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 2009 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 <sys/sid.h> 69 #include "fs/fs_subr.h" 70 #include <sys/zfs_ctldir.h> 71 #include <sys/zfs_fuid.h> 72 #include <sys/dnlc.h> 73 #include <sys/zfs_rlock.h> 74 #include <sys/extdirent.h> 75 #include <sys/kidmap.h> 76 #include <sys/cred_impl.h> 77 #include <sys/attr.h> 78 79 /* 80 * Programming rules. 81 * 82 * Each vnode op performs some logical unit of work. To do this, the ZPL must 83 * properly lock its in-core state, create a DMU transaction, do the work, 84 * record this work in the intent log (ZIL), commit the DMU transaction, 85 * and wait for the intent log to commit if it is a synchronous operation. 86 * Moreover, the vnode ops must work in both normal and log replay context. 87 * The ordering of events is important to avoid deadlocks and references 88 * to freed memory. The example below illustrates the following Big Rules: 89 * 90 * (1) A check must be made in each zfs thread for a mounted file system. 91 * This is done avoiding races using ZFS_ENTER(zfsvfs). 92 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes 93 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros 94 * can return EIO from the calling function. 95 * 96 * (2) VN_RELE() should always be the last thing except for zil_commit() 97 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 98 * First, if it's the last reference, the vnode/znode 99 * can be freed, so the zp may point to freed memory. Second, the last 100 * reference will call zfs_zinactive(), which may induce a lot of work -- 101 * pushing cached pages (which acquires range locks) and syncing out 102 * cached atime changes. Third, zfs_zinactive() may require a new tx, 103 * which could deadlock the system if you were already holding one. 104 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). 105 * 106 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 107 * as they can span dmu_tx_assign() calls. 108 * 109 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign(). 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 * During ZIL replay the zfs_log_* functions will update the sequence 127 * number to indicate the zil transaction has replayed. 128 * 129 * (6) At the end of each vnode op, the DMU tx must always commit, 130 * regardless of whether there were any errors. 131 * 132 * (7) After dropping all locks, invoke zil_commit(zilog, seq, foid) 133 * to ensure that synchronous semantics are provided when necessary. 134 * 135 * In general, this is how things should be ordered in each vnode op: 136 * 137 * ZFS_ENTER(zfsvfs); // exit if unmounted 138 * top: 139 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 140 * rw_enter(...); // grab any other locks you need 141 * tx = dmu_tx_create(...); // get DMU tx 142 * dmu_tx_hold_*(); // hold each object you might modify 143 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign 144 * if (error) { 145 * rw_exit(...); // drop locks 146 * zfs_dirent_unlock(dl); // unlock directory entry 147 * VN_RELE(...); // release held vnodes 148 * if (error == ERESTART) { 149 * dmu_tx_wait(tx); 150 * dmu_tx_abort(tx); 151 * goto top; 152 * } 153 * dmu_tx_abort(tx); // abort DMU tx 154 * ZFS_EXIT(zfsvfs); // finished in zfs 155 * return (error); // really out of space 156 * } 157 * error = do_real_work(); // do whatever this VOP does 158 * if (error == 0) 159 * zfs_log_*(...); // on success, make ZIL entry 160 * dmu_tx_commit(tx); // commit DMU tx -- error or not 161 * rw_exit(...); // drop locks 162 * zfs_dirent_unlock(dl); // unlock directory entry 163 * VN_RELE(...); // release held vnodes 164 * zil_commit(zilog, seq, foid); // synchronous when necessary 165 * ZFS_EXIT(zfsvfs); // finished in zfs 166 * return (error); // done, report error 167 */ 168 169 /* ARGSUSED */ 170 static int 171 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 172 { 173 znode_t *zp = VTOZ(*vpp); 174 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 175 176 ZFS_ENTER(zfsvfs); 177 ZFS_VERIFY_ZP(zp); 178 179 if ((flag & FWRITE) && (zp->z_phys->zp_flags & ZFS_APPENDONLY) && 180 ((flag & FAPPEND) == 0)) { 181 ZFS_EXIT(zfsvfs); 182 return (EPERM); 183 } 184 185 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 186 ZTOV(zp)->v_type == VREG && 187 !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) && 188 zp->z_phys->zp_size > 0) { 189 if (fs_vscan(*vpp, cr, 0) != 0) { 190 ZFS_EXIT(zfsvfs); 191 return (EACCES); 192 } 193 } 194 195 /* Keep a count of the synchronous opens in the znode */ 196 if (flag & (FSYNC | FDSYNC)) 197 atomic_inc_32(&zp->z_sync_cnt); 198 199 ZFS_EXIT(zfsvfs); 200 return (0); 201 } 202 203 /* ARGSUSED */ 204 static int 205 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 206 caller_context_t *ct) 207 { 208 znode_t *zp = VTOZ(vp); 209 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 210 211 /* 212 * Clean up any locks held by this process on the vp. 213 */ 214 cleanlocks(vp, ddi_get_pid(), 0); 215 cleanshares(vp, ddi_get_pid()); 216 217 ZFS_ENTER(zfsvfs); 218 ZFS_VERIFY_ZP(zp); 219 220 /* Decrement the synchronous opens in the znode */ 221 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 222 atomic_dec_32(&zp->z_sync_cnt); 223 224 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 225 ZTOV(zp)->v_type == VREG && 226 !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) && 227 zp->z_phys->zp_size > 0) 228 VERIFY(fs_vscan(vp, cr, 1) == 0); 229 230 ZFS_EXIT(zfsvfs); 231 return (0); 232 } 233 234 /* 235 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 236 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 237 */ 238 static int 239 zfs_holey(vnode_t *vp, int cmd, offset_t *off) 240 { 241 znode_t *zp = VTOZ(vp); 242 uint64_t noff = (uint64_t)*off; /* new offset */ 243 uint64_t file_sz; 244 int error; 245 boolean_t hole; 246 247 file_sz = zp->z_phys->zp_size; 248 if (noff >= file_sz) { 249 return (ENXIO); 250 } 251 252 if (cmd == _FIO_SEEK_HOLE) 253 hole = B_TRUE; 254 else 255 hole = B_FALSE; 256 257 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 258 259 /* end of file? */ 260 if ((error == ESRCH) || (noff > file_sz)) { 261 /* 262 * Handle the virtual hole at the end of file. 263 */ 264 if (hole) { 265 *off = file_sz; 266 return (0); 267 } 268 return (ENXIO); 269 } 270 271 if (noff < *off) 272 return (error); 273 *off = noff; 274 return (error); 275 } 276 277 /* ARGSUSED */ 278 static int 279 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred, 280 int *rvalp, caller_context_t *ct) 281 { 282 offset_t off; 283 int error; 284 zfsvfs_t *zfsvfs; 285 znode_t *zp; 286 287 switch (com) { 288 case _FIOFFS: 289 return (zfs_sync(vp->v_vfsp, 0, cred)); 290 291 /* 292 * The following two ioctls are used by bfu. Faking out, 293 * necessary to avoid bfu errors. 294 */ 295 case _FIOGDIO: 296 case _FIOSDIO: 297 return (0); 298 299 case _FIO_SEEK_DATA: 300 case _FIO_SEEK_HOLE: 301 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 302 return (EFAULT); 303 304 zp = VTOZ(vp); 305 zfsvfs = zp->z_zfsvfs; 306 ZFS_ENTER(zfsvfs); 307 ZFS_VERIFY_ZP(zp); 308 309 /* offset parameter is in/out */ 310 error = zfs_holey(vp, com, &off); 311 ZFS_EXIT(zfsvfs); 312 if (error) 313 return (error); 314 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 315 return (EFAULT); 316 return (0); 317 } 318 return (ENOTTY); 319 } 320 321 /* 322 * Utility functions to map and unmap a single physical page. These 323 * are used to manage the mappable copies of ZFS file data, and therefore 324 * do not update ref/mod bits. 325 */ 326 caddr_t 327 zfs_map_page(page_t *pp, enum seg_rw rw) 328 { 329 if (kpm_enable) 330 return (hat_kpm_mapin(pp, 0)); 331 ASSERT(rw == S_READ || rw == S_WRITE); 332 return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0), 333 (caddr_t)-1)); 334 } 335 336 void 337 zfs_unmap_page(page_t *pp, caddr_t addr) 338 { 339 if (kpm_enable) { 340 hat_kpm_mapout(pp, 0, addr); 341 } else { 342 ppmapout(addr); 343 } 344 } 345 346 /* 347 * When a file is memory mapped, we must keep the IO data synchronized 348 * between the DMU cache and the memory mapped pages. What this means: 349 * 350 * On Write: If we find a memory mapped page, we write to *both* 351 * the page and the dmu buffer. 352 */ 353 static void 354 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid) 355 { 356 int64_t off; 357 358 off = start & PAGEOFFSET; 359 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 360 page_t *pp; 361 uint64_t nbytes = MIN(PAGESIZE - off, len); 362 363 if (pp = page_lookup(vp, start, SE_SHARED)) { 364 caddr_t va; 365 366 va = zfs_map_page(pp, S_WRITE); 367 (void) dmu_read(os, oid, start+off, nbytes, va+off, 368 DMU_READ_PREFETCH); 369 zfs_unmap_page(pp, va); 370 page_unlock(pp); 371 } 372 len -= nbytes; 373 off = 0; 374 } 375 } 376 377 /* 378 * When a file is memory mapped, we must keep the IO data synchronized 379 * between the DMU cache and the memory mapped pages. What this means: 380 * 381 * On Read: We "read" preferentially from memory mapped pages, 382 * else we default from the dmu buffer. 383 * 384 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 385 * the file is memory mapped. 386 */ 387 static int 388 mappedread(vnode_t *vp, int nbytes, uio_t *uio) 389 { 390 znode_t *zp = VTOZ(vp); 391 objset_t *os = zp->z_zfsvfs->z_os; 392 int64_t start, off; 393 int len = nbytes; 394 int error = 0; 395 396 start = uio->uio_loffset; 397 off = start & PAGEOFFSET; 398 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 399 page_t *pp; 400 uint64_t bytes = MIN(PAGESIZE - off, len); 401 402 if (pp = page_lookup(vp, start, SE_SHARED)) { 403 caddr_t va; 404 405 va = zfs_map_page(pp, S_READ); 406 error = uiomove(va + off, bytes, UIO_READ, uio); 407 zfs_unmap_page(pp, va); 408 page_unlock(pp); 409 } else { 410 error = dmu_read_uio(os, zp->z_id, uio, bytes); 411 } 412 len -= bytes; 413 off = 0; 414 if (error) 415 break; 416 } 417 return (error); 418 } 419 420 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 421 422 /* 423 * Read bytes from specified file into supplied buffer. 424 * 425 * IN: vp - vnode of file to be read from. 426 * uio - structure supplying read location, range info, 427 * and return buffer. 428 * ioflag - SYNC flags; used to provide FRSYNC semantics. 429 * cr - credentials of caller. 430 * ct - caller context 431 * 432 * OUT: uio - updated offset and range, buffer filled. 433 * 434 * RETURN: 0 if success 435 * error code if failure 436 * 437 * Side Effects: 438 * vp - atime updated if byte count > 0 439 */ 440 /* ARGSUSED */ 441 static int 442 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 443 { 444 znode_t *zp = VTOZ(vp); 445 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 446 objset_t *os; 447 ssize_t n, nbytes; 448 int error; 449 rl_t *rl; 450 451 ZFS_ENTER(zfsvfs); 452 ZFS_VERIFY_ZP(zp); 453 os = zfsvfs->z_os; 454 455 if (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) { 456 ZFS_EXIT(zfsvfs); 457 return (EACCES); 458 } 459 460 /* 461 * Validate file offset 462 */ 463 if (uio->uio_loffset < (offset_t)0) { 464 ZFS_EXIT(zfsvfs); 465 return (EINVAL); 466 } 467 468 /* 469 * Fasttrack empty reads 470 */ 471 if (uio->uio_resid == 0) { 472 ZFS_EXIT(zfsvfs); 473 return (0); 474 } 475 476 /* 477 * Check for mandatory locks 478 */ 479 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 480 if (error = chklock(vp, FREAD, 481 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 482 ZFS_EXIT(zfsvfs); 483 return (error); 484 } 485 } 486 487 /* 488 * If we're in FRSYNC mode, sync out this znode before reading it. 489 */ 490 if (ioflag & FRSYNC) 491 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 492 493 /* 494 * Lock the range against changes. 495 */ 496 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 497 498 /* 499 * If we are reading past end-of-file we can skip 500 * to the end; but we might still need to set atime. 501 */ 502 if (uio->uio_loffset >= zp->z_phys->zp_size) { 503 error = 0; 504 goto out; 505 } 506 507 ASSERT(uio->uio_loffset < zp->z_phys->zp_size); 508 n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset); 509 510 while (n > 0) { 511 nbytes = MIN(n, zfs_read_chunk_size - 512 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 513 514 if (vn_has_cached_data(vp)) 515 error = mappedread(vp, nbytes, uio); 516 else 517 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 518 if (error) { 519 /* convert checksum errors into IO errors */ 520 if (error == ECKSUM) 521 error = EIO; 522 break; 523 } 524 525 n -= nbytes; 526 } 527 528 out: 529 zfs_range_unlock(rl); 530 531 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 532 ZFS_EXIT(zfsvfs); 533 return (error); 534 } 535 536 /* 537 * Write the bytes to a file. 538 * 539 * IN: vp - vnode of file to be written to. 540 * uio - structure supplying write location, range info, 541 * and data buffer. 542 * ioflag - FAPPEND flag set if in append mode. 543 * cr - credentials of caller. 544 * ct - caller context (NFS/CIFS fem monitor only) 545 * 546 * OUT: uio - updated offset and range. 547 * 548 * RETURN: 0 if success 549 * error code if failure 550 * 551 * Timestamps: 552 * vp - ctime|mtime updated if byte count > 0 553 */ 554 /* ARGSUSED */ 555 static int 556 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 557 { 558 znode_t *zp = VTOZ(vp); 559 rlim64_t limit = uio->uio_llimit; 560 ssize_t start_resid = uio->uio_resid; 561 ssize_t tx_bytes; 562 uint64_t end_size; 563 dmu_tx_t *tx; 564 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 565 zilog_t *zilog; 566 offset_t woff; 567 ssize_t n, nbytes; 568 rl_t *rl; 569 int max_blksz = zfsvfs->z_max_blksz; 570 uint64_t pflags; 571 int error; 572 arc_buf_t *abuf; 573 574 /* 575 * Fasttrack empty write 576 */ 577 n = start_resid; 578 if (n == 0) 579 return (0); 580 581 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 582 limit = MAXOFFSET_T; 583 584 ZFS_ENTER(zfsvfs); 585 ZFS_VERIFY_ZP(zp); 586 587 /* 588 * If immutable or not appending then return EPERM 589 */ 590 pflags = zp->z_phys->zp_flags; 591 if ((pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 592 ((pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 593 (uio->uio_loffset < zp->z_phys->zp_size))) { 594 ZFS_EXIT(zfsvfs); 595 return (EPERM); 596 } 597 598 zilog = zfsvfs->z_log; 599 600 /* 601 * Pre-fault the pages to ensure slow (eg NFS) pages 602 * don't hold up txg. 603 */ 604 uio_prefaultpages(n, uio); 605 606 /* 607 * If in append mode, set the io offset pointer to eof. 608 */ 609 if (ioflag & FAPPEND) { 610 /* 611 * Range lock for a file append: 612 * The value for the start of range will be determined by 613 * zfs_range_lock() (to guarantee append semantics). 614 * If this write will cause the block size to increase, 615 * zfs_range_lock() will lock the entire file, so we must 616 * later reduce the range after we grow the block size. 617 */ 618 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 619 if (rl->r_len == UINT64_MAX) { 620 /* overlocked, zp_size can't change */ 621 woff = uio->uio_loffset = zp->z_phys->zp_size; 622 } else { 623 woff = uio->uio_loffset = rl->r_off; 624 } 625 } else { 626 woff = uio->uio_loffset; 627 /* 628 * Validate file offset 629 */ 630 if (woff < 0) { 631 ZFS_EXIT(zfsvfs); 632 return (EINVAL); 633 } 634 635 /* 636 * If we need to grow the block size then zfs_range_lock() 637 * will lock a wider range than we request here. 638 * Later after growing the block size we reduce the range. 639 */ 640 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 641 } 642 643 if (woff >= limit) { 644 zfs_range_unlock(rl); 645 ZFS_EXIT(zfsvfs); 646 return (EFBIG); 647 } 648 649 if ((woff + n) > limit || woff > (limit - n)) 650 n = limit - woff; 651 652 /* 653 * Check for mandatory locks 654 */ 655 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 656 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 657 zfs_range_unlock(rl); 658 ZFS_EXIT(zfsvfs); 659 return (error); 660 } 661 end_size = MAX(zp->z_phys->zp_size, woff + n); 662 663 /* 664 * Write the file in reasonable size chunks. Each chunk is written 665 * in a separate transaction; this keeps the intent log records small 666 * and allows us to do more fine-grained space accounting. 