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 * Note: we don't call zfs_fuid_map_id() here because 870 * user 0 is not an ephemeral uid. 871 */ 872 mutex_enter(&zp->z_acl_lock); 873 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | 874 (S_IXUSR >> 6))) != 0 && 875 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && 876 secpolicy_vnode_setid_retain(cr, 877 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) { 878 uint64_t newmode; 879 zp->z_mode &= ~(S_ISUID | S_ISGID); 880 newmode = zp->z_mode; 881 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), 882 (void *)&newmode, sizeof (uint64_t), tx); 883 } 884 mutex_exit(&zp->z_acl_lock); 885 886 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 887 B_TRUE); 888 889 /* 890 * Update the file size (zp_size) if it has changed; 891 * account for possible concurrent updates. 892 */ 893 while ((end_size = zp->z_size) < uio->uio_loffset) { 894 (void) atomic_cas_64(&zp->z_size, end_size, 895 uio->uio_loffset); 896 ASSERT(error == 0); 897 } 898 /* 899 * If we are replaying and eof is non zero then force 900 * the file size to the specified eof. Note, there's no 901 * concurrency during replay. 902 */ 903 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) 904 zp->z_size = zfsvfs->z_replay_eof; 905 906 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 907 908 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 909 dmu_tx_commit(tx); 910 911 if (error != 0) 912 break; 913 ASSERT(tx_bytes == nbytes); 914 n -= nbytes; 915 916 if (!xuio && n > 0) 917 uio_prefaultpages(MIN(n, max_blksz), uio); 918 } 919 920 zfs_range_unlock(rl); 921 922 /* 923 * If we're in replay mode, or we made no progress, return error. 924 * Otherwise, it's at least a partial write, so it's successful. 925 */ 926 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 927 ZFS_EXIT(zfsvfs); 928 return (error); 929 } 930 931 if (ioflag & (FSYNC | FDSYNC) || 932 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 933 zil_commit(zilog, zp->z_id); 934 935 ZFS_EXIT(zfsvfs); 936 return (0); 937 } 938 939 void 940 zfs_get_done(zgd_t *zgd, int error) 941 { 942 znode_t *zp = zgd->zgd_private; 943 objset_t *os = zp->z_zfsvfs->z_os; 944 945 if (zgd->zgd_db) 946 dmu_buf_rele(zgd->zgd_db, zgd); 947 948 zfs_range_unlock(zgd->zgd_rl); 949 950 /* 951 * Release the vnode asynchronously as we currently have the 952 * txg stopped from syncing. 953 */ 954 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 955 956 if (error == 0 && zgd->zgd_bp) 957 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); 958 959 kmem_free(zgd, sizeof (zgd_t)); 960 } 961 962 #ifdef DEBUG 963 static int zil_fault_io = 0; 964 #endif 965 966 /* 967 * Get data to generate a TX_WRITE intent log record. 968 */ 969 int 970 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 971 { 972 zfsvfs_t *zfsvfs = arg; 973 objset_t *os = zfsvfs->z_os; 974 znode_t *zp; 975 uint64_t object = lr->lr_foid; 976 uint64_t offset = lr->lr_offset; 977 uint64_t size = lr->lr_length; 978 blkptr_t *bp = &lr->lr_blkptr; 979 dmu_buf_t *db; 980 zgd_t *zgd; 981 int error = 0; 982 983 ASSERT(zio != NULL); 984 ASSERT(size != 0); 985 986 /* 987 * Nothing to do if the file has been removed 988 */ 989 if (zfs_zget(zfsvfs, object, &zp) != 0) 990 return (ENOENT); 991 if (zp->z_unlinked) { 992 /* 993 * Release the vnode asynchronously as we currently have the 994 * txg stopped from syncing. 995 */ 996 VN_RELE_ASYNC(ZTOV(zp), 997 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 998 return (ENOENT); 999 } 1000 1001 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1002 zgd->zgd_zilog = zfsvfs->z_log; 1003 zgd->zgd_private = zp; 1004 1005 /* 1006 * Write records come in two flavors: immediate and indirect. 1007 * For small writes it's cheaper to store the data with the 1008 * log record (immediate); for large writes it's cheaper to 1009 * sync the data and get a pointer to it (indirect) so that 1010 * we don't have to write the data twice. 1011 */ 1012 if (buf != NULL) { /* immediate write */ 1013 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER); 1014 /* test for truncation needs to be done while range locked */ 1015 if (offset >= zp->z_size) { 1016 error = ENOENT; 1017 } else { 1018 error = dmu_read(os, object, offset, size, buf, 1019 DMU_READ_NO_PREFETCH); 1020 } 1021 ASSERT(error == 0 || error == ENOENT); 1022 } else { /* indirect write */ 1023 /* 1024 * Have to lock the whole block to ensure when it's 1025 * written out and it's checksum is being calculated 1026 * that no one can change the data. We need to re-check 1027 * blocksize after we get the lock in case it's changed! 1028 */ 1029 for (;;) { 1030 uint64_t blkoff; 1031 size = zp->z_blksz; 1032 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; 1033 offset -= blkoff; 1034 zgd->zgd_rl = zfs_range_lock(zp, offset, size, 1035 RL_READER); 1036 if (zp->z_blksz == size) 1037 break; 1038 offset += blkoff; 1039 zfs_range_unlock(zgd->zgd_rl); 1040 } 1041 /* test for truncation needs to be done while range locked */ 1042 if (lr->lr_offset >= zp->z_size) 1043 error = ENOENT; 1044 #ifdef DEBUG 1045 if (zil_fault_io) { 1046 error = EIO; 1047 zil_fault_io = 0; 1048 } 1049 #endif 1050 if (error == 0) 1051 error = dmu_buf_hold(os, object, offset, zgd, &db, 1052 DMU_READ_NO_PREFETCH); 1053 1054 if (error == 0) { 1055 zgd->zgd_db = db; 1056 zgd->zgd_bp = bp; 1057 1058 ASSERT(db->db_offset == offset); 1059 ASSERT(db->db_size == size); 1060 1061 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1062 zfs_get_done, zgd); 1063 ASSERT(error || lr->lr_length <= zp->z_blksz); 1064 1065 /* 1066 * On success, we need to wait for the write I/O 1067 * initiated by dmu_sync() to complete before we can 1068 * release this dbuf. We will finish everything up 1069 * in the zfs_get_done() callback. 1070 */ 1071 if (error == 0) 1072 return (0); 1073 1074 if (error == EALREADY) { 1075 lr->lr_common.lrc_txtype = TX_WRITE2; 1076 error = 0; 1077 } 1078 } 1079 } 1080 1081 zfs_get_done(zgd, error); 1082 1083 return (error); 1084 } 1085 1086 /*ARGSUSED*/ 1087 static int 1088 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 1089 caller_context_t *ct) 1090 { 1091 znode_t *zp = VTOZ(vp); 1092 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1093 int error; 1094 1095 ZFS_ENTER(zfsvfs); 1096 ZFS_VERIFY_ZP(zp); 1097 1098 if (flag & V_ACE_MASK) 1099 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 1100 else 1101 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1102 1103 ZFS_EXIT(zfsvfs); 1104 return (error); 1105 } 1106 1107 /* 1108 * If vnode is for a device return a specfs vnode instead. 1109 */ 1110 static int 1111 specvp_check(vnode_t **vpp, cred_t *cr) 1112 { 1113 int error = 0; 1114 1115 if (IS_DEVVP(*vpp)) { 1116 struct vnode *svp; 1117 1118 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1119 VN_RELE(*vpp); 1120 if (svp == NULL) 1121 error = ENOSYS; 1122 *vpp = svp; 1123 } 1124 return (error); 1125 } 1126 1127 1128 /* 1129 * Lookup an entry in a directory, or an extended attribute directory. 1130 * If it exists, return a held vnode reference for it. 1131 * 1132 * IN: dvp - vnode of directory to search. 1133 * nm - name of entry to lookup. 1134 * pnp - full pathname to lookup [UNUSED]. 1135 * flags - LOOKUP_XATTR set if looking for an attribute. 1136 * rdir - root directory vnode [UNUSED]. 1137 * cr - credentials of caller. 1138 * ct - caller context 1139 * direntflags - directory lookup flags 1140 * realpnp - returned pathname. 1141 * 1142 * OUT: vpp - vnode of located entry, NULL if not found. 1143 * 1144 * RETURN: 0 if success 1145 * error code if failure 1146 * 1147 * Timestamps: 1148 * NA 1149 */ 1150 /* ARGSUSED */ 1151 static int 1152 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 1153 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 1154 int *direntflags, pathname_t *realpnp) 1155 { 1156 znode_t *zdp = VTOZ(dvp); 1157 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1158 int error = 0; 1159 1160 /* fast path */ 1161 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 1162 1163 if (dvp->v_type != VDIR) { 1164 return (ENOTDIR); 1165 } else if (zdp->z_sa_hdl == NULL) { 1166 return (EIO); 1167 } 1168 1169 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 1170 error = zfs_fastaccesschk_execute(zdp, cr); 1171 if (!error) { 1172 *vpp = dvp; 1173 VN_HOLD(*vpp); 1174 return (0); 1175 } 1176 return (error); 1177 } else { 1178 vnode_t *tvp = dnlc_lookup(dvp, nm); 1179 1180 if (tvp) { 1181 error = zfs_fastaccesschk_execute(zdp, cr); 1182 if (error) { 1183 VN_RELE(tvp); 1184 return (error); 1185 } 1186 if (tvp == DNLC_NO_VNODE) { 1187 VN_RELE(tvp); 1188 return (ENOENT); 1189 } else { 1190 *vpp = tvp; 1191 return (specvp_check(vpp, cr)); 1192 } 1193 } 1194 } 1195 } 1196 1197 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1198 1199 ZFS_ENTER(zfsvfs); 1200 ZFS_VERIFY_ZP(zdp); 1201 1202 *vpp = NULL; 1203 1204 if (flags & LOOKUP_XATTR) { 1205 /* 1206 * If the xattr property is off, refuse the lookup request. 1207 */ 1208 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1209 ZFS_EXIT(zfsvfs); 1210 return (EINVAL); 1211 } 1212 1213 /* 1214 * We don't allow recursive attributes.. 1215 * Maybe someday we will. 1216 */ 1217 if (zdp->z_pflags & ZFS_XATTR) { 1218 ZFS_EXIT(zfsvfs); 1219 return (EINVAL); 1220 } 1221 1222 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1223 ZFS_EXIT(zfsvfs); 1224 return (error); 1225 } 1226 1227 /* 1228 * Do we have permission to get into attribute directory? 1229 */ 1230 1231 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1232 B_FALSE, cr)) { 1233 VN_RELE(*vpp); 1234 *vpp = NULL; 1235 } 1236 1237 ZFS_EXIT(zfsvfs); 1238 return (error); 1239 } 1240 1241 if (dvp->v_type != VDIR) { 1242 ZFS_EXIT(zfsvfs); 1243 return (ENOTDIR); 1244 } 1245 1246 /* 1247 * Check accessibility of directory. 1248 */ 1249 1250 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1251 ZFS_EXIT(zfsvfs); 1252 return (error); 1253 } 1254 1255 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1256 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1257 ZFS_EXIT(zfsvfs); 1258 return (EILSEQ); 1259 } 1260 1261 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1262 if (error == 0) 1263 error = specvp_check(vpp, cr); 1264 1265 ZFS_EXIT(zfsvfs); 1266 return (error); 1267 } 1268 1269 /* 1270 * Attempt to create a new entry in a directory. If the entry 1271 * already exists, truncate the file if permissible, else return 1272 * an error. Return the vp of the created or trunc'd file. 1273 * 1274 * IN: dvp - vnode of directory to put new file entry in. 1275 * name - name of new file entry. 1276 * vap - attributes of new file. 1277 * excl - flag indicating exclusive or non-exclusive mode. 1278 * mode - mode to open file with. 1279 * cr - credentials of caller. 1280 * flag - large file flag [UNUSED]. 1281 * ct - caller context 1282 * vsecp - ACL to be set 1283 * 1284 * OUT: vpp - vnode of created or trunc'd entry. 1285 * 1286 * RETURN: 0 if success 1287 * error code if failure 1288 * 1289 * Timestamps: 1290 * dvp - ctime|mtime updated if new entry created 1291 * vp - ctime|mtime always, atime if new 1292 */ 1293 1294 /* ARGSUSED */ 1295 static int 1296 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1297 int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct, 1298 vsecattr_t *vsecp) 1299 { 1300 znode_t *zp, *dzp = VTOZ(dvp); 1301 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1302 zilog_t *zilog; 1303 objset_t *os; 1304 zfs_dirlock_t *dl; 1305 dmu_tx_t *tx; 1306 int error; 1307 ksid_t *ksid; 1308 uid_t uid; 1309 gid_t gid = crgetgid(cr); 1310 zfs_acl_ids_t acl_ids; 1311 boolean_t fuid_dirtied; 1312 boolean_t have_acl = B_FALSE; 1313 1314 /* 1315 * If we have an ephemeral id, ACL, or XVATTR then 1316 * make sure file system is at proper version 1317 */ 1318 1319 ksid = crgetsid(cr, KSID_OWNER); 1320 if (ksid) 1321 uid = ksid_getid(ksid); 1322 else 1323 uid = crgetuid(cr); 1324 1325 if (zfsvfs->z_use_fuids == B_FALSE && 1326 (vsecp || (vap->va_mask & AT_XVATTR) || 1327 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1328 return (EINVAL); 1329 1330 ZFS_ENTER(zfsvfs); 1331 ZFS_VERIFY_ZP(dzp); 1332 os = zfsvfs->z_os; 1333 zilog = zfsvfs->z_log; 1334 1335 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1336 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1337 ZFS_EXIT(zfsvfs); 1338 return (EILSEQ); 1339 } 1340 1341 if (vap->va_mask & AT_XVATTR) { 1342 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1343 crgetuid(cr), cr, vap->va_type)) != 0) { 1344 ZFS_EXIT(zfsvfs); 1345 return (error); 1346 } 1347 } 1348 top: 1349 *vpp = NULL; 1350 1351 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1352 vap->va_mode &= ~VSVTX; 1353 1354 if (*name == '\0') { 1355 /* 1356 * Null component name refers to the directory itself. 