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/vfs_opreg.h> 36 #include <sys/vnode.h> 37 #include <sys/file.h> 38 #include <sys/stat.h> 39 #include <sys/kmem.h> 40 #include <sys/taskq.h> 41 #include <sys/uio.h> 42 #include <sys/vmsystm.h> 43 #include <sys/atomic.h> 44 #include <sys/vm.h> 45 #include <vm/seg_vn.h> 46 #include <vm/pvn.h> 47 #include <vm/as.h> 48 #include <vm/kpm.h> 49 #include <vm/seg_kpm.h> 50 #include <sys/mman.h> 51 #include <sys/pathname.h> 52 #include <sys/cmn_err.h> 53 #include <sys/errno.h> 54 #include <sys/unistd.h> 55 #include <sys/zfs_dir.h> 56 #include <sys/zfs_acl.h> 57 #include <sys/zfs_ioctl.h> 58 #include <sys/fs/zfs.h> 59 #include <sys/dmu.h> 60 #include <sys/dmu_objset.h> 61 #include <sys/spa.h> 62 #include <sys/txg.h> 63 #include <sys/dbuf.h> 64 #include <sys/zap.h> 65 #include <sys/sa.h> 66 #include <sys/dirent.h> 67 #include <sys/policy.h> 68 #include <sys/sunddi.h> 69 #include <sys/filio.h> 70 #include <sys/sid.h> 71 #include "fs/fs_subr.h" 72 #include <sys/zfs_ctldir.h> 73 #include <sys/zfs_fuid.h> 74 #include <sys/zfs_sa.h> 75 #include <sys/dnlc.h> 76 #include <sys/zfs_rlock.h> 77 #include <sys/extdirent.h> 78 #include <sys/kidmap.h> 79 #include <sys/cred.h> 80 #include <sys/attr.h> 81 82 /* 83 * Programming rules. 84 * 85 * Each vnode op performs some logical unit of work. To do this, the ZPL must 86 * properly lock its in-core state, create a DMU transaction, do the work, 87 * record this work in the intent log (ZIL), commit the DMU transaction, 88 * and wait for the intent log to commit if it is a synchronous operation. 89 * Moreover, the vnode ops must work in both normal and log replay context. 90 * The ordering of events is important to avoid deadlocks and references 91 * to freed memory. The example below illustrates the following Big Rules: 92 * 93 * (1) A check must be made in each zfs thread for a mounted file system. 94 * This is done avoiding races using ZFS_ENTER(zfsvfs). 95 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes 96 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros 97 * can return EIO from the calling function. 98 * 99 * (2) VN_RELE() should always be the last thing except for zil_commit() 100 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 101 * First, if it's the last reference, the vnode/znode 102 * can be freed, so the zp may point to freed memory. Second, the last 103 * reference will call zfs_zinactive(), which may induce a lot of work -- 104 * pushing cached pages (which acquires range locks) and syncing out 105 * cached atime changes. Third, zfs_zinactive() may require a new tx, 106 * which could deadlock the system if you were already holding one. 107 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). 108 * 109 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 110 * as they can span dmu_tx_assign() calls. 111 * 112 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign(). 113 * This is critical because we don't want to block while holding locks. 114 * Note, in particular, that if a lock is sometimes acquired before 115 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 116 * use a non-blocking assign can deadlock the system. The scenario: 117 * 118 * Thread A has grabbed a lock before calling dmu_tx_assign(). 119 * Thread B is in an already-assigned tx, and blocks for this lock. 120 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 121 * forever, because the previous txg can't quiesce until B's tx commits. 122 * 123 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 124 * then drop all locks, call dmu_tx_wait(), and try again. 125 * 126 * (5) If the operation succeeded, generate the intent log entry for it 127 * before dropping locks. This ensures that the ordering of events 128 * in the intent log matches the order in which they actually occurred. 129 * During ZIL replay the zfs_log_* functions will update the sequence 130 * number to indicate the zil transaction has replayed. 131 * 132 * (6) At the end of each vnode op, the DMU tx must always commit, 133 * regardless of whether there were any errors. 134 * 135 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 136 * to ensure that synchronous semantics are provided when necessary. 137 * 138 * In general, this is how things should be ordered in each vnode op: 139 * 140 * ZFS_ENTER(zfsvfs); // exit if unmounted 141 * top: 142 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 143 * rw_enter(...); // grab any other locks you need 144 * tx = dmu_tx_create(...); // get DMU tx 145 * dmu_tx_hold_*(); // hold each object you might modify 146 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign 147 * if (error) { 148 * rw_exit(...); // drop locks 149 * zfs_dirent_unlock(dl); // unlock directory entry 150 * VN_RELE(...); // release held vnodes 151 * if (error == ERESTART) { 152 * dmu_tx_wait(tx); 153 * dmu_tx_abort(tx); 154 * goto top; 155 * } 156 * dmu_tx_abort(tx); // abort DMU tx 157 * ZFS_EXIT(zfsvfs); // finished in zfs 158 * return (error); // really out of space 159 * } 160 * error = do_real_work(); // do whatever this VOP does 161 * if (error == 0) 162 * zfs_log_*(...); // on success, make ZIL entry 163 * dmu_tx_commit(tx); // commit DMU tx -- error or not 164 * rw_exit(...); // drop locks 165 * zfs_dirent_unlock(dl); // unlock directory entry 166 * VN_RELE(...); // release held vnodes 167 * zil_commit(zilog, foid); // synchronous when necessary 168 * ZFS_EXIT(zfsvfs); // finished in zfs 169 * return (error); // done, report error 170 */ 171 172 /* ARGSUSED */ 173 static int 174 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 175 { 176 znode_t *zp = VTOZ(*vpp); 177 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 178 179 ZFS_ENTER(zfsvfs); 180 ZFS_VERIFY_ZP(zp); 181 182 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 183 ((flag & FAPPEND) == 0)) { 184 ZFS_EXIT(zfsvfs); 185 return (EPERM); 186 } 187 188 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 189 ZTOV(zp)->v_type == VREG && 190 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) { 191 if (fs_vscan(*vpp, cr, 0) != 0) { 192 ZFS_EXIT(zfsvfs); 193 return (EACCES); 194 } 195 } 196 197 /* Keep a count of the synchronous opens in the znode */ 198 if (flag & (FSYNC | FDSYNC)) 199 atomic_inc_32(&zp->z_sync_cnt); 200 201 ZFS_EXIT(zfsvfs); 202 return (0); 203 } 204 205 /* ARGSUSED */ 206 static int 207 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 208 caller_context_t *ct) 209 { 210 znode_t *zp = VTOZ(vp); 211 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 212 213 /* 214 * Clean up any locks held by this process on the vp. 215 */ 216 cleanlocks(vp, ddi_get_pid(), 0); 217 cleanshares(vp, ddi_get_pid()); 218 219 ZFS_ENTER(zfsvfs); 220 ZFS_VERIFY_ZP(zp); 221 222 /* Decrement the synchronous opens in the znode */ 223 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 224 atomic_dec_32(&zp->z_sync_cnt); 225 226 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 227 ZTOV(zp)->v_type == VREG && 228 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) 229 VERIFY(fs_vscan(vp, cr, 1) == 0); 230 231 ZFS_EXIT(zfsvfs); 232 return (0); 233 } 234 235 /* 236 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 237 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 238 */ 239 static int 240 zfs_holey(vnode_t *vp, int cmd, offset_t *off) 241 { 242 znode_t *zp = VTOZ(vp); 243 uint64_t noff = (uint64_t)*off; /* new offset */ 244 uint64_t file_sz; 245 int error; 246 boolean_t hole; 247 248 file_sz = zp->z_size; 249 if (noff >= file_sz) { 250 return (ENXIO); 251 } 252 253 if (cmd == _FIO_SEEK_HOLE) 254 hole = B_TRUE; 255 else 256 hole = B_FALSE; 257 258 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 259 260 /* end of file? */ 261 if ((error == ESRCH) || (noff > file_sz)) { 262 /* 263 * Handle the virtual hole at the end of file. 264 */ 265 if (hole) { 266 *off = file_sz; 267 return (0); 268 } 269 return (ENXIO); 270 } 271 272 if (noff < *off) 273 return (error); 274 *off = noff; 275 return (error); 276 } 277 278 /* ARGSUSED */ 279 static int 280 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred, 281 int *rvalp, caller_context_t *ct) 282 { 283 offset_t off; 284 int error; 285 zfsvfs_t *zfsvfs; 286 znode_t *zp; 287 288 switch (com) { 289 case _FIOFFS: 290 return (zfs_sync(vp->v_vfsp, 0, cred)); 291 292 /* 293 * The following two ioctls are used by bfu. Faking out, 294 * necessary to avoid bfu errors. 295 */ 296 case _FIOGDIO: 297 case _FIOSDIO: 298 return (0); 299 300 case _FIO_SEEK_DATA: 301 case _FIO_SEEK_HOLE: 302 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 303 return (EFAULT); 304 305 zp = VTOZ(vp); 306 zfsvfs = zp->z_zfsvfs; 307 ZFS_ENTER(zfsvfs); 308 ZFS_VERIFY_ZP(zp); 309 310 /* offset parameter is in/out */ 311 error = zfs_holey(vp, com, &off); 312 ZFS_EXIT(zfsvfs); 313 if (error) 314 return (error); 315 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 316 return (EFAULT); 317 return (0); 318 } 319 return (ENOTTY); 320 } 321 322 /* 323 * Utility functions to map and unmap a single physical page. These 324 * are used to manage the mappable copies of ZFS file data, and therefore 325 * do not update ref/mod bits. 326 */ 327 caddr_t 328 zfs_map_page(page_t *pp, enum seg_rw rw) 329 { 330 if (kpm_enable) 331 return (hat_kpm_mapin(pp, 0)); 332 ASSERT(rw == S_READ || rw == S_WRITE); 333 return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0), 334 (caddr_t)-1)); 335 } 336 337 void 338 zfs_unmap_page(page_t *pp, caddr_t addr) 339 { 340 if (kpm_enable) { 341 hat_kpm_mapout(pp, 0, addr); 342 } else { 343 ppmapout(addr); 344 } 345 } 346 347 /* 348 * When a file is memory mapped, we must keep the IO data synchronized 349 * between the DMU cache and the memory mapped pages. What this means: 350 * 351 * On Write: If we find a memory mapped page, we write to *both* 352 * the page and the dmu buffer. 353 */ 354 static void 355 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid) 356 { 357 int64_t off; 358 359 off = start & PAGEOFFSET; 360 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 361 page_t *pp; 362 uint64_t nbytes = MIN(PAGESIZE - off, len); 363 364 if (pp = page_lookup(vp, start, SE_SHARED)) { 365 caddr_t va; 366 367 va = zfs_map_page(pp, S_WRITE); 368 (void) dmu_read(os, oid, start+off, nbytes, va+off, 369 DMU_READ_PREFETCH); 370 zfs_unmap_page(pp, va); 371 page_unlock(pp); 372 } 373 len -= nbytes; 374 off = 0; 375 } 376 } 377 378 /* 379 * When a file is memory mapped, we must keep the IO data synchronized 380 * between the DMU cache and the memory mapped pages. What this means: 381 * 382 * On Read: We "read" preferentially from memory mapped pages, 383 * else we default from the dmu buffer. 384 * 385 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 386 * the file is memory mapped. 387 */ 388 static int 389 mappedread(vnode_t *vp, int nbytes, uio_t *uio) 390 { 391 znode_t *zp = VTOZ(vp); 392 objset_t *os = zp->z_zfsvfs->z_os; 393 int64_t start, off; 394 int len = nbytes; 395 int error = 0; 396 397 start = uio->uio_loffset; 398 off = start & PAGEOFFSET; 399 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 400 page_t *pp; 401 uint64_t bytes = MIN(PAGESIZE - off, len); 402 403 if (pp = page_lookup(vp, start, SE_SHARED)) { 404 caddr_t va; 405 406 va = zfs_map_page(pp, S_READ); 407 error = uiomove(va + off, bytes, UIO_READ, uio); 408 zfs_unmap_page(pp, va); 409 page_unlock(pp); 410 } else { 411 error = dmu_read_uio(os, zp->z_id, uio, bytes); 412 } 413 len -= bytes; 414 off = 0; 415 if (error) 416 break; 417 } 418 return (error); 419 } 420 421 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 422 423 /* 424 * Read bytes from specified file into supplied buffer. 425 * 426 * IN: vp - vnode of file to be read from. 427 * uio - structure supplying read location, range info, 428 * and return buffer. 429 * ioflag - SYNC flags; used to provide FRSYNC semantics. 430 * cr - credentials of caller. 431 * ct - caller context 432 * 433 * OUT: uio - updated offset and range, buffer filled. 434 * 435 * RETURN: 0 if success 436 * error code if failure 437 * 438 * Side Effects: 439 * vp - atime updated if byte count > 0 440 */ 441 /* ARGSUSED */ 442 static int 443 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 444 { 445 znode_t *zp = VTOZ(vp); 446 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 447 objset_t *os; 448 ssize_t n, nbytes; 449 int error; 450 rl_t *rl; 451 xuio_t *xuio = NULL; 452 453 ZFS_ENTER(zfsvfs); 454 ZFS_VERIFY_ZP(zp); 455 os = zfsvfs->z_os; 456 457 if (zp->z_pflags & ZFS_AV_QUARANTINED) { 458 ZFS_EXIT(zfsvfs); 459 return (EACCES); 460 } 461 462 /* 463 * Validate file offset 464 */ 465 if (uio->uio_loffset < (offset_t)0) { 466 ZFS_EXIT(zfsvfs); 467 return (EINVAL); 468 } 469 470 /* 471 * Fasttrack empty reads 472 */ 473 if (uio->uio_resid == 0) { 474 ZFS_EXIT(zfsvfs); 475 return (0); 476 } 477 478 /* 479 * Check for mandatory locks 480 */ 481 if (MANDMODE(zp->z_mode)) { 482 if (error = chklock(vp, FREAD, 483 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 484 ZFS_EXIT(zfsvfs); 485 return (error); 486 } 487 } 488 489 /* 490 * If we're in FRSYNC mode, sync out this znode before reading it. 491 */ 492 if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 493 zil_commit(zfsvfs->z_log, zp->z_id); 494 495 /* 496 * Lock the range against changes. 497 */ 498 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 499 500 /* 501 * If we are reading past end-of-file we can skip 502 * to the end; but we might still need to set atime. 503 */ 504 if (uio->uio_loffset >= zp->z_size) { 505 error = 0; 506 goto out; 507 } 508 509 ASSERT(uio->uio_loffset < zp->z_size); 510 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset); 511 512 if ((uio->uio_extflg == UIO_XUIO) && 513 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) { 514 int nblk; 515 int blksz = zp->z_blksz; 516 uint64_t offset = uio->uio_loffset; 517 518 xuio = (xuio_t *)uio; 519 if ((ISP2(blksz))) { 520 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset, 521 blksz)) / blksz; 522 } else { 523 ASSERT(offset + n <= blksz); 524 nblk = 1; 525 } 526 (void) dmu_xuio_init(xuio, nblk); 527 528 if (vn_has_cached_data(vp)) { 529 /* 530 * For simplicity, we always allocate a full buffer 531 * even if we only expect to read a portion of a block. 532 */ 533 while (--nblk >= 0) { 534 (void) dmu_xuio_add(xuio, 535 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 536 blksz), 0, blksz); 537 } 538 } 539 } 540 541 while (n > 0) { 542 nbytes = MIN(n, zfs_read_chunk_size - 543 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 544 545 if (vn_has_cached_data(vp)) 546 error = mappedread(vp, nbytes, uio); 547 else 548 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 549 if (error) { 550 /* convert checksum errors into IO errors */ 551 if (error == ECKSUM) 552 error = EIO; 553 break; 554 } 555 556 n -= nbytes; 557 } 558 out: 559 zfs_range_unlock(rl); 560 561 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 562 ZFS_EXIT(zfsvfs); 563 return (error); 564 } 565 566 /* 567 * Write the bytes to a file. 568 * 569 * IN: vp - vnode of file to be written to. 570 * uio - structure supplying write location, range info, 571 * and data buffer. 572 * ioflag - FAPPEND flag set if in append mode. 573 * cr - credentials of caller. 574 * ct - caller context (NFS/CIFS fem monitor only) 575 * 576 * OUT: uio - updated offset and range. 577 * 578 * RETURN: 0 if success 579 * error code if failure 580 * 581 * Timestamps: 582 * vp - ctime|mtime updated if byte count > 0 583 */ 584 585 /* ARGSUSED */ 586 static int 587 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 588 { 589 znode_t *zp = VTOZ(vp); 590 rlim64_t limit = uio->uio_llimit; 591 ssize_t start_resid = uio->uio_resid; 592 ssize_t tx_bytes; 593 uint64_t end_size; 594 dmu_tx_t *tx; 595 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 596 zilog_t *zilog; 597 offset_t woff; 598 ssize_t n, nbytes; 599 rl_t *rl; 600 int max_blksz = zfsvfs->z_max_blksz; 601 int error; 602 arc_buf_t *abuf; 603 iovec_t *aiov; 604 xuio_t *xuio = NULL; 605 int i_iov = 0; 606 int iovcnt = uio->uio_iovcnt; 607 iovec_t *iovp = uio->uio_iov; 608 int write_eof; 609 int count = 0; 610 sa_bulk_attr_t bulk[4]; 611 uint64_t mtime[2], ctime[2]; 612 613 /* 614 * Fasttrack empty write 615 */ 616 n = start_resid; 617 if (n == 0) 618 return (0); 619 620 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 621 limit = MAXOFFSET_T; 622 623 ZFS_ENTER(zfsvfs); 624 ZFS_VERIFY_ZP(zp); 625 626 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 627 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 628 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 629 &zp->z_size, 8); 630 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 631 &zp->z_pflags, 8); 632 633 /* 634 * If immutable or not appending then return EPERM 635 */ 636 if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) || 637 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 638 (uio->uio_loffset < zp->z_size))) { 639 ZFS_EXIT(zfsvfs); 640 return (EPERM); 641 } 642 643 zilog = zfsvfs->z_log; 644 645 /* 646 * Validate file offset 647 */ 648 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset; 649 if (woff < 0) { 650 ZFS_EXIT(zfsvfs); 651 return (EINVAL); 652 } 653 654 /* 655 * Check for mandatory locks before calling zfs_range_lock() 656 * in order to prevent a deadlock with locks set via fcntl(). 657 */ 658 if (MANDMODE((mode_t)zp->z_mode) && 659 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 660 ZFS_EXIT(zfsvfs); 661 return (error); 662 } 663 664 /* 665 * Pre-fault the pages to ensure slow (eg NFS) pages 666 * don't hold up txg. 667 * Skip this if uio contains loaned arc_buf. 668 */ 669 if ((uio->uio_extflg == UIO_XUIO) && 670 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) 671 xuio = (xuio_t *)uio; 672 else 673 uio_prefaultpages(MIN(n, max_blksz), uio); 674 675 /* 676 * If in append mode, set the io offset pointer to eof. 677 */ 678 if (ioflag & FAPPEND) { 679 /* 680 * Obtain an appending range lock to guarantee file append 681 * semantics. We reset the write offset once we have the lock. 682 */ 683 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 684 woff = rl->r_off; 685 if (rl->r_len == UINT64_MAX) { 686 /* 687 * We overlocked the file because this write will cause 688 * the file block size to increase. 