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) Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign(). 95 * In normal operation, this will be TXG_NOWAIT. During ZIL replay, 96 * it will be a specific txg. Either way, dmu_tx_assign() never blocks. 97 * This is critical because we don't want to block while holding locks. 98 * Note, in particular, that if a lock is sometimes acquired before 99 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 100 * use a non-blocking assign can deadlock the system. The scenario: 101 * 102 * Thread A has grabbed a lock before calling dmu_tx_assign(). 103 * Thread B is in an already-assigned tx, and blocks for this lock. 104 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 105 * forever, because the previous txg can't quiesce until B's tx commits. 106 * 107 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 108 * then drop all locks, call txg_wait_open(), and try again. 109 * 110 * (4) If the operation succeeded, generate the intent log entry for it 111 * before dropping locks. This ensures that the ordering of events 112 * in the intent log matches the order in which they actually occurred. 113 * 114 * (5) At the end of each vnode op, the DMU tx must always commit, 115 * regardless of whether there were any errors. 116 * 117 * (6) After dropping all locks, invoke zil_commit(zilog, seq, ioflag) 118 * to ensure that synchronous semantics are provided when necessary. 119 * 120 * In general, this is how things should be ordered in each vnode op: 121 * 122 * ZFS_ENTER(zfsvfs); // exit if unmounted 123 * top: 124 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 125 * rw_enter(...); // grab any other locks you need 126 * tx = dmu_tx_create(...); // get DMU tx 127 * dmu_tx_hold_*(); // hold each object you might modify 128 * error = dmu_tx_assign(tx, zfsvfs->z_assign); // try to assign 129 * if (error) { 130 * dmu_tx_abort(tx); // abort DMU tx 131 * rw_exit(...); // drop locks 132 * zfs_dirent_unlock(dl); // unlock directory entry 133 * VN_RELE(...); // release held vnodes 134 * if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 135 * txg_wait_open(dmu_objset_pool(os), 0); 136 * goto top; 137 * } 138 * ZFS_EXIT(zfsvfs); // finished in zfs 139 * return (error); // really out of space 140 * } 141 * error = do_real_work(); // do whatever this VOP does 142 * if (error == 0) 143 * seq = zfs_log_*(...); // on success, make ZIL entry 144 * dmu_tx_commit(tx); // commit DMU tx -- error or not 145 * rw_exit(...); // drop locks 146 * zfs_dirent_unlock(dl); // unlock directory entry 147 * VN_RELE(...); // release held vnodes 148 * zil_commit(zilog, seq, ioflag); // synchronous when necessary 149 * ZFS_EXIT(zfsvfs); // finished in zfs 150 * return (error); // done, report error 151 */ 152 153 /* ARGSUSED */ 154 static int 155 zfs_open(vnode_t **vpp, int flag, cred_t *cr) 156 { 157 return (0); 158 } 159 160 /* ARGSUSED */ 161 static int 162 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr) 163 { 164 /* 165 * Clean up any locks held by this process on the vp. 166 */ 167 cleanlocks(vp, ddi_get_pid(), 0); 168 cleanshares(vp, ddi_get_pid()); 169 170 return (0); 171 } 172 173 /* 174 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 175 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 176 */ 177 static int 178 zfs_holey(vnode_t *vp, int cmd, offset_t *off) 179 { 180 znode_t *zp = VTOZ(vp); 181 uint64_t noff = (uint64_t)*off; /* new offset */ 182 uint64_t file_sz; 183 int error; 184 boolean_t hole; 185 186 file_sz = zp->z_phys->zp_size; 187 if (noff >= file_sz) { 188 return (ENXIO); 189 } 190 191 if (cmd == _FIO_SEEK_HOLE) 192 hole = B_TRUE; 193 else 194 hole = B_FALSE; 195 196 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 197 198 /* end of file? */ 199 if ((error == ESRCH) || (noff > file_sz)) { 200 /* 201 * Handle the virtual hole at the end of file. 202 */ 203 if (hole) { 204 *off = file_sz; 205 return (0); 206 } 207 return (ENXIO); 208 } 209 210 if (noff < *off) 211 return (error); 212 *off = noff; 213 return (error); 214 } 215 216 /* ARGSUSED */ 217 static int 218 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred, 219 int *rvalp) 220 { 221 offset_t off; 222 int error; 223 zfsvfs_t *zfsvfs; 224 225 switch (com) { 226 case _FIOFFS: 227 return (zfs_sync(vp->v_vfsp, 0, cred)); 228 229 /* 230 * The following two ioctls are used by bfu. Faking out, 231 * necessary to avoid bfu errors. 232 */ 233 case _FIOGDIO: 234 case _FIOSDIO: 235 return (0); 236 237 case _FIO_SEEK_DATA: 238 case _FIO_SEEK_HOLE: 239 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 240 return (EFAULT); 241 242 zfsvfs = VTOZ(vp)->z_zfsvfs; 243 ZFS_ENTER(zfsvfs); 244 245 /* offset parameter is in/out */ 246 error = zfs_holey(vp, com, &off); 247 ZFS_EXIT(zfsvfs); 248 if (error) 249 return (error); 250 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 251 return (EFAULT); 252 return (0); 253 } 254 return (ENOTTY); 255 } 256 257 /* 258 * When a file is memory mapped, we must keep the IO data synchronized 259 * between the DMU cache and the memory mapped pages. What this means: 260 * 261 * On Write: If we find a memory mapped page, we write to *both* 262 * the page and the dmu buffer. 263 * 264 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 265 * the file is memory mapped. 266 */ 267 static int 268 mappedwrite(vnode_t *vp, uint64_t woff, int nbytes, uio_t *uio, dmu_tx_t *tx) 269 { 270 znode_t *zp = VTOZ(vp); 271 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 272 int64_t start, off; 273 int len = nbytes; 274 int error = 0; 275 276 start = uio->uio_loffset; 277 off = start & PAGEOFFSET; 278 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 279 page_t *pp; 280 uint64_t bytes = MIN(PAGESIZE - off, len); 281 282 /* 283 * We don't want a new page to "appear" in the middle of 284 * the file update (because it may not get the write 285 * update data), so we grab a lock to block 286 * zfs_getpage(). 287 */ 288 rw_enter(&zp->z_map_lock, RW_WRITER); 289 if (pp = page_lookup(vp, start, SE_SHARED)) { 290 caddr_t va; 291 292 rw_exit(&zp->z_map_lock); 293 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L); 294 error = uiomove(va+off, bytes, UIO_WRITE, uio); 295 if (error == 0) { 296 dmu_write(zfsvfs->z_os, zp->z_id, 297 woff, bytes, va+off, tx); 298 } 299 ppmapout(va); 300 page_unlock(pp); 301 } else { 302 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 303 woff, bytes, uio, tx); 304 rw_exit(&zp->z_map_lock); 305 } 306 len -= bytes; 307 woff += bytes; 308 off = 0; 309 if (error) 310 break; 311 } 312 return (error); 313 } 314 315 /* 316 * When a file is memory mapped, we must keep the IO data synchronized 317 * between the DMU cache and the memory mapped pages. What this means: 318 * 319 * On Read: We "read" preferentially from memory mapped pages, 320 * else we default from the dmu buffer. 321 * 322 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 323 * the file is memory mapped. 324 */ 325 static int 326 mappedread(vnode_t *vp, char *addr, int nbytes, uio_t *uio) 327 { 328 int64_t start, off, bytes; 329 int len = nbytes; 330 int error = 0; 331 332 start = uio->uio_loffset; 333 off = start & PAGEOFFSET; 334 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 335 page_t *pp; 336 337 bytes = MIN(PAGESIZE - off, len); 338 if (pp = page_lookup(vp, start, SE_SHARED)) { 339 caddr_t va; 340 341 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L); 342 error = uiomove(va + off, bytes, UIO_READ, uio); 343 ppmapout(va); 344 page_unlock(pp); 345 } else { 346 /* XXX use dmu_read here? */ 347 error = uiomove(addr, bytes, UIO_READ, uio); 348 } 349 len -= bytes; 350 addr += bytes; 351 off = 0; 352 if (error) 353 break; 354 } 355 return (error); 356 } 357 358 uint_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 359 360 /* 361 * Read bytes from specified file into supplied buffer. 362 * 363 * IN: vp - vnode of file to be read from. 364 * uio - structure supplying read location, range info, 365 * and return buffer. 366 * ioflag - SYNC flags; used to provide FRSYNC semantics. 367 * cr - credentials of caller. 368 * 369 * OUT: uio - updated offset and range, buffer filled. 370 * 371 * RETURN: 0 if success 372 * error code if failure 373 * 374 * Side Effects: 375 * vp - atime updated if byte count > 0 376 */ 377 /* ARGSUSED */ 378 static int 379 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 380 { 381 znode_t *zp = VTOZ(vp); 382 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 383 uint64_t delta; 384 ssize_t n, size, cnt, ndone; 385 int error, i, numbufs; 386 dmu_buf_t *dbp, **dbpp; 387 rl_t *rl; 388 389 ZFS_ENTER(zfsvfs); 390 391 /* 392 * Validate file offset 393 */ 394 if (uio->uio_loffset < (offset_t)0) { 395 ZFS_EXIT(zfsvfs); 396 return (EINVAL); 397 } 398 399 /* 400 * Fasttrack empty reads 401 */ 402 if (uio->uio_resid == 0) { 403 ZFS_EXIT(zfsvfs); 404 return (0); 405 } 406 407 /* 408 * Check for mandatory locks 409 */ 410 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 411 if (error = chklock(vp, FREAD, 412 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 413 ZFS_EXIT(zfsvfs); 414 return (error); 415 } 416 } 417 418 /* 419 * If we're in FRSYNC mode, sync out this znode before reading it. 420 */ 421 zil_commit(zfsvfs->z_log, zp->z_last_itx, ioflag & FRSYNC); 422 423 /* 424 * Lock the range against changes. 425 */ 426 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 427 428 /* 429 * If we are reading past end-of-file we can skip 430 * to the end; but we might still need to set atime. 431 */ 432 if (uio->uio_loffset >= zp->z_phys->zp_size) { 433 cnt = 0; 434 error = 0; 435 goto out; 436 } 437 438 cnt = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset); 439 440 for (ndone = 0; ndone < cnt; ndone += zfs_read_chunk_size) { 441 ASSERT(uio->uio_loffset < zp->z_phys->zp_size); 442 n = MIN(zfs_read_chunk_size, 443 zp->z_phys->zp_size - uio->uio_loffset); 444 n = MIN(n, cnt); 445 error = dmu_buf_hold_array(zfsvfs->z_os, zp->z_id, 446 uio->uio_loffset, n, TRUE, FTAG, &numbufs, &dbpp); 447 if (error) 448 goto out; 449 /* 450 * Compute the adjustment to align the dmu buffers 451 * with the uio buffer. 452 */ 453 delta = uio->uio_loffset - dbpp[0]->db_offset; 454 455 for (i = 0; i < numbufs; i++) { 456 if (n < 0) 457 break; 458 dbp = dbpp[i]; 459 size = dbp->db_size - delta; 460 /* 461 * XXX -- this is correct, but may be suboptimal. 462 * If the pages are all clean, we don't need to 463 * go through mappedread(). Maybe the VMODSORT 464 * stuff can help us here. 465 */ 466 if (vn_has_cached_data(vp)) { 467 error = mappedread(vp, (caddr_t)dbp->db_data + 468 delta, (n < size ? n : size), uio); 469 } else { 470 error = uiomove((caddr_t)dbp->db_data + delta, 471 (n < size ? n : size), UIO_READ, uio); 472 } 473 if (error) { 474 dmu_buf_rele_array(dbpp, numbufs, FTAG); 475 goto out; 476 } 477 n -= dbp->db_size; 478 if (delta) { 479 n += delta; 480 delta = 0; 481 } 482 } 483 dmu_buf_rele_array(dbpp, numbufs, FTAG); 484 } 485 out: 486 zfs_range_unlock(zp, rl); 487 488 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 489 ZFS_EXIT(zfsvfs); 490 return (error); 491 } 492 493 /* 494 * Fault in the pages of the first n bytes specified by the uio structure. 495 * 1 byte in each page is touched and the uio struct is unmodified. 496 * Any error will exit this routine as this is only a best 497 * attempt to get the pages resident. This is a copy of ufs_trans_touch(). 498 */ 499 static void 500 zfs_prefault_write(ssize_t n, struct uio *uio) 501 { 502 struct iovec *iov; 503 ulong_t cnt, incr; 504 caddr_t p; 505 uint8_t tmp; 506 507 iov = uio->uio_iov; 508 509 while (n) { 510 cnt = MIN(iov->iov_len, n); 511 if (cnt == 0) { 512 /* empty iov entry */ 513 iov++; 514 continue; 515 } 516 n -= cnt; 517 /* 518 * touch each page in this segment. 