1357 */ 1358 VN_HOLD(dvp); 1359 zp = dzp; 1360 dl = NULL; 1361 error = 0; 1362 } else { 1363 /* possible VN_HOLD(zp) */ 1364 int zflg = 0; 1365 1366 if (flag & FIGNORECASE) 1367 zflg |= ZCILOOK; 1368 1369 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1370 NULL, NULL); 1371 if (error) { 1372 if (strcmp(name, "..") == 0) 1373 error = EISDIR; 1374 ZFS_EXIT(zfsvfs); 1375 return (error); 1376 } 1377 } 1378 1379 if (zp == NULL) { 1380 uint64_t txtype; 1381 1382 /* 1383 * Create a new file object and update the directory 1384 * to reference it. 1385 */ 1386 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1387 goto out; 1388 } 1389 1390 /* 1391 * We only support the creation of regular files in 1392 * extended attribute directories. 1393 */ 1394 1395 if ((dzp->z_pflags & ZFS_XATTR) && 1396 (vap->va_type != VREG)) { 1397 error = EINVAL; 1398 goto out; 1399 } 1400 1401 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 1402 cr, vsecp, &acl_ids)) != 0) 1403 goto out; 1404 have_acl = B_TRUE; 1405 1406 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1407 zfs_acl_ids_free(&acl_ids); 1408 error = EDQUOT; 1409 goto out; 1410 } 1411 1412 tx = dmu_tx_create(os); 1413 1414 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1415 ZFS_SA_BASE_ATTR_SIZE); 1416 1417 fuid_dirtied = zfsvfs->z_fuid_dirty; 1418 if (fuid_dirtied) 1419 zfs_fuid_txhold(zfsvfs, tx); 1420 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1421 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1422 if (!zfsvfs->z_use_sa && 1423 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1424 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1425 0, acl_ids.z_aclp->z_acl_bytes); 1426 } 1427 error = dmu_tx_assign(tx, TXG_NOWAIT); 1428 if (error) { 1429 zfs_dirent_unlock(dl); 1430 if (error == ERESTART) { 1431 dmu_tx_wait(tx); 1432 dmu_tx_abort(tx); 1433 goto top; 1434 } 1435 zfs_acl_ids_free(&acl_ids); 1436 dmu_tx_abort(tx); 1437 ZFS_EXIT(zfsvfs); 1438 return (error); 1439 } 1440 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1441 1442 if (fuid_dirtied) 1443 zfs_fuid_sync(zfsvfs, tx); 1444 1445 (void) zfs_link_create(dl, zp, tx, ZNEW); 1446 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1447 if (flag & FIGNORECASE) 1448 txtype |= TX_CI; 1449 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1450 vsecp, acl_ids.z_fuidp, vap); 1451 zfs_acl_ids_free(&acl_ids); 1452 dmu_tx_commit(tx); 1453 } else { 1454 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1455 1456 /* 1457 * A directory entry already exists for this name. 1458 */ 1459 /* 1460 * Can't truncate an existing file if in exclusive mode. 1461 */ 1462 if (excl == EXCL) { 1463 error = EEXIST; 1464 goto out; 1465 } 1466 /* 1467 * Can't open a directory for writing. 1468 */ 1469 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1470 error = EISDIR; 1471 goto out; 1472 } 1473 /* 1474 * Verify requested access to file. 1475 */ 1476 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1477 goto out; 1478 } 1479 1480 mutex_enter(&dzp->z_lock); 1481 dzp->z_seq++; 1482 mutex_exit(&dzp->z_lock); 1483 1484 /* 1485 * Truncate regular files if requested. 1486 */ 1487 if ((ZTOV(zp)->v_type == VREG) && 1488 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1489 /* we can't hold any locks when calling zfs_freesp() */ 1490 zfs_dirent_unlock(dl); 1491 dl = NULL; 1492 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1493 if (error == 0) { 1494 vnevent_create(ZTOV(zp), ct); 1495 } 1496 } 1497 } 1498 out: 1499 1500 if (dl) 1501 zfs_dirent_unlock(dl); 1502 1503 if (error) { 1504 if (zp) 1505 VN_RELE(ZTOV(zp)); 1506 } else { 1507 *vpp = ZTOV(zp); 1508 error = specvp_check(vpp, cr); 1509 } 1510 1511 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1512 zil_commit(zilog, 0); 1513 1514 ZFS_EXIT(zfsvfs); 1515 return (error); 1516 } 1517 1518 /* 1519 * Remove an entry from a directory. 1520 * 1521 * IN: dvp - vnode of directory to remove entry from. 1522 * name - name of entry to remove. 1523 * cr - credentials of caller. 1524 * ct - caller context 1525 * flags - case flags 1526 * 1527 * RETURN: 0 if success 1528 * error code if failure 1529 * 1530 * Timestamps: 1531 * dvp - ctime|mtime 1532 * vp - ctime (if nlink > 0) 1533 */ 1534 1535 uint64_t null_xattr = 0; 1536 1537 /*ARGSUSED*/ 1538 static int 1539 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1540 int flags) 1541 { 1542 znode_t *zp, *dzp = VTOZ(dvp); 1543 znode_t *xzp = NULL; 1544 vnode_t *vp; 1545 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1546 zilog_t *zilog; 1547 uint64_t acl_obj, xattr_obj = 0; 1548 uint64_t xattr_obj_unlinked = 0; 1549 uint64_t obj = 0; 1550 zfs_dirlock_t *dl; 1551 dmu_tx_t *tx; 1552 boolean_t may_delete_now, delete_now = FALSE; 1553 boolean_t unlinked, toobig = FALSE; 1554 uint64_t txtype; 1555 pathname_t *realnmp = NULL; 1556 pathname_t realnm; 1557 int error; 1558 int zflg = ZEXISTS; 1559 1560 ZFS_ENTER(zfsvfs); 1561 ZFS_VERIFY_ZP(dzp); 1562 zilog = zfsvfs->z_log; 1563 1564 if (flags & FIGNORECASE) { 1565 zflg |= ZCILOOK; 1566 pn_alloc(&realnm); 1567 realnmp = &realnm; 1568 } 1569 1570 top: 1571 /* 1572 * Attempt to lock directory; fail if entry doesn't exist. 1573 */ 1574 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1575 NULL, realnmp)) { 1576 if (realnmp) 1577 pn_free(realnmp); 1578 ZFS_EXIT(zfsvfs); 1579 return (error); 1580 } 1581 1582 vp = ZTOV(zp); 1583 1584 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1585 goto out; 1586 } 1587 1588 /* 1589 * Need to use rmdir for removing directories. 1590 */ 1591 if (vp->v_type == VDIR) { 1592 error = EPERM; 1593 goto out; 1594 } 1595 1596 vnevent_remove(vp, dvp, name, ct); 1597 1598 if (realnmp) 1599 dnlc_remove(dvp, realnmp->pn_buf); 1600 else 1601 dnlc_remove(dvp, name); 1602 1603 mutex_enter(&vp->v_lock); 1604 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1605 mutex_exit(&vp->v_lock); 1606 1607 /* 1608 * We may delete the znode now, or we may put it in the unlinked set; 1609 * it depends on whether we're the last link, and on whether there are 1610 * other holds on the vnode. So we dmu_tx_hold() the right things to 1611 * allow for either case. 1612 */ 1613 obj = zp->z_id; 1614 tx = dmu_tx_create(zfsvfs->z_os); 1615 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1616 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1617 zfs_sa_upgrade_txholds(tx, zp); 1618 zfs_sa_upgrade_txholds(tx, dzp); 1619 if (may_delete_now) { 1620 toobig = 1621 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 1622 /* if the file is too big, only hold_free a token amount */ 1623 dmu_tx_hold_free(tx, zp->z_id, 0, 1624 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1625 } 1626 1627 /* are there any extended attributes? */ 1628 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1629 &xattr_obj, sizeof (xattr_obj)); 1630 if (xattr_obj) { 1631 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1632 ASSERT3U(error, ==, 0); 1633 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1634 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1635 } 1636 1637 mutex_enter(&zp->z_lock); 1638 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now) 1639 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1640 mutex_exit(&zp->z_lock); 1641 1642 /* charge as an update -- would be nice not to charge at all */ 1643 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1644 1645 error = dmu_tx_assign(tx, TXG_NOWAIT); 1646 if (error) { 1647 zfs_dirent_unlock(dl); 1648 VN_RELE(vp); 1649 if (error == ERESTART) { 1650 dmu_tx_wait(tx); 1651 dmu_tx_abort(tx); 1652 goto top; 1653 } 1654 if (realnmp) 1655 pn_free(realnmp); 1656 dmu_tx_abort(tx); 1657 ZFS_EXIT(zfsvfs); 1658 return (error); 1659 } 1660 1661 /* 1662 * Remove the directory entry. 1663 */ 1664 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1665 1666 if (error) { 1667 dmu_tx_commit(tx); 1668 goto out; 1669 } 1670 1671 if (unlinked) { 1672 1673 /* 1674 * Hold z_lock so that we can make sure that the ACL obj 1675 * hasn't changed. Could have been deleted due to 1676 * zfs_sa_upgrade(). 1677 */ 1678 mutex_enter(&zp->z_lock); 1679 mutex_enter(&vp->v_lock); 1680 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1681 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked)); 1682 delete_now = may_delete_now && !toobig && 1683 vp->v_count == 1 && !vn_has_cached_data(vp) && 1684 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) == 1685 acl_obj; 1686 mutex_exit(&vp->v_lock); 1687 } 1688 1689 if (delete_now) { 1690 if (xattr_obj_unlinked) { 1691 ASSERT3U(xzp->z_links, ==, 2); 1692 mutex_enter(&xzp->z_lock); 1693 xzp->z_unlinked = 1; 1694 xzp->z_links = 0; 1695 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 1696 &xzp->z_links, sizeof (xzp->z_links), tx); 1697 ASSERT3U(error, ==, 0); 1698 mutex_exit(&xzp->z_lock); 1699 zfs_unlinked_add(xzp, tx); 1700 1701 if (zp->z_is_sa) 1702 error = sa_remove(zp->z_sa_hdl, 1703 SA_ZPL_XATTR(zfsvfs), tx); 1704 else 1705 error = sa_update(zp->z_sa_hdl, 1706 SA_ZPL_XATTR(zfsvfs), &null_xattr, 1707 sizeof (uint64_t), tx); 1708 ASSERT3U(error, ==, 0); 1709 } 1710 mutex_enter(&vp->v_lock); 1711 vp->v_count--; 1712 ASSERT3U(vp->v_count, ==, 0); 1713 mutex_exit(&vp->v_lock); 1714 mutex_exit(&zp->z_lock); 1715 zfs_znode_delete(zp, tx); 1716 } else if (unlinked) { 1717 mutex_exit(&zp->z_lock); 1718 zfs_unlinked_add(zp, tx); 1719 } 1720 1721 txtype = TX_REMOVE; 1722 if (flags & FIGNORECASE) 1723 txtype |= TX_CI; 1724 zfs_log_remove(zilog, tx, txtype, dzp, name, obj); 1725 1726 dmu_tx_commit(tx); 1727 out: 1728 if (realnmp) 1729 pn_free(realnmp); 1730 1731 zfs_dirent_unlock(dl); 1732 1733 if (!delete_now) 1734 VN_RELE(vp); 1735 if (xzp) 1736 VN_RELE(ZTOV(xzp)); 1737 1738 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1739 zil_commit(zilog, 0); 1740 1741 ZFS_EXIT(zfsvfs); 1742 return (error); 1743 } 1744 1745 /* 1746 * Create a new directory and insert it into dvp using the name 1747 * provided. Return a pointer to the inserted directory. 1748 * 1749 * IN: dvp - vnode of directory to add subdir to. 1750 * dirname - name of new directory. 1751 * vap - attributes of new directory. 1752 * cr - credentials of caller. 1753 * ct - caller context 1754 * vsecp - ACL to be set 1755 * 1756 * OUT: vpp - vnode of created directory. 1757 * 1758 * RETURN: 0 if success 1759 * error code if failure 1760 * 1761 * Timestamps: 1762 * dvp - ctime|mtime updated 1763 * vp - ctime|mtime|atime updated 1764 */ 1765 /*ARGSUSED*/ 1766 static int 1767 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 1768 caller_context_t *ct, int flags, vsecattr_t *vsecp) 1769 { 1770 znode_t *zp, *dzp = VTOZ(dvp); 1771 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1772 zilog_t *zilog; 1773 zfs_dirlock_t *dl; 1774 uint64_t txtype; 1775 dmu_tx_t *tx; 1776 int error; 1777 int zf = ZNEW; 1778 ksid_t *ksid; 1779 uid_t uid; 1780 gid_t gid = crgetgid(cr); 1781 zfs_acl_ids_t acl_ids; 1782 boolean_t fuid_dirtied; 1783 1784 ASSERT(vap->va_type == VDIR); 1785 1786 /* 1787 * If we have an ephemeral id, ACL, or XVATTR then 1788 * make sure file system is at proper version 1789 */ 1790 1791 ksid = crgetsid(cr, KSID_OWNER); 1792 if (ksid) 1793 uid = ksid_getid(ksid); 1794 else 1795 uid = crgetuid(cr); 1796 if (zfsvfs->z_use_fuids == B_FALSE && 1797 (vsecp || (vap->va_mask & AT_XVATTR) || 1798 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1799 return (EINVAL); 1800 1801 ZFS_ENTER(zfsvfs); 1802 ZFS_VERIFY_ZP(dzp); 1803 zilog = zfsvfs->z_log; 1804 1805 if (dzp->z_pflags & ZFS_XATTR) { 1806 ZFS_EXIT(zfsvfs); 1807 return (EINVAL); 1808 } 1809 1810 if (zfsvfs->z_utf8 && u8_validate(dirname, 1811 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1812 ZFS_EXIT(zfsvfs); 1813 return (EILSEQ); 1814 } 1815 if (flags & FIGNORECASE) 1816 zf |= ZCILOOK; 1817 1818 if (vap->va_mask & AT_XVATTR) { 1819 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1820 crgetuid(cr), cr, vap->va_type)) != 0) { 1821 ZFS_EXIT(zfsvfs); 1822 return (error); 1823 } 1824 } 1825 1826 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 1827 vsecp, &acl_ids)) != 0) { 1828 ZFS_EXIT(zfsvfs); 1829 return (error); 1830 } 1831 /* 1832 * First make sure the new directory doesn't exist. 