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 2006 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <sys/types.h> 29 #include <sys/param.h> 30 #include <sys/time.h> 31 #include <sys/systm.h> 32 #include <sys/sysmacros.h> 33 #include <sys/resource.h> 34 #include <sys/vfs.h> 35 #include <sys/vnode.h> 36 #include <sys/file.h> 37 #include <sys/stat.h> 38 #include <sys/kmem.h> 39 #include <sys/taskq.h> 40 #include <sys/uio.h> 41 #include <sys/vmsystm.h> 42 #include <sys/atomic.h> 43 #include <vm/seg_vn.h> 44 #include <vm/pvn.h> 45 #include <vm/as.h> 46 #include <sys/mman.h> 47 #include <sys/pathname.h> 48 #include <sys/cmn_err.h> 49 #include <sys/errno.h> 50 #include <sys/unistd.h> 51 #include <sys/zfs_vfsops.h> 52 #include <sys/zfs_dir.h> 53 #include <sys/zfs_acl.h> 54 #include <sys/zfs_ioctl.h> 55 #include <sys/fs/zfs.h> 56 #include <sys/dmu.h> 57 #include <sys/spa.h> 58 #include <sys/txg.h> 59 #include <sys/dbuf.h> 60 #include <sys/zap.h> 61 #include <sys/dirent.h> 62 #include <sys/policy.h> 63 #include <sys/sunddi.h> 64 #include <sys/filio.h> 65 #include "fs/fs_subr.h" 66 #include <sys/zfs_ctldir.h> 67 #include <sys/dnlc.h> 68 #include <sys/zfs_rlock.h> 69 70 /* 71 * Programming rules. 72 * 73 * Each vnode op performs some logical unit of work. To do this, the ZPL must 74 * properly lock its in-core state, create a DMU transaction, do the work, 75 * record this work in the intent log (ZIL), commit the DMU transaction, 76 * and wait the the intent log to commit if it's is a synchronous operation. 77 * Morover, the vnode ops must work in both normal and log replay context. 78 * The ordering of events is important to avoid deadlocks and references 79 * to freed memory. The example below illustrates the following Big Rules: 80 * 81 * (1) A check must be made in each zfs thread for a mounted file system. 82 * This is done avoiding races using ZFS_ENTER(zfsvfs). 83 * A ZFS_EXIT(zfsvfs) is needed before all returns. 84 * 85 * (2) VN_RELE() should always be the last thing except for zil_commit() 86 * and ZFS_EXIT(). This is for 3 reasons: 87 * First, if it's the last reference, the vnode/znode 88 * can be freed, so the zp may point to freed memory. Second, the last 89 * reference will call zfs_zinactive(), which may induce a lot of work -- 90 * pushing cached pages (which acquires range locks) and syncing out 91 * cached atime changes. Third, zfs_zinactive() may require a new tx, 92 * which could deadlock the system if you were already holding one. 93 * 94 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 95 * as they can span dmu_tx_assign() calls. 96 * 97 * (4) Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign(). 98 * In normal operation, this will be TXG_NOWAIT. During ZIL replay, 99 * it will be a specific txg. Either way, dmu_tx_assign() never blocks. 100 * This is critical because we don't want to block while holding locks. 101 * Note, in particular, that if a lock is sometimes acquired before 102 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 103 * use a non-blocking assign can deadlock the system. The scenario: 104 * 105 * Thread A has grabbed a lock before calling dmu_tx_assign(). 106 * Thread B is in an already-assigned tx, and blocks for this lock. 107 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 108 * forever, because the previous txg can't quiesce until B's tx commits. 109 * 110 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 111 * then drop all locks, call dmu_tx_wait(), and try again. 112 * 113 * (5) If the operation succeeded, generate the intent log entry for it 114 * before dropping locks. This ensures that the ordering of events 115 * in the intent log matches the order in which they actually occurred. 116 * 117 * (6) At the end of each vnode op, the DMU tx must always commit, 118 * regardless of whether there were any errors. 119 * 120 * (7) After dropping all locks, invoke zil_commit(zilog, seq, ioflag) 121 * to ensure that synchronous semantics are provided when necessary. 122 * 123 * In general, this is how things should be ordered in each vnode op: 124 * 125 * ZFS_ENTER(zfsvfs); // exit if unmounted 126 * top: 127 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 128 * rw_enter(...); // grab any other locks you need 129 * tx = dmu_tx_create(...); // get DMU tx 130 * dmu_tx_hold_*(); // hold each object you might modify 131 * error = dmu_tx_assign(tx, zfsvfs->z_assign); // try to assign 132 * if (error) { 133 * rw_exit(...); // drop locks 134 * zfs_dirent_unlock(dl); // unlock directory entry 135 * VN_RELE(...); // release held vnodes 136 * if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 137 * dmu_tx_wait(tx); 138 * dmu_tx_abort(tx); 139 * goto top; 140 * } 141 * dmu_tx_abort(tx); // abort DMU tx 142 * ZFS_EXIT(zfsvfs); // finished in zfs 143 * return (error); // really out of space 144 * } 145 * error = do_real_work(); // do whatever this VOP does 146 * if (error == 0) 147 * seq = zfs_log_*(...); // on success, make ZIL entry 148 * dmu_tx_commit(tx); // commit DMU tx -- error or not 149 * rw_exit(...); // drop locks 150 * zfs_dirent_unlock(dl); // unlock directory entry 151 * VN_RELE(...); // release held vnodes 152 * zil_commit(zilog, seq, ioflag); // synchronous when necessary 153 * ZFS_EXIT(zfsvfs); // finished in zfs 154 * return (error); // done, report error 155 */ 156 157 /* ARGSUSED */ 158 static int 159 zfs_open(vnode_t **vpp, int flag, cred_t *cr) 160 { 161 return (0); 162 } 163 164 /* ARGSUSED */ 165 static int 166 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr) 167 { 168 /* 169 * Clean up any locks held by this process on the vp. 170 */ 171 cleanlocks(vp, ddi_get_pid(), 0); 172 cleanshares(vp, ddi_get_pid()); 173 174 return (0); 175 } 176 177 /* 178 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 179 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 180 */ 181 static int 182 zfs_holey(vnode_t *vp, int cmd, offset_t *off) 183 { 184 znode_t *zp = VTOZ(vp); 185 uint64_t noff = (uint64_t)*off; /* new offset */ 186 uint64_t file_sz; 187 int error; 188 boolean_t hole; 189 190 file_sz = zp->z_phys->zp_size; 191 if (noff >= file_sz) { 192 return (ENXIO); 193 } 194 195 if (cmd == _FIO_SEEK_HOLE) 196 hole = B_TRUE; 197 else 198 hole = B_FALSE; 199 200 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 201 202 /* end of file? */ 203 if ((error == ESRCH) || (noff > file_sz)) { 204 /* 205 * Handle the virtual hole at the end of file. 206 */ 207 if (hole) { 208 *off = file_sz; 209 return (0); 210 } 211 return (ENXIO); 212 } 213 214 if (noff < *off) 215 return (error); 216 *off = noff; 217 return (error); 218 } 219 220 /* ARGSUSED */ 221 static int 222 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred, 223 int *rvalp) 224 { 225 offset_t off; 226 int error; 227 zfsvfs_t *zfsvfs; 228 229 switch (com) { 230 case _FIOFFS: 231 return (zfs_sync(vp->v_vfsp, 0, cred)); 232 233 /* 234 * The following two ioctls are used by bfu. Faking out, 235 * necessary to avoid bfu errors. 236 */ 237 case _FIOGDIO: 238 case _FIOSDIO: 239 return (0); 240 241 case _FIO_SEEK_DATA: 242 case _FIO_SEEK_HOLE: 243 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 244 return (EFAULT); 245 246 zfsvfs = VTOZ(vp)->z_zfsvfs; 247 ZFS_ENTER(zfsvfs); 248 249 /* offset parameter is in/out */ 250 error = zfs_holey(vp, com, &off); 251 ZFS_EXIT(zfsvfs); 252 if (error) 253 return (error); 254 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 255 return (EFAULT); 256 return (0); 257 } 258 return (ENOTTY); 259 } 260 261 /* 262 * When a file is memory mapped, we must keep the IO data synchronized 263 * between the DMU cache and the memory mapped pages. What this means: 264 * 265 * On Write: If we find a memory mapped page, we write to *both* 266 * the page and the dmu buffer. 267 * 268 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 269 * the file is memory mapped. 270 */ 271 static int 272 mappedwrite(vnode_t *vp, uint64_t woff, int nbytes, uio_t *uio, dmu_tx_t *tx) 273 { 274 znode_t *zp = VTOZ(vp); 275 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 276 int64_t start, off; 277 int len = nbytes; 278 int error = 0; 279 280 start = uio->uio_loffset; 281 off = start & PAGEOFFSET; 282 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 283 page_t *pp; 284 uint64_t bytes = MIN(PAGESIZE - off, len); 285 286 /* 287 * We don't want a new page to "appear" in the middle of 288 * the file update (because it may not get the write 289 * update data), so we grab a lock to block 290 * zfs_getpage(). 291 */ 292 rw_enter(&zp->z_map_lock, RW_WRITER); 293 if (pp = page_lookup(vp, start, SE_SHARED)) { 294 caddr_t va; 295 296 rw_exit(&zp->z_map_lock); 297 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L); 298 error = uiomove(va+off, bytes, UIO_WRITE, uio); 299 if (error == 0) { 300 dmu_write(zfsvfs->z_os, zp->z_id, 301 woff, bytes, va+off, tx); 302 } 303 ppmapout(va); 304 page_unlock(pp); 305 } else { 306 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 307 woff, bytes, uio, tx); 308 rw_exit(&zp->z_map_lock); 309 } 310 len -= bytes; 311 woff += bytes; 312 off = 0; 313 if (error) 314 break; 315 } 316 return (error); 317 } 318 319 /* 320 * When a file is memory mapped, we must keep the IO data synchronized 321 * between the DMU cache and the memory mapped pages. What this means: 322 * 323 * On Read: We "read" preferentially from memory mapped pages, 324 * else we default from the dmu buffer. 325 * 326 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 327 * the file is memory mapped. 328 */ 329 static int 330 mappedread(vnode_t *vp, char *addr, int nbytes, uio_t *uio) 331 { 332 int64_t start, off, bytes; 333 int len = nbytes; 334 int error = 0; 335 336 start = uio->uio_loffset; 337 off = start & PAGEOFFSET; 338 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 339 page_t *pp; 340 341 bytes = MIN(PAGESIZE - off, len); 342 if (pp = page_lookup(vp, start, SE_SHARED)) { 343 caddr_t va; 344 345 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L); 346 error = uiomove(va + off, bytes, UIO_READ, uio); 347 ppmapout(va); 348 page_unlock(pp); 349 } else { 350 /* XXX use dmu_read here? */ 351 error = uiomove(addr, bytes, UIO_READ, uio); 352 } 353 len -= bytes; 354 addr += bytes; 355 off = 0; 356 if (error) 357 break; 358 } 359 return (error); 360 } 361 362 uint_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 363 364 /* 365 * Read bytes from specified file into supplied buffer. 366 * 367 * IN: vp - vnode of file to be read from. 368 * uio - structure supplying read location, range info, 369 * and return buffer. 370 * ioflag - SYNC flags; used to provide FRSYNC semantics. 371 * cr - credentials of caller. 372 * 373 * OUT: uio - updated offset and range, buffer filled. 374 * 375 * RETURN: 0 if success 376 * error code if failure 377 * 378 * Side Effects: 379 * vp - atime updated if byte count > 0 380 */ 381 /* ARGSUSED */ 382 static int 383 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 384 { 385 znode_t *zp = VTOZ(vp); 386 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 387 uint64_t delta; 388 ssize_t n, size, cnt, ndone; 389 int error, i, numbufs; 390 dmu_buf_t *dbp, **dbpp; 391 rl_t *rl; 392 393 ZFS_ENTER(zfsvfs); 394 395 /* 396 * Validate file offset 397 */ 398 if (uio->uio_loffset < (offset_t)0) { 399 ZFS_EXIT(zfsvfs); 400 return (EINVAL); 401 } 402 403 /* 404 * Fasttrack empty reads 405 */ 406 if (uio->uio_resid == 0) { 407 ZFS_EXIT(zfsvfs); 408 return (0); 409 } 410 411 /* 412 * Check for mandatory locks 413 */ 414 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 415 if (error = chklock(vp, FREAD, 416 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 417 ZFS_EXIT(zfsvfs); 418 return (error); 419 } 420 } 421 422 /* 423 * If we're in FRSYNC mode, sync out this znode before reading it. 424 */ 425 zil_commit(zfsvfs->z_log, zp->z_last_itx, ioflag & FRSYNC); 426 427 /* 428 * Lock the range against changes. 429 */ 430 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 431 432 /* 433 * If we are reading past end-of-file we can skip 434 * to the end; but we might still need to set atime. 