zfs_vnops.c revision 224251
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 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 297 return (EFAULT); 298 299 zp = VTOZ(vp); 300 zfsvfs = zp->z_zfsvfs; 301 ZFS_ENTER(zfsvfs); 302 ZFS_VERIFY_ZP(zp); 303 304 /* offset parameter is in/out */ 305 error = zfs_holey(vp, com, &off); 306 ZFS_EXIT(zfsvfs); 307 if (error) 308 return (error); 309 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 310 return (EFAULT); 311 return (0); 312 } 313 return (ENOTTY); 314} 315 316static vm_page_t 317page_lookup(vnode_t *vp, int64_t start, int64_t off, int64_t nbytes) 318{ 319 vm_object_t obj; 320 vm_page_t pp; 321 322 obj = vp->v_object; 323 VM_OBJECT_LOCK_ASSERT(obj, MA_OWNED); 324 325 for (;;) { 326 if ((pp = vm_page_lookup(obj, OFF_TO_IDX(start))) != NULL && 327 vm_page_is_valid(pp, (vm_offset_t)off, nbytes)) { 328 if ((pp->oflags & VPO_BUSY) != 0) { 329 /* 330 * Reference the page before unlocking and 331 * sleeping so that the page daemon is less 332 * likely to reclaim it. 333 */ 334 vm_page_lock_queues(); 335 vm_page_flag_set(pp, PG_REFERENCED); 336 vm_page_sleep(pp, "zfsmwb"); 337 continue; 338 } 339 vm_page_busy(pp); 340 vm_page_undirty(pp); 341 } else { 342 if (__predict_false(obj->cache != NULL)) { 343 vm_page_cache_free(obj, OFF_TO_IDX(start), 344 OFF_TO_IDX(start) + 1); 345 } 346 pp = NULL; 347 } 348 break; 349 } 350 return (pp); 351} 352 353static void 354page_unlock(vm_page_t pp) 355{ 356 357 vm_page_wakeup(pp); 358} 359 360static caddr_t 361zfs_map_page(vm_page_t pp, struct sf_buf **sfp) 362{ 363 364 *sfp = sf_buf_alloc(pp, 0); 365 return ((caddr_t)sf_buf_kva(*sfp)); 366} 367 368static void 369zfs_unmap_page(struct sf_buf *sf) 370{ 371 372 sf_buf_free(sf); 373} 374 375/* 376 * When a file is memory mapped, we must keep the IO data synchronized 377 * between the DMU cache and the memory mapped pages. What this means: 378 * 379 * On Write: If we find a memory mapped page, we write to *both* 380 * the page and the dmu buffer. 381 */ 382static void 383update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid, 384 int segflg, dmu_tx_t *tx) 385{ 386 vm_object_t obj; 387 struct sf_buf *sf; 388 int off; 389 390 ASSERT(vp->v_mount != NULL); 391 obj = vp->v_object; 392 ASSERT(obj != NULL); 393 394 off = start & PAGEOFFSET; 395 VM_OBJECT_LOCK(obj); 396 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 397 vm_page_t pp; 398 int nbytes = MIN(PAGESIZE - off, len); 399 400 if ((pp = page_lookup(vp, start, off, nbytes)) != NULL) { 401 caddr_t va; 402 403 VM_OBJECT_UNLOCK(obj); 404 va = zfs_map_page(pp, &sf); 405 if (segflg == UIO_NOCOPY) { 406 (void) dmu_write(os, oid, start+off, nbytes, 407 va+off, tx); 408 } else { 409 (void) dmu_read(os, oid, start+off, nbytes, 410 va+off, DMU_READ_PREFETCH); 411 } 412 zfs_unmap_page(sf); 413 VM_OBJECT_LOCK(obj); 414 page_unlock(pp); 415 } 416 len -= nbytes; 417 off = 0; 418 } 419 VM_OBJECT_UNLOCK(obj); 420} 421 422/* 423 * Read with UIO_NOCOPY flag means that sendfile(2) requests 424 * ZFS to populate a range of page cache pages with data. 425 * 426 * NOTE: this function could be optimized to pre-allocate 427 * all pages in advance, drain VPO_BUSY on all of them, 428 * map them into contiguous KVA region and populate them 429 * in one single dmu_read() call. 430 */ 431static int 432mappedread_sf(vnode_t *vp, int nbytes, uio_t *uio) 433{ 434 znode_t *zp = VTOZ(vp); 435 objset_t *os = zp->z_zfsvfs->z_os; 436 struct sf_buf *sf; 437 vm_object_t obj; 438 vm_page_t pp; 439 int64_t start; 440 caddr_t va; 441 int len = nbytes; 442 int off; 443 int error = 0; 444 445 ASSERT(uio->uio_segflg == UIO_NOCOPY); 446 ASSERT(vp->v_mount != NULL); 447 obj = vp->v_object; 448 ASSERT(obj != NULL); 449 ASSERT((uio->uio_loffset & PAGEOFFSET) == 0); 450 451 VM_OBJECT_LOCK(obj); 452 for (start = uio->uio_loffset; len > 0; start += PAGESIZE) { 453 int bytes = MIN(PAGESIZE, len); 454 455 pp = vm_page_grab(obj, OFF_TO_IDX(start), VM_ALLOC_NOBUSY | 456 VM_ALLOC_NORMAL | VM_ALLOC_RETRY | VM_ALLOC_IGN_SBUSY); 457 if (pp->valid == 0) { 458 vm_page_io_start(pp); 459 VM_OBJECT_UNLOCK(obj); 460 va = zfs_map_page(pp, &sf); 461 error = dmu_read(os, zp->z_id, start, bytes, va, 462 DMU_READ_PREFETCH); 463 if (bytes != PAGESIZE && error == 0) 464 bzero(va + bytes, PAGESIZE - bytes); 465 zfs_unmap_page(sf); 466 VM_OBJECT_LOCK(obj); 467 vm_page_io_finish(pp); 468 vm_page_lock(pp); 469 if (error) { 470 vm_page_free(pp); 471 } else { 472 pp->valid = VM_PAGE_BITS_ALL; 473 vm_page_activate(pp); 474 } 475 vm_page_unlock(pp); 476 } 477 if (error) 478 break; 479 uio->uio_resid -= bytes; 480 uio->uio_offset += bytes; 481 len -= bytes; 482 } 483 VM_OBJECT_UNLOCK(obj); 484 return (error); 485} 486 487/* 488 * When a file is memory mapped, we must keep the IO data synchronized 489 * between the DMU cache and the memory mapped pages. What this means: 490 * 491 * On Read: We "read" preferentially from memory mapped pages, 492 * else we default from the dmu buffer. 493 * 494 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 495 * the file is memory mapped. 496 */ 497static int 498mappedread(vnode_t *vp, int nbytes, uio_t *uio) 499{ 500 znode_t *zp = VTOZ(vp); 501 objset_t *os = zp->z_zfsvfs->z_os; 502 vm_object_t obj; 503 int64_t start; 504 caddr_t va; 505 int len = nbytes; 506 int off; 507 int error = 0; 508 509 ASSERT(vp->v_mount != NULL); 510 obj = vp->v_object; 511 ASSERT(obj != NULL); 512 513 start = uio->uio_loffset; 514 off = start & PAGEOFFSET; 515 VM_OBJECT_LOCK(obj); 516 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 517 vm_page_t pp; 518 uint64_t bytes = MIN(PAGESIZE - off, len); 519 520 if (pp = page_lookup(vp, start, off, bytes)) { 521 struct sf_buf *sf; 522 caddr_t va; 523 524 VM_OBJECT_UNLOCK(obj); 525 va = zfs_map_page(pp, &sf); 526 error = uiomove(va + off, bytes, UIO_READ, uio); 527 zfs_unmap_page(sf); 528 VM_OBJECT_LOCK(obj); 529 page_unlock(pp); 530 } else { 531 VM_OBJECT_UNLOCK(obj); 532 error = dmu_read_uio(os, zp->z_id, uio, bytes); 533 VM_OBJECT_LOCK(obj); 534 } 535 len -= bytes; 536 off = 0; 537 if (error) 538 break; 539 } 540 VM_OBJECT_UNLOCK(obj); 541 return (error); 542} 543 544offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 545 546/* 547 * Read bytes from specified file into supplied buffer. 548 * 549 * IN: vp - vnode of file to be read from. 550 * uio - structure supplying read location, range info, 551 * and return buffer. 552 * ioflag - SYNC flags; used to provide FRSYNC semantics. 553 * cr - credentials of caller. 554 * ct - caller context 555 * 556 * OUT: uio - updated offset and range, buffer filled. 557 * 558 * RETURN: 0 if success 559 * error code if failure 560 * 561 * Side Effects: 562 * vp - atime updated if byte count > 0 563 */ 564/* ARGSUSED */ 565static int 566zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 567{ 568 znode_t *zp = VTOZ(vp); 569 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 570 objset_t *os; 571 ssize_t n, nbytes; 572 int error; 573 rl_t *rl; 574 xuio_t *xuio = NULL; 575 576 ZFS_ENTER(zfsvfs); 577 ZFS_VERIFY_ZP(zp); 578 os = zfsvfs->z_os; 579 580 if (zp->z_pflags & ZFS_AV_QUARANTINED) { 581 ZFS_EXIT(zfsvfs); 582 return (EACCES); 583 } 584 585 /* 586 * Validate file offset 587 */ 588 if (uio->uio_loffset < (offset_t)0) { 589 ZFS_EXIT(zfsvfs); 590 return (EINVAL); 591 } 592 593 /* 594 * Fasttrack empty reads 595 */ 596 if (uio->uio_resid == 0) { 597 ZFS_EXIT(zfsvfs); 598 return (0); 599 } 600 601 /* 602 * Check for mandatory locks 603 */ 604 if (MANDMODE(zp->z_mode)) { 605 if (error = chklock(vp, FREAD, 606 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 607 ZFS_EXIT(zfsvfs); 608 return (error); 609 } 610 } 611 612 /* 613 * If we're in FRSYNC mode, sync out this znode before reading it. 614 */ 615 if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 616 zil_commit(zfsvfs->z_log, zp->z_id); 617 618 /* 619 * Lock the range against changes. 620 */ 621 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 622 623 /* 624 * If we are reading past end-of-file we can skip 625 * to the end; but we might still need to set atime. 626 */ 627 if (uio->uio_loffset >= zp->z_size) { 628 error = 0; 629 goto out; 630 } 631 632 ASSERT(uio->uio_loffset < zp->z_size); 633 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset); 634 635#ifdef sun 636 if ((uio->uio_extflg == UIO_XUIO) && 637 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) { 638 int nblk; 639 int blksz = zp->z_blksz; 640 uint64_t offset = uio->uio_loffset; 641 642 xuio = (xuio_t *)uio; 643 if ((ISP2(blksz))) { 644 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset, 645 blksz)) / blksz; 646 } else { 647 ASSERT(offset + n <= blksz); 648 nblk = 1; 649 } 650 (void) dmu_xuio_init(xuio, nblk); 651 652 if (vn_has_cached_data(vp)) { 653 /* 654 * For simplicity, we always allocate a full buffer 655 * even if we only expect to read a portion of a block. 656 */ 657 while (--nblk >= 0) { 658 (void) dmu_xuio_add(xuio, 659 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 660 blksz), 0, blksz); 661 } 662 } 663 } 664#endif /* sun */ 665 666 while (n > 0) { 667 nbytes = MIN(n, zfs_read_chunk_size - 668 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 669 670#ifdef __FreeBSD__ 671 if (uio->uio_segflg == UIO_NOCOPY) 672 error = mappedread_sf(vp, nbytes, uio); 673 else 674#endif /* __FreeBSD__ */ 675 if (vn_has_cached_data(vp)) 676 error = mappedread(vp, nbytes, uio); 677 else 678 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 679 if (error) { 680 /* convert checksum errors into IO errors */ 681 if (error == ECKSUM) 682 error = EIO; 683 break; 684 } 685 686 n -= nbytes; 687 } 688out: 689 zfs_range_unlock(rl); 690 691 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 692 ZFS_EXIT(zfsvfs); 693 return (error); 694} 695 696/* 697 * Write the bytes to a file. 698 * 699 * IN: vp - vnode of file to be written to. 700 * uio - structure supplying write location, range info, 701 * and data buffer. 702 * ioflag - FAPPEND flag set if in append mode. 703 * cr - credentials of caller. 704 * ct - caller context (NFS/CIFS fem monitor only) 705 * 706 * OUT: uio - updated offset and range. 707 * 708 * RETURN: 0 if success 709 * error code if failure 710 * 711 * Timestamps: 712 * vp - ctime|mtime updated if byte count > 0 713 */ 714 715/* ARGSUSED */ 716static int 717zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 718{ 719 znode_t *zp = VTOZ(vp); 720 rlim64_t limit = MAXOFFSET_T; 721 ssize_t start_resid = uio->uio_resid; 722 ssize_t tx_bytes; 723 uint64_t end_size; 724 dmu_tx_t *tx; 725 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 726 zilog_t *zilog; 727 offset_t woff; 728 ssize_t n, nbytes; 729 rl_t *rl; 730 int max_blksz = zfsvfs->z_max_blksz; 731 int error; 732 arc_buf_t *abuf; 733 iovec_t *aiov; 734 xuio_t *xuio = NULL; 735 int i_iov = 0; 736 int iovcnt = uio->uio_iovcnt; 737 iovec_t *iovp = uio->uio_iov; 738 int write_eof; 739 int count = 0; 740 sa_bulk_attr_t bulk[4]; 741 uint64_t mtime[2], ctime[2]; 742 743 /* 744 * Fasttrack empty write 745 */ 746 n = start_resid; 747 if (n == 0) 748 return (0); 749 750 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 751 limit = MAXOFFSET_T; 752 753 ZFS_ENTER(zfsvfs); 754 ZFS_VERIFY_ZP(zp); 755 756 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 757 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 758 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 759 &zp->z_size, 8); 760 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 761 &zp->z_pflags, 8); 762 763 /* 764 * If immutable or not appending then return EPERM 765 */ 766 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 767 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 768 (uio->uio_loffset < zp->z_size))) { 769 ZFS_EXIT(zfsvfs); 770 return (EPERM); 771 } 772 773 zilog = zfsvfs->z_log; 774 775 /* 776 * Validate file offset 777 */ 778 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset; 779 if (woff < 0) { 780 ZFS_EXIT(zfsvfs); 781 return (EINVAL); 782 } 783 784 /* 785 * Check for mandatory locks before calling zfs_range_lock() 786 * in order to prevent a deadlock with locks set via fcntl(). 787 */ 788 if (MANDMODE((mode_t)zp->z_mode) && 789 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 790 ZFS_EXIT(zfsvfs); 791 return (error); 792 } 793 794#ifdef sun 795 /* 796 * Pre-fault the pages to ensure slow (eg NFS) pages 797 * don't hold up txg. 798 * Skip this if uio contains loaned arc_buf. 799 */ 800 if ((uio->uio_extflg == UIO_XUIO) && 801 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) 802 xuio = (xuio_t *)uio; 803 else 804 uio_prefaultpages(MIN(n, max_blksz), uio); 805#endif /* sun */ 806 807 /* 808 * If in append mode, set the io offset pointer to eof. 809 */ 810 if (ioflag & FAPPEND) { 811 /* 812 * Obtain an appending range lock to guarantee file append 813 * semantics. We reset the write offset once we have the lock. 814 */ 815 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 816 woff = rl->r_off; 817 if (rl->r_len == UINT64_MAX) { 818 /* 819 * We overlocked the file because this write will cause 820 * the file block size to increase. 821 * Note that zp_size cannot change with this lock held. 822 */ 823 woff = zp->z_size; 824 } 825 uio->uio_loffset = woff; 826 } else { 827 /* 828 * Note that if the file block size will change as a result of 829 * this write, then this range lock will lock the entire file 830 * so that we can re-write the block safely. 831 */ 832 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 833 } 834 835 if (woff >= limit) { 836 zfs_range_unlock(rl); 837 ZFS_EXIT(zfsvfs); 838 return (EFBIG); 839 } 840 841 if ((woff + n) > limit || woff > (limit - n)) 842 n = limit - woff; 843 844 /* Will this write extend the file length? */ 845 write_eof = (woff + n > zp->z_size); 846 847 end_size = MAX(zp->z_size, woff + n); 848 849 /* 850 * Write the file in reasonable size chunks. Each chunk is written 851 * in a separate transaction; this keeps the intent log records small 852 * and allows us to do more fine-grained space accounting. 853 */ 854 while (n > 0) { 855 abuf = NULL; 856 woff = uio->uio_loffset; 857again: 858 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 859 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 860 if (abuf != NULL) 861 dmu_return_arcbuf(abuf); 862 error = EDQUOT; 863 break; 864 } 865 866 if (xuio && abuf == NULL) { 867 ASSERT(i_iov < iovcnt); 868 aiov = &iovp[i_iov]; 869 abuf = dmu_xuio_arcbuf(xuio, i_iov); 870 dmu_xuio_clear(xuio, i_iov); 871 DTRACE_PROBE3(zfs_cp_write, int, i_iov, 872 iovec_t *, aiov, arc_buf_t *, abuf); 873 ASSERT((aiov->iov_base == abuf->b_data) || 874 ((char *)aiov->iov_base - (char *)abuf->b_data + 875 aiov->iov_len == arc_buf_size(abuf))); 876 i_iov++; 877 } else if (abuf == NULL && n >= max_blksz && 878 woff >= zp->z_size && 879 P2PHASE(woff, max_blksz) == 0 && 880 zp->z_blksz == max_blksz) { 881 /* 882 * This write covers a full block. "Borrow" a buffer 883 * from the dmu so that we can fill it before we enter 884 * a transaction. This avoids the possibility of 885 * holding up the transaction if the data copy hangs 886 * up on a pagefault (e.g., from an NFS server mapping). 887 */ 888 size_t cbytes; 889 890 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 891 max_blksz); 892 ASSERT(abuf != NULL); 893 ASSERT(arc_buf_size(abuf) == max_blksz); 894 if (error = uiocopy(abuf->b_data, max_blksz, 895 UIO_WRITE, uio, &cbytes)) { 896 dmu_return_arcbuf(abuf); 897 break; 898 } 899 ASSERT(cbytes == max_blksz); 900 } 901 902 /* 903 * Start a transaction. 904 */ 905 tx = dmu_tx_create(zfsvfs->z_os); 906 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 907 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 908 zfs_sa_upgrade_txholds(tx, zp); 909 error = dmu_tx_assign(tx, TXG_NOWAIT); 910 if (error) { 911 if (error == ERESTART) { 912 dmu_tx_wait(tx); 913 dmu_tx_abort(tx); 914 goto again; 915 } 916 dmu_tx_abort(tx); 917 if (abuf != NULL) 918 dmu_return_arcbuf(abuf); 919 break; 920 } 921 922 /* 923 * If zfs_range_lock() over-locked we grow the blocksize 924 * and then reduce the lock range. This will only happen 925 * on the first iteration since zfs_range_reduce() will 926 * shrink down r_len to the appropriate size. 927 */ 928 if (rl->r_len == UINT64_MAX) { 929 uint64_t new_blksz; 930 931 if (zp->z_blksz > max_blksz) { 932 ASSERT(!ISP2(zp->z_blksz)); 933 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 934 } else { 935 new_blksz = MIN(end_size, max_blksz); 936 } 937 zfs_grow_blocksize(zp, new_blksz, tx); 938 zfs_range_reduce(rl, woff, n); 939 } 940 941 /* 942 * XXX - should we really limit each write to z_max_blksz? 943 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 944 */ 945 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 946 947 if (woff + nbytes > zp->z_size) 948 vnode_pager_setsize(vp, woff + nbytes); 949 950 if (abuf == NULL) { 951 tx_bytes = uio->uio_resid; 952 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), 953 uio, nbytes, tx); 954 tx_bytes -= uio->uio_resid; 955 } else { 956 tx_bytes = nbytes; 957 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len); 958 /* 959 * If this is not a full block write, but we are 960 * extending the file past EOF and this data starts 961 * block-aligned, use assign_arcbuf(). Otherwise, 962 * write via dmu_write(). 963 */ 964 if (tx_bytes < max_blksz && (!write_eof || 965 aiov->iov_base != abuf->b_data)) { 966 ASSERT(xuio); 967 dmu_write(zfsvfs->z_os, zp->z_id, woff, 968 aiov->iov_len, aiov->iov_base, tx); 969 dmu_return_arcbuf(abuf); 970 xuio_stat_wbuf_copied(); 971 } else { 972 ASSERT(xuio || tx_bytes == max_blksz); 973 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), 974 woff, abuf, tx); 975 } 976 ASSERT(tx_bytes <= uio->uio_resid); 977 uioskip(uio, tx_bytes); 978 } 979 if (tx_bytes && vn_has_cached_data(vp)) { 980 update_pages(vp, woff, tx_bytes, zfsvfs->z_os, 981 zp->z_id, uio->uio_segflg, tx); 982 } 983 984 /* 985 * If we made no progress, we're done. If we made even 986 * partial progress, update the znode and ZIL accordingly. 987 */ 988 if (tx_bytes == 0) { 989 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 990 (void *)&zp->z_size, sizeof (uint64_t), tx); 991 dmu_tx_commit(tx); 992 ASSERT(error != 0); 993 break; 994 } 995 996 /* 997 * Clear Set-UID/Set-GID bits on successful write if not 998 * privileged and at least one of the excute bits is set. 999 * 1000 * It would be nice to to this after all writes have 1001 * been done, but that would still expose the ISUID/ISGID 1002 * to another app after the partial write is committed. 1003 * 1004 * Note: we don't call zfs_fuid_map_id() here because 1005 * user 0 is not an ephemeral uid. 1006 */ 1007 mutex_enter(&zp->z_acl_lock); 1008 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | 1009 (S_IXUSR >> 6))) != 0 && 1010 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && 1011 secpolicy_vnode_setid_retain(vp, cr, 1012 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) { 1013 uint64_t newmode; 1014 zp->z_mode &= ~(S_ISUID | S_ISGID); 1015 newmode = zp->z_mode; 1016 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), 1017 (void *)&newmode, sizeof (uint64_t), tx); 1018 } 1019 mutex_exit(&zp->z_acl_lock); 1020 1021 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 1022 B_TRUE); 1023 1024 /* 1025 * Update the file size (zp_size) if it has changed; 1026 * account for possible concurrent updates. 1027 */ 1028 while ((end_size = zp->z_size) < uio->uio_loffset) { 1029 (void) atomic_cas_64(&zp->z_size, end_size, 1030 uio->uio_loffset); 1031 ASSERT(error == 0); 1032 } 1033 /* 1034 * If we are replaying and eof is non zero then force 1035 * the file size to the specified eof. Note, there's no 1036 * concurrency during replay. 1037 */ 1038 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) 1039 zp->z_size = zfsvfs->z_replay_eof; 1040 1041 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1042 1043 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 1044 dmu_tx_commit(tx); 1045 1046 if (error != 0) 1047 break; 1048 ASSERT(tx_bytes == nbytes); 1049 n -= nbytes; 1050 1051#ifdef sun 1052 if (!xuio && n > 0) 1053 uio_prefaultpages(MIN(n, max_blksz), uio); 1054#endif /* sun */ 1055 } 1056 1057 zfs_range_unlock(rl); 1058 1059 /* 1060 * If we're in replay mode, or we made no progress, return error. 1061 * Otherwise, it's at least a partial write, so it's successful. 1062 */ 1063 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 1064 ZFS_EXIT(zfsvfs); 1065 return (error); 1066 } 1067 1068 if (ioflag & (FSYNC | FDSYNC) || 1069 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1070 zil_commit(zilog, zp->z_id); 1071 1072 ZFS_EXIT(zfsvfs); 1073 return (0); 1074} 1075 1076void 1077zfs_get_done(zgd_t *zgd, int error) 1078{ 1079 znode_t *zp = zgd->zgd_private; 1080 objset_t *os = zp->z_zfsvfs->z_os; 1081 int vfslocked; 1082 1083 if (zgd->zgd_db) 1084 dmu_buf_rele(zgd->zgd_db, zgd); 1085 1086 zfs_range_unlock(zgd->zgd_rl); 1087 1088 vfslocked = VFS_LOCK_GIANT(zp->z_zfsvfs->z_vfs); 1089 /* 1090 * Release the vnode asynchronously as we currently have the 1091 * txg stopped from syncing. 1092 */ 1093 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1094 1095 if (error == 0 && zgd->zgd_bp) 1096 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 1097 1098 kmem_free(zgd, sizeof (zgd_t)); 1099 VFS_UNLOCK_GIANT(vfslocked); 1100} 1101 1102#ifdef DEBUG 1103static int zil_fault_io = 0; 1104#endif 1105 1106/* 1107 * Get data to generate a TX_WRITE intent log record. 1108 */ 1109int 1110zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 1111{ 1112 zfsvfs_t *zfsvfs = arg; 1113 objset_t *os = zfsvfs->z_os; 1114 znode_t *zp; 1115 uint64_t object = lr->lr_foid; 1116 uint64_t offset = lr->lr_offset; 1117 uint64_t size = lr->lr_length; 1118 blkptr_t *bp = &lr->lr_blkptr; 1119 dmu_buf_t *db; 1120 zgd_t *zgd; 1121 int error = 0; 1122 1123 ASSERT(zio != NULL); 1124 ASSERT(size != 0); 1125 1126 /* 1127 * Nothing to do if the file has been removed 1128 */ 1129 if (zfs_zget(zfsvfs, object, &zp) != 0) 1130 return (ENOENT); 1131 if (zp->z_unlinked) { 1132 /* 1133 * Release the vnode asynchronously as we currently have the 1134 * txg stopped from syncing. 