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(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 pages to ensure slow (eg NFS) pages 610 * don't hold up txg. 611 */ 612 zfs_prefault_write(n, 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(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 /* 770 * Start another transaction. 771 */ 772 woff = uio->uio_loffset; 773 tx = dmu_tx_create(zfsvfs->z_os); 774 dmu_tx_hold_bonus(tx, zp->z_id); 775 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 776 error = dmu_tx_assign(tx, zfsvfs->z_assign); 777 if (error) { 778 if (error == ERESTART && 779 zfsvfs->z_assign == TXG_NOWAIT) { 780 dmu_tx_wait(tx); 781 dmu_tx_abort(tx); 782 goto top; 783 } 784 dmu_tx_abort(tx); 785 goto no_tx_done; 786 } 787 } 788 789 tx_done: 790 791 if (tx_bytes != 0) { 792 /* 793 * Update the file size if it has changed; account 794 * for possible concurrent updates. 795 */ 796 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) { 797 dmu_buf_will_dirty(zp->z_dbuf, tx); 798 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 799 uio->uio_loffset); 800 } 801 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 802 seq = zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, 803 ioflag, uio); 804 } 805 dmu_tx_commit(tx); 806 807 808 no_tx_done: 809 810 zfs_range_unlock(rl); 811 812 /* 813 * If we're in replay mode, or we made no progress, return error. 814 * Otherwise, it's at least a partial write, so it's successful. 815 */ 816 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 817 ZFS_EXIT(zfsvfs); 818 return (error); 819 } 820 821 zil_commit(zilog, seq, ioflag & (FSYNC | FDSYNC)); 822 823 ZFS_EXIT(zfsvfs); 824 return (0); 825 } 826 827 void 828 zfs_get_done(dmu_buf_t *db, void *vrl) 829 { 830 rl_t *rl = (rl_t *)vrl; 831 vnode_t *vp = ZTOV(rl->r_zp); 832 833 dmu_buf_rele(db, rl); 834 zfs_range_unlock(rl); 835 VN_RELE(vp); 836 } 837 838 /* 839 * Get data to generate a TX_WRITE intent log record. 840 */ 841 int 842 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 843 { 844 zfsvfs_t *zfsvfs = arg; 845 objset_t *os = zfsvfs->z_os; 846 znode_t *zp; 847 uint64_t off = lr->lr_offset; 848 dmu_buf_t *db; 849 rl_t *rl; 850 int dlen = lr->lr_length; /* length of user data */ 851 int error = 0; 852 853 ASSERT(dlen != 0); 854 855 /* 856 * Nothing to do if the file has been removed 857 */ 858 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 859 return (ENOENT); 860 if (zp->z_reap) { 861 VN_RELE(ZTOV(zp)); 862 return (ENOENT); 863 } 864 865 /* 866 * Write records come in two flavors: immediate and indirect. 867 * For small writes it's cheaper to store the data with the 868 * log record (immediate); for large writes it's cheaper to 869 * sync the data and get a pointer to it (indirect) so that 870 * we don't have to write the data twice. 871 */ 872 if (buf != NULL) { /* immediate write */ 873 rl = zfs_range_lock(zp, off, dlen, RL_READER); 874 /* test for truncation needs to be done while range locked */ 875 if (off >= zp->z_phys->zp_size) { 876 error = ENOENT; 877 goto out; 878 } 879 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, off, FTAG, &db)); 880 bcopy((char *)db->db_data + off - db->db_offset, buf, dlen); 881 dmu_buf_rele(db, FTAG); 882 } else { /* indirect write */ 883 uint64_t boff; /* block starting offset */ 884 885 /* 886 * Have to lock the whole block to ensure when it's 887 * written out and it's checksum is being calculated 888 * that no one can change the data. We need to re-check 889 * blocksize after we get the lock in case it's changed! 890 */ 891 for (;;) { 892 if (ISP2(zp->z_blksz)) { 893 boff = P2ALIGN_TYPED(off, zp->z_blksz, 894 uint64_t); 895 } else { 896 boff = 0; 897 } 898 dlen = zp->z_blksz; 899 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 900 if (zp->z_blksz == dlen) 901 break; 902 zfs_range_unlock(rl); 903 } 904 /* test for truncation needs to be done while range locked */ 905 if (off >= zp->z_phys->zp_size) { 906 error = ENOENT; 907 goto out; 908 } 909 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, rl, &db)); 910 ASSERT(boff == db->db_offset); 911 lr->lr_blkoff = off - boff; 912 error = dmu_sync(zio, db, &lr->lr_blkptr, 913 lr->lr_common.lrc_txg, zio ? zfs_get_done : NULL, rl); 914 /* 915 * If we get EINPROGRESS, then we need to wait for a 916 * write IO initiated by dmu_sync() to complete before 917 * we can release this dbuf. We will finish everthing 918 * up in the zfs_get_done() callback. 919 */ 920 if (error == EINPROGRESS) 921 return (0); 922 dmu_buf_rele(db, rl); 923 } 924 out: 925 zfs_range_unlock(rl); 926 VN_RELE(ZTOV(zp)); 927 return (error); 928 } 929 930 /*ARGSUSED*/ 931 static int 932 zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr) 933 { 934 znode_t *zp = VTOZ(vp); 935 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 936 int error; 937 938 ZFS_ENTER(zfsvfs); 939 error = zfs_zaccess_rwx(zp, mode, cr); 940 ZFS_EXIT(zfsvfs); 941 return (error); 942 } 943 944 /* 945 * Lookup an entry in a directory, or an extended attribute directory. 946 * If it exists, return a held vnode reference for it. 947 * 948 * IN: dvp - vnode of directory to search. 949 * nm - name of entry to lookup. 950 * pnp - full pathname to lookup [UNUSED]. 951 * flags - LOOKUP_XATTR set if looking for an attribute. 952 * rdir - root directory vnode [UNUSED]. 953 * cr - credentials of caller. 954 * 955 * OUT: vpp - vnode of located entry, NULL if not found. 956 * 957 * RETURN: 0 if success 958 * error code if failure 959 * 960 * Timestamps: 961 * NA 962 */ 963 /* ARGSUSED */ 964 static int 965 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 966 int flags, vnode_t *rdir, cred_t *cr) 967 { 968 969 znode_t *zdp = VTOZ(dvp); 970 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 971 int error; 972 973 ZFS_ENTER(zfsvfs); 974 975 *vpp = NULL; 976 977 if (flags & LOOKUP_XATTR) { 978 /* 979 * We don't allow recursive attributes.. 980 * Maybe someday we will. 981 */ 982 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 983 ZFS_EXIT(zfsvfs); 984 return (EINVAL); 985 } 986 987 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr)) { 988 ZFS_EXIT(zfsvfs); 989 return (error); 990 } 991 992 /* 993 * Do we have permission to get into attribute directory? 994 */ 995 996 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) { 997 VN_RELE(*vpp); 998 } 999 1000 ZFS_EXIT(zfsvfs); 1001 return (error); 1002 } 1003 1004 if (dvp->v_type != VDIR) { 1005 ZFS_EXIT(zfsvfs); 1006 return (ENOTDIR); 1007 } 1008 1009 /* 1010 * Check accessibility of directory. 1011 */ 1012 1013 if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) { 1014 ZFS_EXIT(zfsvfs); 1015 return (error); 1016 } 1017 1018 if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) { 1019 1020 /* 1021 * Convert device special files 1022 */ 1023 if (IS_DEVVP(*vpp)) { 1024 vnode_t *svp; 1025 1026 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1027 VN_RELE(*vpp); 1028 if (svp == NULL) 1029 error = ENOSYS; 1030 else 1031 *vpp = svp; 1032 } 1033 } 1034 1035 ZFS_EXIT(zfsvfs); 1036 return (error); 1037 } 1038 1039 /* 1040 * Attempt to create a new entry in a directory. If the entry 1041 * already exists, truncate the file if permissible, else return 1042 * an error. Return the vp of the created or trunc'd file. 1043 * 1044 * IN: dvp - vnode of directory to put new file entry in. 1045 * name - name of new file entry. 1046 * vap - attributes of new file. 1047 * excl - flag indicating exclusive or non-exclusive mode. 1048 * mode - mode to open file with. 1049 * cr - credentials of caller. 1050 * flag - large file flag [UNUSED]. 1051 * 1052 * OUT: vpp - vnode of created or trunc'd entry. 1053 * 1054 * RETURN: 0 if success 1055 * error code if failure 1056 * 1057 * Timestamps: 1058 * dvp - ctime|mtime updated if new entry created 1059 * vp - ctime|mtime always, atime if new 1060 */ 1061 /* ARGSUSED */ 1062 static int 1063 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1064 int mode, vnode_t **vpp, cred_t *cr, int flag) 1065 { 1066 znode_t *zp, *dzp = VTOZ(dvp); 1067 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1068 zilog_t *zilog = zfsvfs->z_log; 1069 uint64_t seq = 0; 1070 objset_t *os = zfsvfs->z_os; 1071 zfs_dirlock_t *dl; 1072 dmu_tx_t *tx; 1073 int error; 1074 uint64_t zoid; 1075 1076 ZFS_ENTER(zfsvfs); 1077 1078 top: 1079 *vpp = NULL; 1080 1081 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1082 vap->va_mode &= ~VSVTX; 1083 1084 if (*name == '\0') { 1085 /* 1086 * Null component name refers to the directory itself. 1087 */ 1088 VN_HOLD(dvp); 1089 zp = dzp; 1090 dl = NULL; 1091 error = 0; 1092 } else { 1093 /* possible VN_HOLD(zp) */ 1094 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) { 1095 if (strcmp(name, "..") == 0) 1096 error = EISDIR; 1097 ZFS_EXIT(zfsvfs); 1098 return (error); 1099 } 1100 } 1101 1102 zoid = zp ? zp->z_id : -1ULL; 1103 1104 if (zp == NULL) { 1105 /* 1106 * Create a new file object and update the directory 1107 * to reference it. 1108 */ 1109 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 1110 goto out; 1111 } 1112 1113 /* 1114 * We only support the creation of regular files in 1115 * extended attribute directories. 1116 */ 1117 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1118 (vap->va_type != VREG)) { 1119 error = EINVAL; 1120 goto out; 1121 } 1122 1123 tx = dmu_tx_create(os); 1124 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1125 dmu_tx_hold_bonus(tx, dzp->z_id); 1126 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1127 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1128 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1129 0, SPA_MAXBLOCKSIZE); 1130 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1131 if (error) { 1132 zfs_dirent_unlock(dl); 1133 if (error == ERESTART && 1134 zfsvfs->z_assign == TXG_NOWAIT) { 1135 dmu_tx_wait(tx); 1136 dmu_tx_abort(tx); 1137 goto top; 1138 } 1139 dmu_tx_abort(tx); 1140 ZFS_EXIT(zfsvfs); 1141 return (error); 1142 } 1143 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1144 ASSERT(zp->z_id == zoid); 1145 (void) zfs_link_create(dl, zp, tx, ZNEW); 1146 seq = zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name); 1147 dmu_tx_commit(tx); 1148 } else { 1149 /* 1150 * A directory entry already exists for this name. 1151 */ 1152 /* 1153 * Can't truncate an existing file if in exclusive mode. 1154 */ 1155 if (excl == EXCL) { 1156 error = EEXIST; 1157 goto out; 1158 } 1159 /* 1160 * Can't open a directory for writing. 