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 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* Portions Copyright 2007 Jeremy Teo */ 27 28 #include <sys/types.h> 29 #include <sys/param.h> 30 #include <sys/time.h> 31 #include <sys/systm.h> 32 #include <sys/sysmacros.h> 33 #include <sys/resource.h> 34 #include <sys/vfs.h> 35 #include <sys/vfs_opreg.h> 36 #include <sys/vnode.h> 37 #include <sys/file.h> 38 #include <sys/stat.h> 39 #include <sys/kmem.h> 40 #include <sys/taskq.h> 41 #include <sys/uio.h> 42 #include <sys/vmsystm.h> 43 #include <sys/atomic.h> 44 #include <sys/vm.h> 45 #include <vm/seg_vn.h> 46 #include <vm/pvn.h> 47 #include <vm/as.h> 48 #include <vm/kpm.h> 49 #include <vm/seg_kpm.h> 50 #include <sys/mman.h> 51 #include <sys/pathname.h> 52 #include <sys/cmn_err.h> 53 #include <sys/errno.h> 54 #include <sys/unistd.h> 55 #include <sys/zfs_dir.h> 56 #include <sys/zfs_acl.h> 57 #include <sys/zfs_ioctl.h> 58 #include <sys/fs/zfs.h> 59 #include <sys/dmu.h> 60 #include <sys/spa.h> 61 #include <sys/txg.h> 62 #include <sys/dbuf.h> 63 #include <sys/zap.h> 64 #include <sys/dirent.h> 65 #include <sys/policy.h> 66 #include <sys/sunddi.h> 67 #include <sys/filio.h> 68 #include "fs/fs_subr.h" 69 #include <sys/zfs_ctldir.h> 70 #include <sys/zfs_fuid.h> 71 #include <sys/dnlc.h> 72 #include <sys/zfs_rlock.h> 73 #include <sys/extdirent.h> 74 #include <sys/kidmap.h> 75 #include <sys/cred_impl.h> 76 #include <sys/attr.h> 77 78 /* 79 * Programming rules. 80 * 81 * Each vnode op performs some logical unit of work. To do this, the ZPL must 82 * properly lock its in-core state, create a DMU transaction, do the work, 83 * record this work in the intent log (ZIL), commit the DMU transaction, 84 * and wait for the intent log to commit if it is a synchronous operation. 85 * Moreover, the vnode ops must work in both normal and log replay context. 86 * The ordering of events is important to avoid deadlocks and references 87 * to freed memory. The example below illustrates the following Big Rules: 88 * 89 * (1) A check must be made in each zfs thread for a mounted file system. 90 * This is done avoiding races using ZFS_ENTER(zfsvfs). 91 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes 92 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros 93 * can return EIO from the calling function. 94 * 95 * (2) VN_RELE() should always be the last thing except for zil_commit() 96 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 97 * First, if it's the last reference, the vnode/znode 98 * can be freed, so the zp may point to freed memory. Second, the last 99 * reference will call zfs_zinactive(), which may induce a lot of work -- 100 * pushing cached pages (which acquires range locks) and syncing out 101 * cached atime changes. Third, zfs_zinactive() may require a new tx, 102 * which could deadlock the system if you were already holding one. 103 * 104 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 105 * as they can span dmu_tx_assign() calls. 106 * 107 * (4) Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign(). 108 * In normal operation, this will be TXG_NOWAIT. During ZIL replay, 109 * it will be a specific txg. Either way, dmu_tx_assign() never blocks. 110 * This is critical because we don't want to block while holding locks. 111 * Note, in particular, that if a lock is sometimes acquired before 112 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 113 * use a non-blocking assign can deadlock the system. The scenario: 114 * 115 * Thread A has grabbed a lock before calling dmu_tx_assign(). 116 * Thread B is in an already-assigned tx, and blocks for this lock. 117 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 118 * forever, because the previous txg can't quiesce until B's tx commits. 119 * 120 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 121 * then drop all locks, call dmu_tx_wait(), and try again. 122 * 123 * (5) If the operation succeeded, generate the intent log entry for it 124 * before dropping locks. This ensures that the ordering of events 125 * in the intent log matches the order in which they actually occurred. 126 * 127 * (6) At the end of each vnode op, the DMU tx must always commit, 128 * regardless of whether there were any errors. 129 * 130 * (7) After dropping all locks, invoke zil_commit(zilog, seq, foid) 131 * to ensure that synchronous semantics are provided when necessary. 132 * 133 * In general, this is how things should be ordered in each vnode op: 134 * 135 * ZFS_ENTER(zfsvfs); // exit if unmounted 136 * top: 137 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 138 * rw_enter(...); // grab any other locks you need 139 * tx = dmu_tx_create(...); // get DMU tx 140 * dmu_tx_hold_*(); // hold each object you might modify 141 * error = dmu_tx_assign(tx, zfsvfs->z_assign); // try to assign 142 * if (error) { 143 * rw_exit(...); // drop locks 144 * zfs_dirent_unlock(dl); // unlock directory entry 145 * VN_RELE(...); // release held vnodes 146 * if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 147 * dmu_tx_wait(tx); 148 * dmu_tx_abort(tx); 149 * goto top; 150 * } 151 * dmu_tx_abort(tx); // abort DMU tx 152 * ZFS_EXIT(zfsvfs); // finished in zfs 153 * return (error); // really out of space 154 * } 155 * error = do_real_work(); // do whatever this VOP does 156 * if (error == 0) 157 * zfs_log_*(...); // on success, make ZIL entry 158 * dmu_tx_commit(tx); // commit DMU tx -- error or not 159 * rw_exit(...); // drop locks 160 * zfs_dirent_unlock(dl); // unlock directory entry 161 * VN_RELE(...); // release held vnodes 162 * zil_commit(zilog, seq, foid); // synchronous when necessary 163 * ZFS_EXIT(zfsvfs); // finished in zfs 164 * return (error); // done, report error 165 */ 166 167 /* ARGSUSED */ 168 static int 169 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 170 { 171 znode_t *zp = VTOZ(*vpp); 172 173 if ((flag & FWRITE) && (zp->z_phys->zp_flags & ZFS_APPENDONLY) && 174 ((flag & FAPPEND) == 0)) { 175 return (EPERM); 176 } 177 178 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 179 ZTOV(zp)->v_type == VREG && 180 !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) && 181 zp->z_phys->zp_size > 0) 182 if (fs_vscan(*vpp, cr, 0) != 0) 183 return (EACCES); 184 185 /* Keep a count of the synchronous opens in the znode */ 186 if (flag & (FSYNC | FDSYNC)) 187 atomic_inc_32(&zp->z_sync_cnt); 188 189 return (0); 190 } 191 192 /* ARGSUSED */ 193 static int 194 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 195 caller_context_t *ct) 196 { 197 znode_t *zp = VTOZ(vp); 198 199 /* Decrement the synchronous opens in the znode */ 200 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 201 atomic_dec_32(&zp->z_sync_cnt); 202 203 /* 204 * Clean up any locks held by this process on the vp. 205 */ 206 cleanlocks(vp, ddi_get_pid(), 0); 207 cleanshares(vp, ddi_get_pid()); 208 209 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 210 ZTOV(zp)->v_type == VREG && 211 !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) && 212 zp->z_phys->zp_size > 0) 213 VERIFY(fs_vscan(vp, cr, 1) == 0); 214 215 return (0); 216 } 217 218 /* 219 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 220 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 221 */ 222 static int 223 zfs_holey(vnode_t *vp, int cmd, offset_t *off) 224 { 225 znode_t *zp = VTOZ(vp); 226 uint64_t noff = (uint64_t)*off; /* new offset */ 227 uint64_t file_sz; 228 int error; 229 boolean_t hole; 230 231 file_sz = zp->z_phys->zp_size; 232 if (noff >= file_sz) { 233 return (ENXIO); 234 } 235 236 if (cmd == _FIO_SEEK_HOLE) 237 hole = B_TRUE; 238 else 239 hole = B_FALSE; 240 241 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 242 243 /* end of file? */ 244 if ((error == ESRCH) || (noff > file_sz)) { 245 /* 246 * Handle the virtual hole at the end of file. 247 */ 248 if (hole) { 249 *off = file_sz; 250 return (0); 251 } 252 return (ENXIO); 253 } 254 255 if (noff < *off) 256 return (error); 257 *off = noff; 258 return (error); 259 } 260 261 /* ARGSUSED */ 262 static int 263 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred, 264 int *rvalp, caller_context_t *ct) 265 { 266 offset_t off; 267 int error; 268 zfsvfs_t *zfsvfs; 269 znode_t *zp; 270 271 switch (com) { 272 case _FIOFFS: 273 return (zfs_sync(vp->v_vfsp, 0, cred)); 274 275 /* 276 * The following two ioctls are used by bfu. Faking out, 277 * necessary to avoid bfu errors. 278 */ 279 case _FIOGDIO: 280 case _FIOSDIO: 281 return (0); 282 283 case _FIO_SEEK_DATA: 284 case _FIO_SEEK_HOLE: 285 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 286 return (EFAULT); 287 288 zp = VTOZ(vp); 289 zfsvfs = zp->z_zfsvfs; 290 ZFS_ENTER(zfsvfs); 291 ZFS_VERIFY_ZP(zp); 292 293 /* offset parameter is in/out */ 294 error = zfs_holey(vp, com, &off); 295 ZFS_EXIT(zfsvfs); 296 if (error) 297 return (error); 298 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 299 return (EFAULT); 300 return (0); 301 } 302 return (ENOTTY); 303 } 304 305 /* 306 * Utility functions to map and unmap a single physical page. These 307 * are used to manage the mappable copies of ZFS file data, and therefore 308 * do not update ref/mod bits. 309 */ 310 caddr_t 311 zfs_map_page(page_t *pp, enum seg_rw rw) 312 { 313 if (kpm_enable) 314 return (hat_kpm_mapin(pp, 0)); 315 ASSERT(rw == S_READ || rw == S_WRITE); 316 return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0), 317 (caddr_t)-1)); 318 } 319 320 void 321 zfs_unmap_page(page_t *pp, caddr_t addr) 322 { 323 if (kpm_enable) { 324 hat_kpm_mapout(pp, 0, addr); 325 } else { 326 ppmapout(addr); 327 } 328 } 329 330 /* 331 * When a file is memory mapped, we must keep the IO data synchronized 332 * between the DMU cache and the memory mapped pages. What this means: 333 * 334 * On Write: If we find a memory mapped page, we write to *both* 335 * the page and the dmu buffer. 336 * 337 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 338 * the file is memory mapped. 339 */ 340 static int 341 mappedwrite(vnode_t *vp, int nbytes, uio_t *uio, dmu_tx_t *tx) 342 { 343 znode_t *zp = VTOZ(vp); 344 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 345 int64_t start, off; 346 int len = nbytes; 347 int error = 0; 348 349 start = uio->uio_loffset; 350 off = start & PAGEOFFSET; 351 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 352 page_t *pp; 353 uint64_t bytes = MIN(PAGESIZE - off, len); 354 uint64_t woff = uio->uio_loffset; 355 356 /* 357 * We don't want a new page to "appear" in the middle of 358 * the file update (because it may not get the write 359 * update data), so we grab a lock to block 360 * zfs_getpage(). 361 */ 362 rw_enter(&zp->z_map_lock, RW_WRITER); 363 if (pp = page_lookup(vp, start, SE_SHARED)) { 364 caddr_t va; 365 366 rw_exit(&zp->z_map_lock); 367 va = zfs_map_page(pp, S_WRITE); 368 error = uiomove(va+off, bytes, UIO_WRITE, uio); 369 if (error == 0) { 370 dmu_write(zfsvfs->z_os, zp->z_id, 371 woff, bytes, va+off, tx); 372 } 373 zfs_unmap_page(pp, va); 374 page_unlock(pp); 375 } else { 376 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 377 uio, bytes, tx); 378 rw_exit(&zp->z_map_lock); 379 } 380 len -= bytes; 381 off = 0; 382 if (error) 383 break; 384 } 385 return (error); 386 } 387 388 /* 389 * When a file is memory mapped, we must keep the IO data synchronized 390 * between the DMU cache and the memory mapped pages. What this means: 391 * 392 * On Read: We "read" preferentially from memory mapped pages, 393 * else we default from the dmu buffer. 394 * 395 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 396 * the file is memory mapped. 397 */ 398 static int 399 mappedread(vnode_t *vp, int nbytes, uio_t *uio) 400 { 401 znode_t *zp = VTOZ(vp); 402 objset_t *os = zp->z_zfsvfs->z_os; 403 int64_t start, off; 404 int len = nbytes; 405 int error = 0; 406 407 start = uio->uio_loffset; 408 off = start & PAGEOFFSET; 409 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 410 page_t *pp; 411 uint64_t bytes = MIN(PAGESIZE - off, len); 412 413 if (pp = page_lookup(vp, start, SE_SHARED)) { 414 caddr_t va; 415 416 va = zfs_map_page(pp, S_READ); 417 error = uiomove(va + off, bytes, UIO_READ, uio); 418 zfs_unmap_page(pp, va); 419 page_unlock(pp); 420 } else { 421 error = dmu_read_uio(os, zp->z_id, uio, bytes); 422 } 423 len -= bytes; 424 off = 0; 425 if (error) 426 break; 427 } 428 return (error); 429 } 430 431 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 432 433 /* 434 * Read bytes from specified file into supplied buffer. 435 * 436 * IN: vp - vnode of file to be read from. 437 * uio - structure supplying read location, range info, 438 * and return buffer. 439 * ioflag - SYNC flags; used to provide FRSYNC semantics. 440 * cr - credentials of caller. 441 * ct - caller context 442 * 443 * OUT: uio - updated offset and range, buffer filled. 444 * 445 * RETURN: 0 if success 446 * error code if failure 447 * 448 * Side Effects: 449 * vp - atime updated if byte count > 0 450 */ 451 /* ARGSUSED */ 452 static int 453 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 454 { 455 znode_t *zp = VTOZ(vp); 456 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 457 objset_t *os; 458 ssize_t n, nbytes; 459 int error; 460 rl_t *rl; 461 462 ZFS_ENTER(zfsvfs); 463 ZFS_VERIFY_ZP(zp); 464 os = zfsvfs->z_os; 465 466 if (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) { 467 ZFS_EXIT(zfsvfs); 468 return (EACCES); 469 } 470 471 /* 472 * Validate file offset 473 */ 474 if (uio->uio_loffset < (offset_t)0) { 475 ZFS_EXIT(zfsvfs); 476 return (EINVAL); 477 } 478 479 /* 480 * Fasttrack empty reads 481 */ 482 if (uio->uio_resid == 0) { 483 ZFS_EXIT(zfsvfs); 484 return (0); 485 } 486 487 /* 488 * Check for mandatory locks 489 */ 490 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 491 if (error = chklock(vp, FREAD, 492 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 493 ZFS_EXIT(zfsvfs); 494 return (error); 495 } 496 } 497 498 /* 499 * If we're in FRSYNC mode, sync out this znode before reading it. 500 */ 501 if (ioflag & FRSYNC) 502 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 503 504 /* 505 * Lock the range against changes. 506 */ 507 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 508 509 /* 510 * If we are reading past end-of-file we can skip 511 * to the end; but we might still need to set atime. 512 */ 513 if (uio->uio_loffset >= zp->z_phys->zp_size) { 514 error = 0; 515 goto out; 516 } 517 518 ASSERT(uio->uio_loffset < zp->z_phys->zp_size); 519 n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset); 520 521 while (n > 0) { 522 nbytes = MIN(n, zfs_read_chunk_size - 523 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 524 525 if (vn_has_cached_data(vp)) 526 error = mappedread(vp, nbytes, uio); 527 else 528 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 529 if (error) { 530 /* convert checksum errors into IO errors */ 531 if (error == ECKSUM) 532 error = EIO; 533 break; 534 } 535 536 n -= nbytes; 537 } 538 539 out: 540 zfs_range_unlock(rl); 541 542 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 543 ZFS_EXIT(zfsvfs); 544 return (error); 545 } 546 547 /* 548 * Fault in the pages of the first n bytes specified by the uio structure. 549 * 1 byte in each page is touched and the uio struct is unmodified. 550 * Any error will exit this routine as this is only a best 551 * attempt to get the pages resident. This is a copy of ufs_trans_touch(). 552 */ 553 static void 554 zfs_prefault_write(ssize_t n, struct uio *uio) 555 { 556 struct iovec *iov; 557 ulong_t cnt, incr; 558 caddr_t p; 559 uint8_t tmp; 560 561 iov = uio->uio_iov; 562 563 while (n) { 564 cnt = MIN(iov->iov_len, n); 565 if (cnt == 0) { 566 /* empty iov entry */ 567 iov++; 568 continue; 569 } 570 n -= cnt; 571 /* 572 * touch each page in this segment. 