source are the same, this can be 2971 * done in a single check. 2972 */ 2973 2974 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2975 goto out; 2976 2977 if (ZTOV(szp)->v_type == VDIR) { 2978 /* 2979 * Check to make sure rename is valid. 2980 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2981 */ 2982 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 2983 goto out; 2984 } 2985 2986 /* 2987 * Does target exist? 2988 */ 2989 if (tzp) { 2990 /* 2991 * Source and target must be the same type. 2992 */ 2993 if (ZTOV(szp)->v_type == VDIR) { 2994 if (ZTOV(tzp)->v_type != VDIR) { 2995 error = ENOTDIR; 2996 goto out; 2997 } 2998 } else { 2999 if (ZTOV(tzp)->v_type == VDIR) { 3000 error = EISDIR; 3001 goto out; 3002 } 3003 } 3004 /* 3005 * POSIX dictates that when the source and target 3006 * entries refer to the same file object, rename 3007 * must do nothing and exit without error. 3008 */ 3009 if (szp->z_id == tzp->z_id) { 3010 error = 0; 3011 goto out; 3012 } 3013 } 3014 3015 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3016 if (tzp) 3017 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3018 3019 /* 3020 * notify the target directory if it is not the same 3021 * as source directory. 3022 */ 3023 if (tdvp != sdvp) { 3024 vnevent_rename_dest_dir(tdvp, ct); 3025 } 3026 3027 tx = dmu_tx_create(zfsvfs->z_os); 3028 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 3029 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 3030 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3031 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3032 if (sdzp != tdzp) 3033 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 3034 if (tzp) 3035 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 3036 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3037 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3038 if (error) { 3039 if (zl != NULL) 3040 zfs_rename_unlock(&zl); 3041 zfs_dirent_unlock(sdl); 3042 zfs_dirent_unlock(tdl); 3043 VN_RELE(ZTOV(szp)); 3044 if (tzp) 3045 VN_RELE(ZTOV(tzp)); 3046 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3047 dmu_tx_wait(tx); 3048 dmu_tx_abort(tx); 3049 goto top; 3050 } 3051 dmu_tx_abort(tx); 3052 ZFS_EXIT(zfsvfs); 3053 return (error); 3054 } 3055 3056 if (tzp) /* Attempt to remove the existing target */ 3057 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 3058 3059 if (error == 0) { 3060 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 3061 if (error == 0) { 3062 szp->z_phys->zp_flags |= ZFS_AV_MODIFIED; 3063 3064 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 3065 ASSERT(error == 0); 3066 3067 zfs_log_rename(zilog, tx, 3068 TX_RENAME | (flags & FIGNORECASE ? TX_CI : 0), 3069 sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp); 3070 } 3071 } 3072 3073 dmu_tx_commit(tx); 3074 out: 3075 if (zl != NULL) 3076 zfs_rename_unlock(&zl); 3077 3078 zfs_dirent_unlock(sdl); 3079 zfs_dirent_unlock(tdl); 3080 3081 VN_RELE(ZTOV(szp)); 3082 if (tzp) 3083 VN_RELE(ZTOV(tzp)); 3084 3085 ZFS_EXIT(zfsvfs); 3086 return (error); 3087 } 3088 3089 /* 3090 * Insert the indicated symbolic reference entry into the directory. 3091 * 3092 * IN: dvp - Directory to contain new symbolic link. 3093 * link - Name for new symlink entry. 3094 * vap - Attributes of new entry. 3095 * target - Target path of new symlink. 3096 * cr - credentials of caller. 3097 * ct - caller context 3098 * flags - case flags 3099 * 3100 * RETURN: 0 if success 3101 * error code if failure 3102 * 3103 * Timestamps: 3104 * dvp - ctime|mtime updated 3105 */ 3106 /*ARGSUSED*/ 3107 static int 3108 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr, 3109 caller_context_t *ct, int flags) 3110 { 3111 znode_t *zp, *dzp = VTOZ(dvp); 3112 zfs_dirlock_t *dl; 3113 dmu_tx_t *tx; 3114 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3115 zilog_t *zilog; 3116 uint64_t zoid; 3117 int len = strlen(link); 3118 int error; 3119 int zflg = ZNEW; 3120 zfs_fuid_info_t *fuidp = NULL; 3121 3122 ASSERT(vap->va_type == VLNK); 3123 3124 ZFS_ENTER(zfsvfs); 3125 ZFS_VERIFY_ZP(dzp); 3126 zilog = zfsvfs->z_log; 3127 3128 if (zfsvfs->z_case & ZFS_UTF8_ONLY && u8_validate(name, strlen(name), 3129 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3130 ZFS_EXIT(zfsvfs); 3131 return (EILSEQ); 3132 } 3133 if (flags & FIGNORECASE) 3134 zflg |= ZCILOOK; 3135 top: 3136 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3137 ZFS_EXIT(zfsvfs); 3138 return (error); 3139 } 3140 3141 if (len > MAXPATHLEN) { 3142 ZFS_EXIT(zfsvfs); 3143 return (ENAMETOOLONG); 3144 } 3145 3146 /* 3147 * Attempt to lock directory; fail if entry already exists. 3148 */ 3149 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 3150 if (error) { 3151 ZFS_EXIT(zfsvfs); 3152 return (error); 3153 } 3154 3155 tx = dmu_tx_create(zfsvfs->z_os); 3156 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 3157 dmu_tx_hold_bonus(tx, dzp->z_id); 3158 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3159 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 3160 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 3161 if (zfsvfs->z_fuid_obj == 0) { 3162 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 3163 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 3164 SPA_MAXBLOCKSIZE); 3165 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 3166 } else { 3167 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 3168 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 3169 SPA_MAXBLOCKSIZE); 3170 } 3171 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3172 if (error) { 3173 zfs_dirent_unlock(dl); 3174 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3175 dmu_tx_wait(tx); 3176 dmu_tx_abort(tx); 3177 goto top; 3178 } 3179 dmu_tx_abort(tx); 3180 ZFS_EXIT(zfsvfs); 3181 return (error); 3182 } 3183 3184 dmu_buf_will_dirty(dzp->z_dbuf, tx); 3185 3186 /* 3187 * Create a new object for the symlink. 