689 * Note that zp_size cannot change with this lock held. 690 */ 691 woff = zp->z_size; 692 } 693 uio->uio_loffset = woff; 694 } else { 695 /* 696 * Note that if the file block size will change as a result of 697 * this write, then this range lock will lock the entire file 698 * so that we can re-write the block safely. 699 */ 700 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 701 } 702 703 if (woff >= limit) { 704 zfs_range_unlock(rl); 705 ZFS_EXIT(zfsvfs); 706 return (EFBIG); 707 } 708 709 if ((woff + n) > limit || woff > (limit - n)) 710 n = limit - woff; 711 712 /* Will this write extend the file length? */ 713 write_eof = (woff + n > zp->z_size); 714 715 end_size = MAX(zp->z_size, woff + n); 716 717 /* 718 * Write the file in reasonable size chunks. Each chunk is written 719 * in a separate transaction; this keeps the intent log records small 720 * and allows us to do more fine-grained space accounting. 721 */ 722 while (n > 0) { 723 abuf = NULL; 724 woff = uio->uio_loffset; 725 again: 726 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 727 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 728 if (abuf != NULL) 729 dmu_return_arcbuf(abuf); 730 error = EDQUOT; 731 break; 732 } 733 734 if (xuio && abuf == NULL) { 735 ASSERT(i_iov < iovcnt); 736 aiov = &iovp[i_iov]; 737 abuf = dmu_xuio_arcbuf(xuio, i_iov); 738 dmu_xuio_clear(xuio, i_iov); 739 DTRACE_PROBE3(zfs_cp_write, int, i_iov, 740 iovec_t *, aiov, arc_buf_t *, abuf); 741 ASSERT((aiov->iov_base == abuf->b_data) || 742 ((char *)aiov->iov_base - (char *)abuf->b_data + 743 aiov->iov_len == arc_buf_size(abuf))); 744 i_iov++; 745 } else if (abuf == NULL && n >= max_blksz && 746 woff >= zp->z_size && 747 P2PHASE(woff, max_blksz) == 0 && 748 zp->z_blksz == max_blksz) { 749 /* 750 * This write covers a full block. "Borrow" a buffer 751 * from the dmu so that we can fill it before we enter 752 * a transaction. This avoids the possibility of 753 * holding up the transaction if the data copy hangs 754 * up on a pagefault (e.g., from an NFS server mapping). 755 */ 756 size_t cbytes; 757 758 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 759 max_blksz); 760 ASSERT(abuf != NULL); 761 ASSERT(arc_buf_size(abuf) == max_blksz); 762 if (error = uiocopy(abuf->b_data, max_blksz, 763 UIO_WRITE, uio, &cbytes)) { 764 dmu_return_arcbuf(abuf); 765 break; 766 } 767 ASSERT(cbytes == max_blksz); 768 } 769 770 /* 771 * Start a transaction. 772 */ 773 tx = dmu_tx_create(zfsvfs->z_os); 774 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 775 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 776 zfs_sa_upgrade_txholds(tx, zp); 777 error = dmu_tx_assign(tx, TXG_NOWAIT); 778 if (error) { 779 if (error == ERESTART) { 780 dmu_tx_wait(tx); 781 dmu_tx_abort(tx); 782 goto again; 783 } 784 dmu_tx_abort(tx); 785 if (abuf != NULL) 786 dmu_return_arcbuf(abuf); 787 break; 788 } 789 790 /* 791 * If zfs_range_lock() over-locked we grow the blocksize 792 * and then reduce the lock range. This will only happen 793 * on the first iteration since zfs_range_reduce() will 794 * shrink down r_len to the appropriate size. 795 */ 796 if (rl->r_len == UINT64_MAX) { 797 uint64_t new_blksz; 798 799 if (zp->z_blksz > max_blksz) { 800 ASSERT(!ISP2(zp->z_blksz)); 801 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 802 } else { 803 new_blksz = MIN(end_size, max_blksz); 804 } 805 zfs_grow_blocksize(zp, new_blksz, tx); 806 zfs_range_reduce(rl, woff, n); 807 } 808 809 /* 810 * XXX - should we really limit each write to z_max_blksz? 811 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 812 */ 813 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 814 815 if (abuf == NULL) { 816 tx_bytes = uio->uio_resid; 817 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), 818 uio, nbytes, tx); 819 tx_bytes -= uio->uio_resid; 820 } else { 821 tx_bytes = nbytes; 822 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len); 823 /* 824 * If this is not a full block write, but we are 825 * extending the file past EOF and this data starts 826 * block-aligned, use assign_arcbuf(). Otherwise, 827 * write via dmu_write(). 828 */ 829 if (tx_bytes < max_blksz && (!write_eof || 830 aiov->iov_base != abuf->b_data)) { 831 ASSERT(xuio); 832 dmu_write(zfsvfs->z_os, zp->z_id, woff, 833 aiov->iov_len, aiov->iov_base, tx); 834 dmu_return_arcbuf(abuf); 835 xuio_stat_wbuf_copied(); 836 } else { 837 ASSERT(xuio || tx_bytes == max_blksz); 838 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), 839 woff, abuf, tx); 840 } 841 ASSERT(tx_bytes <= uio->uio_resid); 842 uioskip(uio, tx_bytes); 843 } 844 if (tx_bytes && vn_has_cached_data(vp)) { 845 update_pages(vp, woff, 846 tx_bytes, zfsvfs->z_os, zp->z_id); 847 } 848 849 /* 850 * If we made no progress, we're done. If we made even 851 * partial progress, update the znode and ZIL accordingly. 852 */ 853 if (tx_bytes == 0) { 854 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 855 (void *)&zp->z_size, sizeof (uint64_t), tx); 856 dmu_tx_commit(tx); 857 ASSERT(error != 0); 858 break; 859 } 860 861 /* 862 * Clear Set-UID/Set-GID bits on successful write if not 863 * privileged and at least one of the excute bits is set. 864 * 865 * It would be nice to to this after all writes have 866 * been done, but that would still expose the ISUID/ISGID 867 * to another app after the partial write is committed. 868 * 869 */ 870 mutex_enter(&zp->z_acl_lock); 871 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | 872 (S_IXUSR >> 6))) != 0 && 873 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && 874 secpolicy_vnode_setid_retain(cr, 875 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) { 876 uint64_t newmode; 877 zp->z_mode &= ~(S_ISUID | S_ISGID); 878 newmode = zp->z_mode; 879 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), 880 (void *)&newmode, sizeof (uint64_t), tx); 881 } 882 mutex_exit(&zp->z_acl_lock); 883 884 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 885 B_TRUE); 886 887 /* 888 * Update the file size (zp_size) if it has changed; 889 * account for possible concurrent updates. 890 */ 891 while ((end_size = zp->z_size) < uio->uio_loffset) { 892 (void) atomic_cas_64(&zp->z_size, end_size, 893 uio->uio_loffset); 894 ASSERT(error == 0); 895 } 896 /* 897 * If we are replaying and eof is non zero then force 898 * the file size to the specified eof. Note, there's no 899 * concurrency during replay. 900 */ 901 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) 902 zp->z_size = zfsvfs->z_replay_eof; 903 904 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 905 906 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 907 dmu_tx_commit(tx); 908 909 if (error != 0) 910 break; 911 ASSERT(tx_bytes == nbytes); 912 n -= nbytes; 913 914 if (!xuio && n > 0) 915 uio_prefaultpages(MIN(n, max_blksz), uio); 916 } 917 918 zfs_range_unlock(rl); 919 920 /* 921 * If we're in replay mode, or we made no progress, return error. 922 * Otherwise, it's at least a partial write, so it's successful. 923 */ 924 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 925 ZFS_EXIT(zfsvfs); 926 return (error); 927 } 928 929 if (ioflag & (FSYNC | FDSYNC) || 930 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 931 zil_commit(zilog, zp->z_id); 932 933 ZFS_EXIT(zfsvfs); 934 return (0); 935 } 936 937 void 938 zfs_get_done(zgd_t *zgd, int error) 939 { 940 znode_t *zp = zgd->zgd_private; 941 objset_t *os = zp->z_zfsvfs->z_os; 942 943 if (zgd->zgd_db) 944 dmu_buf_rele(zgd->zgd_db, zgd); 945 946 zfs_range_unlock(zgd->zgd_rl); 947 948 /* 949 * Release the vnode asynchronously as we currently have the 950 * txg stopped from syncing. 951 */ 952 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 953 954 if (error == 0 && zgd->zgd_bp) 955 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 956 957 kmem_free(zgd, sizeof (zgd_t)); 958 } 959 960 #ifdef DEBUG 961 static int zil_fault_io = 0; 962 #endif 963 964 /* 965 * Get data to generate a TX_WRITE intent log record. 966 */ 967 int 968 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 969 { 970 zfsvfs_t *zfsvfs = arg; 971 objset_t *os = zfsvfs->z_os; 972 znode_t *zp; 973 uint64_t object = lr->lr_foid; 974 uint64_t offset = lr->lr_offset; 975 uint64_t size = lr->lr_length; 976 blkptr_t *bp = &lr->lr_blkptr; 977 dmu_buf_t *db; 978 zgd_t *zgd; 979 int error = 0; 980 981 ASSERT(zio != NULL); 982 ASSERT(size != 0); 983 984 /* 985 * Nothing to do if the file has been removed 986 */ 987 if (zfs_zget(zfsvfs, object, &zp) != 0) 988 return (ENOENT); 989 if (zp->z_unlinked) { 990 /* 991 * Release the vnode asynchronously as we currently have the 992 * txg stopped from syncing. 993 */ 994 VN_RELE_ASYNC(ZTOV(zp), 995 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 996 return (ENOENT); 997 } 998 999 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1000 zgd->zgd_zilog = zfsvfs->z_log; 1001 zgd->zgd_private = zp; 1002 1003 /* 1004 * Write records come in two flavors: immediate and indirect. 1005 * For small writes it's cheaper to store the data with the 1006 * log record (immediate); for large writes it's cheaper to 1007 * sync the data and get a pointer to it (indirect) so that 1008 * we don't have to write the data twice. 1009 */ 1010 if (buf != NULL) { /* immediate write */ 1011 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER); 1012 /* test for truncation needs to be done while range locked */ 1013 if (offset >= zp->z_size) { 1014 error = ENOENT; 1015 } else { 1016 error = dmu_read(os, object, offset, size, buf, 1017 DMU_READ_NO_PREFETCH); 1018 } 1019 ASSERT(error == 0 || error == ENOENT); 1020 } else { /* indirect write */ 1021 /* 1022 * Have to lock the whole block to ensure when it's 1023 * written out and it's checksum is being calculated 1024 * that no one can change the data. We need to re-check 1025 * blocksize after we get the lock in case it's changed! 1026 */ 1027 for (;;) { 1028 uint64_t blkoff; 1029 size = zp->z_blksz; 1030 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; 1031 offset -= blkoff; 1032 zgd->zgd_rl = zfs_range_lock(zp, offset, size, 1033 RL_READER); 1034 if (zp->z_blksz == size) 1035 break; 1036 offset += blkoff; 1037 zfs_range_unlock(zgd->zgd_rl); 1038 } 1039 /* test for truncation needs to be done while range locked */ 1040 if (lr->lr_offset >= zp->z_size) 1041 error = ENOENT; 1042 #ifdef DEBUG 1043 if (zil_fault_io) { 1044 error = EIO; 1045 zil_fault_io = 0; 1046 } 1047 #endif 1048 if (error == 0) 1049 error = dmu_buf_hold(os, object, offset, zgd, &db, 1050 DMU_READ_NO_PREFETCH); 1051 1052 if (error == 0) { 1053 zgd->zgd_db = db; 1054 zgd->zgd_bp = bp; 1055 1056 ASSERT(db->db_offset == offset); 1057 ASSERT(db->db_size == size); 1058 1059 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1060 zfs_get_done, zgd); 1061 ASSERT(error || lr->lr_length <= zp->z_blksz); 1062 1063 /* 1064 * On success, we need to wait for the write I/O 1065 * initiated by dmu_sync() to complete before we can 1066 * release this dbuf. We will finish everything up 1067 * in the zfs_get_done() callback. 1068 */ 1069 if (error == 0) 1070 return (0); 1071 1072 if (error == EALREADY) { 1073 lr->lr_common.lrc_txtype = TX_WRITE2; 1074 error = 0; 1075 } 1076 } 1077 } 1078 1079 zfs_get_done(zgd, error); 1080 1081 return (error); 1082 } 1083 1084 /*ARGSUSED*/ 1085 static int 1086 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 1087 caller_context_t *ct) 1088 { 1089 znode_t *zp = VTOZ(vp); 1090 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1091 int error; 1092 1093 ZFS_ENTER(zfsvfs); 1094 ZFS_VERIFY_ZP(zp); 1095 1096 if (flag & V_ACE_MASK) 1097 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 1098 else 1099 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1100 1101 ZFS_EXIT(zfsvfs); 1102 return (error); 1103 } 1104 1105 /* 1106 * If vnode is for a device return a specfs vnode instead. 1107 */ 1108 static int 1109 specvp_check(vnode_t **vpp, cred_t *cr) 1110 { 1111 int error = 0; 1112 1113 if (IS_DEVVP(*vpp)) { 1114 struct vnode *svp; 1115 1116 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1117 VN_RELE(*vpp); 1118 if (svp == NULL) 1119 error = ENOSYS; 1120 *vpp = svp; 1121 } 1122 return (error); 1123 } 1124 1125 1126 /* 1127 * Lookup an entry in a directory, or an extended attribute directory. 1128 * If it exists, return a held vnode reference for it. 1129 * 1130 * IN: dvp - vnode of directory to search. 1131 * nm - name of entry to lookup. 1132 * pnp - full pathname to lookup [UNUSED]. 1133 * flags - LOOKUP_XATTR set if looking for an attribute. 1134 * rdir - root directory vnode [UNUSED]. 1135 * cr - credentials of caller. 1136 * ct - caller context 1137 * direntflags - directory lookup flags 1138 * realpnp - returned pathname. 1139 * 1140 * OUT: vpp - vnode of located entry, NULL if not found. 1141 * 1142 * RETURN: 0 if success 1143 * error code if failure 1144 * 1145 * Timestamps: 1146 * NA 1147 */ 1148 /* ARGSUSED */ 1149 static int 1150 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 1151 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 1152 int *direntflags, pathname_t *realpnp) 1153 { 1154 znode_t *zdp = VTOZ(dvp); 1155 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1156 int error = 0; 1157 1158 /* fast path */ 1159 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 1160 1161 if (dvp->v_type != VDIR) { 1162 return (ENOTDIR); 1163 } else if (zdp->z_sa_hdl == NULL) { 1164 return (EIO); 1165 } 1166 1167 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 1168 error = zfs_fastaccesschk_execute(zdp, cr); 1169 if (!error) { 1170 *vpp = dvp; 1171 VN_HOLD(*vpp); 1172 return (0); 1173 } 1174 return (error); 1175 } else { 1176 vnode_t *tvp = dnlc_lookup(dvp, nm); 1177 1178 if (tvp) { 1179 error = zfs_fastaccesschk_execute(zdp, cr); 1180 if (error) { 1181 VN_RELE(tvp); 1182 return (error); 1183 } 1184 if (tvp == DNLC_NO_VNODE) { 1185 VN_RELE(tvp); 1186 return (ENOENT); 1187 } else { 1188 *vpp = tvp; 1189 return (specvp_check(vpp, cr)); 1190 } 1191 } 1192 } 1193 } 1194 1195 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1196 1197 ZFS_ENTER(zfsvfs); 1198 ZFS_VERIFY_ZP(zdp); 1199 1200 *vpp = NULL; 1201 1202 if (flags & LOOKUP_XATTR) { 1203 /* 1204 * If the xattr property is off, refuse the lookup request. 1205 */ 1206 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1207 ZFS_EXIT(zfsvfs); 1208 return (EINVAL); 1209 } 1210 1211 /* 1212 * We don't allow recursive attributes.. 1213 * Maybe someday we will. 1214 */ 1215 if (zdp->z_pflags & ZFS_XATTR) { 1216 ZFS_EXIT(zfsvfs); 1217 return (EINVAL); 1218 } 1219 1220 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1221 ZFS_EXIT(zfsvfs); 1222 return (error); 1223 } 1224 1225 /* 1226 * Do we have permission to get into attribute directory? 1227 */ 1228 1229 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1230 B_FALSE, cr)) { 1231 VN_RELE(*vpp); 1232 *vpp = NULL; 1233 } 1234 1235 ZFS_EXIT(zfsvfs); 1236 return (error); 1237 } 1238 1239 if (dvp->v_type != VDIR) { 1240 ZFS_EXIT(zfsvfs); 1241 return (ENOTDIR); 1242 } 1243 1244 /* 1245 * Check accessibility of directory. 1246 */ 1247 1248 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1249 ZFS_EXIT(zfsvfs); 1250 return (error); 1251 } 1252 1253 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1254 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1255 ZFS_EXIT(zfsvfs); 1256 return (EILSEQ); 1257 } 1258 1259 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1260 if (error == 0) 1261 error = specvp_check(vpp, cr); 1262 1263 ZFS_EXIT(zfsvfs); 1264 return (error); 1265 } 1266 1267 /* 1268 * Attempt to create a new entry in a directory. If the entry 1269 * already exists, truncate the file if permissible, else return 1270 * an error. Return the vp of the created or trunc'd file. 1271 * 1272 * IN: dvp - vnode of directory to put new file entry in. 1273 * name - name of new file entry. 1274 * vap - attributes of new file. 1275 * excl - flag indicating exclusive or non-exclusive mode. 1276 * mode - mode to open file with. 1277 * cr - credentials of caller. 1278 * flag - large file flag [UNUSED]. 1279 * ct - caller context 1280 * vsecp - ACL to be set 1281 * 1282 * OUT: vpp - vnode of created or trunc'd entry. 1283 * 1284 * RETURN: 0 if success 1285 * error code if failure 1286 * 1287 * Timestamps: 1288 * dvp - ctime|mtime updated if new entry created 1289 * vp - ctime|mtime always, atime if new 1290 */ 1291 1292 /* ARGSUSED */ 1293 static int 1294 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1295 int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct, 1296 vsecattr_t *vsecp) 1297 { 1298 znode_t *zp, *dzp = VTOZ(dvp); 1299 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1300 zilog_t *zilog; 1301 objset_t *os; 1302 zfs_dirlock_t *dl; 1303 dmu_tx_t *tx; 1304 int error; 1305 ksid_t *ksid; 1306 uid_t uid; 1307 gid_t gid = crgetgid(cr); 1308 zfs_acl_ids_t acl_ids; 1309 boolean_t fuid_dirtied; 1310 boolean_t have_acl = B_FALSE; 1311 1312 /* 1313 * If we have an ephemeral id, ACL, or XVATTR then 1314 * make sure file system is at proper version 1315 */ 1316 1317 ksid = crgetsid(cr, KSID_OWNER); 1318 if (ksid) 1319 uid = ksid_getid(ksid); 1320 else 1321 uid = crgetuid(cr); 1322 1323 if (zfsvfs->z_use_fuids == B_FALSE && 1324 (vsecp || (vap->va_mask & AT_XVATTR) || 1325 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1326 return (EINVAL); 1327 1328 ZFS_ENTER(zfsvfs); 1329 ZFS_VERIFY_ZP(dzp); 1330 os = zfsvfs->z_os; 1331 zilog = zfsvfs->z_log; 1332 1333 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1334 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1335 ZFS_EXIT(zfsvfs); 1336 return (EILSEQ); 1337 } 1338 1339 if (vap->va_mask & AT_XVATTR) { 1340 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1341 crgetuid(cr), cr, vap->va_type)) != 0) { 1342 ZFS_EXIT(zfsvfs); 1343 return (error); 1344 } 1345 } 1346 top: 1347 *vpp = NULL; 1348 1349 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1350 vap->va_mode &= ~VSVTX; 1351 1352 if (*name == '\0') { 1353 /* 1354 * Null component name refers to the directory itself. 