573 */ 574 p = iov->iov_base; 575 while (cnt) { 576 switch (uio->uio_segflg) { 577 case UIO_USERSPACE: 578 case UIO_USERISPACE: 579 if (fuword8(p, &tmp)) 580 return; 581 break; 582 case UIO_SYSSPACE: 583 if (kcopy(p, &tmp, 1)) 584 return; 585 break; 586 } 587 incr = MIN(cnt, PAGESIZE); 588 p += incr; 589 cnt -= incr; 590 } 591 /* 592 * touch the last byte in case it straddles a page. 593 */ 594 p--; 595 switch (uio->uio_segflg) { 596 case UIO_USERSPACE: 597 case UIO_USERISPACE: 598 if (fuword8(p, &tmp)) 599 return; 600 break; 601 case UIO_SYSSPACE: 602 if (kcopy(p, &tmp, 1)) 603 return; 604 break; 605 } 606 iov++; 607 } 608 } 609 610 /* 611 * Write the bytes to a file. 612 * 613 * IN: vp - vnode of file to be written to. 614 * uio - structure supplying write location, range info, 615 * and data buffer. 616 * ioflag - FAPPEND flag set if in append mode. 617 * cr - credentials of caller. 618 * ct - caller context (NFS/CIFS fem monitor only) 619 * 620 * OUT: uio - updated offset and range. 621 * 622 * RETURN: 0 if success 623 * error code if failure 624 * 625 * Timestamps: 626 * vp - ctime|mtime updated if byte count > 0 627 */ 628 /* ARGSUSED */ 629 static int 630 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 631 { 632 znode_t *zp = VTOZ(vp); 633 rlim64_t limit = uio->uio_llimit; 634 ssize_t start_resid = uio->uio_resid; 635 ssize_t tx_bytes; 636 uint64_t end_size; 637 dmu_tx_t *tx; 638 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 639 zilog_t *zilog; 640 offset_t woff; 641 ssize_t n, nbytes; 642 rl_t *rl; 643 int max_blksz = zfsvfs->z_max_blksz; 644 uint64_t pflags; 645 int error; 646 647 /* 648 * Fasttrack empty write 649 */ 650 n = start_resid; 651 if (n == 0) 652 return (0); 653 654 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 655 limit = MAXOFFSET_T; 656 657 ZFS_ENTER(zfsvfs); 658 ZFS_VERIFY_ZP(zp); 659 660 /* 661 * If immutable or not appending then return EPERM 662 */ 663 pflags = zp->z_phys->zp_flags; 664 if ((pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 665 ((pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 666 (uio->uio_loffset < zp->z_phys->zp_size))) { 667 ZFS_EXIT(zfsvfs); 668 return (EPERM); 669 } 670 671 zilog = zfsvfs->z_log; 672 673 /* 674 * Pre-fault the pages to ensure slow (eg NFS) pages 675 * don't hold up txg. 676 */ 677 zfs_prefault_write(n, uio); 678 679 /* 680 * If in append mode, set the io offset pointer to eof. 681 */ 682 if (ioflag & FAPPEND) { 683 /* 684 * Range lock for a file append: 685 * The value for the start of range will be determined by 686 * zfs_range_lock() (to guarantee append semantics). 687 * If this write will cause the block size to increase, 688 * zfs_range_lock() will lock the entire file, so we must 689 * later reduce the range after we grow the block size. 690 */ 691 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 692 if (rl->r_len == UINT64_MAX) { 693 /* overlocked, zp_size can't change */ 694 woff = uio->uio_loffset = zp->z_phys->zp_size; 695 } else { 696 woff = uio->uio_loffset = rl->r_off; 697 } 698 } else { 699 woff = uio->uio_loffset; 700 /* 701 * Validate file offset 702 */ 703 if (woff < 0) { 704 ZFS_EXIT(zfsvfs); 705 return (EINVAL); 706 } 707 708 /* 709 * If we need to grow the block size then zfs_range_lock() 710 * will lock a wider range than we request here. 711 * Later after growing the block size we reduce the range. 712 */ 713 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 714 } 715 716 if (woff >= limit) { 717 zfs_range_unlock(rl); 718 ZFS_EXIT(zfsvfs); 719 return (EFBIG); 720 } 721 722 if ((woff + n) > limit || woff > (limit - n)) 723 n = limit - woff; 724 725 /* 726 * Check for mandatory locks 727 */ 728 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 729 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 730 zfs_range_unlock(rl); 731 ZFS_EXIT(zfsvfs); 732 return (error); 733 } 734 end_size = MAX(zp->z_phys->zp_size, woff + n); 735 736 /* 737 * Write the file in reasonable size chunks. Each chunk is written 738 * in a separate transaction; this keeps the intent log records small 739 * and allows us to do more fine-grained space accounting. 740 */ 741 while (n > 0) { 742 /* 743 * Start a transaction. 744 */ 745 woff = uio->uio_loffset; 746 tx = dmu_tx_create(zfsvfs->z_os); 747 dmu_tx_hold_bonus(tx, zp->z_id); 748 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 749 error = dmu_tx_assign(tx, zfsvfs->z_assign); 750 if (error) { 751 if (error == ERESTART && 752 zfsvfs->z_assign == TXG_NOWAIT) { 753 dmu_tx_wait(tx); 754 dmu_tx_abort(tx); 755 continue; 756 } 757 dmu_tx_abort(tx); 758 break; 759 } 760 761 /* 762 * If zfs_range_lock() over-locked we grow the blocksize 763 * and then reduce the lock range. This will only happen 764 * on the first iteration since zfs_range_reduce() will 765 * shrink down r_len to the appropriate size. 766 */ 767 if (rl->r_len == UINT64_MAX) { 768 uint64_t new_blksz; 769 770 if (zp->z_blksz > max_blksz) { 771 ASSERT(!ISP2(zp->z_blksz)); 772 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 773 } else { 774 new_blksz = MIN(end_size, max_blksz); 775 } 776 zfs_grow_blocksize(zp, new_blksz, tx); 777 zfs_range_reduce(rl, woff, n); 778 } 779 780 /* 781 * XXX - should we really limit each write to z_max_blksz? 782 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 783 */ 784 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 785 rw_enter(&zp->z_map_lock, RW_READER); 786 787 tx_bytes = uio->uio_resid; 788 if (vn_has_cached_data(vp)) { 789 rw_exit(&zp->z_map_lock); 790 error = mappedwrite(vp, nbytes, uio, tx); 791 } else { 792 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 793 uio, nbytes, tx); 794 rw_exit(&zp->z_map_lock); 795 } 796 tx_bytes -= uio->uio_resid; 797 798 /* 799 * If we made no progress, we're done. If we made even 800 * partial progress, update the znode and ZIL accordingly. 801 */ 802 if (tx_bytes == 0) { 803 dmu_tx_commit(tx); 804 ASSERT(error != 0); 805 break; 806 } 807 808 /* 809 * Clear Set-UID/Set-GID bits on successful write if not 810 * privileged and at least one of the excute bits is set. 811 * 812 * It would be nice to to this after all writes have 813 * been done, but that would still expose the ISUID/ISGID 814 * to another app after the partial write is committed. 815 * 816 * Note: we don't call zfs_fuid_map_id() here because 817 * user 0 is not an ephemeral uid. 818 */ 819 mutex_enter(&zp->z_acl_lock); 820 if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) | 821 (S_IXUSR >> 6))) != 0 && 822 (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 && 823 secpolicy_vnode_setid_retain(cr, 824 (zp->z_phys->zp_mode & S_ISUID) != 0 && 825 zp->z_phys->zp_uid == 0) != 0) { 826 zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID); 827 } 828 mutex_exit(&zp->z_acl_lock); 829 830 /* 831 * Update time stamp. NOTE: This marks the bonus buffer as 832 * dirty, so we don't have to do it again for zp_size. 833 */ 834 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 835 836 /* 837 * Update the file size (zp_size) if it has changed; 838 * account for possible concurrent updates. 839 */ 840 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) 841 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 842 uio->uio_loffset); 843 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 844 dmu_tx_commit(tx); 845 846 if (error != 0) 847 break; 848 ASSERT(tx_bytes == nbytes); 849 n -= nbytes; 850 } 851 852 zfs_range_unlock(rl); 853 854 /* 855 * If we're in replay mode, or we made no progress, return error. 856 * Otherwise, it's at least a partial write, so it's successful. 857 */ 858 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 859 ZFS_EXIT(zfsvfs); 860 return (error); 861 } 862 863 if (ioflag & (FSYNC | FDSYNC)) 864 zil_commit(zilog, zp->z_last_itx, zp->z_id); 865 866 ZFS_EXIT(zfsvfs); 867 return (0); 868 } 869 870 void 871 zfs_get_done(dmu_buf_t *db, void *vzgd) 872 { 873 zgd_t *zgd = (zgd_t *)vzgd; 874 rl_t *rl = zgd->zgd_rl; 875 vnode_t *vp = ZTOV(rl->r_zp); 876 877 dmu_buf_rele(db, vzgd); 878 zfs_range_unlock(rl); 879 VN_RELE(vp); 880 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 881 kmem_free(zgd, sizeof (zgd_t)); 882 } 883 884 /* 885 * Get data to generate a TX_WRITE intent log record. 886 */ 887 int 888 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 889 { 890 zfsvfs_t *zfsvfs = arg; 891 objset_t *os = zfsvfs->z_os; 892 znode_t *zp; 893 uint64_t off = lr->lr_offset; 894 dmu_buf_t *db; 895 rl_t *rl; 896 zgd_t *zgd; 897 int dlen = lr->lr_length; /* length of user data */ 898 int error = 0; 899 900 ASSERT(zio); 901 ASSERT(dlen != 0); 902 903 /* 904 * Nothing to do if the file has been removed 905 */ 906 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 907 return (ENOENT); 908 if (zp->z_unlinked) { 909 VN_RELE(ZTOV(zp)); 910 return (ENOENT); 911 } 912 913 /* 914 * Write records come in two flavors: immediate and indirect. 915 * For small writes it's cheaper to store the data with the 916 * log record (immediate); for large writes it's cheaper to 917 * sync the data and get a pointer to it (indirect) so that 918 * we don't have to write the data twice. 919 */ 920 if (buf != NULL) { /* immediate write */ 921 rl = zfs_range_lock(zp, off, dlen, RL_READER); 922 /* test for truncation needs to be done while range locked */ 923 if (off >= zp->z_phys->zp_size) { 924 error = ENOENT; 925 goto out; 926 } 927 VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf)); 928 } else { /* indirect write */ 929 uint64_t boff; /* block starting offset */ 930 931 /* 932 * Have to lock the whole block to ensure when it's 933 * written out and it's checksum is being calculated 934 * that no one can change the data. We need to re-check 935 * blocksize after we get the lock in case it's changed! 936 */ 937 for (;;) { 938 if (ISP2(zp->z_blksz)) { 939 boff = P2ALIGN_TYPED(off, zp->z_blksz, 940 uint64_t); 941 } else { 942 boff = 0; 943 } 944 dlen = zp->z_blksz; 945 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 946 if (zp->z_blksz == dlen) 947 break; 948 zfs_range_unlock(rl); 949 } 950 /* test for truncation needs to be done while range locked */ 951 if (off >= zp->z_phys->zp_size) { 952 error = ENOENT; 953 goto out; 954 } 955 zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP); 956 zgd->zgd_rl = rl; 957 zgd->zgd_zilog = zfsvfs->z_log; 958 zgd->zgd_bp = &lr->lr_blkptr; 959 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db)); 960 ASSERT(boff == db->db_offset); 961 lr->lr_blkoff = off - boff; 962 error = dmu_sync(zio, db, &lr->lr_blkptr, 963 lr->lr_common.lrc_txg, zfs_get_done, zgd); 964 ASSERT((error && error != EINPROGRESS) || 965 lr->lr_length <= zp->z_blksz); 966 if (error == 0) 967 zil_add_block(zfsvfs->z_log, &lr->lr_blkptr); 968 /* 969 * If we get EINPROGRESS, then we need to wait for a 970 * write IO initiated by dmu_sync() to complete before 971 * we can release this dbuf. We will finish everything 972 * up in the zfs_get_done() callback. 973 */ 974 if (error == EINPROGRESS) 975 return (0); 976 dmu_buf_rele(db, zgd); 977 kmem_free(zgd, sizeof (zgd_t)); 978 } 979 out: 980 zfs_range_unlock(rl); 981 VN_RELE(ZTOV(zp)); 982 return (error); 983 } 984 985 /*ARGSUSED*/ 986 static int 987 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 988 caller_context_t *ct) 989 { 990 znode_t *zp = VTOZ(vp); 991 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 992 int error; 993 994 ZFS_ENTER(zfsvfs); 995 ZFS_VERIFY_ZP(zp); 996 997 if (flag & V_ACE_MASK) 998 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 999 else 1000 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1001 1002 ZFS_EXIT(zfsvfs); 1003 return (error); 1004 } 1005 1006 /* 1007 * Lookup an entry in a directory, or an extended attribute directory. 1008 * If it exists, return a held vnode reference for it. 1009 * 1010 * IN: dvp - vnode of directory to search. 1011 * nm - name of entry to lookup. 1012 * pnp - full pathname to lookup [UNUSED]. 1013 * flags - LOOKUP_XATTR set if looking for an attribute. 1014 * rdir - root directory vnode [UNUSED]. 1015 * cr - credentials of caller. 1016 * ct - caller context 1017 * direntflags - directory lookup flags 1018 * realpnp - returned pathname. 1019 * 1020 * OUT: vpp - vnode of located entry, NULL if not found. 1021 * 1022 * RETURN: 0 if success 1023 * error code if failure 1024 * 1025 * Timestamps: 1026 * NA 1027 */ 1028 /* ARGSUSED */ 1029 static int 1030 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 1031 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 1032 int *direntflags, pathname_t *realpnp) 1033 { 1034 znode_t *zdp = VTOZ(dvp); 1035 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1036 int error; 1037 1038 ZFS_ENTER(zfsvfs); 1039 ZFS_VERIFY_ZP(zdp); 1040 1041 *vpp = NULL; 1042 1043 if (flags & LOOKUP_XATTR) { 1044 /* 1045 * If the xattr property is off, refuse the lookup request. 1046 */ 1047 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1048 ZFS_EXIT(zfsvfs); 1049 return (EINVAL); 1050 } 1051 1052 /* 1053 * We don't allow recursive attributes.. 1054 * Maybe someday we will. 1055 */ 1056 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 1057 ZFS_EXIT(zfsvfs); 1058 return (EINVAL); 1059 } 1060 1061 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1062 ZFS_EXIT(zfsvfs); 1063 return (error); 1064 } 1065 1066 /* 1067 * Do we have permission to get into attribute directory? 1068 */ 1069 1070 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1071 B_FALSE, cr)) { 1072 VN_RELE(*vpp); 1073 *vpp = NULL; 1074 } 1075 1076 ZFS_EXIT(zfsvfs); 1077 return (error); 1078 } 1079 1080 if (dvp->v_type != VDIR) { 1081 ZFS_EXIT(zfsvfs); 1082 return (ENOTDIR); 1083 } 1084 1085 /* 1086 * Check accessibility of directory. 1087 */ 1088 1089 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1090 ZFS_EXIT(zfsvfs); 1091 return (error); 1092 } 1093 1094 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1095 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1096 ZFS_EXIT(zfsvfs); 1097 return (EILSEQ); 1098 } 1099 1100 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1101 if (error == 0) { 1102 /* 1103 * Convert device special files 1104 */ 1105 if (IS_DEVVP(*vpp)) { 1106 vnode_t *svp; 1107 1108 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1109 VN_RELE(*vpp); 1110 if (svp == NULL) 1111 error = ENOSYS; 1112 else 1113 *vpp = svp; 1114 } 1115 } 1116 1117 ZFS_EXIT(zfsvfs); 1118 return (error); 1119 } 1120 1121 /* 1122 * Attempt to create a new entry in a directory. If the entry 1123 * already exists, truncate the file if permissible, else return 1124 * an error. Return the vp of the created or trunc'd file. 1125 * 1126 * IN: dvp - vnode of directory to put new file entry in. 1127 * name - name of new file entry. 1128 * vap - attributes of new file. 1129 * excl - flag indicating exclusive or non-exclusive mode. 1130 * mode - mode to open file with. 1131 * cr - credentials of caller. 1132 * flag - large file flag [UNUSED]. 1133 * ct - caller context 1134 * vsecp - ACL to be set 1135 * 1136 * OUT: vpp - vnode of created or trunc'd entry. 1137 * 1138 * RETURN: 0 if success 1139 * error code if failure 1140 * 1141 * Timestamps: 1142 * dvp - ctime|mtime updated if new entry created 1143 * vp - ctime|mtime always, atime if new 1144 */ 1145 1146 /* ARGSUSED */ 1147 static int 1148 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1149 int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct, 1150 vsecattr_t *vsecp) 1151 { 1152 znode_t *zp, *dzp = VTOZ(dvp); 1153 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1154 zilog_t *zilog; 1155 objset_t *os; 1156 zfs_dirlock_t *dl; 1157 dmu_tx_t *tx; 1158 int error; 1159 zfs_acl_t *aclp = NULL; 1160 zfs_fuid_info_t *fuidp = NULL; 1161 1162 /* 1163 * If we have an ephemeral id, ACL, or XVATTR then 1164 * make sure file system is at proper version 1165 */ 1166 1167 if (zfsvfs->z_use_fuids == B_FALSE && 1168 (vsecp || (vap->va_mask & AT_XVATTR) || 1169 IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr)))) 1170 return (EINVAL); 1171 1172 ZFS_ENTER(zfsvfs); 1173 ZFS_VERIFY_ZP(dzp); 1174 os = zfsvfs->z_os; 1175 zilog = zfsvfs->z_log; 1176 1177 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1178 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1179 ZFS_EXIT(zfsvfs); 1180 return (EILSEQ); 1181 } 1182 1183 if (vap->va_mask & AT_XVATTR) { 1184 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1185 crgetuid(cr), cr, vap->va_type)) != 0) { 1186 ZFS_EXIT(zfsvfs); 1187 return (error); 1188 } 1189 } 1190 top: 1191 *vpp = NULL; 1192 1193 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1194 vap->va_mode &= ~VSVTX; 1195 1196 if (*name == '\0') { 1197 /* 1198 * Null component name refers to the directory itself. 