3188 * Put the link content into bonus buffer if it will fit; 3189 * otherwise, store it just like any other file data. 3190 */ 3191 zoid = 0; 3192 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 3193 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len, NULL, &fuidp); 3194 if (len != 0) 3195 bcopy(link, zp->z_phys + 1, len); 3196 } else { 3197 dmu_buf_t *dbp; 3198 3199 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0, NULL, &fuidp); 3200 /* 3201 * Nothing can access the znode yet so no locking needed 3202 * for growing the znode's blocksize. 3203 */ 3204 zfs_grow_blocksize(zp, len, tx); 3205 3206 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp)); 3207 dmu_buf_will_dirty(dbp, tx); 3208 3209 ASSERT3U(len, <=, dbp->db_size); 3210 bcopy(link, dbp->db_data, len); 3211 dmu_buf_rele(dbp, FTAG); 3212 } 3213 zp->z_phys->zp_size = len; 3214 3215 /* 3216 * Insert the new object into the directory. 3217 */ 3218 (void) zfs_link_create(dl, zp, tx, ZNEW); 3219 out: 3220 if (error == 0) { 3221 uint64_t txtype = TX_SYMLINK; 3222 if (flags & FIGNORECASE) 3223 txtype |= TX_CI; 3224 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 3225 } 3226 if (fuidp) 3227 zfs_fuid_info_free(fuidp); 3228 3229 dmu_tx_commit(tx); 3230 3231 zfs_dirent_unlock(dl); 3232 3233 VN_RELE(ZTOV(zp)); 3234 3235 ZFS_EXIT(zfsvfs); 3236 return (error); 3237 } 3238 3239 /* 3240 * Return, in the buffer contained in the provided uio structure, 3241 * the symbolic path referred to by vp. 3242 * 3243 * IN: vp - vnode of symbolic link. 3244 * uoip - structure to contain the link path. 3245 * cr - credentials of caller. 3246 * ct - caller context 3247 * 3248 * OUT: uio - structure to contain the link path. 3249 * 3250 * RETURN: 0 if success 3251 * error code if failure 3252 * 3253 * Timestamps: 3254 * vp - atime updated 3255 */ 3256 /* ARGSUSED */ 3257 static int 3258 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 3259 { 3260 znode_t *zp = VTOZ(vp); 3261 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3262 size_t bufsz; 3263 int error; 3264 3265 ZFS_ENTER(zfsvfs); 3266 ZFS_VERIFY_ZP(zp); 3267 3268 bufsz = (size_t)zp->z_phys->zp_size; 3269 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 3270 error = uiomove(zp->z_phys + 1, 3271 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 3272 } else { 3273 dmu_buf_t *dbp; 3274 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 3275 if (error) { 3276 ZFS_EXIT(zfsvfs); 3277 return (error); 3278 } 3279 error = uiomove(dbp->db_data, 3280 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 3281 dmu_buf_rele(dbp, FTAG); 3282 } 3283 3284 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3285 ZFS_EXIT(zfsvfs); 3286 return (error); 3287 } 3288 3289 /* 3290 * Insert a new entry into directory tdvp referencing svp. 3291 * 3292 * IN: tdvp - Directory to contain new entry. 3293 * svp - vnode of new entry. 3294 * name - name of new entry. 3295 * cr - credentials of caller. 3296 * ct - caller context 3297 * 3298 * RETURN: 0 if success 3299 * error code if failure 3300 * 3301 * Timestamps: 3302 * tdvp - ctime|mtime updated 3303 * svp - ctime updated 3304 */ 3305 /* ARGSUSED */ 3306 static int 3307 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 3308 caller_context_t *ct, int flags) 3309 { 3310 znode_t *dzp = VTOZ(tdvp); 3311 znode_t *tzp, *szp; 3312 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3313 zilog_t *zilog; 3314 zfs_dirlock_t *dl; 3315 dmu_tx_t *tx; 3316 vnode_t *realvp; 3317 int error; 3318 int zf = ZNEW; 3319 uid_t owner; 3320 3321 ASSERT(tdvp->v_type == VDIR); 3322 3323 ZFS_ENTER(zfsvfs); 3324 ZFS_VERIFY_ZP(dzp); 3325 zilog = zfsvfs->z_log; 3326 3327 if (VOP_REALVP(svp, &realvp, ct) == 0) 3328 svp = realvp; 3329 3330 if (svp->v_vfsp != tdvp->v_vfsp) { 3331 ZFS_EXIT(zfsvfs); 3332 return (EXDEV); 3333 } 3334 szp = VTOZ(svp); 3335 ZFS_VERIFY_ZP(szp); 3336 3337 if (zfsvfs->z_case & ZFS_UTF8_ONLY && u8_validate(name, 3338 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3339 ZFS_EXIT(zfsvfs); 3340 return (EILSEQ); 3341 } 3342 if (flags & FIGNORECASE) 3343 zf |= ZCILOOK; 3344 3345 top: 3346 /* 3347 * We do not support links between attributes and non-attributes 3348 * because of the potential security risk of creating links 3349 * into "normal" file space in order to circumvent restrictions 3350 * imposed in attribute space. 3351 */ 3352 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 3353 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 3354 ZFS_EXIT(zfsvfs); 3355 return (EINVAL); 3356 } 3357 3358 /* 3359 * POSIX dictates that we return EPERM here. 3360 * Better choices include ENOTSUP or EISDIR. 3361 */ 3362 if (svp->v_type == VDIR) { 3363 ZFS_EXIT(zfsvfs); 3364 return (EPERM); 3365 } 3366 3367 zfs_fuid_map_id(zfsvfs, szp->z_phys->zp_uid, ZFS_OWNER, &owner); 3368 if (owner != crgetuid(cr) && 3369 secpolicy_basic_link(cr) != 0) { 3370 ZFS_EXIT(zfsvfs); 3371 return (EPERM); 3372 } 3373 3374 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3375 ZFS_EXIT(zfsvfs); 3376 return (error); 3377 } 3378 3379 /* 3380 * Attempt to lock directory; fail if entry already exists. 