1161 */ 1162 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1163 error = EISDIR; 1164 goto out; 1165 } 1166 /* 1167 * Verify requested access to file. 1168 */ 1169 if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) { 1170 goto out; 1171 } 1172 1173 mutex_enter(&dzp->z_lock); 1174 dzp->z_seq++; 1175 mutex_exit(&dzp->z_lock); 1176 1177 /* 1178 * Truncate regular files if requested. 1179 */ 1180 if ((ZTOV(zp)->v_type == VREG) && 1181 (zp->z_phys->zp_size != 0) && 1182 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1183 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1184 if (error == ERESTART && 1185 zfsvfs->z_assign == TXG_NOWAIT) { 1186 /* NB: we already did dmu_tx_wait() */ 1187 zfs_dirent_unlock(dl); 1188 goto top; 1189 } 1190 } 1191 } 1192 out: 1193 1194 if (dl) 1195 zfs_dirent_unlock(dl); 1196 1197 if (error) { 1198 if (zp) 1199 VN_RELE(ZTOV(zp)); 1200 } else { 1201 *vpp = ZTOV(zp); 1202 /* 1203 * If vnode is for a device return a specfs vnode instead. 1204 */ 1205 if (IS_DEVVP(*vpp)) { 1206 struct vnode *svp; 1207 1208 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1209 VN_RELE(*vpp); 1210 if (svp == NULL) { 1211 error = ENOSYS; 1212 } 1213 *vpp = svp; 1214 } 1215 } 1216 1217 zil_commit(zilog, seq, 0); 1218 1219 ZFS_EXIT(zfsvfs); 1220 return (error); 1221 } 1222 1223 /* 1224 * Remove an entry from a directory. 1225 * 1226 * IN: dvp - vnode of directory to remove entry from. 1227 * name - name of entry to remove. 1228 * cr - credentials of caller. 1229 * 1230 * RETURN: 0 if success 1231 * error code if failure 1232 * 1233 * Timestamps: 1234 * dvp - ctime|mtime 1235 * vp - ctime (if nlink > 0) 1236 */ 1237 static int 1238 zfs_remove(vnode_t *dvp, char *name, cred_t *cr) 1239 { 1240 znode_t *zp, *dzp = VTOZ(dvp); 1241 znode_t *xzp = NULL; 1242 vnode_t *vp; 1243 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1244 zilog_t *zilog = zfsvfs->z_log; 1245 uint64_t seq = 0; 1246 uint64_t acl_obj, xattr_obj; 1247 zfs_dirlock_t *dl; 1248 dmu_tx_t *tx; 1249 int may_delete_now, delete_now = FALSE; 1250 int reaped; 1251 int error; 1252 1253 ZFS_ENTER(zfsvfs); 1254 1255 top: 1256 /* 1257 * Attempt to lock directory; fail if entry doesn't exist. 1258 */ 1259 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1260 ZFS_EXIT(zfsvfs); 1261 return (error); 1262 } 1263 1264 vp = ZTOV(zp); 1265 1266 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1267 goto out; 1268 } 1269 1270 /* 1271 * Need to use rmdir for removing directories. 1272 */ 1273 if (vp->v_type == VDIR) { 1274 error = EPERM; 1275 goto out; 1276 } 1277 1278 vnevent_remove(vp); 1279 1280 dnlc_remove(dvp, name); 1281 1282 mutex_enter(&vp->v_lock); 1283 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1284 mutex_exit(&vp->v_lock); 1285 1286 /* 1287 * We may delete the znode now, or we may put it on the delete queue; 1288 * it depends on whether we're the last link, and on whether there are 1289 * other holds on the vnode. So we dmu_tx_hold() the right things to 1290 * allow for either case. 1291 */ 1292 tx = dmu_tx_create(zfsvfs->z_os); 1293 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1294 dmu_tx_hold_bonus(tx, zp->z_id); 1295 if (may_delete_now) 1296 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 1297 1298 /* are there any extended attributes? */ 1299 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1300 /* 1301 * XXX - There is a possibility that the delete 1302 * of the parent file could succeed, but then we get 1303 * an ENOSPC when we try to delete the xattrs... 1304 * so we would need to re-try the deletes periodically 1305 */ 1306 /* XXX - do we need this if we are deleting? */ 1307 dmu_tx_hold_bonus(tx, xattr_obj); 1308 } 1309 1310 /* are there any additional acls */ 1311 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1312 may_delete_now) 1313 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1314 1315 /* charge as an update -- would be nice not to charge at all */ 1316 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 1317 1318 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1319 if (error) { 1320 zfs_dirent_unlock(dl); 1321 VN_RELE(vp); 1322 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1323 dmu_tx_wait(tx); 1324 dmu_tx_abort(tx); 1325 goto top; 1326 } 1327 dmu_tx_abort(tx); 1328 ZFS_EXIT(zfsvfs); 1329 return (error); 1330 } 1331 1332 /* 1333 * Remove the directory entry. 1334 */ 1335 error = zfs_link_destroy(dl, zp, tx, 0, &reaped); 1336 1337 if (error) { 1338 dmu_tx_commit(tx); 1339 goto out; 1340 } 1341 1342 if (reaped) { 1343 mutex_enter(&vp->v_lock); 1344 delete_now = may_delete_now && 1345 vp->v_count == 1 && !vn_has_cached_data(vp) && 1346 zp->z_phys->zp_xattr == xattr_obj && 1347 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1348 mutex_exit(&vp->v_lock); 1349 } 1350 1351 if (delete_now) { 1352 if (zp->z_phys->zp_xattr) { 1353 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1354 ASSERT3U(error, ==, 0); 1355 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1356 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1357 mutex_enter(&xzp->z_lock); 1358 xzp->z_reap = 1; 1359 xzp->z_phys->zp_links = 0; 1360 mutex_exit(&xzp->z_lock); 1361 zfs_dq_add(xzp, tx); 1362 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1363 } 1364 mutex_enter(&zp->z_lock); 1365 mutex_enter(&vp->v_lock); 1366 vp->v_count--; 1367 ASSERT3U(vp->v_count, ==, 0); 1368 mutex_exit(&vp->v_lock); 1369 zp->z_active = 0; 1370 mutex_exit(&zp->z_lock); 1371 zfs_znode_delete(zp, tx); 1372 VFS_RELE(zfsvfs->z_vfs); 1373 } else if (reaped) { 1374 zfs_dq_add(zp, tx); 1375 } 1376 1377 seq = zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name); 1378 1379 dmu_tx_commit(tx); 1380 out: 1381 zfs_dirent_unlock(dl); 1382 1383 if (!delete_now) { 1384 VN_RELE(vp); 1385 } else if (xzp) { 1386 /* this rele delayed to prevent nesting transactions */ 1387 VN_RELE(ZTOV(xzp)); 1388 } 1389 1390 zil_commit(zilog, seq, 0); 1391 1392 ZFS_EXIT(zfsvfs); 1393 return (error); 1394 } 1395 1396 /* 1397 * Create a new directory and insert it into dvp using the name 1398 * provided. Return a pointer to the inserted directory. 1399 * 1400 * IN: dvp - vnode of directory to add subdir to. 1401 * dirname - name of new directory. 1402 * vap - attributes of new directory. 1403 * cr - credentials of caller. 1404 * 1405 * OUT: vpp - vnode of created directory. 1406 * 1407 * RETURN: 0 if success 1408 * error code if failure 1409 * 1410 * Timestamps: 1411 * dvp - ctime|mtime updated 1412 * vp - ctime|mtime|atime updated 1413 */ 1414 static int 1415 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr) 1416 { 1417 znode_t *zp, *dzp = VTOZ(dvp); 1418 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1419 zilog_t *zilog = zfsvfs->z_log; 1420 uint64_t seq = 0; 1421 zfs_dirlock_t *dl; 1422 uint64_t zoid = 0; 1423 dmu_tx_t *tx; 1424 int error; 1425 1426 ASSERT(vap->va_type == VDIR); 1427 1428 ZFS_ENTER(zfsvfs); 1429 1430 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1431 ZFS_EXIT(zfsvfs); 1432 return (EINVAL); 1433 } 1434 top: 1435 *vpp = NULL; 1436 1437 /* 1438 * First make sure the new directory doesn't exist. 1439 */ 1440 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) { 1441 ZFS_EXIT(zfsvfs); 1442 return (error); 1443 } 1444 1445 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) { 1446 zfs_dirent_unlock(dl); 1447 ZFS_EXIT(zfsvfs); 1448 return (error); 1449 } 1450 1451 /* 1452 * Add a new entry to the directory. 1453 */ 1454 tx = dmu_tx_create(zfsvfs->z_os); 1455 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1456 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1457 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1458 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1459 0, SPA_MAXBLOCKSIZE); 1460 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1461 if (error) { 1462 zfs_dirent_unlock(dl); 1463 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1464 dmu_tx_wait(tx); 1465 dmu_tx_abort(tx); 1466 goto top; 1467 } 1468 dmu_tx_abort(tx); 1469 ZFS_EXIT(zfsvfs); 1470 return (error); 1471 } 1472 1473 /* 1474 * Create new node. 1475 */ 1476 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1477 1478 /* 1479 * Now put new name in parent dir. 1480 */ 1481 (void) zfs_link_create(dl, zp, tx, ZNEW); 1482 1483 *vpp = ZTOV(zp); 1484 1485 seq = zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname); 1486 dmu_tx_commit(tx); 1487 1488 zfs_dirent_unlock(dl); 1489 1490 zil_commit(zilog, seq, 0); 1491 1492 ZFS_EXIT(zfsvfs); 1493 return (0); 1494 } 1495 1496 /* 1497 * Remove a directory subdir entry. If the current working 1498 * directory is the same as the subdir to be removed, the 1499 * remove will fail. 1500 * 1501 * IN: dvp - vnode of directory to remove from. 1502 * name - name of directory to be removed. 1503 * cwd - vnode of current working directory. 1504 * cr - credentials of caller. 1505 * 1506 * RETURN: 0 if success 1507 * error code if failure 1508 * 1509 * Timestamps: 1510 * dvp - ctime|mtime updated 1511 */ 1512 static int 1513 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr) 1514 { 1515 znode_t *dzp = VTOZ(dvp); 1516 znode_t *zp; 1517 vnode_t *vp; 1518 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1519 zilog_t *zilog = zfsvfs->z_log; 1520 uint64_t seq = 0; 1521 zfs_dirlock_t *dl; 1522 dmu_tx_t *tx; 1523 int error; 1524 1525 ZFS_ENTER(zfsvfs); 1526 1527 top: 1528 zp = NULL; 1529 1530 /* 1531 * Attempt to lock directory; fail if entry doesn't exist. 