zfs_vnops.c revision 234607
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 */ 24 25/* Portions Copyright 2007 Jeremy Teo */ 26/* Portions Copyright 2010 Robert Milkowski */ 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/vnode.h> 36#include <sys/file.h> 37#include <sys/stat.h> 38#include <sys/kmem.h> 39#include <sys/taskq.h> 40#include <sys/uio.h> 41#include <sys/atomic.h> 42#include <sys/namei.h> 43#include <sys/mman.h> 44#include <sys/cmn_err.h> 45#include <sys/errno.h> 46#include <sys/unistd.h> 47#include <sys/zfs_dir.h> 48#include <sys/zfs_ioctl.h> 49#include <sys/fs/zfs.h> 50#include <sys/dmu.h> 51#include <sys/dmu_objset.h> 52#include <sys/spa.h> 53#include <sys/txg.h> 54#include <sys/dbuf.h> 55#include <sys/zap.h> 56#include <sys/sa.h> 57#include <sys/dirent.h> 58#include <sys/policy.h> 59#include <sys/sunddi.h> 60#include <sys/filio.h> 61#include <sys/sid.h> 62#include <sys/zfs_ctldir.h> 63#include <sys/zfs_fuid.h> 64#include <sys/zfs_sa.h> 65#include <sys/dnlc.h> 66#include <sys/zfs_rlock.h> 67#include <sys/extdirent.h> 68#include <sys/kidmap.h> 69#include <sys/bio.h> 70#include <sys/buf.h> 71#include <sys/sf_buf.h> 72#include <sys/sched.h> 73#include <sys/acl.h> 74#include <vm/vm_pageout.h> 75 76/* 77 * Programming rules. 78 * 79 * Each vnode op performs some logical unit of work. To do this, the ZPL must 80 * properly lock its in-core state, create a DMU transaction, do the work, 81 * record this work in the intent log (ZIL), commit the DMU transaction, 82 * and wait for the intent log to commit if it is a synchronous operation. 83 * Moreover, the vnode ops must work in both normal and log replay context. 84 * The ordering of events is important to avoid deadlocks and references 85 * to freed memory. The example below illustrates the following Big Rules: 86 * 87 * (1) A check must be made in each zfs thread for a mounted file system. 88 * This is done avoiding races using ZFS_ENTER(zfsvfs). 89 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes 90 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros 91 * can return EIO from the calling function. 92 * 93 * (2) VN_RELE() should always be the last thing except for zil_commit() 94 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 95 * First, if it's the last reference, the vnode/znode 96 * can be freed, so the zp may point to freed memory. Second, the last 97 * reference will call zfs_zinactive(), which may induce a lot of work -- 98 * pushing cached pages (which acquires range locks) and syncing out 99 * cached atime changes. Third, zfs_zinactive() may require a new tx, 100 * which could deadlock the system if you were already holding one. 101 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). 102 * 103 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 104 * as they can span dmu_tx_assign() calls. 105 * 106 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign(). 107 * This is critical because we don't want to block while holding locks. 108 * Note, in particular, that if a lock is sometimes acquired before 109 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 110 * use a non-blocking assign can deadlock the system. The scenario: 111 * 112 * Thread A has grabbed a lock before calling dmu_tx_assign(). 113 * Thread B is in an already-assigned tx, and blocks for this lock. 114 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 115 * forever, because the previous txg can't quiesce until B's tx commits. 116 * 117 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 118 * then drop all locks, call dmu_tx_wait(), and try again. 119 * 120 * (5) If the operation succeeded, generate the intent log entry for it 121 * before dropping locks. This ensures that the ordering of events 122 * in the intent log matches the order in which they actually occurred. 123 * During ZIL replay the zfs_log_* functions will update the sequence 124 * number to indicate the zil transaction has replayed. 125 * 126 * (6) At the end of each vnode op, the DMU tx must always commit, 127 * regardless of whether there were any errors. 128 * 129 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 130 * to ensure that synchronous semantics are provided when necessary. 131 * 132 * In general, this is how things should be ordered in each vnode op: 133 * 134 * ZFS_ENTER(zfsvfs); // exit if unmounted 135 * top: 136 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 137 * rw_enter(...); // grab any other locks you need 138 * tx = dmu_tx_create(...); // get DMU tx 139 * dmu_tx_hold_*(); // hold each object you might modify 140 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign 141 * if (error) { 142 * rw_exit(...); // drop locks 143 * zfs_dirent_unlock(dl); // unlock directory entry 144 * VN_RELE(...); // release held vnodes 145 * if (error == ERESTART) { 146 * dmu_tx_wait(tx); 147 * dmu_tx_abort(tx); 148 * goto top; 149 * } 150 * dmu_tx_abort(tx); // abort DMU tx 151 * ZFS_EXIT(zfsvfs); // finished in zfs 152 * return (error); // really out of space 153 * } 154 * error = do_real_work(); // do whatever this VOP does 155 * if (error == 0) 156 * zfs_log_*(...); // on success, make ZIL entry 157 * dmu_tx_commit(tx); // commit DMU tx -- error or not 158 * rw_exit(...); // drop locks 159 * zfs_dirent_unlock(dl); // unlock directory entry 160 * VN_RELE(...); // release held vnodes 161 * zil_commit(zilog, foid); // synchronous when necessary 162 * ZFS_EXIT(zfsvfs); // finished in zfs 163 * return (error); // done, report error 164 */ 165 166/* ARGSUSED */ 167static int 168zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 169{ 170 znode_t *zp = VTOZ(*vpp); 171 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 172 173 ZFS_ENTER(zfsvfs); 174 ZFS_VERIFY_ZP(zp); 175 176 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 177 ((flag & FAPPEND) == 0)) { 178 ZFS_EXIT(zfsvfs); 179 return (EPERM); 180 } 181 182 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 183 ZTOV(zp)->v_type == VREG && 184 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) { 185 if (fs_vscan(*vpp, cr, 0) != 0) { 186 ZFS_EXIT(zfsvfs); 187 return (EACCES); 188 } 189 } 190 191 /* Keep a count of the synchronous opens in the znode */ 192 if (flag & (FSYNC | FDSYNC)) 193 atomic_inc_32(&zp->z_sync_cnt); 194 195 ZFS_EXIT(zfsvfs); 196 return (0); 197} 198 199/* ARGSUSED */ 200static int 201zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 202 caller_context_t *ct) 203{ 204 znode_t *zp = VTOZ(vp); 205 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 206 207 /* 208 * Clean up any locks held by this process on the vp. 209 */ 210 cleanlocks(vp, ddi_get_pid(), 0); 211 cleanshares(vp, ddi_get_pid()); 212 213 ZFS_ENTER(zfsvfs); 214 ZFS_VERIFY_ZP(zp); 215 216 /* Decrement the synchronous opens in the znode */ 217 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 218 atomic_dec_32(&zp->z_sync_cnt); 219 220 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 221 ZTOV(zp)->v_type == VREG && 222 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) 223 VERIFY(fs_vscan(vp, cr, 1) == 0); 224 225 ZFS_EXIT(zfsvfs); 226 return (0); 227} 228 229/* 230 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 231 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 232 */ 233static int 234zfs_holey(vnode_t *vp, u_long cmd, offset_t *off) 235{ 236 znode_t *zp = VTOZ(vp); 237 uint64_t noff = (uint64_t)*off; /* new offset */ 238 uint64_t file_sz; 239 int error; 240 boolean_t hole; 241 242 file_sz = zp->z_size; 243 if (noff >= file_sz) { 244 return (ENXIO); 245 } 246 247 if (cmd == _FIO_SEEK_HOLE) 248 hole = B_TRUE; 249 else 250 hole = B_FALSE; 251 252 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 253 254 /* end of file? */ 255 if ((error == ESRCH) || (noff > file_sz)) { 256 /* 257 * Handle the virtual hole at the end of file. 258 */ 259 if (hole) { 260 *off = file_sz; 261 return (0); 262 } 263 return (ENXIO); 264 } 265 266 if (noff < *off) 267 return (error); 268 *off = noff; 269 return (error); 270} 271 272/* ARGSUSED */ 273static int 274zfs_ioctl(vnode_t *vp, u_long com, intptr_t data, int flag, cred_t *cred, 275 int *rvalp, caller_context_t *ct) 276{ 277 offset_t off; 278 int error; 279 zfsvfs_t *zfsvfs; 280 znode_t *zp; 281 282 switch (com) { 283 case _FIOFFS: 284 return (0); 285 286 /* 287 * The following two ioctls are used by bfu. Faking out, 288 * necessary to avoid bfu errors. 289 */ 290 case _FIOGDIO: 291 case _FIOSDIO: 292 return (0); 293 294 case _FIO_SEEK_DATA: 295 case _FIO_SEEK_HOLE: 296#ifdef sun 297 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 298 return (EFAULT); 299#else 300 off = *(offset_t *)data; 301#endif 302 zp = VTOZ(vp); 303 zfsvfs = zp->z_zfsvfs; 304 ZFS_ENTER(zfsvfs); 305 ZFS_VERIFY_ZP(zp); 306 307 /* offset parameter is in/out */ 308 error = zfs_holey(vp, com, &off); 309 ZFS_EXIT(zfsvfs); 310 if (error) 311 return (error); 312#ifdef sun 313 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 314 return (EFAULT); 315#else 316 *(offset_t *)data = off; 317#endif 318 return (0); 319 } 320 return (ENOTTY); 321} 322 323static vm_page_t 324page_lookup(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) 325{ 326 vm_object_t obj; 327 vm_page_t pp; 328 329 obj = vp->v_object; 330 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED); 331 332 for (;;) { 333 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 334 vm_page_is_valid(pp, (vm_offset_t)off, nbytes)) { 335 if ((pp->oflags & VPO_BUSY) != 0) { 336 /* 337 * Reference the page before unlocking and 338 * sleeping so that the page daemon is less 339 * likely to reclaim it. 340 */ 341 vm_page_reference(pp); 342 vm_page_sleep(pp, "zfsmwb"); 343 continue; 344 } 345 vm_page_busy(pp); 346 vm_page_undirty(pp); 347 } else { 348 if (vm_page_is_cached(obj, OFF_TO_IDX(start))) 349 vm_page_cache_free(obj, OFF_TO_IDX(start), 350 OFF_TO_IDX(start) + 1); 351 pp = NULL; 352 } 353 break; 354 } 355 return (pp); 356} 357 358static void 359page_unlock(vm_page_t pp) 360{ 361 362 vm_page_wakeup(pp); 363} 364 365static caddr_t 366zfs_map_page(vm_page_t pp, struct sf_buf **sfp) 367{ 368 369 *sfp = sf_buf_alloc(pp, 0); 370 return ((caddr_t)sf_buf_kva(*sfp)); 371} 372 373static void 374zfs_unmap_page(struct sf_buf *sf) 375{ 376 377 sf_buf_free(sf); 378} 379 380/* 381 * When a file is memory mapped, we must keep the IO data synchronized 382 * between the DMU cache and the memory mapped pages. What this means: 383 * 384 * On Write: If we find a memory mapped page, we write to *both* 385 * the page and the dmu buffer. 386 */ 387static void 388update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid, 389 int segflg, dmu_tx_t *tx) 390{ 391 vm_object_t obj; 392 struct sf_buf *sf; 393 int off; 394 395 ASSERT(vp->v_mount != NULL); 396 obj = vp->v_object; 397 ASSERT(obj != NULL); 398 399 off = start & PAGEOFFSET; 400 VM_OBJECT_LOCK(obj); 401 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 402 vm_page_t pp; 403 int nbytes = MIN(PAGESIZE - off, len); 404 405 if ((pp = page_lookup(vp, start, off, nbytes)) != NULL) { 406 caddr_t va; 407 408 VM_OBJECT_UNLOCK(obj); 409 va = zfs_map_page(pp, &sf); 410 if (segflg == UIO_NOCOPY) { 411 (void) dmu_write(os, oid, start+off, nbytes, 412 va+off, tx); 413 } else { 414 (void) dmu_read(os, oid, start+off, nbytes, 415 va+off, DMU_READ_PREFETCH); 416 } 417 zfs_unmap_page(sf); 418 VM_OBJECT_LOCK(obj); 419 page_unlock(pp); 420 } 421 len -= nbytes; 422 off = 0; 423 } 424 VM_OBJECT_UNLOCK(obj); 425} 426 427/* 428 * Read with UIO_NOCOPY flag means that sendfile(2) requests 429 * ZFS to populate a range of page cache pages with data. 430 * 431 * NOTE: this function could be optimized to pre-allocate 432 * all pages in advance, drain VPO_BUSY on all of them, 433 * map them into contiguous KVA region and populate them 434 * in one single dmu_read() call. 435 */ 436static int 437mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio) 438{ 439 znode_t *zp = VTOZ(vp); 440 objset_t *os = zp->z_zfsvfs->z_os; 441 struct sf_buf *sf; 442 vm_object_t obj; 443 vm_page_t pp; 444 int64_t start; 445 caddr_t va; 446 int len = nbytes; 447 int off; 448 int error = 0; 449 450 ASSERT(uio->uio_segflg == UIO_NOCOPY); 451 ASSERT(vp->v_mount != NULL); 452 obj = vp->v_object; 453 ASSERT(obj != NULL); 454 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0); 455 456 VM_OBJECT_LOCK(obj); 457 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) { 458 int bytes = MIN(PAGESIZE, len); 459 460 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_NOBUSY | 461 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_IGN_SBUSY); 462 if (pp->valid == 0) { 463 vm_page_io_start(pp); 464 VM_OBJECT_UNLOCK(obj); 465 va = zfs_map_page(pp, &sf); 466 error = dmu_read(os, zp->z_id, start, bytes, va, 467 DMU_READ_PREFETCH); 468 if (bytes != PAGESIZE && error == 0) 469 bzero(va + bytes, PAGESIZE - bytes); 470 zfs_unmap_page(sf); 471 VM_OBJECT_LOCK(obj); 472 vm_page_io_finish(pp); 473 vm_page_lock(pp); 474 if (error) { 475 vm_page_free(pp); 476 } else { 477 pp->valid = VM_PAGE_BITS_ALL; 478 vm_page_activate(pp); 479 } 480 vm_page_unlock(pp); 481 } 482 if (error) 483 break; 484 uio->uio_resid -= bytes; 485 uio->uio_offset += bytes; 486 len -= bytes; 487 } 488 VM_OBJECT_UNLOCK(obj); 489 return (error); 490} 491 492/* 493 * When a file is memory mapped, we must keep the IO data synchronized 494 * between the DMU cache and the memory mapped pages. What this means: 495 * 496 * On Read: We "read" preferentially from memory mapped pages, 497 * else we default from the dmu buffer. 498 * 499 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 500 * the file is memory mapped. 501 */ 502static int 503mappedread(vnode_t *vp, int nbytes, uio_t *uio) 504{ 505 znode_t *zp = VTOZ(vp); 506 objset_t *os = zp->z_zfsvfs->z_os; 507 vm_object_t obj; 508 int64_t start; 509 caddr_t va; 510 int len = nbytes; 511 int off; 512 int error = 0; 513 514 ASSERT(vp->v_mount != NULL); 515 obj = vp->v_object; 516 ASSERT(obj != NULL); 517 518 start = uio->uio_loffset; 519 off = start & PAGEOFFSET; 520 VM_OBJECT_LOCK(obj); 521 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 522 vm_page_t pp; 523 uint64_t bytes = MIN(PAGESIZE - off, len); 524 525 if (pp = page_lookup(vp, start, off, bytes)) { 526 struct sf_buf *sf; 527 caddr_t va; 528 529 VM_OBJECT_UNLOCK(obj); 530 va = zfs_map_page(pp, &sf); 531 error = uiomove(va + off, bytes, UIO_READ, uio); 532 zfs_unmap_page(sf); 533 VM_OBJECT_LOCK(obj); 534 page_unlock(pp); 535 } else { 536 VM_OBJECT_UNLOCK(obj); 537 error = dmu_read_uio(os, zp->z_id, uio, bytes); 538 VM_OBJECT_LOCK(obj); 539 } 540 len -= bytes; 541 off = 0; 542 if (error) 543 break; 544 } 545 VM_OBJECT_UNLOCK(obj); 546 return (error); 547} 548 549offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 550 551/* 552 * Read bytes from specified file into supplied buffer. 553 * 554 * IN: vp - vnode of file to be read from. 555 * uio - structure supplying read location, range info, 556 * and return buffer. 557 * ioflag - SYNC flags; used to provide FRSYNC semantics. 558 * cr - credentials of caller. 559 * ct - caller context 560 * 561 * OUT: uio - updated offset and range, buffer filled. 562 * 563 * RETURN: 0 if success 564 * error code if failure 565 * 566 * Side Effects: 567 * vp - atime updated if byte count > 0 568 */ 569/* ARGSUSED */ 570static int 571zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 572{ 573 znode_t *zp = VTOZ(vp); 574 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 575 objset_t *os; 576 ssize_t n, nbytes; 577 int error; 578 rl_t *rl; 579 xuio_t *xuio = NULL; 580 581 ZFS_ENTER(zfsvfs); 582 ZFS_VERIFY_ZP(zp); 583 os = zfsvfs->z_os; 584 585 if (zp->z_pflags & ZFS_AV_QUARANTINED) { 586 ZFS_EXIT(zfsvfs); 587 return (EACCES); 588 } 589 590 /* 591 * Validate file offset 592 */ 593 if (uio->uio_loffset < (offset_t)0) { 594 ZFS_EXIT(zfsvfs); 595 return (EINVAL); 596 } 597 598 /* 599 * Fasttrack empty reads 600 */ 601 if (uio->uio_resid == 0) { 602 ZFS_EXIT(zfsvfs); 603 return (0); 604 } 605 606 /* 607 * Check for mandatory locks 608 */ 609 if (MANDMODE(zp->z_mode)) { 610 if (error = chklock(vp, FREAD, 611 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 612 ZFS_EXIT(zfsvfs); 613 return (error); 614 } 615 } 616 617 /* 618 * If we're in FRSYNC mode, sync out this znode before reading it. 619 */ 620 if (zfsvfs->z_log && 621 (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)) 622 zil_commit(zfsvfs->z_log, zp->z_id); 623 624 /* 625 * Lock the range against changes. 626 */ 627 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 628 629 /* 630 * If we are reading past end-of-file we can skip 631 * to the end; but we might still need to set atime. 632 */ 633 if (uio->uio_loffset >= zp->z_size) { 634 error = 0; 635 goto out; 636 } 637 638 ASSERT(uio->uio_loffset < zp->z_size); 639 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset); 640 641#ifdef sun 642 if ((uio->uio_extflg == UIO_XUIO) && 643 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) { 644 int nblk; 645 int blksz = zp->z_blksz; 646 uint64_t offset = uio->uio_loffset; 647 648 xuio = (xuio_t *)uio; 649 if ((ISP2(blksz))) { 650 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset, 651 blksz)) / blksz; 652 } else { 653 ASSERT(offset + n <= blksz); 654 nblk = 1; 655 } 656 (void) dmu_xuio_init(xuio, nblk); 657 658 if (vn_has_cached_data(vp)) { 659 /* 660 * For simplicity, we always allocate a full buffer 661 * even if we only expect to read a portion of a block. 662 */ 663 while (--nblk >= 0) { 664 (void) dmu_xuio_add(xuio, 665 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 666 blksz), 0, blksz); 667 } 668 } 669 } 670#endif /* sun */ 671 672 while (n > 0) { 673 nbytes = MIN(n, zfs_read_chunk_size - 674 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 675 676#ifdef __FreeBSD__ 677 if (uio->uio_segflg == UIO_NOCOPY) 678 error = mappedread_sf(vp, nbytes, uio); 679 else 680#endif /* __FreeBSD__ */ 681 if (vn_has_cached_data(vp)) 682 error = mappedread(vp, nbytes, uio); 683 else 684 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 685 if (error) { 686 /* convert checksum errors into IO errors */ 687 if (error == ECKSUM) 688 error = EIO; 689 break; 690 } 691 692 n -= nbytes; 693 } 694out: 695 zfs_range_unlock(rl); 696 697 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 698 ZFS_EXIT(zfsvfs); 699 return (error); 700} 701 702/* 703 * Write the bytes to a file. 704 * 705 * IN: vp - vnode of file to be written to. 706 * uio - structure supplying write location, range info, 707 * and data buffer. 708 * ioflag - FAPPEND flag set if in append mode. 709 * cr - credentials of caller. 710 * ct - caller context (NFS/CIFS fem monitor only) 711 * 712 * OUT: uio - updated offset and range. 713 * 714 * RETURN: 0 if success 715 * error code if failure 716 * 717 * Timestamps: 718 * vp - ctime|mtime updated if byte count > 0 719 */ 720 721/* ARGSUSED */ 722static int 723zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 724{ 725 znode_t *zp = VTOZ(vp); 726 rlim64_t limit = MAXOFFSET_T; 727 ssize_t start_resid = uio->uio_resid; 728 ssize_t tx_bytes; 729 uint64_t end_size; 730 dmu_tx_t *tx; 731 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 732 zilog_t *zilog; 733 offset_t woff; 734 ssize_t n, nbytes; 735 rl_t *rl; 736 int max_blksz = zfsvfs->z_max_blksz; 737 int error; 738 arc_buf_t *abuf; 739 iovec_t *aiov; 740 xuio_t *xuio = NULL; 741 int i_iov = 0; 742 int iovcnt = uio->uio_iovcnt; 743 iovec_t *iovp = uio->uio_iov; 744 int write_eof; 745 int count = 0; 746 sa_bulk_attr_t bulk[4]; 747 uint64_t mtime[2], ctime[2]; 748 749 /* 750 * Fasttrack empty write 751 */ 752 n = start_resid; 753 if (n == 0) 754 return (0); 755 756 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 757 limit = MAXOFFSET_T; 758 759 ZFS_ENTER(zfsvfs); 760 ZFS_VERIFY_ZP(zp); 761 762 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 763 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 764 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 765 &zp->z_size, 8); 766 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 767 &zp->z_pflags, 8); 768 769 /* 770 * If immutable or not appending then return EPERM 771 */ 772 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 773 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 774 (uio->uio_loffset < zp->z_size))) { 775 ZFS_EXIT(zfsvfs); 776 return (EPERM); 777 } 778 779 zilog = zfsvfs->z_log; 780 781 /* 782 * Validate file offset 783 */ 784 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset; 785 if (woff < 0) { 786 ZFS_EXIT(zfsvfs); 787 return (EINVAL); 788 } 789 790 /* 791 * Check for mandatory locks before calling zfs_range_lock() 792 * in order to prevent a deadlock with locks set via fcntl(). 793 */ 794 if (MANDMODE((mode_t)zp->z_mode) && 795 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 796 ZFS_EXIT(zfsvfs); 797 return (error); 798 } 799 800#ifdef sun 801 /* 802 * Pre-fault the pages to ensure slow (eg NFS) pages 803 * don't hold up txg. 804 * Skip this if uio contains loaned arc_buf. 805 */ 806 if ((uio->uio_extflg == UIO_XUIO) && 807 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) 808 xuio = (xuio_t *)uio; 809 else 810 uio_prefaultpages(MIN(n, max_blksz), uio); 811#endif /* sun */ 812 813 /* 814 * If in append mode, set the io offset pointer to eof. 815 */ 816 if (ioflag & FAPPEND) { 817 /* 818 * Obtain an appending range lock to guarantee file append 819 * semantics. We reset the write offset once we have the lock. 820 */ 821 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 822 woff = rl->r_off; 823 if (rl->r_len == UINT64_MAX) { 824 /* 825 * We overlocked the file because this write will cause 826 * the file block size to increase. 827 * Note that zp_size cannot change with this lock held. 828 */ 829 woff = zp->z_size; 830 } 831 uio->uio_loffset = woff; 832 } else { 833 /* 834 * Note that if the file block size will change as a result of 835 * this write, then this range lock will lock the entire file 836 * so that we can re-write the block safely. 837 */ 838 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 839 } 840 841 if (woff >= limit) { 842 zfs_range_unlock(rl); 843 ZFS_EXIT(zfsvfs); 844 return (EFBIG); 845 } 846 847 if ((woff + n) > limit || woff > (limit - n)) 848 n = limit - woff; 849 850 /* Will this write extend the file length? */ 851 write_eof = (woff + n > zp->z_size); 852 853 end_size = MAX(zp->z_size, woff + n); 854 855 /* 856 * Write the file in reasonable size chunks. Each chunk is written 857 * in a separate transaction; this keeps the intent log records small 858 * and allows us to do more fine-grained space accounting. 