435 */ 436 if (uio->uio_loffset >= zp->z_phys->zp_size) { 437 cnt = 0; 438 error = 0; 439 goto out; 440 } 441 442 cnt = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset); 443 444 for (ndone = 0; ndone < cnt; ndone += zfs_read_chunk_size) { 445 ASSERT(uio->uio_loffset < zp->z_phys->zp_size); 446 n = MIN(zfs_read_chunk_size, 447 zp->z_phys->zp_size - uio->uio_loffset); 448 n = MIN(n, cnt); 449 error = dmu_buf_hold_array(zfsvfs->z_os, zp->z_id, 450 uio->uio_loffset, n, TRUE, FTAG, &numbufs, &dbpp); 451 if (error) 452 goto out; 453 /* 454 * Compute the adjustment to align the dmu buffers 455 * with the uio buffer. 456 */ 457 delta = uio->uio_loffset - dbpp[0]->db_offset; 458 459 for (i = 0; i < numbufs; i++) { 460 if (n < 0) 461 break; 462 dbp = dbpp[i]; 463 size = dbp->db_size - delta; 464 /* 465 * XXX -- this is correct, but may be suboptimal. 466 * If the pages are all clean, we don't need to 467 * go through mappedread(). Maybe the VMODSORT 468 * stuff can help us here. 469 */ 470 if (vn_has_cached_data(vp)) { 471 error = mappedread(vp, (caddr_t)dbp->db_data + 472 delta, (n < size ? n : size), uio); 473 } else { 474 error = uiomove((caddr_t)dbp->db_data + delta, 475 (n < size ? n : size), UIO_READ, uio); 476 } 477 if (error) { 478 dmu_buf_rele_array(dbpp, numbufs, FTAG); 479 goto out; 480 } 481 n -= dbp->db_size; 482 if (delta) { 483 n += delta; 484 delta = 0; 485 } 486 } 487 dmu_buf_rele_array(dbpp, numbufs, FTAG); 488 } 489 out: 490 zfs_range_unlock(zp, rl); 491 492 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 493 ZFS_EXIT(zfsvfs); 494 return (error); 495 } 496 497 /* 498 * Fault in the pages of the first n bytes specified by the uio structure. 499 * 1 byte in each page is touched and the uio struct is unmodified. 500 * Any error will exit this routine as this is only a best 501 * attempt to get the pages resident. This is a copy of ufs_trans_touch(). 502 */ 503 static void 504 zfs_prefault_write(ssize_t n, struct uio *uio) 505 { 506 struct iovec *iov; 507 ulong_t cnt, incr; 508 caddr_t p; 509 uint8_t tmp; 510 511 iov = uio->uio_iov; 512 513 while (n) { 514 cnt = MIN(iov->iov_len, n); 515 if (cnt == 0) { 516 /* empty iov entry */ 517 iov++; 518 continue; 519 } 520 n -= cnt; 521 /* 522 * touch each page in this segment. 523 */ 524 p = iov->iov_base; 525 while (cnt) { 526 switch (uio->uio_segflg) { 527 case UIO_USERSPACE: 528 case UIO_USERISPACE: 529 if (fuword8(p, &tmp)) 530 return; 531 break; 532 case UIO_SYSSPACE: 533 if (kcopy(p, &tmp, 1)) 534 return; 535 break; 536 } 537 incr = MIN(cnt, PAGESIZE); 538 p += incr; 539 cnt -= incr; 540 } 541 /* 542 * touch the last byte in case it straddles a page. 543 */ 544 p--; 545 switch (uio->uio_segflg) { 546 case UIO_USERSPACE: 547 case UIO_USERISPACE: 548 if (fuword8(p, &tmp)) 549 return; 550 break; 551 case UIO_SYSSPACE: 552 if (kcopy(p, &tmp, 1)) 553 return; 554 break; 555 } 556 iov++; 557 } 558 } 559 560 /* 561 * Write the bytes to a file. 562 * 563 * IN: vp - vnode of file to be written to. 564 * uio - structure supplying write location, range info, 565 * and data buffer. 566 * ioflag - FAPPEND flag set if in append mode. 567 * cr - credentials of caller. 568 * 569 * OUT: uio - updated offset and range. 570 * 571 * RETURN: 0 if success 572 * error code if failure 573 * 574 * Timestamps: 575 * vp - ctime|mtime updated if byte count > 0 576 */ 577 /* ARGSUSED */ 578 static int 579 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 580 { 581 znode_t *zp = VTOZ(vp); 582 rlim64_t limit = uio->uio_llimit; 583 ssize_t start_resid = uio->uio_resid; 584 ssize_t tx_bytes; 585 uint64_t end_size; 586 dmu_tx_t *tx; 587 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 588 zilog_t *zilog = zfsvfs->z_log; 589 uint64_t seq = 0; 590 offset_t woff; 591 ssize_t n, nbytes; 592 rl_t *rl; 593 int max_blksz = zfsvfs->z_max_blksz; 594 int error; 595 596 /* 597 * Fasttrack empty write 598 */ 599 n = start_resid; 600 if (n == 0) 601 return (0); 602 603 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 604 limit = MAXOFFSET_T; 605 606 ZFS_ENTER(zfsvfs); 607 608 /* 609 * Pre-fault the initial pages to ensure slow (eg NFS) pages 610 * don't hold up txg. 611 */ 612 zfs_prefault_write(MIN(start_resid, SPA_MAXBLOCKSIZE), uio); 613 614 /* 615 * If in append mode, set the io offset pointer to eof. 616 */ 617 if (ioflag & FAPPEND) { 618 /* 619 * Range lock for a file append: 620 * The value for the start of range will be determined by 621 * zfs_range_lock() (to guarantee append semantics). 622 * If this write will cause the block size to increase, 623 * zfs_range_lock() will lock the entire file, so we must 624 * later reduce the range after we grow the block size. 625 */ 626 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 627 if (rl->r_len == UINT64_MAX) { 628 /* overlocked, zp_size can't change */ 629 woff = uio->uio_loffset = zp->z_phys->zp_size; 630 } else { 631 woff = uio->uio_loffset = rl->r_off; 632 } 633 } else { 634 woff = uio->uio_loffset; 635 /* 636 * Validate file offset 637 */ 638 if (woff < 0) { 639 ZFS_EXIT(zfsvfs); 640 return (EINVAL); 641 } 642 643 /* 644 * If we need to grow the block size then zfs_range_lock() 645 * will lock a wider range than we request here. 646 * Later after growing the block size we reduce the range. 647 */ 648 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 649 } 650 651 if (woff >= limit) { 652 error = EFBIG; 653 goto no_tx_done; 654 } 655 656 if ((woff + n) > limit || woff > (limit - n)) 657 n = limit - woff; 658 659 /* 660 * Check for mandatory locks 661 */ 662 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 663 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) 664 goto no_tx_done; 665 end_size = MAX(zp->z_phys->zp_size, woff + n); 666 top: 667 tx = dmu_tx_create(zfsvfs->z_os); 668 dmu_tx_hold_bonus(tx, zp->z_id); 669 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 670 error = dmu_tx_assign(tx, zfsvfs->z_assign); 671 if (error) { 672 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 673 dmu_tx_wait(tx); 674 dmu_tx_abort(tx); 675 goto top; 676 } 677 dmu_tx_abort(tx); 678 goto no_tx_done; 679 } 680 681 /* 682 * If zfs_range_lock() over-locked we grow the blocksize 683 * and then reduce the lock range. 684 */ 685 if (rl->r_len == UINT64_MAX) { 686 uint64_t new_blksz; 687 688 if (zp->z_blksz > max_blksz) { 689 ASSERT(!ISP2(zp->z_blksz)); 690 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 691 } else { 692 new_blksz = MIN(end_size, max_blksz); 693 } 694 zfs_grow_blocksize(zp, new_blksz, tx); 695 zfs_range_reduce(zp, rl, woff, n); 696 } 697 698 /* 699 * The file data does not fit in the znode "cache", so we 700 * will be writing to the file block data buffers. 701 * Each buffer will be written in a separate transaction; 702 * this keeps the intent log records small and allows us 703 * to do more fine-grained space accounting. 704 */ 705 while (n > 0) { 706 /* 707 * XXX - should we really limit each write to z_max_blksz? 708 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 709 */ 710 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 711 rw_enter(&zp->z_map_lock, RW_READER); 712 713 tx_bytes = uio->uio_resid; 714 if (vn_has_cached_data(vp)) { 715 rw_exit(&zp->z_map_lock); 716 error = mappedwrite(vp, woff, nbytes, uio, tx); 717 } else { 718 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 719 woff, nbytes, uio, tx); 720 rw_exit(&zp->z_map_lock); 721 } 722 tx_bytes -= uio->uio_resid; 723 724 if (error) { 725 /* XXX - do we need to "clean up" the dmu buffer? */ 726 break; 727 } 728 729 ASSERT(tx_bytes == nbytes); 730 731 /* 732 * Clear Set-UID/Set-GID bits on successful write if not 733 * privileged and at least one of the excute bits is set. 734 * 735 * It would be nice to to this after all writes have 736 * been done, but that would still expose the ISUID/ISGID 737 * to another app after the partial write is committed. 738 */ 739 740 mutex_enter(&zp->z_acl_lock); 741 if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) | 742 (S_IXUSR >> 6))) != 0 && 743 (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 && 744 secpolicy_vnode_setid_retain(cr, 745 (zp->z_phys->zp_mode & S_ISUID) != 0 && 746 zp->z_phys->zp_uid == 0) != 0) { 747 zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID); 748 } 749 mutex_exit(&zp->z_acl_lock); 750 751 n -= nbytes; 752 if (n <= 0) 753 break; 754 755 /* 756 * We have more work ahead of us, so wrap up this transaction 757 * and start another. Exact same logic as tx_done below. 758 */ 759 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) { 760 dmu_buf_will_dirty(zp->z_dbuf, tx); 761 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 762 uio->uio_loffset); 763 } 764 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 765 seq = zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, 766 ioflag, uio); 767 dmu_tx_commit(tx); 768 769 /* Pre-fault the next set of pages */ 770 zfs_prefault_write(MIN(n, SPA_MAXBLOCKSIZE), uio); 771 772 /* 773 * Start another transaction. 774 */ 775 woff = uio->uio_loffset; 776 tx = dmu_tx_create(zfsvfs->z_os); 777 dmu_tx_hold_bonus(tx, zp->z_id); 778 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 779 error = dmu_tx_assign(tx, zfsvfs->z_assign); 780 if (error) { 781 if (error == ERESTART && 782 zfsvfs->z_assign == TXG_NOWAIT) { 783 dmu_tx_wait(tx); 784 dmu_tx_abort(tx); 785 goto top; 786 } 787 dmu_tx_abort(tx); 788 goto no_tx_done; 789 } 790 } 791 792 tx_done: 793 794 if (tx_bytes != 0) { 795 /* 796 * Update the file size if it has changed; account 797 * for possible concurrent updates. 798 */ 799 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) { 800 dmu_buf_will_dirty(zp->z_dbuf, tx); 801 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 802 uio->uio_loffset); 803 } 804 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 805 seq = zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, 806 ioflag, uio); 807 } 808 dmu_tx_commit(tx); 809 810 811 no_tx_done: 812 813 zfs_range_unlock(zp, rl); 814 815 /* 816 * If we're in replay mode, or we made no progress, return error. 817 * Otherwise, it's at least a partial write, so it's successful. 818 */ 819 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 820 ZFS_EXIT(zfsvfs); 821 return (error); 822 } 823 824 zil_commit(zilog, seq, ioflag & (FSYNC | FDSYNC)); 825 826 ZFS_EXIT(zfsvfs); 827 return (0); 828 } 829 830 /* 831 * Get data to generate a TX_WRITE intent log record. 832 */ 833 int 834 zfs_get_data(void *arg, lr_write_t *lr, char *buf) 835 { 836 zfsvfs_t *zfsvfs = arg; 837 objset_t *os = zfsvfs->z_os; 838 znode_t *zp; 839 uint64_t off = lr->lr_offset; 840 rl_t *rl; 841 int dlen = lr->lr_length; /* length of user data */ 842 int error = 0; 843 844 ASSERT(dlen != 0); 845 846 /* 847 * Nothing to do if the file has been removed 848 */ 849 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 850 return (ENOENT); 851 if (zp->z_reap) { 852 VN_RELE(ZTOV(zp)); 853 return (ENOENT); 854 } 855 856 /* 857 * Write records come in two flavors: immediate and indirect. 858 * For small writes it's cheaper to store the data with the 859 * log record (immediate); for large writes it's cheaper to 860 * sync the data and get a pointer to it (indirect) so that 861 * we don't have to write the data twice. 862 */ 863 if (buf != NULL) { /* immediate write */ 864 dmu_buf_t *db; 865 866 rl = zfs_range_lock(zp, off, dlen, RL_READER); 867 /* test for truncation needs to be done while range locked */ 868 if (off >= zp->z_phys->zp_size) { 869 error = ENOENT; 870 goto out; 871 } 872 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, off, FTAG, &db)); 873 bcopy((char *)db->db_data + off - db->db_offset, buf, dlen); 874 dmu_buf_rele(db, FTAG); 875 } else { /* indirect write */ 876 uint64_t boff; /* block starting offset */ 877 878 /* 879 * Have to lock the whole block to ensure when it's 880 * written out and it's checksum is being calculated 881 * that no one can change the data. We need to re-check 882 * blocksize after we get the lock in case it's changed! 883 */ 884 for (;;) { 885 if (ISP2(zp->z_blksz)) { 886 boff = P2ALIGN_TYPED(off, zp->z_blksz, 887 uint64_t); 888 } else { 889 boff = 0; 890 } 891 dlen = zp->z_blksz; 892 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 893 if (zp->z_blksz == dlen) 894 break; 895 zfs_range_unlock(zp, rl); 896 } 897 /* test for truncation needs to be done while range locked */ 898 if (off >= zp->z_phys->zp_size) { 899 error = ENOENT; 900 goto out; 901 } 902 txg_suspend(dmu_objset_pool(os)); 903 error = dmu_sync(os, lr->lr_foid, off, &lr->lr_blkoff, 904 &lr->lr_blkptr, lr->lr_common.lrc_txg); 905 txg_resume(dmu_objset_pool(os)); 906 } 907 out: 908 zfs_range_unlock(zp, rl); 909 VN_RELE(ZTOV(zp)); 910 return (error); 911 } 912 913 /*ARGSUSED*/ 914 static int 915 zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr) 916 { 917 znode_t *zp = VTOZ(vp); 918 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 919 int error; 920 921 ZFS_ENTER(zfsvfs); 922 error = zfs_zaccess_rwx(zp, mode, cr); 923 ZFS_EXIT(zfsvfs); 924 return (error); 925 } 926 927 /* 928 * Lookup an entry in a directory, or an extended attribute directory. 