667 */ 668 while (n > 0) { 669 abuf = NULL; 670 woff = uio->uio_loffset; 671 672 again: 673 if (zfs_usergroup_overquota(zfsvfs, 674 B_FALSE, zp->z_phys->zp_uid) || 675 zfs_usergroup_overquota(zfsvfs, 676 B_TRUE, zp->z_phys->zp_gid)) { 677 if (abuf != NULL) 678 dmu_return_arcbuf(abuf); 679 error = EDQUOT; 680 break; 681 } 682 683 /* 684 * If dmu_assign_arcbuf() is expected to execute with minimum 685 * overhead loan an arc buffer and copy user data to it before 686 * we enter a txg. This avoids holding a txg forever while we 687 * pagefault on a hanging NFS server mapping. 688 */ 689 if (abuf == NULL && n >= max_blksz && 690 woff >= zp->z_phys->zp_size && 691 P2PHASE(woff, max_blksz) == 0 && 692 zp->z_blksz == max_blksz) { 693 size_t cbytes; 694 695 abuf = dmu_request_arcbuf(zp->z_dbuf, max_blksz); 696 ASSERT(abuf != NULL); 697 ASSERT(arc_buf_size(abuf) == max_blksz); 698 if (error = uiocopy(abuf->b_data, max_blksz, 699 UIO_WRITE, uio, &cbytes)) { 700 dmu_return_arcbuf(abuf); 701 break; 702 } 703 ASSERT(cbytes == max_blksz); 704 } 705 706 /* 707 * Start a transaction. 708 */ 709 tx = dmu_tx_create(zfsvfs->z_os); 710 dmu_tx_hold_bonus(tx, zp->z_id); 711 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 712 error = dmu_tx_assign(tx, TXG_NOWAIT); 713 if (error) { 714 if (error == ERESTART) { 715 dmu_tx_wait(tx); 716 dmu_tx_abort(tx); 717 goto again; 718 } 719 dmu_tx_abort(tx); 720 if (abuf != NULL) 721 dmu_return_arcbuf(abuf); 722 break; 723 } 724 725 /* 726 * If zfs_range_lock() over-locked we grow the blocksize 727 * and then reduce the lock range. This will only happen 728 * on the first iteration since zfs_range_reduce() will 729 * shrink down r_len to the appropriate size. 730 */ 731 if (rl->r_len == UINT64_MAX) { 732 uint64_t new_blksz; 733 734 if (zp->z_blksz > max_blksz) { 735 ASSERT(!ISP2(zp->z_blksz)); 736 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 737 } else { 738 new_blksz = MIN(end_size, max_blksz); 739 } 740 zfs_grow_blocksize(zp, new_blksz, tx); 741 zfs_range_reduce(rl, woff, n); 742 } 743 744 /* 745 * XXX - should we really limit each write to z_max_blksz? 746 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 747 */ 748 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 749 750 if (abuf == NULL) { 751 tx_bytes = uio->uio_resid; 752 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, uio, 753 nbytes, tx); 754 tx_bytes -= uio->uio_resid; 755 } else { 756 tx_bytes = nbytes; 757 ASSERT(tx_bytes == max_blksz); 758 dmu_assign_arcbuf(zp->z_dbuf, woff, abuf, tx); 759 ASSERT(tx_bytes <= uio->uio_resid); 760 uioskip(uio, tx_bytes); 761 } 762 if (tx_bytes && vn_has_cached_data(vp)) { 763 update_pages(vp, woff, 764 tx_bytes, zfsvfs->z_os, zp->z_id); 765 } 766 767 /* 768 * If we made no progress, we're done. If we made even 769 * partial progress, update the znode and ZIL accordingly. 770 */ 771 if (tx_bytes == 0) { 772 dmu_tx_commit(tx); 773 ASSERT(error != 0); 774 break; 775 } 776 777 /* 778 * Clear Set-UID/Set-GID bits on successful write if not 779 * privileged and at least one of the excute bits is set. 780 * 781 * It would be nice to to this after all writes have 782 * been done, but that would still expose the ISUID/ISGID 783 * to another app after the partial write is committed. 784 * 785 * Note: we don't call zfs_fuid_map_id() here because 786 * user 0 is not an ephemeral uid. 787 */ 788 mutex_enter(&zp->z_acl_lock); 789 if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) | 790 (S_IXUSR >> 6))) != 0 && 791 (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 && 792 secpolicy_vnode_setid_retain(cr, 793 (zp->z_phys->zp_mode & S_ISUID) != 0 && 794 zp->z_phys->zp_uid == 0) != 0) { 795 zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID); 796 } 797 mutex_exit(&zp->z_acl_lock); 798 799 /* 800 * Update time stamp. NOTE: This marks the bonus buffer as 801 * dirty, so we don't have to do it again for zp_size. 802 */ 803 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 804 805 /* 806 * Update the file size (zp_size) if it has changed; 807 * account for possible concurrent updates. 808 */ 809 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) 810 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 811 uio->uio_loffset); 812 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 813 dmu_tx_commit(tx); 814 815 if (error != 0) 816 break; 817 ASSERT(tx_bytes == nbytes); 818 n -= nbytes; 819 } 820 821 zfs_range_unlock(rl); 822 823 /* 824 * If we're in replay mode, or we made no progress, return error. 825 * Otherwise, it's at least a partial write, so it's successful. 826 */ 827 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 828 ZFS_EXIT(zfsvfs); 829 return (error); 830 } 831 832 if (ioflag & (FSYNC | FDSYNC)) 833 zil_commit(zilog, zp->z_last_itx, zp->z_id); 834 835 ZFS_EXIT(zfsvfs); 836 return (0); 837 } 838 839 void 840 zfs_get_done(dmu_buf_t *db, void *vzgd) 841 { 842 zgd_t *zgd = (zgd_t *)vzgd; 843 rl_t *rl = zgd->zgd_rl; 844 vnode_t *vp = ZTOV(rl->r_zp); 845 objset_t *os = rl->r_zp->z_zfsvfs->z_os; 846 847 dmu_buf_rele(db, vzgd); 848 zfs_range_unlock(rl); 849 /* 850 * Release the vnode asynchronously as we currently have the 851 * txg stopped from syncing. 852 */ 853 VN_RELE_ASYNC(vp, dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 854 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 855 kmem_free(zgd, sizeof (zgd_t)); 856 } 857 858 /* 859 * Get data to generate a TX_WRITE intent log record. 860 */ 861 int 862 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 863 { 864 zfsvfs_t *zfsvfs = arg; 865 objset_t *os = zfsvfs->z_os; 866 znode_t *zp; 867 uint64_t off = lr->lr_offset; 868 dmu_buf_t *db; 869 rl_t *rl; 870 zgd_t *zgd; 871 int dlen = lr->lr_length; /* length of user data */ 872 int error = 0; 873 874 ASSERT(zio); 875 ASSERT(dlen != 0); 876 877 /* 878 * Nothing to do if the file has been removed 879 */ 880 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 881 return (ENOENT); 882 if (zp->z_unlinked) { 883 /* 884 * Release the vnode asynchronously as we currently have the 885 * txg stopped from syncing. 886 */ 887 VN_RELE_ASYNC(ZTOV(zp), 888 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 889 return (ENOENT); 890 } 891 892 /* 893 * Write records come in two flavors: immediate and indirect. 894 * For small writes it's cheaper to store the data with the 895 * log record (immediate); for large writes it's cheaper to 896 * sync the data and get a pointer to it (indirect) so that 897 * we don't have to write the data twice. 898 */ 899 if (buf != NULL) { /* immediate write */ 900 rl = zfs_range_lock(zp, off, dlen, RL_READER); 901 /* test for truncation needs to be done while range locked */ 902 if (off >= zp->z_phys->zp_size) { 903 error = ENOENT; 904 goto out; 905 } 906 VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf, 907 DMU_READ_NO_PREFETCH)); 908 } else { /* indirect write */ 909 uint64_t boff; /* block starting offset */ 910 911 /* 912 * Have to lock the whole block to ensure when it's 913 * written out and it's checksum is being calculated 914 * that no one can change the data. We need to re-check 915 * blocksize after we get the lock in case it's changed! 916 */ 917 for (;;) { 918 if (ISP2(zp->z_blksz)) { 919 boff = P2ALIGN_TYPED(off, zp->z_blksz, 920 uint64_t); 921 } else { 922 boff = 0; 923 } 924 dlen = zp->z_blksz; 925 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 926 if (zp->z_blksz == dlen) 927 break; 928 zfs_range_unlock(rl); 929 } 930 /* test for truncation needs to be done while range locked */ 931 if (off >= zp->z_phys->zp_size) { 932 error = ENOENT; 933 goto out; 934 } 935 zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP); 936 zgd->zgd_rl = rl; 937 zgd->zgd_zilog = zfsvfs->z_log; 938 zgd->zgd_bp = &lr->lr_blkptr; 939 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db)); 940 ASSERT(boff == db->db_offset); 941 lr->lr_blkoff = off - boff; 942 error = dmu_sync(zio, db, &lr->lr_blkptr, 943 lr->lr_common.lrc_txg, zfs_get_done, zgd); 944 ASSERT((error && error != EINPROGRESS) || 945 lr->lr_length <= zp->z_blksz); 946 if (error == 0) 947 zil_add_block(zfsvfs->z_log, &lr->lr_blkptr); 948 /* 949 * If we get EINPROGRESS, then we need to wait for a 950 * write IO initiated by dmu_sync() to complete before 951 * we can release this dbuf. We will finish everything 952 * up in the zfs_get_done() callback. 953 */ 954 if (error == EINPROGRESS) 955 return (0); 956 dmu_buf_rele(db, zgd); 957 kmem_free(zgd, sizeof (zgd_t)); 958 } 959 out: 960 zfs_range_unlock(rl); 961 /* 962 * Release the vnode asynchronously as we currently have the 963 * txg stopped from syncing. 964 */ 965 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 966 return (error); 967 } 968 969 /*ARGSUSED*/ 970 static int 971 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 972 caller_context_t *ct) 973 { 974 znode_t *zp = VTOZ(vp); 975 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 976 int error; 977 978 ZFS_ENTER(zfsvfs); 979 ZFS_VERIFY_ZP(zp); 980 981 if (flag & V_ACE_MASK) 982 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 983 else 984 error = zfs_zaccess_rwx(zp, mode, flag, cr); 985 986 ZFS_EXIT(zfsvfs); 987 return (error); 988 } 989 990 /* 991 * If vnode is for a device return a specfs vnode instead. 992 */ 993 static int 994 specvp_check(vnode_t **vpp, cred_t *cr) 995 { 996 int error = 0; 997 998 if (IS_DEVVP(*vpp)) { 999 struct vnode *svp; 1000 1001 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1002 VN_RELE(*vpp); 1003 if (svp == NULL) 1004 error = ENOSYS; 1005 *vpp = svp; 1006 } 1007 return (error); 1008 } 1009 1010 1011 /* 1012 * Lookup an entry in a directory, or an extended attribute directory. 1013 * If it exists, return a held vnode reference for it. 1014 * 1015 * IN: dvp - vnode of directory to search. 1016 * nm - name of entry to lookup. 1017 * pnp - full pathname to lookup [UNUSED]. 1018 * flags - LOOKUP_XATTR set if looking for an attribute. 1019 * rdir - root directory vnode [UNUSED]. 1020 * cr - credentials of caller. 1021 * ct - caller context 1022 * direntflags - directory lookup flags 1023 * realpnp - returned pathname. 1024 * 1025 * OUT: vpp - vnode of located entry, NULL if not found. 1026 * 1027 * RETURN: 0 if success 1028 * error code if failure 1029 * 1030 * Timestamps: 1031 * NA 1032 */ 1033 /* ARGSUSED */ 1034 static int 1035 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 1036 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 1037 int *direntflags, pathname_t *realpnp) 1038 { 1039 znode_t *zdp = VTOZ(dvp); 1040 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1041 int error = 0; 1042 1043 /* fast path */ 1044 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 1045 1046 if (dvp->v_type != VDIR) { 1047 return (ENOTDIR); 1048 } else if (zdp->z_dbuf == NULL) { 1049 return (EIO); 1050 } 1051 1052 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 1053 error = zfs_fastaccesschk_execute(zdp, cr); 1054 if (!error) { 1055 *vpp = dvp; 1056 VN_HOLD(*vpp); 1057 return (0); 1058 } 1059 return (error); 1060 } else { 1061 vnode_t *tvp = dnlc_lookup(dvp, nm); 1062 1063 if (tvp) { 1064 error = zfs_fastaccesschk_execute(zdp, cr); 1065 if (error) { 1066 VN_RELE(tvp); 1067 return (error); 1068 } 1069 if (tvp == DNLC_NO_VNODE) { 1070 VN_RELE(tvp); 1071 return (ENOENT); 1072 } else { 1073 *vpp = tvp; 1074 return (specvp_check(vpp, cr)); 1075 } 1076 } 1077 } 1078 } 1079 1080 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1081 1082 ZFS_ENTER(zfsvfs); 1083 ZFS_VERIFY_ZP(zdp); 1084 1085 *vpp = NULL; 1086 1087 if (flags & LOOKUP_XATTR) { 1088 /* 1089 * If the xattr property is off, refuse the lookup request. 1090 */ 1091 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1092 ZFS_EXIT(zfsvfs); 1093 return (EINVAL); 1094 } 1095 1096 /* 1097 * We don't allow recursive attributes.. 1098 * Maybe someday we will. 1099 */ 1100 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 1101 ZFS_EXIT(zfsvfs); 1102 return (EINVAL); 1103 } 1104 1105 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1106 ZFS_EXIT(zfsvfs); 1107 return (error); 1108 } 1109 1110 /* 1111 * Do we have permission to get into attribute directory? 1112 */ 1113 1114 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1115 B_FALSE, cr)) { 1116 VN_RELE(*vpp); 1117 *vpp = NULL; 1118 } 1119 1120 ZFS_EXIT(zfsvfs); 1121 return (error); 1122 } 1123 1124 if (dvp->v_type != VDIR) { 1125 ZFS_EXIT(zfsvfs); 1126 return (ENOTDIR); 1127 } 1128 1129 /* 1130 * Check accessibility of directory. 1131 */ 1132 1133 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1134 ZFS_EXIT(zfsvfs); 1135 return (error); 1136 } 1137 1138 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1139 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1140 ZFS_EXIT(zfsvfs); 1141 return (EILSEQ); 1142 } 1143 1144 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1145 if (error == 0) 1146 error = specvp_check(vpp, cr); 1147 1148 ZFS_EXIT(zfsvfs); 1149 return (error); 1150 } 1151 1152 /* 1153 * Attempt to create a new entry in a directory. If the entry 1154 * already exists, truncate the file if permissible, else return 1155 * an error. Return the vp of the created or trunc'd file. 1156 * 1157 * IN: dvp - vnode of directory to put new file entry in. 1158 * name - name of new file entry. 1159 * vap - attributes of new file. 1160 * excl - flag indicating exclusive or non-exclusive mode. 1161 * mode - mode to open file with. 1162 * cr - credentials of caller. 1163 * flag - large file flag [UNUSED]. 1164 * ct - caller context 1165 * vsecp - ACL to be set 1166 * 1167 * OUT: vpp - vnode of created or trunc'd entry. 1168 * 1169 * RETURN: 0 if success 1170 * error code if failure 1171 * 1172 * Timestamps: 1173 * dvp - ctime|mtime updated if new entry created 1174 * vp - ctime|mtime always, atime if new 1175 */ 1176 1177 /* ARGSUSED */ 1178 static int 1179 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1180 int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct, 1181 vsecattr_t *vsecp) 1182 { 1183 znode_t *zp, *dzp = VTOZ(dvp); 1184 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1185 zilog_t *zilog; 1186 objset_t *os; 1187 zfs_dirlock_t *dl; 1188 dmu_tx_t *tx; 1189 int error; 1190 ksid_t *ksid; 1191 uid_t uid; 1192 gid_t gid = crgetgid(cr); 1193 zfs_acl_ids_t acl_ids; 1194 boolean_t fuid_dirtied; 1195 1196 /* 1197 * If we have an ephemeral id, ACL, or XVATTR then 1198 * make sure file system is at proper version 1199 */ 1200 1201 ksid = crgetsid(cr, KSID_OWNER); 1202 if (ksid) 1203 uid = ksid_getid(ksid); 1204 else 1205 uid = crgetuid(cr); 1206 1207 if (zfsvfs->z_use_fuids == B_FALSE && 1208 (vsecp || (vap->va_mask & AT_XVATTR) || 1209 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1210 return (EINVAL); 1211 1212 ZFS_ENTER(zfsvfs); 1213 ZFS_VERIFY_ZP(dzp); 1214 os = zfsvfs->z_os; 1215 zilog = zfsvfs->z_log; 1216 1217 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1218 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1219 ZFS_EXIT(zfsvfs); 1220 return (EILSEQ); 1221 } 1222 1223 if (vap->va_mask & AT_XVATTR) { 1224 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1225 crgetuid(cr), cr, vap->va_type)) != 0) { 1226 ZFS_EXIT(zfsvfs); 1227 return (error); 1228 } 1229 } 1230 top: 1231 *vpp = NULL; 1232 1233 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1234 vap->va_mode &= ~VSVTX; 1235 1236 if (*name == '\0') { 1237 /* 1238 * Null component name refers to the directory itself. 