859 */ 860 while (n > 0) { 861 abuf = NULL; 862 woff = uio->uio_loffset; 863again: 864 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 865 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 866 if (abuf != NULL) 867 dmu_return_arcbuf(abuf); 868 error = EDQUOT; 869 break; 870 } 871 872 if (xuio && abuf == NULL) { 873 ASSERT(i_iov < iovcnt); 874 aiov = &iovp[i_iov]; 875 abuf = dmu_xuio_arcbuf(xuio, i_iov); 876 dmu_xuio_clear(xuio, i_iov); 877 DTRACE_PROBE3(zfs_cp_write, int, i_iov, 878 iovec_t *, aiov, arc_buf_t *, abuf); 879 ASSERT((aiov->iov_base == abuf->b_data) || 880 ((char *)aiov->iov_base - (char *)abuf->b_data + 881 aiov->iov_len == arc_buf_size(abuf))); 882 i_iov++; 883 } else if (abuf == NULL && n >= max_blksz && 884 woff >= zp->z_size && 885 P2PHASE(woff, max_blksz) == 0 && 886 zp->z_blksz == max_blksz) { 887 /* 888 * This write covers a full block. "Borrow" a buffer 889 * from the dmu so that we can fill it before we enter 890 * a transaction. This avoids the possibility of 891 * holding up the transaction if the data copy hangs 892 * up on a pagefault (e.g., from an NFS server mapping). 893 */ 894 size_t cbytes; 895 896 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 897 max_blksz); 898 ASSERT(abuf != NULL); 899 ASSERT(arc_buf_size(abuf) == max_blksz); 900 if (error = uiocopy(abuf->b_data, max_blksz, 901 UIO_WRITE, uio, &cbytes)) { 902 dmu_return_arcbuf(abuf); 903 break; 904 } 905 ASSERT(cbytes == max_blksz); 906 } 907 908 /* 909 * Start a transaction. 910 */ 911 tx = dmu_tx_create(zfsvfs->z_os); 912 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 913 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 914 zfs_sa_upgrade_txholds(tx, zp); 915 error = dmu_tx_assign(tx, TXG_NOWAIT); 916 if (error) { 917 if (error == ERESTART) { 918 dmu_tx_wait(tx); 919 dmu_tx_abort(tx); 920 goto again; 921 } 922 dmu_tx_abort(tx); 923 if (abuf != NULL) 924 dmu_return_arcbuf(abuf); 925 break; 926 } 927 928 /* 929 * If zfs_range_lock() over-locked we grow the blocksize 930 * and then reduce the lock range. This will only happen 931 * on the first iteration since zfs_range_reduce() will 932 * shrink down r_len to the appropriate size. 933 */ 934 if (rl->r_len == UINT64_MAX) { 935 uint64_t new_blksz; 936 937 if (zp->z_blksz > max_blksz) { 938 ASSERT(!ISP2(zp->z_blksz)); 939 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 940 } else { 941 new_blksz = MIN(end_size, max_blksz); 942 } 943 zfs_grow_blocksize(zp, new_blksz, tx); 944 zfs_range_reduce(rl, woff, n); 945 } 946 947 /* 948 * XXX - should we really limit each write to z_max_blksz? 949 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 950 */ 951 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 952 953 if (woff + nbytes > zp->z_size) 954 vnode_pager_setsize(vp, woff + nbytes); 955 956 if (abuf == NULL) { 957 tx_bytes = uio->uio_resid; 958 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), 959 uio, nbytes, tx); 960 tx_bytes -= uio->uio_resid; 961 } else { 962 tx_bytes = nbytes; 963 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len); 964 /* 965 * If this is not a full block write, but we are 966 * extending the file past EOF and this data starts 967 * block-aligned, use assign_arcbuf(). Otherwise, 968 * write via dmu_write(). 969 */ 970 if (tx_bytes < max_blksz && (!write_eof || 971 aiov->iov_base != abuf->b_data)) { 972 ASSERT(xuio); 973 dmu_write(zfsvfs->z_os, zp->z_id, woff, 974 aiov->iov_len, aiov->iov_base, tx); 975 dmu_return_arcbuf(abuf); 976 xuio_stat_wbuf_copied(); 977 } else { 978 ASSERT(xuio || tx_bytes == max_blksz); 979 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), 980 woff, abuf, tx); 981 } 982 ASSERT(tx_bytes <= uio->uio_resid); 983 uioskip(uio, tx_bytes); 984 } 985 if (tx_bytes && vn_has_cached_data(vp)) { 986 update_pages(vp, woff, tx_bytes, zfsvfs->z_os, 987 zp->z_id, uio->uio_segflg, tx); 988 } 989 990 /* 991 * If we made no progress, we're done. If we made even 992 * partial progress, update the znode and ZIL accordingly. 993 */ 994 if (tx_bytes == 0) { 995 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 996 (void *)&zp->z_size, sizeof (uint64_t), tx); 997 dmu_tx_commit(tx); 998 ASSERT(error != 0); 999 break; 1000 } 1001 1002 /* 1003 * Clear Set-UID/Set-GID bits on successful write if not 1004 * privileged and at least one of the excute bits is set. 1005 * 1006 * It would be nice to to this after all writes have 1007 * been done, but that would still expose the ISUID/ISGID 1008 * to another app after the partial write is committed. 1009 * 1010 * Note: we don't call zfs_fuid_map_id() here because 1011 * user 0 is not an ephemeral uid. 1012 */ 1013 mutex_enter(&zp->z_acl_lock); 1014 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | 1015 (S_IXUSR >> 6))) != 0 && 1016 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && 1017 secpolicy_vnode_setid_retain(vp, cr, 1018 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) { 1019 uint64_t newmode; 1020 zp->z_mode &= ~(S_ISUID | S_ISGID); 1021 newmode = zp->z_mode; 1022 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), 1023 (void *)&newmode, sizeof (uint64_t), tx); 1024 } 1025 mutex_exit(&zp->z_acl_lock); 1026 1027 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 1028 B_TRUE); 1029 1030 /* 1031 * Update the file size (zp_size) if it has changed; 1032 * account for possible concurrent updates. 1033 */ 1034 while ((end_size = zp->z_size) < uio->uio_loffset) { 1035 (void) atomic_cas_64(&zp->z_size, end_size, 1036 uio->uio_loffset); 1037 ASSERT(error == 0); 1038 } 1039 /* 1040 * If we are replaying and eof is non zero then force 1041 * the file size to the specified eof. Note, there's no 1042 * concurrency during replay. 1043 */ 1044 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) 1045 zp->z_size = zfsvfs->z_replay_eof; 1046 1047 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1048 1049 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 1050 dmu_tx_commit(tx); 1051 1052 if (error != 0) 1053 break; 1054 ASSERT(tx_bytes == nbytes); 1055 n -= nbytes; 1056 1057#ifdef sun 1058 if (!xuio && n > 0) 1059 uio_prefaultpages(MIN(n, max_blksz), uio); 1060#endif /* sun */ 1061 } 1062 1063 zfs_range_unlock(rl); 1064 1065 /* 1066 * If we're in replay mode, or we made no progress, return error. 1067 * Otherwise, it's at least a partial write, so it's successful. 1068 */ 1069 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 1070 ZFS_EXIT(zfsvfs); 1071 return (error); 1072 } 1073 1074 if (ioflag & (FSYNC | FDSYNC) || 1075 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1076 zil_commit(zilog, zp->z_id); 1077 1078 ZFS_EXIT(zfsvfs); 1079 return (0); 1080} 1081 1082void 1083zfs_get_done(zgd_t *zgd, int error) 1084{ 1085 znode_t *zp = zgd->zgd_private; 1086 objset_t *os = zp->z_zfsvfs->z_os; 1087 int vfslocked; 1088 1089 if (zgd->zgd_db) 1090 dmu_buf_rele(zgd->zgd_db, zgd); 1091 1092 zfs_range_unlock(zgd->zgd_rl); 1093 1094 vfslocked = VFS_LOCK_GIANT(zp->z_zfsvfs->z_vfs); 1095 /* 1096 * Release the vnode asynchronously as we currently have the 1097 * txg stopped from syncing. 1098 */ 1099 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1100 1101 if (error == 0 && zgd->zgd_bp) 1102 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1103 1104 kmem_free(zgd, sizeof (zgd_t)); 1105 VFS_UNLOCK_GIANT(vfslocked); 1106} 1107 1108#ifdef DEBUG 1109static int zil_fault_io = 0; 1110#endif 1111 1112/* 1113 * Get data to generate a TX_WRITE intent log record. 1114 */ 1115int 1116zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1117{ 1118 zfsvfs_t *zfsvfs = arg; 1119 objset_t *os = zfsvfs->z_os; 1120 znode_t *zp; 1121 uint64_t object = lr->lr_foid; 1122 uint64_t offset = lr->lr_offset; 1123 uint64_t size = lr->lr_length; 1124 blkptr_t *bp = &lr->lr_blkptr; 1125 dmu_buf_t *db; 1126 zgd_t *zgd; 1127 int error = 0; 1128 1129 ASSERT(zio != NULL); 1130 ASSERT(size != 0); 1131 1132 /* 1133 * Nothing to do if the file has been removed 1134 */ 1135 if (zfs_zget(zfsvfs, object, &zp) != 0) 1136 return (ENOENT); 1137 if (zp->z_unlinked) { 1138 /* 1139 * Release the vnode asynchronously as we currently have the 1140 * txg stopped from syncing. 1141 */ 1142 VN_RELE_ASYNC(ZTOV(zp), 1143 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1144 return (ENOENT); 1145 } 1146 1147 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1148 zgd->zgd_zilog = zfsvfs->z_log; 1149 zgd->zgd_private = zp; 1150 1151 /* 1152 * Write records come in two flavors: immediate and indirect. 1153 * For small writes it's cheaper to store the data with the 1154 * log record (immediate); for large writes it's cheaper to 1155 * sync the data and get a pointer to it (indirect) so that 1156 * we don't have to write the data twice. 1157 */ 1158 if (buf != NULL) { /* immediate write */ 1159 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER); 1160 /* test for truncation needs to be done while range locked */ 1161 if (offset >= zp->z_size) { 1162 error = ENOENT; 1163 } else { 1164 error = dmu_read(os, object, offset, size, buf, 1165 DMU_READ_NO_PREFETCH); 1166 } 1167 ASSERT(error == 0 || error == ENOENT); 1168 } else { /* indirect write */ 1169 /* 1170 * Have to lock the whole block to ensure when it's 1171 * written out and it's checksum is being calculated 1172 * that no one can change the data. We need to re-check 1173 * blocksize after we get the lock in case it's changed! 1174 */ 1175 for (;;) { 1176 uint64_t blkoff; 1177 size = zp->z_blksz; 1178 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; 1179 offset -= blkoff; 1180 zgd->zgd_rl = zfs_range_lock(zp, offset, size, 1181 RL_READER); 1182 if (zp->z_blksz == size) 1183 break; 1184 offset += blkoff; 1185 zfs_range_unlock(zgd->zgd_rl); 1186 } 1187 /* test for truncation needs to be done while range locked */ 1188 if (lr->lr_offset >= zp->z_size) 1189 error = ENOENT; 1190#ifdef DEBUG 1191 if (zil_fault_io) { 1192 error = EIO; 1193 zil_fault_io = 0; 1194 } 1195#endif 1196 if (error == 0) 1197 error = dmu_buf_hold(os, object, offset, zgd, &db, 1198 DMU_READ_NO_PREFETCH); 1199 1200 if (error == 0) { 1201 zgd->zgd_db = db; 1202 zgd->zgd_bp = bp; 1203 1204 ASSERT(db->db_offset == offset); 1205 ASSERT(db->db_size == size); 1206 1207 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1208 zfs_get_done, zgd); 1209 ASSERT(error || lr->lr_length <= zp->z_blksz); 1210 1211 /* 1212 * On success, we need to wait for the write I/O 1213 * initiated by dmu_sync() to complete before we can 1214 * release this dbuf. We will finish everything up 1215 * in the zfs_get_done() callback. 1216 */ 1217 if (error == 0) 1218 return (0); 1219 1220 if (error == EALREADY) { 1221 lr->lr_common.lrc_txtype = TX_WRITE2; 1222 error = 0; 1223 } 1224 } 1225 } 1226 1227 zfs_get_done(zgd, error); 1228 1229 return (error); 1230} 1231 1232/*ARGSUSED*/ 1233static int 1234zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 1235 caller_context_t *ct) 1236{ 1237 znode_t *zp = VTOZ(vp); 1238 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1239 int error; 1240 1241 ZFS_ENTER(zfsvfs); 1242 ZFS_VERIFY_ZP(zp); 1243 1244 if (flag & V_ACE_MASK) 1245 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 1246 else 1247 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1248 1249 ZFS_EXIT(zfsvfs); 1250 return (error); 1251} 1252 1253/* 1254 * If vnode is for a device return a specfs vnode instead. 1255 */ 1256static int 1257specvp_check(vnode_t **vpp, cred_t *cr) 1258{ 1259 int error = 0; 1260 1261 if (IS_DEVVP(*vpp)) { 1262 struct vnode *svp; 1263 1264 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1265 VN_RELE(*vpp); 1266 if (svp == NULL) 1267 error = ENOSYS; 1268 *vpp = svp; 1269 } 1270 return (error); 1271} 1272 1273 1274/* 1275 * Lookup an entry in a directory, or an extended attribute directory. 1276 * If it exists, return a held vnode reference for it. 1277 * 1278 * IN: dvp - vnode of directory to search. 1279 * nm - name of entry to lookup. 1280 * pnp - full pathname to lookup [UNUSED]. 1281 * flags - LOOKUP_XATTR set if looking for an attribute. 1282 * rdir - root directory vnode [UNUSED]. 1283 * cr - credentials of caller. 1284 * ct - caller context 1285 * direntflags - directory lookup flags 1286 * realpnp - returned pathname. 1287 * 1288 * OUT: vpp - vnode of located entry, NULL if not found. 1289 * 1290 * RETURN: 0 if success 1291 * error code if failure 1292 * 1293 * Timestamps: 1294 * NA 1295 */ 1296/* ARGSUSED */ 1297static int 1298zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 1299 int nameiop, cred_t *cr, kthread_t *td, int flags) 1300{ 1301 znode_t *zdp = VTOZ(dvp); 1302 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1303 int error = 0; 1304 int *direntflags = NULL; 1305 void *realpnp = NULL; 1306 1307 /* fast path */ 1308 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 1309 1310 if (dvp->v_type != VDIR) { 1311 return (ENOTDIR); 1312 } else if (zdp->z_sa_hdl == NULL) { 1313 return (EIO); 1314 } 1315 1316 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 1317 error = zfs_fastaccesschk_execute(zdp, cr); 1318 if (!error) { 1319 *vpp = dvp; 1320 VN_HOLD(*vpp); 1321 return (0); 1322 } 1323 return (error); 1324 } else { 1325 vnode_t *tvp = dnlc_lookup(dvp, nm); 1326 1327 if (tvp) { 1328 error = zfs_fastaccesschk_execute(zdp, cr); 1329 if (error) { 1330 VN_RELE(tvp); 1331 return (error); 1332 } 1333 if (tvp == DNLC_NO_VNODE) { 1334 VN_RELE(tvp); 1335 return (ENOENT); 1336 } else { 1337 *vpp = tvp; 1338 return (specvp_check(vpp, cr)); 1339 } 1340 } 1341 } 1342 } 1343 1344 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1345 1346 ZFS_ENTER(zfsvfs); 1347 ZFS_VERIFY_ZP(zdp); 1348 1349 *vpp = NULL; 1350 1351 if (flags & LOOKUP_XATTR) { 1352#ifdef TODO 1353 /* 1354 * If the xattr property is off, refuse the lookup request. 1355 */ 1356 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1357 ZFS_EXIT(zfsvfs); 1358 return (EINVAL); 1359 } 1360#endif 1361 1362 /* 1363 * We don't allow recursive attributes.. 1364 * Maybe someday we will. 1365 */ 1366 if (zdp->z_pflags & ZFS_XATTR) { 1367 ZFS_EXIT(zfsvfs); 1368 return (EINVAL); 1369 } 1370 1371 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1372 ZFS_EXIT(zfsvfs); 1373 return (error); 1374 } 1375 1376 /* 1377 * Do we have permission to get into attribute directory? 1378 */ 1379 1380 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1381 B_FALSE, cr)) { 1382 VN_RELE(*vpp); 1383 *vpp = NULL; 1384 } 1385 1386 ZFS_EXIT(zfsvfs); 1387 return (error); 1388 } 1389 1390 if (dvp->v_type != VDIR) { 1391 ZFS_EXIT(zfsvfs); 1392 return (ENOTDIR); 1393 } 1394 1395 /* 1396 * Check accessibility of directory. 1397 */ 1398 1399 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1400 ZFS_EXIT(zfsvfs); 1401 return (error); 1402 } 1403 1404 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1405 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1406 ZFS_EXIT(zfsvfs); 1407 return (EILSEQ); 1408 } 1409 1410 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1411 if (error == 0) 1412 error = specvp_check(vpp, cr); 1413 1414 /* Translate errors and add SAVENAME when needed. */ 1415 if (cnp->cn_flags & ISLASTCN) { 1416 switch (nameiop) { 1417 case CREATE: 1418 case RENAME: 1419 if (error == ENOENT) { 1420 error = EJUSTRETURN; 1421 cnp->cn_flags |= SAVENAME; 1422 break; 1423 } 1424 /* FALLTHROUGH */ 1425 case DELETE: 1426 if (error == 0) 1427 cnp->cn_flags |= SAVENAME; 1428 break; 1429 } 1430 } 1431 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) { 1432 int ltype = 0; 1433 1434 if (cnp->cn_flags & ISDOTDOT) { 1435 ltype = VOP_ISLOCKED(dvp); 1436 VOP_UNLOCK(dvp, 0); 1437 } 1438 ZFS_EXIT(zfsvfs); 1439 error = zfs_vnode_lock(*vpp, cnp->cn_lkflags); 1440 if (cnp->cn_flags & ISDOTDOT) 1441 vn_lock(dvp, ltype | LK_RETRY); 1442 if (error != 0) { 1443 VN_RELE(*vpp); 1444 *vpp = NULL; 1445 return (error); 1446 } 1447 } else { 1448 ZFS_EXIT(zfsvfs); 1449 } 1450 1451#ifdef FREEBSD_NAMECACHE 1452 /* 1453 * Insert name into cache (as non-existent) if appropriate. 1454 */ 1455 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE) 1456 cache_enter(dvp, *vpp, cnp); 1457 /* 1458 * Insert name into cache if appropriate. 1459 */ 1460 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1461 if (!(cnp->cn_flags & ISLASTCN) || 1462 (nameiop != DELETE && nameiop != RENAME)) { 1463 cache_enter(dvp, *vpp, cnp); 1464 } 1465 } 1466#endif 1467 1468 return (error); 1469} 1470 1471/* 1472 * Attempt to create a new entry in a directory. If the entry 1473 * already exists, truncate the file if permissible, else return 1474 * an error. Return the vp of the created or trunc'd file. 1475 * 1476 * IN: dvp - vnode of directory to put new file entry in. 1477 * name - name of new file entry. 1478 * vap - attributes of new file. 1479 * excl - flag indicating exclusive or non-exclusive mode. 1480 * mode - mode to open file with. 1481 * cr - credentials of caller. 1482 * flag - large file flag [UNUSED]. 1483 * ct - caller context 1484 * vsecp - ACL to be set 1485 * 1486 * OUT: vpp - vnode of created or trunc'd entry. 1487 * 1488 * RETURN: 0 if success 1489 * error code if failure 1490 * 1491 * Timestamps: 1492 * dvp - ctime|mtime updated if new entry created 1493 * vp - ctime|mtime always, atime if new 1494 */ 1495 1496/* ARGSUSED */ 1497static int 1498zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1499 vnode_t **vpp, cred_t *cr, kthread_t *td) 1500{ 1501 znode_t *zp, *dzp = VTOZ(dvp); 1502 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1503 zilog_t *zilog; 1504 objset_t *os; 1505 zfs_dirlock_t *dl; 1506 dmu_tx_t *tx; 1507 int error; 1508 ksid_t *ksid; 1509 uid_t uid; 1510 gid_t gid = crgetgid(cr); 1511 zfs_acl_ids_t acl_ids; 1512 boolean_t fuid_dirtied; 1513 boolean_t have_acl = B_FALSE; 1514 void *vsecp = NULL; 1515 int flag = 0; 1516 1517 /* 1518 * If we have an ephemeral id, ACL, or XVATTR then 1519 * make sure file system is at proper version 1520 */ 1521 1522 ksid = crgetsid(cr, KSID_OWNER); 1523 if (ksid) 1524 uid = ksid_getid(ksid); 1525 else 1526 uid = crgetuid(cr); 1527 1528 if (zfsvfs->z_use_fuids == B_FALSE && 1529 (vsecp || (vap->va_mask & AT_XVATTR) || 1530 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1531 return (EINVAL); 1532 1533 ZFS_ENTER(zfsvfs); 1534 ZFS_VERIFY_ZP(dzp); 1535 os = zfsvfs->z_os; 1536 zilog = zfsvfs->z_log; 1537 1538 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1539 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1540 ZFS_EXIT(zfsvfs); 1541 return (EILSEQ); 1542 } 1543 1544 if (vap->va_mask & AT_XVATTR) { 1545 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 1546 crgetuid(cr), cr, vap->va_type)) != 0) { 1547 ZFS_EXIT(zfsvfs); 1548 return (error); 1549 } 1550 } 1551top: 1552 *vpp = NULL; 1553 1554 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) 1555 vap->va_mode &= ~S_ISVTX; 1556 1557 if (*name == '\0') { 1558 /* 1559 * Null component name refers to the directory itself. 1560 */ 1561 VN_HOLD(dvp); 1562 zp = dzp; 1563 dl = NULL; 1564 error = 0; 1565 } else { 1566 /* possible VN_HOLD(zp) */ 1567 int zflg = 0; 1568 1569 if (flag & FIGNORECASE) 1570 zflg |= ZCILOOK; 1571 1572 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1573 NULL, NULL); 1574 if (error) { 1575 if (have_acl) 1576 zfs_acl_ids_free(&acl_ids); 1577 if (strcmp(name, "..") == 0) 1578 error = EISDIR; 1579 ZFS_EXIT(zfsvfs); 1580 return (error); 1581 } 1582 } 1583 1584 if (zp == NULL) { 1585 uint64_t txtype; 1586 1587 /* 1588 * Create a new file object and update the directory 1589 * to reference it. 1590 */ 1591 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1592 if (have_acl) 1593 zfs_acl_ids_free(&acl_ids); 1594 goto out; 1595 } 1596 1597 /* 1598 * We only support the creation of regular files in 1599 * extended attribute directories. 1600 */ 1601 1602 if ((dzp->z_pflags & ZFS_XATTR) && 1603 (vap->va_type != VREG)) { 1604 if (have_acl) 1605 zfs_acl_ids_free(&acl_ids); 1606 error = EINVAL; 1607 goto out; 1608 } 1609 1610 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 1611 cr, vsecp, &acl_ids)) != 0) 1612 goto out; 1613 have_acl = B_TRUE; 1614 1615 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1616 zfs_acl_ids_free(&acl_ids); 1617 error = EDQUOT; 1618 goto out; 1619 } 1620 1621 tx = dmu_tx_create(os); 1622 1623 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1624 ZFS_SA_BASE_ATTR_SIZE); 1625 1626 fuid_dirtied = zfsvfs->z_fuid_dirty; 1627 if (fuid_dirtied) 1628 zfs_fuid_txhold(zfsvfs, tx); 1629 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1630 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1631 if (!zfsvfs->z_use_sa && 1632 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1633 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1634 0, acl_ids.z_aclp->z_acl_bytes); 1635 } 1636 error = dmu_tx_assign(tx, TXG_NOWAIT); 1637 if (error) { 1638 zfs_dirent_unlock(dl); 1639 if (error == ERESTART) { 1640 dmu_tx_wait(tx); 1641 dmu_tx_abort(tx); 1642 goto top; 1643 } 1644 zfs_acl_ids_free(&acl_ids); 1645 dmu_tx_abort(tx); 1646 ZFS_EXIT(zfsvfs); 1647 return (error); 1648 } 1649 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1650 1651 if (fuid_dirtied) 1652 zfs_fuid_sync(zfsvfs, tx); 1653 1654 (void) zfs_link_create(dl, zp, tx, ZNEW); 1655 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1656 if (flag & FIGNORECASE) 1657 txtype |= TX_CI; 1658 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1659 vsecp, acl_ids.z_fuidp, vap); 1660 zfs_acl_ids_free(&acl_ids); 1661 dmu_tx_commit(tx); 1662 } else { 1663 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1664 1665 if (have_acl) 1666 zfs_acl_ids_free(&acl_ids); 1667 have_acl = B_FALSE; 1668 1669 /* 1670 * A directory entry already exists for this name. 1671 */ 1672 /* 1673 * Can't truncate an existing file if in exclusive mode. 1674 */ 1675 if (excl == EXCL) { 1676 error = EEXIST; 1677 goto out; 1678 } 1679 /* 1680 * Can't open a directory for writing. 1681 */ 1682 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1683 error = EISDIR; 1684 goto out; 1685 } 1686 /* 1687 * Verify requested access to file. 1688 */ 1689 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1690 goto out; 1691 } 1692 1693 mutex_enter(&dzp->z_lock); 1694 dzp->z_seq++; 1695 mutex_exit(&dzp->z_lock); 1696 1697 /* 1698 * Truncate regular files if requested. 1699 */ 1700 if ((ZTOV(zp)->v_type == VREG) && 1701 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1702 /* we can't hold any locks when calling zfs_freesp() */ 1703 zfs_dirent_unlock(dl); 1704 dl = NULL; 1705 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1706 if (error == 0) { 1707 vnevent_create(ZTOV(zp), ct); 1708 } 1709 } 1710 } 1711out: 1712 if (dl) 1713 zfs_dirent_unlock(dl); 1714 1715 if (error) { 1716 if (zp) 1717 VN_RELE(ZTOV(zp)); 1718 } else { 1719 *vpp = ZTOV(zp); 1720 error = specvp_check(vpp, cr); 1721 } 1722 1723 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1724 zil_commit(zilog, 0); 1725 1726 ZFS_EXIT(zfsvfs); 1727 return (error); 1728} 1729 1730/* 1731 * Remove an entry from a directory. 1732 * 1733 * IN: dvp - vnode of directory to remove entry from. 1734 * name - name of entry to remove. 1735 * cr - credentials of caller. 