1199 */ 1200 VN_HOLD(dvp); 1201 zp = dzp; 1202 dl = NULL; 1203 error = 0; 1204 } else { 1205 /* possible VN_HOLD(zp) */ 1206 int zflg = 0; 1207 1208 if (flag & FIGNORECASE) 1209 zflg |= ZCILOOK; 1210 1211 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1212 NULL, NULL); 1213 if (error) { 1214 if (strcmp(name, "..") == 0) 1215 error = EISDIR; 1216 ZFS_EXIT(zfsvfs); 1217 if (aclp) 1218 zfs_acl_free(aclp); 1219 return (error); 1220 } 1221 } 1222 if (vsecp && aclp == NULL) { 1223 error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp); 1224 if (error) { 1225 ZFS_EXIT(zfsvfs); 1226 if (dl) 1227 zfs_dirent_unlock(dl); 1228 return (error); 1229 } 1230 } 1231 1232 if (zp == NULL) { 1233 uint64_t txtype; 1234 1235 /* 1236 * Create a new file object and update the directory 1237 * to reference it. 1238 */ 1239 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1240 goto out; 1241 } 1242 1243 /* 1244 * We only support the creation of regular files in 1245 * extended attribute directories. 1246 */ 1247 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1248 (vap->va_type != VREG)) { 1249 error = EINVAL; 1250 goto out; 1251 } 1252 1253 tx = dmu_tx_create(os); 1254 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1255 if ((aclp && aclp->z_has_fuids) || IS_EPHEMERAL(crgetuid(cr)) || 1256 IS_EPHEMERAL(crgetgid(cr))) { 1257 if (zfsvfs->z_fuid_obj == 0) { 1258 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1259 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1260 FUID_SIZE_ESTIMATE(zfsvfs)); 1261 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, 1262 FALSE, NULL); 1263 } else { 1264 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 1265 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 1266 FUID_SIZE_ESTIMATE(zfsvfs)); 1267 } 1268 } 1269 dmu_tx_hold_bonus(tx, dzp->z_id); 1270 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1271 if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp) { 1272 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1273 0, SPA_MAXBLOCKSIZE); 1274 } 1275 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1276 if (error) { 1277 zfs_dirent_unlock(dl); 1278 if (error == ERESTART && 1279 zfsvfs->z_assign == TXG_NOWAIT) { 1280 dmu_tx_wait(tx); 1281 dmu_tx_abort(tx); 1282 goto top; 1283 } 1284 dmu_tx_abort(tx); 1285 ZFS_EXIT(zfsvfs); 1286 if (aclp) 1287 zfs_acl_free(aclp); 1288 return (error); 1289 } 1290 zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp); 1291 (void) zfs_link_create(dl, zp, tx, ZNEW); 1292 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1293 if (flag & FIGNORECASE) 1294 txtype |= TX_CI; 1295 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1296 vsecp, fuidp, vap); 1297 if (fuidp) 1298 zfs_fuid_info_free(fuidp); 1299 dmu_tx_commit(tx); 1300 } else { 1301 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1302 1303 /* 1304 * A directory entry already exists for this name. 1305 */ 1306 /* 1307 * Can't truncate an existing file if in exclusive mode. 1308 */ 1309 if (excl == EXCL) { 1310 error = EEXIST; 1311 goto out; 1312 } 1313 /* 1314 * Can't open a directory for writing. 1315 */ 1316 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1317 error = EISDIR; 1318 goto out; 1319 } 1320 /* 1321 * Verify requested access to file. 1322 */ 1323 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1324 goto out; 1325 } 1326 1327 mutex_enter(&dzp->z_lock); 1328 dzp->z_seq++; 1329 mutex_exit(&dzp->z_lock); 1330 1331 /* 1332 * Truncate regular files if requested. 1333 */ 1334 if ((ZTOV(zp)->v_type == VREG) && 1335 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1336 /* we can't hold any locks when calling zfs_freesp() */ 1337 zfs_dirent_unlock(dl); 1338 dl = NULL; 1339 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1340 if (error == 0) { 1341 vnevent_create(ZTOV(zp), ct); 1342 } 1343 } 1344 } 1345 out: 1346 1347 if (dl) 1348 zfs_dirent_unlock(dl); 1349 1350 if (error) { 1351 if (zp) 1352 VN_RELE(ZTOV(zp)); 1353 } else { 1354 *vpp = ZTOV(zp); 1355 /* 1356 * If vnode is for a device return a specfs vnode instead. 1357 */ 1358 if (IS_DEVVP(*vpp)) { 1359 struct vnode *svp; 1360 1361 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1362 VN_RELE(*vpp); 1363 if (svp == NULL) { 1364 error = ENOSYS; 1365 } 1366 *vpp = svp; 1367 } 1368 } 1369 if (aclp) 1370 zfs_acl_free(aclp); 1371 1372 ZFS_EXIT(zfsvfs); 1373 return (error); 1374 } 1375 1376 /* 1377 * Remove an entry from a directory. 1378 * 1379 * IN: dvp - vnode of directory to remove entry from. 1380 * name - name of entry to remove. 1381 * cr - credentials of caller. 1382 * ct - caller context 1383 * flags - case flags 1384 * 1385 * RETURN: 0 if success 1386 * error code if failure 1387 * 1388 * Timestamps: 1389 * dvp - ctime|mtime 1390 * vp - ctime (if nlink > 0) 1391 */ 1392 /*ARGSUSED*/ 1393 static int 1394 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1395 int flags) 1396 { 1397 znode_t *zp, *dzp = VTOZ(dvp); 1398 znode_t *xzp = NULL; 1399 vnode_t *vp; 1400 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1401 zilog_t *zilog; 1402 uint64_t acl_obj, xattr_obj; 1403 zfs_dirlock_t *dl; 1404 dmu_tx_t *tx; 1405 boolean_t may_delete_now, delete_now = FALSE; 1406 boolean_t unlinked, toobig = FALSE; 1407 uint64_t txtype; 1408 pathname_t *realnmp = NULL; 1409 pathname_t realnm; 1410 int error; 1411 int zflg = ZEXISTS; 1412 1413 ZFS_ENTER(zfsvfs); 1414 ZFS_VERIFY_ZP(dzp); 1415 zilog = zfsvfs->z_log; 1416 1417 if (flags & FIGNORECASE) { 1418 zflg |= ZCILOOK; 1419 pn_alloc(&realnm); 1420 realnmp = &realnm; 1421 } 1422 1423 top: 1424 /* 1425 * Attempt to lock directory; fail if entry doesn't exist. 1426 */ 1427 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1428 NULL, realnmp)) { 1429 if (realnmp) 1430 pn_free(realnmp); 1431 ZFS_EXIT(zfsvfs); 1432 return (error); 1433 } 1434 1435 vp = ZTOV(zp); 1436 1437 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1438 goto out; 1439 } 1440 1441 /* 1442 * Need to use rmdir for removing directories. 1443 */ 1444 if (vp->v_type == VDIR) { 1445 error = EPERM; 1446 goto out; 1447 } 1448 1449 vnevent_remove(vp, dvp, name, ct); 1450 1451 if (realnmp) 1452 dnlc_remove(dvp, realnmp->pn_buf); 1453 else 1454 dnlc_remove(dvp, name); 1455 1456 mutex_enter(&vp->v_lock); 1457 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1458 mutex_exit(&vp->v_lock); 1459 1460 /* 1461 * We may delete the znode now, or we may put it in the unlinked set; 1462 * it depends on whether we're the last link, and on whether there are 1463 * other holds on the vnode. So we dmu_tx_hold() the right things to 1464 * allow for either case. 1465 */ 1466 tx = dmu_tx_create(zfsvfs->z_os); 1467 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1468 dmu_tx_hold_bonus(tx, zp->z_id); 1469 if (may_delete_now) { 1470 toobig = 1471 zp->z_phys->zp_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 1472 /* if the file is too big, only hold_free a token amount */ 1473 dmu_tx_hold_free(tx, zp->z_id, 0, 1474 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1475 } 1476 1477 /* are there any extended attributes? */ 1478 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1479 /* XXX - do we need this if we are deleting? */ 1480 dmu_tx_hold_bonus(tx, xattr_obj); 1481 } 1482 1483 /* are there any additional acls */ 1484 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1485 may_delete_now) 1486 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1487 1488 /* charge as an update -- would be nice not to charge at all */ 1489 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1490 1491 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1492 if (error) { 1493 zfs_dirent_unlock(dl); 1494 VN_RELE(vp); 1495 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1496 dmu_tx_wait(tx); 1497 dmu_tx_abort(tx); 1498 goto top; 1499 } 1500 if (realnmp) 1501 pn_free(realnmp); 1502 dmu_tx_abort(tx); 1503 ZFS_EXIT(zfsvfs); 1504 return (error); 1505 } 1506 1507 /* 1508 * Remove the directory entry. 1509 */ 1510 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1511 1512 if (error) { 1513 dmu_tx_commit(tx); 1514 goto out; 1515 } 1516 1517 if (unlinked) { 1518 mutex_enter(&vp->v_lock); 1519 delete_now = may_delete_now && !toobig && 1520 vp->v_count == 1 && !vn_has_cached_data(vp) && 1521 zp->z_phys->zp_xattr == xattr_obj && 1522 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1523 mutex_exit(&vp->v_lock); 1524 } 1525 1526 if (delete_now) { 1527 if (zp->z_phys->zp_xattr) { 1528 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1529 ASSERT3U(error, ==, 0); 1530 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1531 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1532 mutex_enter(&xzp->z_lock); 1533 xzp->z_unlinked = 1; 1534 xzp->z_phys->zp_links = 0; 1535 mutex_exit(&xzp->z_lock); 1536 zfs_unlinked_add(xzp, tx); 1537 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1538 } 1539 mutex_enter(&zp->z_lock); 1540 mutex_enter(&vp->v_lock); 1541 vp->v_count--; 1542 ASSERT3U(vp->v_count, ==, 0); 1543 mutex_exit(&vp->v_lock); 1544 mutex_exit(&zp->z_lock); 1545 zfs_znode_delete(zp, tx); 1546 } else if (unlinked) { 1547 zfs_unlinked_add(zp, tx); 1548 } 1549 1550 txtype = TX_REMOVE; 1551 if (flags & FIGNORECASE) 1552 txtype |= TX_CI; 1553 zfs_log_remove(zilog, tx, txtype, dzp, name); 1554 1555 dmu_tx_commit(tx); 1556 out: 1557 if (realnmp) 1558 pn_free(realnmp); 1559 1560 zfs_dirent_unlock(dl); 1561 1562 if (!delete_now) { 1563 VN_RELE(vp); 1564 } else if (xzp) { 1565 /* this rele is delayed to prevent nesting transactions */ 1566 VN_RELE(ZTOV(xzp)); 1567 } 1568 1569 ZFS_EXIT(zfsvfs); 1570 return (error); 1571 } 1572 1573 /* 1574 * Create a new directory and insert it into dvp using the name 1575 * provided. Return a pointer to the inserted directory. 1576 * 1577 * IN: dvp - vnode of directory to add subdir to. 1578 * dirname - name of new directory. 1579 * vap - attributes of new directory. 1580 * cr - credentials of caller. 1581 * ct - caller context 1582 * vsecp - ACL to be set 1583 * 1584 * OUT: vpp - vnode of created directory. 1585 * 1586 * RETURN: 0 if success 1587 * error code if failure 1588 * 1589 * Timestamps: 1590 * dvp - ctime|mtime updated 1591 * vp - ctime|mtime|atime updated 1592 */ 1593 /*ARGSUSED*/ 1594 static int 1595 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 1596 caller_context_t *ct, int flags, vsecattr_t *vsecp) 1597 { 1598 znode_t *zp, *dzp = VTOZ(dvp); 1599 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1600 zilog_t *zilog; 1601 zfs_dirlock_t *dl; 1602 uint64_t txtype; 1603 dmu_tx_t *tx; 1604 int error; 1605 zfs_acl_t *aclp = NULL; 1606 zfs_fuid_info_t *fuidp = NULL; 1607 int zf = ZNEW; 1608 1609 ASSERT(vap->va_type == VDIR); 1610 1611 /* 1612 * If we have an ephemeral id, ACL, or XVATTR then 1613 * make sure file system is at proper version 1614 */ 1615 1616 if (zfsvfs->z_use_fuids == B_FALSE && 1617 (vsecp || (vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(crgetuid(cr))|| 1618 IS_EPHEMERAL(crgetgid(cr)))) 1619 return (EINVAL); 1620 1621 ZFS_ENTER(zfsvfs); 1622 ZFS_VERIFY_ZP(dzp); 1623 zilog = zfsvfs->z_log; 1624 1625 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1626 ZFS_EXIT(zfsvfs); 1627 return (EINVAL); 1628 } 1629 1630 if (zfsvfs->z_utf8 && u8_validate(dirname, 1631 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1632 ZFS_EXIT(zfsvfs); 1633 return (EILSEQ); 1634 } 1635 if (flags & FIGNORECASE) 1636 zf |= ZCILOOK; 1637 1638 if (vap->va_mask & AT_XVATTR) 1639 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1640 crgetuid(cr), cr, vap->va_type)) != 0) { 1641 ZFS_EXIT(zfsvfs); 1642 return (error); 1643 } 1644 1645 /* 1646 * First make sure the new directory doesn't exist. 1647 */ 1648 top: 1649 *vpp = NULL; 1650 1651 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 1652 NULL, NULL)) { 1653 ZFS_EXIT(zfsvfs); 1654 return (error); 1655 } 1656 1657 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 1658 zfs_dirent_unlock(dl); 1659 ZFS_EXIT(zfsvfs); 1660 return (error); 1661 } 1662 1663 if (vsecp && aclp == NULL) { 1664 error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp); 1665 if (error) { 1666 zfs_dirent_unlock(dl); 1667 ZFS_EXIT(zfsvfs); 1668 return (error); 1669 } 1670 } 1671 /* 1672 * Add a new entry to the directory. 1673 */ 1674 tx = dmu_tx_create(zfsvfs->z_os); 1675 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1676 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1677 if ((aclp && aclp->z_has_fuids) || IS_EPHEMERAL(crgetuid(cr)) || 1678 IS_EPHEMERAL(crgetgid(cr))) { 1679 if (zfsvfs->z_fuid_obj == 0) { 1680 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1681 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1682 FUID_SIZE_ESTIMATE(zfsvfs)); 1683 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 1684 } else { 1685 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 1686 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 1687 FUID_SIZE_ESTIMATE(zfsvfs)); 1688 } 1689 } 1690 if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp) 1691 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1692 0, SPA_MAXBLOCKSIZE); 1693 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1694 if (error) { 1695 zfs_dirent_unlock(dl); 1696 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1697 dmu_tx_wait(tx); 1698 dmu_tx_abort(tx); 1699 goto top; 1700 } 1701 dmu_tx_abort(tx); 1702 ZFS_EXIT(zfsvfs); 1703 if (aclp) 1704 zfs_acl_free(aclp); 1705 return (error); 1706 } 1707 1708 /* 1709 * Create new node. 1710 */ 1711 zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp); 1712 1713 if (aclp) 1714 zfs_acl_free(aclp); 1715 1716 /* 1717 * Now put new name in parent dir. 1718 */ 1719 (void) zfs_link_create(dl, zp, tx, ZNEW); 1720 1721 *vpp = ZTOV(zp); 1722 1723 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 1724 if (flags & FIGNORECASE) 1725 txtype |= TX_CI; 1726 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, fuidp, vap); 1727 1728 if (fuidp) 1729 zfs_fuid_info_free(fuidp); 1730 dmu_tx_commit(tx); 1731 1732 zfs_dirent_unlock(dl); 1733 1734 ZFS_EXIT(zfsvfs); 1735 return (0); 1736 } 1737 1738 /* 1739 * Remove a directory subdir entry. If the current working 1740 * directory is the same as the subdir to be removed, the 1741 * remove will fail. 1742 * 1743 * IN: dvp - vnode of directory to remove from. 1744 * name - name of directory to be removed. 1745 * cwd - vnode of current working directory. 1746 * cr - credentials of caller. 1747 * ct - caller context 1748 * flags - case flags 1749 * 1750 * RETURN: 0 if success 1751 * error code if failure 1752 * 1753 * Timestamps: 1754 * dvp - ctime|mtime updated 1755 */ 1756 /*ARGSUSED*/ 1757 static int 1758 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 1759 caller_context_t *ct, int flags) 1760 { 1761 znode_t *dzp = VTOZ(dvp); 1762 znode_t *zp; 1763 vnode_t *vp; 1764 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1765 zilog_t *zilog; 1766 zfs_dirlock_t *dl; 1767 dmu_tx_t *tx; 1768 int error; 1769 int zflg = ZEXISTS; 1770 1771 ZFS_ENTER(zfsvfs); 1772 ZFS_VERIFY_ZP(dzp); 1773 zilog = zfsvfs->z_log; 1774 1775 if (flags & FIGNORECASE) 1776 zflg |= ZCILOOK; 1777 top: 1778 zp = NULL; 1779 1780 /* 1781 * Attempt to lock directory; fail if entry doesn't exist. 