1135 */ 1136 VN_RELE_ASYNC(ZTOV(zp), 1137 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1138 return (ENOENT); 1139 } 1140 1141 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1142 zgd->zgd_zilog = zfsvfs->z_log; 1143 zgd->zgd_private = zp; 1144 1145 /* 1146 * Write records come in two flavors: immediate and indirect. 1147 * For small writes it's cheaper to store the data with the 1148 * log record (immediate); for large writes it's cheaper to 1149 * sync the data and get a pointer to it (indirect) so that 1150 * we don't have to write the data twice. 1151 */ 1152 if (buf != NULL) { /* immediate write */ 1153 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER); 1154 /* test for truncation needs to be done while range locked */ 1155 if (offset >= zp->z_size) { 1156 error = ENOENT; 1157 } else { 1158 error = dmu_read(os, object, offset, size, buf, 1159 DMU_READ_NO_PREFETCH); 1160 } 1161 ASSERT(error == 0 || error == ENOENT); 1162 } else { /* indirect write */ 1163 /* 1164 * Have to lock the whole block to ensure when it's 1165 * written out and it's checksum is being calculated 1166 * that no one can change the data. We need to re-check 1167 * blocksize after we get the lock in case it's changed! 1168 */ 1169 for (;;) { 1170 uint64_t blkoff; 1171 size = zp->z_blksz; 1172 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; 1173 offset -= blkoff; 1174 zgd->zgd_rl = zfs_range_lock(zp, offset, size, 1175 RL_READER); 1176 if (zp->z_blksz == size) 1177 break; 1178 offset += blkoff; 1179 zfs_range_unlock(zgd->zgd_rl); 1180 } 1181 /* test for truncation needs to be done while range locked */ 1182 if (lr->lr_offset >= zp->z_size) 1183 error = ENOENT; 1184#ifdef DEBUG 1185 if (zil_fault_io) { 1186 error = EIO; 1187 zil_fault_io = 0; 1188 } 1189#endif 1190 if (error == 0) 1191 error = dmu_buf_hold(os, object, offset, zgd, &db, 1192 DMU_READ_NO_PREFETCH); 1193 1194 if (error == 0) { 1195 zgd->zgd_db = db; 1196 zgd->zgd_bp = bp; 1197 1198 ASSERT(db->db_offset == offset); 1199 ASSERT(db->db_size == size); 1200 1201 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1202 zfs_get_done, zgd); 1203 ASSERT(error || lr->lr_length <= zp->z_blksz); 1204 1205 /* 1206 * On success, we need to wait for the write I/O 1207 * initiated by dmu_sync() to complete before we can 1208 * release this dbuf. We will finish everything up 1209 * in the zfs_get_done() callback. 1210 */ 1211 if (error == 0) 1212 return (0); 1213 1214 if (error == EALREADY) { 1215 lr->lr_common.lrc_txtype = TX_WRITE2; 1216 error = 0; 1217 } 1218 } 1219 } 1220 1221 zfs_get_done(zgd, error); 1222 1223 return (error); 1224} 1225 1226/*ARGSUSED*/ 1227static int 1228zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 1229 caller_context_t *ct) 1230{ 1231 znode_t *zp = VTOZ(vp); 1232 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1233 int error; 1234 1235 ZFS_ENTER(zfsvfs); 1236 ZFS_VERIFY_ZP(zp); 1237 1238 if (flag & V_ACE_MASK) 1239 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 1240 else 1241 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1242 1243 ZFS_EXIT(zfsvfs); 1244 return (error); 1245} 1246 1247/* 1248 * If vnode is for a device return a specfs vnode instead. 1249 */ 1250static int 1251specvp_check(vnode_t **vpp, cred_t *cr) 1252{ 1253 int error = 0; 1254 1255 if (IS_DEVVP(*vpp)) { 1256 struct vnode *svp; 1257 1258 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1259 VN_RELE(*vpp); 1260 if (svp == NULL) 1261 error = ENOSYS; 1262 *vpp = svp; 1263 } 1264 return (error); 1265} 1266 1267 1268/* 1269 * Lookup an entry in a directory, or an extended attribute directory. 1270 * If it exists, return a held vnode reference for it. 1271 * 1272 * IN: dvp - vnode of directory to search. 1273 * nm - name of entry to lookup. 1274 * pnp - full pathname to lookup [UNUSED]. 1275 * flags - LOOKUP_XATTR set if looking for an attribute. 1276 * rdir - root directory vnode [UNUSED]. 1277 * cr - credentials of caller. 1278 * ct - caller context 1279 * direntflags - directory lookup flags 1280 * realpnp - returned pathname. 1281 * 1282 * OUT: vpp - vnode of located entry, NULL if not found. 1283 * 1284 * RETURN: 0 if success 1285 * error code if failure 1286 * 1287 * Timestamps: 1288 * NA 1289 */ 1290/* ARGSUSED */ 1291static int 1292zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct componentname *cnp, 1293 int nameiop, cred_t *cr, kthread_t *td, int flags) 1294{ 1295 znode_t *zdp = VTOZ(dvp); 1296 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1297 int error = 0; 1298 int *direntflags = NULL; 1299 void *realpnp = NULL; 1300 1301 /* fast path */ 1302 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 1303 1304 if (dvp->v_type != VDIR) { 1305 return (ENOTDIR); 1306 } else if (zdp->z_sa_hdl == NULL) { 1307 return (EIO); 1308 } 1309 1310 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 1311 error = zfs_fastaccesschk_execute(zdp, cr); 1312 if (!error) { 1313 *vpp = dvp; 1314 VN_HOLD(*vpp); 1315 return (0); 1316 } 1317 return (error); 1318 } else { 1319 vnode_t *tvp = dnlc_lookup(dvp, nm); 1320 1321 if (tvp) { 1322 error = zfs_fastaccesschk_execute(zdp, cr); 1323 if (error) { 1324 VN_RELE(tvp); 1325 return (error); 1326 } 1327 if (tvp == DNLC_NO_VNODE) { 1328 VN_RELE(tvp); 1329 return (ENOENT); 1330 } else { 1331 *vpp = tvp; 1332 return (specvp_check(vpp, cr)); 1333 } 1334 } 1335 } 1336 } 1337 1338 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1339 1340 ZFS_ENTER(zfsvfs); 1341 ZFS_VERIFY_ZP(zdp); 1342 1343 *vpp = NULL; 1344 1345 if (flags & LOOKUP_XATTR) { 1346#ifdef TODO 1347 /* 1348 * If the xattr property is off, refuse the lookup request. 1349 */ 1350 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1351 ZFS_EXIT(zfsvfs); 1352 return (EINVAL); 1353 } 1354#endif 1355 1356 /* 1357 * We don't allow recursive attributes.. 1358 * Maybe someday we will. 1359 */ 1360 if (zdp->z_pflags & ZFS_XATTR) { 1361 ZFS_EXIT(zfsvfs); 1362 return (EINVAL); 1363 } 1364 1365 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1366 ZFS_EXIT(zfsvfs); 1367 return (error); 1368 } 1369 1370 /* 1371 * Do we have permission to get into attribute directory? 1372 */ 1373 1374 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1375 B_FALSE, cr)) { 1376 VN_RELE(*vpp); 1377 *vpp = NULL; 1378 } 1379 1380 ZFS_EXIT(zfsvfs); 1381 return (error); 1382 } 1383 1384 if (dvp->v_type != VDIR) { 1385 ZFS_EXIT(zfsvfs); 1386 return (ENOTDIR); 1387 } 1388 1389 /* 1390 * Check accessibility of directory. 1391 */ 1392 1393 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1394 ZFS_EXIT(zfsvfs); 1395 return (error); 1396 } 1397 1398 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1399 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1400 ZFS_EXIT(zfsvfs); 1401 return (EILSEQ); 1402 } 1403 1404 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1405 if (error == 0) 1406 error = specvp_check(vpp, cr); 1407 1408 /* Translate errors and add SAVENAME when needed. */ 1409 if (cnp->cn_flags & ISLASTCN) { 1410 switch (nameiop) { 1411 case CREATE: 1412 case RENAME: 1413 if (error == ENOENT) { 1414 error = EJUSTRETURN; 1415 cnp->cn_flags |= SAVENAME; 1416 break; 1417 } 1418 /* FALLTHROUGH */ 1419 case DELETE: 1420 if (error == 0) 1421 cnp->cn_flags |= SAVENAME; 1422 break; 1423 } 1424 } 1425 if (error == 0 && (nm[0] != '.' || nm[1] != '\0')) { 1426 int ltype = 0; 1427 1428 if (cnp->cn_flags & ISDOTDOT) { 1429 ltype = VOP_ISLOCKED(dvp); 1430 VOP_UNLOCK(dvp, 0); 1431 } 1432 ZFS_EXIT(zfsvfs); 1433 error = zfs_vnode_lock(*vpp, cnp->cn_lkflags); 1434 if (cnp->cn_flags & ISDOTDOT) 1435 vn_lock(dvp, ltype | LK_RETRY); 1436 if (error != 0) { 1437 VN_RELE(*vpp); 1438 *vpp = NULL; 1439 return (error); 1440 } 1441 } else { 1442 ZFS_EXIT(zfsvfs); 1443 } 1444 1445#ifdef FREEBSD_NAMECACHE 1446 /* 1447 * Insert name into cache (as non-existent) if appropriate. 1448 */ 1449 if (error == ENOENT && (cnp->cn_flags & MAKEENTRY) && nameiop != CREATE) 1450 cache_enter(dvp, *vpp, cnp); 1451 /* 1452 * Insert name into cache if appropriate. 1453 */ 1454 if (error == 0 && (cnp->cn_flags & MAKEENTRY)) { 1455 if (!(cnp->cn_flags & ISLASTCN) || 1456 (nameiop != DELETE && nameiop != RENAME)) { 1457 cache_enter(dvp, *vpp, cnp); 1458 } 1459 } 1460#endif 1461 1462 return (error); 1463} 1464 1465/* 1466 * Attempt to create a new entry in a directory. If the entry 1467 * already exists, truncate the file if permissible, else return 1468 * an error. Return the vp of the created or trunc'd file. 1469 * 1470 * IN: dvp - vnode of directory to put new file entry in. 1471 * name - name of new file entry. 1472 * vap - attributes of new file. 1473 * excl - flag indicating exclusive or non-exclusive mode. 1474 * mode - mode to open file with. 1475 * cr - credentials of caller. 1476 * flag - large file flag [UNUSED]. 1477 * ct - caller context 1478 * vsecp - ACL to be set 1479 * 1480 * OUT: vpp - vnode of created or trunc'd entry. 1481 * 1482 * RETURN: 0 if success 1483 * error code if failure 1484 * 1485 * Timestamps: 1486 * dvp - ctime|mtime updated if new entry created 1487 * vp - ctime|mtime always, atime if new 1488 */ 1489 1490/* ARGSUSED */ 1491static int 1492zfs_create(vnode_t *dvp, char *name, vattr_t *vap, int excl, int mode, 1493 vnode_t **vpp, cred_t *cr, kthread_t *td) 1494{ 1495 znode_t *zp, *dzp = VTOZ(dvp); 1496 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1497 zilog_t *zilog; 1498 objset_t *os; 1499 zfs_dirlock_t *dl; 1500 dmu_tx_t *tx; 1501 int error; 1502 ksid_t *ksid; 1503 uid_t uid; 1504 gid_t gid = crgetgid(cr); 1505 zfs_acl_ids_t acl_ids; 1506 boolean_t fuid_dirtied; 1507 boolean_t have_acl = B_FALSE; 1508 void *vsecp = NULL; 1509 int flag = 0; 1510 1511 /* 1512 * If we have an ephemeral id, ACL, or XVATTR then 1513 * make sure file system is at proper version 1514 */ 1515 1516 ksid = crgetsid(cr, KSID_OWNER); 1517 if (ksid) 1518 uid = ksid_getid(ksid); 1519 else 1520 uid = crgetuid(cr); 1521 1522 if (zfsvfs->z_use_fuids == B_FALSE && 1523 (vsecp || (vap->va_mask & AT_XVATTR) || 1524 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1525 return (EINVAL); 1526 1527 ZFS_ENTER(zfsvfs); 1528 ZFS_VERIFY_ZP(dzp); 1529 os = zfsvfs->z_os; 1530 zilog = zfsvfs->z_log; 1531 1532 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1533 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1534 ZFS_EXIT(zfsvfs); 1535 return (EILSEQ); 1536 } 1537 1538 if (vap->va_mask & AT_XVATTR) { 1539 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 1540 crgetuid(cr), cr, vap->va_type)) != 0) { 1541 ZFS_EXIT(zfsvfs); 1542 return (error); 1543 } 1544 } 1545top: 1546 *vpp = NULL; 1547 1548 if ((vap->va_mode & S_ISVTX) && secpolicy_vnode_stky_modify(cr)) 1549 vap->va_mode &= ~S_ISVTX; 1550 1551 if (*name == '\0') { 1552 /* 1553 * Null component name refers to the directory itself. 1554 */ 1555 VN_HOLD(dvp); 1556 zp = dzp; 1557 dl = NULL; 1558 error = 0; 1559 } else { 1560 /* possible VN_HOLD(zp) */ 1561 int zflg = 0; 1562 1563 if (flag & FIGNORECASE) 1564 zflg |= ZCILOOK; 1565 1566 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1567 NULL, NULL); 1568 if (error) { 1569 if (have_acl) 1570 zfs_acl_ids_free(&acl_ids); 1571 if (strcmp(name, "..") == 0) 1572 error = EISDIR; 1573 ZFS_EXIT(zfsvfs); 1574 return (error); 1575 } 1576 } 1577 1578 if (zp == NULL) { 1579 uint64_t txtype; 1580 1581 /* 1582 * Create a new file object and update the directory 1583 * to reference it. 1584 */ 1585 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1586 if (have_acl) 1587 zfs_acl_ids_free(&acl_ids); 1588 goto out; 1589 } 1590 1591 /* 1592 * We only support the creation of regular files in 1593 * extended attribute directories. 1594 */ 1595 1596 if ((dzp->z_pflags & ZFS_XATTR) && 1597 (vap->va_type != VREG)) { 1598 if (have_acl) 1599 zfs_acl_ids_free(&acl_ids); 1600 error = EINVAL; 1601 goto out; 1602 } 1603 1604 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 1605 cr, vsecp, &acl_ids)) != 0) 1606 goto out; 1607 have_acl = B_TRUE; 1608 1609 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1610 zfs_acl_ids_free(&acl_ids); 1611 error = EDQUOT; 1612 goto out; 1613 } 1614 1615 tx = dmu_tx_create(os); 1616 1617 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1618 ZFS_SA_BASE_ATTR_SIZE); 1619 1620 fuid_dirtied = zfsvfs->z_fuid_dirty; 1621 if (fuid_dirtied) 1622 zfs_fuid_txhold(zfsvfs, tx); 1623 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1624 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1625 if (!zfsvfs->z_use_sa && 1626 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1627 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1628 0, acl_ids.z_aclp->z_acl_bytes); 1629 } 1630 error = dmu_tx_assign(tx, TXG_NOWAIT); 1631 if (error) { 1632 zfs_dirent_unlock(dl); 1633 if (error == ERESTART) { 1634 dmu_tx_wait(tx); 1635 dmu_tx_abort(tx); 1636 goto top; 1637 } 1638 zfs_acl_ids_free(&acl_ids); 1639 dmu_tx_abort(tx); 1640 ZFS_EXIT(zfsvfs); 1641 return (error); 1642 } 1643 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1644 1645 if (fuid_dirtied) 1646 zfs_fuid_sync(zfsvfs, tx); 1647 1648 (void) zfs_link_create(dl, zp, tx, ZNEW); 1649 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1650 if (flag & FIGNORECASE) 1651 txtype |= TX_CI; 1652 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1653 vsecp, acl_ids.z_fuidp, vap); 1654 zfs_acl_ids_free(&acl_ids); 1655 dmu_tx_commit(tx); 1656 } else { 1657 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1658 1659 if (have_acl) 1660 zfs_acl_ids_free(&acl_ids); 1661 have_acl = B_FALSE; 1662 1663 /* 1664 * A directory entry already exists for this name. 1665 */ 1666 /* 1667 * Can't truncate an existing file if in exclusive mode. 1668 */ 1669 if (excl == EXCL) { 1670 error = EEXIST; 1671 goto out; 1672 } 1673 /* 1674 * Can't open a directory for writing. 1675 */ 1676 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1677 error = EISDIR; 1678 goto out; 1679 } 1680 /* 1681 * Verify requested access to file. 1682 */ 1683 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1684 goto out; 1685 } 1686 1687 mutex_enter(&dzp->z_lock); 1688 dzp->z_seq++; 1689 mutex_exit(&dzp->z_lock); 1690 1691 /* 1692 * Truncate regular files if requested. 1693 */ 1694 if ((ZTOV(zp)->v_type == VREG) && 1695 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1696 /* we can't hold any locks when calling zfs_freesp() */ 1697 zfs_dirent_unlock(dl); 1698 dl = NULL; 1699 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1700 if (error == 0) { 1701 vnevent_create(ZTOV(zp), ct); 1702 } 1703 } 1704 } 1705out: 1706 if (dl) 1707 zfs_dirent_unlock(dl); 1708 1709 if (error) { 1710 if (zp) 1711 VN_RELE(ZTOV(zp)); 1712 } else { 1713 *vpp = ZTOV(zp); 1714 error = specvp_check(vpp, cr); 1715 } 1716 1717 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1718 zil_commit(zilog, 0); 1719 1720 ZFS_EXIT(zfsvfs); 1721 return (error); 1722} 1723 1724/* 1725 * Remove an entry from a directory. 1726 * 1727 * IN: dvp - vnode of directory to remove entry from. 1728 * name - name of entry to remove. 1729 * cr - credentials of caller. 1730 * ct - caller context 1731 * flags - case flags 1732 * 1733 * RETURN: 0 if success 1734 * error code if failure 1735 * 1736 * Timestamps: 1737 * dvp - ctime|mtime 1738 * vp - ctime (if nlink > 0) 1739 */ 1740 1741uint64_t null_xattr = 0; 1742 1743/*ARGSUSED*/ 1744static int 1745zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1746 int flags) 1747{ 1748 znode_t *zp, *dzp = VTOZ(dvp); 1749 znode_t *xzp; 1750 vnode_t *vp; 1751 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1752 zilog_t *zilog; 1753 uint64_t acl_obj, xattr_obj; 1754 uint64_t xattr_obj_unlinked = 0; 1755 uint64_t obj = 0; 1756 zfs_dirlock_t *dl; 1757 dmu_tx_t *tx; 1758 boolean_t may_delete_now, delete_now = FALSE; 1759 boolean_t unlinked, toobig = FALSE; 1760 uint64_t txtype; 1761 pathname_t *realnmp = NULL; 1762 pathname_t realnm; 1763 int error; 1764 int zflg = ZEXISTS; 1765 1766 ZFS_ENTER(zfsvfs); 1767 ZFS_VERIFY_ZP(dzp); 1768 zilog = zfsvfs->z_log; 1769 1770 if (flags & FIGNORECASE) { 1771 zflg |= ZCILOOK; 1772 pn_alloc(&realnm); 1773 realnmp = &realnm; 1774 } 1775 1776top: 1777 xattr_obj = 0; 1778 xzp = NULL; 1779 /* 1780 * Attempt to lock directory; fail if entry doesn't exist. 1781 */ 1782 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1783 NULL, realnmp)) { 1784 if (realnmp) 1785 pn_free(realnmp); 1786 ZFS_EXIT(zfsvfs); 1787 return (error); 1788 } 1789 1790 vp = ZTOV(zp); 1791 1792 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1793 goto out; 1794 } 1795 1796 /* 1797 * Need to use rmdir for removing directories. 1798 */ 1799 if (vp->v_type == VDIR) { 1800 error = EPERM; 1801 goto out; 1802 } 1803 1804 vnevent_remove(vp, dvp, name, ct); 1805 1806 if (realnmp) 1807 dnlc_remove(dvp, realnmp->pn_buf); 1808 else 1809 dnlc_remove(dvp, name); 1810 1811 VI_LOCK(vp); 1812 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1813 VI_UNLOCK(vp); 1814 1815 /* 1816 * We may delete the znode now, or we may put it in the unlinked set; 1817 * it depends on whether we're the last link, and on whether there are 1818 * other holds on the vnode. So we dmu_tx_hold() the right things to 1819 * allow for either case. 1820 */ 1821 obj = zp->z_id; 1822 tx = dmu_tx_create(zfsvfs->z_os); 1823 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1824 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1825 zfs_sa_upgrade_txholds(tx, zp); 1826 zfs_sa_upgrade_txholds(tx, dzp); 1827 if (may_delete_now) { 1828 toobig = 1829 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 1830 /* if the file is too big, only hold_free a token amount */ 1831 dmu_tx_hold_free(tx, zp->z_id, 0, 1832 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1833 } 1834 1835 /* are there any extended attributes? */ 1836 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1837 &xattr_obj, sizeof (xattr_obj)); 1838 if (error == 0 && xattr_obj) { 1839 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1840 ASSERT3U(error, ==, 0); 1841 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1842 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1843 } 1844 1845 mutex_enter(&zp->z_lock); 1846 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now) 1847 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1848 mutex_exit(&zp->z_lock); 1849 1850 /* charge as an update -- would be nice not to charge at all */ 1851 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1852 1853 error = dmu_tx_assign(tx, TXG_NOWAIT); 1854 if (error) { 1855 zfs_dirent_unlock(dl); 1856 VN_RELE(vp); 1857 if (xzp) 1858 VN_RELE(ZTOV(xzp)); 1859 if (error == ERESTART) { 1860 dmu_tx_wait(tx); 1861 dmu_tx_abort(tx); 1862 goto top; 1863 } 1864 if (realnmp) 1865 pn_free(realnmp); 1866 dmu_tx_abort(tx); 1867 ZFS_EXIT(zfsvfs); 1868 return (error); 1869 } 1870 1871 /* 1872 * Remove the directory entry. 1873 */ 1874 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1875 1876 if (error) { 1877 dmu_tx_commit(tx); 1878 goto out; 1879 } 1880 1881 if (unlinked) { 1882 1883 /* 1884 * Hold z_lock so that we can make sure that the ACL obj 1885 * hasn't changed. Could have been deleted due to 1886 * zfs_sa_upgrade(). 1887 */ 1888 mutex_enter(&zp->z_lock); 1889 VI_LOCK(vp); 1890 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1891 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked)); 1892 delete_now = may_delete_now && !toobig && 1893 vp->v_count == 1 && !vn_has_cached_data(vp) && 1894 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) == 1895 acl_obj; 1896 VI_UNLOCK(vp); 1897 } 1898 1899 if (delete_now) { 1900 if (xattr_obj_unlinked) { 1901 ASSERT3U(xzp->z_links, ==, 2); 1902 mutex_enter(&xzp->z_lock); 1903 xzp->z_unlinked = 1; 1904 xzp->z_links = 0; 1905 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 1906 &xzp->z_links, sizeof (xzp->z_links), tx); 1907 ASSERT3U(error, ==, 0); 1908 mutex_exit(&xzp->z_lock); 1909 zfs_unlinked_add(xzp, tx); 1910 1911 if (zp->z_is_sa) 1912 error = sa_remove(zp->z_sa_hdl, 1913 SA_ZPL_XATTR(zfsvfs), tx); 1914 else 1915 error = sa_update(zp->z_sa_hdl, 1916 SA_ZPL_XATTR(zfsvfs), &null_xattr, 1917 sizeof (uint64_t), tx); 1918 ASSERT3U(error, ==, 0); 1919 } 1920 VI_LOCK(vp); 1921 vp->v_count--; 1922 ASSERT3U(vp->v_count, ==, 0); 1923 VI_UNLOCK(vp); 1924 mutex_exit(&zp->z_lock); 1925 zfs_znode_delete(zp, tx); 1926 } else if (unlinked) { 1927 mutex_exit(&zp->z_lock); 1928 zfs_unlinked_add(zp, tx); 1929 } 1930 1931 txtype = TX_REMOVE; 1932 if (flags & FIGNORECASE) 1933 txtype |= TX_CI; 1934 zfs_log_remove(zilog, tx, txtype, dzp, name, obj); 1935 1936 dmu_tx_commit(tx); 1937out: 1938 if (realnmp) 1939 pn_free(realnmp); 1940 1941 zfs_dirent_unlock(dl); 1942 1943 if (!delete_now) 1944 VN_RELE(vp); 1945 if (xzp) 1946 VN_RELE(ZTOV(xzp)); 1947 1948 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1949 zil_commit(zilog, 0); 1950 1951 ZFS_EXIT(zfsvfs); 1952 return (error); 1953} 1954 1955/* 1956 * Create a new directory and insert it into dvp using the name 1957 * provided. Return a pointer to the inserted directory. 1958 * 1959 * IN: dvp - vnode of directory to add subdir to. 1960 * dirname - name of new directory. 1961 * vap - attributes of new directory. 1962 * cr - credentials of caller. 1963 * ct - caller context 1964 * vsecp - ACL to be set 1965 * 1966 * OUT: vpp - vnode of created directory. 