3381 */ 3382 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 3383 if (error) { 3384 ZFS_EXIT(zfsvfs); 3385 return (error); 3386 } 3387 3388 tx = dmu_tx_create(zfsvfs->z_os); 3389 dmu_tx_hold_bonus(tx, szp->z_id); 3390 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3391 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3392 if (error) { 3393 zfs_dirent_unlock(dl); 3394 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3395 dmu_tx_wait(tx); 3396 dmu_tx_abort(tx); 3397 goto top; 3398 } 3399 dmu_tx_abort(tx); 3400 ZFS_EXIT(zfsvfs); 3401 return (error); 3402 } 3403 3404 error = zfs_link_create(dl, szp, tx, 0); 3405 3406 if (error == 0) { 3407 uint64_t txtype = TX_LINK; 3408 if (flags & FIGNORECASE) 3409 txtype |= TX_CI; 3410 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 3411 } 3412 3413 dmu_tx_commit(tx); 3414 3415 zfs_dirent_unlock(dl); 3416 3417 if (error == 0) { 3418 vnevent_link(svp, ct); 3419 } 3420 3421 ZFS_EXIT(zfsvfs); 3422 return (error); 3423 } 3424 3425 /* 3426 * zfs_null_putapage() is used when the file system has been force 3427 * unmounted. It just drops the pages. 3428 */ 3429 /* ARGSUSED */ 3430 static int 3431 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 3432 size_t *lenp, int flags, cred_t *cr) 3433 { 3434 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 3435 return (0); 3436 } 3437 3438 /* 3439 * Push a page out to disk, klustering if possible. 3440 * 3441 * IN: vp - file to push page to. 3442 * pp - page to push. 3443 * flags - additional flags. 3444 * cr - credentials of caller. 3445 * 3446 * OUT: offp - start of range pushed. 3447 * lenp - len of range pushed. 3448 * 3449 * RETURN: 0 if success 3450 * error code if failure 3451 * 3452 * NOTE: callers must have locked the page to be pushed. On 3453 * exit, the page (and all other pages in the kluster) must be 3454 * unlocked. 3455 */ 3456 /* ARGSUSED */ 3457 static int 3458 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 3459 size_t *lenp, int flags, cred_t *cr) 3460 { 3461 znode_t *zp = VTOZ(vp); 3462 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3463 zilog_t *zilog = zfsvfs->z_log; 3464 dmu_tx_t *tx; 3465 rl_t *rl; 3466 u_offset_t off, koff; 3467 size_t len, klen; 3468 uint64_t filesz; 3469 int err; 3470 3471 filesz = zp->z_phys->zp_size; 3472 off = pp->p_offset; 3473 len = PAGESIZE; 3474 /* 3475 * If our blocksize is bigger than the page size, try to kluster 3476 * muiltiple pages so that we write a full block (thus avoiding 3477 * a read-modify-write). 3478 */ 3479 if (off < filesz && zp->z_blksz > PAGESIZE) { 3480 if (!ISP2(zp->z_blksz)) { 3481 /* Only one block in the file. */ 3482 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3483 koff = 0; 3484 } else { 3485 klen = zp->z_blksz; 3486 koff = P2ALIGN(off, (u_offset_t)klen); 3487 } 3488 ASSERT(koff <= filesz); 3489 if (koff + klen > filesz) 3490 klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE); 3491 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 3492 } 3493 ASSERT3U(btop(len), ==, btopr(len)); 3494 top: 3495 rl = zfs_range_lock(zp, off, len, RL_WRITER); 3496 /* 3497 * Can't push pages past end-of-file. 3498 */ 3499 filesz = zp->z_phys->zp_size; 3500 if (off >= filesz) { 3501 /* ignore all pages */ 3502 err = 0; 3503 goto out; 3504 } else if (off + len > filesz) { 3505 int npages = btopr(filesz - off); 3506 page_t *trunc; 3507 3508 page_list_break(&pp, &trunc, npages); 3509 /* ignore pages past end of file */ 3510 if (trunc) 3511 pvn_write_done(trunc, flags); 3512 len = filesz - off; 3513 } 3514 3515 tx = dmu_tx_create(zfsvfs->z_os); 3516 dmu_tx_hold_write(tx, zp->z_id, off, len); 3517 dmu_tx_hold_bonus(tx, zp->z_id); 3518 err = dmu_tx_assign(tx, zfsvfs->z_assign); 3519 if (err != 0) { 3520 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3521 zfs_range_unlock(rl); 3522 dmu_tx_wait(tx); 3523 dmu_tx_abort(tx); 3524 err = 0; 3525 goto top; 3526 } 3527 dmu_tx_abort(tx); 3528 goto out; 3529 } 3530 3531 if (zp->z_blksz <= PAGESIZE) { 3532 caddr_t va = ppmapin(pp, PROT_READ, (caddr_t)-1); 3533 ASSERT3U(len, <=, PAGESIZE); 3534 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 3535 ppmapout(va); 3536 } else { 3537 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 3538 } 3539 3540 if (err == 0) { 3541 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 3542 zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0); 3543 dmu_tx_commit(tx); 3544 } 3545 3546 out: 3547 zfs_range_unlock(rl); 3548 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 3549 if (offp) 3550 *offp = off; 3551 if (lenp) 3552 *lenp = len; 3553 3554 return (err); 3555 } 3556 3557 /* 3558 * Copy the portion of the file indicated from pages into the file. 3559 * The pages are stored in a page list attached to the files vnode. 3560 * 3561 * IN: vp - vnode of file to push page data to. 3562 * off - position in file to put data. 3563 * len - amount of data to write. 3564 * flags - flags to control the operation. 3565 * cr - credentials of caller. 3566 * ct - caller context. 