929 * If it exists, return a held vnode reference for it. 930 * 931 * IN: dvp - vnode of directory to search. 932 * nm - name of entry to lookup. 933 * pnp - full pathname to lookup [UNUSED]. 934 * flags - LOOKUP_XATTR set if looking for an attribute. 935 * rdir - root directory vnode [UNUSED]. 936 * cr - credentials of caller. 937 * 938 * OUT: vpp - vnode of located entry, NULL if not found. 939 * 940 * RETURN: 0 if success 941 * error code if failure 942 * 943 * Timestamps: 944 * NA 945 */ 946 /* ARGSUSED */ 947 static int 948 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 949 int flags, vnode_t *rdir, cred_t *cr) 950 { 951 952 znode_t *zdp = VTOZ(dvp); 953 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 954 int error; 955 956 ZFS_ENTER(zfsvfs); 957 958 *vpp = NULL; 959 960 if (flags & LOOKUP_XATTR) { 961 /* 962 * We don't allow recursive attributes.. 963 * Maybe someday we will. 964 */ 965 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 966 ZFS_EXIT(zfsvfs); 967 return (EINVAL); 968 } 969 970 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr)) { 971 ZFS_EXIT(zfsvfs); 972 return (error); 973 } 974 975 /* 976 * Do we have permission to get into attribute directory? 977 */ 978 979 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) { 980 VN_RELE(*vpp); 981 } 982 983 ZFS_EXIT(zfsvfs); 984 return (error); 985 } 986 987 if (dvp->v_type != VDIR) { 988 ZFS_EXIT(zfsvfs); 989 return (ENOTDIR); 990 } 991 992 /* 993 * Check accessibility of directory. 994 */ 995 996 if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) { 997 ZFS_EXIT(zfsvfs); 998 return (error); 999 } 1000 1001 if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) { 1002 1003 /* 1004 * Convert device special files 1005 */ 1006 if (IS_DEVVP(*vpp)) { 1007 vnode_t *svp; 1008 1009 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1010 VN_RELE(*vpp); 1011 if (svp == NULL) 1012 error = ENOSYS; 1013 else 1014 *vpp = svp; 1015 } 1016 } 1017 1018 ZFS_EXIT(zfsvfs); 1019 return (error); 1020 } 1021 1022 /* 1023 * Attempt to create a new entry in a directory. If the entry 1024 * already exists, truncate the file if permissible, else return 1025 * an error. Return the vp of the created or trunc'd file. 1026 * 1027 * IN: dvp - vnode of directory to put new file entry in. 1028 * name - name of new file entry. 1029 * vap - attributes of new file. 1030 * excl - flag indicating exclusive or non-exclusive mode. 1031 * mode - mode to open file with. 1032 * cr - credentials of caller. 1033 * flag - large file flag [UNUSED]. 1034 * 1035 * OUT: vpp - vnode of created or trunc'd entry. 1036 * 1037 * RETURN: 0 if success 1038 * error code if failure 1039 * 1040 * Timestamps: 1041 * dvp - ctime|mtime updated if new entry created 1042 * vp - ctime|mtime always, atime if new 1043 */ 1044 /* ARGSUSED */ 1045 static int 1046 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1047 int mode, vnode_t **vpp, cred_t *cr, int flag) 1048 { 1049 znode_t *zp, *dzp = VTOZ(dvp); 1050 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1051 zilog_t *zilog = zfsvfs->z_log; 1052 uint64_t seq = 0; 1053 objset_t *os = zfsvfs->z_os; 1054 zfs_dirlock_t *dl; 1055 dmu_tx_t *tx; 1056 int error; 1057 uint64_t zoid; 1058 1059 ZFS_ENTER(zfsvfs); 1060 1061 top: 1062 *vpp = NULL; 1063 1064 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1065 vap->va_mode &= ~VSVTX; 1066 1067 if (*name == '\0') { 1068 /* 1069 * Null component name refers to the directory itself. 1070 */ 1071 VN_HOLD(dvp); 1072 zp = dzp; 1073 dl = NULL; 1074 error = 0; 1075 } else { 1076 /* possible VN_HOLD(zp) */ 1077 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) { 1078 if (strcmp(name, "..") == 0) 1079 error = EISDIR; 1080 ZFS_EXIT(zfsvfs); 1081 return (error); 1082 } 1083 } 1084 1085 zoid = zp ? zp->z_id : -1ULL; 1086 1087 if (zp == NULL) { 1088 /* 1089 * Create a new file object and update the directory 1090 * to reference it. 1091 */ 1092 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 1093 goto out; 1094 } 1095 1096 /* 1097 * We only support the creation of regular files in 1098 * extended attribute directories. 1099 */ 1100 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1101 (vap->va_type != VREG)) { 1102 error = EINVAL; 1103 goto out; 1104 } 1105 1106 tx = dmu_tx_create(os); 1107 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1108 dmu_tx_hold_bonus(tx, dzp->z_id); 1109 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1110 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1111 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1112 0, SPA_MAXBLOCKSIZE); 1113 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1114 if (error) { 1115 zfs_dirent_unlock(dl); 1116 if (error == ERESTART && 1117 zfsvfs->z_assign == TXG_NOWAIT) { 1118 dmu_tx_wait(tx); 1119 dmu_tx_abort(tx); 1120 goto top; 1121 } 1122 dmu_tx_abort(tx); 1123 ZFS_EXIT(zfsvfs); 1124 return (error); 1125 } 1126 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1127 ASSERT(zp->z_id == zoid); 1128 (void) zfs_link_create(dl, zp, tx, ZNEW); 1129 seq = zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name); 1130 dmu_tx_commit(tx); 1131 } else { 1132 /* 1133 * A directory entry already exists for this name. 1134 */ 1135 /* 1136 * Can't truncate an existing file if in exclusive mode. 1137 */ 1138 if (excl == EXCL) { 1139 error = EEXIST; 1140 goto out; 1141 } 1142 /* 1143 * Can't open a directory for writing. 1144 */ 1145 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1146 error = EISDIR; 1147 goto out; 1148 } 1149 /* 1150 * Verify requested access to file. 1151 */ 1152 if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) { 1153 goto out; 1154 } 1155 1156 mutex_enter(&dzp->z_lock); 1157 dzp->z_seq++; 1158 mutex_exit(&dzp->z_lock); 1159 1160 /* 1161 * Truncate regular files if requested. 1162 */ 1163 if ((ZTOV(zp)->v_type == VREG) && 1164 (zp->z_phys->zp_size != 0) && 1165 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1166 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1167 if (error == ERESTART && 1168 zfsvfs->z_assign == TXG_NOWAIT) { 1169 /* NB: we already did dmu_tx_wait() */ 1170 zfs_dirent_unlock(dl); 1171 goto top; 1172 } 1173 } 1174 } 1175 out: 1176 1177 if (dl) 1178 zfs_dirent_unlock(dl); 1179 1180 if (error) { 1181 if (zp) 1182 VN_RELE(ZTOV(zp)); 1183 } else { 1184 *vpp = ZTOV(zp); 1185 /* 1186 * If vnode is for a device return a specfs vnode instead. 1187 */ 1188 if (IS_DEVVP(*vpp)) { 1189 struct vnode *svp; 1190 1191 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1192 VN_RELE(*vpp); 1193 if (svp == NULL) { 1194 error = ENOSYS; 1195 } 1196 *vpp = svp; 1197 } 1198 } 1199 1200 zil_commit(zilog, seq, 0); 1201 1202 ZFS_EXIT(zfsvfs); 1203 return (error); 1204 } 1205 1206 /* 1207 * Remove an entry from a directory. 1208 * 1209 * IN: dvp - vnode of directory to remove entry from. 1210 * name - name of entry to remove. 1211 * cr - credentials of caller. 1212 * 1213 * RETURN: 0 if success 1214 * error code if failure 1215 * 1216 * Timestamps: 1217 * dvp - ctime|mtime 1218 * vp - ctime (if nlink > 0) 1219 */ 1220 static int 1221 zfs_remove(vnode_t *dvp, char *name, cred_t *cr) 1222 { 1223 znode_t *zp, *dzp = VTOZ(dvp); 1224 znode_t *xzp = NULL; 1225 vnode_t *vp; 1226 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1227 zilog_t *zilog = zfsvfs->z_log; 1228 uint64_t seq = 0; 1229 uint64_t acl_obj, xattr_obj; 1230 zfs_dirlock_t *dl; 1231 dmu_tx_t *tx; 1232 int may_delete_now, delete_now = FALSE; 1233 int reaped; 1234 int error; 1235 1236 ZFS_ENTER(zfsvfs); 1237 1238 top: 1239 /* 1240 * Attempt to lock directory; fail if entry doesn't exist. 1241 */ 1242 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1243 ZFS_EXIT(zfsvfs); 1244 return (error); 1245 } 1246 1247 vp = ZTOV(zp); 1248 1249 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1250 goto out; 1251 } 1252 1253 /* 1254 * Need to use rmdir for removing directories. 1255 */ 1256 if (vp->v_type == VDIR) { 1257 error = EPERM; 1258 goto out; 1259 } 1260 1261 vnevent_remove(vp); 1262 1263 dnlc_remove(dvp, name); 1264 1265 mutex_enter(&vp->v_lock); 1266 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1267 mutex_exit(&vp->v_lock); 1268 1269 /* 1270 * We may delete the znode now, or we may put it on the delete queue; 1271 * it depends on whether we're the last link, and on whether there are 1272 * other holds on the vnode. So we dmu_tx_hold() the right things to 1273 * allow for either case. 1274 */ 1275 tx = dmu_tx_create(zfsvfs->z_os); 1276 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1277 dmu_tx_hold_bonus(tx, zp->z_id); 1278 if (may_delete_now) 1279 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 1280 1281 /* are there any extended attributes? */ 1282 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1283 /* 1284 * XXX - There is a possibility that the delete 1285 * of the parent file could succeed, but then we get 1286 * an ENOSPC when we try to delete the xattrs... 1287 * so we would need to re-try the deletes periodically 1288 */ 1289 /* XXX - do we need this if we are deleting? */ 1290 dmu_tx_hold_bonus(tx, xattr_obj); 1291 } 1292 1293 /* are there any additional acls */ 1294 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1295 may_delete_now) 1296 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1297 1298 /* charge as an update -- would be nice not to charge at all */ 1299 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 1300 1301 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1302 if (error) { 1303 zfs_dirent_unlock(dl); 1304 VN_RELE(vp); 1305 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1306 dmu_tx_wait(tx); 1307 dmu_tx_abort(tx); 1308 goto top; 1309 } 1310 dmu_tx_abort(tx); 1311 ZFS_EXIT(zfsvfs); 1312 return (error); 1313 } 1314 1315 /* 1316 * Remove the directory entry. 1317 */ 1318 error = zfs_link_destroy(dl, zp, tx, 0, &reaped); 1319 1320 if (error) { 1321 dmu_tx_commit(tx); 1322 goto out; 1323 } 1324 1325 if (reaped) { 1326 mutex_enter(&vp->v_lock); 1327 delete_now = may_delete_now && 1328 vp->v_count == 1 && !vn_has_cached_data(vp) && 1329 zp->z_phys->zp_xattr == xattr_obj && 1330 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1331 mutex_exit(&vp->v_lock); 1332 } 1333 1334 if (delete_now) { 1335 if (zp->z_phys->zp_xattr) { 1336 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1337 ASSERT3U(error, ==, 0); 1338 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1339 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1340 mutex_enter(&xzp->z_lock); 1341 xzp->z_reap = 1; 1342 xzp->z_phys->zp_links = 0; 1343 mutex_exit(&xzp->z_lock); 1344 zfs_dq_add(xzp, tx); 1345 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1346 } 1347 mutex_enter(&zp->z_lock); 1348 mutex_enter(&vp->v_lock); 1349 vp->v_count--; 1350 ASSERT3U(vp->v_count, ==, 0); 1351 mutex_exit(&vp->v_lock); 1352 zp->z_active = 0; 1353 mutex_exit(&zp->z_lock); 1354 zfs_znode_delete(zp, tx); 1355 VFS_RELE(zfsvfs->z_vfs); 1356 } else if (reaped) { 1357 zfs_dq_add(zp, tx); 1358 } 1359 1360 seq = zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name); 1361 1362 dmu_tx_commit(tx); 1363 out: 1364 zfs_dirent_unlock(dl); 1365 1366 if (!delete_now) { 1367 VN_RELE(vp); 1368 } else if (xzp) { 1369 /* this rele delayed to prevent nesting transactions */ 1370 VN_RELE(ZTOV(xzp)); 1371 } 1372 1373 zil_commit(zilog, seq, 0); 1374 1375 ZFS_EXIT(zfsvfs); 1376 return (error); 1377 } 1378 1379 /* 1380 * Create a new directory and insert it into dvp using the name 1381 * provided. Return a pointer to the inserted directory. 1382 * 1383 * IN: dvp - vnode of directory to add subdir to. 1384 * dirname - name of new directory. 1385 * vap - attributes of new directory. 1386 * cr - credentials of caller. 1387 * 1388 * OUT: vpp - vnode of created directory. 