1967 * 1968 * RETURN: 0 if success 1969 * error code if failure 1970 * 1971 * Timestamps: 1972 * dvp - ctime|mtime updated 1973 * vp - ctime|mtime|atime updated 1974 */ 1975/*ARGSUSED*/ 1976static int 1977zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 1978 caller_context_t *ct, int flags, vsecattr_t *vsecp) 1979{ 1980 znode_t *zp, *dzp = VTOZ(dvp); 1981 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1982 zilog_t *zilog; 1983 zfs_dirlock_t *dl; 1984 uint64_t txtype; 1985 dmu_tx_t *tx; 1986 int error; 1987 int zf = ZNEW; 1988 ksid_t *ksid; 1989 uid_t uid; 1990 gid_t gid = crgetgid(cr); 1991 zfs_acl_ids_t acl_ids; 1992 boolean_t fuid_dirtied; 1993 1994 ASSERT(vap->va_type == VDIR); 1995 1996 /* 1997 * If we have an ephemeral id, ACL, or XVATTR then 1998 * make sure file system is at proper version 1999 */ 2000 2001 ksid = crgetsid(cr, KSID_OWNER); 2002 if (ksid) 2003 uid = ksid_getid(ksid); 2004 else 2005 uid = crgetuid(cr); 2006 if (zfsvfs->z_use_fuids == B_FALSE && 2007 (vsecp || (vap->va_mask & AT_XVATTR) || 2008 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 2009 return (EINVAL); 2010 2011 ZFS_ENTER(zfsvfs); 2012 ZFS_VERIFY_ZP(dzp); 2013 zilog = zfsvfs->z_log; 2014 2015 if (dzp->z_pflags & ZFS_XATTR) { 2016 ZFS_EXIT(zfsvfs); 2017 return (EINVAL); 2018 } 2019 2020 if (zfsvfs->z_utf8 && u8_validate(dirname, 2021 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2022 ZFS_EXIT(zfsvfs); 2023 return (EILSEQ); 2024 } 2025 if (flags & FIGNORECASE) 2026 zf |= ZCILOOK; 2027 2028 if (vap->va_mask & AT_XVATTR) { 2029 if ((error = secpolicy_xvattr(dvp, (xvattr_t *)vap, 2030 crgetuid(cr), cr, vap->va_type)) != 0) { 2031 ZFS_EXIT(zfsvfs); 2032 return (error); 2033 } 2034 } 2035 2036 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 2037 vsecp, &acl_ids)) != 0) { 2038 ZFS_EXIT(zfsvfs); 2039 return (error); 2040 } 2041 /* 2042 * First make sure the new directory doesn't exist. 2043 * 2044 * Existence is checked first to make sure we don't return 2045 * EACCES instead of EEXIST which can cause some applications 2046 * to fail. 2047 */ 2048top: 2049 *vpp = NULL; 2050 2051 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 2052 NULL, NULL)) { 2053 zfs_acl_ids_free(&acl_ids); 2054 ZFS_EXIT(zfsvfs); 2055 return (error); 2056 } 2057 2058 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 2059 zfs_acl_ids_free(&acl_ids); 2060 zfs_dirent_unlock(dl); 2061 ZFS_EXIT(zfsvfs); 2062 return (error); 2063 } 2064 2065 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 2066 zfs_acl_ids_free(&acl_ids); 2067 zfs_dirent_unlock(dl); 2068 ZFS_EXIT(zfsvfs); 2069 return (EDQUOT); 2070 } 2071 2072 /* 2073 * Add a new entry to the directory. 2074 */ 2075 tx = dmu_tx_create(zfsvfs->z_os); 2076 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 2077 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2078 fuid_dirtied = zfsvfs->z_fuid_dirty; 2079 if (fuid_dirtied) 2080 zfs_fuid_txhold(zfsvfs, tx); 2081 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2082 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 2083 acl_ids.z_aclp->z_acl_bytes); 2084 } 2085 2086 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 2087 ZFS_SA_BASE_ATTR_SIZE); 2088 2089 error = dmu_tx_assign(tx, TXG_NOWAIT); 2090 if (error) { 2091 zfs_dirent_unlock(dl); 2092 if (error == ERESTART) { 2093 dmu_tx_wait(tx); 2094 dmu_tx_abort(tx); 2095 goto top; 2096 } 2097 zfs_acl_ids_free(&acl_ids); 2098 dmu_tx_abort(tx); 2099 ZFS_EXIT(zfsvfs); 2100 return (error); 2101 } 2102 2103 /* 2104 * Create new node. 2105 */ 2106 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 2107 2108 if (fuid_dirtied) 2109 zfs_fuid_sync(zfsvfs, tx); 2110 2111 /* 2112 * Now put new name in parent dir. 2113 */ 2114 (void) zfs_link_create(dl, zp, tx, ZNEW); 2115 2116 *vpp = ZTOV(zp); 2117 2118 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 2119 if (flags & FIGNORECASE) 2120 txtype |= TX_CI; 2121 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 2122 acl_ids.z_fuidp, vap); 2123 2124 zfs_acl_ids_free(&acl_ids); 2125 2126 dmu_tx_commit(tx); 2127 2128 zfs_dirent_unlock(dl); 2129 2130 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2131 zil_commit(zilog, 0); 2132 2133 ZFS_EXIT(zfsvfs); 2134 return (0); 2135} 2136 2137/* 2138 * Remove a directory subdir entry. If the current working 2139 * directory is the same as the subdir to be removed, the 2140 * remove will fail. 2141 * 2142 * IN: dvp - vnode of directory to remove from. 2143 * name - name of directory to be removed. 2144 * cwd - vnode of current working directory. 2145 * cr - credentials of caller. 2146 * ct - caller context 2147 * flags - case flags 2148 * 2149 * RETURN: 0 if success 2150 * error code if failure 2151 * 2152 * Timestamps: 2153 * dvp - ctime|mtime updated 2154 */ 2155/*ARGSUSED*/ 2156static int 2157zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 2158 caller_context_t *ct, int flags) 2159{ 2160 znode_t *dzp = VTOZ(dvp); 2161 znode_t *zp; 2162 vnode_t *vp; 2163 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2164 zilog_t *zilog; 2165 zfs_dirlock_t *dl; 2166 dmu_tx_t *tx; 2167 int error; 2168 int zflg = ZEXISTS; 2169 2170 ZFS_ENTER(zfsvfs); 2171 ZFS_VERIFY_ZP(dzp); 2172 zilog = zfsvfs->z_log; 2173 2174 if (flags & FIGNORECASE) 2175 zflg |= ZCILOOK; 2176top: 2177 zp = NULL; 2178 2179 /* 2180 * Attempt to lock directory; fail if entry doesn't exist. 2181 */ 2182 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 2183 NULL, NULL)) { 2184 ZFS_EXIT(zfsvfs); 2185 return (error); 2186 } 2187 2188 vp = ZTOV(zp); 2189 2190 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 2191 goto out; 2192 } 2193 2194 if (vp->v_type != VDIR) { 2195 error = ENOTDIR; 2196 goto out; 2197 } 2198 2199 if (vp == cwd) { 2200 error = EINVAL; 2201 goto out; 2202 } 2203 2204 vnevent_rmdir(vp, dvp, name, ct); 2205 2206 /* 2207 * Grab a lock on the directory to make sure that noone is 2208 * trying to add (or lookup) entries while we are removing it. 2209 */ 2210 rw_enter(&zp->z_name_lock, RW_WRITER); 2211 2212 /* 2213 * Grab a lock on the parent pointer to make sure we play well 2214 * with the treewalk and directory rename code. 2215 */ 2216 rw_enter(&zp->z_parent_lock, RW_WRITER); 2217 2218 tx = dmu_tx_create(zfsvfs->z_os); 2219 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2220 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2221 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2222 zfs_sa_upgrade_txholds(tx, zp); 2223 zfs_sa_upgrade_txholds(tx, dzp); 2224 error = dmu_tx_assign(tx, TXG_NOWAIT); 2225 if (error) { 2226 rw_exit(&zp->z_parent_lock); 2227 rw_exit(&zp->z_name_lock); 2228 zfs_dirent_unlock(dl); 2229 VN_RELE(vp); 2230 if (error == ERESTART) { 2231 dmu_tx_wait(tx); 2232 dmu_tx_abort(tx); 2233 goto top; 2234 } 2235 dmu_tx_abort(tx); 2236 ZFS_EXIT(zfsvfs); 2237 return (error); 2238 } 2239 2240#ifdef FREEBSD_NAMECACHE 2241 cache_purge(dvp); 2242#endif 2243 2244 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 2245 2246 if (error == 0) { 2247 uint64_t txtype = TX_RMDIR; 2248 if (flags & FIGNORECASE) 2249 txtype |= TX_CI; 2250 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT); 2251 } 2252 2253 dmu_tx_commit(tx); 2254 2255 rw_exit(&zp->z_parent_lock); 2256 rw_exit(&zp->z_name_lock); 2257#ifdef FREEBSD_NAMECACHE 2258 cache_purge(vp); 2259#endif 2260out: 2261 zfs_dirent_unlock(dl); 2262 2263 VN_RELE(vp); 2264 2265 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2266 zil_commit(zilog, 0); 2267 2268 ZFS_EXIT(zfsvfs); 2269 return (error); 2270} 2271 2272/* 2273 * Read as many directory entries as will fit into the provided 2274 * buffer from the given directory cursor position (specified in 2275 * the uio structure. 2276 * 2277 * IN: vp - vnode of directory to read. 2278 * uio - structure supplying read location, range info, 2279 * and return buffer. 2280 * cr - credentials of caller. 2281 * ct - caller context 2282 * flags - case flags 2283 * 2284 * OUT: uio - updated offset and range, buffer filled. 2285 * eofp - set to true if end-of-file detected. 2286 * 2287 * RETURN: 0 if success 2288 * error code if failure 2289 * 2290 * Timestamps: 2291 * vp - atime updated 2292 * 2293 * Note that the low 4 bits of the cookie returned by zap is always zero. 2294 * This allows us to use the low range for "special" directory entries: 2295 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 2296 * we use the offset 2 for the '.zfs' directory. 2297 */ 2298/* ARGSUSED */ 2299static int 2300zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, int *ncookies, u_long **cookies) 2301{ 2302 znode_t *zp = VTOZ(vp); 2303 iovec_t *iovp; 2304 edirent_t *eodp; 2305 dirent64_t *odp; 2306 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2307 objset_t *os; 2308 caddr_t outbuf; 2309 size_t bufsize; 2310 zap_cursor_t zc; 2311 zap_attribute_t zap; 2312 uint_t bytes_wanted; 2313 uint64_t offset; /* must be unsigned; checks for < 1 */ 2314 uint64_t parent; 2315 int local_eof; 2316 int outcount; 2317 int error; 2318 uint8_t prefetch; 2319 boolean_t check_sysattrs; 2320 uint8_t type; 2321 int ncooks; 2322 u_long *cooks = NULL; 2323 int flags = 0; 2324 2325 ZFS_ENTER(zfsvfs); 2326 ZFS_VERIFY_ZP(zp); 2327 2328 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 2329 &parent, sizeof (parent))) != 0) { 2330 ZFS_EXIT(zfsvfs); 2331 return (error); 2332 } 2333 2334 /* 2335 * If we are not given an eof variable, 2336 * use a local one. 2337 */ 2338 if (eofp == NULL) 2339 eofp = &local_eof; 2340 2341 /* 2342 * Check for valid iov_len. 2343 */ 2344 if (uio->uio_iov->iov_len <= 0) { 2345 ZFS_EXIT(zfsvfs); 2346 return (EINVAL); 2347 } 2348 2349 /* 2350 * Quit if directory has been removed (posix) 2351 */ 2352 if ((*eofp = zp->z_unlinked) != 0) { 2353 ZFS_EXIT(zfsvfs); 2354 return (0); 2355 } 2356 2357 error = 0; 2358 os = zfsvfs->z_os; 2359 offset = uio->uio_loffset; 2360 prefetch = zp->z_zn_prefetch; 2361 2362 /* 2363 * Initialize the iterator cursor. 2364 */ 2365 if (offset <= 3) { 2366 /* 2367 * Start iteration from the beginning of the directory. 2368 */ 2369 zap_cursor_init(&zc, os, zp->z_id); 2370 } else { 2371 /* 2372 * The offset is a serialized cursor. 2373 */ 2374 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 2375 } 2376 2377 /* 2378 * Get space to change directory entries into fs independent format. 2379 */ 2380 iovp = uio->uio_iov; 2381 bytes_wanted = iovp->iov_len; 2382 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 2383 bufsize = bytes_wanted; 2384 outbuf = kmem_alloc(bufsize, KM_SLEEP); 2385 odp = (struct dirent64 *)outbuf; 2386 } else { 2387 bufsize = bytes_wanted; 2388 odp = (struct dirent64 *)iovp->iov_base; 2389 } 2390 eodp = (struct edirent *)odp; 2391 2392 if (ncookies != NULL) { 2393 /* 2394 * Minimum entry size is dirent size and 1 byte for a file name. 2395 */ 2396 ncooks = uio->uio_resid / (sizeof(struct dirent) - sizeof(((struct dirent *)NULL)->d_name) + 1); 2397 cooks = malloc(ncooks * sizeof(u_long), M_TEMP, M_WAITOK); 2398 *cookies = cooks; 2399 *ncookies = ncooks; 2400 } 2401 /* 2402 * If this VFS supports the system attribute view interface; and 2403 * we're looking at an extended attribute directory; and we care 2404 * about normalization conflicts on this vfs; then we must check 2405 * for normalization conflicts with the sysattr name space. 2406 */ 2407#ifdef TODO 2408 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 2409 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 2410 (flags & V_RDDIR_ENTFLAGS); 2411#else 2412 check_sysattrs = 0; 2413#endif 2414 2415 /* 2416 * Transform to file-system independent format 2417 */ 2418 outcount = 0; 2419 while (outcount < bytes_wanted) { 2420 ino64_t objnum; 2421 ushort_t reclen; 2422 off64_t *next = NULL; 2423 2424 /* 2425 * Special case `.', `..', and `.zfs'. 2426 */ 2427 if (offset == 0) { 2428 (void) strcpy(zap.za_name, "."); 2429 zap.za_normalization_conflict = 0; 2430 objnum = zp->z_id; 2431 type = DT_DIR; 2432 } else if (offset == 1) { 2433 (void) strcpy(zap.za_name, ".."); 2434 zap.za_normalization_conflict = 0; 2435 objnum = parent; 2436 type = DT_DIR; 2437 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2438 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2439 zap.za_normalization_conflict = 0; 2440 objnum = ZFSCTL_INO_ROOT; 2441 type = DT_DIR; 2442 } else { 2443 /* 2444 * Grab next entry. 2445 */ 2446 if (error = zap_cursor_retrieve(&zc, &zap)) { 2447 if ((*eofp = (error == ENOENT)) != 0) 2448 break; 2449 else 2450 goto update; 2451 } 2452 2453 if (zap.za_integer_length != 8 || 2454 zap.za_num_integers != 1) { 2455 cmn_err(CE_WARN, "zap_readdir: bad directory " 2456 "entry, obj = %lld, offset = %lld\n", 2457 (u_longlong_t)zp->z_id, 2458 (u_longlong_t)offset); 2459 error = ENXIO; 2460 goto update; 2461 } 2462 2463 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2464 /* 2465 * MacOS X can extract the object type here such as: 2466 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2467 */ 2468 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2469 2470 if (check_sysattrs && !zap.za_normalization_conflict) { 2471#ifdef TODO 2472 zap.za_normalization_conflict = 2473 xattr_sysattr_casechk(zap.za_name); 2474#else 2475 panic("%s:%u: TODO", __func__, __LINE__); 2476#endif 2477 } 2478 } 2479 2480 if (flags & V_RDDIR_ACCFILTER) { 2481 /* 2482 * If we have no access at all, don't include 2483 * this entry in the returned information 2484 */ 2485 znode_t *ezp; 2486 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2487 goto skip_entry; 2488 if (!zfs_has_access(ezp, cr)) { 2489 VN_RELE(ZTOV(ezp)); 2490 goto skip_entry; 2491 } 2492 VN_RELE(ZTOV(ezp)); 2493 } 2494 2495 if (flags & V_RDDIR_ENTFLAGS) 2496 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2497 else 2498 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2499 2500 /* 2501 * Will this entry fit in the buffer? 2502 */ 2503 if (outcount + reclen > bufsize) { 2504 /* 2505 * Did we manage to fit anything in the buffer? 2506 */ 2507 if (!outcount) { 2508 error = EINVAL; 2509 goto update; 2510 } 2511 break; 2512 } 2513 if (flags & V_RDDIR_ENTFLAGS) { 2514 /* 2515 * Add extended flag entry: 2516 */ 2517 eodp->ed_ino = objnum; 2518 eodp->ed_reclen = reclen; 2519 /* NOTE: ed_off is the offset for the *next* entry */ 2520 next = &(eodp->ed_off); 2521 eodp->ed_eflags = zap.za_normalization_conflict ? 2522 ED_CASE_CONFLICT : 0; 2523 (void) strncpy(eodp->ed_name, zap.za_name, 2524 EDIRENT_NAMELEN(reclen)); 2525 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2526 } else { 2527 /* 2528 * Add normal entry: 2529 */ 2530 odp->d_ino = objnum; 2531 odp->d_reclen = reclen; 2532 odp->d_namlen = strlen(zap.za_name); 2533 (void) strlcpy(odp->d_name, zap.za_name, odp->d_namlen + 1); 2534 odp->d_type = type; 2535 odp = (dirent64_t *)((intptr_t)odp + reclen); 2536 } 2537 outcount += reclen; 2538 2539 ASSERT(outcount <= bufsize); 2540 2541 /* Prefetch znode */ 2542 if (prefetch) 2543 dmu_prefetch(os, objnum, 0, 0); 2544 2545 skip_entry: 2546 /* 2547 * Move to the next entry, fill in the previous offset. 2548 */ 2549 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2550 zap_cursor_advance(&zc); 2551 offset = zap_cursor_serialize(&zc); 2552 } else { 2553 offset += 1; 2554 } 2555 2556 if (cooks != NULL) { 2557 *cooks++ = offset; 2558 ncooks--; 2559 KASSERT(ncooks >= 0, ("ncookies=%d", ncooks)); 2560 } 2561 } 2562 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2563 2564 /* Subtract unused cookies */ 2565 if (ncookies != NULL) 2566 *ncookies -= ncooks; 2567 2568 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2569 iovp->iov_base += outcount; 2570 iovp->iov_len -= outcount; 2571 uio->uio_resid -= outcount; 2572 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2573 /* 2574 * Reset the pointer. 2575 */ 2576 offset = uio->uio_loffset; 2577 } 2578 2579update: 2580 zap_cursor_fini(&zc); 2581 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2582 kmem_free(outbuf, bufsize); 2583 2584 if (error == ENOENT) 2585 error = 0; 2586 2587 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2588 2589 uio->uio_loffset = offset; 2590 ZFS_EXIT(zfsvfs); 2591 if (error != 0 && cookies != NULL) { 2592 free(*cookies, M_TEMP); 2593 *cookies = NULL; 2594 *ncookies = 0; 2595 } 2596 return (error); 2597} 2598 2599ulong_t zfs_fsync_sync_cnt = 4; 2600 2601static int 2602zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2603{ 2604 znode_t *zp = VTOZ(vp); 2605 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2606 2607 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2608 2609 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { 2610 ZFS_ENTER(zfsvfs); 2611 ZFS_VERIFY_ZP(zp); 2612 zil_commit(zfsvfs->z_log, zp->z_id); 2613 ZFS_EXIT(zfsvfs); 2614 } 2615 return (0); 2616} 2617 2618 2619/* 2620 * Get the requested file attributes and place them in the provided 2621 * vattr structure. 2622 * 2623 * IN: vp - vnode of file. 2624 * vap - va_mask identifies requested attributes. 2625 * If AT_XVATTR set, then optional attrs are requested 2626 * flags - ATTR_NOACLCHECK (CIFS server context) 2627 * cr - credentials of caller. 2628 * ct - caller context 2629 * 2630 * OUT: vap - attribute values. 2631 * 2632 * RETURN: 0 (always succeeds) 2633 */ 2634/* ARGSUSED */ 2635static int 2636zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2637 caller_context_t *ct) 2638{ 2639 znode_t *zp = VTOZ(vp); 2640 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2641 int error = 0; 2642 uint32_t blksize; 2643 u_longlong_t nblocks; 2644 uint64_t links; 2645 uint64_t mtime[2], ctime[2], crtime[2], rdev; 2646 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2647 xoptattr_t *xoap = NULL; 2648 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2649 sa_bulk_attr_t bulk[4]; 2650 int count = 0; 2651 2652 ZFS_ENTER(zfsvfs); 2653 ZFS_VERIFY_ZP(zp); 2654 2655 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2656 2657 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 2658 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 2659 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &crtime, 16); 2660 if (vp->v_type == VBLK || vp->v_type == VCHR) 2661 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_RDEV(zfsvfs), NULL, 2662 &rdev, 8); 2663 2664 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 2665 ZFS_EXIT(zfsvfs); 2666 return (error); 2667 } 2668 2669 /* 2670 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2671 * Also, if we are the owner don't bother, since owner should 2672 * always be allowed to read basic attributes of file. 2673 */ 2674 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 2675 (vap->va_uid != crgetuid(cr))) { 2676 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2677 skipaclchk, cr)) { 2678 ZFS_EXIT(zfsvfs); 2679 return (error); 2680 } 2681 } 2682 2683 /* 2684 * Return all attributes. It's cheaper to provide the answer 2685 * than to determine whether we were asked the question. 2686 */ 2687 2688 mutex_enter(&zp->z_lock); 2689 vap->va_type = IFTOVT(zp->z_mode); 2690 vap->va_mode = zp->z_mode & ~S_IFMT; 2691// vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2692 vap->va_nodeid = zp->z_id; 2693 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2694 links = zp->z_links + 1; 2695 else 2696 links = zp->z_links; 2697 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */ 2698 vap->va_size = zp->z_size; 2699 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 2700 if (vp->v_type == VBLK || vp->v_type == VCHR) 2701 vap->va_rdev = zfs_cmpldev(rdev); 2702 vap->va_seq = zp->z_seq; 2703 vap->va_flags = 0; /* FreeBSD: Reset chflags(2) flags. */ 2704 2705 /* 2706 * Add in any requested optional attributes and the create time. 2707 * Also set the corresponding bits in the returned attribute bitmap. 2708 */ 2709 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2710 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2711 xoap->xoa_archive = 2712 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2713 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2714 } 2715 2716 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2717 xoap->xoa_readonly = 2718 ((zp->z_pflags & ZFS_READONLY) != 0); 2719 XVA_SET_RTN(xvap, XAT_READONLY); 2720 } 2721 2722 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2723 xoap->xoa_system = 2724 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2725 XVA_SET_RTN(xvap, XAT_SYSTEM); 2726 } 2727 2728 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2729 xoap->xoa_hidden = 2730 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2731 XVA_SET_RTN(xvap, XAT_HIDDEN); 2732 } 2733 2734 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2735 xoap->xoa_nounlink = 2736 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2737 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2738 } 2739 2740 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2741 xoap->xoa_immutable = 2742 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2743 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2744 } 2745 2746 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2747 xoap->xoa_appendonly = 2748 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2749 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2750 } 2751 2752 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2753 xoap->xoa_nodump = 2754 ((zp->z_pflags & ZFS_NODUMP) != 0); 2755 XVA_SET_RTN(xvap, XAT_NODUMP); 2756 } 2757 2758 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2759 xoap->xoa_opaque = 2760 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2761 XVA_SET_RTN(xvap, XAT_OPAQUE); 2762 } 2763 2764 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2765 xoap->xoa_av_quarantined = 2766 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2767 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2768 } 2769 2770 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2771 xoap->xoa_av_modified = 2772 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2773 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2774 } 2775 2776 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2777 vp->v_type == VREG) { 2778 zfs_sa_get_scanstamp(zp, xvap); 2779 } 2780 2781 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2782 uint64_t times[2]; 2783 2784 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 2785 times, sizeof (times)); 2786 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 2787 XVA_SET_RTN(xvap, XAT_CREATETIME); 2788 } 2789 2790 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2791 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2792 XVA_SET_RTN(xvap, XAT_REPARSE); 2793 } 2794 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2795 xoap->xoa_generation = zp->z_gen; 2796 XVA_SET_RTN(xvap, XAT_GEN); 2797 } 2798 2799 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2800 xoap->xoa_offline = 2801 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2802 XVA_SET_RTN(xvap, XAT_OFFLINE); 2803 } 2804 2805 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2806 xoap->xoa_sparse = 2807 ((zp->z_pflags & ZFS_SPARSE) != 0); 2808 XVA_SET_RTN(xvap, XAT_SPARSE); 2809 } 2810 } 2811 2812 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2813 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2814 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2815 ZFS_TIME_DECODE(&vap->va_birthtime, crtime); 2816 2817 mutex_exit(&zp->z_lock); 2818 2819 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 2820 vap->va_blksize = blksize; 2821 vap->va_bytes = nblocks << 9; /* nblocks * 512 */ 2822 2823 if (zp->z_blksz == 0) { 2824 /* 2825 * Block size hasn't been set; suggest maximal I/O transfers. 