1833 * 1834 * Existence is checked first to make sure we don't return 1835 * EACCES instead of EEXIST which can cause some applications 1836 * to fail. 1837 */ 1838 top: 1839 *vpp = NULL; 1840 1841 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 1842 NULL, NULL)) { 1843 zfs_acl_ids_free(&acl_ids); 1844 ZFS_EXIT(zfsvfs); 1845 return (error); 1846 } 1847 1848 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 1849 zfs_acl_ids_free(&acl_ids); 1850 zfs_dirent_unlock(dl); 1851 ZFS_EXIT(zfsvfs); 1852 return (error); 1853 } 1854 1855 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1856 zfs_acl_ids_free(&acl_ids); 1857 zfs_dirent_unlock(dl); 1858 ZFS_EXIT(zfsvfs); 1859 return (EDQUOT); 1860 } 1861 1862 /* 1863 * Add a new entry to the directory. 1864 */ 1865 tx = dmu_tx_create(zfsvfs->z_os); 1866 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1867 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1868 fuid_dirtied = zfsvfs->z_fuid_dirty; 1869 if (fuid_dirtied) 1870 zfs_fuid_txhold(zfsvfs, tx); 1871 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1872 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1873 acl_ids.z_aclp->z_acl_bytes); 1874 } 1875 1876 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1877 ZFS_SA_BASE_ATTR_SIZE); 1878 1879 error = dmu_tx_assign(tx, TXG_NOWAIT); 1880 if (error) { 1881 zfs_dirent_unlock(dl); 1882 if (error == ERESTART) { 1883 dmu_tx_wait(tx); 1884 dmu_tx_abort(tx); 1885 goto top; 1886 } 1887 zfs_acl_ids_free(&acl_ids); 1888 dmu_tx_abort(tx); 1889 ZFS_EXIT(zfsvfs); 1890 return (error); 1891 } 1892 1893 /* 1894 * Create new node. 1895 */ 1896 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1897 1898 if (fuid_dirtied) 1899 zfs_fuid_sync(zfsvfs, tx); 1900 1901 /* 1902 * Now put new name in parent dir. 1903 */ 1904 (void) zfs_link_create(dl, zp, tx, ZNEW); 1905 1906 *vpp = ZTOV(zp); 1907 1908 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 1909 if (flags & FIGNORECASE) 1910 txtype |= TX_CI; 1911 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 1912 acl_ids.z_fuidp, vap); 1913 1914 zfs_acl_ids_free(&acl_ids); 1915 1916 dmu_tx_commit(tx); 1917 1918 zfs_dirent_unlock(dl); 1919 1920 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1921 zil_commit(zilog, 0); 1922 1923 ZFS_EXIT(zfsvfs); 1924 return (0); 1925 } 1926 1927 /* 1928 * Remove a directory subdir entry. If the current working 1929 * directory is the same as the subdir to be removed, the 1930 * remove will fail. 1931 * 1932 * IN: dvp - vnode of directory to remove from. 1933 * name - name of directory to be removed. 1934 * cwd - vnode of current working directory. 1935 * cr - credentials of caller. 1936 * ct - caller context 1937 * flags - case flags 1938 * 1939 * RETURN: 0 if success 1940 * error code if failure 1941 * 1942 * Timestamps: 1943 * dvp - ctime|mtime updated 1944 */ 1945 /*ARGSUSED*/ 1946 static int 1947 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 1948 caller_context_t *ct, int flags) 1949 { 1950 znode_t *dzp = VTOZ(dvp); 1951 znode_t *zp; 1952 vnode_t *vp; 1953 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1954 zilog_t *zilog; 1955 zfs_dirlock_t *dl; 1956 dmu_tx_t *tx; 1957 int error; 1958 int zflg = ZEXISTS; 1959 1960 ZFS_ENTER(zfsvfs); 1961 ZFS_VERIFY_ZP(dzp); 1962 zilog = zfsvfs->z_log; 1963 1964 if (flags & FIGNORECASE) 1965 zflg |= ZCILOOK; 1966 top: 1967 zp = NULL; 1968 1969 /* 1970 * Attempt to lock directory; fail if entry doesn't exist. 1971 */ 1972 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1973 NULL, NULL)) { 1974 ZFS_EXIT(zfsvfs); 1975 return (error); 1976 } 1977 1978 vp = ZTOV(zp); 1979 1980 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1981 goto out; 1982 } 1983 1984 if (vp->v_type != VDIR) { 1985 error = ENOTDIR; 1986 goto out; 1987 } 1988 1989 if (vp == cwd) { 1990 error = EINVAL; 1991 goto out; 1992 } 1993 1994 vnevent_rmdir(vp, dvp, name, ct); 1995 1996 /* 1997 * Grab a lock on the directory to make sure that noone is 1998 * trying to add (or lookup) entries while we are removing it. 1999 */ 2000 rw_enter(&zp->z_name_lock, RW_WRITER); 2001 2002 /* 2003 * Grab a lock on the parent pointer to make sure we play well 2004 * with the treewalk and directory rename code. 2005 */ 2006 rw_enter(&zp->z_parent_lock, RW_WRITER); 2007 2008 tx = dmu_tx_create(zfsvfs->z_os); 2009 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2010 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2011 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2012 zfs_sa_upgrade_txholds(tx, zp); 2013 zfs_sa_upgrade_txholds(tx, dzp); 2014 error = dmu_tx_assign(tx, TXG_NOWAIT); 2015 if (error) { 2016 rw_exit(&zp->z_parent_lock); 2017 rw_exit(&zp->z_name_lock); 2018 zfs_dirent_unlock(dl); 2019 VN_RELE(vp); 2020 if (error == ERESTART) { 2021 dmu_tx_wait(tx); 2022 dmu_tx_abort(tx); 2023 goto top; 2024 } 2025 dmu_tx_abort(tx); 2026 ZFS_EXIT(zfsvfs); 2027 return (error); 2028 } 2029 2030 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 2031 2032 if (error == 0) { 2033 uint64_t txtype = TX_RMDIR; 2034 if (flags & FIGNORECASE) 2035 txtype |= TX_CI; 2036 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT); 2037 } 2038 2039 dmu_tx_commit(tx); 2040 2041 rw_exit(&zp->z_parent_lock); 2042 rw_exit(&zp->z_name_lock); 2043 out: 2044 zfs_dirent_unlock(dl); 2045 2046 VN_RELE(vp); 2047 2048 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2049 zil_commit(zilog, 0); 2050 2051 ZFS_EXIT(zfsvfs); 2052 return (error); 2053 } 2054 2055 /* 2056 * Read as many directory entries as will fit into the provided 2057 * buffer from the given directory cursor position (specified in 2058 * the uio structure. 2059 * 2060 * IN: vp - vnode of directory to read. 2061 * uio - structure supplying read location, range info, 2062 * and return buffer. 2063 * cr - credentials of caller. 2064 * ct - caller context 2065 * flags - case flags 2066 * 2067 * OUT: uio - updated offset and range, buffer filled. 2068 * eofp - set to true if end-of-file detected. 2069 * 2070 * RETURN: 0 if success 2071 * error code if failure 2072 * 2073 * Timestamps: 2074 * vp - atime updated 2075 * 2076 * Note that the low 4 bits of the cookie returned by zap is always zero. 2077 * This allows us to use the low range for "special" directory entries: 2078 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 2079 * we use the offset 2 for the '.zfs' directory. 2080 */ 2081 /* ARGSUSED */ 2082 static int 2083 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, 2084 caller_context_t *ct, int flags) 2085 { 2086 znode_t *zp = VTOZ(vp); 2087 iovec_t *iovp; 2088 edirent_t *eodp; 2089 dirent64_t *odp; 2090 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2091 objset_t *os; 2092 caddr_t outbuf; 2093 size_t bufsize; 2094 zap_cursor_t zc; 2095 zap_attribute_t zap; 2096 uint_t bytes_wanted; 2097 uint64_t offset; /* must be unsigned; checks for < 1 */ 2098 uint64_t parent; 2099 int local_eof; 2100 int outcount; 2101 int error; 2102 uint8_t prefetch; 2103 boolean_t check_sysattrs; 2104 2105 ZFS_ENTER(zfsvfs); 2106 ZFS_VERIFY_ZP(zp); 2107 2108 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 2109 &parent, sizeof (parent))) != 0) { 2110 ZFS_EXIT(zfsvfs); 2111 return (error); 2112 } 2113 2114 /* 2115 * If we are not given an eof variable, 2116 * use a local one. 2117 */ 2118 if (eofp == NULL) 2119 eofp = &local_eof; 2120 2121 /* 2122 * Check for valid iov_len. 2123 */ 2124 if (uio->uio_iov->iov_len <= 0) { 2125 ZFS_EXIT(zfsvfs); 2126 return (EINVAL); 2127 } 2128 2129 /* 2130 * Quit if directory has been removed (posix) 2131 */ 2132 if ((*eofp = zp->z_unlinked) != 0) { 2133 ZFS_EXIT(zfsvfs); 2134 return (0); 2135 } 2136 2137 error = 0; 2138 os = zfsvfs->z_os; 2139 offset = uio->uio_loffset; 2140 prefetch = zp->z_zn_prefetch; 2141 2142 /* 2143 * Initialize the iterator cursor. 2144 */ 2145 if (offset <= 3) { 2146 /* 2147 * Start iteration from the beginning of the directory. 2148 */ 2149 zap_cursor_init(&zc, os, zp->z_id); 2150 } else { 2151 /* 2152 * The offset is a serialized cursor. 2153 */ 2154 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 2155 } 2156 2157 /* 2158 * Get space to change directory entries into fs independent format. 2159 */ 2160 iovp = uio->uio_iov; 2161 bytes_wanted = iovp->iov_len; 2162 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 2163 bufsize = bytes_wanted; 2164 outbuf = kmem_alloc(bufsize, KM_SLEEP); 2165 odp = (struct dirent64 *)outbuf; 2166 } else { 2167 bufsize = bytes_wanted; 2168 odp = (struct dirent64 *)iovp->iov_base; 2169 } 2170 eodp = (struct edirent *)odp; 2171 2172 /* 2173 * If this VFS supports the system attribute view interface; and 2174 * we're looking at an extended attribute directory; and we care 2175 * about normalization conflicts on this vfs; then we must check 2176 * for normalization conflicts with the sysattr name space. 2177 */ 2178 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 2179 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 2180 (flags & V_RDDIR_ENTFLAGS); 2181 2182 /* 2183 * Transform to file-system independent format 2184 */ 2185 outcount = 0; 2186 while (outcount < bytes_wanted) { 2187 ino64_t objnum; 2188 ushort_t reclen; 2189 off64_t *next = NULL; 2190 2191 /* 2192 * Special case `.', `..', and `.zfs'. 2193 */ 2194 if (offset == 0) { 2195 (void) strcpy(zap.za_name, "."); 2196 zap.za_normalization_conflict = 0; 2197 objnum = zp->z_id; 2198 } else if (offset == 1) { 2199 (void) strcpy(zap.za_name, ".."); 2200 zap.za_normalization_conflict = 0; 2201 objnum = parent; 2202 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2203 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2204 zap.za_normalization_conflict = 0; 2205 objnum = ZFSCTL_INO_ROOT; 2206 } else { 2207 /* 2208 * Grab next entry. 2209 */ 2210 if (error = zap_cursor_retrieve(&zc, &zap)) { 2211 if ((*eofp = (error == ENOENT)) != 0) 2212 break; 2213 else 2214 goto update; 2215 } 2216 2217 if (zap.za_integer_length != 8 || 2218 zap.za_num_integers != 1) { 2219 cmn_err(CE_WARN, "zap_readdir: bad directory " 2220 "entry, obj = %lld, offset = %lld\n", 2221 (u_longlong_t)zp->z_id, 2222 (u_longlong_t)offset); 2223 error = ENXIO; 2224 goto update; 2225 } 2226 2227 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2228 /* 2229 * MacOS X can extract the object type here such as: 2230 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2231 */ 2232 2233 if (check_sysattrs && !zap.za_normalization_conflict) { 2234 zap.za_normalization_conflict = 2235 xattr_sysattr_casechk(zap.za_name); 2236 } 2237 } 2238 2239 if (flags & V_RDDIR_ACCFILTER) { 2240 /* 2241 * If we have no access at all, don't include 2242 * this entry in the returned information 2243 */ 2244 znode_t *ezp; 2245 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2246 goto skip_entry; 2247 if (!zfs_has_access(ezp, cr)) { 2248 VN_RELE(ZTOV(ezp)); 2249 goto skip_entry; 2250 } 2251 VN_RELE(ZTOV(ezp)); 2252 } 2253 2254 if (flags & V_RDDIR_ENTFLAGS) 2255 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2256 else 2257 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2258 2259 /* 2260 * Will this entry fit in the buffer? 2261 */ 2262 if (outcount + reclen > bufsize) { 2263 /* 2264 * Did we manage to fit anything in the buffer? 2265 */ 2266 if (!outcount) { 2267 error = EINVAL; 2268 goto update; 2269 } 2270 break; 2271 } 2272 if (flags & V_RDDIR_ENTFLAGS) { 2273 /* 2274 * Add extended flag entry: 2275 */ 2276 eodp->ed_ino = objnum; 2277 eodp->ed_reclen = reclen; 2278 /* NOTE: ed_off is the offset for the *next* entry */ 2279 next = &(eodp->ed_off); 2280 eodp->ed_eflags = zap.za_normalization_conflict ? 