1389 * 1390 * RETURN: 0 if success 1391 * error code if failure 1392 * 1393 * Timestamps: 1394 * dvp - ctime|mtime updated 1395 * vp - ctime|mtime|atime updated 1396 */ 1397 static int 1398 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr) 1399 { 1400 znode_t *zp, *dzp = VTOZ(dvp); 1401 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1402 zilog_t *zilog = zfsvfs->z_log; 1403 uint64_t seq = 0; 1404 zfs_dirlock_t *dl; 1405 uint64_t zoid = 0; 1406 dmu_tx_t *tx; 1407 int error; 1408 1409 ASSERT(vap->va_type == VDIR); 1410 1411 ZFS_ENTER(zfsvfs); 1412 1413 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1414 ZFS_EXIT(zfsvfs); 1415 return (EINVAL); 1416 } 1417 top: 1418 *vpp = NULL; 1419 1420 /* 1421 * First make sure the new directory doesn't exist. 1422 */ 1423 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) { 1424 ZFS_EXIT(zfsvfs); 1425 return (error); 1426 } 1427 1428 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) { 1429 zfs_dirent_unlock(dl); 1430 ZFS_EXIT(zfsvfs); 1431 return (error); 1432 } 1433 1434 /* 1435 * Add a new entry to the directory. 1436 */ 1437 tx = dmu_tx_create(zfsvfs->z_os); 1438 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1439 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1440 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1441 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1442 0, SPA_MAXBLOCKSIZE); 1443 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1444 if (error) { 1445 zfs_dirent_unlock(dl); 1446 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1447 dmu_tx_wait(tx); 1448 dmu_tx_abort(tx); 1449 goto top; 1450 } 1451 dmu_tx_abort(tx); 1452 ZFS_EXIT(zfsvfs); 1453 return (error); 1454 } 1455 1456 /* 1457 * Create new node. 1458 */ 1459 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1460 1461 /* 1462 * Now put new name in parent dir. 1463 */ 1464 (void) zfs_link_create(dl, zp, tx, ZNEW); 1465 1466 *vpp = ZTOV(zp); 1467 1468 seq = zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname); 1469 dmu_tx_commit(tx); 1470 1471 zfs_dirent_unlock(dl); 1472 1473 zil_commit(zilog, seq, 0); 1474 1475 ZFS_EXIT(zfsvfs); 1476 return (0); 1477 } 1478 1479 /* 1480 * Remove a directory subdir entry. If the current working 1481 * directory is the same as the subdir to be removed, the 1482 * remove will fail. 1483 * 1484 * IN: dvp - vnode of directory to remove from. 1485 * name - name of directory to be removed. 1486 * cwd - vnode of current working directory. 1487 * cr - credentials of caller. 1488 * 1489 * RETURN: 0 if success 1490 * error code if failure 1491 * 1492 * Timestamps: 1493 * dvp - ctime|mtime updated 1494 */ 1495 static int 1496 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr) 1497 { 1498 znode_t *dzp = VTOZ(dvp); 1499 znode_t *zp; 1500 vnode_t *vp; 1501 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1502 zilog_t *zilog = zfsvfs->z_log; 1503 uint64_t seq = 0; 1504 zfs_dirlock_t *dl; 1505 dmu_tx_t *tx; 1506 int error; 1507 1508 ZFS_ENTER(zfsvfs); 1509 1510 top: 1511 zp = NULL; 1512 1513 /* 1514 * Attempt to lock directory; fail if entry doesn't exist. 1515 */ 1516 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1517 ZFS_EXIT(zfsvfs); 1518 return (error); 1519 } 1520 1521 vp = ZTOV(zp); 1522 1523 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1524 goto out; 1525 } 1526 1527 if (vp->v_type != VDIR) { 1528 error = ENOTDIR; 1529 goto out; 1530 } 1531 1532 if (vp == cwd) { 1533 error = EINVAL; 1534 goto out; 1535 } 1536 1537 vnevent_rmdir(vp); 1538 1539 /* 1540 * Grab a lock on the parent pointer make sure we play well 1541 * with the treewalk and directory rename code. 1542 */ 1543 rw_enter(&zp->z_parent_lock, RW_WRITER); 1544 1545 tx = dmu_tx_create(zfsvfs->z_os); 1546 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1547 dmu_tx_hold_bonus(tx, zp->z_id); 1548 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 1549 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1550 if (error) { 1551 rw_exit(&zp->z_parent_lock); 1552 zfs_dirent_unlock(dl); 1553 VN_RELE(vp); 1554 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1555 dmu_tx_wait(tx); 1556 dmu_tx_abort(tx); 1557 goto top; 1558 } 1559 dmu_tx_abort(tx); 1560 ZFS_EXIT(zfsvfs); 1561 return (error); 1562 } 1563 1564 error = zfs_link_destroy(dl, zp, tx, 0, NULL); 1565 1566 if (error == 0) 1567 seq = zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name); 1568 1569 dmu_tx_commit(tx); 1570 1571 rw_exit(&zp->z_parent_lock); 1572 out: 1573 zfs_dirent_unlock(dl); 1574 1575 VN_RELE(vp); 1576 1577 zil_commit(zilog, seq, 0); 1578 1579 ZFS_EXIT(zfsvfs); 1580 return (error); 1581 } 1582 1583 /* 1584 * Read as many directory entries as will fit into the provided 1585 * buffer from the given directory cursor position (specified in 1586 * the uio structure. 1587 * 1588 * IN: vp - vnode of directory to read. 1589 * uio - structure supplying read location, range info, 1590 * and return buffer. 1591 * cr - credentials of caller. 1592 * 1593 * OUT: uio - updated offset and range, buffer filled. 1594 * eofp - set to true if end-of-file detected. 1595 * 1596 * RETURN: 0 if success 1597 * error code if failure 1598 * 1599 * Timestamps: 1600 * vp - atime updated 1601 * 1602 * Note that the low 4 bits of the cookie returned by zap is always zero. 1603 * This allows us to use the low range for "special" directory entries: 1604 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1605 * we use the offset 2 for the '.zfs' directory. 1606 */ 1607 /* ARGSUSED */ 1608 static int 1609 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp) 1610 { 1611 znode_t *zp = VTOZ(vp); 1612 iovec_t *iovp; 1613 dirent64_t *odp; 1614 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1615 objset_t *os; 1616 caddr_t outbuf; 1617 size_t bufsize; 1618 zap_cursor_t zc; 1619 zap_attribute_t zap; 1620 uint_t bytes_wanted; 1621 ushort_t this_reclen; 1622 uint64_t offset; /* must be unsigned; checks for < 1 */ 1623 off64_t *next; 1624 int local_eof; 1625 int outcount; 1626 int error; 1627 uint8_t prefetch; 1628 1629 ZFS_ENTER(zfsvfs); 1630 1631 /* 1632 * If we are not given an eof variable, 1633 * use a local one. 1634 */ 1635 if (eofp == NULL) 1636 eofp = &local_eof; 1637 1638 /* 1639 * Check for valid iov_len. 1640 */ 1641 if (uio->uio_iov->iov_len <= 0) { 1642 ZFS_EXIT(zfsvfs); 1643 return (EINVAL); 1644 } 1645 1646 /* 1647 * Quit if directory has been removed (posix) 1648 */ 1649 if ((*eofp = zp->z_reap) != 0) { 1650 ZFS_EXIT(zfsvfs); 1651 return (0); 1652 } 1653 1654 error = 0; 1655 os = zfsvfs->z_os; 1656 offset = uio->uio_loffset; 1657 prefetch = zp->z_zn_prefetch; 1658 1659 /* 1660 * Initialize the iterator cursor. 1661 */ 1662 if (offset <= 3) { 1663 /* 1664 * Start iteration from the beginning of the directory. 1665 */ 1666 zap_cursor_init(&zc, os, zp->z_id); 1667 } else { 1668 /* 1669 * The offset is a serialized cursor. 1670 */ 1671 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1672 } 1673 1674 /* 1675 * Get space to change directory entries into fs independent format. 1676 */ 1677 iovp = uio->uio_iov; 1678 bytes_wanted = iovp->iov_len; 1679 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1680 bufsize = bytes_wanted; 1681 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1682 odp = (struct dirent64 *)outbuf; 1683 } else { 1684 bufsize = bytes_wanted; 1685 odp = (struct dirent64 *)iovp->iov_base; 1686 } 1687 1688 /* 1689 * Transform to file-system independent format 1690 */ 1691 outcount = 0; 1692 while (outcount < bytes_wanted) { 1693 /* 1694 * Special case `.', `..', and `.zfs'. 1695 */ 1696 if (offset == 0) { 1697 (void) strcpy(zap.za_name, "."); 1698 zap.za_first_integer = zp->z_id; 1699 this_reclen = DIRENT64_RECLEN(1); 1700 } else if (offset == 1) { 1701 (void) strcpy(zap.za_name, ".."); 1702 zap.za_first_integer = zp->z_phys->zp_parent; 1703 this_reclen = DIRENT64_RECLEN(2); 1704 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1705 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1706 zap.za_first_integer = ZFSCTL_INO_ROOT; 1707 this_reclen = 1708 DIRENT64_RECLEN(sizeof (ZFS_CTLDIR_NAME) - 1); 1709 } else { 1710 /* 1711 * Grab next entry. 1712 */ 1713 if (error = zap_cursor_retrieve(&zc, &zap)) { 1714 if ((*eofp = (error == ENOENT)) != 0) 1715 break; 1716 else 1717 goto update; 1718 } 1719 1720 if (zap.za_integer_length != 8 || 1721 zap.za_num_integers != 1) { 1722 cmn_err(CE_WARN, "zap_readdir: bad directory " 1723 "entry, obj = %lld, offset = %lld\n", 1724 (u_longlong_t)zp->z_id, 1725 (u_longlong_t)offset); 1726 error = ENXIO; 1727 goto update; 1728 } 1729 this_reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1730 } 1731 1732 /* 1733 * Will this entry fit in the buffer? 1734 */ 1735 if (outcount + this_reclen > bufsize) { 1736 /* 1737 * Did we manage to fit anything in the buffer? 1738 */ 1739 if (!outcount) { 1740 error = EINVAL; 1741 goto update; 1742 } 1743 break; 1744 } 1745 /* 1746 * Add this entry: 1747 */ 1748 odp->d_ino = (ino64_t)zap.za_first_integer; 1749 odp->d_reclen = (ushort_t)this_reclen; 1750 /* NOTE: d_off is the offset for the *next* entry */ 1751 next = &(odp->d_off); 1752 (void) strncpy(odp->d_name, zap.za_name, 1753 DIRENT64_NAMELEN(this_reclen)); 1754 outcount += this_reclen; 1755 odp = (dirent64_t *)((intptr_t)odp + this_reclen); 1756 1757 ASSERT(outcount <= bufsize); 1758 1759 /* Prefetch znode */ 1760 if (prefetch) 1761 dmu_prefetch(os, zap.za_first_integer, 0, 0); 1762 1763 /* 1764 * Move to the next entry, fill in the previous offset. 1765 */ 1766 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1767 zap_cursor_advance(&zc); 1768 offset = zap_cursor_serialize(&zc); 1769 } else { 1770 offset += 1; 1771 } 1772 *next = offset; 1773 } 1774 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1775 1776 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 1777 iovp->iov_base += outcount; 1778 iovp->iov_len -= outcount; 1779 uio->uio_resid -= outcount; 1780 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 1781 /* 1782 * Reset the pointer. 1783 */ 1784 offset = uio->uio_loffset; 1785 } 1786 1787 update: 1788 zap_cursor_fini(&zc); 1789 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 1790 kmem_free(outbuf, bufsize); 1791 1792 if (error == ENOENT) 1793 error = 0; 1794 1795 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1796 1797 uio->uio_loffset = offset; 1798 ZFS_EXIT(zfsvfs); 1799 return (error); 1800 } 1801 1802 static int 1803 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr) 1804 { 1805 znode_t *zp = VTOZ(vp); 1806 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1807 1808 /* 1809 * Regardless of whether this is required for standards conformance, 1810 * this is the logical behavior when fsync() is called on a file with 1811 * dirty pages. We use B_ASYNC since the ZIL transactions are already 1812 * going to be pushed out as part of the zil_commit(). 1813 */ 1814 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 1815 (vp->v_type == VREG) && !(IS_SWAPVP(vp))) 1816 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr); 1817 1818 ZFS_ENTER(zfsvfs); 1819 zil_commit(zfsvfs->z_log, zp->z_last_itx, FSYNC); 1820 ZFS_EXIT(zfsvfs); 1821 return (0); 1822 } 1823 1824 /* 1825 * Get the requested file attributes and place them in the provided 1826 * vattr structure. 1827 * 1828 * IN: vp - vnode of file. 1829 * vap - va_mask identifies requested attributes. 1830 * flags - [UNUSED] 1831 * cr - credentials of caller. 1832 * 1833 * OUT: vap - attribute values. 1834 * 1835 * RETURN: 0 (always succeeds) 1836 */ 1837 /* ARGSUSED */ 1838 static int 1839 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1840 { 1841 znode_t *zp = VTOZ(vp); 1842 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1843 znode_phys_t *pzp = zp->z_phys; 1844 int error; 1845 1846 ZFS_ENTER(zfsvfs); 1847 1848 /* 1849 * Return all attributes. It's cheaper to provide the answer 1850 * than to determine whether we were asked the question. 