1736 * ct - caller context 1737 * flags - case flags 1738 * 1739 * RETURN: 0 if success 1740 * error code if failure 1741 * 1742 * Timestamps: 1743 * dvp - ctime|mtime 1744 * vp - ctime (if nlink > 0) 1745 */ 1746 1747uint64_t null_xattr = 0; 1748 1749/*ARGSUSED*/ 1750static int 1751zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1752 int flags) 1753{ 1754 znode_t *zp, *dzp = VTOZ(dvp); 1755 znode_t *xzp; 1756 vnode_t *vp; 1757 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1758 zilog_t *zilog; 1759 uint64_t acl_obj, xattr_obj; 1760 uint64_t xattr_obj_unlinked = 0; 1761 uint64_t obj = 0; 1762 zfs_dirlock_t *dl; 1763 dmu_tx_t *tx; 1764 boolean_t may_delete_now, delete_now = FALSE; 1765 boolean_t unlinked, toobig = FALSE; 1766 uint64_t txtype; 1767 pathname_t *realnmp = NULL; 1768 pathname_t realnm; 1769 int error; 1770 int zflg = ZEXISTS; 1771 1772 ZFS_ENTER(zfsvfs); 1773 ZFS_VERIFY_ZP(dzp); 1774 zilog = zfsvfs->z_log; 1775 1776 if (flags & FIGNORECASE) { 1777 zflg |= ZCILOOK; 1778 pn_alloc(&realnm); 1779 realnmp = &realnm; 1780 } 1781 1782top: 1783 xattr_obj = 0; 1784 xzp = NULL; 1785 /* 1786 * Attempt to lock directory; fail if entry doesn't exist. 1787 */ 1788 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1789 NULL, realnmp)) { 1790 if (realnmp) 1791 pn_free(realnmp); 1792 ZFS_EXIT(zfsvfs); 1793 return (error); 1794 } 1795 1796 vp = ZTOV(zp); 1797 1798 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1799 goto out; 1800 } 1801 1802 /* 1803 * Need to use rmdir for removing directories. 1804 */ 1805 if (vp->v_type == VDIR) { 1806 error = EPERM; 1807 goto out; 1808 } 1809 1810 vnevent_remove(vp, dvp, name, ct); 1811 1812 if (realnmp) 1813 dnlc_remove(dvp, realnmp->pn_buf); 1814 else 1815 dnlc_remove(dvp, name); 1816 1817 VI_LOCK(vp); 1818 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1819 VI_UNLOCK(vp); 1820 1821 /* 1822 * We may delete the znode now, or we may put it in the unlinked set; 1823 * it depends on whether we're the last link, and on whether there are 1824 * other holds on the vnode. So we dmu_tx_hold() the right things to 1825 * allow for either case. 1826 */ 1827 obj = zp->z_id; 1828 tx = dmu_tx_create(zfsvfs->z_os); 1829 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1830 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1831 zfs_sa_upgrade_txholds(tx, zp); 1832 zfs_sa_upgrade_txholds(tx, dzp); 1833 if (may_delete_now) { 1834 toobig = 1835 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 1836 /* if the file is too big, only hold_free a token amount */ 1837 dmu_tx_hold_free(tx, zp->z_id, 0, 1838 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1839 } 1840 1841 /* are there any extended attributes? */ 1842 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1843 &xattr_obj, sizeof (xattr_obj)); 1844 if (error == 0 && xattr_obj) { 1845 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1846 ASSERT3U(error, ==, 0); 1847 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1848 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1849 } 1850 1851 mutex_enter(&zp->z_lock); 1852 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now) 1853 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1854 mutex_exit(&zp->z_lock); 1855 1856 /* charge as an update -- would be nice not to charge at all */ 1857 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1858 1859 error = dmu_tx_assign(tx, TXG_NOWAIT); 1860 if (error) { 1861 zfs_dirent_unlock(dl); 1862 VN_RELE(vp); 1863 if (xzp) 1864 VN_RELE(ZTOV(xzp)); 1865 if (error == ERESTART) { 1866 dmu_tx_wait(tx); 1867 dmu_tx_abort(tx); 1868 goto top; 1869 } 1870 if (realnmp) 1871 pn_free(realnmp); 1872 dmu_tx_abort(tx); 1873 ZFS_EXIT(zfsvfs); 1874 return (error); 1875 } 1876 1877 /* 1878 * Remove the directory entry. 1879 */ 1880 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1881 1882 if (error) { 1883 dmu_tx_commit(tx); 1884 goto out; 1885 } 1886 1887 if (unlinked) { 1888 1889 /* 1890 * Hold z_lock so that we can make sure that the ACL obj 1891 * hasn't changed. Could have been deleted due to 1892 * zfs_sa_upgrade(). 1893 */ 1894 mutex_enter(&zp->z_lock); 1895 VI_LOCK(vp); 1896 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1897 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked)); 1898 delete_now = may_delete_now && !toobig && 1899 vp->v_count == 1 && !vn_has_cached_data(vp) && 1900 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) == 1901 acl_obj; 1902 VI_UNLOCK(vp); 1903 } 1904 1905 if (delete_now) { 1906 if (xattr_obj_unlinked) { 1907 ASSERT3U(xzp->z_links, ==, 2); 1908 mutex_enter(&xzp->z_lock); 1909 xzp->z_unlinked = 1; 1910 xzp->z_links = 0; 1911 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 1912 &xzp->z_links, sizeof (xzp->z_links), tx); 1913 ASSERT3U(error, ==, 0); 1914 mutex_exit(&xzp->z_lock); 1915 zfs_unlinked_add(xzp, tx); 1916 1917 if (zp->z_is_sa) 1918 error = sa_remove(zp->z_sa_hdl, 1919 SA_ZPL_XATTR(zfsvfs), tx); 1920 else 1921 error = sa_update(zp->z_sa_hdl, 1922 SA_ZPL_XATTR(zfsvfs), &null_xattr, 1923 sizeof (uint64_t), tx); 1924 ASSERT3U(error, ==, 0); 1925 } 1926 VI_LOCK(vp); 1927 vp->v_count--; 1928 ASSERT3U(vp->v_count, ==, 0); 1929 VI_UNLOCK(vp); 1930 mutex_exit(&zp->z_lock); 1931 zfs_znode_delete(zp, tx); 1932 } else if (unlinked) { 1933 mutex_exit(&zp->z_lock); 1934 zfs_unlinked_add(zp, tx); 1935 } 1936 1937 txtype = TX_REMOVE; 1938 if (flags & FIGNORECASE) 1939 txtype |= TX_CI; 1940 zfs_log_remove(zilog, tx, txtype, dzp, name, obj); 1941 1942 dmu_tx_commit(tx); 1943out: 1944 if (realnmp) 1945 pn_free(realnmp); 1946 1947 zfs_dirent_unlock(dl); 1948 1949 if (!delete_now) 1950 VN_RELE(vp); 1951 if (xzp) 1952 VN_RELE(ZTOV(xzp)); 1953 1954 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1955 zil_commit(zilog, 0); 1956 1957 ZFS_EXIT(zfsvfs); 1958 return (error); 1959} 1960 1961/* 1962 * Create a new directory and insert it into dvp using the name 1963 * provided. Return a pointer to the inserted directory. 1964 * 1965 * IN: dvp - vnode of directory to add subdir to. 1966 * dirname - name of new directory. 1967 * vap - attributes of new directory. 1968 * cr - credentials of caller. 1969 * ct - caller context 1970 * vsecp - ACL to be set 1971 * 1972 * OUT: vpp - vnode of created directory. 1973 * 1974 * RETURN: 0 if success 1975 * error code if failure 1976 * 1977 * Timestamps: 1978 * dvp - ctime|mtime updated 1979 * vp - ctime|mtime|atime updated 1980 */ 1981/*ARGSUSED*/ 1982static int 1983zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 1984 caller_context_t *ct, int flags, vsecattr_t *vsecp) 1985{ 1986 znode_t *zp, *dzp = VTOZ(dvp); 1987 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1988 zilog_t *zilog; 1989 zfs_dirlock_t *dl; 1990 uint64_t txtype; 1991 dmu_tx_t *tx; 1992 int error; 1993 int zf = ZNEW; 1994 ksid_t *ksid; 1995 uid_t uid; 1996 gid_t gid = crgetgid(cr); 1997 zfs_acl_ids_t acl_ids; 1998 boolean_t fuid_dirtied; 1999 2000 ASSERT(vap->va_type == VDIR); 2001 2002 /* 2003 * If we have an ephemeral id, ACL, or XVATTR then 2004 * make sure file system is at proper version 2005 */ 2006 2007 ksid = crgetsid(cr, KSID_OWNER); 2008 if (ksid) 2009 uid = ksid_getid(ksid); 2010 else 2011 uid = crgetuid(cr); 2012 if (zfsvfs->z_use_fuids == B_FALSE && 2013 (vsecp || (vap->va_mask & AT_XVATTR) || 2014 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 2015 return (EINVAL); 2016 2017 ZFS_ENTER(zfsvfs); 2018 ZFS_VERIFY_ZP(dzp); 2019 zilog = zfsvfs->z_log; 2020 2021 if (dzp->z_pflags & ZFS_XATTR) { 2022 ZFS_EXIT(zfsvfs); 2023 return (EINVAL); 2024 } 2025 2026 if (zfsvfs->z_utf8 && u8_validate(dirname, 2027 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2028 ZFS_EXIT(zfsvfs); 2029 return (EILSEQ); 2030 } 2031 if (flags & FIGNORECASE) 2032 zf |= ZCILOOK; 2033 2034 if (vap->va_mask & AT_XVATTR) { 2035 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 2036 crgetuid(cr), cr, vap->va_type)) != 0) { 2037 ZFS_EXIT(zfsvfs); 2038 return (error); 2039 } 2040 } 2041 2042 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 2043 vsecp, &acl_ids)) != 0) { 2044 ZFS_EXIT(zfsvfs); 2045 return (error); 2046 } 2047 /* 2048 * First make sure the new directory doesn't exist. 2049 * 2050 * Existence is checked first to make sure we don't return 2051 * EACCES instead of EEXIST which can cause some applications 2052 * to fail. 2053 */ 2054top: 2055 *vpp = NULL; 2056 2057 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 2058 NULL, NULL)) { 2059 zfs_acl_ids_free(&acl_ids); 2060 ZFS_EXIT(zfsvfs); 2061 return (error); 2062 } 2063 2064 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 2065 zfs_acl_ids_free(&acl_ids); 2066 zfs_dirent_unlock(dl); 2067 ZFS_EXIT(zfsvfs); 2068 return (error); 2069 } 2070 2071 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 2072 zfs_acl_ids_free(&acl_ids); 2073 zfs_dirent_unlock(dl); 2074 ZFS_EXIT(zfsvfs); 2075 return (EDQUOT); 2076 } 2077 2078 /* 2079 * Add a new entry to the directory. 2080 */ 2081 tx = dmu_tx_create(zfsvfs->z_os); 2082 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 2083 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2084 fuid_dirtied = zfsvfs->z_fuid_dirty; 2085 if (fuid_dirtied) 2086 zfs_fuid_txhold(zfsvfs, tx); 2087 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2088 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2089 acl_ids.z_aclp->z_acl_bytes); 2090 } 2091 2092 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 2093 ZFS_SA_BASE_ATTR_SIZE); 2094 2095 error = dmu_tx_assign(tx, TXG_NOWAIT); 2096 if (error) { 2097 zfs_dirent_unlock(dl); 2098 if (error == ERESTART) { 2099 dmu_tx_wait(tx); 2100 dmu_tx_abort(tx); 2101 goto top; 2102 } 2103 zfs_acl_ids_free(&acl_ids); 2104 dmu_tx_abort(tx); 2105 ZFS_EXIT(zfsvfs); 2106 return (error); 2107 } 2108 2109 /* 2110 * Create new node. 2111 */ 2112 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 2113 2114 if (fuid_dirtied) 2115 zfs_fuid_sync(zfsvfs, tx); 2116 2117 /* 2118 * Now put new name in parent dir. 2119 */ 2120 (void) zfs_link_create(dl, zp, tx, ZNEW); 2121 2122 *vpp = ZTOV(zp); 2123 2124 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 2125 if (flags & FIGNORECASE) 2126 txtype |= TX_CI; 2127 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 2128 acl_ids.z_fuidp, vap); 2129 2130 zfs_acl_ids_free(&acl_ids); 2131 2132 dmu_tx_commit(tx); 2133 2134 zfs_dirent_unlock(dl); 2135 2136 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2137 zil_commit(zilog, 0); 2138 2139 ZFS_EXIT(zfsvfs); 2140 return (0); 2141} 2142 2143/* 2144 * Remove a directory subdir entry. If the current working 2145 * directory is the same as the subdir to be removed, the 2146 * remove will fail. 2147 * 2148 * IN: dvp - vnode of directory to remove from. 2149 * name - name of directory to be removed. 2150 * cwd - vnode of current working directory. 2151 * cr - credentials of caller. 2152 * ct - caller context 2153 * flags - case flags 2154 * 2155 * RETURN: 0 if success 2156 * error code if failure 2157 * 2158 * Timestamps: 2159 * dvp - ctime|mtime updated 2160 */ 2161/*ARGSUSED*/ 2162static int 2163zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 2164 caller_context_t *ct, int flags) 2165{ 2166 znode_t *dzp = VTOZ(dvp); 2167 znode_t *zp; 2168 vnode_t *vp; 2169 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2170 zilog_t *zilog; 2171 zfs_dirlock_t *dl; 2172 dmu_tx_t *tx; 2173 int error; 2174 int zflg = ZEXISTS; 2175 2176 ZFS_ENTER(zfsvfs); 2177 ZFS_VERIFY_ZP(dzp); 2178 zilog = zfsvfs->z_log; 2179 2180 if (flags & FIGNORECASE) 2181 zflg |= ZCILOOK; 2182top: 2183 zp = NULL; 2184 2185 /* 2186 * Attempt to lock directory; fail if entry doesn't exist. 2187 */ 2188 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 2189 NULL, NULL)) { 2190 ZFS_EXIT(zfsvfs); 2191 return (error); 2192 } 2193 2194 vp = ZTOV(zp); 2195 2196 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 2197 goto out; 2198 } 2199 2200 if (vp->v_type != VDIR) { 2201 error = ENOTDIR; 2202 goto out; 2203 } 2204 2205 if (vp == cwd) { 2206 error = EINVAL; 2207 goto out; 2208 } 2209 2210 vnevent_rmdir(vp, dvp, name, ct); 2211 2212 /* 2213 * Grab a lock on the directory to make sure that noone is 2214 * trying to add (or lookup) entries while we are removing it. 2215 */ 2216 rw_enter(&zp->z_name_lock, RW_WRITER); 2217 2218 /* 2219 * Grab a lock on the parent pointer to make sure we play well 2220 * with the treewalk and directory rename code. 2221 */ 2222 rw_enter(&zp->z_parent_lock, RW_WRITER); 2223 2224 tx = dmu_tx_create(zfsvfs->z_os); 2225 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2226 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2227 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2228 zfs_sa_upgrade_txholds(tx, zp); 2229 zfs_sa_upgrade_txholds(tx, dzp); 2230 error = dmu_tx_assign(tx, TXG_NOWAIT); 2231 if (error) { 2232 rw_exit(&zp->z_parent_lock); 2233 rw_exit(&zp->z_name_lock); 2234 zfs_dirent_unlock(dl); 2235 VN_RELE(vp); 2236 if (error == ERESTART) { 2237 dmu_tx_wait(tx); 2238 dmu_tx_abort(tx); 2239 goto top; 2240 } 2241 dmu_tx_abort(tx); 2242 ZFS_EXIT(zfsvfs); 2243 return (error); 2244 } 2245 2246#ifdef FREEBSD_NAMECACHE 2247 cache_purge(dvp); 2248#endif 2249 2250 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 2251 2252 if (error == 0) { 2253 uint64_t txtype = TX_RMDIR; 2254 if (flags & FIGNORECASE) 2255 txtype |= TX_CI; 2256 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT); 2257 } 2258 2259 dmu_tx_commit(tx); 2260 2261 rw_exit(&zp->z_parent_lock); 2262 rw_exit(&zp->z_name_lock); 2263#ifdef FREEBSD_NAMECACHE 2264 cache_purge(vp); 2265#endif 2266out: 2267 zfs_dirent_unlock(dl); 2268 2269 VN_RELE(vp); 2270 2271 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2272 zil_commit(zilog, 0); 2273 2274 ZFS_EXIT(zfsvfs); 2275 return (error); 2276} 2277 2278/* 2279 * Read as many directory entries as will fit into the provided 2280 * buffer from the given directory cursor position (specified in 2281 * the uio structure. 2282 * 2283 * IN: vp - vnode of directory to read. 2284 * uio - structure supplying read location, range info, 2285 * and return buffer. 2286 * cr - credentials of caller. 2287 * ct - caller context 2288 * flags - case flags 2289 * 2290 * OUT: uio - updated offset and range, buffer filled. 2291 * eofp - set to true if end-of-file detected. 2292 * 2293 * RETURN: 0 if success 2294 * error code if failure 2295 * 2296 * Timestamps: 2297 * vp - atime updated 2298 * 2299 * Note that the low 4 bits of the cookie returned by zap is always zero. 2300 * This allows us to use the low range for "special" directory entries: 2301 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 2302 * we use the offset 2 for the '.zfs' directory. 2303 */ 2304/* ARGSUSED */ 2305static int 2306zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 2307{ 2308 znode_t *zp = VTOZ(vp); 2309 iovec_t *iovp; 2310 edirent_t *eodp; 2311 dirent64_t *odp; 2312 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2313 objset_t *os; 2314 caddr_t outbuf; 2315 size_t bufsize; 2316 zap_cursor_t zc; 2317 zap_attribute_t zap; 2318 uint_t bytes_wanted; 2319 uint64_t offset; /* must be unsigned; checks for < 1 */ 2320 uint64_t parent; 2321 int local_eof; 2322 int outcount; 2323 int error; 2324 uint8_t prefetch; 2325 boolean_t check_sysattrs; 2326 uint8_t type; 2327 int ncooks; 2328 u_long *cooks = NULL; 2329 int flags = 0; 2330 2331 ZFS_ENTER(zfsvfs); 2332 ZFS_VERIFY_ZP(zp); 2333 2334 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 2335 &parent, sizeof (parent))) != 0) { 2336 ZFS_EXIT(zfsvfs); 2337 return (error); 2338 } 2339 2340 /* 2341 * If we are not given an eof variable, 2342 * use a local one. 2343 */ 2344 if (eofp == NULL) 2345 eofp = &local_eof; 2346 2347 /* 2348 * Check for valid iov_len. 2349 */ 2350 if (uio->uio_iov->iov_len <= 0) { 2351 ZFS_EXIT(zfsvfs); 2352 return (EINVAL); 2353 } 2354 2355 /* 2356 * Quit if directory has been removed (posix) 2357 */ 2358 if ((*eofp = zp->z_unlinked) != 0) { 2359 ZFS_EXIT(zfsvfs); 2360 return (0); 2361 } 2362 2363 error = 0; 2364 os = zfsvfs->z_os; 2365 offset = uio->uio_loffset; 2366 prefetch = zp->z_zn_prefetch; 2367 2368 /* 2369 * Initialize the iterator cursor. 2370 */ 2371 if (offset <= 3) { 2372 /* 2373 * Start iteration from the beginning of the directory. 2374 */ 2375 zap_cursor_init(&zc, os, zp->z_id); 2376 } else { 2377 /* 2378 * The offset is a serialized cursor. 2379 */ 2380 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 2381 } 2382 2383 /* 2384 * Get space to change directory entries into fs independent format. 2385 */ 2386 iovp = uio->uio_iov; 2387 bytes_wanted = iovp->iov_len; 2388 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 2389 bufsize = bytes_wanted; 2390 outbuf = kmem_alloc(bufsize, KM_SLEEP); 2391 odp = (struct dirent64 *)outbuf; 2392 } else { 2393 bufsize = bytes_wanted; 2394 odp = (struct dirent64 *)iovp->iov_base; 2395 } 2396 eodp = (struct edirent *)odp; 2397 2398 if (ncookies != NULL) { 2399 /* 2400 * Minimum entry size is dirent size and 1 byte for a file name. 2401 */ 2402 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 2403 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 2404 *cookies = cooks; 2405 *ncookies = ncooks; 2406 } 2407 /* 2408 * If this VFS supports the system attribute view interface; and 2409 * we're looking at an extended attribute directory; and we care 2410 * about normalization conflicts on this vfs; then we must check 2411 * for normalization conflicts with the sysattr name space. 2412 */ 2413#ifdef TODO 2414 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 2415 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 2416 (flags & V_RDDIR_ENTFLAGS); 2417#else 2418 check_sysattrs = 0; 2419#endif 2420 2421 /* 2422 * Transform to file-system independent format 2423 */ 2424 outcount = 0; 2425 while (outcount < bytes_wanted) { 2426 ino64_t objnum; 2427 ushort_t reclen; 2428 off64_t *next = NULL; 2429 2430 /* 2431 * Special case `.', `..', and `.zfs'. 2432 */ 2433 if (offset == 0) { 2434 (void) strcpy(zap.za_name, "."); 2435 zap.za_normalization_conflict = 0; 2436 objnum = zp->z_id; 2437 type = DT_DIR; 2438 } else if (offset == 1) { 2439 (void) strcpy(zap.za_name, ".."); 2440 zap.za_normalization_conflict = 0; 2441 objnum = parent; 2442 type = DT_DIR; 2443 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2444 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2445 zap.za_normalization_conflict = 0; 2446 objnum = ZFSCTL_INO_ROOT; 2447 type = DT_DIR; 2448 } else { 2449 /* 2450 * Grab next entry. 2451 */ 2452 if (error = zap_cursor_retrieve(&zc, &zap)) { 2453 if ((*eofp = (error == ENOENT)) != 0) 2454 break; 2455 else 2456 goto update; 2457 } 2458 2459 if (zap.za_integer_length != 8 || 2460 zap.za_num_integers != 1) { 2461 cmn_err(CE_WARN, "zap_readdir: bad directory " 2462 "entry, obj = %lld, offset = %lld\n", 2463 (u_longlong_t)zp->z_id, 2464 (u_longlong_t)offset); 2465 error = ENXIO; 2466 goto update; 2467 } 2468 2469 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2470 /* 2471 * MacOS X can extract the object type here such as: 2472 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2473 */ 2474 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2475 2476 if (check_sysattrs && !zap.za_normalization_conflict) { 2477#ifdef TODO 2478 zap.za_normalization_conflict = 2479 xattr_sysattr_casechk(zap.za_name); 2480#else 2481 panic("%s:%u: TODO", __func__, __LINE__); 2482#endif 2483 } 2484 } 2485 2486 if (flags & V_RDDIR_ACCFILTER) { 2487 /* 2488 * If we have no access at all, don't include 2489 * this entry in the returned information 2490 */ 2491 znode_t *ezp; 2492 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2493 goto skip_entry; 2494 if (!zfs_has_access(ezp, cr)) { 2495 VN_RELE(ZTOV(ezp)); 2496 goto skip_entry; 2497 } 2498 VN_RELE(ZTOV(ezp)); 2499 } 2500 2501 if (flags & V_RDDIR_ENTFLAGS) 2502 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2503 else 2504 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2505 2506 /* 2507 * Will this entry fit in the buffer? 2508 */ 2509 if (outcount + reclen > bufsize) { 2510 /* 2511 * Did we manage to fit anything in the buffer? 2512 */ 2513 if (!outcount) { 2514 error = EINVAL; 2515 goto update; 2516 } 2517 break; 2518 } 2519 if (flags & V_RDDIR_ENTFLAGS) { 2520 /* 2521 * Add extended flag entry: 2522 */ 2523 eodp->ed_ino = objnum; 2524 eodp->ed_reclen = reclen; 2525 /* NOTE: ed_off is the offset for the *next* entry */ 2526 next = &(eodp->ed_off); 2527 eodp->ed_eflags = zap.za_normalization_conflict ? 2528 ED_CASE_CONFLICT : 0; 2529 (void) strncpy(eodp->ed_name, zap.za_name, 2530 EDIRENT_NAMELEN(reclen)); 2531 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2532 } else { 2533 /* 2534 * Add normal entry: 2535 */ 2536 odp->d_ino = objnum; 2537 odp->d_reclen = reclen; 2538 odp->d_namlen = strlen(zap.za_name); 2539 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 2540 odp->d_type = type; 2541 odp = (dirent64_t *)((intptr_t)odp + reclen); 2542 } 2543 outcount += reclen; 2544 2545 ASSERT(outcount <= bufsize); 2546 2547 /* Prefetch znode */ 2548 if (prefetch) 2549 dmu_prefetch(os, objnum, 0, 0); 2550 2551 skip_entry: 2552 /* 2553 * Move to the next entry, fill in the previous offset. 2554 */ 2555 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2556 zap_cursor_advance(&zc); 2557 offset = zap_cursor_serialize(&zc); 2558 } else { 2559 offset += 1; 2560 } 2561 2562 if (cooks != NULL) { 2563 *cooks++ = offset; 2564 ncooks--; 2565 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 2566 } 2567 } 2568 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2569 2570 /* Subtract unused cookies */ 2571 if (ncookies != NULL) 2572 *ncookies -= ncooks; 2573 2574 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2575 iovp->iov_base += outcount; 2576 iovp->iov_len -= outcount; 2577 uio->uio_resid -= outcount; 2578 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2579 /* 2580 * Reset the pointer. 2581 */ 2582 offset = uio->uio_loffset; 2583 } 2584 2585update: 2586 zap_cursor_fini(&zc); 2587 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2588 kmem_free(outbuf, bufsize); 2589 2590 if (error == ENOENT) 2591 error = 0; 2592 2593 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2594 2595 uio->uio_loffset = offset; 2596 ZFS_EXIT(zfsvfs); 2597 if (error != 0 && cookies != NULL) { 2598 free(*cookies, M_TEMP); 2599 *cookies = NULL; 2600 *ncookies = 0; 2601 } 2602 return (error); 2603} 2604 2605ulong_t zfs_fsync_sync_cnt = 4; 2606 2607static int 2608zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2609{ 2610 znode_t *zp = VTOZ(vp); 2611 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2612 2613 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2614 2615 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { 2616 ZFS_ENTER(zfsvfs); 2617 ZFS_VERIFY_ZP(zp); 2618 zil_commit(zfsvfs->z_log, zp->z_id); 2619 ZFS_EXIT(zfsvfs); 2620 } 2621 return (0); 2622} 2623 2624 2625/* 2626 * Get the requested file attributes and place them in the provided 2627 * vattr structure. 2628 * 2629 * IN: vp - vnode of file. 2630 * vap - va_mask identifies requested attributes. 