1239 */ 1240 VN_HOLD(dvp); 1241 zp = dzp; 1242 dl = NULL; 1243 error = 0; 1244 } else { 1245 /* possible VN_HOLD(zp) */ 1246 int zflg = 0; 1247 1248 if (flag & FIGNORECASE) 1249 zflg |= ZCILOOK; 1250 1251 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1252 NULL, NULL); 1253 if (error) { 1254 if (strcmp(name, "..") == 0) 1255 error = EISDIR; 1256 ZFS_EXIT(zfsvfs); 1257 return (error); 1258 } 1259 } 1260 if (zp == NULL) { 1261 uint64_t txtype; 1262 1263 /* 1264 * Create a new file object and update the directory 1265 * to reference it. 1266 */ 1267 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1268 goto out; 1269 } 1270 1271 /* 1272 * We only support the creation of regular files in 1273 * extended attribute directories. 1274 */ 1275 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1276 (vap->va_type != VREG)) { 1277 error = EINVAL; 1278 goto out; 1279 } 1280 1281 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, vsecp, 1282 &acl_ids)) != 0) 1283 goto out; 1284 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1285 zfs_acl_ids_free(&acl_ids); 1286 error = EDQUOT; 1287 goto out; 1288 } 1289 1290 tx = dmu_tx_create(os); 1291 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1292 fuid_dirtied = zfsvfs->z_fuid_dirty; 1293 if (fuid_dirtied) 1294 zfs_fuid_txhold(zfsvfs, tx); 1295 dmu_tx_hold_bonus(tx, dzp->z_id); 1296 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1297 if (acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1298 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1299 0, SPA_MAXBLOCKSIZE); 1300 } 1301 error = dmu_tx_assign(tx, TXG_NOWAIT); 1302 if (error) { 1303 zfs_acl_ids_free(&acl_ids); 1304 zfs_dirent_unlock(dl); 1305 if (error == ERESTART) { 1306 dmu_tx_wait(tx); 1307 dmu_tx_abort(tx); 1308 goto top; 1309 } 1310 dmu_tx_abort(tx); 1311 ZFS_EXIT(zfsvfs); 1312 return (error); 1313 } 1314 zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, &acl_ids); 1315 1316 if (fuid_dirtied) 1317 zfs_fuid_sync(zfsvfs, tx); 1318 1319 (void) zfs_link_create(dl, zp, tx, ZNEW); 1320 1321 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1322 if (flag & FIGNORECASE) 1323 txtype |= TX_CI; 1324 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1325 vsecp, acl_ids.z_fuidp, vap); 1326 zfs_acl_ids_free(&acl_ids); 1327 dmu_tx_commit(tx); 1328 } else { 1329 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1330 1331 /* 1332 * A directory entry already exists for this name. 1333 */ 1334 /* 1335 * Can't truncate an existing file if in exclusive mode. 1336 */ 1337 if (excl == EXCL) { 1338 error = EEXIST; 1339 goto out; 1340 } 1341 /* 1342 * Can't open a directory for writing. 1343 */ 1344 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1345 error = EISDIR; 1346 goto out; 1347 } 1348 /* 1349 * Verify requested access to file. 1350 */ 1351 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1352 goto out; 1353 } 1354 1355 mutex_enter(&dzp->z_lock); 1356 dzp->z_seq++; 1357 mutex_exit(&dzp->z_lock); 1358 1359 /* 1360 * Truncate regular files if requested. 1361 */ 1362 if ((ZTOV(zp)->v_type == VREG) && 1363 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1364 /* we can't hold any locks when calling zfs_freesp() */ 1365 zfs_dirent_unlock(dl); 1366 dl = NULL; 1367 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1368 if (error == 0) { 1369 vnevent_create(ZTOV(zp), ct); 1370 } 1371 } 1372 } 1373 out: 1374 1375 if (dl) 1376 zfs_dirent_unlock(dl); 1377 1378 if (error) { 1379 if (zp) 1380 VN_RELE(ZTOV(zp)); 1381 } else { 1382 *vpp = ZTOV(zp); 1383 error = specvp_check(vpp, cr); 1384 } 1385 1386 ZFS_EXIT(zfsvfs); 1387 return (error); 1388 } 1389 1390 /* 1391 * Remove an entry from a directory. 1392 * 1393 * IN: dvp - vnode of directory to remove entry from. 1394 * name - name of entry to remove. 1395 * cr - credentials of caller. 1396 * ct - caller context 1397 * flags - case flags 1398 * 1399 * RETURN: 0 if success 1400 * error code if failure 1401 * 1402 * Timestamps: 1403 * dvp - ctime|mtime 1404 * vp - ctime (if nlink > 0) 1405 */ 1406 /*ARGSUSED*/ 1407 static int 1408 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1409 int flags) 1410 { 1411 znode_t *zp, *dzp = VTOZ(dvp); 1412 znode_t *xzp = NULL; 1413 vnode_t *vp; 1414 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1415 zilog_t *zilog; 1416 uint64_t acl_obj, xattr_obj; 1417 zfs_dirlock_t *dl; 1418 dmu_tx_t *tx; 1419 boolean_t may_delete_now, delete_now = FALSE; 1420 boolean_t unlinked, toobig = FALSE; 1421 uint64_t txtype; 1422 pathname_t *realnmp = NULL; 1423 pathname_t realnm; 1424 int error; 1425 int zflg = ZEXISTS; 1426 1427 ZFS_ENTER(zfsvfs); 1428 ZFS_VERIFY_ZP(dzp); 1429 zilog = zfsvfs->z_log; 1430 1431 if (flags & FIGNORECASE) { 1432 zflg |= ZCILOOK; 1433 pn_alloc(&realnm); 1434 realnmp = &realnm; 1435 } 1436 1437 top: 1438 /* 1439 * Attempt to lock directory; fail if entry doesn't exist. 1440 */ 1441 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1442 NULL, realnmp)) { 1443 if (realnmp) 1444 pn_free(realnmp); 1445 ZFS_EXIT(zfsvfs); 1446 return (error); 1447 } 1448 1449 vp = ZTOV(zp); 1450 1451 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1452 goto out; 1453 } 1454 1455 /* 1456 * Need to use rmdir for removing directories. 1457 */ 1458 if (vp->v_type == VDIR) { 1459 error = EPERM; 1460 goto out; 1461 } 1462 1463 vnevent_remove(vp, dvp, name, ct); 1464 1465 if (realnmp) 1466 dnlc_remove(dvp, realnmp->pn_buf); 1467 else 1468 dnlc_remove(dvp, name); 1469 1470 mutex_enter(&vp->v_lock); 1471 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1472 mutex_exit(&vp->v_lock); 1473 1474 /* 1475 * We may delete the znode now, or we may put it in the unlinked set; 1476 * it depends on whether we're the last link, and on whether there are 1477 * other holds on the vnode. So we dmu_tx_hold() the right things to 1478 * allow for either case. 1479 */ 1480 tx = dmu_tx_create(zfsvfs->z_os); 1481 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1482 dmu_tx_hold_bonus(tx, zp->z_id); 1483 if (may_delete_now) { 1484 toobig = 1485 zp->z_phys->zp_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 1486 /* if the file is too big, only hold_free a token amount */ 1487 dmu_tx_hold_free(tx, zp->z_id, 0, 1488 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1489 } 1490 1491 /* are there any extended attributes? */ 1492 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1493 /* XXX - do we need this if we are deleting? */ 1494 dmu_tx_hold_bonus(tx, xattr_obj); 1495 } 1496 1497 /* are there any additional acls */ 1498 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1499 may_delete_now) 1500 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1501 1502 /* charge as an update -- would be nice not to charge at all */ 1503 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1504 1505 error = dmu_tx_assign(tx, TXG_NOWAIT); 1506 if (error) { 1507 zfs_dirent_unlock(dl); 1508 VN_RELE(vp); 1509 if (error == ERESTART) { 1510 dmu_tx_wait(tx); 1511 dmu_tx_abort(tx); 1512 goto top; 1513 } 1514 if (realnmp) 1515 pn_free(realnmp); 1516 dmu_tx_abort(tx); 1517 ZFS_EXIT(zfsvfs); 1518 return (error); 1519 } 1520 1521 /* 1522 * Remove the directory entry. 1523 */ 1524 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1525 1526 if (error) { 1527 dmu_tx_commit(tx); 1528 goto out; 1529 } 1530 1531 if (unlinked) { 1532 mutex_enter(&vp->v_lock); 1533 delete_now = may_delete_now && !toobig && 1534 vp->v_count == 1 && !vn_has_cached_data(vp) && 1535 zp->z_phys->zp_xattr == xattr_obj && 1536 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1537 mutex_exit(&vp->v_lock); 1538 } 1539 1540 if (delete_now) { 1541 if (zp->z_phys->zp_xattr) { 1542 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1543 ASSERT3U(error, ==, 0); 1544 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1545 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1546 mutex_enter(&xzp->z_lock); 1547 xzp->z_unlinked = 1; 1548 xzp->z_phys->zp_links = 0; 1549 mutex_exit(&xzp->z_lock); 1550 zfs_unlinked_add(xzp, tx); 1551 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1552 } 1553 mutex_enter(&zp->z_lock); 1554 mutex_enter(&vp->v_lock); 1555 vp->v_count--; 1556 ASSERT3U(vp->v_count, ==, 0); 1557 mutex_exit(&vp->v_lock); 1558 mutex_exit(&zp->z_lock); 1559 zfs_znode_delete(zp, tx); 1560 } else if (unlinked) { 1561 zfs_unlinked_add(zp, tx); 1562 } 1563 1564 txtype = TX_REMOVE; 1565 if (flags & FIGNORECASE) 1566 txtype |= TX_CI; 1567 zfs_log_remove(zilog, tx, txtype, dzp, name); 1568 1569 dmu_tx_commit(tx); 1570 out: 1571 if (realnmp) 1572 pn_free(realnmp); 1573 1574 zfs_dirent_unlock(dl); 1575 1576 if (!delete_now) { 1577 VN_RELE(vp); 1578 } else if (xzp) { 1579 /* this rele is delayed to prevent nesting transactions */ 1580 VN_RELE(ZTOV(xzp)); 1581 } 1582 1583 ZFS_EXIT(zfsvfs); 1584 return (error); 1585 } 1586 1587 /* 1588 * Create a new directory and insert it into dvp using the name 1589 * provided. Return a pointer to the inserted directory. 1590 * 1591 * IN: dvp - vnode of directory to add subdir to. 1592 * dirname - name of new directory. 1593 * vap - attributes of new directory. 1594 * cr - credentials of caller. 1595 * ct - caller context 1596 * vsecp - ACL to be set 1597 * 1598 * OUT: vpp - vnode of created directory. 1599 * 1600 * RETURN: 0 if success 1601 * error code if failure 1602 * 1603 * Timestamps: 1604 * dvp - ctime|mtime updated 1605 * vp - ctime|mtime|atime updated 1606 */ 1607 /*ARGSUSED*/ 1608 static int 1609 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 1610 caller_context_t *ct, int flags, vsecattr_t *vsecp) 1611 { 1612 znode_t *zp, *dzp = VTOZ(dvp); 1613 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1614 zilog_t *zilog; 1615 zfs_dirlock_t *dl; 1616 uint64_t txtype; 1617 dmu_tx_t *tx; 1618 int error; 1619 int zf = ZNEW; 1620 ksid_t *ksid; 1621 uid_t uid; 1622 gid_t gid = crgetgid(cr); 1623 zfs_acl_ids_t acl_ids; 1624 boolean_t fuid_dirtied; 1625 1626 ASSERT(vap->va_type == VDIR); 1627 1628 /* 1629 * If we have an ephemeral id, ACL, or XVATTR then 1630 * make sure file system is at proper version 1631 */ 1632 1633 ksid = crgetsid(cr, KSID_OWNER); 1634 if (ksid) 1635 uid = ksid_getid(ksid); 1636 else 1637 uid = crgetuid(cr); 1638 if (zfsvfs->z_use_fuids == B_FALSE && 1639 (vsecp || (vap->va_mask & AT_XVATTR) || 1640 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1641 return (EINVAL); 1642 1643 ZFS_ENTER(zfsvfs); 1644 ZFS_VERIFY_ZP(dzp); 1645 zilog = zfsvfs->z_log; 1646 1647 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1648 ZFS_EXIT(zfsvfs); 1649 return (EINVAL); 1650 } 1651 1652 if (zfsvfs->z_utf8 && u8_validate(dirname, 1653 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1654 ZFS_EXIT(zfsvfs); 1655 return (EILSEQ); 1656 } 1657 if (flags & FIGNORECASE) 1658 zf |= ZCILOOK; 1659 1660 if (vap->va_mask & AT_XVATTR) 1661 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1662 crgetuid(cr), cr, vap->va_type)) != 0) { 1663 ZFS_EXIT(zfsvfs); 1664 return (error); 1665 } 1666 1667 /* 1668 * First make sure the new directory doesn't exist. 1669 */ 1670 top: 1671 *vpp = NULL; 1672 1673 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 1674 NULL, NULL)) { 1675 ZFS_EXIT(zfsvfs); 1676 return (error); 1677 } 1678 1679 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 1680 zfs_dirent_unlock(dl); 1681 ZFS_EXIT(zfsvfs); 1682 return (error); 1683 } 1684 1685 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, vsecp, 1686 &acl_ids)) != 0) { 1687 zfs_dirent_unlock(dl); 1688 ZFS_EXIT(zfsvfs); 1689 return (error); 1690 } 1691 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1692 zfs_acl_ids_free(&acl_ids); 1693 zfs_dirent_unlock(dl); 1694 ZFS_EXIT(zfsvfs); 1695 return (EDQUOT); 1696 } 1697 1698 /* 1699 * Add a new entry to the directory. 1700 */ 1701 tx = dmu_tx_create(zfsvfs->z_os); 1702 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1703 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1704 fuid_dirtied = zfsvfs->z_fuid_dirty; 1705 if (fuid_dirtied) 1706 zfs_fuid_txhold(zfsvfs, tx); 1707 if (acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) 1708 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1709 0, SPA_MAXBLOCKSIZE); 1710 error = dmu_tx_assign(tx, TXG_NOWAIT); 1711 if (error) { 1712 zfs_acl_ids_free(&acl_ids); 1713 zfs_dirent_unlock(dl); 1714 if (error == ERESTART) { 1715 dmu_tx_wait(tx); 1716 dmu_tx_abort(tx); 1717 goto top; 1718 } 1719 dmu_tx_abort(tx); 1720 ZFS_EXIT(zfsvfs); 1721 return (error); 1722 } 1723 1724 /* 1725 * Create new node. 1726 */ 1727 zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, &acl_ids); 1728 1729 if (fuid_dirtied) 1730 zfs_fuid_sync(zfsvfs, tx); 1731 /* 1732 * Now put new name in parent dir. 1733 */ 1734 (void) zfs_link_create(dl, zp, tx, ZNEW); 1735 1736 *vpp = ZTOV(zp); 1737 1738 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 1739 if (flags & FIGNORECASE) 1740 txtype |= TX_CI; 1741 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 1742 acl_ids.z_fuidp, vap); 1743 1744 zfs_acl_ids_free(&acl_ids); 1745 dmu_tx_commit(tx); 1746 1747 zfs_dirent_unlock(dl); 1748 1749 ZFS_EXIT(zfsvfs); 1750 return (0); 1751 } 1752 1753 /* 1754 * Remove a directory subdir entry. If the current working 1755 * directory is the same as the subdir to be removed, the 1756 * remove will fail. 1757 * 1758 * IN: dvp - vnode of directory to remove from. 1759 * name - name of directory to be removed. 1760 * cwd - vnode of current working directory. 1761 * cr - credentials of caller. 1762 * ct - caller context 1763 * flags - case flags 1764 * 1765 * RETURN: 0 if success 1766 * error code if failure 1767 * 1768 * Timestamps: 1769 * dvp - ctime|mtime updated 1770 */ 1771 /*ARGSUSED*/ 1772 static int 1773 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 1774 caller_context_t *ct, int flags) 1775 { 1776 znode_t *dzp = VTOZ(dvp); 1777 znode_t *zp; 1778 vnode_t *vp; 1779 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1780 zilog_t *zilog; 1781 zfs_dirlock_t *dl; 1782 dmu_tx_t *tx; 1783 int error; 1784 int zflg = ZEXISTS; 1785 1786 ZFS_ENTER(zfsvfs); 1787 ZFS_VERIFY_ZP(dzp); 1788 zilog = zfsvfs->z_log; 1789 1790 if (flags & FIGNORECASE) 1791 zflg |= ZCILOOK; 1792 top: 1793 zp = NULL; 1794 1795 /* 1796 * Attempt to lock directory; fail if entry doesn't exist. 1797 */ 1798 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1799 NULL, NULL)) { 1800 ZFS_EXIT(zfsvfs); 1801 return (error); 1802 } 1803 1804 vp = ZTOV(zp); 1805 1806 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1807 goto out; 1808 } 1809 1810 if (vp->v_type != VDIR) { 1811 error = ENOTDIR; 1812 goto out; 1813 } 1814 1815 if (vp == cwd) { 1816 error = EINVAL; 1817 goto out; 1818 } 1819 1820 vnevent_rmdir(vp, dvp, name, ct); 1821 1822 /* 1823 * Grab a lock on the directory to make sure that noone is 1824 * trying to add (or lookup) entries while we are removing it. 1825 */ 1826 rw_enter(&zp->z_name_lock, RW_WRITER); 1827 1828 /* 1829 * Grab a lock on the parent pointer to make sure we play well 1830 * with the treewalk and directory rename code. 1831 */ 1832 rw_enter(&zp->z_parent_lock, RW_WRITER); 1833 1834 tx = dmu_tx_create(zfsvfs->z_os); 1835 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1836 dmu_tx_hold_bonus(tx, zp->z_id); 1837 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1838 error = dmu_tx_assign(tx, TXG_NOWAIT); 1839 if (error) { 1840 rw_exit(&zp->z_parent_lock); 1841 rw_exit(&zp->z_name_lock); 1842 zfs_dirent_unlock(dl); 1843 VN_RELE(vp); 1844 if (error == ERESTART) { 1845 dmu_tx_wait(tx); 1846 dmu_tx_abort(tx); 1847 goto top; 1848 } 1849 dmu_tx_abort(tx); 1850 ZFS_EXIT(zfsvfs); 1851 return (error); 1852 } 1853 1854 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 1855 1856 if (error == 0) { 1857 uint64_t txtype = TX_RMDIR; 1858 if (flags & FIGNORECASE) 1859 txtype |= TX_CI; 1860 zfs_log_remove(zilog, tx, txtype, dzp, name); 1861 } 1862 1863 dmu_tx_commit(tx); 1864 1865 rw_exit(&zp->z_parent_lock); 1866 rw_exit(&zp->z_name_lock); 1867 out: 1868 zfs_dirent_unlock(dl); 1869 1870 VN_RELE(vp); 1871 1872 ZFS_EXIT(zfsvfs); 1873 return (error); 1874 } 1875 1876 /* 1877 * Read as many directory entries as will fit into the provided 1878 * buffer from the given directory cursor position (specified in 1879 * the uio structure. 