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 zfs_acl_t *aclp = NULL; 1125 zfs_fuid_info_t *fuidp = NULL; 1126 1127 /* 1128 * If we have an ephemeral id, ACL, or XVATTR then 1129 * make sure file system is at proper version 1130 */ 1131 1132 if (zfsvfs->z_use_fuids == B_FALSE && 1133 (vsecp || (vap->va_mask & AT_XVATTR) || 1134 IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr)))) 1135 return (EINVAL); 1136 1137 ZFS_ENTER(zfsvfs); 1138 ZFS_VERIFY_ZP(dzp); 1139 os = zfsvfs->z_os; 1140 zilog = zfsvfs->z_log; 1141 1142 if (zfsvfs->z_case & ZFS_UTF8_ONLY && u8_validate(name, strlen(name), 1143 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1144 ZFS_EXIT(zfsvfs); 1145 return (EILSEQ); 1146 } 1147 1148 if (vap->va_mask & AT_XVATTR) { 1149 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1150 crgetuid(cr), cr, vap->va_type)) != 0) { 1151 ZFS_EXIT(zfsvfs); 1152 return (error); 1153 } 1154 } 1155 top: 1156 *vpp = NULL; 1157 1158 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1159 vap->va_mode &= ~VSVTX; 1160 1161 if (*name == '\0') { 1162 /* 1163 * Null component name refers to the directory itself. 1164 */ 1165 VN_HOLD(dvp); 1166 zp = dzp; 1167 dl = NULL; 1168 error = 0; 1169 } else { 1170 /* possible VN_HOLD(zp) */ 1171 int zflg = 0; 1172 1173 if (flag & FIGNORECASE) 1174 zflg |= ZCILOOK; 1175 1176 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1177 NULL, NULL); 1178 if (error) { 1179 if (strcmp(name, "..") == 0) 1180 error = EISDIR; 1181 ZFS_EXIT(zfsvfs); 1182 if (aclp) 1183 zfs_acl_free(aclp); 1184 return (error); 1185 } 1186 } 1187 if (vsecp && aclp == NULL) { 1188 error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp); 1189 if (error) { 1190 ZFS_EXIT(zfsvfs); 1191 if (dl) 1192 zfs_dirent_unlock(dl); 1193 return (error); 1194 } 1195 } 1196 1197 if (zp == NULL) { 1198 uint64_t txtype; 1199 1200 /* 1201 * Create a new file object and update the directory 1202 * to reference it. 1203 */ 1204 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1205 goto out; 1206 } 1207 1208 /* 1209 * We only support the creation of regular files in 1210 * extended attribute directories. 1211 */ 1212 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1213 (vap->va_type != VREG)) { 1214 error = EINVAL; 1215 goto out; 1216 } 1217 1218 tx = dmu_tx_create(os); 1219 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1220 if (zfsvfs->z_fuid_obj == 0) { 1221 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1222 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1223 SPA_MAXBLOCKSIZE); 1224 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 1225 } else { 1226 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 1227 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 1228 SPA_MAXBLOCKSIZE); 1229 } 1230 dmu_tx_hold_bonus(tx, dzp->z_id); 1231 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1232 if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp) { 1233 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1234 0, SPA_MAXBLOCKSIZE); 1235 } 1236 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1237 if (error) { 1238 zfs_dirent_unlock(dl); 1239 if (error == ERESTART && 1240 zfsvfs->z_assign == TXG_NOWAIT) { 1241 dmu_tx_wait(tx); 1242 dmu_tx_abort(tx); 1243 goto top; 1244 } 1245 dmu_tx_abort(tx); 1246 ZFS_EXIT(zfsvfs); 1247 if (aclp) 1248 zfs_acl_free(aclp); 1249 return (error); 1250 } 1251 zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp); 1252 (void) zfs_link_create(dl, zp, tx, ZNEW); 1253 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1254 if (flag & FIGNORECASE) 1255 txtype |= TX_CI; 1256 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1257 vsecp, fuidp, vap); 1258 if (fuidp) 1259 zfs_fuid_info_free(fuidp); 1260 dmu_tx_commit(tx); 1261 } else { 1262 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1263 1264 /* 1265 * A directory entry already exists for this name. 1266 */ 1267 /* 1268 * Can't truncate an existing file if in exclusive mode. 1269 */ 1270 if (excl == EXCL) { 1271 error = EEXIST; 1272 goto out; 1273 } 1274 /* 1275 * Can't open a directory for writing. 1276 */ 1277 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1278 error = EISDIR; 1279 goto out; 1280 } 1281 /* 1282 * Verify requested access to file. 1283 */ 1284 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1285 goto out; 1286 } 1287 1288 mutex_enter(&dzp->z_lock); 1289 dzp->z_seq++; 1290 mutex_exit(&dzp->z_lock); 1291 1292 /* 1293 * Truncate regular files if requested. 1294 */ 1295 if ((ZTOV(zp)->v_type == VREG) && 1296 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1297 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1298 if (error == ERESTART && 1299 zfsvfs->z_assign == TXG_NOWAIT) { 1300 /* NB: we already did dmu_tx_wait() */ 1301 zfs_dirent_unlock(dl); 1302 VN_RELE(ZTOV(zp)); 1303 goto top; 1304 } 1305 1306 if (error == 0) { 1307 vnevent_create(ZTOV(zp), ct); 1308 } 1309 } 1310 } 1311 out: 1312 1313 if (dl) 1314 zfs_dirent_unlock(dl); 1315 1316 if (error) { 1317 if (zp) 1318 VN_RELE(ZTOV(zp)); 1319 } else { 1320 *vpp = ZTOV(zp); 1321 /* 1322 * If vnode is for a device return a specfs vnode instead. 1323 */ 1324 if (IS_DEVVP(*vpp)) { 1325 struct vnode *svp; 1326 1327 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1328 VN_RELE(*vpp); 1329 if (svp == NULL) { 1330 error = ENOSYS; 1331 } 1332 *vpp = svp; 1333 } 1334 } 1335 if (aclp) 1336 zfs_acl_free(aclp); 1337 1338 ZFS_EXIT(zfsvfs); 1339 return (error); 1340 } 1341 1342 /* 1343 * Remove an entry from a directory. 1344 * 1345 * IN: dvp - vnode of directory to remove entry from. 1346 * name - name of entry to remove. 1347 * cr - credentials of caller. 1348 * ct - caller context 1349 * flags - case flags 1350 * 1351 * RETURN: 0 if success 1352 * error code if failure 1353 * 1354 * Timestamps: 1355 * dvp - ctime|mtime 1356 * vp - ctime (if nlink > 0) 1357 */ 1358 /*ARGSUSED*/ 1359 static int 1360 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1361 int flags) 1362 { 1363 znode_t *zp, *dzp = VTOZ(dvp); 1364 znode_t *xzp = NULL; 1365 vnode_t *vp; 1366 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1367 zilog_t *zilog; 1368 uint64_t acl_obj, xattr_obj; 1369 zfs_dirlock_t *dl; 1370 dmu_tx_t *tx; 1371 boolean_t may_delete_now, delete_now = FALSE; 1372 boolean_t unlinked; 1373 uint64_t txtype; 1374 pathname_t *realnmp = NULL; 1375 pathname_t realnm; 1376 int error; 1377 int zflg = ZEXISTS; 1378 1379 ZFS_ENTER(zfsvfs); 1380 ZFS_VERIFY_ZP(dzp); 1381 zilog = zfsvfs->z_log; 1382 1383 if (flags & FIGNORECASE) { 1384 zflg |= ZCILOOK; 1385 pn_alloc(&realnm); 1386 realnmp = &realnm; 1387 } 1388 1389 top: 1390 /* 1391 * Attempt to lock directory; fail if entry doesn't exist. 1392 */ 1393 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1394 NULL, realnmp)) { 1395 if (realnmp) 1396 pn_free(realnmp); 1397 ZFS_EXIT(zfsvfs); 1398 return (error); 1399 } 1400 1401 vp = ZTOV(zp); 1402 1403 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1404 goto out; 1405 } 1406 1407 /* 1408 * Need to use rmdir for removing directories. 1409 */ 1410 if (vp->v_type == VDIR) { 1411 error = EPERM; 1412 goto out; 1413 } 1414 1415 vnevent_remove(vp, dvp, name, ct); 1416 1417 if (realnmp) 1418 dnlc_remove(dvp, realnmp->pn_path); 1419 else 1420 dnlc_remove(dvp, name); 1421 1422 mutex_enter(&vp->v_lock); 1423 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1424 mutex_exit(&vp->v_lock); 1425 1426 /* 1427 * We may delete the znode now, or we may put it in the unlinked set; 1428 * it depends on whether we're the last link, and on whether there are 1429 * other holds on the vnode. So we dmu_tx_hold() the right things to 1430 * allow for either case. 1431 */ 1432 tx = dmu_tx_create(zfsvfs->z_os); 1433 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1434 dmu_tx_hold_bonus(tx, zp->z_id); 1435 if (may_delete_now) 1436 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 1437 1438 /* are there any extended attributes? */ 1439 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1440 /* XXX - do we need this if we are deleting? */ 1441 dmu_tx_hold_bonus(tx, xattr_obj); 1442 } 1443 1444 /* are there any additional acls */ 1445 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1446 may_delete_now) 1447 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1448 1449 /* charge as an update -- would be nice not to charge at all */ 1450 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1451 1452 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1453 if (error) { 1454 zfs_dirent_unlock(dl); 1455 VN_RELE(vp); 1456 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1457 dmu_tx_wait(tx); 1458 dmu_tx_abort(tx); 1459 goto top; 1460 } 1461 if (realnmp) 1462 pn_free(realnmp); 1463 dmu_tx_abort(tx); 1464 ZFS_EXIT(zfsvfs); 1465 return (error); 1466 } 1467 1468 /* 1469 * Remove the directory entry. 1470 */ 1471 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1472 1473 if (error) { 1474 dmu_tx_commit(tx); 1475 goto out; 1476 } 1477 1478 if (unlinked) { 1479 mutex_enter(&vp->v_lock); 1480 delete_now = may_delete_now && 1481 vp->v_count == 1 && !vn_has_cached_data(vp) && 1482 zp->z_phys->zp_xattr == xattr_obj && 1483 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1484 mutex_exit(&vp->v_lock); 1485 } 1486 1487 if (delete_now) { 1488 if (zp->z_phys->zp_xattr) { 1489 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1490 ASSERT3U(error, ==, 0); 1491 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1492 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1493 mutex_enter(&xzp->z_lock); 1494 xzp->z_unlinked = 1; 1495 xzp->z_phys->zp_links = 0; 1496 mutex_exit(&xzp->z_lock); 1497 zfs_unlinked_add(xzp, tx); 1498 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1499 } 1500 mutex_enter(&zp->z_lock); 1501 mutex_enter(&vp->v_lock); 1502 vp->v_count--; 1503 ASSERT3U(vp->v_count, ==, 0); 1504 mutex_exit(&vp->v_lock); 1505 mutex_exit(&zp->z_lock); 1506 zfs_znode_delete(zp, tx); 1507 VFS_RELE(zfsvfs->z_vfs); 1508 } else if (unlinked) { 1509 zfs_unlinked_add(zp, tx); 1510 } 1511 1512 txtype = TX_REMOVE; 1513 if (flags & FIGNORECASE) 1514 txtype |= TX_CI; 1515 zfs_log_remove(zilog, tx, txtype, dzp, name); 1516 1517 dmu_tx_commit(tx); 1518 out: 1519 if (realnmp) 1520 pn_free(realnmp); 1521 1522 zfs_dirent_unlock(dl); 1523 1524 if (!delete_now) { 1525 VN_RELE(vp); 1526 } else if (xzp) { 1527 /* this rele delayed to prevent nesting transactions */ 1528 VN_RELE(ZTOV(xzp)); 1529 } 1530 1531 ZFS_EXIT(zfsvfs); 1532 return (error); 1533 } 1534 1535 /* 1536 * Create a new directory and insert it into dvp using the name 1537 * provided. Return a pointer to the inserted directory. 1538 * 1539 * IN: dvp - vnode of directory to add subdir to. 1540 * dirname - name of new directory. 1541 * vap - attributes of new directory. 1542 * cr - credentials of caller. 1543 * ct - caller context 1544 * vsecp - ACL to be set 1545 * 1546 * OUT: vpp - vnode of created directory. 