1782 */ 1783 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1784 NULL, NULL)) { 1785 ZFS_EXIT(zfsvfs); 1786 return (error); 1787 } 1788 1789 vp = ZTOV(zp); 1790 1791 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1792 goto out; 1793 } 1794 1795 if (vp->v_type != VDIR) { 1796 error = ENOTDIR; 1797 goto out; 1798 } 1799 1800 if (vp == cwd) { 1801 error = EINVAL; 1802 goto out; 1803 } 1804 1805 vnevent_rmdir(vp, dvp, name, ct); 1806 1807 /* 1808 * Grab a lock on the directory to make sure that noone is 1809 * trying to add (or lookup) entries while we are removing it. 1810 */ 1811 rw_enter(&zp->z_name_lock, RW_WRITER); 1812 1813 /* 1814 * Grab a lock on the parent pointer to make sure we play well 1815 * with the treewalk and directory rename code. 1816 */ 1817 rw_enter(&zp->z_parent_lock, RW_WRITER); 1818 1819 tx = dmu_tx_create(zfsvfs->z_os); 1820 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1821 dmu_tx_hold_bonus(tx, zp->z_id); 1822 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1823 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1824 if (error) { 1825 rw_exit(&zp->z_parent_lock); 1826 rw_exit(&zp->z_name_lock); 1827 zfs_dirent_unlock(dl); 1828 VN_RELE(vp); 1829 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1830 dmu_tx_wait(tx); 1831 dmu_tx_abort(tx); 1832 goto top; 1833 } 1834 dmu_tx_abort(tx); 1835 ZFS_EXIT(zfsvfs); 1836 return (error); 1837 } 1838 1839 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 1840 1841 if (error == 0) { 1842 uint64_t txtype = TX_RMDIR; 1843 if (flags & FIGNORECASE) 1844 txtype |= TX_CI; 1845 zfs_log_remove(zilog, tx, txtype, dzp, name); 1846 } 1847 1848 dmu_tx_commit(tx); 1849 1850 rw_exit(&zp->z_parent_lock); 1851 rw_exit(&zp->z_name_lock); 1852 out: 1853 zfs_dirent_unlock(dl); 1854 1855 VN_RELE(vp); 1856 1857 ZFS_EXIT(zfsvfs); 1858 return (error); 1859 } 1860 1861 /* 1862 * Read as many directory entries as will fit into the provided 1863 * buffer from the given directory cursor position (specified in 1864 * the uio structure. 1865 * 1866 * IN: vp - vnode of directory to read. 1867 * uio - structure supplying read location, range info, 1868 * and return buffer. 1869 * cr - credentials of caller. 1870 * ct - caller context 1871 * flags - case flags 1872 * 1873 * OUT: uio - updated offset and range, buffer filled. 1874 * eofp - set to true if end-of-file detected. 1875 * 1876 * RETURN: 0 if success 1877 * error code if failure 1878 * 1879 * Timestamps: 1880 * vp - atime updated 1881 * 1882 * Note that the low 4 bits of the cookie returned by zap is always zero. 1883 * This allows us to use the low range for "special" directory entries: 1884 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1885 * we use the offset 2 for the '.zfs' directory. 1886 */ 1887 /* ARGSUSED */ 1888 static int 1889 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, 1890 caller_context_t *ct, int flags) 1891 { 1892 znode_t *zp = VTOZ(vp); 1893 iovec_t *iovp; 1894 edirent_t *eodp; 1895 dirent64_t *odp; 1896 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1897 objset_t *os; 1898 caddr_t outbuf; 1899 size_t bufsize; 1900 zap_cursor_t zc; 1901 zap_attribute_t zap; 1902 uint_t bytes_wanted; 1903 uint64_t offset; /* must be unsigned; checks for < 1 */ 1904 int local_eof; 1905 int outcount; 1906 int error; 1907 uint8_t prefetch; 1908 boolean_t check_sysattrs; 1909 1910 ZFS_ENTER(zfsvfs); 1911 ZFS_VERIFY_ZP(zp); 1912 1913 /* 1914 * If we are not given an eof variable, 1915 * use a local one. 1916 */ 1917 if (eofp == NULL) 1918 eofp = &local_eof; 1919 1920 /* 1921 * Check for valid iov_len. 1922 */ 1923 if (uio->uio_iov->iov_len <= 0) { 1924 ZFS_EXIT(zfsvfs); 1925 return (EINVAL); 1926 } 1927 1928 /* 1929 * Quit if directory has been removed (posix) 1930 */ 1931 if ((*eofp = zp->z_unlinked) != 0) { 1932 ZFS_EXIT(zfsvfs); 1933 return (0); 1934 } 1935 1936 error = 0; 1937 os = zfsvfs->z_os; 1938 offset = uio->uio_loffset; 1939 prefetch = zp->z_zn_prefetch; 1940 1941 /* 1942 * Initialize the iterator cursor. 1943 */ 1944 if (offset <= 3) { 1945 /* 1946 * Start iteration from the beginning of the directory. 1947 */ 1948 zap_cursor_init(&zc, os, zp->z_id); 1949 } else { 1950 /* 1951 * The offset is a serialized cursor. 1952 */ 1953 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1954 } 1955 1956 /* 1957 * Get space to change directory entries into fs independent format. 1958 */ 1959 iovp = uio->uio_iov; 1960 bytes_wanted = iovp->iov_len; 1961 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1962 bufsize = bytes_wanted; 1963 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1964 odp = (struct dirent64 *)outbuf; 1965 } else { 1966 bufsize = bytes_wanted; 1967 odp = (struct dirent64 *)iovp->iov_base; 1968 } 1969 eodp = (struct edirent *)odp; 1970 1971 /* 1972 * If this VFS supports the system attribute view interface; and 1973 * we're looking at an extended attribute directory; and we care 1974 * about normalization conflicts on this vfs; then we must check 1975 * for normalization conflicts with the sysattr name space. 1976 */ 1977 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 1978 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 1979 (flags & V_RDDIR_ENTFLAGS); 1980 1981 /* 1982 * Transform to file-system independent format 1983 */ 1984 outcount = 0; 1985 while (outcount < bytes_wanted) { 1986 ino64_t objnum; 1987 ushort_t reclen; 1988 off64_t *next; 1989 1990 /* 1991 * Special case `.', `..', and `.zfs'. 1992 */ 1993 if (offset == 0) { 1994 (void) strcpy(zap.za_name, "."); 1995 zap.za_normalization_conflict = 0; 1996 objnum = zp->z_id; 1997 } else if (offset == 1) { 1998 (void) strcpy(zap.za_name, ".."); 1999 zap.za_normalization_conflict = 0; 2000 objnum = zp->z_phys->zp_parent; 2001 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2002 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2003 zap.za_normalization_conflict = 0; 2004 objnum = ZFSCTL_INO_ROOT; 2005 } else { 2006 /* 2007 * Grab next entry. 2008 */ 2009 if (error = zap_cursor_retrieve(&zc, &zap)) { 2010 if ((*eofp = (error == ENOENT)) != 0) 2011 break; 2012 else 2013 goto update; 2014 } 2015 2016 if (zap.za_integer_length != 8 || 2017 zap.za_num_integers != 1) { 2018 cmn_err(CE_WARN, "zap_readdir: bad directory " 2019 "entry, obj = %lld, offset = %lld\n", 2020 (u_longlong_t)zp->z_id, 2021 (u_longlong_t)offset); 2022 error = ENXIO; 2023 goto update; 2024 } 2025 2026 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2027 /* 2028 * MacOS X can extract the object type here such as: 2029 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2030 */ 2031 2032 if (check_sysattrs && !zap.za_normalization_conflict) { 2033 zap.za_normalization_conflict = 2034 xattr_sysattr_casechk(zap.za_name); 2035 } 2036 } 2037 2038 if (flags & V_RDDIR_ENTFLAGS) 2039 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2040 else 2041 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2042 2043 /* 2044 * Will this entry fit in the buffer? 2045 */ 2046 if (outcount + reclen > bufsize) { 2047 /* 2048 * Did we manage to fit anything in the buffer? 2049 */ 2050 if (!outcount) { 2051 error = EINVAL; 2052 goto update; 2053 } 2054 break; 2055 } 2056 if (flags & V_RDDIR_ENTFLAGS) { 2057 /* 2058 * Add extended flag entry: 2059 */ 2060 eodp->ed_ino = objnum; 2061 eodp->ed_reclen = reclen; 2062 /* NOTE: ed_off is the offset for the *next* entry */ 2063 next = &(eodp->ed_off); 2064 eodp->ed_eflags = zap.za_normalization_conflict ? 2065 ED_CASE_CONFLICT : 0; 2066 (void) strncpy(eodp->ed_name, zap.za_name, 2067 EDIRENT_NAMELEN(reclen)); 2068 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2069 } else { 2070 /* 2071 * Add normal entry: 2072 */ 2073 odp->d_ino = objnum; 2074 odp->d_reclen = reclen; 2075 /* NOTE: d_off is the offset for the *next* entry */ 2076 next = &(odp->d_off); 2077 (void) strncpy(odp->d_name, zap.za_name, 2078 DIRENT64_NAMELEN(reclen)); 2079 odp = (dirent64_t *)((intptr_t)odp + reclen); 2080 } 2081 outcount += reclen; 2082 2083 ASSERT(outcount <= bufsize); 2084 2085 /* Prefetch znode */ 2086 if (prefetch) 2087 dmu_prefetch(os, objnum, 0, 0); 2088 2089 /* 2090 * Move to the next entry, fill in the previous offset. 2091 */ 2092 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2093 zap_cursor_advance(&zc); 2094 offset = zap_cursor_serialize(&zc); 2095 } else { 2096 offset += 1; 2097 } 2098 *next = offset; 2099 } 2100 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2101 2102 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2103 iovp->iov_base += outcount; 2104 iovp->iov_len -= outcount; 2105 uio->uio_resid -= outcount; 2106 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2107 /* 2108 * Reset the pointer. 2109 */ 2110 offset = uio->uio_loffset; 2111 } 2112 2113 update: 2114 zap_cursor_fini(&zc); 2115 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2116 kmem_free(outbuf, bufsize); 2117 2118 if (error == ENOENT) 2119 error = 0; 2120 2121 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2122 2123 uio->uio_loffset = offset; 2124 ZFS_EXIT(zfsvfs); 2125 return (error); 2126 } 2127 2128 ulong_t zfs_fsync_sync_cnt = 4; 2129 2130 static int 2131 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2132 { 2133 znode_t *zp = VTOZ(vp); 2134 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2135 2136 /* 2137 * Regardless of whether this is required for standards conformance, 2138 * this is the logical behavior when fsync() is called on a file with 2139 * dirty pages. We use B_ASYNC since the ZIL transactions are already 2140 * going to be pushed out as part of the zil_commit(). 2141 */ 2142 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 2143 (vp->v_type == VREG) && !(IS_SWAPVP(vp))) 2144 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct); 2145 2146 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2147 2148 ZFS_ENTER(zfsvfs); 2149 ZFS_VERIFY_ZP(zp); 2150 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 2151 ZFS_EXIT(zfsvfs); 2152 return (0); 2153 } 2154 2155 2156 /* 2157 * Get the requested file attributes and place them in the provided 2158 * vattr structure. 2159 * 2160 * IN: vp - vnode of file. 2161 * vap - va_mask identifies requested attributes. 2162 * If AT_XVATTR set, then optional attrs are requested 2163 * flags - ATTR_NOACLCHECK (CIFS server context) 2164 * cr - credentials of caller. 2165 * ct - caller context 2166 * 2167 * OUT: vap - attribute values. 2168 * 2169 * RETURN: 0 (always succeeds) 2170 */ 2171 /* ARGSUSED */ 2172 static int 2173 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2174 caller_context_t *ct) 2175 { 2176 znode_t *zp = VTOZ(vp); 2177 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2178 znode_phys_t *pzp; 2179 int error = 0; 2180 uint64_t links; 2181 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2182 xoptattr_t *xoap = NULL; 2183 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2184 2185 ZFS_ENTER(zfsvfs); 2186 ZFS_VERIFY_ZP(zp); 2187 pzp = zp->z_phys; 2188 2189 mutex_enter(&zp->z_lock); 2190 2191 /* 2192 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2193 * Also, if we are the owner don't bother, since owner should 2194 * always be allowed to read basic attributes of file. 2195 */ 2196 if (!(pzp->zp_flags & ZFS_ACL_TRIVIAL) && 2197 (pzp->zp_uid != crgetuid(cr))) { 2198 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2199 skipaclchk, cr)) { 2200 mutex_exit(&zp->z_lock); 2201 ZFS_EXIT(zfsvfs); 2202 return (error); 2203 } 2204 } 2205 2206 /* 2207 * Return all attributes. It's cheaper to provide the answer 2208 * than to determine whether we were asked the question. 2209 */ 2210 2211 vap->va_type = vp->v_type; 2212 vap->va_mode = pzp->zp_mode & MODEMASK; 2213 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2214 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2215 vap->va_nodeid = zp->z_id; 2216 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2217 links = pzp->zp_links + 1; 2218 else 2219 links = pzp->zp_links; 2220 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */ 2221 vap->va_size = pzp->zp_size; 2222 vap->va_rdev = vp->v_rdev; 2223 vap->va_seq = zp->z_seq; 2224 2225 /* 2226 * Add in any requested optional attributes and the create time. 2227 * Also set the corresponding bits in the returned attribute bitmap. 2228 */ 2229 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2230 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2231 xoap->xoa_archive = 2232 ((pzp->zp_flags & ZFS_ARCHIVE) != 0); 2233 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2234 } 2235 2236 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2237 xoap->xoa_readonly = 2238 ((pzp->zp_flags & ZFS_READONLY) != 0); 2239 XVA_SET_RTN(xvap, XAT_READONLY); 2240 } 2241 2242 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2243 xoap->xoa_system = 2244 ((pzp->zp_flags & ZFS_SYSTEM) != 0); 2245 XVA_SET_RTN(xvap, XAT_SYSTEM); 2246 } 2247 2248 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2249 xoap->xoa_hidden = 2250 ((pzp->zp_flags & ZFS_HIDDEN) != 0); 2251 XVA_SET_RTN(xvap, XAT_HIDDEN); 2252 } 2253 2254 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2255 xoap->xoa_nounlink = 2256 ((pzp->zp_flags & ZFS_NOUNLINK) != 0); 2257 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2258 } 2259 2260 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2261 xoap->xoa_immutable = 2262 ((pzp->zp_flags & ZFS_IMMUTABLE) != 0); 2263 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2264 } 2265 2266 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2267 xoap->xoa_appendonly = 2268 ((pzp->zp_flags & ZFS_APPENDONLY) != 0); 2269 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2270 } 2271 2272 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2273 xoap->xoa_nodump = 2274 ((pzp->zp_flags & ZFS_NODUMP) != 0); 2275 XVA_SET_RTN(xvap, XAT_NODUMP); 2276 } 2277 2278 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2279 xoap->xoa_opaque = 2280 ((pzp->zp_flags & ZFS_OPAQUE) != 0); 2281 XVA_SET_RTN(xvap, XAT_OPAQUE); 2282 } 2283 2284 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2285 xoap->xoa_av_quarantined = 2286 ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0); 2287 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2288 } 2289 2290 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2291 xoap->xoa_av_modified = 2292 ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0); 2293 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2294 } 2295 2296 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2297 vp->v_type == VREG && 2298 (pzp->zp_flags & ZFS_BONUS_SCANSTAMP)) { 2299 size_t len; 2300 dmu_object_info_t doi; 2301 2302 /* 2303 * Only VREG files have anti-virus scanstamps, so we 2304 * won't conflict with symlinks in the bonus buffer. 2305 */ 2306 dmu_object_info_from_db(zp->z_dbuf, &doi); 2307 len = sizeof (xoap->xoa_av_scanstamp) + 2308 sizeof (znode_phys_t); 2309 if (len <= doi.doi_bonus_size) { 2310 /* 2311 * pzp points to the start of the 2312 * znode_phys_t. pzp + 1 points to the 2313 * first byte after the znode_phys_t. 2314 */ 2315 (void) memcpy(xoap->xoa_av_scanstamp, 2316 pzp + 1, 2317 sizeof (xoap->xoa_av_scanstamp)); 2318 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 2319 } 2320 } 2321 2322 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2323 ZFS_TIME_DECODE(&xoap->xoa_createtime, pzp->zp_crtime); 2324 XVA_SET_RTN(xvap, XAT_CREATETIME); 2325 } 2326 } 2327 2328 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 2329 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 2330 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 2331 2332 mutex_exit(&zp->z_lock); 2333 2334 dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks); 2335 2336 if (zp->z_blksz == 0) { 2337 /* 2338 * Block size hasn't been set; suggest maximal I/O transfers. 2339 */ 2340 vap->va_blksize = zfsvfs->z_max_blksz; 2341 } 2342 2343 ZFS_EXIT(zfsvfs); 2344 return (0); 2345 } 2346 2347 /* 2348 * Set the file attributes to the values contained in the 2349 * vattr structure. 