1880 * 1881 * IN: vp - vnode of directory to read. 1882 * uio - structure supplying read location, range info, 1883 * and return buffer. 1884 * cr - credentials of caller. 1885 * ct - caller context 1886 * flags - case flags 1887 * 1888 * OUT: uio - updated offset and range, buffer filled. 1889 * eofp - set to true if end-of-file detected. 1890 * 1891 * RETURN: 0 if success 1892 * error code if failure 1893 * 1894 * Timestamps: 1895 * vp - atime updated 1896 * 1897 * Note that the low 4 bits of the cookie returned by zap is always zero. 1898 * This allows us to use the low range for "special" directory entries: 1899 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1900 * we use the offset 2 for the '.zfs' directory. 1901 */ 1902 /* ARGSUSED */ 1903 static int 1904 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, 1905 caller_context_t *ct, int flags) 1906 { 1907 znode_t *zp = VTOZ(vp); 1908 iovec_t *iovp; 1909 edirent_t *eodp; 1910 dirent64_t *odp; 1911 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1912 objset_t *os; 1913 caddr_t outbuf; 1914 size_t bufsize; 1915 zap_cursor_t zc; 1916 zap_attribute_t zap; 1917 uint_t bytes_wanted; 1918 uint64_t offset; /* must be unsigned; checks for < 1 */ 1919 int local_eof; 1920 int outcount; 1921 int error; 1922 uint8_t prefetch; 1923 boolean_t check_sysattrs; 1924 1925 ZFS_ENTER(zfsvfs); 1926 ZFS_VERIFY_ZP(zp); 1927 1928 /* 1929 * If we are not given an eof variable, 1930 * use a local one. 1931 */ 1932 if (eofp == NULL) 1933 eofp = &local_eof; 1934 1935 /* 1936 * Check for valid iov_len. 1937 */ 1938 if (uio->uio_iov->iov_len <= 0) { 1939 ZFS_EXIT(zfsvfs); 1940 return (EINVAL); 1941 } 1942 1943 /* 1944 * Quit if directory has been removed (posix) 1945 */ 1946 if ((*eofp = zp->z_unlinked) != 0) { 1947 ZFS_EXIT(zfsvfs); 1948 return (0); 1949 } 1950 1951 error = 0; 1952 os = zfsvfs->z_os; 1953 offset = uio->uio_loffset; 1954 prefetch = zp->z_zn_prefetch; 1955 1956 /* 1957 * Initialize the iterator cursor. 1958 */ 1959 if (offset <= 3) { 1960 /* 1961 * Start iteration from the beginning of the directory. 1962 */ 1963 zap_cursor_init(&zc, os, zp->z_id); 1964 } else { 1965 /* 1966 * The offset is a serialized cursor. 1967 */ 1968 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1969 } 1970 1971 /* 1972 * Get space to change directory entries into fs independent format. 1973 */ 1974 iovp = uio->uio_iov; 1975 bytes_wanted = iovp->iov_len; 1976 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1977 bufsize = bytes_wanted; 1978 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1979 odp = (struct dirent64 *)outbuf; 1980 } else { 1981 bufsize = bytes_wanted; 1982 odp = (struct dirent64 *)iovp->iov_base; 1983 } 1984 eodp = (struct edirent *)odp; 1985 1986 /* 1987 * If this VFS supports the system attribute view interface; and 1988 * we're looking at an extended attribute directory; and we care 1989 * about normalization conflicts on this vfs; then we must check 1990 * for normalization conflicts with the sysattr name space. 1991 */ 1992 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 1993 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 1994 (flags & V_RDDIR_ENTFLAGS); 1995 1996 /* 1997 * Transform to file-system independent format 1998 */ 1999 outcount = 0; 2000 while (outcount < bytes_wanted) { 2001 ino64_t objnum; 2002 ushort_t reclen; 2003 off64_t *next; 2004 2005 /* 2006 * Special case `.', `..', and `.zfs'. 2007 */ 2008 if (offset == 0) { 2009 (void) strcpy(zap.za_name, "."); 2010 zap.za_normalization_conflict = 0; 2011 objnum = zp->z_id; 2012 } else if (offset == 1) { 2013 (void) strcpy(zap.za_name, ".."); 2014 zap.za_normalization_conflict = 0; 2015 objnum = zp->z_phys->zp_parent; 2016 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2017 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2018 zap.za_normalization_conflict = 0; 2019 objnum = ZFSCTL_INO_ROOT; 2020 } else { 2021 /* 2022 * Grab next entry. 2023 */ 2024 if (error = zap_cursor_retrieve(&zc, &zap)) { 2025 if ((*eofp = (error == ENOENT)) != 0) 2026 break; 2027 else 2028 goto update; 2029 } 2030 2031 if (zap.za_integer_length != 8 || 2032 zap.za_num_integers != 1) { 2033 cmn_err(CE_WARN, "zap_readdir: bad directory " 2034 "entry, obj = %lld, offset = %lld\n", 2035 (u_longlong_t)zp->z_id, 2036 (u_longlong_t)offset); 2037 error = ENXIO; 2038 goto update; 2039 } 2040 2041 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2042 /* 2043 * MacOS X can extract the object type here such as: 2044 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2045 */ 2046 2047 if (check_sysattrs && !zap.za_normalization_conflict) { 2048 zap.za_normalization_conflict = 2049 xattr_sysattr_casechk(zap.za_name); 2050 } 2051 } 2052 2053 if (flags & V_RDDIR_ACCFILTER) { 2054 /* 2055 * If we have no access at all, don't include 2056 * this entry in the returned information 2057 */ 2058 znode_t *ezp; 2059 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2060 goto skip_entry; 2061 if (!zfs_has_access(ezp, cr)) { 2062 VN_RELE(ZTOV(ezp)); 2063 goto skip_entry; 2064 } 2065 VN_RELE(ZTOV(ezp)); 2066 } 2067 2068 if (flags & V_RDDIR_ENTFLAGS) 2069 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2070 else 2071 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2072 2073 /* 2074 * Will this entry fit in the buffer? 2075 */ 2076 if (outcount + reclen > bufsize) { 2077 /* 2078 * Did we manage to fit anything in the buffer? 2079 */ 2080 if (!outcount) { 2081 error = EINVAL; 2082 goto update; 2083 } 2084 break; 2085 } 2086 if (flags & V_RDDIR_ENTFLAGS) { 2087 /* 2088 * Add extended flag entry: 2089 */ 2090 eodp->ed_ino = objnum; 2091 eodp->ed_reclen = reclen; 2092 /* NOTE: ed_off is the offset for the *next* entry */ 2093 next = &(eodp->ed_off); 2094 eodp->ed_eflags = zap.za_normalization_conflict ? 2095 ED_CASE_CONFLICT : 0; 2096 (void) strncpy(eodp->ed_name, zap.za_name, 2097 EDIRENT_NAMELEN(reclen)); 2098 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2099 } else { 2100 /* 2101 * Add normal entry: 2102 */ 2103 odp->d_ino = objnum; 2104 odp->d_reclen = reclen; 2105 /* NOTE: d_off is the offset for the *next* entry */ 2106 next = &(odp->d_off); 2107 (void) strncpy(odp->d_name, zap.za_name, 2108 DIRENT64_NAMELEN(reclen)); 2109 odp = (dirent64_t *)((intptr_t)odp + reclen); 2110 } 2111 outcount += reclen; 2112 2113 ASSERT(outcount <= bufsize); 2114 2115 /* Prefetch znode */ 2116 if (prefetch) 2117 dmu_prefetch(os, objnum, 0, 0); 2118 2119 skip_entry: 2120 /* 2121 * Move to the next entry, fill in the previous offset. 2122 */ 2123 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2124 zap_cursor_advance(&zc); 2125 offset = zap_cursor_serialize(&zc); 2126 } else { 2127 offset += 1; 2128 } 2129 *next = offset; 2130 } 2131 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2132 2133 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2134 iovp->iov_base += outcount; 2135 iovp->iov_len -= outcount; 2136 uio->uio_resid -= outcount; 2137 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2138 /* 2139 * Reset the pointer. 2140 */ 2141 offset = uio->uio_loffset; 2142 } 2143 2144 update: 2145 zap_cursor_fini(&zc); 2146 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2147 kmem_free(outbuf, bufsize); 2148 2149 if (error == ENOENT) 2150 error = 0; 2151 2152 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2153 2154 uio->uio_loffset = offset; 2155 ZFS_EXIT(zfsvfs); 2156 return (error); 2157 } 2158 2159 ulong_t zfs_fsync_sync_cnt = 4; 2160 2161 static int 2162 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2163 { 2164 znode_t *zp = VTOZ(vp); 2165 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2166 2167 /* 2168 * Regardless of whether this is required for standards conformance, 2169 * this is the logical behavior when fsync() is called on a file with 2170 * dirty pages. We use B_ASYNC since the ZIL transactions are already 2171 * going to be pushed out as part of the zil_commit(). 2172 */ 2173 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 2174 (vp->v_type == VREG) && !(IS_SWAPVP(vp))) 2175 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct); 2176 2177 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2178 2179 ZFS_ENTER(zfsvfs); 2180 ZFS_VERIFY_ZP(zp); 2181 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 2182 ZFS_EXIT(zfsvfs); 2183 return (0); 2184 } 2185 2186 2187 /* 2188 * Get the requested file attributes and place them in the provided 2189 * vattr structure. 2190 * 2191 * IN: vp - vnode of file. 2192 * vap - va_mask identifies requested attributes. 2193 * If AT_XVATTR set, then optional attrs are requested 2194 * flags - ATTR_NOACLCHECK (CIFS server context) 2195 * cr - credentials of caller. 2196 * ct - caller context 2197 * 2198 * OUT: vap - attribute values. 2199 * 2200 * RETURN: 0 (always succeeds) 2201 */ 2202 /* ARGSUSED */ 2203 static int 2204 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2205 caller_context_t *ct) 2206 { 2207 znode_t *zp = VTOZ(vp); 2208 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2209 znode_phys_t *pzp; 2210 int error = 0; 2211 uint64_t links; 2212 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2213 xoptattr_t *xoap = NULL; 2214 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2215 2216 ZFS_ENTER(zfsvfs); 2217 ZFS_VERIFY_ZP(zp); 2218 pzp = zp->z_phys; 2219 2220 /* 2221 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2222 * Also, if we are the owner don't bother, since owner should 2223 * always be allowed to read basic attributes of file. 2224 */ 2225 if (!(pzp->zp_flags & ZFS_ACL_TRIVIAL) && 2226 (pzp->zp_uid != crgetuid(cr))) { 2227 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2228 skipaclchk, cr)) { 2229 ZFS_EXIT(zfsvfs); 2230 return (error); 2231 } 2232 } 2233 2234 /* 2235 * Return all attributes. It's cheaper to provide the answer 2236 * than to determine whether we were asked the question. 2237 */ 2238 2239 mutex_enter(&zp->z_lock); 2240 vap->va_type = vp->v_type; 2241 vap->va_mode = pzp->zp_mode & MODEMASK; 2242 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2243 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2244 vap->va_nodeid = zp->z_id; 2245 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2246 links = pzp->zp_links + 1; 2247 else 2248 links = pzp->zp_links; 2249 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */ 2250 vap->va_size = pzp->zp_size; 2251 vap->va_rdev = vp->v_rdev; 2252 vap->va_seq = zp->z_seq; 2253 2254 /* 2255 * Add in any requested optional attributes and the create time. 2256 * Also set the corresponding bits in the returned attribute bitmap. 2257 */ 2258 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2259 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2260 xoap->xoa_archive = 2261 ((pzp->zp_flags & ZFS_ARCHIVE) != 0); 2262 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2263 } 2264 2265 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2266 xoap->xoa_readonly = 2267 ((pzp->zp_flags & ZFS_READONLY) != 0); 2268 XVA_SET_RTN(xvap, XAT_READONLY); 2269 } 2270 2271 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2272 xoap->xoa_system = 2273 ((pzp->zp_flags & ZFS_SYSTEM) != 0); 2274 XVA_SET_RTN(xvap, XAT_SYSTEM); 2275 } 2276 2277 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2278 xoap->xoa_hidden = 2279 ((pzp->zp_flags & ZFS_HIDDEN) != 0); 2280 XVA_SET_RTN(xvap, XAT_HIDDEN); 2281 } 2282 2283 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2284 xoap->xoa_nounlink = 2285 ((pzp->zp_flags & ZFS_NOUNLINK) != 0); 2286 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2287 } 2288 2289 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2290 xoap->xoa_immutable = 2291 ((pzp->zp_flags & ZFS_IMMUTABLE) != 0); 2292 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2293 } 2294 2295 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2296 xoap->xoa_appendonly = 2297 ((pzp->zp_flags & ZFS_APPENDONLY) != 0); 2298 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2299 } 2300 2301 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2302 xoap->xoa_nodump = 2303 ((pzp->zp_flags & ZFS_NODUMP) != 0); 2304 XVA_SET_RTN(xvap, XAT_NODUMP); 2305 } 2306 2307 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2308 xoap->xoa_opaque = 2309 ((pzp->zp_flags & ZFS_OPAQUE) != 0); 2310 XVA_SET_RTN(xvap, XAT_OPAQUE); 2311 } 2312 2313 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2314 xoap->xoa_av_quarantined = 2315 ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0); 2316 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2317 } 2318 2319 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2320 xoap->xoa_av_modified = 2321 ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0); 2322 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2323 } 2324 2325 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2326 vp->v_type == VREG && 2327 (pzp->zp_flags & ZFS_BONUS_SCANSTAMP)) { 2328 size_t len; 2329 dmu_object_info_t doi; 2330 2331 /* 2332 * Only VREG files have anti-virus scanstamps, so we 2333 * won't conflict with symlinks in the bonus buffer. 2334 */ 2335 dmu_object_info_from_db(zp->z_dbuf, &doi); 2336 len = sizeof (xoap->xoa_av_scanstamp) + 2337 sizeof (znode_phys_t); 2338 if (len <= doi.doi_bonus_size) { 2339 /* 2340 * pzp points to the start of the 2341 * znode_phys_t. pzp + 1 points to the 2342 * first byte after the znode_phys_t. 2343 */ 2344 (void) memcpy(xoap->xoa_av_scanstamp, 2345 pzp + 1, 2346 sizeof (xoap->xoa_av_scanstamp)); 2347 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 2348 } 2349 } 2350 2351 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2352 ZFS_TIME_DECODE(&xoap->xoa_createtime, pzp->zp_crtime); 2353 XVA_SET_RTN(xvap, XAT_CREATETIME); 2354 } 2355 } 2356 2357 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 2358 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 2359 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 2360 2361 mutex_exit(&zp->z_lock); 2362 2363 dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks); 2364 2365 if (zp->z_blksz == 0) { 2366 /* 2367 * Block size hasn't been set; suggest maximal I/O transfers. 2368 */ 2369 vap->va_blksize = zfsvfs->z_max_blksz; 2370 } 2371 2372 ZFS_EXIT(zfsvfs); 2373 return (0); 2374 } 2375 2376 /* 2377 * Set the file attributes to the values contained in the 2378 * vattr structure. 2379 * 2380 * IN: vp - vnode of file to be modified. 2381 * vap - new attribute values. 2382 * If AT_XVATTR set, then optional attrs are being set 2383 * flags - ATTR_UTIME set if non-default time values provided. 2384 * - ATTR_NOACLCHECK (CIFS context only). 2385 * cr - credentials of caller. 2386 * ct - caller context 2387 * 2388 * RETURN: 0 if success 2389 * error code if failure 2390 * 2391 * Timestamps: 2392 * vp - ctime updated, mtime updated if size changed. 