1547 * 1548 * RETURN: 0 if success 1549 * error code if failure 1550 * 1551 * Timestamps: 1552 * dvp - ctime|mtime updated 1553 * vp - ctime|mtime|atime updated 1554 */ 1555 /*ARGSUSED*/ 1556 static int 1557 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 1558 caller_context_t *ct, int flags, vsecattr_t *vsecp) 1559 { 1560 znode_t *zp, *dzp = VTOZ(dvp); 1561 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1562 zilog_t *zilog; 1563 zfs_dirlock_t *dl; 1564 uint64_t txtype; 1565 dmu_tx_t *tx; 1566 int error; 1567 zfs_acl_t *aclp = NULL; 1568 zfs_fuid_info_t *fuidp = NULL; 1569 int zf = ZNEW; 1570 1571 ASSERT(vap->va_type == VDIR); 1572 1573 /* 1574 * If we have an ephemeral id, ACL, or XVATTR then 1575 * make sure file system is at proper version 1576 */ 1577 1578 if (zfsvfs->z_use_fuids == B_FALSE && 1579 (vsecp || (vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(crgetuid(cr))|| 1580 IS_EPHEMERAL(crgetgid(cr)))) 1581 return (EINVAL); 1582 1583 ZFS_ENTER(zfsvfs); 1584 ZFS_VERIFY_ZP(dzp); 1585 zilog = zfsvfs->z_log; 1586 1587 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1588 ZFS_EXIT(zfsvfs); 1589 return (EINVAL); 1590 } 1591 1592 if (zfsvfs->z_case & ZFS_UTF8_ONLY && u8_validate(dirname, 1593 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1594 ZFS_EXIT(zfsvfs); 1595 return (EILSEQ); 1596 } 1597 if (flags & FIGNORECASE) 1598 zf |= ZCILOOK; 1599 1600 if (vap->va_mask & AT_XVATTR) 1601 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1602 crgetuid(cr), cr, vap->va_type)) != 0) { 1603 ZFS_EXIT(zfsvfs); 1604 return (error); 1605 } 1606 1607 /* 1608 * First make sure the new directory doesn't exist. 1609 */ 1610 top: 1611 *vpp = NULL; 1612 1613 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 1614 NULL, NULL)) { 1615 ZFS_EXIT(zfsvfs); 1616 return (error); 1617 } 1618 1619 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 1620 zfs_dirent_unlock(dl); 1621 ZFS_EXIT(zfsvfs); 1622 return (error); 1623 } 1624 1625 if (vsecp && aclp == NULL) { 1626 error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp); 1627 if (error) { 1628 zfs_dirent_unlock(dl); 1629 ZFS_EXIT(zfsvfs); 1630 return (error); 1631 } 1632 } 1633 /* 1634 * Add a new entry to the directory. 1635 */ 1636 tx = dmu_tx_create(zfsvfs->z_os); 1637 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1638 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1639 if (zfsvfs->z_fuid_obj == 0) { 1640 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1641 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1642 SPA_MAXBLOCKSIZE); 1643 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 1644 } else { 1645 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 1646 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 1647 SPA_MAXBLOCKSIZE); 1648 } 1649 if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp) 1650 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1651 0, SPA_MAXBLOCKSIZE); 1652 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1653 if (error) { 1654 zfs_dirent_unlock(dl); 1655 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1656 dmu_tx_wait(tx); 1657 dmu_tx_abort(tx); 1658 goto top; 1659 } 1660 dmu_tx_abort(tx); 1661 ZFS_EXIT(zfsvfs); 1662 if (aclp) 1663 zfs_acl_free(aclp); 1664 return (error); 1665 } 1666 1667 /* 1668 * Create new node. 1669 */ 1670 zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp); 1671 1672 if (aclp) 1673 zfs_acl_free(aclp); 1674 1675 /* 1676 * Now put new name in parent dir. 1677 */ 1678 (void) zfs_link_create(dl, zp, tx, ZNEW); 1679 1680 *vpp = ZTOV(zp); 1681 1682 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 1683 if (flags & FIGNORECASE) 1684 txtype |= TX_CI; 1685 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, fuidp, vap); 1686 1687 if (fuidp) 1688 zfs_fuid_info_free(fuidp); 1689 dmu_tx_commit(tx); 1690 1691 zfs_dirent_unlock(dl); 1692 1693 ZFS_EXIT(zfsvfs); 1694 return (0); 1695 } 1696 1697 /* 1698 * Remove a directory subdir entry. If the current working 1699 * directory is the same as the subdir to be removed, the 1700 * remove will fail. 1701 * 1702 * IN: dvp - vnode of directory to remove from. 1703 * name - name of directory to be removed. 1704 * cwd - vnode of current working directory. 1705 * cr - credentials of caller. 1706 * ct - caller context 1707 * flags - case flags 1708 * 1709 * RETURN: 0 if success 1710 * error code if failure 1711 * 1712 * Timestamps: 1713 * dvp - ctime|mtime updated 1714 */ 1715 /*ARGSUSED*/ 1716 static int 1717 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 1718 caller_context_t *ct, int flags) 1719 { 1720 znode_t *dzp = VTOZ(dvp); 1721 znode_t *zp; 1722 vnode_t *vp; 1723 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1724 zilog_t *zilog; 1725 zfs_dirlock_t *dl; 1726 dmu_tx_t *tx; 1727 int error; 1728 int zflg = ZEXISTS; 1729 1730 ZFS_ENTER(zfsvfs); 1731 ZFS_VERIFY_ZP(dzp); 1732 zilog = zfsvfs->z_log; 1733 1734 if (flags & FIGNORECASE) 1735 zflg |= ZCILOOK; 1736 top: 1737 zp = NULL; 1738 1739 /* 1740 * Attempt to lock directory; fail if entry doesn't exist. 1741 */ 1742 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1743 NULL, NULL)) { 1744 ZFS_EXIT(zfsvfs); 1745 return (error); 1746 } 1747 1748 vp = ZTOV(zp); 1749 1750 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1751 goto out; 1752 } 1753 1754 if (vp->v_type != VDIR) { 1755 error = ENOTDIR; 1756 goto out; 1757 } 1758 1759 if (vp == cwd) { 1760 error = EINVAL; 1761 goto out; 1762 } 1763 1764 vnevent_rmdir(vp, dvp, name, ct); 1765 1766 /* 1767 * Grab a lock on the directory to make sure that noone is 1768 * trying to add (or lookup) entries while we are removing it. 1769 */ 1770 rw_enter(&zp->z_name_lock, RW_WRITER); 1771 1772 /* 1773 * Grab a lock on the parent pointer to make sure we play well 1774 * with the treewalk and directory rename code. 1775 */ 1776 rw_enter(&zp->z_parent_lock, RW_WRITER); 1777 1778 tx = dmu_tx_create(zfsvfs->z_os); 1779 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1780 dmu_tx_hold_bonus(tx, zp->z_id); 1781 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1782 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1783 if (error) { 1784 rw_exit(&zp->z_parent_lock); 1785 rw_exit(&zp->z_name_lock); 1786 zfs_dirent_unlock(dl); 1787 VN_RELE(vp); 1788 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1789 dmu_tx_wait(tx); 1790 dmu_tx_abort(tx); 1791 goto top; 1792 } 1793 dmu_tx_abort(tx); 1794 ZFS_EXIT(zfsvfs); 1795 return (error); 1796 } 1797 1798 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 1799 1800 if (error == 0) { 1801 uint64_t txtype = TX_RMDIR; 1802 if (flags & FIGNORECASE) 1803 txtype |= TX_CI; 1804 zfs_log_remove(zilog, tx, txtype, dzp, name); 1805 } 1806 1807 dmu_tx_commit(tx); 1808 1809 rw_exit(&zp->z_parent_lock); 1810 rw_exit(&zp->z_name_lock); 1811 out: 1812 zfs_dirent_unlock(dl); 1813 1814 VN_RELE(vp); 1815 1816 ZFS_EXIT(zfsvfs); 1817 return (error); 1818 } 1819 1820 /* 1821 * Read as many directory entries as will fit into the provided 1822 * buffer from the given directory cursor position (specified in 1823 * the uio structure. 1824 * 1825 * IN: vp - vnode of directory to read. 1826 * uio - structure supplying read location, range info, 1827 * and return buffer. 1828 * cr - credentials of caller. 1829 * ct - caller context 1830 * flags - case flags 1831 * 1832 * OUT: uio - updated offset and range, buffer filled. 1833 * eofp - set to true if end-of-file detected. 1834 * 1835 * RETURN: 0 if success 1836 * error code if failure 1837 * 1838 * Timestamps: 1839 * vp - atime updated 1840 * 1841 * Note that the low 4 bits of the cookie returned by zap is always zero. 1842 * This allows us to use the low range for "special" directory entries: 1843 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1844 * we use the offset 2 for the '.zfs' directory. 1845 */ 1846 /* ARGSUSED */ 1847 static int 1848 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, 1849 caller_context_t *ct, int flags) 1850 { 1851 znode_t *zp = VTOZ(vp); 1852 iovec_t *iovp; 1853 edirent_t *eodp; 1854 dirent64_t *odp; 1855 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1856 objset_t *os; 1857 caddr_t outbuf; 1858 size_t bufsize; 1859 zap_cursor_t zc; 1860 zap_attribute_t zap; 1861 uint_t bytes_wanted; 1862 uint64_t offset; /* must be unsigned; checks for < 1 */ 1863 int local_eof; 1864 int outcount; 1865 int error; 1866 uint8_t prefetch; 1867 1868 ZFS_ENTER(zfsvfs); 1869 ZFS_VERIFY_ZP(zp); 1870 1871 /* 1872 * If we are not given an eof variable, 1873 * use a local one. 1874 */ 1875 if (eofp == NULL) 1876 eofp = &local_eof; 1877 1878 /* 1879 * Check for valid iov_len. 1880 */ 1881 if (uio->uio_iov->iov_len <= 0) { 1882 ZFS_EXIT(zfsvfs); 1883 return (EINVAL); 1884 } 1885 1886 /* 1887 * Quit if directory has been removed (posix) 1888 */ 1889 if ((*eofp = zp->z_unlinked) != 0) { 1890 ZFS_EXIT(zfsvfs); 1891 return (0); 1892 } 1893 1894 error = 0; 1895 os = zfsvfs->z_os; 1896 offset = uio->uio_loffset; 1897 prefetch = zp->z_zn_prefetch; 1898 1899 /* 1900 * Initialize the iterator cursor. 1901 */ 1902 if (offset <= 3) { 1903 /* 1904 * Start iteration from the beginning of the directory. 1905 */ 1906 zap_cursor_init(&zc, os, zp->z_id); 1907 } else { 1908 /* 1909 * The offset is a serialized cursor. 1910 */ 1911 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1912 } 1913 1914 /* 1915 * Get space to change directory entries into fs independent format. 1916 */ 1917 iovp = uio->uio_iov; 1918 bytes_wanted = iovp->iov_len; 1919 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1920 bufsize = bytes_wanted; 1921 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1922 odp = (struct dirent64 *)outbuf; 1923 } else { 1924 bufsize = bytes_wanted; 1925 odp = (struct dirent64 *)iovp->iov_base; 1926 } 1927 eodp = (struct edirent *)odp; 1928 1929 /* 1930 * Transform to file-system independent format 1931 */ 1932 outcount = 0; 1933 while (outcount < bytes_wanted) { 1934 ino64_t objnum; 1935 ushort_t reclen; 1936 off64_t *next; 1937 1938 /* 1939 * Special case `.', `..', and `.zfs'. 1940 */ 1941 if (offset == 0) { 1942 (void) strcpy(zap.za_name, "."); 1943 zap.za_normalization_conflict = 0; 1944 objnum = zp->z_id; 1945 } else if (offset == 1) { 1946 (void) strcpy(zap.za_name, ".."); 1947 zap.za_normalization_conflict = 0; 1948 objnum = zp->z_phys->zp_parent; 1949 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1950 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1951 zap.za_normalization_conflict = 0; 1952 objnum = ZFSCTL_INO_ROOT; 1953 } else { 1954 /* 1955 * Grab next entry. 1956 */ 1957 if (error = zap_cursor_retrieve(&zc, &zap)) { 1958 if ((*eofp = (error == ENOENT)) != 0) 1959 break; 1960 else 1961 goto update; 1962 } 1963 1964 if (zap.za_integer_length != 8 || 1965 zap.za_num_integers != 1) { 1966 cmn_err(CE_WARN, "zap_readdir: bad directory " 1967 "entry, obj = %lld, offset = %lld\n", 1968 (u_longlong_t)zp->z_id, 1969 (u_longlong_t)offset); 1970 error = ENXIO; 1971 goto update; 1972 } 1973 1974 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 1975 /* 1976 * MacOS X can extract the object type here such as: 1977 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1978 */ 1979 } 1980 1981 if (flags & V_RDDIR_ENTFLAGS) 1982 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 1983 else 1984 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1985 1986 /* 1987 * Will this entry fit in the buffer? 1988 */ 1989 if (outcount + reclen > bufsize) { 1990 /* 1991 * Did we manage to fit anything in the buffer? 1992 */ 1993 if (!outcount) { 1994 error = EINVAL; 1995 goto update; 1996 } 1997 break; 1998 } 1999 if (flags & V_RDDIR_ENTFLAGS) { 2000 /* 2001 * Add extended flag entry: 2002 */ 2003 eodp->ed_ino = objnum; 2004 eodp->ed_reclen = reclen; 2005 /* NOTE: ed_off is the offset for the *next* entry */ 2006 next = &(eodp->ed_off); 2007 eodp->ed_eflags = zap.za_normalization_conflict ? 2008 ED_CASE_CONFLICT : 0; 2009 (void) strncpy(eodp->ed_name, zap.za_name, 2010 EDIRENT_NAMELEN(reclen)); 2011 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2012 } else { 2013 /* 2014 * Add normal entry: 2015 */ 2016 odp->d_ino = objnum; 2017 odp->d_reclen = reclen; 2018 /* NOTE: d_off is the offset for the *next* entry */ 2019 next = &(odp->d_off); 2020 (void) strncpy(odp->d_name, zap.za_name, 2021 DIRENT64_NAMELEN(reclen)); 2022 odp = (dirent64_t *)((intptr_t)odp + reclen); 2023 } 2024 outcount += reclen; 2025 2026 ASSERT(outcount <= bufsize); 2027 2028 /* Prefetch znode */ 2029 if (prefetch) 2030 dmu_prefetch(os, objnum, 0, 0); 2031 2032 /* 2033 * Move to the next entry, fill in the previous offset. 