519 */ 520 p = iov->iov_base; 521 while (cnt) { 522 switch (uio->uio_segflg) { 523 case UIO_USERSPACE: 524 case UIO_USERISPACE: 525 if (fuword8(p, &tmp)) 526 return; 527 break; 528 case UIO_SYSSPACE: 529 if (kcopy(p, &tmp, 1)) 530 return; 531 break; 532 } 533 incr = MIN(cnt, PAGESIZE); 534 p += incr; 535 cnt -= incr; 536 } 537 /* 538 * touch the last byte in case it straddles a page. 539 */ 540 p--; 541 switch (uio->uio_segflg) { 542 case UIO_USERSPACE: 543 case UIO_USERISPACE: 544 if (fuword8(p, &tmp)) 545 return; 546 break; 547 case UIO_SYSSPACE: 548 if (kcopy(p, &tmp, 1)) 549 return; 550 break; 551 } 552 iov++; 553 } 554 } 555 556 /* 557 * Write the bytes to a file. 558 * 559 * IN: vp - vnode of file to be written to. 560 * uio - structure supplying write location, range info, 561 * and data buffer. 562 * ioflag - FAPPEND flag set if in append mode. 563 * cr - credentials of caller. 564 * 565 * OUT: uio - updated offset and range. 566 * 567 * RETURN: 0 if success 568 * error code if failure 569 * 570 * Timestamps: 571 * vp - ctime|mtime updated if byte count > 0 572 */ 573 /* ARGSUSED */ 574 static int 575 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 576 { 577 znode_t *zp = VTOZ(vp); 578 rlim64_t limit = uio->uio_llimit; 579 ssize_t start_resid = uio->uio_resid; 580 ssize_t tx_bytes; 581 uint64_t end_size; 582 dmu_tx_t *tx; 583 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 584 zilog_t *zilog = zfsvfs->z_log; 585 uint64_t seq = 0; 586 offset_t woff; 587 ssize_t n, nbytes; 588 rl_t *rl; 589 int max_blksz = zfsvfs->z_max_blksz; 590 int error; 591 592 /* 593 * Fasttrack empty write 594 */ 595 n = start_resid; 596 if (n == 0) 597 return (0); 598 599 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 600 limit = MAXOFFSET_T; 601 602 ZFS_ENTER(zfsvfs); 603 604 /* 605 * Pre-fault the initial pages to ensure slow (eg NFS) pages 606 * don't hold up txg. 607 */ 608 zfs_prefault_write(MIN(start_resid, SPA_MAXBLOCKSIZE), uio); 609 610 /* 611 * If in append mode, set the io offset pointer to eof. 612 */ 613 if (ioflag & FAPPEND) { 614 /* 615 * Range lock for a file append: 616 * The value for the start of range will be determined by 617 * zfs_range_lock() (to guarantee append semantics). 618 * If this write will cause the block size to increase, 619 * zfs_range_lock() will lock the entire file, so we must 620 * later reduce the range after we grow the block size. 621 */ 622 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 623 if (rl->r_len == UINT64_MAX) { 624 /* overlocked, zp_size can't change */ 625 woff = uio->uio_loffset = zp->z_phys->zp_size; 626 } else { 627 woff = uio->uio_loffset = rl->r_off; 628 } 629 } else { 630 woff = uio->uio_loffset; 631 /* 632 * Validate file offset 633 */ 634 if (woff < 0) { 635 ZFS_EXIT(zfsvfs); 636 return (EINVAL); 637 } 638 639 /* 640 * If we need to grow the block size then zfs_range_lock() 641 * will lock a wider range than we request here. 642 * Later after growing the block size we reduce the range. 643 */ 644 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 645 } 646 647 if (woff >= limit) { 648 error = EFBIG; 649 goto no_tx_done; 650 } 651 652 if ((woff + n) > limit || woff > (limit - n)) 653 n = limit - woff; 654 655 /* 656 * Check for mandatory locks 657 */ 658 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 659 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) 660 goto no_tx_done; 661 end_size = MAX(zp->z_phys->zp_size, woff + n); 662 top: 663 tx = dmu_tx_create(zfsvfs->z_os); 664 dmu_tx_hold_bonus(tx, zp->z_id); 665 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 666 error = dmu_tx_assign(tx, zfsvfs->z_assign); 667 if (error) { 668 dmu_tx_abort(tx); 669 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 670 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 671 goto top; 672 } 673 goto no_tx_done; 674 } 675 676 /* 677 * If zfs_range_lock() over-locked we grow the blocksize 678 * and then reduce the lock range. 679 */ 680 if (rl->r_len == UINT64_MAX) { 681 uint64_t new_blksz; 682 683 if (zp->z_blksz > max_blksz) { 684 ASSERT(!ISP2(zp->z_blksz)); 685 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 686 } else { 687 new_blksz = MIN(end_size, max_blksz); 688 } 689 zfs_grow_blocksize(zp, new_blksz, tx); 690 zfs_range_reduce(zp, rl, woff, n); 691 } 692 693 /* 694 * The file data does not fit in the znode "cache", so we 695 * will be writing to the file block data buffers. 696 * Each buffer will be written in a separate transaction; 697 * this keeps the intent log records small and allows us 698 * to do more fine-grained space accounting. 699 */ 700 while (n > 0) { 701 /* 702 * XXX - should we really limit each write to z_max_blksz? 703 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 704 */ 705 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 706 rw_enter(&zp->z_map_lock, RW_READER); 707 708 tx_bytes = uio->uio_resid; 709 if (vn_has_cached_data(vp)) { 710 rw_exit(&zp->z_map_lock); 711 error = mappedwrite(vp, woff, nbytes, uio, tx); 712 } else { 713 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 714 woff, nbytes, uio, tx); 715 rw_exit(&zp->z_map_lock); 716 } 717 tx_bytes -= uio->uio_resid; 718 719 if (error) { 720 /* XXX - do we need to "clean up" the dmu buffer? */ 721 break; 722 } 723 724 ASSERT(tx_bytes == nbytes); 725 726 /* 727 * Clear Set-UID/Set-GID bits on successful write if not 728 * privileged and at least one of the excute bits is set. 729 * 730 * It would be nice to to this after all writes have 731 * been done, but that would still expose the ISUID/ISGID 732 * to another app after the partial write is committed. 733 */ 734 735 mutex_enter(&zp->z_acl_lock); 736 if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) | 737 (S_IXUSR >> 6))) != 0 && 738 (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 && 739 secpolicy_vnode_setid_retain(cr, 740 (zp->z_phys->zp_mode & S_ISUID) != 0 && 741 zp->z_phys->zp_uid == 0) != 0) { 742 zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID); 743 } 744 mutex_exit(&zp->z_acl_lock); 745 746 n -= nbytes; 747 if (n <= 0) 748 break; 749 750 /* 751 * We have more work ahead of us, so wrap up this transaction 752 * and start another. Exact same logic as tx_done below. 753 */ 754 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) { 755 dmu_buf_will_dirty(zp->z_dbuf, tx); 756 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 757 uio->uio_loffset); 758 } 759 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 760 seq = zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, 761 ioflag, uio); 762 dmu_tx_commit(tx); 763 764 /* Pre-fault the next set of pages */ 765 zfs_prefault_write(MIN(n, SPA_MAXBLOCKSIZE), uio); 766 767 /* 768 * Start another transaction. 769 */ 770 woff = uio->uio_loffset; 771 tx = dmu_tx_create(zfsvfs->z_os); 772 dmu_tx_hold_bonus(tx, zp->z_id); 773 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 774 error = dmu_tx_assign(tx, zfsvfs->z_assign); 775 if (error) { 776 dmu_tx_abort(tx); 777 if (error == ERESTART && 778 zfsvfs->z_assign == TXG_NOWAIT) { 779 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 780 goto top; 781 } 782 goto no_tx_done; 783 } 784 } 785 786 tx_done: 787 788 if (tx_bytes != 0) { 789 /* 790 * Update the file size if it has changed; account 791 * for possible concurrent updates. 792 */ 793 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) { 794 dmu_buf_will_dirty(zp->z_dbuf, tx); 795 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 796 uio->uio_loffset); 797 } 798 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 799 seq = zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, 800 ioflag, uio); 801 } 802 dmu_tx_commit(tx); 803 804 805 no_tx_done: 806 807 zfs_range_unlock(zp, rl); 808 809 /* 810 * If we're in replay mode, or we made no progress, return error. 811 * Otherwise, it's at least a partial write, so it's successful. 812 */ 813 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 814 ZFS_EXIT(zfsvfs); 815 return (error); 816 } 817 818 zil_commit(zilog, seq, ioflag & (FSYNC | FDSYNC)); 819 820 ZFS_EXIT(zfsvfs); 821 return (0); 822 } 823 824 /* 825 * Get data to generate a TX_WRITE intent log record. 826 */ 827 int 828 zfs_get_data(void *arg, lr_write_t *lr, char *buf) 829 { 830 zfsvfs_t *zfsvfs = arg; 831 objset_t *os = zfsvfs->z_os; 832 znode_t *zp; 833 uint64_t off = lr->lr_offset; 834 rl_t *rl; 835 int dlen = lr->lr_length; /* length of user data */ 836 int error = 0; 837 838 ASSERT(dlen != 0); 839 840 /* 841 * Nothing to do if the file has been removed 842 */ 843 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 844 return (ENOENT); 845 if (zp->z_reap) { 846 VN_RELE(ZTOV(zp)); 847 return (ENOENT); 848 } 849 850 /* 851 * Write records come in two flavors: immediate and indirect. 852 * For small writes it's cheaper to store the data with the 853 * log record (immediate); for large writes it's cheaper to 854 * sync the data and get a pointer to it (indirect) so that 855 * we don't have to write the data twice. 856 */ 857 if (buf != NULL) { /* immediate write */ 858 dmu_buf_t *db; 859 860 rl = zfs_range_lock(zp, off, dlen, RL_READER); 861 /* test for truncation needs to be done while range locked */ 862 if (off >= zp->z_phys->zp_size) { 863 error = ENOENT; 864 goto out; 865 } 866 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, off, FTAG, &db)); 867 bcopy((char *)db->db_data + off - db->db_offset, buf, dlen); 868 dmu_buf_rele(db, FTAG); 869 } else { /* indirect write */ 870 uint64_t boff; /* block starting offset */ 871 872 /* 873 * Have to lock the whole block to ensure when it's 874 * written out and it's checksum is being calculated 875 * that no one can change the data. We need to re-check 876 * blocksize after we get the lock in case it's changed! 877 */ 878 for (;;) { 879 boff = off & ~(zp->z_blksz - 1); 880 dlen = zp->z_blksz; 881 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 882 if (zp->z_blksz == dlen) 883 break; 884 zfs_range_unlock(zp, rl); 885 } 886 /* test for truncation needs to be done while range locked */ 887 if (off >= zp->z_phys->zp_size) { 888 error = ENOENT; 889 goto out; 890 } 891 txg_suspend(dmu_objset_pool(os)); 892 error = dmu_sync(os, lr->lr_foid, off, &lr->lr_blkoff, 893 &lr->lr_blkptr, lr->lr_common.lrc_txg); 894 txg_resume(dmu_objset_pool(os)); 895 } 896 out: 897 zfs_range_unlock(zp, rl); 898 VN_RELE(ZTOV(zp)); 899 return (error); 900 } 901 902 /*ARGSUSED*/ 903 static int 904 zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr) 905 { 906 znode_t *zp = VTOZ(vp); 907 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 908 int error; 909 910 ZFS_ENTER(zfsvfs); 911 error = zfs_zaccess_rwx(zp, mode, cr); 912 ZFS_EXIT(zfsvfs); 913 return (error); 914 } 915 916 /* 917 * Lookup an entry in a directory, or an extended attribute directory. 918 * If it exists, return a held vnode reference for it. 919 * 920 * IN: dvp - vnode of directory to search. 921 * nm - name of entry to lookup. 922 * pnp - full pathname to lookup [UNUSED]. 923 * flags - LOOKUP_XATTR set if looking for an attribute. 924 * rdir - root directory vnode [UNUSED]. 925 * cr - credentials of caller. 926 * 927 * OUT: vpp - vnode of located entry, NULL if not found. 928 * 929 * RETURN: 0 if success 930 * error code if failure 931 * 932 * Timestamps: 933 * NA 934 */ 935 /* ARGSUSED */ 936 static int 937 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 938 int flags, vnode_t *rdir, cred_t *cr) 939 { 940 941 znode_t *zdp = VTOZ(dvp); 942 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 943 int error; 944 945 ZFS_ENTER(zfsvfs); 946 947 *vpp = NULL; 948 949 if (flags & LOOKUP_XATTR) { 950 /* 951 * We don't allow recursive attributes.. 952 * Maybe someday we will. 953 */ 954 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 955 ZFS_EXIT(zfsvfs); 956 return (EINVAL); 957 } 958 959 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr)) { 960 ZFS_EXIT(zfsvfs); 961 return (error); 962 } 963 964 /* 965 * Do we have permission to get into attribute directory? 