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