2034 */ 2035 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2036 zap_cursor_advance(&zc); 2037 offset = zap_cursor_serialize(&zc); 2038 } else { 2039 offset += 1; 2040 } 2041 *next = offset; 2042 } 2043 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2044 2045 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2046 iovp->iov_base += outcount; 2047 iovp->iov_len -= outcount; 2048 uio->uio_resid -= outcount; 2049 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2050 /* 2051 * Reset the pointer. 2052 */ 2053 offset = uio->uio_loffset; 2054 } 2055 2056 update: 2057 zap_cursor_fini(&zc); 2058 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2059 kmem_free(outbuf, bufsize); 2060 2061 if (error == ENOENT) 2062 error = 0; 2063 2064 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2065 2066 uio->uio_loffset = offset; 2067 ZFS_EXIT(zfsvfs); 2068 return (error); 2069 } 2070 2071 ulong_t zfs_fsync_sync_cnt = 4; 2072 2073 static int 2074 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2075 { 2076 znode_t *zp = VTOZ(vp); 2077 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2078 2079 /* 2080 * Regardless of whether this is required for standards conformance, 2081 * this is the logical behavior when fsync() is called on a file with 2082 * dirty pages. We use B_ASYNC since the ZIL transactions are already 2083 * going to be pushed out as part of the zil_commit(). 2084 */ 2085 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 2086 (vp->v_type == VREG) && !(IS_SWAPVP(vp))) 2087 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct); 2088 2089 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2090 2091 ZFS_ENTER(zfsvfs); 2092 ZFS_VERIFY_ZP(zp); 2093 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 2094 ZFS_EXIT(zfsvfs); 2095 return (0); 2096 } 2097 2098 2099 /* 2100 * Get the requested file attributes and place them in the provided 2101 * vattr structure. 2102 * 2103 * IN: vp - vnode of file. 2104 * vap - va_mask identifies requested attributes. 2105 * If AT_XVATTR set, then optional attrs are requested 2106 * flags - ATTR_NOACLCHECK (CIFS server context) 2107 * cr - credentials of caller. 2108 * ct - caller context 2109 * 2110 * OUT: vap - attribute values. 2111 * 2112 * RETURN: 0 (always succeeds) 2113 */ 2114 /* ARGSUSED */ 2115 static int 2116 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2117 caller_context_t *ct) 2118 { 2119 znode_t *zp = VTOZ(vp); 2120 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2121 znode_phys_t *pzp; 2122 int error = 0; 2123 uint64_t links; 2124 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2125 xoptattr_t *xoap = NULL; 2126 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2127 2128 ZFS_ENTER(zfsvfs); 2129 ZFS_VERIFY_ZP(zp); 2130 pzp = zp->z_phys; 2131 2132 mutex_enter(&zp->z_lock); 2133 2134 /* 2135 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2136 * Also, if we are the owner don't bother, since owner should 2137 * always be allowed to read basic attributes of file. 2138 */ 2139 if (!(pzp->zp_flags & ZFS_ACL_TRIVIAL) && 2140 (pzp->zp_uid != crgetuid(cr))) { 2141 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2142 skipaclchk, cr)) { 2143 mutex_exit(&zp->z_lock); 2144 ZFS_EXIT(zfsvfs); 2145 return (error); 2146 } 2147 } 2148 2149 /* 2150 * Return all attributes. It's cheaper to provide the answer 2151 * than to determine whether we were asked the question. 2152 */ 2153 2154 vap->va_type = vp->v_type; 2155 vap->va_mode = pzp->zp_mode & MODEMASK; 2156 zfs_fuid_map_ids(zp, &vap->va_uid, &vap->va_gid); 2157 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2158 vap->va_nodeid = zp->z_id; 2159 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2160 links = pzp->zp_links + 1; 2161 else 2162 links = pzp->zp_links; 2163 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */ 2164 vap->va_size = pzp->zp_size; 2165 vap->va_rdev = vp->v_rdev; 2166 vap->va_seq = zp->z_seq; 2167 2168 /* 2169 * Add in any requested optional attributes and the create time. 2170 * Also set the corresponding bits in the returned attribute bitmap. 2171 */ 2172 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2173 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2174 xoap->xoa_archive = 2175 ((pzp->zp_flags & ZFS_ARCHIVE) != 0); 2176 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2177 } 2178 2179 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2180 xoap->xoa_readonly = 2181 ((pzp->zp_flags & ZFS_READONLY) != 0); 2182 XVA_SET_RTN(xvap, XAT_READONLY); 2183 } 2184 2185 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2186 xoap->xoa_system = 2187 ((pzp->zp_flags & ZFS_SYSTEM) != 0); 2188 XVA_SET_RTN(xvap, XAT_SYSTEM); 2189 } 2190 2191 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2192 xoap->xoa_hidden = 2193 ((pzp->zp_flags & ZFS_HIDDEN) != 0); 2194 XVA_SET_RTN(xvap, XAT_HIDDEN); 2195 } 2196 2197 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2198 xoap->xoa_nounlink = 2199 ((pzp->zp_flags & ZFS_NOUNLINK) != 0); 2200 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2201 } 2202 2203 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2204 xoap->xoa_immutable = 2205 ((pzp->zp_flags & ZFS_IMMUTABLE) != 0); 2206 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2207 } 2208 2209 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2210 xoap->xoa_appendonly = 2211 ((pzp->zp_flags & ZFS_APPENDONLY) != 0); 2212 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2213 } 2214 2215 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2216 xoap->xoa_nodump = 2217 ((pzp->zp_flags & ZFS_NODUMP) != 0); 2218 XVA_SET_RTN(xvap, XAT_NODUMP); 2219 } 2220 2221 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2222 xoap->xoa_opaque = 2223 ((pzp->zp_flags & ZFS_OPAQUE) != 0); 2224 XVA_SET_RTN(xvap, XAT_OPAQUE); 2225 } 2226 2227 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2228 xoap->xoa_av_quarantined = 2229 ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0); 2230 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2231 } 2232 2233 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2234 xoap->xoa_av_modified = 2235 ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0); 2236 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2237 } 2238 2239 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2240 vp->v_type == VREG && 2241 (pzp->zp_flags & ZFS_BONUS_SCANSTAMP)) { 2242 size_t len; 2243 dmu_object_info_t doi; 2244 2245 /* 2246 * Only VREG files have anti-virus scanstamps, so we 2247 * won't conflict with symlinks in the bonus buffer. 2248 */ 2249 dmu_object_info_from_db(zp->z_dbuf, &doi); 2250 len = sizeof (xoap->xoa_av_scanstamp) + 2251 sizeof (znode_phys_t); 2252 if (len <= doi.doi_bonus_size) { 2253 /* 2254 * pzp points to the start of the 2255 * znode_phys_t. pzp + 1 points to the 2256 * first byte after the znode_phys_t. 2257 */ 2258 (void) memcpy(xoap->xoa_av_scanstamp, 2259 pzp + 1, 2260 sizeof (xoap->xoa_av_scanstamp)); 2261 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 2262 } 2263 } 2264 2265 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2266 ZFS_TIME_DECODE(&xoap->xoa_createtime, pzp->zp_crtime); 2267 XVA_SET_RTN(xvap, XAT_CREATETIME); 2268 } 2269 } 2270 2271 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 2272 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 2273 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 2274 2275 mutex_exit(&zp->z_lock); 2276 2277 dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks); 2278 2279 if (zp->z_blksz == 0) { 2280 /* 2281 * Block size hasn't been set; suggest maximal I/O transfers. 2282 */ 2283 vap->va_blksize = zfsvfs->z_max_blksz; 2284 } 2285 2286 ZFS_EXIT(zfsvfs); 2287 return (0); 2288 } 2289 2290 /* 2291 * Set the file attributes to the values contained in the 2292 * vattr structure. 2293 * 2294 * IN: vp - vnode of file to be modified. 2295 * vap - new attribute values. 2296 * If AT_XVATTR set, then optional attrs are being set 2297 * flags - ATTR_UTIME set if non-default time values provided. 2298 * - ATTR_NOACLCHECK (CIFS context only). 2299 * cr - credentials of caller. 2300 * ct - caller context 2301 * 2302 * RETURN: 0 if success 2303 * error code if failure 2304 * 2305 * Timestamps: 2306 * vp - ctime updated, mtime updated if size changed. 2307 */ 2308 /* ARGSUSED */ 2309 static int 2310 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2311 caller_context_t *ct) 2312 { 2313 znode_t *zp = VTOZ(vp); 2314 znode_phys_t *pzp; 2315 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2316 zilog_t *zilog; 2317 dmu_tx_t *tx; 2318 vattr_t oldva; 2319 uint_t mask = vap->va_mask; 2320 uint_t saved_mask; 2321 int trim_mask = 0; 2322 uint64_t new_mode; 2323 znode_t *attrzp; 2324 int need_policy = FALSE; 2325 int err; 2326 zfs_fuid_info_t *fuidp = NULL; 2327 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2328 xoptattr_t *xoap; 2329 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2330 2331 if (mask == 0) 2332 return (0); 2333 2334 if (mask & AT_NOSET) 2335 return (EINVAL); 2336 2337 ZFS_ENTER(zfsvfs); 2338 ZFS_VERIFY_ZP(zp); 2339 2340 pzp = zp->z_phys; 2341 zilog = zfsvfs->z_log; 2342 2343 /* 2344 * Make sure that if we have ephemeral uid/gid or xvattr specified 2345 * that file system is at proper version level 2346 */ 2347 2348 if (zfsvfs->z_use_fuids == B_FALSE && 2349 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2350 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2351 (mask & AT_XVATTR))) { 2352 ZFS_EXIT(zfsvfs); 2353 return (EINVAL); 2354 } 2355 2356 if (mask & AT_SIZE && vp->v_type == VDIR) { 2357 ZFS_EXIT(zfsvfs); 2358 return (EISDIR); 2359 } 2360 2361 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 2362 ZFS_EXIT(zfsvfs); 2363 return (EINVAL); 2364 } 2365 2366 /* 2367 * If this is an xvattr_t, then get a pointer to the structure of 2368 * optional attributes. If this is NULL, then we have a vattr_t. 2369 */ 2370 xoap = xva_getxoptattr(xvap); 2371 2372 /* 2373 * Immutable files can only alter immutable bit and atime 2374 */ 2375 if ((pzp->zp_flags & ZFS_IMMUTABLE) && 2376 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2377 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 2378 ZFS_EXIT(zfsvfs); 2379 return (EPERM); 2380 } 2381 2382 if ((mask & AT_SIZE) && (pzp->zp_flags & ZFS_READONLY)) { 2383 ZFS_EXIT(zfsvfs); 2384 return (EPERM); 2385 } 2386 2387 top: 2388 attrzp = NULL; 2389 2390 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2391 ZFS_EXIT(zfsvfs); 2392 return (EROFS); 2393 } 2394 2395 /* 2396 * First validate permissions 2397 */ 2398 2399 if (mask & AT_SIZE) { 2400 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr); 2401 if (err) { 2402 ZFS_EXIT(zfsvfs); 2403 return (err); 2404 } 2405 /* 2406 * XXX - Note, we are not providing any open 2407 * mode flags here (like FNDELAY), so we may 2408 * block if there are locks present... this 2409 * should be addressed in openat(). 2410 */ 2411 do { 2412 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2413 /* NB: we already did dmu_tx_wait() if necessary */ 2414 } while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 2415 if (err) { 2416 ZFS_EXIT(zfsvfs); 2417 return (err); 2418 } 2419 } 2420 2421 if (mask & (AT_ATIME|AT_MTIME) || 2422 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2423 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2424 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2425 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2426 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) 2427 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2428 skipaclchk, cr); 2429 2430 if (mask & (AT_UID|AT_GID)) { 2431 int idmask = (mask & (AT_UID|AT_GID)); 2432 int take_owner; 2433 int take_group; 2434 2435 /* 2436 * NOTE: even if a new mode is being set, 2437 * we may clear S_ISUID/S_ISGID bits. 2438 */ 2439 2440 if (!