2350 * 2351 * IN: vp - vnode of file to be modified. 2352 * vap - new attribute values. 2353 * If AT_XVATTR set, then optional attrs are being set 2354 * flags - ATTR_UTIME set if non-default time values provided. 2355 * - ATTR_NOACLCHECK (CIFS context only). 2356 * cr - credentials of caller. 2357 * ct - caller context 2358 * 2359 * RETURN: 0 if success 2360 * error code if failure 2361 * 2362 * Timestamps: 2363 * vp - ctime updated, mtime updated if size changed. 2364 */ 2365 /* ARGSUSED */ 2366 static int 2367 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2368 caller_context_t *ct) 2369 { 2370 znode_t *zp = VTOZ(vp); 2371 znode_phys_t *pzp; 2372 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2373 zilog_t *zilog; 2374 dmu_tx_t *tx; 2375 vattr_t oldva; 2376 uint_t mask = vap->va_mask; 2377 uint_t saved_mask; 2378 int trim_mask = 0; 2379 uint64_t new_mode; 2380 znode_t *attrzp; 2381 int need_policy = FALSE; 2382 int err; 2383 zfs_fuid_info_t *fuidp = NULL; 2384 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2385 xoptattr_t *xoap; 2386 zfs_acl_t *aclp = NULL; 2387 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2388 2389 if (mask == 0) 2390 return (0); 2391 2392 if (mask & AT_NOSET) 2393 return (EINVAL); 2394 2395 ZFS_ENTER(zfsvfs); 2396 ZFS_VERIFY_ZP(zp); 2397 2398 pzp = zp->z_phys; 2399 zilog = zfsvfs->z_log; 2400 2401 /* 2402 * Make sure that if we have ephemeral uid/gid or xvattr specified 2403 * that file system is at proper version level 2404 */ 2405 2406 if (zfsvfs->z_use_fuids == B_FALSE && 2407 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2408 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2409 (mask & AT_XVATTR))) { 2410 ZFS_EXIT(zfsvfs); 2411 return (EINVAL); 2412 } 2413 2414 if (mask & AT_SIZE && vp->v_type == VDIR) { 2415 ZFS_EXIT(zfsvfs); 2416 return (EISDIR); 2417 } 2418 2419 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 2420 ZFS_EXIT(zfsvfs); 2421 return (EINVAL); 2422 } 2423 2424 /* 2425 * If this is an xvattr_t, then get a pointer to the structure of 2426 * optional attributes. If this is NULL, then we have a vattr_t. 2427 */ 2428 xoap = xva_getxoptattr(xvap); 2429 2430 /* 2431 * Immutable files can only alter immutable bit and atime 2432 */ 2433 if ((pzp->zp_flags & ZFS_IMMUTABLE) && 2434 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2435 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 2436 ZFS_EXIT(zfsvfs); 2437 return (EPERM); 2438 } 2439 2440 if ((mask & AT_SIZE) && (pzp->zp_flags & ZFS_READONLY)) { 2441 ZFS_EXIT(zfsvfs); 2442 return (EPERM); 2443 } 2444 2445 /* 2446 * Verify timestamps doesn't overflow 32 bits. 2447 * ZFS can handle large timestamps, but 32bit syscalls can't 2448 * handle times greater than 2039. This check should be removed 2449 * once large timestamps are fully supported. 2450 */ 2451 if (mask & (AT_ATIME | AT_MTIME)) { 2452 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 2453 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 2454 ZFS_EXIT(zfsvfs); 2455 return (EOVERFLOW); 2456 } 2457 } 2458 2459 top: 2460 attrzp = NULL; 2461 2462 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2463 ZFS_EXIT(zfsvfs); 2464 return (EROFS); 2465 } 2466 2467 /* 2468 * First validate permissions 2469 */ 2470 2471 if (mask & AT_SIZE) { 2472 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr); 2473 if (err) { 2474 ZFS_EXIT(zfsvfs); 2475 return (err); 2476 } 2477 /* 2478 * XXX - Note, we are not providing any open 2479 * mode flags here (like FNDELAY), so we may 2480 * block if there are locks present... this 2481 * should be addressed in openat(). 2482 */ 2483 /* XXX - would it be OK to generate a log record here? */ 2484 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2485 if (err) { 2486 ZFS_EXIT(zfsvfs); 2487 return (err); 2488 } 2489 } 2490 2491 if (mask & (AT_ATIME|AT_MTIME) || 2492 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2493 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2494 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2495 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2496 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) 2497 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2498 skipaclchk, cr); 2499 2500 if (mask & (AT_UID|AT_GID)) { 2501 int idmask = (mask & (AT_UID|AT_GID)); 2502 int take_owner; 2503 int take_group; 2504 2505 /* 2506 * NOTE: even if a new mode is being set, 2507 * we may clear S_ISUID/S_ISGID bits. 2508 */ 2509 2510 if (!(mask & AT_MODE)) 2511 vap->va_mode = pzp->zp_mode; 2512 2513 /* 2514 * Take ownership or chgrp to group we are a member of 2515 */ 2516 2517 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2518 take_group = (mask & AT_GID) && 2519 zfs_groupmember(zfsvfs, vap->va_gid, cr); 2520 2521 /* 2522 * If both AT_UID and AT_GID are set then take_owner and 2523 * take_group must both be set in order to allow taking 2524 * ownership. 2525 * 2526 * Otherwise, send the check through secpolicy_vnode_setattr() 2527 * 2528 */ 2529 2530 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2531 ((idmask == AT_UID) && take_owner) || 2532 ((idmask == AT_GID) && take_group)) { 2533 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 2534 skipaclchk, cr) == 0) { 2535 /* 2536 * Remove setuid/setgid for non-privileged users 2537 */ 2538 secpolicy_setid_clear(vap, cr); 2539 trim_mask = (mask & (AT_UID|AT_GID)); 2540 } else { 2541 need_policy = TRUE; 2542 } 2543 } else { 2544 need_policy = TRUE; 2545 } 2546 } 2547 2548 mutex_enter(&zp->z_lock); 2549 oldva.va_mode = pzp->zp_mode; 2550 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 2551 if (mask & AT_XVATTR) { 2552 if ((need_policy == FALSE) && 2553 (XVA_ISSET_REQ(xvap, XAT_APPENDONLY) && 2554 xoap->xoa_appendonly != 2555 ((pzp->zp_flags & ZFS_APPENDONLY) != 0)) || 2556 (XVA_ISSET_REQ(xvap, XAT_NOUNLINK) && 2557 xoap->xoa_nounlink != 2558 ((pzp->zp_flags & ZFS_NOUNLINK) != 0)) || 2559 (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE) && 2560 xoap->xoa_immutable != 2561 ((pzp->zp_flags & ZFS_IMMUTABLE) != 0)) || 2562 (XVA_ISSET_REQ(xvap, XAT_NODUMP) && 2563 xoap->xoa_nodump != 2564 ((pzp->zp_flags & ZFS_NODUMP) != 0)) || 2565 (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED) && 2566 xoap->xoa_av_modified != 2567 ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0)) || 2568 ((XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED) && 2569 ((vp->v_type != VREG && xoap->xoa_av_quarantined) || 2570 xoap->xoa_av_quarantined != 2571 ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0)))) || 2572 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) || 2573 (XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 2574 need_policy = TRUE; 2575 } 2576 } 2577 2578 mutex_exit(&zp->z_lock); 2579 2580 if (mask & AT_MODE) { 2581 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 2582 err = secpolicy_setid_setsticky_clear(vp, vap, 2583 &oldva, cr); 2584 if (err) { 2585 ZFS_EXIT(zfsvfs); 2586 return (err); 2587 } 2588 trim_mask |= AT_MODE; 2589 } else { 2590 need_policy = TRUE; 2591 } 2592 } 2593 2594 if (need_policy) { 2595 /* 2596 * If trim_mask is set then take ownership 2597 * has been granted or write_acl is present and user 2598 * has the ability to modify mode. In that case remove 2599 * UID|GID and or MODE from mask so that 2600 * secpolicy_vnode_setattr() doesn't revoke it. 2601 */ 2602 2603 if (trim_mask) { 2604 saved_mask = vap->va_mask; 2605 vap->va_mask &= ~trim_mask; 2606 } 2607 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2608 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 2609 if (err) { 2610 ZFS_EXIT(zfsvfs); 2611 return (err); 2612 } 2613 2614 if (trim_mask) 2615 vap->va_mask |= saved_mask; 2616 } 2617 2618 /* 2619 * secpolicy_vnode_setattr, or take ownership may have 2620 * changed va_mask 2621 */ 2622 mask = vap->va_mask; 2623 2624 tx = dmu_tx_create(zfsvfs->z_os); 2625 dmu_tx_hold_bonus(tx, zp->z_id); 2626 if (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2627 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid))) { 2628 if (zfsvfs->z_fuid_obj == 0) { 2629 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 2630 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2631 FUID_SIZE_ESTIMATE(zfsvfs)); 2632 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 2633 } else { 2634 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 2635 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 2636 FUID_SIZE_ESTIMATE(zfsvfs)); 2637 } 2638 } 2639 2640 if (mask & AT_MODE) { 2641 uint64_t pmode = pzp->zp_mode; 2642 2643 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2644 2645 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) { 2646 dmu_tx_abort(tx); 2647 ZFS_EXIT(zfsvfs); 2648 return (err); 2649 } 2650 if (pzp->zp_acl.z_acl_extern_obj) { 2651 /* Are we upgrading ACL from old V0 format to new V1 */ 2652 if (zfsvfs->z_version <= ZPL_VERSION_FUID && 2653 pzp->zp_acl.z_acl_version == 2654 ZFS_ACL_VERSION_INITIAL) { 2655 dmu_tx_hold_free(tx, 2656 pzp->zp_acl.z_acl_extern_obj, 0, 2657 DMU_OBJECT_END); 2658 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2659 0, aclp->z_acl_bytes); 2660 } else { 2661 dmu_tx_hold_write(tx, 2662 pzp->zp_acl.z_acl_extern_obj, 0, 2663 aclp->z_acl_bytes); 2664 } 2665 } else if (aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2666 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2667 0, aclp->z_acl_bytes); 2668 } 2669 } 2670 2671 if ((mask & (AT_UID | AT_GID)) && pzp->zp_xattr != 0) { 2672 err = zfs_zget(zp->z_zfsvfs, pzp->zp_xattr, &attrzp); 2673 if (err) { 2674 dmu_tx_abort(tx); 2675 ZFS_EXIT(zfsvfs); 2676 if (aclp) 2677 zfs_acl_free(aclp); 2678 return (err); 2679 } 2680 dmu_tx_hold_bonus(tx, attrzp->z_id); 2681 } 2682 2683 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2684 if (err) { 2685 if (attrzp) 2686 VN_RELE(ZTOV(attrzp)); 2687 2688 if (aclp) { 2689 zfs_acl_free(aclp); 2690 aclp = NULL; 2691 } 2692 2693 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2694 dmu_tx_wait(tx); 2695 dmu_tx_abort(tx); 2696 goto top; 2697 } 2698 dmu_tx_abort(tx); 2699 ZFS_EXIT(zfsvfs); 2700 return (err); 2701 } 2702 2703 dmu_buf_will_dirty(zp->z_dbuf, tx); 2704 2705 /* 2706 * Set each attribute requested. 2707 * We group settings according to the locks they need to acquire. 2708 * 2709 * Note: you cannot set ctime directly, although it will be 2710 * updated as a side-effect of calling this function. 2711 */ 2712 2713 mutex_enter(&zp->z_lock); 2714 2715 if (mask & AT_MODE) { 2716 mutex_enter(&zp->z_acl_lock); 2717 zp->z_phys->zp_mode = new_mode; 2718 err = zfs_aclset_common(zp, aclp, cr, &fuidp, tx); 2719 ASSERT3U(err, ==, 0); 2720 mutex_exit(&zp->z_acl_lock); 2721 } 2722 2723 if (attrzp) 2724 mutex_enter(&attrzp->z_lock); 2725 2726 if (mask & AT_UID) { 2727 pzp->zp_uid = zfs_fuid_create(zfsvfs, 2728 vap->va_uid, cr, ZFS_OWNER, tx, &fuidp); 2729 if (attrzp) { 2730 attrzp->z_phys->zp_uid = zfs_fuid_create(zfsvfs, 2731 vap->va_uid, cr, ZFS_OWNER, tx, &fuidp); 2732 } 2733 } 2734 2735 if (mask & AT_GID) { 2736 pzp->zp_gid = zfs_fuid_create(zfsvfs, vap->va_gid, 2737 cr, ZFS_GROUP, tx, &fuidp); 2738 if (attrzp) 2739 attrzp->z_phys->zp_gid = zfs_fuid_create(zfsvfs, 2740 vap->va_gid, cr, ZFS_GROUP, tx, &fuidp); 2741 } 2742 2743 if (aclp) 2744 zfs_acl_free(aclp); 2745 2746 if (attrzp) 2747 mutex_exit(&attrzp->z_lock); 2748 2749 if (mask & AT_ATIME) 2750 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2751 2752 if (mask & AT_MTIME) 2753 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2754 2755 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 2756 if (mask & AT_SIZE) 2757 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2758 else if (mask != 0) 2759 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2760 /* 2761 * Do this after setting timestamps to prevent timestamp 2762 * update from toggling bit 2763 */ 2764 2765 if (xoap && (mask & AT_XVATTR)) { 2766 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 2767 size_t len; 2768 dmu_object_info_t doi; 2769 2770 ASSERT(vp->v_type == VREG); 2771 2772 /* Grow the bonus buffer if necessary. */ 2773 dmu_object_info_from_db(zp->z_dbuf, &doi); 2774 len = sizeof (xoap->xoa_av_scanstamp) + 2775 sizeof (znode_phys_t); 2776 if (len > doi.doi_bonus_size) 2777 VERIFY(dmu_set_bonus(zp->z_dbuf, len, tx) == 0); 2778 } 2779 zfs_xvattr_set(zp, xvap); 2780 } 2781 2782 if (mask != 0) 2783 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 2784 2785 if (fuidp) 2786 zfs_fuid_info_free(fuidp); 2787 mutex_exit(&zp->z_lock); 2788 2789 if (attrzp) 2790 VN_RELE(ZTOV(attrzp)); 2791 2792 dmu_tx_commit(tx); 2793 2794 ZFS_EXIT(zfsvfs); 2795 return (err); 2796 } 2797 2798 typedef struct zfs_zlock { 2799 krwlock_t *zl_rwlock; /* lock we acquired */ 2800 znode_t *zl_znode; /* znode we held */ 2801 struct zfs_zlock *zl_next; /* next in list */ 2802 } zfs_zlock_t; 2803 2804 /* 2805 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2806 */ 2807 static void 2808 zfs_rename_unlock(zfs_zlock_t **zlpp) 2809 { 2810 zfs_zlock_t *zl; 2811 2812 while ((zl = *zlpp) != NULL) { 2813 if (zl->zl_znode != NULL) 2814 VN_RELE(ZTOV(zl->zl_znode)); 2815 rw_exit(zl->zl_rwlock); 2816 *zlpp = zl->zl_next; 2817 kmem_free(zl, sizeof (*zl)); 2818 } 2819 } 2820 2821 /* 2822 * Search back through the directory tree, using the ".." entries. 2823 * Lock each directory in the chain to prevent concurrent renames. 2824 * Fail any attempt to move a directory into one of its own descendants. 2825 * XXX - z_parent_lock can overlap with map or grow locks 2826 */ 2827 static int 2828 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2829 { 2830 zfs_zlock_t *zl; 2831 znode_t *zp = tdzp; 2832 uint64_t rootid = zp->z_zfsvfs->z_root; 2833 uint64_t *oidp = &zp->z_id; 2834 krwlock_t *rwlp = &szp->z_parent_lock; 2835 krw_t rw = RW_WRITER; 2836 2837 /* 2838 * First pass write-locks szp and compares to zp->z_id. 2839 * Later passes read-lock zp and compare to zp->z_parent. 2840 */ 2841 do { 2842 if (!rw_tryenter(rwlp, rw)) { 2843 /* 2844 * Another thread is renaming in this path. 2845 * Note that if we are a WRITER, we don't have any 2846 * parent_locks held yet. 2847 */ 2848 if (rw == RW_READER && zp->z_id > szp->z_id) { 2849 /* 2850 * Drop our locks and restart 2851 */ 2852 zfs_rename_unlock(&zl); 2853 *zlpp = NULL; 2854 zp = tdzp; 2855 oidp = &zp->z_id; 2856 rwlp = &szp->z_parent_lock; 2857 rw = RW_WRITER; 2858 continue; 2859 } else { 2860 /* 2861 * Wait for other thread to drop its locks 2862 */ 2863 rw_enter(rwlp, rw); 2864 } 2865 } 2866 2867 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2868 zl->zl_rwlock = rwlp; 2869 zl->zl_znode = NULL; 2870 zl->zl_next = *zlpp; 2871 *zlpp = zl; 2872 2873 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2874 return (EINVAL); 2875 2876 if (*oidp == rootid) /* We've hit the top */ 2877 return (0); 2878 2879 if (rw == RW_READER) { /* i.e. not the first pass */ 2880 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2881 if (error) 2882 return (error); 2883 zl->zl_znode = zp; 2884 } 2885 oidp = &zp->z_phys->zp_parent; 2886 rwlp = &zp->z_parent_lock; 2887 rw = RW_READER; 2888 2889 } while (zp->z_id != sdzp->z_id); 2890 2891 return (0); 2892 } 2893 2894 /* 2895 * Move an entry from the provided source directory to the target 2896 * directory. Change the entry name as indicated. 2897 * 2898 * IN: sdvp - Source directory containing the "old entry". 2899 * snm - Old entry name. 2900 * tdvp - Target directory to contain the "new entry". 2901 * tnm - New entry name. 2902 * cr - credentials of caller. 