2393 */ 2394 /* ARGSUSED */ 2395 static int 2396 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2397 caller_context_t *ct) 2398 { 2399 znode_t *zp = VTOZ(vp); 2400 znode_phys_t *pzp; 2401 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2402 zilog_t *zilog; 2403 dmu_tx_t *tx; 2404 vattr_t oldva; 2405 xvattr_t tmpxvattr; 2406 uint_t mask = vap->va_mask; 2407 uint_t saved_mask; 2408 int trim_mask = 0; 2409 uint64_t new_mode; 2410 uint64_t new_uid, new_gid; 2411 znode_t *attrzp; 2412 int need_policy = FALSE; 2413 int err; 2414 zfs_fuid_info_t *fuidp = NULL; 2415 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2416 xoptattr_t *xoap; 2417 zfs_acl_t *aclp = NULL; 2418 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2419 boolean_t fuid_dirtied = B_FALSE; 2420 2421 if (mask == 0) 2422 return (0); 2423 2424 if (mask & AT_NOSET) 2425 return (EINVAL); 2426 2427 ZFS_ENTER(zfsvfs); 2428 ZFS_VERIFY_ZP(zp); 2429 2430 pzp = zp->z_phys; 2431 zilog = zfsvfs->z_log; 2432 2433 /* 2434 * Make sure that if we have ephemeral uid/gid or xvattr specified 2435 * that file system is at proper version level 2436 */ 2437 2438 if (zfsvfs->z_use_fuids == B_FALSE && 2439 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2440 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2441 (mask & AT_XVATTR))) { 2442 ZFS_EXIT(zfsvfs); 2443 return (EINVAL); 2444 } 2445 2446 if (mask & AT_SIZE && vp->v_type == VDIR) { 2447 ZFS_EXIT(zfsvfs); 2448 return (EISDIR); 2449 } 2450 2451 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 2452 ZFS_EXIT(zfsvfs); 2453 return (EINVAL); 2454 } 2455 2456 /* 2457 * If this is an xvattr_t, then get a pointer to the structure of 2458 * optional attributes. If this is NULL, then we have a vattr_t. 2459 */ 2460 xoap = xva_getxoptattr(xvap); 2461 2462 xva_init(&tmpxvattr); 2463 2464 /* 2465 * Immutable files can only alter immutable bit and atime 2466 */ 2467 if ((pzp->zp_flags & ZFS_IMMUTABLE) && 2468 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2469 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 2470 ZFS_EXIT(zfsvfs); 2471 return (EPERM); 2472 } 2473 2474 if ((mask & AT_SIZE) && (pzp->zp_flags & ZFS_READONLY)) { 2475 ZFS_EXIT(zfsvfs); 2476 return (EPERM); 2477 } 2478 2479 /* 2480 * Verify timestamps doesn't overflow 32 bits. 2481 * ZFS can handle large timestamps, but 32bit syscalls can't 2482 * handle times greater than 2039. This check should be removed 2483 * once large timestamps are fully supported. 2484 */ 2485 if (mask & (AT_ATIME | AT_MTIME)) { 2486 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 2487 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 2488 ZFS_EXIT(zfsvfs); 2489 return (EOVERFLOW); 2490 } 2491 } 2492 2493 top: 2494 attrzp = NULL; 2495 2496 /* Can this be moved to before the top label? */ 2497 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2498 ZFS_EXIT(zfsvfs); 2499 return (EROFS); 2500 } 2501 2502 /* 2503 * First validate permissions 2504 */ 2505 2506 if (mask & AT_SIZE) { 2507 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr); 2508 if (err) { 2509 ZFS_EXIT(zfsvfs); 2510 return (err); 2511 } 2512 /* 2513 * XXX - Note, we are not providing any open 2514 * mode flags here (like FNDELAY), so we may 2515 * block if there are locks present... this 2516 * should be addressed in openat(). 2517 */ 2518 /* XXX - would it be OK to generate a log record here? */ 2519 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2520 if (err) { 2521 ZFS_EXIT(zfsvfs); 2522 return (err); 2523 } 2524 } 2525 2526 if (mask & (AT_ATIME|AT_MTIME) || 2527 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2528 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2529 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2530 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2531 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) 2532 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2533 skipaclchk, cr); 2534 2535 if (mask & (AT_UID|AT_GID)) { 2536 int idmask = (mask & (AT_UID|AT_GID)); 2537 int take_owner; 2538 int take_group; 2539 2540 /* 2541 * NOTE: even if a new mode is being set, 2542 * we may clear S_ISUID/S_ISGID bits. 2543 */ 2544 2545 if (!(mask & AT_MODE)) 2546 vap->va_mode = pzp->zp_mode; 2547 2548 /* 2549 * Take ownership or chgrp to group we are a member of 2550 */ 2551 2552 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2553 take_group = (mask & AT_GID) && 2554 zfs_groupmember(zfsvfs, vap->va_gid, cr); 2555 2556 /* 2557 * If both AT_UID and AT_GID are set then take_owner and 2558 * take_group must both be set in order to allow taking 2559 * ownership. 2560 * 2561 * Otherwise, send the check through secpolicy_vnode_setattr() 2562 * 2563 */ 2564 2565 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2566 ((idmask == AT_UID) && take_owner) || 2567 ((idmask == AT_GID) && take_group)) { 2568 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 2569 skipaclchk, cr) == 0) { 2570 /* 2571 * Remove setuid/setgid for non-privileged users 2572 */ 2573 secpolicy_setid_clear(vap, cr); 2574 trim_mask = (mask & (AT_UID|AT_GID)); 2575 } else { 2576 need_policy = TRUE; 2577 } 2578 } else { 2579 need_policy = TRUE; 2580 } 2581 } 2582 2583 mutex_enter(&zp->z_lock); 2584 oldva.va_mode = pzp->zp_mode; 2585 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 2586 if (mask & AT_XVATTR) { 2587 /* 2588 * Update xvattr mask to include only those attributes 2589 * that are actually changing. 2590 * 2591 * the bits will be restored prior to actually setting 2592 * the attributes so the caller thinks they were set. 2593 */ 2594 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2595 if (xoap->xoa_appendonly != 2596 ((pzp->zp_flags & ZFS_APPENDONLY) != 0)) { 2597 need_policy = TRUE; 2598 } else { 2599 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 2600 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 2601 } 2602 } 2603 2604 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2605 if (xoap->xoa_nounlink != 2606 ((pzp->zp_flags & ZFS_NOUNLINK) != 0)) { 2607 need_policy = TRUE; 2608 } else { 2609 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 2610 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 2611 } 2612 } 2613 2614 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2615 if (xoap->xoa_immutable != 2616 ((pzp->zp_flags & ZFS_IMMUTABLE) != 0)) { 2617 need_policy = TRUE; 2618 } else { 2619 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 2620 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 2621 } 2622 } 2623 2624 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2625 if (xoap->xoa_nodump != 2626 ((pzp->zp_flags & ZFS_NODUMP) != 0)) { 2627 need_policy = TRUE; 2628 } else { 2629 XVA_CLR_REQ(xvap, XAT_NODUMP); 2630 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 2631 } 2632 } 2633 2634 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2635 if (xoap->xoa_av_modified != 2636 ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0)) { 2637 need_policy = TRUE; 2638 } else { 2639 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 2640 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 2641 } 2642 } 2643 2644 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2645 if ((vp->v_type != VREG && 2646 xoap->xoa_av_quarantined) || 2647 xoap->xoa_av_quarantined != 2648 ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0)) { 2649 need_policy = TRUE; 2650 } else { 2651 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 2652 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 2653 } 2654 } 2655 2656 if (need_policy == FALSE && 2657 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 2658 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 2659 need_policy = TRUE; 2660 } 2661 } 2662 2663 mutex_exit(&zp->z_lock); 2664 2665 if (mask & AT_MODE) { 2666 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 2667 err = secpolicy_setid_setsticky_clear(vp, vap, 2668 &oldva, cr); 2669 if (err) { 2670 ZFS_EXIT(zfsvfs); 2671 return (err); 2672 } 2673 trim_mask |= AT_MODE; 2674 } else { 2675 need_policy = TRUE; 2676 } 2677 } 2678 2679 if (need_policy) { 2680 /* 2681 * If trim_mask is set then take ownership 2682 * has been granted or write_acl is present and user 2683 * has the ability to modify mode. In that case remove 2684 * UID|GID and or MODE from mask so that 2685 * secpolicy_vnode_setattr() doesn't revoke it. 2686 */ 2687 2688 if (trim_mask) { 2689 saved_mask = vap->va_mask; 2690 vap->va_mask &= ~trim_mask; 2691 } 2692 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2693 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 2694 if (err) { 2695 ZFS_EXIT(zfsvfs); 2696 return (err); 2697 } 2698 2699 if (trim_mask) 2700 vap->va_mask |= saved_mask; 2701 } 2702 2703 /* 2704 * secpolicy_vnode_setattr, or take ownership may have 2705 * changed va_mask 2706 */ 2707 mask = vap->va_mask; 2708 2709 tx = dmu_tx_create(zfsvfs->z_os); 2710 dmu_tx_hold_bonus(tx, zp->z_id); 2711 2712 if (mask & AT_MODE) { 2713 uint64_t pmode = pzp->zp_mode; 2714 2715 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2716 2717 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 2718 goto out; 2719 if (pzp->zp_acl.z_acl_extern_obj) { 2720 /* Are we upgrading ACL from old V0 format to new V1 */ 2721 if (zfsvfs->z_version <= ZPL_VERSION_FUID && 2722 pzp->zp_acl.z_acl_version == 2723 ZFS_ACL_VERSION_INITIAL) { 2724 dmu_tx_hold_free(tx, 2725 pzp->zp_acl.z_acl_extern_obj, 0, 2726 DMU_OBJECT_END); 2727 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2728 0, aclp->z_acl_bytes); 2729 } else { 2730 dmu_tx_hold_write(tx, 2731 pzp->zp_acl.z_acl_extern_obj, 0, 2732 aclp->z_acl_bytes); 2733 } 2734 } else if (aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2735 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2736 0, aclp->z_acl_bytes); 2737 } 2738 } 2739 2740 if (mask & (AT_UID | AT_GID)) { 2741 if (pzp->zp_xattr) { 2742 err = zfs_zget(zp->z_zfsvfs, pzp->zp_xattr, &attrzp); 2743 if (err) 2744 goto out; 2745 dmu_tx_hold_bonus(tx, attrzp->z_id); 2746 } 2747 if (mask & AT_UID) { 2748 new_uid = zfs_fuid_create(zfsvfs, 2749 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 2750 if (new_uid != pzp->zp_uid && 2751 zfs_usergroup_overquota(zfsvfs, B_FALSE, new_uid)) { 2752 err = EDQUOT; 2753 goto out; 2754 } 2755 } 2756 2757 if (mask & AT_GID) { 2758 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 2759 cr, ZFS_GROUP, &fuidp); 2760 if (new_gid != pzp->zp_gid && 2761 zfs_usergroup_overquota(zfsvfs, B_TRUE, new_gid)) { 2762 err = EDQUOT; 2763 goto out; 2764 } 2765 } 2766 fuid_dirtied = zfsvfs->z_fuid_dirty; 2767 if (fuid_dirtied) { 2768 if (zfsvfs->z_fuid_obj == 0) { 2769 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 2770 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2771 FUID_SIZE_ESTIMATE(zfsvfs)); 2772 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, 2773 FALSE, NULL); 2774 } else { 2775 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 2776 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 2777 FUID_SIZE_ESTIMATE(zfsvfs)); 2778 } 2779 } 2780 } 2781 2782 err = dmu_tx_assign(tx, TXG_NOWAIT); 2783 if (err) { 2784 if (err == ERESTART) 2785 dmu_tx_wait(tx); 2786 goto out; 2787 } 2788 2789 dmu_buf_will_dirty(zp->z_dbuf, tx); 2790 2791 /* 2792 * Set each attribute requested. 2793 * We group settings according to the locks they need to acquire. 2794 * 2795 * Note: you cannot set ctime directly, although it will be 2796 * updated as a side-effect of calling this function. 2797 */ 2798 2799 mutex_enter(&zp->z_lock); 2800 2801 if (mask & AT_MODE) { 2802 mutex_enter(&zp->z_acl_lock); 2803 zp->z_phys->zp_mode = new_mode; 2804 err = zfs_aclset_common(zp, aclp, cr, tx); 2805 ASSERT3U(err, ==, 0); 2806 zp->z_acl_cached = aclp; 2807 aclp = NULL; 2808 mutex_exit(&zp->z_acl_lock); 2809 } 2810 2811 if (attrzp) 2812 mutex_enter(&attrzp->z_lock); 2813 2814 if (mask & AT_UID) { 2815 pzp->zp_uid = new_uid; 2816 if (attrzp) 2817 attrzp->z_phys->zp_uid = new_uid; 2818 } 2819 2820 if (mask & AT_GID) { 2821 pzp->zp_gid = new_gid; 2822 if (attrzp) 2823 attrzp->z_phys->zp_gid = new_gid; 2824 } 2825 2826 if (attrzp) 2827 mutex_exit(&attrzp->z_lock); 2828 2829 if (mask & AT_ATIME) 2830 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2831 2832 if (mask & AT_MTIME) 2833 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2834 2835 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 2836 if (mask & AT_SIZE) 2837 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2838 else if (mask != 0) 2839 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2840 /* 2841 * Do this after setting timestamps to prevent timestamp 2842 * update from toggling bit 2843 */ 2844 2845 if (xoap && (mask & AT_XVATTR)) { 2846 2847 /* 2848 * restore trimmed off masks 2849 * so that return masks can be set for caller. 2850 */ 2851 2852 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 2853 XVA_SET_REQ(xvap, XAT_APPENDONLY); 2854 } 2855 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 2856 XVA_SET_REQ(xvap, XAT_NOUNLINK); 2857 } 2858 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 2859 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 2860 } 2861 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 2862 XVA_SET_REQ(xvap, XAT_NODUMP); 2863 } 2864 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 2865 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 2866 } 2867 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 2868 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 2869 } 2870 2871 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 2872 size_t len; 2873 dmu_object_info_t doi; 2874 2875 ASSERT(vp->v_type == VREG); 2876 2877 /* Grow the bonus buffer if necessary. */ 2878 dmu_object_info_from_db(zp->z_dbuf, &doi); 2879 len = sizeof (xoap->xoa_av_scanstamp) + 2880 sizeof (znode_phys_t); 2881 if (len > doi.doi_bonus_size) 2882 VERIFY(dmu_set_bonus(zp->z_dbuf, len, tx) == 0); 2883 } 2884 zfs_xvattr_set(zp, xvap); 2885 } 2886 2887 if (fuid_dirtied) 2888 zfs_fuid_sync(zfsvfs, tx); 2889 2890 if (mask != 0) 2891 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 2892 2893 mutex_exit(&zp->z_lock); 2894 2895 out: 2896 if (attrzp) 2897 VN_RELE(ZTOV(attrzp)); 2898 2899 if (aclp) 2900 zfs_acl_free(aclp); 2901 2902 if (fuidp) { 2903 zfs_fuid_info_free(fuidp); 2904 fuidp = NULL; 2905 } 2906 2907 if (err) 2908 dmu_tx_abort(tx); 2909 else 2910 dmu_tx_commit(tx); 2911 2912 if (err == ERESTART) 2913 goto top; 2914 2915 ZFS_EXIT(zfsvfs); 2916 return (err); 2917 } 2918 2919 typedef struct zfs_zlock { 2920 krwlock_t *zl_rwlock; /* lock we acquired */ 2921 znode_t *zl_znode; /* znode we held */ 2922 struct zfs_zlock *zl_next; /* next in list */ 2923 } zfs_zlock_t; 2924 2925 /* 2926 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2927 */ 2928 static void 2929 zfs_rename_unlock(zfs_zlock_t **zlpp) 2930 { 2931 zfs_zlock_t *zl; 2932 2933 while ((zl = *zlpp) != NULL) { 2934 if (zl->zl_znode != NULL) 2935 VN_RELE(ZTOV(zl->zl_znode)); 2936 rw_exit(zl->zl_rwlock); 2937 *zlpp = zl->zl_next; 2938 kmem_free(zl, sizeof (*zl)); 2939 } 2940 } 2941 2942 /* 2943 * Search back through the directory tree, using the ".." entries. 2944 * Lock each directory in the chain to prevent concurrent renames. 2945 * Fail any attempt to move a directory into one of its own descendants. 2946 * XXX - z_parent_lock can overlap with map or grow locks 2947 */ 2948 static int 2949 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2950 { 2951 zfs_zlock_t *zl; 2952 znode_t *zp = tdzp; 2953 uint64_t rootid = zp->z_zfsvfs->z_root; 2954 uint64_t *oidp = &zp->z_id; 2955 krwlock_t *rwlp = &szp->z_parent_lock; 2956 krw_t rw = RW_WRITER; 2957 2958 /* 2959 * First pass write-locks szp and compares to zp->z_id. 2960 * Later passes read-lock zp and compare to zp->z_parent. 2961 */ 2962 do { 2963 if (!rw_tryenter(rwlp, rw)) { 2964 /* 2965 * Another thread is renaming in this path. 2966 * Note that if we are a WRITER, we don't have any 2967 * parent_locks held yet. 2968 */ 2969 if (rw == RW_READER && zp->z_id > szp->z_id) { 2970 /* 2971 * Drop our locks and restart 2972 */ 2973 zfs_rename_unlock(&zl); 2974 *zlpp = NULL; 2975 zp = tdzp; 2976 oidp = &zp->z_id; 2977 rwlp = &szp->z_parent_lock; 2978 rw = RW_WRITER; 2979 continue; 2980 } else { 2981 /* 2982 * Wait for other thread to drop its locks 2983 */ 2984 rw_enter(rwlp, rw); 2985 } 2986 } 2987 2988 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2989 zl->zl_rwlock = rwlp; 2990 zl->zl_znode = NULL; 2991 zl->zl_next = *zlpp; 2992 *zlpp = zl; 2993 2994 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2995 return (EINVAL); 2996 2997 if (*oidp == rootid) /* We've hit the top */ 2998 return (0); 2999 3000 if (rw == RW_READER) { /* i.e. not the first pass */ 3001 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 3002 if (error) 3003 return (error); 3004 zl->zl_znode = zp; 3005 } 3006 oidp = &zp->z_phys->zp_parent; 3007 rwlp = &zp->z_parent_lock; 3008 rw = RW_READER; 3009 3010 } while (zp->z_id != sdzp->z_id); 3011 3012 return (0); 3013 } 3014 3015 /* 3016 * Move an entry from the provided source directory to the target 3017 * directory. Change the entry name as indicated. 