966 */ 967 968 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) { 969 VN_RELE(*vpp); 970 } 971 972 ZFS_EXIT(zfsvfs); 973 return (error); 974 } 975 976 if (dvp->v_type != VDIR) { 977 ZFS_EXIT(zfsvfs); 978 return (ENOTDIR); 979 } 980 981 /* 982 * Check accessibility of directory. 983 */ 984 985 if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) { 986 ZFS_EXIT(zfsvfs); 987 return (error); 988 } 989 990 if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) { 991 992 /* 993 * Convert device special files 994 */ 995 if (IS_DEVVP(*vpp)) { 996 vnode_t *svp; 997 998 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 999 VN_RELE(*vpp); 1000 if (svp == NULL) 1001 error = ENOSYS; 1002 else 1003 *vpp = svp; 1004 } 1005 } 1006 1007 ZFS_EXIT(zfsvfs); 1008 return (error); 1009 } 1010 1011 /* 1012 * Attempt to create a new entry in a directory. If the entry 1013 * already exists, truncate the file if permissible, else return 1014 * an error. Return the vp of the created or trunc'd file. 1015 * 1016 * IN: dvp - vnode of directory to put new file entry in. 1017 * name - name of new file entry. 1018 * vap - attributes of new file. 1019 * excl - flag indicating exclusive or non-exclusive mode. 1020 * mode - mode to open file with. 1021 * cr - credentials of caller. 1022 * flag - large file flag [UNUSED]. 1023 * 1024 * OUT: vpp - vnode of created or trunc'd entry. 1025 * 1026 * RETURN: 0 if success 1027 * error code if failure 1028 * 1029 * Timestamps: 1030 * dvp - ctime|mtime updated if new entry created 1031 * vp - ctime|mtime always, atime if new 1032 */ 1033 /* ARGSUSED */ 1034 static int 1035 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1036 int mode, vnode_t **vpp, cred_t *cr, int flag) 1037 { 1038 znode_t *zp, *dzp = VTOZ(dvp); 1039 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1040 zilog_t *zilog = zfsvfs->z_log; 1041 uint64_t seq = 0; 1042 objset_t *os = zfsvfs->z_os; 1043 zfs_dirlock_t *dl; 1044 dmu_tx_t *tx; 1045 rl_t *rl; 1046 int error; 1047 uint64_t zoid; 1048 1049 ZFS_ENTER(zfsvfs); 1050 1051 top: 1052 *vpp = NULL; 1053 1054 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1055 vap->va_mode &= ~VSVTX; 1056 1057 if (*name == '\0') { 1058 /* 1059 * Null component name refers to the directory itself. 1060 */ 1061 VN_HOLD(dvp); 1062 zp = dzp; 1063 dl = NULL; 1064 error = 0; 1065 } else { 1066 /* possible VN_HOLD(zp) */ 1067 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) { 1068 if (strcmp(name, "..") == 0) 1069 error = EISDIR; 1070 ZFS_EXIT(zfsvfs); 1071 return (error); 1072 } 1073 } 1074 1075 zoid = zp ? zp->z_id : -1ULL; 1076 1077 if (zp == NULL) { 1078 /* 1079 * Create a new file object and update the directory 1080 * to reference it. 1081 */ 1082 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 1083 goto out; 1084 } 1085 1086 /* 1087 * We only support the creation of regular files in 1088 * extended attribute directories. 1089 */ 1090 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1091 (vap->va_type != VREG)) { 1092 error = EINVAL; 1093 goto out; 1094 } 1095 1096 tx = dmu_tx_create(os); 1097 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1098 dmu_tx_hold_bonus(tx, dzp->z_id); 1099 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1100 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1101 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1102 0, SPA_MAXBLOCKSIZE); 1103 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1104 if (error) { 1105 dmu_tx_abort(tx); 1106 zfs_dirent_unlock(dl); 1107 if (error == ERESTART && 1108 zfsvfs->z_assign == TXG_NOWAIT) { 1109 txg_wait_open(dmu_objset_pool(os), 0); 1110 goto top; 1111 } 1112 ZFS_EXIT(zfsvfs); 1113 return (error); 1114 } 1115 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1116 ASSERT(zp->z_id == zoid); 1117 (void) zfs_link_create(dl, zp, tx, ZNEW); 1118 seq = zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name); 1119 dmu_tx_commit(tx); 1120 } else { 1121 /* 1122 * A directory entry already exists for this name. 1123 */ 1124 /* 1125 * Can't truncate an existing file if in exclusive mode. 1126 */ 1127 if (excl == EXCL) { 1128 error = EEXIST; 1129 goto out; 1130 } 1131 /* 1132 * Can't open a directory for writing. 1133 */ 1134 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1135 error = EISDIR; 1136 goto out; 1137 } 1138 /* 1139 * Verify requested access to file. 1140 */ 1141 if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) { 1142 goto out; 1143 } 1144 /* 1145 * Truncate regular files if requested. 1146 */ 1147 1148 /* 1149 * Need to update dzp->z_seq? 1150 */ 1151 1152 mutex_enter(&dzp->z_lock); 1153 dzp->z_seq++; 1154 mutex_exit(&dzp->z_lock); 1155 1156 if ((ZTOV(zp)->v_type == VREG) && (zp->z_phys->zp_size != 0) && 1157 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1158 /* 1159 * Truncate the file. 1160 */ 1161 tx = dmu_tx_create(os); 1162 dmu_tx_hold_bonus(tx, zoid); 1163 dmu_tx_hold_free(tx, zoid, 0, DMU_OBJECT_END); 1164 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1165 if (error) { 1166 dmu_tx_abort(tx); 1167 if (dl) 1168 zfs_dirent_unlock(dl); 1169 VN_RELE(ZTOV(zp)); 1170 if (error == ERESTART && 1171 zfsvfs->z_assign == TXG_NOWAIT) { 1172 txg_wait_open(dmu_objset_pool(os), 0); 1173 goto top; 1174 } 1175 ZFS_EXIT(zfsvfs); 1176 return (error); 1177 } 1178 /* 1179 * Lock the whole range of the file 1180 */ 1181 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1182 error = zfs_freesp(zp, 0, 0, mode, tx, cr); 1183 if (error == 0) { 1184 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 1185 seq = zfs_log_truncate(zilog, tx, 1186 TX_TRUNCATE, zp, 0, 0); 1187 } 1188 zfs_range_unlock(zp, rl); 1189 dmu_tx_commit(tx); 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 dmu_tx_abort(tx); 1321 zfs_dirent_unlock(dl); 1322 VN_RELE(vp); 1323 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1324 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 1325 goto top; 1326 } 1327 ZFS_EXIT(zfsvfs); 1328 return (error); 1329 } 1330 1331 /* 1332 * Remove the directory entry. 1333 */ 1334 error = zfs_link_destroy(dl, zp, tx, 0, &reaped); 1335 1336 if (error) { 1337 dmu_tx_commit(tx); 1338 goto out; 1339 } 1340 1341 if (reaped) { 1342 mutex_enter(&vp->v_lock); 1343 delete_now = may_delete_now && 1344 vp->v_count == 1 && !vn_has_cached_data(vp) && 1345 zp->z_phys->zp_xattr == xattr_obj && 1346 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1347 mutex_exit(&vp->v_lock); 1348 } 1349 1350 if (delete_now) { 1351 if (zp->z_phys->zp_xattr) { 1352 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1353 ASSERT3U(error, ==, 0); 1354 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1355 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1356 mutex_enter(&xzp->z_lock); 1357 xzp->z_reap = 1; 1358 xzp->z_phys->zp_links = 0; 1359 mutex_exit(&xzp->z_lock); 1360 zfs_dq_add(xzp, tx); 1361 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1362 } 1363 mutex_enter(&zp->z_lock); 1364 mutex_enter(&vp->v_lock); 1365 vp->v_count--; 1366 ASSERT3U(vp->v_count, ==, 0); 1367 mutex_exit(&vp->v_lock); 1368 zp->z_active = 0; 1369 mutex_exit(&zp->z_lock); 1370 zfs_znode_delete(zp, tx); 1371 VFS_RELE(zfsvfs->z_vfs); 1372 } else if (reaped) { 1373 zfs_dq_add(zp, tx); 1374 } 1375 1376 seq = zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name); 1377 1378 dmu_tx_commit(tx); 1379 out: 1380 zfs_dirent_unlock(dl); 1381 1382 if (!delete_now) { 1383 VN_RELE(vp); 1384 } else if (xzp) { 1385 /* this rele delayed to prevent nesting transactions */ 1386 VN_RELE(ZTOV(xzp)); 1387 } 1388 1389 zil_commit(zilog, seq, 0); 1390 1391 ZFS_EXIT(zfsvfs); 1392 return (error); 1393 } 1394 1395 /* 1396 * Create a new directory and insert it into dvp using the name 1397 * provided. Return a pointer to the inserted directory. 1398 * 1399 * IN: dvp - vnode of directory to add subdir to. 1400 * dirname - name of new directory. 1401 * vap - attributes of new directory. 1402 * cr - credentials of caller. 1403 * 1404 * OUT: vpp - vnode of created directory. 1405 * 1406 * RETURN: 0 if success 1407 * error code if failure 1408 * 1409 * Timestamps: 1410 * dvp - ctime|mtime updated 1411 * vp - ctime|mtime|atime updated 1412 */ 1413 static int 1414 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr) 1415 { 1416 znode_t *zp, *dzp = VTOZ(dvp); 1417 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1418 zilog_t *zilog = zfsvfs->z_log; 1419 uint64_t seq = 0; 1420 zfs_dirlock_t *dl; 1421 uint64_t zoid = 0; 1422 dmu_tx_t *tx; 1423 int error; 1424 1425 ASSERT(vap->va_type == VDIR); 1426 1427 ZFS_ENTER(zfsvfs); 1428 1429 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1430 ZFS_EXIT(zfsvfs); 1431 return (EINVAL); 1432 } 1433 top: 1434 *vpp = NULL; 1435 1436 /* 1437 * First make sure the new directory doesn't exist. 1438 */ 1439 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) { 1440 ZFS_EXIT(zfsvfs); 1441 return (error); 1442 } 1443 1444 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) { 1445 zfs_dirent_unlock(dl); 1446 ZFS_EXIT(zfsvfs); 1447 return (error); 1448 } 1449 1450 /* 1451 * Add a new entry to the directory. 1452 */ 1453 tx = dmu_tx_create(zfsvfs->z_os); 1454 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1455 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1456 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1457 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1458 0, SPA_MAXBLOCKSIZE); 1459 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1460 if (error) { 1461 dmu_tx_abort(tx); 1462 zfs_dirent_unlock(dl); 1463 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1464 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 1465 goto top; 1466 } 1467 ZFS_EXIT(zfsvfs); 1468 return (error); 1469 } 1470 1471 /* 1472 * Create new node. 1473 */ 1474 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1475 1476 /* 1477 * Now put new name in parent dir. 1478 */ 1479 (void) zfs_link_create(dl, zp, tx, ZNEW); 1480 1481 *vpp = ZTOV(zp); 1482 1483 seq = zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname); 1484 dmu_tx_commit(tx); 1485 1486 zfs_dirent_unlock(dl); 1487 1488 zil_commit(zilog, seq, 0); 1489 1490 ZFS_EXIT(zfsvfs); 1491 return (0); 1492 } 1493 1494 /* 1495 * Remove a directory subdir entry. If the current working 1496 * directory is the same as the subdir to be removed, the 1497 * remove will fail. 1498 * 1499 * IN: dvp - vnode of directory to remove from. 1500 * name - name of directory to be removed. 1501 * cwd - vnode of current working directory. 1502 * cr - credentials of caller. 1503 * 1504 * RETURN: 0 if success 1505 * error code if failure 1506 * 1507 * Timestamps: 1508 * dvp - ctime|mtime updated 1509 */ 1510 static int 1511 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr) 1512 { 1513 znode_t *dzp = VTOZ(dvp); 1514 znode_t *zp; 1515 vnode_t *vp; 1516 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1517 zilog_t *zilog = zfsvfs->z_log; 1518 uint64_t seq = 0; 1519 zfs_dirlock_t *dl; 1520 dmu_tx_t *tx; 1521 int error; 1522 1523 ZFS_ENTER(zfsvfs); 1524 1525 top: 1526 zp = NULL; 1527 1528 /* 1529 * Attempt to lock directory; fail if entry doesn't exist. 