1532 */ 1533 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1534 ZFS_EXIT(zfsvfs); 1535 return (error); 1536 } 1537 1538 vp = ZTOV(zp); 1539 1540 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1541 goto out; 1542 } 1543 1544 if (vp->v_type != VDIR) { 1545 error = ENOTDIR; 1546 goto out; 1547 } 1548 1549 if (vp == cwd) { 1550 error = EINVAL; 1551 goto out; 1552 } 1553 1554 vnevent_rmdir(vp); 1555 1556 /* 1557 * Grab a lock on the parent pointer make sure we play well 1558 * with the treewalk and directory rename code. 1559 */ 1560 rw_enter(&zp->z_parent_lock, RW_WRITER); 1561 1562 tx = dmu_tx_create(zfsvfs->z_os); 1563 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1564 dmu_tx_hold_bonus(tx, zp->z_id); 1565 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 1566 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1567 if (error) { 1568 rw_exit(&zp->z_parent_lock); 1569 zfs_dirent_unlock(dl); 1570 VN_RELE(vp); 1571 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1572 dmu_tx_wait(tx); 1573 dmu_tx_abort(tx); 1574 goto top; 1575 } 1576 dmu_tx_abort(tx); 1577 ZFS_EXIT(zfsvfs); 1578 return (error); 1579 } 1580 1581 error = zfs_link_destroy(dl, zp, tx, 0, NULL); 1582 1583 if (error == 0) 1584 seq = zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name); 1585 1586 dmu_tx_commit(tx); 1587 1588 rw_exit(&zp->z_parent_lock); 1589 out: 1590 zfs_dirent_unlock(dl); 1591 1592 VN_RELE(vp); 1593 1594 zil_commit(zilog, seq, 0); 1595 1596 ZFS_EXIT(zfsvfs); 1597 return (error); 1598 } 1599 1600 /* 1601 * Read as many directory entries as will fit into the provided 1602 * buffer from the given directory cursor position (specified in 1603 * the uio structure. 1604 * 1605 * IN: vp - vnode of directory to read. 1606 * uio - structure supplying read location, range info, 1607 * and return buffer. 1608 * cr - credentials of caller. 1609 * 1610 * OUT: uio - updated offset and range, buffer filled. 1611 * eofp - set to true if end-of-file detected. 1612 * 1613 * RETURN: 0 if success 1614 * error code if failure 1615 * 1616 * Timestamps: 1617 * vp - atime updated 1618 * 1619 * Note that the low 4 bits of the cookie returned by zap is always zero. 1620 * This allows us to use the low range for "special" directory entries: 1621 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1622 * we use the offset 2 for the '.zfs' directory. 1623 */ 1624 /* ARGSUSED */ 1625 static int 1626 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp) 1627 { 1628 znode_t *zp = VTOZ(vp); 1629 iovec_t *iovp; 1630 dirent64_t *odp; 1631 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1632 objset_t *os; 1633 caddr_t outbuf; 1634 size_t bufsize; 1635 zap_cursor_t zc; 1636 zap_attribute_t zap; 1637 uint_t bytes_wanted; 1638 ushort_t this_reclen; 1639 uint64_t offset; /* must be unsigned; checks for < 1 */ 1640 off64_t *next; 1641 int local_eof; 1642 int outcount; 1643 int error; 1644 uint8_t prefetch; 1645 1646 ZFS_ENTER(zfsvfs); 1647 1648 /* 1649 * If we are not given an eof variable, 1650 * use a local one. 1651 */ 1652 if (eofp == NULL) 1653 eofp = &local_eof; 1654 1655 /* 1656 * Check for valid iov_len. 1657 */ 1658 if (uio->uio_iov->iov_len <= 0) { 1659 ZFS_EXIT(zfsvfs); 1660 return (EINVAL); 1661 } 1662 1663 /* 1664 * Quit if directory has been removed (posix) 1665 */ 1666 if ((*eofp = zp->z_reap) != 0) { 1667 ZFS_EXIT(zfsvfs); 1668 return (0); 1669 } 1670 1671 error = 0; 1672 os = zfsvfs->z_os; 1673 offset = uio->uio_loffset; 1674 prefetch = zp->z_zn_prefetch; 1675 1676 /* 1677 * Initialize the iterator cursor. 1678 */ 1679 if (offset <= 3) { 1680 /* 1681 * Start iteration from the beginning of the directory. 1682 */ 1683 zap_cursor_init(&zc, os, zp->z_id); 1684 } else { 1685 /* 1686 * The offset is a serialized cursor. 1687 */ 1688 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1689 } 1690 1691 /* 1692 * Get space to change directory entries into fs independent format. 1693 */ 1694 iovp = uio->uio_iov; 1695 bytes_wanted = iovp->iov_len; 1696 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1697 bufsize = bytes_wanted; 1698 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1699 odp = (struct dirent64 *)outbuf; 1700 } else { 1701 bufsize = bytes_wanted; 1702 odp = (struct dirent64 *)iovp->iov_base; 1703 } 1704 1705 /* 1706 * Transform to file-system independent format 1707 */ 1708 outcount = 0; 1709 while (outcount < bytes_wanted) { 1710 /* 1711 * Special case `.', `..', and `.zfs'. 1712 */ 1713 if (offset == 0) { 1714 (void) strcpy(zap.za_name, "."); 1715 zap.za_first_integer = zp->z_id; 1716 this_reclen = DIRENT64_RECLEN(1); 1717 } else if (offset == 1) { 1718 (void) strcpy(zap.za_name, ".."); 1719 zap.za_first_integer = zp->z_phys->zp_parent; 1720 this_reclen = DIRENT64_RECLEN(2); 1721 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1722 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1723 zap.za_first_integer = ZFSCTL_INO_ROOT; 1724 this_reclen = 1725 DIRENT64_RECLEN(sizeof (ZFS_CTLDIR_NAME) - 1); 1726 } else { 1727 /* 1728 * Grab next entry. 1729 */ 1730 if (error = zap_cursor_retrieve(&zc, &zap)) { 1731 if ((*eofp = (error == ENOENT)) != 0) 1732 break; 1733 else 1734 goto update; 1735 } 1736 1737 if (zap.za_integer_length != 8 || 1738 zap.za_num_integers != 1) { 1739 cmn_err(CE_WARN, "zap_readdir: bad directory " 1740 "entry, obj = %lld, offset = %lld\n", 1741 (u_longlong_t)zp->z_id, 1742 (u_longlong_t)offset); 1743 error = ENXIO; 1744 goto update; 1745 } 1746 this_reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1747 } 1748 1749 /* 1750 * Will this entry fit in the buffer? 1751 */ 1752 if (outcount + this_reclen > bufsize) { 1753 /* 1754 * Did we manage to fit anything in the buffer? 1755 */ 1756 if (!outcount) { 1757 error = EINVAL; 1758 goto update; 1759 } 1760 break; 1761 } 1762 /* 1763 * Add this entry: 1764 */ 1765 odp->d_ino = (ino64_t)zap.za_first_integer; 1766 odp->d_reclen = (ushort_t)this_reclen; 1767 /* NOTE: d_off is the offset for the *next* entry */ 1768 next = &(odp->d_off); 1769 (void) strncpy(odp->d_name, zap.za_name, 1770 DIRENT64_NAMELEN(this_reclen)); 1771 outcount += this_reclen; 1772 odp = (dirent64_t *)((intptr_t)odp + this_reclen); 1773 1774 ASSERT(outcount <= bufsize); 1775 1776 /* Prefetch znode */ 1777 if (prefetch) 1778 dmu_prefetch(os, zap.za_first_integer, 0, 0); 1779 1780 /* 1781 * Move to the next entry, fill in the previous offset. 1782 */ 1783 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1784 zap_cursor_advance(&zc); 1785 offset = zap_cursor_serialize(&zc); 1786 } else { 1787 offset += 1; 1788 } 1789 *next = offset; 1790 } 1791 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1792 1793 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 1794 iovp->iov_base += outcount; 1795 iovp->iov_len -= outcount; 1796 uio->uio_resid -= outcount; 1797 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 1798 /* 1799 * Reset the pointer. 1800 */ 1801 offset = uio->uio_loffset; 1802 } 1803 1804 update: 1805 zap_cursor_fini(&zc); 1806 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 1807 kmem_free(outbuf, bufsize); 1808 1809 if (error == ENOENT) 1810 error = 0; 1811 1812 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1813 1814 uio->uio_loffset = offset; 1815 ZFS_EXIT(zfsvfs); 1816 return (error); 1817 } 1818 1819 static int 1820 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr) 1821 { 1822 znode_t *zp = VTOZ(vp); 1823 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1824 1825 /* 1826 * Regardless of whether this is required for standards conformance, 1827 * this is the logical behavior when fsync() is called on a file with 1828 * dirty pages. We use B_ASYNC since the ZIL transactions are already 1829 * going to be pushed out as part of the zil_commit(). 1830 */ 1831 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 1832 (vp->v_type == VREG) && !(IS_SWAPVP(vp))) 1833 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr); 1834 1835 ZFS_ENTER(zfsvfs); 1836 zil_commit(zfsvfs->z_log, zp->z_last_itx, FSYNC); 1837 ZFS_EXIT(zfsvfs); 1838 return (0); 1839 } 1840 1841 /* 1842 * Get the requested file attributes and place them in the provided 1843 * vattr structure. 1844 * 1845 * IN: vp - vnode of file. 1846 * vap - va_mask identifies requested attributes. 1847 * flags - [UNUSED] 1848 * cr - credentials of caller. 1849 * 1850 * OUT: vap - attribute values. 1851 * 1852 * RETURN: 0 (always succeeds) 1853 */ 1854 /* ARGSUSED */ 1855 static int 1856 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1857 { 1858 znode_t *zp = VTOZ(vp); 1859 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1860 znode_phys_t *pzp = zp->z_phys; 1861 int error; 1862 1863 ZFS_ENTER(zfsvfs); 1864 1865 /* 1866 * Return all attributes. It's cheaper to provide the answer 1867 * than to determine whether we were asked the question. 1868 */ 1869 mutex_enter(&zp->z_lock); 1870 1871 vap->va_type = vp->v_type; 1872 vap->va_mode = pzp->zp_mode & MODEMASK; 1873 vap->va_uid = zp->z_phys->zp_uid; 1874 vap->va_gid = zp->z_phys->zp_gid; 1875 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 1876 vap->va_nodeid = zp->z_id; 1877 vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX); /* nlink_t limit! */ 1878 vap->va_size = pzp->zp_size; 1879 vap->va_rdev = vp->v_rdev; 1880 vap->va_seq = zp->z_seq; 1881 1882 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 1883 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 1884 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 1885 1886 /* 1887 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 1888 * Also, if we are the owner don't bother, since owner should 1889 * always be allowed to read basic attributes of file. 1890 */ 1891 if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) && 1892 (zp->z_phys->zp_uid != crgetuid(cr))) { 1893 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) { 1894 mutex_exit(&zp->z_lock); 1895 ZFS_EXIT(zfsvfs); 1896 return (error); 1897 } 1898 } 1899 1900 mutex_exit(&zp->z_lock); 1901 1902 dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks); 1903 1904 if (zp->z_blksz == 0) { 1905 /* 1906 * Block size hasn't been set; suggest maximal I/O transfers. 