3018 * 3019 * IN: sdvp - Source directory containing the "old entry". 3020 * snm - Old entry name. 3021 * tdvp - Target directory to contain the "new entry". 3022 * tnm - New entry name. 3023 * cr - credentials of caller. 3024 * ct - caller context 3025 * flags - case flags 3026 * 3027 * RETURN: 0 if success 3028 * error code if failure 3029 * 3030 * Timestamps: 3031 * sdvp,tdvp - ctime|mtime updated 3032 */ 3033 /*ARGSUSED*/ 3034 static int 3035 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 3036 caller_context_t *ct, int flags) 3037 { 3038 znode_t *tdzp, *szp, *tzp; 3039 znode_t *sdzp = VTOZ(sdvp); 3040 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 3041 zilog_t *zilog; 3042 vnode_t *realvp; 3043 zfs_dirlock_t *sdl, *tdl; 3044 dmu_tx_t *tx; 3045 zfs_zlock_t *zl; 3046 int cmp, serr, terr; 3047 int error = 0; 3048 int zflg = 0; 3049 3050 ZFS_ENTER(zfsvfs); 3051 ZFS_VERIFY_ZP(sdzp); 3052 zilog = zfsvfs->z_log; 3053 3054 /* 3055 * Make sure we have the real vp for the target directory. 3056 */ 3057 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 3058 tdvp = realvp; 3059 3060 if (tdvp->v_vfsp != sdvp->v_vfsp) { 3061 ZFS_EXIT(zfsvfs); 3062 return (EXDEV); 3063 } 3064 3065 tdzp = VTOZ(tdvp); 3066 ZFS_VERIFY_ZP(tdzp); 3067 if (zfsvfs->z_utf8 && u8_validate(tnm, 3068 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3069 ZFS_EXIT(zfsvfs); 3070 return (EILSEQ); 3071 } 3072 3073 if (flags & FIGNORECASE) 3074 zflg |= ZCILOOK; 3075 3076 top: 3077 szp = NULL; 3078 tzp = NULL; 3079 zl = NULL; 3080 3081 /* 3082 * This is to prevent the creation of links into attribute space 3083 * by renaming a linked file into/outof an attribute directory. 3084 * See the comment in zfs_link() for why this is considered bad. 3085 */ 3086 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 3087 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 3088 ZFS_EXIT(zfsvfs); 3089 return (EINVAL); 3090 } 3091 3092 /* 3093 * Lock source and target directory entries. To prevent deadlock, 3094 * a lock ordering must be defined. We lock the directory with 3095 * the smallest object id first, or if it's a tie, the one with 3096 * the lexically first name. 3097 */ 3098 if (sdzp->z_id < tdzp->z_id) { 3099 cmp = -1; 3100 } else if (sdzp->z_id > tdzp->z_id) { 3101 cmp = 1; 3102 } else { 3103 /* 3104 * First compare the two name arguments without 3105 * considering any case folding. 3106 */ 3107 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 3108 3109 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 3110 ASSERT(error == 0 || !zfsvfs->z_utf8); 3111 if (cmp == 0) { 3112 /* 3113 * POSIX: "If the old argument and the new argument 3114 * both refer to links to the same existing file, 3115 * the rename() function shall return successfully 3116 * and perform no other action." 3117 */ 3118 ZFS_EXIT(zfsvfs); 3119 return (0); 3120 } 3121 /* 3122 * If the file system is case-folding, then we may 3123 * have some more checking to do. A case-folding file 3124 * system is either supporting mixed case sensitivity 3125 * access or is completely case-insensitive. Note 3126 * that the file system is always case preserving. 3127 * 3128 * In mixed sensitivity mode case sensitive behavior 3129 * is the default. FIGNORECASE must be used to 3130 * explicitly request case insensitive behavior. 3131 * 3132 * If the source and target names provided differ only 3133 * by case (e.g., a request to rename 'tim' to 'Tim'), 3134 * we will treat this as a special case in the 3135 * case-insensitive mode: as long as the source name 3136 * is an exact match, we will allow this to proceed as 3137 * a name-change request. 3138 */ 3139 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3140 (zfsvfs->z_case == ZFS_CASE_MIXED && 3141 flags & FIGNORECASE)) && 3142 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3143 &error) == 0) { 3144 /* 3145 * case preserving rename request, require exact 3146 * name matches 3147 */ 3148 zflg |= ZCIEXACT; 3149 zflg &= ~ZCILOOK; 3150 } 3151 } 3152 3153 if (cmp < 0) { 3154 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3155 ZEXISTS | zflg, NULL, NULL); 3156 terr = zfs_dirent_lock(&tdl, 3157 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3158 } else { 3159 terr = zfs_dirent_lock(&tdl, 3160 tdzp, tnm, &tzp, zflg, NULL, NULL); 3161 serr = zfs_dirent_lock(&sdl, 3162 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3163 NULL, NULL); 3164 } 3165 3166 if (serr) { 3167 /* 3168 * Source entry invalid or not there. 3169 */ 3170 if (!terr) { 3171 zfs_dirent_unlock(tdl); 3172 if (tzp) 3173 VN_RELE(ZTOV(tzp)); 3174 } 3175 if (strcmp(snm, "..") == 0) 3176 serr = EINVAL; 3177 ZFS_EXIT(zfsvfs); 3178 return (serr); 3179 } 3180 if (terr) { 3181 zfs_dirent_unlock(sdl); 3182 VN_RELE(ZTOV(szp)); 3183 if (strcmp(tnm, "..") == 0) 3184 terr = EINVAL; 3185 ZFS_EXIT(zfsvfs); 3186 return (terr); 3187 } 3188 3189 /* 3190 * Must have write access at the source to remove the old entry 3191 * and write access at the target to create the new entry. 3192 * Note that if target and source are the same, this can be 3193 * done in a single check. 3194 */ 3195 3196 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3197 goto out; 3198 3199 if (ZTOV(szp)->v_type == VDIR) { 3200 /* 3201 * Check to make sure rename is valid. 3202 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3203 */ 3204 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3205 goto out; 3206 } 3207 3208 /* 3209 * Does target exist? 3210 */ 3211 if (tzp) { 3212 /* 3213 * Source and target must be the same type. 3214 */ 3215 if (ZTOV(szp)->v_type == VDIR) { 3216 if (ZTOV(tzp)->v_type != VDIR) { 3217 error = ENOTDIR; 3218 goto out; 3219 } 3220 } else { 3221 if (ZTOV(tzp)->v_type == VDIR) { 3222 error = EISDIR; 3223 goto out; 3224 } 3225 } 3226 /* 3227 * POSIX dictates that when the source and target 3228 * entries refer to the same file object, rename 3229 * must do nothing and exit without error. 3230 */ 3231 if (szp->z_id == tzp->z_id) { 3232 error = 0; 3233 goto out; 3234 } 3235 } 3236 3237 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3238 if (tzp) 3239 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3240 3241 /* 3242 * notify the target directory if it is not the same 3243 * as source directory. 3244 */ 3245 if (tdvp != sdvp) { 3246 vnevent_rename_dest_dir(tdvp, ct); 3247 } 3248 3249 tx = dmu_tx_create(zfsvfs->z_os); 3250 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 3251 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 3252 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3253 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3254 if (sdzp != tdzp) 3255 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 3256 if (tzp) 3257 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 3258 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3259 error = dmu_tx_assign(tx, TXG_NOWAIT); 3260 if (error) { 3261 if (zl != NULL) 3262 zfs_rename_unlock(&zl); 3263 zfs_dirent_unlock(sdl); 3264 zfs_dirent_unlock(tdl); 3265 VN_RELE(ZTOV(szp)); 3266 if (tzp) 3267 VN_RELE(ZTOV(tzp)); 3268 if (error == ERESTART) { 3269 dmu_tx_wait(tx); 3270 dmu_tx_abort(tx); 3271 goto top; 3272 } 3273 dmu_tx_abort(tx); 3274 ZFS_EXIT(zfsvfs); 3275 return (error); 3276 } 3277 3278 if (tzp) /* Attempt to remove the existing target */ 3279 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 3280 3281 if (error == 0) { 3282 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 3283 if (error == 0) { 3284 szp->z_phys->zp_flags |= ZFS_AV_MODIFIED; 3285 3286 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 3287 ASSERT(error == 0); 3288 3289 zfs_log_rename(zilog, tx, 3290 TX_RENAME | (flags & FIGNORECASE ? TX_CI : 0), 3291 sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp); 3292 3293 /* Update path information for the target vnode */ 3294 vn_renamepath(tdvp, ZTOV(szp), tnm, strlen(tnm)); 3295 } 3296 } 3297 3298 dmu_tx_commit(tx); 3299 out: 3300 if (zl != NULL) 3301 zfs_rename_unlock(&zl); 3302 3303 zfs_dirent_unlock(sdl); 3304 zfs_dirent_unlock(tdl); 3305 3306 VN_RELE(ZTOV(szp)); 3307 if (tzp) 3308 VN_RELE(ZTOV(tzp)); 3309 3310 ZFS_EXIT(zfsvfs); 3311 return (error); 3312 } 3313 3314 /* 3315 * Insert the indicated symbolic reference entry into the directory. 3316 * 3317 * IN: dvp - Directory to contain new symbolic link. 3318 * link - Name for new symlink entry. 3319 * vap - Attributes of new entry. 3320 * target - Target path of new symlink. 3321 * cr - credentials of caller. 3322 * ct - caller context 3323 * flags - case flags 3324 * 3325 * RETURN: 0 if success 3326 * error code if failure 3327 * 3328 * Timestamps: 3329 * dvp - ctime|mtime updated 3330 */ 3331 /*ARGSUSED*/ 3332 static int 3333 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr, 3334 caller_context_t *ct, int flags) 3335 { 3336 znode_t *zp, *dzp = VTOZ(dvp); 3337 zfs_dirlock_t *dl; 3338 dmu_tx_t *tx; 3339 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3340 zilog_t *zilog; 3341 int len = strlen(link); 3342 int error; 3343 int zflg = ZNEW; 3344 zfs_acl_ids_t acl_ids; 3345 boolean_t fuid_dirtied; 3346 3347 ASSERT(vap->va_type == VLNK); 3348 3349 ZFS_ENTER(zfsvfs); 3350 ZFS_VERIFY_ZP(dzp); 3351 zilog = zfsvfs->z_log; 3352 3353 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 3354 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3355 ZFS_EXIT(zfsvfs); 3356 return (EILSEQ); 3357 } 3358 if (flags & FIGNORECASE) 3359 zflg |= ZCILOOK; 3360 top: 3361 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3362 ZFS_EXIT(zfsvfs); 3363 return (error); 3364 } 3365 3366 if (len > MAXPATHLEN) { 3367 ZFS_EXIT(zfsvfs); 3368 return (ENAMETOOLONG); 3369 } 3370 3371 /* 3372 * Attempt to lock directory; fail if entry already exists. 3373 */ 3374 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 3375 if (error) { 3376 ZFS_EXIT(zfsvfs); 3377 return (error); 3378 } 3379 3380 VERIFY(0 == zfs_acl_ids_create(dzp, 0, vap, cr, NULL, &acl_ids)); 3381 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 3382 zfs_acl_ids_free(&acl_ids); 3383 zfs_dirent_unlock(dl); 3384 ZFS_EXIT(zfsvfs); 3385 return (EDQUOT); 3386 } 3387 tx = dmu_tx_create(zfsvfs->z_os); 3388 fuid_dirtied = zfsvfs->z_fuid_dirty; 3389 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 3390 dmu_tx_hold_bonus(tx, dzp->z_id); 3391 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3392 if (acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) 3393 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 3394 if (fuid_dirtied) 3395 zfs_fuid_txhold(zfsvfs, tx); 3396 error = dmu_tx_assign(tx, TXG_NOWAIT); 3397 if (error) { 3398 zfs_acl_ids_free(&acl_ids); 3399 zfs_dirent_unlock(dl); 3400 if (error == ERESTART) { 3401 dmu_tx_wait(tx); 3402 dmu_tx_abort(tx); 3403 goto top; 3404 } 3405 dmu_tx_abort(tx); 3406 ZFS_EXIT(zfsvfs); 3407 return (error); 3408 } 3409 3410 dmu_buf_will_dirty(dzp->z_dbuf, tx); 3411 3412 /* 3413 * Create a new object for the symlink. 3414 * Put the link content into bonus buffer if it will fit; 3415 * otherwise, store it just like any other file data. 3416 */ 3417 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 3418 zfs_mknode(dzp, vap, tx, cr, 0, &zp, len, &acl_ids); 3419 if (len != 0) 3420 bcopy(link, zp->z_phys + 1, len); 3421 } else { 3422 dmu_buf_t *dbp; 3423 3424 zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, &acl_ids); 3425 3426 if (fuid_dirtied) 3427 zfs_fuid_sync(zfsvfs, tx); 3428 /* 3429 * Nothing can access the znode yet so no locking needed 3430 * for growing the znode's blocksize. 3431 */ 3432 zfs_grow_blocksize(zp, len, tx); 3433 3434 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, 3435 zp->z_id, 0, FTAG, &dbp)); 3436 dmu_buf_will_dirty(dbp, tx); 3437 3438 ASSERT3U(len, <=, dbp->db_size); 3439 bcopy(link, dbp->db_data, len); 3440 dmu_buf_rele(dbp, FTAG); 3441 } 3442 zp->z_phys->zp_size = len; 3443 3444 /* 3445 * Insert the new object into the directory. 3446 */ 3447 (void) zfs_link_create(dl, zp, tx, ZNEW); 3448 if (error == 0) { 3449 uint64_t txtype = TX_SYMLINK; 3450 if (flags & FIGNORECASE) 3451 txtype |= TX_CI; 3452 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 3453 } 3454 3455 zfs_acl_ids_free(&acl_ids); 3456 3457 dmu_tx_commit(tx); 3458 3459 zfs_dirent_unlock(dl); 3460 3461 VN_RELE(ZTOV(zp)); 3462 3463 ZFS_EXIT(zfsvfs); 3464 return (error); 3465 } 3466 3467 /* 3468 * Return, in the buffer contained in the provided uio structure, 3469 * the symbolic path referred to by vp. 3470 * 3471 * IN: vp - vnode of symbolic link. 3472 * uoip - structure to contain the link path. 3473 * cr - credentials of caller. 3474 * ct - caller context 3475 * 3476 * OUT: uio - structure to contain the link path. 3477 * 3478 * RETURN: 0 if success 3479 * error code if failure 3480 * 3481 * Timestamps: 3482 * vp - atime updated 3483 */ 3484 /* ARGSUSED */ 3485 static int 3486 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 3487 { 3488 znode_t *zp = VTOZ(vp); 3489 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3490 size_t bufsz; 3491 int error; 3492 3493 ZFS_ENTER(zfsvfs); 3494 ZFS_VERIFY_ZP(zp); 3495 3496 bufsz = (size_t)zp->z_phys->zp_size; 3497 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 3498 error = uiomove(zp->z_phys + 1, 3499 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 3500 } else { 3501 dmu_buf_t *dbp; 3502 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 3503 if (error) { 3504 ZFS_EXIT(zfsvfs); 3505 return (error); 3506 } 3507 error = uiomove(dbp->db_data, 3508 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 3509 dmu_buf_rele(dbp, FTAG); 3510 } 3511 3512 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3513 ZFS_EXIT(zfsvfs); 3514 return (error); 3515 } 3516 3517 /* 3518 * Insert a new entry into directory tdvp referencing svp. 3519 * 3520 * IN: tdvp - Directory to contain new entry. 3521 * svp - vnode of new entry. 3522 * name - name of new entry. 3523 * cr - credentials of caller. 3524 * ct - caller context 3525 * 3526 * RETURN: 0 if success 3527 * error code if failure 3528 * 3529 * Timestamps: 3530 * tdvp - ctime|mtime updated 3531 * svp - ctime updated 3532 */ 3533 /* ARGSUSED */ 3534 static int 3535 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 3536 caller_context_t *ct, int flags) 3537 { 3538 znode_t *dzp = VTOZ(tdvp); 3539 znode_t *tzp, *szp; 3540 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3541 zilog_t *zilog; 3542 zfs_dirlock_t *dl; 3543 dmu_tx_t *tx; 3544 vnode_t *realvp; 3545 int error; 3546 int zf = ZNEW; 3547 uid_t owner; 3548 3549 ASSERT(tdvp->v_type == VDIR); 3550 3551 ZFS_ENTER(zfsvfs); 3552 ZFS_VERIFY_ZP(dzp); 3553 zilog = zfsvfs->z_log; 3554 3555 if (VOP_REALVP(svp, &realvp, ct) == 0) 3556 svp = realvp; 3557 3558 if (svp->v_vfsp != tdvp->v_vfsp) { 3559 ZFS_EXIT(zfsvfs); 3560 return (EXDEV); 3561 } 3562 szp = VTOZ(svp); 3563 ZFS_VERIFY_ZP(szp); 3564 3565 if (zfsvfs->z_utf8 && u8_validate(name, 3566 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3567 ZFS_EXIT(zfsvfs); 3568 return (EILSEQ); 3569 } 3570 if (flags & FIGNORECASE) 3571 zf |= ZCILOOK; 3572 3573 top: 3574 /* 3575 * We do not support links between attributes and non-attributes 3576 * because of the potential security risk of creating links 3577 * into "normal" file space in order to circumvent restrictions 3578 * imposed in attribute space. 3579 */ 3580 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 3581 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 3582 ZFS_EXIT(zfsvfs); 3583 return (EINVAL); 3584 } 3585 3586 /* 3587 * POSIX dictates that we return EPERM here. 3588 * Better choices include ENOTSUP or EISDIR. 