(mask & AT_MODE)) 2441 vap->va_mode = pzp->zp_mode; 2442 2443 /* 2444 * Take ownership or chgrp to group we are a member of 2445 */ 2446 2447 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2448 take_group = (mask & AT_GID) && 2449 zfs_groupmember(zfsvfs, vap->va_gid, cr); 2450 2451 /* 2452 * If both AT_UID and AT_GID are set then take_owner and 2453 * take_group must both be set in order to allow taking 2454 * ownership. 2455 * 2456 * Otherwise, send the check through secpolicy_vnode_setattr() 2457 * 2458 */ 2459 2460 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2461 ((idmask == AT_UID) && take_owner) || 2462 ((idmask == AT_GID) && take_group)) { 2463 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 2464 skipaclchk, cr) == 0) { 2465 /* 2466 * Remove setuid/setgid for non-privileged users 2467 */ 2468 secpolicy_setid_clear(vap, cr); 2469 trim_mask = (mask & (AT_UID|AT_GID)); 2470 } else { 2471 need_policy = TRUE; 2472 } 2473 } else { 2474 need_policy = TRUE; 2475 } 2476 } 2477 2478 mutex_enter(&zp->z_lock); 2479 oldva.va_mode = pzp->zp_mode; 2480 zfs_fuid_map_ids(zp, &oldva.va_uid, &oldva.va_gid); 2481 if (mask & AT_XVATTR) { 2482 if ((need_policy == FALSE) && 2483 (XVA_ISSET_REQ(xvap, XAT_APPENDONLY) && 2484 xoap->xoa_appendonly != 2485 ((pzp->zp_flags & ZFS_APPENDONLY) != 0)) || 2486 (XVA_ISSET_REQ(xvap, XAT_NOUNLINK) && 2487 xoap->xoa_nounlink != 2488 ((pzp->zp_flags & ZFS_NOUNLINK) != 0)) || 2489 (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE) && 2490 xoap->xoa_immutable != 2491 ((pzp->zp_flags & ZFS_IMMUTABLE) != 0)) || 2492 (XVA_ISSET_REQ(xvap, XAT_NODUMP) && 2493 xoap->xoa_nodump != 2494 ((pzp->zp_flags & ZFS_NODUMP) != 0)) || 2495 (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED) && 2496 xoap->xoa_av_modified != 2497 ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0)) || 2498 (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED) && 2499 xoap->xoa_av_quarantined != 2500 ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0)) || 2501 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) || 2502 (XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 2503 need_policy = TRUE; 2504 } 2505 } 2506 2507 mutex_exit(&zp->z_lock); 2508 2509 if (mask & AT_MODE) { 2510 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 2511 err = secpolicy_setid_setsticky_clear(vp, vap, 2512 &oldva, cr); 2513 if (err) { 2514 ZFS_EXIT(zfsvfs); 2515 return (err); 2516 } 2517 trim_mask |= AT_MODE; 2518 } else { 2519 need_policy = TRUE; 2520 } 2521 } 2522 2523 if (need_policy) { 2524 /* 2525 * If trim_mask is set then take ownership 2526 * has been granted or write_acl is present and user 2527 * has the ability to modify mode. In that case remove 2528 * UID|GID and or MODE from mask so that 2529 * secpolicy_vnode_setattr() doesn't revoke it. 2530 */ 2531 2532 if (trim_mask) { 2533 saved_mask = vap->va_mask; 2534 vap->va_mask &= ~trim_mask; 2535 } 2536 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2537 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 2538 if (err) { 2539 ZFS_EXIT(zfsvfs); 2540 return (err); 2541 } 2542 2543 if (trim_mask) 2544 vap->va_mask |= saved_mask; 2545 } 2546 2547 /* 2548 * secpolicy_vnode_setattr, or take ownership may have 2549 * changed va_mask 2550 */ 2551 mask = vap->va_mask; 2552 2553 tx = dmu_tx_create(zfsvfs->z_os); 2554 dmu_tx_hold_bonus(tx, zp->z_id); 2555 if (zfsvfs->z_fuid_obj == 0) { 2556 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 2557 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2558 SPA_MAXBLOCKSIZE); 2559 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 2560 } else { 2561 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 2562 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 2563 SPA_MAXBLOCKSIZE); 2564 } 2565 2566 if (mask & AT_MODE) { 2567 uint64_t pmode = pzp->zp_mode; 2568 2569 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2570 2571 if (pzp->zp_acl.z_acl_extern_obj) { 2572 /* Are we upgrading ACL from old V0 format to new V1 */ 2573 if (zfsvfs->z_version <= ZPL_VERSION_FUID && 2574 pzp->zp_acl.z_acl_version == 2575 ZFS_ACL_VERSION_INITIAL) { 2576 dmu_tx_hold_free(tx, 2577 pzp->zp_acl.z_acl_extern_obj, 0, 2578 DMU_OBJECT_END); 2579 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2580 0, sizeof (zfs_object_ace_t) * 2048 + 6); 2581 } else { 2582 dmu_tx_hold_write(tx, 2583 pzp->zp_acl.z_acl_extern_obj, 0, 2584 SPA_MAXBLOCKSIZE); 2585 } 2586 } else { 2587 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2588 0, sizeof (zfs_object_ace_t) * 2048 + 6); 2589 } 2590 } 2591 2592 if ((mask & (AT_UID | AT_GID)) && pzp->zp_xattr != 0) { 2593 err = zfs_zget(zp->z_zfsvfs, pzp->zp_xattr, &attrzp); 2594 if (err) { 2595 dmu_tx_abort(tx); 2596 ZFS_EXIT(zfsvfs); 2597 return (err); 2598 } 2599 dmu_tx_hold_bonus(tx, attrzp->z_id); 2600 } 2601 2602 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2603 if (err) { 2604 if (attrzp) 2605 VN_RELE(ZTOV(attrzp)); 2606 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2607 dmu_tx_wait(tx); 2608 dmu_tx_abort(tx); 2609 goto top; 2610 } 2611 dmu_tx_abort(tx); 2612 ZFS_EXIT(zfsvfs); 2613 return (err); 2614 } 2615 2616 dmu_buf_will_dirty(zp->z_dbuf, tx); 2617 2618 /* 2619 * Set each attribute requested. 2620 * We group settings according to the locks they need to acquire. 2621 * 2622 * Note: you cannot set ctime directly, although it will be 2623 * updated as a side-effect of calling this function. 2624 */ 2625 2626 mutex_enter(&zp->z_lock); 2627 2628 if (mask & AT_MODE) { 2629 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2630 ASSERT3U(err, ==, 0); 2631 } 2632 2633 if (attrzp) 2634 mutex_enter(&attrzp->z_lock); 2635 2636 if (mask & AT_UID) { 2637 pzp->zp_uid = zfs_fuid_create(zfsvfs, 2638 vap->va_uid, ZFS_OWNER, tx, &fuidp); 2639 if (attrzp) { 2640 attrzp->z_phys->zp_uid = zfs_fuid_create(zfsvfs, 2641 vap->va_uid, ZFS_OWNER, tx, &fuidp); 2642 } 2643 } 2644 2645 if (mask & AT_GID) { 2646 pzp->zp_gid = zfs_fuid_create(zfsvfs, vap->va_gid, 2647 ZFS_GROUP, tx, &fuidp); 2648 if (attrzp) 2649 attrzp->z_phys->zp_gid = zfs_fuid_create(zfsvfs, 2650 vap->va_gid, ZFS_GROUP, tx, &fuidp); 2651 } 2652 2653 if (attrzp) 2654 mutex_exit(&attrzp->z_lock); 2655 2656 if (mask & AT_ATIME) 2657 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2658 2659 if (mask & AT_MTIME) 2660 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2661 2662 if (mask & AT_SIZE) 2663 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2664 else if (mask != 0) 2665 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2666 /* 2667 * Do this after setting timestamps to prevent timestamp 2668 * update from toggling bit 2669 */ 2670 2671 if (xoap && (mask & AT_XVATTR)) { 2672 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 2673 size_t len; 2674 dmu_object_info_t doi; 2675 2676 ASSERT(vp->v_type == VREG); 2677 2678 /* Grow the bonus buffer if necessary. */ 2679 dmu_object_info_from_db(zp->z_dbuf, &doi); 2680 len = sizeof (xoap->xoa_av_scanstamp) + 2681 sizeof (znode_phys_t); 2682 if (len > doi.doi_bonus_size) 2683 VERIFY(dmu_set_bonus(zp->z_dbuf, len, tx) == 0); 2684 } 2685 zfs_xvattr_set(zp, xvap); 2686 } 2687 2688 if (mask != 0) 2689 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 2690 2691 if (fuidp) 2692 zfs_fuid_info_free(fuidp); 2693 mutex_exit(&zp->z_lock); 2694 2695 if (attrzp) 2696 VN_RELE(ZTOV(attrzp)); 2697 2698 dmu_tx_commit(tx); 2699 2700 ZFS_EXIT(zfsvfs); 2701 return (err); 2702 } 2703 2704 typedef struct zfs_zlock { 2705 krwlock_t *zl_rwlock; /* lock we acquired */ 2706 znode_t *zl_znode; /* znode we held */ 2707 struct zfs_zlock *zl_next; /* next in list */ 2708 } zfs_zlock_t; 2709 2710 /* 2711 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2712 */ 2713 static void 2714 zfs_rename_unlock(zfs_zlock_t **zlpp) 2715 { 2716 zfs_zlock_t *zl; 2717 2718 while ((zl = *zlpp) != NULL) { 2719 if (zl->zl_znode != NULL) 2720 VN_RELE(ZTOV(zl->zl_znode)); 2721 rw_exit(zl->zl_rwlock); 2722 *zlpp = zl->zl_next; 2723 kmem_free(zl, sizeof (*zl)); 2724 } 2725 } 2726 2727 /* 2728 * Search back through the directory tree, using the ".." entries. 2729 * Lock each directory in the chain to prevent concurrent renames. 2730 * Fail any attempt to move a directory into one of its own descendants. 2731 * XXX - z_parent_lock can overlap with map or grow locks 2732 */ 2733 static int 2734 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2735 { 2736 zfs_zlock_t *zl; 2737 znode_t *zp = tdzp; 2738 uint64_t rootid = zp->z_zfsvfs->z_root; 2739 uint64_t *oidp = &zp->z_id; 2740 krwlock_t *rwlp = &szp->z_parent_lock; 2741 krw_t rw = RW_WRITER; 2742 2743 /* 2744 * First pass write-locks szp and compares to zp->z_id. 2745 * Later passes read-lock zp and compare to zp->z_parent. 2746 */ 2747 do { 2748 if (!rw_tryenter(rwlp, rw)) { 2749 /* 2750 * Another thread is renaming in this path. 2751 * Note that if we are a WRITER, we don't have any 2752 * parent_locks held yet. 2753 */ 2754 if (rw == RW_READER && zp->z_id > szp->z_id) { 2755 /* 2756 * Drop our locks and restart 2757 */ 2758 zfs_rename_unlock(&zl); 2759 *zlpp = NULL; 2760 zp = tdzp; 2761 oidp = &zp->z_id; 2762 rwlp = &szp->z_parent_lock; 2763 rw = RW_WRITER; 2764 continue; 2765 } else { 2766 /* 2767 * Wait for other thread to drop its locks 2768 */ 2769 rw_enter(rwlp, rw); 2770 } 2771 } 2772 2773 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2774 zl->zl_rwlock = rwlp; 2775 zl->zl_znode = NULL; 2776 zl->zl_next = *zlpp; 2777 *zlpp = zl; 2778 2779 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2780 return (EINVAL); 2781 2782 if (*oidp == rootid) /* We've hit the top */ 2783 return (0); 2784 2785 if (rw == RW_READER) { /* i.e. not the first pass */ 2786 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2787 if (error) 2788 return (error); 2789 zl->zl_znode = zp; 2790 } 2791 oidp = &zp->z_phys->zp_parent; 2792 rwlp = &zp->z_parent_lock; 2793 rw = RW_READER; 2794 2795 } while (zp->z_id != sdzp->z_id); 2796 2797 return (0); 2798 } 2799 2800 /* 2801 * Move an entry from the provided source directory to the target 2802 * directory. Change the entry name as indicated. 2803 * 2804 * IN: sdvp - Source directory containing the "old entry". 2805 * snm - Old entry name. 2806 * tdvp - Target directory to contain the "new entry". 2807 * tnm - New entry name. 2808 * cr - credentials of caller. 2809 * ct - caller context 2810 * flags - case flags 2811 * 2812 * RETURN: 0 if success 2813 * error code if failure 2814 * 2815 * Timestamps: 2816 * sdvp,tdvp - ctime|mtime updated 2817 */ 2818 /*ARGSUSED*/ 2819 static int 2820 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 2821 caller_context_t *ct, int flags) 2822 { 2823 znode_t *tdzp, *szp, *tzp; 2824 znode_t *sdzp = VTOZ(sdvp); 2825 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2826 zilog_t *zilog; 2827 vnode_t *realvp; 2828 zfs_dirlock_t *sdl, *tdl; 2829 dmu_tx_t *tx; 2830 zfs_zlock_t *zl; 2831 int cmp, serr, terr; 2832 int error = 0; 2833 int zflg = 0; 2834 2835 ZFS_ENTER(zfsvfs); 2836 ZFS_VERIFY_ZP(sdzp); 2837 zilog = zfsvfs->z_log; 2838 2839 /* 2840 * Make sure we have the real vp for the target directory. 