1907 */ 1908 vap->va_blksize = zfsvfs->z_max_blksz; 1909 } 1910 1911 ZFS_EXIT(zfsvfs); 1912 return (0); 1913 } 1914 1915 /* 1916 * Set the file attributes to the values contained in the 1917 * vattr structure. 1918 * 1919 * IN: vp - vnode of file to be modified. 1920 * vap - new attribute values. 1921 * flags - ATTR_UTIME set if non-default time values provided. 1922 * cr - credentials of caller. 1923 * 1924 * RETURN: 0 if success 1925 * error code if failure 1926 * 1927 * Timestamps: 1928 * vp - ctime updated, mtime updated if size changed. 1929 */ 1930 /* ARGSUSED */ 1931 static int 1932 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 1933 caller_context_t *ct) 1934 { 1935 struct znode *zp = VTOZ(vp); 1936 znode_phys_t *pzp = zp->z_phys; 1937 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1938 zilog_t *zilog = zfsvfs->z_log; 1939 uint64_t seq = 0; 1940 dmu_tx_t *tx; 1941 vattr_t oldva; 1942 uint_t mask = vap->va_mask; 1943 uint_t saved_mask; 1944 int trim_mask = FALSE; 1945 uint64_t new_mode; 1946 znode_t *attrzp; 1947 int need_policy = FALSE; 1948 int err; 1949 1950 if (mask == 0) 1951 return (0); 1952 1953 if (mask & AT_NOSET) 1954 return (EINVAL); 1955 1956 if (mask & AT_SIZE && vp->v_type == VDIR) 1957 return (EISDIR); 1958 1959 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) 1960 return (EINVAL); 1961 1962 ZFS_ENTER(zfsvfs); 1963 1964 top: 1965 attrzp = NULL; 1966 1967 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 1968 ZFS_EXIT(zfsvfs); 1969 return (EROFS); 1970 } 1971 1972 /* 1973 * First validate permissions 1974 */ 1975 1976 if (mask & AT_SIZE) { 1977 err = zfs_zaccess(zp, ACE_WRITE_DATA, cr); 1978 if (err) { 1979 ZFS_EXIT(zfsvfs); 1980 return (err); 1981 } 1982 /* 1983 * XXX - Note, we are not providing any open 1984 * mode flags here (like FNDELAY), so we may 1985 * block if there are locks present... this 1986 * should be addressed in openat(). 1987 */ 1988 do { 1989 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 1990 /* NB: we already did dmu_tx_wait() if necessary */ 1991 } while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 1992 if (err) { 1993 ZFS_EXIT(zfsvfs); 1994 return (err); 1995 } 1996 } 1997 1998 if (mask & (AT_ATIME|AT_MTIME)) 1999 need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr); 2000 2001 if (mask & (AT_UID|AT_GID)) { 2002 int idmask = (mask & (AT_UID|AT_GID)); 2003 int take_owner; 2004 int take_group; 2005 2006 /* 2007 * NOTE: even if a new mode is being set, 2008 * we may clear S_ISUID/S_ISGID bits. 2009 */ 2010 2011 if (!(mask & AT_MODE)) 2012 vap->va_mode = pzp->zp_mode; 2013 2014 /* 2015 * Take ownership or chgrp to group we are a member of 2016 */ 2017 2018 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2019 take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr); 2020 2021 /* 2022 * If both AT_UID and AT_GID are set then take_owner and 2023 * take_group must both be set in order to allow taking 2024 * ownership. 2025 * 2026 * Otherwise, send the check through secpolicy_vnode_setattr() 2027 * 2028 */ 2029 2030 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2031 ((idmask == AT_UID) && take_owner) || 2032 ((idmask == AT_GID) && take_group)) { 2033 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) { 2034 /* 2035 * Remove setuid/setgid for non-privileged users 2036 */ 2037 secpolicy_setid_clear(vap, cr); 2038 trim_mask = TRUE; 2039 saved_mask = vap->va_mask; 2040 } else { 2041 need_policy = TRUE; 2042 } 2043 } else { 2044 need_policy = TRUE; 2045 } 2046 } 2047 2048 if (mask & AT_MODE) 2049 need_policy = TRUE; 2050 2051 if (need_policy) { 2052 mutex_enter(&zp->z_lock); 2053 oldva.va_mode = pzp->zp_mode; 2054 oldva.va_uid = zp->z_phys->zp_uid; 2055 oldva.va_gid = zp->z_phys->zp_gid; 2056 mutex_exit(&zp->z_lock); 2057 2058 /* 2059 * If trim_mask is set then take ownership 2060 * has been granted. In that case remove 2061 * UID|GID from mask so that 2062 * secpolicy_vnode_setattr() doesn't revoke it. 2063 */ 2064 if (trim_mask) 2065 vap->va_mask &= ~(AT_UID|AT_GID); 2066 2067 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2068 (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp); 2069 if (err) { 2070 ZFS_EXIT(zfsvfs); 2071 return (err); 2072 } 2073 2074 if (trim_mask) 2075 vap->va_mask |= (saved_mask & (AT_UID|AT_GID)); 2076 } 2077 2078 /* 2079 * secpolicy_vnode_setattr, or take ownership may have 2080 * changed va_mask 2081 */ 2082 mask = vap->va_mask; 2083 2084 tx = dmu_tx_create(zfsvfs->z_os); 2085 dmu_tx_hold_bonus(tx, zp->z_id); 2086 2087 if (mask & AT_MODE) { 2088 uint64_t pmode = pzp->zp_mode; 2089 2090 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2091 2092 if (zp->z_phys->zp_acl.z_acl_extern_obj) 2093 dmu_tx_hold_write(tx, 2094 pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE); 2095 else 2096 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2097 0, ZFS_ACL_SIZE(MAX_ACL_SIZE)); 2098 } 2099 2100 if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) { 2101 err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp); 2102 if (err) { 2103 dmu_tx_abort(tx); 2104 ZFS_EXIT(zfsvfs); 2105 return (err); 2106 } 2107 dmu_tx_hold_bonus(tx, attrzp->z_id); 2108 } 2109 2110 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2111 if (err) { 2112 if (attrzp) 2113 VN_RELE(ZTOV(attrzp)); 2114 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2115 dmu_tx_wait(tx); 2116 dmu_tx_abort(tx); 2117 goto top; 2118 } 2119 dmu_tx_abort(tx); 2120 ZFS_EXIT(zfsvfs); 2121 return (err); 2122 } 2123 2124 dmu_buf_will_dirty(zp->z_dbuf, tx); 2125 2126 /* 2127 * Set each attribute requested. 2128 * We group settings according to the locks they need to acquire. 2129 * 2130 * Note: you cannot set ctime directly, although it will be 2131 * updated as a side-effect of calling this function. 2132 */ 2133 2134 mutex_enter(&zp->z_lock); 2135 2136 if (mask & AT_MODE) { 2137 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2138 ASSERT3U(err, ==, 0); 2139 } 2140 2141 if (attrzp) 2142 mutex_enter(&attrzp->z_lock); 2143 2144 if (mask & AT_UID) { 2145 zp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2146 if (attrzp) { 2147 attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2148 } 2149 } 2150 2151 if (mask & AT_GID) { 2152 zp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2153 if (attrzp) 2154 attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2155 } 2156 2157 if (attrzp) 2158 mutex_exit(&attrzp->z_lock); 2159 2160 if (mask & AT_ATIME) 2161 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2162 2163 if (mask & AT_MTIME) 2164 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2165 2166 if (mask & AT_SIZE) 2167 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2168 else if (mask != 0) 2169 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2170 2171 if (mask != 0) 2172 seq = zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask); 2173 2174 mutex_exit(&zp->z_lock); 2175 2176 if (attrzp) 2177 VN_RELE(ZTOV(attrzp)); 2178 2179 dmu_tx_commit(tx); 2180 2181 zil_commit(zilog, seq, 0); 2182 2183 ZFS_EXIT(zfsvfs); 2184 return (err); 2185 } 2186 2187 /* 2188 * Search back through the directory tree, using the ".." entries. 2189 * Lock each directory in the chain to prevent concurrent renames. 2190 * Fail any attempt to move a directory into one of its own descendants. 2191 * XXX - z_parent_lock can overlap with map or grow locks 2192 */ 2193 typedef struct zfs_zlock { 2194 krwlock_t *zl_rwlock; /* lock we acquired */ 2195 znode_t *zl_znode; /* znode we held */ 2196 struct zfs_zlock *zl_next; /* next in list */ 2197 } zfs_zlock_t; 2198 2199 static int 2200 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2201 { 2202 zfs_zlock_t *zl; 2203 znode_t *zp = tdzp; 2204 uint64_t rootid = zp->z_zfsvfs->z_root; 2205 uint64_t *oidp = &zp->z_id; 2206 krwlock_t *rwlp = &szp->z_parent_lock; 2207 krw_t rw = RW_WRITER; 2208 2209 /* 2210 * First pass write-locks szp and compares to zp->z_id. 2211 * Later passes read-lock zp and compare to zp->z_parent. 2212 */ 2213 do { 2214 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2215 zl->zl_rwlock = rwlp; 2216 zl->zl_znode = NULL; 2217 zl->zl_next = *zlpp; 2218 *zlpp = zl; 2219 2220 rw_enter(rwlp, rw); 2221 2222 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2223 return (EINVAL); 2224 2225 if (*oidp == rootid) /* We've hit the top */ 2226 return (0); 2227 2228 if (rw == RW_READER) { /* i.e. not the first pass */ 2229 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2230 if (error) 2231 return (error); 2232 zl->zl_znode = zp; 2233 } 2234 oidp = &zp->z_phys->zp_parent; 2235 rwlp = &zp->z_parent_lock; 2236 rw = RW_READER; 2237 2238 } while (zp->z_id != sdzp->z_id); 2239 2240 return (0); 2241 } 2242 2243 /* 2244 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2245 */ 2246 static void 2247 zfs_rename_unlock(zfs_zlock_t **zlpp) 2248 { 2249 zfs_zlock_t *zl; 2250 2251 while ((zl = *zlpp) != NULL) { 2252 if (zl->zl_znode != NULL) 2253 VN_RELE(ZTOV(zl->zl_znode)); 2254 rw_exit(zl->zl_rwlock); 2255 *zlpp = zl->zl_next; 2256 kmem_free(zl, sizeof (*zl)); 2257 } 2258 } 2259 2260 /* 2261 * Move an entry from the provided source directory to the target 2262 * directory. Change the entry name as indicated. 2263 * 2264 * IN: sdvp - Source directory containing the "old entry". 