2281 ED_CASE_CONFLICT : 0; 2282 (void) strncpy(eodp->ed_name, zap.za_name, 2283 EDIRENT_NAMELEN(reclen)); 2284 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2285 } else { 2286 /* 2287 * Add normal entry: 2288 */ 2289 odp->d_ino = objnum; 2290 odp->d_reclen = reclen; 2291 /* NOTE: d_off is the offset for the *next* entry */ 2292 next = &(odp->d_off); 2293 (void) strncpy(odp->d_name, zap.za_name, 2294 DIRENT64_NAMELEN(reclen)); 2295 odp = (dirent64_t *)((intptr_t)odp + reclen); 2296 } 2297 outcount += reclen; 2298 2299 ASSERT(outcount <= bufsize); 2300 2301 /* Prefetch znode */ 2302 if (prefetch) 2303 dmu_prefetch(os, objnum, 0, 0); 2304 2305 skip_entry: 2306 /* 2307 * Move to the next entry, fill in the previous offset. 2308 */ 2309 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2310 zap_cursor_advance(&zc); 2311 offset = zap_cursor_serialize(&zc); 2312 } else { 2313 offset += 1; 2314 } 2315 if (next) 2316 *next = offset; 2317 } 2318 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2319 2320 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2321 iovp->iov_base += outcount; 2322 iovp->iov_len -= outcount; 2323 uio->uio_resid -= outcount; 2324 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2325 /* 2326 * Reset the pointer. 2327 */ 2328 offset = uio->uio_loffset; 2329 } 2330 2331 update: 2332 zap_cursor_fini(&zc); 2333 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2334 kmem_free(outbuf, bufsize); 2335 2336 if (error == ENOENT) 2337 error = 0; 2338 2339 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2340 2341 uio->uio_loffset = offset; 2342 ZFS_EXIT(zfsvfs); 2343 return (error); 2344 } 2345 2346 ulong_t zfs_fsync_sync_cnt = 4; 2347 2348 static int 2349 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2350 { 2351 znode_t *zp = VTOZ(vp); 2352 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2353 2354 /* 2355 * Regardless of whether this is required for standards conformance, 2356 * this is the logical behavior when fsync() is called on a file with 2357 * dirty pages. We use B_ASYNC since the ZIL transactions are already 2358 * going to be pushed out as part of the zil_commit(). 2359 */ 2360 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 2361 (vp->v_type == VREG) && !(IS_SWAPVP(vp))) 2362 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct); 2363 2364 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2365 2366 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { 2367 ZFS_ENTER(zfsvfs); 2368 ZFS_VERIFY_ZP(zp); 2369 zil_commit(zfsvfs->z_log, zp->z_id); 2370 ZFS_EXIT(zfsvfs); 2371 } 2372 return (0); 2373 } 2374 2375 2376 /* 2377 * Get the requested file attributes and place them in the provided 2378 * vattr structure. 2379 * 2380 * IN: vp - vnode of file. 2381 * vap - va_mask identifies requested attributes. 2382 * If AT_XVATTR set, then optional attrs are requested 2383 * flags - ATTR_NOACLCHECK (CIFS server context) 2384 * cr - credentials of caller. 2385 * ct - caller context 2386 * 2387 * OUT: vap - attribute values. 2388 * 2389 * RETURN: 0 (always succeeds) 2390 */ 2391 /* ARGSUSED */ 2392 static int 2393 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2394 caller_context_t *ct) 2395 { 2396 znode_t *zp = VTOZ(vp); 2397 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2398 int error = 0; 2399 uint64_t links; 2400 uint64_t mtime[2], ctime[2]; 2401 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2402 xoptattr_t *xoap = NULL; 2403 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2404 sa_bulk_attr_t bulk[2]; 2405 int count = 0; 2406 2407 ZFS_ENTER(zfsvfs); 2408 ZFS_VERIFY_ZP(zp); 2409 2410 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2411 2412 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 2413 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 2414 2415 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 2416 ZFS_EXIT(zfsvfs); 2417 return (error); 2418 } 2419 2420 /* 2421 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2422 * Also, if we are the owner don't bother, since owner should 2423 * always be allowed to read basic attributes of file. 2424 */ 2425 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 2426 (vap->va_uid != crgetuid(cr))) { 2427 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2428 skipaclchk, cr)) { 2429 ZFS_EXIT(zfsvfs); 2430 return (error); 2431 } 2432 } 2433 2434 /* 2435 * Return all attributes. It's cheaper to provide the answer 2436 * than to determine whether we were asked the question. 2437 */ 2438 2439 mutex_enter(&zp->z_lock); 2440 vap->va_type = vp->v_type; 2441 vap->va_mode = zp->z_mode & MODEMASK; 2442 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2443 vap->va_nodeid = zp->z_id; 2444 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2445 links = zp->z_links + 1; 2446 else 2447 links = zp->z_links; 2448 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */ 2449 vap->va_size = zp->z_size; 2450 vap->va_rdev = vp->v_rdev; 2451 vap->va_seq = zp->z_seq; 2452 2453 /* 2454 * Add in any requested optional attributes and the create time. 2455 * Also set the corresponding bits in the returned attribute bitmap. 2456 */ 2457 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2458 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2459 xoap->xoa_archive = 2460 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2461 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2462 } 2463 2464 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2465 xoap->xoa_readonly = 2466 ((zp->z_pflags & ZFS_READONLY) != 0); 2467 XVA_SET_RTN(xvap, XAT_READONLY); 2468 } 2469 2470 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2471 xoap->xoa_system = 2472 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2473 XVA_SET_RTN(xvap, XAT_SYSTEM); 2474 } 2475 2476 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2477 xoap->xoa_hidden = 2478 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2479 XVA_SET_RTN(xvap, XAT_HIDDEN); 2480 } 2481 2482 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2483 xoap->xoa_nounlink = 2484 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2485 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2486 } 2487 2488 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2489 xoap->xoa_immutable = 2490 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2491 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2492 } 2493 2494 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2495 xoap->xoa_appendonly = 2496 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2497 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2498 } 2499 2500 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2501 xoap->xoa_nodump = 2502 ((zp->z_pflags & ZFS_NODUMP) != 0); 2503 XVA_SET_RTN(xvap, XAT_NODUMP); 2504 } 2505 2506 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2507 xoap->xoa_opaque = 2508 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2509 XVA_SET_RTN(xvap, XAT_OPAQUE); 2510 } 2511 2512 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2513 xoap->xoa_av_quarantined = 2514 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2515 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2516 } 2517 2518 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2519 xoap->xoa_av_modified = 2520 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2521 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2522 } 2523 2524 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2525 vp->v_type == VREG) { 2526 zfs_sa_get_scanstamp(zp, xvap); 2527 } 2528 2529 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2530 uint64_t times[2]; 2531 2532 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 2533 times, sizeof (times)); 2534 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 2535 XVA_SET_RTN(xvap, XAT_CREATETIME); 2536 } 2537 2538 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2539 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2540 XVA_SET_RTN(xvap, XAT_REPARSE); 2541 } 2542 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2543 xoap->xoa_generation = zp->z_gen; 2544 XVA_SET_RTN(xvap, XAT_GEN); 2545 } 2546 } 2547 2548 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2549 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2550 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2551 2552 mutex_exit(&zp->z_lock); 2553 2554 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks); 2555 2556 if (zp->z_blksz == 0) { 2557 /* 2558 * Block size hasn't been set; suggest maximal I/O transfers. 2559 */ 2560 vap->va_blksize = zfsvfs->z_max_blksz; 2561 } 2562 2563 ZFS_EXIT(zfsvfs); 2564 return (0); 2565 } 2566 2567 /* 2568 * Set the file attributes to the values contained in the 2569 * vattr structure. 2570 * 2571 * IN: vp - vnode of file to be modified. 2572 * vap - new attribute values. 2573 * If AT_XVATTR set, then optional attrs are being set 2574 * flags - ATTR_UTIME set if non-default time values provided. 2575 * - ATTR_NOACLCHECK (CIFS context only). 2576 * cr - credentials of caller. 2577 * ct - caller context 2578 * 2579 * RETURN: 0 if success 2580 * error code if failure 2581 * 2582 * Timestamps: 2583 * vp - ctime updated, mtime updated if size changed. 2584 */ 2585 /* ARGSUSED */ 2586 static int 2587 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2588 caller_context_t *ct) 2589 { 2590 znode_t *zp = VTOZ(vp); 2591 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2592 zilog_t *zilog; 2593 dmu_tx_t *tx; 2594 vattr_t oldva; 2595 xvattr_t tmpxvattr; 2596 uint_t mask = vap->va_mask; 2597 uint_t saved_mask; 2598 int trim_mask = 0; 2599 uint64_t new_mode; 2600 uint64_t new_uid, new_gid; 2601 uint64_t xattr_obj = 0; 2602 uint64_t mtime[2], ctime[2]; 2603 znode_t *attrzp; 2604 int need_policy = FALSE; 2605 int err, err2; 2606 zfs_fuid_info_t *fuidp = NULL; 2607 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2608 xoptattr_t *xoap; 2609 zfs_acl_t *aclp = NULL; 2610 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2611 boolean_t fuid_dirtied = B_FALSE; 2612 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 2613 int count = 0, xattr_count = 0; 2614 2615 if (mask == 0) 2616 return (0); 2617 2618 if (mask & AT_NOSET) 2619 return (EINVAL); 2620 2621 ZFS_ENTER(zfsvfs); 2622 ZFS_VERIFY_ZP(zp); 2623 2624 zilog = zfsvfs->z_log; 2625 2626 /* 2627 * Make sure that if we have ephemeral uid/gid or xvattr specified 2628 * that file system is at proper version level 2629 */ 2630 2631 if (zfsvfs->z_use_fuids == B_FALSE && 2632 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2633 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2634 (mask & AT_XVATTR))) { 2635 ZFS_EXIT(zfsvfs); 2636 return (EINVAL); 2637 } 2638 2639 if (mask & AT_SIZE && vp->v_type == VDIR) { 2640 ZFS_EXIT(zfsvfs); 2641 return (EISDIR); 2642 } 2643 2644 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 2645 ZFS_EXIT(zfsvfs); 2646 return (EINVAL); 2647 } 2648 2649 /* 2650 * If this is an xvattr_t, then get a pointer to the structure of 2651 * optional attributes. If this is NULL, then we have a vattr_t. 2652 */ 2653 xoap = xva_getxoptattr(xvap); 2654 2655 xva_init(&tmpxvattr); 2656 2657 /* 2658 * Immutable files can only alter immutable bit and atime 2659 */ 2660 if ((zp->z_pflags & ZFS_IMMUTABLE) && 2661 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2662 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 2663 ZFS_EXIT(zfsvfs); 2664 return (EPERM); 2665 } 2666 2667 if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 2668 ZFS_EXIT(zfsvfs); 2669 return (EPERM); 2670 } 2671 2672 /* 2673 * Verify timestamps doesn't overflow 32 bits. 2674 * ZFS can handle large timestamps, but 32bit syscalls can't 2675 * handle times greater than 2039. This check should be removed 2676 * once large timestamps are fully supported. 