1355 */ 1356 VN_HOLD(dvp); 1357 zp = dzp; 1358 dl = NULL; 1359 error = 0; 1360 } else { 1361 /* possible VN_HOLD(zp) */ 1362 int zflg = 0; 1363 1364 if (flag & FIGNORECASE) 1365 zflg |= ZCILOOK; 1366 1367 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1368 NULL, NULL); 1369 if (error) { 1370 if (strcmp(name, "..") == 0) 1371 error = EISDIR; 1372 ZFS_EXIT(zfsvfs); 1373 return (error); 1374 } 1375 } 1376 1377 if (zp == NULL) { 1378 uint64_t txtype; 1379 1380 /* 1381 * Create a new file object and update the directory 1382 * to reference it. 1383 */ 1384 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1385 goto out; 1386 } 1387 1388 /* 1389 * We only support the creation of regular files in 1390 * extended attribute directories. 1391 */ 1392 1393 if ((dzp->z_pflags & ZFS_XATTR) && 1394 (vap->va_type != VREG)) { 1395 error = EINVAL; 1396 goto out; 1397 } 1398 1399 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 1400 cr, vsecp, &acl_ids)) != 0) 1401 goto out; 1402 have_acl = B_TRUE; 1403 1404 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1405 zfs_acl_ids_free(&acl_ids); 1406 error = EDQUOT; 1407 goto out; 1408 } 1409 1410 tx = dmu_tx_create(os); 1411 1412 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1413 ZFS_SA_BASE_ATTR_SIZE); 1414 1415 fuid_dirtied = zfsvfs->z_fuid_dirty; 1416 if (fuid_dirtied) 1417 zfs_fuid_txhold(zfsvfs, tx); 1418 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1419 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1420 if (!zfsvfs->z_use_sa && 1421 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1422 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1423 0, acl_ids.z_aclp->z_acl_bytes); 1424 } 1425 error = dmu_tx_assign(tx, TXG_NOWAIT); 1426 if (error) { 1427 zfs_dirent_unlock(dl); 1428 if (error == ERESTART) { 1429 dmu_tx_wait(tx); 1430 dmu_tx_abort(tx); 1431 goto top; 1432 } 1433 zfs_acl_ids_free(&acl_ids); 1434 dmu_tx_abort(tx); 1435 ZFS_EXIT(zfsvfs); 1436 return (error); 1437 } 1438 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1439 1440 if (fuid_dirtied) 1441 zfs_fuid_sync(zfsvfs, tx); 1442 1443 (void) zfs_link_create(dl, zp, tx, ZNEW); 1444 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1445 if (flag & FIGNORECASE) 1446 txtype |= TX_CI; 1447 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1448 vsecp, acl_ids.z_fuidp, vap); 1449 zfs_acl_ids_free(&acl_ids); 1450 dmu_tx_commit(tx); 1451 } else { 1452 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1453 1454 /* 1455 * A directory entry already exists for this name. 1456 */ 1457 /* 1458 * Can't truncate an existing file if in exclusive mode. 1459 */ 1460 if (excl == EXCL) { 1461 error = EEXIST; 1462 goto out; 1463 } 1464 /* 1465 * Can't open a directory for writing. 1466 */ 1467 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1468 error = EISDIR; 1469 goto out; 1470 } 1471 /* 1472 * Verify requested access to file. 1473 */ 1474 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1475 goto out; 1476 } 1477 1478 mutex_enter(&dzp->z_lock); 1479 dzp->z_seq++; 1480 mutex_exit(&dzp->z_lock); 1481 1482 /* 1483 * Truncate regular files if requested. 1484 */ 1485 if ((ZTOV(zp)->v_type == VREG) && 1486 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1487 /* we can't hold any locks when calling zfs_freesp() */ 1488 zfs_dirent_unlock(dl); 1489 dl = NULL; 1490 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1491 if (error == 0) { 1492 vnevent_create(ZTOV(zp), ct); 1493 } 1494 } 1495 } 1496 out: 1497 1498 if (dl) 1499 zfs_dirent_unlock(dl); 1500 1501 if (error) { 1502 if (zp) 1503 VN_RELE(ZTOV(zp)); 1504 } else { 1505 *vpp = ZTOV(zp); 1506 error = specvp_check(vpp, cr); 1507 } 1508 1509 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1510 zil_commit(zilog, 0); 1511 1512 ZFS_EXIT(zfsvfs); 1513 return (error); 1514 } 1515 1516 /* 1517 * Remove an entry from a directory. 1518 * 1519 * IN: dvp - vnode of directory to remove entry from. 1520 * name - name of entry to remove. 1521 * cr - credentials of caller. 1522 * ct - caller context 1523 * flags - case flags 1524 * 1525 * RETURN: 0 if success 1526 * error code if failure 1527 * 1528 * Timestamps: 1529 * dvp - ctime|mtime 1530 * vp - ctime (if nlink > 0) 1531 */ 1532 1533 uint64_t null_xattr = 0; 1534 1535 /*ARGSUSED*/ 1536 static int 1537 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1538 int flags) 1539 { 1540 znode_t *zp, *dzp = VTOZ(dvp); 1541 znode_t *xzp = NULL; 1542 vnode_t *vp; 1543 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1544 zilog_t *zilog; 1545 uint64_t acl_obj, xattr_obj = 0; 1546 uint64_t xattr_obj_unlinked = 0; 1547 uint64_t obj = 0; 1548 zfs_dirlock_t *dl; 1549 dmu_tx_t *tx; 1550 boolean_t may_delete_now, delete_now = FALSE; 1551 boolean_t unlinked, toobig = FALSE; 1552 uint64_t txtype; 1553 pathname_t *realnmp = NULL; 1554 pathname_t realnm; 1555 int error; 1556 int zflg = ZEXISTS; 1557 1558 ZFS_ENTER(zfsvfs); 1559 ZFS_VERIFY_ZP(dzp); 1560 zilog = zfsvfs->z_log; 1561 1562 if (flags & FIGNORECASE) { 1563 zflg |= ZCILOOK; 1564 pn_alloc(&realnm); 1565 realnmp = &realnm; 1566 } 1567 1568 top: 1569 /* 1570 * Attempt to lock directory; fail if entry doesn't exist. 1571 */ 1572 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1573 NULL, realnmp)) { 1574 if (realnmp) 1575 pn_free(realnmp); 1576 ZFS_EXIT(zfsvfs); 1577 return (error); 1578 } 1579 1580 vp = ZTOV(zp); 1581 1582 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1583 goto out; 1584 } 1585 1586 /* 1587 * Need to use rmdir for removing directories. 1588 */ 1589 if (vp->v_type == VDIR) { 1590 error = EPERM; 1591 goto out; 1592 } 1593 1594 vnevent_remove(vp, dvp, name, ct); 1595 1596 if (realnmp) 1597 dnlc_remove(dvp, realnmp->pn_buf); 1598 else 1599 dnlc_remove(dvp, name); 1600 1601 mutex_enter(&vp->v_lock); 1602 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1603 mutex_exit(&vp->v_lock); 1604 1605 /* 1606 * We may delete the znode now, or we may put it in the unlinked set; 1607 * it depends on whether we're the last link, and on whether there are 1608 * other holds on the vnode. So we dmu_tx_hold() the right things to 1609 * allow for either case. 1610 */ 1611 obj = zp->z_id; 1612 tx = dmu_tx_create(zfsvfs->z_os); 1613 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1614 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1615 zfs_sa_upgrade_txholds(tx, zp); 1616 zfs_sa_upgrade_txholds(tx, dzp); 1617 if (may_delete_now) { 1618 toobig = 1619 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 1620 /* if the file is too big, only hold_free a token amount */ 1621 dmu_tx_hold_free(tx, zp->z_id, 0, 1622 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1623 } 1624 1625 /* are there any extended attributes? */ 1626 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1627 &xattr_obj, sizeof (xattr_obj)); 1628 if (xattr_obj) { 1629 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1630 ASSERT3U(error, ==, 0); 1631 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1632 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1633 } 1634 1635 mutex_enter(&zp->z_lock); 1636 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now) 1637 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1638 mutex_exit(&zp->z_lock); 1639 1640 /* charge as an update -- would be nice not to charge at all */ 1641 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1642 1643 error = dmu_tx_assign(tx, TXG_NOWAIT); 1644 if (error) { 1645 zfs_dirent_unlock(dl); 1646 VN_RELE(vp); 1647 if (error == ERESTART) { 1648 dmu_tx_wait(tx); 1649 dmu_tx_abort(tx); 1650 goto top; 1651 } 1652 if (realnmp) 1653 pn_free(realnmp); 1654 dmu_tx_abort(tx); 1655 ZFS_EXIT(zfsvfs); 1656 return (error); 1657 } 1658 1659 /* 1660 * Remove the directory entry. 1661 */ 1662 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1663 1664 if (error) { 1665 dmu_tx_commit(tx); 1666 goto out; 1667 } 1668 1669 if (unlinked) { 1670 1671 /* 1672 * Hold z_lock so that we can make sure that the ACL obj 1673 * hasn't changed. Could have been deleted due to 1674 * zfs_sa_upgrade(). 1675 */ 1676 mutex_enter(&zp->z_lock); 1677 mutex_enter(&vp->v_lock); 1678 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1679 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked)); 1680 delete_now = may_delete_now && !toobig && 1681 vp->v_count == 1 && !vn_has_cached_data(vp) && 1682 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) == 1683 acl_obj; 1684 mutex_exit(&vp->v_lock); 1685 } 1686 1687 if (delete_now) { 1688 if (xattr_obj_unlinked) { 1689 ASSERT3U(xzp->z_links, ==, 2); 1690 mutex_enter(&xzp->z_lock); 1691 xzp->z_unlinked = 1; 1692 xzp->z_links = 0; 1693 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 1694 &xzp->z_links, sizeof (xzp->z_links), tx); 1695 ASSERT3U(error, ==, 0); 1696 mutex_exit(&xzp->z_lock); 1697 zfs_unlinked_add(xzp, tx); 1698 1699 if (zp->z_is_sa) 1700 error = sa_remove(zp->z_sa_hdl, 1701 SA_ZPL_XATTR(zfsvfs), tx); 1702 else 1703 error = sa_update(zp->z_sa_hdl, 1704 SA_ZPL_XATTR(zfsvfs), &null_xattr, 1705 sizeof (uint64_t), tx); 1706 ASSERT3U(error, ==, 0); 1707 } 1708 mutex_enter(&vp->v_lock); 1709 vp->v_count--; 1710 ASSERT3U(vp->v_count, ==, 0); 1711 mutex_exit(&vp->v_lock); 1712 mutex_exit(&zp->z_lock); 1713 zfs_znode_delete(zp, tx); 1714 } else if (unlinked) { 1715 mutex_exit(&zp->z_lock); 1716 zfs_unlinked_add(zp, tx); 1717 } 1718 1719 txtype = TX_REMOVE; 1720 if (flags & FIGNORECASE) 1721 txtype |= TX_CI; 1722 zfs_log_remove(zilog, tx, txtype, dzp, name, obj); 1723 1724 dmu_tx_commit(tx); 1725 out: 1726 if (realnmp) 1727 pn_free(realnmp); 1728 1729 zfs_dirent_unlock(dl); 1730 1731 if (!delete_now) 1732 VN_RELE(vp); 1733 if (xzp) 1734 VN_RELE(ZTOV(xzp)); 1735 1736 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1737 zil_commit(zilog, 0); 1738 1739 ZFS_EXIT(zfsvfs); 1740 return (error); 1741 } 1742 1743 /* 1744 * Create a new directory and insert it into dvp using the name 1745 * provided. Return a pointer to the inserted directory. 1746 * 1747 * IN: dvp - vnode of directory to add subdir to. 1748 * dirname - name of new directory. 1749 * vap - attributes of new directory. 1750 * cr - credentials of caller. 1751 * ct - caller context 1752 * vsecp - ACL to be set 1753 * 1754 * OUT: vpp - vnode of created directory. 1755 * 1756 * RETURN: 0 if success 1757 * error code if failure 1758 * 1759 * Timestamps: 1760 * dvp - ctime|mtime updated 1761 * vp - ctime|mtime|atime updated 1762 */ 1763 /*ARGSUSED*/ 1764 static int 1765 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 1766 caller_context_t *ct, int flags, vsecattr_t *vsecp) 1767 { 1768 znode_t *zp, *dzp = VTOZ(dvp); 1769 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1770 zilog_t *zilog; 1771 zfs_dirlock_t *dl; 1772 uint64_t txtype; 1773 dmu_tx_t *tx; 1774 int error; 1775 int zf = ZNEW; 1776 ksid_t *ksid; 1777 uid_t uid; 1778 gid_t gid = crgetgid(cr); 1779 zfs_acl_ids_t acl_ids; 1780 boolean_t fuid_dirtied; 1781 1782 ASSERT(vap->va_type == VDIR); 1783 1784 /* 1785 * If we have an ephemeral id, ACL, or XVATTR then 1786 * make sure file system is at proper version 1787 */ 1788 1789 ksid = crgetsid(cr, KSID_OWNER); 1790 if (ksid) 1791 uid = ksid_getid(ksid); 1792 else 1793 uid = crgetuid(cr); 1794 if (zfsvfs->z_use_fuids == B_FALSE && 1795 (vsecp || (vap->va_mask & AT_XVATTR) || 1796 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1797 return (EINVAL); 1798 1799 ZFS_ENTER(zfsvfs); 1800 ZFS_VERIFY_ZP(dzp); 1801 zilog = zfsvfs->z_log; 1802 1803 if (dzp->z_pflags & ZFS_XATTR) { 1804 ZFS_EXIT(zfsvfs); 1805 return (EINVAL); 1806 } 1807 1808 if (zfsvfs->z_utf8 && u8_validate(dirname, 1809 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1810 ZFS_EXIT(zfsvfs); 1811 return (EILSEQ); 1812 } 1813 if (flags & FIGNORECASE) 1814 zf |= ZCILOOK; 1815 1816 if (vap->va_mask & AT_XVATTR) { 1817 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1818 crgetuid(cr), cr, vap->va_type)) != 0) { 1819 ZFS_EXIT(zfsvfs); 1820 return (error); 1821 } 1822 } 1823 1824 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 1825 vsecp, &acl_ids)) != 0) { 1826 ZFS_EXIT(zfsvfs); 1827 return (error); 1828 } 1829 /* 1830 * First make sure the new directory doesn't exist. 1831 * 1832 * Existence is checked first to make sure we don't return 1833 * EACCES instead of EEXIST which can cause some applications 1834 * to fail. 1835 */ 1836 top: 1837 *vpp = NULL; 1838 1839 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 1840 NULL, NULL)) { 1841 zfs_acl_ids_free(&acl_ids); 1842 ZFS_EXIT(zfsvfs); 1843 return (error); 1844 } 1845 1846 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 1847 zfs_acl_ids_free(&acl_ids); 1848 zfs_dirent_unlock(dl); 1849 ZFS_EXIT(zfsvfs); 1850 return (error); 1851 } 1852 1853 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1854 zfs_acl_ids_free(&acl_ids); 1855 zfs_dirent_unlock(dl); 1856 ZFS_EXIT(zfsvfs); 1857 return (EDQUOT); 1858 } 1859 1860 /* 1861 * Add a new entry to the directory. 1862 */ 1863 tx = dmu_tx_create(zfsvfs->z_os); 1864 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1865 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1866 fuid_dirtied = zfsvfs->z_fuid_dirty; 1867 if (fuid_dirtied) 1868 zfs_fuid_txhold(zfsvfs, tx); 1869 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1870 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1871 acl_ids.z_aclp->z_acl_bytes); 1872 } 1873 1874 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1875 ZFS_SA_BASE_ATTR_SIZE); 1876 1877 error = dmu_tx_assign(tx, TXG_NOWAIT); 1878 if (error) { 1879 zfs_dirent_unlock(dl); 1880 if (error == ERESTART) { 1881 dmu_tx_wait(tx); 1882 dmu_tx_abort(tx); 1883 goto top; 1884 } 1885 zfs_acl_ids_free(&acl_ids); 1886 dmu_tx_abort(tx); 1887 ZFS_EXIT(zfsvfs); 1888 return (error); 1889 } 1890 1891 /* 1892 * Create new node. 1893 */ 1894 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1895 1896 if (fuid_dirtied) 1897 zfs_fuid_sync(zfsvfs, tx); 1898 1899 /* 1900 * Now put new name in parent dir. 1901 */ 1902 (void) zfs_link_create(dl, zp, tx, ZNEW); 1903 1904 *vpp = ZTOV(zp); 1905 1906 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 1907 if (flags & FIGNORECASE) 1908 txtype |= TX_CI; 1909 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 1910 acl_ids.z_fuidp, vap); 1911 1912 zfs_acl_ids_free(&acl_ids); 1913 1914 dmu_tx_commit(tx); 1915 1916 zfs_dirent_unlock(dl); 1917 1918 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1919 zil_commit(zilog, 0); 1920 1921 ZFS_EXIT(zfsvfs); 1922 return (0); 1923 } 1924 1925 /* 1926 * Remove a directory subdir entry. If the current working 1927 * directory is the same as the subdir to be removed, the 1928 * remove will fail. 1929 * 1930 * IN: dvp - vnode of directory to remove from. 1931 * name - name of directory to be removed. 1932 * cwd - vnode of current working directory. 1933 * cr - credentials of caller. 1934 * ct - caller context 1935 * flags - case flags 1936 * 1937 * RETURN: 0 if success 1938 * error code if failure 1939 * 1940 * Timestamps: 1941 * dvp - ctime|mtime updated 1942 */ 1943 /*ARGSUSED*/ 1944 static int 1945 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 1946 caller_context_t *ct, int flags) 1947 { 1948 znode_t *dzp = VTOZ(dvp); 1949 znode_t *zp; 1950 vnode_t *vp; 1951 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1952 zilog_t *zilog; 1953 zfs_dirlock_t *dl; 1954 dmu_tx_t *tx; 1955 int error; 1956 int zflg = ZEXISTS; 1957 1958 ZFS_ENTER(zfsvfs); 1959 ZFS_VERIFY_ZP(dzp); 1960 zilog = zfsvfs->z_log; 1961 1962 if (flags & FIGNORECASE) 1963 zflg |= ZCILOOK; 1964 top: 1965 zp = NULL; 1966 1967 /* 1968 * Attempt to lock directory; fail if entry doesn't exist. 1969 */ 1970 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1971 NULL, NULL)) { 1972 ZFS_EXIT(zfsvfs); 1973 return (error); 1974 } 1975 1976 vp = ZTOV(zp); 1977 1978 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1979 goto out; 1980 } 1981 1982 if (vp->v_type != VDIR) { 1983 error = ENOTDIR; 1984 goto out; 1985 } 1986 1987 if (vp == cwd) { 1988 error = EINVAL; 1989 goto out; 1990 } 1991 1992 vnevent_rmdir(vp, dvp, name, ct); 1993 1994 /* 1995 * Grab a lock on the directory to make sure that noone is 1996 * trying to add (or lookup) entries while we are removing it. 1997 */ 1998 rw_enter(&zp->z_name_lock, RW_WRITER); 1999 2000 /* 2001 * Grab a lock on the parent pointer to make sure we play well 2002 * with the treewalk and directory rename code. 2003 */ 2004 rw_enter(&zp->z_parent_lock, RW_WRITER); 2005 2006 tx = dmu_tx_create(zfsvfs->z_os); 2007 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2008 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2009 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2010 zfs_sa_upgrade_txholds(tx, zp); 2011 zfs_sa_upgrade_txholds(tx, dzp); 2012 error = dmu_tx_assign(tx, TXG_NOWAIT); 2013 if (error) { 2014 rw_exit(&zp->z_parent_lock); 2015 rw_exit(&zp->z_name_lock); 2016 zfs_dirent_unlock(dl); 2017 VN_RELE(vp); 2018 if (error == ERESTART) { 2019 dmu_tx_wait(tx); 2020 dmu_tx_abort(tx); 2021 goto top; 2022 } 2023 dmu_tx_abort(tx); 2024 ZFS_EXIT(zfsvfs); 2025 return (error); 2026 } 2027 2028 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 2029 2030 if (error == 0) { 2031 uint64_t txtype = TX_RMDIR; 2032 if (flags & FIGNORECASE) 2033 txtype |= TX_CI; 2034 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT); 2035 } 2036 2037 dmu_tx_commit(tx); 2038 2039 rw_exit(&zp->z_parent_lock); 2040 rw_exit(&zp->z_name_lock); 2041 out: 2042 zfs_dirent_unlock(dl); 2043 2044 VN_RELE(vp); 2045 2046 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2047 zil_commit(zilog, 0); 2048 2049 ZFS_EXIT(zfsvfs); 2050 return (error); 2051 } 2052 2053 /* 2054 * Read as many directory entries as will fit into the provided 2055 * buffer from the given directory cursor position (specified in 2056 * the uio structure. 