2841 */ 2842 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 2843 tdvp = realvp; 2844 2845 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2846 ZFS_EXIT(zfsvfs); 2847 return (EXDEV); 2848 } 2849 2850 tdzp = VTOZ(tdvp); 2851 ZFS_VERIFY_ZP(tdzp); 2852 if (zfsvfs->z_case & ZFS_UTF8_ONLY && u8_validate(tnm, 2853 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2854 ZFS_EXIT(zfsvfs); 2855 return (EILSEQ); 2856 } 2857 2858 if (flags & FIGNORECASE) 2859 zflg |= ZCILOOK; 2860 2861 top: 2862 szp = NULL; 2863 tzp = NULL; 2864 zl = NULL; 2865 2866 /* 2867 * This is to prevent the creation of links into attribute space 2868 * by renaming a linked file into/outof an attribute directory. 2869 * See the comment in zfs_link() for why this is considered bad. 2870 */ 2871 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2872 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2873 ZFS_EXIT(zfsvfs); 2874 return (EINVAL); 2875 } 2876 2877 /* 2878 * Lock source and target directory entries. To prevent deadlock, 2879 * a lock ordering must be defined. We lock the directory with 2880 * the smallest object id first, or if it's a tie, the one with 2881 * the lexically first name. 2882 */ 2883 if (sdzp->z_id < tdzp->z_id) { 2884 cmp = -1; 2885 } else if (sdzp->z_id > tdzp->z_id) { 2886 cmp = 1; 2887 } else { 2888 /* 2889 * First compare the two name arguments without 2890 * considering any case folding. 2891 */ 2892 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 2893 2894 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 2895 ASSERT(error == 0 || !(zfsvfs->z_case & ZFS_UTF8_ONLY)); 2896 if (cmp == 0) { 2897 /* 2898 * POSIX: "If the old argument and the new argument 2899 * both refer to links to the same existing file, 2900 * the rename() function shall return successfully 2901 * and perform no other action." 2902 */ 2903 ZFS_EXIT(zfsvfs); 2904 return (0); 2905 } 2906 /* 2907 * If the file system is case-folding, then we may 2908 * have some more checking to do. A case-folding file 2909 * system is either supporting mixed case sensitivity 2910 * access or is completely case-insensitive. Note 2911 * that the file system is always case preserving. 2912 * 2913 * In mixed sensitivity mode case sensitive behavior 2914 * is the default. FIGNORECASE must be used to 2915 * explicitly request case insensitive behavior. 2916 * 2917 * If the source and target names provided differ only 2918 * by case (e.g., a request to rename 'tim' to 'Tim'), 2919 * we will treat this as a special case in the 2920 * case-insensitive mode: as long as the source name 2921 * is an exact match, we will allow this to proceed as 2922 * a name-change request. 2923 */ 2924 if ((zfsvfs->z_case & ZFS_CI_ONLY || 2925 (zfsvfs->z_case & ZFS_CI_MIXD && flags & FIGNORECASE)) && 2926 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 2927 &error) == 0) { 2928 /* 2929 * case preserving rename request, require exact 2930 * name matches 2931 */ 2932 zflg |= ZCIEXACT; 2933 zflg &= ~ZCILOOK; 2934 } 2935 } 2936 2937 if (cmp < 0) { 2938 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 2939 ZEXISTS | zflg, NULL, NULL); 2940 terr = zfs_dirent_lock(&tdl, 2941 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 2942 } else { 2943 terr = zfs_dirent_lock(&tdl, 2944 tdzp, tnm, &tzp, zflg, NULL, NULL); 2945 serr = zfs_dirent_lock(&sdl, 2946 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 2947 NULL, NULL); 2948 } 2949 2950 if (serr) { 2951 /* 2952 * Source entry invalid or not there. 2953 */ 2954 if (!terr) { 2955 zfs_dirent_unlock(tdl); 2956 if (tzp) 2957 VN_RELE(ZTOV(tzp)); 2958 } 2959 if (strcmp(snm, "..") == 0) 2960 serr = EINVAL; 2961 ZFS_EXIT(zfsvfs); 2962 return (serr); 2963 } 2964 if (terr) { 2965 zfs_dirent_unlock(sdl); 2966 VN_RELE(ZTOV(szp)); 2967 if (strcmp(tnm, "..") == 0) 2968 terr = EINVAL; 2969 ZFS_EXIT(zfsvfs); 2970 return (terr); 2971 } 2972 2973 /* 2974 * Must have write access at the source to remove the old entry 2975 * and write access at the target to create the new entry. 2976 * Note that if target and source are the same, this can be 2977 * done in a single check. 2978 */ 2979 2980 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2981 goto out; 2982 2983 if (ZTOV(szp)->v_type == VDIR) { 2984 /* 2985 * Check to make sure rename is valid. 2986 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2987 */ 2988 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 2989 goto out; 2990 } 2991 2992 /* 2993 * Does target exist? 2994 */ 2995 if (tzp) { 2996 /* 2997 * Source and target must be the same type. 2998 */ 2999 if (ZTOV(szp)->v_type == VDIR) { 3000 if (ZTOV(tzp)->v_type != VDIR) { 3001 error = ENOTDIR; 3002 goto out; 3003 } 3004 } else { 3005 if (ZTOV(tzp)->v_type == VDIR) { 3006 error = EISDIR; 3007 goto out; 3008 } 3009 } 3010 /* 3011 * POSIX dictates that when the source and target 3012 * entries refer to the same file object, rename 3013 * must do nothing and exit without error. 3014 */ 3015 if (szp->z_id == tzp->z_id) { 3016 error = 0; 3017 goto out; 3018 } 3019 } 3020 3021 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3022 if (tzp) 3023 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3024 3025 /* 3026 * notify the target directory if it is not the same 3027 * as source directory. 3028 */ 3029 if (tdvp != sdvp) { 3030 vnevent_rename_dest_dir(tdvp, ct); 3031 } 3032 3033 tx = dmu_tx_create(zfsvfs->z_os); 3034 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 3035 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 3036 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3037 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3038 if (sdzp != tdzp) 3039 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 3040 if (tzp) 3041 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 3042 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3043 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3044 if (error) { 3045 if (zl != NULL) 3046 zfs_rename_unlock(&zl); 3047 zfs_dirent_unlock(sdl); 3048 zfs_dirent_unlock(tdl); 3049 VN_RELE(ZTOV(szp)); 3050 if (tzp) 3051 VN_RELE(ZTOV(tzp)); 3052 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3053 dmu_tx_wait(tx); 3054 dmu_tx_abort(tx); 3055 goto top; 3056 } 3057 dmu_tx_abort(tx); 3058 ZFS_EXIT(zfsvfs); 3059 return (error); 3060 } 3061 3062 if (tzp) /* Attempt to remove the existing target */ 3063 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 3064 3065 if (error == 0) { 3066 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 3067 if (error == 0) { 3068 szp->z_phys->zp_flags |= ZFS_AV_MODIFIED; 3069 3070 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 3071 ASSERT(error == 0); 3072 3073 zfs_log_rename(zilog, tx, 3074 TX_RENAME | (flags & FIGNORECASE ? TX_CI : 0), 3075 sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp); 3076 } 3077 } 3078 3079 dmu_tx_commit(tx); 3080 out: 3081 if (zl != NULL) 3082 zfs_rename_unlock(&zl); 3083 3084 zfs_dirent_unlock(sdl); 3085 zfs_dirent_unlock(tdl); 3086 3087 VN_RELE(ZTOV(szp)); 3088 if (tzp) 3089 VN_RELE(ZTOV(tzp)); 3090 3091 ZFS_EXIT(zfsvfs); 3092 return (error); 3093 } 3094 3095 /* 3096 * Insert the indicated symbolic reference entry into the directory. 3097 * 3098 * IN: dvp - Directory to contain new symbolic link. 3099 * link - Name for new symlink entry. 3100 * vap - Attributes of new entry. 3101 * target - Target path of new symlink. 3102 * cr - credentials of caller. 3103 * ct - caller context 3104 * flags - case flags 3105 * 3106 * RETURN: 0 if success 3107 * error code if failure 3108 * 3109 * Timestamps: 3110 * dvp - ctime|mtime updated 3111 */ 3112 /*ARGSUSED*/ 3113 static int 3114 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr, 3115 caller_context_t *ct, int flags) 3116 { 3117 znode_t *zp, *dzp = VTOZ(dvp); 3118 zfs_dirlock_t *dl; 3119 dmu_tx_t *tx; 3120 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3121 zilog_t *zilog; 3122 int len = strlen(link); 3123 int error; 3124 int zflg = ZNEW; 3125 zfs_fuid_info_t *fuidp = NULL; 3126 3127 ASSERT(vap->va_type == VLNK); 3128 3129 ZFS_ENTER(zfsvfs); 3130 ZFS_VERIFY_ZP(dzp); 3131 zilog = zfsvfs->z_log; 3132 3133 if (zfsvfs->z_case & ZFS_UTF8_ONLY && u8_validate(name, strlen(name), 3134 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3135 ZFS_EXIT(zfsvfs); 3136 return (EILSEQ); 3137 } 3138 if (flags & FIGNORECASE) 3139 zflg |= ZCILOOK; 3140 top: 3141 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3142 ZFS_EXIT(zfsvfs); 3143 return (error); 3144 } 3145 3146 if (len > MAXPATHLEN) { 3147 ZFS_EXIT(zfsvfs); 3148 return (ENAMETOOLONG); 3149 } 3150 3151 /* 3152 * Attempt to lock directory; fail if entry already exists. 3153 */ 3154 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 3155 if (error) { 3156 ZFS_EXIT(zfsvfs); 3157 return (error); 3158 } 3159 3160 tx = dmu_tx_create(zfsvfs->z_os); 3161 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 3162 dmu_tx_hold_bonus(tx, dzp->z_id); 3163 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3164 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 3165 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 3166 if (zfsvfs->z_fuid_obj == 0) { 3167 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 3168 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 3169 SPA_MAXBLOCKSIZE); 3170 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 3171 } else { 3172 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 3173 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 3174 SPA_MAXBLOCKSIZE); 3175 } 3176 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3177 if (error) { 3178 zfs_dirent_unlock(dl); 3179 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3180 dmu_tx_wait(tx); 3181 dmu_tx_abort(tx); 3182 goto top; 3183 } 3184 dmu_tx_abort(tx); 3185 ZFS_EXIT(zfsvfs); 3186 return (error); 3187 } 3188 3189 dmu_buf_will_dirty(dzp->z_dbuf, tx); 3190 3191 /* 3192 * Create a new object for the symlink. 3193 * Put the link content into bonus buffer if it will fit; 3194 * otherwise, store it just like any other file data. 3195 */ 3196 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 3197 zfs_mknode(dzp, vap, tx, cr, 0, &zp, len, NULL, &fuidp); 3198 if (len != 0) 3199 bcopy(link, zp->z_phys + 1, len); 3200 } else { 3201 dmu_buf_t *dbp; 3202 3203 zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, NULL, &fuidp); 3204 /* 3205 * Nothing can access the znode yet so no locking needed 3206 * for growing the znode's blocksize. 3207 */ 3208 zfs_grow_blocksize(zp, len, tx); 3209 3210 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, 3211 zp->z_id, 0, FTAG, &dbp)); 3212 dmu_buf_will_dirty(dbp, tx); 3213 3214 ASSERT3U(len, <=, dbp->db_size); 3215 bcopy(link, dbp->db_data, len); 3216 dmu_buf_rele(dbp, FTAG); 3217 } 3218 zp->z_phys->zp_size = len; 3219 3220 /* 3221 * Insert the new object into the directory. 3222 */ 3223 (void) zfs_link_create(dl, zp, tx, ZNEW); 3224 out: 3225 if (error == 0) { 3226 uint64_t txtype = TX_SYMLINK; 3227 if (flags & FIGNORECASE) 3228 txtype |= TX_CI; 3229 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 3230 } 3231 if (fuidp) 3232 zfs_fuid_info_free(fuidp); 3233 3234 dmu_tx_commit(tx); 3235 3236 zfs_dirent_unlock(dl); 3237 3238 VN_RELE(ZTOV(zp)); 3239 3240 ZFS_EXIT(zfsvfs); 3241 return (error); 3242 } 3243 3244 /* 3245 * Return, in the buffer contained in the provided uio structure, 3246 * the symbolic path referred to by vp. 3247 * 3248 * IN: vp - vnode of symbolic link. 3249 * uoip - structure to contain the link path. 3250 * cr - credentials of caller. 