2677 */ 2678 if (mask & (AT_ATIME | AT_MTIME)) { 2679 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 2680 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 2681 ZFS_EXIT(zfsvfs); 2682 return (EOVERFLOW); 2683 } 2684 } 2685 2686 top: 2687 attrzp = NULL; 2688 2689 /* Can this be moved to before the top label? */ 2690 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2691 ZFS_EXIT(zfsvfs); 2692 return (EROFS); 2693 } 2694 2695 /* 2696 * First validate permissions 2697 */ 2698 2699 if (mask & AT_SIZE) { 2700 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr); 2701 if (err) { 2702 ZFS_EXIT(zfsvfs); 2703 return (err); 2704 } 2705 /* 2706 * XXX - Note, we are not providing any open 2707 * mode flags here (like FNDELAY), so we may 2708 * block if there are locks present... this 2709 * should be addressed in openat(). 2710 */ 2711 /* XXX - would it be OK to generate a log record here? */ 2712 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2713 if (err) { 2714 ZFS_EXIT(zfsvfs); 2715 return (err); 2716 } 2717 } 2718 2719 if (mask & (AT_ATIME|AT_MTIME) || 2720 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2721 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2722 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2723 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2724 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 2725 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2726 skipaclchk, cr); 2727 } 2728 2729 if (mask & (AT_UID|AT_GID)) { 2730 int idmask = (mask & (AT_UID|AT_GID)); 2731 int take_owner; 2732 int take_group; 2733 2734 /* 2735 * NOTE: even if a new mode is being set, 2736 * we may clear S_ISUID/S_ISGID bits. 2737 */ 2738 2739 if (!(mask & AT_MODE)) 2740 vap->va_mode = zp->z_mode; 2741 2742 /* 2743 * Take ownership or chgrp to group we are a member of 2744 */ 2745 2746 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2747 take_group = (mask & AT_GID) && 2748 zfs_groupmember(zfsvfs, vap->va_gid, cr); 2749 2750 /* 2751 * If both AT_UID and AT_GID are set then take_owner and 2752 * take_group must both be set in order to allow taking 2753 * ownership. 2754 * 2755 * Otherwise, send the check through secpolicy_vnode_setattr() 2756 * 2757 */ 2758 2759 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2760 ((idmask == AT_UID) && take_owner) || 2761 ((idmask == AT_GID) && take_group)) { 2762 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 2763 skipaclchk, cr) == 0) { 2764 /* 2765 * Remove setuid/setgid for non-privileged users 2766 */ 2767 secpolicy_setid_clear(vap, cr); 2768 trim_mask = (mask & (AT_UID|AT_GID)); 2769 } else { 2770 need_policy = TRUE; 2771 } 2772 } else { 2773 need_policy = TRUE; 2774 } 2775 } 2776 2777 mutex_enter(&zp->z_lock); 2778 oldva.va_mode = zp->z_mode; 2779 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 2780 if (mask & AT_XVATTR) { 2781 /* 2782 * Update xvattr mask to include only those attributes 2783 * that are actually changing. 2784 * 2785 * the bits will be restored prior to actually setting 2786 * the attributes so the caller thinks they were set. 2787 */ 2788 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2789 if (xoap->xoa_appendonly != 2790 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 2791 need_policy = TRUE; 2792 } else { 2793 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 2794 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 2795 } 2796 } 2797 2798 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2799 if (xoap->xoa_nounlink != 2800 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 2801 need_policy = TRUE; 2802 } else { 2803 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 2804 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 2805 } 2806 } 2807 2808 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2809 if (xoap->xoa_immutable != 2810 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 2811 need_policy = TRUE; 2812 } else { 2813 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 2814 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 2815 } 2816 } 2817 2818 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2819 if (xoap->xoa_nodump != 2820 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 2821 need_policy = TRUE; 2822 } else { 2823 XVA_CLR_REQ(xvap, XAT_NODUMP); 2824 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 2825 } 2826 } 2827 2828 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2829 if (xoap->xoa_av_modified != 2830 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 2831 need_policy = TRUE; 2832 } else { 2833 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 2834 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 2835 } 2836 } 2837 2838 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2839 if ((vp->v_type != VREG && 2840 xoap->xoa_av_quarantined) || 2841 xoap->xoa_av_quarantined != 2842 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 2843 need_policy = TRUE; 2844 } else { 2845 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 2846 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 2847 } 2848 } 2849 2850 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2851 mutex_exit(&zp->z_lock); 2852 ZFS_EXIT(zfsvfs); 2853 return (EPERM); 2854 } 2855 2856 if (need_policy == FALSE && 2857 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 2858 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 2859 need_policy = TRUE; 2860 } 2861 } 2862 2863 mutex_exit(&zp->z_lock); 2864 2865 if (mask & AT_MODE) { 2866 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 2867 err = secpolicy_setid_setsticky_clear(vp, vap, 2868 &oldva, cr); 2869 if (err) { 2870 ZFS_EXIT(zfsvfs); 2871 return (err); 2872 } 2873 trim_mask |= AT_MODE; 2874 } else { 2875 need_policy = TRUE; 2876 } 2877 } 2878 2879 if (need_policy) { 2880 /* 2881 * If trim_mask is set then take ownership 2882 * has been granted or write_acl is present and user 2883 * has the ability to modify mode. In that case remove 2884 * UID|GID and or MODE from mask so that 2885 * secpolicy_vnode_setattr() doesn't revoke it. 2886 */ 2887 2888 if (trim_mask) { 2889 saved_mask = vap->va_mask; 2890 vap->va_mask &= ~trim_mask; 2891 } 2892 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2893 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 2894 if (err) { 2895 ZFS_EXIT(zfsvfs); 2896 return (err); 2897 } 2898 2899 if (trim_mask) 2900 vap->va_mask |= saved_mask; 2901 } 2902 2903 /* 2904 * secpolicy_vnode_setattr, or take ownership may have 2905 * changed va_mask 2906 */ 2907 mask = vap->va_mask; 2908 2909 if ((mask & (AT_UID | AT_GID))) { 2910 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xattr_obj, 2911 sizeof (xattr_obj)); 2912 2913 if (xattr_obj) { 2914 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 2915 if (err) 2916 goto out2; 2917 } 2918 if (mask & AT_UID) { 2919 new_uid = zfs_fuid_create(zfsvfs, 2920 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 2921 if (new_uid != zp->z_uid && 2922 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 2923 err = EDQUOT; 2924 goto out2; 2925 } 2926 } 2927 2928 if (mask & AT_GID) { 2929 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 2930 cr, ZFS_GROUP, &fuidp); 2931 if (new_gid != zp->z_gid && 2932 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 2933 err = EDQUOT; 2934 goto out2; 2935 } 2936 } 2937 } 2938 tx = dmu_tx_create(zfsvfs->z_os); 2939 2940 if (mask & AT_MODE) { 2941 uint64_t pmode = zp->z_mode; 2942 uint64_t acl_obj; 2943 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2944 2945 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 2946 goto out; 2947 2948 mutex_enter(&zp->z_lock); 2949 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 2950 /* 2951 * Are we upgrading ACL from old V0 format 2952 * to V1 format? 2953 */ 2954 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 2955 zfs_znode_acl_version(zp) == 2956 ZFS_ACL_VERSION_INITIAL) { 2957 dmu_tx_hold_free(tx, acl_obj, 0, 2958 DMU_OBJECT_END); 2959 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2960 0, aclp->z_acl_bytes); 2961 } else { 2962 dmu_tx_hold_write(tx, acl_obj, 0, 2963 aclp->z_acl_bytes); 2964 } 2965 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2966 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2967 0, aclp->z_acl_bytes); 2968 } 2969 mutex_exit(&zp->z_lock); 2970 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2971 } else { 2972 if ((mask & AT_XVATTR) && 2973 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 2974 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2975 else 2976 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2977 } 2978 2979 if (attrzp) { 2980 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 2981 } 2982 2983 fuid_dirtied = zfsvfs->z_fuid_dirty; 2984 if (fuid_dirtied) 2985 zfs_fuid_txhold(zfsvfs, tx); 2986 2987 zfs_sa_upgrade_txholds(tx, zp); 2988 2989 err = dmu_tx_assign(tx, TXG_NOWAIT); 2990 if (err) { 2991 if (err == ERESTART) 2992 dmu_tx_wait(tx); 2993 goto out; 2994 } 2995 2996 count = 0; 2997 /* 2998 * Set each attribute requested. 2999 * We group settings according to the locks they need to acquire. 3000 * 3001 * Note: you cannot set ctime directly, although it will be 3002 * updated as a side-effect of calling this function. 3003 */ 3004 3005 3006 if (mask & (AT_UID|AT_GID|AT_MODE)) 3007 mutex_enter(&zp->z_acl_lock); 3008 mutex_enter(&zp->z_lock); 3009 3010 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3011 &zp->z_pflags, sizeof (zp->z_pflags)); 3012 3013 if (attrzp) { 3014 if (mask & (AT_UID|AT_GID|AT_MODE)) 3015 mutex_enter(&attrzp->z_acl_lock); 3016 mutex_enter(&attrzp->z_lock); 3017 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3018 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3019 sizeof (attrzp->z_pflags)); 3020 } 3021 3022 if (mask & (AT_UID|AT_GID)) { 3023 3024 if (mask & AT_UID) { 3025 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3026 &new_uid, sizeof (new_uid)); 3027 zp->z_uid = new_uid; 3028 if (attrzp) { 3029 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3030 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3031 sizeof (new_uid)); 3032 attrzp->z_uid = new_uid; 3033 } 3034 } 3035 3036 if (mask & AT_GID) { 3037 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3038 NULL, &new_gid, sizeof (new_gid)); 3039 zp->z_gid = new_gid; 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 = new_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 uid_t owner; 3870 3871 ASSERT(tdvp->v_type == VDIR); 3872 3873 ZFS_ENTER(zfsvfs); 3874 ZFS_VERIFY_ZP(dzp); 3875 zilog = zfsvfs->z_log; 3876 3877 if (VOP_REALVP(svp, &realvp, ct) == 0) 3878 svp = realvp; 3879 3880 /* 3881 * POSIX dictates that we return EPERM here. 3882 * Better choices include ENOTSUP or EISDIR. 3883 */ 3884 if (svp->v_type == VDIR) { 3885 ZFS_EXIT(zfsvfs); 3886 return (EPERM); 3887 } 3888 3889 if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) { 3890 ZFS_EXIT(zfsvfs); 3891 return (EXDEV); 3892 } 3893 3894 szp = VTOZ(svp); 3895 ZFS_VERIFY_ZP(szp); 3896 3897 /* Prevent links to .zfs/shares files */ 3898 3899 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 3900 &parent, sizeof (uint64_t))) != 0) { 3901 ZFS_EXIT(zfsvfs); 3902 return (error); 3903 } 3904 if (parent == zfsvfs->z_shares_dir) { 3905 ZFS_EXIT(zfsvfs); 3906 return (EPERM); 3907 } 3908 3909 if (zfsvfs->z_utf8 && u8_validate(name, 3910 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3911 ZFS_EXIT(zfsvfs); 3912 return (EILSEQ); 3913 } 3914 if (flags & FIGNORECASE) 3915 zf |= ZCILOOK; 3916 3917 /* 3918 * We do not support links between attributes and non-attributes 3919 * because of the potential security risk of creating links 3920 * into "normal" file space in order to circumvent restrictions 3921 * imposed in attribute space. 3922 */ 3923 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 3924 ZFS_EXIT(zfsvfs); 3925 return (EINVAL); 3926 } 3927 3928 3929 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 3930 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) { 3931 ZFS_EXIT(zfsvfs); 3932 return (EPERM); 3933 } 3934 3935 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3936 ZFS_EXIT(zfsvfs); 3937 return (error); 3938 } 3939 3940 top: 3941 /* 3942 * Attempt to lock directory; fail if entry already exists. 