2903 * ct - caller context 2904 * flags - case flags 2905 * 2906 * RETURN: 0 if success 2907 * error code if failure 2908 * 2909 * Timestamps: 2910 * sdvp,tdvp - ctime|mtime updated 2911 */ 2912 /*ARGSUSED*/ 2913 static int 2914 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 2915 caller_context_t *ct, int flags) 2916 { 2917 znode_t *tdzp, *szp, *tzp; 2918 znode_t *sdzp = VTOZ(sdvp); 2919 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2920 zilog_t *zilog; 2921 vnode_t *realvp; 2922 zfs_dirlock_t *sdl, *tdl; 2923 dmu_tx_t *tx; 2924 zfs_zlock_t *zl; 2925 int cmp, serr, terr; 2926 int error = 0; 2927 int zflg = 0; 2928 2929 ZFS_ENTER(zfsvfs); 2930 ZFS_VERIFY_ZP(sdzp); 2931 zilog = zfsvfs->z_log; 2932 2933 /* 2934 * Make sure we have the real vp for the target directory. 2935 */ 2936 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 2937 tdvp = realvp; 2938 2939 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2940 ZFS_EXIT(zfsvfs); 2941 return (EXDEV); 2942 } 2943 2944 tdzp = VTOZ(tdvp); 2945 ZFS_VERIFY_ZP(tdzp); 2946 if (zfsvfs->z_utf8 && u8_validate(tnm, 2947 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2948 ZFS_EXIT(zfsvfs); 2949 return (EILSEQ); 2950 } 2951 2952 if (flags & FIGNORECASE) 2953 zflg |= ZCILOOK; 2954 2955 top: 2956 szp = NULL; 2957 tzp = NULL; 2958 zl = NULL; 2959 2960 /* 2961 * This is to prevent the creation of links into attribute space 2962 * by renaming a linked file into/outof an attribute directory. 2963 * See the comment in zfs_link() for why this is considered bad. 2964 */ 2965 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2966 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2967 ZFS_EXIT(zfsvfs); 2968 return (EINVAL); 2969 } 2970 2971 /* 2972 * Lock source and target directory entries. To prevent deadlock, 2973 * a lock ordering must be defined. We lock the directory with 2974 * the smallest object id first, or if it's a tie, the one with 2975 * the lexically first name. 2976 */ 2977 if (sdzp->z_id < tdzp->z_id) { 2978 cmp = -1; 2979 } else if (sdzp->z_id > tdzp->z_id) { 2980 cmp = 1; 2981 } else { 2982 /* 2983 * First compare the two name arguments without 2984 * considering any case folding. 2985 */ 2986 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 2987 2988 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 2989 ASSERT(error == 0 || !zfsvfs->z_utf8); 2990 if (cmp == 0) { 2991 /* 2992 * POSIX: "If the old argument and the new argument 2993 * both refer to links to the same existing file, 2994 * the rename() function shall return successfully 2995 * and perform no other action." 2996 */ 2997 ZFS_EXIT(zfsvfs); 2998 return (0); 2999 } 3000 /* 3001 * If the file system is case-folding, then we may 3002 * have some more checking to do. A case-folding file 3003 * system is either supporting mixed case sensitivity 3004 * access or is completely case-insensitive. Note 3005 * that the file system is always case preserving. 3006 * 3007 * In mixed sensitivity mode case sensitive behavior 3008 * is the default. FIGNORECASE must be used to 3009 * explicitly request case insensitive behavior. 3010 * 3011 * If the source and target names provided differ only 3012 * by case (e.g., a request to rename 'tim' to 'Tim'), 3013 * we will treat this as a special case in the 3014 * case-insensitive mode: as long as the source name 3015 * is an exact match, we will allow this to proceed as 3016 * a name-change request. 3017 */ 3018 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3019 (zfsvfs->z_case == ZFS_CASE_MIXED && 3020 flags & FIGNORECASE)) && 3021 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3022 &error) == 0) { 3023 /* 3024 * case preserving rename request, require exact 3025 * name matches 3026 */ 3027 zflg |= ZCIEXACT; 3028 zflg &= ~ZCILOOK; 3029 } 3030 } 3031 3032 if (cmp < 0) { 3033 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3034 ZEXISTS | zflg, NULL, NULL); 3035 terr = zfs_dirent_lock(&tdl, 3036 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3037 } else { 3038 terr = zfs_dirent_lock(&tdl, 3039 tdzp, tnm, &tzp, zflg, NULL, NULL); 3040 serr = zfs_dirent_lock(&sdl, 3041 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3042 NULL, NULL); 3043 } 3044 3045 if (serr) { 3046 /* 3047 * Source entry invalid or not there. 3048 */ 3049 if (!terr) { 3050 zfs_dirent_unlock(tdl); 3051 if (tzp) 3052 VN_RELE(ZTOV(tzp)); 3053 } 3054 if (strcmp(snm, "..") == 0) 3055 serr = EINVAL; 3056 ZFS_EXIT(zfsvfs); 3057 return (serr); 3058 } 3059 if (terr) { 3060 zfs_dirent_unlock(sdl); 3061 VN_RELE(ZTOV(szp)); 3062 if (strcmp(tnm, "..") == 0) 3063 terr = EINVAL; 3064 ZFS_EXIT(zfsvfs); 3065 return (terr); 3066 } 3067 3068 /* 3069 * Must have write access at the source to remove the old entry 3070 * and write access at the target to create the new entry. 3071 * Note that if target and source are the same, this can be 3072 * done in a single check. 3073 */ 3074 3075 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3076 goto out; 3077 3078 if (ZTOV(szp)->v_type == VDIR) { 3079 /* 3080 * Check to make sure rename is valid. 3081 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3082 */ 3083 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3084 goto out; 3085 } 3086 3087 /* 3088 * Does target exist? 3089 */ 3090 if (tzp) { 3091 /* 3092 * Source and target must be the same type. 3093 */ 3094 if (ZTOV(szp)->v_type == VDIR) { 3095 if (ZTOV(tzp)->v_type != VDIR) { 3096 error = ENOTDIR; 3097 goto out; 3098 } 3099 } else { 3100 if (ZTOV(tzp)->v_type == VDIR) { 3101 error = EISDIR; 3102 goto out; 3103 } 3104 } 3105 /* 3106 * POSIX dictates that when the source and target 3107 * entries refer to the same file object, rename 3108 * must do nothing and exit without error. 3109 */ 3110 if (szp->z_id == tzp->z_id) { 3111 error = 0; 3112 goto out; 3113 } 3114 } 3115 3116 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3117 if (tzp) 3118 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3119 3120 /* 3121 * notify the target directory if it is not the same 3122 * as source directory. 3123 */ 3124 if (tdvp != sdvp) { 3125 vnevent_rename_dest_dir(tdvp, ct); 3126 } 3127 3128 tx = dmu_tx_create(zfsvfs->z_os); 3129 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 3130 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 3131 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3132 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3133 if (sdzp != tdzp) 3134 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 3135 if (tzp) 3136 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 3137 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3138 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3139 if (error) { 3140 if (zl != NULL) 3141 zfs_rename_unlock(&zl); 3142 zfs_dirent_unlock(sdl); 3143 zfs_dirent_unlock(tdl); 3144 VN_RELE(ZTOV(szp)); 3145 if (tzp) 3146 VN_RELE(ZTOV(tzp)); 3147 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3148 dmu_tx_wait(tx); 3149 dmu_tx_abort(tx); 3150 goto top; 3151 } 3152 dmu_tx_abort(tx); 3153 ZFS_EXIT(zfsvfs); 3154 return (error); 3155 } 3156 3157 if (tzp) /* Attempt to remove the existing target */ 3158 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 3159 3160 if (error == 0) { 3161 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 3162 if (error == 0) { 3163 szp->z_phys->zp_flags |= ZFS_AV_MODIFIED; 3164 3165 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 3166 ASSERT(error == 0); 3167 3168 zfs_log_rename(zilog, tx, 3169 TX_RENAME | (flags & FIGNORECASE ? TX_CI : 0), 3170 sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp); 3171 3172 /* Update path information for the target vnode */ 3173 vn_renamepath(tdvp, ZTOV(szp), tnm, strlen(tnm)); 3174 } 3175 } 3176 3177 dmu_tx_commit(tx); 3178 out: 3179 if (zl != NULL) 3180 zfs_rename_unlock(&zl); 3181 3182 zfs_dirent_unlock(sdl); 3183 zfs_dirent_unlock(tdl); 3184 3185 VN_RELE(ZTOV(szp)); 3186 if (tzp) 3187 VN_RELE(ZTOV(tzp)); 3188 3189 ZFS_EXIT(zfsvfs); 3190 return (error); 3191 } 3192 3193 /* 3194 * Insert the indicated symbolic reference entry into the directory. 3195 * 3196 * IN: dvp - Directory to contain new symbolic link. 3197 * link - Name for new symlink entry. 3198 * vap - Attributes of new entry. 3199 * target - Target path of new symlink. 3200 * cr - credentials of caller. 3201 * ct - caller context 3202 * flags - case flags 3203 * 3204 * RETURN: 0 if success 3205 * error code if failure 3206 * 3207 * Timestamps: 3208 * dvp - ctime|mtime updated 3209 */ 3210 /*ARGSUSED*/ 3211 static int 3212 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr, 3213 caller_context_t *ct, int flags) 3214 { 3215 znode_t *zp, *dzp = VTOZ(dvp); 3216 zfs_dirlock_t *dl; 3217 dmu_tx_t *tx; 3218 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3219 zilog_t *zilog; 3220 int len = strlen(link); 3221 int error; 3222 int zflg = ZNEW; 3223 zfs_fuid_info_t *fuidp = NULL; 3224 3225 ASSERT(vap->va_type == VLNK); 3226 3227 ZFS_ENTER(zfsvfs); 3228 ZFS_VERIFY_ZP(dzp); 3229 zilog = zfsvfs->z_log; 3230 3231 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 3232 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3233 ZFS_EXIT(zfsvfs); 3234 return (EILSEQ); 3235 } 3236 if (flags & FIGNORECASE) 3237 zflg |= ZCILOOK; 3238 top: 3239 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3240 ZFS_EXIT(zfsvfs); 3241 return (error); 3242 } 3243 3244 if (len > MAXPATHLEN) { 3245 ZFS_EXIT(zfsvfs); 3246 return (ENAMETOOLONG); 3247 } 3248 3249 /* 3250 * Attempt to lock directory; fail if entry already exists. 3251 */ 3252 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 3253 if (error) { 3254 ZFS_EXIT(zfsvfs); 3255 return (error); 3256 } 3257 3258 tx = dmu_tx_create(zfsvfs->z_os); 3259 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 3260 dmu_tx_hold_bonus(tx, dzp->z_id); 3261 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3262 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 3263 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 3264 if (IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr))) { 3265 if (zfsvfs->z_fuid_obj == 0) { 3266 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 3267 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 3268 FUID_SIZE_ESTIMATE(zfsvfs)); 3269 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 3270 } else { 3271 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 3272 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 3273 FUID_SIZE_ESTIMATE(zfsvfs)); 3274 } 3275 } 3276 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3277 if (error) { 3278 zfs_dirent_unlock(dl); 3279 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3280 dmu_tx_wait(tx); 3281 dmu_tx_abort(tx); 3282 goto top; 3283 } 3284 dmu_tx_abort(tx); 3285 ZFS_EXIT(zfsvfs); 3286 return (error); 3287 } 3288 3289 dmu_buf_will_dirty(dzp->z_dbuf, tx); 3290 3291 /* 3292 * Create a new object for the symlink. 3293 * Put the link content into bonus buffer if it will fit; 3294 * otherwise, store it just like any other file data. 3295 */ 3296 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 3297 zfs_mknode(dzp, vap, tx, cr, 0, &zp, len, NULL, &fuidp); 3298 if (len != 0) 3299 bcopy(link, zp->z_phys + 1, len); 3300 } else { 3301 dmu_buf_t *dbp; 3302 3303 zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, NULL, &fuidp); 3304 /* 3305 * Nothing can access the znode yet so no locking needed 3306 * for growing the znode's blocksize. 3307 */ 3308 zfs_grow_blocksize(zp, len, tx); 3309 3310 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, 3311 zp->z_id, 0, FTAG, &dbp)); 3312 dmu_buf_will_dirty(dbp, tx); 3313 3314 ASSERT3U(len, <=, dbp->db_size); 3315 bcopy(link, dbp->db_data, len); 3316 dmu_buf_rele(dbp, FTAG); 3317 } 3318 zp->z_phys->zp_size = len; 3319 3320 /* 3321 * Insert the new object into the directory. 3322 */ 3323 (void) zfs_link_create(dl, zp, tx, ZNEW); 3324 out: 3325 if (error == 0) { 3326 uint64_t txtype = TX_SYMLINK; 3327 if (flags & FIGNORECASE) 3328 txtype |= TX_CI; 3329 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 3330 } 3331 if (fuidp) 3332 zfs_fuid_info_free(fuidp); 3333 3334 dmu_tx_commit(tx); 3335 3336 zfs_dirent_unlock(dl); 3337 3338 VN_RELE(ZTOV(zp)); 3339 3340 ZFS_EXIT(zfsvfs); 3341 return (error); 3342 } 3343 3344 /* 3345 * Return, in the buffer contained in the provided uio structure, 3346 * the symbolic path referred to by vp. 3347 * 3348 * IN: vp - vnode of symbolic link. 3349 * uoip - structure to contain the link path. 3350 * cr - credentials of caller. 3351 * ct - caller context 3352 * 3353 * OUT: uio - structure to contain the link path. 3354 * 3355 * RETURN: 0 if success 3356 * error code if failure 3357 * 3358 * Timestamps: 3359 * vp - atime updated 3360 */ 3361 /* ARGSUSED */ 3362 static int 3363 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 3364 { 3365 znode_t *zp = VTOZ(vp); 3366 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3367 size_t bufsz; 3368 int error; 3369 3370 ZFS_ENTER(zfsvfs); 3371 ZFS_VERIFY_ZP(zp); 3372 3373 bufsz = (size_t)zp->z_phys->zp_size; 3374 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 3375 error = uiomove(zp->z_phys + 1, 3376 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 3377 } else { 3378 dmu_buf_t *dbp; 3379 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 3380 if (error) { 3381 ZFS_EXIT(zfsvfs); 3382 return (error); 3383 } 3384 error = uiomove(dbp->db_data, 3385 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 3386 dmu_buf_rele(dbp, FTAG); 3387 } 3388 3389 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3390 ZFS_EXIT(zfsvfs); 3391 return (error); 3392 } 3393 3394 /* 3395 * Insert a new entry into directory tdvp referencing svp. 3396 * 3397 * IN: tdvp - Directory to contain new entry. 3398 * svp - vnode of new entry. 3399 * name - name of new entry. 3400 * cr - credentials of caller. 3401 * ct - caller context 3402 * 3403 * RETURN: 0 if success 3404 * error code if failure 3405 * 3406 * Timestamps: 3407 * tdvp - ctime|mtime updated 3408 * svp - ctime updated 3409 */ 3410 /* ARGSUSED */ 3411 static int 3412 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 3413 caller_context_t *ct, int flags) 3414 { 3415 znode_t *dzp = VTOZ(tdvp); 3416 znode_t *tzp, *szp; 3417 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3418 zilog_t *zilog; 3419 zfs_dirlock_t *dl; 3420 dmu_tx_t *tx; 3421 vnode_t *realvp; 3422 int error; 3423 int zf = ZNEW; 3424 uid_t owner; 3425 3426 ASSERT(tdvp->v_type == VDIR); 3427 3428 ZFS_ENTER(zfsvfs); 3429 ZFS_VERIFY_ZP(dzp); 3430 zilog = zfsvfs->z_log; 3431 3432 if (VOP_REALVP(svp, &realvp, ct) == 0) 3433 svp = realvp; 3434 3435 if (svp->v_vfsp != tdvp->v_vfsp) { 3436 ZFS_EXIT(zfsvfs); 3437 return (EXDEV); 3438 } 3439 szp = VTOZ(svp); 3440 ZFS_VERIFY_ZP(szp); 3441 3442 if (zfsvfs->z_utf8 && u8_validate(name, 3443 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3444 ZFS_EXIT(zfsvfs); 3445 return (EILSEQ); 3446 } 3447 if (flags & FIGNORECASE) 3448 zf |= ZCILOOK; 3449 3450 top: 3451 /* 3452 * We do not support links between attributes and non-attributes 3453 * because of the potential security risk of creating links 3454 * into "normal" file space in order to circumvent restrictions 3455 * imposed in attribute space. 3456 */ 3457 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 3458 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 3459 ZFS_EXIT(zfsvfs); 3460 return (EINVAL); 3461 } 3462 3463 /* 3464 * POSIX dictates that we return EPERM here. 3465 * Better choices include ENOTSUP or EISDIR. 3466 */ 3467 if (svp->v_type == VDIR) { 3468 ZFS_EXIT(zfsvfs); 3469 return (EPERM); 3470 } 3471 3472 owner = zfs_fuid_map_id(zfsvfs, szp->z_phys->zp_uid, cr, ZFS_OWNER); 3473 if (owner != crgetuid(cr) && 3474 secpolicy_basic_link(cr) != 0) { 3475 ZFS_EXIT(zfsvfs); 3476 return (EPERM); 3477 } 3478 3479 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3480 ZFS_EXIT(zfsvfs); 3481 return (error); 3482 } 3483 3484 /* 3485 * Attempt to lock directory; fail if entry already exists. 