3251 * ct - caller context 3252 * 3253 * OUT: uio - structure to contain the link path. 3254 * 3255 * RETURN: 0 if success 3256 * error code if failure 3257 * 3258 * Timestamps: 3259 * vp - atime updated 3260 */ 3261 /* ARGSUSED */ 3262 static int 3263 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 3264 { 3265 znode_t *zp = VTOZ(vp); 3266 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3267 size_t bufsz; 3268 int error; 3269 3270 ZFS_ENTER(zfsvfs); 3271 ZFS_VERIFY_ZP(zp); 3272 3273 bufsz = (size_t)zp->z_phys->zp_size; 3274 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 3275 error = uiomove(zp->z_phys + 1, 3276 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 3277 } else { 3278 dmu_buf_t *dbp; 3279 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 3280 if (error) { 3281 ZFS_EXIT(zfsvfs); 3282 return (error); 3283 } 3284 error = uiomove(dbp->db_data, 3285 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 3286 dmu_buf_rele(dbp, FTAG); 3287 } 3288 3289 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3290 ZFS_EXIT(zfsvfs); 3291 return (error); 3292 } 3293 3294 /* 3295 * Insert a new entry into directory tdvp referencing svp. 3296 * 3297 * IN: tdvp - Directory to contain new entry. 3298 * svp - vnode of new entry. 3299 * name - name of new entry. 3300 * cr - credentials of caller. 3301 * ct - caller context 3302 * 3303 * RETURN: 0 if success 3304 * error code if failure 3305 * 3306 * Timestamps: 3307 * tdvp - ctime|mtime updated 3308 * svp - ctime updated 3309 */ 3310 /* ARGSUSED */ 3311 static int 3312 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 3313 caller_context_t *ct, int flags) 3314 { 3315 znode_t *dzp = VTOZ(tdvp); 3316 znode_t *tzp, *szp; 3317 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3318 zilog_t *zilog; 3319 zfs_dirlock_t *dl; 3320 dmu_tx_t *tx; 3321 vnode_t *realvp; 3322 int error; 3323 int zf = ZNEW; 3324 uid_t owner; 3325 3326 ASSERT(tdvp->v_type == VDIR); 3327 3328 ZFS_ENTER(zfsvfs); 3329 ZFS_VERIFY_ZP(dzp); 3330 zilog = zfsvfs->z_log; 3331 3332 if (VOP_REALVP(svp, &realvp, ct) == 0) 3333 svp = realvp; 3334 3335 if (svp->v_vfsp != tdvp->v_vfsp) { 3336 ZFS_EXIT(zfsvfs); 3337 return (EXDEV); 3338 } 3339 szp = VTOZ(svp); 3340 ZFS_VERIFY_ZP(szp); 3341 3342 if (zfsvfs->z_case & ZFS_UTF8_ONLY && u8_validate(name, 3343 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3344 ZFS_EXIT(zfsvfs); 3345 return (EILSEQ); 3346 } 3347 if (flags & FIGNORECASE) 3348 zf |= ZCILOOK; 3349 3350 top: 3351 /* 3352 * We do not support links between attributes and non-attributes 3353 * because of the potential security risk of creating links 3354 * into "normal" file space in order to circumvent restrictions 3355 * imposed in attribute space. 3356 */ 3357 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 3358 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 3359 ZFS_EXIT(zfsvfs); 3360 return (EINVAL); 3361 } 3362 3363 /* 3364 * POSIX dictates that we return EPERM here. 3365 * Better choices include ENOTSUP or EISDIR. 3366 */ 3367 if (svp->v_type == VDIR) { 3368 ZFS_EXIT(zfsvfs); 3369 return (EPERM); 3370 } 3371 3372 zfs_fuid_map_id(zfsvfs, szp->z_phys->zp_uid, ZFS_OWNER, &owner); 3373 if (owner != crgetuid(cr) && 3374 secpolicy_basic_link(cr) != 0) { 3375 ZFS_EXIT(zfsvfs); 3376 return (EPERM); 3377 } 3378 3379 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3380 ZFS_EXIT(zfsvfs); 3381 return (error); 3382 } 3383 3384 /* 3385 * Attempt to lock directory; fail if entry already exists. 3386 */ 3387 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 3388 if (error) { 3389 ZFS_EXIT(zfsvfs); 3390 return (error); 3391 } 3392 3393 tx = dmu_tx_create(zfsvfs->z_os); 3394 dmu_tx_hold_bonus(tx, szp->z_id); 3395 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3396 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3397 if (error) { 3398 zfs_dirent_unlock(dl); 3399 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3400 dmu_tx_wait(tx); 3401 dmu_tx_abort(tx); 3402 goto top; 3403 } 3404 dmu_tx_abort(tx); 3405 ZFS_EXIT(zfsvfs); 3406 return (error); 3407 } 3408 3409 error = zfs_link_create(dl, szp, tx, 0); 3410 3411 if (error == 0) { 3412 uint64_t txtype = TX_LINK; 3413 if (flags & FIGNORECASE) 3414 txtype |= TX_CI; 3415 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 3416 } 3417 3418 dmu_tx_commit(tx); 3419 3420 zfs_dirent_unlock(dl); 3421 3422 if (error == 0) { 3423 vnevent_link(svp, ct); 3424 } 3425 3426 ZFS_EXIT(zfsvfs); 3427 return (error); 3428 } 3429 3430 /* 3431 * zfs_null_putapage() is used when the file system has been force 3432 * unmounted. It just drops the pages. 3433 */ 3434 /* ARGSUSED */ 3435 static int 3436 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 3437 size_t *lenp, int flags, cred_t *cr) 3438 { 3439 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 3440 return (0); 3441 } 3442 3443 /* 3444 * Push a page out to disk, klustering if possible. 3445 * 3446 * IN: vp - file to push page to. 3447 * pp - page to push. 3448 * flags - additional flags. 3449 * cr - credentials of caller. 3450 * 3451 * OUT: offp - start of range pushed. 3452 * lenp - len of range pushed. 3453 * 3454 * RETURN: 0 if success 3455 * error code if failure 3456 * 3457 * NOTE: callers must have locked the page to be pushed. On 3458 * exit, the page (and all other pages in the kluster) must be 3459 * unlocked. 3460 */ 3461 /* ARGSUSED */ 3462 static int 3463 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 3464 size_t *lenp, int flags, cred_t *cr) 3465 { 3466 znode_t *zp = VTOZ(vp); 3467 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3468 zilog_t *zilog = zfsvfs->z_log; 3469 dmu_tx_t *tx; 3470 rl_t *rl; 3471 u_offset_t off, koff; 3472 size_t len, klen; 3473 uint64_t filesz; 3474 int err; 3475 3476 filesz = zp->z_phys->zp_size; 3477 off = pp->p_offset; 3478 len = PAGESIZE; 3479 /* 3480 * If our blocksize is bigger than the page size, try to kluster 3481 * muiltiple pages so that we write a full block (thus avoiding 3482 * a read-modify-write). 3483 */ 3484 if (off < filesz && zp->z_blksz > PAGESIZE) { 3485 if (!ISP2(zp->z_blksz)) { 3486 /* Only one block in the file. */ 3487 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3488 koff = 0; 3489 } else { 3490 klen = zp->z_blksz; 3491 koff = P2ALIGN(off, (u_offset_t)klen); 3492 } 3493 ASSERT(koff <= filesz); 3494 if (koff + klen > filesz) 3495 klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE); 3496 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 3497 } 3498 ASSERT3U(btop(len), ==, btopr(len)); 3499 top: 3500 rl = zfs_range_lock(zp, off, len, RL_WRITER); 3501 /* 3502 * Can't push pages past end-of-file. 3503 */ 3504 filesz = zp->z_phys->zp_size; 3505 if (off >= filesz) { 3506 /* ignore all pages */ 3507 err = 0; 3508 goto out; 3509 } else if (off + len > filesz) { 3510 int npages = btopr(filesz - off); 3511 page_t *trunc; 3512 3513 page_list_break(&pp, &trunc, npages); 3514 /* ignore pages past end of file */ 3515 if (trunc) 3516 pvn_write_done(trunc, flags); 3517 len = filesz - off; 3518 } 3519 3520 tx = dmu_tx_create(zfsvfs->z_os); 3521 dmu_tx_hold_write(tx, zp->z_id, off, len); 3522 dmu_tx_hold_bonus(tx, zp->z_id); 3523 err = dmu_tx_assign(tx, zfsvfs->z_assign); 3524 if (err != 0) { 3525 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3526 zfs_range_unlock(rl); 3527 dmu_tx_wait(tx); 3528 dmu_tx_abort(tx); 3529 err = 0; 3530 goto top; 3531 } 3532 dmu_tx_abort(tx); 3533 goto out; 3534 } 3535 3536 if (zp->z_blksz <= PAGESIZE) { 3537 caddr_t va = ppmapin(pp, PROT_READ, (caddr_t)-1); 3538 ASSERT3U(len, <=, PAGESIZE); 3539 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 3540 ppmapout(va); 3541 } else { 3542 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 3543 } 3544 3545 if (err == 0) { 3546 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 3547 zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0); 3548 dmu_tx_commit(tx); 3549 } 3550 3551 out: 3552 zfs_range_unlock(rl); 3553 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 3554 if (offp) 3555 *offp = off; 3556 if (lenp) 3557 *lenp = len; 3558 3559 return (err); 3560 } 3561 3562 /* 3563 * Copy the portion of the file indicated from pages into the file. 3564 * The pages are stored in a page list attached to the files vnode. 3565 * 3566 * IN: vp - vnode of file to push page data to. 3567 * off - position in file to put data. 3568 * len - amount of data to write. 3569 * flags - flags to control the operation. 3570 * cr - credentials of caller. 3571 * ct - caller context. 3572 * 3573 * RETURN: 0 if success 3574 * error code if failure 3575 * 3576 * Timestamps: 3577 * vp - ctime|mtime updated 3578 */ 3579 /*ARGSUSED*/ 3580 static int 3581 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 3582 caller_context_t *ct) 3583 { 3584 znode_t *zp = VTOZ(vp); 3585 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3586 page_t *pp; 3587 size_t io_len; 3588 u_offset_t io_off; 3589 uint64_t filesz; 3590 int error = 0; 3591 3592 ZFS_ENTER(zfsvfs); 3593 ZFS_VERIFY_ZP(zp); 3594 3595 if (len == 0) { 3596 /* 3597 * Search the entire vp list for pages >= off. 3598 */ 3599 error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage, 3600 flags, cr); 3601 goto out; 3602 } 3603 3604 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3605 if (off > filesz) { 3606 /* past end of file */ 3607 ZFS_EXIT(zfsvfs); 3608 return (0); 3609 } 3610 3611 len = MIN(len, filesz - off); 3612 3613 for (io_off = off; io_off < off + len; io_off += io_len) { 3614 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 3615 pp = page_lookup(vp, io_off, 3616 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 3617 } else { 3618 pp = page_lookup_nowait(vp, io_off, 3619 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 3620 } 3621 3622 if (pp != NULL && pvn_getdirty(pp, flags)) { 3623 int err; 3624 3625 /* 3626 * Found a dirty page to push 3627 */ 3628 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 3629 if (err) 3630 error = err; 3631 } else { 3632 io_len = PAGESIZE; 3633 } 3634 } 3635 out: 3636 if ((flags & B_ASYNC) == 0) 3637 zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id); 3638 ZFS_EXIT(zfsvfs); 3639 return (error); 3640 } 3641 3642 /*ARGSUSED*/ 3643 void 3644 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 3645 { 3646 znode_t *zp = VTOZ(vp); 3647 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3648 int error; 3649 3650 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 3651 if (zp->z_dbuf == NULL) { 3652 /* 3653 * This fs has been unmounted, or we did 3654 * zfs_suspend/resume and it no longer exists. 3655 */ 3656 if (vn_has_cached_data(vp)) { 3657 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 3658 B_INVAL, cr); 3659 } 3660 3661 mutex_enter(&zp->z_lock); 3662 vp->v_count = 0; /* count arrives as 1 */ 3663 mutex_exit(&zp->z_lock); 3664 zfs_znode_free(zp); 3665 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3666 VFS_RELE(zfsvfs->z_vfs); 3667 return; 3668 } 3669 3670 /* 3671 * Attempt to push any data in the page cache. If this fails 3672 * we will get kicked out later in zfs_zinactive(). 3673 */ 3674 if (vn_has_cached_data(vp)) { 3675 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 3676 cr); 3677 } 3678 3679 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 3680 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 3681 3682 dmu_tx_hold_bonus(tx, zp->z_id); 3683 error = dmu_tx_assign(tx, TXG_WAIT); 3684 if (error) { 3685 dmu_tx_abort(tx); 3686 } else { 3687 dmu_buf_will_dirty(zp->z_dbuf, tx); 3688 mutex_enter(&zp->z_lock); 3689 zp->z_atime_dirty = 0; 3690 mutex_exit(&zp->z_lock); 3691 dmu_tx_commit(tx); 3692 } 3693 } 3694 3695 zfs_zinactive(zp); 3696 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3697 } 3698 3699 /* 3700 * Bounds-check the seek operation. 