3567 * 3568 * RETURN: 0 if success 3569 * error code if failure 3570 * 3571 * Timestamps: 3572 * vp - ctime|mtime updated 3573 */ 3574 /*ARGSUSED*/ 3575 static int 3576 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 3577 caller_context_t *ct) 3578 { 3579 znode_t *zp = VTOZ(vp); 3580 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3581 page_t *pp; 3582 size_t io_len; 3583 u_offset_t io_off; 3584 uint64_t filesz; 3585 int error = 0; 3586 3587 ZFS_ENTER(zfsvfs); 3588 ZFS_VERIFY_ZP(zp); 3589 3590 if (len == 0) { 3591 /* 3592 * Search the entire vp list for pages >= off. 3593 */ 3594 error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage, 3595 flags, cr); 3596 goto out; 3597 } 3598 3599 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3600 if (off > filesz) { 3601 /* past end of file */ 3602 ZFS_EXIT(zfsvfs); 3603 return (0); 3604 } 3605 3606 len = MIN(len, filesz - off); 3607 3608 for (io_off = off; io_off < off + len; io_off += io_len) { 3609 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 3610 pp = page_lookup(vp, io_off, 3611 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 3612 } else { 3613 pp = page_lookup_nowait(vp, io_off, 3614 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 3615 } 3616 3617 if (pp != NULL && pvn_getdirty(pp, flags)) { 3618 int err; 3619 3620 /* 3621 * Found a dirty page to push 3622 */ 3623 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 3624 if (err) 3625 error = err; 3626 } else { 3627 io_len = PAGESIZE; 3628 } 3629 } 3630 out: 3631 if ((flags & B_ASYNC) == 0) 3632 zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id); 3633 ZFS_EXIT(zfsvfs); 3634 return (error); 3635 } 3636 3637 /*ARGSUSED*/ 3638 void 3639 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 3640 { 3641 znode_t *zp = VTOZ(vp); 3642 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3643 int error; 3644 3645 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 3646 if (zp->z_dbuf_held == 0) { 3647 if (vn_has_cached_data(vp)) { 3648 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 3649 B_INVAL, cr); 3650 } 3651 3652 mutex_enter(&zp->z_lock); 3653 vp->v_count = 0; /* count arrives as 1 */ 3654 if (zp->z_dbuf == NULL) { 3655 mutex_exit(&zp->z_lock); 3656 zfs_znode_free(zp); 3657 } else { 3658 mutex_exit(&zp->z_lock); 3659 } 3660 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3661 VFS_RELE(zfsvfs->z_vfs); 3662 return; 3663 } 3664 3665 /* 3666 * Attempt to push any data in the page cache. If this fails 3667 * we will get kicked out later in zfs_zinactive(). 3668 */ 3669 if (vn_has_cached_data(vp)) { 3670 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 3671 cr); 3672 } 3673 3674 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 3675 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 3676 3677 dmu_tx_hold_bonus(tx, zp->z_id); 3678 error = dmu_tx_assign(tx, TXG_WAIT); 3679 if (error) { 3680 dmu_tx_abort(tx); 3681 } else { 3682 dmu_buf_will_dirty(zp->z_dbuf, tx); 3683 mutex_enter(&zp->z_lock); 3684 zp->z_atime_dirty = 0; 3685 mutex_exit(&zp->z_lock); 3686 dmu_tx_commit(tx); 3687 } 3688 } 3689 3690 zfs_zinactive(zp); 3691 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3692 } 3693 3694 /* 3695 * Bounds-check the seek operation. 3696 * 3697 * IN: vp - vnode seeking within 3698 * ooff - old file offset 3699 * noffp - pointer to new file offset 3700 * ct - caller context 3701 * 3702 * RETURN: 0 if success 3703 * EINVAL if new offset invalid 3704 */ 3705 /* ARGSUSED */ 3706 static int 3707 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 3708 caller_context_t *ct) 3709 { 3710 if (vp->v_type == VDIR) 3711 return (0); 3712 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 3713 } 3714 3715 /* 3716 * Pre-filter the generic locking function to trap attempts to place 3717 * a mandatory lock on a memory mapped file. 3718 */ 3719 static int 3720 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 3721 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 3722 { 3723 znode_t *zp = VTOZ(vp); 3724 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3725 int error; 3726 3727 ZFS_ENTER(zfsvfs); 3728 ZFS_VERIFY_ZP(zp); 3729 3730 /* 3731 * We are following the UFS semantics with respect to mapcnt 3732 * here: If we see that the file is mapped already, then we will 3733 * return an error, but we don't worry about races between this 3734 * function and zfs_map(). 3735 */ 3736 if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) { 3737 ZFS_EXIT(zfsvfs); 3738 return (EAGAIN); 3739 } 3740 error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct); 3741 ZFS_EXIT(zfsvfs); 3742 return (error); 3743 } 3744 3745 /* 3746 * If we can't find a page in the cache, we will create a new page 3747 * and fill it with file data. For efficiency, we may try to fill 3748 * multiple pages at once (klustering). 3749 */ 3750 static int 3751 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 3752 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 3753 { 3754 znode_t *zp = VTOZ(vp); 3755 page_t *pp, *cur_pp; 3756 objset_t *os = zp->z_zfsvfs->z_os; 3757 caddr_t va; 3758 u_offset_t io_off, total; 3759 uint64_t oid = zp->z_id; 3760 size_t io_len; 3761 uint64_t filesz; 3762 int err; 3763 3764 /* 3765 * If we are only asking for a single page don't bother klustering. 3766 */ 3767 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3768 if (off >= filesz) 3769 return (EFAULT); 3770 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 3771 io_off = off; 3772 io_len = PAGESIZE; 3773 pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr); 3774 } else { 3775 /* 3776 * Try to fill a kluster of pages (a blocks worth). 