3589 */ 3590 if (svp->v_type == VDIR) { 3591 ZFS_EXIT(zfsvfs); 3592 return (EPERM); 3593 } 3594 3595 owner = zfs_fuid_map_id(zfsvfs, szp->z_phys->zp_uid, cr, ZFS_OWNER); 3596 if (owner != crgetuid(cr) && 3597 secpolicy_basic_link(cr) != 0) { 3598 ZFS_EXIT(zfsvfs); 3599 return (EPERM); 3600 } 3601 3602 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3603 ZFS_EXIT(zfsvfs); 3604 return (error); 3605 } 3606 3607 /* 3608 * Attempt to lock directory; fail if entry already exists. 3609 */ 3610 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 3611 if (error) { 3612 ZFS_EXIT(zfsvfs); 3613 return (error); 3614 } 3615 3616 tx = dmu_tx_create(zfsvfs->z_os); 3617 dmu_tx_hold_bonus(tx, szp->z_id); 3618 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3619 error = dmu_tx_assign(tx, TXG_NOWAIT); 3620 if (error) { 3621 zfs_dirent_unlock(dl); 3622 if (error == ERESTART) { 3623 dmu_tx_wait(tx); 3624 dmu_tx_abort(tx); 3625 goto top; 3626 } 3627 dmu_tx_abort(tx); 3628 ZFS_EXIT(zfsvfs); 3629 return (error); 3630 } 3631 3632 error = zfs_link_create(dl, szp, tx, 0); 3633 3634 if (error == 0) { 3635 uint64_t txtype = TX_LINK; 3636 if (flags & FIGNORECASE) 3637 txtype |= TX_CI; 3638 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 3639 } 3640 3641 dmu_tx_commit(tx); 3642 3643 zfs_dirent_unlock(dl); 3644 3645 if (error == 0) { 3646 vnevent_link(svp, ct); 3647 } 3648 3649 ZFS_EXIT(zfsvfs); 3650 return (error); 3651 } 3652 3653 /* 3654 * zfs_null_putapage() is used when the file system has been force 3655 * unmounted. It just drops the pages. 3656 */ 3657 /* ARGSUSED */ 3658 static int 3659 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 3660 size_t *lenp, int flags, cred_t *cr) 3661 { 3662 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 3663 return (0); 3664 } 3665 3666 /* 3667 * Push a page out to disk, klustering if possible. 3668 * 3669 * IN: vp - file to push page to. 3670 * pp - page to push. 3671 * flags - additional flags. 3672 * cr - credentials of caller. 3673 * 3674 * OUT: offp - start of range pushed. 3675 * lenp - len of range pushed. 3676 * 3677 * RETURN: 0 if success 3678 * error code if failure 3679 * 3680 * NOTE: callers must have locked the page to be pushed. On 3681 * exit, the page (and all other pages in the kluster) must be 3682 * unlocked. 3683 */ 3684 /* ARGSUSED */ 3685 static int 3686 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 3687 size_t *lenp, int flags, cred_t *cr) 3688 { 3689 znode_t *zp = VTOZ(vp); 3690 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3691 dmu_tx_t *tx; 3692 u_offset_t off, koff; 3693 size_t len, klen; 3694 uint64_t filesz; 3695 int err; 3696 3697 filesz = zp->z_phys->zp_size; 3698 off = pp->p_offset; 3699 len = PAGESIZE; 3700 /* 3701 * If our blocksize is bigger than the page size, try to kluster 3702 * multiple pages so that we write a full block (thus avoiding 3703 * a read-modify-write). 3704 */ 3705 if (off < filesz && zp->z_blksz > PAGESIZE) { 3706 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3707 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 3708 ASSERT(koff <= filesz); 3709 if (koff + klen > filesz) 3710 klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE); 3711 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 3712 } 3713 ASSERT3U(btop(len), ==, btopr(len)); 3714 3715 /* 3716 * Can't push pages past end-of-file. 3717 */ 3718 if (off >= filesz) { 3719 /* ignore all pages */ 3720 err = 0; 3721 goto out; 3722 } else if (off + len > filesz) { 3723 int npages = btopr(filesz - off); 3724 page_t *trunc; 3725 3726 page_list_break(&pp, &trunc, npages); 3727 /* ignore pages past end of file */ 3728 if (trunc) 3729 pvn_write_done(trunc, flags); 3730 len = filesz - off; 3731 } 3732 3733 if (zfs_usergroup_overquota(zfsvfs, B_FALSE, zp->z_phys->zp_uid) || 3734 zfs_usergroup_overquota(zfsvfs, B_TRUE, zp->z_phys->zp_gid)) { 3735 err = EDQUOT; 3736 goto out; 3737 } 3738 top: 3739 tx = dmu_tx_create(zfsvfs->z_os); 3740 dmu_tx_hold_write(tx, zp->z_id, off, len); 3741 dmu_tx_hold_bonus(tx, zp->z_id); 3742 err = dmu_tx_assign(tx, TXG_NOWAIT); 3743 if (err != 0) { 3744 if (err == ERESTART) { 3745 dmu_tx_wait(tx); 3746 dmu_tx_abort(tx); 3747 goto top; 3748 } 3749 dmu_tx_abort(tx); 3750 goto out; 3751 } 3752 3753 if (zp->z_blksz <= PAGESIZE) { 3754 caddr_t va = zfs_map_page(pp, S_READ); 3755 ASSERT3U(len, <=, PAGESIZE); 3756 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 3757 zfs_unmap_page(pp, va); 3758 } else { 3759 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 3760 } 3761 3762 if (err == 0) { 3763 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 3764 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 3765 } 3766 dmu_tx_commit(tx); 3767 3768 out: 3769 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 3770 if (offp) 3771 *offp = off; 3772 if (lenp) 3773 *lenp = len; 3774 3775 return (err); 3776 } 3777 3778 /* 3779 * Copy the portion of the file indicated from pages into the file. 3780 * The pages are stored in a page list attached to the files vnode. 3781 * 3782 * IN: vp - vnode of file to push page data to. 3783 * off - position in file to put data. 3784 * len - amount of data to write. 3785 * flags - flags to control the operation. 3786 * cr - credentials of caller. 3787 * ct - caller context. 3788 * 3789 * RETURN: 0 if success 3790 * error code if failure 3791 * 3792 * Timestamps: 3793 * vp - ctime|mtime updated 3794 */ 3795 /*ARGSUSED*/ 3796 static int 3797 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 3798 caller_context_t *ct) 3799 { 3800 znode_t *zp = VTOZ(vp); 3801 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3802 page_t *pp; 3803 size_t io_len; 3804 u_offset_t io_off; 3805 uint_t blksz; 3806 rl_t *rl; 3807 int error = 0; 3808 3809 ZFS_ENTER(zfsvfs); 3810 ZFS_VERIFY_ZP(zp); 3811 3812 /* 3813 * Align this request to the file block size in case we kluster. 3814 * XXX - this can result in pretty aggresive locking, which can 3815 * impact simultanious read/write access. One option might be 3816 * to break up long requests (len == 0) into block-by-block 3817 * operations to get narrower locking. 3818 */ 3819 blksz = zp->z_blksz; 3820 if (ISP2(blksz)) 3821 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 3822 else 3823 io_off = 0; 3824 if (len > 0 && ISP2(blksz)) 3825 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 3826 else 3827 io_len = 0; 3828 3829 if (io_len == 0) { 3830 /* 3831 * Search the entire vp list for pages >= io_off. 3832 */ 3833 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 3834 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 3835 goto out; 3836 } 3837 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 3838 3839 if (off > zp->z_phys->zp_size) { 3840 /* past end of file */ 3841 zfs_range_unlock(rl); 3842 ZFS_EXIT(zfsvfs); 3843 return (0); 3844 } 3845 3846 len = MIN(io_len, P2ROUNDUP(zp->z_phys->zp_size, PAGESIZE) - io_off); 3847 3848 for (off = io_off; io_off < off + len; io_off += io_len) { 3849 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 3850 pp = page_lookup(vp, io_off, 3851 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 3852 } else { 3853 pp = page_lookup_nowait(vp, io_off, 3854 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 3855 } 3856 3857 if (pp != NULL && pvn_getdirty(pp, flags)) { 3858 int err; 3859 3860 /* 3861 * Found a dirty page to push 3862 */ 3863 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 3864 if (err) 3865 error = err; 3866 } else { 3867 io_len = PAGESIZE; 3868 } 3869 } 3870 out: 3871 zfs_range_unlock(rl); 3872 if ((flags & B_ASYNC) == 0) 3873 zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id); 3874 ZFS_EXIT(zfsvfs); 3875 return (error); 3876 } 3877 3878 /*ARGSUSED*/ 3879 void 3880 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 3881 { 3882 znode_t *zp = VTOZ(vp); 3883 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3884 int error; 3885 3886 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 3887 if (zp->z_dbuf == NULL) { 3888 /* 3889 * The fs has been unmounted, or we did a 3890 * suspend/resume and this file no longer exists. 3891 */ 3892 if (vn_has_cached_data(vp)) { 3893 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 3894 B_INVAL, cr); 3895 } 3896 3897 mutex_enter(&zp->z_lock); 3898 vp->v_count = 0; /* count arrives as 1 */ 3899 mutex_exit(&zp->z_lock); 3900 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3901 zfs_znode_free(zp); 3902 return; 3903 } 3904 3905 /* 3906 * Attempt to push any data in the page cache. If this fails 3907 * we will get kicked out later in zfs_zinactive(). 3908 */ 3909 if (vn_has_cached_data(vp)) { 3910 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 3911 cr); 3912 } 3913 3914 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 3915 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 3916 3917 dmu_tx_hold_bonus(tx, zp->z_id); 3918 error = dmu_tx_assign(tx, TXG_WAIT); 3919 if (error) { 3920 dmu_tx_abort(tx); 3921 } else { 3922 dmu_buf_will_dirty(zp->z_dbuf, tx); 3923 mutex_enter(&zp->z_lock); 3924 zp->z_atime_dirty = 0; 3925 mutex_exit(&zp->z_lock); 3926 dmu_tx_commit(tx); 3927 } 3928 } 3929 3930 zfs_zinactive(zp); 3931 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3932 } 3933 3934 /* 3935 * Bounds-check the seek operation. 3936 * 3937 * IN: vp - vnode seeking within 3938 * ooff - old file offset 3939 * noffp - pointer to new file offset 3940 * ct - caller context 3941 * 3942 * RETURN: 0 if success 3943 * EINVAL if new offset invalid 3944 */ 3945 /* ARGSUSED */ 3946 static int 3947 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 3948 caller_context_t *ct) 3949 { 3950 if (vp->v_type == VDIR) 3951 return (0); 3952 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 3953 } 3954 3955 /* 3956 * Pre-filter the generic locking function to trap attempts to place 3957 * a mandatory lock on a memory mapped file. 3958 */ 3959 static int 3960 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 3961 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 3962 { 3963 znode_t *zp = VTOZ(vp); 3964 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3965 int error; 3966 3967 ZFS_ENTER(zfsvfs); 3968 ZFS_VERIFY_ZP(zp); 3969 3970 /* 3971 * We are following the UFS semantics with respect to mapcnt 3972 * here: If we see that the file is mapped already, then we will 3973 * return an error, but we don't worry about races between this 3974 * function and zfs_map(). 3975 */ 3976 if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) { 3977 ZFS_EXIT(zfsvfs); 3978 return (EAGAIN); 3979 } 3980 error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct); 3981 ZFS_EXIT(zfsvfs); 3982 return (error); 3983 } 3984 3985 /* 3986 * If we can't find a page in the cache, we will create a new page 3987 * and fill it with file data. For efficiency, we may try to fill 3988 * multiple pages at once (klustering) to fill up the supplied page 3989 * list. Note that the pages to be filled are held with an exclusive 3990 * lock to prevent access by other threads while they are being filled. 3991 */ 3992 static int 3993 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 3994 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 3995 { 3996 znode_t *zp = VTOZ(vp); 3997 page_t *pp, *cur_pp; 3998 objset_t *os = zp->z_zfsvfs->z_os; 3999 u_offset_t io_off, total; 4000 size_t io_len; 4001 int err; 4002 4003 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4004 /* 4005 * We only have a single page, don't bother klustering 4006 */ 4007 io_off = off; 4008 io_len = PAGESIZE; 4009 pp = page_create_va(vp, io_off, io_len, 4010 PG_EXCL | PG_WAIT, seg, addr); 4011 } else { 4012 /* 4013 * Try to find enough pages to fill the page list 4014 */ 4015 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4016 &io_len, off, plsz, 0); 4017 } 4018 if (pp == NULL) { 4019 /* 4020 * The page already exists, nothing to do here. 4021 */ 4022 *pl = NULL; 4023 return (0); 4024 } 4025 4026 /* 4027 * Fill the pages in the kluster. 4028 */ 4029 cur_pp = pp; 4030 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4031 caddr_t va; 4032 4033 ASSERT3U(io_off, ==, cur_pp->p_offset); 4034 va = zfs_map_page(cur_pp, S_WRITE); 4035 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4036 DMU_READ_PREFETCH); 4037 zfs_unmap_page(cur_pp, va); 4038 if (err) { 4039 /* On error, toss the entire kluster */ 4040 pvn_read_done(pp, B_ERROR); 4041 /* convert checksum errors into IO errors */ 4042 if (err == ECKSUM) 4043 err = EIO; 4044 return (err); 4045 } 4046 cur_pp = cur_pp->p_next; 4047 } 4048 4049 /* 4050 * Fill in the page list array from the kluster starting 4051 * from the desired offset `off'. 4052 * NOTE: the page list will always be null terminated. 4053 */ 4054 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4055 ASSERT(pl == NULL || (*pl)->p_offset == off); 4056 4057 return (0); 4058 } 4059 4060 /* 4061 * Return pointers to the pages for the file region [off, off + len] 4062 * in the pl array. If plsz is greater than len, this function may 4063 * also return page pointers from after the specified region 4064 * (i.e. the region [off, off + plsz]). These additional pages are 4065 * only returned if they are already in the cache, or were created as 4066 * part of a klustered read. 4067 * 4068 * IN: vp - vnode of file to get data from. 4069 * off - position in file to get data from. 4070 * len - amount of data to retrieve. 4071 * plsz - length of provided page list. 4072 * seg - segment to obtain pages for. 4073 * addr - virtual address of fault. 4074 * rw - mode of created pages. 4075 * cr - credentials of caller. 4076 * ct - caller context. 4077 * 4078 * OUT: protp - protection mode of created pages. 4079 * pl - list of pages created. 4080 * 4081 * RETURN: 0 if success 4082 * error code if failure 4083 * 4084 * Timestamps: 4085 * vp - atime updated 4086 */ 4087 /* ARGSUSED */ 4088 static int 4089 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4090 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4091 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4092 { 4093 znode_t *zp = VTOZ(vp); 4094 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4095 page_t **pl0 = pl; 4096 int err = 0; 4097 4098 /* we do our own caching, faultahead is unnecessary */ 4099 if (pl == NULL) 4100 return (0); 4101 else if (len > plsz) 4102 len = plsz; 4103 else 4104 len = P2ROUNDUP(len, PAGESIZE); 4105 ASSERT(plsz >= len); 4106 4107 ZFS_ENTER(zfsvfs); 4108 ZFS_VERIFY_ZP(zp); 4109 4110 if (protp) 4111 *protp = PROT_ALL; 4112 4113 /* 4114 * Loop through the requested range [off, off + len) looking 4115 * for pages. If we don't find a page, we will need to create 4116 * a new page and fill it with data from the file. 4117 */ 4118 while (len > 0) { 4119 if (*pl = page_lookup(vp, off, SE_SHARED)) 4120 *(pl+1) = NULL; 4121 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 4122 goto out; 4123 while (*pl) { 4124 ASSERT3U((*pl)->p_offset, ==, off); 4125 off += PAGESIZE; 4126 addr += PAGESIZE; 4127 if (len > 0) { 4128 ASSERT3U(len, >=, PAGESIZE); 4129 len -= PAGESIZE; 4130 } 4131 ASSERT3U(plsz, >=, PAGESIZE); 4132 plsz -= PAGESIZE; 4133 pl++; 4134 } 4135 } 4136 4137 /* 4138 * Fill out the page array with any pages already in the cache. 4139 */ 4140 while (plsz > 0 && 4141 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 4142 off += PAGESIZE; 4143 plsz -= PAGESIZE; 4144 } 4145 out: 4146 if (err) { 4147 /* 4148 * Release any pages we have previously locked. 