1851 */ 1852 mutex_enter(&zp->z_lock); 1853 1854 vap->va_type = vp->v_type; 1855 vap->va_mode = pzp->zp_mode & MODEMASK; 1856 vap->va_uid = zp->z_phys->zp_uid; 1857 vap->va_gid = zp->z_phys->zp_gid; 1858 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 1859 vap->va_nodeid = zp->z_id; 1860 vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX); /* nlink_t limit! */ 1861 vap->va_size = pzp->zp_size; 1862 vap->va_rdev = vp->v_rdev; 1863 vap->va_seq = zp->z_seq; 1864 1865 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 1866 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 1867 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 1868 1869 /* 1870 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 1871 * Also, if we are the owner don't bother, since owner should 1872 * always be allowed to read basic attributes of file. 1873 */ 1874 if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) && 1875 (zp->z_phys->zp_uid != crgetuid(cr))) { 1876 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) { 1877 mutex_exit(&zp->z_lock); 1878 ZFS_EXIT(zfsvfs); 1879 return (error); 1880 } 1881 } 1882 1883 mutex_exit(&zp->z_lock); 1884 1885 dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks); 1886 1887 if (zp->z_blksz == 0) { 1888 /* 1889 * Block size hasn't been set; suggest maximal I/O transfers. 1890 */ 1891 vap->va_blksize = zfsvfs->z_max_blksz; 1892 } 1893 1894 ZFS_EXIT(zfsvfs); 1895 return (0); 1896 } 1897 1898 /* 1899 * Set the file attributes to the values contained in the 1900 * vattr structure. 1901 * 1902 * IN: vp - vnode of file to be modified. 1903 * vap - new attribute values. 1904 * flags - ATTR_UTIME set if non-default time values provided. 1905 * cr - credentials of caller. 1906 * 1907 * RETURN: 0 if success 1908 * error code if failure 1909 * 1910 * Timestamps: 1911 * vp - ctime updated, mtime updated if size changed. 1912 */ 1913 /* ARGSUSED */ 1914 static int 1915 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 1916 caller_context_t *ct) 1917 { 1918 struct znode *zp = VTOZ(vp); 1919 znode_phys_t *pzp = zp->z_phys; 1920 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1921 zilog_t *zilog = zfsvfs->z_log; 1922 uint64_t seq = 0; 1923 dmu_tx_t *tx; 1924 vattr_t oldva; 1925 uint_t mask = vap->va_mask; 1926 uint_t saved_mask; 1927 int trim_mask = FALSE; 1928 uint64_t new_mode; 1929 znode_t *attrzp; 1930 int need_policy = FALSE; 1931 int err; 1932 1933 if (mask == 0) 1934 return (0); 1935 1936 if (mask & AT_NOSET) 1937 return (EINVAL); 1938 1939 if (mask & AT_SIZE && vp->v_type == VDIR) 1940 return (EISDIR); 1941 1942 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) 1943 return (EINVAL); 1944 1945 ZFS_ENTER(zfsvfs); 1946 1947 top: 1948 attrzp = NULL; 1949 1950 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 1951 ZFS_EXIT(zfsvfs); 1952 return (EROFS); 1953 } 1954 1955 /* 1956 * First validate permissions 1957 */ 1958 1959 if (mask & AT_SIZE) { 1960 err = zfs_zaccess(zp, ACE_WRITE_DATA, cr); 1961 if (err) { 1962 ZFS_EXIT(zfsvfs); 1963 return (err); 1964 } 1965 /* 1966 * XXX - Note, we are not providing any open 1967 * mode flags here (like FNDELAY), so we may 1968 * block if there are locks present... this 1969 * should be addressed in openat(). 1970 */ 1971 do { 1972 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 1973 /* NB: we already did dmu_tx_wait() if necessary */ 1974 } while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 1975 if (err) { 1976 ZFS_EXIT(zfsvfs); 1977 return (err); 1978 } 1979 } 1980 1981 if (mask & (AT_ATIME|AT_MTIME)) 1982 need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr); 1983 1984 if (mask & (AT_UID|AT_GID)) { 1985 int idmask = (mask & (AT_UID|AT_GID)); 1986 int take_owner; 1987 int take_group; 1988 1989 /* 1990 * NOTE: even if a new mode is being set, 1991 * we may clear S_ISUID/S_ISGID bits. 1992 */ 1993 1994 if (!(mask & AT_MODE)) 1995 vap->va_mode = pzp->zp_mode; 1996 1997 /* 1998 * Take ownership or chgrp to group we are a member of 1999 */ 2000 2001 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2002 take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr); 2003 2004 /* 2005 * If both AT_UID and AT_GID are set then take_owner and 2006 * take_group must both be set in order to allow taking 2007 * ownership. 2008 * 2009 * Otherwise, send the check through secpolicy_vnode_setattr() 2010 * 2011 */ 2012 2013 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2014 ((idmask == AT_UID) && take_owner) || 2015 ((idmask == AT_GID) && take_group)) { 2016 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) { 2017 /* 2018 * Remove setuid/setgid for non-privileged users 2019 */ 2020 secpolicy_setid_clear(vap, cr); 2021 trim_mask = TRUE; 2022 saved_mask = vap->va_mask; 2023 } else { 2024 need_policy = TRUE; 2025 } 2026 } else { 2027 need_policy = TRUE; 2028 } 2029 } 2030 2031 if (mask & AT_MODE) 2032 need_policy = TRUE; 2033 2034 if (need_policy) { 2035 mutex_enter(&zp->z_lock); 2036 oldva.va_mode = pzp->zp_mode; 2037 oldva.va_uid = zp->z_phys->zp_uid; 2038 oldva.va_gid = zp->z_phys->zp_gid; 2039 mutex_exit(&zp->z_lock); 2040 2041 /* 2042 * If trim_mask is set then take ownership 2043 * has been granted. In that case remove 2044 * UID|GID from mask so that 2045 * secpolicy_vnode_setattr() doesn't revoke it. 2046 */ 2047 if (trim_mask) 2048 vap->va_mask &= ~(AT_UID|AT_GID); 2049 2050 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2051 (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp); 2052 if (err) { 2053 ZFS_EXIT(zfsvfs); 2054 return (err); 2055 } 2056 2057 if (trim_mask) 2058 vap->va_mask |= (saved_mask & (AT_UID|AT_GID)); 2059 } 2060 2061 /* 2062 * secpolicy_vnode_setattr, or take ownership may have 2063 * changed va_mask 2064 */ 2065 mask = vap->va_mask; 2066 2067 tx = dmu_tx_create(zfsvfs->z_os); 2068 dmu_tx_hold_bonus(tx, zp->z_id); 2069 2070 if (mask & AT_MODE) { 2071 uint64_t pmode = pzp->zp_mode; 2072 2073 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2074 2075 if (zp->z_phys->zp_acl.z_acl_extern_obj) 2076 dmu_tx_hold_write(tx, 2077 pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE); 2078 else 2079 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2080 0, ZFS_ACL_SIZE(MAX_ACL_SIZE)); 2081 } 2082 2083 if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) { 2084 err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp); 2085 if (err) { 2086 dmu_tx_abort(tx); 2087 ZFS_EXIT(zfsvfs); 2088 return (err); 2089 } 2090 dmu_tx_hold_bonus(tx, attrzp->z_id); 2091 } 2092 2093 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2094 if (err) { 2095 if (attrzp) 2096 VN_RELE(ZTOV(attrzp)); 2097 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2098 dmu_tx_wait(tx); 2099 dmu_tx_abort(tx); 2100 goto top; 2101 } 2102 dmu_tx_abort(tx); 2103 ZFS_EXIT(zfsvfs); 2104 return (err); 2105 } 2106 2107 dmu_buf_will_dirty(zp->z_dbuf, tx); 2108 2109 /* 2110 * Set each attribute requested. 2111 * We group settings according to the locks they need to acquire. 2112 * 2113 * Note: you cannot set ctime directly, although it will be 2114 * updated as a side-effect of calling this function. 2115 */ 2116 2117 mutex_enter(&zp->z_lock); 2118 2119 if (mask & AT_MODE) { 2120 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2121 ASSERT3U(err, ==, 0); 2122 } 2123 2124 if (attrzp) 2125 mutex_enter(&attrzp->z_lock); 2126 2127 if (mask & AT_UID) { 2128 zp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2129 if (attrzp) { 2130 attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2131 } 2132 } 2133 2134 if (mask & AT_GID) { 2135 zp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2136 if (attrzp) 2137 attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2138 } 2139 2140 if (attrzp) 2141 mutex_exit(&attrzp->z_lock); 2142 2143 if (mask & AT_ATIME) 2144 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2145 2146 if (mask & AT_MTIME) 2147 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2148 2149 if (mask & AT_SIZE) 2150 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2151 else if (mask != 0) 2152 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2153 2154 if (mask != 0) 2155 seq = zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask); 2156 2157 mutex_exit(&zp->z_lock); 2158 2159 if (attrzp) 2160 VN_RELE(ZTOV(attrzp)); 2161 2162 dmu_tx_commit(tx); 2163 2164 zil_commit(zilog, seq, 0); 2165 2166 ZFS_EXIT(zfsvfs); 2167 return (err); 2168 } 2169 2170 /* 2171 * Search back through the directory tree, using the ".." entries. 2172 * Lock each directory in the chain to prevent concurrent renames. 2173 * Fail any attempt to move a directory into one of its own descendants. 2174 * XXX - z_parent_lock can overlap with map or grow locks 2175 */ 2176 typedef struct zfs_zlock { 2177 krwlock_t *zl_rwlock; /* lock we acquired */ 2178 znode_t *zl_znode; /* znode we held */ 2179 struct zfs_zlock *zl_next; /* next in list */ 2180 } zfs_zlock_t; 2181 2182 static int 2183 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2184 { 2185 zfs_zlock_t *zl; 2186 znode_t *zp = tdzp; 2187 uint64_t rootid = zp->z_zfsvfs->z_root; 2188 uint64_t *oidp = &zp->z_id; 2189 krwlock_t *rwlp = &szp->z_parent_lock; 2190 krw_t rw = RW_WRITER; 2191 2192 /* 2193 * First pass write-locks szp and compares to zp->z_id. 2194 * Later passes read-lock zp and compare to zp->z_parent. 2195 */ 2196 do { 2197 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2198 zl->zl_rwlock = rwlp; 2199 zl->zl_znode = NULL; 2200 zl->zl_next = *zlpp; 2201 *zlpp = zl; 2202 2203 rw_enter(rwlp, rw); 2204 2205 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2206 return (EINVAL); 2207 2208 if (*oidp == rootid) /* We've hit the top */ 2209 return (0); 2210 2211 if (rw == RW_READER) { /* i.e. not the first pass */ 2212 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2213 if (error) 2214 return (error); 2215 zl->zl_znode = zp; 2216 } 2217 oidp = &zp->z_phys->zp_parent; 2218 rwlp = &zp->z_parent_lock; 2219 rw = RW_READER; 2220 2221 } while (zp->z_id != sdzp->z_id); 2222 2223 return (0); 2224 } 2225 2226 /* 2227 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2228 */ 2229 static void 2230 zfs_rename_unlock(zfs_zlock_t **zlpp) 2231 { 2232 zfs_zlock_t *zl; 2233 2234 while ((zl = *zlpp) != NULL) { 2235 if (zl->zl_znode != NULL) 2236 VN_RELE(ZTOV(zl->zl_znode)); 2237 rw_exit(zl->zl_rwlock); 2238 *zlpp = zl->zl_next; 2239 kmem_free(zl, sizeof (*zl)); 2240 } 2241 } 2242 2243 /* 2244 * Move an entry from the provided source directory to the target 2245 * directory. Change the entry name as indicated. 2246 * 2247 * IN: sdvp - Source directory containing the "old entry". 2248 * snm - Old entry name. 2249 * tdvp - Target directory to contain the "new entry". 2250 * tnm - New entry name. 2251 * cr - credentials of caller. 2252 * 2253 * RETURN: 0 if success 2254 * error code if failure 2255 * 2256 * Timestamps: 2257 * sdvp,tdvp - ctime|mtime updated 2258 */ 2259 static int 2260 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr) 2261 { 2262 znode_t *tdzp, *szp, *tzp; 2263 znode_t *sdzp = VTOZ(sdvp); 2264 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2265 zilog_t *zilog = zfsvfs->z_log; 2266 uint64_t seq = 0; 2267 vnode_t *realvp; 2268 zfs_dirlock_t *sdl, *tdl; 2269 dmu_tx_t *tx; 2270 zfs_zlock_t *zl; 2271 int cmp, serr, terr, error; 2272 2273 ZFS_ENTER(zfsvfs); 2274 2275 /* 2276 * Make sure we have the real vp for the target directory. 2277 */ 2278 if (VOP_REALVP(tdvp, &realvp) == 0) 2279 tdvp = realvp; 2280 2281 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2282 ZFS_EXIT(zfsvfs); 2283 return (EXDEV); 2284 } 2285 2286 tdzp = VTOZ(tdvp); 2287 top: 2288 szp = NULL; 2289 tzp = NULL; 2290 zl = NULL; 2291 2292 /* 2293 * This is to prevent the creation of links into attribute space 2294 * by renaming a linked file into/outof an attribute directory. 2295 * See the comment in zfs_link() for why this is considered bad. 2296 */ 2297 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2298 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2299 ZFS_EXIT(zfsvfs); 2300 return (EINVAL); 2301 } 2302 2303 /* 2304 * Lock source and target directory entries. To prevent deadlock, 2305 * a lock ordering must be defined. We lock the directory with 2306 * the smallest object id first, or if it's a tie, the one with 2307 * the lexically first name. 2308 */ 2309 if (sdzp->z_id < tdzp->z_id) { 2310 cmp = -1; 2311 } else if (sdzp->z_id > tdzp->z_id) { 2312 cmp = 1; 2313 } else { 2314 cmp = strcmp(snm, tnm); 2315 if (cmp == 0) { 2316 /* 2317 * POSIX: "If the old argument and the new argument 2318 * both refer to links to the same existing file, 2319 * the rename() function shall return successfully 2320 * and perform no other action." 