2826 */ 2827 vap->va_blksize = zfsvfs->z_max_blksz; 2828 } 2829 2830 ZFS_EXIT(zfsvfs); 2831 return (0); 2832} 2833 2834/* 2835 * Set the file attributes to the values contained in the 2836 * vattr structure. 2837 * 2838 * IN: vp - vnode of file to be modified. 2839 * vap - new attribute values. 2840 * If AT_XVATTR set, then optional attrs are being set 2841 * flags - ATTR_UTIME set if non-default time values provided. 2842 * - ATTR_NOACLCHECK (CIFS context only). 2843 * cr - credentials of caller. 2844 * ct - caller context 2845 * 2846 * RETURN: 0 if success 2847 * error code if failure 2848 * 2849 * Timestamps: 2850 * vp - ctime updated, mtime updated if size changed. 2851 */ 2852/* ARGSUSED */ 2853static int 2854zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2855 caller_context_t *ct) 2856{ 2857 znode_t *zp = VTOZ(vp); 2858 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2859 zilog_t *zilog; 2860 dmu_tx_t *tx; 2861 vattr_t oldva; 2862 xvattr_t tmpxvattr; 2863 uint_t mask = vap->va_mask; 2864 uint_t saved_mask; 2865 uint64_t saved_mode; 2866 int trim_mask = 0; 2867 uint64_t new_mode; 2868 uint64_t new_uid, new_gid; 2869 uint64_t xattr_obj; 2870 uint64_t mtime[2], ctime[2]; 2871 znode_t *attrzp; 2872 int need_policy = FALSE; 2873 int err, err2; 2874 zfs_fuid_info_t *fuidp = NULL; 2875 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2876 xoptattr_t *xoap; 2877 zfs_acl_t *aclp; 2878 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2879 boolean_t fuid_dirtied = B_FALSE; 2880 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 2881 int count = 0, xattr_count = 0; 2882 2883 if (mask == 0) 2884 return (0); 2885 2886 if (mask & AT_NOSET) 2887 return (EINVAL); 2888 2889 ZFS_ENTER(zfsvfs); 2890 ZFS_VERIFY_ZP(zp); 2891 2892 zilog = zfsvfs->z_log; 2893 2894 /* 2895 * Make sure that if we have ephemeral uid/gid or xvattr specified 2896 * that file system is at proper version level 2897 */ 2898 2899 if (zfsvfs->z_use_fuids == B_FALSE && 2900 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2901 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2902 (mask & AT_XVATTR))) { 2903 ZFS_EXIT(zfsvfs); 2904 return (EINVAL); 2905 } 2906 2907 if (mask & AT_SIZE && vp->v_type == VDIR) { 2908 ZFS_EXIT(zfsvfs); 2909 return (EISDIR); 2910 } 2911 2912 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 2913 ZFS_EXIT(zfsvfs); 2914 return (EINVAL); 2915 } 2916 2917 /* 2918 * If this is an xvattr_t, then get a pointer to the structure of 2919 * optional attributes. If this is NULL, then we have a vattr_t. 2920 */ 2921 xoap = xva_getxoptattr(xvap); 2922 2923 xva_init(&tmpxvattr); 2924 2925 /* 2926 * Immutable files can only alter immutable bit and atime 2927 */ 2928 if ((zp->z_pflags & ZFS_IMMUTABLE) && 2929 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2930 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 2931 ZFS_EXIT(zfsvfs); 2932 return (EPERM); 2933 } 2934 2935 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 2936 ZFS_EXIT(zfsvfs); 2937 return (EPERM); 2938 } 2939 2940 /* 2941 * Verify timestamps doesn't overflow 32 bits. 2942 * ZFS can handle large timestamps, but 32bit syscalls can't 2943 * handle times greater than 2039. This check should be removed 2944 * once large timestamps are fully supported. 2945 */ 2946 if (mask & (AT_ATIME | AT_MTIME)) { 2947 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 2948 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 2949 ZFS_EXIT(zfsvfs); 2950 return (EOVERFLOW); 2951 } 2952 } 2953 2954top: 2955 attrzp = NULL; 2956 aclp = NULL; 2957 2958 /* Can this be moved to before the top label? */ 2959 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2960 ZFS_EXIT(zfsvfs); 2961 return (EROFS); 2962 } 2963 2964 /* 2965 * First validate permissions 2966 */ 2967 2968 if (mask & AT_SIZE) { 2969 /* 2970 * XXX - Note, we are not providing any open 2971 * mode flags here (like FNDELAY), so we may 2972 * block if there are locks present... this 2973 * should be addressed in openat(). 2974 */ 2975 /* XXX - would it be OK to generate a log record here? */ 2976 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2977 if (err) { 2978 ZFS_EXIT(zfsvfs); 2979 return (err); 2980 } 2981 } 2982 2983 if (mask & (AT_ATIME|AT_MTIME) || 2984 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2985 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2986 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2987 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 2988 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 2989 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2990 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 2991 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2992 skipaclchk, cr); 2993 } 2994 2995 if (mask & (AT_UID|AT_GID)) { 2996 int idmask = (mask & (AT_UID|AT_GID)); 2997 int take_owner; 2998 int take_group; 2999 3000 /* 3001 * NOTE: even if a new mode is being set, 3002 * we may clear S_ISUID/S_ISGID bits. 3003 */ 3004 3005 if (!(mask & AT_MODE)) 3006 vap->va_mode = zp->z_mode; 3007 3008 /* 3009 * Take ownership or chgrp to group we are a member of 3010 */ 3011 3012 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 3013 take_group = (mask & AT_GID) && 3014 zfs_groupmember(zfsvfs, vap->va_gid, cr); 3015 3016 /* 3017 * If both AT_UID and AT_GID are set then take_owner and 3018 * take_group must both be set in order to allow taking 3019 * ownership. 3020 * 3021 * Otherwise, send the check through secpolicy_vnode_setattr() 3022 * 3023 */ 3024 3025 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 3026 ((idmask == AT_UID) && take_owner) || 3027 ((idmask == AT_GID) && take_group)) { 3028 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 3029 skipaclchk, cr) == 0) { 3030 /* 3031 * Remove setuid/setgid for non-privileged users 3032 */ 3033 secpolicy_setid_clear(vap, vp, cr); 3034 trim_mask = (mask & (AT_UID|AT_GID)); 3035 } else { 3036 need_policy = TRUE; 3037 } 3038 } else { 3039 need_policy = TRUE; 3040 } 3041 } 3042 3043 mutex_enter(&zp->z_lock); 3044 oldva.va_mode = zp->z_mode; 3045 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 3046 if (mask & AT_XVATTR) { 3047 /* 3048 * Update xvattr mask to include only those attributes 3049 * that are actually changing. 3050 * 3051 * the bits will be restored prior to actually setting 3052 * the attributes so the caller thinks they were set. 3053 */ 3054 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 3055 if (xoap->xoa_appendonly != 3056 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 3057 need_policy = TRUE; 3058 } else { 3059 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 3060 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 3061 } 3062 } 3063 3064 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 3065 if (xoap->xoa_nounlink != 3066 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 3067 need_policy = TRUE; 3068 } else { 3069 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 3070 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 3071 } 3072 } 3073 3074 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 3075 if (xoap->xoa_immutable != 3076 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 3077 need_policy = TRUE; 3078 } else { 3079 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 3080 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 3081 } 3082 } 3083 3084 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 3085 if (xoap->xoa_nodump != 3086 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 3087 need_policy = TRUE; 3088 } else { 3089 XVA_CLR_REQ(xvap, XAT_NODUMP); 3090 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 3091 } 3092 } 3093 3094 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 3095 if (xoap->xoa_av_modified != 3096 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 3097 need_policy = TRUE; 3098 } else { 3099 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 3100 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 3101 } 3102 } 3103 3104 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 3105 if ((vp->v_type != VREG && 3106 xoap->xoa_av_quarantined) || 3107 xoap->xoa_av_quarantined != 3108 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 3109 need_policy = TRUE; 3110 } else { 3111 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 3112 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 3113 } 3114 } 3115 3116 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 3117 mutex_exit(&zp->z_lock); 3118 ZFS_EXIT(zfsvfs); 3119 return (EPERM); 3120 } 3121 3122 if (need_policy == FALSE && 3123 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 3124 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 3125 need_policy = TRUE; 3126 } 3127 } 3128 3129 mutex_exit(&zp->z_lock); 3130 3131 if (mask & AT_MODE) { 3132 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 3133 err = secpolicy_setid_setsticky_clear(vp, vap, 3134 &oldva, cr); 3135 if (err) { 3136 ZFS_EXIT(zfsvfs); 3137 return (err); 3138 } 3139 trim_mask |= AT_MODE; 3140 } else { 3141 need_policy = TRUE; 3142 } 3143 } 3144 3145 if (need_policy) { 3146 /* 3147 * If trim_mask is set then take ownership 3148 * has been granted or write_acl is present and user 3149 * has the ability to modify mode. In that case remove 3150 * UID|GID and or MODE from mask so that 3151 * secpolicy_vnode_setattr() doesn't revoke it. 3152 */ 3153 3154 if (trim_mask) { 3155 saved_mask = vap->va_mask; 3156 vap->va_mask &= ~trim_mask; 3157 if (trim_mask & AT_MODE) { 3158 /* 3159 * Save the mode, as secpolicy_vnode_setattr() 3160 * will overwrite it with ova.va_mode. 3161 */ 3162 saved_mode = vap->va_mode; 3163 } 3164 } 3165 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 3166 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 3167 if (err) { 3168 ZFS_EXIT(zfsvfs); 3169 return (err); 3170 } 3171 3172 if (trim_mask) { 3173 vap->va_mask |= saved_mask; 3174 if (trim_mask & AT_MODE) { 3175 /* 3176 * Recover the mode after 3177 * secpolicy_vnode_setattr(). 3178 */ 3179 vap->va_mode = saved_mode; 3180 } 3181 } 3182 } 3183 3184 /* 3185 * secpolicy_vnode_setattr, or take ownership may have 3186 * changed va_mask 3187 */ 3188 mask = vap->va_mask; 3189 3190 if ((mask & (AT_UID | AT_GID))) { 3191 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 3192 &xattr_obj, sizeof (xattr_obj)); 3193 3194 if (err == 0 && xattr_obj) { 3195 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 3196 if (err) 3197 goto out2; 3198 } 3199 if (mask & AT_UID) { 3200 new_uid = zfs_fuid_create(zfsvfs, 3201 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 3202 if (new_uid != zp->z_uid && 3203 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 3204 if (attrzp) 3205 VN_RELE(ZTOV(attrzp)); 3206 err = EDQUOT; 3207 goto out2; 3208 } 3209 } 3210 3211 if (mask & AT_GID) { 3212 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 3213 cr, ZFS_GROUP, &fuidp); 3214 if (new_gid != zp->z_gid && 3215 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 3216 if (attrzp) 3217 VN_RELE(ZTOV(attrzp)); 3218 err = EDQUOT; 3219 goto out2; 3220 } 3221 } 3222 } 3223 tx = dmu_tx_create(zfsvfs->z_os); 3224 3225 if (mask & AT_MODE) { 3226 uint64_t pmode = zp->z_mode; 3227 uint64_t acl_obj; 3228 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 3229 3230 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 3231 goto out; 3232 3233 mutex_enter(&zp->z_lock); 3234 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 3235 /* 3236 * Are we upgrading ACL from old V0 format 3237 * to V1 format? 3238 */ 3239 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 3240 zfs_znode_acl_version(zp) == 3241 ZFS_ACL_VERSION_INITIAL) { 3242 dmu_tx_hold_free(tx, acl_obj, 0, 3243 DMU_OBJECT_END); 3244 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3245 0, aclp->z_acl_bytes); 3246 } else { 3247 dmu_tx_hold_write(tx, acl_obj, 0, 3248 aclp->z_acl_bytes); 3249 } 3250 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3251 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3252 0, aclp->z_acl_bytes); 3253 } 3254 mutex_exit(&zp->z_lock); 3255 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3256 } else { 3257 if ((mask & AT_XVATTR) && 3258 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3259 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3260 else 3261 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3262 } 3263 3264 if (attrzp) { 3265 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 3266 } 3267 3268 fuid_dirtied = zfsvfs->z_fuid_dirty; 3269 if (fuid_dirtied) 3270 zfs_fuid_txhold(zfsvfs, tx); 3271 3272 zfs_sa_upgrade_txholds(tx, zp); 3273 3274 err = dmu_tx_assign(tx, TXG_NOWAIT); 3275 if (err) { 3276 if (err == ERESTART) 3277 dmu_tx_wait(tx); 3278 goto out; 3279 } 3280 3281 count = 0; 3282 /* 3283 * Set each attribute requested. 3284 * We group settings according to the locks they need to acquire. 3285 * 3286 * Note: you cannot set ctime directly, although it will be 3287 * updated as a side-effect of calling this function. 3288 */ 3289 3290 3291 if (mask & (AT_UID|AT_GID|AT_MODE)) 3292 mutex_enter(&zp->z_acl_lock); 3293 mutex_enter(&zp->z_lock); 3294 3295 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3296 &zp->z_pflags, sizeof (zp->z_pflags)); 3297 3298 if (attrzp) { 3299 if (mask & (AT_UID|AT_GID|AT_MODE)) 3300 mutex_enter(&attrzp->z_acl_lock); 3301 mutex_enter(&attrzp->z_lock); 3302 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3303 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3304 sizeof (attrzp->z_pflags)); 3305 } 3306 3307 if (mask & (AT_UID|AT_GID)) { 3308 3309 if (mask & AT_UID) { 3310 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3311 &new_uid, sizeof (new_uid)); 3312 zp->z_uid = new_uid; 3313 if (attrzp) { 3314 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3315 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3316 sizeof (new_uid)); 3317 attrzp->z_uid = new_uid; 3318 } 3319 } 3320 3321 if (mask & AT_GID) { 3322 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3323 NULL, &new_gid, sizeof (new_gid)); 3324 zp->z_gid = new_gid; 3325 if (attrzp) { 3326 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3327 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 3328 sizeof (new_gid)); 3329 attrzp->z_gid = new_gid; 3330 } 3331 } 3332 if (!(mask & AT_MODE)) { 3333 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 3334 NULL, &new_mode, sizeof (new_mode)); 3335 new_mode = zp->z_mode; 3336 } 3337 err = zfs_acl_chown_setattr(zp); 3338 ASSERT(err == 0); 3339 if (attrzp) { 3340 err = zfs_acl_chown_setattr(attrzp); 3341 ASSERT(err == 0); 3342 } 3343 } 3344 3345 if (mask & AT_MODE) { 3346 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 3347 &new_mode, sizeof (new_mode)); 3348 zp->z_mode = new_mode; 3349 ASSERT3U((uintptr_t)aclp, !=, 0); 3350 err = zfs_aclset_common(zp, aclp, cr, tx); 3351 ASSERT3U(err, ==, 0); 3352 if (zp->z_acl_cached) 3353 zfs_acl_free(zp->z_acl_cached); 3354 zp->z_acl_cached = aclp; 3355 aclp = NULL; 3356 } 3357 3358 3359 if (mask & AT_ATIME) { 3360 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 3361 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 3362 &zp->z_atime, sizeof (zp->z_atime)); 3363 } 3364 3365 if (mask & AT_MTIME) { 3366 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 3367 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 3368 mtime, sizeof (mtime)); 3369 } 3370 3371 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 3372 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 3373 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 3374 NULL, mtime, sizeof (mtime)); 3375 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3376 &ctime, sizeof (ctime)); 3377 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 3378 B_TRUE); 3379 } else if (mask != 0) { 3380 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3381 &ctime, sizeof (ctime)); 3382 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime, 3383 B_TRUE); 3384 if (attrzp) { 3385 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3386 SA_ZPL_CTIME(zfsvfs), NULL, 3387 &ctime, sizeof (ctime)); 3388 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 3389 mtime, ctime, B_TRUE); 3390 } 3391 } 3392 /* 3393 * Do this after setting timestamps to prevent timestamp 3394 * update from toggling bit 3395 */ 3396 3397 if (xoap && (mask & AT_XVATTR)) { 3398 3399 /* 3400 * restore trimmed off masks 3401 * so that return masks can be set for caller. 3402 */ 3403 3404 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 3405 XVA_SET_REQ(xvap, XAT_APPENDONLY); 3406 } 3407 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 3408 XVA_SET_REQ(xvap, XAT_NOUNLINK); 3409 } 3410 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 3411 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 3412 } 3413 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 3414 XVA_SET_REQ(xvap, XAT_NODUMP); 3415 } 3416 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 3417 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 3418 } 3419 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 3420 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 3421 } 3422 3423 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3424 ASSERT(vp->v_type == VREG); 3425 3426 zfs_xvattr_set(zp, xvap, tx); 3427 } 3428 3429 if (fuid_dirtied) 3430 zfs_fuid_sync(zfsvfs, tx); 3431 3432 if (mask != 0) 3433 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 3434 3435 mutex_exit(&zp->z_lock); 3436 if (mask & (AT_UID|AT_GID|AT_MODE)) 3437 mutex_exit(&zp->z_acl_lock); 3438 3439 if (attrzp) { 3440 if (mask & (AT_UID|AT_GID|AT_MODE)) 3441 mutex_exit(&attrzp->z_acl_lock); 3442 mutex_exit(&attrzp->z_lock); 3443 } 3444out: 3445 if (err == 0 && attrzp) { 3446 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 3447 xattr_count, tx); 3448 ASSERT(err2 == 0); 3449 } 3450 3451 if (attrzp) 3452 VN_RELE(ZTOV(attrzp)); 3453 if (aclp) 3454 zfs_acl_free(aclp); 3455 3456 if (fuidp) { 3457 zfs_fuid_info_free(fuidp); 3458 fuidp = NULL; 3459 } 3460 3461 if (err) { 3462 dmu_tx_abort(tx); 3463 if (err == ERESTART) 3464 goto top; 3465 } else { 3466 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3467 dmu_tx_commit(tx); 3468 } 3469 3470out2: 3471 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3472 zil_commit(zilog, 0); 3473 3474 ZFS_EXIT(zfsvfs); 3475 return (err); 3476} 3477 3478typedef struct zfs_zlock { 3479 krwlock_t *zl_rwlock; /* lock we acquired */ 3480 znode_t *zl_znode; /* znode we held */ 3481 struct zfs_zlock *zl_next; /* next in list */ 3482} zfs_zlock_t; 3483 3484/* 3485 * Drop locks and release vnodes that were held by zfs_rename_lock(). 3486 */ 3487static void 3488zfs_rename_unlock(zfs_zlock_t **zlpp) 3489{ 3490 zfs_zlock_t *zl; 3491 3492 while ((zl = *zlpp) != NULL) { 3493 if (zl->zl_znode != NULL) 3494 VN_RELE(ZTOV(zl->zl_znode)); 3495 rw_exit(zl->zl_rwlock); 3496 *zlpp = zl->zl_next; 3497 kmem_free(zl, sizeof (*zl)); 3498 } 3499} 3500 3501/* 3502 * Search back through the directory tree, using the ".." entries. 3503 * Lock each directory in the chain to prevent concurrent renames. 3504 * Fail any attempt to move a directory into one of its own descendants. 3505 * XXX - z_parent_lock can overlap with map or grow locks 3506 */ 3507static int 3508zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 3509{ 3510 zfs_zlock_t *zl; 3511 znode_t *zp = tdzp; 3512 uint64_t rootid = zp->z_zfsvfs->z_root; 3513 uint64_t oidp = zp->z_id; 3514 krwlock_t *rwlp = &szp->z_parent_lock; 3515 krw_t rw = RW_WRITER; 3516 3517 /* 3518 * First pass write-locks szp and compares to zp->z_id. 3519 * Later passes read-lock zp and compare to zp->z_parent. 3520 */ 3521 do { 3522 if (!rw_tryenter(rwlp, rw)) { 3523 /* 3524 * Another thread is renaming in this path. 3525 * Note that if we are a WRITER, we don't have any 3526 * parent_locks held yet. 3527 */ 3528 if (rw == RW_READER && zp->z_id > szp->z_id) { 3529 /* 3530 * Drop our locks and restart 3531 */ 3532 zfs_rename_unlock(&zl); 3533 *zlpp = NULL; 3534 zp = tdzp; 3535 oidp = zp->z_id; 3536 rwlp = &szp->z_parent_lock; 3537 rw = RW_WRITER; 3538 continue; 3539 } else { 3540 /* 3541 * Wait for other thread to drop its locks 3542 */ 3543 rw_enter(rwlp, rw); 3544 } 3545 } 3546 3547 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 3548 zl->zl_rwlock = rwlp; 3549 zl->zl_znode = NULL; 3550 zl->zl_next = *zlpp; 3551 *zlpp = zl; 3552 3553 if (oidp == szp->z_id) /* We're a descendant of szp */ 3554 return (EINVAL); 3555 3556 if (oidp == rootid) /* We've hit the top */ 3557 return (0); 3558 3559 if (rw == RW_READER) { /* i.e. not the first pass */ 3560 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp); 3561 if (error) 3562 return (error); 3563 zl->zl_znode = zp; 3564 } 3565 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs), 3566 &oidp, sizeof (oidp)); 3567 rwlp = &zp->z_parent_lock; 3568 rw = RW_READER; 3569 3570 } while (zp->z_id != sdzp->z_id); 3571 3572 return (0); 3573} 3574 3575/* 3576 * Move an entry from the provided source directory to the target 3577 * directory. Change the entry name as indicated. 