3486 */ 3487 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 3488 if (error) { 3489 ZFS_EXIT(zfsvfs); 3490 return (error); 3491 } 3492 3493 tx = dmu_tx_create(zfsvfs->z_os); 3494 dmu_tx_hold_bonus(tx, szp->z_id); 3495 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3496 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3497 if (error) { 3498 zfs_dirent_unlock(dl); 3499 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3500 dmu_tx_wait(tx); 3501 dmu_tx_abort(tx); 3502 goto top; 3503 } 3504 dmu_tx_abort(tx); 3505 ZFS_EXIT(zfsvfs); 3506 return (error); 3507 } 3508 3509 error = zfs_link_create(dl, szp, tx, 0); 3510 3511 if (error == 0) { 3512 uint64_t txtype = TX_LINK; 3513 if (flags & FIGNORECASE) 3514 txtype |= TX_CI; 3515 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 3516 } 3517 3518 dmu_tx_commit(tx); 3519 3520 zfs_dirent_unlock(dl); 3521 3522 if (error == 0) { 3523 vnevent_link(svp, ct); 3524 } 3525 3526 ZFS_EXIT(zfsvfs); 3527 return (error); 3528 } 3529 3530 /* 3531 * zfs_null_putapage() is used when the file system has been force 3532 * unmounted. It just drops the pages. 3533 */ 3534 /* ARGSUSED */ 3535 static int 3536 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 3537 size_t *lenp, int flags, cred_t *cr) 3538 { 3539 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 3540 return (0); 3541 } 3542 3543 /* 3544 * Push a page out to disk, klustering if possible. 3545 * 3546 * IN: vp - file to push page to. 3547 * pp - page to push. 3548 * flags - additional flags. 3549 * cr - credentials of caller. 3550 * 3551 * OUT: offp - start of range pushed. 3552 * lenp - len of range pushed. 3553 * 3554 * RETURN: 0 if success 3555 * error code if failure 3556 * 3557 * NOTE: callers must have locked the page to be pushed. On 3558 * exit, the page (and all other pages in the kluster) must be 3559 * unlocked. 3560 */ 3561 /* ARGSUSED */ 3562 static int 3563 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 3564 size_t *lenp, int flags, cred_t *cr) 3565 { 3566 znode_t *zp = VTOZ(vp); 3567 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3568 zilog_t *zilog = zfsvfs->z_log; 3569 dmu_tx_t *tx; 3570 rl_t *rl; 3571 u_offset_t off, koff; 3572 size_t len, klen; 3573 uint64_t filesz; 3574 int err; 3575 3576 filesz = zp->z_phys->zp_size; 3577 off = pp->p_offset; 3578 len = PAGESIZE; 3579 /* 3580 * If our blocksize is bigger than the page size, try to kluster 3581 * muiltiple pages so that we write a full block (thus avoiding 3582 * a read-modify-write). 3583 */ 3584 if (off < filesz && zp->z_blksz > PAGESIZE) { 3585 if (!ISP2(zp->z_blksz)) { 3586 /* Only one block in the file. */ 3587 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3588 koff = 0; 3589 } else { 3590 klen = zp->z_blksz; 3591 koff = P2ALIGN(off, (u_offset_t)klen); 3592 } 3593 ASSERT(koff <= filesz); 3594 if (koff + klen > filesz) 3595 klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE); 3596 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 3597 } 3598 ASSERT3U(btop(len), ==, btopr(len)); 3599 top: 3600 rl = zfs_range_lock(zp, off, len, RL_WRITER); 3601 /* 3602 * Can't push pages past end-of-file. 3603 */ 3604 filesz = zp->z_phys->zp_size; 3605 if (off >= filesz) { 3606 /* ignore all pages */ 3607 err = 0; 3608 goto out; 3609 } else if (off + len > filesz) { 3610 int npages = btopr(filesz - off); 3611 page_t *trunc; 3612 3613 page_list_break(&pp, &trunc, npages); 3614 /* ignore pages past end of file */ 3615 if (trunc) 3616 pvn_write_done(trunc, flags); 3617 len = filesz - off; 3618 } 3619 3620 tx = dmu_tx_create(zfsvfs->z_os); 3621 dmu_tx_hold_write(tx, zp->z_id, off, len); 3622 dmu_tx_hold_bonus(tx, zp->z_id); 3623 err = dmu_tx_assign(tx, zfsvfs->z_assign); 3624 if (err != 0) { 3625 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3626 zfs_range_unlock(rl); 3627 dmu_tx_wait(tx); 3628 dmu_tx_abort(tx); 3629 err = 0; 3630 goto top; 3631 } 3632 dmu_tx_abort(tx); 3633 goto out; 3634 } 3635 3636 if (zp->z_blksz <= PAGESIZE) { 3637 caddr_t va = zfs_map_page(pp, S_READ); 3638 ASSERT3U(len, <=, PAGESIZE); 3639 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 3640 zfs_unmap_page(pp, va); 3641 } else { 3642 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 3643 } 3644 3645 if (err == 0) { 3646 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 3647 zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0); 3648 dmu_tx_commit(tx); 3649 } 3650 3651 out: 3652 zfs_range_unlock(rl); 3653 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 3654 if (offp) 3655 *offp = off; 3656 if (lenp) 3657 *lenp = len; 3658 3659 return (err); 3660 } 3661 3662 /* 3663 * Copy the portion of the file indicated from pages into the file. 3664 * The pages are stored in a page list attached to the files vnode. 3665 * 3666 * IN: vp - vnode of file to push page data to. 3667 * off - position in file to put data. 3668 * len - amount of data to write. 3669 * flags - flags to control the operation. 3670 * cr - credentials of caller. 3671 * ct - caller context. 3672 * 3673 * RETURN: 0 if success 3674 * error code if failure 3675 * 3676 * Timestamps: 3677 * vp - ctime|mtime updated 3678 */ 3679 /*ARGSUSED*/ 3680 static int 3681 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 3682 caller_context_t *ct) 3683 { 3684 znode_t *zp = VTOZ(vp); 3685 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3686 page_t *pp; 3687 size_t io_len; 3688 u_offset_t io_off; 3689 uint64_t filesz; 3690 int error = 0; 3691 3692 ZFS_ENTER(zfsvfs); 3693 ZFS_VERIFY_ZP(zp); 3694 3695 if (len == 0) { 3696 /* 3697 * Search the entire vp list for pages >= off. 3698 */ 3699 error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage, 3700 flags, cr); 3701 goto out; 3702 } 3703 3704 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3705 if (off > filesz) { 3706 /* past end of file */ 3707 ZFS_EXIT(zfsvfs); 3708 return (0); 3709 } 3710 3711 len = MIN(len, filesz - off); 3712 3713 for (io_off = off; io_off < off + len; io_off += io_len) { 3714 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 3715 pp = page_lookup(vp, io_off, 3716 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 3717 } else { 3718 pp = page_lookup_nowait(vp, io_off, 3719 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 3720 } 3721 3722 if (pp != NULL && pvn_getdirty(pp, flags)) { 3723 int err; 3724 3725 /* 3726 * Found a dirty page to push 3727 */ 3728 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 3729 if (err) 3730 error = err; 3731 } else { 3732 io_len = PAGESIZE; 3733 } 3734 } 3735 out: 3736 if ((flags & B_ASYNC) == 0) 3737 zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id); 3738 ZFS_EXIT(zfsvfs); 3739 return (error); 3740 } 3741 3742 /*ARGSUSED*/ 3743 void 3744 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 3745 { 3746 znode_t *zp = VTOZ(vp); 3747 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3748 int error; 3749 3750 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 3751 if (zp->z_dbuf == NULL) { 3752 /* 3753 * The fs has been unmounted, or we did a 3754 * suspend/resume and this file no longer exists. 3755 */ 3756 if (vn_has_cached_data(vp)) { 3757 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 3758 B_INVAL, cr); 3759 } 3760 3761 mutex_enter(&zp->z_lock); 3762 vp->v_count = 0; /* count arrives as 1 */ 3763 mutex_exit(&zp->z_lock); 3764 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3765 zfs_znode_free(zp); 3766 return; 3767 } 3768 3769 /* 3770 * Attempt to push any data in the page cache. If this fails 3771 * we will get kicked out later in zfs_zinactive(). 3772 */ 3773 if (vn_has_cached_data(vp)) { 3774 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 3775 cr); 3776 } 3777 3778 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 3779 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 3780 3781 dmu_tx_hold_bonus(tx, zp->z_id); 3782 error = dmu_tx_assign(tx, TXG_WAIT); 3783 if (error) { 3784 dmu_tx_abort(tx); 3785 } else { 3786 dmu_buf_will_dirty(zp->z_dbuf, tx); 3787 mutex_enter(&zp->z_lock); 3788 zp->z_atime_dirty = 0; 3789 mutex_exit(&zp->z_lock); 3790 dmu_tx_commit(tx); 3791 } 3792 } 3793 3794 zfs_zinactive(zp); 3795 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3796 } 3797 3798 /* 3799 * Bounds-check the seek operation. 3800 * 3801 * IN: vp - vnode seeking within 3802 * ooff - old file offset 3803 * noffp - pointer to new file offset 3804 * ct - caller context 3805 * 3806 * RETURN: 0 if success 3807 * EINVAL if new offset invalid 3808 */ 3809 /* ARGSUSED */ 3810 static int 3811 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 3812 caller_context_t *ct) 3813 { 3814 if (vp->v_type == VDIR) 3815 return (0); 3816 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 3817 } 3818 3819 /* 3820 * Pre-filter the generic locking function to trap attempts to place 3821 * a mandatory lock on a memory mapped file. 3822 */ 3823 static int 3824 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 3825 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 3826 { 3827 znode_t *zp = VTOZ(vp); 3828 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3829 int error; 3830 3831 ZFS_ENTER(zfsvfs); 3832 ZFS_VERIFY_ZP(zp); 3833 3834 /* 3835 * We are following the UFS semantics with respect to mapcnt 3836 * here: If we see that the file is mapped already, then we will 3837 * return an error, but we don't worry about races between this 3838 * function and zfs_map(). 3839 */ 3840 if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) { 3841 ZFS_EXIT(zfsvfs); 3842 return (EAGAIN); 3843 } 3844 error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct); 3845 ZFS_EXIT(zfsvfs); 3846 return (error); 3847 } 3848 3849 /* 3850 * If we can't find a page in the cache, we will create a new page 3851 * and fill it with file data. For efficiency, we may try to fill 3852 * multiple pages at once (klustering). 3853 */ 3854 static int 3855 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 3856 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 3857 { 3858 znode_t *zp = VTOZ(vp); 3859 page_t *pp, *cur_pp; 3860 objset_t *os = zp->z_zfsvfs->z_os; 3861 caddr_t va; 3862 u_offset_t io_off, total; 3863 uint64_t oid = zp->z_id; 3864 size_t io_len; 3865 uint64_t filesz; 3866 int err; 3867 3868 /* 3869 * If we are only asking for a single page don't bother klustering. 3870 */ 3871 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3872 if (off >= filesz) 3873 return (EFAULT); 3874 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 3875 io_off = off; 3876 io_len = PAGESIZE; 3877 pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr); 3878 } else { 3879 /* 3880 * Try to fill a kluster of pages (a blocks worth). 3881 */ 3882 size_t klen; 3883 u_offset_t koff; 3884 3885 if (!ISP2(zp->z_blksz)) { 3886 /* Only one block in the file. */ 3887 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3888 koff = 0; 3889 } else { 3890 /* 3891 * It would be ideal to align our offset to the 3892 * blocksize but doing so has resulted in some 3893 * strange application crashes. For now, we 3894 * leave the offset as is and only adjust the 3895 * length if we are off the end of the file. 3896 */ 3897 koff = off; 3898 klen = plsz; 3899 } 3900 ASSERT(koff <= filesz); 3901 if (koff + klen > filesz) 3902 klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff; 3903 ASSERT3U(off, >=, koff); 3904 ASSERT3U(off, <, koff + klen); 3905 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 3906 &io_len, koff, klen, 0); 3907 } 3908 if (pp == NULL) { 3909 /* 3910 * Some other thread entered the page before us. 3911 * Return to zfs_getpage to retry the lookup. 3912 */ 3913 *pl = NULL; 3914 return (0); 3915 } 3916 3917 /* 3918 * Fill the pages in the kluster. 3919 */ 3920 cur_pp = pp; 3921 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 3922 ASSERT3U(io_off, ==, cur_pp->p_offset); 3923 va = zfs_map_page(cur_pp, S_WRITE); 3924 err = dmu_read(os, oid, io_off, PAGESIZE, va); 3925 zfs_unmap_page(cur_pp, va); 3926 if (err) { 3927 /* On error, toss the entire kluster */ 3928 pvn_read_done(pp, B_ERROR); 3929 /* convert checksum errors into IO errors */ 3930 if (err == ECKSUM) 3931 err = EIO; 3932 return (err); 3933 } 3934 cur_pp = cur_pp->p_next; 3935 } 3936 out: 3937 /* 3938 * Fill in the page list array from the kluster. If 3939 * there are too many pages in the kluster, return 3940 * as many pages as possible starting from the desired 3941 * offset `off'. 3942 * NOTE: the page list will always be null terminated. 3943 */ 3944 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 3945 3946 return (0); 3947 } 3948 3949 /* 3950 * Return pointers to the pages for the file region [off, off + len] 3951 * in the pl array. If plsz is greater than len, this function may 3952 * also return page pointers from before or after the specified 3953 * region (i.e. some region [off', off' + plsz]). These additional 3954 * pages are only returned if they are already in the cache, or were 3955 * created as part of a klustered read. 3956 * 3957 * IN: vp - vnode of file to get data from. 3958 * off - position in file to get data from. 3959 * len - amount of data to retrieve. 3960 * plsz - length of provided page list. 3961 * seg - segment to obtain pages for. 3962 * addr - virtual address of fault. 3963 * rw - mode of created pages. 3964 * cr - credentials of caller. 3965 * ct - caller context. 3966 * 3967 * OUT: protp - protection mode of created pages. 3968 * pl - list of pages created. 3969 * 3970 * RETURN: 0 if success 3971 * error code if failure 3972 * 3973 * Timestamps: 3974 * vp - atime updated 3975 */ 3976 /* ARGSUSED */ 3977 static int 3978 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 3979 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 3980 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 3981 { 3982 znode_t *zp = VTOZ(vp); 3983 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3984 page_t *pp, **pl0 = pl; 3985 int need_unlock = 0, err = 0; 3986 offset_t orig_off; 3987 3988 ZFS_ENTER(zfsvfs); 3989 ZFS_VERIFY_ZP(zp); 3990 3991 if (protp) 3992 *protp = PROT_ALL; 3993 3994 /* no faultahead (for now) */ 3995 if (pl == NULL) { 3996 ZFS_EXIT(zfsvfs); 3997 return (0); 3998 } 3999 4000 /* can't fault past EOF */ 4001 if (off >= zp->z_phys->zp_size) { 4002 ZFS_EXIT(zfsvfs); 4003 return (EFAULT); 4004 } 4005 orig_off = off; 4006 4007 /* 4008 * If we already own the lock, then we must be page faulting 4009 * in the middle of a write to this file (i.e., we are writing 4010 * to this file using data from a mapped region of the file). 4011 */ 4012 if (rw_owner(&zp->z_map_lock) != curthread) { 4013 rw_enter(&zp->z_map_lock, RW_WRITER); 4014 need_unlock = TRUE; 4015 } 4016 4017 /* 4018 * Loop through the requested range [off, off + len] looking 4019 * for pages. If we don't find a page, we will need to create 4020 * a new page and fill it with data from the file. 4021 */ 4022 while (len > 0) { 4023 if (plsz < PAGESIZE) 4024 break; 4025 if (pp = page_lookup(vp, off, SE_SHARED)) { 4026 *pl++ = pp; 4027 off += PAGESIZE; 4028 addr += PAGESIZE; 4029 len -= PAGESIZE; 4030 plsz -= PAGESIZE; 4031 } else { 4032 err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw); 4033 if (err) 4034 goto out; 4035 /* 4036 * klustering may have changed our region 4037 * to be block aligned. 4038 */ 4039 if (((pp = *pl) != 0) && (off != pp->p_offset)) { 4040 int delta = off - pp->p_offset; 4041 len += delta; 4042 off -= delta; 4043 addr -= delta; 4044 } 4045 while (*pl) { 4046 pl++; 4047 off += PAGESIZE; 4048 addr += PAGESIZE; 4049 plsz -= PAGESIZE; 4050 if (len > PAGESIZE) 4051 len -= PAGESIZE; 4052 else 4053 len = 0; 4054 } 4055 } 4056 } 4057 4058 /* 4059 * Fill out the page array with any pages already in the cache. 4060 */ 4061 while (plsz > 0) { 4062 pp = page_lookup_nowait(vp, off, SE_SHARED); 4063 if (pp == NULL) 4064 break; 4065 *pl++ = pp; 4066 off += PAGESIZE; 4067 plsz -= PAGESIZE; 4068 } 4069 4070 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4071 out: 4072 /* 4073 * We can't grab the range lock for the page as reader which would 4074 * stop truncation as this leads to deadlock. So we need to recheck 4075 * the file size. 4076 */ 4077 if (orig_off >= zp->z_phys->zp_size) 4078 err = EFAULT; 4079 if (err) { 4080 /* 4081 * Release any pages we have previously locked. 