3701 * 3702 * IN: vp - vnode seeking within 3703 * ooff - old file offset 3704 * noffp - pointer to new file offset 3705 * ct - caller context 3706 * 3707 * RETURN: 0 if success 3708 * EINVAL if new offset invalid 3709 */ 3710 /* ARGSUSED */ 3711 static int 3712 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 3713 caller_context_t *ct) 3714 { 3715 if (vp->v_type == VDIR) 3716 return (0); 3717 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 3718 } 3719 3720 /* 3721 * Pre-filter the generic locking function to trap attempts to place 3722 * a mandatory lock on a memory mapped file. 3723 */ 3724 static int 3725 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 3726 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 3727 { 3728 znode_t *zp = VTOZ(vp); 3729 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3730 int error; 3731 3732 ZFS_ENTER(zfsvfs); 3733 ZFS_VERIFY_ZP(zp); 3734 3735 /* 3736 * We are following the UFS semantics with respect to mapcnt 3737 * here: If we see that the file is mapped already, then we will 3738 * return an error, but we don't worry about races between this 3739 * function and zfs_map(). 3740 */ 3741 if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) { 3742 ZFS_EXIT(zfsvfs); 3743 return (EAGAIN); 3744 } 3745 error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct); 3746 ZFS_EXIT(zfsvfs); 3747 return (error); 3748 } 3749 3750 /* 3751 * If we can't find a page in the cache, we will create a new page 3752 * and fill it with file data. For efficiency, we may try to fill 3753 * multiple pages at once (klustering). 3754 */ 3755 static int 3756 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 3757 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 3758 { 3759 znode_t *zp = VTOZ(vp); 3760 page_t *pp, *cur_pp; 3761 objset_t *os = zp->z_zfsvfs->z_os; 3762 caddr_t va; 3763 u_offset_t io_off, total; 3764 uint64_t oid = zp->z_id; 3765 size_t io_len; 3766 uint64_t filesz; 3767 int err; 3768 3769 /* 3770 * If we are only asking for a single page don't bother klustering. 3771 */ 3772 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3773 if (off >= filesz) 3774 return (EFAULT); 3775 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 3776 io_off = off; 3777 io_len = PAGESIZE; 3778 pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr); 3779 } else { 3780 /* 3781 * Try to fill a kluster of pages (a blocks worth). 3782 */ 3783 size_t klen; 3784 u_offset_t koff; 3785 3786 if (!ISP2(zp->z_blksz)) { 3787 /* Only one block in the file. */ 3788 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3789 koff = 0; 3790 } else { 3791 /* 3792 * It would be ideal to align our offset to the 3793 * blocksize but doing so has resulted in some 3794 * strange application crashes. For now, we 3795 * leave the offset as is and only adjust the 3796 * length if we are off the end of the file. 3797 */ 3798 koff = off; 3799 klen = plsz; 3800 } 3801 ASSERT(koff <= filesz); 3802 if (koff + klen > filesz) 3803 klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff; 3804 ASSERT3U(off, >=, koff); 3805 ASSERT3U(off, <, koff + klen); 3806 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 3807 &io_len, koff, klen, 0); 3808 } 3809 if (pp == NULL) { 3810 /* 3811 * Some other thread entered the page before us. 3812 * Return to zfs_getpage to retry the lookup. 3813 */ 3814 *pl = NULL; 3815 return (0); 3816 } 3817 3818 /* 3819 * Fill the pages in the kluster. 3820 */ 3821 cur_pp = pp; 3822 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 3823 ASSERT3U(io_off, ==, cur_pp->p_offset); 3824 va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 3825 err = dmu_read(os, oid, io_off, PAGESIZE, va); 3826 ppmapout(va); 3827 if (err) { 3828 /* On error, toss the entire kluster */ 3829 pvn_read_done(pp, B_ERROR); 3830 return (err); 3831 } 3832 cur_pp = cur_pp->p_next; 3833 } 3834 out: 3835 /* 3836 * Fill in the page list array from the kluster. If 3837 * there are too many pages in the kluster, return 3838 * as many pages as possible starting from the desired 3839 * offset `off'. 3840 * NOTE: the page list will always be null terminated. 3841 */ 3842 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 3843 3844 return (0); 3845 } 3846 3847 /* 3848 * Return pointers to the pages for the file region [off, off + len] 3849 * in the pl array. If plsz is greater than len, this function may 3850 * also return page pointers from before or after the specified 3851 * region (i.e. some region [off', off' + plsz]). These additional 3852 * pages are only returned if they are already in the cache, or were 3853 * created as part of a klustered read. 3854 * 3855 * IN: vp - vnode of file to get data from. 3856 * off - position in file to get data from. 3857 * len - amount of data to retrieve. 3858 * plsz - length of provided page list. 3859 * seg - segment to obtain pages for. 3860 * addr - virtual address of fault. 3861 * rw - mode of created pages. 3862 * cr - credentials of caller. 3863 * ct - caller context. 3864 * 3865 * OUT: protp - protection mode of created pages. 3866 * pl - list of pages created. 3867 * 3868 * RETURN: 0 if success 3869 * error code if failure 3870 * 3871 * Timestamps: 3872 * vp - atime updated 3873 */ 3874 /* ARGSUSED */ 3875 static int 3876 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 3877 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 3878 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 3879 { 3880 znode_t *zp = VTOZ(vp); 3881 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3882 page_t *pp, **pl0 = pl; 3883 int need_unlock = 0, err = 0; 3884 offset_t orig_off; 3885 3886 ZFS_ENTER(zfsvfs); 3887 ZFS_VERIFY_ZP(zp); 3888 3889 if (protp) 3890 *protp = PROT_ALL; 3891 3892 /* no faultahead (for now) */ 3893 if (pl == NULL) { 3894 ZFS_EXIT(zfsvfs); 3895 return (0); 3896 } 3897 3898 /* can't fault past EOF */ 3899 if (off >= zp->z_phys->zp_size) { 3900 ZFS_EXIT(zfsvfs); 3901 return (EFAULT); 3902 } 3903 orig_off = off; 3904 3905 /* 3906 * If we already own the lock, then we must be page faulting 3907 * in the middle of a write to this file (i.e., we are writing 3908 * to this file using data from a mapped region of the file). 3909 */ 3910 if (rw_owner(&zp->z_map_lock) != curthread) { 3911 rw_enter(&zp->z_map_lock, RW_WRITER); 3912 need_unlock = TRUE; 3913 } 3914 3915 /* 3916 * Loop through the requested range [off, off + len] looking 3917 * for pages. If we don't find a page, we will need to create 3918 * a new page and fill it with data from the file. 3919 */ 3920 while (len > 0) { 3921 if (plsz < PAGESIZE) 3922 break; 3923 if (pp = page_lookup(vp, off, SE_SHARED)) { 3924 *pl++ = pp; 3925 off += PAGESIZE; 3926 addr += PAGESIZE; 3927 len -= PAGESIZE; 3928 plsz -= PAGESIZE; 3929 } else { 3930 err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw); 3931 if (err) 3932 goto out; 3933 /* 3934 * klustering may have changed our region 3935 * to be block aligned. 3936 */ 3937 if (((pp = *pl) != 0) && (off != pp->p_offset)) { 3938 int delta = off - pp->p_offset; 3939 len += delta; 3940 off -= delta; 3941 addr -= delta; 3942 } 3943 while (*pl) { 3944 pl++; 3945 off += PAGESIZE; 3946 addr += PAGESIZE; 3947 plsz -= PAGESIZE; 3948 if (len > PAGESIZE) 3949 len -= PAGESIZE; 3950 else 3951 len = 0; 3952 } 3953 } 3954 } 3955 3956 /* 3957 * Fill out the page array with any pages already in the cache. 3958 */ 3959 while (plsz > 0) { 3960 pp = page_lookup_nowait(vp, off, SE_SHARED); 3961 if (pp == NULL) 3962 break; 3963 *pl++ = pp; 3964 off += PAGESIZE; 3965 plsz -= PAGESIZE; 3966 } 3967 3968 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3969 out: 3970 /* 3971 * We can't grab the range lock for the page as reader which would 3972 * stop truncation as this leads to deadlock. So we need to recheck 3973 * the file size. 3974 */ 3975 if (orig_off >= zp->z_phys->zp_size) 3976 err = EFAULT; 3977 if (err) { 3978 /* 3979 * Release any pages we have previously locked. 3980 */ 3981 while (pl > pl0) 3982 page_unlock(*--pl); 3983 } 3984 3985 *pl = NULL; 3986 3987 if (need_unlock) 3988 rw_exit(&zp->z_map_lock); 3989 3990 ZFS_EXIT(zfsvfs); 3991 return (err); 3992 } 3993 3994 /* 3995 * Request a memory map for a section of a file. This code interacts 3996 * with common code and the VM system as follows: 3997 * 3998 * common code calls mmap(), which ends up in smmap_common() 3999 * 4000 * this calls VOP_MAP(), which takes you into (say) zfs 4001 * 4002 * zfs_map() calls as_map(), passing segvn_create() as the callback 4003 * 4004 * segvn_create() creates the new segment and calls VOP_ADDMAP() 4005 * 4006 * zfs_addmap() updates z_mapcnt 4007 */ 4008 /*ARGSUSED*/ 4009 static int 4010 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4011 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4012 caller_context_t *ct) 4013 { 4014 znode_t *zp = VTOZ(vp); 4015 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4016 segvn_crargs_t vn_a; 4017 int error; 4018 4019 if ((prot & PROT_WRITE) && 4020 (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_READONLY | 4021 ZFS_APPENDONLY))) 4022 return (EPERM); 4023 4024 ZFS_ENTER(zfsvfs); 4025 ZFS_VERIFY_ZP(zp); 4026 4027 if (vp->v_flag & VNOMAP) { 4028 ZFS_EXIT(zfsvfs); 4029 return (ENOSYS); 4030 } 4031 4032 if (off < 0 || len > MAXOFFSET_T - off) { 4033 ZFS_EXIT(zfsvfs); 4034 return (ENXIO); 4035 } 4036 4037 if (vp->v_type != VREG) { 4038 ZFS_EXIT(zfsvfs); 4039 return (ENODEV); 4040 } 4041 4042 /* 4043 * If file is locked, disallow mapping. 4044 */ 4045 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) { 4046 ZFS_EXIT(zfsvfs); 4047 return (EAGAIN); 4048 } 4049 4050 as_rangelock(as); 4051 if ((flags & MAP_FIXED) == 0) { 4052 map_addr(addrp, len, off, 1, flags); 4053 if (*addrp == NULL) { 4054 as_rangeunlock(as); 4055 ZFS_EXIT(zfsvfs); 4056 return (ENOMEM); 4057 } 4058 } else { 4059 /* 4060 * User specified address - blow away any previous mappings 4061 */ 4062 (void) as_unmap(as, *addrp, len); 4063 } 4064 4065 vn_a.vp = vp; 4066 vn_a.offset = (u_offset_t)off; 4067 vn_a.type = flags & MAP_TYPE; 4068 vn_a.prot = prot; 4069 vn_a.maxprot = maxprot; 4070 vn_a.cred = cr; 4071 vn_a.amp = NULL; 4072 vn_a.flags = flags & ~MAP_TYPE; 4073 vn_a.szc = 0; 4074 vn_a.lgrp_mem_policy_flags = 0; 4075 4076 error = as_map(as, *addrp, len, segvn_create, &vn_a); 4077 4078 as_rangeunlock(as); 4079 ZFS_EXIT(zfsvfs); 4080 return (error); 4081 } 4082 4083 /* ARGSUSED */ 4084 static int 4085 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4086 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4087 caller_context_t *ct) 4088 { 4089 uint64_t pages = btopr(len); 4090 4091 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 4092 return (0); 4093 } 4094 4095 /* 4096 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 4097 * more accurate mtime for the associated file. Since we don't have a way of 4098 * detecting when the data was actually modified, we have to resort to 4099 * heuristics. If an explicit msync() is done, then we mark the mtime when the 4100 * last page is pushed. The problem occurs when the msync() call is omitted, 4101 * which by far the most common case: 4102 * 4103 * open() 4104 * mmap() 4105 * <modify memory> 4106 * munmap() 4107 * close() 4108 * <time lapse> 4109 * putpage() via fsflush 4110 * 4111 * If we wait until fsflush to come along, we can have a modification time that 4112 * is some arbitrary point in the future. In order to prevent this in the 4113 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 4114 * torn down. 4115 */ 4116 /* ARGSUSED */ 4117 static int 4118 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4119 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 4120 caller_context_t *ct) 4121 { 4122 uint64_t pages = btopr(len); 4123 4124 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 4125 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 4126 4127 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 4128 vn_has_cached_data(vp)) 4129 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 4130 4131 return (0); 4132 } 4133 4134 /* 4135 * Free or allocate space in a file. Currently, this function only 4136 * supports the `F_FREESP' command. However, this command is somewhat 4137 * misnamed, as its functionality includes the ability to allocate as 4138 * well as free space. 