3943 */ 3944 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 3945 if (error) { 3946 ZFS_EXIT(zfsvfs); 3947 return (error); 3948 } 3949 3950 tx = dmu_tx_create(zfsvfs->z_os); 3951 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3952 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3953 zfs_sa_upgrade_txholds(tx, szp); 3954 zfs_sa_upgrade_txholds(tx, dzp); 3955 error = dmu_tx_assign(tx, TXG_NOWAIT); 3956 if (error) { 3957 zfs_dirent_unlock(dl); 3958 if (error == ERESTART) { 3959 dmu_tx_wait(tx); 3960 dmu_tx_abort(tx); 3961 goto top; 3962 } 3963 dmu_tx_abort(tx); 3964 ZFS_EXIT(zfsvfs); 3965 return (error); 3966 } 3967 3968 error = zfs_link_create(dl, szp, tx, 0); 3969 3970 if (error == 0) { 3971 uint64_t txtype = TX_LINK; 3972 if (flags & FIGNORECASE) 3973 txtype |= TX_CI; 3974 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 3975 } 3976 3977 dmu_tx_commit(tx); 3978 3979 zfs_dirent_unlock(dl); 3980 3981 if (error == 0) { 3982 vnevent_link(svp, ct); 3983 } 3984 3985 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3986 zil_commit(zilog, 0); 3987 3988 ZFS_EXIT(zfsvfs); 3989 return (error); 3990 } 3991 3992 /* 3993 * zfs_null_putapage() is used when the file system has been force 3994 * unmounted. It just drops the pages. 3995 */ 3996 /* ARGSUSED */ 3997 static int 3998 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 3999 size_t *lenp, int flags, cred_t *cr) 4000 { 4001 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 4002 return (0); 4003 } 4004 4005 /* 4006 * Push a page out to disk, klustering if possible. 4007 * 4008 * IN: vp - file to push page to. 4009 * pp - page to push. 4010 * flags - additional flags. 4011 * cr - credentials of caller. 4012 * 4013 * OUT: offp - start of range pushed. 4014 * lenp - len of range pushed. 4015 * 4016 * RETURN: 0 if success 4017 * error code if failure 4018 * 4019 * NOTE: callers must have locked the page to be pushed. On 4020 * exit, the page (and all other pages in the kluster) must be 4021 * unlocked. 4022 */ 4023 /* ARGSUSED */ 4024 static int 4025 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4026 size_t *lenp, int flags, cred_t *cr) 4027 { 4028 znode_t *zp = VTOZ(vp); 4029 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4030 dmu_tx_t *tx; 4031 u_offset_t off, koff; 4032 size_t len, klen; 4033 int err; 4034 4035 off = pp->p_offset; 4036 len = PAGESIZE; 4037 /* 4038 * If our blocksize is bigger than the page size, try to kluster 4039 * multiple pages so that we write a full block (thus avoiding 4040 * a read-modify-write). 4041 */ 4042 if (off < zp->z_size && zp->z_blksz > PAGESIZE) { 4043 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 4044 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 4045 ASSERT(koff <= zp->z_size); 4046 if (koff + klen > zp->z_size) 4047 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); 4048 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 4049 } 4050 ASSERT3U(btop(len), ==, btopr(len)); 4051 4052 /* 4053 * Can't push pages past end-of-file. 4054 */ 4055 if (off >= zp->z_size) { 4056 /* ignore all pages */ 4057 err = 0; 4058 goto out; 4059 } else if (off + len > zp->z_size) { 4060 int npages = btopr(zp->z_size - off); 4061 page_t *trunc; 4062 4063 page_list_break(&pp, &trunc, npages); 4064 /* ignore pages past end of file */ 4065 if (trunc) 4066 pvn_write_done(trunc, flags); 4067 len = zp->z_size - off; 4068 } 4069 4070 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4071 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4072 err = EDQUOT; 4073 goto out; 4074 } 4075 top: 4076 tx = dmu_tx_create(zfsvfs->z_os); 4077 dmu_tx_hold_write(tx, zp->z_id, off, len); 4078 4079 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4080 zfs_sa_upgrade_txholds(tx, zp); 4081 err = dmu_tx_assign(tx, TXG_NOWAIT); 4082 if (err != 0) { 4083 if (err == ERESTART) { 4084 dmu_tx_wait(tx); 4085 dmu_tx_abort(tx); 4086 goto top; 4087 } 4088 dmu_tx_abort(tx); 4089 goto out; 4090 } 4091 4092 if (zp->z_blksz <= PAGESIZE) { 4093 caddr_t va = zfs_map_page(pp, S_READ); 4094 ASSERT3U(len, <=, PAGESIZE); 4095 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 4096 zfs_unmap_page(pp, va); 4097 } else { 4098 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 4099 } 4100 4101 if (err == 0) { 4102 uint64_t mtime[2], ctime[2]; 4103 sa_bulk_attr_t bulk[3]; 4104 int count = 0; 4105 4106 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4107 &mtime, 16); 4108 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4109 &ctime, 16); 4110 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4111 &zp->z_pflags, 8); 4112 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4113 B_TRUE); 4114 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4115 } 4116 dmu_tx_commit(tx); 4117 4118 out: 4119 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 4120 if (offp) 4121 *offp = off; 4122 if (lenp) 4123 *lenp = len; 4124 4125 return (err); 4126 } 4127 4128 /* 4129 * Copy the portion of the file indicated from pages into the file. 4130 * The pages are stored in a page list attached to the files vnode. 4131 * 4132 * IN: vp - vnode of file to push page data to. 4133 * off - position in file to put data. 4134 * len - amount of data to write. 4135 * flags - flags to control the operation. 4136 * cr - credentials of caller. 4137 * ct - caller context. 4138 * 4139 * RETURN: 0 if success 4140 * error code if failure 4141 * 4142 * Timestamps: 4143 * vp - ctime|mtime updated 4144 */ 4145 /*ARGSUSED*/ 4146 static int 4147 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4148 caller_context_t *ct) 4149 { 4150 znode_t *zp = VTOZ(vp); 4151 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4152 page_t *pp; 4153 size_t io_len; 4154 u_offset_t io_off; 4155 uint_t blksz; 4156 rl_t *rl; 4157 int error = 0; 4158 4159 ZFS_ENTER(zfsvfs); 4160 ZFS_VERIFY_ZP(zp); 4161 4162 /* 4163 * Align this request to the file block size in case we kluster. 4164 * XXX - this can result in pretty aggresive locking, which can 4165 * impact simultanious read/write access. One option might be 4166 * to break up long requests (len == 0) into block-by-block 4167 * operations to get narrower locking. 4168 */ 4169 blksz = zp->z_blksz; 4170 if (ISP2(blksz)) 4171 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 4172 else 4173 io_off = 0; 4174 if (len > 0 && ISP2(blksz)) 4175 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 4176 else 4177 io_len = 0; 4178 4179 if (io_len == 0) { 4180 /* 4181 * Search the entire vp list for pages >= io_off. 4182 */ 4183 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 4184 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 4185 goto out; 4186 } 4187 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 4188 4189 if (off > zp->z_size) { 4190 /* past end of file */ 4191 zfs_range_unlock(rl); 4192 ZFS_EXIT(zfsvfs); 4193 return (0); 4194 } 4195 4196 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); 4197 4198 for (off = io_off; io_off < off + len; io_off += io_len) { 4199 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 4200 pp = page_lookup(vp, io_off, 4201 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 4202 } else { 4203 pp = page_lookup_nowait(vp, io_off, 4204 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 4205 } 4206 4207 if (pp != NULL && pvn_getdirty(pp, flags)) { 4208 int err; 4209 4210 /* 4211 * Found a dirty page to push 4212 */ 4213 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 4214 if (err) 4215 error = err; 4216 } else { 4217 io_len = PAGESIZE; 4218 } 4219 } 4220 out: 4221 zfs_range_unlock(rl); 4222 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4223 zil_commit(zfsvfs->z_log, zp->z_id); 4224 ZFS_EXIT(zfsvfs); 4225 return (error); 4226 } 4227 4228 /*ARGSUSED*/ 4229 void 4230 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4231 { 4232 znode_t *zp = VTOZ(vp); 4233 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4234 int error; 4235 4236 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4237 if (zp->z_sa_hdl == NULL) { 4238 /* 4239 * The fs has been unmounted, or we did a 4240 * suspend/resume and this file no longer exists. 4241 */ 4242 if (vn_has_cached_data(vp)) { 4243 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 4244 B_INVAL, cr); 4245 } 4246 4247 mutex_enter(&zp->z_lock); 4248 mutex_enter(&vp->v_lock); 4249 ASSERT(vp->v_count == 1); 4250 vp->v_count = 0; 4251 mutex_exit(&vp->v_lock); 4252 mutex_exit(&zp->z_lock); 4253 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4254 zfs_znode_free(zp); 4255 return; 4256 } 4257 4258 /* 4259 * Attempt to push any data in the page cache. If this fails 4260 * we will get kicked out later in zfs_zinactive(). 4261 */ 4262 if (vn_has_cached_data(vp)) { 4263 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 4264 cr); 4265 } 4266 4267 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4268 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4269 4270 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4271 zfs_sa_upgrade_txholds(tx, zp); 4272 error = dmu_tx_assign(tx, TXG_WAIT); 4273 if (error) { 4274 dmu_tx_abort(tx); 4275 } else { 4276 mutex_enter(&zp->z_lock); 4277 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4278 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4279 zp->z_atime_dirty = 0; 4280 mutex_exit(&zp->z_lock); 4281 dmu_tx_commit(tx); 4282 } 4283 } 4284 4285 zfs_zinactive(zp); 4286 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4287 } 4288 4289 /* 4290 * Bounds-check the seek operation. 4291 * 4292 * IN: vp - vnode seeking within 4293 * ooff - old file offset 4294 * noffp - pointer to new file offset 4295 * ct - caller context 4296 * 4297 * RETURN: 0 if success 4298 * EINVAL if new offset invalid 4299 */ 4300 /* ARGSUSED */ 4301 static int 4302 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 4303 caller_context_t *ct) 4304 { 4305 if (vp->v_type == VDIR) 4306 return (0); 4307 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4308 } 4309 4310 /* 4311 * Pre-filter the generic locking function to trap attempts to place 4312 * a mandatory lock on a memory mapped file. 4313 */ 4314 static int 4315 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 4316 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 4317 { 4318 znode_t *zp = VTOZ(vp); 4319 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4320 4321 ZFS_ENTER(zfsvfs); 4322 ZFS_VERIFY_ZP(zp); 4323 4324 /* 4325 * We are following the UFS semantics with respect to mapcnt 4326 * here: If we see that the file is mapped already, then we will 4327 * return an error, but we don't worry about races between this 4328 * function and zfs_map(). 4329 */ 4330 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { 4331 ZFS_EXIT(zfsvfs); 4332 return (EAGAIN); 4333 } 4334 ZFS_EXIT(zfsvfs); 4335 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4336 } 4337 4338 /* 4339 * If we can't find a page in the cache, we will create a new page 4340 * and fill it with file data. For efficiency, we may try to fill 4341 * multiple pages at once (klustering) to fill up the supplied page 4342 * list. Note that the pages to be filled are held with an exclusive 4343 * lock to prevent access by other threads while they are being filled. 4344 */ 4345 static int 4346 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 4347 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 4348 { 4349 znode_t *zp = VTOZ(vp); 4350 page_t *pp, *cur_pp; 4351 objset_t *os = zp->z_zfsvfs->z_os; 4352 u_offset_t io_off, total; 4353 size_t io_len; 4354 int err; 4355 4356 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4357 /* 4358 * We only have a single page, don't bother klustering 4359 */ 4360 io_off = off; 4361 io_len = PAGESIZE; 4362 pp = page_create_va(vp, io_off, io_len, 4363 PG_EXCL | PG_WAIT, seg, addr); 4364 } else { 4365 /* 4366 * Try to find enough pages to fill the page list 4367 */ 4368 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4369 &io_len, off, plsz, 0); 4370 } 4371 if (pp == NULL) { 4372 /* 4373 * The page already exists, nothing to do here. 4374 */ 4375 *pl = NULL; 4376 return (0); 4377 } 4378 4379 /* 4380 * Fill the pages in the kluster. 4381 */ 4382 cur_pp = pp; 4383 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4384 caddr_t va; 4385 4386 ASSERT3U(io_off, ==, cur_pp->p_offset); 4387 va = zfs_map_page(cur_pp, S_WRITE); 4388 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4389 DMU_READ_PREFETCH); 4390 zfs_unmap_page(cur_pp, va); 4391 if (err) { 4392 /* On error, toss the entire kluster */ 4393 pvn_read_done(pp, B_ERROR); 4394 /* convert checksum errors into IO errors */ 4395 if (err == ECKSUM) 4396 err = EIO; 4397 return (err); 4398 } 4399 cur_pp = cur_pp->p_next; 4400 } 4401 4402 /* 4403 * Fill in the page list array from the kluster starting 4404 * from the desired offset `off'. 