3578 * 3579 * IN: sdvp - Source directory containing the "old entry". 3580 * snm - Old entry name. 3581 * tdvp - Target directory to contain the "new entry". 3582 * tnm - New entry name. 3583 * cr - credentials of caller. 3584 * ct - caller context 3585 * flags - case flags 3586 * 3587 * RETURN: 0 if success 3588 * error code if failure 3589 * 3590 * Timestamps: 3591 * sdvp,tdvp - ctime|mtime updated 3592 */ 3593/*ARGSUSED*/ 3594static int 3595zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 3596 caller_context_t *ct, int flags) 3597{ 3598 znode_t *tdzp, *szp, *tzp; 3599 znode_t *sdzp = VTOZ(sdvp); 3600 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 3601 zilog_t *zilog; 3602 vnode_t *realvp; 3603 zfs_dirlock_t *sdl, *tdl; 3604 dmu_tx_t *tx; 3605 zfs_zlock_t *zl; 3606 int cmp, serr, terr; 3607 int error = 0; 3608 int zflg = 0; 3609 3610 ZFS_ENTER(zfsvfs); 3611 ZFS_VERIFY_ZP(sdzp); 3612 zilog = zfsvfs->z_log; 3613 3614 /* 3615 * Make sure we have the real vp for the target directory. 3616 */ 3617 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 3618 tdvp = realvp; 3619 3620 if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) { 3621 ZFS_EXIT(zfsvfs); 3622 return (EXDEV); 3623 } 3624 3625 tdzp = VTOZ(tdvp); 3626 ZFS_VERIFY_ZP(tdzp); 3627 if (zfsvfs->z_utf8 && u8_validate(tnm, 3628 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3629 ZFS_EXIT(zfsvfs); 3630 return (EILSEQ); 3631 } 3632 3633 if (flags & FIGNORECASE) 3634 zflg |= ZCILOOK; 3635 3636top: 3637 szp = NULL; 3638 tzp = NULL; 3639 zl = NULL; 3640 3641 /* 3642 * This is to prevent the creation of links into attribute space 3643 * by renaming a linked file into/outof an attribute directory. 3644 * See the comment in zfs_link() for why this is considered bad. 3645 */ 3646 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3647 ZFS_EXIT(zfsvfs); 3648 return (EINVAL); 3649 } 3650 3651 /* 3652 * Lock source and target directory entries. To prevent deadlock, 3653 * a lock ordering must be defined. We lock the directory with 3654 * the smallest object id first, or if it's a tie, the one with 3655 * the lexically first name. 3656 */ 3657 if (sdzp->z_id < tdzp->z_id) { 3658 cmp = -1; 3659 } else if (sdzp->z_id > tdzp->z_id) { 3660 cmp = 1; 3661 } else { 3662 /* 3663 * First compare the two name arguments without 3664 * considering any case folding. 3665 */ 3666 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 3667 3668 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 3669 ASSERT(error == 0 || !zfsvfs->z_utf8); 3670 if (cmp == 0) { 3671 /* 3672 * POSIX: "If the old argument and the new argument 3673 * both refer to links to the same existing file, 3674 * the rename() function shall return successfully 3675 * and perform no other action." 3676 */ 3677 ZFS_EXIT(zfsvfs); 3678 return (0); 3679 } 3680 /* 3681 * If the file system is case-folding, then we may 3682 * have some more checking to do. A case-folding file 3683 * system is either supporting mixed case sensitivity 3684 * access or is completely case-insensitive. Note 3685 * that the file system is always case preserving. 3686 * 3687 * In mixed sensitivity mode case sensitive behavior 3688 * is the default. FIGNORECASE must be used to 3689 * explicitly request case insensitive behavior. 3690 * 3691 * If the source and target names provided differ only 3692 * by case (e.g., a request to rename 'tim' to 'Tim'), 3693 * we will treat this as a special case in the 3694 * case-insensitive mode: as long as the source name 3695 * is an exact match, we will allow this to proceed as 3696 * a name-change request. 3697 */ 3698 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3699 (zfsvfs->z_case == ZFS_CASE_MIXED && 3700 flags & FIGNORECASE)) && 3701 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3702 &error) == 0) { 3703 /* 3704 * case preserving rename request, require exact 3705 * name matches 3706 */ 3707 zflg |= ZCIEXACT; 3708 zflg &= ~ZCILOOK; 3709 } 3710 } 3711 3712 /* 3713 * If the source and destination directories are the same, we should 3714 * grab the z_name_lock of that directory only once. 3715 */ 3716 if (sdzp == tdzp) { 3717 zflg |= ZHAVELOCK; 3718 rw_enter(&sdzp->z_name_lock, RW_READER); 3719 } 3720 3721 if (cmp < 0) { 3722 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3723 ZEXISTS | zflg, NULL, NULL); 3724 terr = zfs_dirent_lock(&tdl, 3725 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3726 } else { 3727 terr = zfs_dirent_lock(&tdl, 3728 tdzp, tnm, &tzp, zflg, NULL, NULL); 3729 serr = zfs_dirent_lock(&sdl, 3730 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3731 NULL, NULL); 3732 } 3733 3734 if (serr) { 3735 /* 3736 * Source entry invalid or not there. 3737 */ 3738 if (!terr) { 3739 zfs_dirent_unlock(tdl); 3740 if (tzp) 3741 VN_RELE(ZTOV(tzp)); 3742 } 3743 3744 if (sdzp == tdzp) 3745 rw_exit(&sdzp->z_name_lock); 3746 3747 /* 3748 * FreeBSD: In OpenSolaris they only check if rename source is 3749 * ".." here, because "." is handled in their lookup. This is 3750 * not the case for FreeBSD, so we check for "." explicitly. 3751 */ 3752 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0) 3753 serr = EINVAL; 3754 ZFS_EXIT(zfsvfs); 3755 return (serr); 3756 } 3757 if (terr) { 3758 zfs_dirent_unlock(sdl); 3759 VN_RELE(ZTOV(szp)); 3760 3761 if (sdzp == tdzp) 3762 rw_exit(&sdzp->z_name_lock); 3763 3764 if (strcmp(tnm, "..") == 0) 3765 terr = EINVAL; 3766 ZFS_EXIT(zfsvfs); 3767 return (terr); 3768 } 3769 3770 /* 3771 * Must have write access at the source to remove the old entry 3772 * and write access at the target to create the new entry. 3773 * Note that if target and source are the same, this can be 3774 * done in a single check. 3775 */ 3776 3777 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3778 goto out; 3779 3780 if (ZTOV(szp)->v_type == VDIR) { 3781 /* 3782 * Check to make sure rename is valid. 3783 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3784 */ 3785 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3786 goto out; 3787 } 3788 3789 /* 3790 * Does target exist? 3791 */ 3792 if (tzp) { 3793 /* 3794 * Source and target must be the same type. 3795 */ 3796 if (ZTOV(szp)->v_type == VDIR) { 3797 if (ZTOV(tzp)->v_type != VDIR) { 3798 error = ENOTDIR; 3799 goto out; 3800 } 3801 } else { 3802 if (ZTOV(tzp)->v_type == VDIR) { 3803 error = EISDIR; 3804 goto out; 3805 } 3806 } 3807 /* 3808 * POSIX dictates that when the source and target 3809 * entries refer to the same file object, rename 3810 * must do nothing and exit without error. 3811 */ 3812 if (szp->z_id == tzp->z_id) { 3813 error = 0; 3814 goto out; 3815 } 3816 } 3817 3818 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3819 if (tzp) 3820 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3821 3822 /* 3823 * notify the target directory if it is not the same 3824 * as source directory. 3825 */ 3826 if (tdvp != sdvp) { 3827 vnevent_rename_dest_dir(tdvp, ct); 3828 } 3829 3830 tx = dmu_tx_create(zfsvfs->z_os); 3831 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3832 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 3833 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3834 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3835 if (sdzp != tdzp) { 3836 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 3837 zfs_sa_upgrade_txholds(tx, tdzp); 3838 } 3839 if (tzp) { 3840 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 3841 zfs_sa_upgrade_txholds(tx, tzp); 3842 } 3843 3844 zfs_sa_upgrade_txholds(tx, szp); 3845 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3846 error = dmu_tx_assign(tx, TXG_NOWAIT); 3847 if (error) { 3848 if (zl != NULL) 3849 zfs_rename_unlock(&zl); 3850 zfs_dirent_unlock(sdl); 3851 zfs_dirent_unlock(tdl); 3852 3853 if (sdzp == tdzp) 3854 rw_exit(&sdzp->z_name_lock); 3855 3856 VN_RELE(ZTOV(szp)); 3857 if (tzp) 3858 VN_RELE(ZTOV(tzp)); 3859 if (error == ERESTART) { 3860 dmu_tx_wait(tx); 3861 dmu_tx_abort(tx); 3862 goto top; 3863 } 3864 dmu_tx_abort(tx); 3865 ZFS_EXIT(zfsvfs); 3866 return (error); 3867 } 3868 3869 if (tzp) /* Attempt to remove the existing target */ 3870 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 3871 3872 if (error == 0) { 3873 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 3874 if (error == 0) { 3875 szp->z_pflags |= ZFS_AV_MODIFIED; 3876 3877 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 3878 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 3879 ASSERT3U(error, ==, 0); 3880 3881 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 3882 if (error == 0) { 3883 zfs_log_rename(zilog, tx, TX_RENAME | 3884 (flags & FIGNORECASE ? TX_CI : 0), sdzp, 3885 sdl->dl_name, tdzp, tdl->dl_name, szp); 3886 3887 /* 3888 * Update path information for the target vnode 3889 */ 3890 vn_renamepath(tdvp, ZTOV(szp), tnm, 3891 strlen(tnm)); 3892 } else { 3893 /* 3894 * At this point, we have successfully created 3895 * the target name, but have failed to remove 3896 * the source name. Since the create was done 3897 * with the ZRENAMING flag, there are 3898 * complications; for one, the link count is 3899 * wrong. The easiest way to deal with this 3900 * is to remove the newly created target, and 3901 * return the original error. This must 3902 * succeed; fortunately, it is very unlikely to 3903 * fail, since we just created it. 3904 */ 3905 VERIFY3U(zfs_link_destroy(tdl, szp, tx, 3906 ZRENAMING, NULL), ==, 0); 3907 } 3908 } 3909#ifdef FREEBSD_NAMECACHE 3910 if (error == 0) { 3911 cache_purge(sdvp); 3912 cache_purge(tdvp); 3913 } 3914#endif 3915 } 3916 3917 dmu_tx_commit(tx); 3918out: 3919 if (zl != NULL) 3920 zfs_rename_unlock(&zl); 3921 3922 zfs_dirent_unlock(sdl); 3923 zfs_dirent_unlock(tdl); 3924 3925 if (sdzp == tdzp) 3926 rw_exit(&sdzp->z_name_lock); 3927 3928 3929 VN_RELE(ZTOV(szp)); 3930 if (tzp) 3931 VN_RELE(ZTOV(tzp)); 3932 3933 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3934 zil_commit(zilog, 0); 3935 3936 ZFS_EXIT(zfsvfs); 3937 3938 return (error); 3939} 3940 3941/* 3942 * Insert the indicated symbolic reference entry into the directory. 3943 * 3944 * IN: dvp - Directory to contain new symbolic link. 3945 * link - Name for new symlink entry. 3946 * vap - Attributes of new entry. 3947 * target - Target path of new symlink. 3948 * cr - credentials of caller. 3949 * ct - caller context 3950 * flags - case flags 3951 * 3952 * RETURN: 0 if success 3953 * error code if failure 3954 * 3955 * Timestamps: 3956 * dvp - ctime|mtime updated 3957 */ 3958/*ARGSUSED*/ 3959static int 3960zfs_symlink(vnode_t *dvp, vnode_t **vpp, char *name, vattr_t *vap, char *link, 3961 cred_t *cr, kthread_t *td) 3962{ 3963 znode_t *zp, *dzp = VTOZ(dvp); 3964 zfs_dirlock_t *dl; 3965 dmu_tx_t *tx; 3966 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3967 zilog_t *zilog; 3968 uint64_t len = strlen(link); 3969 int error; 3970 int zflg = ZNEW; 3971 zfs_acl_ids_t acl_ids; 3972 boolean_t fuid_dirtied; 3973 uint64_t txtype = TX_SYMLINK; 3974 int flags = 0; 3975 3976 ASSERT(vap->va_type == VLNK); 3977 3978 ZFS_ENTER(zfsvfs); 3979 ZFS_VERIFY_ZP(dzp); 3980 zilog = zfsvfs->z_log; 3981 3982 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 3983 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3984 ZFS_EXIT(zfsvfs); 3985 return (EILSEQ); 3986 } 3987 if (flags & FIGNORECASE) 3988 zflg |= ZCILOOK; 3989 3990 if (len > MAXPATHLEN) { 3991 ZFS_EXIT(zfsvfs); 3992 return (ENAMETOOLONG); 3993 } 3994 3995 if ((error = zfs_acl_ids_create(dzp, 0, 3996 vap, cr, NULL, &acl_ids)) != 0) { 3997 ZFS_EXIT(zfsvfs); 3998 return (error); 3999 } 4000top: 4001 /* 4002 * Attempt to lock directory; fail if entry already exists. 4003 */ 4004 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 4005 if (error) { 4006 zfs_acl_ids_free(&acl_ids); 4007 ZFS_EXIT(zfsvfs); 4008 return (error); 4009 } 4010 4011 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4012 zfs_acl_ids_free(&acl_ids); 4013 zfs_dirent_unlock(dl); 4014 ZFS_EXIT(zfsvfs); 4015 return (error); 4016 } 4017 4018 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 4019 zfs_acl_ids_free(&acl_ids); 4020 zfs_dirent_unlock(dl); 4021 ZFS_EXIT(zfsvfs); 4022 return (EDQUOT); 4023 } 4024 tx = dmu_tx_create(zfsvfs->z_os); 4025 fuid_dirtied = zfsvfs->z_fuid_dirty; 4026 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 4027 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4028 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 4029 ZFS_SA_BASE_ATTR_SIZE + len); 4030 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 4031 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 4032 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 4033 acl_ids.z_aclp->z_acl_bytes); 4034 } 4035 if (fuid_dirtied) 4036 zfs_fuid_txhold(zfsvfs, tx); 4037 error = dmu_tx_assign(tx, TXG_NOWAIT); 4038 if (error) { 4039 zfs_dirent_unlock(dl); 4040 if (error == ERESTART) { 4041 dmu_tx_wait(tx); 4042 dmu_tx_abort(tx); 4043 goto top; 4044 } 4045 zfs_acl_ids_free(&acl_ids); 4046 dmu_tx_abort(tx); 4047 ZFS_EXIT(zfsvfs); 4048 return (error); 4049 } 4050 4051 /* 4052 * Create a new object for the symlink. 4053 * for version 4 ZPL datsets the symlink will be an SA attribute 4054 */ 4055 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 4056 4057 if (fuid_dirtied) 4058 zfs_fuid_sync(zfsvfs, tx); 4059 4060 mutex_enter(&zp->z_lock); 4061 if (zp->z_is_sa) 4062 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 4063 link, len, tx); 4064 else 4065 zfs_sa_symlink(zp, link, len, tx); 4066 mutex_exit(&zp->z_lock); 4067 4068 zp->z_size = len; 4069 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 4070 &zp->z_size, sizeof (zp->z_size), tx); 4071 /* 4072 * Insert the new object into the directory. 4073 */ 4074 (void) zfs_link_create(dl, zp, tx, ZNEW); 4075 4076 if (flags & FIGNORECASE) 4077 txtype |= TX_CI; 4078 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 4079 *vpp = ZTOV(zp); 4080 4081 zfs_acl_ids_free(&acl_ids); 4082 4083 dmu_tx_commit(tx); 4084 4085 zfs_dirent_unlock(dl); 4086 4087 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4088 zil_commit(zilog, 0); 4089 4090 ZFS_EXIT(zfsvfs); 4091 return (error); 4092} 4093 4094/* 4095 * Return, in the buffer contained in the provided uio structure, 4096 * the symbolic path referred to by vp. 4097 * 4098 * IN: vp - vnode of symbolic link. 4099 * uoip - structure to contain the link path. 4100 * cr - credentials of caller. 4101 * ct - caller context 4102 * 4103 * OUT: uio - structure to contain the link path. 4104 * 4105 * RETURN: 0 if success 4106 * error code if failure 4107 * 4108 * Timestamps: 4109 * vp - atime updated 4110 */ 4111/* ARGSUSED */ 4112static int 4113zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 4114{ 4115 znode_t *zp = VTOZ(vp); 4116 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4117 int error; 4118 4119 ZFS_ENTER(zfsvfs); 4120 ZFS_VERIFY_ZP(zp); 4121 4122 mutex_enter(&zp->z_lock); 4123 if (zp->z_is_sa) 4124 error = sa_lookup_uio(zp->z_sa_hdl, 4125 SA_ZPL_SYMLINK(zfsvfs), uio); 4126 else 4127 error = zfs_sa_readlink(zp, uio); 4128 mutex_exit(&zp->z_lock); 4129 4130 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4131 4132 ZFS_EXIT(zfsvfs); 4133 return (error); 4134} 4135 4136/* 4137 * Insert a new entry into directory tdvp referencing svp. 4138 * 4139 * IN: tdvp - Directory to contain new entry. 4140 * svp - vnode of new entry. 4141 * name - name of new entry. 4142 * cr - credentials of caller. 4143 * ct - caller context 4144 * 4145 * RETURN: 0 if success 4146 * error code if failure 4147 * 4148 * Timestamps: 4149 * tdvp - ctime|mtime updated 4150 * svp - ctime updated 4151 */ 4152/* ARGSUSED */ 4153static int 4154zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 4155 caller_context_t *ct, int flags) 4156{ 4157 znode_t *dzp = VTOZ(tdvp); 4158 znode_t *tzp, *szp; 4159 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4160 zilog_t *zilog; 4161 zfs_dirlock_t *dl; 4162 dmu_tx_t *tx; 4163 vnode_t *realvp; 4164 int error; 4165 int zf = ZNEW; 4166 uint64_t parent; 4167 uid_t owner; 4168 4169 ASSERT(tdvp->v_type == VDIR); 4170 4171 ZFS_ENTER(zfsvfs); 4172 ZFS_VERIFY_ZP(dzp); 4173 zilog = zfsvfs->z_log; 4174 4175 if (VOP_REALVP(svp, &realvp, ct) == 0) 4176 svp = realvp; 4177 4178 /* 4179 * POSIX dictates that we return EPERM here. 4180 * Better choices include ENOTSUP or EISDIR. 4181 */ 4182 if (svp->v_type == VDIR) { 4183 ZFS_EXIT(zfsvfs); 4184 return (EPERM); 4185 } 4186 4187 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) { 4188 ZFS_EXIT(zfsvfs); 4189 return (EXDEV); 4190 } 4191 4192 szp = VTOZ(svp); 4193 ZFS_VERIFY_ZP(szp); 4194 4195 /* Prevent links to .zfs/shares files */ 4196 4197 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 4198 &parent, sizeof (uint64_t))) != 0) { 4199 ZFS_EXIT(zfsvfs); 4200 return (error); 4201 } 4202 if (parent == zfsvfs->z_shares_dir) { 4203 ZFS_EXIT(zfsvfs); 4204 return (EPERM); 4205 } 4206 4207 if (zfsvfs->z_utf8 && u8_validate(name, 4208 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4209 ZFS_EXIT(zfsvfs); 4210 return (EILSEQ); 4211 } 4212 if (flags & FIGNORECASE) 4213 zf |= ZCILOOK; 4214 4215 /* 4216 * We do not support links between attributes and non-attributes 4217 * because of the potential security risk of creating links 4218 * into "normal" file space in order to circumvent restrictions 4219 * imposed in attribute space. 4220 */ 4221 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 4222 ZFS_EXIT(zfsvfs); 4223 return (EINVAL); 4224 } 4225 4226 4227 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 4228 if (owner != crgetuid(cr) && secpolicy_basic_link(svp, cr) != 0) { 4229 ZFS_EXIT(zfsvfs); 4230 return (EPERM); 4231 } 4232 4233 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4234 ZFS_EXIT(zfsvfs); 4235 return (error); 4236 } 4237 4238top: 4239 /* 4240 * Attempt to lock directory; fail if entry already exists. 4241 */ 4242 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 4243 if (error) { 4244 ZFS_EXIT(zfsvfs); 4245 return (error); 4246 } 4247 4248 tx = dmu_tx_create(zfsvfs->z_os); 4249 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4250 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4251 zfs_sa_upgrade_txholds(tx, szp); 4252 zfs_sa_upgrade_txholds(tx, dzp); 4253 error = dmu_tx_assign(tx, TXG_NOWAIT); 4254 if (error) { 4255 zfs_dirent_unlock(dl); 4256 if (error == ERESTART) { 4257 dmu_tx_wait(tx); 4258 dmu_tx_abort(tx); 4259 goto top; 4260 } 4261 dmu_tx_abort(tx); 4262 ZFS_EXIT(zfsvfs); 4263 return (error); 4264 } 4265 4266 error = zfs_link_create(dl, szp, tx, 0); 4267 4268 if (error == 0) { 4269 uint64_t txtype = TX_LINK; 4270 if (flags & FIGNORECASE) 4271 txtype |= TX_CI; 4272 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 4273 } 4274 4275 dmu_tx_commit(tx); 4276 4277 zfs_dirent_unlock(dl); 4278 4279 if (error == 0) { 4280 vnevent_link(svp, ct); 4281 } 4282 4283 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4284 zil_commit(zilog, 0); 4285 4286 ZFS_EXIT(zfsvfs); 4287 return (error); 4288} 4289 4290#ifdef sun 4291/* 4292 * zfs_null_putapage() is used when the file system has been force 4293 * unmounted. It just drops the pages. 4294 */ 4295/* ARGSUSED */ 4296static int 4297zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4298 size_t *lenp, int flags, cred_t *cr) 4299{ 4300 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 4301 return (0); 4302} 4303 4304/* 4305 * Push a page out to disk, klustering if possible. 4306 * 4307 * IN: vp - file to push page to. 4308 * pp - page to push. 4309 * flags - additional flags. 4310 * cr - credentials of caller. 4311 * 4312 * OUT: offp - start of range pushed. 4313 * lenp - len of range pushed. 4314 * 4315 * RETURN: 0 if success 4316 * error code if failure 4317 * 4318 * NOTE: callers must have locked the page to be pushed. On 4319 * exit, the page (and all other pages in the kluster) must be 4320 * unlocked. 4321 */ 4322/* ARGSUSED */ 4323static int 4324zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4325 size_t *lenp, int flags, cred_t *cr) 4326{ 4327 znode_t *zp = VTOZ(vp); 4328 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4329 dmu_tx_t *tx; 4330 u_offset_t off, koff; 4331 size_t len, klen; 4332 int err; 4333 4334 off = pp->p_offset; 4335 len = PAGESIZE; 4336 /* 4337 * If our blocksize is bigger than the page size, try to kluster 4338 * multiple pages so that we write a full block (thus avoiding 4339 * a read-modify-write). 4340 */ 4341 if (off < zp->z_size && zp->z_blksz > PAGESIZE) { 4342 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 4343 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 4344 ASSERT(koff <= zp->z_size); 4345 if (koff + klen > zp->z_size) 4346 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); 4347 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 4348 } 4349 ASSERT3U(btop(len), ==, btopr(len)); 4350 4351 /* 4352 * Can't push pages past end-of-file. 4353 */ 4354 if (off >= zp->z_size) { 4355 /* ignore all pages */ 4356 err = 0; 4357 goto out; 4358 } else if (off + len > zp->z_size) { 4359 int npages = btopr(zp->z_size - off); 4360 page_t *trunc; 4361 4362 page_list_break(&pp, &trunc, npages); 4363 /* ignore pages past end of file */ 4364 if (trunc) 4365 pvn_write_done(trunc, flags); 4366 len = zp->z_size - off; 4367 } 4368 4369 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4370 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4371 err = EDQUOT; 4372 goto out; 4373 } 4374top: 4375 tx = dmu_tx_create(zfsvfs->z_os); 4376 dmu_tx_hold_write(tx, zp->z_id, off, len); 4377 4378 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4379 zfs_sa_upgrade_txholds(tx, zp); 4380 err = dmu_tx_assign(tx, TXG_NOWAIT); 4381 if (err != 0) { 4382 if (err == ERESTART) { 4383 dmu_tx_wait(tx); 4384 dmu_tx_abort(tx); 4385 goto top; 4386 } 4387 dmu_tx_abort(tx); 4388 goto out; 4389 } 4390 4391 if (zp->z_blksz <= PAGESIZE) { 4392 caddr_t va = zfs_map_page(pp, S_READ); 4393 ASSERT3U(len, <=, PAGESIZE); 4394 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 4395 zfs_unmap_page(pp, va); 4396 } else { 4397 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 4398 } 4399 4400 if (err == 0) { 4401 uint64_t mtime[2], ctime[2]; 4402 sa_bulk_attr_t bulk[3]; 4403 int count = 0; 4404 4405 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4406 &mtime, 16); 4407 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4408 &ctime, 16); 4409 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4410 &zp->z_pflags, 8); 4411 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4412 B_TRUE); 4413 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4414 } 4415 dmu_tx_commit(tx); 4416 4417out: 4418 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 4419 if (offp) 4420 *offp = off; 4421 if (lenp) 4422 *lenp = len; 4423 4424 return (err); 4425} 4426 4427/* 4428 * Copy the portion of the file indicated from pages into the file. 4429 * The pages are stored in a page list attached to the files vnode. 