2057 * 2058 * IN: vp - vnode of directory to read. 2059 * uio - structure supplying read location, range info, 2060 * and return buffer. 2061 * cr - credentials of caller. 2062 * ct - caller context 2063 * flags - case flags 2064 * 2065 * OUT: uio - updated offset and range, buffer filled. 2066 * eofp - set to true if end-of-file detected. 2067 * 2068 * RETURN: 0 if success 2069 * error code if failure 2070 * 2071 * Timestamps: 2072 * vp - atime updated 2073 * 2074 * Note that the low 4 bits of the cookie returned by zap is always zero. 2075 * This allows us to use the low range for "special" directory entries: 2076 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 2077 * we use the offset 2 for the '.zfs' directory. 2078 */ 2079 /* ARGSUSED */ 2080 static int 2081 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, 2082 caller_context_t *ct, int flags) 2083 { 2084 znode_t *zp = VTOZ(vp); 2085 iovec_t *iovp; 2086 edirent_t *eodp; 2087 dirent64_t *odp; 2088 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2089 objset_t *os; 2090 caddr_t outbuf; 2091 size_t bufsize; 2092 zap_cursor_t zc; 2093 zap_attribute_t zap; 2094 uint_t bytes_wanted; 2095 uint64_t offset; /* must be unsigned; checks for < 1 */ 2096 uint64_t parent; 2097 int local_eof; 2098 int outcount; 2099 int error; 2100 uint8_t prefetch; 2101 boolean_t check_sysattrs; 2102 2103 ZFS_ENTER(zfsvfs); 2104 ZFS_VERIFY_ZP(zp); 2105 2106 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 2107 &parent, sizeof (parent))) != 0) { 2108 ZFS_EXIT(zfsvfs); 2109 return (error); 2110 } 2111 2112 /* 2113 * If we are not given an eof variable, 2114 * use a local one. 2115 */ 2116 if (eofp == NULL) 2117 eofp = &local_eof; 2118 2119 /* 2120 * Check for valid iov_len. 2121 */ 2122 if (uio->uio_iov->iov_len <= 0) { 2123 ZFS_EXIT(zfsvfs); 2124 return (EINVAL); 2125 } 2126 2127 /* 2128 * Quit if directory has been removed (posix) 2129 */ 2130 if ((*eofp = zp->z_unlinked) != 0) { 2131 ZFS_EXIT(zfsvfs); 2132 return (0); 2133 } 2134 2135 error = 0; 2136 os = zfsvfs->z_os; 2137 offset = uio->uio_loffset; 2138 prefetch = zp->z_zn_prefetch; 2139 2140 /* 2141 * Initialize the iterator cursor. 2142 */ 2143 if (offset <= 3) { 2144 /* 2145 * Start iteration from the beginning of the directory. 2146 */ 2147 zap_cursor_init(&zc, os, zp->z_id); 2148 } else { 2149 /* 2150 * The offset is a serialized cursor. 2151 */ 2152 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 2153 } 2154 2155 /* 2156 * Get space to change directory entries into fs independent format. 2157 */ 2158 iovp = uio->uio_iov; 2159 bytes_wanted = iovp->iov_len; 2160 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 2161 bufsize = bytes_wanted; 2162 outbuf = kmem_alloc(bufsize, KM_SLEEP); 2163 odp = (struct dirent64 *)outbuf; 2164 } else { 2165 bufsize = bytes_wanted; 2166 odp = (struct dirent64 *)iovp->iov_base; 2167 } 2168 eodp = (struct edirent *)odp; 2169 2170 /* 2171 * If this VFS supports the system attribute view interface; and 2172 * we're looking at an extended attribute directory; and we care 2173 * about normalization conflicts on this vfs; then we must check 2174 * for normalization conflicts with the sysattr name space. 2175 */ 2176 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 2177 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 2178 (flags & V_RDDIR_ENTFLAGS); 2179 2180 /* 2181 * Transform to file-system independent format 2182 */ 2183 outcount = 0; 2184 while (outcount < bytes_wanted) { 2185 ino64_t objnum; 2186 ushort_t reclen; 2187 off64_t *next = NULL; 2188 2189 /* 2190 * Special case `.', `..', and `.zfs'. 2191 */ 2192 if (offset == 0) { 2193 (void) strcpy(zap.za_name, "."); 2194 zap.za_normalization_conflict = 0; 2195 objnum = zp->z_id; 2196 } else if (offset == 1) { 2197 (void) strcpy(zap.za_name, ".."); 2198 zap.za_normalization_conflict = 0; 2199 objnum = parent; 2200 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2201 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2202 zap.za_normalization_conflict = 0; 2203 objnum = ZFSCTL_INO_ROOT; 2204 } else { 2205 /* 2206 * Grab next entry. 2207 */ 2208 if (error = zap_cursor_retrieve(&zc, &zap)) { 2209 if ((*eofp = (error == ENOENT)) != 0) 2210 break; 2211 else 2212 goto update; 2213 } 2214 2215 if (zap.za_integer_length != 8 || 2216 zap.za_num_integers != 1) { 2217 cmn_err(CE_WARN, "zap_readdir: bad directory " 2218 "entry, obj = %lld, offset = %lld\n", 2219 (u_longlong_t)zp->z_id, 2220 (u_longlong_t)offset); 2221 error = ENXIO; 2222 goto update; 2223 } 2224 2225 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2226 /* 2227 * MacOS X can extract the object type here such as: 2228 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2229 */ 2230 2231 if (check_sysattrs && !zap.za_normalization_conflict) { 2232 zap.za_normalization_conflict = 2233 xattr_sysattr_casechk(zap.za_name); 2234 } 2235 } 2236 2237 if (flags & V_RDDIR_ACCFILTER) { 2238 /* 2239 * If we have no access at all, don't include 2240 * this entry in the returned information 2241 */ 2242 znode_t *ezp; 2243 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2244 goto skip_entry; 2245 if (!zfs_has_access(ezp, cr)) { 2246 VN_RELE(ZTOV(ezp)); 2247 goto skip_entry; 2248 } 2249 VN_RELE(ZTOV(ezp)); 2250 } 2251 2252 if (flags & V_RDDIR_ENTFLAGS) 2253 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2254 else 2255 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2256 2257 /* 2258 * Will this entry fit in the buffer? 2259 */ 2260 if (outcount + reclen > bufsize) { 2261 /* 2262 * Did we manage to fit anything in the buffer? 2263 */ 2264 if (!outcount) { 2265 error = EINVAL; 2266 goto update; 2267 } 2268 break; 2269 } 2270 if (flags & V_RDDIR_ENTFLAGS) { 2271 /* 2272 * Add extended flag entry: 2273 */ 2274 eodp->ed_ino = objnum; 2275 eodp->ed_reclen = reclen; 2276 /* NOTE: ed_off is the offset for the *next* entry */ 2277 next = &(eodp->ed_off); 2278 eodp->ed_eflags = zap.za_normalization_conflict ? 2279 ED_CASE_CONFLICT : 0; 2280 (void) strncpy(eodp->ed_name, zap.za_name, 2281 EDIRENT_NAMELEN(reclen)); 2282 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2283 } else { 2284 /* 2285 * Add normal entry: 2286 */ 2287 odp->d_ino = objnum; 2288 odp->d_reclen = reclen; 2289 /* NOTE: d_off is the offset for the *next* entry */ 2290 next = &(odp->d_off); 2291 (void) strncpy(odp->d_name, zap.za_name, 2292 DIRENT64_NAMELEN(reclen)); 2293 odp = (dirent64_t *)((intptr_t)odp + reclen); 2294 } 2295 outcount += reclen; 2296 2297 ASSERT(outcount <= bufsize); 2298 2299 /* Prefetch znode */ 2300 if (prefetch) 2301 dmu_prefetch(os, objnum, 0, 0); 2302 2303 skip_entry: 2304 /* 2305 * Move to the next entry, fill in the previous offset. 2306 */ 2307 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2308 zap_cursor_advance(&zc); 2309 offset = zap_cursor_serialize(&zc); 2310 } else { 2311 offset += 1; 2312 } 2313 if (next) 2314 *next = offset; 2315 } 2316 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2317 2318 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2319 iovp->iov_base += outcount; 2320 iovp->iov_len -= outcount; 2321 uio->uio_resid -= outcount; 2322 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2323 /* 2324 * Reset the pointer. 2325 */ 2326 offset = uio->uio_loffset; 2327 } 2328 2329 update: 2330 zap_cursor_fini(&zc); 2331 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2332 kmem_free(outbuf, bufsize); 2333 2334 if (error == ENOENT) 2335 error = 0; 2336 2337 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2338 2339 uio->uio_loffset = offset; 2340 ZFS_EXIT(zfsvfs); 2341 return (error); 2342 } 2343 2344 ulong_t zfs_fsync_sync_cnt = 4; 2345 2346 static int 2347 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2348 { 2349 znode_t *zp = VTOZ(vp); 2350 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2351 2352 /* 2353 * Regardless of whether this is required for standards conformance, 2354 * this is the logical behavior when fsync() is called on a file with 2355 * dirty pages. We use B_ASYNC since the ZIL transactions are already 2356 * going to be pushed out as part of the zil_commit(). 2357 */ 2358 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 2359 (vp->v_type == VREG) && !(IS_SWAPVP(vp))) 2360 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct); 2361 2362 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2363 2364 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { 2365 ZFS_ENTER(zfsvfs); 2366 ZFS_VERIFY_ZP(zp); 2367 zil_commit(zfsvfs->z_log, zp->z_id); 2368 ZFS_EXIT(zfsvfs); 2369 } 2370 return (0); 2371 } 2372 2373 2374 /* 2375 * Get the requested file attributes and place them in the provided 2376 * vattr structure. 2377 * 2378 * IN: vp - vnode of file. 2379 * vap - va_mask identifies requested attributes. 2380 * If AT_XVATTR set, then optional attrs are requested 2381 * flags - ATTR_NOACLCHECK (CIFS server context) 2382 * cr - credentials of caller. 2383 * ct - caller context 2384 * 2385 * OUT: vap - attribute values. 2386 * 2387 * RETURN: 0 (always succeeds) 2388 */ 2389 /* ARGSUSED */ 2390 static int 2391 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2392 caller_context_t *ct) 2393 { 2394 znode_t *zp = VTOZ(vp); 2395 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2396 int error = 0; 2397 uint64_t links; 2398 uint64_t mtime[2], ctime[2]; 2399 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2400 xoptattr_t *xoap = NULL; 2401 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2402 sa_bulk_attr_t bulk[2]; 2403 int count = 0; 2404 2405 ZFS_ENTER(zfsvfs); 2406 ZFS_VERIFY_ZP(zp); 2407 2408 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 2409 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 2410 2411 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 2412 ZFS_EXIT(zfsvfs); 2413 return (error); 2414 } 2415 2416 /* 2417 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2418 * Also, if we are the owner don't bother, since owner should 2419 * always be allowed to read basic attributes of file. 2420 */ 2421 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && (zp->z_uid != crgetuid(cr))) { 2422 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2423 skipaclchk, cr)) { 2424 ZFS_EXIT(zfsvfs); 2425 return (error); 2426 } 2427 } 2428 2429 /* 2430 * Return all attributes. It's cheaper to provide the answer 2431 * than to determine whether we were asked the question. 2432 */ 2433 2434 mutex_enter(&zp->z_lock); 2435 vap->va_type = vp->v_type; 2436 vap->va_mode = zp->z_mode & MODEMASK; 2437 vap->va_uid = zp->z_uid; 2438 vap->va_gid = zp->z_gid; 2439 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2440 vap->va_nodeid = zp->z_id; 2441 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2442 links = zp->z_links + 1; 2443 else 2444 links = zp->z_links; 2445 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */ 2446 vap->va_size = zp->z_size; 2447 vap->va_rdev = vp->v_rdev; 2448 vap->va_seq = zp->z_seq; 2449 2450 /* 2451 * Add in any requested optional attributes and the create time. 2452 * Also set the corresponding bits in the returned attribute bitmap. 2453 */ 2454 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2455 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2456 xoap->xoa_archive = 2457 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2458 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2459 } 2460 2461 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2462 xoap->xoa_readonly = 2463 ((zp->z_pflags & ZFS_READONLY) != 0); 2464 XVA_SET_RTN(xvap, XAT_READONLY); 2465 } 2466 2467 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2468 xoap->xoa_system = 2469 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2470 XVA_SET_RTN(xvap, XAT_SYSTEM); 2471 } 2472 2473 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2474 xoap->xoa_hidden = 2475 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2476 XVA_SET_RTN(xvap, XAT_HIDDEN); 2477 } 2478 2479 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2480 xoap->xoa_nounlink = 2481 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2482 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2483 } 2484 2485 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2486 xoap->xoa_immutable = 2487 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2488 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2489 } 2490 2491 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2492 xoap->xoa_appendonly = 2493 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2494 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2495 } 2496 2497 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2498 xoap->xoa_nodump = 2499 ((zp->z_pflags & ZFS_NODUMP) != 0); 2500 XVA_SET_RTN(xvap, XAT_NODUMP); 2501 } 2502 2503 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2504 xoap->xoa_opaque = 2505 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2506 XVA_SET_RTN(xvap, XAT_OPAQUE); 2507 } 2508 2509 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2510 xoap->xoa_av_quarantined = 2511 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2512 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2513 } 2514 2515 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2516 xoap->xoa_av_modified = 2517 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2518 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2519 } 2520 2521 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2522 vp->v_type == VREG) { 2523 zfs_sa_get_scanstamp(zp, xvap); 2524 } 2525 2526 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2527 uint64_t times[2]; 2528 2529 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 2530 times, sizeof (times)); 2531 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 2532 XVA_SET_RTN(xvap, XAT_CREATETIME); 2533 } 2534 2535 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2536 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2537 XVA_SET_RTN(xvap, XAT_REPARSE); 2538 } 2539 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2540 xoap->xoa_generation = zp->z_gen; 2541 XVA_SET_RTN(xvap, XAT_GEN); 2542 } 2543 } 2544 2545 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2546 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2547 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2548 2549 mutex_exit(&zp->z_lock); 2550 2551 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks); 2552 2553 if (zp->z_blksz == 0) { 2554 /* 2555 * Block size hasn't been set; suggest maximal I/O transfers. 2556 */ 2557 vap->va_blksize = zfsvfs->z_max_blksz; 2558 } 2559 2560 ZFS_EXIT(zfsvfs); 2561 return (0); 2562 } 2563 2564 /* 2565 * Set the file attributes to the values contained in the 2566 * vattr structure. 2567 * 2568 * IN: vp - vnode of file to be modified. 2569 * vap - new attribute values. 2570 * If AT_XVATTR set, then optional attrs are being set 2571 * flags - ATTR_UTIME set if non-default time values provided. 2572 * - ATTR_NOACLCHECK (CIFS context only). 2573 * cr - credentials of caller. 2574 * ct - caller context 2575 * 2576 * RETURN: 0 if success 2577 * error code if failure 2578 * 2579 * Timestamps: 2580 * vp - ctime updated, mtime updated if size changed. 2581 */ 2582 /* ARGSUSED */ 2583 static int 2584 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2585 caller_context_t *ct) 2586 { 2587 znode_t *zp = VTOZ(vp); 2588 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2589 zilog_t *zilog; 2590 dmu_tx_t *tx; 2591 vattr_t oldva; 2592 xvattr_t tmpxvattr; 2593 uint_t mask = vap->va_mask; 2594 uint_t saved_mask; 2595 int trim_mask = 0; 2596 uint64_t new_mode; 2597 uint64_t new_uid, new_gid; 2598 uint64_t xattr_obj = 0; 2599 uint64_t mtime[2], ctime[2]; 2600 znode_t *attrzp; 2601 int need_policy = FALSE; 2602 int err, err2; 2603 zfs_fuid_info_t *fuidp = NULL; 2604 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2605 xoptattr_t *xoap; 2606 zfs_acl_t *aclp = NULL; 2607 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2608 boolean_t fuid_dirtied = B_FALSE; 2609 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 2610 int count = 0, xattr_count = 0; 2611 2612 if (mask == 0) 2613 return (0); 2614 2615 if (mask & AT_NOSET) 2616 return (EINVAL); 2617 2618 ZFS_ENTER(zfsvfs); 2619 ZFS_VERIFY_ZP(zp); 2620 2621 zilog = zfsvfs->z_log; 2622 2623 /* 2624 * Make sure that if we have ephemeral uid/gid or xvattr specified 2625 * that file system is at proper version level 2626 */ 2627 2628 if (zfsvfs->z_use_fuids == B_FALSE && 2629 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2630 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2631 (mask & AT_XVATTR))) { 2632 ZFS_EXIT(zfsvfs); 2633 return (EINVAL); 2634 } 2635 2636 if (mask & AT_SIZE && vp->v_type == VDIR) { 2637 ZFS_EXIT(zfsvfs); 2638 return (EISDIR); 2639 } 2640 2641 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 2642 ZFS_EXIT(zfsvfs); 2643 return (EINVAL); 2644 } 2645 2646 /* 2647 * If this is an xvattr_t, then get a pointer to the structure of 2648 * optional attributes. If this is NULL, then we have a vattr_t. 2649 */ 2650 xoap = xva_getxoptattr(xvap); 2651 2652 xva_init(&tmpxvattr); 2653 2654 /* 2655 * Immutable files can only alter immutable bit and atime 2656 */ 2657 if ((zp->z_pflags & ZFS_IMMUTABLE) && 2658 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2659 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 2660 ZFS_EXIT(zfsvfs); 2661 return (EPERM); 2662 } 2663 2664 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 2665 ZFS_EXIT(zfsvfs); 2666 return (EPERM); 2667 } 2668 2669 /* 2670 * Verify timestamps doesn't overflow 32 bits. 2671 * ZFS can handle large timestamps, but 32bit syscalls can't 2672 * handle times greater than 2039. This check should be removed 2673 * once large timestamps are fully supported. 