1530 */ 1531 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1532 ZFS_EXIT(zfsvfs); 1533 return (error); 1534 } 1535 1536 vp = ZTOV(zp); 1537 1538 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1539 goto out; 1540 } 1541 1542 if (vp->v_type != VDIR) { 1543 error = ENOTDIR; 1544 goto out; 1545 } 1546 1547 if (vp == cwd) { 1548 error = EINVAL; 1549 goto out; 1550 } 1551 1552 vnevent_rmdir(vp); 1553 1554 /* 1555 * Grab a lock on the parent pointer make sure we play well 1556 * with the treewalk and directory rename code. 1557 */ 1558 rw_enter(&zp->z_parent_lock, RW_WRITER); 1559 1560 tx = dmu_tx_create(zfsvfs->z_os); 1561 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1562 dmu_tx_hold_bonus(tx, zp->z_id); 1563 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 1564 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1565 if (error) { 1566 dmu_tx_abort(tx); 1567 rw_exit(&zp->z_parent_lock); 1568 zfs_dirent_unlock(dl); 1569 VN_RELE(vp); 1570 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1571 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 1572 goto top; 1573 } 1574 ZFS_EXIT(zfsvfs); 1575 return (error); 1576 } 1577 1578 error = zfs_link_destroy(dl, zp, tx, 0, NULL); 1579 1580 if (error == 0) 1581 seq = zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name); 1582 1583 dmu_tx_commit(tx); 1584 1585 rw_exit(&zp->z_parent_lock); 1586 out: 1587 zfs_dirent_unlock(dl); 1588 1589 VN_RELE(vp); 1590 1591 zil_commit(zilog, seq, 0); 1592 1593 ZFS_EXIT(zfsvfs); 1594 return (error); 1595 } 1596 1597 /* 1598 * Read as many directory entries as will fit into the provided 1599 * buffer from the given directory cursor position (specified in 1600 * the uio structure. 1601 * 1602 * IN: vp - vnode of directory to read. 1603 * uio - structure supplying read location, range info, 1604 * and return buffer. 1605 * cr - credentials of caller. 1606 * 1607 * OUT: uio - updated offset and range, buffer filled. 1608 * eofp - set to true if end-of-file detected. 1609 * 1610 * RETURN: 0 if success 1611 * error code if failure 1612 * 1613 * Timestamps: 1614 * vp - atime updated 1615 * 1616 * Note that the low 4 bits of the cookie returned by zap is always zero. 1617 * This allows us to use the low range for "special" directory entries: 1618 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1619 * we use the offset 2 for the '.zfs' directory. 1620 */ 1621 /* ARGSUSED */ 1622 static int 1623 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp) 1624 { 1625 znode_t *zp = VTOZ(vp); 1626 iovec_t *iovp; 1627 dirent64_t *odp; 1628 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1629 objset_t *os; 1630 caddr_t outbuf; 1631 size_t bufsize; 1632 zap_cursor_t zc; 1633 zap_attribute_t zap; 1634 uint_t bytes_wanted; 1635 ushort_t this_reclen; 1636 uint64_t offset; /* must be unsigned; checks for < 1 */ 1637 off64_t *next; 1638 int local_eof; 1639 int outcount; 1640 int error; 1641 uint8_t prefetch; 1642 1643 ZFS_ENTER(zfsvfs); 1644 1645 /* 1646 * If we are not given an eof variable, 1647 * use a local one. 1648 */ 1649 if (eofp == NULL) 1650 eofp = &local_eof; 1651 1652 /* 1653 * Check for valid iov_len. 1654 */ 1655 if (uio->uio_iov->iov_len <= 0) { 1656 ZFS_EXIT(zfsvfs); 1657 return (EINVAL); 1658 } 1659 1660 /* 1661 * Quit if directory has been removed (posix) 1662 */ 1663 if ((*eofp = zp->z_reap) != 0) { 1664 ZFS_EXIT(zfsvfs); 1665 return (0); 1666 } 1667 1668 error = 0; 1669 os = zfsvfs->z_os; 1670 offset = uio->uio_loffset; 1671 prefetch = zp->z_zn_prefetch; 1672 1673 /* 1674 * Initialize the iterator cursor. 1675 */ 1676 if (offset <= 3) { 1677 /* 1678 * Start iteration from the beginning of the directory. 1679 */ 1680 zap_cursor_init(&zc, os, zp->z_id); 1681 } else { 1682 /* 1683 * The offset is a serialized cursor. 1684 */ 1685 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1686 } 1687 1688 /* 1689 * Get space to change directory entries into fs independent format. 1690 */ 1691 iovp = uio->uio_iov; 1692 bytes_wanted = iovp->iov_len; 1693 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1694 bufsize = bytes_wanted; 1695 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1696 odp = (struct dirent64 *)outbuf; 1697 } else { 1698 bufsize = bytes_wanted; 1699 odp = (struct dirent64 *)iovp->iov_base; 1700 } 1701 1702 /* 1703 * Transform to file-system independent format 1704 */ 1705 outcount = 0; 1706 while (outcount < bytes_wanted) { 1707 /* 1708 * Special case `.', `..', and `.zfs'. 1709 */ 1710 if (offset == 0) { 1711 (void) strcpy(zap.za_name, "."); 1712 zap.za_first_integer = zp->z_id; 1713 this_reclen = DIRENT64_RECLEN(1); 1714 } else if (offset == 1) { 1715 (void) strcpy(zap.za_name, ".."); 1716 zap.za_first_integer = zp->z_phys->zp_parent; 1717 this_reclen = DIRENT64_RECLEN(2); 1718 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1719 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1720 zap.za_first_integer = ZFSCTL_INO_ROOT; 1721 this_reclen = 1722 DIRENT64_RECLEN(sizeof (ZFS_CTLDIR_NAME) - 1); 1723 } else { 1724 /* 1725 * Grab next entry. 1726 */ 1727 if (error = zap_cursor_retrieve(&zc, &zap)) { 1728 if ((*eofp = (error == ENOENT)) != 0) 1729 break; 1730 else 1731 goto update; 1732 } 1733 1734 if (zap.za_integer_length != 8 || 1735 zap.za_num_integers != 1) { 1736 cmn_err(CE_WARN, "zap_readdir: bad directory " 1737 "entry, obj = %lld, offset = %lld\n", 1738 (u_longlong_t)zp->z_id, 1739 (u_longlong_t)offset); 1740 error = ENXIO; 1741 goto update; 1742 } 1743 this_reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1744 } 1745 1746 /* 1747 * Will this entry fit in the buffer? 1748 */ 1749 if (outcount + this_reclen > bufsize) { 1750 /* 1751 * Did we manage to fit anything in the buffer? 1752 */ 1753 if (!outcount) { 1754 error = EINVAL; 1755 goto update; 1756 } 1757 break; 1758 } 1759 /* 1760 * Add this entry: 1761 */ 1762 odp->d_ino = (ino64_t)zap.za_first_integer; 1763 odp->d_reclen = (ushort_t)this_reclen; 1764 /* NOTE: d_off is the offset for the *next* entry */ 1765 next = &(odp->d_off); 1766 (void) strncpy(odp->d_name, zap.za_name, 1767 DIRENT64_NAMELEN(this_reclen)); 1768 outcount += this_reclen; 1769 odp = (dirent64_t *)((intptr_t)odp + this_reclen); 1770 1771 ASSERT(outcount <= bufsize); 1772 1773 /* Prefetch znode */ 1774 if (prefetch) 1775 dmu_prefetch(os, zap.za_first_integer, 0, 0); 1776 1777 /* 1778 * Move to the next entry, fill in the previous offset. 1779 */ 1780 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1781 zap_cursor_advance(&zc); 1782 offset = zap_cursor_serialize(&zc); 1783 } else { 1784 offset += 1; 1785 } 1786 *next = offset; 1787 } 1788 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1789 1790 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 1791 iovp->iov_base += outcount; 1792 iovp->iov_len -= outcount; 1793 uio->uio_resid -= outcount; 1794 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 1795 /* 1796 * Reset the pointer. 1797 */ 1798 offset = uio->uio_loffset; 1799 } 1800 1801 update: 1802 zap_cursor_fini(&zc); 1803 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 1804 kmem_free(outbuf, bufsize); 1805 1806 if (error == ENOENT) 1807 error = 0; 1808 1809 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1810 1811 uio->uio_loffset = offset; 1812 ZFS_EXIT(zfsvfs); 1813 return (error); 1814 } 1815 1816 /* ARGSUSED */ 1817 static int 1818 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr) 1819 { 1820 znode_t *zp = VTOZ(vp); 1821 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1822 1823 ZFS_ENTER(zfsvfs); 1824 zil_commit(zfsvfs->z_log, zp->z_last_itx, FSYNC); 1825 ZFS_EXIT(zfsvfs); 1826 return (0); 1827 } 1828 1829 /* 1830 * Get the requested file attributes and place them in the provided 1831 * vattr structure. 1832 * 1833 * IN: vp - vnode of file. 1834 * vap - va_mask identifies requested attributes. 1835 * flags - [UNUSED] 1836 * cr - credentials of caller. 1837 * 1838 * OUT: vap - attribute values. 1839 * 1840 * RETURN: 0 (always succeeds) 1841 */ 1842 /* ARGSUSED */ 1843 static int 1844 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1845 { 1846 znode_t *zp = VTOZ(vp); 1847 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1848 znode_phys_t *pzp = zp->z_phys; 1849 int error; 1850 1851 ZFS_ENTER(zfsvfs); 1852 1853 /* 1854 * Return all attributes. It's cheaper to provide the answer 1855 * than to determine whether we were asked the question. 1856 */ 1857 mutex_enter(&zp->z_lock); 1858 1859 vap->va_type = vp->v_type; 1860 vap->va_mode = pzp->zp_mode & MODEMASK; 1861 vap->va_uid = zp->z_phys->zp_uid; 1862 vap->va_gid = zp->z_phys->zp_gid; 1863 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 1864 vap->va_nodeid = zp->z_id; 1865 vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX); /* nlink_t limit! */ 1866 vap->va_size = pzp->zp_size; 1867 vap->va_rdev = pzp->zp_rdev; 1868 vap->va_seq = zp->z_seq; 1869 1870 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 1871 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 1872 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 1873 1874 /* 1875 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 1876 * Also, if we are the owner don't bother, since owner should 1877 * always be allowed to read basic attributes of file. 1878 */ 1879 if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) && 1880 (zp->z_phys->zp_uid != crgetuid(cr))) { 1881 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) { 1882 mutex_exit(&zp->z_lock); 1883 ZFS_EXIT(zfsvfs); 1884 return (error); 1885 } 1886 } 1887 1888 mutex_exit(&zp->z_lock); 1889 1890 dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks); 1891 1892 if (zp->z_blksz == 0) { 1893 /* 1894 * Block size hasn't been set; suggest maximal I/O transfers. 1895 */ 1896 vap->va_blksize = zfsvfs->z_max_blksz; 1897 } 1898 1899 ZFS_EXIT(zfsvfs); 1900 return (0); 1901 } 1902 1903 /* 1904 * Set the file attributes to the values contained in the 1905 * vattr structure. 1906 * 1907 * IN: vp - vnode of file to be modified. 1908 * vap - new attribute values. 1909 * flags - ATTR_UTIME set if non-default time values provided. 1910 * cr - credentials of caller. 1911 * 1912 * RETURN: 0 if success 1913 * error code if failure 1914 * 1915 * Timestamps: 1916 * vp - ctime updated, mtime updated if size changed. 1917 */ 1918 /* ARGSUSED */ 1919 static int 1920 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 1921 caller_context_t *ct) 1922 { 1923 struct znode *zp = VTOZ(vp); 1924 znode_phys_t *pzp = zp->z_phys; 1925 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1926 zilog_t *zilog = zfsvfs->z_log; 1927 uint64_t seq = 0; 1928 dmu_tx_t *tx; 1929 rl_t *rl; 1930 uint_t mask = vap->va_mask; 1931 uint_t mask_applied = 0; 1932 vattr_t oldva; 1933 int trim_mask = FALSE; 1934 int saved_mask; 1935 uint64_t new_mode; 1936 znode_t *attrzp; 1937 int need_policy = FALSE; 1938 int err; 1939 1940 if (mask == 0) 1941 return (0); 1942 1943 if (mask & AT_NOSET) 1944 return (EINVAL); 1945 1946 if (mask & AT_SIZE && vp->v_type == VDIR) 1947 return (EISDIR); 1948 1949 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) 1950 return (EINVAL); 1951 1952 ZFS_ENTER(zfsvfs); 1953 1954 top: 1955 rl = NULL; 1956 attrzp = NULL; 1957 1958 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 1959 ZFS_EXIT(zfsvfs); 1960 return (EROFS); 1961 } 1962 1963 /* 1964 * First validate permissions 1965 */ 1966 1967 if (mask & AT_SIZE) { 1968 err = zfs_zaccess(zp, ACE_WRITE_DATA, cr); 1969 if (err) { 1970 ZFS_EXIT(zfsvfs); 1971 return (err); 1972 } 1973 } 1974 1975 if (mask & (AT_ATIME|AT_MTIME)) 1976 need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr); 1977 1978 if (mask & (AT_UID|AT_GID)) { 1979 int idmask = (mask & (AT_UID|AT_GID)); 1980 int take_owner; 1981 int take_group; 1982 1983 /* 1984 * NOTE: even if a new mode is being set, 1985 * we may clear S_ISUID/S_ISGID bits. 1986 */ 1987 1988 if (!(mask & AT_MODE)) 1989 vap->va_mode = pzp->zp_mode; 1990 1991 /* 1992 * Take ownership or chgrp to group we are a member of 1993 */ 1994 1995 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 1996 take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr); 1997 1998 /* 1999 * If both AT_UID and AT_GID are set then take_owner and 2000 * take_group must both be set in order to allow taking 2001 * ownership. 