4139 * 4140 * IN: vp - vnode of file to free data in. 4141 * cmd - action to take (only F_FREESP supported). 4142 * bfp - section of file to free/alloc. 4143 * flag - current file open mode flags. 4144 * offset - current file offset. 4145 * cr - credentials of caller [UNUSED]. 4146 * ct - caller context. 4147 * 4148 * RETURN: 0 if success 4149 * error code if failure 4150 * 4151 * Timestamps: 4152 * vp - ctime|mtime updated 4153 */ 4154 /* ARGSUSED */ 4155 static int 4156 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 4157 offset_t offset, cred_t *cr, caller_context_t *ct) 4158 { 4159 znode_t *zp = VTOZ(vp); 4160 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4161 uint64_t off, len; 4162 int error; 4163 4164 ZFS_ENTER(zfsvfs); 4165 ZFS_VERIFY_ZP(zp); 4166 4167 top: 4168 if (cmd != F_FREESP) { 4169 ZFS_EXIT(zfsvfs); 4170 return (EINVAL); 4171 } 4172 4173 if (error = convoff(vp, bfp, 0, offset)) { 4174 ZFS_EXIT(zfsvfs); 4175 return (error); 4176 } 4177 4178 if (bfp->l_len < 0) { 4179 ZFS_EXIT(zfsvfs); 4180 return (EINVAL); 4181 } 4182 4183 off = bfp->l_start; 4184 len = bfp->l_len; /* 0 means from off to end of file */ 4185 4186 do { 4187 error = zfs_freesp(zp, off, len, flag, TRUE); 4188 /* NB: we already did dmu_tx_wait() if necessary */ 4189 } while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 4190 4191 ZFS_EXIT(zfsvfs); 4192 return (error); 4193 } 4194 4195 /*ARGSUSED*/ 4196 static int 4197 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 4198 { 4199 znode_t *zp = VTOZ(vp); 4200 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4201 uint32_t gen; 4202 uint64_t object = zp->z_id; 4203 zfid_short_t *zfid; 4204 int size, i; 4205 4206 ZFS_ENTER(zfsvfs); 4207 ZFS_VERIFY_ZP(zp); 4208 gen = (uint32_t)zp->z_gen; 4209 4210 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 4211 if (fidp->fid_len < size) { 4212 fidp->fid_len = size; 4213 ZFS_EXIT(zfsvfs); 4214 return (ENOSPC); 4215 } 4216 4217 zfid = (zfid_short_t *)fidp; 4218 4219 zfid->zf_len = size; 4220 4221 for (i = 0; i < sizeof (zfid->zf_object); i++) 4222 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 4223 4224 /* Must have a non-zero generation number to distinguish from .zfs */ 4225 if (gen == 0) 4226 gen = 1; 4227 for (i = 0; i < sizeof (zfid->zf_gen); i++) 4228 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 4229 4230 if (size == LONG_FID_LEN) { 4231 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 4232 zfid_long_t *zlfid; 4233 4234 zlfid = (zfid_long_t *)fidp; 4235 4236 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 4237 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 4238 4239 /* XXX - this should be the generation number for the objset */ 4240 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 4241 zlfid->zf_setgen[i] = 0; 4242 } 4243 4244 ZFS_EXIT(zfsvfs); 4245 return (0); 4246 } 4247 4248 static int 4249 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 4250 caller_context_t *ct) 4251 { 4252 znode_t *zp, *xzp; 4253 zfsvfs_t *zfsvfs; 4254 zfs_dirlock_t *dl; 4255 int error; 4256 4257 switch (cmd) { 4258 case _PC_LINK_MAX: 4259 *valp = ULONG_MAX; 4260 return (0); 4261 4262 case _PC_FILESIZEBITS: 4263 *valp = 64; 4264 return (0); 4265 4266 case _PC_XATTR_EXISTS: 4267 zp = VTOZ(vp); 4268 zfsvfs = zp->z_zfsvfs; 4269 ZFS_ENTER(zfsvfs); 4270 ZFS_VERIFY_ZP(zp); 4271 *valp = 0; 4272 error = zfs_dirent_lock(&dl, zp, "", &xzp, 4273 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 4274 if (error == 0) { 4275 zfs_dirent_unlock(dl); 4276 if (!zfs_dirempty(xzp)) 4277 *valp = 1; 4278 VN_RELE(ZTOV(xzp)); 4279 } else if (error == ENOENT) { 4280 /* 4281 * If there aren't extended attributes, it's the 4282 * same as having zero of them. 4283 */ 4284 error = 0; 4285 } 4286 ZFS_EXIT(zfsvfs); 4287 return (error); 4288 4289 case _PC_SATTR_ENABLED: 4290 case _PC_SATTR_EXISTS: 4291 zp = VTOZ(vp); 4292 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) && 4293 (vp->v_type == VREG || vp->v_type == VDIR); 4294 return (0); 4295 4296 case _PC_ACL_ENABLED: 4297 *valp = _ACL_ACE_ENABLED; 4298 return (0); 4299 4300 case _PC_MIN_HOLE_SIZE: 4301 *valp = (ulong_t)SPA_MINBLOCKSIZE; 4302 return (0); 4303 4304 default: 4305 return (fs_pathconf(vp, cmd, valp, cr, ct)); 4306 } 4307 } 4308 4309 /*ARGSUSED*/ 4310 static int 4311 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4312 caller_context_t *ct) 4313 { 4314 znode_t *zp = VTOZ(vp); 4315 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4316 int error; 4317 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4318 4319 ZFS_ENTER(zfsvfs); 4320 ZFS_VERIFY_ZP(zp); 4321 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 4322 ZFS_EXIT(zfsvfs); 4323 4324 return (error); 4325 } 4326 4327 /*ARGSUSED*/ 4328 static int 4329 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4330 caller_context_t *ct) 4331 { 4332 znode_t *zp = VTOZ(vp); 4333 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4334 int error; 4335 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4336 4337 ZFS_ENTER(zfsvfs); 4338 ZFS_VERIFY_ZP(zp); 4339 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 4340 ZFS_EXIT(zfsvfs); 4341 return (error); 4342 } 4343 4344 /* 4345 * Predeclare these here so that the compiler assumes that 4346 * this is an "old style" function declaration that does 4347 * not include arguments => we won't get type mismatch errors 4348 * in the initializations that follow. 4349 */ 4350 static int zfs_inval(); 4351 static int zfs_isdir(); 4352 4353 static int 4354 zfs_inval() 4355 { 4356 return (EINVAL); 4357 } 4358 4359 static int 4360 zfs_isdir() 4361 { 4362 return (EISDIR); 4363 } 4364 /* 4365 * Directory vnode operations template 4366 */ 4367 vnodeops_t *zfs_dvnodeops; 4368 const fs_operation_def_t zfs_dvnodeops_template[] = { 4369 VOPNAME_OPEN, { .vop_open = zfs_open }, 4370 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4371 VOPNAME_READ, { .error = zfs_isdir }, 4372 VOPNAME_WRITE, { .error = zfs_isdir }, 4373 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4374 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4375 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4376 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4377 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4378 VOPNAME_CREATE, { .vop_create = zfs_create }, 4379 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 4380 VOPNAME_LINK, { .vop_link = zfs_link }, 4381 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4382 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 4383 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 4384 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 4385 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 4386 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4387 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4388 VOPNAME_FID, { .vop_fid = zfs_fid }, 4389 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4390 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4391 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4392 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4393 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4394 NULL, NULL 4395 }; 4396 4397 /* 4398 * Regular file vnode operations template 4399 */ 4400 vnodeops_t *zfs_fvnodeops; 4401 const fs_operation_def_t zfs_fvnodeops_template[] = { 4402 VOPNAME_OPEN, { .vop_open = zfs_open }, 4403 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4404 VOPNAME_READ, { .vop_read = zfs_read }, 4405 VOPNAME_WRITE, { .vop_write = zfs_write }, 4406 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4407 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4408 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4409 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4410 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4411 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4412 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4413 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4414 VOPNAME_FID, { .vop_fid = zfs_fid }, 4415 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4416 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 4417 VOPNAME_SPACE, { .vop_space = zfs_space }, 4418 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 4419 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 4420 VOPNAME_MAP, { .vop_map = zfs_map }, 4421 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 4422 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 4423 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4424 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4425 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4426 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4427 NULL, NULL 4428 }; 4429 4430 /* 4431 * Symbolic link vnode operations template 4432 */ 4433 vnodeops_t *zfs_symvnodeops; 4434 const fs_operation_def_t zfs_symvnodeops_template[] = { 4435 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4436 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4437 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4438 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4439 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 4440 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4441 VOPNAME_FID, { .vop_fid = zfs_fid }, 4442 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4443 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4444 NULL, NULL 4445 }; 4446 4447 /* 4448 * Extended attribute directory vnode operations template 4449 * This template is identical to the directory vnodes 4450 * operation template except for restricted operations: 4451 * VOP_MKDIR() 4452 * VOP_SYMLINK() 4453 * Note that there are other restrictions embedded in: 4454 * zfs_create() - restrict type to VREG 4455 * zfs_link() - no links into/out of attribute space 4456 * zfs_rename() - no moves into/out of attribute space 4457 */ 4458 vnodeops_t *zfs_xdvnodeops; 4459 const fs_operation_def_t zfs_xdvnodeops_template[] = { 4460 VOPNAME_OPEN, { .vop_open = zfs_open }, 4461 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4462 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4463 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4464 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4465 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4466 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4467 VOPNAME_CREATE, { .vop_create = zfs_create }, 4468 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 4469 VOPNAME_LINK, { .vop_link = zfs_link }, 4470 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4471 VOPNAME_MKDIR, { .error = zfs_inval }, 4472 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 4473 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 4474 VOPNAME_SYMLINK, { .error = zfs_inval }, 4475 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4476 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4477 VOPNAME_FID, { .vop_fid = zfs_fid }, 4478 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4479 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4480 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4481 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4482 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4483 NULL, NULL 4484 }; 4485 4486 /* 4487 * Error vnode operations template 4488 */ 4489 vnodeops_t *zfs_evnodeops; 4490 const fs_operation_def_t zfs_evnodeops_template[] = { 4491 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4492 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4493 NULL, NULL 4494 }; 4495