4430 * 4431 * IN: vp - vnode of file to push page data to. 4432 * off - position in file to put data. 4433 * len - amount of data to write. 4434 * flags - flags to control the operation. 4435 * cr - credentials of caller. 4436 * ct - caller context. 4437 * 4438 * RETURN: 0 if success 4439 * error code if failure 4440 * 4441 * Timestamps: 4442 * vp - ctime|mtime updated 4443 */ 4444/*ARGSUSED*/ 4445static int 4446zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4447 caller_context_t *ct) 4448{ 4449 znode_t *zp = VTOZ(vp); 4450 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4451 page_t *pp; 4452 size_t io_len; 4453 u_offset_t io_off; 4454 uint_t blksz; 4455 rl_t *rl; 4456 int error = 0; 4457 4458 ZFS_ENTER(zfsvfs); 4459 ZFS_VERIFY_ZP(zp); 4460 4461 /* 4462 * Align this request to the file block size in case we kluster. 4463 * XXX - this can result in pretty aggresive locking, which can 4464 * impact simultanious read/write access. One option might be 4465 * to break up long requests (len == 0) into block-by-block 4466 * operations to get narrower locking. 4467 */ 4468 blksz = zp->z_blksz; 4469 if (ISP2(blksz)) 4470 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 4471 else 4472 io_off = 0; 4473 if (len > 0 && ISP2(blksz)) 4474 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 4475 else 4476 io_len = 0; 4477 4478 if (io_len == 0) { 4479 /* 4480 * Search the entire vp list for pages >= io_off. 4481 */ 4482 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 4483 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 4484 goto out; 4485 } 4486 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 4487 4488 if (off > zp->z_size) { 4489 /* past end of file */ 4490 zfs_range_unlock(rl); 4491 ZFS_EXIT(zfsvfs); 4492 return (0); 4493 } 4494 4495 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); 4496 4497 for (off = io_off; io_off < off + len; io_off += io_len) { 4498 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 4499 pp = page_lookup(vp, io_off, 4500 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 4501 } else { 4502 pp = page_lookup_nowait(vp, io_off, 4503 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 4504 } 4505 4506 if (pp != NULL && pvn_getdirty(pp, flags)) { 4507 int err; 4508 4509 /* 4510 * Found a dirty page to push 4511 */ 4512 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 4513 if (err) 4514 error = err; 4515 } else { 4516 io_len = PAGESIZE; 4517 } 4518 } 4519out: 4520 zfs_range_unlock(rl); 4521 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4522 zil_commit(zfsvfs->z_log, zp->z_id); 4523 ZFS_EXIT(zfsvfs); 4524 return (error); 4525} 4526#endif /* sun */ 4527 4528/*ARGSUSED*/ 4529void 4530zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4531{ 4532 znode_t *zp = VTOZ(vp); 4533 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4534 int error; 4535 4536 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4537 if (zp->z_sa_hdl == NULL) { 4538 /* 4539 * The fs has been unmounted, or we did a 4540 * suspend/resume and this file no longer exists. 4541 */ 4542 VI_LOCK(vp); 4543 ASSERT(vp->v_count <= 1); 4544 vp->v_count = 0; 4545 VI_UNLOCK(vp); 4546 vrecycle(vp, curthread); 4547 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4548 return; 4549 } 4550 4551 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4552 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4553 4554 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4555 zfs_sa_upgrade_txholds(tx, zp); 4556 error = dmu_tx_assign(tx, TXG_WAIT); 4557 if (error) { 4558 dmu_tx_abort(tx); 4559 } else { 4560 mutex_enter(&zp->z_lock); 4561 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4562 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4563 zp->z_atime_dirty = 0; 4564 mutex_exit(&zp->z_lock); 4565 dmu_tx_commit(tx); 4566 } 4567 } 4568 4569 zfs_zinactive(zp); 4570 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4571} 4572 4573#ifdef sun 4574/* 4575 * Bounds-check the seek operation. 4576 * 4577 * IN: vp - vnode seeking within 4578 * ooff - old file offset 4579 * noffp - pointer to new file offset 4580 * ct - caller context 4581 * 4582 * RETURN: 0 if success 4583 * EINVAL if new offset invalid 4584 */ 4585/* ARGSUSED */ 4586static int 4587zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 4588 caller_context_t *ct) 4589{ 4590 if (vp->v_type == VDIR) 4591 return (0); 4592 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4593} 4594 4595/* 4596 * Pre-filter the generic locking function to trap attempts to place 4597 * a mandatory lock on a memory mapped file. 4598 */ 4599static int 4600zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 4601 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 4602{ 4603 znode_t *zp = VTOZ(vp); 4604 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4605 4606 ZFS_ENTER(zfsvfs); 4607 ZFS_VERIFY_ZP(zp); 4608 4609 /* 4610 * We are following the UFS semantics with respect to mapcnt 4611 * here: If we see that the file is mapped already, then we will 4612 * return an error, but we don't worry about races between this 4613 * function and zfs_map(). 4614 */ 4615 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { 4616 ZFS_EXIT(zfsvfs); 4617 return (EAGAIN); 4618 } 4619 ZFS_EXIT(zfsvfs); 4620 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4621} 4622 4623/* 4624 * If we can't find a page in the cache, we will create a new page 4625 * and fill it with file data. For efficiency, we may try to fill 4626 * multiple pages at once (klustering) to fill up the supplied page 4627 * list. Note that the pages to be filled are held with an exclusive 4628 * lock to prevent access by other threads while they are being filled. 4629 */ 4630static int 4631zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 4632 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 4633{ 4634 znode_t *zp = VTOZ(vp); 4635 page_t *pp, *cur_pp; 4636 objset_t *os = zp->z_zfsvfs->z_os; 4637 u_offset_t io_off, total; 4638 size_t io_len; 4639 int err; 4640 4641 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4642 /* 4643 * We only have a single page, don't bother klustering 4644 */ 4645 io_off = off; 4646 io_len = PAGESIZE; 4647 pp = page_create_va(vp, io_off, io_len, 4648 PG_EXCL | PG_WAIT, seg, addr); 4649 } else { 4650 /* 4651 * Try to find enough pages to fill the page list 4652 */ 4653 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4654 &io_len, off, plsz, 0); 4655 } 4656 if (pp == NULL) { 4657 /* 4658 * The page already exists, nothing to do here. 4659 */ 4660 *pl = NULL; 4661 return (0); 4662 } 4663 4664 /* 4665 * Fill the pages in the kluster. 4666 */ 4667 cur_pp = pp; 4668 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4669 caddr_t va; 4670 4671 ASSERT3U(io_off, ==, cur_pp->p_offset); 4672 va = zfs_map_page(cur_pp, S_WRITE); 4673 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4674 DMU_READ_PREFETCH); 4675 zfs_unmap_page(cur_pp, va); 4676 if (err) { 4677 /* On error, toss the entire kluster */ 4678 pvn_read_done(pp, B_ERROR); 4679 /* convert checksum errors into IO errors */ 4680 if (err == ECKSUM) 4681 err = EIO; 4682 return (err); 4683 } 4684 cur_pp = cur_pp->p_next; 4685 } 4686 4687 /* 4688 * Fill in the page list array from the kluster starting 4689 * from the desired offset `off'. 4690 * NOTE: the page list will always be null terminated. 4691 */ 4692 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4693 ASSERT(pl == NULL || (*pl)->p_offset == off); 4694 4695 return (0); 4696} 4697 4698/* 4699 * Return pointers to the pages for the file region [off, off + len] 4700 * in the pl array. If plsz is greater than len, this function may 4701 * also return page pointers from after the specified region 4702 * (i.e. the region [off, off + plsz]). These additional pages are 4703 * only returned if they are already in the cache, or were created as 4704 * part of a klustered read. 4705 * 4706 * IN: vp - vnode of file to get data from. 4707 * off - position in file to get data from. 4708 * len - amount of data to retrieve. 4709 * plsz - length of provided page list. 4710 * seg - segment to obtain pages for. 4711 * addr - virtual address of fault. 4712 * rw - mode of created pages. 4713 * cr - credentials of caller. 4714 * ct - caller context. 4715 * 4716 * OUT: protp - protection mode of created pages. 4717 * pl - list of pages created. 4718 * 4719 * RETURN: 0 if success 4720 * error code if failure 4721 * 4722 * Timestamps: 4723 * vp - atime updated 4724 */ 4725/* ARGSUSED */ 4726static int 4727zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4728 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4729 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4730{ 4731 znode_t *zp = VTOZ(vp); 4732 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4733 page_t **pl0 = pl; 4734 int err = 0; 4735 4736 /* we do our own caching, faultahead is unnecessary */ 4737 if (pl == NULL) 4738 return (0); 4739 else if (len > plsz) 4740 len = plsz; 4741 else 4742 len = P2ROUNDUP(len, PAGESIZE); 4743 ASSERT(plsz >= len); 4744 4745 ZFS_ENTER(zfsvfs); 4746 ZFS_VERIFY_ZP(zp); 4747 4748 if (protp) 4749 *protp = PROT_ALL; 4750 4751 /* 4752 * Loop through the requested range [off, off + len) looking 4753 * for pages. If we don't find a page, we will need to create 4754 * a new page and fill it with data from the file. 4755 */ 4756 while (len > 0) { 4757 if (*pl = page_lookup(vp, off, SE_SHARED)) 4758 *(pl+1) = NULL; 4759 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 4760 goto out; 4761 while (*pl) { 4762 ASSERT3U((*pl)->p_offset, ==, off); 4763 off += PAGESIZE; 4764 addr += PAGESIZE; 4765 if (len > 0) { 4766 ASSERT3U(len, >=, PAGESIZE); 4767 len -= PAGESIZE; 4768 } 4769 ASSERT3U(plsz, >=, PAGESIZE); 4770 plsz -= PAGESIZE; 4771 pl++; 4772 } 4773 } 4774 4775 /* 4776 * Fill out the page array with any pages already in the cache. 4777 */ 4778 while (plsz > 0 && 4779 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 4780 off += PAGESIZE; 4781 plsz -= PAGESIZE; 4782 } 4783out: 4784 if (err) { 4785 /* 4786 * Release any pages we have previously locked. 4787 */ 4788 while (pl > pl0) 4789 page_unlock(*--pl); 4790 } else { 4791 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4792 } 4793 4794 *pl = NULL; 4795 4796 ZFS_EXIT(zfsvfs); 4797 return (err); 4798} 4799 4800/* 4801 * Request a memory map for a section of a file. This code interacts 4802 * with common code and the VM system as follows: 4803 * 4804 * common code calls mmap(), which ends up in smmap_common() 4805 * 4806 * this calls VOP_MAP(), which takes you into (say) zfs 4807 * 4808 * zfs_map() calls as_map(), passing segvn_create() as the callback 4809 * 4810 * segvn_create() creates the new segment and calls VOP_ADDMAP() 4811 * 4812 * zfs_addmap() updates z_mapcnt 4813 */ 4814/*ARGSUSED*/ 4815static int 4816zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4817 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4818 caller_context_t *ct) 4819{ 4820 znode_t *zp = VTOZ(vp); 4821 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4822 segvn_crargs_t vn_a; 4823 int error; 4824 4825 ZFS_ENTER(zfsvfs); 4826 ZFS_VERIFY_ZP(zp); 4827 4828 if ((prot & PROT_WRITE) && (zp->z_pflags & 4829 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 4830 ZFS_EXIT(zfsvfs); 4831 return (EPERM); 4832 } 4833 4834 if ((prot & (PROT_READ | PROT_EXEC)) && 4835 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 4836 ZFS_EXIT(zfsvfs); 4837 return (EACCES); 4838 } 4839 4840 if (vp->v_flag & VNOMAP) { 4841 ZFS_EXIT(zfsvfs); 4842 return (ENOSYS); 4843 } 4844 4845 if (off < 0 || len > MAXOFFSET_T - off) { 4846 ZFS_EXIT(zfsvfs); 4847 return (ENXIO); 4848 } 4849 4850 if (vp->v_type != VREG) { 4851 ZFS_EXIT(zfsvfs); 4852 return (ENODEV); 4853 } 4854 4855 /* 4856 * If file is locked, disallow mapping. 4857 */ 4858 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { 4859 ZFS_EXIT(zfsvfs); 4860 return (EAGAIN); 4861 } 4862 4863 as_rangelock(as); 4864 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 4865 if (error != 0) { 4866 as_rangeunlock(as); 4867 ZFS_EXIT(zfsvfs); 4868 return (error); 4869 } 4870 4871 vn_a.vp = vp; 4872 vn_a.offset = (u_offset_t)off; 4873 vn_a.type = flags & MAP_TYPE; 4874 vn_a.prot = prot; 4875 vn_a.maxprot = maxprot; 4876 vn_a.cred = cr; 4877 vn_a.amp = NULL; 4878 vn_a.flags = flags & ~MAP_TYPE; 4879 vn_a.szc = 0; 4880 vn_a.lgrp_mem_policy_flags = 0; 4881 4882 error = as_map(as, *addrp, len, segvn_create, &vn_a); 4883 4884 as_rangeunlock(as); 4885 ZFS_EXIT(zfsvfs); 4886 return (error); 4887} 4888 4889/* ARGSUSED */ 4890static int 4891zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4892 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4893 caller_context_t *ct) 4894{ 4895 uint64_t pages = btopr(len); 4896 4897 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 4898 return (0); 4899} 4900 4901/* 4902 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 4903 * more accurate mtime for the associated file. Since we don't have a way of 4904 * detecting when the data was actually modified, we have to resort to 4905 * heuristics. If an explicit msync() is done, then we mark the mtime when the 4906 * last page is pushed. The problem occurs when the msync() call is omitted, 4907 * which by far the most common case: 4908 * 4909 * open() 4910 * mmap() 4911 * <modify memory> 4912 * munmap() 4913 * close() 4914 * <time lapse> 4915 * putpage() via fsflush 4916 * 4917 * If we wait until fsflush to come along, we can have a modification time that 4918 * is some arbitrary point in the future. In order to prevent this in the 4919 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 4920 * torn down. 4921 */ 4922/* ARGSUSED */ 4923static int 4924zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4925 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 4926 caller_context_t *ct) 4927{ 4928 uint64_t pages = btopr(len); 4929 4930 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 4931 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 4932 4933 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 4934 vn_has_cached_data(vp)) 4935 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 4936 4937 return (0); 4938} 4939 4940/* 4941 * Free or allocate space in a file. Currently, this function only 4942 * supports the `F_FREESP' command. However, this command is somewhat 4943 * misnamed, as its functionality includes the ability to allocate as 4944 * well as free space. 4945 * 4946 * IN: vp - vnode of file to free data in. 4947 * cmd - action to take (only F_FREESP supported). 4948 * bfp - section of file to free/alloc. 4949 * flag - current file open mode flags. 4950 * offset - current file offset. 4951 * cr - credentials of caller [UNUSED]. 4952 * ct - caller context. 4953 * 4954 * RETURN: 0 if success 4955 * error code if failure 4956 * 4957 * Timestamps: 4958 * vp - ctime|mtime updated 4959 */ 4960/* ARGSUSED */ 4961static int 4962zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 4963 offset_t offset, cred_t *cr, caller_context_t *ct) 4964{ 4965 znode_t *zp = VTOZ(vp); 4966 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4967 uint64_t off, len; 4968 int error; 4969 4970 ZFS_ENTER(zfsvfs); 4971 ZFS_VERIFY_ZP(zp); 4972 4973 if (cmd != F_FREESP) { 4974 ZFS_EXIT(zfsvfs); 4975 return (EINVAL); 4976 } 4977 4978 if (error = convoff(vp, bfp, 0, offset)) { 4979 ZFS_EXIT(zfsvfs); 4980 return (error); 4981 } 4982 4983 if (bfp->l_len < 0) { 4984 ZFS_EXIT(zfsvfs); 4985 return (EINVAL); 4986 } 4987 4988 off = bfp->l_start; 4989 len = bfp->l_len; /* 0 means from off to end of file */ 4990 4991 error = zfs_freesp(zp, off, len, flag, TRUE); 4992 4993 ZFS_EXIT(zfsvfs); 4994 return (error); 4995} 4996#endif /* sun */ 4997 4998CTASSERT(sizeof(struct zfid_short) <= sizeof(struct fid)); 4999CTASSERT(sizeof(struct zfid_long) <= sizeof(struct fid)); 5000 5001/*ARGSUSED*/ 5002static int 5003zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 5004{ 5005 znode_t *zp = VTOZ(vp); 5006 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5007 uint32_t gen; 5008 uint64_t gen64; 5009 uint64_t object = zp->z_id; 5010 zfid_short_t *zfid; 5011 int size, i, error; 5012 5013 ZFS_ENTER(zfsvfs); 5014 ZFS_VERIFY_ZP(zp); 5015 5016 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 5017 &gen64, sizeof (uint64_t))) != 0) { 5018 ZFS_EXIT(zfsvfs); 5019 return (error); 5020 } 5021 5022 gen = (uint32_t)gen64; 5023 5024 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 5025 fidp->fid_len = size; 5026 5027 zfid = (zfid_short_t *)fidp; 5028 5029 zfid->zf_len = size; 5030 5031 for (i = 0; i < sizeof (zfid->zf_object); i++) 5032 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 5033 5034 /* Must have a non-zero generation number to distinguish from .zfs */ 5035 if (gen == 0) 5036 gen = 1; 5037 for (i = 0; i < sizeof (zfid->zf_gen); i++) 5038 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 5039 5040 if (size == LONG_FID_LEN) { 5041 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 5042 zfid_long_t *zlfid; 5043 5044 zlfid = (zfid_long_t *)fidp; 5045 5046 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 5047 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 5048 5049 /* XXX - this should be the generation number for the objset */ 5050 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 5051 zlfid->zf_setgen[i] = 0; 5052 } 5053 5054 ZFS_EXIT(zfsvfs); 5055 return (0); 5056} 5057 5058static int 5059zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 5060 caller_context_t *ct) 5061{ 5062 znode_t *zp, *xzp; 5063 zfsvfs_t *zfsvfs; 5064 zfs_dirlock_t *dl; 5065 int error; 5066 5067 switch (cmd) { 5068 case _PC_LINK_MAX: 5069 *valp = INT_MAX; 5070 return (0); 5071 5072 case _PC_FILESIZEBITS: 5073 *valp = 64; 5074 return (0); 5075#ifdef sun 5076 case _PC_XATTR_EXISTS: 5077 zp = VTOZ(vp); 5078 zfsvfs = zp->z_zfsvfs; 5079 ZFS_ENTER(zfsvfs); 5080 ZFS_VERIFY_ZP(zp); 5081 *valp = 0; 5082 error = zfs_dirent_lock(&dl, zp, "", &xzp, 5083 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 5084 if (error == 0) { 5085 zfs_dirent_unlock(dl); 5086 if (!zfs_dirempty(xzp)) 5087 *valp = 1; 5088 VN_RELE(ZTOV(xzp)); 5089 } else if (error == ENOENT) { 5090 /* 5091 * If there aren't extended attributes, it's the 5092 * same as having zero of them. 5093 */ 5094 error = 0; 5095 } 5096 ZFS_EXIT(zfsvfs); 5097 return (error); 5098 5099 case _PC_SATTR_ENABLED: 5100 case _PC_SATTR_EXISTS: 5101 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 5102 (vp->v_type == VREG || vp->v_type == VDIR); 5103 return (0); 5104 5105 case _PC_ACCESS_FILTERING: 5106 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 5107 vp->v_type == VDIR; 5108 return (0); 5109 5110 case _PC_ACL_ENABLED: 5111 *valp = _ACL_ACE_ENABLED; 5112 return (0); 5113#endif /* sun */ 5114 case _PC_MIN_HOLE_SIZE: 5115 *valp = (int)SPA_MINBLOCKSIZE; 5116 return (0); 5117#ifdef sun 5118 case _PC_TIMESTAMP_RESOLUTION: 5119 /* nanosecond timestamp resolution */ 5120 *valp = 1L; 5121 return (0); 5122#endif /* sun */ 5123 case _PC_ACL_EXTENDED: 5124 *valp = 0; 5125 return (0); 5126 5127 case _PC_ACL_NFS4: 5128 *valp = 1; 5129 return (0); 5130 5131 case _PC_ACL_PATH_MAX: 5132 *valp = ACL_MAX_ENTRIES; 5133 return (0); 5134 5135 default: 5136 return (EOPNOTSUPP); 5137 } 5138} 5139 5140/*ARGSUSED*/ 5141static int 5142zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5143 caller_context_t *ct) 5144{ 5145 znode_t *zp = VTOZ(vp); 5146 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5147 int error; 5148 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5149 5150 ZFS_ENTER(zfsvfs); 5151 ZFS_VERIFY_ZP(zp); 5152 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 5153 ZFS_EXIT(zfsvfs); 5154 5155 return (error); 5156} 5157 5158/*ARGSUSED*/ 5159static int 5160zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5161 caller_context_t *ct) 5162{ 5163 znode_t *zp = VTOZ(vp); 5164 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5165 int error; 5166 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5167 zilog_t *zilog = zfsvfs->z_log; 5168 5169 ZFS_ENTER(zfsvfs); 5170 ZFS_VERIFY_ZP(zp); 5171 5172 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 5173 5174 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 5175 zil_commit(zilog, 0); 5176 5177 ZFS_EXIT(zfsvfs); 5178 return (error); 5179} 5180 5181#ifdef sun 5182/* 5183 * Tunable, both must be a power of 2. 