2002 * 2003 * Otherwise, send the check through secpolicy_vnode_setattr() 2004 * 2005 */ 2006 2007 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2008 ((idmask == AT_UID) && take_owner) || 2009 ((idmask == AT_GID) && take_group)) { 2010 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) { 2011 /* 2012 * Remove setuid/setgid for non-privileged users 2013 */ 2014 secpolicy_setid_clear(vap, cr); 2015 trim_mask = TRUE; 2016 saved_mask = vap->va_mask; 2017 } else { 2018 need_policy = TRUE; 2019 } 2020 } else { 2021 need_policy = TRUE; 2022 } 2023 } 2024 2025 if (mask & AT_MODE) 2026 need_policy = TRUE; 2027 2028 if (need_policy) { 2029 mutex_enter(&zp->z_lock); 2030 oldva.va_mode = pzp->zp_mode; 2031 oldva.va_uid = zp->z_phys->zp_uid; 2032 oldva.va_gid = zp->z_phys->zp_gid; 2033 mutex_exit(&zp->z_lock); 2034 2035 /* 2036 * If trim_mask is set then take ownership 2037 * has been granted. In that case remove 2038 * UID|GID from mask so that 2039 * secpolicy_vnode_setattr() doesn't revoke it. 2040 */ 2041 if (trim_mask) 2042 vap->va_mask &= ~(AT_UID|AT_GID); 2043 2044 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2045 (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp); 2046 if (err) { 2047 ZFS_EXIT(zfsvfs); 2048 return (err); 2049 } 2050 2051 if (trim_mask) 2052 vap->va_mask |= (saved_mask & (AT_UID|AT_GID)); 2053 } 2054 2055 /* 2056 * secpolicy_vnode_setattr, or take ownership may have 2057 * changed va_mask 2058 */ 2059 mask = vap->va_mask; 2060 2061 tx = dmu_tx_create(zfsvfs->z_os); 2062 dmu_tx_hold_bonus(tx, zp->z_id); 2063 2064 if (mask & AT_MODE) { 2065 uint64_t pmode = pzp->zp_mode; 2066 2067 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2068 2069 if (zp->z_phys->zp_acl.z_acl_extern_obj) 2070 dmu_tx_hold_write(tx, 2071 pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE); 2072 else 2073 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2074 0, ZFS_ACL_SIZE(MAX_ACL_SIZE)); 2075 } 2076 2077 if (mask & AT_SIZE) { 2078 uint64_t off = vap->va_size; 2079 /* 2080 * Range lock the entire file, to ensure the truncate 2081 * is serialised. 2082 */ 2083 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 2084 ASSERT(rl != NULL); 2085 if (off < zp->z_phys->zp_size) 2086 dmu_tx_hold_free(tx, zp->z_id, off, DMU_OBJECT_END); 2087 else if (zp->z_blksz < zfsvfs->z_max_blksz && off > zp->z_blksz) 2088 /* we will rewrite this block if we grow */ 2089 dmu_tx_hold_write(tx, zp->z_id, 0, zp->z_phys->zp_size); 2090 } 2091 2092 if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) { 2093 err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp); 2094 if (err) { 2095 dmu_tx_abort(tx); 2096 if (rl != NULL) 2097 zfs_range_unlock(zp, rl); 2098 ZFS_EXIT(zfsvfs); 2099 return (err); 2100 } 2101 dmu_tx_hold_bonus(tx, attrzp->z_id); 2102 } 2103 2104 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2105 if (err) { 2106 if (attrzp) 2107 VN_RELE(ZTOV(attrzp)); 2108 dmu_tx_abort(tx); 2109 if (rl != NULL) 2110 zfs_range_unlock(zp, rl); 2111 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2112 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2113 goto top; 2114 } 2115 ZFS_EXIT(zfsvfs); 2116 return (err); 2117 } 2118 2119 dmu_buf_will_dirty(zp->z_dbuf, tx); 2120 2121 /* 2122 * Set each attribute requested. 2123 * We group settings according to the locks they need to acquire. 2124 * 2125 * Note: you cannot set ctime directly, although it will be 2126 * updated as a side-effect of calling this function. 2127 */ 2128 if (mask & AT_SIZE) { 2129 /* 2130 * XXX - Note, we are not providing any open 2131 * mode flags here (like FNDELAY), so we may 2132 * block if there are locks present... this 2133 * should be addressed in openat(). 2134 */ 2135 err = zfs_freesp(zp, vap->va_size, 0, 0, tx, cr); 2136 if (err) { 2137 mutex_enter(&zp->z_lock); 2138 goto out; 2139 } 2140 mask_applied |= AT_SIZE; 2141 } 2142 2143 mask_applied = mask; /* no errors after this point */ 2144 2145 mutex_enter(&zp->z_lock); 2146 2147 if (mask & AT_MODE) { 2148 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2149 ASSERT3U(err, ==, 0); 2150 } 2151 2152 if (attrzp) 2153 mutex_enter(&attrzp->z_lock); 2154 2155 if (mask & AT_UID) { 2156 zp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2157 if (attrzp) { 2158 attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2159 } 2160 } 2161 2162 if (mask & AT_GID) { 2163 zp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2164 if (attrzp) 2165 attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2166 } 2167 2168 if (attrzp) 2169 mutex_exit(&attrzp->z_lock); 2170 2171 if (mask & AT_ATIME) 2172 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2173 2174 if (mask & AT_MTIME) 2175 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2176 2177 if (mask_applied & AT_SIZE) 2178 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2179 else if (mask_applied != 0) 2180 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2181 2182 out: 2183 2184 if (mask_applied != 0) 2185 seq = zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, 2186 mask_applied); 2187 2188 mutex_exit(&zp->z_lock); 2189 2190 if (attrzp) 2191 VN_RELE(ZTOV(attrzp)); 2192 2193 if (rl != NULL) 2194 zfs_range_unlock(zp, rl); 2195 2196 dmu_tx_commit(tx); 2197 2198 zil_commit(zilog, seq, 0); 2199 2200 ZFS_EXIT(zfsvfs); 2201 return (err); 2202 } 2203 2204 /* 2205 * Search back through the directory tree, using the ".." entries. 2206 * Lock each directory in the chain to prevent concurrent renames. 2207 * Fail any attempt to move a directory into one of its own descendants. 2208 * XXX - z_parent_lock can overlap with map or grow locks 2209 */ 2210 typedef struct zfs_zlock { 2211 krwlock_t *zl_rwlock; /* lock we acquired */ 2212 znode_t *zl_znode; /* znode we held */ 2213 struct zfs_zlock *zl_next; /* next in list */ 2214 } zfs_zlock_t; 2215 2216 static int 2217 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2218 { 2219 zfs_zlock_t *zl; 2220 znode_t *zp = tdzp; 2221 uint64_t rootid = zp->z_zfsvfs->z_root; 2222 uint64_t *oidp = &zp->z_id; 2223 krwlock_t *rwlp = &szp->z_parent_lock; 2224 krw_t rw = RW_WRITER; 2225 2226 /* 2227 * First pass write-locks szp and compares to zp->z_id. 2228 * Later passes read-lock zp and compare to zp->z_parent. 2229 */ 2230 do { 2231 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2232 zl->zl_rwlock = rwlp; 2233 zl->zl_znode = NULL; 2234 zl->zl_next = *zlpp; 2235 *zlpp = zl; 2236 2237 rw_enter(rwlp, rw); 2238 2239 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2240 return (EINVAL); 2241 2242 if (*oidp == rootid) /* We've hit the top */ 2243 return (0); 2244 2245 if (rw == RW_READER) { /* i.e. not the first pass */ 2246 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2247 if (error) 2248 return (error); 2249 zl->zl_znode = zp; 2250 } 2251 oidp = &zp->z_phys->zp_parent; 2252 rwlp = &zp->z_parent_lock; 2253 rw = RW_READER; 2254 2255 } while (zp->z_id != sdzp->z_id); 2256 2257 return (0); 2258 } 2259 2260 /* 2261 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2262 */ 2263 static void 2264 zfs_rename_unlock(zfs_zlock_t **zlpp) 2265 { 2266 zfs_zlock_t *zl; 2267 2268 while ((zl = *zlpp) != NULL) { 2269 if (zl->zl_znode != NULL) 2270 VN_RELE(ZTOV(zl->zl_znode)); 2271 rw_exit(zl->zl_rwlock); 2272 *zlpp = zl->zl_next; 2273 kmem_free(zl, sizeof (*zl)); 2274 } 2275 } 2276 2277 /* 2278 * Move an entry from the provided source directory to the target 2279 * directory. Change the entry name as indicated. 2280 * 2281 * IN: sdvp - Source directory containing the "old entry". 2282 * snm - Old entry name. 2283 * tdvp - Target directory to contain the "new entry". 2284 * tnm - New entry name. 2285 * cr - credentials of caller. 2286 * 2287 * RETURN: 0 if success 2288 * error code if failure 2289 * 2290 * Timestamps: 2291 * sdvp,tdvp - ctime|mtime updated 2292 */ 2293 static int 2294 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr) 2295 { 2296 znode_t *tdzp, *szp, *tzp; 2297 znode_t *sdzp = VTOZ(sdvp); 2298 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2299 zilog_t *zilog = zfsvfs->z_log; 2300 uint64_t seq = 0; 2301 vnode_t *realvp; 2302 zfs_dirlock_t *sdl, *tdl; 2303 dmu_tx_t *tx; 2304 zfs_zlock_t *zl; 2305 int cmp, serr, terr, error; 2306 2307 ZFS_ENTER(zfsvfs); 2308 2309 /* 2310 * Make sure we have the real vp for the target directory. 2311 */ 2312 if (VOP_REALVP(tdvp, &realvp) == 0) 2313 tdvp = realvp; 2314 2315 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2316 ZFS_EXIT(zfsvfs); 2317 return (EXDEV); 2318 } 2319 2320 tdzp = VTOZ(tdvp); 2321 top: 2322 szp = NULL; 2323 tzp = NULL; 2324 zl = NULL; 2325 2326 /* 2327 * This is to prevent the creation of links into attribute space 2328 * by renaming a linked file into/outof an attribute directory. 2329 * See the comment in zfs_link() for why this is considered bad. 2330 */ 2331 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2332 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2333 ZFS_EXIT(zfsvfs); 2334 return (EINVAL); 2335 } 2336 2337 /* 2338 * Lock source and target directory entries. To prevent deadlock, 2339 * a lock ordering must be defined. We lock the directory with 2340 * the smallest object id first, or if it's a tie, the one with 2341 * the lexically first name. 2342 */ 2343 if (sdzp->z_id < tdzp->z_id) { 2344 cmp = -1; 2345 } else if (sdzp->z_id > tdzp->z_id) { 2346 cmp = 1; 2347 } else { 2348 cmp = strcmp(snm, tnm); 2349 if (cmp == 0) { 2350 /* 2351 * POSIX: "If the old argument and the new argument 2352 * both refer to links to the same existing file, 2353 * the rename() function shall return successfully 2354 * and perform no other action." 2355 */ 2356 ZFS_EXIT(zfsvfs); 2357 return (0); 2358 } 2359 } 2360 if (cmp < 0) { 2361 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2362 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2363 } else { 2364 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2365 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2366 } 2367 2368 if (serr) { 2369 /* 2370 * Source entry invalid or not there. 2371 */ 2372 if (!terr) { 2373 zfs_dirent_unlock(tdl); 2374 if (tzp) 2375 VN_RELE(ZTOV(tzp)); 2376 } 2377 if (strcmp(snm, "..") == 0) 2378 serr = EINVAL; 2379 ZFS_EXIT(zfsvfs); 2380 return (serr); 2381 } 2382 if (terr) { 2383 zfs_dirent_unlock(sdl); 2384 VN_RELE(ZTOV(szp)); 2385 if (strcmp(tnm, "..") == 0) 2386 terr = EINVAL; 2387 ZFS_EXIT(zfsvfs); 2388 return (terr); 2389 } 2390 2391 /* 2392 * Must have write access at the source to remove the old entry 2393 * and write access at the target to create the new entry. 2394 * Note that if target and source are the same, this can be 2395 * done in a single check. 2396 */ 2397 2398 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2399 goto out; 2400 2401 if (ZTOV(szp)->v_type == VDIR) { 2402 /* 2403 * Check to make sure rename is valid. 2404 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2405 */ 2406 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 2407 goto out; 2408 } 2409 2410 /* 2411 * Does target exist? 2412 */ 2413 if (tzp) { 2414 /* 2415 * Source and target must be the same type. 2416 */ 2417 if (ZTOV(szp)->v_type == VDIR) { 2418 if (ZTOV(tzp)->v_type != VDIR) { 2419 error = ENOTDIR; 2420 goto out; 2421 } 2422 } else { 2423 if (ZTOV(tzp)->v_type == VDIR) { 2424 error = EISDIR; 2425 goto out; 2426 } 2427 } 2428 /* 2429 * POSIX dictates that when the source and target 2430 * entries refer to the same file object, rename 2431 * must do nothing and exit without error. 