2631 * If AT_XVATTR set, then optional attrs are requested 2632 * flags - ATTR_NOACLCHECK (CIFS server context) 2633 * cr - credentials of caller. 2634 * ct - caller context 2635 * 2636 * OUT: vap - attribute values. 2637 * 2638 * RETURN: 0 (always succeeds) 2639 */ 2640/* ARGSUSED */ 2641static int 2642zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2643 caller_context_t *ct) 2644{ 2645 znode_t *zp = VTOZ(vp); 2646 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2647 int error = 0; 2648 uint32_t blksize; 2649 u_longlong_t nblocks; 2650 uint64_t links; 2651 uint64_t mtime[2], ctime[2], crtime[2], rdev; 2652 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2653 xoptattr_t *xoap = NULL; 2654 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2655 sa_bulk_attr_t bulk[4]; 2656 int count = 0; 2657 2658 ZFS_ENTER(zfsvfs); 2659 ZFS_VERIFY_ZP(zp); 2660 2661 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2662 2663 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 2664 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 2665 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &crtime, 16); 2666 if (vp->v_type == VBLK || vp->v_type == VCHR) 2667 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, 2668 &rdev, 8); 2669 2670 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 2671 ZFS_EXIT(zfsvfs); 2672 return (error); 2673 } 2674 2675 /* 2676 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2677 * Also, if we are the owner don't bother, since owner should 2678 * always be allowed to read basic attributes of file. 2679 */ 2680 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 2681 (vap->va_uid != crgetuid(cr))) { 2682 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2683 skipaclchk, cr)) { 2684 ZFS_EXIT(zfsvfs); 2685 return (error); 2686 } 2687 } 2688 2689 /* 2690 * Return all attributes. It's cheaper to provide the answer 2691 * than to determine whether we were asked the question. 2692 */ 2693 2694 mutex_enter(&zp->z_lock); 2695 vap->va_type = IFTOVT(zp->z_mode); 2696 vap->va_mode = zp->z_mode & ~S_IFMT; 2697#ifdef sun 2698 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2699#else 2700 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 2701#endif 2702 vap->va_nodeid = zp->z_id; 2703 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2704 links = zp->z_links + 1; 2705 else 2706 links = zp->z_links; 2707 vap->va_nlink = MIN(links, LINK_MAX); /* nlink_t limit! */ 2708 vap->va_size = zp->z_size; 2709#ifdef sun 2710 vap->va_rdev = vp->v_rdev; 2711#else 2712 if (vp->v_type == VBLK || vp->v_type == VCHR) 2713 vap->va_rdev = zfs_cmpldev(rdev); 2714#endif 2715 vap->va_seq = zp->z_seq; 2716 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2717 2718 /* 2719 * Add in any requested optional attributes and the create time. 2720 * Also set the corresponding bits in the returned attribute bitmap. 2721 */ 2722 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2723 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2724 xoap->xoa_archive = 2725 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2726 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2727 } 2728 2729 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2730 xoap->xoa_readonly = 2731 ((zp->z_pflags & ZFS_READONLY) != 0); 2732 XVA_SET_RTN(xvap, XAT_READONLY); 2733 } 2734 2735 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2736 xoap->xoa_system = 2737 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2738 XVA_SET_RTN(xvap, XAT_SYSTEM); 2739 } 2740 2741 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2742 xoap->xoa_hidden = 2743 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2744 XVA_SET_RTN(xvap, XAT_HIDDEN); 2745 } 2746 2747 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2748 xoap->xoa_nounlink = 2749 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2750 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2751 } 2752 2753 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2754 xoap->xoa_immutable = 2755 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2756 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2757 } 2758 2759 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2760 xoap->xoa_appendonly = 2761 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2762 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2763 } 2764 2765 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2766 xoap->xoa_nodump = 2767 ((zp->z_pflags & ZFS_NODUMP) != 0); 2768 XVA_SET_RTN(xvap, XAT_NODUMP); 2769 } 2770 2771 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2772 xoap->xoa_opaque = 2773 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2774 XVA_SET_RTN(xvap, XAT_OPAQUE); 2775 } 2776 2777 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2778 xoap->xoa_av_quarantined = 2779 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2780 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2781 } 2782 2783 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2784 xoap->xoa_av_modified = 2785 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2786 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2787 } 2788 2789 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2790 vp->v_type == VREG) { 2791 zfs_sa_get_scanstamp(zp, xvap); 2792 } 2793 2794 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2795 uint64_t times[2]; 2796 2797 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 2798 times, sizeof (times)); 2799 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 2800 XVA_SET_RTN(xvap, XAT_CREATETIME); 2801 } 2802 2803 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2804 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2805 XVA_SET_RTN(xvap, XAT_REPARSE); 2806 } 2807 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2808 xoap->xoa_generation = zp->z_gen; 2809 XVA_SET_RTN(xvap, XAT_GEN); 2810 } 2811 2812 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2813 xoap->xoa_offline = 2814 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2815 XVA_SET_RTN(xvap, XAT_OFFLINE); 2816 } 2817 2818 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2819 xoap->xoa_sparse = 2820 ((zp->z_pflags & ZFS_SPARSE) != 0); 2821 XVA_SET_RTN(xvap, XAT_SPARSE); 2822 } 2823 } 2824 2825 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2826 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2827 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2828 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 2829 2830 mutex_exit(&zp->z_lock); 2831 2832 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 2833 vap->va_blksize = blksize; 2834 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2835 2836 if (zp->z_blksz == 0) { 2837 /* 2838 * Block size hasn't been set; suggest maximal I/O transfers. 2839 */ 2840 vap->va_blksize = zfsvfs->z_max_blksz; 2841 } 2842 2843 ZFS_EXIT(zfsvfs); 2844 return (0); 2845} 2846 2847/* 2848 * Set the file attributes to the values contained in the 2849 * vattr structure. 2850 * 2851 * IN: vp - vnode of file to be modified. 2852 * vap - new attribute values. 2853 * If AT_XVATTR set, then optional attrs are being set 2854 * flags - ATTR_UTIME set if non-default time values provided. 2855 * - ATTR_NOACLCHECK (CIFS context only). 2856 * cr - credentials of caller. 2857 * ct - caller context 2858 * 2859 * RETURN: 0 if success 2860 * error code if failure 2861 * 2862 * Timestamps: 2863 * vp - ctime updated, mtime updated if size changed. 2864 */ 2865/* ARGSUSED */ 2866static int 2867zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2868 caller_context_t *ct) 2869{ 2870 znode_t *zp = VTOZ(vp); 2871 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2872 zilog_t *zilog; 2873 dmu_tx_t *tx; 2874 vattr_t oldva; 2875 xvattr_t tmpxvattr; 2876 uint_t mask = vap->va_mask; 2877 uint_t saved_mask; 2878 uint64_t saved_mode; 2879 int trim_mask = 0; 2880 uint64_t new_mode; 2881 uint64_t new_uid, new_gid; 2882 uint64_t xattr_obj; 2883 uint64_t mtime[2], ctime[2]; 2884 znode_t *attrzp; 2885 int need_policy = FALSE; 2886 int err, err2; 2887 zfs_fuid_info_t *fuidp = NULL; 2888 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2889 xoptattr_t *xoap; 2890 zfs_acl_t *aclp; 2891 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2892 boolean_t fuid_dirtied = B_FALSE; 2893 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 2894 int count = 0, xattr_count = 0; 2895 2896 if (mask == 0) 2897 return (0); 2898 2899 if (mask & AT_NOSET) 2900 return (EINVAL); 2901 2902 ZFS_ENTER(zfsvfs); 2903 ZFS_VERIFY_ZP(zp); 2904 2905 zilog = zfsvfs->z_log; 2906 2907 /* 2908 * Make sure that if we have ephemeral uid/gid or xvattr specified 2909 * that file system is at proper version level 2910 */ 2911 2912 if (zfsvfs->z_use_fuids == B_FALSE && 2913 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2914 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2915 (mask & AT_XVATTR))) { 2916 ZFS_EXIT(zfsvfs); 2917 return (EINVAL); 2918 } 2919 2920 if (mask & AT_SIZE && vp->v_type == VDIR) { 2921 ZFS_EXIT(zfsvfs); 2922 return (EISDIR); 2923 } 2924 2925 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 2926 ZFS_EXIT(zfsvfs); 2927 return (EINVAL); 2928 } 2929 2930 /* 2931 * If this is an xvattr_t, then get a pointer to the structure of 2932 * optional attributes. If this is NULL, then we have a vattr_t. 2933 */ 2934 xoap = xva_getxoptattr(xvap); 2935 2936 xva_init(&tmpxvattr); 2937 2938 /* 2939 * Immutable files can only alter immutable bit and atime 2940 */ 2941 if ((zp->z_pflags & ZFS_IMMUTABLE) && 2942 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2943 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 2944 ZFS_EXIT(zfsvfs); 2945 return (EPERM); 2946 } 2947 2948 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 2949 ZFS_EXIT(zfsvfs); 2950 return (EPERM); 2951 } 2952 2953 /* 2954 * Verify timestamps doesn't overflow 32 bits. 2955 * ZFS can handle large timestamps, but 32bit syscalls can't 2956 * handle times greater than 2039. This check should be removed 2957 * once large timestamps are fully supported. 2958 */ 2959 if (mask & (AT_ATIME | AT_MTIME)) { 2960 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 2961 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 2962 ZFS_EXIT(zfsvfs); 2963 return (EOVERFLOW); 2964 } 2965 } 2966 2967top: 2968 attrzp = NULL; 2969 aclp = NULL; 2970 2971 /* Can this be moved to before the top label? */ 2972 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2973 ZFS_EXIT(zfsvfs); 2974 return (EROFS); 2975 } 2976 2977 /* 2978 * First validate permissions 2979 */ 2980 2981 if (mask & AT_SIZE) { 2982 /* 2983 * XXX - Note, we are not providing any open 2984 * mode flags here (like FNDELAY), so we may 2985 * block if there are locks present... this 2986 * should be addressed in openat(). 2987 */ 2988 /* XXX - would it be OK to generate a log record here? */ 2989 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2990 if (err) { 2991 ZFS_EXIT(zfsvfs); 2992 return (err); 2993 } 2994 } 2995 2996 if (mask & (AT_ATIME|AT_MTIME) || 2997 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2998 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2999 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 3000 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 3001 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 3002 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 3003 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 3004 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 3005 skipaclchk, cr); 3006 } 3007 3008 if (mask & (AT_UID|AT_GID)) { 3009 int idmask = (mask & (AT_UID|AT_GID)); 3010 int take_owner; 3011 int take_group; 3012 3013 /* 3014 * NOTE: even if a new mode is being set, 3015 * we may clear S_ISUID/S_ISGID bits. 3016 */ 3017 3018 if (!(mask & AT_MODE)) 3019 vap->va_mode = zp->z_mode; 3020 3021 /* 3022 * Take ownership or chgrp to group we are a member of 3023 */ 3024 3025 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 3026 take_group = (mask & AT_GID) && 3027 zfs_groupmember(zfsvfs, vap->va_gid, cr); 3028 3029 /* 3030 * If both AT_UID and AT_GID are set then take_owner and 3031 * take_group must both be set in order to allow taking 3032 * ownership. 3033 * 3034 * Otherwise, send the check through secpolicy_vnode_setattr() 3035 * 3036 */ 3037 3038 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 3039 ((idmask == AT_UID) && take_owner) || 3040 ((idmask == AT_GID) && take_group)) { 3041 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 3042 skipaclchk, cr) == 0) { 3043 /* 3044 * Remove setuid/setgid for non-privileged users 3045 */ 3046 secpolicy_setid_clear(vap, vp, cr); 3047 trim_mask = (mask & (AT_UID|AT_GID)); 3048 } else { 3049 need_policy = TRUE; 3050 } 3051 } else { 3052 need_policy = TRUE; 3053 } 3054 } 3055 3056 mutex_enter(&zp->z_lock); 3057 oldva.va_mode = zp->z_mode; 3058 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 3059 if (mask & AT_XVATTR) { 3060 /* 3061 * Update xvattr mask to include only those attributes 3062 * that are actually changing. 3063 * 3064 * the bits will be restored prior to actually setting 3065 * the attributes so the caller thinks they were set. 3066 */ 3067 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 3068 if (xoap->xoa_appendonly != 3069 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 3070 need_policy = TRUE; 3071 } else { 3072 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 3073 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 3074 } 3075 } 3076 3077 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 3078 if (xoap->xoa_nounlink != 3079 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 3080 need_policy = TRUE; 3081 } else { 3082 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 3083 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 3084 } 3085 } 3086 3087 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 3088 if (xoap->xoa_immutable != 3089 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 3090 need_policy = TRUE; 3091 } else { 3092 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 3093 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 3094 } 3095 } 3096 3097 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 3098 if (xoap->xoa_nodump != 3099 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 3100 need_policy = TRUE; 3101 } else { 3102 XVA_CLR_REQ(xvap, XAT_NODUMP); 3103 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 3104 } 3105 } 3106 3107 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 3108 if (xoap->xoa_av_modified != 3109 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 3110 need_policy = TRUE; 3111 } else { 3112 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 3113 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 3114 } 3115 } 3116 3117 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 3118 if ((vp->v_type != VREG && 3119 xoap->xoa_av_quarantined) || 3120 xoap->xoa_av_quarantined != 3121 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 3122 need_policy = TRUE; 3123 } else { 3124 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 3125 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 3126 } 3127 } 3128 3129 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 3130 mutex_exit(&zp->z_lock); 3131 ZFS_EXIT(zfsvfs); 3132 return (EPERM); 3133 } 3134 3135 if (need_policy == FALSE && 3136 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 3137 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 3138 need_policy = TRUE; 3139 } 3140 } 3141 3142 mutex_exit(&zp->z_lock); 3143 3144 if (mask & AT_MODE) { 3145 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 3146 err = secpolicy_setid_setsticky_clear(vp, vap, 3147 &oldva, cr); 3148 if (err) { 3149 ZFS_EXIT(zfsvfs); 3150 return (err); 3151 } 3152 trim_mask |= AT_MODE; 3153 } else { 3154 need_policy = TRUE; 3155 } 3156 } 3157 3158 if (need_policy) { 3159 /* 3160 * If trim_mask is set then take ownership 3161 * has been granted or write_acl is present and user 3162 * has the ability to modify mode. In that case remove 3163 * UID|GID and or MODE from mask so that 3164 * secpolicy_vnode_setattr() doesn't revoke it. 3165 */ 3166 3167 if (trim_mask) { 3168 saved_mask = vap->va_mask; 3169 vap->va_mask &= ~trim_mask; 3170 if (trim_mask & AT_MODE) { 3171 /* 3172 * Save the mode, as secpolicy_vnode_setattr() 3173 * will overwrite it with ova.va_mode. 3174 */ 3175 saved_mode = vap->va_mode; 3176 } 3177 } 3178 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 3179 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 3180 if (err) { 3181 ZFS_EXIT(zfsvfs); 3182 return (err); 3183 } 3184 3185 if (trim_mask) { 3186 vap->va_mask |= saved_mask; 3187 if (trim_mask & AT_MODE) { 3188 /* 3189 * Recover the mode after 3190 * secpolicy_vnode_setattr(). 3191 */ 3192 vap->va_mode = saved_mode; 3193 } 3194 } 3195 } 3196 3197 /* 3198 * secpolicy_vnode_setattr, or take ownership may have 3199 * changed va_mask 3200 */ 3201 mask = vap->va_mask; 3202 3203 if ((mask & (AT_UID | AT_GID))) { 3204 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 3205 &xattr_obj, sizeof (xattr_obj)); 3206 3207 if (err == 0 && xattr_obj) { 3208 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 3209 if (err) 3210 goto out2; 3211 } 3212 if (mask & AT_UID) { 3213 new_uid = zfs_fuid_create(zfsvfs, 3214 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 3215 if (new_uid != zp->z_uid && 3216 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 3217 if (attrzp) 3218 VN_RELE(ZTOV(attrzp)); 3219 err = EDQUOT; 3220 goto out2; 3221 } 3222 } 3223 3224 if (mask & AT_GID) { 3225 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 3226 cr, ZFS_GROUP, &fuidp); 3227 if (new_gid != zp->z_gid && 3228 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 3229 if (attrzp) 3230 VN_RELE(ZTOV(attrzp)); 3231 err = EDQUOT; 3232 goto out2; 3233 } 3234 } 3235 } 3236 tx = dmu_tx_create(zfsvfs->z_os); 3237 3238 if (mask & AT_MODE) { 3239 uint64_t pmode = zp->z_mode; 3240 uint64_t acl_obj; 3241 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 3242 3243 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 3244 goto out; 3245 3246 mutex_enter(&zp->z_lock); 3247 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 3248 /* 3249 * Are we upgrading ACL from old V0 format 3250 * to V1 format? 3251 */ 3252 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 3253 zfs_znode_acl_version(zp) == 3254 ZFS_ACL_VERSION_INITIAL) { 3255 dmu_tx_hold_free(tx, acl_obj, 0, 3256 DMU_OBJECT_END); 3257 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3258 0, aclp->z_acl_bytes); 3259 } else { 3260 dmu_tx_hold_write(tx, acl_obj, 0, 3261 aclp->z_acl_bytes); 3262 } 3263 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3264 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3265 0, aclp->z_acl_bytes); 3266 } 3267 mutex_exit(&zp->z_lock); 3268 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3269 } else { 3270 if ((mask & AT_XVATTR) && 3271 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3272 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3273 else 3274 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3275 } 3276 3277 if (attrzp) { 3278 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 3279 } 3280 3281 fuid_dirtied = zfsvfs->z_fuid_dirty; 3282 if (fuid_dirtied) 3283 zfs_fuid_txhold(zfsvfs, tx); 3284 3285 zfs_sa_upgrade_txholds(tx, zp); 3286 3287 err = dmu_tx_assign(tx, TXG_NOWAIT); 3288 if (err) { 3289 if (err == ERESTART) 3290 dmu_tx_wait(tx); 3291 goto out; 3292 } 3293 3294 count = 0; 3295 /* 3296 * Set each attribute requested. 3297 * We group settings according to the locks they need to acquire. 3298 * 3299 * Note: you cannot set ctime directly, although it will be 3300 * updated as a side-effect of calling this function. 3301 */ 3302 3303 3304 if (mask & (AT_UID|AT_GID|AT_MODE)) 3305 mutex_enter(&zp->z_acl_lock); 3306 mutex_enter(&zp->z_lock); 3307 3308 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3309 &zp->z_pflags, sizeof (zp->z_pflags)); 3310 3311 if (attrzp) { 3312 if (mask & (AT_UID|AT_GID|AT_MODE)) 3313 mutex_enter(&attrzp->z_acl_lock); 3314 mutex_enter(&attrzp->z_lock); 3315 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3316 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3317 sizeof (attrzp->z_pflags)); 3318 } 3319 3320 if (mask & (AT_UID|AT_GID)) { 3321 3322 if (mask & AT_UID) { 3323 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3324 &new_uid, sizeof (new_uid)); 3325 zp->z_uid = new_uid; 3326 if (attrzp) { 3327 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3328 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3329 sizeof (new_uid)); 3330 attrzp->z_uid = new_uid; 3331 } 3332 } 3333 3334 if (mask & AT_GID) { 3335 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3336 NULL, &new_gid, sizeof (new_gid)); 3337 zp->z_gid = new_gid; 3338 if (attrzp) { 3339 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3340 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 3341 sizeof (new_gid)); 3342 attrzp->z_gid = new_gid; 3343 } 3344 } 3345 if (!(mask & AT_MODE)) { 3346 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 3347 NULL, &new_mode, sizeof (new_mode)); 3348 new_mode = zp->z_mode; 3349 } 3350 err = zfs_acl_chown_setattr(zp); 3351 ASSERT(err == 0); 3352 if (attrzp) { 3353 err = zfs_acl_chown_setattr(attrzp); 3354 ASSERT(err == 0); 3355 } 3356 } 3357 3358 if (mask & AT_MODE) { 3359 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 3360 &new_mode, sizeof (new_mode)); 3361 zp->z_mode = new_mode; 3362 ASSERT3U((uintptr_t)aclp, !=, 0); 3363 err = zfs_aclset_common(zp, aclp, cr, tx); 3364 ASSERT3U(err, ==, 0); 3365 if (zp->z_acl_cached) 3366 zfs_acl_free(zp->z_acl_cached); 3367 zp->z_acl_cached = aclp; 3368 aclp = NULL; 3369 } 3370 3371 3372 if (mask & AT_ATIME) { 3373 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 3374 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 3375 &zp->z_atime, sizeof (zp->z_atime)); 3376 } 3377 3378 if (mask & AT_MTIME) { 3379 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 3380 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 3381 mtime, sizeof (mtime)); 3382 } 3383 3384 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 3385 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 3386 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 3387 NULL, mtime, sizeof (mtime)); 3388 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3389 &ctime, sizeof (ctime)); 3390 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 3391 B_TRUE); 3392 } else if (mask != 0) { 3393 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3394 &ctime, sizeof (ctime)); 3395 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime, 3396 B_TRUE); 3397 if (attrzp) { 3398 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3399 SA_ZPL_CTIME(zfsvfs), NULL, 3400 &ctime, sizeof (ctime)); 3401 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 3402 mtime, ctime, B_TRUE); 3403 } 3404 } 3405 /* 3406 * Do this after setting timestamps to prevent timestamp 3407 * update from toggling bit 3408 */ 3409 3410 if (xoap && (mask & AT_XVATTR)) { 3411 3412 /* 3413 * restore trimmed off masks 3414 * so that return masks can be set for caller. 