4405 * NOTE: the page list will always be null terminated. 4406 */ 4407 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4408 ASSERT(pl == NULL || (*pl)->p_offset == off); 4409 4410 return (0); 4411 } 4412 4413 /* 4414 * Return pointers to the pages for the file region [off, off + len] 4415 * in the pl array. If plsz is greater than len, this function may 4416 * also return page pointers from after the specified region 4417 * (i.e. the region [off, off + plsz]). These additional pages are 4418 * only returned if they are already in the cache, or were created as 4419 * part of a klustered read. 4420 * 4421 * IN: vp - vnode of file to get data from. 4422 * off - position in file to get data from. 4423 * len - amount of data to retrieve. 4424 * plsz - length of provided page list. 4425 * seg - segment to obtain pages for. 4426 * addr - virtual address of fault. 4427 * rw - mode of created pages. 4428 * cr - credentials of caller. 4429 * ct - caller context. 4430 * 4431 * OUT: protp - protection mode of created pages. 4432 * pl - list of pages created. 4433 * 4434 * RETURN: 0 if success 4435 * error code if failure 4436 * 4437 * Timestamps: 4438 * vp - atime updated 4439 */ 4440 /* ARGSUSED */ 4441 static int 4442 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4443 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4444 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4445 { 4446 znode_t *zp = VTOZ(vp); 4447 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4448 page_t **pl0 = pl; 4449 int err = 0; 4450 4451 /* we do our own caching, faultahead is unnecessary */ 4452 if (pl == NULL) 4453 return (0); 4454 else if (len > plsz) 4455 len = plsz; 4456 else 4457 len = P2ROUNDUP(len, PAGESIZE); 4458 ASSERT(plsz >= len); 4459 4460 ZFS_ENTER(zfsvfs); 4461 ZFS_VERIFY_ZP(zp); 4462 4463 if (protp) 4464 *protp = PROT_ALL; 4465 4466 /* 4467 * Loop through the requested range [off, off + len) looking 4468 * for pages. If we don't find a page, we will need to create 4469 * a new page and fill it with data from the file. 4470 */ 4471 while (len > 0) { 4472 if (*pl = page_lookup(vp, off, SE_SHARED)) 4473 *(pl+1) = NULL; 4474 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 4475 goto out; 4476 while (*pl) { 4477 ASSERT3U((*pl)->p_offset, ==, off); 4478 off += PAGESIZE; 4479 addr += PAGESIZE; 4480 if (len > 0) { 4481 ASSERT3U(len, >=, PAGESIZE); 4482 len -= PAGESIZE; 4483 } 4484 ASSERT3U(plsz, >=, PAGESIZE); 4485 plsz -= PAGESIZE; 4486 pl++; 4487 } 4488 } 4489 4490 /* 4491 * Fill out the page array with any pages already in the cache. 4492 */ 4493 while (plsz > 0 && 4494 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 4495 off += PAGESIZE; 4496 plsz -= PAGESIZE; 4497 } 4498 out: 4499 if (err) { 4500 /* 4501 * Release any pages we have previously locked. 4502 */ 4503 while (pl > pl0) 4504 page_unlock(*--pl); 4505 } else { 4506 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4507 } 4508 4509 *pl = NULL; 4510 4511 ZFS_EXIT(zfsvfs); 4512 return (err); 4513 } 4514 4515 /* 4516 * Request a memory map for a section of a file. This code interacts 4517 * with common code and the VM system as follows: 4518 * 4519 * common code calls mmap(), which ends up in smmap_common() 4520 * 4521 * this calls VOP_MAP(), which takes you into (say) zfs 4522 * 4523 * zfs_map() calls as_map(), passing segvn_create() as the callback 4524 * 4525 * segvn_create() creates the new segment and calls VOP_ADDMAP() 4526 * 4527 * zfs_addmap() updates z_mapcnt 4528 */ 4529 /*ARGSUSED*/ 4530 static int 4531 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4532 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4533 caller_context_t *ct) 4534 { 4535 znode_t *zp = VTOZ(vp); 4536 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4537 segvn_crargs_t vn_a; 4538 int error; 4539 4540 ZFS_ENTER(zfsvfs); 4541 ZFS_VERIFY_ZP(zp); 4542 4543 if ((prot & PROT_WRITE) && (zp->z_pflags & 4544 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 4545 ZFS_EXIT(zfsvfs); 4546 return (EPERM); 4547 } 4548 4549 if ((prot & (PROT_READ | PROT_EXEC)) && 4550 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 4551 ZFS_EXIT(zfsvfs); 4552 return (EACCES); 4553 } 4554 4555 if (vp->v_flag & VNOMAP) { 4556 ZFS_EXIT(zfsvfs); 4557 return (ENOSYS); 4558 } 4559 4560 if (off < 0 || len > MAXOFFSET_T - off) { 4561 ZFS_EXIT(zfsvfs); 4562 return (ENXIO); 4563 } 4564 4565 if (vp->v_type != VREG) { 4566 ZFS_EXIT(zfsvfs); 4567 return (ENODEV); 4568 } 4569 4570 /* 4571 * If file is locked, disallow mapping. 4572 */ 4573 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { 4574 ZFS_EXIT(zfsvfs); 4575 return (EAGAIN); 4576 } 4577 4578 as_rangelock(as); 4579 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 4580 if (error != 0) { 4581 as_rangeunlock(as); 4582 ZFS_EXIT(zfsvfs); 4583 return (error); 4584 } 4585 4586 vn_a.vp = vp; 4587 vn_a.offset = (u_offset_t)off; 4588 vn_a.type = flags & MAP_TYPE; 4589 vn_a.prot = prot; 4590 vn_a.maxprot = maxprot; 4591 vn_a.cred = cr; 4592 vn_a.amp = NULL; 4593 vn_a.flags = flags & ~MAP_TYPE; 4594 vn_a.szc = 0; 4595 vn_a.lgrp_mem_policy_flags = 0; 4596 4597 error = as_map(as, *addrp, len, segvn_create, &vn_a); 4598 4599 as_rangeunlock(as); 4600 ZFS_EXIT(zfsvfs); 4601 return (error); 4602 } 4603 4604 /* ARGSUSED */ 4605 static int 4606 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4607 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4608 caller_context_t *ct) 4609 { 4610 uint64_t pages = btopr(len); 4611 4612 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 4613 return (0); 4614 } 4615 4616 /* 4617 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 4618 * more accurate mtime for the associated file. Since we don't have a way of 4619 * detecting when the data was actually modified, we have to resort to 4620 * heuristics. If an explicit msync() is done, then we mark the mtime when the 4621 * last page is pushed. The problem occurs when the msync() call is omitted, 4622 * which by far the most common case: 4623 * 4624 * open() 4625 * mmap() 4626 * <modify memory> 4627 * munmap() 4628 * close() 4629 * <time lapse> 4630 * putpage() via fsflush 4631 * 4632 * If we wait until fsflush to come along, we can have a modification time that 4633 * is some arbitrary point in the future. In order to prevent this in the 4634 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 4635 * torn down. 4636 */ 4637 /* ARGSUSED */ 4638 static int 4639 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4640 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 4641 caller_context_t *ct) 4642 { 4643 uint64_t pages = btopr(len); 4644 4645 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 4646 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 4647 4648 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 4649 vn_has_cached_data(vp)) 4650 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 4651 4652 return (0); 4653 } 4654 4655 /* 4656 * Free or allocate space in a file. Currently, this function only 4657 * supports the `F_FREESP' command. However, this command is somewhat 4658 * misnamed, as its functionality includes the ability to allocate as 4659 * well as free space. 4660 * 4661 * IN: vp - vnode of file to free data in. 4662 * cmd - action to take (only F_FREESP supported). 4663 * bfp - section of file to free/alloc. 4664 * flag - current file open mode flags. 4665 * offset - current file offset. 4666 * cr - credentials of caller [UNUSED]. 4667 * ct - caller context. 4668 * 4669 * RETURN: 0 if success 4670 * error code if failure 4671 * 4672 * Timestamps: 4673 * vp - ctime|mtime updated 4674 */ 4675 /* ARGSUSED */ 4676 static int 4677 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 4678 offset_t offset, cred_t *cr, caller_context_t *ct) 4679 { 4680 znode_t *zp = VTOZ(vp); 4681 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4682 uint64_t off, len; 4683 int error; 4684 4685 ZFS_ENTER(zfsvfs); 4686 ZFS_VERIFY_ZP(zp); 4687 4688 if (cmd != F_FREESP) { 4689 ZFS_EXIT(zfsvfs); 4690 return (EINVAL); 4691 } 4692 4693 if (error = convoff(vp, bfp, 0, offset)) { 4694 ZFS_EXIT(zfsvfs); 4695 return (error); 4696 } 4697 4698 if (bfp->l_len < 0) { 4699 ZFS_EXIT(zfsvfs); 4700 return (EINVAL); 4701 } 4702 4703 off = bfp->l_start; 4704 len = bfp->l_len; /* 0 means from off to end of file */ 4705 4706 error = zfs_freesp(zp, off, len, flag, TRUE); 4707 4708 ZFS_EXIT(zfsvfs); 4709 return (error); 4710 } 4711 4712 /*ARGSUSED*/ 4713 static int 4714 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 4715 { 4716 znode_t *zp = VTOZ(vp); 4717 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4718 uint32_t gen; 4719 uint64_t gen64; 4720 uint64_t object = zp->z_id; 4721 zfid_short_t *zfid; 4722 int size, i, error; 4723 4724 ZFS_ENTER(zfsvfs); 4725 ZFS_VERIFY_ZP(zp); 4726 4727 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 4728 &gen64, sizeof (uint64_t))) != 0) { 4729 ZFS_EXIT(zfsvfs); 4730 return (error); 4731 } 4732 4733 gen = (uint32_t)gen64; 4734 4735 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 4736 if (fidp->fid_len < size) { 4737 fidp->fid_len = size; 4738 ZFS_EXIT(zfsvfs); 4739 return (ENOSPC); 4740 } 4741 4742 zfid = (zfid_short_t *)fidp; 4743 4744 zfid->zf_len = size; 4745 4746 for (i = 0; i < sizeof (zfid->zf_object); i++) 4747 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 4748 4749 /* Must have a non-zero generation number to distinguish from .zfs */ 4750 if (gen == 0) 4751 gen = 1; 4752 for (i = 0; i < sizeof (zfid->zf_gen); i++) 4753 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 4754 4755 if (size == LONG_FID_LEN) { 4756 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 4757 zfid_long_t *zlfid; 4758 4759 zlfid = (zfid_long_t *)fidp; 4760 4761 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 4762 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 4763 4764 /* XXX - this should be the generation number for the objset */ 4765 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 4766 zlfid->zf_setgen[i] = 0; 4767 } 4768 4769 ZFS_EXIT(zfsvfs); 4770 return (0); 4771 } 4772 4773 static int 4774 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 4775 caller_context_t *ct) 4776 { 4777 znode_t *zp, *xzp; 4778 zfsvfs_t *zfsvfs; 4779 zfs_dirlock_t *dl; 4780 int error; 4781 4782 switch (cmd) { 4783 case _PC_LINK_MAX: 4784 *valp = ULONG_MAX; 4785 return (0); 4786 4787 case _PC_FILESIZEBITS: 4788 *valp = 64; 4789 return (0); 4790 4791 case _PC_XATTR_EXISTS: 4792 zp = VTOZ(vp); 4793 zfsvfs = zp->z_zfsvfs; 4794 ZFS_ENTER(zfsvfs); 4795 ZFS_VERIFY_ZP(zp); 4796 *valp = 0; 4797 error = zfs_dirent_lock(&dl, zp, "", &xzp, 4798 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 4799 if (error == 0) { 4800 zfs_dirent_unlock(dl); 4801 if (!zfs_dirempty(xzp)) 4802 *valp = 1; 4803 VN_RELE(ZTOV(xzp)); 4804 } else if (error == ENOENT) { 4805 /* 4806 * If there aren't extended attributes, it's the 4807 * same as having zero of them. 4808 */ 4809 error = 0; 4810 } 4811 ZFS_EXIT(zfsvfs); 4812 return (error); 4813 4814 case _PC_SATTR_ENABLED: 4815 case _PC_SATTR_EXISTS: 4816 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 4817 (vp->v_type == VREG || vp->v_type == VDIR); 4818 return (0); 4819 4820 case _PC_ACCESS_FILTERING: 4821 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 4822 vp->v_type == VDIR; 4823 return (0); 4824 4825 case _PC_ACL_ENABLED: 4826 *valp = _ACL_ACE_ENABLED; 4827 return (0); 4828 4829 case _PC_MIN_HOLE_SIZE: 4830 *valp = (ulong_t)SPA_MINBLOCKSIZE; 4831 return (0); 4832 4833 case _PC_TIMESTAMP_RESOLUTION: 4834 /* nanosecond timestamp resolution */ 4835 *valp = 1L; 4836 return (0); 4837 4838 default: 4839 return (fs_pathconf(vp, cmd, valp, cr, ct)); 4840 } 4841 } 4842 4843 /*ARGSUSED*/ 4844 static int 4845 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4846 caller_context_t *ct) 4847 { 4848 znode_t *zp = VTOZ(vp); 4849 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4850 int error; 4851 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4852 4853 ZFS_ENTER(zfsvfs); 4854 ZFS_VERIFY_ZP(zp); 4855 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 4856 ZFS_EXIT(zfsvfs); 4857 4858 return (error); 4859 } 4860 4861 /*ARGSUSED*/ 4862 static int 4863 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 4864 caller_context_t *ct) 4865 { 4866 znode_t *zp = VTOZ(vp); 4867 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4868 int error; 4869 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 4870 zilog_t *zilog = zfsvfs->z_log; 4871 4872 ZFS_ENTER(zfsvfs); 4873 ZFS_VERIFY_ZP(zp); 4874 4875 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 4876 4877 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4878 zil_commit(zilog, 0); 4879 4880 ZFS_EXIT(zfsvfs); 4881 return (error); 4882 } 4883 4884 /* 4885 * Tunable, both must be a power of 2. 