4082 */ 4083 while (pl > pl0) 4084 page_unlock(*--pl); 4085 } 4086 4087 *pl = NULL; 4088 4089 if (need_unlock) 4090 rw_exit(&zp->z_map_lock); 4091 4092 ZFS_EXIT(zfsvfs); 4093 return (err); 4094 } 4095 4096 /* 4097 * Request a memory map for a section of a file. This code interacts 4098 * with common code and the VM system as follows: 4099 * 4100 * common code calls mmap(), which ends up in smmap_common() 4101 * 4102 * this calls VOP_MAP(), which takes you into (say) zfs 4103 * 4104 * zfs_map() calls as_map(), passing segvn_create() as the callback 4105 * 4106 * segvn_create() creates the new segment and calls VOP_ADDMAP() 4107 * 4108 * zfs_addmap() updates z_mapcnt 4109 */ 4110 /*ARGSUSED*/ 4111 static int 4112 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4113 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4114 caller_context_t *ct) 4115 { 4116 znode_t *zp = VTOZ(vp); 4117 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4118 segvn_crargs_t vn_a; 4119 int error; 4120 4121 ZFS_ENTER(zfsvfs); 4122 ZFS_VERIFY_ZP(zp); 4123 4124 if ((prot & PROT_WRITE) && 4125 (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_READONLY | 4126 ZFS_APPENDONLY))) { 4127 ZFS_EXIT(zfsvfs); 4128 return (EPERM); 4129 } 4130 4131 if ((prot & (PROT_READ | PROT_EXEC)) && 4132 (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED)) { 4133 ZFS_EXIT(zfsvfs); 4134 return (EACCES); 4135 } 4136 4137 if (vp->v_flag & VNOMAP) { 4138 ZFS_EXIT(zfsvfs); 4139 return (ENOSYS); 4140 } 4141 4142 if (off < 0 || len > MAXOFFSET_T - off) { 4143 ZFS_EXIT(zfsvfs); 4144 return (ENXIO); 4145 } 4146 4147 if (vp->v_type != VREG) { 4148 ZFS_EXIT(zfsvfs); 4149 return (ENODEV); 4150 } 4151 4152 /* 4153 * If file is locked, disallow mapping. 4154 */ 4155 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) { 4156 ZFS_EXIT(zfsvfs); 4157 return (EAGAIN); 4158 } 4159 4160 as_rangelock(as); 4161 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 4162 if (error != 0) { 4163 as_rangeunlock(as); 4164 ZFS_EXIT(zfsvfs); 4165 return (error); 4166 } 4167 4168 vn_a.vp = vp; 4169 vn_a.offset = (u_offset_t)off; 4170 vn_a.type = flags & MAP_TYPE; 4171 vn_a.prot = prot; 4172 vn_a.maxprot = maxprot; 4173 vn_a.cred = cr; 4174 vn_a.amp = NULL; 4175 vn_a.flags = flags & ~MAP_TYPE; 4176 vn_a.szc = 0; 4177 vn_a.lgrp_mem_policy_flags = 0; 4178 4179 error = as_map(as, *addrp, len, segvn_create, &vn_a); 4180 4181 as_rangeunlock(as); 4182 ZFS_EXIT(zfsvfs); 4183 return (error); 4184 } 4185 4186 /* ARGSUSED */ 4187 static int 4188 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4189 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4190 caller_context_t *ct) 4191 { 4192 uint64_t pages = btopr(len); 4193 4194 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 4195 return (0); 4196 } 4197 4198 /* 4199 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 4200 * more accurate mtime for the associated file. Since we don't have a way of 4201 * detecting when the data was actually modified, we have to resort to 4202 * heuristics. If an explicit msync() is done, then we mark the mtime when the 4203 * last page is pushed. The problem occurs when the msync() call is omitted, 4204 * which by far the most common case: 4205 * 4206 * open() 4207 * mmap() 4208 * <modify memory> 4209 * munmap() 4210 * close() 4211 * <time lapse> 4212 * putpage() via fsflush 4213 * 4214 * If we wait until fsflush to come along, we can have a modification time that 4215 * is some arbitrary point in the future. In order to prevent this in the 4216 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 4217 * torn down. 4218 */ 4219 /* ARGSUSED */ 4220 static int 4221 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4222 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 4223 caller_context_t *ct) 4224 { 4225 uint64_t pages = btopr(len); 4226 4227 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 4228 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 4229 4230 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 4231 vn_has_cached_data(vp)) 4232 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 4233 4234 return (0); 4235 } 4236 4237 /* 4238 * Free or allocate space in a file. Currently, this function only 4239 * supports the `F_FREESP' command. However, this command is somewhat 4240 * misnamed, as its functionality includes the ability to allocate as 4241 * well as free space. 4242 * 4243 * IN: vp - vnode of file to free data in. 4244 * cmd - action to take (only F_FREESP supported). 4245 * bfp - section of file to free/alloc. 4246 * flag - current file open mode flags. 4247 * offset - current file offset. 4248 * cr - credentials of caller [UNUSED]. 4249 * ct - caller context. 4250 * 4251 * RETURN: 0 if success 4252 * error code if failure 4253 * 4254 * Timestamps: 4255 * vp - ctime|mtime updated 4256 */ 4257 /* ARGSUSED */ 4258 static int 4259 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 4260 offset_t offset, cred_t *cr, caller_context_t *ct) 4261 { 4262 znode_t *zp = VTOZ(vp); 4263 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4264 uint64_t off, len; 4265 int error; 4266 4267 ZFS_ENTER(zfsvfs); 4268 ZFS_VERIFY_ZP(zp); 4269 4270 if (cmd != F_FREESP) { 4271 ZFS_EXIT(zfsvfs); 4272 return (EINVAL); 4273 } 4274 4275 if (error = convoff(vp, bfp, 0, offset)) { 4276 ZFS_EXIT(zfsvfs); 4277 return (error); 4278 } 4279 4280 if (bfp->l_len < 0) { 4281 ZFS_EXIT(zfsvfs); 4282 return (EINVAL); 4283 } 4284 4285 off = bfp->l_start; 4286 len = bfp->l_len; /* 0 means from off to end of file */ 4287 4288 error = zfs_freesp(zp, off, len, flag, TRUE); 4289 4290 ZFS_EXIT(zfsvfs); 4291 return (error); 4292 } 4293 4294 /*ARGSUSED*/ 4295 static int 4296 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 4297 { 4298 znode_t *zp = VTOZ(vp); 4299 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4300 uint32_t gen; 4301 uint64_t object = zp->z_id; 4302 zfid_short_t *zfid; 4303 int size, i; 4304 4305 ZFS_ENTER(zfsvfs); 4306 ZFS_VERIFY_ZP(zp); 4307 gen = (uint32_t)zp->z_gen; 4308 4309 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 4310 if (fidp->fid_len < size) { 4311 fidp->fid_len = size; 4312 ZFS_EXIT(zfsvfs); 4313 return (ENOSPC); 4314 } 4315 4316 zfid = (zfid_short_t *)fidp; 4317 4318 zfid->zf_len = size; 4319 4320 for (i = 0; i < sizeof (zfid->zf_object); i++) 4321 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 4322 4323 /* Must have a non-zero generation number to distinguish from .zfs */ 4324 if (gen == 0) 4325 gen = 1; 4326 for (i = 0; i < sizeof (zfid->zf_gen); i++) 4327 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 4328 4329 if (size == LONG_FID_LEN) { 4330 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 4331 zfid_long_t *zlfid; 4332 4333 zlfid = (zfid_long_t *)fidp; 4334 4335 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 4336 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 4337 4338 /* XXX - this should be the generation number for the objset */ 4339 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 4340 zlfid->zf_setgen[i] = 0; 4341 } 4342 4343 ZFS_EXIT(zfsvfs); 4344 return (0); 4345 } 4346 4347 static int 4348 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 4349 caller_context_t *ct) 4350 { 4351 znode_t *zp, *xzp; 4352 zfsvfs_t *zfsvfs; 4353 zfs_dirlock_t *dl; 4354 int error; 4355 4356 switch (cmd) { 4357 case _PC_LINK_MAX: 4358 *valp = ULONG_MAX; 4359 return (0); 4360 4361 case _PC_FILESIZEBITS: 4362 *valp = 64; 4363 return (0); 4364 4365 case _PC_XATTR_EXISTS: 4366 zp = VTOZ(vp); 4367 zfsvfs = zp->z_zfsvfs; 4368 ZFS_ENTER(zfsvfs); 4369 ZFS_VERIFY_ZP(zp); 4370 *valp = 0; 4371 error = zfs_dirent_lock(&dl, zp, "", &xzp, 4372 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 4373 if (error == 0) { 4374 zfs_dirent_unlock(dl); 4375 if (!zfs_dirempty(xzp)) 4376 *valp = 1; 4377 VN_RELE(ZTOV(xzp)); 4378 } else if (error == ENOENT) { 4379 /* 4380 * If there aren't extended attributes, it's the 4381 * same as having zero of them. 4382 */ 4383 error = 0; 4384 } 4385 ZFS_EXIT(zfsvfs); 4386 return (error); 4387 4388 case _PC_SATTR_ENABLED: 4389 case _PC_SATTR_EXISTS: 4390 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 4391 (vp->v_type == VREG || vp->v_type == VDIR); 4392 return (0); 4393 4394 case _PC_ACL_ENABLED: 4395 *valp = _ACL_ACE_ENABLED; 4396 return (0); 4397 4398 case _PC_MIN_HOLE_SIZE: 4399 *valp = (ulong_t)SPA_MINBLOCKSIZE; 4400 return (0); 4401 4402 default: 4403 return (fs_pathconf(vp, cmd, valp, cr, ct)); 4404 } 4405 } 4406 4407 /*ARGSUSED*/ 4408 static int 4409 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4410 caller_context_t *ct) 4411 { 4412 znode_t *zp = VTOZ(vp); 4413 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4414 int error; 4415 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4416 4417 ZFS_ENTER(zfsvfs); 4418 ZFS_VERIFY_ZP(zp); 4419 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 4420 ZFS_EXIT(zfsvfs); 4421 4422 return (error); 4423 } 4424 4425 /*ARGSUSED*/ 4426 static int 4427 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4428 caller_context_t *ct) 4429 { 4430 znode_t *zp = VTOZ(vp); 4431 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4432 int error; 4433 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4434 4435 ZFS_ENTER(zfsvfs); 4436 ZFS_VERIFY_ZP(zp); 4437 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 4438 ZFS_EXIT(zfsvfs); 4439 return (error); 4440 } 4441 4442 /* 4443 * Predeclare these here so that the compiler assumes that 4444 * this is an "old style" function declaration that does 4445 * not include arguments => we won't get type mismatch errors 4446 * in the initializations that follow. 4447 */ 4448 static int zfs_inval(); 4449 static int zfs_isdir(); 4450 4451 static int 4452 zfs_inval() 4453 { 4454 return (EINVAL); 4455 } 4456 4457 static int 4458 zfs_isdir() 4459 { 4460 return (EISDIR); 4461 } 4462 /* 4463 * Directory vnode operations template 4464 */ 4465 vnodeops_t *zfs_dvnodeops; 4466 const fs_operation_def_t zfs_dvnodeops_template[] = { 4467 VOPNAME_OPEN, { .vop_open = zfs_open }, 4468 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4469 VOPNAME_READ, { .error = zfs_isdir }, 4470 VOPNAME_WRITE, { .error = zfs_isdir }, 4471 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4472 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4473 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4474 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4475 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4476 VOPNAME_CREATE, { .vop_create = zfs_create }, 4477 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 4478 VOPNAME_LINK, { .vop_link = zfs_link }, 4479 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4480 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 4481 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 4482 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 4483 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 4484 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4485 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4486 VOPNAME_FID, { .vop_fid = zfs_fid }, 4487 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4488 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4489 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4490 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4491 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4492 NULL, NULL 4493 }; 4494 4495 /* 4496 * Regular file vnode operations template 4497 */ 4498 vnodeops_t *zfs_fvnodeops; 4499 const fs_operation_def_t zfs_fvnodeops_template[] = { 4500 VOPNAME_OPEN, { .vop_open = zfs_open }, 4501 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4502 VOPNAME_READ, { .vop_read = zfs_read }, 4503 VOPNAME_WRITE, { .vop_write = zfs_write }, 4504 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4505 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4506 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4507 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4508 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4509 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4510 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4511 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4512 VOPNAME_FID, { .vop_fid = zfs_fid }, 4513 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4514 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 4515 VOPNAME_SPACE, { .vop_space = zfs_space }, 4516 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 4517 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 4518 VOPNAME_MAP, { .vop_map = zfs_map }, 4519 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 4520 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 4521 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4522 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4523 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4524 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4525 NULL, NULL 4526 }; 4527 4528 /* 4529 * Symbolic link vnode operations template 4530 */ 4531 vnodeops_t *zfs_symvnodeops; 4532 const fs_operation_def_t zfs_symvnodeops_template[] = { 4533 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4534 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4535 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4536 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4537 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 4538 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4539 VOPNAME_FID, { .vop_fid = zfs_fid }, 4540 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4541 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4542 NULL, NULL 4543 }; 4544 4545 /* 4546 * Extended attribute directory vnode operations template 4547 * This template is identical to the directory vnodes 4548 * operation template except for restricted operations: 4549 * VOP_MKDIR() 4550 * VOP_SYMLINK() 4551 * Note that there are other restrictions embedded in: 4552 * zfs_create() - restrict type to VREG 4553 * zfs_link() - no links into/out of attribute space 4554 * zfs_rename() - no moves into/out of attribute space 4555 */ 4556 vnodeops_t *zfs_xdvnodeops; 4557 const fs_operation_def_t zfs_xdvnodeops_template[] = { 4558 VOPNAME_OPEN, { .vop_open = zfs_open }, 4559 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4560 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4561 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4562 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4563 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4564 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4565 VOPNAME_CREATE, { .vop_create = zfs_create }, 4566 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 4567 VOPNAME_LINK, { .vop_link = zfs_link }, 4568 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4569 VOPNAME_MKDIR, { .error = zfs_inval }, 4570 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 4571 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 4572 VOPNAME_SYMLINK, { .error = zfs_inval }, 4573 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4574 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4575 VOPNAME_FID, { .vop_fid = zfs_fid }, 4576 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4577 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4578 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4579 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4580 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4581 NULL, NULL 4582 }; 4583 4584 /* 4585 * Error vnode operations template 4586 */ 4587 vnodeops_t *zfs_evnodeops; 4588 const fs_operation_def_t zfs_evnodeops_template[] = { 4589 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4590 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4591 NULL, NULL 4592 }; 4593