2674 */ 2675 if (mask & (AT_ATIME | AT_MTIME)) { 2676 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 2677 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 2678 ZFS_EXIT(zfsvfs); 2679 return (EOVERFLOW); 2680 } 2681 } 2682 2683 top: 2684 attrzp = NULL; 2685 2686 /* Can this be moved to before the top label? */ 2687 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2688 ZFS_EXIT(zfsvfs); 2689 return (EROFS); 2690 } 2691 2692 /* 2693 * First validate permissions 2694 */ 2695 2696 if (mask & AT_SIZE) { 2697 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr); 2698 if (err) { 2699 ZFS_EXIT(zfsvfs); 2700 return (err); 2701 } 2702 /* 2703 * XXX - Note, we are not providing any open 2704 * mode flags here (like FNDELAY), so we may 2705 * block if there are locks present... this 2706 * should be addressed in openat(). 2707 */ 2708 /* XXX - would it be OK to generate a log record here? */ 2709 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2710 if (err) { 2711 ZFS_EXIT(zfsvfs); 2712 return (err); 2713 } 2714 } 2715 2716 if (mask & (AT_ATIME|AT_MTIME) || 2717 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2718 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2719 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2720 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2721 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 2722 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2723 skipaclchk, cr); 2724 } 2725 2726 if (mask & (AT_UID|AT_GID)) { 2727 int idmask = (mask & (AT_UID|AT_GID)); 2728 int take_owner; 2729 int take_group; 2730 2731 /* 2732 * NOTE: even if a new mode is being set, 2733 * we may clear S_ISUID/S_ISGID bits. 2734 */ 2735 2736 if (!(mask & AT_MODE)) 2737 vap->va_mode = zp->z_mode; 2738 2739 /* 2740 * Take ownership or chgrp to group we are a member of 2741 */ 2742 2743 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2744 take_group = (mask & AT_GID) && 2745 zfs_groupmember(zfsvfs, vap->va_gid, cr); 2746 2747 /* 2748 * If both AT_UID and AT_GID are set then take_owner and 2749 * take_group must both be set in order to allow taking 2750 * ownership. 2751 * 2752 * Otherwise, send the check through secpolicy_vnode_setattr() 2753 * 2754 */ 2755 2756 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2757 ((idmask == AT_UID) && take_owner) || 2758 ((idmask == AT_GID) && take_group)) { 2759 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 2760 skipaclchk, cr) == 0) { 2761 /* 2762 * Remove setuid/setgid for non-privileged users 2763 */ 2764 secpolicy_setid_clear(vap, cr); 2765 trim_mask = (mask & (AT_UID|AT_GID)); 2766 } else { 2767 need_policy = TRUE; 2768 } 2769 } else { 2770 need_policy = TRUE; 2771 } 2772 } 2773 2774 mutex_enter(&zp->z_lock); 2775 oldva.va_mode = zp->z_mode; 2776 oldva.va_uid = zp->z_uid; 2777 oldva.va_gid = zp->z_gid; 2778 if (mask & AT_XVATTR) { 2779 /* 2780 * Update xvattr mask to include only those attributes 2781 * that are actually changing. 2782 * 2783 * the bits will be restored prior to actually setting 2784 * the attributes so the caller thinks they were set. 2785 */ 2786 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2787 if (xoap->xoa_appendonly != 2788 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 2789 need_policy = TRUE; 2790 } else { 2791 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 2792 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 2793 } 2794 } 2795 2796 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2797 if (xoap->xoa_nounlink != 2798 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 2799 need_policy = TRUE; 2800 } else { 2801 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 2802 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 2803 } 2804 } 2805 2806 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2807 if (xoap->xoa_immutable != 2808 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 2809 need_policy = TRUE; 2810 } else { 2811 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 2812 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 2813 } 2814 } 2815 2816 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2817 if (xoap->xoa_nodump != 2818 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 2819 need_policy = TRUE; 2820 } else { 2821 XVA_CLR_REQ(xvap, XAT_NODUMP); 2822 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 2823 } 2824 } 2825 2826 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2827 if (xoap->xoa_av_modified != 2828 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 2829 need_policy = TRUE; 2830 } else { 2831 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 2832 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 2833 } 2834 } 2835 2836 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2837 if ((vp->v_type != VREG && 2838 xoap->xoa_av_quarantined) || 2839 xoap->xoa_av_quarantined != 2840 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 2841 need_policy = TRUE; 2842 } else { 2843 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 2844 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 2845 } 2846 } 2847 2848 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2849 mutex_exit(&zp->z_lock); 2850 ZFS_EXIT(zfsvfs); 2851 return (EPERM); 2852 } 2853 2854 if (need_policy == FALSE && 2855 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 2856 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 2857 need_policy = TRUE; 2858 } 2859 } 2860 2861 mutex_exit(&zp->z_lock); 2862 2863 if (mask & AT_MODE) { 2864 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 2865 err = secpolicy_setid_setsticky_clear(vp, vap, 2866 &oldva, cr); 2867 if (err) { 2868 ZFS_EXIT(zfsvfs); 2869 return (err); 2870 } 2871 trim_mask |= AT_MODE; 2872 } else { 2873 need_policy = TRUE; 2874 } 2875 } 2876 2877 if (need_policy) { 2878 /* 2879 * If trim_mask is set then take ownership 2880 * has been granted or write_acl is present and user 2881 * has the ability to modify mode. In that case remove 2882 * UID|GID and or MODE from mask so that 2883 * secpolicy_vnode_setattr() doesn't revoke it. 2884 */ 2885 2886 if (trim_mask) { 2887 saved_mask = vap->va_mask; 2888 vap->va_mask &= ~trim_mask; 2889 } 2890 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2891 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 2892 if (err) { 2893 ZFS_EXIT(zfsvfs); 2894 return (err); 2895 } 2896 2897 if (trim_mask) 2898 vap->va_mask |= saved_mask; 2899 } 2900 2901 /* 2902 * secpolicy_vnode_setattr, or take ownership may have 2903 * changed va_mask 2904 */ 2905 mask = vap->va_mask; 2906 2907 if ((mask & (AT_UID | AT_GID))) { 2908 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xattr_obj, 2909 sizeof (xattr_obj)); 2910 2911 if (xattr_obj) { 2912 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 2913 if (err) 2914 goto out2; 2915 } 2916 if (mask & AT_UID) { 2917 new_uid = zfs_fuid_create(zfsvfs, 2918 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 2919 if (vap->va_uid != zp->z_uid && 2920 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 2921 err = EDQUOT; 2922 goto out2; 2923 } 2924 } 2925 2926 if (mask & AT_GID) { 2927 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 2928 cr, ZFS_GROUP, &fuidp); 2929 if (new_gid != zp->z_gid && 2930 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 2931 err = EDQUOT; 2932 goto out2; 2933 } 2934 } 2935 } 2936 tx = dmu_tx_create(zfsvfs->z_os); 2937 2938 if (mask & AT_MODE) { 2939 uint64_t pmode = zp->z_mode; 2940 uint64_t acl_obj; 2941 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2942 2943 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 2944 goto out; 2945 2946 mutex_enter(&zp->z_lock); 2947 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 2948 /* 2949 * Are we upgrading ACL from old V0 format 2950 * to V1 format? 2951 */ 2952 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 2953 zfs_znode_acl_version(zp) == 2954 ZFS_ACL_VERSION_INITIAL) { 2955 dmu_tx_hold_free(tx, acl_obj, 0, 2956 DMU_OBJECT_END); 2957 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2958 0, aclp->z_acl_bytes); 2959 } else { 2960 dmu_tx_hold_write(tx, acl_obj, 0, 2961 aclp->z_acl_bytes); 2962 } 2963 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2964 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2965 0, aclp->z_acl_bytes); 2966 } 2967 mutex_exit(&zp->z_lock); 2968 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2969 } else { 2970 if ((mask & AT_XVATTR) && 2971 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 2972 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2973 else 2974 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2975 } 2976 2977 if (attrzp) { 2978 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 2979 } 2980 2981 fuid_dirtied = zfsvfs->z_fuid_dirty; 2982 if (fuid_dirtied) 2983 zfs_fuid_txhold(zfsvfs, tx); 2984 2985 zfs_sa_upgrade_txholds(tx, zp); 2986 2987 err = dmu_tx_assign(tx, TXG_NOWAIT); 2988 if (err) { 2989 if (err == ERESTART) 2990 dmu_tx_wait(tx); 2991 goto out; 2992 } 2993 2994 count = 0; 2995 /* 2996 * Set each attribute requested. 2997 * We group settings according to the locks they need to acquire. 2998 * 2999 * Note: you cannot set ctime directly, although it will be 3000 * updated as a side-effect of calling this function. 3001 */ 3002 3003 3004 if (mask & (AT_UID|AT_GID|AT_MODE)) 3005 mutex_enter(&zp->z_acl_lock); 3006 mutex_enter(&zp->z_lock); 3007 3008 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3009 &zp->z_pflags, sizeof (zp->z_pflags)); 3010 3011 if (attrzp) { 3012 if (mask & (AT_UID|AT_GID|AT_MODE)) 3013 mutex_enter(&attrzp->z_acl_lock); 3014 mutex_enter(&attrzp->z_lock); 3015 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3016 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3017 sizeof (attrzp->z_pflags)); 3018 } 3019 3020 if (mask & (AT_UID|AT_GID)) { 3021 3022 if (mask & AT_UID) { 3023 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3024 &new_uid, sizeof (new_uid)); 3025 zp->z_uid = zfs_fuid_map_id(zfsvfs, new_uid, 3026 cr, ZFS_OWNER); 3027 if (attrzp) { 3028 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3029 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3030 sizeof (new_uid)); 3031 attrzp->z_uid = zp->z_uid; 3032 } 3033 } 3034 3035 if (mask & AT_GID) { 3036 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3037 NULL, &new_gid, sizeof (new_gid)); 3038 zp->z_gid = zfs_fuid_map_id(zfsvfs, new_gid, cr, 3039 ZFS_GROUP); 3040 if (attrzp) { 3041 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3042 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 3043 sizeof (new_gid)); 3044 attrzp->z_gid = zp->z_gid; 3045 } 3046 } 3047 if (!(mask & AT_MODE)) { 3048 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 3049 NULL, &new_mode, sizeof (new_mode)); 3050 new_mode = zp->z_mode; 3051 } 3052 err = zfs_acl_chown_setattr(zp); 3053 ASSERT(err == 0); 3054 if (attrzp) { 3055 err = zfs_acl_chown_setattr(attrzp); 3056 ASSERT(err == 0); 3057 } 3058 } 3059 3060 if (mask & AT_MODE) { 3061 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 3062 &new_mode, sizeof (new_mode)); 3063 zp->z_mode = new_mode; 3064 ASSERT3U((uintptr_t)aclp, !=, NULL); 3065 err = zfs_aclset_common(zp, aclp, cr, tx); 3066 ASSERT3U(err, ==, 0); 3067 zp->z_acl_cached = aclp; 3068 aclp = NULL; 3069 } 3070 3071 3072 if (mask & AT_ATIME) { 3073 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 3074 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 3075 &zp->z_atime, sizeof (zp->z_atime)); 3076 } 3077 3078 if (mask & AT_MTIME) { 3079 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 3080 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 3081 mtime, sizeof (mtime)); 3082 } 3083 3084 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 3085 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 3086 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 3087 NULL, mtime, sizeof (mtime)); 3088 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3089 &ctime, sizeof (ctime)); 3090 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 3091 B_TRUE); 3092 } else if (mask != 0) { 3093 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3094 &ctime, sizeof (ctime)); 3095 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime, 3096 B_TRUE); 3097 if (attrzp) { 3098 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3099 SA_ZPL_CTIME(zfsvfs), NULL, 3100 &ctime, sizeof (ctime)); 3101 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 3102 mtime, ctime, B_TRUE); 3103 } 3104 } 3105 /* 3106 * Do this after setting timestamps to prevent timestamp 3107 * update from toggling bit 3108 */ 3109 3110 if (xoap && (mask & AT_XVATTR)) { 3111 3112 /* 3113 * restore trimmed off masks 3114 * so that return masks can be set for caller. 3115 */ 3116 3117 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 3118 XVA_SET_REQ(xvap, XAT_APPENDONLY); 3119 } 3120 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 3121 XVA_SET_REQ(xvap, XAT_NOUNLINK); 3122 } 3123 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 3124 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 3125 } 3126 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 3127 XVA_SET_REQ(xvap, XAT_NODUMP); 3128 } 3129 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 3130 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 3131 } 3132 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 3133 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 3134 } 3135 3136 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3137 ASSERT(vp->v_type == VREG); 3138 3139 zfs_xvattr_set(zp, xvap, tx); 3140 } 3141 3142 if (fuid_dirtied) 3143 zfs_fuid_sync(zfsvfs, tx); 3144 3145 if (mask != 0) 3146 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 3147 3148 mutex_exit(&zp->z_lock); 3149 if (mask & (AT_UID|AT_GID|AT_MODE)) 3150 mutex_exit(&zp->z_acl_lock); 3151 3152 if (attrzp) { 3153 if (mask & (AT_UID|AT_GID|AT_MODE)) 3154 mutex_exit(&attrzp->z_acl_lock); 3155 mutex_exit(&attrzp->z_lock); 3156 } 3157 out: 3158 if (err == 0 && attrzp) { 3159 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 3160 xattr_count, tx); 3161 ASSERT(err2 == 0); 3162 } 3163 3164 if (attrzp) 3165 VN_RELE(ZTOV(attrzp)); 3166 if (aclp) 3167 zfs_acl_free(aclp); 3168 3169 if (fuidp) { 3170 zfs_fuid_info_free(fuidp); 3171 fuidp = NULL; 3172 } 3173 3174 if (err) { 3175 dmu_tx_abort(tx); 3176 if (err == ERESTART) 3177 goto top; 3178 } else { 3179 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3180 dmu_tx_commit(tx); 3181 } 3182 3183 3184 out2: 3185 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3186 zil_commit(zilog, 0); 3187 3188 ZFS_EXIT(zfsvfs); 3189 return (err); 3190 } 3191 3192 typedef struct zfs_zlock { 3193 krwlock_t *zl_rwlock; /* lock we acquired */ 3194 znode_t *zl_znode; /* znode we held */ 3195 struct zfs_zlock *zl_next; /* next in list */ 3196 } zfs_zlock_t; 3197 3198 /* 3199 * Drop locks and release vnodes that were held by zfs_rename_lock(). 3200 */ 3201 static void 3202 zfs_rename_unlock(zfs_zlock_t **zlpp) 3203 { 3204 zfs_zlock_t *zl; 3205 3206 while ((zl = *zlpp) != NULL) { 3207 if (zl->zl_znode != NULL) 3208 VN_RELE(ZTOV(zl->zl_znode)); 3209 rw_exit(zl->zl_rwlock); 3210 *zlpp = zl->zl_next; 3211 kmem_free(zl, sizeof (*zl)); 3212 } 3213 } 3214 3215 /* 3216 * Search back through the directory tree, using the ".." entries. 3217 * Lock each directory in the chain to prevent concurrent renames. 3218 * Fail any attempt to move a directory into one of its own descendants. 3219 * XXX - z_parent_lock can overlap with map or grow locks 3220 */ 3221 static int 3222 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 3223 { 3224 zfs_zlock_t *zl; 3225 znode_t *zp = tdzp; 3226 uint64_t rootid = zp->z_zfsvfs->z_root; 3227 uint64_t oidp = zp->z_id; 3228 krwlock_t *rwlp = &szp->z_parent_lock; 3229 krw_t rw = RW_WRITER; 3230 3231 /* 3232 * First pass write-locks szp and compares to zp->z_id. 3233 * Later passes read-lock zp and compare to zp->z_parent. 3234 */ 3235 do { 3236 if (!rw_tryenter(rwlp, rw)) { 3237 /* 3238 * Another thread is renaming in this path. 3239 * Note that if we are a WRITER, we don't have any 3240 * parent_locks held yet. 3241 */ 3242 if (rw == RW_READER && zp->z_id > szp->z_id) { 3243 /* 3244 * Drop our locks and restart 3245 */ 3246 zfs_rename_unlock(&zl); 3247 *zlpp = NULL; 3248 zp = tdzp; 3249 oidp = zp->z_id; 3250 rwlp = &szp->z_parent_lock; 3251 rw = RW_WRITER; 3252 continue; 3253 } else { 3254 /* 3255 * Wait for other thread to drop its locks 3256 */ 3257 rw_enter(rwlp, rw); 3258 } 3259 } 3260 3261 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 3262 zl->zl_rwlock = rwlp; 3263 zl->zl_znode = NULL; 3264 zl->zl_next = *zlpp; 3265 *zlpp = zl; 3266 3267 if (oidp == szp->z_id) /* We're a descendant of szp */ 3268 return (EINVAL); 3269 3270 if (oidp == rootid) /* We've hit the top */ 3271 return (0); 3272 3273 if (rw == RW_READER) { /* i.e. not the first pass */ 3274 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp); 3275 if (error) 3276 return (error); 3277 zl->zl_znode = zp; 3278 } 3279 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs), 3280 &oidp, sizeof (oidp)); 3281 rwlp = &zp->z_parent_lock; 3282 rw = RW_READER; 3283 3284 } while (zp->z_id != sdzp->z_id); 3285 3286 return (0); 3287 } 3288 3289 /* 3290 * Move an entry from the provided source directory to the target 3291 * directory. Change the entry name as indicated. 3292 * 3293 * IN: sdvp - Source directory containing the "old entry". 3294 * snm - Old entry name. 3295 * tdvp - Target directory to contain the "new entry". 3296 * tnm - New entry name. 3297 * cr - credentials of caller. 