3777 */ 3778 size_t klen; 3779 u_offset_t koff; 3780 3781 if (!ISP2(zp->z_blksz)) { 3782 /* Only one block in the file. */ 3783 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3784 koff = 0; 3785 } else { 3786 /* 3787 * It would be ideal to align our offset to the 3788 * blocksize but doing so has resulted in some 3789 * strange application crashes. For now, we 3790 * leave the offset as is and only adjust the 3791 * length if we are off the end of the file. 3792 */ 3793 koff = off; 3794 klen = plsz; 3795 } 3796 ASSERT(koff <= filesz); 3797 if (koff + klen > filesz) 3798 klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff; 3799 ASSERT3U(off, >=, koff); 3800 ASSERT3U(off, <, koff + klen); 3801 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 3802 &io_len, koff, klen, 0); 3803 } 3804 if (pp == NULL) { 3805 /* 3806 * Some other thread entered the page before us. 3807 * Return to zfs_getpage to retry the lookup. 3808 */ 3809 *pl = NULL; 3810 return (0); 3811 } 3812 3813 /* 3814 * Fill the pages in the kluster. 3815 */ 3816 cur_pp = pp; 3817 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 3818 ASSERT3U(io_off, ==, cur_pp->p_offset); 3819 va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 3820 err = dmu_read(os, oid, io_off, PAGESIZE, va); 3821 ppmapout(va); 3822 if (err) { 3823 /* On error, toss the entire kluster */ 3824 pvn_read_done(pp, B_ERROR); 3825 return (err); 3826 } 3827 cur_pp = cur_pp->p_next; 3828 } 3829 out: 3830 /* 3831 * Fill in the page list array from the kluster. If 3832 * there are too many pages in the kluster, return 3833 * as many pages as possible starting from the desired 3834 * offset `off'. 3835 * NOTE: the page list will always be null terminated. 3836 */ 3837 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 3838 3839 return (0); 3840 } 3841 3842 /* 3843 * Return pointers to the pages for the file region [off, off + len] 3844 * in the pl array. If plsz is greater than len, this function may 3845 * also return page pointers from before or after the specified 3846 * region (i.e. some region [off', off' + plsz]). These additional 3847 * pages are only returned if they are already in the cache, or were 3848 * created as part of a klustered read. 3849 * 3850 * IN: vp - vnode of file to get data from. 3851 * off - position in file to get data from. 3852 * len - amount of data to retrieve. 3853 * plsz - length of provided page list. 3854 * seg - segment to obtain pages for. 3855 * addr - virtual address of fault. 3856 * rw - mode of created pages. 3857 * cr - credentials of caller. 3858 * ct - caller context. 3859 * 3860 * OUT: protp - protection mode of created pages. 3861 * pl - list of pages created. 3862 * 3863 * RETURN: 0 if success 3864 * error code if failure 3865 * 3866 * Timestamps: 3867 * vp - atime updated 3868 */ 3869 /* ARGSUSED */ 3870 static int 3871 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 3872 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 3873 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 3874 { 3875 znode_t *zp = VTOZ(vp); 3876 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3877 page_t *pp, **pl0 = pl; 3878 int need_unlock = 0, err = 0; 3879 offset_t orig_off; 3880 3881 ZFS_ENTER(zfsvfs); 3882 ZFS_VERIFY_ZP(zp); 3883 3884 if (protp) 3885 *protp = PROT_ALL; 3886 3887 /* no faultahead (for now) */ 3888 if (pl == NULL) { 3889 ZFS_EXIT(zfsvfs); 3890 return (0); 3891 } 3892 3893 /* can't fault past EOF */ 3894 if (off >= zp->z_phys->zp_size) { 3895 ZFS_EXIT(zfsvfs); 3896 return (EFAULT); 3897 } 3898 orig_off = off; 3899 3900 /* 3901 * If we already own the lock, then we must be page faulting 3902 * in the middle of a write to this file (i.e., we are writing 3903 * to this file using data from a mapped region of the file). 3904 */ 3905 if (rw_owner(&zp->z_map_lock) != curthread) { 3906 rw_enter(&zp->z_map_lock, RW_WRITER); 3907 need_unlock = TRUE; 3908 } 3909 3910 /* 3911 * Loop through the requested range [off, off + len] looking 3912 * for pages. If we don't find a page, we will need to create 3913 * a new page and fill it with data from the file. 3914 */ 3915 while (len > 0) { 3916 if (plsz < PAGESIZE) 3917 break; 3918 if (pp = page_lookup(vp, off, SE_SHARED)) { 3919 *pl++ = pp; 3920 off += PAGESIZE; 3921 addr += PAGESIZE; 3922 len -= PAGESIZE; 3923 plsz -= PAGESIZE; 3924 } else { 3925 err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw); 3926 if (err) 3927 goto out; 3928 /* 3929 * klustering may have changed our region 3930 * to be block aligned. 3931 */ 3932 if (((pp = *pl) != 0) && (off != pp->p_offset)) { 3933 int delta = off - pp->p_offset; 3934 len += delta; 3935 off -= delta; 3936 addr -= delta; 3937 } 3938 while (*pl) { 3939 pl++; 3940 off += PAGESIZE; 3941 addr += PAGESIZE; 3942 plsz -= PAGESIZE; 3943 if (len > PAGESIZE) 3944 len -= PAGESIZE; 3945 else 3946 len = 0; 3947 } 3948 } 3949 } 3950 3951 /* 3952 * Fill out the page array with any pages already in the cache. 3953 */ 3954 while (plsz > 0) { 3955 pp = page_lookup_nowait(vp, off, SE_SHARED); 3956 if (pp == NULL) 3957 break; 3958 *pl++ = pp; 3959 off += PAGESIZE; 3960 plsz -= PAGESIZE; 3961 } 3962 3963 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3964 out: 3965 /* 3966 * We can't grab the range lock for the page as reader which would 3967 * stop truncation as this leads to deadlock. So we need to recheck 3968 * the file size. 3969 */ 3970 if (orig_off >= zp->z_phys->zp_size) 3971 err = EFAULT; 3972 if (err) { 3973 /* 3974 * Release any pages we have previously locked. 