2432 */ 2433 if (szp->z_id == tzp->z_id) { 2434 error = 0; 2435 goto out; 2436 } 2437 } 2438 2439 vnevent_rename_src(ZTOV(szp)); 2440 if (tzp) 2441 vnevent_rename_dest(ZTOV(tzp)); 2442 2443 tx = dmu_tx_create(zfsvfs->z_os); 2444 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 2445 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 2446 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 2447 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 2448 if (sdzp != tdzp) 2449 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 2450 if (tzp) 2451 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 2452 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 2453 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2454 if (error) { 2455 dmu_tx_abort(tx); 2456 if (zl != NULL) 2457 zfs_rename_unlock(&zl); 2458 zfs_dirent_unlock(sdl); 2459 zfs_dirent_unlock(tdl); 2460 VN_RELE(ZTOV(szp)); 2461 if (tzp) 2462 VN_RELE(ZTOV(tzp)); 2463 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2464 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2465 goto top; 2466 } 2467 ZFS_EXIT(zfsvfs); 2468 return (error); 2469 } 2470 2471 if (tzp) /* Attempt to remove the existing target */ 2472 error = zfs_link_destroy(tdl, tzp, tx, 0, NULL); 2473 2474 if (error == 0) { 2475 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 2476 if (error == 0) { 2477 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 2478 ASSERT(error == 0); 2479 seq = zfs_log_rename(zilog, tx, TX_RENAME, 2480 sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp); 2481 } 2482 } 2483 2484 dmu_tx_commit(tx); 2485 out: 2486 if (zl != NULL) 2487 zfs_rename_unlock(&zl); 2488 2489 zfs_dirent_unlock(sdl); 2490 zfs_dirent_unlock(tdl); 2491 2492 VN_RELE(ZTOV(szp)); 2493 if (tzp) 2494 VN_RELE(ZTOV(tzp)); 2495 2496 zil_commit(zilog, seq, 0); 2497 2498 ZFS_EXIT(zfsvfs); 2499 return (error); 2500 } 2501 2502 /* 2503 * Insert the indicated symbolic reference entry into the directory. 2504 * 2505 * IN: dvp - Directory to contain new symbolic link. 2506 * link - Name for new symlink entry. 2507 * vap - Attributes of new entry. 2508 * target - Target path of new symlink. 2509 * cr - credentials of caller. 2510 * 2511 * RETURN: 0 if success 2512 * error code if failure 2513 * 2514 * Timestamps: 2515 * dvp - ctime|mtime updated 2516 */ 2517 static int 2518 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr) 2519 { 2520 znode_t *zp, *dzp = VTOZ(dvp); 2521 zfs_dirlock_t *dl; 2522 dmu_tx_t *tx; 2523 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2524 zilog_t *zilog = zfsvfs->z_log; 2525 uint64_t seq = 0; 2526 uint64_t zoid; 2527 int len = strlen(link); 2528 int error; 2529 2530 ASSERT(vap->va_type == VLNK); 2531 2532 ZFS_ENTER(zfsvfs); 2533 top: 2534 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2535 ZFS_EXIT(zfsvfs); 2536 return (error); 2537 } 2538 2539 if (len > MAXPATHLEN) { 2540 ZFS_EXIT(zfsvfs); 2541 return (ENAMETOOLONG); 2542 } 2543 2544 /* 2545 * Attempt to lock directory; fail if entry already exists. 2546 */ 2547 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) { 2548 ZFS_EXIT(zfsvfs); 2549 return (error); 2550 } 2551 2552 tx = dmu_tx_create(zfsvfs->z_os); 2553 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 2554 dmu_tx_hold_bonus(tx, dzp->z_id); 2555 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2556 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 2557 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 2558 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2559 if (error) { 2560 dmu_tx_abort(tx); 2561 zfs_dirent_unlock(dl); 2562 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2563 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2564 goto top; 2565 } 2566 ZFS_EXIT(zfsvfs); 2567 return (error); 2568 } 2569 2570 dmu_buf_will_dirty(dzp->z_dbuf, tx); 2571 2572 /* 2573 * Create a new object for the symlink. 2574 * Put the link content into bonus buffer if it will fit; 2575 * otherwise, store it just like any other file data. 2576 */ 2577 zoid = 0; 2578 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 2579 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len); 2580 if (len != 0) 2581 bcopy(link, zp->z_phys + 1, len); 2582 } else { 2583 dmu_buf_t *dbp; 2584 2585 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 2586 2587 /* 2588 * Nothing can access the znode yet so no locking needed 2589 * for growing the znode's blocksize. 2590 */ 2591 zfs_grow_blocksize(zp, len, tx); 2592 2593 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp)); 2594 dmu_buf_will_dirty(dbp, tx); 2595 2596 ASSERT3U(len, <=, dbp->db_size); 2597 bcopy(link, dbp->db_data, len); 2598 dmu_buf_rele(dbp, FTAG); 2599 } 2600 zp->z_phys->zp_size = len; 2601 2602 /* 2603 * Insert the new object into the directory. 2604 */ 2605 (void) zfs_link_create(dl, zp, tx, ZNEW); 2606 out: 2607 if (error == 0) 2608 seq = zfs_log_symlink(zilog, tx, TX_SYMLINK, 2609 dzp, zp, name, link); 2610 2611 dmu_tx_commit(tx); 2612 2613 zfs_dirent_unlock(dl); 2614 2615 VN_RELE(ZTOV(zp)); 2616 2617 zil_commit(zilog, seq, 0); 2618 2619 ZFS_EXIT(zfsvfs); 2620 return (error); 2621 } 2622 2623 /* 2624 * Return, in the buffer contained in the provided uio structure, 2625 * the symbolic path referred to by vp. 2626 * 2627 * IN: vp - vnode of symbolic link. 2628 * uoip - structure to contain the link path. 2629 * cr - credentials of caller. 2630 * 2631 * OUT: uio - structure to contain the link path. 2632 * 2633 * RETURN: 0 if success 2634 * error code if failure 2635 * 2636 * Timestamps: 2637 * vp - atime updated 2638 */ 2639 /* ARGSUSED */ 2640 static int 2641 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr) 2642 { 2643 znode_t *zp = VTOZ(vp); 2644 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2645 size_t bufsz; 2646 int error; 2647 2648 ZFS_ENTER(zfsvfs); 2649 2650 bufsz = (size_t)zp->z_phys->zp_size; 2651 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 2652 error = uiomove(zp->z_phys + 1, 2653 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2654 } else { 2655 dmu_buf_t *dbp; 2656 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 2657 if (error) { 2658 ZFS_EXIT(zfsvfs); 2659 return (error); 2660 } 2661 error = uiomove(dbp->db_data, 2662 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2663 dmu_buf_rele(dbp, FTAG); 2664 } 2665 2666 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2667 ZFS_EXIT(zfsvfs); 2668 return (error); 2669 } 2670 2671 /* 2672 * Insert a new entry into directory tdvp referencing svp. 2673 * 2674 * IN: tdvp - Directory to contain new entry. 2675 * svp - vnode of new entry. 2676 * name - name of new entry. 2677 * cr - credentials of caller. 2678 * 2679 * RETURN: 0 if success 2680 * error code if failure 2681 * 2682 * Timestamps: 2683 * tdvp - ctime|mtime updated 2684 * svp - ctime updated 2685 */ 2686 /* ARGSUSED */ 2687 static int 2688 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr) 2689 { 2690 znode_t *dzp = VTOZ(tdvp); 2691 znode_t *tzp, *szp; 2692 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2693 zilog_t *zilog = zfsvfs->z_log; 2694 uint64_t seq = 0; 2695 zfs_dirlock_t *dl; 2696 dmu_tx_t *tx; 2697 vnode_t *realvp; 2698 int error; 2699 2700 ASSERT(tdvp->v_type == VDIR); 2701 2702 ZFS_ENTER(zfsvfs); 2703 2704 if (VOP_REALVP(svp, &realvp) == 0) 2705 svp = realvp; 2706 2707 if (svp->v_vfsp != tdvp->v_vfsp) { 2708 ZFS_EXIT(zfsvfs); 2709 return (EXDEV); 2710 } 2711 2712 szp = VTOZ(svp); 2713 top: 2714 /* 2715 * We do not support links between attributes and non-attributes 2716 * because of the potential security risk of creating links 2717 * into "normal" file space in order to circumvent restrictions 2718 * imposed in attribute space. 2719 */ 2720 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 2721 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 2722 ZFS_EXIT(zfsvfs); 2723 return (EINVAL); 2724 } 2725 2726 /* 2727 * POSIX dictates that we return EPERM here. 2728 * Better choices include ENOTSUP or EISDIR. 2729 */ 2730 if (svp->v_type == VDIR) { 2731 ZFS_EXIT(zfsvfs); 2732 return (EPERM); 2733 } 2734 2735 if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) && 2736 secpolicy_basic_link(cr) != 0) { 2737 ZFS_EXIT(zfsvfs); 2738 return (EPERM); 2739 } 2740 2741 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2742 ZFS_EXIT(zfsvfs); 2743 return (error); 2744 } 2745 2746 /* 2747 * Attempt to lock directory; fail if entry already exists. 2748 */ 2749 if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) { 2750 ZFS_EXIT(zfsvfs); 2751 return (error); 2752 } 2753 2754 tx = dmu_tx_create(zfsvfs->z_os); 2755 dmu_tx_hold_bonus(tx, szp->z_id); 2756 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2757 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2758 if (error) { 2759 dmu_tx_abort(tx); 2760 zfs_dirent_unlock(dl); 2761 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2762 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2763 goto top; 2764 } 2765 ZFS_EXIT(zfsvfs); 2766 return (error); 2767 } 2768 2769 error = zfs_link_create(dl, szp, tx, 0); 2770 2771 if (error == 0) 2772 seq = zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name); 2773 2774 dmu_tx_commit(tx); 2775 2776 zfs_dirent_unlock(dl); 2777 2778 zil_commit(zilog, seq, 0); 2779 2780 ZFS_EXIT(zfsvfs); 2781 return (error); 2782 } 2783 2784 /* 2785 * zfs_null_putapage() is used when the file system has been force 2786 * unmounted. It just drops the pages. 2787 */ 2788 /* ARGSUSED */ 2789 static int 2790 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2791 size_t *lenp, int flags, cred_t *cr) 2792 { 2793 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 2794 return (0); 2795 } 2796 2797 /* ARGSUSED */ 2798 static int 2799 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2800 size_t *lenp, int flags, cred_t *cr) 2801 { 2802 znode_t *zp = VTOZ(vp); 2803 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2804 zilog_t *zilog = zfsvfs->z_log; 2805 dmu_tx_t *tx; 2806 rl_t *rl; 2807 u_offset_t off; 2808 ssize_t len; 2809 caddr_t va; 2810 int err; 2811 2812 top: 2813 off = pp->p_offset; 2814 rl = zfs_range_lock(zp, off, PAGESIZE, RL_WRITER); 2815 len = MIN(PAGESIZE, zp->z_phys->zp_size - off); 2816 2817 tx = dmu_tx_create(zfsvfs->z_os); 2818 dmu_tx_hold_write(tx, zp->z_id, off, len); 2819 dmu_tx_hold_bonus(tx, zp->z_id); 2820 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2821 if (err != 0) { 2822 dmu_tx_abort(tx); 2823 zfs_range_unlock(zp, rl); 2824 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2825 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2826 goto top; 2827 } 2828 goto out; 2829 } 2830 2831 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 2832 2833 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 2834 2835 ppmapout(va); 2836 2837 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 2838 (void) zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0, NULL); 2839 dmu_tx_commit(tx); 2840 2841 zfs_range_unlock(zp, rl); 2842 2843 pvn_write_done(pp, B_WRITE | flags); 2844 if (offp) 2845 *offp = off; 2846 if (lenp) 2847 *lenp = len; 2848 2849 out: 2850 return (err); 2851 } 2852 2853 /* 2854 * Copy the portion of the file indicated from pages into the file. 2855 * The pages are stored in a page list attached to the files vnode. 2856 * 2857 * IN: vp - vnode of file to push page data to. 2858 * off - position in file to put data. 2859 * len - amount of data to write. 2860 * flags - flags to control the operation. 2861 * cr - credentials of caller. 2862 * 2863 * RETURN: 0 if success 2864 * error code if failure 2865 * 2866 * Timestamps: 2867 * vp - ctime|mtime updated 2868 */ 2869 static int 2870 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr) 2871 { 2872 znode_t *zp = VTOZ(vp); 2873 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2874 page_t *pp; 2875 size_t io_len; 2876 u_offset_t io_off; 2877 uint64_t filesz; 2878 int error = 0; 2879 2880 ZFS_ENTER(zfsvfs); 2881 2882 ASSERT(zp->z_dbuf_held && zp->z_phys); 2883 2884 if (len == 0) { 2885 /* 2886 * Search the entire vp list for pages >= off. 2887 */ 2888 error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage, 2889 flags, cr); 2890 goto out; 2891 } 2892 2893 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 2894 if (off > filesz) { 2895 /* past end of file */ 2896 ZFS_EXIT(zfsvfs); 2897 return (0); 2898 } 2899 2900 len = MIN(len, filesz - off); 2901 2902 for (io_off = off; io_off < off + len; io_off += io_len) { 2903 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 2904 pp = page_lookup(vp, io_off, 2905 (flags & (B_INVAL | B_FREE)) ? 