3298 * ct - caller context 3299 * flags - case flags 3300 * 3301 * RETURN: 0 if success 3302 * error code if failure 3303 * 3304 * Timestamps: 3305 * sdvp,tdvp - ctime|mtime updated 3306 */ 3307 /*ARGSUSED*/ 3308 static int 3309 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 3310 caller_context_t *ct, int flags) 3311 { 3312 znode_t *tdzp, *szp, *tzp; 3313 znode_t *sdzp = VTOZ(sdvp); 3314 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 3315 zilog_t *zilog; 3316 vnode_t *realvp; 3317 zfs_dirlock_t *sdl, *tdl; 3318 dmu_tx_t *tx; 3319 zfs_zlock_t *zl; 3320 int cmp, serr, terr; 3321 int error = 0; 3322 int zflg = 0; 3323 3324 ZFS_ENTER(zfsvfs); 3325 ZFS_VERIFY_ZP(sdzp); 3326 zilog = zfsvfs->z_log; 3327 3328 /* 3329 * Make sure we have the real vp for the target directory. 3330 */ 3331 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 3332 tdvp = realvp; 3333 3334 if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) { 3335 ZFS_EXIT(zfsvfs); 3336 return (EXDEV); 3337 } 3338 3339 tdzp = VTOZ(tdvp); 3340 ZFS_VERIFY_ZP(tdzp); 3341 if (zfsvfs->z_utf8 && u8_validate(tnm, 3342 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3343 ZFS_EXIT(zfsvfs); 3344 return (EILSEQ); 3345 } 3346 3347 if (flags & FIGNORECASE) 3348 zflg |= ZCILOOK; 3349 3350 top: 3351 szp = NULL; 3352 tzp = NULL; 3353 zl = NULL; 3354 3355 /* 3356 * This is to prevent the creation of links into attribute space 3357 * by renaming a linked file into/outof an attribute directory. 3358 * See the comment in zfs_link() for why this is considered bad. 3359 */ 3360 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3361 ZFS_EXIT(zfsvfs); 3362 return (EINVAL); 3363 } 3364 3365 /* 3366 * Lock source and target directory entries. To prevent deadlock, 3367 * a lock ordering must be defined. We lock the directory with 3368 * the smallest object id first, or if it's a tie, the one with 3369 * the lexically first name. 3370 */ 3371 if (sdzp->z_id < tdzp->z_id) { 3372 cmp = -1; 3373 } else if (sdzp->z_id > tdzp->z_id) { 3374 cmp = 1; 3375 } else { 3376 /* 3377 * First compare the two name arguments without 3378 * considering any case folding. 3379 */ 3380 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 3381 3382 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 3383 ASSERT(error == 0 || !zfsvfs->z_utf8); 3384 if (cmp == 0) { 3385 /* 3386 * POSIX: "If the old argument and the new argument 3387 * both refer to links to the same existing file, 3388 * the rename() function shall return successfully 3389 * and perform no other action." 3390 */ 3391 ZFS_EXIT(zfsvfs); 3392 return (0); 3393 } 3394 /* 3395 * If the file system is case-folding, then we may 3396 * have some more checking to do. A case-folding file 3397 * system is either supporting mixed case sensitivity 3398 * access or is completely case-insensitive. Note 3399 * that the file system is always case preserving. 3400 * 3401 * In mixed sensitivity mode case sensitive behavior 3402 * is the default. FIGNORECASE must be used to 3403 * explicitly request case insensitive behavior. 3404 * 3405 * If the source and target names provided differ only 3406 * by case (e.g., a request to rename 'tim' to 'Tim'), 3407 * we will treat this as a special case in the 3408 * case-insensitive mode: as long as the source name 3409 * is an exact match, we will allow this to proceed as 3410 * a name-change request. 3411 */ 3412 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3413 (zfsvfs->z_case == ZFS_CASE_MIXED && 3414 flags & FIGNORECASE)) && 3415 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3416 &error) == 0) { 3417 /* 3418 * case preserving rename request, require exact 3419 * name matches 3420 */ 3421 zflg |= ZCIEXACT; 3422 zflg &= ~ZCILOOK; 3423 } 3424 } 3425 3426 /* 3427 * If the source and destination directories are the same, we should 3428 * grab the z_name_lock of that directory only once. 3429 */ 3430 if (sdzp == tdzp) { 3431 zflg |= ZHAVELOCK; 3432 rw_enter(&sdzp->z_name_lock, RW_READER); 3433 } 3434 3435 if (cmp < 0) { 3436 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3437 ZEXISTS | zflg, NULL, NULL); 3438 terr = zfs_dirent_lock(&tdl, 3439 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3440 } else { 3441 terr = zfs_dirent_lock(&tdl, 3442 tdzp, tnm, &tzp, zflg, NULL, NULL); 3443 serr = zfs_dirent_lock(&sdl, 3444 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3445 NULL, NULL); 3446 } 3447 3448 if (serr) { 3449 /* 3450 * Source entry invalid or not there. 3451 */ 3452 if (!terr) { 3453 zfs_dirent_unlock(tdl); 3454 if (tzp) 3455 VN_RELE(ZTOV(tzp)); 3456 } 3457 3458 if (sdzp == tdzp) 3459 rw_exit(&sdzp->z_name_lock); 3460 3461 if (strcmp(snm, "..") == 0) 3462 serr = EINVAL; 3463 ZFS_EXIT(zfsvfs); 3464 return (serr); 3465 } 3466 if (terr) { 3467 zfs_dirent_unlock(sdl); 3468 VN_RELE(ZTOV(szp)); 3469 3470 if (sdzp == tdzp) 3471 rw_exit(&sdzp->z_name_lock); 3472 3473 if (strcmp(tnm, "..") == 0) 3474 terr = EINVAL; 3475 ZFS_EXIT(zfsvfs); 3476 return (terr); 3477 } 3478 3479 /* 3480 * Must have write access at the source to remove the old entry 3481 * and write access at the target to create the new entry. 3482 * Note that if target and source are the same, this can be 3483 * done in a single check. 3484 */ 3485 3486 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3487 goto out; 3488 3489 if (ZTOV(szp)->v_type == VDIR) { 3490 /* 3491 * Check to make sure rename is valid. 3492 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3493 */ 3494 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3495 goto out; 3496 } 3497 3498 /* 3499 * Does target exist? 3500 */ 3501 if (tzp) { 3502 /* 3503 * Source and target must be the same type. 3504 */ 3505 if (ZTOV(szp)->v_type == VDIR) { 3506 if (ZTOV(tzp)->v_type != VDIR) { 3507 error = ENOTDIR; 3508 goto out; 3509 } 3510 } else { 3511 if (ZTOV(tzp)->v_type == VDIR) { 3512 error = EISDIR; 3513 goto out; 3514 } 3515 } 3516 /* 3517 * POSIX dictates that when the source and target 3518 * entries refer to the same file object, rename 3519 * must do nothing and exit without error. 3520 */ 3521 if (szp->z_id == tzp->z_id) { 3522 error = 0; 3523 goto out; 3524 } 3525 } 3526 3527 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3528 if (tzp) 3529 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3530 3531 /* 3532 * notify the target directory if it is not the same 3533 * as source directory. 3534 */ 3535 if (tdvp != sdvp) { 3536 vnevent_rename_dest_dir(tdvp, ct); 3537 } 3538 3539 tx = dmu_tx_create(zfsvfs->z_os); 3540 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3541 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 3542 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3543 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3544 if (sdzp != tdzp) { 3545 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 3546 zfs_sa_upgrade_txholds(tx, tdzp); 3547 } 3548 if (tzp) { 3549 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 3550 zfs_sa_upgrade_txholds(tx, tzp); 3551 } 3552 3553 zfs_sa_upgrade_txholds(tx, szp); 3554 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3555 error = dmu_tx_assign(tx, TXG_NOWAIT); 3556 if (error) { 3557 if (zl != NULL) 3558 zfs_rename_unlock(&zl); 3559 zfs_dirent_unlock(sdl); 3560 zfs_dirent_unlock(tdl); 3561 3562 if (sdzp == tdzp) 3563 rw_exit(&sdzp->z_name_lock); 3564 3565 VN_RELE(ZTOV(szp)); 3566 if (tzp) 3567 VN_RELE(ZTOV(tzp)); 3568 if (error == ERESTART) { 3569 dmu_tx_wait(tx); 3570 dmu_tx_abort(tx); 3571 goto top; 3572 } 3573 dmu_tx_abort(tx); 3574 ZFS_EXIT(zfsvfs); 3575 return (error); 3576 } 3577 3578 if (tzp) /* Attempt to remove the existing target */ 3579 error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 3580 3581 if (error == 0) { 3582 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 3583 if (error == 0) { 3584 szp->z_pflags |= ZFS_AV_MODIFIED; 3585 3586 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 3587 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 3588 ASSERT3U(error, ==, 0); 3589 3590 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 3591 if (error == 0) { 3592 zfs_log_rename(zilog, tx, TX_RENAME | 3593 (flags & FIGNORECASE ? TX_CI : 0), sdzp, 3594 sdl->dl_name, tdzp, tdl->dl_name, szp); 3595 3596 /* 3597 * Update path information for the target vnode 3598 */ 3599 vn_renamepath(tdvp, ZTOV(szp), tnm, 3600 strlen(tnm)); 3601 } else { 3602 /* 3603 * At this point, we have successfully created 3604 * the target name, but have failed to remove 3605 * the source name. Since the create was done 3606 * with the ZRENAMING flag, there are 3607 * complications; for one, the link count is 3608 * wrong. The easiest way to deal with this 3609 * is to remove the newly created target, and 3610 * return the original error. This must 3611 * succeed; fortunately, it is very unlikely to 3612 * fail, since we just created it. 3613 */ 3614 VERIFY3U(zfs_link_destroy(tdl, szp, tx, 3615 ZRENAMING, NULL), ==, 0); 3616 } 3617 } 3618 } 3619 3620 dmu_tx_commit(tx); 3621 out: 3622 if (zl != NULL) 3623 zfs_rename_unlock(&zl); 3624 3625 zfs_dirent_unlock(sdl); 3626 zfs_dirent_unlock(tdl); 3627 3628 if (sdzp == tdzp) 3629 rw_exit(&sdzp->z_name_lock); 3630 3631 3632 VN_RELE(ZTOV(szp)); 3633 if (tzp) 3634 VN_RELE(ZTOV(tzp)); 3635 3636 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3637 zil_commit(zilog, 0); 3638 3639 ZFS_EXIT(zfsvfs); 3640 return (error); 3641 } 3642 3643 /* 3644 * Insert the indicated symbolic reference entry into the directory. 3645 * 3646 * IN: dvp - Directory to contain new symbolic link. 3647 * link - Name for new symlink entry. 3648 * vap - Attributes of new entry. 3649 * target - Target path of new symlink. 3650 * cr - credentials of caller. 3651 * ct - caller context 3652 * flags - case flags 3653 * 3654 * RETURN: 0 if success 3655 * error code if failure 3656 * 3657 * Timestamps: 3658 * dvp - ctime|mtime updated 3659 */ 3660 /*ARGSUSED*/ 3661 static int 3662 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr, 3663 caller_context_t *ct, int flags) 3664 { 3665 znode_t *zp, *dzp = VTOZ(dvp); 3666 zfs_dirlock_t *dl; 3667 dmu_tx_t *tx; 3668 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3669 zilog_t *zilog; 3670 uint64_t len = strlen(link); 3671 int error; 3672 int zflg = ZNEW; 3673 zfs_acl_ids_t acl_ids; 3674 boolean_t fuid_dirtied; 3675 uint64_t txtype = TX_SYMLINK; 3676 3677 ASSERT(vap->va_type == VLNK); 3678 3679 ZFS_ENTER(zfsvfs); 3680 ZFS_VERIFY_ZP(dzp); 3681 zilog = zfsvfs->z_log; 3682 3683 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 3684 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3685 ZFS_EXIT(zfsvfs); 3686 return (EILSEQ); 3687 } 3688 if (flags & FIGNORECASE) 3689 zflg |= ZCILOOK; 3690 3691 if (len > MAXPATHLEN) { 3692 ZFS_EXIT(zfsvfs); 3693 return (ENAMETOOLONG); 3694 } 3695 3696 if ((error = zfs_acl_ids_create(dzp, 0, 3697 vap, cr, NULL, &acl_ids)) != 0) { 3698 ZFS_EXIT(zfsvfs); 3699 return (error); 3700 } 3701 top: 3702 /* 3703 * Attempt to lock directory; fail if entry already exists. 3704 */ 3705 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 3706 if (error) { 3707 zfs_acl_ids_free(&acl_ids); 3708 ZFS_EXIT(zfsvfs); 3709 return (error); 3710 } 3711 3712 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3713 zfs_acl_ids_free(&acl_ids); 3714 zfs_dirent_unlock(dl); 3715 ZFS_EXIT(zfsvfs); 3716 return (error); 3717 } 3718 3719 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 3720 zfs_acl_ids_free(&acl_ids); 3721 zfs_dirent_unlock(dl); 3722 ZFS_EXIT(zfsvfs); 3723 return (EDQUOT); 3724 } 3725 tx = dmu_tx_create(zfsvfs->z_os); 3726 fuid_dirtied = zfsvfs->z_fuid_dirty; 3727 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 3728 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3729 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 3730 ZFS_SA_BASE_ATTR_SIZE + len); 3731 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 3732 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3733 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 3734 acl_ids.z_aclp->z_acl_bytes); 3735 } 3736 if (fuid_dirtied) 3737 zfs_fuid_txhold(zfsvfs, tx); 3738 error = dmu_tx_assign(tx, TXG_NOWAIT); 3739 if (error) { 3740 zfs_dirent_unlock(dl); 3741 if (error == ERESTART) { 3742 dmu_tx_wait(tx); 3743 dmu_tx_abort(tx); 3744 goto top; 3745 } 3746 zfs_acl_ids_free(&acl_ids); 3747 dmu_tx_abort(tx); 3748 ZFS_EXIT(zfsvfs); 3749 return (error); 3750 } 3751 3752 /* 3753 * Create a new object for the symlink. 3754 * for version 4 ZPL datsets the symlink will be an SA attribute 3755 */ 3756 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 3757 3758 if (fuid_dirtied) 3759 zfs_fuid_sync(zfsvfs, tx); 3760 3761 mutex_enter(&zp->z_lock); 3762 if (zp->z_is_sa) 3763 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 3764 link, len, tx); 3765 else 3766 zfs_sa_symlink(zp, link, len, tx); 3767 mutex_exit(&zp->z_lock); 3768 3769 zp->z_size = len; 3770 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 3771 &zp->z_size, sizeof (zp->z_size), tx); 3772 /* 3773 * Insert the new object into the directory. 3774 */ 3775 (void) zfs_link_create(dl, zp, tx, ZNEW); 3776 3777 if (flags & FIGNORECASE) 3778 txtype |= TX_CI; 3779 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 3780 3781 zfs_acl_ids_free(&acl_ids); 3782 3783 dmu_tx_commit(tx); 3784 3785 zfs_dirent_unlock(dl); 3786 3787 VN_RELE(ZTOV(zp)); 3788 3789 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3790 zil_commit(zilog, 0); 3791 3792 ZFS_EXIT(zfsvfs); 3793 return (error); 3794 } 3795 3796 /* 3797 * Return, in the buffer contained in the provided uio structure, 3798 * the symbolic path referred to by vp. 3799 * 3800 * IN: vp - vnode of symbolic link. 3801 * uoip - structure to contain the link path. 3802 * cr - credentials of caller. 3803 * ct - caller context 3804 * 3805 * OUT: uio - structure to contain the link path. 3806 * 3807 * RETURN: 0 if success 3808 * error code if failure 3809 * 3810 * Timestamps: 3811 * vp - atime updated 3812 */ 3813 /* ARGSUSED */ 3814 static int 3815 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 3816 { 3817 znode_t *zp = VTOZ(vp); 3818 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3819 int error; 3820 3821 ZFS_ENTER(zfsvfs); 3822 ZFS_VERIFY_ZP(zp); 3823 3824 mutex_enter(&zp->z_lock); 3825 if (zp->z_is_sa) 3826 error = sa_lookup_uio(zp->z_sa_hdl, 3827 SA_ZPL_SYMLINK(zfsvfs), uio); 3828 else 3829 error = zfs_sa_readlink(zp, uio); 3830 mutex_exit(&zp->z_lock); 3831 3832 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3833 3834 ZFS_EXIT(zfsvfs); 3835 return (error); 3836 } 3837 3838 /* 3839 * Insert a new entry into directory tdvp referencing svp. 3840 * 3841 * IN: tdvp - Directory to contain new entry. 3842 * svp - vnode of new entry. 3843 * name - name of new entry. 3844 * cr - credentials of caller. 3845 * ct - caller context 3846 * 3847 * RETURN: 0 if success 3848 * error code if failure 3849 * 3850 * Timestamps: 3851 * tdvp - ctime|mtime updated 3852 * svp - ctime updated 3853 */ 3854 /* ARGSUSED */ 3855 static int 3856 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 3857 caller_context_t *ct, int flags) 3858 { 3859 znode_t *dzp = VTOZ(tdvp); 3860 znode_t *tzp, *szp; 3861 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3862 zilog_t *zilog; 3863 zfs_dirlock_t *dl; 3864 dmu_tx_t *tx; 3865 vnode_t *realvp; 3866 int error; 3867 int zf = ZNEW; 3868 uint64_t parent; 3869 3870 ASSERT(tdvp->v_type == VDIR); 3871 3872 ZFS_ENTER(zfsvfs); 3873 ZFS_VERIFY_ZP(dzp); 3874 zilog = zfsvfs->z_log; 3875 3876 if (VOP_REALVP(svp, &realvp, ct) == 0) 3877 svp = realvp; 3878 3879 /* 3880 * POSIX dictates that we return EPERM here. 3881 * Better choices include ENOTSUP or EISDIR. 3882 */ 3883 if (svp->v_type == VDIR) { 3884 ZFS_EXIT(zfsvfs); 3885 return (EPERM); 3886 } 3887 3888 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) { 3889 ZFS_EXIT(zfsvfs); 3890 return (EXDEV); 3891 } 3892 3893 szp = VTOZ(svp); 3894 ZFS_VERIFY_ZP(szp); 3895 3896 /* Prevent links to .zfs/shares files */ 3897 3898 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 3899 &parent, sizeof (uint64_t))) != 0) { 3900 ZFS_EXIT(zfsvfs); 3901 return (error); 3902 } 3903 if (parent == zfsvfs->z_shares_dir) { 3904 ZFS_EXIT(zfsvfs); 3905 return (EPERM); 3906 } 3907 3908 if (zfsvfs->z_utf8 && u8_validate(name, 3909 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3910 ZFS_EXIT(zfsvfs); 3911 return (EILSEQ); 3912 } 3913 if (flags & FIGNORECASE) 3914 zf |= ZCILOOK; 3915 3916 /* 3917 * We do not support links between attributes and non-attributes 3918 * because of the potential security risk of creating links 3919 * into "normal" file space in order to circumvent restrictions 3920 * imposed in attribute space. 3921 */ 3922 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 3923 ZFS_EXIT(zfsvfs); 3924 return (EINVAL); 3925 } 3926 3927 3928 if (szp->z_uid != crgetuid(cr) && 3929 secpolicy_basic_link(cr) != 0) { 3930 ZFS_EXIT(zfsvfs); 3931 return (EPERM); 3932 } 3933 3934 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3935 ZFS_EXIT(zfsvfs); 3936 return (error); 3937 } 3938 3939 top: 3940 /* 3941 * Attempt to lock directory; fail if entry already exists. 3942 */ 3943 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 3944 if (error) { 3945 ZFS_EXIT(zfsvfs); 3946 return (error); 3947 } 3948 3949 tx = dmu_tx_create(zfsvfs->z_os); 3950 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3951 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3952 zfs_sa_upgrade_txholds(tx, szp); 3953 zfs_sa_upgrade_txholds(tx, dzp); 3954 error = dmu_tx_assign(tx, TXG_NOWAIT); 3955 if (error) { 3956 zfs_dirent_unlock(dl); 3957 if (error == ERESTART) { 3958 dmu_tx_wait(tx); 3959 dmu_tx_abort(tx); 3960 goto top; 3961 } 3962 dmu_tx_abort(tx); 3963 ZFS_EXIT(zfsvfs); 3964 return (error); 3965 } 3966 3967 error = zfs_link_create(dl, szp, tx, 0); 3968 3969 if (error == 0) { 3970 uint64_t txtype = TX_LINK; 3971 if (flags & FIGNORECASE) 3972 txtype |= TX_CI; 3973 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 3974 } 3975 3976 dmu_tx_commit(tx); 3977 3978 zfs_dirent_unlock(dl); 3979 3980 if (error == 0) { 3981 vnevent_link(svp, ct); 3982 } 3983 3984 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3985 zil_commit(zilog, 0); 3986 3987 ZFS_EXIT(zfsvfs); 3988 return (error); 3989 } 3990 3991 /* 3992 * zfs_null_putapage() is used when the file system has been force 3993 * unmounted. It just drops the pages. 