4886 * 4887 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf 4888 * zcr_blksz_max: if set to less than the file block size, allow loaning out of 4889 * an arcbuf for a partial block read 4890 */ 4891 int zcr_blksz_min = (1 << 10); /* 1K */ 4892 int zcr_blksz_max = (1 << 17); /* 128K */ 4893 4894 /*ARGSUSED*/ 4895 static int 4896 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr, 4897 caller_context_t *ct) 4898 { 4899 znode_t *zp = VTOZ(vp); 4900 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4901 int max_blksz = zfsvfs->z_max_blksz; 4902 uio_t *uio = &xuio->xu_uio; 4903 ssize_t size = uio->uio_resid; 4904 offset_t offset = uio->uio_loffset; 4905 int blksz; 4906 int fullblk, i; 4907 arc_buf_t *abuf; 4908 ssize_t maxsize; 4909 int preamble, postamble; 4910 4911 if (xuio->xu_type != UIOTYPE_ZEROCOPY) 4912 return (EINVAL); 4913 4914 ZFS_ENTER(zfsvfs); 4915 ZFS_VERIFY_ZP(zp); 4916 switch (ioflag) { 4917 case UIO_WRITE: 4918 /* 4919 * Loan out an arc_buf for write if write size is bigger than 4920 * max_blksz, and the file's block size is also max_blksz. 4921 */ 4922 blksz = max_blksz; 4923 if (size < blksz || zp->z_blksz != blksz) { 4924 ZFS_EXIT(zfsvfs); 4925 return (EINVAL); 4926 } 4927 /* 4928 * Caller requests buffers for write before knowing where the 4929 * write offset might be (e.g. NFS TCP write). 4930 */ 4931 if (offset == -1) { 4932 preamble = 0; 4933 } else { 4934 preamble = P2PHASE(offset, blksz); 4935 if (preamble) { 4936 preamble = blksz - preamble; 4937 size -= preamble; 4938 } 4939 } 4940 4941 postamble = P2PHASE(size, blksz); 4942 size -= postamble; 4943 4944 fullblk = size / blksz; 4945 (void) dmu_xuio_init(xuio, 4946 (preamble != 0) + fullblk + (postamble != 0)); 4947 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble, 4948 int, postamble, int, 4949 (preamble != 0) + fullblk + (postamble != 0)); 4950 4951 /* 4952 * Have to fix iov base/len for partial buffers. They 4953 * currently represent full arc_buf's. 4954 */ 4955 if (preamble) { 4956 /* data begins in the middle of the arc_buf */ 4957 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 4958 blksz); 4959 ASSERT(abuf); 4960 (void) dmu_xuio_add(xuio, abuf, 4961 blksz - preamble, preamble); 4962 } 4963 4964 for (i = 0; i < fullblk; i++) { 4965 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 4966 blksz); 4967 ASSERT(abuf); 4968 (void) dmu_xuio_add(xuio, abuf, 0, blksz); 4969 } 4970 4971 if (postamble) { 4972 /* data ends in the middle of the arc_buf */ 4973 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 4974 blksz); 4975 ASSERT(abuf); 4976 (void) dmu_xuio_add(xuio, abuf, 0, postamble); 4977 } 4978 break; 4979 case UIO_READ: 4980 /* 4981 * Loan out an arc_buf for read if the read size is larger than 4982 * the current file block size. Block alignment is not 4983 * considered. Partial arc_buf will be loaned out for read. 4984 */ 4985 blksz = zp->z_blksz; 4986 if (blksz < zcr_blksz_min) 4987 blksz = zcr_blksz_min; 4988 if (blksz > zcr_blksz_max) 4989 blksz = zcr_blksz_max; 4990 /* avoid potential complexity of dealing with it */ 4991 if (blksz > max_blksz) { 4992 ZFS_EXIT(zfsvfs); 4993 return (EINVAL); 4994 } 4995 4996 maxsize = zp->z_size - uio->uio_loffset; 4997 if (size > maxsize) 4998 size = maxsize; 4999 5000 if (size < blksz || vn_has_cached_data(vp)) { 5001 ZFS_EXIT(zfsvfs); 5002 return (EINVAL); 5003 } 5004 break; 5005 default: 5006 ZFS_EXIT(zfsvfs); 5007 return (EINVAL); 5008 } 5009 5010 uio->uio_extflg = UIO_XUIO; 5011 XUIO_XUZC_RW(xuio) = ioflag; 5012 ZFS_EXIT(zfsvfs); 5013 return (0); 5014 } 5015 5016 /*ARGSUSED*/ 5017 static int 5018 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct) 5019 { 5020 int i; 5021 arc_buf_t *abuf; 5022 int ioflag = XUIO_XUZC_RW(xuio); 5023 5024 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY); 5025 5026 i = dmu_xuio_cnt(xuio); 5027 while (i-- > 0) { 5028 abuf = dmu_xuio_arcbuf(xuio, i); 5029 /* 5030 * if abuf == NULL, it must be a write buffer 5031 * that has been returned in zfs_write(). 5032 */ 5033 if (abuf) 5034 dmu_return_arcbuf(abuf); 5035 ASSERT(abuf || ioflag == UIO_WRITE); 5036 } 5037 5038 dmu_xuio_fini(xuio); 5039 return (0); 5040 } 5041 5042 /* 5043 * Predeclare these here so that the compiler assumes that 5044 * this is an "old style" function declaration that does 5045 * not include arguments => we won't get type mismatch errors 5046 * in the initializations that follow. 5047 */ 5048 static int zfs_inval(); 5049 static int zfs_isdir(); 5050 5051 static int 5052 zfs_inval() 5053 { 5054 return (EINVAL); 5055 } 5056 5057 static int 5058 zfs_isdir() 5059 { 5060 return (EISDIR); 5061 } 5062 /* 5063 * Directory vnode operations template 5064 */ 5065 vnodeops_t *zfs_dvnodeops; 5066 const fs_operation_def_t zfs_dvnodeops_template[] = { 5067 VOPNAME_OPEN, { .vop_open = zfs_open }, 5068 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5069 VOPNAME_READ, { .error = zfs_isdir }, 5070 VOPNAME_WRITE, { .error = zfs_isdir }, 5071 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5072 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5073 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5074 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5075 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5076 VOPNAME_CREATE, { .vop_create = zfs_create }, 5077 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5078 VOPNAME_LINK, { .vop_link = zfs_link }, 5079 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5080 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 5081 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5082 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5083 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 5084 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5085 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5086 VOPNAME_FID, { .vop_fid = zfs_fid }, 5087 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5088 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5089 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5090 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5091 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5092 NULL, NULL 5093 }; 5094 5095 /* 5096 * Regular file vnode operations template 5097 */ 5098 vnodeops_t *zfs_fvnodeops; 5099 const fs_operation_def_t zfs_fvnodeops_template[] = { 5100 VOPNAME_OPEN, { .vop_open = zfs_open }, 5101 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5102 VOPNAME_READ, { .vop_read = zfs_read }, 5103 VOPNAME_WRITE, { .vop_write = zfs_write }, 5104 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5105 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5106 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5107 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5108 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5109 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5110 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5111 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5112 VOPNAME_FID, { .vop_fid = zfs_fid }, 5113 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5114 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 5115 VOPNAME_SPACE, { .vop_space = zfs_space }, 5116 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 5117 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 5118 VOPNAME_MAP, { .vop_map = zfs_map }, 5119 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 5120 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 5121 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5122 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5123 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5124 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5125 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf }, 5126 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf }, 5127 NULL, NULL 5128 }; 5129 5130 /* 5131 * Symbolic link vnode operations template 5132 */ 5133 vnodeops_t *zfs_symvnodeops; 5134 const fs_operation_def_t zfs_symvnodeops_template[] = { 5135 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5136 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5137 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5138 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5139 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 5140 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5141 VOPNAME_FID, { .vop_fid = zfs_fid }, 5142 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5143 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5144 NULL, NULL 5145 }; 5146 5147 /* 5148 * special share hidden files vnode operations template 5149 */ 5150 vnodeops_t *zfs_sharevnodeops; 5151 const fs_operation_def_t zfs_sharevnodeops_template[] = { 5152 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5153 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5154 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5155 VOPNAME_FID, { .vop_fid = zfs_fid }, 5156 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5157 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5158 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5159 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5160 NULL, NULL 5161 }; 5162 5163 /* 5164 * Extended attribute directory vnode operations template 5165 * This template is identical to the directory vnodes 5166 * operation template except for restricted operations: 5167 * VOP_MKDIR() 5168 * VOP_SYMLINK() 5169 * Note that there are other restrictions embedded in: 5170 * zfs_create() - restrict type to VREG 5171 * zfs_link() - no links into/out of attribute space 5172 * zfs_rename() - no moves into/out of attribute space 5173 */ 5174 vnodeops_t *zfs_xdvnodeops; 5175 const fs_operation_def_t zfs_xdvnodeops_template[] = { 5176 VOPNAME_OPEN, { .vop_open = zfs_open }, 5177 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5178 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5179 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5180 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5181 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5182 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5183 VOPNAME_CREATE, { .vop_create = zfs_create }, 5184 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5185 VOPNAME_LINK, { .vop_link = zfs_link }, 5186 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5187 VOPNAME_MKDIR, { .error = zfs_inval }, 5188 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5189 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5190 VOPNAME_SYMLINK, { .error = zfs_inval }, 5191 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5192 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5193 VOPNAME_FID, { .vop_fid = zfs_fid }, 5194 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5195 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5196 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5197 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5198 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5199 NULL, NULL 5200 }; 5201 5202 /* 5203 * Error vnode operations template 5204 */ 5205 vnodeops_t *zfs_evnodeops; 5206 const fs_operation_def_t zfs_evnodeops_template[] = { 5207 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5208 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5209 NULL, NULL 5210 }; 5211