2321 */ 2322 ZFS_EXIT(zfsvfs); 2323 return (0); 2324 } 2325 } 2326 if (cmp < 0) { 2327 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2328 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2329 } else { 2330 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2331 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2332 } 2333 2334 if (serr) { 2335 /* 2336 * Source entry invalid or not there. 2337 */ 2338 if (!terr) { 2339 zfs_dirent_unlock(tdl); 2340 if (tzp) 2341 VN_RELE(ZTOV(tzp)); 2342 } 2343 if (strcmp(snm, "..") == 0) 2344 serr = EINVAL; 2345 ZFS_EXIT(zfsvfs); 2346 return (serr); 2347 } 2348 if (terr) { 2349 zfs_dirent_unlock(sdl); 2350 VN_RELE(ZTOV(szp)); 2351 if (strcmp(tnm, "..") == 0) 2352 terr = EINVAL; 2353 ZFS_EXIT(zfsvfs); 2354 return (terr); 2355 } 2356 2357 /* 2358 * Must have write access at the source to remove the old entry 2359 * and write access at the target to create the new entry. 2360 * Note that if target and source are the same, this can be 2361 * done in a single check. 2362 */ 2363 2364 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2365 goto out; 2366 2367 if (ZTOV(szp)->v_type == VDIR) { 2368 /* 2369 * Check to make sure rename is valid. 2370 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2371 */ 2372 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 2373 goto out; 2374 } 2375 2376 /* 2377 * Does target exist? 2378 */ 2379 if (tzp) { 2380 /* 2381 * Source and target must be the same type. 2382 */ 2383 if (ZTOV(szp)->v_type == VDIR) { 2384 if (ZTOV(tzp)->v_type != VDIR) { 2385 error = ENOTDIR; 2386 goto out; 2387 } 2388 } else { 2389 if (ZTOV(tzp)->v_type == VDIR) { 2390 error = EISDIR; 2391 goto out; 2392 } 2393 } 2394 /* 2395 * POSIX dictates that when the source and target 2396 * entries refer to the same file object, rename 2397 * must do nothing and exit without error. 2398 */ 2399 if (szp->z_id == tzp->z_id) { 2400 error = 0; 2401 goto out; 2402 } 2403 } 2404 2405 vnevent_rename_src(ZTOV(szp)); 2406 if (tzp) 2407 vnevent_rename_dest(ZTOV(tzp)); 2408 2409 tx = dmu_tx_create(zfsvfs->z_os); 2410 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 2411 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 2412 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 2413 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 2414 if (sdzp != tdzp) 2415 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 2416 if (tzp) 2417 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 2418 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 2419 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2420 if (error) { 2421 if (zl != NULL) 2422 zfs_rename_unlock(&zl); 2423 zfs_dirent_unlock(sdl); 2424 zfs_dirent_unlock(tdl); 2425 VN_RELE(ZTOV(szp)); 2426 if (tzp) 2427 VN_RELE(ZTOV(tzp)); 2428 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2429 dmu_tx_wait(tx); 2430 dmu_tx_abort(tx); 2431 goto top; 2432 } 2433 dmu_tx_abort(tx); 2434 ZFS_EXIT(zfsvfs); 2435 return (error); 2436 } 2437 2438 if (tzp) /* Attempt to remove the existing target */ 2439 error = zfs_link_destroy(tdl, tzp, tx, 0, NULL); 2440 2441 if (error == 0) { 2442 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 2443 if (error == 0) { 2444 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 2445 ASSERT(error == 0); 2446 seq = zfs_log_rename(zilog, tx, TX_RENAME, 2447 sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp); 2448 } 2449 } 2450 2451 dmu_tx_commit(tx); 2452 out: 2453 if (zl != NULL) 2454 zfs_rename_unlock(&zl); 2455 2456 zfs_dirent_unlock(sdl); 2457 zfs_dirent_unlock(tdl); 2458 2459 VN_RELE(ZTOV(szp)); 2460 if (tzp) 2461 VN_RELE(ZTOV(tzp)); 2462 2463 zil_commit(zilog, seq, 0); 2464 2465 ZFS_EXIT(zfsvfs); 2466 return (error); 2467 } 2468 2469 /* 2470 * Insert the indicated symbolic reference entry into the directory. 2471 * 2472 * IN: dvp - Directory to contain new symbolic link. 2473 * link - Name for new symlink entry. 2474 * vap - Attributes of new entry. 2475 * target - Target path of new symlink. 2476 * cr - credentials of caller. 2477 * 2478 * RETURN: 0 if success 2479 * error code if failure 2480 * 2481 * Timestamps: 2482 * dvp - ctime|mtime updated 2483 */ 2484 static int 2485 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr) 2486 { 2487 znode_t *zp, *dzp = VTOZ(dvp); 2488 zfs_dirlock_t *dl; 2489 dmu_tx_t *tx; 2490 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2491 zilog_t *zilog = zfsvfs->z_log; 2492 uint64_t seq = 0; 2493 uint64_t zoid; 2494 int len = strlen(link); 2495 int error; 2496 2497 ASSERT(vap->va_type == VLNK); 2498 2499 ZFS_ENTER(zfsvfs); 2500 top: 2501 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2502 ZFS_EXIT(zfsvfs); 2503 return (error); 2504 } 2505 2506 if (len > MAXPATHLEN) { 2507 ZFS_EXIT(zfsvfs); 2508 return (ENAMETOOLONG); 2509 } 2510 2511 /* 2512 * Attempt to lock directory; fail if entry already exists. 2513 */ 2514 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) { 2515 ZFS_EXIT(zfsvfs); 2516 return (error); 2517 } 2518 2519 tx = dmu_tx_create(zfsvfs->z_os); 2520 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 2521 dmu_tx_hold_bonus(tx, dzp->z_id); 2522 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2523 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 2524 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 2525 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2526 if (error) { 2527 zfs_dirent_unlock(dl); 2528 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2529 dmu_tx_wait(tx); 2530 dmu_tx_abort(tx); 2531 goto top; 2532 } 2533 dmu_tx_abort(tx); 2534 ZFS_EXIT(zfsvfs); 2535 return (error); 2536 } 2537 2538 dmu_buf_will_dirty(dzp->z_dbuf, tx); 2539 2540 /* 2541 * Create a new object for the symlink. 2542 * Put the link content into bonus buffer if it will fit; 2543 * otherwise, store it just like any other file data. 2544 */ 2545 zoid = 0; 2546 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 2547 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len); 2548 if (len != 0) 2549 bcopy(link, zp->z_phys + 1, len); 2550 } else { 2551 dmu_buf_t *dbp; 2552 2553 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 2554 2555 /* 2556 * Nothing can access the znode yet so no locking needed 2557 * for growing the znode's blocksize. 2558 */ 2559 zfs_grow_blocksize(zp, len, tx); 2560 2561 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp)); 2562 dmu_buf_will_dirty(dbp, tx); 2563 2564 ASSERT3U(len, <=, dbp->db_size); 2565 bcopy(link, dbp->db_data, len); 2566 dmu_buf_rele(dbp, FTAG); 2567 } 2568 zp->z_phys->zp_size = len; 2569 2570 /* 2571 * Insert the new object into the directory. 2572 */ 2573 (void) zfs_link_create(dl, zp, tx, ZNEW); 2574 out: 2575 if (error == 0) 2576 seq = zfs_log_symlink(zilog, tx, TX_SYMLINK, 2577 dzp, zp, name, link); 2578 2579 dmu_tx_commit(tx); 2580 2581 zfs_dirent_unlock(dl); 2582 2583 VN_RELE(ZTOV(zp)); 2584 2585 zil_commit(zilog, seq, 0); 2586 2587 ZFS_EXIT(zfsvfs); 2588 return (error); 2589 } 2590 2591 /* 2592 * Return, in the buffer contained in the provided uio structure, 2593 * the symbolic path referred to by vp. 2594 * 2595 * IN: vp - vnode of symbolic link. 2596 * uoip - structure to contain the link path. 2597 * cr - credentials of caller. 2598 * 2599 * OUT: uio - structure to contain the link path. 2600 * 2601 * RETURN: 0 if success 2602 * error code if failure 2603 * 2604 * Timestamps: 2605 * vp - atime updated 2606 */ 2607 /* ARGSUSED */ 2608 static int 2609 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr) 2610 { 2611 znode_t *zp = VTOZ(vp); 2612 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2613 size_t bufsz; 2614 int error; 2615 2616 ZFS_ENTER(zfsvfs); 2617 2618 bufsz = (size_t)zp->z_phys->zp_size; 2619 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 2620 error = uiomove(zp->z_phys + 1, 2621 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2622 } else { 2623 dmu_buf_t *dbp; 2624 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 2625 if (error) { 2626 ZFS_EXIT(zfsvfs); 2627 return (error); 2628 } 2629 error = uiomove(dbp->db_data, 2630 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2631 dmu_buf_rele(dbp, FTAG); 2632 } 2633 2634 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2635 ZFS_EXIT(zfsvfs); 2636 return (error); 2637 } 2638 2639 /* 2640 * Insert a new entry into directory tdvp referencing svp. 2641 * 2642 * IN: tdvp - Directory to contain new entry. 2643 * svp - vnode of new entry. 2644 * name - name of new entry. 2645 * cr - credentials of caller. 2646 * 2647 * RETURN: 0 if success 2648 * error code if failure 2649 * 2650 * Timestamps: 2651 * tdvp - ctime|mtime updated 2652 * svp - ctime updated 2653 */ 2654 /* ARGSUSED */ 2655 static int 2656 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr) 2657 { 2658 znode_t *dzp = VTOZ(tdvp); 2659 znode_t *tzp, *szp; 2660 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2661 zilog_t *zilog = zfsvfs->z_log; 2662 uint64_t seq = 0; 2663 zfs_dirlock_t *dl; 2664 dmu_tx_t *tx; 2665 vnode_t *realvp; 2666 int error; 2667 2668 ASSERT(tdvp->v_type == VDIR); 2669 2670 ZFS_ENTER(zfsvfs); 2671 2672 if (VOP_REALVP(svp, &realvp) == 0) 2673 svp = realvp; 2674 2675 if (svp->v_vfsp != tdvp->v_vfsp) { 2676 ZFS_EXIT(zfsvfs); 2677 return (EXDEV); 2678 } 2679 2680 szp = VTOZ(svp); 2681 top: 2682 /* 2683 * We do not support links between attributes and non-attributes 2684 * because of the potential security risk of creating links 2685 * into "normal" file space in order to circumvent restrictions 2686 * imposed in attribute space. 2687 */ 2688 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 2689 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 2690 ZFS_EXIT(zfsvfs); 2691 return (EINVAL); 2692 } 2693 2694 /* 2695 * POSIX dictates that we return EPERM here. 2696 * Better choices include ENOTSUP or EISDIR. 2697 */ 2698 if (svp->v_type == VDIR) { 2699 ZFS_EXIT(zfsvfs); 2700 return (EPERM); 2701 } 2702 2703 if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) && 2704 secpolicy_basic_link(cr) != 0) { 2705 ZFS_EXIT(zfsvfs); 2706 return (EPERM); 2707 } 2708 2709 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2710 ZFS_EXIT(zfsvfs); 2711 return (error); 2712 } 2713 2714 /* 2715 * Attempt to lock directory; fail if entry already exists. 2716 */ 2717 if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) { 2718 ZFS_EXIT(zfsvfs); 2719 return (error); 2720 } 2721 2722 tx = dmu_tx_create(zfsvfs->z_os); 2723 dmu_tx_hold_bonus(tx, szp->z_id); 2724 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2725 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2726 if (error) { 2727 zfs_dirent_unlock(dl); 2728 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2729 dmu_tx_wait(tx); 2730 dmu_tx_abort(tx); 2731 goto top; 2732 } 2733 dmu_tx_abort(tx); 2734 ZFS_EXIT(zfsvfs); 2735 return (error); 2736 } 2737 2738 error = zfs_link_create(dl, szp, tx, 0); 2739 2740 if (error == 0) 2741 seq = zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name); 2742 2743 dmu_tx_commit(tx); 2744 2745 zfs_dirent_unlock(dl); 2746 2747 zil_commit(zilog, seq, 0); 2748 2749 ZFS_EXIT(zfsvfs); 2750 return (error); 2751 } 2752 2753 /* 2754 * zfs_null_putapage() is used when the file system has been force 2755 * unmounted. It just drops the pages. 2756 */ 2757 /* ARGSUSED */ 2758 static int 2759 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2760 size_t *lenp, int flags, cred_t *cr) 2761 { 2762 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 2763 return (0); 2764 } 2765 2766 /* ARGSUSED */ 2767 static int 2768 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2769 size_t *lenp, int flags, cred_t *cr) 2770 { 2771 znode_t *zp = VTOZ(vp); 2772 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2773 zilog_t *zilog = zfsvfs->z_log; 2774 dmu_tx_t *tx; 2775 rl_t *rl; 2776 u_offset_t off; 2777 ssize_t len; 2778 caddr_t va; 2779 int err; 2780 2781 top: 2782 off = pp->p_offset; 2783 rl = zfs_range_lock(zp, off, PAGESIZE, RL_WRITER); 2784 /* 2785 * Can't push pages past end-of-file. 2786 */ 2787 if (off >= zp->z_phys->zp_size) { 2788 zfs_range_unlock(zp, rl); 2789 return (EIO); 2790 } 2791 len = MIN(PAGESIZE, zp->z_phys->zp_size - off); 2792 2793 tx = dmu_tx_create(zfsvfs->z_os); 2794 dmu_tx_hold_write(tx, zp->z_id, off, len); 2795 dmu_tx_hold_bonus(tx, zp->z_id); 2796 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2797 if (err != 0) { 2798 zfs_range_unlock(zp, rl); 2799 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2800 dmu_tx_wait(tx); 2801 dmu_tx_abort(tx); 2802 goto top; 2803 } 2804 dmu_tx_abort(tx); 2805 goto out; 2806 } 2807 2808 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 2809 2810 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 2811 2812 ppmapout(va); 2813 2814 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 2815 (void) zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0, NULL); 2816 dmu_tx_commit(tx); 2817 2818 zfs_range_unlock(zp, rl); 2819 2820 pvn_write_done(pp, B_WRITE | flags); 2821 if (offp) 2822 *offp = off; 2823 if (lenp) 2824 *lenp = len; 2825 2826 out: 2827 return (err); 2828 } 2829 2830 /* 2831 * Copy the portion of the file indicated from pages into the file. 2832 * The pages are stored in a page list attached to the files vnode. 