3994 */ 3995 /* ARGSUSED */ 3996 static int 3997 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 3998 size_t *lenp, int flags, cred_t *cr) 3999 { 4000 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 4001 return (0); 4002 } 4003 4004 /* 4005 * Push a page out to disk, klustering if possible. 4006 * 4007 * IN: vp - file to push page to. 4008 * pp - page to push. 4009 * flags - additional flags. 4010 * cr - credentials of caller. 4011 * 4012 * OUT: offp - start of range pushed. 4013 * lenp - len of range pushed. 4014 * 4015 * RETURN: 0 if success 4016 * error code if failure 4017 * 4018 * NOTE: callers must have locked the page to be pushed. On 4019 * exit, the page (and all other pages in the kluster) must be 4020 * unlocked. 4021 */ 4022 /* ARGSUSED */ 4023 static int 4024 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4025 size_t *lenp, int flags, cred_t *cr) 4026 { 4027 znode_t *zp = VTOZ(vp); 4028 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4029 dmu_tx_t *tx; 4030 u_offset_t off, koff; 4031 size_t len, klen; 4032 int err; 4033 4034 off = pp->p_offset; 4035 len = PAGESIZE; 4036 /* 4037 * If our blocksize is bigger than the page size, try to kluster 4038 * multiple pages so that we write a full block (thus avoiding 4039 * a read-modify-write). 4040 */ 4041 if (off < zp->z_size && zp->z_blksz > PAGESIZE) { 4042 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 4043 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 4044 ASSERT(koff <= zp->z_size); 4045 if (koff + klen > zp->z_size) 4046 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); 4047 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 4048 } 4049 ASSERT3U(btop(len), ==, btopr(len)); 4050 4051 /* 4052 * Can't push pages past end-of-file. 4053 */ 4054 if (off >= zp->z_size) { 4055 /* ignore all pages */ 4056 err = 0; 4057 goto out; 4058 } else if (off + len > zp->z_size) { 4059 int npages = btopr(zp->z_size - off); 4060 page_t *trunc; 4061 4062 page_list_break(&pp, &trunc, npages); 4063 /* ignore pages past end of file */ 4064 if (trunc) 4065 pvn_write_done(trunc, flags); 4066 len = zp->z_size - off; 4067 } 4068 4069 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4070 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4071 err = EDQUOT; 4072 goto out; 4073 } 4074 top: 4075 tx = dmu_tx_create(zfsvfs->z_os); 4076 dmu_tx_hold_write(tx, zp->z_id, off, len); 4077 4078 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4079 zfs_sa_upgrade_txholds(tx, zp); 4080 err = dmu_tx_assign(tx, TXG_NOWAIT); 4081 if (err != 0) { 4082 if (err == ERESTART) { 4083 dmu_tx_wait(tx); 4084 dmu_tx_abort(tx); 4085 goto top; 4086 } 4087 dmu_tx_abort(tx); 4088 goto out; 4089 } 4090 4091 if (zp->z_blksz <= PAGESIZE) { 4092 caddr_t va = zfs_map_page(pp, S_READ); 4093 ASSERT3U(len, <=, PAGESIZE); 4094 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 4095 zfs_unmap_page(pp, va); 4096 } else { 4097 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 4098 } 4099 4100 if (err == 0) { 4101 uint64_t mtime[2], ctime[2]; 4102 sa_bulk_attr_t bulk[3]; 4103 int count = 0; 4104 4105 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4106 &mtime, 16); 4107 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4108 &ctime, 16); 4109 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4110 &zp->z_pflags, 8); 4111 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4112 B_TRUE); 4113 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4114 } 4115 dmu_tx_commit(tx); 4116 4117 out: 4118 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 4119 if (offp) 4120 *offp = off; 4121 if (lenp) 4122 *lenp = len; 4123 4124 return (err); 4125 } 4126 4127 /* 4128 * Copy the portion of the file indicated from pages into the file. 4129 * The pages are stored in a page list attached to the files vnode. 4130 * 4131 * IN: vp - vnode of file to push page data to. 4132 * off - position in file to put data. 4133 * len - amount of data to write. 4134 * flags - flags to control the operation. 4135 * cr - credentials of caller. 4136 * ct - caller context. 4137 * 4138 * RETURN: 0 if success 4139 * error code if failure 4140 * 4141 * Timestamps: 4142 * vp - ctime|mtime updated 4143 */ 4144 /*ARGSUSED*/ 4145 static int 4146 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4147 caller_context_t *ct) 4148 { 4149 znode_t *zp = VTOZ(vp); 4150 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4151 page_t *pp; 4152 size_t io_len; 4153 u_offset_t io_off; 4154 uint_t blksz; 4155 rl_t *rl; 4156 int error = 0; 4157 4158 ZFS_ENTER(zfsvfs); 4159 ZFS_VERIFY_ZP(zp); 4160 4161 /* 4162 * Align this request to the file block size in case we kluster. 4163 * XXX - this can result in pretty aggresive locking, which can 4164 * impact simultanious read/write access. One option might be 4165 * to break up long requests (len == 0) into block-by-block 4166 * operations to get narrower locking. 4167 */ 4168 blksz = zp->z_blksz; 4169 if (ISP2(blksz)) 4170 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 4171 else 4172 io_off = 0; 4173 if (len > 0 && ISP2(blksz)) 4174 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 4175 else 4176 io_len = 0; 4177 4178 if (io_len == 0) { 4179 /* 4180 * Search the entire vp list for pages >= io_off. 4181 */ 4182 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 4183 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 4184 goto out; 4185 } 4186 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 4187 4188 if (off > zp->z_size) { 4189 /* past end of file */ 4190 zfs_range_unlock(rl); 4191 ZFS_EXIT(zfsvfs); 4192 return (0); 4193 } 4194 4195 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); 4196 4197 for (off = io_off; io_off < off + len; io_off += io_len) { 4198 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 4199 pp = page_lookup(vp, io_off, 4200 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 4201 } else { 4202 pp = page_lookup_nowait(vp, io_off, 4203 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 4204 } 4205 4206 if (pp != NULL && pvn_getdirty(pp, flags)) { 4207 int err; 4208 4209 /* 4210 * Found a dirty page to push 4211 */ 4212 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 4213 if (err) 4214 error = err; 4215 } else { 4216 io_len = PAGESIZE; 4217 } 4218 } 4219 out: 4220 zfs_range_unlock(rl); 4221 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4222 zil_commit(zfsvfs->z_log, zp->z_id); 4223 ZFS_EXIT(zfsvfs); 4224 return (error); 4225 } 4226 4227 /*ARGSUSED*/ 4228 void 4229 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4230 { 4231 znode_t *zp = VTOZ(vp); 4232 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4233 int error; 4234 4235 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4236 if (zp->z_sa_hdl == NULL) { 4237 /* 4238 * The fs has been unmounted, or we did a 4239 * suspend/resume and this file no longer exists. 4240 */ 4241 if (vn_has_cached_data(vp)) { 4242 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 4243 B_INVAL, cr); 4244 } 4245 4246 mutex_enter(&zp->z_lock); 4247 mutex_enter(&vp->v_lock); 4248 ASSERT(vp->v_count == 1); 4249 vp->v_count = 0; 4250 mutex_exit(&vp->v_lock); 4251 mutex_exit(&zp->z_lock); 4252 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4253 zfs_znode_free(zp); 4254 return; 4255 } 4256 4257 /* 4258 * Attempt to push any data in the page cache. If this fails 4259 * we will get kicked out later in zfs_zinactive(). 4260 */ 4261 if (vn_has_cached_data(vp)) { 4262 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 4263 cr); 4264 } 4265 4266 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4267 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4268 4269 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4270 zfs_sa_upgrade_txholds(tx, zp); 4271 error = dmu_tx_assign(tx, TXG_WAIT); 4272 if (error) { 4273 dmu_tx_abort(tx); 4274 } else { 4275 mutex_enter(&zp->z_lock); 4276 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4277 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4278 zp->z_atime_dirty = 0; 4279 mutex_exit(&zp->z_lock); 4280 dmu_tx_commit(tx); 4281 } 4282 } 4283 4284 zfs_zinactive(zp); 4285 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4286 } 4287 4288 /* 4289 * Bounds-check the seek operation. 4290 * 4291 * IN: vp - vnode seeking within 4292 * ooff - old file offset 4293 * noffp - pointer to new file offset 4294 * ct - caller context 4295 * 4296 * RETURN: 0 if success 4297 * EINVAL if new offset invalid 4298 */ 4299 /* ARGSUSED */ 4300 static int 4301 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 4302 caller_context_t *ct) 4303 { 4304 if (vp->v_type == VDIR) 4305 return (0); 4306 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4307 } 4308 4309 /* 4310 * Pre-filter the generic locking function to trap attempts to place 4311 * a mandatory lock on a memory mapped file. 4312 */ 4313 static int 4314 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 4315 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 4316 { 4317 znode_t *zp = VTOZ(vp); 4318 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4319 4320 ZFS_ENTER(zfsvfs); 4321 ZFS_VERIFY_ZP(zp); 4322 4323 /* 4324 * We are following the UFS semantics with respect to mapcnt 4325 * here: If we see that the file is mapped already, then we will 4326 * return an error, but we don't worry about races between this 4327 * function and zfs_map(). 4328 */ 4329 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { 4330 ZFS_EXIT(zfsvfs); 4331 return (EAGAIN); 4332 } 4333 ZFS_EXIT(zfsvfs); 4334 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4335 } 4336 4337 /* 4338 * If we can't find a page in the cache, we will create a new page 4339 * and fill it with file data. For efficiency, we may try to fill 4340 * multiple pages at once (klustering) to fill up the supplied page 4341 * list. Note that the pages to be filled are held with an exclusive 4342 * lock to prevent access by other threads while they are being filled. 4343 */ 4344 static int 4345 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 4346 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 4347 { 4348 znode_t *zp = VTOZ(vp); 4349 page_t *pp, *cur_pp; 4350 objset_t *os = zp->z_zfsvfs->z_os; 4351 u_offset_t io_off, total; 4352 size_t io_len; 4353 int err; 4354 4355 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4356 /* 4357 * We only have a single page, don't bother klustering 4358 */ 4359 io_off = off; 4360 io_len = PAGESIZE; 4361 pp = page_create_va(vp, io_off, io_len, 4362 PG_EXCL | PG_WAIT, seg, addr); 4363 } else { 4364 /* 4365 * Try to find enough pages to fill the page list 4366 */ 4367 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4368 &io_len, off, plsz, 0); 4369 } 4370 if (pp == NULL) { 4371 /* 4372 * The page already exists, nothing to do here. 4373 */ 4374 *pl = NULL; 4375 return (0); 4376 } 4377 4378 /* 4379 * Fill the pages in the kluster. 4380 */ 4381 cur_pp = pp; 4382 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4383 caddr_t va; 4384 4385 ASSERT3U(io_off, ==, cur_pp->p_offset); 4386 va = zfs_map_page(cur_pp, S_WRITE); 4387 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4388 DMU_READ_PREFETCH); 4389 zfs_unmap_page(cur_pp, va); 4390 if (err) { 4391 /* On error, toss the entire kluster */ 4392 pvn_read_done(pp, B_ERROR); 4393 /* convert checksum errors into IO errors */ 4394 if (err == ECKSUM) 4395 err = EIO; 4396 return (err); 4397 } 4398 cur_pp = cur_pp->p_next; 4399 } 4400 4401 /* 4402 * Fill in the page list array from the kluster starting 4403 * from the desired offset `off'. 4404 * NOTE: the page list will always be null terminated. 4405 */ 4406 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4407 ASSERT(pl == NULL || (*pl)->p_offset == off); 4408 4409 return (0); 4410 } 4411 4412 /* 4413 * Return pointers to the pages for the file region [off, off + len] 4414 * in the pl array. If plsz is greater than len, this function may 4415 * also return page pointers from after the specified region 4416 * (i.e. the region [off, off + plsz]). These additional pages are 4417 * only returned if they are already in the cache, or were created as 4418 * part of a klustered read. 4419 * 4420 * IN: vp - vnode of file to get data from. 4421 * off - position in file to get data from. 4422 * len - amount of data to retrieve. 4423 * plsz - length of provided page list. 4424 * seg - segment to obtain pages for. 4425 * addr - virtual address of fault. 4426 * rw - mode of created pages. 4427 * cr - credentials of caller. 4428 * ct - caller context. 4429 * 4430 * OUT: protp - protection mode of created pages. 4431 * pl - list of pages created. 4432 * 4433 * RETURN: 0 if success 4434 * error code if failure 4435 * 4436 * Timestamps: 4437 * vp - atime updated 4438 */ 4439 /* ARGSUSED */ 4440 static int 4441 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4442 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4443 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4444 { 4445 znode_t *zp = VTOZ(vp); 4446 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4447 page_t **pl0 = pl; 4448 int err = 0; 4449 4450 /* we do our own caching, faultahead is unnecessary */ 4451 if (pl == NULL) 4452 return (0); 4453 else if (len > plsz) 4454 len = plsz; 4455 else 4456 len = P2ROUNDUP(len, PAGESIZE); 4457 ASSERT(plsz >= len); 4458 4459 ZFS_ENTER(zfsvfs); 4460 ZFS_VERIFY_ZP(zp); 4461 4462 if (protp) 4463 *protp = PROT_ALL; 4464 4465 /* 4466 * Loop through the requested range [off, off + len) looking 4467 * for pages. If we don't find a page, we will need to create 4468 * a new page and fill it with data from the file. 4469 */ 4470 while (len > 0) { 4471 if (*pl = page_lookup(vp, off, SE_SHARED)) 4472 *(pl+1) = NULL; 4473 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 4474 goto out; 4475 while (*pl) { 4476 ASSERT3U((*pl)->p_offset, ==, off); 4477 off += PAGESIZE; 4478 addr += PAGESIZE; 4479 if (len > 0) { 4480 ASSERT3U(len, >=, PAGESIZE); 4481 len -= PAGESIZE; 4482 } 4483 ASSERT3U(plsz, >=, PAGESIZE); 4484 plsz -= PAGESIZE; 4485 pl++; 4486 } 4487 } 4488 4489 /* 4490 * Fill out the page array with any pages already in the cache. 4491 */ 4492 while (plsz > 0 && 4493 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 4494 off += PAGESIZE; 4495 plsz -= PAGESIZE; 4496 } 4497 out: 4498 if (err) { 4499 /* 4500 * Release any pages we have previously locked. 4501 */ 4502 while (pl > pl0) 4503 page_unlock(*--pl); 4504 } else { 4505 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4506 } 4507 4508 *pl = NULL; 4509 4510 ZFS_EXIT(zfsvfs); 4511 return (err); 4512 } 4513 4514 /* 4515 * Request a memory map for a section of a file. This code interacts 4516 * with common code and the VM system as follows: 4517 * 4518 * common code calls mmap(), which ends up in smmap_common() 4519 * 4520 * this calls VOP_MAP(), which takes you into (say) zfs 4521 * 4522 * zfs_map() calls as_map(), passing segvn_create() as the callback 4523 * 4524 * segvn_create() creates the new segment and calls VOP_ADDMAP() 4525 * 4526 * zfs_addmap() updates z_mapcnt 4527 */ 4528 /*ARGSUSED*/ 4529 static int 4530 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4531 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4532 caller_context_t *ct) 4533 { 4534 znode_t *zp = VTOZ(vp); 4535 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4536 segvn_crargs_t vn_a; 4537 int error; 4538 4539 ZFS_ENTER(zfsvfs); 4540 ZFS_VERIFY_ZP(zp); 4541 4542 if ((prot & PROT_WRITE) && (zp->z_pflags & 4543 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 4544 ZFS_EXIT(zfsvfs); 4545 return (EPERM); 4546 } 4547 4548 if ((prot & (PROT_READ | PROT_EXEC)) && 4549 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 4550 ZFS_EXIT(zfsvfs); 4551 return (EACCES); 4552 } 4553 4554 if (vp->v_flag & VNOMAP) { 4555 ZFS_EXIT(zfsvfs); 4556 return (ENOSYS); 4557 } 4558 4559 if (off < 0 || len > MAXOFFSET_T - off) { 4560 ZFS_EXIT(zfsvfs); 4561 return (ENXIO); 4562 } 4563 4564 if (vp->v_type != VREG) { 4565 ZFS_EXIT(zfsvfs); 4566 return (ENODEV); 4567 } 4568 4569 /* 4570 * If file is locked, disallow mapping. 4571 */ 4572 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { 4573 ZFS_EXIT(zfsvfs); 4574 return (EAGAIN); 4575 } 4576 4577 as_rangelock(as); 4578 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 4579 if (error != 0) { 4580 as_rangeunlock(as); 4581 ZFS_EXIT(zfsvfs); 4582 return (error); 4583 } 4584 4585 vn_a.vp = vp; 4586 vn_a.offset = (u_offset_t)off; 4587 vn_a.type = flags & MAP_TYPE; 4588 vn_a.prot = prot; 4589 vn_a.maxprot = maxprot; 4590 vn_a.cred = cr; 4591 vn_a.amp = NULL; 4592 vn_a.flags = flags & ~MAP_TYPE; 4593 vn_a.szc = 0; 4594 vn_a.lgrp_mem_policy_flags = 0; 4595 4596 error = as_map(as, *addrp, len, segvn_create, &vn_a); 4597 4598 as_rangeunlock(as); 4599 ZFS_EXIT(zfsvfs); 4600 return (error); 4601 } 4602 4603 /* ARGSUSED */ 4604 static int 4605 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4606 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4607 caller_context_t *ct) 4608 { 4609 uint64_t pages = btopr(len); 4610 4611 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 4612 return (0); 4613 } 4614 4615 /* 4616 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 4617 * more accurate mtime for the associated file. Since we don't have a way of 4618 * detecting when the data was actually modified, we have to resort to 4619 * heuristics. If an explicit msync() is done, then we mark the mtime when the 4620 * last page is pushed. The problem occurs when the msync() call is omitted, 4621 * which by far the most common case: 4622 * 4623 * open() 4624 * mmap() 4625 * <modify memory> 4626 * munmap() 4627 * close() 4628 * <time lapse> 4629 * putpage() via fsflush 4630 * 4631 * If we wait until fsflush to come along, we can have a modification time that 4632 * is some arbitrary point in the future. In order to prevent this in the 4633 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 4634 * torn down. 4635 */ 4636 /* ARGSUSED */ 4637 static int 4638 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4639 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 4640 caller_context_t *ct) 4641 { 4642 uint64_t pages = btopr(len); 4643 4644 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 4645 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 4646 4647 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 4648 vn_has_cached_data(vp)) 4649 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 4650 4651 return (0); 4652 } 4653 4654 /* 4655 * Free or allocate space in a file. Currently, this function only 4656 * supports the `F_FREESP' command. However, this command is somewhat 4657 * misnamed, as its functionality includes the ability to allocate as 4658 * well as free space. 