3415 */ 3416 3417 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 3418 XVA_SET_REQ(xvap, XAT_APPENDONLY); 3419 } 3420 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 3421 XVA_SET_REQ(xvap, XAT_NOUNLINK); 3422 } 3423 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 3424 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 3425 } 3426 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 3427 XVA_SET_REQ(xvap, XAT_NODUMP); 3428 } 3429 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 3430 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 3431 } 3432 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 3433 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 3434 } 3435 3436 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3437 ASSERT(vp->v_type == VREG); 3438 3439 zfs_xvattr_set(zp, xvap, tx); 3440 } 3441 3442 if (fuid_dirtied) 3443 zfs_fuid_sync(zfsvfs, tx); 3444 3445 if (mask != 0) 3446 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 3447 3448 mutex_exit(&zp->z_lock); 3449 if (mask & (AT_UID|AT_GID|AT_MODE)) 3450 mutex_exit(&zp->z_acl_lock); 3451 3452 if (attrzp) { 3453 if (mask & (AT_UID|AT_GID|AT_MODE)) 3454 mutex_exit(&attrzp->z_acl_lock); 3455 mutex_exit(&attrzp->z_lock); 3456 } 3457out: 3458 if (err == 0 && attrzp) { 3459 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 3460 xattr_count, tx); 3461 ASSERT(err2 == 0); 3462 } 3463 3464 if (attrzp) 3465 VN_RELE(ZTOV(attrzp)); 3466 if (aclp) 3467 zfs_acl_free(aclp); 3468 3469 if (fuidp) { 3470 zfs_fuid_info_free(fuidp); 3471 fuidp = NULL; 3472 } 3473 3474 if (err) { 3475 dmu_tx_abort(tx); 3476 if (err == ERESTART) 3477 goto top; 3478 } else { 3479 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3480 dmu_tx_commit(tx); 3481 } 3482 3483out2: 3484 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3485 zil_commit(zilog, 0); 3486 3487 ZFS_EXIT(zfsvfs); 3488 return (err); 3489} 3490 3491typedef struct zfs_zlock { 3492 krwlock_t *zl_rwlock; /* lock we acquired */ 3493 znode_t *zl_znode; /* znode we held */ 3494 struct zfs_zlock *zl_next; /* next in list */ 3495} zfs_zlock_t; 3496 3497/* 3498 * Drop locks and release vnodes that were held by zfs_rename_lock(). 3499 */ 3500static void 3501zfs_rename_unlock(zfs_zlock_t **zlpp) 3502{ 3503 zfs_zlock_t *zl; 3504 3505 while ((zl = *zlpp) != NULL) { 3506 if (zl->zl_znode != NULL) 3507 VN_RELE(ZTOV(zl->zl_znode)); 3508 rw_exit(zl->zl_rwlock); 3509 *zlpp = zl->zl_next; 3510 kmem_free(zl, sizeof (*zl)); 3511 } 3512} 3513 3514/* 3515 * Search back through the directory tree, using the ".." entries. 3516 * Lock each directory in the chain to prevent concurrent renames. 3517 * Fail any attempt to move a directory into one of its own descendants. 3518 * XXX - z_parent_lock can overlap with map or grow locks 3519 */ 3520static int 3521zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 3522{ 3523 zfs_zlock_t *zl; 3524 znode_t *zp = tdzp; 3525 uint64_t rootid = zp->z_zfsvfs->z_root; 3526 uint64_t oidp = zp->z_id; 3527 krwlock_t *rwlp = &szp->z_parent_lock; 3528 krw_t rw = RW_WRITER; 3529 3530 /* 3531 * First pass write-locks szp and compares to zp->z_id. 3532 * Later passes read-lock zp and compare to zp->z_parent. 3533 */ 3534 do { 3535 if (!rw_tryenter(rwlp, rw)) { 3536 /* 3537 * Another thread is renaming in this path. 3538 * Note that if we are a WRITER, we don't have any 3539 * parent_locks held yet. 3540 */ 3541 if (rw == RW_READER && zp->z_id > szp->z_id) { 3542 /* 3543 * Drop our locks and restart 3544 */ 3545 zfs_rename_unlock(&zl); 3546 *zlpp = NULL; 3547 zp = tdzp; 3548 oidp = zp->z_id; 3549 rwlp = &szp->z_parent_lock; 3550 rw = RW_WRITER; 3551 continue; 3552 } else { 3553 /* 3554 * Wait for other thread to drop its locks 3555 */ 3556 rw_enter(rwlp, rw); 3557 } 3558 } 3559 3560 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 3561 zl->zl_rwlock = rwlp; 3562 zl->zl_znode = NULL; 3563 zl->zl_next = *zlpp; 3564 *zlpp = zl; 3565 3566 if (oidp == szp->z_id) /* We're a descendant of szp */ 3567 return (EINVAL); 3568 3569 if (oidp == rootid) /* We've hit the top */ 3570 return (0); 3571 3572 if (rw == RW_READER) { /* i.e. not the first pass */ 3573 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp); 3574 if (error) 3575 return (error); 3576 zl->zl_znode = zp; 3577 } 3578 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs), 3579 &oidp, sizeof (oidp)); 3580 rwlp = &zp->z_parent_lock; 3581 rw = RW_READER; 3582 3583 } while (zp->z_id != sdzp->z_id); 3584 3585 return (0); 3586} 3587 3588/* 3589 * Move an entry from the provided source directory to the target 3590 * directory. Change the entry name as indicated. 3591 * 3592 * IN: sdvp - Source directory containing the "old entry". 3593 * snm - Old entry name. 3594 * tdvp - Target directory to contain the "new entry". 3595 * tnm - New entry name. 3596 * cr - credentials of caller. 3597 * ct - caller context 3598 * flags - case flags 3599 * 3600 * RETURN: 0 if success 3601 * error code if failure 3602 * 3603 * Timestamps: 3604 * sdvp,tdvp - ctime|mtime updated 3605 */ 3606/*ARGSUSED*/ 3607static int 3608zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 3609 caller_context_t *ct, int flags) 3610{ 3611 znode_t *tdzp, *szp, *tzp; 3612 znode_t *sdzp = VTOZ(sdvp); 3613 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 3614 zilog_t *zilog; 3615 vnode_t *realvp; 3616 zfs_dirlock_t *sdl, *tdl; 3617 dmu_tx_t *tx; 3618 zfs_zlock_t *zl; 3619 int cmp, serr, terr; 3620 int error = 0; 3621 int zflg = 0; 3622 3623 ZFS_ENTER(zfsvfs); 3624 ZFS_VERIFY_ZP(sdzp); 3625 zilog = zfsvfs->z_log; 3626 3627 /* 3628 * Make sure we have the real vp for the target directory. 3629 */ 3630 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 3631 tdvp = realvp; 3632 3633 if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) { 3634 ZFS_EXIT(zfsvfs); 3635 return (EXDEV); 3636 } 3637 3638 tdzp = VTOZ(tdvp); 3639 ZFS_VERIFY_ZP(tdzp); 3640 if (zfsvfs->z_utf8 && u8_validate(tnm, 3641 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3642 ZFS_EXIT(zfsvfs); 3643 return (EILSEQ); 3644 } 3645 3646 if (flags & FIGNORECASE) 3647 zflg |= ZCILOOK; 3648 3649top: 3650 szp = NULL; 3651 tzp = NULL; 3652 zl = NULL; 3653 3654 /* 3655 * This is to prevent the creation of links into attribute space 3656 * by renaming a linked file into/outof an attribute directory. 3657 * See the comment in zfs_link() for why this is considered bad. 3658 */ 3659 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3660 ZFS_EXIT(zfsvfs); 3661 return (EINVAL); 3662 } 3663 3664 /* 3665 * Lock source and target directory entries. To prevent deadlock, 3666 * a lock ordering must be defined. We lock the directory with 3667 * the smallest object id first, or if it's a tie, the one with 3668 * the lexically first name. 3669 */ 3670 if (sdzp->z_id < tdzp->z_id) { 3671 cmp = -1; 3672 } else if (sdzp->z_id > tdzp->z_id) { 3673 cmp = 1; 3674 } else { 3675 /* 3676 * First compare the two name arguments without 3677 * considering any case folding. 3678 */ 3679 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 3680 3681 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 3682 ASSERT(error == 0 || !zfsvfs->z_utf8); 3683 if (cmp == 0) { 3684 /* 3685 * POSIX: "If the old argument and the new argument 3686 * both refer to links to the same existing file, 3687 * the rename() function shall return successfully 3688 * and perform no other action." 3689 */ 3690 ZFS_EXIT(zfsvfs); 3691 return (0); 3692 } 3693 /* 3694 * If the file system is case-folding, then we may 3695 * have some more checking to do. A case-folding file 3696 * system is either supporting mixed case sensitivity 3697 * access or is completely case-insensitive. Note 3698 * that the file system is always case preserving. 3699 * 3700 * In mixed sensitivity mode case sensitive behavior 3701 * is the default. FIGNORECASE must be used to 3702 * explicitly request case insensitive behavior. 3703 * 3704 * If the source and target names provided differ only 3705 * by case (e.g., a request to rename 'tim' to 'Tim'), 3706 * we will treat this as a special case in the 3707 * case-insensitive mode: as long as the source name 3708 * is an exact match, we will allow this to proceed as 3709 * a name-change request. 3710 */ 3711 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3712 (zfsvfs->z_case == ZFS_CASE_MIXED && 3713 flags & FIGNORECASE)) && 3714 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3715 &error) == 0) { 3716 /* 3717 * case preserving rename request, require exact 3718 * name matches 3719 */ 3720 zflg |= ZCIEXACT; 3721 zflg &= ~ZCILOOK; 3722 } 3723 } 3724 3725 /* 3726 * If the source and destination directories are the same, we should 3727 * grab the z_name_lock of that directory only once. 3728 */ 3729 if (sdzp == tdzp) { 3730 zflg |= ZHAVELOCK; 3731 rw_enter(&sdzp->z_name_lock, RW_READER); 3732 } 3733 3734 if (cmp < 0) { 3735 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3736 ZEXISTS | zflg, NULL, NULL); 3737 terr = zfs_dirent_lock(&tdl, 3738 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3739 } else { 3740 terr = zfs_dirent_lock(&tdl, 3741 tdzp, tnm, &tzp, zflg, NULL, NULL); 3742 serr = zfs_dirent_lock(&sdl, 3743 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3744 NULL, NULL); 3745 } 3746 3747 if (serr) { 3748 /* 3749 * Source entry invalid or not there. 3750 */ 3751 if (!terr) { 3752 zfs_dirent_unlock(tdl); 3753 if (tzp) 3754 VN_RELE(ZTOV(tzp)); 3755 } 3756 3757 if (sdzp == tdzp) 3758 rw_exit(&sdzp->z_name_lock); 3759 3760 /* 3761 * FreeBSD: In OpenSolaris they only check if rename source is 3762 * ".." here, because "." is handled in their lookup. This is 3763 * not the case for FreeBSD, so we check for "." explicitly. 3764 */ 3765 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 3766 serr = EINVAL; 3767 ZFS_EXIT(zfsvfs); 3768 return (serr); 3769 } 3770 if (terr) { 3771 zfs_dirent_unlock(sdl); 3772 VN_RELE(ZTOV(szp)); 3773 3774 if (sdzp == tdzp) 3775 rw_exit(&sdzp->z_name_lock); 3776 3777 if (strcmp(tnm, "..") == 0) 3778 terr = EINVAL; 3779 ZFS_EXIT(zfsvfs); 3780 return (terr); 3781 } 3782 3783 /* 3784 * Must have write access at the source to remove the old entry 3785 * and write access at the target to create the new entry. 3786 * Note that if target and source are the same, this can be 3787 * done in a single check. 3788 */ 3789 3790 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3791 goto out; 3792 3793 if (ZTOV(szp)->v_type == VDIR) { 3794 /* 3795 * Check to make sure rename is valid. 3796 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3797 */ 3798 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3799 goto out; 3800 } 3801 3802 /* 3803 * Does target exist? 3804 */ 3805 if (tzp) { 3806 /* 3807 * Source and target must be the same type. 3808 */ 3809 if (ZTOV(szp)->v_type == VDIR) { 3810 if (ZTOV(tzp)->v_type != VDIR) { 3811 error = ENOTDIR; 3812 goto out; 3813 } 3814 } else { 3815 if (ZTOV(tzp)->v_type == VDIR) { 3816 error = EISDIR; 3817 goto out; 3818 } 3819 } 3820 /* 3821 * POSIX dictates that when the source and target 3822 * entries refer to the same file object, rename 3823 * must do nothing and exit without error. 3824 */ 3825 if (szp->z_id == tzp->z_id) { 3826 error = 0; 3827 goto out; 3828 } 3829 } 3830 3831 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3832 if (tzp) 3833 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3834 3835 /* 3836 * notify the target directory if it is not the same 3837 * as source directory. 3838 */ 3839 if (tdvp != sdvp) { 3840 vnevent_rename_dest_dir(tdvp, ct); 3841 } 3842 3843 tx = dmu_tx_create(zfsvfs->z_os); 3844 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3845 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 3846 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3847 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3848 if (sdzp != tdzp) { 3849 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 3850 zfs_sa_upgrade_txholds(tx, tdzp); 3851 } 3852 if (tzp) { 3853 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 3854 zfs_sa_upgrade_txholds(tx, tzp); 3855 } 3856 3857 zfs_sa_upgrade_txholds(tx, szp); 3858 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3859 error = dmu_tx_assign(tx, TXG_NOWAIT); 3860 if (error) { 3861 if (zl != NULL) 3862 zfs_rename_unlock(&zl); 3863 zfs_dirent_unlock(sdl); 3864 zfs_dirent_unlock(tdl); 3865 3866 if (sdzp == tdzp) 3867 rw_exit(&sdzp->z_name_lock); 3868 3869 VN_RELE(ZTOV(szp)); 3870 if (tzp) 3871 VN_RELE(ZTOV(tzp)); 3872 if (error == ERESTART) { 3873 dmu_tx_wait(tx); 3874 dmu_tx_abort(tx); 3875 goto top; 3876 } 3877 dmu_tx_abort(tx); 3878 ZFS_EXIT(zfsvfs); 3879 return (error); 3880 } 3881 3882 if (tzp) /* Attempt to remove the existing target */ 3883 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 3884 3885 if (error == 0) { 3886 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 3887 if (error == 0) { 3888 szp->z_pflags |= ZFS_AV_MODIFIED; 3889 3890 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 3891 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 3892 ASSERT3U(error, ==, 0); 3893 3894 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 3895 if (error == 0) { 3896 zfs_log_rename(zilog, tx, TX_RENAME | 3897 (flags & FIGNORECASE ? TX_CI : 0), sdzp, 3898 sdl->dl_name, tdzp, tdl->dl_name, szp); 3899 3900 /* 3901 * Update path information for the target vnode 3902 */ 3903 vn_renamepath(tdvp, ZTOV(szp), tnm, 3904 strlen(tnm)); 3905 } else { 3906 /* 3907 * At this point, we have successfully created 3908 * the target name, but have failed to remove 3909 * the source name. Since the create was done 3910 * with the ZRENAMING flag, there are 3911 * complications; for one, the link count is 3912 * wrong. The easiest way to deal with this 3913 * is to remove the newly created target, and 3914 * return the original error. This must 3915 * succeed; fortunately, it is very unlikely to 3916 * fail, since we just created it. 3917 */ 3918 VERIFY3U(zfs_link_destroy(tdl, szp, tx, 3919 ZRENAMING, NULL), ==, 0); 3920 } 3921 } 3922#ifdef FREEBSD_NAMECACHE 3923 if (error == 0) { 3924 cache_purge(sdvp); 3925 cache_purge(tdvp); 3926 } 3927#endif 3928 } 3929 3930 dmu_tx_commit(tx); 3931out: 3932 if (zl != NULL) 3933 zfs_rename_unlock(&zl); 3934 3935 zfs_dirent_unlock(sdl); 3936 zfs_dirent_unlock(tdl); 3937 3938 if (sdzp == tdzp) 3939 rw_exit(&sdzp->z_name_lock); 3940 3941 3942 VN_RELE(ZTOV(szp)); 3943 if (tzp) 3944 VN_RELE(ZTOV(tzp)); 3945 3946 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3947 zil_commit(zilog, 0); 3948 3949 ZFS_EXIT(zfsvfs); 3950 3951 return (error); 3952} 3953 3954/* 3955 * Insert the indicated symbolic reference entry into the directory. 3956 * 3957 * IN: dvp - Directory to contain new symbolic link. 3958 * link - Name for new symlink entry. 3959 * vap - Attributes of new entry. 3960 * target - Target path of new symlink. 3961 * cr - credentials of caller. 3962 * ct - caller context 3963 * flags - case flags 3964 * 3965 * RETURN: 0 if success 3966 * error code if failure 3967 * 3968 * Timestamps: 3969 * dvp - ctime|mtime updated 3970 */ 3971/*ARGSUSED*/ 3972static int 3973zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, 3974 cred_t *cr, kthread_t *td) 3975{ 3976 znode_t *zp, *dzp = VTOZ(dvp); 3977 zfs_dirlock_t *dl; 3978 dmu_tx_t *tx; 3979 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3980 zilog_t *zilog; 3981 uint64_t len = strlen(link); 3982 int error; 3983 int zflg = ZNEW; 3984 zfs_acl_ids_t acl_ids; 3985 boolean_t fuid_dirtied; 3986 uint64_t txtype = TX_SYMLINK; 3987 int flags = 0; 3988 3989 ASSERT(vap->va_type == VLNK); 3990 3991 ZFS_ENTER(zfsvfs); 3992 ZFS_VERIFY_ZP(dzp); 3993 zilog = zfsvfs->z_log; 3994 3995 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 3996 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3997 ZFS_EXIT(zfsvfs); 3998 return (EILSEQ); 3999 } 4000 if (flags & FIGNORECASE) 4001 zflg |= ZCILOOK; 4002 4003 if (len > MAXPATHLEN) { 4004 ZFS_EXIT(zfsvfs); 4005 return (ENAMETOOLONG); 4006 } 4007 4008 if ((error = zfs_acl_ids_create(dzp, 0, 4009 vap, cr, NULL, &acl_ids)) != 0) { 4010 ZFS_EXIT(zfsvfs); 4011 return (error); 4012 } 4013top: 4014 /* 4015 * Attempt to lock directory; fail if entry already exists. 4016 */ 4017 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 4018 if (error) { 4019 zfs_acl_ids_free(&acl_ids); 4020 ZFS_EXIT(zfsvfs); 4021 return (error); 4022 } 4023 4024 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4025 zfs_acl_ids_free(&acl_ids); 4026 zfs_dirent_unlock(dl); 4027 ZFS_EXIT(zfsvfs); 4028 return (error); 4029 } 4030 4031 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 4032 zfs_acl_ids_free(&acl_ids); 4033 zfs_dirent_unlock(dl); 4034 ZFS_EXIT(zfsvfs); 4035 return (EDQUOT); 4036 } 4037 tx = dmu_tx_create(zfsvfs->z_os); 4038 fuid_dirtied = zfsvfs->z_fuid_dirty; 4039 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 4040 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4041 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 4042 ZFS_SA_BASE_ATTR_SIZE + len); 4043 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 4044 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 4045 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 4046 acl_ids.z_aclp->z_acl_bytes); 4047 } 4048 if (fuid_dirtied) 4049 zfs_fuid_txhold(zfsvfs, tx); 4050 error = dmu_tx_assign(tx, TXG_NOWAIT); 4051 if (error) { 4052 zfs_dirent_unlock(dl); 4053 if (error == ERESTART) { 4054 dmu_tx_wait(tx); 4055 dmu_tx_abort(tx); 4056 goto top; 4057 } 4058 zfs_acl_ids_free(&acl_ids); 4059 dmu_tx_abort(tx); 4060 ZFS_EXIT(zfsvfs); 4061 return (error); 4062 } 4063 4064 /* 4065 * Create a new object for the symlink. 4066 * for version 4 ZPL datsets the symlink will be an SA attribute 4067 */ 4068 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 4069 4070 if (fuid_dirtied) 4071 zfs_fuid_sync(zfsvfs, tx); 4072 4073 mutex_enter(&zp->z_lock); 4074 if (zp->z_is_sa) 4075 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 4076 link, len, tx); 4077 else 4078 zfs_sa_symlink(zp, link, len, tx); 4079 mutex_exit(&zp->z_lock); 4080 4081 zp->z_size = len; 4082 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 4083 &zp->z_size, sizeof (zp->z_size), tx); 4084 /* 4085 * Insert the new object into the directory. 4086 */ 4087 (void) zfs_link_create(dl, zp, tx, ZNEW); 4088 4089 if (flags & FIGNORECASE) 4090 txtype |= TX_CI; 4091 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 4092 *vpp = ZTOV(zp); 4093 4094 zfs_acl_ids_free(&acl_ids); 4095 4096 dmu_tx_commit(tx); 4097 4098 zfs_dirent_unlock(dl); 4099 4100 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4101 zil_commit(zilog, 0); 4102 4103 ZFS_EXIT(zfsvfs); 4104 return (error); 4105} 4106 4107/* 4108 * Return, in the buffer contained in the provided uio structure, 4109 * the symbolic path referred to by vp. 4110 * 4111 * IN: vp - vnode of symbolic link. 4112 * uoip - structure to contain the link path. 4113 * cr - credentials of caller. 4114 * ct - caller context 4115 * 4116 * OUT: uio - structure to contain the link path. 4117 * 4118 * RETURN: 0 if success 4119 * error code if failure 4120 * 4121 * Timestamps: 4122 * vp - atime updated 4123 */ 4124/* ARGSUSED */ 4125static int 4126zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 4127{ 4128 znode_t *zp = VTOZ(vp); 4129 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4130 int error; 4131 4132 ZFS_ENTER(zfsvfs); 4133 ZFS_VERIFY_ZP(zp); 4134 4135 mutex_enter(&zp->z_lock); 4136 if (zp->z_is_sa) 4137 error = sa_lookup_uio(zp->z_sa_hdl, 4138 SA_ZPL_SYMLINK(zfsvfs), uio); 4139 else 4140 error = zfs_sa_readlink(zp, uio); 4141 mutex_exit(&zp->z_lock); 4142 4143 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4144 4145 ZFS_EXIT(zfsvfs); 4146 return (error); 4147} 4148 4149/* 4150 * Insert a new entry into directory tdvp referencing svp. 4151 * 4152 * IN: tdvp - Directory to contain new entry. 4153 * svp - vnode of new entry. 4154 * name - name of new entry. 4155 * cr - credentials of caller. 4156 * ct - caller context 4157 * 4158 * RETURN: 0 if success 4159 * error code if failure 4160 * 4161 * Timestamps: 4162 * tdvp - ctime|mtime updated 4163 * svp - ctime updated 4164 */ 4165/* ARGSUSED */ 4166static int 4167zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 4168 caller_context_t *ct, int flags) 4169{ 4170 znode_t *dzp = VTOZ(tdvp); 4171 znode_t *tzp, *szp; 4172 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4173 zilog_t *zilog; 4174 zfs_dirlock_t *dl; 4175 dmu_tx_t *tx; 4176 vnode_t *realvp; 4177 int error; 4178 int zf = ZNEW; 4179 uint64_t parent; 4180 uid_t owner; 4181 4182 ASSERT(tdvp->v_type == VDIR); 4183 4184 ZFS_ENTER(zfsvfs); 4185 ZFS_VERIFY_ZP(dzp); 4186 zilog = zfsvfs->z_log; 4187 4188 if (VOP_REALVP(svp, &realvp, ct) == 0) 4189 svp = realvp; 4190 4191 /* 4192 * POSIX dictates that we return EPERM here. 4193 * Better choices include ENOTSUP or EISDIR. 4194 */ 4195 if (svp->v_type == VDIR) { 4196 ZFS_EXIT(zfsvfs); 4197 return (EPERM); 4198 } 4199 4200 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) { 4201 ZFS_EXIT(zfsvfs); 4202 return (EXDEV); 4203 } 4204 4205 szp = VTOZ(svp); 4206 ZFS_VERIFY_ZP(szp); 4207 4208 /* Prevent links to .zfs/shares files */ 4209 4210 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 4211 &parent, sizeof (uint64_t))) != 0) { 4212 ZFS_EXIT(zfsvfs); 4213 return (error); 4214 } 4215 if (parent == zfsvfs->z_shares_dir) { 4216 ZFS_EXIT(zfsvfs); 4217 return (EPERM); 4218 } 4219 4220 if (zfsvfs->z_utf8 && u8_validate(name, 4221 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4222 ZFS_EXIT(zfsvfs); 4223 return (EILSEQ); 4224 } 4225 if (flags & FIGNORECASE) 4226 zf |= ZCILOOK; 4227 4228 /* 4229 * We do not support links between attributes and non-attributes 4230 * because of the potential security risk of creating links 4231 * into "normal" file space in order to circumvent restrictions 4232 * imposed in attribute space. 4233 */ 4234 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 4235 ZFS_EXIT(zfsvfs); 4236 return (EINVAL); 4237 } 4238 4239 4240 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 4241 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) { 4242 ZFS_EXIT(zfsvfs); 4243 return (EPERM); 4244 } 4245 4246 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4247 ZFS_EXIT(zfsvfs); 4248 return (error); 4249 } 4250 4251top: 4252 /* 4253 * Attempt to lock directory; fail if entry already exists. 