2265 * snm - Old entry name. 2266 * tdvp - Target directory to contain the "new entry". 2267 * tnm - New entry name. 2268 * cr - credentials of caller. 2269 * 2270 * RETURN: 0 if success 2271 * error code if failure 2272 * 2273 * Timestamps: 2274 * sdvp,tdvp - ctime|mtime updated 2275 */ 2276 static int 2277 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr) 2278 { 2279 znode_t *tdzp, *szp, *tzp; 2280 znode_t *sdzp = VTOZ(sdvp); 2281 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2282 zilog_t *zilog = zfsvfs->z_log; 2283 uint64_t seq = 0; 2284 vnode_t *realvp; 2285 zfs_dirlock_t *sdl, *tdl; 2286 dmu_tx_t *tx; 2287 zfs_zlock_t *zl; 2288 int cmp, serr, terr, error; 2289 2290 ZFS_ENTER(zfsvfs); 2291 2292 /* 2293 * Make sure we have the real vp for the target directory. 2294 */ 2295 if (VOP_REALVP(tdvp, &realvp) == 0) 2296 tdvp = realvp; 2297 2298 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2299 ZFS_EXIT(zfsvfs); 2300 return (EXDEV); 2301 } 2302 2303 tdzp = VTOZ(tdvp); 2304 top: 2305 szp = NULL; 2306 tzp = NULL; 2307 zl = NULL; 2308 2309 /* 2310 * This is to prevent the creation of links into attribute space 2311 * by renaming a linked file into/outof an attribute directory. 2312 * See the comment in zfs_link() for why this is considered bad. 2313 */ 2314 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2315 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2316 ZFS_EXIT(zfsvfs); 2317 return (EINVAL); 2318 } 2319 2320 /* 2321 * Lock source and target directory entries. To prevent deadlock, 2322 * a lock ordering must be defined. We lock the directory with 2323 * the smallest object id first, or if it's a tie, the one with 2324 * the lexically first name. 2325 */ 2326 if (sdzp->z_id < tdzp->z_id) { 2327 cmp = -1; 2328 } else if (sdzp->z_id > tdzp->z_id) { 2329 cmp = 1; 2330 } else { 2331 cmp = strcmp(snm, tnm); 2332 if (cmp == 0) { 2333 /* 2334 * POSIX: "If the old argument and the new argument 2335 * both refer to links to the same existing file, 2336 * the rename() function shall return successfully 2337 * and perform no other action." 2338 */ 2339 ZFS_EXIT(zfsvfs); 2340 return (0); 2341 } 2342 } 2343 if (cmp < 0) { 2344 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2345 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2346 } else { 2347 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2348 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2349 } 2350 2351 if (serr) { 2352 /* 2353 * Source entry invalid or not there. 2354 */ 2355 if (!terr) { 2356 zfs_dirent_unlock(tdl); 2357 if (tzp) 2358 VN_RELE(ZTOV(tzp)); 2359 } 2360 if (strcmp(snm, "..") == 0) 2361 serr = EINVAL; 2362 ZFS_EXIT(zfsvfs); 2363 return (serr); 2364 } 2365 if (terr) { 2366 zfs_dirent_unlock(sdl); 2367 VN_RELE(ZTOV(szp)); 2368 if (strcmp(tnm, "..") == 0) 2369 terr = EINVAL; 2370 ZFS_EXIT(zfsvfs); 2371 return (terr); 2372 } 2373 2374 /* 2375 * Must have write access at the source to remove the old entry 2376 * and write access at the target to create the new entry. 2377 * Note that if target and source are the same, this can be 2378 * done in a single check. 2379 */ 2380 2381 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2382 goto out; 2383 2384 if (ZTOV(szp)->v_type == VDIR) { 2385 /* 2386 * Check to make sure rename is valid. 2387 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2388 */ 2389 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 2390 goto out; 2391 } 2392 2393 /* 2394 * Does target exist? 2395 */ 2396 if (tzp) { 2397 /* 2398 * Source and target must be the same type. 2399 */ 2400 if (ZTOV(szp)->v_type == VDIR) { 2401 if (ZTOV(tzp)->v_type != VDIR) { 2402 error = ENOTDIR; 2403 goto out; 2404 } 2405 } else { 2406 if (ZTOV(tzp)->v_type == VDIR) { 2407 error = EISDIR; 2408 goto out; 2409 } 2410 } 2411 /* 2412 * POSIX dictates that when the source and target 2413 * entries refer to the same file object, rename 2414 * must do nothing and exit without error. 2415 */ 2416 if (szp->z_id == tzp->z_id) { 2417 error = 0; 2418 goto out; 2419 } 2420 } 2421 2422 vnevent_rename_src(ZTOV(szp)); 2423 if (tzp) 2424 vnevent_rename_dest(ZTOV(tzp)); 2425 2426 tx = dmu_tx_create(zfsvfs->z_os); 2427 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 2428 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 2429 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 2430 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 2431 if (sdzp != tdzp) 2432 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 2433 if (tzp) 2434 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 2435 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 2436 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2437 if (error) { 2438 if (zl != NULL) 2439 zfs_rename_unlock(&zl); 2440 zfs_dirent_unlock(sdl); 2441 zfs_dirent_unlock(tdl); 2442 VN_RELE(ZTOV(szp)); 2443 if (tzp) 2444 VN_RELE(ZTOV(tzp)); 2445 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2446 dmu_tx_wait(tx); 2447 dmu_tx_abort(tx); 2448 goto top; 2449 } 2450 dmu_tx_abort(tx); 2451 ZFS_EXIT(zfsvfs); 2452 return (error); 2453 } 2454 2455 if (tzp) /* Attempt to remove the existing target */ 2456 error = zfs_link_destroy(tdl, tzp, tx, 0, NULL); 2457 2458 if (error == 0) { 2459 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 2460 if (error == 0) { 2461 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 2462 ASSERT(error == 0); 2463 seq = zfs_log_rename(zilog, tx, TX_RENAME, 2464 sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp); 2465 } 2466 } 2467 2468 dmu_tx_commit(tx); 2469 out: 2470 if (zl != NULL) 2471 zfs_rename_unlock(&zl); 2472 2473 zfs_dirent_unlock(sdl); 2474 zfs_dirent_unlock(tdl); 2475 2476 VN_RELE(ZTOV(szp)); 2477 if (tzp) 2478 VN_RELE(ZTOV(tzp)); 2479 2480 zil_commit(zilog, seq, 0); 2481 2482 ZFS_EXIT(zfsvfs); 2483 return (error); 2484 } 2485 2486 /* 2487 * Insert the indicated symbolic reference entry into the directory. 2488 * 2489 * IN: dvp - Directory to contain new symbolic link. 2490 * link - Name for new symlink entry. 2491 * vap - Attributes of new entry. 2492 * target - Target path of new symlink. 2493 * cr - credentials of caller. 2494 * 2495 * RETURN: 0 if success 2496 * error code if failure 2497 * 2498 * Timestamps: 2499 * dvp - ctime|mtime updated 2500 */ 2501 static int 2502 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr) 2503 { 2504 znode_t *zp, *dzp = VTOZ(dvp); 2505 zfs_dirlock_t *dl; 2506 dmu_tx_t *tx; 2507 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2508 zilog_t *zilog = zfsvfs->z_log; 2509 uint64_t seq = 0; 2510 uint64_t zoid; 2511 int len = strlen(link); 2512 int error; 2513 2514 ASSERT(vap->va_type == VLNK); 2515 2516 ZFS_ENTER(zfsvfs); 2517 top: 2518 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2519 ZFS_EXIT(zfsvfs); 2520 return (error); 2521 } 2522 2523 if (len > MAXPATHLEN) { 2524 ZFS_EXIT(zfsvfs); 2525 return (ENAMETOOLONG); 2526 } 2527 2528 /* 2529 * Attempt to lock directory; fail if entry already exists. 2530 */ 2531 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) { 2532 ZFS_EXIT(zfsvfs); 2533 return (error); 2534 } 2535 2536 tx = dmu_tx_create(zfsvfs->z_os); 2537 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 2538 dmu_tx_hold_bonus(tx, dzp->z_id); 2539 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2540 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 2541 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 2542 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2543 if (error) { 2544 zfs_dirent_unlock(dl); 2545 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2546 dmu_tx_wait(tx); 2547 dmu_tx_abort(tx); 2548 goto top; 2549 } 2550 dmu_tx_abort(tx); 2551 ZFS_EXIT(zfsvfs); 2552 return (error); 2553 } 2554 2555 dmu_buf_will_dirty(dzp->z_dbuf, tx); 2556 2557 /* 2558 * Create a new object for the symlink. 2559 * Put the link content into bonus buffer if it will fit; 2560 * otherwise, store it just like any other file data. 2561 */ 2562 zoid = 0; 2563 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 2564 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len); 2565 if (len != 0) 2566 bcopy(link, zp->z_phys + 1, len); 2567 } else { 2568 dmu_buf_t *dbp; 2569 2570 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 2571 2572 /* 2573 * Nothing can access the znode yet so no locking needed 2574 * for growing the znode's blocksize. 2575 */ 2576 zfs_grow_blocksize(zp, len, tx); 2577 2578 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp)); 2579 dmu_buf_will_dirty(dbp, tx); 2580 2581 ASSERT3U(len, <=, dbp->db_size); 2582 bcopy(link, dbp->db_data, len); 2583 dmu_buf_rele(dbp, FTAG); 2584 } 2585 zp->z_phys->zp_size = len; 2586 2587 /* 2588 * Insert the new object into the directory. 2589 */ 2590 (void) zfs_link_create(dl, zp, tx, ZNEW); 2591 out: 2592 if (error == 0) 2593 seq = zfs_log_symlink(zilog, tx, TX_SYMLINK, 2594 dzp, zp, name, link); 2595 2596 dmu_tx_commit(tx); 2597 2598 zfs_dirent_unlock(dl); 2599 2600 VN_RELE(ZTOV(zp)); 2601 2602 zil_commit(zilog, seq, 0); 2603 2604 ZFS_EXIT(zfsvfs); 2605 return (error); 2606 } 2607 2608 /* 2609 * Return, in the buffer contained in the provided uio structure, 2610 * the symbolic path referred to by vp. 2611 * 2612 * IN: vp - vnode of symbolic link. 2613 * uoip - structure to contain the link path. 2614 * cr - credentials of caller. 