4659 * 4660 * IN: vp - vnode of file to free data in. 4661 * cmd - action to take (only F_FREESP supported). 4662 * bfp - section of file to free/alloc. 4663 * flag - current file open mode flags. 4664 * offset - current file offset. 4665 * cr - credentials of caller [UNUSED]. 4666 * ct - caller context. 4667 * 4668 * RETURN: 0 if success 4669 * error code if failure 4670 * 4671 * Timestamps: 4672 * vp - ctime|mtime updated 4673 */ 4674 /* ARGSUSED */ 4675 static int 4676 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 4677 offset_t offset, cred_t *cr, caller_context_t *ct) 4678 { 4679 znode_t *zp = VTOZ(vp); 4680 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4681 uint64_t off, len; 4682 int error; 4683 4684 ZFS_ENTER(zfsvfs); 4685 ZFS_VERIFY_ZP(zp); 4686 4687 if (cmd != F_FREESP) { 4688 ZFS_EXIT(zfsvfs); 4689 return (EINVAL); 4690 } 4691 4692 if (error = convoff(vp, bfp, 0, offset)) { 4693 ZFS_EXIT(zfsvfs); 4694 return (error); 4695 } 4696 4697 if (bfp->l_len < 0) { 4698 ZFS_EXIT(zfsvfs); 4699 return (EINVAL); 4700 } 4701 4702 off = bfp->l_start; 4703 len = bfp->l_len; /* 0 means from off to end of file */ 4704 4705 error = zfs_freesp(zp, off, len, flag, TRUE); 4706 4707 ZFS_EXIT(zfsvfs); 4708 return (error); 4709 } 4710 4711 /*ARGSUSED*/ 4712 static int 4713 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 4714 { 4715 znode_t *zp = VTOZ(vp); 4716 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4717 uint32_t gen; 4718 uint64_t gen64; 4719 uint64_t object = zp->z_id; 4720 zfid_short_t *zfid; 4721 int size, i, error; 4722 4723 ZFS_ENTER(zfsvfs); 4724 ZFS_VERIFY_ZP(zp); 4725 4726 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 4727 &gen64, sizeof (uint64_t))) != 0) { 4728 ZFS_EXIT(zfsvfs); 4729 return (error); 4730 } 4731 4732 gen = (uint32_t)gen64; 4733 4734 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 4735 if (fidp->fid_len < size) { 4736 fidp->fid_len = size; 4737 ZFS_EXIT(zfsvfs); 4738 return (ENOSPC); 4739 } 4740 4741 zfid = (zfid_short_t *)fidp; 4742 4743 zfid->zf_len = size; 4744 4745 for (i = 0; i < sizeof (zfid->zf_object); i++) 4746 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 4747 4748 /* Must have a non-zero generation number to distinguish from .zfs */ 4749 if (gen == 0) 4750 gen = 1; 4751 for (i = 0; i < sizeof (zfid->zf_gen); i++) 4752 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 4753 4754 if (size == LONG_FID_LEN) { 4755 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 4756 zfid_long_t *zlfid; 4757 4758 zlfid = (zfid_long_t *)fidp; 4759 4760 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 4761 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 4762 4763 /* XXX - this should be the generation number for the objset */ 4764 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 4765 zlfid->zf_setgen[i] = 0; 4766 } 4767 4768 ZFS_EXIT(zfsvfs); 4769 return (0); 4770 } 4771 4772 static int 4773 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 4774 caller_context_t *ct) 4775 { 4776 znode_t *zp, *xzp; 4777 zfsvfs_t *zfsvfs; 4778 zfs_dirlock_t *dl; 4779 int error; 4780 4781 switch (cmd) { 4782 case _PC_LINK_MAX: 4783 *valp = ULONG_MAX; 4784 return (0); 4785 4786 case _PC_FILESIZEBITS: 4787 *valp = 64; 4788 return (0); 4789 4790 case _PC_XATTR_EXISTS: 4791 zp = VTOZ(vp); 4792 zfsvfs = zp->z_zfsvfs; 4793 ZFS_ENTER(zfsvfs); 4794 ZFS_VERIFY_ZP(zp); 4795 *valp = 0; 4796 error = zfs_dirent_lock(&dl, zp, "", &xzp, 4797 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 4798 if (error == 0) { 4799 zfs_dirent_unlock(dl); 4800 if (!zfs_dirempty(xzp)) 4801 *valp = 1; 4802 VN_RELE(ZTOV(xzp)); 4803 } else if (error == ENOENT) { 4804 /* 4805 * If there aren't extended attributes, it's the 4806 * same as having zero of them. 4807 */ 4808 error = 0; 4809 } 4810 ZFS_EXIT(zfsvfs); 4811 return (error); 4812 4813 case _PC_SATTR_ENABLED: 4814 case _PC_SATTR_EXISTS: 4815 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 4816 (vp->v_type == VREG || vp->v_type == VDIR); 4817 return (0); 4818 4819 case _PC_ACCESS_FILTERING: 4820 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 4821 vp->v_type == VDIR; 4822 return (0); 4823 4824 case _PC_ACL_ENABLED: 4825 *valp = _ACL_ACE_ENABLED; 4826 return (0); 4827 4828 case _PC_MIN_HOLE_SIZE: 4829 *valp = (ulong_t)SPA_MINBLOCKSIZE; 4830 return (0); 4831 4832 case _PC_TIMESTAMP_RESOLUTION: 4833 /* nanosecond timestamp resolution */ 4834 *valp = 1L; 4835 return (0); 4836 4837 default: 4838 return (fs_pathconf(vp, cmd, valp, cr, ct)); 4839 } 4840 } 4841 4842 /*ARGSUSED*/ 4843 static int 4844 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4845 caller_context_t *ct) 4846 { 4847 znode_t *zp = VTOZ(vp); 4848 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4849 int error; 4850 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4851 4852 ZFS_ENTER(zfsvfs); 4853 ZFS_VERIFY_ZP(zp); 4854 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 4855 ZFS_EXIT(zfsvfs); 4856 4857 return (error); 4858 } 4859 4860 /*ARGSUSED*/ 4861 static int 4862 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4863 caller_context_t *ct) 4864 { 4865 znode_t *zp = VTOZ(vp); 4866 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4867 int error; 4868 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4869 zilog_t *zilog = zfsvfs->z_log; 4870 4871 ZFS_ENTER(zfsvfs); 4872 ZFS_VERIFY_ZP(zp); 4873 4874 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 4875 4876 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4877 zil_commit(zilog, 0); 4878 4879 ZFS_EXIT(zfsvfs); 4880 return (error); 4881 } 4882 4883 /* 4884 * Tunable, both must be a power of 2. 4885 * 4886 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf 4887 * zcr_blksz_max: if set to less than the file block size, allow loaning out of 4888 * an arcbuf for a partial block read 4889 */ 4890 int zcr_blksz_min = (1 << 10); /* 1K */ 4891 int zcr_blksz_max = (1 << 17); /* 128K */ 4892 4893 /*ARGSUSED*/ 4894 static int 4895 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr, 4896 caller_context_t *ct) 4897 { 4898 znode_t *zp = VTOZ(vp); 4899 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4900 int max_blksz = zfsvfs->z_max_blksz; 4901 uio_t *uio = &xuio->xu_uio; 4902 ssize_t size = uio->uio_resid; 4903 offset_t offset = uio->uio_loffset; 4904 int blksz; 4905 int fullblk, i; 4906 arc_buf_t *abuf; 4907 ssize_t maxsize; 4908 int preamble, postamble; 4909 4910 if (xuio->xu_type != UIOTYPE_ZEROCOPY) 4911 return (EINVAL); 4912 4913 ZFS_ENTER(zfsvfs); 4914 ZFS_VERIFY_ZP(zp); 4915 switch (ioflag) { 4916 case UIO_WRITE: 4917 /* 4918 * Loan out an arc_buf for write if write size is bigger than 4919 * max_blksz, and the file's block size is also max_blksz. 4920 */ 4921 blksz = max_blksz; 4922 if (size < blksz || zp->z_blksz != blksz) { 4923 ZFS_EXIT(zfsvfs); 4924 return (EINVAL); 4925 } 4926 /* 4927 * Caller requests buffers for write before knowing where the 4928 * write offset might be (e.g. NFS TCP write). 4929 */ 4930 if (offset == -1) { 4931 preamble = 0; 4932 } else { 4933 preamble = P2PHASE(offset, blksz); 4934 if (preamble) { 4935 preamble = blksz - preamble; 4936 size -= preamble; 4937 } 4938 } 4939 4940 postamble = P2PHASE(size, blksz); 4941 size -= postamble; 4942 4943 fullblk = size / blksz; 4944 (void) dmu_xuio_init(xuio, 4945 (preamble != 0) + fullblk + (postamble != 0)); 4946 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble, 4947 int, postamble, int, 4948 (preamble != 0) + fullblk + (postamble != 0)); 4949 4950 /* 4951 * Have to fix iov base/len for partial buffers. They 4952 * currently represent full arc_buf's. 4953 */ 4954 if (preamble) { 4955 /* data begins in the middle of the arc_buf */ 4956 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 4957 blksz); 4958 ASSERT(abuf); 4959 (void) dmu_xuio_add(xuio, abuf, 4960 blksz - preamble, preamble); 4961 } 4962 4963 for (i = 0; i < fullblk; i++) { 4964 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 4965 blksz); 4966 ASSERT(abuf); 4967 (void) dmu_xuio_add(xuio, abuf, 0, blksz); 4968 } 4969 4970 if (postamble) { 4971 /* data ends in the middle of the arc_buf */ 4972 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 4973 blksz); 4974 ASSERT(abuf); 4975 (void) dmu_xuio_add(xuio, abuf, 0, postamble); 4976 } 4977 break; 4978 case UIO_READ: 4979 /* 4980 * Loan out an arc_buf for read if the read size is larger than 4981 * the current file block size. Block alignment is not 4982 * considered. Partial arc_buf will be loaned out for read. 4983 */ 4984 blksz = zp->z_blksz; 4985 if (blksz < zcr_blksz_min) 4986 blksz = zcr_blksz_min; 4987 if (blksz > zcr_blksz_max) 4988 blksz = zcr_blksz_max; 4989 /* avoid potential complexity of dealing with it */ 4990 if (blksz > max_blksz) { 4991 ZFS_EXIT(zfsvfs); 4992 return (EINVAL); 4993 } 4994 4995 maxsize = zp->z_size - uio->uio_loffset; 4996 if (size > maxsize) 4997 size = maxsize; 4998 4999 if (size < blksz || vn_has_cached_data(vp)) { 5000 ZFS_EXIT(zfsvfs); 5001 return (EINVAL); 5002 } 5003 break; 5004 default: 5005 ZFS_EXIT(zfsvfs); 5006 return (EINVAL); 5007 } 5008 5009 uio->uio_extflg = UIO_XUIO; 5010 XUIO_XUZC_RW(xuio) = ioflag; 5011 ZFS_EXIT(zfsvfs); 5012 return (0); 5013 } 5014 5015 /*ARGSUSED*/ 5016 static int 5017 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct) 5018 { 5019 int i; 5020 arc_buf_t *abuf; 5021 int ioflag = XUIO_XUZC_RW(xuio); 5022 5023 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY); 5024 5025 i = dmu_xuio_cnt(xuio); 5026 while (i-- > 0) { 5027 abuf = dmu_xuio_arcbuf(xuio, i); 5028 /* 5029 * if abuf == NULL, it must be a write buffer 5030 * that has been returned in zfs_write(). 5031 */ 5032 if (abuf) 5033 dmu_return_arcbuf(abuf); 5034 ASSERT(abuf || ioflag == UIO_WRITE); 5035 } 5036 5037 dmu_xuio_fini(xuio); 5038 return (0); 5039 } 5040 5041 /* 5042 * Predeclare these here so that the compiler assumes that 5043 * this is an "old style" function declaration that does 5044 * not include arguments => we won't get type mismatch errors 5045 * in the initializations that follow. 5046 */ 5047 static int zfs_inval(); 5048 static int zfs_isdir(); 5049 5050 static int 5051 zfs_inval() 5052 { 5053 return (EINVAL); 5054 } 5055 5056 static int 5057 zfs_isdir() 5058 { 5059 return (EISDIR); 5060 } 5061 /* 5062 * Directory vnode operations template 5063 */ 5064 vnodeops_t *zfs_dvnodeops; 5065 const fs_operation_def_t zfs_dvnodeops_template[] = { 5066 VOPNAME_OPEN, { .vop_open = zfs_open }, 5067 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5068 VOPNAME_READ, { .error = zfs_isdir }, 5069 VOPNAME_WRITE, { .error = zfs_isdir }, 5070 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5071 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5072 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5073 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5074 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5075 VOPNAME_CREATE, { .vop_create = zfs_create }, 5076 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5077 VOPNAME_LINK, { .vop_link = zfs_link }, 5078 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5079 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 5080 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5081 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5082 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 5083 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5084 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5085 VOPNAME_FID, { .vop_fid = zfs_fid }, 5086 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5087 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5088 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5089 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5090 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5091 NULL, NULL 5092 }; 5093 5094 /* 5095 * Regular file vnode operations template 5096 */ 5097 vnodeops_t *zfs_fvnodeops; 5098 const fs_operation_def_t zfs_fvnodeops_template[] = { 5099 VOPNAME_OPEN, { .vop_open = zfs_open }, 5100 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5101 VOPNAME_READ, { .vop_read = zfs_read }, 5102 VOPNAME_WRITE, { .vop_write = zfs_write }, 5103 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5104 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5105 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5106 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5107 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5108 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5109 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5110 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5111 VOPNAME_FID, { .vop_fid = zfs_fid }, 5112 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5113 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 5114 VOPNAME_SPACE, { .vop_space = zfs_space }, 5115 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 5116 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 5117 VOPNAME_MAP, { .vop_map = zfs_map }, 5118 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 5119 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 5120 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5121 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5122 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5123 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5124 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf }, 5125 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf }, 5126 NULL, NULL 5127 }; 5128 5129 /* 5130 * Symbolic link vnode operations template 5131 */ 5132 vnodeops_t *zfs_symvnodeops; 5133 const fs_operation_def_t zfs_symvnodeops_template[] = { 5134 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5135 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5136 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5137 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5138 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 5139 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5140 VOPNAME_FID, { .vop_fid = zfs_fid }, 5141 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5142 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5143 NULL, NULL 5144 }; 5145 5146 /* 5147 * special share hidden files vnode operations template 5148 */ 5149 vnodeops_t *zfs_sharevnodeops; 5150 const fs_operation_def_t zfs_sharevnodeops_template[] = { 5151 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5152 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5153 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5154 VOPNAME_FID, { .vop_fid = zfs_fid }, 5155 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5156 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5157 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5158 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5159 NULL, NULL 5160 }; 5161 5162 /* 5163 * Extended attribute directory vnode operations template 5164 * This template is identical to the directory vnodes 5165 * operation template except for restricted operations: 5166 * VOP_MKDIR() 5167 * VOP_SYMLINK() 5168 * Note that there are other restrictions embedded in: 5169 * zfs_create() - restrict type to VREG 5170 * zfs_link() - no links into/out of attribute space 5171 * zfs_rename() - no moves into/out of attribute space 5172 */ 5173 vnodeops_t *zfs_xdvnodeops; 5174 const fs_operation_def_t zfs_xdvnodeops_template[] = { 5175 VOPNAME_OPEN, { .vop_open = zfs_open }, 5176 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5177 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5178 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5179 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5180 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5181 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5182 VOPNAME_CREATE, { .vop_create = zfs_create }, 5183 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5184 VOPNAME_LINK, { .vop_link = zfs_link }, 5185 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5186 VOPNAME_MKDIR, { .error = zfs_inval }, 5187 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5188 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5189 VOPNAME_SYMLINK, { .error = zfs_inval }, 5190 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5191 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5192 VOPNAME_FID, { .vop_fid = zfs_fid }, 5193 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5194 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5195 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5196 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5197 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5198 NULL, NULL 5199 }; 5200 5201 /* 5202 * Error vnode operations template 5203 */ 5204 vnodeops_t *zfs_evnodeops; 5205 const fs_operation_def_t zfs_evnodeops_template[] = { 5206 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5207 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5208 NULL, NULL 5209 }; 5210