2833 * 2834 * IN: vp - vnode of file to push page data to. 2835 * off - position in file to put data. 2836 * len - amount of data to write. 2837 * flags - flags to control the operation. 2838 * cr - credentials of caller. 2839 * 2840 * RETURN: 0 if success 2841 * error code if failure 2842 * 2843 * Timestamps: 2844 * vp - ctime|mtime updated 2845 */ 2846 static int 2847 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr) 2848 { 2849 znode_t *zp = VTOZ(vp); 2850 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2851 page_t *pp; 2852 size_t io_len; 2853 u_offset_t io_off; 2854 uint64_t filesz; 2855 int error = 0; 2856 2857 ZFS_ENTER(zfsvfs); 2858 2859 ASSERT(zp->z_dbuf_held && zp->z_phys); 2860 2861 if (len == 0) { 2862 /* 2863 * Search the entire vp list for pages >= off. 2864 */ 2865 error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage, 2866 flags, cr); 2867 goto out; 2868 } 2869 2870 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 2871 if (off > filesz) { 2872 /* past end of file */ 2873 ZFS_EXIT(zfsvfs); 2874 return (0); 2875 } 2876 2877 len = MIN(len, filesz - off); 2878 2879 for (io_off = off; io_off < off + len; io_off += io_len) { 2880 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 2881 pp = page_lookup(vp, io_off, 2882 (flags & (B_INVAL | B_FREE)) ? 2883 SE_EXCL : SE_SHARED); 2884 } else { 2885 pp = page_lookup_nowait(vp, io_off, 2886 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 2887 } 2888 2889 if (pp != NULL && pvn_getdirty(pp, flags)) { 2890 int err; 2891 2892 /* 2893 * Found a dirty page to push 2894 */ 2895 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 2896 if (err) 2897 error = err; 2898 } else { 2899 io_len = PAGESIZE; 2900 } 2901 } 2902 out: 2903 zil_commit(zfsvfs->z_log, UINT64_MAX, (flags & B_ASYNC) ? 0 : FDSYNC); 2904 ZFS_EXIT(zfsvfs); 2905 return (error); 2906 } 2907 2908 void 2909 zfs_inactive(vnode_t *vp, cred_t *cr) 2910 { 2911 znode_t *zp = VTOZ(vp); 2912 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2913 int error; 2914 2915 rw_enter(&zfsvfs->z_um_lock, RW_READER); 2916 if (zfsvfs->z_unmounted2) { 2917 ASSERT(zp->z_dbuf_held == 0); 2918 2919 if (vn_has_cached_data(vp)) { 2920 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 2921 B_INVAL, cr); 2922 } 2923 2924 mutex_enter(&zp->z_lock); 2925 vp->v_count = 0; /* count arrives as 1 */ 2926 if (zp->z_dbuf == NULL) { 2927 mutex_exit(&zp->z_lock); 2928 zfs_znode_free(zp); 2929 } else { 2930 mutex_exit(&zp->z_lock); 2931 } 2932 rw_exit(&zfsvfs->z_um_lock); 2933 VFS_RELE(zfsvfs->z_vfs); 2934 return; 2935 } 2936 2937 /* 2938 * Attempt to push any data in the page cache. If this fails 2939 * we will get kicked out later in zfs_zinactive(). 2940 */ 2941 if (vn_has_cached_data(vp)) { 2942 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 2943 cr); 2944 } 2945 2946 if (zp->z_atime_dirty && zp->z_reap == 0) { 2947 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 2948 2949 dmu_tx_hold_bonus(tx, zp->z_id); 2950 error = dmu_tx_assign(tx, TXG_WAIT); 2951 if (error) { 2952 dmu_tx_abort(tx); 2953 } else { 2954 dmu_buf_will_dirty(zp->z_dbuf, tx); 2955 mutex_enter(&zp->z_lock); 2956 zp->z_atime_dirty = 0; 2957 mutex_exit(&zp->z_lock); 2958 dmu_tx_commit(tx); 2959 } 2960 } 2961 2962 zfs_zinactive(zp); 2963 rw_exit(&zfsvfs->z_um_lock); 2964 } 2965 2966 /* 2967 * Bounds-check the seek operation. 2968 * 2969 * IN: vp - vnode seeking within 2970 * ooff - old file offset 2971 * noffp - pointer to new file offset 2972 * 2973 * RETURN: 0 if success 2974 * EINVAL if new offset invalid 2975 */ 2976 /* ARGSUSED */ 2977 static int 2978 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp) 2979 { 2980 if (vp->v_type == VDIR) 2981 return (0); 2982 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 2983 } 2984 2985 /* 2986 * Pre-filter the generic locking function to trap attempts to place 2987 * a mandatory lock on a memory mapped file. 2988 */ 2989 static int 2990 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 2991 flk_callback_t *flk_cbp, cred_t *cr) 2992 { 2993 znode_t *zp = VTOZ(vp); 2994 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2995 int error; 2996 2997 ZFS_ENTER(zfsvfs); 2998 2999 /* 3000 * We are following the UFS semantics with respect to mapcnt 3001 * here: If we see that the file is mapped already, then we will 3002 * return an error, but we don't worry about races between this 3003 * function and zfs_map(). 3004 */ 3005 if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) { 3006 ZFS_EXIT(zfsvfs); 3007 return (EAGAIN); 3008 } 3009 error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr); 3010 ZFS_EXIT(zfsvfs); 3011 return (error); 3012 } 3013 3014 /* 3015 * If we can't find a page in the cache, we will create a new page 3016 * and fill it with file data. For efficiency, we may try to fill 3017 * multiple pages at once (klustering). 3018 */ 3019 static int 3020 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 3021 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 3022 { 3023 znode_t *zp = VTOZ(vp); 3024 page_t *pp, *cur_pp; 3025 objset_t *os = zp->z_zfsvfs->z_os; 3026 caddr_t va; 3027 u_offset_t io_off, total; 3028 uint64_t oid = zp->z_id; 3029 size_t io_len; 3030 uint64_t filesz; 3031 int err; 3032 3033 /* 3034 * If we are only asking for a single page don't bother klustering. 3035 */ 3036 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3037 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE || off > filesz) { 3038 io_off = off; 3039 io_len = PAGESIZE; 3040 pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr); 3041 } else { 3042 /* 3043 * Try to fill a kluster of pages (a blocks worth). 3044 */ 3045 size_t klen; 3046 u_offset_t koff; 3047 3048 if (!ISP2(zp->z_blksz)) { 3049 /* Only one block in the file. */ 3050 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3051 koff = 0; 3052 } else { 3053 klen = plsz; 3054 koff = P2ALIGN(off, (u_offset_t)klen); 3055 } 3056 ASSERT(koff <= filesz); 3057 if (koff + klen > filesz) 3058 klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff; 3059 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 3060 &io_len, koff, klen, 0); 3061 } 3062 if (pp == NULL) { 3063 /* 3064 * Some other thread entered the page before us. 3065 * Return to zfs_getpage to retry the lookup. 3066 */ 3067 *pl = NULL; 3068 return (0); 3069 } 3070 3071 /* 3072 * Fill the pages in the kluster. 3073 */ 3074 cur_pp = pp; 3075 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 3076 ASSERT(io_off == cur_pp->p_offset); 3077 va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 3078 err = dmu_read(os, oid, io_off, PAGESIZE, va); 3079 ppmapout(va); 3080 if (err) { 3081 /* On error, toss the entire kluster */ 3082 pvn_read_done(pp, B_ERROR); 3083 return (err); 3084 } 3085 cur_pp = cur_pp->p_next; 3086 } 3087 out: 3088 /* 3089 * Fill in the page list array from the kluster. If 3090 * there are too many pages in the kluster, return 3091 * as many pages as possible starting from the desired 3092 * offset `off'. 3093 * NOTE: the page list will always be null terminated. 3094 */ 3095 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 3096 3097 return (0); 3098 } 3099 3100 /* 3101 * Return pointers to the pages for the file region [off, off + len] 3102 * in the pl array. If plsz is greater than len, this function may 3103 * also return page pointers from before or after the specified 3104 * region (i.e. some region [off', off' + plsz]). These additional 3105 * pages are only returned if they are already in the cache, or were 3106 * created as part of a klustered read. 3107 * 3108 * IN: vp - vnode of file to get data from. 3109 * off - position in file to get data from. 3110 * len - amount of data to retrieve. 3111 * plsz - length of provided page list. 3112 * seg - segment to obtain pages for. 3113 * addr - virtual address of fault. 3114 * rw - mode of created pages. 3115 * cr - credentials of caller. 3116 * 3117 * OUT: protp - protection mode of created pages. 3118 * pl - list of pages created. 3119 * 3120 * RETURN: 0 if success 3121 * error code if failure 3122 * 3123 * Timestamps: 3124 * vp - atime updated 3125 */ 3126 /* ARGSUSED */ 3127 static int 3128 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 3129 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 3130 enum seg_rw rw, cred_t *cr) 3131 { 3132 znode_t *zp = VTOZ(vp); 3133 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3134 page_t *pp, **pl0 = pl; 3135 rl_t *rl; 3136 int cnt = 0, need_unlock = 0, err = 0; 3137 3138 ZFS_ENTER(zfsvfs); 3139 3140 if (protp) 3141 *protp = PROT_ALL; 3142 3143 ASSERT(zp->z_dbuf_held && zp->z_phys); 3144 3145 /* no faultahead (for now) */ 3146 if (pl == NULL) { 3147 ZFS_EXIT(zfsvfs); 3148 return (0); 3149 } 3150 3151 /* 3152 * Make sure nobody restructures the file in the middle of the getpage. 3153 */ 3154 rl = zfs_range_lock(zp, off, len, RL_READER); 3155 3156 /* can't fault past EOF */ 3157 if (off >= zp->z_phys->zp_size) { 3158 zfs_range_unlock(zp, rl); 3159 ZFS_EXIT(zfsvfs); 3160 return (EFAULT); 3161 } 3162 3163 /* 3164 * If we already own the lock, then we must be page faulting 3165 * in the middle of a write to this file (i.e., we are writing 3166 * to this file using data from a mapped region of the file). 3167 */ 3168 if (!rw_owner(&zp->z_map_lock)) { 3169 rw_enter(&zp->z_map_lock, RW_WRITER); 3170 need_unlock = TRUE; 3171 } 3172 3173 /* 3174 * Loop through the requested range [off, off + len] looking 3175 * for pages. If we don't find a page, we will need to create 3176 * a new page and fill it with data from the file. 3177 */ 3178 while (len > 0) { 3179 if (plsz < PAGESIZE) 3180 break; 3181 if (pp = page_lookup(vp, off, SE_SHARED)) { 3182 *pl++ = pp; 3183 off += PAGESIZE; 3184 addr += PAGESIZE; 3185 len -= PAGESIZE; 3186 plsz -= PAGESIZE; 3187 } else { 3188 err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw); 3189 /* 3190 * klustering may have changed our region 3191 * to be block aligned. 3192 */ 3193 if (((pp = *pl) != 0) && (off != pp->p_offset)) { 3194 int delta = off - pp->p_offset; 3195 len += delta; 3196 off -= delta; 3197 addr -= delta; 3198 } 3199 while (*pl) { 3200 pl++; 3201 cnt++; 3202 off += PAGESIZE; 3203 addr += PAGESIZE; 3204 plsz -= PAGESIZE; 3205 if (len > PAGESIZE) 3206 len -= PAGESIZE; 3207 else 3208 len = 0; 3209 } 3210 if (err) { 3211 /* 3212 * Release any pages we have locked. 3213 */ 3214 while (pl > pl0) 3215 page_unlock(*--pl); 3216 goto out; 3217 } 3218 } 3219 } 3220 3221 /* 3222 * Fill out the page array with any pages already in the cache. 3223 */ 3224 while (plsz > 0) { 3225 pp = page_lookup_nowait(vp, off, SE_SHARED); 3226 if (pp == NULL) 3227 break; 3228 *pl++ = pp; 3229 off += PAGESIZE; 3230 plsz -= PAGESIZE; 3231 } 3232 3233 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3234 out: 3235 *pl = NULL; 3236 3237 if (need_unlock) 3238 rw_exit(&zp->z_map_lock); 3239 zfs_range_unlock(zp, rl); 3240 3241 ZFS_EXIT(zfsvfs); 3242 return (err); 3243 } 3244 3245 /* 3246 * Request a memory map for a section of a file. This code interacts 3247 * with common code and the VM system as follows: 3248 * 3249 * common code calls mmap(), which ends up in smmap_common() 3250 * 3251 * this calls VOP_MAP(), which takes you into (say) zfs 3252 * 3253 * zfs_map() calls as_map(), passing segvn_create() as the callback 3254 * 3255 * segvn_create() creates the new segment and calls VOP_ADDMAP() 3256 * 3257 * zfs_addmap() updates z_mapcnt 3258 */ 3259 static int 3260 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 3261 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3262 { 3263 znode_t *zp = VTOZ(vp); 3264 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3265 segvn_crargs_t vn_a; 3266 int error; 3267 3268 ZFS_ENTER(zfsvfs); 3269 3270 if (vp->v_flag & VNOMAP) { 3271 ZFS_EXIT(zfsvfs); 3272 return (ENOSYS); 3273 } 3274 3275 if (off < 0 || len > MAXOFFSET_T - off) { 3276 ZFS_EXIT(zfsvfs); 3277 return (ENXIO); 3278 } 3279 3280 if (vp->v_type != VREG) { 3281 ZFS_EXIT(zfsvfs); 3282 return (ENODEV); 3283 } 3284 3285 /* 3286 * If file is locked, disallow mapping. 