2615 * 2616 * OUT: uio - structure to contain the link path. 2617 * 2618 * RETURN: 0 if success 2619 * error code if failure 2620 * 2621 * Timestamps: 2622 * vp - atime updated 2623 */ 2624 /* ARGSUSED */ 2625 static int 2626 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr) 2627 { 2628 znode_t *zp = VTOZ(vp); 2629 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2630 size_t bufsz; 2631 int error; 2632 2633 ZFS_ENTER(zfsvfs); 2634 2635 bufsz = (size_t)zp->z_phys->zp_size; 2636 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 2637 error = uiomove(zp->z_phys + 1, 2638 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2639 } else { 2640 dmu_buf_t *dbp; 2641 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 2642 if (error) { 2643 ZFS_EXIT(zfsvfs); 2644 return (error); 2645 } 2646 error = uiomove(dbp->db_data, 2647 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2648 dmu_buf_rele(dbp, FTAG); 2649 } 2650 2651 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2652 ZFS_EXIT(zfsvfs); 2653 return (error); 2654 } 2655 2656 /* 2657 * Insert a new entry into directory tdvp referencing svp. 2658 * 2659 * IN: tdvp - Directory to contain new entry. 2660 * svp - vnode of new entry. 2661 * name - name of new entry. 2662 * cr - credentials of caller. 2663 * 2664 * RETURN: 0 if success 2665 * error code if failure 2666 * 2667 * Timestamps: 2668 * tdvp - ctime|mtime updated 2669 * svp - ctime updated 2670 */ 2671 /* ARGSUSED */ 2672 static int 2673 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr) 2674 { 2675 znode_t *dzp = VTOZ(tdvp); 2676 znode_t *tzp, *szp; 2677 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2678 zilog_t *zilog = zfsvfs->z_log; 2679 uint64_t seq = 0; 2680 zfs_dirlock_t *dl; 2681 dmu_tx_t *tx; 2682 vnode_t *realvp; 2683 int error; 2684 2685 ASSERT(tdvp->v_type == VDIR); 2686 2687 ZFS_ENTER(zfsvfs); 2688 2689 if (VOP_REALVP(svp, &realvp) == 0) 2690 svp = realvp; 2691 2692 if (svp->v_vfsp != tdvp->v_vfsp) { 2693 ZFS_EXIT(zfsvfs); 2694 return (EXDEV); 2695 } 2696 2697 szp = VTOZ(svp); 2698 top: 2699 /* 2700 * We do not support links between attributes and non-attributes 2701 * because of the potential security risk of creating links 2702 * into "normal" file space in order to circumvent restrictions 2703 * imposed in attribute space. 2704 */ 2705 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 2706 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 2707 ZFS_EXIT(zfsvfs); 2708 return (EINVAL); 2709 } 2710 2711 /* 2712 * POSIX dictates that we return EPERM here. 2713 * Better choices include ENOTSUP or EISDIR. 2714 */ 2715 if (svp->v_type == VDIR) { 2716 ZFS_EXIT(zfsvfs); 2717 return (EPERM); 2718 } 2719 2720 if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) && 2721 secpolicy_basic_link(cr) != 0) { 2722 ZFS_EXIT(zfsvfs); 2723 return (EPERM); 2724 } 2725 2726 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2727 ZFS_EXIT(zfsvfs); 2728 return (error); 2729 } 2730 2731 /* 2732 * Attempt to lock directory; fail if entry already exists. 2733 */ 2734 if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) { 2735 ZFS_EXIT(zfsvfs); 2736 return (error); 2737 } 2738 2739 tx = dmu_tx_create(zfsvfs->z_os); 2740 dmu_tx_hold_bonus(tx, szp->z_id); 2741 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2742 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2743 if (error) { 2744 zfs_dirent_unlock(dl); 2745 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2746 dmu_tx_wait(tx); 2747 dmu_tx_abort(tx); 2748 goto top; 2749 } 2750 dmu_tx_abort(tx); 2751 ZFS_EXIT(zfsvfs); 2752 return (error); 2753 } 2754 2755 error = zfs_link_create(dl, szp, tx, 0); 2756 2757 if (error == 0) 2758 seq = zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name); 2759 2760 dmu_tx_commit(tx); 2761 2762 zfs_dirent_unlock(dl); 2763 2764 zil_commit(zilog, seq, 0); 2765 2766 ZFS_EXIT(zfsvfs); 2767 return (error); 2768 } 2769 2770 /* 2771 * zfs_null_putapage() is used when the file system has been force 2772 * unmounted. It just drops the pages. 2773 */ 2774 /* ARGSUSED */ 2775 static int 2776 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2777 size_t *lenp, int flags, cred_t *cr) 2778 { 2779 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 2780 return (0); 2781 } 2782 2783 /* ARGSUSED */ 2784 static int 2785 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2786 size_t *lenp, int flags, cred_t *cr) 2787 { 2788 znode_t *zp = VTOZ(vp); 2789 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2790 zilog_t *zilog = zfsvfs->z_log; 2791 dmu_tx_t *tx; 2792 rl_t *rl; 2793 u_offset_t off; 2794 ssize_t len; 2795 caddr_t va; 2796 int err; 2797 2798 top: 2799 off = pp->p_offset; 2800 rl = zfs_range_lock(zp, off, PAGESIZE, RL_WRITER); 2801 /* 2802 * Can't push pages past end-of-file. 2803 */ 2804 if (off >= zp->z_phys->zp_size) { 2805 zfs_range_unlock(rl); 2806 return (EIO); 2807 } 2808 len = MIN(PAGESIZE, zp->z_phys->zp_size - off); 2809 2810 tx = dmu_tx_create(zfsvfs->z_os); 2811 dmu_tx_hold_write(tx, zp->z_id, off, len); 2812 dmu_tx_hold_bonus(tx, zp->z_id); 2813 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2814 if (err != 0) { 2815 zfs_range_unlock(rl); 2816 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2817 dmu_tx_wait(tx); 2818 dmu_tx_abort(tx); 2819 goto top; 2820 } 2821 dmu_tx_abort(tx); 2822 goto out; 2823 } 2824 2825 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 2826 2827 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 2828 2829 ppmapout(va); 2830 2831 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 2832 (void) zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0, NULL); 2833 dmu_tx_commit(tx); 2834 2835 zfs_range_unlock(rl); 2836 2837 pvn_write_done(pp, B_WRITE | flags); 2838 if (offp) 2839 *offp = off; 2840 if (lenp) 2841 *lenp = len; 2842 2843 out: 2844 return (err); 2845 } 2846 2847 /* 2848 * Copy the portion of the file indicated from pages into the file. 2849 * The pages are stored in a page list attached to the files vnode. 2850 * 2851 * IN: vp - vnode of file to push page data to. 2852 * off - position in file to put data. 2853 * len - amount of data to write. 2854 * flags - flags to control the operation. 2855 * cr - credentials of caller. 2856 * 2857 * RETURN: 0 if success 2858 * error code if failure 2859 * 2860 * Timestamps: 2861 * vp - ctime|mtime updated 2862 */ 2863 static int 2864 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr) 2865 { 2866 znode_t *zp = VTOZ(vp); 2867 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2868 page_t *pp; 2869 size_t io_len; 2870 u_offset_t io_off; 2871 uint64_t filesz; 2872 int error = 0; 2873 2874 ZFS_ENTER(zfsvfs); 2875 2876 ASSERT(zp->z_dbuf_held && zp->z_phys); 2877 2878 if (len == 0) { 2879 /* 2880 * Search the entire vp list for pages >= off. 2881 */ 2882 error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage, 2883 flags, cr); 2884 goto out; 2885 } 2886 2887 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 2888 if (off > filesz) { 2889 /* past end of file */ 2890 ZFS_EXIT(zfsvfs); 2891 return (0); 2892 } 2893 2894 len = MIN(len, filesz - off); 2895 2896 for (io_off = off; io_off < off + len; io_off += io_len) { 2897 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 2898 pp = page_lookup(vp, io_off, 2899 (flags & (B_INVAL | B_FREE)) ? 2900 SE_EXCL : SE_SHARED); 2901 } else { 2902 pp = page_lookup_nowait(vp, io_off, 2903 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 2904 } 2905 2906 if (pp != NULL && pvn_getdirty(pp, flags)) { 2907 int err; 2908 2909 /* 2910 * Found a dirty page to push 2911 */ 2912 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 2913 if (err) 2914 error = err; 2915 } else { 2916 io_len = PAGESIZE; 2917 } 2918 } 2919 out: 2920 zil_commit(zfsvfs->z_log, UINT64_MAX, (flags & B_ASYNC) ? 0 : FDSYNC); 2921 ZFS_EXIT(zfsvfs); 2922 return (error); 2923 } 2924 2925 void 2926 zfs_inactive(vnode_t *vp, cred_t *cr) 2927 { 2928 znode_t *zp = VTOZ(vp); 2929 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2930 int error; 2931 2932 rw_enter(&zfsvfs->z_um_lock, RW_READER); 2933 if (zfsvfs->z_unmounted2) { 2934 ASSERT(zp->z_dbuf_held == 0); 2935 2936 if (vn_has_cached_data(vp)) { 2937 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 2938 B_INVAL, cr); 2939 } 2940 2941 mutex_enter(&zp->z_lock); 2942 vp->v_count = 0; /* count arrives as 1 */ 2943 if (zp->z_dbuf == NULL) { 2944 mutex_exit(&zp->z_lock); 2945 zfs_znode_free(zp); 2946 } else { 2947 mutex_exit(&zp->z_lock); 2948 } 2949 rw_exit(&zfsvfs->z_um_lock); 2950 VFS_RELE(zfsvfs->z_vfs); 2951 return; 2952 } 2953 2954 /* 2955 * Attempt to push any data in the page cache. If this fails 2956 * we will get kicked out later in zfs_zinactive(). 2957 */ 2958 if (vn_has_cached_data(vp)) { 2959 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 2960 cr); 2961 } 2962 2963 if (zp->z_atime_dirty && zp->z_reap == 0) { 2964 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 2965 2966 dmu_tx_hold_bonus(tx, zp->z_id); 2967 error = dmu_tx_assign(tx, TXG_WAIT); 2968 if (error) { 2969 dmu_tx_abort(tx); 2970 } else { 2971 dmu_buf_will_dirty(zp->z_dbuf, tx); 2972 mutex_enter(&zp->z_lock); 2973 zp->z_atime_dirty = 0; 2974 mutex_exit(&zp->z_lock); 2975 dmu_tx_commit(tx); 2976 } 2977 } 2978 2979 zfs_zinactive(zp); 2980 rw_exit(&zfsvfs->z_um_lock); 2981 } 2982 2983 /* 2984 * Bounds-check the seek operation. 2985 * 2986 * IN: vp - vnode seeking within 2987 * ooff - old file offset 2988 * noffp - pointer to new file offset 2989 * 2990 * RETURN: 0 if success 2991 * EINVAL if new offset invalid 2992 */ 2993 /* ARGSUSED */ 2994 static int 2995 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp) 2996 { 2997 if (vp->v_type == VDIR) 2998 return (0); 2999 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 3000 } 3001 3002 /* 3003 * Pre-filter the generic locking function to trap attempts to place 3004 * a mandatory lock on a memory mapped file. 3005 */ 3006 static int 3007 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 3008 flk_callback_t *flk_cbp, cred_t *cr) 3009 { 3010 znode_t *zp = VTOZ(vp); 3011 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3012 int error; 3013 3014 ZFS_ENTER(zfsvfs); 3015 3016 /* 3017 * We are following the UFS semantics with respect to mapcnt 3018 * here: If we see that the file is mapped already, then we will 3019 * return an error, but we don't worry about races between this 3020 * function and zfs_map(). 