3975 */ 3976 while (pl > pl0) 3977 page_unlock(*--pl); 3978 } 3979 3980 *pl = NULL; 3981 3982 if (need_unlock) 3983 rw_exit(&zp->z_map_lock); 3984 3985 ZFS_EXIT(zfsvfs); 3986 return (err); 3987 } 3988 3989 /* 3990 * Request a memory map for a section of a file. This code interacts 3991 * with common code and the VM system as follows: 3992 * 3993 * common code calls mmap(), which ends up in smmap_common() 3994 * 3995 * this calls VOP_MAP(), which takes you into (say) zfs 3996 * 3997 * zfs_map() calls as_map(), passing segvn_create() as the callback 3998 * 3999 * segvn_create() creates the new segment and calls VOP_ADDMAP() 4000 * 4001 * zfs_addmap() updates z_mapcnt 4002 */ 4003 /*ARGSUSED*/ 4004 static int 4005 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4006 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4007 caller_context_t *ct) 4008 { 4009 znode_t *zp = VTOZ(vp); 4010 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4011 segvn_crargs_t vn_a; 4012 int error; 4013 4014 if ((prot & PROT_WRITE) && 4015 (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_READONLY | 4016 ZFS_APPENDONLY))) 4017 return (EPERM); 4018 4019 ZFS_ENTER(zfsvfs); 4020 ZFS_VERIFY_ZP(zp); 4021 4022 if (vp->v_flag & VNOMAP) { 4023 ZFS_EXIT(zfsvfs); 4024 return (ENOSYS); 4025 } 4026 4027 if (off < 0 || len > MAXOFFSET_T - off) { 4028 ZFS_EXIT(zfsvfs); 4029 return (ENXIO); 4030 } 4031 4032 if (vp->v_type != VREG) { 4033 ZFS_EXIT(zfsvfs); 4034 return (ENODEV); 4035 } 4036 4037 /* 4038 * If file is locked, disallow mapping. 4039 */ 4040 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) { 4041 ZFS_EXIT(zfsvfs); 4042 return (EAGAIN); 4043 } 4044 4045 as_rangelock(as); 4046 if ((flags & MAP_FIXED) == 0) { 4047 map_addr(addrp, len, off, 1, flags); 4048 if (*addrp == NULL) { 4049 as_rangeunlock(as); 4050 ZFS_EXIT(zfsvfs); 4051 return (ENOMEM); 4052 } 4053 } else { 4054 /* 4055 * User specified address - blow away any previous mappings 4056 */ 4057 (void) as_unmap(as, *addrp, len); 4058 } 4059 4060 vn_a.vp = vp; 4061 vn_a.offset = (u_offset_t)off; 4062 vn_a.type = flags & MAP_TYPE; 4063 vn_a.prot = prot; 4064 vn_a.maxprot = maxprot; 4065 vn_a.cred = cr; 4066 vn_a.amp = NULL; 4067 vn_a.flags = flags & ~MAP_TYPE; 4068 vn_a.szc = 0; 4069 vn_a.lgrp_mem_policy_flags = 0; 4070 4071 error = as_map(as, *addrp, len, segvn_create, &vn_a); 4072 4073 as_rangeunlock(as); 4074 ZFS_EXIT(zfsvfs); 4075 return (error); 4076 } 4077 4078 /* ARGSUSED */ 4079 static int 4080 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4081 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4082 caller_context_t *ct) 4083 { 4084 uint64_t pages = btopr(len); 4085 4086 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 4087 return (0); 4088 } 4089 4090 /* 4091 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 4092 * more accurate mtime for the associated file. Since we don't have a way of 4093 * detecting when the data was actually modified, we have to resort to 4094 * heuristics. If an explicit msync() is done, then we mark the mtime when the 4095 * last page is pushed. The problem occurs when the msync() call is omitted, 4096 * which by far the most common case: 4097 * 4098 * open() 4099 * mmap() 4100 * <modify memory> 4101 * munmap() 4102 * close() 4103 * <time lapse> 4104 * putpage() via fsflush 4105 * 4106 * If we wait until fsflush to come along, we can have a modification time that 4107 * is some arbitrary point in the future. In order to prevent this in the 4108 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 4109 * torn down. 4110 */ 4111 /* ARGSUSED */ 4112 static int 4113 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4114 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 4115 caller_context_t *ct) 4116 { 4117 uint64_t pages = btopr(len); 4118 4119 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 4120 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 4121 4122 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 4123 vn_has_cached_data(vp)) 4124 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 4125 4126 return (0); 4127 } 4128 4129 /* 4130 * Free or allocate space in a file. Currently, this function only 4131 * supports the `F_FREESP' command. However, this command is somewhat 4132 * misnamed, as its functionality includes the ability to allocate as 4133 * well as free space. 4134 * 4135 * IN: vp - vnode of file to free data in. 4136 * cmd - action to take (only F_FREESP supported). 4137 * bfp - section of file to free/alloc. 4138 * flag - current file open mode flags. 4139 * offset - current file offset. 4140 * cr - credentials of caller [UNUSED]. 4141 * ct - caller context. 