5184 * 5185 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf 5186 * zcr_blksz_max: if set to less than the file block size, allow loaning out of 5187 * an arcbuf for a partial block read 5188 */ 5189int zcr_blksz_min = (1 << 10); /* 1K */ 5190int zcr_blksz_max = (1 << 17); /* 128K */ 5191 5192/*ARGSUSED*/ 5193static int 5194zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr, 5195 caller_context_t *ct) 5196{ 5197 znode_t *zp = VTOZ(vp); 5198 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5199 int max_blksz = zfsvfs->z_max_blksz; 5200 uio_t *uio = &xuio->xu_uio; 5201 ssize_t size = uio->uio_resid; 5202 offset_t offset = uio->uio_loffset; 5203 int blksz; 5204 int fullblk, i; 5205 arc_buf_t *abuf; 5206 ssize_t maxsize; 5207 int preamble, postamble; 5208 5209 if (xuio->xu_type != UIOTYPE_ZEROCOPY) 5210 return (EINVAL); 5211 5212 ZFS_ENTER(zfsvfs); 5213 ZFS_VERIFY_ZP(zp); 5214 switch (ioflag) { 5215 case UIO_WRITE: 5216 /* 5217 * Loan out an arc_buf for write if write size is bigger than 5218 * max_blksz, and the file's block size is also max_blksz. 5219 */ 5220 blksz = max_blksz; 5221 if (size < blksz || zp->z_blksz != blksz) { 5222 ZFS_EXIT(zfsvfs); 5223 return (EINVAL); 5224 } 5225 /* 5226 * Caller requests buffers for write before knowing where the 5227 * write offset might be (e.g. NFS TCP write). 5228 */ 5229 if (offset == -1) { 5230 preamble = 0; 5231 } else { 5232 preamble = P2PHASE(offset, blksz); 5233 if (preamble) { 5234 preamble = blksz - preamble; 5235 size -= preamble; 5236 } 5237 } 5238 5239 postamble = P2PHASE(size, blksz); 5240 size -= postamble; 5241 5242 fullblk = size / blksz; 5243 (void) dmu_xuio_init(xuio, 5244 (preamble != 0) + fullblk + (postamble != 0)); 5245 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble, 5246 int, postamble, int, 5247 (preamble != 0) + fullblk + (postamble != 0)); 5248 5249 /* 5250 * Have to fix iov base/len for partial buffers. They 5251 * currently represent full arc_buf's. 5252 */ 5253 if (preamble) { 5254 /* data begins in the middle of the arc_buf */ 5255 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5256 blksz); 5257 ASSERT(abuf); 5258 (void) dmu_xuio_add(xuio, abuf, 5259 blksz - preamble, preamble); 5260 } 5261 5262 for (i = 0; i < fullblk; i++) { 5263 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5264 blksz); 5265 ASSERT(abuf); 5266 (void) dmu_xuio_add(xuio, abuf, 0, blksz); 5267 } 5268 5269 if (postamble) { 5270 /* data ends in the middle of the arc_buf */ 5271 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5272 blksz); 5273 ASSERT(abuf); 5274 (void) dmu_xuio_add(xuio, abuf, 0, postamble); 5275 } 5276 break; 5277 case UIO_READ: 5278 /* 5279 * Loan out an arc_buf for read if the read size is larger than 5280 * the current file block size. Block alignment is not 5281 * considered. Partial arc_buf will be loaned out for read. 5282 */ 5283 blksz = zp->z_blksz; 5284 if (blksz < zcr_blksz_min) 5285 blksz = zcr_blksz_min; 5286 if (blksz > zcr_blksz_max) 5287 blksz = zcr_blksz_max; 5288 /* avoid potential complexity of dealing with it */ 5289 if (blksz > max_blksz) { 5290 ZFS_EXIT(zfsvfs); 5291 return (EINVAL); 5292 } 5293 5294 maxsize = zp->z_size - uio->uio_loffset; 5295 if (size > maxsize) 5296 size = maxsize; 5297 5298 if (size < blksz || vn_has_cached_data(vp)) { 5299 ZFS_EXIT(zfsvfs); 5300 return (EINVAL); 5301 } 5302 break; 5303 default: 5304 ZFS_EXIT(zfsvfs); 5305 return (EINVAL); 5306 } 5307 5308 uio->uio_extflg = UIO_XUIO; 5309 XUIO_XUZC_RW(xuio) = ioflag; 5310 ZFS_EXIT(zfsvfs); 5311 return (0); 5312} 5313 5314/*ARGSUSED*/ 5315static int 5316zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct) 5317{ 5318 int i; 5319 arc_buf_t *abuf; 5320 int ioflag = XUIO_XUZC_RW(xuio); 5321 5322 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY); 5323 5324 i = dmu_xuio_cnt(xuio); 5325 while (i-- > 0) { 5326 abuf = dmu_xuio_arcbuf(xuio, i); 5327 /* 5328 * if abuf == NULL, it must be a write buffer 5329 * that has been returned in zfs_write(). 5330 */ 5331 if (abuf) 5332 dmu_return_arcbuf(abuf); 5333 ASSERT(abuf || ioflag == UIO_WRITE); 5334 } 5335 5336 dmu_xuio_fini(xuio); 5337 return (0); 5338} 5339 5340/* 5341 * Predeclare these here so that the compiler assumes that 5342 * this is an "old style" function declaration that does 5343 * not include arguments => we won't get type mismatch errors 5344 * in the initializations that follow. 5345 */ 5346static int zfs_inval(); 5347static int zfs_isdir(); 5348 5349static int 5350zfs_inval() 5351{ 5352 return (EINVAL); 5353} 5354 5355static int 5356zfs_isdir() 5357{ 5358 return (EISDIR); 5359} 5360/* 5361 * Directory vnode operations template 5362 */ 5363vnodeops_t *zfs_dvnodeops; 5364const fs_operation_def_t zfs_dvnodeops_template[] = { 5365 VOPNAME_OPEN, { .vop_open = zfs_open }, 5366 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5367 VOPNAME_READ, { .error = zfs_isdir }, 5368 VOPNAME_WRITE, { .error = zfs_isdir }, 5369 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5370 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5371 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5372 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5373 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5374 VOPNAME_CREATE, { .vop_create = zfs_create }, 5375 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5376 VOPNAME_LINK, { .vop_link = zfs_link }, 5377 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5378 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 5379 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5380 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5381 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 5382 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5383 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5384 VOPNAME_FID, { .vop_fid = zfs_fid }, 5385 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5386 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5387 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5388 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5389 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5390 NULL, NULL 5391}; 5392 5393/* 5394 * Regular file vnode operations template 5395 */ 5396vnodeops_t *zfs_fvnodeops; 5397const fs_operation_def_t zfs_fvnodeops_template[] = { 5398 VOPNAME_OPEN, { .vop_open = zfs_open }, 5399 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5400 VOPNAME_READ, { .vop_read = zfs_read }, 5401 VOPNAME_WRITE, { .vop_write = zfs_write }, 5402 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5403 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5404 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5405 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5406 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5407 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5408 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5409 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5410 VOPNAME_FID, { .vop_fid = zfs_fid }, 5411 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5412 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 5413 VOPNAME_SPACE, { .vop_space = zfs_space }, 5414 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 5415 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 5416 VOPNAME_MAP, { .vop_map = zfs_map }, 5417 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 5418 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 5419 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5420 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5421 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5422 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5423 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf }, 5424 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf }, 5425 NULL, NULL 5426}; 5427 5428/* 5429 * Symbolic link vnode operations template 5430 */ 5431vnodeops_t *zfs_symvnodeops; 5432const fs_operation_def_t zfs_symvnodeops_template[] = { 5433 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5434 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5435 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5436 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5437 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 5438 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5439 VOPNAME_FID, { .vop_fid = zfs_fid }, 5440 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5441 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5442 NULL, NULL 5443}; 5444 5445/* 5446 * special share hidden files vnode operations template 5447 */ 5448vnodeops_t *zfs_sharevnodeops; 5449const fs_operation_def_t zfs_sharevnodeops_template[] = { 5450 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5451 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5452 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5453 VOPNAME_FID, { .vop_fid = zfs_fid }, 5454 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5455 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5456 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5457 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5458 NULL, NULL 5459}; 5460 5461/* 5462 * Extended attribute directory vnode operations template 5463 * This template is identical to the directory vnodes 5464 * operation template except for restricted operations: 5465 * VOP_MKDIR() 5466 * VOP_SYMLINK() 5467 * Note that there are other restrictions embedded in: 5468 * zfs_create() - restrict type to VREG 5469 * zfs_link() - no links into/out of attribute space 5470 * zfs_rename() - no moves into/out of attribute space 5471 */ 5472vnodeops_t *zfs_xdvnodeops; 5473const fs_operation_def_t zfs_xdvnodeops_template[] = { 5474 VOPNAME_OPEN, { .vop_open = zfs_open }, 5475 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5476 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5477 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5478 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5479 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5480 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5481 VOPNAME_CREATE, { .vop_create = zfs_create }, 5482 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5483 VOPNAME_LINK, { .vop_link = zfs_link }, 5484 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5485 VOPNAME_MKDIR, { .error = zfs_inval }, 5486 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5487 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5488 VOPNAME_SYMLINK, { .error = zfs_inval }, 5489 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5490 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5491 VOPNAME_FID, { .vop_fid = zfs_fid }, 5492 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5493 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5494 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5495 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5496 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5497 NULL, NULL 5498}; 5499 5500/* 5501 * Error vnode operations template 5502 */ 5503vnodeops_t *zfs_evnodeops; 5504const fs_operation_def_t zfs_evnodeops_template[] = { 5505 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5506 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5507 NULL, NULL 5508}; 5509#endif /* sun */ 5510 5511static int 5512ioflags(int ioflags) 5513{ 5514 int flags = 0; 5515 5516 if (ioflags & IO_APPEND) 5517 flags |= FAPPEND; 5518 if (ioflags & IO_NDELAY) 5519 flags |= FNONBLOCK; 5520 if (ioflags & IO_SYNC) 5521 flags |= (FSYNC | FDSYNC | FRSYNC); 5522 5523 return (flags); 5524} 5525 5526static int 5527zfs_getpages(struct vnode *vp, vm_page_t *m, int count, int reqpage) 5528{ 5529 znode_t *zp = VTOZ(vp); 5530 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5531 objset_t *os = zp->z_zfsvfs->z_os; 5532 vm_page_t mreq; 5533 vm_object_t object; 5534 caddr_t va; 5535 struct sf_buf *sf; 5536 int i, error; 5537 int pcount, size; 5538 5539 ZFS_ENTER(zfsvfs); 5540 ZFS_VERIFY_ZP(zp); 5541 5542 pcount = round_page(count) / PAGE_SIZE; 5543 mreq = m[reqpage]; 5544 object = mreq->object; 5545 error = 0; 5546 5547 KASSERT(vp->v_object == object, ("mismatching object")); 5548 5549 VM_OBJECT_LOCK(object); 5550 5551 for (i = 0; i < pcount; i++) { 5552 if (i != reqpage) { 5553 vm_page_lock(m[i]); 5554 vm_page_free(m[i]); 5555 vm_page_unlock(m[i]); 5556 } 5557 } 5558 5559 if (mreq->valid) { 5560 if (mreq->valid != VM_PAGE_BITS_ALL) 5561 vm_page_zero_invalid(mreq, TRUE); 5562 VM_OBJECT_UNLOCK(object); 5563 ZFS_EXIT(zfsvfs); 5564 return (VM_PAGER_OK); 5565 } 5566 5567 PCPU_INC(cnt.v_vnodein); 5568 PCPU_INC(cnt.v_vnodepgsin); 5569 5570 if (IDX_TO_OFF(mreq->pindex) >= object->un_pager.vnp.vnp_size) { 5571 VM_OBJECT_UNLOCK(object); 5572 ZFS_EXIT(zfsvfs); 5573 return (VM_PAGER_BAD); 5574 } 5575 5576 size = PAGE_SIZE; 5577 if (IDX_TO_OFF(mreq->pindex) + size > object->un_pager.vnp.vnp_size) 5578 size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(mreq->pindex); 5579 5580 VM_OBJECT_UNLOCK(object); 5581 va = zfs_map_page(mreq, &sf); 5582 error = dmu_read(os, zp->z_id, IDX_TO_OFF(mreq->pindex), 5583 size, va, DMU_READ_PREFETCH); 5584 if (size != PAGE_SIZE) 5585 bzero(va + size, PAGE_SIZE - size); 5586 zfs_unmap_page(sf); 5587 VM_OBJECT_LOCK(object); 5588 5589 if (!error) 5590 mreq->valid = VM_PAGE_BITS_ALL; 5591 KASSERT(mreq->dirty == 0, ("zfs_getpages: page %p is dirty", mreq)); 5592 5593 VM_OBJECT_UNLOCK(object); 5594 5595 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 5596 ZFS_EXIT(zfsvfs); 5597 return (error ? VM_PAGER_ERROR : VM_PAGER_OK); 5598} 5599 5600static int 5601zfs_freebsd_getpages(ap) 5602 struct vop_getpages_args /* { 5603 struct vnode *a_vp; 5604 vm_page_t *a_m; 5605 int a_count; 5606 int a_reqpage; 5607 vm_ooffset_t a_offset; 5608 } */ *ap; 5609{ 5610 5611 return (zfs_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage)); 5612} 5613 5614static int 5615zfs_freebsd_open(ap) 5616 struct vop_open_args /* { 5617 struct vnode *a_vp; 5618 int a_mode; 5619 struct ucred *a_cred; 5620 struct thread *a_td; 5621 } */ *ap; 5622{ 5623 vnode_t *vp = ap->a_vp; 5624 znode_t *zp = VTOZ(vp); 5625 int error; 5626 5627 error = zfs_open(&vp, ap->a_mode, ap->a_cred, NULL); 5628 if (error == 0) 5629 vnode_create_vobject(vp, zp->z_size, ap->a_td); 5630 return (error); 5631} 5632 5633static int 5634zfs_freebsd_close(ap) 5635 struct vop_close_args /* { 5636 struct vnode *a_vp; 5637 int a_fflag; 5638 struct ucred *a_cred; 5639 struct thread *a_td; 5640 } */ *ap; 5641{ 5642 5643 return (zfs_close(ap->a_vp, ap->a_fflag, 0, 0, ap->a_cred, NULL)); 5644} 5645 5646static int 5647zfs_freebsd_ioctl(ap) 5648 struct vop_ioctl_args /* { 5649 struct vnode *a_vp; 5650 u_long a_command; 5651 caddr_t a_data; 5652 int a_fflag; 5653 struct ucred *cred; 5654 struct thread *td; 5655 } */ *ap; 5656{ 5657 5658 return (zfs_ioctl(ap->a_vp, ap->a_command, (intptr_t)ap->a_data, 5659 ap->a_fflag, ap->a_cred, NULL, NULL)); 5660} 5661 5662static int 5663zfs_freebsd_read(ap) 5664 struct vop_read_args /* { 5665 struct vnode *a_vp; 5666 struct uio *a_uio; 5667 int a_ioflag; 5668 struct ucred *a_cred; 5669 } */ *ap; 5670{ 5671 5672 return (zfs_read(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 5673 ap->a_cred, NULL)); 5674} 5675 5676static int 5677zfs_freebsd_write(ap) 5678 struct vop_write_args /* { 5679 struct vnode *a_vp; 5680 struct uio *a_uio; 5681 int a_ioflag; 5682 struct ucred *a_cred; 5683 } */ *ap; 5684{ 5685 5686 if (vn_rlimit_fsize(ap->a_vp, ap->a_uio, ap->a_uio->uio_td)) 5687 return (EFBIG); 5688 5689 return (zfs_write(ap->a_vp, ap->a_uio, ioflags(ap->a_ioflag), 5690 ap->a_cred, NULL)); 5691} 5692 5693static int 5694zfs_freebsd_access(ap) 5695 struct vop_access_args /* { 5696 struct vnode *a_vp; 5697 accmode_t a_accmode; 5698 struct ucred *a_cred; 5699 struct thread *a_td; 5700 } */ *ap; 5701{ 5702 vnode_t *vp = ap->a_vp; 5703 znode_t *zp = VTOZ(vp); 5704 accmode_t accmode; 5705 int error = 0; 5706 5707 /* 5708 * ZFS itself only knowns about VREAD, VWRITE, VEXEC and VAPPEND, 5709 */ 5710 accmode = ap->a_accmode & (VREAD|VWRITE|VEXEC|VAPPEND); 5711 if (accmode != 0) 5712 error = zfs_access(ap->a_vp, accmode, 0, ap->a_cred, NULL); 5713 5714 /* 5715 * VADMIN has to be handled by vaccess(). 5716 */ 5717 if (error == 0) { 5718 accmode = ap->a_accmode & ~(VREAD|VWRITE|VEXEC|VAPPEND); 5719 if (accmode != 0) { 5720 error = vaccess(vp->v_type, zp->z_mode, zp->z_uid, 5721 zp->z_gid, accmode, ap->a_cred, NULL); 5722 } 5723 } 5724 5725 /* 5726 * For VEXEC, ensure that at least one execute bit is set for 5727 * non-directories. 5728 */ 5729 if (error == 0 && (ap->a_accmode & VEXEC) != 0 && vp->v_type != VDIR && 5730 (zp->z_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0) { 5731 error = EACCES; 5732 } 5733 5734 return (error); 5735} 5736 5737static int 5738zfs_freebsd_lookup(ap) 5739 struct vop_lookup_args /* { 5740 struct vnode *a_dvp; 5741 struct vnode **a_vpp; 5742 struct componentname *a_cnp; 5743 } */ *ap; 5744{ 5745 struct componentname *cnp = ap->a_cnp; 5746 char nm[NAME_MAX + 1]; 5747 5748 ASSERT(cnp->cn_namelen < sizeof(nm)); 5749 strlcpy(nm, cnp->cn_nameptr, MIN(cnp->cn_namelen + 1, sizeof(nm))); 5750 5751 return (zfs_lookup(ap->a_dvp, nm, ap->a_vpp, cnp, cnp->cn_nameiop, 5752 cnp->cn_cred, cnp->cn_thread, 0)); 5753} 5754 5755static int 5756zfs_freebsd_create(ap) 5757 struct vop_create_args /* { 5758 struct vnode *a_dvp; 5759 struct vnode **a_vpp; 5760 struct componentname *a_cnp; 5761 struct vattr *a_vap; 5762 } */ *ap; 5763{ 5764 struct componentname *cnp = ap->a_cnp; 5765 vattr_t *vap = ap->a_vap; 5766 int mode; 5767 5768 ASSERT(cnp->cn_flags & SAVENAME); 5769 5770 vattr_init_mask(vap); 5771 mode = vap->va_mode & ALLPERMS; 5772 5773 return (zfs_create(ap->a_dvp, cnp->cn_nameptr, vap, !EXCL, mode, 5774 ap->a_vpp, cnp->cn_cred, cnp->cn_thread)); 5775} 5776 5777static int 5778zfs_freebsd_remove(ap) 5779 struct vop_remove_args /* { 5780 struct vnode *a_dvp; 5781 struct vnode *a_vp; 5782 struct componentname *a_cnp; 5783 } */ *ap; 5784{ 5785 5786 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 5787 5788 return (zfs_remove(ap->a_dvp, ap->a_cnp->cn_nameptr, 5789 ap->a_cnp->cn_cred, NULL, 0)); 5790} 5791 5792static int 5793zfs_freebsd_mkdir(ap) 5794 struct vop_mkdir_args /* { 5795 struct vnode *a_dvp; 5796 struct vnode **a_vpp; 5797 struct componentname *a_cnp; 5798 struct vattr *a_vap; 5799 } */ *ap; 5800{ 5801 vattr_t *vap = ap->a_vap; 5802 5803 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 5804 5805 vattr_init_mask(vap); 5806 5807 return (zfs_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, vap, ap->a_vpp, 5808 ap->a_cnp->cn_cred, NULL, 0, NULL)); 5809} 5810 5811static int 5812zfs_freebsd_rmdir(ap) 5813 struct vop_rmdir_args /* { 5814 struct vnode *a_dvp; 5815 struct vnode *a_vp; 5816 struct componentname *a_cnp; 5817 } */ *ap; 5818{ 5819 struct componentname *cnp = ap->a_cnp; 5820 5821 ASSERT(cnp->cn_flags & SAVENAME); 5822 5823 return (zfs_rmdir(ap->a_dvp, cnp->cn_nameptr, NULL, cnp->cn_cred, NULL, 0)); 5824} 5825 5826static int 5827zfs_freebsd_readdir(ap) 5828 struct vop_readdir_args /* { 5829 struct vnode *a_vp; 5830 struct uio *a_uio; 5831 struct ucred *a_cred; 5832 int *a_eofflag; 5833 int *a_ncookies; 5834 u_long **a_cookies; 5835 } */ *ap; 5836{ 5837 5838 return (zfs_readdir(ap->a_vp, ap->a_uio, ap->a_cred, ap->a_eofflag, 5839 ap->a_ncookies, ap->a_cookies)); 5840} 5841 5842static int 5843zfs_freebsd_fsync(ap) 5844 struct vop_fsync_args /* { 5845 struct vnode *a_vp; 5846 int a_waitfor; 5847 struct thread *a_td; 5848 } */ *ap; 5849{ 5850 5851 vop_stdfsync(ap); 5852 return (zfs_fsync(ap->a_vp, 0, ap->a_td->td_ucred, NULL)); 5853} 5854 5855static int 5856zfs_freebsd_getattr(ap) 5857 struct vop_getattr_args /* { 5858 struct vnode *a_vp; 5859 struct vattr *a_vap; 5860 struct ucred *a_cred; 5861 } */ *ap; 5862{ 5863 vattr_t *vap = ap->a_vap; 5864 xvattr_t xvap; 5865 u_long fflags = 0; 5866 int error; 5867 5868 xva_init(&xvap); 5869 xvap.xva_vattr = *vap; 5870 xvap.xva_vattr.va_mask |= AT_XVATTR; 5871 5872 /* Convert chflags into ZFS-type flags. */ 5873 /* XXX: what about SF_SETTABLE?. */ 5874 XVA_SET_REQ(&xvap, XAT_IMMUTABLE); 5875 XVA_SET_REQ(&xvap, XAT_APPENDONLY); 5876 XVA_SET_REQ(&xvap, XAT_NOUNLINK); 5877 XVA_SET_REQ(&xvap, XAT_NODUMP); 5878 error = zfs_getattr(ap->a_vp, (vattr_t *)&xvap, 0, ap->a_cred, NULL); 5879 if (error != 0) 5880 return (error); 5881 5882 /* Convert ZFS xattr into chflags. */ 5883#define FLAG_CHECK(fflag, xflag, xfield) do { \ 5884 if (XVA_ISSET_RTN(&xvap, (xflag)) && (xfield) != 0) \ 5885 fflags |= (fflag); \ 5886} while (0) 5887 FLAG_CHECK(SF_IMMUTABLE, XAT_IMMUTABLE, 5888 xvap.xva_xoptattrs.xoa_immutable); 5889 FLAG_CHECK(SF_APPEND, XAT_APPENDONLY, 5890 xvap.xva_xoptattrs.xoa_appendonly); 5891 FLAG_CHECK(SF_NOUNLINK, XAT_NOUNLINK, 5892 xvap.xva_xoptattrs.xoa_nounlink); 5893 FLAG_CHECK(UF_NODUMP, XAT_NODUMP, 5894 xvap.xva_xoptattrs.xoa_nodump); 5895#undef FLAG_CHECK 5896 *vap = xvap.xva_vattr; 5897 vap->va_flags = fflags; 5898 return (0); 5899} 5900 5901static int 5902zfs_freebsd_setattr(ap) 5903 struct vop_setattr_args /* { 5904 struct vnode *a_vp; 5905 struct vattr *a_vap; 5906 struct ucred *a_cred; 5907 } */ *ap; 5908{ 5909 vnode_t *vp = ap->a_vp; 5910 vattr_t *vap = ap->a_vap; 5911 cred_t *cred = ap->a_cred; 5912 xvattr_t xvap; 5913 u_long fflags; 5914 uint64_t zflags; 5915 5916 vattr_init_mask(vap); 5917 vap->va_mask &= ~AT_NOSET; 5918 5919 xva_init(&xvap); 5920 xvap.xva_vattr = *vap; 5921 5922 zflags = VTOZ(vp)->z_pflags; 5923 5924 if (vap->va_flags != VNOVAL) { 5925 zfsvfs_t *zfsvfs = VTOZ(vp)->z_zfsvfs; 5926 int error; 5927 5928 if (zfsvfs->z_use_fuids == B_FALSE) 5929 return (EOPNOTSUPP); 5930 5931 fflags = vap->va_flags; 5932 if ((fflags & ~(SF_IMMUTABLE|SF_APPEND|SF_NOUNLINK|UF_NODUMP)) != 0) 5933 return (EOPNOTSUPP); 5934 /* 5935 * Unprivileged processes are not permitted to unset system 5936 * flags, or modify flags if any system flags are set. 5937 * Privileged non-jail processes may not modify system flags 5938 * if securelevel > 0 and any existing system flags are set. 5939 * Privileged jail processes behave like privileged non-jail 5940 * processes if the security.jail.chflags_allowed sysctl is 5941 * is non-zero; otherwise, they behave like unprivileged 5942 * processes. 