2906 SE_EXCL : SE_SHARED); 2907 } else { 2908 pp = page_lookup_nowait(vp, io_off, 2909 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 2910 } 2911 2912 if (pp != NULL && pvn_getdirty(pp, flags)) { 2913 int err; 2914 2915 /* 2916 * Found a dirty page to push 2917 */ 2918 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 2919 if (err) 2920 error = err; 2921 } else { 2922 io_len = PAGESIZE; 2923 } 2924 } 2925 out: 2926 zil_commit(zfsvfs->z_log, UINT64_MAX, (flags & B_ASYNC) ? 0 : FDSYNC); 2927 ZFS_EXIT(zfsvfs); 2928 return (error); 2929 } 2930 2931 void 2932 zfs_inactive(vnode_t *vp, cred_t *cr) 2933 { 2934 znode_t *zp = VTOZ(vp); 2935 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2936 int error; 2937 2938 rw_enter(&zfsvfs->z_um_lock, RW_READER); 2939 if (zfsvfs->z_unmounted2) { 2940 ASSERT(zp->z_dbuf_held == 0); 2941 2942 if (vn_has_cached_data(vp)) { 2943 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 2944 B_INVAL, cr); 2945 } 2946 2947 mutex_enter(&zp->z_lock); 2948 vp->v_count = 0; /* count arrives as 1 */ 2949 if (zp->z_dbuf == NULL) { 2950 mutex_exit(&zp->z_lock); 2951 zfs_znode_free(zp); 2952 } else { 2953 mutex_exit(&zp->z_lock); 2954 } 2955 rw_exit(&zfsvfs->z_um_lock); 2956 VFS_RELE(zfsvfs->z_vfs); 2957 return; 2958 } 2959 2960 /* 2961 * Attempt to push any data in the page cache. If this fails 2962 * we will get kicked out later in zfs_zinactive(). 2963 */ 2964 if (vn_has_cached_data(vp)) { 2965 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 2966 cr); 2967 } 2968 2969 if (zp->z_atime_dirty && zp->z_reap == 0) { 2970 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 2971 2972 dmu_tx_hold_bonus(tx, zp->z_id); 2973 error = dmu_tx_assign(tx, TXG_WAIT); 2974 if (error) { 2975 dmu_tx_abort(tx); 2976 } else { 2977 dmu_buf_will_dirty(zp->z_dbuf, tx); 2978 mutex_enter(&zp->z_lock); 2979 zp->z_atime_dirty = 0; 2980 mutex_exit(&zp->z_lock); 2981 dmu_tx_commit(tx); 2982 } 2983 } 2984 2985 zfs_zinactive(zp); 2986 rw_exit(&zfsvfs->z_um_lock); 2987 } 2988 2989 /* 2990 * Bounds-check the seek operation. 2991 * 2992 * IN: vp - vnode seeking within 2993 * ooff - old file offset 2994 * noffp - pointer to new file offset 2995 * 2996 * RETURN: 0 if success 2997 * EINVAL if new offset invalid 2998 */ 2999 /* ARGSUSED */ 3000 static int 3001 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp) 3002 { 3003 if (vp->v_type == VDIR) 3004 return (0); 3005 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 3006 } 3007 3008 /* 3009 * Pre-filter the generic locking function to trap attempts to place 3010 * a mandatory lock on a memory mapped file. 3011 */ 3012 static int 3013 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 3014 flk_callback_t *flk_cbp, cred_t *cr) 3015 { 3016 znode_t *zp = VTOZ(vp); 3017 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3018 int error; 3019 3020 ZFS_ENTER(zfsvfs); 3021 3022 /* 3023 * We are following the UFS semantics with respect to mapcnt 3024 * here: If we see that the file is mapped already, then we will 3025 * return an error, but we don't worry about races between this 3026 * function and zfs_map(). 3027 */ 3028 if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) { 3029 ZFS_EXIT(zfsvfs); 3030 return (EAGAIN); 3031 } 3032 error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr); 3033 ZFS_EXIT(zfsvfs); 3034 return (error); 3035 } 3036 3037 /* 3038 * If we can't find a page in the cache, we will create a new page 3039 * and fill it with file data. For efficiency, we may try to fill 3040 * multiple pages at once (klustering). 3041 */ 3042 static int 3043 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 3044 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 3045 { 3046 znode_t *zp = VTOZ(vp); 3047 page_t *pp, *cur_pp; 3048 objset_t *os = zp->z_zfsvfs->z_os; 3049 caddr_t va; 3050 u_offset_t io_off, total; 3051 uint64_t oid = zp->z_id; 3052 size_t io_len; 3053 uint64_t filesz; 3054 int err; 3055 3056 /* 3057 * If we are only asking for a single page don't bother klustering. 3058 */ 3059 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3060 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE || off > filesz) { 3061 io_off = off; 3062 io_len = PAGESIZE; 3063 pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr); 3064 } else { 3065 /* 3066 * Try to fill a kluster of pages (a blocks worth). 3067 */ 3068 size_t klen; 3069 u_offset_t koff; 3070 3071 if (!ISP2(zp->z_blksz)) { 3072 /* Only one block in the file. */ 3073 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3074 koff = 0; 3075 } else { 3076 klen = plsz; 3077 koff = P2ALIGN(off, (u_offset_t)klen); 3078 } 3079 if (klen > filesz) 3080 klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE); 3081 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 3082 &io_len, koff, klen, 0); 3083 } 3084 if (pp == NULL) { 3085 /* 3086 * Some other thread entered the page before us. 3087 * Return to zfs_getpage to retry the lookup. 3088 */ 3089 *pl = NULL; 3090 return (0); 3091 } 3092 3093 /* 3094 * Fill the pages in the kluster. 3095 */ 3096 cur_pp = pp; 3097 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 3098 ASSERT(io_off == cur_pp->p_offset); 3099 va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 3100 err = dmu_read(os, oid, io_off, PAGESIZE, va); 3101 ppmapout(va); 3102 if (err) { 3103 /* On error, toss the entire kluster */ 3104 pvn_read_done(pp, B_ERROR); 3105 return (err); 3106 } 3107 cur_pp = cur_pp->p_next; 3108 } 3109 out: 3110 /* 3111 * Fill in the page list array from the kluster. If 3112 * there are too many pages in the kluster, return 3113 * as many pages as possible starting from the desired 3114 * offset `off'. 3115 * NOTE: the page list will always be null terminated. 3116 */ 3117 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 3118 3119 return (0); 3120 } 3121 3122 /* 3123 * Return pointers to the pages for the file region [off, off + len] 3124 * in the pl array. If plsz is greater than len, this function may 3125 * also return page pointers from before or after the specified 3126 * region (i.e. some region [off', off' + plsz]). These additional 3127 * pages are only returned if they are already in the cache, or were 3128 * created as part of a klustered read. 3129 * 3130 * IN: vp - vnode of file to get data from. 3131 * off - position in file to get data from. 3132 * len - amount of data to retrieve. 3133 * plsz - length of provided page list. 3134 * seg - segment to obtain pages for. 3135 * addr - virtual address of fault. 3136 * rw - mode of created pages. 3137 * cr - credentials of caller. 3138 * 3139 * OUT: protp - protection mode of created pages. 3140 * pl - list of pages created. 3141 * 3142 * RETURN: 0 if success 3143 * error code if failure 3144 * 3145 * Timestamps: 3146 * vp - atime updated 3147 */ 3148 /* ARGSUSED */ 3149 static int 3150 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 3151 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 3152 enum seg_rw rw, cred_t *cr) 3153 { 3154 znode_t *zp = VTOZ(vp); 3155 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3156 page_t *pp, **pl0 = pl; 3157 rl_t *rl; 3158 int cnt = 0, need_unlock = 0, err = 0; 3159 3160 ZFS_ENTER(zfsvfs); 3161 3162 if (protp) 3163 *protp = PROT_ALL; 3164 3165 ASSERT(zp->z_dbuf_held && zp->z_phys); 3166 3167 /* no faultahead (for now) */ 3168 if (pl == NULL) { 3169 ZFS_EXIT(zfsvfs); 3170 return (0); 3171 } 3172 3173 /* 3174 * Make sure nobody restructures the file in the middle of the getpage. 3175 */ 3176 rl = zfs_range_lock(zp, off, len, RL_READER); 3177 3178 /* can't fault past EOF */ 3179 if (off >= zp->z_phys->zp_size) { 3180 zfs_range_unlock(zp, rl); 3181 ZFS_EXIT(zfsvfs); 3182 return (EFAULT); 3183 } 3184 3185 /* 3186 * If we already own the lock, then we must be page faulting 3187 * in the middle of a write to this file (i.e., we are writing 3188 * to this file using data from a mapped region of the file). 3189 */ 3190 if (!rw_owner(&zp->z_map_lock)) { 3191 rw_enter(&zp->z_map_lock, RW_WRITER); 3192 need_unlock = TRUE; 3193 } 3194 3195 /* 3196 * Loop through the requested range [off, off + len] looking 3197 * for pages. If we don't find a page, we will need to create 3198 * a new page and fill it with data from the file. 3199 */ 3200 while (len > 0) { 3201 if (plsz < PAGESIZE) 3202 break; 3203 if (pp = page_lookup(vp, off, SE_SHARED)) { 3204 *pl++ = pp; 3205 off += PAGESIZE; 3206 addr += PAGESIZE; 3207 len -= PAGESIZE; 3208 plsz -= PAGESIZE; 3209 } else { 3210 err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw); 3211 /* 3212 * klustering may have changed our region 3213 * to be block aligned. 3214 */ 3215 if (((pp = *pl) != 0) && (off != pp->p_offset)) { 3216 int delta = off - pp->p_offset; 3217 len += delta; 3218 off -= delta; 3219 addr -= delta; 3220 } 3221 while (*pl) { 3222 pl++; 3223 cnt++; 3224 off += PAGESIZE; 3225 addr += PAGESIZE; 3226 plsz -= PAGESIZE; 3227 if (len > PAGESIZE) 3228 len -= PAGESIZE; 3229 else 3230 len = 0; 3231 } 3232 if (err) { 3233 /* 3234 * Release any pages we have locked. 3235 */ 3236 while (pl > pl0) 3237 page_unlock(*--pl); 3238 goto out; 3239 } 3240 } 3241 } 3242 3243 /* 3244 * Fill out the page array with any pages already in the cache. 3245 */ 3246 while (plsz > 0) { 3247 pp = page_lookup_nowait(vp, off, SE_SHARED); 3248 if (pp == NULL) 3249 break; 3250 *pl++ = pp; 3251 off += PAGESIZE; 3252 plsz -= PAGESIZE; 3253 } 3254 3255 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3256 out: 3257 *pl = NULL; 3258 3259 if (need_unlock) 3260 rw_exit(&zp->z_map_lock); 3261 zfs_range_unlock(zp, rl); 3262 3263 ZFS_EXIT(zfsvfs); 3264 return (err); 3265 } 3266 3267 /* 3268 * Request a memory map for a section of a file. This code interacts 3269 * with common code and the VM system as follows: 3270 * 3271 * common code calls mmap(), which ends up in smmap_common() 3272 * 3273 * this calls VOP_MAP(), which takes you into (say) zfs 3274 * 3275 * zfs_map() calls as_map(), passing segvn_create() as the callback 3276 * 3277 * segvn_create() creates the new segment and calls VOP_ADDMAP() 3278 * 3279 * zfs_addmap() updates z_mapcnt 3280 */ 3281 static int 3282 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 3283 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3284 { 3285 znode_t *zp = VTOZ(vp); 3286 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3287 segvn_crargs_t vn_a; 3288 int error; 3289 3290 ZFS_ENTER(zfsvfs); 3291 3292 if (vp->v_flag & VNOMAP) { 3293 ZFS_EXIT(zfsvfs); 3294 return (ENOSYS); 3295 } 3296 3297 if (off < 0 || len > MAXOFFSET_T - off) { 3298 ZFS_EXIT(zfsvfs); 3299 return (ENXIO); 3300 } 3301 3302 if (vp->v_type != VREG) { 3303 ZFS_EXIT(zfsvfs); 3304 return (ENODEV); 3305 } 3306 3307 /* 3308 * If file is locked, disallow mapping. 