4142 * 4143 * RETURN: 0 if success 4144 * error code if failure 4145 * 4146 * Timestamps: 4147 * vp - ctime|mtime updated 4148 */ 4149 /* ARGSUSED */ 4150 static int 4151 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 4152 offset_t offset, cred_t *cr, caller_context_t *ct) 4153 { 4154 znode_t *zp = VTOZ(vp); 4155 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4156 uint64_t off, len; 4157 int error; 4158 4159 ZFS_ENTER(zfsvfs); 4160 ZFS_VERIFY_ZP(zp); 4161 4162 top: 4163 if (cmd != F_FREESP) { 4164 ZFS_EXIT(zfsvfs); 4165 return (EINVAL); 4166 } 4167 4168 if (error = convoff(vp, bfp, 0, offset)) { 4169 ZFS_EXIT(zfsvfs); 4170 return (error); 4171 } 4172 4173 if (bfp->l_len < 0) { 4174 ZFS_EXIT(zfsvfs); 4175 return (EINVAL); 4176 } 4177 4178 off = bfp->l_start; 4179 len = bfp->l_len; /* 0 means from off to end of file */ 4180 4181 do { 4182 error = zfs_freesp(zp, off, len, flag, TRUE); 4183 /* NB: we already did dmu_tx_wait() if necessary */ 4184 } while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 4185 4186 ZFS_EXIT(zfsvfs); 4187 return (error); 4188 } 4189 4190 /*ARGSUSED*/ 4191 static int 4192 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 4193 { 4194 znode_t *zp = VTOZ(vp); 4195 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4196 uint32_t gen; 4197 uint64_t object = zp->z_id; 4198 zfid_short_t *zfid; 4199 int size, i; 4200 4201 ZFS_ENTER(zfsvfs); 4202 ZFS_VERIFY_ZP(zp); 4203 gen = (uint32_t)zp->z_gen; 4204 4205 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 4206 if (fidp->fid_len < size) { 4207 fidp->fid_len = size; 4208 ZFS_EXIT(zfsvfs); 4209 return (ENOSPC); 4210 } 4211 4212 zfid = (zfid_short_t *)fidp; 4213 4214 zfid->zf_len = size; 4215 4216 for (i = 0; i < sizeof (zfid->zf_object); i++) 4217 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 4218 4219 /* Must have a non-zero generation number to distinguish from .zfs */ 4220 if (gen == 0) 4221 gen = 1; 4222 for (i = 0; i < sizeof (zfid->zf_gen); i++) 4223 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 4224 4225 if (size == LONG_FID_LEN) { 4226 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 4227 zfid_long_t *zlfid; 4228 4229 zlfid = (zfid_long_t *)fidp; 4230 4231 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 4232 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 4233 4234 /* XXX - this should be the generation number for the objset */ 4235 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 4236 zlfid->zf_setgen[i] = 0; 4237 } 4238 4239 ZFS_EXIT(zfsvfs); 4240 return (0); 4241 } 4242 4243 static int 4244 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 4245 caller_context_t *ct) 4246 { 4247 znode_t *zp, *xzp; 4248 zfsvfs_t *zfsvfs; 4249 zfs_dirlock_t *dl; 4250 int error; 4251 4252 switch (cmd) { 4253 case _PC_LINK_MAX: 4254 *valp = ULONG_MAX; 4255 return (0); 4256 4257 case _PC_FILESIZEBITS: 4258 *valp = 64; 4259 return (0); 4260 4261 case _PC_XATTR_EXISTS: 4262 zp = VTOZ(vp); 4263 zfsvfs = zp->z_zfsvfs; 4264 ZFS_ENTER(zfsvfs); 4265 ZFS_VERIFY_ZP(zp); 4266 *valp = 0; 4267 error = zfs_dirent_lock(&dl, zp, "", &xzp, 4268 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 4269 if (error == 0) { 4270 zfs_dirent_unlock(dl); 4271 if (!zfs_dirempty(xzp)) 4272 *valp = 1; 4273 VN_RELE(ZTOV(xzp)); 4274 } else if (error == ENOENT) { 4275 /* 4276 * If there aren't extended attributes, it's the 4277 * same as having zero of them. 4278 */ 4279 error = 0; 4280 } 4281 ZFS_EXIT(zfsvfs); 4282 return (error); 4283 4284 case _PC_SATTR_ENABLED: 4285 case _PC_SATTR_EXISTS: 4286 zp = VTOZ(vp); 4287 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) && 4288 (vp->v_type == VREG || vp->v_type == VDIR); 4289 return (0); 4290 4291 case _PC_ACL_ENABLED: 4292 *valp = _ACL_ACE_ENABLED; 4293 return (0); 4294 4295 case _PC_MIN_HOLE_SIZE: 4296 *valp = (ulong_t)SPA_MINBLOCKSIZE; 4297 return (0); 4298 4299 default: 4300 return (fs_pathconf(vp, cmd, valp, cr, ct)); 4301 } 4302 } 4303 4304 /*ARGSUSED*/ 4305 static int 4306 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4307 caller_context_t *ct) 4308 { 4309 znode_t *zp = VTOZ(vp); 4310 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4311 int error; 4312 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4313 4314 ZFS_ENTER(zfsvfs); 4315 ZFS_VERIFY_ZP(zp); 4316 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 4317 ZFS_EXIT(zfsvfs); 4318 4319 return (error); 4320 } 4321 4322 /*ARGSUSED*/ 4323 static int 4324 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4325 caller_context_t *ct) 4326 { 4327 znode_t *zp = VTOZ(vp); 4328 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4329 int error; 4330 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4331 4332 ZFS_ENTER(zfsvfs); 4333 ZFS_VERIFY_ZP(zp); 4334 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 4335 ZFS_EXIT(zfsvfs); 4336 return (error); 4337 } 4338 4339 /* 4340 * Predeclare these here so that the compiler assumes that 4341 * this is an "old style" function declaration that does 4342 * not include arguments => we won't get type mismatch errors 4343 * in the initializations that follow. 