4254 */ 4255 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 4256 if (error) { 4257 ZFS_EXIT(zfsvfs); 4258 return (error); 4259 } 4260 4261 tx = dmu_tx_create(zfsvfs->z_os); 4262 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4263 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4264 zfs_sa_upgrade_txholds(tx, szp); 4265 zfs_sa_upgrade_txholds(tx, dzp); 4266 error = dmu_tx_assign(tx, TXG_NOWAIT); 4267 if (error) { 4268 zfs_dirent_unlock(dl); 4269 if (error == ERESTART) { 4270 dmu_tx_wait(tx); 4271 dmu_tx_abort(tx); 4272 goto top; 4273 } 4274 dmu_tx_abort(tx); 4275 ZFS_EXIT(zfsvfs); 4276 return (error); 4277 } 4278 4279 error = zfs_link_create(dl, szp, tx, 0); 4280 4281 if (error == 0) { 4282 uint64_t txtype = TX_LINK; 4283 if (flags & FIGNORECASE) 4284 txtype |= TX_CI; 4285 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 4286 } 4287 4288 dmu_tx_commit(tx); 4289 4290 zfs_dirent_unlock(dl); 4291 4292 if (error == 0) { 4293 vnevent_link(svp, ct); 4294 } 4295 4296 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4297 zil_commit(zilog, 0); 4298 4299 ZFS_EXIT(zfsvfs); 4300 return (error); 4301} 4302 4303#ifdef sun 4304/* 4305 * zfs_null_putapage() is used when the file system has been force 4306 * unmounted. It just drops the pages. 4307 */ 4308/* ARGSUSED */ 4309static int 4310zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4311 size_t *lenp, int flags, cred_t *cr) 4312{ 4313 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 4314 return (0); 4315} 4316 4317/* 4318 * Push a page out to disk, klustering if possible. 4319 * 4320 * IN: vp - file to push page to. 4321 * pp - page to push. 4322 * flags - additional flags. 4323 * cr - credentials of caller. 4324 * 4325 * OUT: offp - start of range pushed. 4326 * lenp - len of range pushed. 4327 * 4328 * RETURN: 0 if success 4329 * error code if failure 4330 * 4331 * NOTE: callers must have locked the page to be pushed. On 4332 * exit, the page (and all other pages in the kluster) must be 4333 * unlocked. 4334 */ 4335/* ARGSUSED */ 4336static int 4337zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4338 size_t *lenp, int flags, cred_t *cr) 4339{ 4340 znode_t *zp = VTOZ(vp); 4341 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4342 dmu_tx_t *tx; 4343 u_offset_t off, koff; 4344 size_t len, klen; 4345 int err; 4346 4347 off = pp->p_offset; 4348 len = PAGESIZE; 4349 /* 4350 * If our blocksize is bigger than the page size, try to kluster 4351 * multiple pages so that we write a full block (thus avoiding 4352 * a read-modify-write). 4353 */ 4354 if (off < zp->z_size && zp->z_blksz > PAGESIZE) { 4355 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 4356 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 4357 ASSERT(koff <= zp->z_size); 4358 if (koff + klen > zp->z_size) 4359 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); 4360 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 4361 } 4362 ASSERT3U(btop(len), ==, btopr(len)); 4363 4364 /* 4365 * Can't push pages past end-of-file. 4366 */ 4367 if (off >= zp->z_size) { 4368 /* ignore all pages */ 4369 err = 0; 4370 goto out; 4371 } else if (off + len > zp->z_size) { 4372 int npages = btopr(zp->z_size - off); 4373 page_t *trunc; 4374 4375 page_list_break(&pp, &trunc, npages); 4376 /* ignore pages past end of file */ 4377 if (trunc) 4378 pvn_write_done(trunc, flags); 4379 len = zp->z_size - off; 4380 } 4381 4382 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4383 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4384 err = EDQUOT; 4385 goto out; 4386 } 4387top: 4388 tx = dmu_tx_create(zfsvfs->z_os); 4389 dmu_tx_hold_write(tx, zp->z_id, off, len); 4390 4391 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4392 zfs_sa_upgrade_txholds(tx, zp); 4393 err = dmu_tx_assign(tx, TXG_NOWAIT); 4394 if (err != 0) { 4395 if (err == ERESTART) { 4396 dmu_tx_wait(tx); 4397 dmu_tx_abort(tx); 4398 goto top; 4399 } 4400 dmu_tx_abort(tx); 4401 goto out; 4402 } 4403 4404 if (zp->z_blksz <= PAGESIZE) { 4405 caddr_t va = zfs_map_page(pp, S_READ); 4406 ASSERT3U(len, <=, PAGESIZE); 4407 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 4408 zfs_unmap_page(pp, va); 4409 } else { 4410 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 4411 } 4412 4413 if (err == 0) { 4414 uint64_t mtime[2], ctime[2]; 4415 sa_bulk_attr_t bulk[3]; 4416 int count = 0; 4417 4418 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4419 &mtime, 16); 4420 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4421 &ctime, 16); 4422 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4423 &zp->z_pflags, 8); 4424 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4425 B_TRUE); 4426 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4427 } 4428 dmu_tx_commit(tx); 4429 4430out: 4431 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 4432 if (offp) 4433 *offp = off; 4434 if (lenp) 4435 *lenp = len; 4436 4437 return (err); 4438} 4439 4440/* 4441 * Copy the portion of the file indicated from pages into the file. 4442 * The pages are stored in a page list attached to the files vnode. 4443 * 4444 * IN: vp - vnode of file to push page data to. 4445 * off - position in file to put data. 4446 * len - amount of data to write. 4447 * flags - flags to control the operation. 4448 * cr - credentials of caller. 4449 * ct - caller context. 4450 * 4451 * RETURN: 0 if success 4452 * error code if failure 4453 * 4454 * Timestamps: 4455 * vp - ctime|mtime updated 4456 */ 4457/*ARGSUSED*/ 4458static int 4459zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4460 caller_context_t *ct) 4461{ 4462 znode_t *zp = VTOZ(vp); 4463 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4464 page_t *pp; 4465 size_t io_len; 4466 u_offset_t io_off; 4467 uint_t blksz; 4468 rl_t *rl; 4469 int error = 0; 4470 4471 ZFS_ENTER(zfsvfs); 4472 ZFS_VERIFY_ZP(zp); 4473 4474 /* 4475 * Align this request to the file block size in case we kluster. 4476 * XXX - this can result in pretty aggresive locking, which can 4477 * impact simultanious read/write access. One option might be 4478 * to break up long requests (len == 0) into block-by-block 4479 * operations to get narrower locking. 4480 */ 4481 blksz = zp->z_blksz; 4482 if (ISP2(blksz)) 4483 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 4484 else 4485 io_off = 0; 4486 if (len > 0 && ISP2(blksz)) 4487 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 4488 else 4489 io_len = 0; 4490 4491 if (io_len == 0) { 4492 /* 4493 * Search the entire vp list for pages >= io_off. 4494 */ 4495 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 4496 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 4497 goto out; 4498 } 4499 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 4500 4501 if (off > zp->z_size) { 4502 /* past end of file */ 4503 zfs_range_unlock(rl); 4504 ZFS_EXIT(zfsvfs); 4505 return (0); 4506 } 4507 4508 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); 4509 4510 for (off = io_off; io_off < off + len; io_off += io_len) { 4511 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 4512 pp = page_lookup(vp, io_off, 4513 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 4514 } else { 4515 pp = page_lookup_nowait(vp, io_off, 4516 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 4517 } 4518 4519 if (pp != NULL && pvn_getdirty(pp, flags)) { 4520 int err; 4521 4522 /* 4523 * Found a dirty page to push 4524 */ 4525 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 4526 if (err) 4527 error = err; 4528 } else { 4529 io_len = PAGESIZE; 4530 } 4531 } 4532out: 4533 zfs_range_unlock(rl); 4534 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4535 zil_commit(zfsvfs->z_log, zp->z_id); 4536 ZFS_EXIT(zfsvfs); 4537 return (error); 4538} 4539#endif /* sun */ 4540 4541/*ARGSUSED*/ 4542void 4543zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4544{ 4545 znode_t *zp = VTOZ(vp); 4546 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4547 int error; 4548 4549 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4550 if (zp->z_sa_hdl == NULL) { 4551 /* 4552 * The fs has been unmounted, or we did a 4553 * suspend/resume and this file no longer exists. 4554 */ 4555 VI_LOCK(vp); 4556 ASSERT(vp->v_count <= 1); 4557 vp->v_count = 0; 4558 VI_UNLOCK(vp); 4559 vrecycle(vp); 4560 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4561 return; 4562 } 4563 4564 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4565 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4566 4567 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4568 zfs_sa_upgrade_txholds(tx, zp); 4569 error = dmu_tx_assign(tx, TXG_WAIT); 4570 if (error) { 4571 dmu_tx_abort(tx); 4572 } else { 4573 mutex_enter(&zp->z_lock); 4574 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4575 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4576 zp->z_atime_dirty = 0; 4577 mutex_exit(&zp->z_lock); 4578 dmu_tx_commit(tx); 4579 } 4580 } 4581 4582 zfs_zinactive(zp); 4583 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4584} 4585 4586#ifdef sun 4587/* 4588 * Bounds-check the seek operation. 4589 * 4590 * IN: vp - vnode seeking within 4591 * ooff - old file offset 4592 * noffp - pointer to new file offset 4593 * ct - caller context 4594 * 4595 * RETURN: 0 if success 4596 * EINVAL if new offset invalid 4597 */ 4598/* ARGSUSED */ 4599static int 4600zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 4601 caller_context_t *ct) 4602{ 4603 if (vp->v_type == VDIR) 4604 return (0); 4605 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4606} 4607 4608/* 4609 * Pre-filter the generic locking function to trap attempts to place 4610 * a mandatory lock on a memory mapped file. 4611 */ 4612static int 4613zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 4614 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 4615{ 4616 znode_t *zp = VTOZ(vp); 4617 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4618 4619 ZFS_ENTER(zfsvfs); 4620 ZFS_VERIFY_ZP(zp); 4621 4622 /* 4623 * We are following the UFS semantics with respect to mapcnt 4624 * here: If we see that the file is mapped already, then we will 4625 * return an error, but we don't worry about races between this 4626 * function and zfs_map(). 4627 */ 4628 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { 4629 ZFS_EXIT(zfsvfs); 4630 return (EAGAIN); 4631 } 4632 ZFS_EXIT(zfsvfs); 4633 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4634} 4635 4636/* 4637 * If we can't find a page in the cache, we will create a new page 4638 * and fill it with file data. For efficiency, we may try to fill 4639 * multiple pages at once (klustering) to fill up the supplied page 4640 * list. Note that the pages to be filled are held with an exclusive 4641 * lock to prevent access by other threads while they are being filled. 4642 */ 4643static int 4644zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 4645 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 4646{ 4647 znode_t *zp = VTOZ(vp); 4648 page_t *pp, *cur_pp; 4649 objset_t *os = zp->z_zfsvfs->z_os; 4650 u_offset_t io_off, total; 4651 size_t io_len; 4652 int err; 4653 4654 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4655 /* 4656 * We only have a single page, don't bother klustering 4657 */ 4658 io_off = off; 4659 io_len = PAGESIZE; 4660 pp = page_create_va(vp, io_off, io_len, 4661 PG_EXCL | PG_WAIT, seg, addr); 4662 } else { 4663 /* 4664 * Try to find enough pages to fill the page list 4665 */ 4666 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4667 &io_len, off, plsz, 0); 4668 } 4669 if (pp == NULL) { 4670 /* 4671 * The page already exists, nothing to do here. 4672 */ 4673 *pl = NULL; 4674 return (0); 4675 } 4676 4677 /* 4678 * Fill the pages in the kluster. 4679 */ 4680 cur_pp = pp; 4681 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4682 caddr_t va; 4683 4684 ASSERT3U(io_off, ==, cur_pp->p_offset); 4685 va = zfs_map_page(cur_pp, S_WRITE); 4686 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4687 DMU_READ_PREFETCH); 4688 zfs_unmap_page(cur_pp, va); 4689 if (err) { 4690 /* On error, toss the entire kluster */ 4691 pvn_read_done(pp, B_ERROR); 4692 /* convert checksum errors into IO errors */ 4693 if (err == ECKSUM) 4694 err = EIO; 4695 return (err); 4696 } 4697 cur_pp = cur_pp->p_next; 4698 } 4699 4700 /* 4701 * Fill in the page list array from the kluster starting 4702 * from the desired offset `off'. 4703 * NOTE: the page list will always be null terminated. 4704 */ 4705 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4706 ASSERT(pl == NULL || (*pl)->p_offset == off); 4707 4708 return (0); 4709} 4710 4711/* 4712 * Return pointers to the pages for the file region [off, off + len] 4713 * in the pl array. If plsz is greater than len, this function may 4714 * also return page pointers from after the specified region 4715 * (i.e. the region [off, off + plsz]). These additional pages are 4716 * only returned if they are already in the cache, or were created as 4717 * part of a klustered read. 4718 * 4719 * IN: vp - vnode of file to get data from. 4720 * off - position in file to get data from. 4721 * len - amount of data to retrieve. 4722 * plsz - length of provided page list. 4723 * seg - segment to obtain pages for. 4724 * addr - virtual address of fault. 4725 * rw - mode of created pages. 4726 * cr - credentials of caller. 4727 * ct - caller context. 4728 * 4729 * OUT: protp - protection mode of created pages. 4730 * pl - list of pages created. 4731 * 4732 * RETURN: 0 if success 4733 * error code if failure 4734 * 4735 * Timestamps: 4736 * vp - atime updated 4737 */ 4738/* ARGSUSED */ 4739static int 4740zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4741 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4742 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4743{ 4744 znode_t *zp = VTOZ(vp); 4745 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4746 page_t **pl0 = pl; 4747 int err = 0; 4748 4749 /* we do our own caching, faultahead is unnecessary */ 4750 if (pl == NULL) 4751 return (0); 4752 else if (len > plsz) 4753 len = plsz; 4754 else 4755 len = P2ROUNDUP(len, PAGESIZE); 4756 ASSERT(plsz >= len); 4757 4758 ZFS_ENTER(zfsvfs); 4759 ZFS_VERIFY_ZP(zp); 4760 4761 if (protp) 4762 *protp = PROT_ALL; 4763 4764 /* 4765 * Loop through the requested range [off, off + len) looking 4766 * for pages. If we don't find a page, we will need to create 4767 * a new page and fill it with data from the file. 4768 */ 4769 while (len > 0) { 4770 if (*pl = page_lookup(vp, off, SE_SHARED)) 4771 *(pl+1) = NULL; 4772 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 4773 goto out; 4774 while (*pl) { 4775 ASSERT3U((*pl)->p_offset, ==, off); 4776 off += PAGESIZE; 4777 addr += PAGESIZE; 4778 if (len > 0) { 4779 ASSERT3U(len, >=, PAGESIZE); 4780 len -= PAGESIZE; 4781 } 4782 ASSERT3U(plsz, >=, PAGESIZE); 4783 plsz -= PAGESIZE; 4784 pl++; 4785 } 4786 } 4787 4788 /* 4789 * Fill out the page array with any pages already in the cache. 4790 */ 4791 while (plsz > 0 && 4792 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 4793 off += PAGESIZE; 4794 plsz -= PAGESIZE; 4795 } 4796out: 4797 if (err) { 4798 /* 4799 * Release any pages we have previously locked. 4800 */ 4801 while (pl > pl0) 4802 page_unlock(*--pl); 4803 } else { 4804 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4805 } 4806 4807 *pl = NULL; 4808 4809 ZFS_EXIT(zfsvfs); 4810 return (err); 4811} 4812 4813/* 4814 * Request a memory map for a section of a file. This code interacts 4815 * with common code and the VM system as follows: 4816 * 4817 * common code calls mmap(), which ends up in smmap_common() 4818 * 4819 * this calls VOP_MAP(), which takes you into (say) zfs 4820 * 4821 * zfs_map() calls as_map(), passing segvn_create() as the callback 4822 * 4823 * segvn_create() creates the new segment and calls VOP_ADDMAP() 4824 * 4825 * zfs_addmap() updates z_mapcnt 4826 */ 4827/*ARGSUSED*/ 4828static int 4829zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4830 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4831 caller_context_t *ct) 4832{ 4833 znode_t *zp = VTOZ(vp); 4834 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4835 segvn_crargs_t vn_a; 4836 int error; 4837 4838 ZFS_ENTER(zfsvfs); 4839 ZFS_VERIFY_ZP(zp); 4840 4841 if ((prot & PROT_WRITE) && (zp->z_pflags & 4842 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 4843 ZFS_EXIT(zfsvfs); 4844 return (EPERM); 4845 } 4846 4847 if ((prot & (PROT_READ | PROT_EXEC)) && 4848 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 4849 ZFS_EXIT(zfsvfs); 4850 return (EACCES); 4851 } 4852 4853 if (vp->v_flag & VNOMAP) { 4854 ZFS_EXIT(zfsvfs); 4855 return (ENOSYS); 4856 } 4857 4858 if (off < 0 || len > MAXOFFSET_T - off) { 4859 ZFS_EXIT(zfsvfs); 4860 return (ENXIO); 4861 } 4862 4863 if (vp->v_type != VREG) { 4864 ZFS_EXIT(zfsvfs); 4865 return (ENODEV); 4866 } 4867 4868 /* 4869 * If file is locked, disallow mapping. 4870 */ 4871 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { 4872 ZFS_EXIT(zfsvfs); 4873 return (EAGAIN); 4874 } 4875 4876 as_rangelock(as); 4877 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 4878 if (error != 0) { 4879 as_rangeunlock(as); 4880 ZFS_EXIT(zfsvfs); 4881 return (error); 4882 } 4883 4884 vn_a.vp = vp; 4885 vn_a.offset = (u_offset_t)off; 4886 vn_a.type = flags & MAP_TYPE; 4887 vn_a.prot = prot; 4888 vn_a.maxprot = maxprot; 4889 vn_a.cred = cr; 4890 vn_a.amp = NULL; 4891 vn_a.flags = flags & ~MAP_TYPE; 4892 vn_a.szc = 0; 4893 vn_a.lgrp_mem_policy_flags = 0; 4894 4895 error = as_map(as, *addrp, len, segvn_create, &vn_a); 4896 4897 as_rangeunlock(as); 4898 ZFS_EXIT(zfsvfs); 4899 return (error); 4900} 4901 4902/* ARGSUSED */ 4903static int 4904zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4905 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4906 caller_context_t *ct) 4907{ 4908 uint64_t pages = btopr(len); 4909 4910 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 4911 return (0); 4912} 4913 4914/* 4915 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 4916 * more accurate mtime for the associated file. Since we don't have a way of 4917 * detecting when the data was actually modified, we have to resort to 4918 * heuristics. If an explicit msync() is done, then we mark the mtime when the 4919 * last page is pushed. The problem occurs when the msync() call is omitted, 4920 * which by far the most common case: 4921 * 4922 * open() 4923 * mmap() 4924 * <modify memory> 4925 * munmap() 4926 * close() 4927 * <time lapse> 4928 * putpage() via fsflush 4929 * 4930 * If we wait until fsflush to come along, we can have a modification time that 4931 * is some arbitrary point in the future. In order to prevent this in the 4932 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 4933 * torn down. 4934 */ 4935/* ARGSUSED */ 4936static int 4937zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4938 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 4939 caller_context_t *ct) 4940{ 4941 uint64_t pages = btopr(len); 4942 4943 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 4944 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 4945 4946 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 4947 vn_has_cached_data(vp)) 4948 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 4949 4950 return (0); 4951} 4952 4953/* 4954 * Free or allocate space in a file. Currently, this function only 4955 * supports the `F_FREESP' command. However, this command is somewhat 4956 * misnamed, as its functionality includes the ability to allocate as 4957 * well as free space. 4958 * 4959 * IN: vp - vnode of file to free data in. 4960 * cmd - action to take (only F_FREESP supported). 4961 * bfp - section of file to free/alloc. 4962 * flag - current file open mode flags. 4963 * offset - current file offset. 4964 * cr - credentials of caller [UNUSED]. 4965 * ct - caller context. 4966 * 4967 * RETURN: 0 if success 4968 * error code if failure 4969 * 4970 * Timestamps: 4971 * vp - ctime|mtime updated 4972 */ 4973/* ARGSUSED */ 4974static int 4975zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 4976 offset_t offset, cred_t *cr, caller_context_t *ct) 4977{ 4978 znode_t *zp = VTOZ(vp); 4979 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4980 uint64_t off, len; 4981 int error; 4982 4983 ZFS_ENTER(zfsvfs); 4984 ZFS_VERIFY_ZP(zp); 4985 4986 if (cmd != F_FREESP) { 4987 ZFS_EXIT(zfsvfs); 4988 return (EINVAL); 4989 } 4990 4991 if (error = convoff(vp, bfp, 0, offset)) { 4992 ZFS_EXIT(zfsvfs); 4993 return (error); 4994 } 4995 4996 if (bfp->l_len < 0) { 4997 ZFS_EXIT(zfsvfs); 4998 return (EINVAL); 4999 } 5000 5001 off = bfp->l_start; 5002 len = bfp->l_len; /* 0 means from off to end of file */ 5003 5004 error = zfs_freesp(zp, off, len, flag, TRUE); 5005 5006 ZFS_EXIT(zfsvfs); 5007 return (error); 5008} 5009#endif /* sun */ 5010 5011CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 5012CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 5013 5014/*ARGSUSED*/ 5015static int 5016zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 5017{ 5018 znode_t *zp = VTOZ(vp); 5019 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5020 uint32_t gen; 5021 uint64_t gen64; 5022 uint64_t object = zp->z_id; 5023 zfid_short_t *zfid; 5024 int size, i, error; 5025 5026 ZFS_ENTER(zfsvfs); 5027 ZFS_VERIFY_ZP(zp); 5028 5029 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 5030 &gen64, sizeof (uint64_t))) != 0) { 5031 ZFS_EXIT(zfsvfs); 5032 return (error); 5033 } 5034 5035 gen = (uint32_t)gen64; 5036 5037 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 5038 fidp->fid_len = size; 5039 5040 zfid = (zfid_short_t *)fidp; 5041 5042 zfid->zf_len = size; 5043 5044 for (i = 0; i < sizeof (zfid->zf_object); i++) 5045 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 5046 5047 /* Must have a non-zero generation number to distinguish from .zfs */ 5048 if (gen == 0) 5049 gen = 1; 5050 for (i = 0; i < sizeof (zfid->zf_gen); i++) 5051 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 5052 5053 if (size == LONG_FID_LEN) { 5054 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 5055 zfid_long_t *zlfid; 5056 5057 zlfid = (zfid_long_t *)fidp; 5058 5059 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 5060 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 5061 5062 /* XXX - this should be the generation number for the objset */ 5063 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 5064 zlfid->zf_setgen[i] = 0; 5065 } 5066 5067 ZFS_EXIT(zfsvfs); 5068 return (0); 5069} 5070 5071static int 5072zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 5073 caller_context_t *ct) 5074{ 5075 znode_t *zp, *xzp; 5076 zfsvfs_t *zfsvfs; 5077 zfs_dirlock_t *dl; 5078 int error; 5079 5080 switch (cmd) { 5081 case _PC_LINK_MAX: 5082 *valp = INT_MAX; 5083 return (0); 5084 5085 case _PC_FILESIZEBITS: 5086 *valp = 64; 5087 return (0); 5088#ifdef sun 5089 case _PC_XATTR_EXISTS: 5090 zp = VTOZ(vp); 5091 zfsvfs = zp->z_zfsvfs; 5092 ZFS_ENTER(zfsvfs); 5093 ZFS_VERIFY_ZP(zp); 5094 *valp = 0; 5095 error = zfs_dirent_lock(&dl, zp, "", &xzp, 5096 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 5097 if (error == 0) { 5098 zfs_dirent_unlock(dl); 5099 if (!zfs_dirempty(xzp)) 5100 *valp = 1; 5101 VN_RELE(ZTOV(xzp)); 5102 } else if (error == ENOENT) { 5103 /* 5104 * If there aren't extended attributes, it's the 5105 * same as having zero of them. 5106 */ 5107 error = 0; 5108 } 5109 ZFS_EXIT(zfsvfs); 5110 return (error); 5111 5112 case _PC_SATTR_ENABLED: 5113 case _PC_SATTR_EXISTS: 5114 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 5115 (vp->v_type == VREG || vp->v_type == VDIR); 5116 return (0); 5117 5118 case _PC_ACCESS_FILTERING: 5119 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 5120 vp->v_type == VDIR; 5121 return (0); 5122 5123 case _PC_ACL_ENABLED: 5124 *valp = _ACL_ACE_ENABLED; 5125 return (0); 5126#endif /* sun */ 5127 case _PC_MIN_HOLE_SIZE: 5128 *valp = (int)SPA_MINBLOCKSIZE; 5129 return (0); 5130#ifdef sun 5131 case _PC_TIMESTAMP_RESOLUTION: 5132 /* nanosecond timestamp resolution */ 5133 *valp = 1L; 5134 return (0); 5135#endif /* sun */ 5136 case _PC_ACL_EXTENDED: 5137 *valp = 0; 5138 return (0); 5139 5140 case _PC_ACL_NFS4: 5141 *valp = 1; 5142 return (0); 5143 5144 case _PC_ACL_PATH_MAX: 5145 *valp = ACL_MAX_ENTRIES; 5146 return (0); 5147 5148 default: 5149 return (EOPNOTSUPP); 5150 } 5151} 5152 5153/*ARGSUSED*/ 5154static int 5155zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5156 caller_context_t *ct) 5157{ 5158 znode_t *zp = VTOZ(vp); 5159 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5160 int error; 5161 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5162 5163 ZFS_ENTER(zfsvfs); 5164 ZFS_VERIFY_ZP(zp); 5165 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 5166 ZFS_EXIT(zfsvfs); 5167 5168 return (error); 5169} 5170 5171/*ARGSUSED*/ 5172int 5173zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5174 caller_context_t *ct) 5175{ 5176 znode_t *zp = VTOZ(vp); 5177 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5178 int error; 5179 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5180 zilog_t *zilog = zfsvfs->z_log; 5181 5182 ZFS_ENTER(zfsvfs); 5183 ZFS_VERIFY_ZP(zp); 5184 5185 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 5186 5187 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 5188 zil_commit(zilog, 0); 5189 5190 ZFS_EXIT(zfsvfs); 5191 return (error); 5192} 5193 5194#ifdef sun 5195/* 5196 * Tunable, both must be a power of 2. 