3309 */ 3310 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) { 3311 ZFS_EXIT(zfsvfs); 3312 return (EAGAIN); 3313 } 3314 3315 as_rangelock(as); 3316 if ((flags & MAP_FIXED) == 0) { 3317 map_addr(addrp, len, off, 1, flags); 3318 if (*addrp == NULL) { 3319 as_rangeunlock(as); 3320 ZFS_EXIT(zfsvfs); 3321 return (ENOMEM); 3322 } 3323 } else { 3324 /* 3325 * User specified address - blow away any previous mappings 3326 */ 3327 (void) as_unmap(as, *addrp, len); 3328 } 3329 3330 vn_a.vp = vp; 3331 vn_a.offset = (u_offset_t)off; 3332 vn_a.type = flags & MAP_TYPE; 3333 vn_a.prot = prot; 3334 vn_a.maxprot = maxprot; 3335 vn_a.cred = cr; 3336 vn_a.amp = NULL; 3337 vn_a.flags = flags & ~MAP_TYPE; 3338 vn_a.szc = 0; 3339 vn_a.lgrp_mem_policy_flags = 0; 3340 3341 error = as_map(as, *addrp, len, segvn_create, &vn_a); 3342 3343 as_rangeunlock(as); 3344 ZFS_EXIT(zfsvfs); 3345 return (error); 3346 } 3347 3348 /* ARGSUSED */ 3349 static int 3350 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3351 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3352 { 3353 uint64_t pages = btopr(len); 3354 3355 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 3356 return (0); 3357 } 3358 3359 /* ARGSUSED */ 3360 static int 3361 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3362 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr) 3363 { 3364 uint64_t pages = btopr(len); 3365 3366 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 3367 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 3368 return (0); 3369 } 3370 3371 /* 3372 * Free or allocate space in a file. Currently, this function only 3373 * supports the `F_FREESP' command. However, this command is somewhat 3374 * misnamed, as its functionality includes the ability to allocate as 3375 * well as free space. 3376 * 3377 * IN: vp - vnode of file to free data in. 3378 * cmd - action to take (only F_FREESP supported). 3379 * bfp - section of file to free/alloc. 3380 * flag - current file open mode flags. 3381 * offset - current file offset. 3382 * cr - credentials of caller [UNUSED]. 3383 * 3384 * RETURN: 0 if success 3385 * error code if failure 3386 * 3387 * Timestamps: 3388 * vp - ctime|mtime updated 3389 */ 3390 /* ARGSUSED */ 3391 static int 3392 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 3393 offset_t offset, cred_t *cr, caller_context_t *ct) 3394 { 3395 dmu_tx_t *tx; 3396 znode_t *zp = VTOZ(vp); 3397 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3398 zilog_t *zilog = zfsvfs->z_log; 3399 rl_t *rl; 3400 uint64_t seq = 0; 3401 uint64_t off, len; 3402 int error; 3403 3404 ZFS_ENTER(zfsvfs); 3405 3406 top: 3407 if (cmd != F_FREESP) { 3408 ZFS_EXIT(zfsvfs); 3409 return (EINVAL); 3410 } 3411 3412 if (error = convoff(vp, bfp, 0, offset)) { 3413 ZFS_EXIT(zfsvfs); 3414 return (error); 3415 } 3416 3417 if (bfp->l_len < 0) { 3418 ZFS_EXIT(zfsvfs); 3419 return (EINVAL); 3420 } 3421 3422 off = bfp->l_start; 3423 len = bfp->l_len; /* 0 means from off to end of file */ 3424 tx = dmu_tx_create(zfsvfs->z_os); 3425 dmu_tx_hold_bonus(tx, zp->z_id); 3426 /* 3427 * If we will change zp_size (in zfs_freesp) then lock the whole file, 3428 * otherwise just lock the range being freed. 3429 */ 3430 if (len == 0 || off + len > zp->z_phys->zp_size) { 3431 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 3432 } else { 3433 rl = zfs_range_lock(zp, off, len, RL_WRITER); 3434 /* recheck, in case zp_size changed */ 3435 if (off + len > zp->z_phys->zp_size) { 3436 /* lost race: file size changed, lock whole file */ 3437 zfs_range_unlock(zp, rl); 3438 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 3439 } 3440 } 3441 3442 if (off + len > zp->z_blksz && zp->z_blksz < zfsvfs->z_max_blksz && 3443 off >= zp->z_phys->zp_size) { 3444 /* 3445 * We are increasing the length of the file, 3446 * and this may mean a block size increase. 3447 */ 3448 dmu_tx_hold_write(tx, zp->z_id, 0, 3449 MIN(off + len, zfsvfs->z_max_blksz)); 3450 } else if (off < zp->z_phys->zp_size) { 3451 /* 3452 * If len == 0, we are truncating the file. 3453 */ 3454 dmu_tx_hold_free(tx, zp->z_id, off, len ? len : DMU_OBJECT_END); 3455 } 3456 3457 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3458 if (error) { 3459 dmu_tx_abort(tx); 3460 zfs_range_unlock(zp, rl); 3461 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3462 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 3463 goto top; 3464 } 3465 ZFS_EXIT(zfsvfs); 3466 return (error); 3467 } 3468 3469 error = zfs_freesp(zp, off, len, flag, tx, cr); 3470 3471 if (error == 0) { 3472 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 3473 seq = zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 3474 } 3475 3476 zfs_range_unlock(zp, rl); 3477 3478 dmu_tx_commit(tx); 3479 3480 zil_commit(zilog, seq, 0); 3481 3482 ZFS_EXIT(zfsvfs); 3483 return (error); 3484 } 3485 3486 static int 3487 zfs_fid(vnode_t *vp, fid_t *fidp) 3488 { 3489 znode_t *zp = VTOZ(vp); 3490 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3491 uint32_t gen = (uint32_t)zp->z_phys->zp_gen; 3492 uint64_t object = zp->z_id; 3493 zfid_short_t *zfid; 3494 int size, i; 3495 3496 ZFS_ENTER(zfsvfs); 3497 3498 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 3499 if (fidp->fid_len < size) { 3500 fidp->fid_len = size; 3501 ZFS_EXIT(zfsvfs); 3502 return (ENOSPC); 3503 } 3504 3505 zfid = (zfid_short_t *)fidp; 3506 3507 zfid->zf_len = size; 3508 3509 for (i = 0; i < sizeof (zfid->zf_object); i++) 3510 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 3511 3512 /* Must have a non-zero generation number to distinguish from .zfs */ 3513 if (gen == 0) 3514 gen = 1; 3515 for (i = 0; i < sizeof (zfid->zf_gen); i++) 3516 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 3517 3518 if (size == LONG_FID_LEN) { 3519 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 3520 zfid_long_t *zlfid; 3521 3522 zlfid = (zfid_long_t *)fidp; 3523 3524 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 3525 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 3526 3527 /* XXX - this should be the generation number for the objset */ 3528 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 3529 zlfid->zf_setgen[i] = 0; 3530 } 3531 3532 ZFS_EXIT(zfsvfs); 3533 return (0); 3534 } 3535 3536 static int 3537 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr) 3538 { 3539 znode_t *zp, *xzp; 3540 zfsvfs_t *zfsvfs; 3541 zfs_dirlock_t *dl; 3542 int error; 3543 3544 switch (cmd) { 3545 case _PC_LINK_MAX: 3546 *valp = ULONG_MAX; 3547 return (0); 3548 3549 case _PC_FILESIZEBITS: 3550 *valp = 64; 3551 return (0); 3552 3553 case _PC_XATTR_EXISTS: 3554 zp = VTOZ(vp); 3555 zfsvfs = zp->z_zfsvfs; 3556 ZFS_ENTER(zfsvfs); 3557 *valp = 0; 3558 error = zfs_dirent_lock(&dl, zp, "", &xzp, 3559 ZXATTR | ZEXISTS | ZSHARED); 3560 if (error == 0) { 3561 zfs_dirent_unlock(dl); 3562 if (!zfs_dirempty(xzp)) 3563 *valp = 1; 3564 VN_RELE(ZTOV(xzp)); 3565 } else if (error == ENOENT) { 3566 /* 3567 * If there aren't extended attributes, it's the 3568 * same as having zero of them. 3569 */ 3570 error = 0; 3571 } 3572 ZFS_EXIT(zfsvfs); 3573 return (error); 3574 3575 case _PC_ACL_ENABLED: 3576 *valp = _ACL_ACE_ENABLED; 3577 return (0); 3578 3579 case _PC_MIN_HOLE_SIZE: 3580 *valp = (ulong_t)SPA_MINBLOCKSIZE; 3581 return (0); 3582 3583 default: 3584 return (fs_pathconf(vp, cmd, valp, cr)); 3585 } 3586 } 3587 3588 /*ARGSUSED*/ 3589 static int 3590 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3591 { 3592 znode_t *zp = VTOZ(vp); 3593 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3594 int error; 3595 3596 ZFS_ENTER(zfsvfs); 3597 error = zfs_getacl(zp, vsecp, cr); 3598 ZFS_EXIT(zfsvfs); 3599 3600 return (error); 3601 } 3602 3603 /*ARGSUSED*/ 3604 static int 3605 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3606 { 3607 znode_t *zp = VTOZ(vp); 3608 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3609 int error; 3610 3611 ZFS_ENTER(zfsvfs); 3612 error = zfs_setacl(zp, vsecp, cr); 3613 ZFS_EXIT(zfsvfs); 3614 return (error); 3615 } 3616 3617 /* 3618 * Predeclare these here so that the compiler assumes that 3619 * this is an "old style" function declaration that does 3620 * not include arguments => we won't get type mismatch errors 3621 * in the initializations that follow. 3622 */ 3623 static int zfs_inval(); 3624 static int zfs_isdir(); 3625 3626 static int 3627 zfs_inval() 3628 { 3629 return (EINVAL); 3630 } 3631 3632 static int 3633 zfs_isdir() 3634 { 3635 return (EISDIR); 3636 } 3637 /* 3638 * Directory vnode operations template 3639 */ 3640 vnodeops_t *zfs_dvnodeops; 3641 const fs_operation_def_t zfs_dvnodeops_template[] = { 3642 VOPNAME_OPEN, zfs_open, 3643 VOPNAME_CLOSE, zfs_close, 3644 VOPNAME_READ, zfs_isdir, 3645 VOPNAME_WRITE, zfs_isdir, 3646 VOPNAME_IOCTL, zfs_ioctl, 3647 VOPNAME_GETATTR, zfs_getattr, 3648 VOPNAME_SETATTR, zfs_setattr, 3649 VOPNAME_ACCESS, zfs_access, 3650 VOPNAME_LOOKUP, zfs_lookup, 3651 VOPNAME_CREATE, zfs_create, 3652 VOPNAME_REMOVE, zfs_remove, 3653 VOPNAME_LINK, zfs_link, 3654 VOPNAME_RENAME, zfs_rename, 3655 VOPNAME_MKDIR, zfs_mkdir, 3656 VOPNAME_RMDIR, zfs_rmdir, 3657 VOPNAME_READDIR, zfs_readdir, 3658 VOPNAME_SYMLINK, zfs_symlink, 3659 VOPNAME_FSYNC, zfs_fsync, 3660 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3661 VOPNAME_FID, zfs_fid, 3662 VOPNAME_SEEK, zfs_seek, 3663 VOPNAME_PATHCONF, zfs_pathconf, 3664 VOPNAME_GETSECATTR, zfs_getsecattr, 3665 VOPNAME_SETSECATTR, zfs_setsecattr, 3666 NULL, NULL 3667 }; 3668 3669 /* 3670 * Regular file vnode operations template 3671 */ 3672 vnodeops_t *zfs_fvnodeops; 3673 const fs_operation_def_t zfs_fvnodeops_template[] = { 3674 VOPNAME_OPEN, zfs_open, 3675 VOPNAME_CLOSE, zfs_close, 3676 VOPNAME_READ, zfs_read, 3677 VOPNAME_WRITE, zfs_write, 3678 VOPNAME_IOCTL, zfs_ioctl, 3679 VOPNAME_GETATTR, zfs_getattr, 3680 VOPNAME_SETATTR, zfs_setattr, 3681 VOPNAME_ACCESS, zfs_access, 3682 VOPNAME_LOOKUP, zfs_lookup, 3683 VOPNAME_RENAME, zfs_rename, 3684 VOPNAME_FSYNC, zfs_fsync, 3685 VOPNAME_INACTIVE, (fs_generic_func_p)zfs_inactive, 3686 VOPNAME_FID, zfs_fid, 3687 VOPNAME_SEEK, zfs_seek, 3688 VOPNAME_FRLOCK, zfs_frlock, 3689 VOPNAME_SPACE, zfs_space, 3690 VOPNAME_GETPAGE, zfs_getpage, 3691 VOPNAME_PUTPAGE, zfs_putpage, 3692 VOPNAME_MAP, (fs_generic_func_p) zfs_map, 3693 VOPNAME_ADDMAP, (fs_generic_func_p) zfs_addmap, 3694 VOPNAME_DELMAP, zfs_delmap, 3695 VOPNAME_PATHCONF, zfs_pathconf, 3696 VOPNAME_GETSECATTR, zfs_getsecattr, 3697 VOPNAME_SETSECATTR, zfs_setsecattr, 3698 VOPNAME_VNEVENT, fs_vnevent_support, 3699 NULL, NULL 3700 }; 3701 3702 /* 3703 * Symbolic link vnode operations template 3704 */ 3705 vnodeops_t *zfs_symvnodeops; 3706 const fs_operation_def_t zfs_symvnodeops_template[] = { 3707 VOPNAME_GETATTR, zfs_getattr, 3708 VOPNAME_SETATTR, zfs_setattr, 3709 VOPNAME_ACCESS, zfs_access, 3710 VOPNAME_RENAME, zfs_rename, 3711 VOPNAME_READLINK, zfs_readlink, 3712 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3713 VOPNAME_FID, zfs_fid, 3714 VOPNAME_PATHCONF, zfs_pathconf, 3715 VOPNAME_VNEVENT, fs_vnevent_support, 3716 NULL, NULL 3717 }; 3718 3719 /* 3720 * Extended attribute directory vnode operations template 3721 * This template is identical to the directory vnodes 3722 * operation template except for restricted operations: 3723 * VOP_MKDIR() 3724 * VOP_SYMLINK() 3725 * Note that there are other restrictions embedded in: 3726 * zfs_create() - restrict type to VREG 3727 * zfs_link() - no links into/out of attribute space 3728 * zfs_rename() - no moves into/out of attribute space 3729 */ 3730 vnodeops_t *zfs_xdvnodeops; 3731 const fs_operation_def_t zfs_xdvnodeops_template[] = { 3732 VOPNAME_OPEN, zfs_open, 3733 VOPNAME_CLOSE, zfs_close, 3734 VOPNAME_IOCTL, zfs_ioctl, 3735 VOPNAME_GETATTR, zfs_getattr, 3736 VOPNAME_SETATTR, zfs_setattr, 3737 VOPNAME_ACCESS, zfs_access, 3738 VOPNAME_LOOKUP, zfs_lookup, 3739 VOPNAME_CREATE, zfs_create, 3740 VOPNAME_REMOVE, zfs_remove, 3741 VOPNAME_LINK, zfs_link, 3742 VOPNAME_RENAME, zfs_rename, 3743 VOPNAME_MKDIR, zfs_inval, 3744 VOPNAME_RMDIR, zfs_rmdir, 3745 VOPNAME_READDIR, zfs_readdir, 3746 VOPNAME_SYMLINK, zfs_inval, 3747 VOPNAME_FSYNC, zfs_fsync, 3748 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3749 VOPNAME_FID, zfs_fid, 3750 VOPNAME_SEEK, zfs_seek, 3751 VOPNAME_PATHCONF, zfs_pathconf, 3752 VOPNAME_GETSECATTR, zfs_getsecattr, 3753 VOPNAME_SETSECATTR, zfs_setsecattr, 3754 VOPNAME_VNEVENT, fs_vnevent_support, 3755 NULL, NULL 3756 }; 3757 3758 /* 3759 * Error vnode operations template 3760 */ 3761 vnodeops_t *zfs_evnodeops; 3762 const fs_operation_def_t zfs_evnodeops_template[] = { 3763 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3764 VOPNAME_PATHCONF, zfs_pathconf, 3765 NULL, NULL 3766 }; 3767