4344 */ 4345 static int zfs_inval(); 4346 static int zfs_isdir(); 4347 4348 static int 4349 zfs_inval() 4350 { 4351 return (EINVAL); 4352 } 4353 4354 static int 4355 zfs_isdir() 4356 { 4357 return (EISDIR); 4358 } 4359 /* 4360 * Directory vnode operations template 4361 */ 4362 vnodeops_t *zfs_dvnodeops; 4363 const fs_operation_def_t zfs_dvnodeops_template[] = { 4364 VOPNAME_OPEN, { .vop_open = zfs_open }, 4365 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4366 VOPNAME_READ, { .error = zfs_isdir }, 4367 VOPNAME_WRITE, { .error = zfs_isdir }, 4368 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4369 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4370 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4371 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4372 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4373 VOPNAME_CREATE, { .vop_create = zfs_create }, 4374 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 4375 VOPNAME_LINK, { .vop_link = zfs_link }, 4376 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4377 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 4378 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 4379 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 4380 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 4381 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4382 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4383 VOPNAME_FID, { .vop_fid = zfs_fid }, 4384 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4385 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4386 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4387 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4388 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4389 NULL, NULL 4390 }; 4391 4392 /* 4393 * Regular file vnode operations template 4394 */ 4395 vnodeops_t *zfs_fvnodeops; 4396 const fs_operation_def_t zfs_fvnodeops_template[] = { 4397 VOPNAME_OPEN, { .vop_open = zfs_open }, 4398 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4399 VOPNAME_READ, { .vop_read = zfs_read }, 4400 VOPNAME_WRITE, { .vop_write = zfs_write }, 4401 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4402 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4403 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4404 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4405 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4406 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4407 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4408 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4409 VOPNAME_FID, { .vop_fid = zfs_fid }, 4410 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4411 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 4412 VOPNAME_SPACE, { .vop_space = zfs_space }, 4413 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 4414 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 4415 VOPNAME_MAP, { .vop_map = zfs_map }, 4416 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 4417 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 4418 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4419 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4420 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4421 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4422 NULL, NULL 4423 }; 4424 4425 /* 4426 * Symbolic link vnode operations template 4427 */ 4428 vnodeops_t *zfs_symvnodeops; 4429 const fs_operation_def_t zfs_symvnodeops_template[] = { 4430 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4431 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4432 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4433 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4434 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 4435 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4436 VOPNAME_FID, { .vop_fid = zfs_fid }, 4437 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4438 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4439 NULL, NULL 4440 }; 4441 4442 /* 4443 * Extended attribute directory vnode operations template 4444 * This template is identical to the directory vnodes 4445 * operation template except for restricted operations: 4446 * VOP_MKDIR() 4447 * VOP_SYMLINK() 4448 * Note that there are other restrictions embedded in: 4449 * zfs_create() - restrict type to VREG 4450 * zfs_link() - no links into/out of attribute space 4451 * zfs_rename() - no moves into/out of attribute space 4452 */ 4453 vnodeops_t *zfs_xdvnodeops; 4454 const fs_operation_def_t zfs_xdvnodeops_template[] = { 4455 VOPNAME_OPEN, { .vop_open = zfs_open }, 4456 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4457 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4458 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4459 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4460 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4461 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4462 VOPNAME_CREATE, { .vop_create = zfs_create }, 4463 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 4464 VOPNAME_LINK, { .vop_link = zfs_link }, 4465 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4466 VOPNAME_MKDIR, { .error = zfs_inval }, 4467 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 4468 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 4469 VOPNAME_SYMLINK, { .error = zfs_inval }, 4470 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4471 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4472 VOPNAME_FID, { .vop_fid = zfs_fid }, 4473 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4474 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4475 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4476 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4477 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4478 NULL, NULL 4479 }; 4480 4481 /* 4482 * Error vnode operations template 4483 */ 4484 vnodeops_t *zfs_evnodeops; 4485 const fs_operation_def_t zfs_evnodeops_template[] = { 4486 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4487 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4488 NULL, NULL 4489 }; 4490