3021 */ 3022 if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) { 3023 ZFS_EXIT(zfsvfs); 3024 return (EAGAIN); 3025 } 3026 error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr); 3027 ZFS_EXIT(zfsvfs); 3028 return (error); 3029 } 3030 3031 /* 3032 * If we can't find a page in the cache, we will create a new page 3033 * and fill it with file data. For efficiency, we may try to fill 3034 * multiple pages at once (klustering). 3035 */ 3036 static int 3037 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 3038 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 3039 { 3040 znode_t *zp = VTOZ(vp); 3041 page_t *pp, *cur_pp; 3042 objset_t *os = zp->z_zfsvfs->z_os; 3043 caddr_t va; 3044 u_offset_t io_off, total; 3045 uint64_t oid = zp->z_id; 3046 size_t io_len; 3047 uint64_t filesz; 3048 int err; 3049 3050 /* 3051 * If we are only asking for a single page don't bother klustering. 3052 */ 3053 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3054 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE || off > filesz) { 3055 io_off = off; 3056 io_len = PAGESIZE; 3057 pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr); 3058 } else { 3059 /* 3060 * Try to fill a kluster of pages (a blocks worth). 3061 */ 3062 size_t klen; 3063 u_offset_t koff; 3064 3065 if (!ISP2(zp->z_blksz)) { 3066 /* Only one block in the file. */ 3067 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3068 koff = 0; 3069 } else { 3070 klen = plsz; 3071 koff = P2ALIGN(off, (u_offset_t)klen); 3072 } 3073 ASSERT(koff <= filesz); 3074 if (koff + klen > filesz) 3075 klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff; 3076 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 3077 &io_len, koff, klen, 0); 3078 } 3079 if (pp == NULL) { 3080 /* 3081 * Some other thread entered the page before us. 3082 * Return to zfs_getpage to retry the lookup. 3083 */ 3084 *pl = NULL; 3085 return (0); 3086 } 3087 3088 /* 3089 * Fill the pages in the kluster. 3090 */ 3091 cur_pp = pp; 3092 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 3093 ASSERT(io_off == cur_pp->p_offset); 3094 va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 3095 err = dmu_read(os, oid, io_off, PAGESIZE, va); 3096 ppmapout(va); 3097 if (err) { 3098 /* On error, toss the entire kluster */ 3099 pvn_read_done(pp, B_ERROR); 3100 return (err); 3101 } 3102 cur_pp = cur_pp->p_next; 3103 } 3104 out: 3105 /* 3106 * Fill in the page list array from the kluster. If 3107 * there are too many pages in the kluster, return 3108 * as many pages as possible starting from the desired 3109 * offset `off'. 3110 * NOTE: the page list will always be null terminated. 3111 */ 3112 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 3113 3114 return (0); 3115 } 3116 3117 /* 3118 * Return pointers to the pages for the file region [off, off + len] 3119 * in the pl array. If plsz is greater than len, this function may 3120 * also return page pointers from before or after the specified 3121 * region (i.e. some region [off', off' + plsz]). These additional 3122 * pages are only returned if they are already in the cache, or were 3123 * created as part of a klustered read. 3124 * 3125 * IN: vp - vnode of file to get data from. 3126 * off - position in file to get data from. 3127 * len - amount of data to retrieve. 3128 * plsz - length of provided page list. 3129 * seg - segment to obtain pages for. 3130 * addr - virtual address of fault. 3131 * rw - mode of created pages. 3132 * cr - credentials of caller. 3133 * 3134 * OUT: protp - protection mode of created pages. 3135 * pl - list of pages created. 3136 * 3137 * RETURN: 0 if success 3138 * error code if failure 3139 * 3140 * Timestamps: 3141 * vp - atime updated 3142 */ 3143 /* ARGSUSED */ 3144 static int 3145 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 3146 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 3147 enum seg_rw rw, cred_t *cr) 3148 { 3149 znode_t *zp = VTOZ(vp); 3150 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3151 page_t *pp, **pl0 = pl; 3152 rl_t *rl; 3153 int cnt = 0, need_unlock = 0, err = 0; 3154 3155 ZFS_ENTER(zfsvfs); 3156 3157 if (protp) 3158 *protp = PROT_ALL; 3159 3160 ASSERT(zp->z_dbuf_held && zp->z_phys); 3161 3162 /* no faultahead (for now) */ 3163 if (pl == NULL) { 3164 ZFS_EXIT(zfsvfs); 3165 return (0); 3166 } 3167 3168 /* 3169 * Make sure nobody restructures the file in the middle of the getpage. 3170 */ 3171 rl = zfs_range_lock(zp, off, len, RL_READER); 3172 3173 /* can't fault past EOF */ 3174 if (off >= zp->z_phys->zp_size) { 3175 zfs_range_unlock(rl); 3176 ZFS_EXIT(zfsvfs); 3177 return (EFAULT); 3178 } 3179 3180 /* 3181 * If we already own the lock, then we must be page faulting 3182 * in the middle of a write to this file (i.e., we are writing 3183 * to this file using data from a mapped region of the file). 3184 */ 3185 if (!rw_owner(&zp->z_map_lock)) { 3186 rw_enter(&zp->z_map_lock, RW_WRITER); 3187 need_unlock = TRUE; 3188 } 3189 3190 /* 3191 * Loop through the requested range [off, off + len] looking 3192 * for pages. If we don't find a page, we will need to create 3193 * a new page and fill it with data from the file. 3194 */ 3195 while (len > 0) { 3196 if (plsz < PAGESIZE) 3197 break; 3198 if (pp = page_lookup(vp, off, SE_SHARED)) { 3199 *pl++ = pp; 3200 off += PAGESIZE; 3201 addr += PAGESIZE; 3202 len -= PAGESIZE; 3203 plsz -= PAGESIZE; 3204 } else { 3205 err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw); 3206 /* 3207 * klustering may have changed our region 3208 * to be block aligned. 3209 */ 3210 if (((pp = *pl) != 0) && (off != pp->p_offset)) { 3211 int delta = off - pp->p_offset; 3212 len += delta; 3213 off -= delta; 3214 addr -= delta; 3215 } 3216 while (*pl) { 3217 pl++; 3218 cnt++; 3219 off += PAGESIZE; 3220 addr += PAGESIZE; 3221 plsz -= PAGESIZE; 3222 if (len > PAGESIZE) 3223 len -= PAGESIZE; 3224 else 3225 len = 0; 3226 } 3227 if (err) { 3228 /* 3229 * Release any pages we have locked. 3230 */ 3231 while (pl > pl0) 3232 page_unlock(*--pl); 3233 goto out; 3234 } 3235 } 3236 } 3237 3238 /* 3239 * Fill out the page array with any pages already in the cache. 3240 */ 3241 while (plsz > 0) { 3242 pp = page_lookup_nowait(vp, off, SE_SHARED); 3243 if (pp == NULL) 3244 break; 3245 *pl++ = pp; 3246 off += PAGESIZE; 3247 plsz -= PAGESIZE; 3248 } 3249 3250 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3251 out: 3252 *pl = NULL; 3253 3254 if (need_unlock) 3255 rw_exit(&zp->z_map_lock); 3256 zfs_range_unlock(rl); 3257 3258 ZFS_EXIT(zfsvfs); 3259 return (err); 3260 } 3261 3262 /* 3263 * Request a memory map for a section of a file. This code interacts 3264 * with common code and the VM system as follows: 3265 * 3266 * common code calls mmap(), which ends up in smmap_common() 3267 * 3268 * this calls VOP_MAP(), which takes you into (say) zfs 3269 * 3270 * zfs_map() calls as_map(), passing segvn_create() as the callback 3271 * 3272 * segvn_create() creates the new segment and calls VOP_ADDMAP() 3273 * 3274 * zfs_addmap() updates z_mapcnt 3275 */ 3276 static int 3277 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 3278 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3279 { 3280 znode_t *zp = VTOZ(vp); 3281 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3282 segvn_crargs_t vn_a; 3283 int error; 3284 3285 ZFS_ENTER(zfsvfs); 3286 3287 if (vp->v_flag & VNOMAP) { 3288 ZFS_EXIT(zfsvfs); 3289 return (ENOSYS); 3290 } 3291 3292 if (off < 0 || len > MAXOFFSET_T - off) { 3293 ZFS_EXIT(zfsvfs); 3294 return (ENXIO); 3295 } 3296 3297 if (vp->v_type != VREG) { 3298 ZFS_EXIT(zfsvfs); 3299 return (ENODEV); 3300 } 3301 3302 /* 3303 * If file is locked, disallow mapping. 3304 */ 3305 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) { 3306 ZFS_EXIT(zfsvfs); 3307 return (EAGAIN); 3308 } 3309 3310 as_rangelock(as); 3311 if ((flags & MAP_FIXED) == 0) { 3312 map_addr(addrp, len, off, 1, flags); 3313 if (*addrp == NULL) { 3314 as_rangeunlock(as); 3315 ZFS_EXIT(zfsvfs); 3316 return (ENOMEM); 3317 } 3318 } else { 3319 /* 3320 * User specified address - blow away any previous mappings 3321 */ 3322 (void) as_unmap(as, *addrp, len); 3323 } 3324 3325 vn_a.vp = vp; 3326 vn_a.offset = (u_offset_t)off; 3327 vn_a.type = flags & MAP_TYPE; 3328 vn_a.prot = prot; 3329 vn_a.maxprot = maxprot; 3330 vn_a.cred = cr; 3331 vn_a.amp = NULL; 3332 vn_a.flags = flags & ~MAP_TYPE; 3333 vn_a.szc = 0; 3334 vn_a.lgrp_mem_policy_flags = 0; 3335 3336 error = as_map(as, *addrp, len, segvn_create, &vn_a); 3337 3338 as_rangeunlock(as); 3339 ZFS_EXIT(zfsvfs); 3340 return (error); 3341 } 3342 3343 /* ARGSUSED */ 3344 static int 3345 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3346 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3347 { 3348 uint64_t pages = btopr(len); 3349 3350 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 3351 return (0); 3352 } 3353 3354 /* 3355 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 3356 * more accurate mtime for the associated file. Since we don't have a way of 3357 * detecting when the data was actually modified, we have to resort to 3358 * heuristics. If an explicit msync() is done, then we mark the mtime when the 3359 * last page is pushed. The problem occurs when the msync() call is omitted, 3360 * which by far the most common case: 3361 * 3362 * open() 3363 * mmap() 3364 * <modify memory> 3365 * munmap() 3366 * close() 3367 * <time lapse> 3368 * putpage() via fsflush 3369 * 3370 * If we wait until fsflush to come along, we can have a modification time that 3371 * is some arbitrary point in the future. In order to prevent this in the 3372 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 3373 * torn down. 3374 */ 3375 /* ARGSUSED */ 3376 static int 3377 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3378 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr) 3379 { 3380 uint64_t pages = btopr(len); 3381 3382 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 3383 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 3384 3385 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 3386 vn_has_cached_data(vp)) 3387 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr); 3388 3389 return (0); 3390 } 3391 3392 /* 3393 * Free or allocate space in a file. Currently, this function only 3394 * supports the `F_FREESP' command. However, this command is somewhat 3395 * misnamed, as its functionality includes the ability to allocate as 3396 * well as free space. 