4149 */ 4150 while (pl > pl0) 4151 page_unlock(*--pl); 4152 } else { 4153 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4154 } 4155 4156 *pl = NULL; 4157 4158 ZFS_EXIT(zfsvfs); 4159 return (err); 4160 } 4161 4162 /* 4163 * Request a memory map for a section of a file. This code interacts 4164 * with common code and the VM system as follows: 4165 * 4166 * common code calls mmap(), which ends up in smmap_common() 4167 * 4168 * this calls VOP_MAP(), which takes you into (say) zfs 4169 * 4170 * zfs_map() calls as_map(), passing segvn_create() as the callback 4171 * 4172 * segvn_create() creates the new segment and calls VOP_ADDMAP() 4173 * 4174 * zfs_addmap() updates z_mapcnt 4175 */ 4176 /*ARGSUSED*/ 4177 static int 4178 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4179 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4180 caller_context_t *ct) 4181 { 4182 znode_t *zp = VTOZ(vp); 4183 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4184 segvn_crargs_t vn_a; 4185 int error; 4186 4187 ZFS_ENTER(zfsvfs); 4188 ZFS_VERIFY_ZP(zp); 4189 4190 if ((prot & PROT_WRITE) && 4191 (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_READONLY | 4192 ZFS_APPENDONLY))) { 4193 ZFS_EXIT(zfsvfs); 4194 return (EPERM); 4195 } 4196 4197 if ((prot & (PROT_READ | PROT_EXEC)) && 4198 (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED)) { 4199 ZFS_EXIT(zfsvfs); 4200 return (EACCES); 4201 } 4202 4203 if (vp->v_flag & VNOMAP) { 4204 ZFS_EXIT(zfsvfs); 4205 return (ENOSYS); 4206 } 4207 4208 if (off < 0 || len > MAXOFFSET_T - off) { 4209 ZFS_EXIT(zfsvfs); 4210 return (ENXIO); 4211 } 4212 4213 if (vp->v_type != VREG) { 4214 ZFS_EXIT(zfsvfs); 4215 return (ENODEV); 4216 } 4217 4218 /* 4219 * If file is locked, disallow mapping. 4220 */ 4221 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) { 4222 ZFS_EXIT(zfsvfs); 4223 return (EAGAIN); 4224 } 4225 4226 as_rangelock(as); 4227 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 4228 if (error != 0) { 4229 as_rangeunlock(as); 4230 ZFS_EXIT(zfsvfs); 4231 return (error); 4232 } 4233 4234 vn_a.vp = vp; 4235 vn_a.offset = (u_offset_t)off; 4236 vn_a.type = flags & MAP_TYPE; 4237 vn_a.prot = prot; 4238 vn_a.maxprot = maxprot; 4239 vn_a.cred = cr; 4240 vn_a.amp = NULL; 4241 vn_a.flags = flags & ~MAP_TYPE; 4242 vn_a.szc = 0; 4243 vn_a.lgrp_mem_policy_flags = 0; 4244 4245 error = as_map(as, *addrp, len, segvn_create, &vn_a); 4246 4247 as_rangeunlock(as); 4248 ZFS_EXIT(zfsvfs); 4249 return (error); 4250 } 4251 4252 /* ARGSUSED */ 4253 static int 4254 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4255 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4256 caller_context_t *ct) 4257 { 4258 uint64_t pages = btopr(len); 4259 4260 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 4261 return (0); 4262 } 4263 4264 /* 4265 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 4266 * more accurate mtime for the associated file. Since we don't have a way of 4267 * detecting when the data was actually modified, we have to resort to 4268 * heuristics. If an explicit msync() is done, then we mark the mtime when the 4269 * last page is pushed. The problem occurs when the msync() call is omitted, 4270 * which by far the most common case: 4271 * 4272 * open() 4273 * mmap() 4274 * <modify memory> 4275 * munmap() 4276 * close() 4277 * <time lapse> 4278 * putpage() via fsflush 4279 * 4280 * If we wait until fsflush to come along, we can have a modification time that 4281 * is some arbitrary point in the future. In order to prevent this in the 4282 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 4283 * torn down. 4284 */ 4285 /* ARGSUSED */ 4286 static int 4287 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4288 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 4289 caller_context_t *ct) 4290 { 4291 uint64_t pages = btopr(len); 4292 4293 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 4294 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 4295 4296 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 4297 vn_has_cached_data(vp)) 4298 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 4299 4300 return (0); 4301 } 4302 4303 /* 4304 * Free or allocate space in a file. Currently, this function only 4305 * supports the `F_FREESP' command. However, this command is somewhat 4306 * misnamed, as its functionality includes the ability to allocate as 4307 * well as free space. 4308 * 4309 * IN: vp - vnode of file to free data in. 4310 * cmd - action to take (only F_FREESP supported). 4311 * bfp - section of file to free/alloc. 4312 * flag - current file open mode flags. 4313 * offset - current file offset. 4314 * cr - credentials of caller [UNUSED]. 4315 * ct - caller context. 4316 * 4317 * RETURN: 0 if success 4318 * error code if failure 4319 * 4320 * Timestamps: 4321 * vp - ctime|mtime updated 4322 */ 4323 /* ARGSUSED */ 4324 static int 4325 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 4326 offset_t offset, cred_t *cr, caller_context_t *ct) 4327 { 4328 znode_t *zp = VTOZ(vp); 4329 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4330 uint64_t off, len; 4331 int error; 4332 4333 ZFS_ENTER(zfsvfs); 4334 ZFS_VERIFY_ZP(zp); 4335 4336 if (cmd != F_FREESP) { 4337 ZFS_EXIT(zfsvfs); 4338 return (EINVAL); 4339 } 4340 4341 if (error = convoff(vp, bfp, 0, offset)) { 4342 ZFS_EXIT(zfsvfs); 4343 return (error); 4344 } 4345 4346 if (bfp->l_len < 0) { 4347 ZFS_EXIT(zfsvfs); 4348 return (EINVAL); 4349 } 4350 4351 off = bfp->l_start; 4352 len = bfp->l_len; /* 0 means from off to end of file */ 4353 4354 error = zfs_freesp(zp, off, len, flag, TRUE); 4355 4356 ZFS_EXIT(zfsvfs); 4357 return (error); 4358 } 4359 4360 /*ARGSUSED*/ 4361 static int 4362 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 4363 { 4364 znode_t *zp = VTOZ(vp); 4365 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4366 uint32_t gen; 4367 uint64_t object = zp->z_id; 4368 zfid_short_t *zfid; 4369 int size, i; 4370 4371 ZFS_ENTER(zfsvfs); 4372 ZFS_VERIFY_ZP(zp); 4373 gen = (uint32_t)zp->z_gen; 4374 4375 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 4376 if (fidp->fid_len < size) { 4377 fidp->fid_len = size; 4378 ZFS_EXIT(zfsvfs); 4379 return (ENOSPC); 4380 } 4381 4382 zfid = (zfid_short_t *)fidp; 4383 4384 zfid->zf_len = size; 4385 4386 for (i = 0; i < sizeof (zfid->zf_object); i++) 4387 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 4388 4389 /* Must have a non-zero generation number to distinguish from .zfs */ 4390 if (gen == 0) 4391 gen = 1; 4392 for (i = 0; i < sizeof (zfid->zf_gen); i++) 4393 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 4394 4395 if (size == LONG_FID_LEN) { 4396 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 4397 zfid_long_t *zlfid; 4398 4399 zlfid = (zfid_long_t *)fidp; 4400 4401 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 4402 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 4403 4404 /* XXX - this should be the generation number for the objset */ 4405 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 4406 zlfid->zf_setgen[i] = 0; 4407 } 4408 4409 ZFS_EXIT(zfsvfs); 4410 return (0); 4411 } 4412 4413 static int 4414 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 4415 caller_context_t *ct) 4416 { 4417 znode_t *zp, *xzp; 4418 zfsvfs_t *zfsvfs; 4419 zfs_dirlock_t *dl; 4420 int error; 4421 4422 switch (cmd) { 4423 case _PC_LINK_MAX: 4424 *valp = ULONG_MAX; 4425 return (0); 4426 4427 case _PC_FILESIZEBITS: 4428 *valp = 64; 4429 return (0); 4430 4431 case _PC_XATTR_EXISTS: 4432 zp = VTOZ(vp); 4433 zfsvfs = zp->z_zfsvfs; 4434 ZFS_ENTER(zfsvfs); 4435 ZFS_VERIFY_ZP(zp); 4436 *valp = 0; 4437 error = zfs_dirent_lock(&dl, zp, "", &xzp, 4438 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 4439 if (error == 0) { 4440 zfs_dirent_unlock(dl); 4441 if (!zfs_dirempty(xzp)) 4442 *valp = 1; 4443 VN_RELE(ZTOV(xzp)); 4444 } else if (error == ENOENT) { 4445 /* 4446 * If there aren't extended attributes, it's the 4447 * same as having zero of them. 4448 */ 4449 error = 0; 4450 } 4451 ZFS_EXIT(zfsvfs); 4452 return (error); 4453 4454 case _PC_SATTR_ENABLED: 4455 case _PC_SATTR_EXISTS: 4456 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 4457 (vp->v_type == VREG || vp->v_type == VDIR); 4458 return (0); 4459 4460 case _PC_ACCESS_FILTERING: 4461 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 4462 vp->v_type == VDIR; 4463 return (0); 4464 4465 case _PC_ACL_ENABLED: 4466 *valp = _ACL_ACE_ENABLED; 4467 return (0); 4468 4469 case _PC_MIN_HOLE_SIZE: 4470 *valp = (ulong_t)SPA_MINBLOCKSIZE; 4471 return (0); 4472 4473 default: 4474 return (fs_pathconf(vp, cmd, valp, cr, ct)); 4475 } 4476 } 4477 4478 /*ARGSUSED*/ 4479 static int 4480 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4481 caller_context_t *ct) 4482 { 4483 znode_t *zp = VTOZ(vp); 4484 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4485 int error; 4486 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4487 4488 ZFS_ENTER(zfsvfs); 4489 ZFS_VERIFY_ZP(zp); 4490 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 4491 ZFS_EXIT(zfsvfs); 4492 4493 return (error); 4494 } 4495 4496 /*ARGSUSED*/ 4497 static int 4498 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4499 caller_context_t *ct) 4500 { 4501 znode_t *zp = VTOZ(vp); 4502 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4503 int error; 4504 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4505 4506 ZFS_ENTER(zfsvfs); 4507 ZFS_VERIFY_ZP(zp); 4508 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 4509 ZFS_EXIT(zfsvfs); 4510 return (error); 4511 } 4512 4513 /* 4514 * Predeclare these here so that the compiler assumes that 4515 * this is an "old style" function declaration that does 4516 * not include arguments => we won't get type mismatch errors 4517 * in the initializations that follow. 4518 */ 4519 static int zfs_inval(); 4520 static int zfs_isdir(); 4521 4522 static int 4523 zfs_inval() 4524 { 4525 return (EINVAL); 4526 } 4527 4528 static int 4529 zfs_isdir() 4530 { 4531 return (EISDIR); 4532 } 4533 /* 4534 * Directory vnode operations template 4535 */ 4536 vnodeops_t *zfs_dvnodeops; 4537 const fs_operation_def_t zfs_dvnodeops_template[] = { 4538 VOPNAME_OPEN, { .vop_open = zfs_open }, 4539 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4540 VOPNAME_READ, { .error = zfs_isdir }, 4541 VOPNAME_WRITE, { .error = zfs_isdir }, 4542 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4543 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4544 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4545 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4546 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4547 VOPNAME_CREATE, { .vop_create = zfs_create }, 4548 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 4549 VOPNAME_LINK, { .vop_link = zfs_link }, 4550 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4551 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 4552 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 4553 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 4554 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 4555 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4556 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4557 VOPNAME_FID, { .vop_fid = zfs_fid }, 4558 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4559 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4560 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4561 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4562 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4563 NULL, NULL 4564 }; 4565 4566 /* 4567 * Regular file vnode operations template 4568 */ 4569 vnodeops_t *zfs_fvnodeops; 4570 const fs_operation_def_t zfs_fvnodeops_template[] = { 4571 VOPNAME_OPEN, { .vop_open = zfs_open }, 4572 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4573 VOPNAME_READ, { .vop_read = zfs_read }, 4574 VOPNAME_WRITE, { .vop_write = zfs_write }, 4575 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4576 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4577 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4578 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4579 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4580 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4581 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4582 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4583 VOPNAME_FID, { .vop_fid = zfs_fid }, 4584 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4585 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 4586 VOPNAME_SPACE, { .vop_space = zfs_space }, 4587 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 4588 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 4589 VOPNAME_MAP, { .vop_map = zfs_map }, 4590 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 4591 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 4592 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4593 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4594 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4595 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4596 NULL, NULL 4597 }; 4598 4599 /* 4600 * Symbolic link vnode operations template 4601 */ 4602 vnodeops_t *zfs_symvnodeops; 4603 const fs_operation_def_t zfs_symvnodeops_template[] = { 4604 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4605 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4606 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4607 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4608 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 4609 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4610 VOPNAME_FID, { .vop_fid = zfs_fid }, 4611 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4612 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4613 NULL, NULL 4614 }; 4615 4616 /* 4617 * special share hidden files vnode operations template 4618 */ 4619 vnodeops_t *zfs_sharevnodeops; 4620 const fs_operation_def_t zfs_sharevnodeops_template[] = { 4621 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4622 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4623 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4624 VOPNAME_FID, { .vop_fid = zfs_fid }, 4625 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4626 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4627 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4628 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4629 NULL, NULL 4630 }; 4631 4632 /* 4633 * Extended attribute directory vnode operations template 4634 * This template is identical to the directory vnodes 4635 * operation template except for restricted operations: 4636 * VOP_MKDIR() 4637 * VOP_SYMLINK() 4638 * Note that there are other restrictions embedded in: 4639 * zfs_create() - restrict type to VREG 4640 * zfs_link() - no links into/out of attribute space 4641 * zfs_rename() - no moves into/out of attribute space 4642 */ 4643 vnodeops_t *zfs_xdvnodeops; 4644 const fs_operation_def_t zfs_xdvnodeops_template[] = { 4645 VOPNAME_OPEN, { .vop_open = zfs_open }, 4646 VOPNAME_CLOSE, { .vop_close = zfs_close }, 4647 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 4648 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 4649 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 4650 VOPNAME_ACCESS, { .vop_access = zfs_access }, 4651 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 4652 VOPNAME_CREATE, { .vop_create = zfs_create }, 4653 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 4654 VOPNAME_LINK, { .vop_link = zfs_link }, 4655 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 4656 VOPNAME_MKDIR, { .error = zfs_inval }, 4657 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 4658 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 4659 VOPNAME_SYMLINK, { .error = zfs_inval }, 4660 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 4661 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4662 VOPNAME_FID, { .vop_fid = zfs_fid }, 4663 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 4664 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4665 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 4666 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 4667 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 4668 NULL, NULL 4669 }; 4670 4671 /* 4672 * Error vnode operations template 4673 */ 4674 vnodeops_t *zfs_evnodeops; 4675 const fs_operation_def_t zfs_evnodeops_template[] = { 4676 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 4677 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 4678 NULL, NULL 4679 }; 4680