3287 */ 3288 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) { 3289 ZFS_EXIT(zfsvfs); 3290 return (EAGAIN); 3291 } 3292 3293 as_rangelock(as); 3294 if ((flags & MAP_FIXED) == 0) { 3295 map_addr(addrp, len, off, 1, flags); 3296 if (*addrp == NULL) { 3297 as_rangeunlock(as); 3298 ZFS_EXIT(zfsvfs); 3299 return (ENOMEM); 3300 } 3301 } else { 3302 /* 3303 * User specified address - blow away any previous mappings 3304 */ 3305 (void) as_unmap(as, *addrp, len); 3306 } 3307 3308 vn_a.vp = vp; 3309 vn_a.offset = (u_offset_t)off; 3310 vn_a.type = flags & MAP_TYPE; 3311 vn_a.prot = prot; 3312 vn_a.maxprot = maxprot; 3313 vn_a.cred = cr; 3314 vn_a.amp = NULL; 3315 vn_a.flags = flags & ~MAP_TYPE; 3316 vn_a.szc = 0; 3317 vn_a.lgrp_mem_policy_flags = 0; 3318 3319 error = as_map(as, *addrp, len, segvn_create, &vn_a); 3320 3321 as_rangeunlock(as); 3322 ZFS_EXIT(zfsvfs); 3323 return (error); 3324 } 3325 3326 /* ARGSUSED */ 3327 static int 3328 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3329 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3330 { 3331 uint64_t pages = btopr(len); 3332 3333 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 3334 return (0); 3335 } 3336 3337 /* 3338 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 3339 * more accurate mtime for the associated file. Since we don't have a way of 3340 * detecting when the data was actually modified, we have to resort to 3341 * heuristics. If an explicit msync() is done, then we mark the mtime when the 3342 * last page is pushed. The problem occurs when the msync() call is omitted, 3343 * which by far the most common case: 3344 * 3345 * open() 3346 * mmap() 3347 * <modify memory> 3348 * munmap() 3349 * close() 3350 * <time lapse> 3351 * putpage() via fsflush 3352 * 3353 * If we wait until fsflush to come along, we can have a modification time that 3354 * is some arbitrary point in the future. In order to prevent this in the 3355 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 3356 * torn down. 3357 */ 3358 /* ARGSUSED */ 3359 static int 3360 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3361 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr) 3362 { 3363 uint64_t pages = btopr(len); 3364 3365 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 3366 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 3367 3368 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 3369 vn_has_cached_data(vp)) 3370 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr); 3371 3372 return (0); 3373 } 3374 3375 /* 3376 * Free or allocate space in a file. Currently, this function only 3377 * supports the `F_FREESP' command. However, this command is somewhat 3378 * misnamed, as its functionality includes the ability to allocate as 3379 * well as free space. 3380 * 3381 * IN: vp - vnode of file to free data in. 3382 * cmd - action to take (only F_FREESP supported). 3383 * bfp - section of file to free/alloc. 3384 * flag - current file open mode flags. 3385 * offset - current file offset. 3386 * cr - credentials of caller [UNUSED]. 3387 * 3388 * RETURN: 0 if success 3389 * error code if failure 3390 * 3391 * Timestamps: 3392 * vp - ctime|mtime updated 3393 */ 3394 /* ARGSUSED */ 3395 static int 3396 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 3397 offset_t offset, cred_t *cr, caller_context_t *ct) 3398 { 3399 znode_t *zp = VTOZ(vp); 3400 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3401 uint64_t off, len; 3402 int error; 3403 3404 ZFS_ENTER(zfsvfs); 3405 3406 top: 3407 if (cmd != F_FREESP) { 3408 ZFS_EXIT(zfsvfs); 3409 return (EINVAL); 3410 } 3411 3412 if (error = convoff(vp, bfp, 0, offset)) { 3413 ZFS_EXIT(zfsvfs); 3414 return (error); 3415 } 3416 3417 if (bfp->l_len < 0) { 3418 ZFS_EXIT(zfsvfs); 3419 return (EINVAL); 3420 } 3421 3422 off = bfp->l_start; 3423 len = bfp->l_len; /* 0 means from off to end of file */ 3424 3425 do { 3426 error = zfs_freesp(zp, off, len, flag, TRUE); 3427 /* NB: we already did dmu_tx_wait() if necessary */ 3428 } while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 3429 3430 ZFS_EXIT(zfsvfs); 3431 return (error); 3432 } 3433 3434 static int 3435 zfs_fid(vnode_t *vp, fid_t *fidp) 3436 { 3437 znode_t *zp = VTOZ(vp); 3438 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3439 uint32_t gen = (uint32_t)zp->z_phys->zp_gen; 3440 uint64_t object = zp->z_id; 3441 zfid_short_t *zfid; 3442 int size, i; 3443 3444 ZFS_ENTER(zfsvfs); 3445 3446 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 3447 if (fidp->fid_len < size) { 3448 fidp->fid_len = size; 3449 ZFS_EXIT(zfsvfs); 3450 return (ENOSPC); 3451 } 3452 3453 zfid = (zfid_short_t *)fidp; 3454 3455 zfid->zf_len = size; 3456 3457 for (i = 0; i < sizeof (zfid->zf_object); i++) 3458 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 3459 3460 /* Must have a non-zero generation number to distinguish from .zfs */ 3461 if (gen == 0) 3462 gen = 1; 3463 for (i = 0; i < sizeof (zfid->zf_gen); i++) 3464 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 3465 3466 if (size == LONG_FID_LEN) { 3467 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 3468 zfid_long_t *zlfid; 3469 3470 zlfid = (zfid_long_t *)fidp; 3471 3472 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 3473 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 3474 3475 /* XXX - this should be the generation number for the objset */ 3476 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 3477 zlfid->zf_setgen[i] = 0; 3478 } 3479 3480 ZFS_EXIT(zfsvfs); 3481 return (0); 3482 } 3483 3484 static int 3485 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr) 3486 { 3487 znode_t *zp, *xzp; 3488 zfsvfs_t *zfsvfs; 3489 zfs_dirlock_t *dl; 3490 int error; 3491 3492 switch (cmd) { 3493 case _PC_LINK_MAX: 3494 *valp = ULONG_MAX; 3495 return (0); 3496 3497 case _PC_FILESIZEBITS: 3498 *valp = 64; 3499 return (0); 3500 3501 case _PC_XATTR_EXISTS: 3502 zp = VTOZ(vp); 3503 zfsvfs = zp->z_zfsvfs; 3504 ZFS_ENTER(zfsvfs); 3505 *valp = 0; 3506 error = zfs_dirent_lock(&dl, zp, "", &xzp, 3507 ZXATTR | ZEXISTS | ZSHARED); 3508 if (error == 0) { 3509 zfs_dirent_unlock(dl); 3510 if (!zfs_dirempty(xzp)) 3511 *valp = 1; 3512 VN_RELE(ZTOV(xzp)); 3513 } else if (error == ENOENT) { 3514 /* 3515 * If there aren't extended attributes, it's the 3516 * same as having zero of them. 3517 */ 3518 error = 0; 3519 } 3520 ZFS_EXIT(zfsvfs); 3521 return (error); 3522 3523 case _PC_ACL_ENABLED: 3524 *valp = _ACL_ACE_ENABLED; 3525 return (0); 3526 3527 case _PC_MIN_HOLE_SIZE: 3528 *valp = (ulong_t)SPA_MINBLOCKSIZE; 3529 return (0); 3530 3531 default: 3532 return (fs_pathconf(vp, cmd, valp, cr)); 3533 } 3534 } 3535 3536 /*ARGSUSED*/ 3537 static int 3538 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3539 { 3540 znode_t *zp = VTOZ(vp); 3541 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3542 int error; 3543 3544 ZFS_ENTER(zfsvfs); 3545 error = zfs_getacl(zp, vsecp, cr); 3546 ZFS_EXIT(zfsvfs); 3547 3548 return (error); 3549 } 3550 3551 /*ARGSUSED*/ 3552 static int 3553 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3554 { 3555 znode_t *zp = VTOZ(vp); 3556 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3557 int error; 3558 3559 ZFS_ENTER(zfsvfs); 3560 error = zfs_setacl(zp, vsecp, cr); 3561 ZFS_EXIT(zfsvfs); 3562 return (error); 3563 } 3564 3565 /* 3566 * Predeclare these here so that the compiler assumes that 3567 * this is an "old style" function declaration that does 3568 * not include arguments => we won't get type mismatch errors 3569 * in the initializations that follow. 3570 */ 3571 static int zfs_inval(); 3572 static int zfs_isdir(); 3573 3574 static int 3575 zfs_inval() 3576 { 3577 return (EINVAL); 3578 } 3579 3580 static int 3581 zfs_isdir() 3582 { 3583 return (EISDIR); 3584 } 3585 /* 3586 * Directory vnode operations template 3587 */ 3588 vnodeops_t *zfs_dvnodeops; 3589 const fs_operation_def_t zfs_dvnodeops_template[] = { 3590 VOPNAME_OPEN, zfs_open, 3591 VOPNAME_CLOSE, zfs_close, 3592 VOPNAME_READ, zfs_isdir, 3593 VOPNAME_WRITE, zfs_isdir, 3594 VOPNAME_IOCTL, zfs_ioctl, 3595 VOPNAME_GETATTR, zfs_getattr, 3596 VOPNAME_SETATTR, zfs_setattr, 3597 VOPNAME_ACCESS, zfs_access, 3598 VOPNAME_LOOKUP, zfs_lookup, 3599 VOPNAME_CREATE, zfs_create, 3600 VOPNAME_REMOVE, zfs_remove, 3601 VOPNAME_LINK, zfs_link, 3602 VOPNAME_RENAME, zfs_rename, 3603 VOPNAME_MKDIR, zfs_mkdir, 3604 VOPNAME_RMDIR, zfs_rmdir, 3605 VOPNAME_READDIR, zfs_readdir, 3606 VOPNAME_SYMLINK, zfs_symlink, 3607 VOPNAME_FSYNC, zfs_fsync, 3608 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3609 VOPNAME_FID, zfs_fid, 3610 VOPNAME_SEEK, zfs_seek, 3611 VOPNAME_PATHCONF, zfs_pathconf, 3612 VOPNAME_GETSECATTR, zfs_getsecattr, 3613 VOPNAME_SETSECATTR, zfs_setsecattr, 3614 NULL, NULL 3615 }; 3616 3617 /* 3618 * Regular file vnode operations template 3619 */ 3620 vnodeops_t *zfs_fvnodeops; 3621 const fs_operation_def_t zfs_fvnodeops_template[] = { 3622 VOPNAME_OPEN, zfs_open, 3623 VOPNAME_CLOSE, zfs_close, 3624 VOPNAME_READ, zfs_read, 3625 VOPNAME_WRITE, zfs_write, 3626 VOPNAME_IOCTL, zfs_ioctl, 3627 VOPNAME_GETATTR, zfs_getattr, 3628 VOPNAME_SETATTR, zfs_setattr, 3629 VOPNAME_ACCESS, zfs_access, 3630 VOPNAME_LOOKUP, zfs_lookup, 3631 VOPNAME_RENAME, zfs_rename, 3632 VOPNAME_FSYNC, zfs_fsync, 3633 VOPNAME_INACTIVE, (fs_generic_func_p)zfs_inactive, 3634 VOPNAME_FID, zfs_fid, 3635 VOPNAME_SEEK, zfs_seek, 3636 VOPNAME_FRLOCK, zfs_frlock, 3637 VOPNAME_SPACE, zfs_space, 3638 VOPNAME_GETPAGE, zfs_getpage, 3639 VOPNAME_PUTPAGE, zfs_putpage, 3640 VOPNAME_MAP, (fs_generic_func_p) zfs_map, 3641 VOPNAME_ADDMAP, (fs_generic_func_p) zfs_addmap, 3642 VOPNAME_DELMAP, zfs_delmap, 3643 VOPNAME_PATHCONF, zfs_pathconf, 3644 VOPNAME_GETSECATTR, zfs_getsecattr, 3645 VOPNAME_SETSECATTR, zfs_setsecattr, 3646 VOPNAME_VNEVENT, fs_vnevent_support, 3647 NULL, NULL 3648 }; 3649 3650 /* 3651 * Symbolic link vnode operations template 3652 */ 3653 vnodeops_t *zfs_symvnodeops; 3654 const fs_operation_def_t zfs_symvnodeops_template[] = { 3655 VOPNAME_GETATTR, zfs_getattr, 3656 VOPNAME_SETATTR, zfs_setattr, 3657 VOPNAME_ACCESS, zfs_access, 3658 VOPNAME_RENAME, zfs_rename, 3659 VOPNAME_READLINK, zfs_readlink, 3660 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3661 VOPNAME_FID, zfs_fid, 3662 VOPNAME_PATHCONF, zfs_pathconf, 3663 VOPNAME_VNEVENT, fs_vnevent_support, 3664 NULL, NULL 3665 }; 3666 3667 /* 3668 * Extended attribute directory vnode operations template 3669 * This template is identical to the directory vnodes 3670 * operation template except for restricted operations: 3671 * VOP_MKDIR() 3672 * VOP_SYMLINK() 3673 * Note that there are other restrictions embedded in: 3674 * zfs_create() - restrict type to VREG 3675 * zfs_link() - no links into/out of attribute space 3676 * zfs_rename() - no moves into/out of attribute space 3677 */ 3678 vnodeops_t *zfs_xdvnodeops; 3679 const fs_operation_def_t zfs_xdvnodeops_template[] = { 3680 VOPNAME_OPEN, zfs_open, 3681 VOPNAME_CLOSE, zfs_close, 3682 VOPNAME_IOCTL, zfs_ioctl, 3683 VOPNAME_GETATTR, zfs_getattr, 3684 VOPNAME_SETATTR, zfs_setattr, 3685 VOPNAME_ACCESS, zfs_access, 3686 VOPNAME_LOOKUP, zfs_lookup, 3687 VOPNAME_CREATE, zfs_create, 3688 VOPNAME_REMOVE, zfs_remove, 3689 VOPNAME_LINK, zfs_link, 3690 VOPNAME_RENAME, zfs_rename, 3691 VOPNAME_MKDIR, zfs_inval, 3692 VOPNAME_RMDIR, zfs_rmdir, 3693 VOPNAME_READDIR, zfs_readdir, 3694 VOPNAME_SYMLINK, zfs_inval, 3695 VOPNAME_FSYNC, zfs_fsync, 3696 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3697 VOPNAME_FID, zfs_fid, 3698 VOPNAME_SEEK, zfs_seek, 3699 VOPNAME_PATHCONF, zfs_pathconf, 3700 VOPNAME_GETSECATTR, zfs_getsecattr, 3701 VOPNAME_SETSECATTR, zfs_setsecattr, 3702 VOPNAME_VNEVENT, fs_vnevent_support, 3703 NULL, NULL 3704 }; 3705 3706 /* 3707 * Error vnode operations template 3708 */ 3709 vnodeops_t *zfs_evnodeops; 3710 const fs_operation_def_t zfs_evnodeops_template[] = { 3711 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3712 VOPNAME_PATHCONF, zfs_pathconf, 3713 NULL, NULL 3714 }; 3715