3397 * 3398 * IN: vp - vnode of file to free data in. 3399 * cmd - action to take (only F_FREESP supported). 3400 * bfp - section of file to free/alloc. 3401 * flag - current file open mode flags. 3402 * offset - current file offset. 3403 * cr - credentials of caller [UNUSED]. 3404 * 3405 * RETURN: 0 if success 3406 * error code if failure 3407 * 3408 * Timestamps: 3409 * vp - ctime|mtime updated 3410 */ 3411 /* ARGSUSED */ 3412 static int 3413 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 3414 offset_t offset, cred_t *cr, caller_context_t *ct) 3415 { 3416 znode_t *zp = VTOZ(vp); 3417 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3418 uint64_t off, len; 3419 int error; 3420 3421 ZFS_ENTER(zfsvfs); 3422 3423 top: 3424 if (cmd != F_FREESP) { 3425 ZFS_EXIT(zfsvfs); 3426 return (EINVAL); 3427 } 3428 3429 if (error = convoff(vp, bfp, 0, offset)) { 3430 ZFS_EXIT(zfsvfs); 3431 return (error); 3432 } 3433 3434 if (bfp->l_len < 0) { 3435 ZFS_EXIT(zfsvfs); 3436 return (EINVAL); 3437 } 3438 3439 off = bfp->l_start; 3440 len = bfp->l_len; /* 0 means from off to end of file */ 3441 3442 do { 3443 error = zfs_freesp(zp, off, len, flag, TRUE); 3444 /* NB: we already did dmu_tx_wait() if necessary */ 3445 } while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 3446 3447 ZFS_EXIT(zfsvfs); 3448 return (error); 3449 } 3450 3451 static int 3452 zfs_fid(vnode_t *vp, fid_t *fidp) 3453 { 3454 znode_t *zp = VTOZ(vp); 3455 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3456 uint32_t gen = (uint32_t)zp->z_phys->zp_gen; 3457 uint64_t object = zp->z_id; 3458 zfid_short_t *zfid; 3459 int size, i; 3460 3461 ZFS_ENTER(zfsvfs); 3462 3463 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 3464 if (fidp->fid_len < size) { 3465 fidp->fid_len = size; 3466 ZFS_EXIT(zfsvfs); 3467 return (ENOSPC); 3468 } 3469 3470 zfid = (zfid_short_t *)fidp; 3471 3472 zfid->zf_len = size; 3473 3474 for (i = 0; i < sizeof (zfid->zf_object); i++) 3475 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 3476 3477 /* Must have a non-zero generation number to distinguish from .zfs */ 3478 if (gen == 0) 3479 gen = 1; 3480 for (i = 0; i < sizeof (zfid->zf_gen); i++) 3481 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 3482 3483 if (size == LONG_FID_LEN) { 3484 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 3485 zfid_long_t *zlfid; 3486 3487 zlfid = (zfid_long_t *)fidp; 3488 3489 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 3490 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 3491 3492 /* XXX - this should be the generation number for the objset */ 3493 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 3494 zlfid->zf_setgen[i] = 0; 3495 } 3496 3497 ZFS_EXIT(zfsvfs); 3498 return (0); 3499 } 3500 3501 static int 3502 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr) 3503 { 3504 znode_t *zp, *xzp; 3505 zfsvfs_t *zfsvfs; 3506 zfs_dirlock_t *dl; 3507 int error; 3508 3509 switch (cmd) { 3510 case _PC_LINK_MAX: 3511 *valp = ULONG_MAX; 3512 return (0); 3513 3514 case _PC_FILESIZEBITS: 3515 *valp = 64; 3516 return (0); 3517 3518 case _PC_XATTR_EXISTS: 3519 zp = VTOZ(vp); 3520 zfsvfs = zp->z_zfsvfs; 3521 ZFS_ENTER(zfsvfs); 3522 *valp = 0; 3523 error = zfs_dirent_lock(&dl, zp, "", &xzp, 3524 ZXATTR | ZEXISTS | ZSHARED); 3525 if (error == 0) { 3526 zfs_dirent_unlock(dl); 3527 if (!zfs_dirempty(xzp)) 3528 *valp = 1; 3529 VN_RELE(ZTOV(xzp)); 3530 } else if (error == ENOENT) { 3531 /* 3532 * If there aren't extended attributes, it's the 3533 * same as having zero of them. 3534 */ 3535 error = 0; 3536 } 3537 ZFS_EXIT(zfsvfs); 3538 return (error); 3539 3540 case _PC_ACL_ENABLED: 3541 *valp = _ACL_ACE_ENABLED; 3542 return (0); 3543 3544 case _PC_MIN_HOLE_SIZE: 3545 *valp = (ulong_t)SPA_MINBLOCKSIZE; 3546 return (0); 3547 3548 default: 3549 return (fs_pathconf(vp, cmd, valp, cr)); 3550 } 3551 } 3552 3553 /*ARGSUSED*/ 3554 static int 3555 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3556 { 3557 znode_t *zp = VTOZ(vp); 3558 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3559 int error; 3560 3561 ZFS_ENTER(zfsvfs); 3562 error = zfs_getacl(zp, vsecp, cr); 3563 ZFS_EXIT(zfsvfs); 3564 3565 return (error); 3566 } 3567 3568 /*ARGSUSED*/ 3569 static int 3570 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3571 { 3572 znode_t *zp = VTOZ(vp); 3573 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3574 int error; 3575 3576 ZFS_ENTER(zfsvfs); 3577 error = zfs_setacl(zp, vsecp, cr); 3578 ZFS_EXIT(zfsvfs); 3579 return (error); 3580 } 3581 3582 /* 3583 * Predeclare these here so that the compiler assumes that 3584 * this is an "old style" function declaration that does 3585 * not include arguments => we won't get type mismatch errors 3586 * in the initializations that follow. 3587 */ 3588 static int zfs_inval(); 3589 static int zfs_isdir(); 3590 3591 static int 3592 zfs_inval() 3593 { 3594 return (EINVAL); 3595 } 3596 3597 static int 3598 zfs_isdir() 3599 { 3600 return (EISDIR); 3601 } 3602 /* 3603 * Directory vnode operations template 3604 */ 3605 vnodeops_t *zfs_dvnodeops; 3606 const fs_operation_def_t zfs_dvnodeops_template[] = { 3607 VOPNAME_OPEN, zfs_open, 3608 VOPNAME_CLOSE, zfs_close, 3609 VOPNAME_READ, zfs_isdir, 3610 VOPNAME_WRITE, zfs_isdir, 3611 VOPNAME_IOCTL, zfs_ioctl, 3612 VOPNAME_GETATTR, zfs_getattr, 3613 VOPNAME_SETATTR, zfs_setattr, 3614 VOPNAME_ACCESS, zfs_access, 3615 VOPNAME_LOOKUP, zfs_lookup, 3616 VOPNAME_CREATE, zfs_create, 3617 VOPNAME_REMOVE, zfs_remove, 3618 VOPNAME_LINK, zfs_link, 3619 VOPNAME_RENAME, zfs_rename, 3620 VOPNAME_MKDIR, zfs_mkdir, 3621 VOPNAME_RMDIR, zfs_rmdir, 3622 VOPNAME_READDIR, zfs_readdir, 3623 VOPNAME_SYMLINK, zfs_symlink, 3624 VOPNAME_FSYNC, zfs_fsync, 3625 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3626 VOPNAME_FID, zfs_fid, 3627 VOPNAME_SEEK, zfs_seek, 3628 VOPNAME_PATHCONF, zfs_pathconf, 3629 VOPNAME_GETSECATTR, zfs_getsecattr, 3630 VOPNAME_SETSECATTR, zfs_setsecattr, 3631 NULL, NULL 3632 }; 3633 3634 /* 3635 * Regular file vnode operations template 3636 */ 3637 vnodeops_t *zfs_fvnodeops; 3638 const fs_operation_def_t zfs_fvnodeops_template[] = { 3639 VOPNAME_OPEN, zfs_open, 3640 VOPNAME_CLOSE, zfs_close, 3641 VOPNAME_READ, zfs_read, 3642 VOPNAME_WRITE, zfs_write, 3643 VOPNAME_IOCTL, zfs_ioctl, 3644 VOPNAME_GETATTR, zfs_getattr, 3645 VOPNAME_SETATTR, zfs_setattr, 3646 VOPNAME_ACCESS, zfs_access, 3647 VOPNAME_LOOKUP, zfs_lookup, 3648 VOPNAME_RENAME, zfs_rename, 3649 VOPNAME_FSYNC, zfs_fsync, 3650 VOPNAME_INACTIVE, (fs_generic_func_p)zfs_inactive, 3651 VOPNAME_FID, zfs_fid, 3652 VOPNAME_SEEK, zfs_seek, 3653 VOPNAME_FRLOCK, zfs_frlock, 3654 VOPNAME_SPACE, zfs_space, 3655 VOPNAME_GETPAGE, zfs_getpage, 3656 VOPNAME_PUTPAGE, zfs_putpage, 3657 VOPNAME_MAP, (fs_generic_func_p) zfs_map, 3658 VOPNAME_ADDMAP, (fs_generic_func_p) zfs_addmap, 3659 VOPNAME_DELMAP, zfs_delmap, 3660 VOPNAME_PATHCONF, zfs_pathconf, 3661 VOPNAME_GETSECATTR, zfs_getsecattr, 3662 VOPNAME_SETSECATTR, zfs_setsecattr, 3663 VOPNAME_VNEVENT, fs_vnevent_support, 3664 NULL, NULL 3665 }; 3666 3667 /* 3668 * Symbolic link vnode operations template 3669 */ 3670 vnodeops_t *zfs_symvnodeops; 3671 const fs_operation_def_t zfs_symvnodeops_template[] = { 3672 VOPNAME_GETATTR, zfs_getattr, 3673 VOPNAME_SETATTR, zfs_setattr, 3674 VOPNAME_ACCESS, zfs_access, 3675 VOPNAME_RENAME, zfs_rename, 3676 VOPNAME_READLINK, zfs_readlink, 3677 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3678 VOPNAME_FID, zfs_fid, 3679 VOPNAME_PATHCONF, zfs_pathconf, 3680 VOPNAME_VNEVENT, fs_vnevent_support, 3681 NULL, NULL 3682 }; 3683 3684 /* 3685 * Extended attribute directory vnode operations template 3686 * This template is identical to the directory vnodes 3687 * operation template except for restricted operations: 3688 * VOP_MKDIR() 3689 * VOP_SYMLINK() 3690 * Note that there are other restrictions embedded in: 3691 * zfs_create() - restrict type to VREG 3692 * zfs_link() - no links into/out of attribute space 3693 * zfs_rename() - no moves into/out of attribute space 3694 */ 3695 vnodeops_t *zfs_xdvnodeops; 3696 const fs_operation_def_t zfs_xdvnodeops_template[] = { 3697 VOPNAME_OPEN, zfs_open, 3698 VOPNAME_CLOSE, zfs_close, 3699 VOPNAME_IOCTL, zfs_ioctl, 3700 VOPNAME_GETATTR, zfs_getattr, 3701 VOPNAME_SETATTR, zfs_setattr, 3702 VOPNAME_ACCESS, zfs_access, 3703 VOPNAME_LOOKUP, zfs_lookup, 3704 VOPNAME_CREATE, zfs_create, 3705 VOPNAME_REMOVE, zfs_remove, 3706 VOPNAME_LINK, zfs_link, 3707 VOPNAME_RENAME, zfs_rename, 3708 VOPNAME_MKDIR, zfs_inval, 3709 VOPNAME_RMDIR, zfs_rmdir, 3710 VOPNAME_READDIR, zfs_readdir, 3711 VOPNAME_SYMLINK, zfs_inval, 3712 VOPNAME_FSYNC, zfs_fsync, 3713 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3714 VOPNAME_FID, zfs_fid, 3715 VOPNAME_SEEK, zfs_seek, 3716 VOPNAME_PATHCONF, zfs_pathconf, 3717 VOPNAME_GETSECATTR, zfs_getsecattr, 3718 VOPNAME_SETSECATTR, zfs_setsecattr, 3719 VOPNAME_VNEVENT, fs_vnevent_support, 3720 NULL, NULL 3721 }; 3722 3723 /* 3724 * Error vnode operations template 3725 */ 3726 vnodeops_t *zfs_evnodeops; 3727 const fs_operation_def_t zfs_evnodeops_template[] = { 3728 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3729 VOPNAME_PATHCONF, zfs_pathconf, 3730 NULL, NULL 3731 }; 3732