5943 */ 5944 if (secpolicy_fs_owner(vp->v_mount, cred) == 0 || 5945 priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0) == 0) { 5946 if (zflags & 5947 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 5948 error = securelevel_gt(cred, 0); 5949 if (error != 0) 5950 return (error); 5951 } 5952 } else { 5953 /* 5954 * Callers may only modify the file flags on objects they 5955 * have VADMIN rights for. 5956 */ 5957 if ((error = VOP_ACCESS(vp, VADMIN, cred, curthread)) != 0) 5958 return (error); 5959 if (zflags & 5960 (ZFS_IMMUTABLE | ZFS_APPENDONLY | ZFS_NOUNLINK)) { 5961 return (EPERM); 5962 } 5963 if (fflags & 5964 (SF_IMMUTABLE | SF_APPEND | SF_NOUNLINK)) { 5965 return (EPERM); 5966 } 5967 } 5968 5969#define FLAG_CHANGE(fflag, zflag, xflag, xfield) do { \ 5970 if (((fflags & (fflag)) && !(zflags & (zflag))) || \ 5971 ((zflags & (zflag)) && !(fflags & (fflag)))) { \ 5972 XVA_SET_REQ(&xvap, (xflag)); \ 5973 (xfield) = ((fflags & (fflag)) != 0); \ 5974 } \ 5975} while (0) 5976 /* Convert chflags into ZFS-type flags. */ 5977 /* XXX: what about SF_SETTABLE?. */ 5978 FLAG_CHANGE(SF_IMMUTABLE, ZFS_IMMUTABLE, XAT_IMMUTABLE, 5979 xvap.xva_xoptattrs.xoa_immutable); 5980 FLAG_CHANGE(SF_APPEND, ZFS_APPENDONLY, XAT_APPENDONLY, 5981 xvap.xva_xoptattrs.xoa_appendonly); 5982 FLAG_CHANGE(SF_NOUNLINK, ZFS_NOUNLINK, XAT_NOUNLINK, 5983 xvap.xva_xoptattrs.xoa_nounlink); 5984 FLAG_CHANGE(UF_NODUMP, ZFS_NODUMP, XAT_NODUMP, 5985 xvap.xva_xoptattrs.xoa_nodump); 5986#undef FLAG_CHANGE 5987 } 5988 return (zfs_setattr(vp, (vattr_t *)&xvap, 0, cred, NULL)); 5989} 5990 5991static int 5992zfs_freebsd_rename(ap) 5993 struct vop_rename_args /* { 5994 struct vnode *a_fdvp; 5995 struct vnode *a_fvp; 5996 struct componentname *a_fcnp; 5997 struct vnode *a_tdvp; 5998 struct vnode *a_tvp; 5999 struct componentname *a_tcnp; 6000 } */ *ap; 6001{ 6002 vnode_t *fdvp = ap->a_fdvp; 6003 vnode_t *fvp = ap->a_fvp; 6004 vnode_t *tdvp = ap->a_tdvp; 6005 vnode_t *tvp = ap->a_tvp; 6006 int error; 6007 6008 ASSERT(ap->a_fcnp->cn_flags & (SAVENAME|SAVESTART)); 6009 ASSERT(ap->a_tcnp->cn_flags & (SAVENAME|SAVESTART)); 6010 6011 error = zfs_rename(fdvp, ap->a_fcnp->cn_nameptr, tdvp, 6012 ap->a_tcnp->cn_nameptr, ap->a_fcnp->cn_cred, NULL, 0); 6013 6014 if (tdvp == tvp) 6015 VN_RELE(tdvp); 6016 else 6017 VN_URELE(tdvp); 6018 if (tvp) 6019 VN_URELE(tvp); 6020 VN_RELE(fdvp); 6021 VN_RELE(fvp); 6022 6023 return (error); 6024} 6025 6026static int 6027zfs_freebsd_symlink(ap) 6028 struct vop_symlink_args /* { 6029 struct vnode *a_dvp; 6030 struct vnode **a_vpp; 6031 struct componentname *a_cnp; 6032 struct vattr *a_vap; 6033 char *a_target; 6034 } */ *ap; 6035{ 6036 struct componentname *cnp = ap->a_cnp; 6037 vattr_t *vap = ap->a_vap; 6038 6039 ASSERT(cnp->cn_flags & SAVENAME); 6040 6041 vap->va_type = VLNK; /* FreeBSD: Syscall only sets va_mode. */ 6042 vattr_init_mask(vap); 6043 6044 return (zfs_symlink(ap->a_dvp, ap->a_vpp, cnp->cn_nameptr, vap, 6045 ap->a_target, cnp->cn_cred, cnp->cn_thread)); 6046} 6047 6048static int 6049zfs_freebsd_readlink(ap) 6050 struct vop_readlink_args /* { 6051 struct vnode *a_vp; 6052 struct uio *a_uio; 6053 struct ucred *a_cred; 6054 } */ *ap; 6055{ 6056 6057 return (zfs_readlink(ap->a_vp, ap->a_uio, ap->a_cred, NULL)); 6058} 6059 6060static int 6061zfs_freebsd_link(ap) 6062 struct vop_link_args /* { 6063 struct vnode *a_tdvp; 6064 struct vnode *a_vp; 6065 struct componentname *a_cnp; 6066 } */ *ap; 6067{ 6068 struct componentname *cnp = ap->a_cnp; 6069 6070 ASSERT(cnp->cn_flags & SAVENAME); 6071 6072 return (zfs_link(ap->a_tdvp, ap->a_vp, cnp->cn_nameptr, cnp->cn_cred, NULL, 0)); 6073} 6074 6075static int 6076zfs_freebsd_inactive(ap) 6077 struct vop_inactive_args /* { 6078 struct vnode *a_vp; 6079 struct thread *a_td; 6080 } */ *ap; 6081{ 6082 vnode_t *vp = ap->a_vp; 6083 6084 zfs_inactive(vp, ap->a_td->td_ucred, NULL); 6085 return (0); 6086} 6087 6088static void 6089zfs_reclaim_complete(void *arg, int pending) 6090{ 6091 znode_t *zp = arg; 6092 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6093 6094 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 6095 if (zp->z_sa_hdl != NULL) { 6096 ZFS_OBJ_HOLD_ENTER(zfsvfs, zp->z_id); 6097 zfs_znode_dmu_fini(zp); 6098 ZFS_OBJ_HOLD_EXIT(zfsvfs, zp->z_id); 6099 } 6100 zfs_znode_free(zp); 6101 rw_exit(&zfsvfs->z_teardown_inactive_lock); 6102 /* 6103 * If the file system is being unmounted, there is a process waiting 6104 * for us, wake it up. 6105 */ 6106 if (zfsvfs->z_unmounted) 6107 wakeup_one(zfsvfs); 6108} 6109 6110static int 6111zfs_freebsd_reclaim(ap) 6112 struct vop_reclaim_args /* { 6113 struct vnode *a_vp; 6114 struct thread *a_td; 6115 } */ *ap; 6116{ 6117 vnode_t *vp = ap->a_vp; 6118 znode_t *zp = VTOZ(vp); 6119 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 6120 boolean_t rlocked; 6121 6122 rlocked = rw_tryenter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 6123 6124 ASSERT(zp != NULL); 6125 6126 /* 6127 * Destroy the vm object and flush associated pages. 6128 */ 6129 vnode_destroy_vobject(vp); 6130 6131 mutex_enter(&zp->z_lock); 6132 zp->z_vnode = NULL; 6133 mutex_exit(&zp->z_lock); 6134 6135 if (zp->z_unlinked) { 6136 ; /* Do nothing. */ 6137 } else if (!rlocked) { 6138 TASK_INIT(&zp->z_task, 0, zfs_reclaim_complete, zp); 6139 taskqueue_enqueue(taskqueue_thread, &zp->z_task); 6140 } else if (zp->z_sa_hdl == NULL) { 6141 zfs_znode_free(zp); 6142 } else /* if (!zp->z_unlinked && zp->z_dbuf != NULL) */ { 6143 int locked; 6144 6145 locked = MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)) ? 2 : 6146 ZFS_OBJ_HOLD_TRYENTER(zfsvfs, zp->z_id); 6147 if (locked == 0) { 6148 /* 6149 * Lock can't be obtained due to deadlock possibility, 6150 * so defer znode destruction. 6151 */ 6152 TASK_INIT(&zp->z_task, 0, zfs_reclaim_complete, zp); 6153 taskqueue_enqueue(taskqueue_thread, &zp->z_task); 6154 } else { 6155 zfs_znode_dmu_fini(zp); 6156 if (locked == 1) 6157 ZFS_OBJ_HOLD_EXIT(zfsvfs, zp->z_id); 6158 zfs_znode_free(zp); 6159 } 6160 } 6161 VI_LOCK(vp); 6162 vp->v_data = NULL; 6163 ASSERT(vp->v_holdcnt >= 1); 6164 VI_UNLOCK(vp); 6165 if (rlocked) 6166 rw_exit(&zfsvfs->z_teardown_inactive_lock); 6167 return (0); 6168} 6169 6170static int 6171zfs_freebsd_fid(ap) 6172 struct vop_fid_args /* { 6173 struct vnode *a_vp; 6174 struct fid *a_fid; 6175 } */ *ap; 6176{ 6177 6178 return (zfs_fid(ap->a_vp, (void *)ap->a_fid, NULL)); 6179} 6180 6181static int 6182zfs_freebsd_pathconf(ap) 6183 struct vop_pathconf_args /* { 6184 struct vnode *a_vp; 6185 int a_name; 6186 register_t *a_retval; 6187 } */ *ap; 6188{ 6189 ulong_t val; 6190 int error; 6191 6192 error = zfs_pathconf(ap->a_vp, ap->a_name, &val, curthread->td_ucred, NULL); 6193 if (error == 0) 6194 *ap->a_retval = val; 6195 else if (error == EOPNOTSUPP) 6196 error = vop_stdpathconf(ap); 6197 return (error); 6198} 6199 6200static int 6201zfs_freebsd_fifo_pathconf(ap) 6202 struct vop_pathconf_args /* { 6203 struct vnode *a_vp; 6204 int a_name; 6205 register_t *a_retval; 6206 } */ *ap; 6207{ 6208 6209 switch (ap->a_name) { 6210 case _PC_ACL_EXTENDED: 6211 case _PC_ACL_NFS4: 6212 case _PC_ACL_PATH_MAX: 6213 case _PC_MAC_PRESENT: 6214 return (zfs_freebsd_pathconf(ap)); 6215 default: 6216 return (fifo_specops.vop_pathconf(ap)); 6217 } 6218} 6219 6220/* 6221 * FreeBSD's extended attributes namespace defines file name prefix for ZFS' 6222 * extended attribute name: 6223 * 6224 * NAMESPACE PREFIX 6225 * system freebsd:system: 6226 * user (none, can be used to access ZFS fsattr(5) attributes 6227 * created on Solaris) 6228 */ 6229static int 6230zfs_create_attrname(int attrnamespace, const char *name, char *attrname, 6231 size_t size) 6232{ 6233 const char *namespace, *prefix, *suffix; 6234 6235 /* We don't allow '/' character in attribute name. */ 6236 if (strchr(name, '/') != NULL) 6237 return (EINVAL); 6238 /* We don't allow attribute names that start with "freebsd:" string. */ 6239 if (strncmp(name, "freebsd:", 8) == 0) 6240 return (EINVAL); 6241 6242 bzero(attrname, size); 6243 6244 switch (attrnamespace) { 6245 case EXTATTR_NAMESPACE_USER: 6246#if 0 6247 prefix = "freebsd:"; 6248 namespace = EXTATTR_NAMESPACE_USER_STRING; 6249 suffix = ":"; 6250#else 6251 /* 6252 * This is the default namespace by which we can access all 6253 * attributes created on Solaris. 6254 */ 6255 prefix = namespace = suffix = ""; 6256#endif 6257 break; 6258 case EXTATTR_NAMESPACE_SYSTEM: 6259 prefix = "freebsd:"; 6260 namespace = EXTATTR_NAMESPACE_SYSTEM_STRING; 6261 suffix = ":"; 6262 break; 6263 case EXTATTR_NAMESPACE_EMPTY: 6264 default: 6265 return (EINVAL); 6266 } 6267 if (snprintf(attrname, size, "%s%s%s%s", prefix, namespace, suffix, 6268 name) >= size) { 6269 return (ENAMETOOLONG); 6270 } 6271 return (0); 6272} 6273 6274/* 6275 * Vnode operating to retrieve a named extended attribute. 6276 */ 6277static int 6278zfs_getextattr(struct vop_getextattr_args *ap) 6279/* 6280vop_getextattr { 6281 IN struct vnode *a_vp; 6282 IN int a_attrnamespace; 6283 IN const char *a_name; 6284 INOUT struct uio *a_uio; 6285 OUT size_t *a_size; 6286 IN struct ucred *a_cred; 6287 IN struct thread *a_td; 6288}; 6289*/ 6290{ 6291 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6292 struct thread *td = ap->a_td; 6293 struct nameidata nd; 6294 char attrname[255]; 6295 struct vattr va; 6296 vnode_t *xvp = NULL, *vp; 6297 int error, flags; 6298 6299 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6300 ap->a_cred, ap->a_td, VREAD); 6301 if (error != 0) 6302 return (error); 6303 6304 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6305 sizeof(attrname)); 6306 if (error != 0) 6307 return (error); 6308 6309 ZFS_ENTER(zfsvfs); 6310 6311 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6312 LOOKUP_XATTR); 6313 if (error != 0) { 6314 ZFS_EXIT(zfsvfs); 6315 return (error); 6316 } 6317 6318 flags = FREAD; 6319 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname, 6320 xvp, td); 6321 error = vn_open_cred(&nd, &flags, 0, 0, ap->a_cred, NULL); 6322 vp = nd.ni_vp; 6323 NDFREE(&nd, NDF_ONLY_PNBUF); 6324 if (error != 0) { 6325 ZFS_EXIT(zfsvfs); 6326 if (error == ENOENT) 6327 error = ENOATTR; 6328 return (error); 6329 } 6330 6331 if (ap->a_size != NULL) { 6332 error = VOP_GETATTR(vp, &va, ap->a_cred); 6333 if (error == 0) 6334 *ap->a_size = (size_t)va.va_size; 6335 } else if (ap->a_uio != NULL) 6336 error = VOP_READ(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred); 6337 6338 VOP_UNLOCK(vp, 0); 6339 vn_close(vp, flags, ap->a_cred, td); 6340 ZFS_EXIT(zfsvfs); 6341 6342 return (error); 6343} 6344 6345/* 6346 * Vnode operation to remove a named attribute. 6347 */ 6348int 6349zfs_deleteextattr(struct vop_deleteextattr_args *ap) 6350/* 6351vop_deleteextattr { 6352 IN struct vnode *a_vp; 6353 IN int a_attrnamespace; 6354 IN const char *a_name; 6355 IN struct ucred *a_cred; 6356 IN struct thread *a_td; 6357}; 6358*/ 6359{ 6360 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6361 struct thread *td = ap->a_td; 6362 struct nameidata nd; 6363 char attrname[255]; 6364 struct vattr va; 6365 vnode_t *xvp = NULL, *vp; 6366 int error, flags; 6367 6368 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6369 ap->a_cred, ap->a_td, VWRITE); 6370 if (error != 0) 6371 return (error); 6372 6373 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6374 sizeof(attrname)); 6375 if (error != 0) 6376 return (error); 6377 6378 ZFS_ENTER(zfsvfs); 6379 6380 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6381 LOOKUP_XATTR); 6382 if (error != 0) { 6383 ZFS_EXIT(zfsvfs); 6384 return (error); 6385 } 6386 6387 NDINIT_ATVP(&nd, DELETE, NOFOLLOW | LOCKPARENT | LOCKLEAF | MPSAFE, 6388 UIO_SYSSPACE, attrname, xvp, td); 6389 error = namei(&nd); 6390 vp = nd.ni_vp; 6391 NDFREE(&nd, NDF_ONLY_PNBUF); 6392 if (error != 0) { 6393 ZFS_EXIT(zfsvfs); 6394 if (error == ENOENT) 6395 error = ENOATTR; 6396 return (error); 6397 } 6398 error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); 6399 6400 vput(nd.ni_dvp); 6401 if (vp == nd.ni_dvp) 6402 vrele(vp); 6403 else 6404 vput(vp); 6405 ZFS_EXIT(zfsvfs); 6406 6407 return (error); 6408} 6409 6410/* 6411 * Vnode operation to set a named attribute. 6412 */ 6413static int 6414zfs_setextattr(struct vop_setextattr_args *ap) 6415/* 6416vop_setextattr { 6417 IN struct vnode *a_vp; 6418 IN int a_attrnamespace; 6419 IN const char *a_name; 6420 INOUT struct uio *a_uio; 6421 IN struct ucred *a_cred; 6422 IN struct thread *a_td; 6423}; 6424*/ 6425{ 6426 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6427 struct thread *td = ap->a_td; 6428 struct nameidata nd; 6429 char attrname[255]; 6430 struct vattr va; 6431 vnode_t *xvp = NULL, *vp; 6432 int error, flags; 6433 6434 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6435 ap->a_cred, ap->a_td, VWRITE); 6436 if (error != 0) 6437 return (error); 6438 6439 error = zfs_create_attrname(ap->a_attrnamespace, ap->a_name, attrname, 6440 sizeof(attrname)); 6441 if (error != 0) 6442 return (error); 6443 6444 ZFS_ENTER(zfsvfs); 6445 6446 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6447 LOOKUP_XATTR | CREATE_XATTR_DIR); 6448 if (error != 0) { 6449 ZFS_EXIT(zfsvfs); 6450 return (error); 6451 } 6452 6453 flags = FFLAGS(O_WRONLY | O_CREAT); 6454 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_SYSSPACE, attrname, 6455 xvp, td); 6456 error = vn_open_cred(&nd, &flags, 0600, 0, ap->a_cred, NULL); 6457 vp = nd.ni_vp; 6458 NDFREE(&nd, NDF_ONLY_PNBUF); 6459 if (error != 0) { 6460 ZFS_EXIT(zfsvfs); 6461 return (error); 6462 } 6463 6464 VATTR_NULL(&va); 6465 va.va_size = 0; 6466 error = VOP_SETATTR(vp, &va, ap->a_cred); 6467 if (error == 0) 6468 VOP_WRITE(vp, ap->a_uio, IO_UNIT | IO_SYNC, ap->a_cred); 6469 6470 VOP_UNLOCK(vp, 0); 6471 vn_close(vp, flags, ap->a_cred, td); 6472 ZFS_EXIT(zfsvfs); 6473 6474 return (error); 6475} 6476 6477/* 6478 * Vnode operation to retrieve extended attributes on a vnode. 6479 */ 6480static int 6481zfs_listextattr(struct vop_listextattr_args *ap) 6482/* 6483vop_listextattr { 6484 IN struct vnode *a_vp; 6485 IN int a_attrnamespace; 6486 INOUT struct uio *a_uio; 6487 OUT size_t *a_size; 6488 IN struct ucred *a_cred; 6489 IN struct thread *a_td; 6490}; 6491*/ 6492{ 6493 zfsvfs_t *zfsvfs = VTOZ(ap->a_vp)->z_zfsvfs; 6494 struct thread *td = ap->a_td; 6495 struct nameidata nd; 6496 char attrprefix[16]; 6497 u_char dirbuf[sizeof(struct dirent)]; 6498 struct dirent *dp; 6499 struct iovec aiov; 6500 struct uio auio, *uio = ap->a_uio; 6501 size_t *sizep = ap->a_size; 6502 size_t plen; 6503 vnode_t *xvp = NULL, *vp; 6504 int done, error, eof, pos; 6505 6506 error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, 6507 ap->a_cred, ap->a_td, VREAD); 6508 if (error != 0) 6509 return (error); 6510 6511 error = zfs_create_attrname(ap->a_attrnamespace, "", attrprefix, 6512 sizeof(attrprefix)); 6513 if (error != 0) 6514 return (error); 6515 plen = strlen(attrprefix); 6516 6517 ZFS_ENTER(zfsvfs); 6518 6519 if (sizep != NULL) 6520 *sizep = 0; 6521 6522 error = zfs_lookup(ap->a_vp, NULL, &xvp, NULL, 0, ap->a_cred, td, 6523 LOOKUP_XATTR); 6524 if (error != 0) { 6525 ZFS_EXIT(zfsvfs); 6526 /* 6527 * ENOATTR means that the EA directory does not yet exist, 6528 * i.e. there are no extended attributes there. 6529 */ 6530 if (error == ENOATTR) 6531 error = 0; 6532 return (error); 6533 } 6534 6535 NDINIT_ATVP(&nd, LOOKUP, NOFOLLOW | LOCKLEAF | LOCKSHARED | MPSAFE, 6536 UIO_SYSSPACE, ".", xvp, td); 6537 error = namei(&nd); 6538 vp = nd.ni_vp; 6539 NDFREE(&nd, NDF_ONLY_PNBUF); 6540 if (error != 0) { 6541 ZFS_EXIT(zfsvfs); 6542 return (error); 6543 } 6544 6545 auio.uio_iov = &aiov; 6546 auio.uio_iovcnt = 1; 6547 auio.uio_segflg = UIO_SYSSPACE; 6548 auio.uio_td = td; 6549 auio.uio_rw = UIO_READ; 6550 auio.uio_offset = 0; 6551 6552 do { 6553 u_char nlen; 6554 6555 aiov.iov_base = (void *)dirbuf; 6556 aiov.iov_len = sizeof(dirbuf); 6557 auio.uio_resid = sizeof(dirbuf); 6558 error = VOP_READDIR(vp, &auio, ap->a_cred, &eof, NULL, NULL); 6559 done = sizeof(dirbuf) - auio.uio_resid; 6560 if (error != 0) 6561 break; 6562 for (pos = 0; pos < done;) { 6563 dp = (struct dirent *)(dirbuf + pos); 6564 pos += dp->d_reclen; 6565 /* 6566 * XXX: Temporarily we also accept DT_UNKNOWN, as this 6567 * is what we get when attribute was created on Solaris. 6568 */ 6569 if (dp->d_type != DT_REG && dp->d_type != DT_UNKNOWN) 6570 continue; 6571 if (plen == 0 && strncmp(dp->d_name, "freebsd:", 8) == 0) 6572 continue; 6573 else if (strncmp(dp->d_name, attrprefix, plen) != 0) 6574 continue; 6575 nlen = dp->d_namlen - plen; 6576 if (sizep != NULL) 6577 *sizep += 1 + nlen; 6578 else if (uio != NULL) { 6579 /* 6580 * Format of extattr name entry is one byte for 6581 * length and the rest for name. 6582 */ 6583 error = uiomove(&nlen, 1, uio->uio_rw, uio); 6584 if (error == 0) { 6585 error = uiomove(dp->d_name + plen, nlen, 6586 uio->uio_rw, uio); 6587 } 6588 if (error != 0) 6589 break; 6590 } 6591 } 6592 } while (!eof && error == 0); 6593 6594 vput(vp); 6595 ZFS_EXIT(zfsvfs); 6596 6597 return (error); 6598} 6599 6600int 6601zfs_freebsd_getacl(ap) 6602 struct vop_getacl_args /* { 6603 struct vnode *vp; 6604 acl_type_t type; 6605 struct acl *aclp; 6606 struct ucred *cred; 6607 struct thread *td; 6608 } */ *ap; 6609{ 6610 int error; 6611 vsecattr_t vsecattr; 6612 6613 if (ap->a_type != ACL_TYPE_NFS4) 6614 return (EINVAL); 6615 6616 vsecattr.vsa_mask = VSA_ACE | VSA_ACECNT; 6617 if (error = zfs_getsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL)) 6618 return (error); 6619 6620 error = acl_from_aces(ap->a_aclp, vsecattr.vsa_aclentp, vsecattr.vsa_aclcnt); 6621 if (vsecattr.vsa_aclentp != NULL) 6622 kmem_free(vsecattr.vsa_aclentp, vsecattr.vsa_aclentsz); 6623 6624 return (error); 6625} 6626 6627int 6628zfs_freebsd_setacl(ap) 6629 struct vop_setacl_args /* { 6630 struct vnode *vp; 6631 acl_type_t type; 6632 struct acl *aclp; 6633 struct ucred *cred; 6634 struct thread *td; 6635 } */ *ap; 6636{ 6637 int error; 6638 vsecattr_t vsecattr; 6639 int aclbsize; /* size of acl list in bytes */ 6640 aclent_t *aaclp; 6641 6642 if (ap->a_type != ACL_TYPE_NFS4) 6643 return (EINVAL); 6644 6645 if (ap->a_aclp->acl_cnt < 1 || ap->a_aclp->acl_cnt > MAX_ACL_ENTRIES) 6646 return (EINVAL); 6647 6648 /* 6649 * With NFSv4 ACLs, chmod(2) may need to add additional entries, 6650 * splitting every entry into two and appending "canonical six" 6651 * entries at the end. Don't allow for setting an ACL that would 6652 * cause chmod(2) to run out of ACL entries. 6653 */ 6654 if (ap->a_aclp->acl_cnt * 2 + 6 > ACL_MAX_ENTRIES) 6655 return (ENOSPC); 6656 6657 error = acl_nfs4_check(ap->a_aclp, ap->a_vp->v_type == VDIR); 6658 if (error != 0) 6659 return (error); 6660 6661 vsecattr.vsa_mask = VSA_ACE; 6662 aclbsize = ap->a_aclp->acl_cnt * sizeof(ace_t); 6663 vsecattr.vsa_aclentp = kmem_alloc(aclbsize, KM_SLEEP); 6664 aaclp = vsecattr.vsa_aclentp; 6665 vsecattr.vsa_aclentsz = aclbsize; 6666 6667 aces_from_acl(vsecattr.vsa_aclentp, &vsecattr.vsa_aclcnt, ap->a_aclp); 6668 error = zfs_setsecattr(ap->a_vp, &vsecattr, 0, ap->a_cred, NULL); 6669 kmem_free(aaclp, aclbsize); 6670 6671 return (error); 6672} 6673 6674int 6675zfs_freebsd_aclcheck(ap) 6676 struct vop_aclcheck_args /* { 6677 struct vnode *vp; 6678 acl_type_t type; 6679 struct acl *aclp; 6680 struct ucred *cred; 6681 struct thread *td; 6682 } */ *ap; 6683{ 6684 6685 return (EOPNOTSUPP); 6686} 6687 6688struct vop_vector zfs_vnodeops; 6689struct vop_vector zfs_fifoops; 6690struct vop_vector zfs_shareops; 6691 6692struct vop_vector zfs_vnodeops = { 6693 .vop_default = &default_vnodeops, 6694 .vop_inactive = zfs_freebsd_inactive, 6695 .vop_reclaim = zfs_freebsd_reclaim, 6696 .vop_access = zfs_freebsd_access, 6697#ifdef FREEBSD_NAMECACHE 6698 .vop_lookup = vfs_cache_lookup, 6699 .vop_cachedlookup = zfs_freebsd_lookup, 6700#else 6701 .vop_lookup = zfs_freebsd_lookup, 6702#endif 6703 .vop_getattr = zfs_freebsd_getattr, 6704 .vop_setattr = zfs_freebsd_setattr, 6705 .vop_create = zfs_freebsd_create, 6706 .vop_mknod = zfs_freebsd_create, 6707 .vop_mkdir = zfs_freebsd_mkdir, 6708 .vop_readdir = zfs_freebsd_readdir, 6709 .vop_fsync = zfs_freebsd_fsync, 6710 .vop_open = zfs_freebsd_open, 6711 .vop_close = zfs_freebsd_close, 6712 .vop_rmdir = zfs_freebsd_rmdir, 6713 .vop_ioctl = zfs_freebsd_ioctl, 6714 .vop_link = zfs_freebsd_link, 6715 .vop_symlink = zfs_freebsd_symlink, 6716 .vop_readlink = zfs_freebsd_readlink, 6717 .vop_read = zfs_freebsd_read, 6718 .vop_write = zfs_freebsd_write, 6719 .vop_remove = zfs_freebsd_remove, 6720 .vop_rename = zfs_freebsd_rename, 6721 .vop_pathconf = zfs_freebsd_pathconf, 6722 .vop_bmap = VOP_EOPNOTSUPP, 6723 .vop_fid = zfs_freebsd_fid, 6724 .vop_getextattr = zfs_getextattr, 6725 .vop_deleteextattr = zfs_deleteextattr, 6726 .vop_setextattr = zfs_setextattr, 6727 .vop_listextattr = zfs_listextattr, 6728 .vop_getacl = zfs_freebsd_getacl, 6729 .vop_setacl = zfs_freebsd_setacl, 6730 .vop_aclcheck = zfs_freebsd_aclcheck, 6731 .vop_getpages = zfs_freebsd_getpages, 6732}; 6733 6734struct vop_vector zfs_fifoops = { 6735 .vop_default = &fifo_specops, 6736 .vop_fsync = zfs_freebsd_fsync, 6737 .vop_access = zfs_freebsd_access, 6738 .vop_getattr = zfs_freebsd_getattr, 6739 .vop_inactive = zfs_freebsd_inactive, 6740 .vop_read = VOP_PANIC, 6741 .vop_reclaim = zfs_freebsd_reclaim, 6742 .vop_setattr = zfs_freebsd_setattr, 6743 .vop_write = VOP_PANIC, 6744 .vop_pathconf = zfs_freebsd_fifo_pathconf, 6745 .vop_fid = zfs_freebsd_fid, 6746 .vop_getacl = zfs_freebsd_getacl, 6747 .vop_setacl = zfs_freebsd_setacl, 6748 .vop_aclcheck = zfs_freebsd_aclcheck, 6749}; 6750 6751/* 6752 * special share hidden files vnode operations template 6753 */ 6754struct vop_vector zfs_shareops = { 6755 .vop_default = &default_vnodeops, 6756 .vop_access = zfs_freebsd_access, 6757 .vop_inactive = zfs_freebsd_inactive, 6758 .vop_reclaim = zfs_freebsd_reclaim, 6759 .vop_fid = zfs_freebsd_fid, 6760 .vop_pathconf = zfs_freebsd_pathconf, 6761}; 6762