5197 * 5198 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf 5199 * zcr_blksz_max: if set to less than the file block size, allow loaning out of 5200 * an arcbuf for a partial block read 5201 */ 5202int zcr_blksz_min = (1 << 10); /* 1K */ 5203int zcr_blksz_max = (1 << 17); /* 128K */ 5204 5205/*ARGSUSED*/ 5206static int 5207zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr, 5208 caller_context_t *ct) 5209{ 5210 znode_t *zp = VTOZ(vp); 5211 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5212 int max_blksz = zfsvfs->z_max_blksz; 5213 uio_t *uio = &xuio->xu_uio; 5214 ssize_t size = uio->uio_resid; 5215 offset_t offset = uio->uio_loffset; 5216 int blksz; 5217 int fullblk, i; 5218 arc_buf_t *abuf; 5219 ssize_t maxsize; 5220 int preamble, postamble; 5221 5222 if (xuio->xu_type != UIOTYPE_ZEROCOPY) 5223 return (EINVAL); 5224 5225 ZFS_ENTER(zfsvfs); 5226 ZFS_VERIFY_ZP(zp); 5227 switch (ioflag) { 5228 case UIO_WRITE: 5229 /* 5230 * Loan out an arc_buf for write if write size is bigger than 5231 * max_blksz, and the file's block size is also max_blksz. 5232 */ 5233 blksz = max_blksz; 5234 if (size < blksz || zp->z_blksz != blksz) { 5235 ZFS_EXIT(zfsvfs); 5236 return (EINVAL); 5237 } 5238 /* 5239 * Caller requests buffers for write before knowing where the 5240 * write offset might be (e.g. NFS TCP write). 5241 */ 5242 if (offset == -1) { 5243 preamble = 0; 5244 } else { 5245 preamble = P2PHASE(offset, blksz); 5246 if (preamble) { 5247 preamble = blksz - preamble; 5248 size -= preamble; 5249 } 5250 } 5251 5252 postamble = P2PHASE(size, blksz); 5253 size -= postamble; 5254 5255 fullblk = size / blksz; 5256 (void) dmu_xuio_init(xuio, 5257 (preamble != 0) + fullblk + (postamble != 0)); 5258 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble, 5259 int, postamble, int, 5260 (preamble != 0) + fullblk + (postamble != 0)); 5261 5262 /* 5263 * Have to fix iov base/len for partial buffers. They 5264 * currently represent full arc_buf's. 5265 */ 5266 if (preamble) { 5267 /* data begins in the middle of the arc_buf */ 5268 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5269 blksz); 5270 ASSERT(abuf); 5271 (void) dmu_xuio_add(xuio, abuf, 5272 blksz - preamble, preamble); 5273 } 5274 5275 for (i = 0; i < fullblk; i++) { 5276 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5277 blksz); 5278 ASSERT(abuf); 5279 (void) dmu_xuio_add(xuio, abuf, 0, blksz); 5280 } 5281 5282 if (postamble) { 5283 /* data ends in the middle of the arc_buf */ 5284 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5285 blksz); 5286 ASSERT(abuf); 5287 (void) dmu_xuio_add(xuio, abuf, 0, postamble); 5288 } 5289 break; 5290 case UIO_READ: 5291 /* 5292 * Loan out an arc_buf for read if the read size is larger than 5293 * the current file block size. Block alignment is not 5294 * considered. Partial arc_buf will be loaned out for read. 5295 */ 5296 blksz = zp->z_blksz; 5297 if (blksz < zcr_blksz_min) 5298 blksz = zcr_blksz_min; 5299 if (blksz > zcr_blksz_max) 5300 blksz = zcr_blksz_max; 5301 /* avoid potential complexity of dealing with it */ 5302 if (blksz > max_blksz) { 5303 ZFS_EXIT(zfsvfs); 5304 return (EINVAL); 5305 } 5306 5307 maxsize = zp->z_size - uio->uio_loffset; 5308 if (size > maxsize) 5309 size = maxsize; 5310 5311 if (size < blksz || vn_has_cached_data(vp)) { 5312 ZFS_EXIT(zfsvfs); 5313 return (EINVAL); 5314 } 5315 break; 5316 default: 5317 ZFS_EXIT(zfsvfs); 5318 return (EINVAL); 5319 } 5320 5321 uio->uio_extflg = UIO_XUIO; 5322 XUIO_XUZC_RW(xuio) = ioflag; 5323 ZFS_EXIT(zfsvfs); 5324 return (0); 5325} 5326 5327/*ARGSUSED*/ 5328static int 5329zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct) 5330{ 5331 int i; 5332 arc_buf_t *abuf; 5333 int ioflag = XUIO_XUZC_RW(xuio); 5334 5335 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY); 5336 5337 i = dmu_xuio_cnt(xuio); 5338 while (i-- > 0) { 5339 abuf = dmu_xuio_arcbuf(xuio, i); 5340 /* 5341 * if abuf == NULL, it must be a write buffer 5342 * that has been returned in zfs_write(). 5343 */ 5344 if (abuf) 5345 dmu_return_arcbuf(abuf); 5346 ASSERT(abuf || ioflag == UIO_WRITE); 5347 } 5348 5349 dmu_xuio_fini(xuio); 5350 return (0); 5351} 5352 5353/* 5354 * Predeclare these here so that the compiler assumes that 5355 * this is an "old style" function declaration that does 5356 * not include arguments => we won't get type mismatch errors 5357 * in the initializations that follow. 5358 */ 5359static int zfs_inval(); 5360static int zfs_isdir(); 5361 5362static int 5363zfs_inval() 5364{ 5365 return (EINVAL); 5366} 5367 5368static int 5369zfs_isdir() 5370{ 5371 return (EISDIR); 5372} 5373/* 5374 * Directory vnode operations template 5375 */ 5376vnodeops_t *zfs_dvnodeops; 5377const fs_operation_def_t zfs_dvnodeops_template[] = { 5378 VOPNAME_OPEN, { .vop_open = zfs_open }, 5379 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5380 VOPNAME_READ, { .error = zfs_isdir }, 5381 VOPNAME_WRITE, { .error = zfs_isdir }, 5382 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5383 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5384 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5385 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5386 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5387 VOPNAME_CREATE, { .vop_create = zfs_create }, 5388 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5389 VOPNAME_LINK, { .vop_link = zfs_link }, 5390 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5391 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 5392 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5393 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5394 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 5395 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5396 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5397 VOPNAME_FID, { .vop_fid = zfs_fid }, 5398 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5399 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5400 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5401 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5402 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5403 NULL, NULL 5404}; 5405 5406/* 5407 * Regular file vnode operations template 5408 */ 5409vnodeops_t *zfs_fvnodeops; 5410const fs_operation_def_t zfs_fvnodeops_template[] = { 5411 VOPNAME_OPEN, { .vop_open = zfs_open }, 5412 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5413 VOPNAME_READ, { .vop_read = zfs_read }, 5414 VOPNAME_WRITE, { .vop_write = zfs_write }, 5415 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5416 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5417 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5418 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5419 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5420 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5421 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5422 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5423 VOPNAME_FID, { .vop_fid = zfs_fid }, 5424 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5425 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 5426 VOPNAME_SPACE, { .vop_space = zfs_space }, 5427 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 5428 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 5429 VOPNAME_MAP, { .vop_map = zfs_map }, 5430 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 5431 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 5432 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5433 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5434 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5435 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5436 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf }, 5437 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf }, 5438 NULL, NULL 5439}; 5440 5441/* 5442 * Symbolic link vnode operations template 5443 */ 5444vnodeops_t *zfs_symvnodeops; 5445const fs_operation_def_t zfs_symvnodeops_template[] = { 5446 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5447 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5448 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5449 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5450 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 5451 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5452 VOPNAME_FID, { .vop_fid = zfs_fid }, 5453 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5454 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5455 NULL, NULL 5456}; 5457 5458/* 5459 * special share hidden files vnode operations template 5460 */ 5461vnodeops_t *zfs_sharevnodeops; 5462const fs_operation_def_t zfs_sharevnodeops_template[] = { 5463 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5464 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5465 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5466 VOPNAME_FID, { .vop_fid = zfs_fid }, 5467 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5468 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5469 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5470 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5471 NULL, NULL 5472}; 5473 5474/* 5475 * Extended attribute directory vnode operations template 5476 * This template is identical to the directory vnodes 5477 * operation template except for restricted operations: 5478 * VOP_MKDIR() 5479 * VOP_SYMLINK() 5480 * Note that there are other restrictions embedded in: 5481 * zfs_create() - restrict type to VREG 5482 * zfs_link() - no links into/out of attribute space 5483 * zfs_rename() - no moves into/out of attribute space 5484 */ 5485vnodeops_t *zfs_xdvnodeops; 5486const fs_operation_def_t zfs_xdvnodeops_template[] = { 5487 VOPNAME_OPEN, { .vop_open = zfs_open }, 5488 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5489 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5490 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5491 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5492 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5493 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5494 VOPNAME_CREATE, { .vop_create = zfs_create }, 5495 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5496 VOPNAME_LINK, { .vop_link = zfs_link }, 5497 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5498 VOPNAME_MKDIR, { .error = zfs_inval }, 5499 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5500 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5501 VOPNAME_SYMLINK, { .error = zfs_inval }, 5502 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5503 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5504 VOPNAME_FID, { .vop_fid = zfs_fid }, 5505 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5506 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5507 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5508 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5509 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5510 NULL, NULL 5511}; 5512 5513/* 5514 * Error vnode operations template 5515 */ 5516vnodeops_t *zfs_evnodeops; 5517const fs_operation_def_t zfs_evnodeops_template[] = { 5518 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5519 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5520 NULL, NULL 5521}; 5522#endif /* sun */ 5523 5524static int 5525ioflags(int ioflags) 5526{ 5527 int flags = 0; 5528 5529 if (ioflags & IO_APPEND) 5530 flags |= FAPPEND; 5531 if (ioflags & IO_NDELAY) 5532 flags |= FNONBLOCK; 5533 if (ioflags & IO_SYNC) 5534 flags |= (FSYNC | FDSYNC | FRSYNC); 5535 5536 return (flags); 5537} 5538 5539static int 5540zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage) 5541{ 5542 znode_t *zp = VTOZ(vp); 5543 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5544 objset_t *os = zp->z_zfsvfs->z_os; 5545 vm_page_t mreq; 5546 vm_object_t object; 5547 caddr_t va; 5548 struct sf_buf *sf; 5549 int i, error; 5550 int pcount, size; 5551 5552 ZFS_ENTER(zfsvfs); 5553 ZFS_VERIFY_ZP(zp); 5554 5555 pcount = round_page(count) / PAGE_SIZE; 5556 mreq = m[reqpage]; 5557 object = mreq->object; 5558 error = 0; 5559 5560 KASSERT(vp->v_object == object, ("mismatching object")); 5561 5562 VM_OBJECT_LOCK(object); 5563 5564 for (i = 0; i < pcount; i++) { 5565 if (i != reqpage) { 5566 vm_page_lock(m[i]); 5567 vm_page_free(m[i]); 5568 vm_page_unlock(m[i]); 5569 } 5570 } 5571 5572 if (mreq->valid) { 5573 if (mreq->valid != VM_PAGE_BITS_ALL) 5574 vm_page_zero_invalid(mreq, TRUE); 5575 VM_OBJECT_UNLOCK(object); 5576 ZFS_EXIT(zfsvfs); 5577 return (VM_PAGER_OK); 5578 } 5579 5580 PCPU_INC(cnt.v_vnodein); 5581 PCPU_INC(cnt.v_vnodepgsin); 5582 5583 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) { 5584 VM_OBJECT_UNLOCK(object); 5585 ZFS_EXIT(zfsvfs); 5586 return (VM_PAGER_BAD); 5587 } 5588 5589 size = PAGE_SIZE; 5590 if (IDX_TO_OFF(mreq->pindex) + size > object->un_pager.vnp.vnp_size) 5591 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mreq->pindex); 5592 5593 VM_OBJECT_UNLOCK(object); 5594 va = zfs_map_page(mreq, &sf); 5595 error = dmu_read(os, zp->z_id, IDX_TO_OFF(mreq->pindex), 5596 size, va, DMU_READ_PREFETCH); 5597 if (size != PAGE_SIZE) 5598 bzero(va + size, PAGE_SIZE - size); 5599 zfs_unmap_page(sf); 5600 VM_OBJECT_LOCK(object); 5601 5602 if (!error) 5603 mreq->valid = VM_PAGE_BITS_ALL; 5604 KASSERT(mreq->dirty == 0, ("zfs_getpages: page %p is dirty", mreq)); 5605 5606 VM_OBJECT_UNLOCK(object); 5607 5608 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 5609 ZFS_EXIT(zfsvfs); 5610 return (error ? VM_PAGER_ERROR : VM_PAGER_OK); 5611} 5612 5613static int 5614zfs_freebsd_getpages(ap) 5615 struct vop_getpages_args /* { 5616 struct vnode *a_vp; 5617 vm_page_t *a_m; 5618 int a_count; 5619 int a_reqpage; 5620 vm_ooffset_t a_offset; 5621 } */ *ap; 5622{ 5623 5624 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage)); 5625} 5626 5627static int 5628zfs_freebsd_open(ap) 5629 struct vop_open_args /* { 5630 struct vnode *a_vp; 5631 int a_mode; 5632 struct ucred *a_cred; 5633 struct thread *a_td; 5634 } */ *ap; 5635{ 5636 vnode_t *vp = ap->a_vp; 5637 znode_t *zp = VTOZ(vp); 5638 int error; 5639 5640 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL); 5641 if (error == 0) 5642 vnode_create_vobject(vp, zp->z_size, ap->a_td); 5643 return (error); 5644} 5645 5646static int 5647zfs_freebsd_close(ap) 5648 struct vop_close_args /* { 5649 struct vnode *a_vp; 5650 int a_fflag; 5651 struct ucred *a_cred; 5652 struct thread *a_td; 5653 } */ *ap; 5654{ 5655 5656 return (zfs_close(ap->a_vp, ap->a_fflag, 0, 0, ap->a_cred, NULL)); 5657} 5658 5659static int 5660zfs_freebsd_ioctl(ap) 5661 struct vop_ioctl_args /* { 5662 struct vnode *a_vp; 5663 u_long a_command; 5664 caddr_t a_data; 5665 int a_fflag; 5666 struct ucred *cred; 5667 struct thread *td; 5668 } */ *ap; 5669{ 5670 5671 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 5672 ap->a_fflag, ap->a_cred, NULL, NULL)); 5673} 5674 5675static int 5676zfs_freebsd_read(ap) 5677 struct vop_read_args /* { 5678 struct vnode *a_vp; 5679 struct uio *a_uio; 5680 int a_ioflag; 5681 struct ucred *a_cred; 5682 } */ *ap; 5683{ 5684 5685 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 5686 ap->a_cred, NULL)); 5687} 5688 5689static int 5690zfs_freebsd_write(ap) 5691 struct vop_write_args /* { 5692 struct vnode *a_vp; 5693 struct uio *a_uio; 5694 int a_ioflag; 5695 struct ucred *a_cred; 5696 } */ *ap; 5697{ 5698 5699 if (vn_rlimit_fsize(ap->a_vp, ap->a_uio, ap->a_uio->uio_td)) 5700 return (EFBIG); 5701 5702 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 5703 ap->a_cred, NULL)); 5704} 5705 5706static int 5707zfs_freebsd_access(ap) 5708 struct vop_access_args /* { 5709 struct vnode *a_vp; 5710 accmode_t a_accmode; 5711 struct ucred *a_cred; 5712 struct thread *a_td; 5713 } */ *ap; 5714{ 5715 vnode_t *vp = ap->a_vp; 5716 znode_t *zp = VTOZ(vp); 5717 accmode_t accmode; 5718 int error = 0; 5719 5720 /* 5721 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 5722 */ 5723 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 5724 if (accmode != 0) 5725 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL); 5726 5727 /* 5728 * VADMIN has to be handled by vaccess(). 5729 */ 5730 if (error == 0) { 5731 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 5732 if (accmode != 0) { 5733 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 5734 zp->z_gid, accmode, ap->a_cred, NULL); 5735 } 5736 } 5737 5738 /* 5739 * For VEXEC, ensure that at least one execute bit is set for 5740 * non-directories. 5741 */ 5742 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 5743 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 5744 error = EACCES; 5745 } 5746 5747 return (error); 5748} 5749 5750static int 5751zfs_freebsd_lookup(ap) 5752 struct vop_lookup_args /* { 5753 struct vnode *a_dvp; 5754 struct vnode **a_vpp; 5755 struct componentname *a_cnp; 5756 } */ *ap; 5757{ 5758 struct componentname *cnp = ap->a_cnp; 5759 char nm[NAME_MAX + 1]; 5760 5761 ASSERT(cnp->cn_namelen < sizeof(nm)); 5762 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 5763 5764 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 5765 cnp->cn_cred, cnp->cn_thread, 0)); 5766} 5767 5768static int 5769zfs_freebsd_create(ap) 5770 struct vop_create_args /* { 5771 struct vnode *a_dvp; 5772 struct vnode **a_vpp; 5773 struct componentname *a_cnp; 5774 struct vattr *a_vap; 5775 } */ *ap; 5776{ 5777 struct componentname *cnp = ap->a_cnp; 5778 vattr_t *vap = ap->a_vap; 5779 int mode; 5780 5781 ASSERT(cnp->cn_flags & SAVENAME); 5782 5783 vattr_init_mask(vap); 5784 mode = vap->va_mode & ALLPERMS; 5785 5786 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 5787 ap->a_vpp, cnp->cn_cred, cnp->cn_thread)); 5788} 5789 5790static int 5791zfs_freebsd_remove(ap) 5792 struct vop_remove_args /* { 5793 struct vnode *a_dvp; 5794 struct vnode *a_vp; 5795 struct componentname *a_cnp; 5796 } */ *ap; 5797{ 5798 5799 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 5800 5801 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 5802 ap->a_cnp->cn_cred, NULL, 0)); 5803} 5804 5805static int 5806zfs_freebsd_mkdir(ap) 5807 struct vop_mkdir_args /* { 5808 struct vnode *a_dvp; 5809 struct vnode **a_vpp; 5810 struct componentname *a_cnp; 5811 struct vattr *a_vap; 5812 } */ *ap; 5813{ 5814 vattr_t *vap = ap->a_vap; 5815 5816 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 5817 5818 vattr_init_mask(vap); 5819 5820 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 5821 ap->a_cnp->cn_cred, NULL, 0, NULL)); 5822} 5823 5824static int 5825zfs_freebsd_rmdir(ap) 5826 struct vop_rmdir_args /* { 5827 struct vnode *a_dvp; 5828 struct vnode *a_vp; 5829 struct componentname *a_cnp; 5830 } */ *ap; 5831{ 5832 struct componentname *cnp = ap->a_cnp; 5833 5834 ASSERT(cnp->cn_flags & SAVENAME); 5835 5836 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0)); 5837} 5838 5839static int 5840zfs_freebsd_readdir(ap) 5841 struct vop_readdir_args /* { 5842 struct vnode *a_vp; 5843 struct uio *a_uio; 5844 struct ucred *a_cred; 5845 int *a_eofflag; 5846 int *a_ncookies; 5847 u_long **a_cookies; 5848 } */ *ap; 5849{ 5850 5851 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 5852 ap->a_ncookies, ap->a_cookies)); 5853} 5854 5855static int 5856zfs_freebsd_fsync(ap) 5857 struct vop_fsync_args /* { 5858 struct vnode *a_vp; 5859 int a_waitfor; 5860 struct thread *a_td; 5861 } */ *ap; 5862{ 5863 5864 vop_stdfsync(ap); 5865 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL)); 5866} 5867 5868static int 5869zfs_freebsd_getattr(ap) 5870 struct vop_getattr_args /* { 5871 struct vnode *a_vp; 5872 struct vattr *a_vap; 5873 struct ucred *a_cred; 5874 } */ *ap; 5875{ 5876 vattr_t *vap = ap->a_vap; 5877 xvattr_t xvap; 5878 u_long fflags = 0; 5879 int error; 5880 5881 xva_init(&xvap); 5882 xvap.xva_vattr = *vap; 5883 xvap.xva_vattr.va_mask |= AT_XVATTR; 5884 5885 /* Convert chflags into ZFS-type flags. */ 5886 /* XXX: what about SF_SETTABLE?. */ 5887 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 5888 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 5889 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 5890 XVA_SET_REQ(&xvap, XAT_NODUMP); 5891 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL); 5892 if (error != 0) 5893 return (error); 5894 5895 /* Convert ZFS xattr into chflags. */ 5896#define FLAG_CHECK(fflag, xflag, xfield) do { \ 5897 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 5898 fflags |= (fflag); \ 5899} while (0) 5900 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 5901 xvap.xva_xoptattrs.xoa_immutable); 5902 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 5903 xvap.xva_xoptattrs.xoa_appendonly); 5904 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 5905 xvap.xva_xoptattrs.xoa_nounlink); 5906 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 5907 xvap.xva_xoptattrs.xoa_nodump); 5908#undef FLAG_CHECK 5909 *vap = xvap.xva_vattr; 5910 vap->va_flags = fflags; 5911 return (0); 5912} 5913 5914static int 5915zfs_freebsd_setattr(ap) 5916 struct vop_setattr_args /* { 5917 struct vnode *a_vp; 5918 struct vattr *a_vap; 5919 struct ucred *a_cred; 5920 } */ *ap; 5921{ 5922 vnode_t *vp = ap->a_vp; 5923 vattr_t *vap = ap->a_vap; 5924 cred_t *cred = ap->a_cred; 5925 xvattr_t xvap; 5926 u_long fflags; 5927 uint64_t zflags; 5928 5929 vattr_init_mask(vap); 5930 vap->va_mask &= ~AT_NOSET; 5931 5932 xva_init(&xvap); 5933 xvap.xva_vattr = *vap; 5934 5935 zflags = VTOZ(vp)->z_pflags; 5936 5937 if (vap->va_flags != VNOVAL) { 5938 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 5939 int error; 5940 5941 if (zfsvfs->z_use_fuids == B_FALSE) 5942 return (EOPNOTSUPP); 5943 5944 fflags = vap->va_flags; 5945 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_NODUMP)) != 0) 5946 return (EOPNOTSUPP); 5947 /* 5948 * Unprivileged processes are not permitted to unset system 5949 * flags, or modify flags if any system flags are set. 5950 * Privileged non-jail processes may not modify system flags 5951 * if securelevel > 0 and any existing system flags are set. 5952 * Privileged jail processes behave like privileged non-jail 5953 * processes if the security.jail.chflags_allowed sysctl is 5954 * is non-zero; otherwise, they behave like unprivileged 5955 * processes. 5956 */ 5957 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 5958 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) { 5959 if (zflags & 5960 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 5961 error = securelevel_gt(cred, 0); 5962 if (error != 0) 5963 return (error); 5964 } 5965 } else { 5966 /* 5967 * Callers may only modify the file flags on objects they 5968 * have VADMIN rights for. 5969 */ 5970 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0) 5971 return (error); 5972 if (zflags & 5973 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 5974 return (EPERM); 5975 } 5976 if (fflags & 5977 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 5978 return (EPERM); 5979 } 5980 } 5981 5982#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 5983 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 5984 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 5985 XVA_SET_REQ(&xvap, (xflag)); \ 5986 (xfield) = ((fflags & (fflag)) != 0); \ 5987 } \ 5988} while (0) 5989 /* Convert chflags into ZFS-type flags. */ 5990 /* XXX: what about SF_SETTABLE?. */ 5991 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 5992 xvap.xva_xoptattrs.xoa_immutable); 5993 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 5994 xvap.xva_xoptattrs.xoa_appendonly); 5995 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 5996 xvap.xva_xoptattrs.xoa_nounlink); 5997 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 5998 xvap.xva_xoptattrs.xoa_nodump); 5999#undef FLAG_CHANGE 6000 } 6001 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL)); 6002} 6003 6004static int 6005zfs_freebsd_rename(ap) 6006 struct vop_rename_args /* { 6007 struct vnode *a_fdvp; 6008 struct vnode *a_fvp; 6009 struct componentname *a_fcnp; 6010 struct vnode *a_tdvp; 6011 struct vnode *a_tvp; 6012 struct componentname *a_tcnp; 6013 } */ *ap; 6014{ 6015 vnode_t *fdvp = ap->a_fdvp; 6016 vnode_t *fvp = ap->a_fvp; 6017 vnode_t *tdvp = ap->a_tdvp; 6018 vnode_t *tvp = ap->a_tvp; 6019 int error; 6020 6021 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 6022 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 6023 6024 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 6025 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0); 6026 6027 if (tdvp == tvp) 6028 VN_RELE(tdvp); 6029 else 6030 VN_URELE(tdvp); 6031 if (tvp) 6032 VN_URELE(tvp); 6033 VN_RELE(fdvp); 6034 VN_RELE(fvp); 6035 6036 return (error); 6037} 6038 6039static int 6040zfs_freebsd_symlink(ap) 6041 struct vop_symlink_args /* { 6042 struct vnode *a_dvp; 6043 struct vnode **a_vpp; 6044 struct componentname *a_cnp; 6045 struct vattr *a_vap; 6046 char *a_target; 6047 } */ *ap; 6048{ 6049 struct componentname *cnp = ap->a_cnp; 6050 vattr_t *vap = ap->a_vap; 6051 6052 ASSERT(cnp->cn_flags & SAVENAME); 6053 6054 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 6055 vattr_init_mask(vap); 6056 6057 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 6058 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 6059} 6060 6061static int 6062zfs_freebsd_readlink(ap) 6063 struct vop_readlink_args /* { 6064 struct vnode *a_vp; 6065 struct uio *a_uio; 6066 struct ucred *a_cred; 6067 } */ *ap; 6068{ 6069 6070 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL)); 6071} 6072 6073static int 6074zfs_freebsd_link(ap) 6075 struct vop_link_args /* { 6076 struct vnode *a_tdvp; 6077 struct vnode *a_vp; 6078 struct componentname *a_cnp; 6079 } */ *ap; 6080{ 6081 struct componentname *cnp = ap->a_cnp; 6082 6083 ASSERT(cnp->cn_flags & SAVENAME); 6084 6085 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0)); 6086} 6087 6088static int 6089zfs_freebsd_inactive(ap) 6090 struct vop_inactive_args /* { 6091 struct vnode *a_vp; 6092 struct thread *a_td; 6093 } */ *ap; 6094{ 6095 vnode_t *vp = ap->a_vp; 6096 6097 zfs_inactive(vp, ap->a_td->td_ucred, NULL); 6098 return (0); 6099} 6100 6101static void 6102zfs_reclaim_complete(void *arg, int pending) 6103{ 6104 znode_t *zp = arg; 6105 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6106 6107 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 6108 if (zp->z_sa_hdl != NULL) { 6109 ZFS_OBJ_HOLD_ENTER(zfsvfs, zp->z_id); 6110 zfs_znode_dmu_fini(zp); 6111 ZFS_OBJ_HOLD_EXIT(zfsvfs, zp->z_id); 6112 } 6113 zfs_znode_free(zp); 6114 rw_exit(&zfsvfs->z_teardown_inactive_lock); 6115 /* 6116 * If the file system is being unmounted, there is a process waiting 6117 * for us, wake it up. 6118 */ 6119 if (zfsvfs->z_unmounted) 6120 wakeup_one(zfsvfs); 6121} 6122 6123static int 6124zfs_freebsd_reclaim(ap) 6125 struct vop_reclaim_args /* { 6126 struct vnode *a_vp; 6127 struct thread *a_td; 6128 } */ *ap; 6129{ 6130 vnode_t *vp = ap->a_vp; 6131 znode_t *zp = VTOZ(vp); 6132 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6133 boolean_t rlocked; 6134 6135 rlocked = rw_tryenter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 6136 6137 ASSERT(zp != NULL); 6138 6139 /* 6140 * Destroy the vm object and flush associated pages. 6141 */ 6142 vnode_destroy_vobject(vp); 6143 6144 mutex_enter(&zp->z_lock); 6145 zp->z_vnode = NULL; 6146 mutex_exit(&zp->z_lock); 6147 6148 if (zp->z_unlinked) { 6149 ; /* Do nothing. */ 6150 } else if (!rlocked) { 6151 TASK_INIT(&zp->z_task, 0, zfs_reclaim_complete, zp); 6152 taskqueue_enqueue(taskqueue_thread, &zp->z_task); 6153 } else if (zp->z_sa_hdl == NULL) { 6154 zfs_znode_free(zp); 6155 } else /* if (!zp->z_unlinked && zp->z_dbuf != NULL) */ { 6156 int locked; 6157 6158 locked = MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)) ? 2 : 6159 ZFS_OBJ_HOLD_TRYENTER(zfsvfs, zp->z_id); 6160 if (locked == 0) { 6161 /* 6162 * Lock can't be obtained due to deadlock possibility, 6163 * so defer znode destruction. 6164 */ 6165 TASK_INIT(&zp->z_task, 0, zfs_reclaim_complete, zp); 6166 taskqueue_enqueue(taskqueue_thread, &zp->z_task); 6167 } else { 6168 zfs_znode_dmu_fini(zp); 6169 if (locked == 1) 6170 ZFS_OBJ_HOLD_EXIT(zfsvfs, zp->z_id); 6171 zfs_znode_free(zp); 6172 } 6173 } 6174 VI_LOCK(vp); 6175 vp->v_data = NULL; 6176 ASSERT(vp->v_holdcnt >= 1); 6177 VI_UNLOCK(vp); 6178 if (rlocked) 6179 rw_exit(&zfsvfs->z_teardown_inactive_lock); 6180 return (0); 6181} 6182 6183static int 6184zfs_freebsd_fid(ap) 6185 struct vop_fid_args /* { 6186 struct vnode *a_vp; 6187 struct fid *a_fid; 6188 } */ *ap; 6189{ 6190 6191 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 6192} 6193 6194static int 6195zfs_freebsd_pathconf(ap) 6196 struct vop_pathconf_args /* { 6197 struct vnode *a_vp; 6198 int a_name; 6199 register_t *a_retval; 6200 } */ *ap; 6201{ 6202 ulong_t val; 6203 int error; 6204 6205 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); 6206 if (error == 0) 6207 *ap->a_retval = val; 6208 else if (error == EOPNOTSUPP) 6209 error = vop_stdpathconf(ap); 6210 return (error); 6211} 6212 6213static int 6214zfs_freebsd_fifo_pathconf(ap) 6215 struct vop_pathconf_args /* { 6216 struct vnode *a_vp; 6217 int a_name; 6218 register_t *a_retval; 6219 } */ *ap; 6220{ 6221 6222 switch (ap->a_name) { 6223 case _PC_ACL_EXTENDED: 6224 case _PC_ACL_NFS4: 6225 case _PC_ACL_PATH_MAX: 6226 case _PC_MAC_PRESENT: 6227 return (zfs_freebsd_pathconf(ap)); 6228 default: 6229 return (fifo_specops.vop_pathconf(ap)); 6230 } 6231} 6232 6233/* 6234 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 6235 * extended attribute name: 6236 * 6237 * NAMESPACE PREFIX 6238 * system freebsd:system: 6239 * user (none, can be used to access ZFS fsattr(5) attributes 6240 * created on Solaris) 6241 */ 6242static int 6243zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 6244 size_t size) 6245{ 6246 const char *namespace, *prefix, *suffix; 6247 6248 /* We don't allow '/' character in attribute name. */ 6249 if (strchr(name, '/') != NULL) 6250 return (EINVAL); 6251 /* We don't allow attribute names that start with "freebsd:" string. */ 6252 if (strncmp(name, "freebsd:", 8) == 0) 6253 return (EINVAL); 6254 6255 bzero(attrname, size); 6256 6257 switch (attrnamespace) { 6258 case EXTATTR_NAMESPACE_USER: 6259#if 0 6260 prefix = "freebsd:"; 6261 namespace = EXTATTR_NAMESPACE_USER_STRING; 6262 suffix = ":"; 6263#else 6264 /* 6265 * This is the default namespace by which we can access all 6266 * attributes created on Solaris. 6267 */ 6268 prefix = namespace = suffix = ""; 6269#endif 6270 break; 6271 case EXTATTR_NAMESPACE_SYSTEM: 6272 prefix = "freebsd:"; 6273 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 6274 suffix = ":"; 6275 break; 6276 case EXTATTR_NAMESPACE_EMPTY: 6277 default: 6278 return (EINVAL); 6279 } 6280 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 6281 name) >= size) { 6282 return (ENAMETOOLONG); 6283 } 6284 return (0); 6285} 6286 6287/* 6288 * Vnode operating to retrieve a named extended attribute. 6289 */ 6290static int 6291zfs_getextattr(struct vop_getextattr_args *ap) 6292/* 6293vop_getextattr { 6294 IN struct vnode *a_vp; 6295 IN int a_attrnamespace; 6296 IN const char *a_name; 6297 INOUT struct uio *a_uio; 6298 OUT size_t *a_size; 6299 IN struct ucred *a_cred; 6300 IN struct thread *a_td; 6301}; 6302*/ 6303{ 6304 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6305 struct thread *td = ap->a_td; 6306 struct nameidata nd; 6307 char attrname[255]; 6308 struct vattr va; 6309 vnode_t *xvp = NULL, *vp; 6310 int error, flags; 6311 6312 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6313 ap->a_cred, ap->a_td, VREAD); 6314 if (error != 0) 6315 return (error); 6316 6317 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6318 sizeof(attrname)); 6319 if (error != 0) 6320 return (error); 6321 6322 ZFS_ENTER(zfsvfs); 6323 6324 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6325 LOOKUP_XATTR); 6326 if (error != 0) { 6327 ZFS_EXIT(zfsvfs); 6328 return (error); 6329 } 6330 6331 flags = FREAD; 6332 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname, 6333 xvp, td); 6334 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL); 6335 vp = nd.ni_vp; 6336 NDFREE(&nd, NDF_ONLY_PNBUF); 6337 if (error != 0) { 6338 ZFS_EXIT(zfsvfs); 6339 if (error == ENOENT) 6340 error = ENOATTR; 6341 return (error); 6342 } 6343 6344 if (ap->a_size != NULL) { 6345 error = VOP_GETATTR(vp, &va, ap->a_cred); 6346 if (error == 0) 6347 *ap->a_size = (size_t)va.va_size; 6348 } else if (ap->a_uio != NULL) 6349 error = VOP_READ(vp, ap->a_uio, IO_UNIT, ap->a_cred); 6350 6351 VOP_UNLOCK(vp, 0); 6352 vn_close(vp, flags, ap->a_cred, td); 6353 ZFS_EXIT(zfsvfs); 6354 6355 return (error); 6356} 6357 6358/* 6359 * Vnode operation to remove a named attribute. 6360 */ 6361int 6362zfs_deleteextattr(struct vop_deleteextattr_args *ap) 6363/* 6364vop_deleteextattr { 6365 IN struct vnode *a_vp; 6366 IN int a_attrnamespace; 6367 IN const char *a_name; 6368 IN struct ucred *a_cred; 6369 IN struct thread *a_td; 6370}; 6371*/ 6372{ 6373 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6374 struct thread *td = ap->a_td; 6375 struct nameidata nd; 6376 char attrname[255]; 6377 struct vattr va; 6378 vnode_t *xvp = NULL, *vp; 6379 int error, flags; 6380 6381 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6382 ap->a_cred, ap->a_td, VWRITE); 6383 if (error != 0) 6384 return (error); 6385 6386 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6387 sizeof(attrname)); 6388 if (error != 0) 6389 return (error); 6390 6391 ZFS_ENTER(zfsvfs); 6392 6393 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6394 LOOKUP_XATTR); 6395 if (error != 0) { 6396 ZFS_EXIT(zfsvfs); 6397 return (error); 6398 } 6399 6400 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF | MPSAFE, 6401 UIO_SYSSPACE, attrname, xvp, td); 6402 error = namei(&nd); 6403 vp = nd.ni_vp; 6404 NDFREE(&nd, NDF_ONLY_PNBUF); 6405 if (error != 0) { 6406 ZFS_EXIT(zfsvfs); 6407 if (error == ENOENT) 6408 error = ENOATTR; 6409 return (error); 6410 } 6411 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 6412 6413 vput(nd.ni_dvp); 6414 if (vp == nd.ni_dvp) 6415 vrele(vp); 6416 else 6417 vput(vp); 6418 ZFS_EXIT(zfsvfs); 6419 6420 return (error); 6421} 6422 6423/* 6424 * Vnode operation to set a named attribute. 6425 */ 6426static int 6427zfs_setextattr(struct vop_setextattr_args *ap) 6428/* 6429vop_setextattr { 6430 IN struct vnode *a_vp; 6431 IN int a_attrnamespace; 6432 IN const char *a_name; 6433 INOUT struct uio *a_uio; 6434 IN struct ucred *a_cred; 6435 IN struct thread *a_td; 6436}; 6437*/ 6438{ 6439 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6440 struct thread *td = ap->a_td; 6441 struct nameidata nd; 6442 char attrname[255]; 6443 struct vattr va; 6444 vnode_t *xvp = NULL, *vp; 6445 int error, flags; 6446 6447 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6448 ap->a_cred, ap->a_td, VWRITE); 6449 if (error != 0) 6450 return (error); 6451 6452 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6453 sizeof(attrname)); 6454 if (error != 0) 6455 return (error); 6456 6457 ZFS_ENTER(zfsvfs); 6458 6459 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6460 LOOKUP_XATTR | CREATE_XATTR_DIR); 6461 if (error != 0) { 6462 ZFS_EXIT(zfsvfs); 6463 return (error); 6464 } 6465 6466 flags = FFLAGS(O_WRONLY | O_CREAT); 6467 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname, 6468 xvp, td); 6469 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL); 6470 vp = nd.ni_vp; 6471 NDFREE(&nd, NDF_ONLY_PNBUF); 6472 if (error != 0) { 6473 ZFS_EXIT(zfsvfs); 6474 return (error); 6475 } 6476 6477 VATTR_NULL(&va); 6478 va.va_size = 0; 6479 error = VOP_SETATTR(vp, &va, ap->a_cred); 6480 if (error == 0) 6481 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred); 6482 6483 VOP_UNLOCK(vp, 0); 6484 vn_close(vp, flags, ap->a_cred, td); 6485 ZFS_EXIT(zfsvfs); 6486 6487 return (error); 6488} 6489 6490/* 6491 * Vnode operation to retrieve extended attributes on a vnode. 6492 */ 6493static int 6494zfs_listextattr(struct vop_listextattr_args *ap) 6495/* 6496vop_listextattr { 6497 IN struct vnode *a_vp; 6498 IN int a_attrnamespace; 6499 INOUT struct uio *a_uio; 6500 OUT size_t *a_size; 6501 IN struct ucred *a_cred; 6502 IN struct thread *a_td; 6503}; 6504*/ 6505{ 6506 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6507 struct thread *td = ap->a_td; 6508 struct nameidata nd; 6509 char attrprefix[16]; 6510 u_char dirbuf[sizeof(struct dirent)]; 6511 struct dirent *dp; 6512 struct iovec aiov; 6513 struct uio auio, *uio = ap->a_uio; 6514 size_t *sizep = ap->a_size; 6515 size_t plen; 6516 vnode_t *xvp = NULL, *vp; 6517 int done, error, eof, pos; 6518 6519 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6520 ap->a_cred, ap->a_td, VREAD); 6521 if (error != 0) 6522 return (error); 6523 6524 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 6525 sizeof(attrprefix)); 6526 if (error != 0) 6527 return (error); 6528 plen = strlen(attrprefix); 6529 6530 ZFS_ENTER(zfsvfs); 6531 6532 if (sizep != NULL) 6533 *sizep = 0; 6534 6535 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6536 LOOKUP_XATTR); 6537 if (error != 0) { 6538 ZFS_EXIT(zfsvfs); 6539 /* 6540 * ENOATTR means that the EA directory does not yet exist, 6541 * i.e. there are no extended attributes there. 6542 */ 6543 if (error == ENOATTR) 6544 error = 0; 6545 return (error); 6546 } 6547 6548 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED | MPSAFE, 6549 UIO_SYSSPACE, ".", xvp, td); 6550 error = namei(&nd); 6551 vp = nd.ni_vp; 6552 NDFREE(&nd, NDF_ONLY_PNBUF); 6553 if (error != 0) { 6554 ZFS_EXIT(zfsvfs); 6555 return (error); 6556 } 6557 6558 auio.uio_iov = &aiov; 6559 auio.uio_iovcnt = 1; 6560 auio.uio_segflg = UIO_SYSSPACE; 6561 auio.uio_td = td; 6562 auio.uio_rw = UIO_READ; 6563 auio.uio_offset = 0; 6564 6565 do { 6566 u_char nlen; 6567 6568 aiov.iov_base = (void *)dirbuf; 6569 aiov.iov_len = sizeof(dirbuf); 6570 auio.uio_resid = sizeof(dirbuf); 6571 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 6572 done = sizeof(dirbuf) - auio.uio_resid; 6573 if (error != 0) 6574 break; 6575 for (pos = 0; pos < done;) { 6576 dp = (struct dirent *)(dirbuf + pos); 6577 pos += dp->d_reclen; 6578 /* 6579 * XXX: Temporarily we also accept DT_UNKNOWN, as this 6580 * is what we get when attribute was created on Solaris. 6581 */ 6582 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 6583 continue; 6584 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0) 6585 continue; 6586 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 6587 continue; 6588 nlen = dp->d_namlen - plen; 6589 if (sizep != NULL) 6590 *sizep += 1 + nlen; 6591 else if (uio != NULL) { 6592 /* 6593 * Format of extattr name entry is one byte for 6594 * length and the rest for name. 6595 */ 6596 error = uiomove(&nlen, 1, uio->uio_rw, uio); 6597 if (error == 0) { 6598 error = uiomove(dp->d_name + plen, nlen, 6599 uio->uio_rw, uio); 6600 } 6601 if (error != 0) 6602 break; 6603 } 6604 } 6605 } while (!eof && error == 0); 6606 6607 vput(vp); 6608 ZFS_EXIT(zfsvfs); 6609 6610 return (error); 6611} 6612 6613int 6614zfs_freebsd_getacl(ap) 6615 struct vop_getacl_args /* { 6616 struct vnode *vp; 6617 acl_type_t type; 6618 struct acl *aclp; 6619 struct ucred *cred; 6620 struct thread *td; 6621 } */ *ap; 6622{ 6623 int error; 6624 vsecattr_t vsecattr; 6625 6626 if (ap->a_type != ACL_TYPE_NFS4) 6627 return (EINVAL); 6628 6629 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 6630 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL)) 6631 return (error); 6632 6633 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt); 6634 if (vsecattr.vsa_aclentp != NULL) 6635 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 6636 6637 return (error); 6638} 6639 6640int 6641zfs_freebsd_setacl(ap) 6642 struct vop_setacl_args /* { 6643 struct vnode *vp; 6644 acl_type_t type; 6645 struct acl *aclp; 6646 struct ucred *cred; 6647 struct thread *td; 6648 } */ *ap; 6649{ 6650 int error; 6651 vsecattr_t vsecattr; 6652 int aclbsize; /* size of acl list in bytes */ 6653 aclent_t *aaclp; 6654 6655 if (ap->a_type != ACL_TYPE_NFS4) 6656 return (EINVAL); 6657 6658 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 6659 return (EINVAL); 6660 6661 /* 6662 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 6663 * splitting every entry into two and appending "canonical six" 6664 * entries at the end. Don't allow for setting an ACL that would 6665 * cause chmod(2) to run out of ACL entries. 6666 */ 6667 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 6668 return (ENOSPC); 6669 6670 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 6671 if (error != 0) 6672 return (error); 6673 6674 vsecattr.vsa_mask = VSA_ACE; 6675 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t); 6676 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 6677 aaclp = vsecattr.vsa_aclentp; 6678 vsecattr.vsa_aclentsz = aclbsize; 6679 6680 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 6681 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL); 6682 kmem_free(aaclp, aclbsize); 6683 6684 return (error); 6685} 6686 6687int 6688zfs_freebsd_aclcheck(ap) 6689 struct vop_aclcheck_args /* { 6690 struct vnode *vp; 6691 acl_type_t type; 6692 struct acl *aclp; 6693 struct ucred *cred; 6694 struct thread *td; 6695 } */ *ap; 6696{ 6697 6698 return (EOPNOTSUPP); 6699} 6700 6701struct vop_vector zfs_vnodeops; 6702struct vop_vector zfs_fifoops; 6703struct vop_vector zfs_shareops; 6704 6705struct vop_vector zfs_vnodeops = { 6706 .vop_default = &default_vnodeops, 6707 .vop_inactive = zfs_freebsd_inactive, 6708 .vop_reclaim = zfs_freebsd_reclaim, 6709 .vop_access = zfs_freebsd_access, 6710#ifdef FREEBSD_NAMECACHE 6711 .vop_lookup = vfs_cache_lookup, 6712 .vop_cachedlookup = zfs_freebsd_lookup, 6713#else 6714 .vop_lookup = zfs_freebsd_lookup, 6715#endif 6716 .vop_getattr = zfs_freebsd_getattr, 6717 .vop_setattr = zfs_freebsd_setattr, 6718 .vop_create = zfs_freebsd_create, 6719 .vop_mknod = zfs_freebsd_create, 6720 .vop_mkdir = zfs_freebsd_mkdir, 6721 .vop_readdir = zfs_freebsd_readdir, 6722 .vop_fsync = zfs_freebsd_fsync, 6723 .vop_open = zfs_freebsd_open, 6724 .vop_close = zfs_freebsd_close, 6725 .vop_rmdir = zfs_freebsd_rmdir, 6726 .vop_ioctl = zfs_freebsd_ioctl, 6727 .vop_link = zfs_freebsd_link, 6728 .vop_symlink = zfs_freebsd_symlink, 6729 .vop_readlink = zfs_freebsd_readlink, 6730 .vop_read = zfs_freebsd_read, 6731 .vop_write = zfs_freebsd_write, 6732 .vop_remove = zfs_freebsd_remove, 6733 .vop_rename = zfs_freebsd_rename, 6734 .vop_pathconf = zfs_freebsd_pathconf, 6735 .vop_bmap = VOP_EOPNOTSUPP, 6736 .vop_fid = zfs_freebsd_fid, 6737 .vop_getextattr = zfs_getextattr, 6738 .vop_deleteextattr = zfs_deleteextattr, 6739 .vop_setextattr = zfs_setextattr, 6740 .vop_listextattr = zfs_listextattr, 6741 .vop_getacl = zfs_freebsd_getacl, 6742 .vop_setacl = zfs_freebsd_setacl, 6743 .vop_aclcheck = zfs_freebsd_aclcheck, 6744 .vop_getpages = zfs_freebsd_getpages, 6745}; 6746 6747struct vop_vector zfs_fifoops = { 6748 .vop_default = &fifo_specops, 6749 .vop_fsync = zfs_freebsd_fsync, 6750 .vop_access = zfs_freebsd_access, 6751 .vop_getattr = zfs_freebsd_getattr, 6752 .vop_inactive = zfs_freebsd_inactive, 6753 .vop_read = VOP_PANIC, 6754 .vop_reclaim = zfs_freebsd_reclaim, 6755 .vop_setattr = zfs_freebsd_setattr, 6756 .vop_write = VOP_PANIC, 6757 .vop_pathconf = zfs_freebsd_fifo_pathconf, 6758 .vop_fid = zfs_freebsd_fid, 6759 .vop_getacl = zfs_freebsd_getacl, 6760 .vop_setacl = zfs_freebsd_setacl, 6761 .vop_aclcheck = zfs_freebsd_aclcheck, 6762}; 6763 6764/* 6765 * special share hidden files vnode operations template 6766 */ 6767struct vop_vector zfs_shareops = { 6768 .vop_default = &default_vnodeops, 6769 .vop_access = zfs_freebsd_access, 6770 .vop_inactive = zfs_freebsd_inactive, 6771 .vop_reclaim = zfs_freebsd_reclaim, 6772 .vop_fid = zfs_freebsd_fid, 6773 .vop_pathconf = zfs_freebsd_pathconf, 6774}; 6775